tial misreporting including non-registered businesses, multiple businesses at residential

Issuer: Federal Reserve Bank of Atlanta, Hong
Document type: Exhibit
Date: 2021-12-06

Full text

Did FinTech Lenders Facilitate PPP Fraud?∗
John M. Griﬃn†
Samuel Kruger‡
Prateek Mahajan§
December 6, 2021
Abstract
In the distribution of the Paycheck Protection Program’s (PPP) $803 billion in funds,
FinTech lenders began minimally but ramped up their market share to over 80% of
originated loans by the end of the program. We examine metrics related to poten-
tial misreporting including non-registered businesses, multiple businesses at residential
addresses, abnormally high implied compensation per employee, and large inconsisten-
cies in jobs reported with another government program. We assess these four metrics
with ﬁve supplemental measures and extensive supporting analysis. Suspicious loans
exhibit sharp and discontinuous increases in misreporting at maximum loan thresholds
and round loan amounts with discontinuities more pronounced among FinTechs. Fin-
Tech loans are more than 3.5 times as likely to be initiated by someone with a felony
record, strongly cluster in industry-county pairs to a degree that is infeasible based
on U.S. Census data, and frequently exhibit similar loan features within lender-county
pairs. Diﬀerences across lenders are persistent with certain FinTech lenders seemingly
specializing in questionable loans. Few of these loans have been prosecuted by author-
ities or repaid. FinTech lenders in round three rapidly increased both their market
share and the fraction of their loans with potential misreporting, particularly in zip
codes with the highest levels of questionable lending in earlier rounds. From the ﬁrst
round of the program in April 2020 to the last month of the program in May 2021,
the amount of potential misreporting increased more than four-fold. While FinTech
lenders likely expand PPP access, this may come at the cost of facilitating fraudulent
credit.
JEL classiﬁcation: G21, G23, G28, H12
keywords: FinTech, Paycheck Protection Program (PPP), Misreporting, Fraud.
∗We are thankful for comments from seminar and conference participants at the Federal Reserve Bank of Atlanta, Hong
Kong Baptist University, the Lone Star Finance Conference, the Oﬃce of the Comptroller of the Currency, the Texas Finance
Festival, the U.S. Government Accountability Oﬃce, The University of Texas at Austin, and the UT Dallas Fall Finance
Conference. We also thank Andres Almazan, Bruce Carlin, Jonathan Cohn, William Fuchs, Benjamin Herbert, Zack Liu, Alex
Priest, Alberto Rossi, Laura Starks, Brian Wolfe, Yao Zeng, Harold Zhang, Capital Plus, Blue Acorn, Itria Ventures, and many
others for helpful comments. We thank Integra Research Group for research support and the University of Texas McCombs
Research Excellence Grant. Griﬃn is an owner of Integra Research Group, Integra FEC, and Integra REC, which engage in
research, ﬁnancial consulting, and recovery on a variety of issues related to investigation of ﬁnancial fraud.
†McCombs School of Business, University of Texas at Austin. Email: John.Griﬃn@utexas.edu.
‡McCombs School of Business, University of Texas at Austin. Email: Sam.Kruger@mccombs.utexas.edu.
§McCombs School of Business, University of Texas at Austin. Email: Prateek.Mahajan@mccombs.utexas.edu.

Did FinTech Lenders Facilitate PPP Fraud?
December 6, 2021
Abstract
In the distribution of the Paycheck Protection Program’s (PPP) $803 billion in funds,
FinTech lenders began minimally but ramped up their market share to over 80% of
originated loans by the end of the program. We examine metrics related to poten-
tial misreporting including non-registered businesses, multiple businesses at residential
addresses, abnormally high implied compensation per employee, and large inconsisten-
cies in jobs reported with another government program. We assess these four metrics
with ﬁve supplemental measures and extensive supporting analysis. Suspicious loans
exhibit sharp and discontinuous increases in misreporting at maximum loan thresholds
and round loan amounts with discontinuities more pronounced among FinTechs. Fin-
Tech loans are more than 3.5 times as likely to be initiated by someone with a felony
record, strongly cluster in industry-county pairs to a degree that is infeasible based
on U.S. Census data, and frequently exhibit similar loan features within lender-county
pairs. Diﬀerences across lenders are persistent with certain FinTech lenders seemingly
specializing in questionable loans. Few of these loans have been prosecuted by author-
ities or repaid. FinTech lenders in round three rapidly increased both their market
share and the fraction of their loans with potential misreporting, particularly in zip
codes with the highest levels of questionable lending in earlier rounds. From the ﬁrst
round of the program in April 2020 to the last month of the program in May 2021,
the amount of potential misreporting increased more than four-fold. While FinTech
lenders likely expand PPP access, this may come at the cost of facilitating fraudulent
credit.
JEL classiﬁcation: G21, G23, G28, H12
keywords: FinTech, Paycheck Protection Program (PPP), Misreporting, Fraud.
2

The melding of ﬁnancial technology and banking, also known as FinTech lending, has
emerged at a rapid pace in the aftermath of the ﬁnancial crisis. Buchak et al. (2018) ﬁnd that
an increase in regulatory burdens for traditional banks is the predominant driver in the rise
of FinTech lending. A large aspect of the scrutiny and regulation of traditional banking was
their perceived role in the ﬁnancial crisis, which included facilitating wide-scale mortgage
fraud as partially evidenced by over $137 billion in government ﬁnes and settlements. Fin-
Tech lenders oﬀer a new banking model that replaces traditional lending relationships with
online advertisements, app interfaces, and loan screening algorithms. Are FinTech lenders
able to harness the power of technology to reduce loan maleﬁcence?
The Paycheck Protection Program (PPP), a historic COVID-19 relief program for busi-
nesses, rapidly distributed over $803 billion in funds through 11.7 million loans in three short
rounds spread between April 2020 and May 2021. Although FinTech lenders began with a
slow start with less than 10% of loans in round 1, they ramped up their market share to
over 80% of loans by May 2021, highlighting their growing importance. FinTech lending
was recognized for broadening access to PPP loans, particularly to smaller ﬁrms without
pre-existing lending relationships with traditional banks, and for facilitating quick and eﬃ-
cient lending at a time when many small businesses were in dire need due to the COVID-19
pandemic. However, the rapid expansion of FinTech lending may have come at the expense
of underwriting standards. Whereas traditional banks have established borrower relation-
ships and extensive Bank Secrecy Act (BSA) compliance programs, many FinTech lenders
had few established relationships and may have been less diligent when establishing formal
procedures with little reputation to protect.
Alternatively, FinTech lenders have been shown to use ﬁnancial data with increased speed
and accuracy.
Fuster et al. (2019) ﬁnd that FinTech mortgage lenders not only process
government agency loans faster than traditional banks but have fewer defaults, indicating
potentially superior loan screening. Peer-to-peer FinTech platforms utilize a rich set of alter-
native data and machine learning to optimize credit decisions (Jagtiani and Lemieux 2019).
If used eﬀectively, this enhanced technology and increased data access may be able to detect
and prevent PPP applications from ﬁctitious businesses and individuals. Do traditional or
FinTech loans exhibit more features consistent with potential PPP misreporting? And how
does this potential misreporting vary across individual traditional and FinTech lenders?
To investigate these questions, we perform a big data analysis of loan features on over
eleven million PPP loans with six disparate datasets. We introduce four primary and ﬁve
secondary indicators of whether a loan is potentially misstated. Each indicator creates an in-
ference that a loan is suspicious but is not proof of misreporting on its own. The four primary
1

measures are non-registered businesses, multiple loans at a residential address, abnormally
high implied compensation relative to industry and CBSA averages, and large inconsisten-
cies with diﬀerences as large as tenfold between the jobs reported by a borrower on its PPP
application and jobs reported to another contemporaneous government program application
with a diﬀerent incentive structure. The ﬁve secondary measures are discontinuities around
the maximum PPP compensation level of $100,000, rounded loan amounts, overrepresenta-
tion of PPP loans relative to U.S. Census data on the number of business establishments in a
particular industry and county, clustering of loans with similar features within lender-county
pairs, and criminal records for PPP borrowers.
We assess each of the four primary indicators with multiple discontinuity and comparative
analyses. First, there is substantial cross-validation across the four main indicators. For ex-
ample, loans that report abnormally high compensation relative to the U.S. Census’s average
compensation in the loan’s industry and CBSA also have higher incidences of non-registered
businesses and multiple loans at the same address. These patterns are substantially elevated
in FinTech lenders but are also present for traditional lenders, indicating that misreporting
is not just conﬁned to FinTech. Second, someone receiving a ﬁctitious loan might wish to
maximize or come close to maximizing their proceeds. We ﬁnd monotonically increasing
levels of indicators when approaching the perceived maximum compensation threshold, and
a sharp discontinuity at the threshold with lower levels of suspicious lending just above the
threshold. These patterns are present for all four main indicators; discontinuities are present
both for traditional and FinTech lenders, but are much more pronounced for FinTechs with
an increase in potential misreporting of seven times when approaching the threshold from
below.
Third, even though PPP loan amounts were supposed to be based on historical
past compensation with requirements for detailed supporting documentation, loans cluster
at rounded monthly compensation values, and these spikes coincide with higher levels of
each of the four primary misreporting indicators. These patterns are most pronounced for
FinTech lenders.
Fourth, PPP lending at the industry-county level frequently exceeds the number of es-
tablishments listed for that industry and county in U.S. Census data. For FinTech lenders,
40.4% of loans exceed industry-county establishment counts, and 33.7% of loans exceed
industry-county establishment counts by a factor of more than two.1 Measures of misreport-
ing monotonically increase as the ratio of PPP loans to businesses documented by the U.S.
Census increases, particularly for FinTech lenders. Fifth, based on the idea that networks
in a region may use recurring loan features, we construct a concentration ratio to measure
1The corresponding ﬁgures for traditional lenders are 14.9% and 8.8%, respectively.
2

clustering in loan amounts, number of jobs, and industries within each lender-county pair.
Like the other secondary measures, FinTech lenders have higher levels of clustering along
loan features, and clustering is monotonically associated with higher levels of potential mis-
reporting. Finally, we collect criminal background data for a sample of 150,000 individuals.
FinTech borrowers are more than 3.5 times as likely to have a felony record, and borrowers
ﬂagged for potential misreporting based on the primary and other secondary measures are
also more likely to have felony records.
Overall, we ﬁnd more than 1.51 million questionable loans representing over $68.9 billion
in capital. FinTech loans are more than 3.17 times as likely to have at least one primary
indicator of misreporting and 6.16 times as likely to have a primary indicator that is con-
ﬁrmed by an additional primary or secondary indicator. There is also substantial geographic
heterogeneity in the rates of suspicious lending across counties and across zip codes within
the same county. Suspicious lending rates also vary substantially across lenders, with poten-
tial misreporting rates in excess of 32% for two large FinTech lenders. Moreover, potential
misreporting increases over time with particularly high rates in the last month of round 3
(25.8%), even after the Oﬃce of the Inspector General for the Small Business Administra-
tion (SBA) ﬂagged PPP fraud as a concern. Network graphs of the lending space indicate
that many FinTech portals switch lenders and utilize multiple lenders even for the same
individual. Several of the FinTech lenders with the highest suspicious loan rates are new
lenders that did not start making PPP loans until round 3, and there is no evidence that
lenders attempted to decrease misreporting over time. Instead, second-draw loans to bor-
rowers with suspicious ﬁrst-draw loans by the same lender are common, and lenders with
high rates of misreporting in rounds 1 and 2 increased both their misreporting rates and
their loan volume in round 3. For example, the largest four FinTech lenders, Prestamos,
Cross River, Capital Plus, and Harvest exhibited high and increasing rates of misreporting
and lending volume while receiving over a billion dollars in processing fees each. Finally,
FinTech lenders often doubled, tripled, or even quadrupled their lending in zip codes with
high levels of potential misreporting in rounds 1 and 2 while also substantially increasing
their misreporting percentages.
Our work is related to four main literatures. First, there is a rapidly emerging literature
on FinTech lending that highlights its growing importance and positive economic eﬀects
through ﬁlling gaps left by traditional banks in both residential (Buchak et al. 2018) and
business lending (Gopal and Schnabl 2020). Fuster et al. (2019) ﬁnd that FinTech mortgage
lenders process loans faster and increase the odds of borrowers reﬁnancing their loans at
lower rates, all with fewer defaults, indicating that FinTechs are not simply engaged in lax
3

screening as was the case for securitized lending in the run-up to the ﬁnancial crisis (Keys
et al. 2010; Purnanandam 2011). Erel and Liebersohn (2021) examine FinTech lending in the
PPP and ﬁnds that FinTech lenders increased access to the PPP by lending more in zip codes
with fewer traditional banks, lower incomes, and higher minority percentages. Chernenko
and Scharfstein (2021) ﬁnd that black- and Hispanic-owned ﬁrms were less likely to receive
PPP loans. Howell et al. (2021) ﬁnd that FinTechs were more likely to provide PPP loans
to black-owned businesses, and Atkins et al. (2021) ﬁnd that FinTechs helped close a gap
in loan size between black- and white-owned businesses.2 With respect to FinTech lending
before the PPP, Gopal and Schnabl (2020) ﬁnd that FinTech lenders have positive economic
eﬀects by ﬁlling in gaps in lending to small businesses left by traditional banks following the
ﬁnancial crisis. While most of the FinTech literature ﬁnds beneﬁts to FinTech lending such
as increased competition, broader ﬁnancial access, less discrimination, faster lending speed,
and lower defaults, our paper analyzes a potential cost of FinTech expansion and diﬀerential
practices across FinTechs. We are not anti- or pro-FinTech and leave overall welfare analysis
to future research.
Second, regarding the eﬃcacy of the PPP, Chetty et al. (2020) ﬁnd that the PPP increased
employment at participating ﬁrms by only 2% at a cost of $377,000 per job saved, and Autor
et al. (2020) ﬁnd only slightly higher employment beneﬁts of 2% to 4.5%. Granja et al.
(2020) ﬁnd small employment eﬀects due to the PPP and a low correlation between regional
COVID variation and PPP funding allocation.3 In contrast, Faulkender et al. (2021) ﬁnds
that the program was much more eﬀective with an estimated 18.6 million jobs saved at an
average cost of $28,000. Additionally, there is evidence of diﬀerential access to the PPP based
on knowledge of the program, distance to the closest bank branch, banking relationships,
and personal banking connections (Amiram and Rabetti 2020; Bartik et al. 2020; Neilson
et al. 2020; Duchin et al. 2021; Glancy 2021; Li and Strahan 2021). Our evidence adds
an additional concern regarding the PPP’s eﬃcacy and fairness. We are the ﬁrst academic
paper to examine wide-scale potential PPP loan misreporting, but there have been interesting
press and investigative reports regarding suspicious PPP loans (Miami Herald 2020; The Wall
Street Journal 2020; Bloomberg Businessweek 2020; Project on Government Oversight 2020;
ProPublica 2021), some of which feature FinTech lenders.4
Third, assessment of the PPP program also relates to a broader literature on fraud,
2In contrast, Bartlett et al. (2021) ﬁnd that FinTech algorithms charge higher interest rates to minorities in residential
mortgage lending but price discriminate less than traditional banks. Begly et al. (2021) ﬁnd that the SBA disaster-relief home
loan program denies more loans to minorities and subprime borrowers due to the program’s risk-insensitive pricing.
3Relatedly, Meirer and Smith (2021) ﬁnd that the PPP was a windfall for some ﬁrms.
4Concerns about PPP fraud have been ﬂagged by the Oﬃce of the Inspector General for the Small Business Administration
(SBA) (see report here). Beggs and Harvison (2021) ﬁnd that among the 2,999 registered investment advisors who took PPP
loans, those with a history of ﬁnancial misconduct received unusually large PPP loan allocations.
4

waste, and abuse in government programs. While most of this literature focuses on active
corruption within the government (e.g., Shleifer and Vishny 1993; Svensson 2003; Glaeser
and Goldin 2006; Avis et al. 2018), waste due to poor program design or administration can
also be costly (Bandiera et al. 2009). Hart et al. (1997) model tradeoﬀs between internal
and external provision of government services, and Deﬂo (2017) emphasizes the importance
of program details. In the PPP program, lack of direct tools for validating eligibility and
limited incentives for high-quality underwriting by lenders with no skin in the game may
have signiﬁcantly increased fraud and abuse. Hanson et al. (2020) argue that direct relief
for small businesses from the Internal Revenue Service could have been more targeted and
eﬃcient than the PPP’s external lender model.
Finally, our work relates to forensic economics and loan misreporting. Zitzewitz (2012)
surveys the literature on forensic economics, noting that a common thread in this literature
is quantifying activity about which there was previously only anecdotal evidence, in large
part because agents have an incentive to keep it hidden. Widescale mortgage fraud and
misreporting in securitized mortgages prior to the ﬁnancial crisis included second-lien and
owner-occupancy status misreporting (Piskorski et al. 2015; Griﬃn and Maturana 2016),
misreported income (Jiang et al. 2014; Mian and Suﬁ2017), misreported assets (Garmaise
2015), and inﬂated appraisals (Ben-David 2011; Kruger and Maturana 2020). This fraud
involved both smaller, less-known mortgage originators and large bank underwriters who
knowingly passed along these misrepresentations in mortgage-backed securities.
FinTech
lending emerged and grew against this backdrop as related regulation increased for traditional
banks (Buchak et al. 2018). Our ﬁndings indicate that replacing traditional lending with
FinTech lending may amplify misreporting problems, at least with respect to the PPP.
Our ﬁndings also have important practical implications regarding the extent and nature of
PPP misreporting, the expanding role of FinTech lending, waste in the PPP, the proliferation
of ﬁctitious lending, and the insuﬃcient deterrence of current policies and enforcement. The
potential policy and practical implications of these ﬁndings are further discussed in the
conclusion.
1
Data and Summary Statistics
1.1
Data Sources
The basis for our sample is loan-level PPP data released on June 30, 2021 by the Small
Business Administration (SBA). This dataset covers all PPP loans issued from the start of
the program on April 3, 2020 through the end of the program on June 30, 2021 that had
5

not been repaid as of June 30, 2021. At the loan-level, the data include business name,
address, business type (e.g., corporation, LLC, self-employed, etc.), NAICS code (industry),
loan amount, number of employees, date approved, loan draw (i.e., initial, ﬁrst-draw loan
or repeat, second draw loan), and lender for 11,768,689 loans originated by 4,890 diﬀerent
lenders and with a total value of $803 billion. We follow Erel and Liebersohn (2021) and
classify lenders as either traditional or FinTech (consisting of online banks and non-bank
lenders) based on automated name matching with bank identiﬁers from the Federal Financial
Institutions Examination Council (FFIEC) with hand matching for remaining lenders.5
Concurrently with the PPP, the SBA provided businesses and individuals with the ability
to receive an Economic Injury Disaster Loan (EIDL), with forgivable advances of up to $10k.
EIDL Advance loan-level data was released on December 1, 2020 and covers all EIDL Advance
issued in 2020.6 To check for inconsistencies between the information borrowers provided on
their PPP and the EIDL applications, we match borrowers in the PPP and EIDL loan-level
datasets based on business name and zip code.
We also match PPP borrowers with business registry data from OpenCorporates, a non-
proﬁt that maintains a database of companies around the world. OpenCorporates collects
its data directly from state governments and covers 76 million businesses across all US states
except Illinois. The data include incorporation dates, dissolution dates (if applicable), and,
implicitly, whether the business has ever been registered. We match OpenCorporates data
to the PPP loan-level data based on business name and state.
To examine previous criminal and ﬁnancial activity, we collect criminal background data
from LexisNexis based on the borrower’s name and address for a random sample of 150,000
round 1 and 2 loans made to individuals (12.9% of rounds 1 and 2 PPP loans made to
individuals).7
Finally, we use several U.S. governmental data sources for address and demographic
information. We standardize addresses and distinguish residential and commercial addresses
from one another based on the Address Validation Application Programming Interface from
the United States Postal Service. For data on the number of establishments and average
5We use Erel and Liebersohn’s (2021) classiﬁcations for lenders that were active in rounds 1 and 2 (the sample period for
Erel and Liebersohn (2021)), and we use the same methodology for classifying round 3 lenders that were not active enough to
be classiﬁed in the earlier rounds. We also note that the classiﬁcation of FinTech lenders can be diﬃcult because traditional
banks with multiple branch locations may also originate loans from other lenders or online portals. See the Internet Appendix
for additional details.
6The SBA has not released updated EIDL Advance data for 2021.
7Because the LexisNexis searches require an individual’s name, only loans with an individual name listed as the borrower
(rather than a business name) and where the business type is a self-employed individual, an independent contractor, or a sole-
proprietor are included in this criminal search. The criminal records data is collected only from rounds 1 and 2 loans because
round 3 data was released after the criminal records data was collected.
6

compensation, we use the 2019 County Business Patterns (CBP) data from the US Census
Bureau, aggregated by region (either core-based statistical area (CBSA) or county) and
North American Industry Classiﬁcation (NAICS) code. The CBP data include the number
of establishments, number of employees, and total wages for a given industry in a county or
CBSA. Similarly, for data on total receipts by non-employer businesses, we use data from the
Nonemployer Statistics (NES) data from the US Census. Matching between the loan-level
data and the CBP and NES datasets is based on the business’s zip code and the ﬁrst four
digits of its NAICS code.
1.2
Summary Statistics
Panel A of Figure 1 shows the number of loans originated on the left axis and the total
amount lent on the right axis by each of the top 75 PPP lenders.
FinTech lenders are
highlighted in red (non-bank FinTech lenders) and cream color (online banks). Six of the
ten top lenders by number of loans are FinTech, with Prestamos, Cross River, and Capital
Plus in the top ﬁve alongside Bank of America and JP Morgan Chase.8 Due to their larger
average loan size (Erel and Liebersohn 2021), dollar lending volume tends to be higher for
traditional banks.
Panel B of Figure 1 shows the total FinTech market share during each week throughout
the three rounds of PPP lending. Total FinTech market share grew from only 1.4% of loans
in the ﬁrst week of round 1 to 7.6% in the last week of round 1. Round 2 continued the
PPP after a short break of ten days in May 2020 with new funding for borrowers who did
not receive a loan in round 1. By the end of round 2 in August 2020, FinTech market share
grew to 49.0% of loans, over the last two weeks, for an overall market share of 4.8% in round
1 and 20.4% in round 2. Round 3 of the PPP, which includes both ﬁrst-draw loans for new
borrowers and second-draw loans for borrowers that already obtained loans in round 1 or 2,
started in January 2021 with a low FinTech market share of less than 20% for the ﬁrst three
weeks as traditional lenders were once again the fastest to originate PPP loans. However,
FinTech share grew rapidly during round 3, reaching over 80% of loans by the end of May
2021 for an overall round 3 market share of 47.9%.9
Table I reports summary statistics for the 4.0 million FinTech and 7.8 million traditional
bank loans in our sample. FinTech loans have an average loan amount of $25 thousand
compared to $90 thousand for traditional bank loans.
Despite these large diﬀerences in
means, the median loan sizes of $19 and $21 thousand are similar. The average FinTech
8Comparing this ﬁgure to Panel A of Figure IA.1 shows how the top lenders diﬀer between the entire sample and solely
rounds 1 and 2. In particular, the growth of Capital Plus and Harvest in round 3 is apparent.
9Panel B of Figure IA.1 shows the number of loans originated each week of the PPP by type of lender.
7

loan reports supporting 2.4 jobs compared to 10.3 for traditional banks. After normalizing
loan size relative to reported jobs, FinTech loans have higher average ($64 thousand) and
median implied compensation ($71 thousand) than traditional bank loans ($47 thousand
average and $39 thousand median). For FinTech loans, 18.5% of borrowers are organized
as corporations, S-corporations, or limited liability companies (LLC) compared to 64.5% for
traditional banks lenders. FinTech loans were also less likely to be repeat loans, with 26.8%
of round 3 FinTech loans going to borrowers with previous PPP loans, compared to 59.4%
for round 3 traditional bank loans.
2
Suspicious Loan Measures
We introduce four primary indicators that a loan is potentially misstated. In this section,
we deﬁne and introduce the indicators. Each indicator creates an inference that a loan is
suspicious but is not deﬁnitive proof of misreporting on its own. In subsequent sections, we
validate the measures and explore how they relate to one another and other misreporting
indicators.
2.1
Business Registry Flag
Businesses organized as corporations, S-corporations, and LLCs are required to ﬁle an
article of incorporation or LLC ﬁling with a state, either as a domestic company in their
home state or as a foreign company in another state. Further, the SBA required businesses
to be “in operation on February 15, 2020... [and] not permanently closed.”10 Based on these
requirements, we check the following conditions for all corporation, S-corporation, and LLC
borrowers:
1. Is the business found in the business registry for its home state or in another state
while listing an address in its home state? (“Missing Business”)
2. Was the business dissolved and inactive before being approved for a PPP loan? (“Dis-
solved Business”)11
3. Is the earliest incorporation or initial ﬁlling date for the business after February 15,
2020? (“Late Incorporation/Filing”)
These three subﬂags are combined to form an overall business registry ﬂag.12
10See loan application here.
11To be ﬂagged, the dissolution date of the business must be before the PPP loan approval date and, to screen out businesses
that may be administratively dissolved (e.g., for not ﬁlling some paperwork), the business status must be listed as inactive.
12As external validation of this ﬂag for a smaller sample, we also compare PPP borrower names to data from the Florida
Department of Business and Professional Regulation following Chernenko and Scharfstein (2021) (see Figure IA.2).
Loans
ﬂagged as a missing business based on the overall business registry are over 6.7 times less likely to have a potential match in
8

Panel A of Figure 2 plots the proportion of corporate and LLC borrowers with missing,
dissolved, or late business registrations. The ﬂag is plotted as a percent of corporation, S-
corporation, and LLC loans because other PPP business entities, such as sole-proprietorships,
partnerships, and independent contractors, do not require business registrations.13 Missing
registrations are the most common type of business registry ﬂag, representing 4.2% of cor-
porate and LLC loans. Another 0.7% of corporate and LLC loans are to dissolved entities,
and 0.2% have late registrations for a total business registry ﬂag percentage of 5.1%. Nine of
the ten lenders with the highest rates of business registry ﬂags are FinTech lenders. These
lenders have 7.6% to 26.7% of their corporate and LLC loans ﬂagged for one of the three
business registry issues, with the vast majority of the ﬂagged loans simply not appearing in
the business registry data. It is possible that there are errors in the data or that some busi-
nesses have names that are diﬃcult to match; which may explain why all of the lenders have
at least some missing registrations, with business registry ﬂag rates of one to ﬁve percent
common across many lenders. However, there is not an obvious explanation for why certain
lenders, who also have elevated levels of other indicators, would have disproportionately high
matching issues.
2.2
Multiple Loan Flag
While it is possible that a business owner may have multiple businesses registered to the
same address, the presence of multiple loans at a residential address during the same draw is
also a potential sign of ﬁctitious operations. Using the business address disclosed in the PPP
loan-level data, we identify individual residential addresses associated with three or more
loans during the same draw. To do so, we ﬁrst standardize addresses and identify addresses
that are known business or central addresses (e.g., oﬃce and apartment buildings) using
the Address Validation Application Programming Interface from the United States Postal
Service. Then, we ﬁnd residential (i.e., non-business, non-central) standardized addresses
with three or more loans within the same draw.
As an example, Panel A of Exhibit 1 shows 14 loans given to a single address, all with
colorful business names, almost all in the same industry, most with the same loan amount,
and all backing ten jobs. The address associated with all 14 loans is a modest single-family
home in suburban Chicago (estimated to have a value of $170k per Zillow). The borrower
associated with the ﬁrst loan is an LLC that was registered in 2018, but the 13 subsequent
loans during July and August of 2020 are to LLCs that were registered only shortly before the
the restaurant data as compared to loans that are not ﬂagged. Loans ﬂagged as having missing business registrations that also
have another ﬂag (primary or secondary) are over 12.5 times less likely to have a potential match in the restaurant data.
13Businesses in Illinois are excluded because Illinois is missing from the business registry data due to restrictive terms and
conditions (see regulation here).
9

loans were approved, well after the February 15 eligibility cutoﬀ. Detailed internet searches
did not produce information for any of the other 13 business names or any indication of
employees other than the owner. Panel B of Exhibit 1 shows another multiple-loan example,
this one involving loans to four people in the same household, again in a modest suburban
Chicago home, all of whom received loans for the same amount, $20,833, which corresponds
to the PPP’s maximum annual compensation of $100,000.14 This income is at the top of
the spectrum for the indicated industries, which have average compensation $25-46k in the
Chicago CBSA according to the US Census CBP. Random loan-level inspections of the data
reveal numerous other examples of multiple suspicious loans ﬂowing to addresses that do not
seem to be the locations of identiﬁable businesses. The multiple loan ﬂag functions as a way
to systematically analyze these loans.
Panel B of Figure 2 shows the percentage of PPP loans that involve at least three loans
to the same residential address in the same draw by lender. Nine of the ten lenders with the
highest multiple loan ﬂag rates are FinTech lenders. For these lenders, 3.7% to 5.1% of their
loans involve multiple loans to the same residential address, and most of their ﬂagged loans
are to individual borrowers identiﬁed as independent contractors, self-employed, or sole-
proprietors. This contrasts with traditional banks, which have fewer ﬂagged loans (1.0% on
average).15 Interestingly, three FinTech lenders, Capital One, Square, and Intuit, have lower
than median levels of multiple loans at the same address.
2.3
High Implied Compensation Flag
PPP loan size is limited to 2.5 times a business’s average monthly payroll expenses,
including up to $100,000 in annual compensation per employee. PPP loan applications report
how many employees the business has based on the same time period used to calculate average
payroll expenses (2019 in most cases). Using loan size and number of reported employees,
we are able to impute implied average annual compensation. Implied compensation at the
borrower level is strongly related to average compensation in the borrower’s industry (NAICS
4-digit) and CBSA (e.g., see Figure IA.4).16
14All four of these individuals also received second draw loans for the same amount. SBA guidelines asks the borrower for
their business address. The industries themselves are also somewhat suspicious in that two are equipment manufacturing and
one is auto repair despite no evidence of these businesses in photos of the property. Further, the borrower in the nail salon
industry does not appear to have an Illinois nail technician license. One of the equipment manufacturing borrowers also switched
to the nail salon industry during round 3 despite also not having a nail technician license.
15Large diﬀerences between traditional and FinTech lenders hold even if we limit the sample to only loans with residential
addresses and are even larger if we ﬂag loans if there are at least two loans at the same residential address with the same
draw (see Figure IA.3). Flagged loans for traditional banks are mainly to formally registered corporations, S-corporations, and
LLCs, consistent with the incentives for owners of multiple legitimate businesses to formally register their businesses for tax
and limited liability purposes. We control for diﬀerences in loan composition across lenders in subsequent regression analysis.
16Schedule C ﬁlers also had the option to use gross income instead of net income for owner compensation after March 3,
2021. To conservatively account for the option, we compare implied compensation for sole proprietor, independent contractor,
10

To capture abnormally high compensation, we examine the kernel density of implied
average compensation for the borrower normalized by mean compensation across all ﬁrms
in the borrower’s industry and CBSA based on US Census Bureau CBP data separately
for FinTech and traditional loans in Panel A of Figure 3. FinTech borrowers have a much
fatter right tail of the abnormal compensation distribution. Speciﬁcally, in the lower plot,
which includes all loans, 17.8% of FinTech borrowers have normalized compensation above
3, compared to 3.5% of traditional borrowers.17
It is instructive to examine how normalized compensation relates to our ﬁrst two sus-
picious loan ﬂags.
On the right axis of Panel A of Figure 3, we plot the percentage of
loans with the business registry and multiple loan ﬂags across the distribution of normalized
compensation separately for FinTech and traditional lenders.18 Both ﬂags increase signiﬁ-
cantly as normalized compensation increases for loans made by FinTech lenders. Whereas
7.8% of corporate and LLC FinTech loans with normalized compensation below one have the
business registry ﬂag, 26.7% of loans with normalized compensation above three have the
ﬂag. Similarly, the multiple loan ﬂag increases from 2.7% for FinTech loans with normalized
compensation below one to 5.1% when normalized compensation is above three. Impor-
tantly, while FinTech loans exhibit a stronger relation between normalized compensation
and the other loan ﬂags, this pattern is not limited to FinTechs. Traditional bank loans
also have more business registry and multiple loan ﬂags when normalized compensation is
higher suggesting that they also have loan misreporting issues, though at a much lower scale.
Overall, the results show that while some variation in normalized compensation across ﬁrms
is to be expected, high implied compensation is strongly related to other suspicious loan
characteristics, particularly for FinTech loans.
For our main measure of high implied compensation, we conservatively only ﬂag loans
where the implied compensation per job reported is more than three times the industry-
CBSA average compensation/receipts (“high implied compensation”). Because compensa-
tion is censored at $100,000 for most borrowers, this ﬂag is only possible in industry-CBSA
pairs with average annual compensation/receipts below $33,333.33.19
Within this set of
self-employed, and single member LLC loans after March 3, 2021 to the greater of industry/CBSA average compensation
and industry/CBSA average receipts for single-employee ﬁrms. This adjustment is also used in calculating the high implied
compensation measure below.
17Most of this is due to round 3 FinTech loans, as evident in Figure IA.4, Panel A. FinTech implied compensation is much
higher round 3 and appears to be almost completely disconnected from average industry-CBSA compensation and receipts.
Panel B of Figure IA.4 shows that this diﬀerential pattern for ﬁntech and traditional lenders in round 3 is also evident even
when the sample is restricted to Schedule C borrowers, both before and after Schedule C borrowers were permitted to use gross
income, starting on March 3, 2021.
18The business registry ﬂag plot includes only corporation, S-corporation, and LLC loans because the business registry ﬂag
can only be determined for these business types.
19Some loans are also outside of a CBSA or in an industry-CBSA pair that is too small to be included in the US Census CBP
data. In total, 3,416,620 loans are in industry-CBSA pairs with average annual compensation/receipts below $33,333.33.
11

industry-CBSA pairs, 48.9% of FinTech loans and 9.4% of traditional loans have normalized
compensation above three. Panel B of Figure 3 plots how the percentage of loans with the
high compensation ﬂag varies across lenders. Eight of the ten lenders with the highest ab-
normal compensation percentages are FinTech. For all of these lenders, more than 50% of
applicable loans have the high implied compensation ﬂag. By contrast, 46 of the 71 largest
lenders have less than 10% of their loans ﬂagged, and 43 of these lenders are traditional
banks. Although most of the FinTech lenders cluster with high rates of abnormally high
compensation, Capital One and Square are near the very low end of ﬂagged loans, perhaps
indicating these FinTechs were relying on existing relationships with the borrowers.
2.4
EIDL Advance Jobs > PPP Jobs Flag
Concurrently with the PPP, the SBA provided businesses and individuals with the ability
to receive a forgivable Economic Injury Disaster Loan (EIDL) Advance of up to $10,000.20
For all EIDL Advances issued in 2020, the advance amount was calculated as $1,000 per
employee (up to the $10,000 maximum).21 Thus, there was an incentive for borrowers to
inﬂate the number of jobs reported on their EIDL applications.22 We focus on cases where
EIDL jobs exceed PPP jobs because the job inﬂation incentive is provided by the EIDL
Advance program (PPP loans are based on total payroll as opposed to number of jobs).
Panel A of Figure 4 plots the distribution of diﬀerences between EIDL and PPP jobs.
Three patterns stand out. First, consistent with the incentive to inﬂate EIDL jobs as opposed
to PPP jobs, EIDL exceeds PPP by three or more jobs 9.1% of the time, whereas PPP exceeds
EIDL by three or more jobs only 3.6% of the time. Second, the most common discrepancy
between the programs is a diﬀerence of nine jobs, which implies that the borrower claimed
10 or more jobs and took out the maximum EIDL Advance of $10,000 despite only reporting
one PPP job. Third, EIDL job inﬂation is much more pronounced in FinTech loans than
in traditional loans. In particular, EIDL data exceeds PPP data by nine jobs 14.5% of the
time for FinTech loans compared to 0.6% for traditional loans.
Panel B of Figure 4 shows the prevalence of EIDL job inﬂation by lender as a percentage
of PPP loans with matching EIDL Advances. To be conservative, we only include cases
20While the EIDL Advance program was billed as a forgivable advance with the potential for a larger non-forgivable loan,
65.8% of EIDL Advances involved no additional EIDL loan. EIDL Advances were immediately forgiven by the SBA.
21The EIDL Advance rules changed for 2021 to: A) provide the entire $10,000 regardless of employee count, and B) to target
the advances to low-income communities and those with a demonstrated decrease in revenue. The SBA has not yet reported
data on 2021 EIDL Advances.
22For borrowers who take out the maximum EIDL Advance of $10,000, we can infer that the borrower claimed at least 10
employees on their EIDL Advance application. Subsequent to the original public version of this paper, an October 7, 2021
report by the SBA OIG found that over 700,000 EIDL recipients applied for and received advances for multiple employees even
though they only had a single employee, resulting in $4.5 billion of improper EIDL Advance payments (see report here).
12

where the EIDL implied number of jobs is at least three more than the PPP reported
number of jobs. All ten lenders with the most frequent discrepancies between EIDL and
PPP jobs are FinTech. In particular, Capital Plus, Prestamos, Lendistry, Harvest, Benworth,
Fountainhead, Itria, MBE, Cross River, and Leader Bank (all FinTechs) have job reporting
inconsistencies ranging from 12.8% to 54.6%.
For all of these lenders except MBE and
Leader Bank, most of the inconsistencies are a full nine jobs. Five FinTech lenders (Intuit,
WebBank, Square, Capital One, and Live Oak) have levels of inconsistencies that are similar
to traditional banks, with nine-job diﬀerences only in rare cases.
2.5
Are the Suspicious Loan Flags Related to One Another?
If the above indicators of potential misreporting are due to random data errors or honest
mistakes, one might expect diﬀerent types of indicators to occur randomly across loans
and lenders. Therefore, multiple ﬂags for the same loan create a heightened misreporting
inference, and high lender ﬂag rates across multiple indicators may be due to policies and
practices that facilitate more misreporting.
In Table II, we examine how the four ﬂags relate to one another by reporting odds ratios
between each pair of ﬂags. The odds ratios are calculated based on loans for which data to
calculate both ﬂags are available (e.g., corporate and LLC loans for the business registry ﬂag
and loans with matched EIDL Advances for the EIDL > PPP Jobs ﬂag), with z-statistics
calculated based on standard errors double-clustered by zip code and lender in parentheses.
Panel A reports odds ratios for the full sample, all of which are above 1.50 and highly
signiﬁcant. In particular, the odds ratio between the high implied compensation and EIDL
> PPP jobs ﬂags is 14.75 and has a z-statistic of 19.90. Panel B reports odds ratios separately
for FinTech and traditional loans with FinTech loans in the lower triangle and traditional
loans in the upper triangle. The odds ratios are all positive and highly signiﬁcant with
consistently higher ratios for FinTech loans. To check that these relations are independent
of one another and not explained by loan characteristics, Table IA.I regresses each of the
ﬂags jointly on the other ﬂags, controlling for loan size and number of jobs with zip code,
business type, and industry × CBSA ﬁxed eﬀects, with and without lender ﬁxed eﬀects.
Except for the relation between the business registry ﬂag and EIDL > PPP Jobs ﬂags, the
coeﬃcients between the ﬂags are all positive, economically large relative to the mean ﬂag
rates, and highly statistically signiﬁcant.
We also ﬁnd that ﬂag rates are signiﬁcantly correlated with one another across lenders,
and the same lenders frequently have high ﬂag rates across all four indicators (as shown in
Figure IA.5). In particular, FinTech lenders Capital Plus, Prestamos, MBE, and Harvest
13

have ﬂagged rates in the top 10 for all four ﬂags, and Itria and Benworth are in the top ten
for three of the four ﬂags. In contrast, no traditional lender is consistently in the top 10
for more than two ﬂags. This pattern is exactly what we would expect if some lenders have
looser underwriting standards and is diﬃcult to explain with random mistakes or errors in
the data.
As an additional validation of the primary ﬂags, we also compare them to direct evidence
of loan size inﬂation for nonproﬁts based on comparing loan sizes to non-proﬁt compensation
disclosed on IRS Form 990.
Figure IA.6 shows that loan size inﬂation by non-proﬁts is
increasing and highly related to the primary ﬂags. See the Internet Appendix for details on
this analysis.
2.6
FinTech Diﬀerences?
Table III summarizes the percentage of loans with each of the four ﬂags separately for
FinTech and traditional lenders. The table also summarizes the percent of all loans with at
least one ﬂag and with two or more ﬂags. For each individual measure, the denominator
is the loans that could have such ﬂag (i.e., only corporate and LLC loans for the business
registry ﬂag, only loans in industry-CBSA pairs with average compensation below $33,333.33
for high implied compensation, and only loans with a matched EIDL Advance loan for the
comparison to EIDL). For the overall ﬂag measures, the denominator is all loans in the
sample, which understates the incidence of suspicious loans since most of the ﬂags are only
applicable to a minority of the loans.
Diﬀerences between FinTech and traditional ﬂag
percentages are reported in column (3). For all four individual measures, FinTech lenders
have ﬂag rates that are 2.35 to 5.19 times as high as traditional lenders, with particularly
large diﬀerences for the high implied compensation and EIDL > PPP jobs ﬂags. Overall,
23.6% of FinTech loans have at least one of the ﬂags, compared to 7.4% for traditional loans.
These diﬀerences are all highly signiﬁcant with standard errors double-clustered by zip code
and lender to conservatively allow for potential geographic and within-lender correlations.
To account for potential compositional diﬀerences between FinTech and traditional lenders,
column (4) reports adjusted diﬀerences that control for geography, business type, and indus-
try based on regressions controlling for loan size and number of jobs with zip code, business
type and industry × CBSA ﬁxed eﬀects.23 After accounting for these eﬀects, the adjusted
diﬀerence between FinTech and traditional ﬂag rates is 3.5 percentage points (ppt) for the
business registry ﬂag (which is 81% of the rate for traditional loans), 1.1 ppt (106%) for the
multiple loan ﬂag, 9.4 ppt (99%) for the high implied compensation ﬂag, and 6.5 ppt (134%)
23Corresponding regressions results with and without the control variables and ﬁxed eﬀects are reported in Table IA.II.
14

for the EIDL > PPP jobs ﬂag. These results indicate that even though loan composition
explains part of the diﬀerence between FinTech and traditional loans, ﬂag rates remain much
higher for FinTech loans even after controlling for all observable characteristics. To further
control for potentially non-linear loan characteristic eﬀects, we match FinTech loans with
traditional loans based on loan size, industry, county, and business type in column (5) Fin-
Tech with similar results.24 It remains possible that other omitted variables or unobserved
loan characteristics could explain some of the diﬀerence between FinTech and traditional
loans, but these eﬀects would have to be large to explain the results. To control for any
unobserved diﬀerences across households, Table IA.III considers a restricted sample of res-
idential addresses with loans from both traditional and FinTech lenders. Consistent with
results in Table III, ﬂag rates are elevated for FinTech loans across all of the potential misre-
porting measures, with highly statistically signiﬁcant diﬀerences in all but one speciﬁcation.
Tests in the next section, including grouping around discontinuities and clustering, also help
to address omitted variable concerns.
3
Suspicious Loans or Mistakes?
While they are suggestive of misreporting, the misreporting indicators in the previous
section also have potentially innocent explanations. In this section, we develop and ana-
lyze ﬁve additional measures as external veriﬁcation to assess the plausibility of alternative
explanations. The additional measures involve discontinuities, rounded compensation lev-
els, abnormal numbers of loans in industry-county pairs, clustering of loan features within
lender-county pairs, and criminal records. We also explore diﬀerences between FinTech and
traditional lenders.
3.1
Discontinuities at $100,000 Compensation
PPP loan size is calculated as 2.5 times a borrower’s average monthly payroll, includ-
ing up to $100,000 in wages per employee.25 This $100,000 cutoﬀis a hard maximum for
self-employment compensation. For other employees, payroll expenses also include employer
insurance and retirement contributions and unemployment taxes, which can push included
payroll expenses above $100,000 per employee. Someone ﬁlling out a fraudulent PPP appli-
cation and who may not have carefully read the PPP rules, might want to maximize their
loan amount by submitting payroll expenses at or close to the $100,000 per employee limit
without the additional expenses that are eligible with proper payroll details.
24Details on the matching process are provided in the Internet Appendix.
25See the Internet Appendix for details on the SBA guidance for how to calculate loan size.
15

Figure 5 plots the distribution of implied compensation per employee and shows how
it relates to the misreporting indicators from the previous section. The implied compensa-
tion distributions (up to $130,000) for FinTech and traditional loans are plotted as orange
and gray bars, respectively. Panel A shows that FinTech loans stand out as having more
loans with high implied compensation right at and slightly under $100,000, and traditional
banks have more loans with implied compensation in the range of $10,000 to $75,000. The
percent of loans with one of the four primary ﬂags for FinTech and traditional loans are
plotted as orange and gray dots along with third-degree polynomials and their associated
95% conﬁdence intervals estimated separately above and below the $100,000 compensation
bin. As compensation increase from $40,000 to $100,000, the prevalence of the primary ﬂags
for FinTech loans increases from 6.1% to 43.1%. For traditional lenders, the increase is also
present but much smaller. For FinTech loans with implied compensation above $100,000,
there is a sharp drop-oﬀin the ﬂag rate, which indicates that businesses that followed the
detailed SBA guidelines for including non-wage payroll expenses for employees with wages
above $100,000 are less likely to have one of the primary misreporting ﬂags.
Panel B repeat the same analysis separately for the business registry, multiple loan,
high implied compensation, and EIDL > PPP Jobs ﬂags.
For each indicator there is a
much steeper slope for FinTech loans than for traditional loans, and FinTech loans have
a sharp discontinuity above $100,000. The drop in the ﬂags above $100,000 indicate that
borrowers with detailed insurance and tax expenses in excess of $100,000 are also more likely
to be reporting correctly than those at or below the threshold. In Table IA.IV, we formally
test for discontinuities at $100,000 of compensation after controlling for loan characteristics
including number of employees, loan size, business type, zip code, and industry × CBSA and
ﬁnd large and highly economically signiﬁcant discontinuities for FinTech loans for all four
measures. Diﬀerences are also statistically signiﬁcant for traditional loans, though the eﬀects
are economically smaller. Overall, the increasing ﬂag rates as compensation increases and the
discontinuities around $100,000 in annualized compensation are consistent with suspicious
loans maximizing loan amounts.
Additionally, the SBA used a loan amount cutoﬀof $150,000 for a more streamlined
processing (fewer calculations and less documentation) of loan forgiveness applications.26
Consistent with applicants or lenders being aware of the threshold and trying to avoid
scrutiny, the percentage of ﬂagged loans is high for loans up to $150,000 and decreases
after the threshold. This is true for both traditional and FinTech lenders, but much more
pronounced for FinTech lenders (see Figure IA.7).
26The shorter form and reduced requirements for loans of $150,000 or below to receive forgiveness are outlined here.
16

3.2
Rounded Loan Amounts
The PPP loan application instructs borrowers to enter their average monthly compensa-
tion and to calculate their loan amount as
Loan Amount = Average Monthly Payroll × 2.5 + EIDL Reﬁnance Amount.
Applicants are instructed to calculate average monthly payroll based on historical compen-
sation (in 2019 in most cases) with detailed supporting documentation.27 It is unlikely that
actual monthly payroll would be a round number, especially after including unemployment
insurance and employer insurance and retirement contributions.
Rounded loan amounts
suggest that the numbers are potentially ﬁctitious as opposed to being based on actual doc-
umented data. If the ﬂags we have previously identiﬁed reﬂect misreporting issues, then
one might expect both a clustering of loans at round numbers and elevated ﬂags at round
numbers. However, if round numbers are simply a result of a borrower with valid documen-
tation rounding numbers slightly downward to simplify calculations, then one would expect
no elevated reporting issues at round numbers.
In Panel A of Figure 6, we ﬁrst examine the distribution of the last four digits of loan
amounts, excluding EIDL Reﬁnancing, for FinTech and traditional loans. Loan amounts
within 50 cents of a $1,250 increment (which corresponds to $500 of implied monthly payroll)
are plotted as thicker and slightly darker bars with all other loans binned into $1 wide
bins plotted as the thinner, lighter bars. Loans with an implied compensation within ±
$1,000 of $100,000 are excluded to make sure these results are distinct from the maximum
compensation result shown in Figure 5. Both FinTech and traditional loans exhibit rounding
at $1,250 increments, particularly at increments of $2,500 (corresponding to $500 and $1,000
increments of implied monthly payroll). FinTech lenders have moderately more rounding
with 10.2% of loans rounded to $1,250 increments compared to 7.6% for traditional lenders.
The right axis of Panel A examines the prevalence of the primary misreporting ﬂags. The
percent of loans with a primary ﬂag is plotted as solid dots at the $1,250 loan increments and
as hollow dots at other loan amounts (shown in $250 wide bins). If rounded loans are more
likely to be misreported, one would expect an elevated ﬂag rates at round number thresholds.
For FinTech loans, this is exactly what we observe. At rounded increments, the ﬂag rates are
consistently higher, by 2.41 ppt on average. This diﬀerence is highly signiﬁcant, which can
be seen by comparison to the dotted lines plotting a 95% conﬁdence interval estimated with a
third-degree polynomial estimated based on the non-rounded loans. For traditional lenders,
27See the Internet Appendix for details on how the loan size was to be calculated and exclusions.
17

there is only small and weak evidence of elevated ﬂags in some of the rounded bins. Thus,
rounding appears to capture suspicious loans for FinTech lenders but less so for traditional
lenders, which is also consistent with our ﬁndings in Figures 2, 3, 4, and Figure 5.
In Panel B, we consider each of the four primary misreporting indicators separately. The
top left subpanel plots corporate and LLC loans, the top right subpanel plots all loans, the
bottom left subpanel plots loans with an industry-CBSA pairs with average compensation
of less than $33,333.33, and the bottom right subpanel plots loans with matched EIDL
Advances. The four plots show that rounded loans by FinTech lenders have elevated levels
of all four primary misreporting ﬂags. For traditional lenders, the business registry ﬂag is
slightly elevated at some round numbers, the multiple loan ﬂag is slightly elevated levels at
every other $1,250 increment (but not at $2,500 increments), and there is no evidence of
elevation for the other ﬂags. Overall, the fact that all of the loan ﬂags are elevated at round
loan amounts for FinTech loans provides additional validation for the suspicious behavior
underlying these loans.
3.3
Loan Overrepresentation
If there is an organized eﬀort to obtain funds for non-existent businesses, networks of
illegitimate borrowers may ﬁll out multiple applications in a similar manner and could cluster
on characteristics such as industry and geography. Exhibit 2 shows examples from 4,300
$20,000 ﬁrst draw loans made by Cross River to businesses in the “Insurance Agencies and
Brokerage Industry” in Illinois, mainly in the Chicago area, almost all of which have one
employee. These are followed by examples from 938 $20,000 ﬁrst draw loans by Cross River to
businesses engaged in “All Other Miscellaneous Crop Farming,” most of which have exactly
one or eight employees.28
Most of these loans are in urban areas of Chicago, frequently
in apartment dwellings, where it is diﬃcult to see how crop farming is performed. There
are also another 3,068 $20,000 ﬁrst draw loans by Cross River in Illinois to borrowers in
other industries (including 706 to business in “All Other Personal Services,” 351 to “General
Freight Trucking, Local,” 337 to “Other Performing Arts Companies,” and 300 to “New
Single-Family Housing Construction (except For-Sale Builders)”). In addition to having the
same loan amount and similar industries, these $20,000 loans were almost non-existent until
the very end of round 2. Speciﬁcally, 40.8% were originated in late July and early August of
2020 during the ﬁnal two weeks of round 2, and 56.8% were originated in round 3. Overall,
48.9% of Cross River’s Illinois loans between July 21, 2020 and August 8, 2020 and 17.6% of
Cross River’s Illinois round 3 loans are for $20,000, compared to 1.1% of Cross River’s Illinois
28There are an additional 1,574 loans for amounts besides $20,000 by Cross River in Illinois to business in “Insurance Agencies
and Brokerage Industry” and 646 to the “All Other Miscellaneous Crop Farming” industry.
18

loans before July 21, 2020 and 2.1% of Cross River’s loans in other states. This pattern is
particularly suspicious given that the US Census CBP reports 2,207 “Insurance Agencies
and Brokerage Industry” establishments in Cook County, Illinois, which is about half the
number of ﬁrst draw loans made in this industry by Cross River alone (4,384 loans, of which
3,321 are for exactly $20,000).29 To systematically look for similar patterns throughout the
PPP data, we compare PPP numbers to overall establishment counts in the 2019 US Census
CBP database.
Because the CBP data does not include self-employed and independent
contractors as establishments, we exclude loans to these business types from our analysis.
Panel A of Figure 7 plots histograms of FinTech (red bars) and traditional (gray bars)
lender loans by the ratio of ﬁrst-draw PPP loans to census establishments in the loan’s indus-
try and county. For FinTech lenders, 40.4% of loans exceed industry-county establishment
counts, and this occurs 14.9% of the time for traditional lenders. For loans in the tails the
diﬀerences are even more extreme with 33.7% of loans exceed industry-county establishment
counts by a factor of more than two for FinTech lenders and 8.8% for traditional lenders.
Even further in the right tail, 8.2% of FinTech loans exceed industry-county establishment
counts by a factor of more than ten as compared to 0.9% of traditional loans.30 It is possible
that some excess PPP loans may be due to the missing establishments in the CBP data,
industry misclassiﬁcations, or other errors in the data. Nonetheless, the large excess loan
rate for FinTech lenders is diﬃcult to explain, particularly since it is so much higher than
traditional lenders.
Panel A of Figure 7 also plots, for FinTech and traditional lenders separately, the per-
centage of loans ﬂagged by one of the four primary suspicious loan ﬂags by the ratio between
PPP ﬁrst-draw loans and CBP establishments. The ﬂag rate increases substantially as the
loan-to-establishment ratio increases, particularly for FinTech lenders. Whereas 13.8% of
FinTech and 5.8% of traditional loans in industry-county pairs with a loan-to-establishment
ratio at or below one are ﬂagged by at least one of the primary misreporting indicators, the
ﬂag rates are 43.1% for FinTech and 11.9% for traditional loans where loan-to-establishment
ratios are above two.
Panel B of Figure 7 plots separate rates for each of the four suspicious loan ﬂags, with
consistent results for all measures. As one moves to ratios above one, indicating more PPP
29Excluding Cross River’s loans, there are 1,700 ﬁrst draw loans to Cook County businesses in this industry, which is already
77% of the establishment count provided by the CBP. Loan counts for “All Other Miscellaneous Crop Farming” also appear
to be high, but the CBP data does not have a comparable establishment count for this industry because it does not include
agricultural establishments.
30Excess loan percentages are calculated by assigning a weight to each loan based on the inverse of its industry-county’s
loan-to-establishment ratio. Speciﬁcally, let r be the loan-to-establishment ratio in the loan’s industry-county pair, the weight
is 0 if r ≤1 and 1 – 1/r if r > 1. The interval limits are changed to 2 (10) instead of 1 for the 33.7% (8.2%) and 8.8% (0.9%)
ﬁgures.
19

loans in an industry-county than listed in the CBP, the number of suspicious loans ﬂagged
increases substantially for all of the suspicious loan measures. This is true for both FinTech
and traditional lenders, but the increase is generally steeper for FinTech lenders, consistent
with FinTech loans in industry-county pairs with high loan-to-establishment ratio being
particularly suspicious.
3.4
Loan Clustering
In addition to exhibiting geographic and industry clustering, many of the examples dis-
cussed above also feature identical loan amounts and job numbers. If networks submitting
ﬁctitious loan applications repeat the same application information across multiple loans,
lenders may have many loans in a geographic region with similar industries, loan amounts,
or jobs reported. There will clearly be some loan similarities by chance and due to lender
specialization, but it is instructive to quantify how frequently loans cluster. For each lender-
county pair with at least 25 loans, we calculate concentration ratios for the industry, loan
amount (rounded to $100), and reported jobs (excluding one because it is common across
all lenders and counties). The concentration ratios are based on the sum of squared shares
of loans with a characteristic.31 Then, we rescale each of the concentration ratios to have
a median of 1,000 and an interquartile range (IQR) of 300 so that the three concentration
ratios have similar impacts on the overall concentration measure. Finally, we average the
three concentration ratios for each lender-county pair.
The bars in Panel A of Figure 8 plot the distribution of scaled concentration ratios
separately for FinTech and traditional loans.
High concentration ratios are much more
common for FinTech loans with 88.4% of FinTech loans in lender-county pairs with a scaled
concentration ratio above 1,000, compared to 21.3% of loans for traditional banks. The
dots in Panel A of Figure 8 plot how the incidence of the four primary suspicious loan ﬂags
changes with concentration ratio. When the scaled concentration ratio is below 1,000, 10.2%
of FinTech loans and 6.8% of traditional loans have at least one ﬂag. However, when the
scaled concentration ratio is above 1,300, this grows to 36.2% for FinTech loans and 7.3% for
traditional loans. Panel B of Figure 8 shows that similar patterns hold for each of the four
suspicious loan ﬂags individually. The overall pattern is similar to the previous measures:
FinTech lenders have much higher loan concentration ratios, and high concentration ratios
are highly related to the suspicious loan ﬂags, particularly for FinTech loans. This pattern
is exactly what one would expect if the indicators are picking up misreported FinTech loans
31For example, let i = 1, 2, ..., n represent the n industries in a given lender-county pair, then Concentrationindustry =
Pn
i=1 s2
i where si is the percentage of loans in the lender-county that are in industry i times 100 (e.g., 6.2 for 6.2%). Note
that this concentration ratio is the same as a Herﬁndahl-Hirschman Index (HHI), which is commonly used to measure market
concentration.
20

and is diﬃcult to explain with innocent mistakes or errors in the data.
3.5
Criminal Records
Recidivism statistics show that individuals with past criminal histories are more likely
to commit crimes in the future (Alper et al. 2018). The PPP originally prohibited loans
to businesses more than 20% owned by individuals currently subject to criminal charges,
incarceration, probation, or parole or who had been convicted of a felony within the past ﬁve
years. These restrictions were relaxed somewhat in June 2020 to permit loans to businesses
owned by individuals facing misdemeanor charges and those with convictions, probation, or
parole for most felonies more than a year in the past.32 To assess the prevalence of criminal
records among PPP borrowers, we collect criminal histories for a random sample of 150,000
round 1 and 2 loans to individual names in the PPP data that can be matched to LexisNexis
public records data.
Panel A of Figure 9 plots the percentage of borrowers with felony criminal records in
2000–2020 within the sample of 150,000 individual borrowers for whom we collected back-
ground information.33 Felony criminal records are present for 4.9% of non-bank FinTech
borrowers and 4.6% of online bank FinTech borrowers have criminal records compared to
only 1.3% of traditional borrowers. There is also a strong relation between criminal records
and both the primary and secondary indicators. The EIDL misreporting indicator seems to
be capturing the highest percentage of criminals. We conﬁrm that these relations are robust
and statistically signiﬁcant by regressing an indicator for having a criminal record on the
other primary and secondary risk ﬂags for loans originated by FinTech lenders.34
Panel B of Figure 9 examines how criminal records vary across lenders with a clear positive
relation between the percentage of a lender’s sampled borrowers with criminal records and
the percentage of its overall loans with one of the primary suspicious loan ﬂags. In particular,
the four lenders with the highest criminal record percentages (MBE, Cross River, Fundbox,
and Kabbage, all of which are FinTech) also have the highest primary ﬂag rates.35
32The ﬁve-year criminal record prohibition was only retained for ﬁnancial crimes such as fraud and embezzlement. As a
result, many individuals with criminal records were legally eligible for PPP loans. Nonetheless, a criminal record is a potential
risk factor.
33Ninety-ﬁve percent conﬁdence intervals based on standard errors clustered by zip code and lender are plotted on top of
the bars. Panel A of Figure IA.8 replicates this ﬁgure using felonies from 2015-2020. While the percentage of borrowers with
felonies is lower across the board, the relative results remain.
34Results are reported in Table IA.V. The regressions control for loan size and number of jobs with business type, industry
× CBSA, and lender ﬁxed eﬀects. Standard errors are double-clustered by zip code and lender. In all cases, the coeﬃcients
are positive, statistically signiﬁcant, and economically large for FinTech loans (Panel A) with almost no relation between the
misreporting indicators and criminal records for traditional loans (Panel B).
35Panel B of Figure IA.8 replicates this ﬁgure using felonies post-2005, post-2010, and post-2015.
While the percentage
of borrowers with felonies decreases as the time period is decreased, the relative results remain.
Additionally, Panel C of
Figure IA.8 replicates this ﬁgure using bankruptcy ﬁllings post-2015 and ﬁnds similar results.
21

3.6
Relation Between Primary and Secondary Flags
We have already seen that the primary ﬂags are strongly predictive of one another, and
the evidence in Figures 5–9 show strong relations between the primary and secondary ﬂags.
In Table IV, we more formally assess these relations with regression analysis controlling for
loan size and number of jobs with zip code, business type, industry × CBSA, and lender
ﬁxed eﬀects. The dependent variable in the regressions is an indicator variable for the loan
having at least one of the primary ﬂags. Standard errors are double clustered by zip code and
lender. The secondary ﬂags are all interacted with an indicator variable for FinTech loans,
so the direct coeﬃcients represent eﬀects for traditional loans. Four of these ﬁve eﬀects are
positive and signiﬁcant with magnitudes ranging from 6.4% to 25.0% of mean misreporting
rate.
Further, all of the interactions between the secondary ﬂags and the indicator for
FinTech loans are large and positive, and almost all are signiﬁcant. As a result, all of the
secondary ﬂags strongly relate to the primary ﬂags for FinTech loans, with relations that
are much stronger than for traditional loans. For compensation near $100,000 and rounded
compensation, the eﬀects for FinTech loans are 5.75 and 3.77 times as high as the eﬀects
for traditional loans, respectively. For criminal records and high loan concentration, the
eﬀects for FinTech loans are over 2.41 and 2.13 times as large as the traditional loan eﬀects,
respectively. Lastly, for industry overrepresentation, there is a strong FinTech eﬀect despite
essentially no relation for traditional loans. We also examine relations between the primary
and secondary ﬂags at the lender level (see Figure IA.9) and ﬁnd that except for monthly
rounding, lenders with high levels of each secondary ﬂags tend to be the same lenders that
have high levels of the primary ﬂags.
4
How Many PPP Loans Are Suspicious?
In this section, we quantify ranges of suspicious loans based on the primary and secondary
ﬂags developed in the previous two sections. Panel A of Figure 10 plots ﬂag rates for each
of the four primary ﬂags along with overall suspicious lending rates. Our primary measure
consists of loans that have at least one primary ﬂag, plotted as the total height of the bars.
By this measure, 1,515,887 loans representing 12.9% of the PPP and totaling $68.9B are
suspicious.36
FinTech lenders are responsible for a disproportionate share of suspicious loans. Com-
bined, non-bank FinTech and online bank FinTech originated 936,084 suspicious FinTech
loans totaling $21.6B. This means FinTech lenders originated 61.8% of ﬂagged loans despite,
36In addition, the EIDL > PPP ﬂag also provides an indication of misreporting in the EIDL and EIDL Advance program.
In particular, 207,595 EIDL Advances (10.3% of those matched to a PPP loan), totaling $1.79B, have potential misreporting.
22

substantially outpacing their overall FinTech 33.7% market share of loans.37
As a share
of loans originated by each lender type, 7.4% of traditional loans have at least one of the
primary suspicious loan ﬂags compared to 25.2% for non-bank FinTech and 20.7% for online
bank FinTech.
While some of the loans ﬂagged as suspicious by the primary measures may be sincere
mistakes or errors in the data, these four measures also surely miss many fraudulent loans.
This is particularly true for the business registry and EIDL > PPP Jobs ﬂags, which only
apply to subsets of loans (corporate/LLC loans and loans with matched EIDL Advances,
respectively).38 Thus, despite having much higher rates for these ﬂags within the relevant
subsets of loans, these ﬂags are relatively uncommon overall, especially for FinTech lenders.
As a more lenient measure of suspicious lending that is less sensitive to these restrictions,
Panel A of Figure IA.10 plots suspicious loan rates including all loans with any primary or
secondary ﬂag. By this measure, 5,826,006 loans totaling $300B are suspicious with 2,679,848
suspicious FinTech loans (46.0% of ﬂagged loans) totaling $61.5B.
As a more conservative estimate, we consider loans that have at least one primary ﬂag
plus an additional primary or secondary ﬂag. While this measure almost certainly misses
considerable misreporting, it has the beneﬁt of dropping sincere mistakes or errors in the
data that are isolated to a single measure. Under this more conservative measure, 1,057,613
loans totaling $35.4B are suspicious. Of these loans, 777,490 ($17.5B) are FinTech. This is
an even larger FinTech share than for the primary measure because 83.1% of FinTech loans
with a primary ﬂag are further conﬁrmed by an additional ﬂag while the corresponding
ﬁgure is only 45.2% for traditional loans. The higher conﬁrmation rate for FinTech loans is
consistent with ﬂagged FinTech loans being far more likely to be fraudulent as opposed to
simply reﬂecting honest explanations or errors in the data.
The last three bars of Panel A plot suspicious lending rates by rounds of the program
with the clear pattern that suspicious lending increased over time. In round 1, 6.4% are
suspicious, compared to 7.9% in round 2 and 17.1% in round 3. The conservative measure
with an additional conﬁrmatory ﬂag follows the same pattern.
In Panel B of Figure 10, we plot suspicious loan rates by lender. The total height of the
bars plots the percent of loans with at least one primary ﬂag, and the solid part of the bars
plots the percent of loans with a primary ﬂag that is conﬁrmed with an additional primary or
37FinTech represents a larger share of suspicious loans than suspicious loan dollar volume because FinTech loans tend to be
smaller. The same pattern is reﬂected in FinTech overall market share, which is 33.7% of PPP loans and 12.4% of PPP dollar
lending volume.
38Corporation and LLC loans constitute 18.5% and 64.5% of FinTech and traditional loans, respectively, and 17.8% and
26.7% of FinTech and traditional loans have a matched EIDL Advance, respectively.
23

secondary ﬂag. Average rates for the two measures are plotted as solid and dashed horizontal
lines, respectively. Disparities across lenders are striking. Using the at least one primary
ﬂag measure, 13 out of 21 FinTech lenders have above average suspicious loan rates, and the
10 lenders with the most suspicious loans (eight of which are FinTech) all have at least a
quarter of their loans implicated compared to the overall average of 12.9%. In the extreme,
Lendistry, Capital Plus, and Prestamos have primary ﬂag rates of 34.3%, 32.9%, and 30.7%,
respectively. Even with the more conservative measure, requiring an additional primary or
secondary ﬂag, these three lenders have ﬂag rates of 30.0%, 31.0%, and 29.2%, respectively.
Prestamos is particularly striking because it is largest lender overall with 495,545 loans. Cross
River (second largest FinTech lender and third largest overall lender with 479,869 loans) and
Harvest (fourth largest FinTech lender and sixth largest overall lender with 433,305 loans) are
also well above the average ﬂag rate with primary ﬂag rates of 20.2% and 28.0%, respectively.
While most of the FinTech lenders cluster among the lenders with the most suspicious loans,
there are a few exceptions. In particular, Square, Capital One, and Intuit have misreporting
rates that are well under the average misreporting rates across all lenders.
4.1
Geography of Suspicious Lending
In addition to varying across lenders, suspicious lending also varies geographically. Panel
A of Figure 11 plots the percent of loans with at least one primary ﬂag in each county across
the U.S with considerable variation.39 Areas with a particularly high percentage of ﬂagged
loans cluster near New Orleans, Atlanta, and surrounding areas in Louisiana, Mississippi,
Georgia. Chicago and parts of South Carolina also exhibit elevated levels. Many counties in
these areas have suspicious lending rates in excess of 25% whereas large parts of the country
have suspicious loan rates under 10%. The geographic pattern is somewhat regional with
elevated fraud rates in the Southeast, but there are elevated counties scattered across the
country. There are also big diﬀerences across large cities. For example, Cook County, IL
has a suspicious loan rate of 31.7% compared to suspicious loan rates of 8.8% in New York
County and 6.1% in Los Angeles County.
In Panel B of Figure 11, we examine the relation between FinTech market share and
suspicious loan rates across counties and zip codes. Each dot represents a zip code. The
horizontal axis plots the percent of loans ﬂagged at the county level, and the vertical axis
plots the percent of loans ﬂagged at the zip code level. There is signiﬁcant variation across
zip codes within counties, with ﬂagged loan rates varying from 20% to 50% in many counties.
Additionally, FinTech market share (represented by the color of the dots) is strongly related
39Panel A Figure IA.11 shows geographic variation in FinTech market share.
24

to the percent of ﬂagged loans not only across counties, but also across zip codes within
counties; zip codes with the highest ﬂagged loan rates consistently have the highest FinTech
market share.40
Is geographic variation in suspicious lending related to poverty, crime, or culture? Or
does suspicious lending cluster in other ways? Table V further analyzes the geography of
suspicious PPP lending by considering relations with demographic and cultural measures
that are associated with other forms of ﬁnancial misconduct (Grullon et al. 2010; Parsons
et al. 2018; Griﬃn et al. 2019). The dependent variable is an indicator for whether a loan is
ﬂagged by at least one primary ﬂag and the explanatory variables are county-level cultural
and demographic measures.41
In column (1), public corruption convictions and religious
aﬃliation, have a positive relation with the probability of a loan being ﬂagged as suspicious
and usage of a marital inﬁdelity website (Ashley Madison) has a negative relation. The
strongest relation is for the public corruption measure. A one standard deviation increase in
per capita public corruption convictions is associated with a 1.23 ppt increase in the suspi-
cious loan rate, which is 9.5% of the mean. The other two cultural variables are economically
much less important. In column (2), we add county-level demographic control variables for
population density, median income, percentage of the population that is non-white, per-
centage of adults who are college educated, and pre-pandemic unemployment. Suspicious
lending rates decrease with population density, median income, and the percentage of college
educated adults and increase with the percentage of the population that is non-white and
pre-pandemic unemployment. Coeﬃcients on all but the percentage of the population that
is non-white are economically small.
In column (3), we add county-level FinTech market share. A one standard deviation
increase (15.7 ppt) in FinTech market share in the county is associated with a 2.69 ppt
increase in the suspicious loan rate, which is 20.9% of the mean misreporting rate. This
is a much stronger relation than any of the other county variables, and coeﬃcients for the
cultural and demographic variables generally decrease or become statistically insigniﬁcant
once FinTech market share is added to the regression. While this regression speciﬁcation
is not conducive to a causal interpretation, it indicates a strong FinTech association and
indicates that cultural variables play a minor role.
40Within a county, a 10 ppt rise in FinTech market share in a zip code is associated with a 3.21 ppt rise in suspicious lending
(see Table IA.VII).
41The regressions are at the loan level to control for jobs reported, loan size, business type, and industry code × state ﬁxed
eﬀects. Standard errors are double clustered by zip code and lender. The independent variables are standardized to have a mean
of 0 and a standard deviation of 1 at the county level. Thus, the coeﬃcients can be interpreted as the change in the probability
of a loan being ﬂagged for a one standard deviation change in the variable. Table IA.VIII shows equivalent regressions at the
county level.
25

Why does suspicious lending vary so much across geographies?
Strong clustering in
certain counties and zip codes suggests that suspicious borrowing is driven by more than
just the idiosyncratic decisions of individual borrowers. One possibility is that referral fee
programs, agent fees, kickback schemes, or local networks may arise in certain areas to
systematically attract and facilitate suspicious lending.42 Because we do not observe the
identity of agents directing or assisting the PPP borrowers, this possibility is diﬃcult to
directly test. Nonetheless, the geographic clustering of suspicious loans shown in Figure 11
is what one would expect from agents steering suspicious borrowers to FinTech lenders. If
agents are utilizing more than one FinTech lender to originate suspicious loans, one might
expect counties with many potentially misreported loans by one lender to have elevated levels
of suspicious loans and FinTech lending more generally. Consistent with this premise, when
one FinTech lender has a high ﬂagged loan rate in a county, other FinTech lenders also tend to
have elevated ﬂagged loan rates (as shown in the lower triangle of Figure IA.12). Additionally,
FinTech lenders with the highest overall ﬂagged loan rates have a high correlation in their
market shares at the county level, whereas market-share correlations between most other
lenders are slightly negative (as shown in the upper triangular of Figure IA.12).
5
Why Does Suspicious Lending Concentrate in Fin-
Tech?
FinTech lenders on average have much higher suspicious lending rates than traditional
lenders, and Tables III, IA.II, and IA.VI show that their elevated suspicious lending is not
explained by observable facets of loan composition. What could be driving the elevated ﬂag
rates for FinTech lenders?
5.1
FinTech Lender Background and Incentives
5.1.1
FinTech Lender Background
Diﬀerences across FinTech lenders give a ﬁrst clue to this puzzle. While most FinTech
lenders have high suspicious loan rates, Square and Intuit have among the lowest suspicious
loan rates of all lenders. Online lending does not appear to be the problem in and of itself.
One thing that sets Square and Intuit apart is that they have established relationships with
customers based on a broad suite of payment, accounting, payroll, and other ﬁnancial support
services.
42For example, Amur Equipment oﬀered a referral fee program and explicitly stated that the referral program required “zero-
touch and follow up on your end.” (see Tweet here). The PPP allowed lenders to pay agents 1% on loans up to $350k, 0.5% on
loans between $350k and $2M, and 0.25% on loans above $2M (see instructions here). Further, some people ﬁled PPP loans in
return for upfront and backend fees or kickbacks, which was against SBA rules (e.g., see here, here, and here).
26

By contrast, the largest FinTech PPP lender, Prestamos, is a Community Development
Financial Institution with locations in Arizona, Nevada, and New Mexico.
The second
largest FinTech lender, Cross River, is a small community bank in New Jersey that acts as
a conduit for partner FinTechs. Similarly, Capital Plus Financial, the third largest FinTech
PPP lender, is a small mortgage lender in Texas that traditionally focused on supporting
Hispanic homeownership but now appears to be almost entirely focused on PPP lending. The
number four FinTech lender, Harvest Small Business Finance, is also a small lender with
limited history. Now that the PPP has ended, Harvest’s only current product appears to
be SBA 7(a) commercial real estate loans. Benworth and Fountainhead, the other FinTech
lenders in the top ten by number of PPP loans originated, follow a similar pattern with
limited business outside of PPP lending. We systematically examine this relation and ﬁnd
that lenders who have fewer SBA loans pre-pandemic, have lent in SBA programs for fewer
years, and for whom the PPP was their ﬁrst experience with SBA lending (in particular
new FinTechs) all have higher rates of ﬂagged loans (as shown in Table IA.IX). FinTech
lenders also relied more heavily on liquidity support from the Federal Reserve than traditional
lenders.43
The six largest FinTech lenders primarily originated loans that were sourced from other
FinTech platforms. Cross River adopted this business model early in round 1 by partnering
with other FinTechs such as Intuit and Kabbage to originate PPP loans (New York Times
2020). The other large FinTech lenders originated loans sourced by two marketing FinTechs
that did not do any PPP lending until round 3, Womply and BlueAcorn (New York Times
2021a). Womply is a marketing technology ﬁrm with no lending history before participating
in the PPP. It launched a platform called Fast Lane to facilitate PPP applications that
were then originated by partner lenders including Harvest, Capital Plus, Benworth, and
Fountainhead. BlueAcorn was founded in April 2020 exclusively to source PPP loans in
partnership with Capital Plus and Prestamos. Both ﬁrms relied heavily on online advertising
promoting easy access to PPP money.
5.1.2
FinTech Fluidity
To examine relationships between these lenders, Panel A, Figure 12 plots the network
of relationships between lenders based on originating loans in the same draw to the same
non-commercial address as identiﬁed by the multiple loan ﬂag. The edges between lenders
represent the number of addresses to which both lenders originated a loan within the same
43Financing for FinTech PPP loans was in part provided with a credit facility, the Paycheck Protection Program Liquidity
Facility (PPPLF), in which the Federal Reserve extended credit to lenders using PPP loans as collateral. Figure IA.13 plots
ﬂagged loan rates relative to PPPLF advance volume for the 100 largest PPP lenders. Whereas most traditional lenders did
not use the PPPLF at all, it was a major source of funding for some of the large FinTech lenders.
27

draw. Node size is based on the number of loans at addresses ﬂagged for having multiple
loans. The thicker edges between FinTech lenders show that FinTech borrowers with mul-
tiple loans often received funds from more than one FinTech lender even within the same
draw. Speciﬁcally, 61.2% of FinTech borrowers with multiple loans to the same address
split their loans across multiple lenders. Shared FinTech lending to the same address is
largely explained by FinTech portals sourcing loans for multiple lenders. For example, the
plot shows a strong relationship between Prestamos and Capital Plus, the two lenders that
partnered with BlueAcorn. This is likely from borrowers applying for multiple loans through
BlueAcorn, some of which were originated by Prestamos while others were directed to Capital
Plus. Similarly, there are strong relationships between Harvest, Benworth, Capital Plus, and
Fountainhead, all of which are Womply partners. By contrast, for traditional lenders 81.4%
percent of traditional borrowers who took out multiple loans received all their loans from
the same lender. Some exceptions include Customers Bank, Amur Equipment, and Bank of
America. Customers Bank has known FinTech aﬃliations even though it does not meet the
formal FinTech criteria and Amur Equipment went from specializing in equipment ﬁnancing
to becoming one of the largest PPP lenders by advertising their “lighting fast portal” on
social media.44
As another way to examine relationships between lenders, we track borrowers switching
lenders between their ﬁrst and second PPP loan draws. If a borrower already received a ﬁrst
draw from the same lender, obtaining second draw only required refreshing the application
with some additional information. This provided a strong incentive for borrowers to use
the same lender. For borrowers with ﬁrst draw loans ﬂagged for potential misreporting that
subsequently took out second draw loans, Panel B of Figure 12 shows the movement of these
loans across rounds to diﬀerent lenders.45 The large movements and connections between
FinTech lenders likely reﬂect online lending portals switching lenders. Overall, the graphs
highlight the ﬂuid nature of the FinTech space where online portals originating loans can
easily originate their loans through diﬀerent lenders, and suspicious borrowers can utilize
several platforms or switch platforms. The lack of relationship banking within the FinTech
space may be advantageous to expand access to capital (Erel and Liebersohn 2021), but it
also appears to be expedient for dubious lending.
44Customers Bank directly worked with multiple FinTech lenders, in particular Kabbage and Cross River (see press release
here). See one of many posts on social media by Amur Equipment here.
45The thickness of the edges between lenders is proportional to the number of ﬂagged loans that changed lenders between
the ﬁrst and second draws. The switches between the ﬁrst and second draw are clockwise. Node size is based on the number of
ﬁrst draw loans (with a matching second draw) and second draw loans by each lender. Figure IA.14 replicates this ﬁgure using
all second draw loans.
28

5.1.3
FinTech Revenue
PPP lending had the potential to be a proﬁtable business for lenders.
Lenders were
initially compensated with processing fees of 5% for loans up to $350,000, 3% for loans
between $350,000 and $2,000,000, and 1% for loans of $2,000,000 or more. For loans made
in 2021, fees for small loans were increased to the lesser of 50% or $2,500 for loans below
$50,000.46
Based on this fee schedule, we estimate that PPP lending generated $38.8B
of lender processing fees, $9.4B of which went to FinTech lenders (see Table IA.X). The
top four FinTech lenders alone likely generated $4.4B in processing fees, including $1.18B
to Prestamos, $1.10 to Capital Plus, $1.06B to Harvest, and $1.04B to Cross River. The
average processing fee for FinTech PPP loans was 18.3% of the loan balance, largely driven
by the high processing fees for small loans in round 3.47 We lack data on cost structure
associated with PPP lending and do not observe how lender fees are shared with partner
organizations used to source the loans such as Womply and BlueAcorn.48
5.2
Did FinTech Lenders Improve Standards Over Time?
We consider two potential scenarios under which suspicious lending could arise:
• Scenario A: The lender does not want to facilitate ﬁctitious loans but is not performing
great due diligence. As it learns over time, the lender cracks down on the fraud.
• Scenario B: The lender is aware of the existence of or potential for fraud within its
PPP loans but ignores this risk because there is little downside for the lender. This
may be particularly true for lenders with little reputation or other business to protect.
Under scenario A, when lenders are new to PPP lending they may facilitate questionable
loans, but over time as they experience more loans with improbable features, they should
originate fewer of these loans. In this case, borrowers who wish to commit loan fraud would
need to rotate among lenders. In scenario B, the amount of suspicious lending could grow
through time as lenders develop a reputation for rapid and unquestioning approval.
Scenario A predicts:
1. Loan misreporting will decrease over time as lenders become more aware and develop
46See fee schedule here.
47The average FinTech processing fee for round 1 and 2 was 4.96%. This dramatically increased to 21.7% in round 3.
48Capital Plus (the second largest FinTech lender and fourth largest lender overall) received a PPP loan of $376,800, reportedly
to cover payroll for its 28 employees. The loan was approved in April 2020, potentially before their business opportunities as a
PPP lender, most of which occurred in round 3, were apparent. Similarly, Benworth Capital Partners (fourth largest FinTech
and eighth largest lender overall) received a PPP loan of $100,600 for its 13 employees on April 5, 2020, DreamSpring received a
PPP loan of $757,753 for its 54 employees on April 27, 2020, and Amur Equipment was approved for a PPP loan of $2,817,846
on May 2, 2020 but then repaid/canceled its loan 12 days later.
29

systems to screen out suspicious loans.
2. Suspicious borrowers will be less likely to receive a repeat loan from the same lender
compared to other borrowers.
3. Regions with high misreporting in rounds 1 and 2 will face extra scrutiny from lenders,
which will decrease round 3 misreporting.
Scenario B predicts:
1. Loan misreporting will grow over time as borrowers learn about the potential for fraud.
2. Borrowers with suspicious ﬁrst draw loans in rounds 1 and 2 will be able to obtain
second draw loans in round 3 from the same lenders.
3. Regions with high misreporting in rounds 1 and 2 will have the same or more misre-
porting in round 3.
Did lenders improve their loan screening over time? While we do not observe denied
applications or speciﬁc lender practices, we can observe how loans that were approved and
funded changed over time. We have already seen that the overall rate of suspicious lending
grew over time from round 1 to round 3. Panel A of Figure 13 plots more granular suspi-
cious loan rates on a weekly basis separately for non-bank FinTech, online bank FinTech,
and traditional lenders. For the FinTech lenders, loans became more suspicious over time
throughout rounds 1 and 2. The rate of suspicious lending dropped at the beginning of
round 3, likely due to pent up demand for second draw loans from legitimate borrowers.
Most round 3 FinTech lending occurred later in round 3 (see Figure 1), and as round 3
progressed, the suspicious loan rate rose dramatically to around 30% of loans ﬂagged as
suspicious during April and early May of 2021. PPP funds for most loans were exhausted on
May 4, 2021 (New York Times 2021b). The suspicious loan rate fell after this date, but this
could be due to loan composition since funding after May 4 was only available for prioritized
community ﬁnancial institutions and some loans that were already under review prior to
May 4. Suspicious lending by traditional lenders also grew over time, but at a much lower
rate. FinTech and traditional lenders both started the PPP with suspicious loan rates of
around 10%, but by the end of the program the FinTech suspicious loan rate was close to
30%, more consistent with scenario B.49
We also examine lending growth and changes in suspicious loan rates across rounds at the
lender level. Panel A of Figure IA.16 shows that most lenders had higher suspicious lending
rates in round 3 than in rounds 1 and 2. Additionally, many of the FinTech lenders with the
49Panel B of Figure IA.10 shows similar trends for each primary ﬂag individually.
30

highest suspicious loan rates in rounds 1 and 2 also had the most growth in lending and the
most growth in suspicious loans in round 3. In Table VI, we ask whether lenders appear to
be learning by regressing indicators for the four primary ﬂags in round 3, individually and
combined, on lenders’ rounds 1 and 2 misreporting rates for the same ﬂags. As in previous
regressions, we control for loan size and jobs with zip code, business type, industry × CBSA
ﬁxed eﬀects. For FinTech lenders, we ﬁnd highly economically and statistically signiﬁcant
positive relations across the board with largely insigniﬁcant and one negative relationship
for high implied compensation for traditional lenders. For traditional banks there is overall
no relationship between the lenders suspicious loan percentages in rounds 1 and 2 and their
lending in Round 3. By contrast, FinTech lenders have persistent and increasing levels of
suspicious loans through time, consistent with scenario B above.
To assess prediction 2, if lenders are taking steps to screen out questionable loans, then
their borrowers with questionable ﬁrst draw loans in rounds 1 and 2 may get rejected when
they apply for a second draw loan in round 3. To examine this, we estimate regressions to
determine whether a ﬁrst draw borrower is more or less likely to receive a second draw loan
from the same lender if its ﬁrst draw loan is ﬂagged by one of the primary misreporting indi-
cators. Table VII shows that traditional loans which are ﬂagged in the ﬁrst two rounds have
a statistically signiﬁcant decrease in the probability of receiving a second draw loan from the
same lender of 3.10 ppt (with t-stat of -11.94) and FinTechs have a smaller and statistically
insigniﬁcant decrease of 0.98 ppt (with t-stat of -0.83).50 This provides some indication that
traditional banks were less likely to continue lending to borrowers with previous suspicious
borrowing, but FinTechs do not seem to be screening or implementing procedures which
make it less likely for questionable borrowers to continue receiving funds in the form of a
second draw. Columns (3) and (4) of Table VII condition on the borrower receiving a second
draw (either from the same or diﬀerent lender) with similar results.51
To assess prediction 3, we examine whether areas with high misreporting in rounds 1 and
2 had higher or lower misreporting in round 3. Panel B Figure 13 plots the percentage of
loans ﬂagged in rounds 1 and 2 in each zip code on the horizontal axis, and the percentage
of ﬂagged loans in round 3 in the same zip code on the vertical axis. The left subpanel
uses all loans, the middle uses FinTech loans, and the right uses traditional loans. Each dot
represents a zip code, and the size of the dots corresponds to the number of loans in the
50These results are based on SameLenderi being set to 0 if the borrower did not get a second draw at all.
51In Figure IA.15, we show results separately for individual lenders with lender ﬁxed eﬀects and lender interactions. The
inclusion of the lender ﬁxed eﬀects ensures that the reported coeﬃcient is due solely to diﬀerences in the lender’s behavior
towards ﬂagged and nonﬂagged loans rather than systematic changes in the lender’s behavior. For most traditional lenders,
borrowers with a ﬂagged ﬁrst-draw loan are less likely to receive a second draw loan from the same lender, but for several
FinTech lenders, suspicious ﬁrst-draw borrowers are slightly more likely to receive a second draw.
31

zip code. Purple to blue colors indicate that a zip code had fewer loans in round 3 than in
rounds 1 and 2, yellow colors indicate that a zip code had about the same number of loans
in round 3 compared to rounds 1 and 2, and orange to red colors correspond to an increase
in the number of loans in the zip code.
The ﬁgure displays three interesting ﬁndings. First, most zip codes (85.4%) are above
the 45-degree line in the left subpanel, indicating that misreporting rates increased in round
3 almost everywhere. In many zip codes (35.0%), the percentage of loans that are ﬂagged
as suspicious in round 3 is more than twice as high as in rounds 1 and 2. Second, the zip
codes with the highest suspicious loan rates experienced the most growth in lending. Many
zip codes with the highest level of ﬂagged loans in round 3 have more than three times the
number of loans in round 3 compared to rounds 1 and 2, suggesting that signiﬁcant portions
of zip code-level loan growth in round 3 may be due to suspicious lending practices. Third,
the middle and right subpanels diﬀerentiate between FinTech and traditional lenders and
show that lending growth and increased misreporting rates are almost entirely from FinTech
lenders. Traditional lenders had only small increases in suspicious loan indicators, and their
lending generally decreased. In contrast, FinTech lenders increased the number of loans they
originated and increased their suspicious lending rates in almost all zip codes. Additionally,
FinTech lending growth was highest in zip codes with the highest misreporting rates.52 We
also test these results at the zip code-lender level in Table IA.XI with zip code and lender
ﬁxed eﬀects and ﬁnd that a 10 ppt increase in ﬂagged loans in a zip code-lender pair in
rounds 1 and 2 is associated with a 22.1 ppt increase in lending for a FinTech lender and an
insigniﬁcant increase of 1.3 ppt for a traditional lender. There is also strong persistence of
suspicious lending across rounds within zip code-lender pairs.
5.3
Repayments and Enforcement Actions
The economics of crime depends crucially on a crime’s expected penalty and probability
of detection (Becker 1968). The US Department of Justice is pursuing criminal complaints
alleging PPP fraud, and some borrowers have voluntarily repaid their loans without applying
for loan forgiveness or had their loan canceled.53 However, the magnitude of these enforce-
ment actions is tiny. Compared to the 2.3 million loans we identify as suspicious, the DOJ
52Results are similar at the county and state level. See Panels C and D of Figure IA.16. Panel B of Figure IA.11 also plots
lending growth by county.
53An earlier version of this paper based on the May 3, 2021 SBA data release was strongly criticized by several PPP lenders
for including loans that were eventually canceled. Dropping the canceled loans led to only minuscule reductions in suspicious
loans. The current version of the paper is entirely based on the June 30, 2021 SBA data release and thus does not include these
canceled loans.
32

has publicized 162 criminal complaints regarding only 355 loans.54 SBA data indicates that
only 16,930 round 1 and 2 loans were repaid between December 1, 2020 and June 30, 2021.
Repayment, enforcement action, and cancelation rates are all elevated for ﬂagged loans (see
Table IA.XII). While more enforcement actions may be forthcoming, there appears to be
little penalty for most suspicious lending thus far.
6
Conclusion
We examine four primary and ﬁve secondary metrics related to potentially misreported
loans. FinTech loans are highly suspicious at a rate of over six times that for traditional
lenders. Eight of the ten lenders with the highest rates of suspicious loans are FinTech lenders
and the remaining two traditional banks function much like FinTechs, including one with a
“lighting fast portal.” We estimate the total amount of potential misreporting as 1.52 million
loans with a balance of $68.9 billion based on the four primary metrics, and $35.4 billion
(1.06 million loans) under a more conservative estimate requiring an additional indicator.
The total amount of misreporting is likely larger than either estimate because many of our
indicators are only available for a subset of loans, and a much larger set of loans has at least
one of the nine indicators. In the early stages of the PPP, less than 10% of FinTech loans
were potentially misreported, but the percentage of suspicious FinTech loans increased to
more than 30% by April 2021. Extremely few of the suspicious loans have been prosecuted
by authorities or repaid.
Our ﬁndings have important policy implications. First, the PPP did not include robust
veriﬁcation requirements, but traditional banks may have been more apt to follow standard
lending practices. The lack of rigorous veriﬁcation, seems to have led to substantial costs
to taxpayers, especially in later rounds. Second, FinTech lending, though quite successful
at adapting to new environments and quickly disbursing funds, seemingly needs to improve
due diligence practices.
Two established FinTech lenders persistently have low rates of
misreporting, indicating that FinTech lending need not be substandard. Third, our evidence,
along with evidence that the PPP saved relatively few jobs at a high cost (Autor et al. 2020;
Chetty et al. 2020; Granja et al. 2020), provides growing evidence that the PPP may not
have been an eﬃcient source of capital allocation. Fourth, incentives in the PPP appear
misaligned in that FinTech lenders with widespread indicators of misreporting made billions
of dollars dispersing loans with lax oversight procedures.
54See the Internet Appendix for additional details on the repayment and enforcement action data. Of the DOJ enforcement
action loans with enough data to be matched to the PPP loan-level data, 153 loans were originated by FinTech lenders and 126
were originated by traditional banks. There are likely other cases that are still sealed, are in early stages of investigation, or
are not included on the DOJ website for other reasons. We focus on loans from rounds 1 and 2 for this analysis to allow more
time for repayments and enforcement actions.
33

Finally, the increasing scale of misreporting through time indicates that current penalty
and enforcement systems are not eﬀective. If the system is not changed, the most likely
outcome is even more of the same. This paper is also an example of how forensic research
(Zitzewitz 2012) can more fully investigate the rent-seeking dimension of ﬁnance (Zingales
2015).
Government agencies can assist this transparency goal by making detailed data
available to the public. We hope to see future research with additional forensic investigation
of the PPP as well as other recent government and private lending programs.
34

References
Alper, Mariel, Matthew R. Durose, and Joshua Markman, 2018, 2018 update on prisoner
recidivism: A 9-year follow-up period (2005-2014), Special Report by Bureau of Justice
Statistics: Oﬃce of Justice Programs.
Amiram, Dan, and Daniel Rabetti, 2020, The relevance of relationship lending in times of
crisis, Working Paper.
Atkins, Rachel M. B., Lisa D. Cook, and Robert Seamans, 2021, Discrimination in lending?
Evidence from the Paycheck Protection Program, Working Paper.
Autor, David, David Choz, Leland D. Cranez, Mita Goldarx, Byron Lutzz, Joshua Montesz,
William B. Petermanz, David Ratnerz, Daniel Villarz, and Ahu Yildirmazx, 2020, An
evaluation of the Paycheck Protection Program using administrative payroll microdata,
Working Paper.
Avis, Eric, Claudio Ferraz, and Frederico Finan, 2018, Do government audits reduce corrup-
tion? Estimating the impacts of exposing corrupt politicians, Journal of Political Economy
126, 1912–1964.
Bandiera, Oriana, Andrea Prat, and Tommaso Valletti, 2009, Active and passive waste in
government spending: Evidence from a policy experiment, American Economic Review
99, 1278–1308.
Bartik, Alexander W., Marianne Bertrand, Zoe Cullen, Edward L. Glaeser, Michael Luca,
and Christopher Stanton, 2020, The impact of COVID-19 on small business outcomes and
expectations, Proceedings of the National Academy of Sciences 117, 17656–17666.
Bartlett, Robert, Adair Morse, Richard Stanton, and Nancy Wallace, 2021, Consumer-
lending discrimination in the FinTech era, Journal of Financial Economics.
Becker, Gary S., 1968, Crime and punishment: An economic approach, Journal of Political
Economy 76, 169–217.
Beggs, William, and Thuong Harvison, 2021, Fraud and abuse in the Paycheck Protection
Program? Evidence from investment advisory ﬁrms, Working Paper.
Begly, Taylor A., Umit G. Gurun, Amiyatosh Purnanandam, and Daniel Weagley, 2021,
Disaster lending: “fair" prices, but “unfair" access, Working Paper.
Ben-David, Itzhak, 2011, Financial constraints and inﬂated home prices during the real
estate boom, American Economic Journal: Applied Economics 3, 55–87.
Bloomberg Businessweek, 2020, PPP scammers made ﬁntech companies their lenders of
choice, Michelle F. Davis, October 7.
Buchak, Greg, Gregor Matvos, Tomasz Piskorski, and Amit Seru, 2018, Fintech, regulatory
arbitrage, and the rise of shadow banks, Journal of Financial Economics 130, 453–483.
Chernenko, Sergey, and David S. Scharfstein, 2021, Racial disparities in the Paycheck Pro-
tection Program, Working Paper.
35

Chetty, Raj, John N. Friedman, Nathaniel Hendren, Michael Stepner, and The Opportunity
Insights Team, 2020, The economic impacts of COVID-19: Evidence from a new public
database built using private sector data, Working Paper.
Deﬂo, Esther, 2017, The economist as plumber, American Economic Review 107, 1–26.
Duchin, Ran, Xiumin Martin, Roni Michaely, and Ivy Wang, 2021, Concierge treatment
from banks: Evidence from the Paycheck Protection Program, Working Paper.
Erel, Isil, and Jack Liebersohn, 2021, Does ﬁntech substitute for banks? Evidence from the
Paycheck Protection Program, Working Paper.
Faulkender, Michael W., Robert Jackman, and Stephen Miran, 2021, The job preservation
eﬀects of Paycheck Protection Program loans, Working Paper.
Fuster, Andreas, Matthew Plosser, Philipp Schnabl, and James Vickery, 2019, The role of
technology in mortgage lending, The Review of Financial Studies 32, 1854–1899.
Garmaise, Mark J., 2015, Borrower misreporting and loan performance, The Journal of
Finance 70, 449–484.
Glaeser, Edward L., and Claudia Goldin, 2006, Corruption and Reform: Lessons from Amer-
ica’s Economic History (University of Chicago Press).
Glancy, David, 2021, Bank relationships and the geography of PPP lending, Working Paper.
Gopal, Manasa, and Philipp Schnabl, 2020, The rise of ﬁnance companies and ﬁntech lenders
in small business lending, Working Paper.
Granja, João, Christos Makridis, Constantine Yannelis, and Eric Zwick, 2020, Did the Pay-
check Protection Program hit the target?, Working Paper.
Griﬃn, John M., Samuel Kruger, and Gonzalo Maturana, 2019, Personal inﬁdelity and
professional conduct in 4 settings, Proceedings of the National Academy of Sciences 116,
16268–16273.
Griﬃn, John M., and Gonzalo Maturana, 2016, Who facilitated misreporting in securitized
loans?, The Review of Financial Studies 20, 384–419.
Grullon, Gustavo, George Kanatas, and James Weston, 2010, Religion and corporate
(mis)behavior, Working Paper.
Hanson, Samuel G., Jeremy Stein, Adi Sunderam, and Eric Zwick, 2020, Business credit
programs in the pandemic era, Brookings Papers on Economic Activity 3–60.
Hart, Oliver D., Andrei Shleifer, and Robert W. Vishny, 1997, The proper scope of gov-
ernment: Theory and an application to prisons, Quarterly Journal of Economics 112,
1127–1261.
Howell, Sabrina T., Theresa Kuchler, David Snitkof, Johannes Stroebe, and Jun Wong,
2021, Racial disparities in access to small business credit: Evidence from the Paycheck
Protection Program, Working Paper.
36

Jagtiani, Julapa, and Catharine Lemieux, 2019, The roles of alternative data and machine
learning in ﬁntech lending: Evidence from the LendingClub consumer platform, Financial
Management 48, 1009–1029.
Jiang, Wei, Ashlyn A. Nelson, and Edward Vytlacil, 2014, Liar’s loan? eﬀects of origination
channel and information falsiﬁcation on mortgage delinquency, The Review of Economics
and Statistics 96, 1–18.
Keys, Benjamin J., Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2010, Did securitization
lead to lax screening? Evidence from subprime loans, The Quarterly Journal of Economics
125, 307–362.
Kruger, Samuel, and Gonzalo Maturana, 2020, Collateral misreporting in the residential
mortgage-backed security market, Management Science 67, 2657–3320.
Li, Lei, and Philip E. Strahan, 2021, Who supplies PPP loans (and does it matter)? Banks,
relationships and the COVID crisis, Journal of Financial and Quantitative Analysis, Forth-
coming.
Meirer, Jean-Marie, and Jake Smith, 2021, The COVID-19 bailouts, Working Paper.
Miami Herald, 2020, Copycat websites, newly minted companies, double dippers cashed in
on PPP loans, Ben Wieder and Meghan Bobrowsky, September 11.
Mian, Atif R., and Amir Suﬁ, 2017, Fraudulent income overstatement on mortgage ap-
plications during the credit expansion of 2002 to 2005, Review of Financial Studies 30,
1832–1864.
Neilson, Christopher, John Eric Humphries, and Gabriel Ulyssea, 2020, Information frictions
and access to the Paycheck Protection Program, Journal of Public Economics 190.
New York Times, 2020, The tiny bank that got pandemic aid to 100,000 small businesses,
Stacy Cowley, June 23.
New York Times, 2021a, How two start-ups reaped billions in fees on small business relief
loans, Stacy Cowley and Ella Koeze, June 27.
New York Times, 2021b, The Paycheck Protection Program is out of money., Stacy Cowley,
May 4.
Parsons, Christopher A., Johan Sulaeman, and Sheridan Titman, 2018, The geography of
ﬁnancial misconduct, The Jorrnal of Finance 73, 2087–2137.
Piskorski, Tomsaz, Amit Seru, and James Witkin, 2015, Asset quality misrepresentation by
ﬁnancial intermediaries: Evidence from the RMBS market, The Journal of Finance 70,
2635–2678.
Project on Government Oversight, 2020, Lamborghinis, strip clubs, bogus companies, and
lies: The ﬁrst 56 Paycheck Protection Program fraud cases, Ryan Summers and Nick
Schwellenbach, October 8.
37

ProPublica, 2021, Hundreds of PPP loans went to fake farms in absurd places, Derek Willis
and Lydia DePillis, May 18.
Purnanandam, Amiyatosh, 2011, Originate-to-distribute model and the subprime mortgage
crisis, Review of Financial Studies 24, 1181–1915.
Shleifer, Andrei, and Robert W. Vishny, 1993, Corruption, The Quarterly Journal of Eco-
nomics 108, 599–617.
Svensson, Jakob, 2003, Who must pay bribes and how much? evidence from a cross section
of ﬁrms, Quarterly Journal of Economics 118, 207–230.
The Wall Street Journal, 2020, Evidence of PPP fraud mounts, oﬃcials say, Ryan Tracy,
November 8.
Zingales, Luigi, 2015, Presidential address: Does ﬁnance beneﬁt society?, The Journal of
Finance 70, 1327–63.
Zitzewitz, Eric, 2012, Forensic economics, Journal of Economic Literature 50, 731–769.
38

Figure 1. Fintech Market Share
This ﬁgure shows the role that ﬁntech lenders played in the PPP. Panel A shows the number of
loans (bars) and dollar value of loans (dots) originated by the top 75 lenders (by number of loans).
Panel B shows the percentage of loans originated by ﬁntech lenders during each week of Round 1,
2, and 3 of the PPP on the left axis and the total number of loans originated each week on the the
right axis. In both panels, red represents non-bank ﬁntech lenders, cream represents online bank
ﬁntech lenders, and grey represents traditional lenders. Note that mid-August through December
2020 is not shown in Panel B since no PPP loans were originated during this period.
Panel A. Number of Loans and Dollar Value of Loans, by Lender (Top 75)
Panel B. Fintech Market Share, by Week
39

Figure 2. Business Registry and Multiple Loans Flags
This ﬁgure shows the prevalence of the business registry and multiple loans ﬂagged loans by lender.
Panel A shows the percentage of loans ﬂagged for being incorporated after February 15, 2020 (“Late
Incorporation/Filling”), being dissolved and inactive before approved for a PPP loan (“Dissolved
Business”), or not being found in the business registry for its home state or in any other state while
listing an address in its home state (“Missing Business”). Panel B shows the percentage of loans
ﬂagged as being located at a non-business, non-central (e.g., not an apartment or oﬃce building)
address that received at least three loan within the given loan’s draw (i.e., the ﬁrst or second
draw). For Panel A, only loans to businesses organized as a corporation, subchapter S corporation,
or LLC and not based in Illinois or a territory are considered; lenders originating at least 10,000
loans ﬁtting these criteria are shown. For Panel B, all loans are considered and lenders originating
at least 15,000 loans are shown. In both panels, red represents non-bank ﬁntech lenders, cream
represents online bank ﬁntech lenders, and grey represents traditional lenders.
Panel A. Business Registry Flagged Loans, by Lender
Panel B. Multiple Loans Flagged Loans, by Lender
40

Figure 3. High Implied Compensation Flag
This ﬁgure shows the relationship between a loan’s implied compensation per employee and the
average compensation in the loan’s industry (represented by NAICS [North American Industry
Classiﬁcation System] code) and region (represented by CBSA [core-based statistical area]). We
deﬁne normalized compensation as the implied compensation of the loan divided by the average
compensation in the loan’s industry-CBSA. Panel A shows the relationship between normalized
compensation and the business registry (top subpanel) and multiple loans ﬂags (bottom subpanel).
Panel B shows the percentage of loans with normalized compensation above 3 (i.e., implied com-
pensation is more than three times the industry-CBSA average) by lender. For Panel A, the left
axis shows the kernel density of loans (distribution) and the right axis shows the percentage of
ﬂagged loans in each bin (dots), where each bin is 0.2 units wide. The solid lines are third-degree
polynomial ﬁts for the percentage ﬂagged and the dashed lines are 95% conﬁdence intervals. For
the business registry subpanel, only loans to businesses organized as a corporation, subchapter S
corporation, or LLC and not based in Illinois are considered. For Panel B, only loans where the
average compensation in the loan’s industry-CBSA is less than $33,333.33 are considered; lenders
with at least 5,000 loans ﬁtting this criterion are shown.
Panel A. Business Registry and Multiple Loans Flags, by Normalized Compensation
Panel B. High Implied Compensation, by Lender

Figure 4. EIDL > PPP Jobs Flag
This ﬁgure shows the diﬀerence between the number of employees implied by a business’s EIDL
Advance amount (“EIDL Implied Jobs”) and the number of jobs reported by the business on its
PPP application (“PPP Reported Jobs”). Panel A shows the lower bound (in the absolute value
sense) of the diﬀerence between the EIDL implied jobs and PPP reported jobs by lender type.
Panel B shows the percentage of loans where the EIDL implied jobs is at least three more than
the PPP reported jobs by lender. In both panels, only loans with a matched EIDL Advance are
consider; lenders with at least 5,000 loans ﬁtting this criterion are shown in Panel B. In both
panels, red represents non-bank ﬁntech lenders, cream represents online bank ﬁntech lenders, and
grey represents traditional lenders.
Panel A. Diﬀerence Between EIDL Implied Jobs and PPP Reported Jobs
Panel B. EIDL > PPP Jobs, by Lender
42

Figure 5. Discontinuities at $100,000
This ﬁgure shows the prevalence of the primary ﬂags by the implied compensation per employee.
Panel A shows the relationship between implied compensation per employee and our four main
ﬂags combined together as at least one ﬂag and Panel B shows the relationship for each primary
ﬂag separately. For both panels, loans are binned into $2,000 wide bins (i.e., ($0k, $2k], ... , ($98k,
$100k], ... ,($128, $130k]). For Panel A, the left axis shows the percentage of loans that in each
bin (bars) and the right axis shows the percentage of the loans in the bin that are ﬂagged by the
given ﬂag (dots). In Panel B, loans are ﬁltered to corporation, S-corporation, and LLC loans for the
Business Registry subpanel, loans for which we can determine industry-CBSA average compensation
for the High Implied Compensation subpanel, and loans with a matched EIDL Advance for the EIDL
> PPP subppanel. The solid lines are third-degree polynomial ﬁts (weighted based on number of
loans in the each bin), which are separately ﬁtted for loans below $98,000 and loans above $100,000,
and the dashed lines are 95% conﬁdence intervals. Red represent ﬁntech loans and grey represent
traditional loans.
Panel A. Percentage Flagged, by Implied Compensation
Panel B. Individual Flags, by Implied Compensation
43

Figure 6. Rounded Loan Amounts
This ﬁgure shows the prevalence of the primary ﬂags by whether the total monthly implied com-
pensation of a loan is rounded to an interval of $500 (i.e., loan amount is within ± 50 cents of an
interval of $1,250) and lender type. Panel A shows the relationship between implied compensation
per employee and our four main ﬂags combined together as at least one ﬂag and Panel B shows the
relationship for each primary ﬂag separately. For both panels, the last four digits of the loan amount
is considered (i.e., $123,456.78 →$3,456.78). In Panel A, the top subpanel shows ﬁntech loans and
the bottom shows traditional loans. Further, the left axis shows the percentage of loans in each $1
wide bin (bars for rounded compensation are thickened) and the right axis shows the percentage of
loans that are ﬂagged within each $1 bins for monthly rounded (solid dots) and $250 wide bins for
non-rounded (hollow dots). In Panel B, loans are ﬁltered to corporation, S-corporation, and LLC
loans for the Business Registry subpanel, loans with industry-CBSA average compensation less than
$33,333.33 for the High Implied Compensation subpanel, and loans with a matched EIDL Advance
for the EIDL > PPP Jobs subpanel. Additionally for both panels, loans with one job reported,
loans with implied compensation within ± $1,000 of $100,000, and second draw loans to hospitality
businesses are excluded. The solid lines are third-degree polynomial ﬁts for the percentage ﬂagged
in the non-rounded bins and the dashed lines are 95% conﬁdence intervals.
Panel A. Percentage Flagged, by Lender Type and Rounding
Panel B. Individual Flags, by Lender Type and Rounding

Figure 7. Overrepresentation of Industries in Counties
This ﬁgure shows overrepresenation of loans withib industry-county pairs. We deﬁne the normalized
number of loans as the number of ﬁrst draw loans divided by the number of establishments (per the
2019 US Census County Business Patterns dataset) in an industry (represented by NAICS [North
American Industry Classiﬁcation System] code) and county pair. Panel A shows the relationship
between normalized number of loans and our four main ﬂags combined together as at least one
ﬂag and Panel B shows the relationship for each ﬂag separately. Since the CBP does not include
self-employed and independent contractors as establishments, we exclude loans to these business
types. Note that 6.20% of ﬁntech and 0.72% of traditional loans are in industry-county pairs with
ratios of at least 10; these loans are represented in Panel A by the bars and dots at the far right
labeled “≥10”. In both panels, loans are binned into 0.25 unit wide bins. The solid lines are third-
degree polynomial ﬁts for the percentage of ﬂagged loans and the dashed lines are 95% conﬁdence
intervals.
Panel A. Percentage Flagged, by Normalized Number of Loans in Industry-County Pair
Panel B. Individual Flags, by Normalized Number of Loans in Industry-County Pair
45

Figure 8. Clustering Within Lenders and Counties
This ﬁgure shows clustering of loans within lenders-county pairs.
We calculate the concentra-
tion ratios of industries, loan amount (rounded to $100), and jobs reported (excluding 1) for ﬁrst
draw loans in each lender-county pair, rescaled each concentration ratio to a median of 1,000 and
IQR (interquartile range) of 300, and then take the average of the three rescaled concentration
ratios. For example, let i = 1, 2, ..., n represent the n industries in a given lender-county pair,
then Concentrationindustry = Pn
i=1 s2
i where si is the percentage of loans in the lender-county
pair that are in industry i times 100 (e.g., 6.2 for 6.2%). Then, Rescaled Concentrationindustry =
Concentrationindustry−Median[Concentrationindustry]
75thPercentile[Concentrationindustry]−25thPercentile[Concentrationindustry] ∗300 + 1000. Panel A shows the rela-
tionship between the average rescaled concentration ratio and our four main ﬂags combined together
as at least one ﬂag and Panel B shows the relationship for each ﬂag separately. In both panels, only
lender-county pairs with at least 25 loans are considered. Note that 2.4% of ﬁntech loans and 0.5%
traditional loans are outside the range of average rescaled concentration ratio shown in Panel A. In
both panels, loans are binned into 50 unit wide bins; in Panel B, bins with fewer than 100 loans for
which the given ﬂag can be determined are excluded. The solid lines are third-degree polynomial
ﬁts for the percentage of ﬂagged loans and the dashed lines are 95% conﬁdence intervals.
Panel A. Percentage Flagged, by Average Rescaled Concentration Ratio in Lender-County Pair
Panel B. Individual Flags, by Average Rescaled Concentration Ratio in Lender-County Pair
46

Figure 9. Criminal Records
This ﬁgure shows criminal records for a sample of 150,000 Round 1 and 2 loans to self-employed
individuals, independent contractors, and sole-proprietors. Panel A shows the percentage of loans
where the borrower has a felony from 2000-2020 on their record by lender type and the presence
of misreporting indicators. The error bars denote 95% conﬁdence intervals (based on standard
errors double clustered by zip code and lender) for each percentage. Panel B shows the relationship
between the percentage of loans in this sample that are ﬂagged by at least one primary ﬂags and the
percentage of borrowers that have a felony from 2000-2020 on their record by lender. Lenders with
at least 0.2% of the sample (300 loans) are shown. The dashed line is a linear ﬁt and correlation
is shown in the bottom left corner. In both panels, red represents non-bank ﬁntech lenders, cream
represents online bank ﬁntech lenders, and grey represents traditional lenders.
Panel A. Percentage Felony, by Lender Type and Presence of Misreporting Indicators
Panel B. Percentage Flagged vs. Felonies
47

Figure 10. Overall Misreporting Flag Rates
This ﬁgure shows the variation in percentage of loans ﬂagged.
Panel A shows the percentage
and dollar amounts of ﬂagged loans overall, by lender type, and by round. Panel B shows the
percentage of ﬂagged loans by lender for the top 75 lenders (by number of loans). In both panels,
the plain section of each bar represents the percentage of loans ﬂagged by one primary ﬂag and
an additional ﬂag (either another primary or a secondary) and the entire bar (plain and stripped
sections combined) represents loans ﬂagged by at least one primary ﬂag. In Panel A, the set of
numbers to the left of each bar represent the number of loans and dollar value of loans ﬂagged one
primary ﬂag and an additional ﬂag and the set on top of each bar by at least one primary ﬂag.
The markers within each bar represent the percentage of loans ﬂagged by each of the primary ﬂags
(unconditional of whether a ﬂag can be determine for a given loan). In Panel B, the two horizontal
lines represent the percentage of loans ﬂagged by each measure across the entire sample (dashed
for loans ﬂagged by one primary ﬂag and an additional ﬂag and solid by at least one primary). In
both panels, red represents non-bank ﬁntech lenders, cream represents online bank ﬁntech lenders,
and grey represents traditional lenders.
Panel A. Percentage of Loans Flagged, by Lender Type and Rounds
Panel B. Percentage of Loans Flagged, by Lender

Figure 11. Geography of Flagged Loans
This ﬁgure shows geographic variation in the percentage of ﬂagged loans.
Panel A shows the
percentage of ﬂagged loans in each county and Panel B shows within county variation. In Panel
A, counties are colored based on the color scheme shown by the bar to the right of the map and
counties with fewer than 100 loans are colored grey. Panel B shows the percentage of ﬂagged loans
in each zip code on the vertical axis and the percentage of ﬂagged loans in the corresponding county
on the horizontal axis. Dots are colored by the percentage of ﬁntech loans in each zip code and
sized based on the number of loans in the zip code. Zip codes with at least 100 loans are shown.
The dashed line is a linear ﬁt and the correlation is shown in at the bottom left corner.
Panel A. Percentage of Flagged Loans, by County
Panel B. Within County Variation
49

Figure 12. Lender Network
This ﬁgure shows connections between lenders. Panel A shows connections between lenders that
were used by borrowers at the same address within the same draw. Panel B shows connections
between lenders used by the ﬂagged ﬁrst draw borrowers across draws. In Panel A, node size is
proportional to the number of loans with the multiple loans ﬂag originated by each lender, edges
are not directed, and edge width is proportional to the number of addresses that used both lenders.
In Panel B, node size is proportional to the number of ﬁrst draw loans (that also got a second
draw loan from the same or diﬀerent lender) and second draws originated by the lender, edges are
directed, edge width is proportional to the number of ﬂagged ﬁrst draw borrowers moving clockwise
from the ﬁrst draw lender to the second draw lender. In both panels, red nodes are ﬁntech lenders
and grey nodes are traditional lenders. Top 100 lenders (by the same measure that node size is
based on) are shown and the remainder are combined into the “Other” nodes (one for other ﬁntech
lenders and one for other traditional lenders).
Panel A. Multiple Lenders at Same Address
Panel B. Changes Between Draws

Figure 13. Persistence and Growth Across Rounds
This ﬁgure shows the persistence and growth of ﬂagged loans across lending rounds. Panel A shows
this by lender type and Panel B by zip code. In Panel A, each subpanel shows a lender type and
each series is the percentage of loans ﬂagged by the given measure across time. In Panels B, the
percentage of loans ﬂagged in rounds 1 and 2 are shown on the horizontal axis and in round 3 on
the vertical axis. For Panel A, the vertical dotted lines split each subpanel into the three lending
rounds. The solid lines are loans ﬂagged by at least one primary ﬂag and the dashed line is loans
ﬂagged by at least one primary ﬂag and an additional ﬂag (either another primary or a secondary).
For Panel B, the left subpanel uses all loans, the middle uses ﬁntech loans, and the right uses
traditional loans. Zip codes with at least 100 loans and, for the ﬁntech and traditional subpanels,
25 loans by the given lender type are shown. The black line is a 45-degree line and the correlation
is presented in the bottom of each panel. The circle size corresponds to the number of loans in the
zip code by the given lender type and color corresponds the growth/decline in lending in the zip
code by the given lender type.
Panel A. By Lender Type
Panel B. By Zip Code
51

Exhibit 1. Examples of Suspicious Loans
This exhibit shows some examples of suspicious loans.
Panel A. 14 Loans to The Same Address, 13 Incorporated Late
Business Name
Date
Date
Jobs
(Redacted)
Incorporated
Approved
Lender
Industry
Loan Amount
Reported
FDML
4/2/2018
5/13/2020
Celtic
Indep. Artists, Writers, Performers
$62,083
10
DREL
7/7/2020
7/8/2020
Kabbage
Musical Groups & Artists
$53,125
10
LTTBTL
7/15/2020
7/15/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
KYHUL
7/15/2020
7/16/2020
Kabbage
Misc. Schools & Instruction
$91,770
10
STWL
7/16/2020
7/17/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
STWL
7/16/2020
7/17/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
ATYL
7/19/2020
7/21/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
BLNL
7/19/2020
7/21/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
CAYL
7/21/2020
7/22/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
CTWIL
7/23/2020
7/23/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
DYJNL
7/22/2020
7/26/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
FML
7/27/2020
7/30/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
EIEL
7/22/2020
7/30/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
GGITL
7/30/2020
8/1/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
JTBCL
8/5/2020
8/5/2020
Kabbage
Misc. Schools & Instruction
$53,229
10
Panel B. Four Loans to Same Household
Individual Name
Date
Jobs
(Redacted)
Age
Approved
Lender
Industry
Loan Amount
Reported
O. P.
21
7/10/2020
Kabbage
Lawn/Garden Equipment Manuf.
$20,833
1
T. P.
49
7/10/2020
Kabbage
Lawn/Garden Equipment Manuf.
$20,833
1
A. P.
20
7/15/2020
Kabbage
Nail Salons
$20,833
1
G. P.
46
7/15/2020
Kabbage
Other Automotive Repair
$20,833
1
Exhibit 2. Cross River Case Study
This exhibit shows examples of $20,000 ﬁrst draw loans by Cross River Bank in Illinois.
Individual Name
Date
Jobs
(Redacted)
Approved
Lender
Industry
Loan Amount
Reported
J. C.
7/29/2020
Cross River
Insurance Agencies and Brokerages
$20,000
1
R. J.
7/29/2020
Cross River
Insurance Agencies and Brokerages
$20,000
1
4,300 $20k Loans by Cross River in Illinois (Mostly in Chicago Area) to Individuals/Businesses
(98% with 1 Employee) in “Insurance Agencies and Brokerages” Industry.
C.C.
7/29/2020
Cross River
All Other Miscellaneous Crop Farming
$20,000
8
M. A.
7/29/2020
Cross River
All Other Miscellaneous Crop Farming
$20,000
8
938 $20k Loans by Cross River in Illinois (Mostly in Chicago Area) to Individuals/Businesses
(56% with 8 Employee and 22% with 1 Employee) in “All Other Miscellaneous Crop Farming” Industry.
K. K.
7/30/2020
Cross River
All Other Miscellaneous Manufacturing
$20,000
9
C.M.
8/7/2020
Cross River
Other General Government Support
$20,000
50
3,068 $20k Loans by Cross River in Illinois (Mostly in Chicago Area) to Individuals/Businesses
in Various Industries (Including 706 to “All Other Personal Services,” 351 to “General Freight Trucking, Local,”
337 to “Other Performing Arts Companies,” and 300 to “New Single-Family Housing Construction.”)
52

Table I. Summary Statistics
This table presents summary statistics for our sample. The sample includes all PPP loans approved
from the start of the program (March 2020) through most of Round 3 (April 2021) that have not
been repaid as of May 3, 2021. Fintech lenders are determined following Erel and Liebersohn (2021).
Loan Amount is the initial approved amount minus any portion used to reﬁnance an EIDL loan.
Implied Comp. is determined following the guidelines in place when the loan was approved and
is based on loan amount and jobs reported. CBSA/NAICS Avg. Comp. is average compensation
in the loan’s industry-CBSA based on the US Census CBP data, CBSA/NAICS Avg. Receipts
is average receipts (for business types that were able to use gross income to calculate loan size)
to nonemployer businesses in the loan’s industry-CBSA based on the US Census NES data, and
Normalized Comp. is the ratio of Implied Comp. and either CBSA/NAICS Avg. Comp. or, if the
business was able to use gross income to calculate their loan amount, the larger of CBSA/NAICS
Avg. Comp. and CBSA/NAICS Avg. Receipts. Loans (Within Draw) at Address is the number
of loans (within the loan’s draw) at the same residential address. Frac. Corp, S Corp, LLC is
the percentage of loans to these business types, Frac. Second Draw is the percentage of Round 3
loans that are the borrower’s second draw from the PPP, and Frac. Matched EIDL Advance is the
percentage of loans with a matching EIDL Advance. Frac. Fintech (Either Type), Frac. Non-bank
Fintech, and Frac. Online Bank Fintech are the percentages of loans that are originated by the
given type of lender.
Fintech
Traditional
Mean
SD
Median
Mean
SD
Median
Num. Loans [Pct. Loans]
3,969,845 [33.73%]
7,798,844 [66.37%]
Loan Amount
25,112
99,717
19,052
90,123
305,553
20,833
Jobs Reported
2.448
9.895
1.000
10.331
29.299
3.000
Implied Comp.
64,217
39,474
70,666
47,206
66,709
38,932
CBSA/NAICS Avg. Comp
46,911
38,966
36,837
49,657
37,302
42,698
CBSA/NAICS Avg. Receipts
41,917
30,714
30,437
48,908
33,596
39,172
Normalized Comp.
1.494
1.452
1.066
1.043
1.903
0.854
Num. Loans (Within Draw) at Address
1.335
0.843
1.000
1.197
0.792
1.000
Frac. Corp, S Corp, LLC
0.185
0.645
Frac. Second Draw (Round 3 Loans)
0.268
0.594
Frac. Matched EIDL Advance
0.178
0.267
Round 1
Round 2
Round 3
Num. Loans [Pct. Loans]
1,619,201 [13.8%]
3,518,979 [29.9%]
6,630,509 [56.3%]
Loan Amount
198,864
57,849
41,773
Jobs Reported
20.468
7.914
4.418
Implied Comp.
48,023
43,838
58,978
CBSA/NAICS Avg. Comp
49,339
51,832
46,795
CBSA/NAICS Avg. Receipts
-
-
43,523
Normalized Comp.
1.178
1.055
1.293
Num. Loans (Within Draw) at Address
1.277
1.215
1.271
Frac. Fintech (Either Type)
0.0479
0.204
0.479
Frac. Non-bank Fintech
0.0249
0.0748
0.333
Frac. Online Bank Fintech
0.230
0.130
0.145
Frac. Corp, S Corp, LLC
0.829
0.657
0.319
Frac. Second Draw
-
-
0.438
Frac. Matched EIDL Advance
0.292
0.301
0.190

Table II. Odds Ratios
In this table, we present the odds ratios between each of our four main indicators. Panel A shows
the odds ratios for ﬁntech and traditional loans combined. Panel B shows the odds ratios for ﬁntech
loans only in the lower triangular and traditional loans only in the upper triangular. Note that
odds ratios are symmetric, which is why only values for the lower triangular are provided. Robust
standard errors are double clustered by zip code and lender.
Panel A. Fintech and Traditional Loans Combined
Business
Multiple
High Implied
EIDL >
Registry
Loans
Comp.
PPP Jobs
Business Registry
-
Multiple Loans
1.904***
-
(7.49)
High Implied Comp.
2.965***
3.241***
-
(9.34)
(16.17)
EIDL > PPP Jobs
1.496***
4.531***
14.746***
-
(5.95)
(19.27)
(19.90)
Panel B. Fintech Loans (Lower Triangular) and Traditional Loans (Upper Triangular)
Business
Multiple
High Implied
EIDL >
Registry
Loans
Comp.
PPP Jobs
Business Registry
-
1.338***
1.898***
1.184***
(8.10)
(12.06)
(6.13)
Multiple Loans
2.351***
-
1.655***
2.136***
(17.55)
(4.40)
(11.59)
High Implied Comp.
3.655***
2.085***
-
5.830***
(13.78)
(6.28)
(22.53)
EIDL > PPP Jobs
2.177***
3.093***
16.007***
-
(6.03)
(11.07)
(38.45)
z-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
54

Table III. Prevalence of Flags by Lender Type
This table presents the percentage of loans ﬂagged by the four main ﬂags, at least one of four ﬂags,
and at least two of the four ﬂags. In Panel A, column (1) shows the percentage of ﬁntech loans with
the given ﬂag, column (2) shows the percentage of traditional loans with the given ﬂag, and column
(3) shows the diﬀerence between the ﬁntech and traditional percentages, column (4) shows the
adjusted diﬀerences with zip code, business type, and NAICS × CBSA ﬁxed eﬀects and controlling
for jobs and loan size, and column (5) shows the diﬀerences between matched pairs of ﬁntech and
traditional loans. The N values show is the number of loans for which the ﬂag can be determined
and robust standard errors are double clustered by zip code and lender. For the matched diﬀerences,
robust standard errors are four-way clustered by the zip code and lender of both matched loans.
The full regression results for the unadjusted diﬀerences and adjusted diﬀerences are reported in
Panel A and Panel B, respectively, of Table IA.II.
(1)
(2)
(3)
(4)
(5)
Fintech
Traditional
Unadjusted
Adjusted
Matched
Diﬀerence
Diﬀerence
Diﬀerence
Business
0.103
0.0434
0.0587***
0.0350***
0.0268***
Registry
N = 739,657
N = 4,825,093
(3.09)
(3.55)
(3.12)
Multiple
0.0344
0.0100
0.0244***
0.0106***
0.0170***
Loans
N = 3,969,845
N = 7,798,844
(8.71)
(5.19)
(3.76)
High Implied
0.489
0.0942
0.395***
0.0935***
0.0818***
Comp.
N = 1,600,474
N = 2,133,022
(7.83)
(7.34)
(3.91)
EIDL > PPP
0.216
0.0487
0.168***
0.0651***
0.0794***
Jobs
N = 705,837
N = 2,083,055
(4.34)
(4.13)
(4.62)
At Least One
0.236
0.0743
0.161***
0.0560***
0.0535***
Flag
N = 3,969,845
N = 7,798,844
(8.12)
(8.81)
(6.71)
At Least Two
0.0246
0.00362
0.0210***
0.00686***
0.0107***
Flags
N = 3,969,845
N = 7,798,844
(7.39)
(7.00)
(5.99)
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
55

Table IV. Secondary Flags
In this table, we examine the relationship between our four main ﬂags, which we combine to form
At Least One Flag, and the secondary ﬂags. We estimate OLS regressions with At Least One Flag
as the dependent variable and the ﬁve secondary ﬂags as independent variables. Each speciﬁcation
also include an interaction between the secondary ﬂag and an indicator for whether the loan was
originated by a ﬁntech lender. $100k Implied Comp./Receipts is a dummy variable equal 1 if the
implied compensation/receipts per job is within ± $1,000 of $100,000.
Monthly Rounding is a
dummy variable equal 1 if the loan amount is within ± 50 cents of an interval of $1,250. Overrep.
in County/NAICS is a dummy variable equal 1 if the number of ﬁrst draw loans to businesses not
listed as self-employed and independent contractors in a loan’s industry-county pair exceeds the
number of establishments in the industry-county pair according to the US Census CBP data. High
Concentration is a dummy variable equal 1 if the average rescaled concentration ratio in the loan’s
lender-county pair is above the 75th percentile. Felony Post-2000 is a dummy variable equal 1 if the
borrower has a felony on their criminal record from 2000 or after. 1(Fintech) is a dummy variable
equal 1 if the loan was originated by a ﬁntech lender. For all speciﬁcations, loans are ﬁltered to the
sets for which we can determine the secondary ﬂag. Further, for speciﬁcation (2), one job loans and
loans where 1($100k Implied Comp.) = 1 are excluded. Fixed eﬀects are as indicated at bottom of
each column. Robust standard errors are double clustered by zip code and lender.
Dep. Variable: At Least One Primary Flag
(1)
(2)
(3)
(4)
(5)
$100k Implied Comp.
0.0238***
(2.65)
Monthly Rounding
0.00426***
(5.84)
Overrep. in County/NAICS
-0.00739**
(-2.19)
High Concentration
0.0141***
(4.23)
Felony Post-2000
0.0265***
(2.71)
× 1(Fintech)
0.113***
0.0118**
0.0604***
0.0159***
0.0374***
(14.30)
(2.40)
(8.87)
(2.77)
(3.01)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Yes
Zip Code FE
Yes
Yes
Yes
Yes
No
Business Type FE
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Yes
Lender FE
Yes
Yes
Yes
Yes
Yes
Observations
11,085,989
5,120,313
6,215,638
7,615,688
123,670
Num. Lenders
4,849
4,768
4,798
4,221
2,578
R2
0.324
0.103
0.347
0.346
0.383
Mean of Dep. Variable
0.134
0.0671
0.145
0.143
0.106
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010

Table V. County Cultural Features
In this table, we examine the relationship between our four main ﬂags, which we combine to form
At Least One Flag, and cultural/regional features. We estimate OLS regressions with At Least One
Flag as the dependent variable and the cultural/regional features as independent variables. All
independent variables are rescaled at the county level to have a mean of 0 and a standard deviation
of 1. Political Corruption is the number of public corruption convictions in 2004-2013 per million
residents (as reported by the DOJ). Religious Adherence is the percent of the county’s population
with a religious aﬃliation as of 2010 (as reported by the Association of Religious Data Archives).
Ashley Madison Usage is the paid Ashley Madison usage rate in the county (as reported by Griﬃn
et al. (2019)). Population Density is the population per square mile as of 2019, Median Income
is the median household income as of 2019, Pct. Non-White is the percentage of the population
that is non-white as of 2019, College Educated is the percentage adults with a bachelor’s degree or
higher as of 2015-19, and 2019 Unemployment is the unemployment rate as of 2019 (all from the
Economic Research Service of the U.S. Department of Agriculture). Pct. Fintech is the percentage
of PPP loans in the county originated by a ﬁntech lender. Fixed eﬀects are as indicated at bottom
of each column. Robust standard errors are double clustered by zip code and lender.
Dep. Variable: At Least One Primary Flag
(1)
(2)
(3)
Public Corruption
0.0123***
0.00863***
0.00447***
(7.08)
(5.78)
(3.82)
Religious Aﬃliation
0.00122
-0.00136*
-0.0000799
(1.54)
(-1.66)
(-0.10))
Ashley Madison Usage
-0.00304***
0.00747***
0.000160
(-5.61)
(4.79)
(0.13)
Population Density
-0.00112***
-0.000861***
(-7.01)
(-4.40)
Median Income
-0.00704***
-0.00383***
(-7.27)
(-4.19)
Pct. Non-White
0.0193***
-0.00316**
(13.57)
(-2.27)
College Educated
-0.00446***
0.000720
(-3.52)
(0.84)
2019 Unemployment
0.00441***
0.00522***
(3.57)
(4.53)
Pct. Fintech
0.0269***
(9.94)
ln(Jobs Reported)
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
NAICS × State FE
Yes
Yes
Yes
Observations
11,651,511
11,651,511
11,651,511
Num. Lenders
4,874
4,874
4,874
R2
0.248
0.250
0.252
Mean of Dep. Variable
0.129
0.129
0.129
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010

Table VI. Persistence of Lender Behavior Across Rounds
In this table, we examine the persistence of lender behavior across rounds. We estimate OLS regres-
sions with dummies for whether each Round 3 loan is ﬂagged by our four main ﬂags individually
(speciﬁcations (1) through (4)) and at least one of them (speciﬁcation (5)) as the dependent vari-
ables and the percentage of the lender’s loans were ﬂagged by the same ﬂag in rounds 1 and 2 as the
independent variable. Interactions with whether the loam was originated by a ﬁntech or traditional
lender are include in all speciﬁcation. For speciﬁcations (1) through (4), loans are ﬁltered to the
sets for which we can determine the ﬂag (same as in Figures 2- 4). Further, to ensure we have
accurate measures of past behavior, we require that each lender have at least 100 loans in Round
1 and 2 (combined) for which we can determine the given ﬂag. Fixed eﬀects are as indicated at
bottom of each column. Robust standard errors are double clustered by zip code and lender.
(1)
(2)
(3)
(4)
(5)
Dep. Variable:
Business
Multiple
High Implied
EIDL >
At Least
Registry
Loans
Comp.
PPP Jobs
One Flag
Past Pct. This Flag
× 1(Fintech)
0.930***
0.640***
0.417***
0.872***
0.516***
(4.78)
(3.45)
(3.00)
(2.94)
(5.79)
× 1(Traditional)
0.201
0.225*
-0.303**
-0.125
-0.00962
(0.92)
(1.94)
(-2.23)
(-0.91)
(-0.09)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Yes
Zip Code FE
Yes
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Yes
Observations
1,853,605
5,503,222
1,649,381
1,029,259
5,503,222
Num. Lenders
2,475
3,133
1,574
1,439
3,133
R2
0.142
0.052
0.638
0.350
0.377
Mean of Dep. Variable
0.0720
0.0220
0.351
0.139
0.171
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
58

Table VII. Likelihood of Receiving a Second Draw Loan
In this table, we examine whether lenders were more/less likely to provide a second draw loan to
a borrower who’s ﬁrst draw loan is ﬂagged by at least one of our primary ﬂags. We estimate OLS
regressions with a dummy for whether the same lender provided the ﬁrst and second draw loans as
the dependent variable and a dummy for whether the ﬁrst draw loan was ﬂagged by at least one
of the primary ﬂags as the independent variable. In speciﬁcations (1) and (2), if a borrower did
not receive a second draw loan, the dependent variable is set to 0, and in speciﬁcations (3) and
(4), only borrowers that received both a ﬁrst and second draw loans are included in the sample. In
the even speciﬁcation, an interaction with whether the ﬁrst draw loan was originated by a ﬁntech
lender is included. Fixed eﬀects are as indicated at bottom of each column. Robust standard errors
are double clustered by zip code and lender.
Dep. Variable: 1(First and Second Draw by Same Lender)
(1)
(2)
(3)
(4)
Unconditional of Receiving
Conditional on Receiving
Second Draw
Second Draw
First Draw Flagged
-0.0260***
-0.0310***
-0.00972**
-0.0161***
(-6.64)
(-11.94)
(-2.06)
(-5.00)
× 1(Fintech)
0.0212*
0.0254
(1.66)
(1.08)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Lender FE
Yes
Yes
Yes
Yes
Observations
4,849,427
4,849,427
1,584,875
1,584,875
Num. Lenders
4,727
4,727
4,518
4,518
R2
0.122
0.122
0.420
0.420
Mean of Dep. Var.
0.278
0.278
0.834
0.834
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
59

Internet Appendix
Classifying Lenders as FinTechs
We use Erel and Liebersohn’s (2021) classiﬁcations for lenders that were active in rounds 1
and 2 (the sample period for Erel and Liebersohn (2021)), and we use the same methodology for
classifying round 3 lenders that were not active enough to be classiﬁed in the earlier rounds. The
method used by Erel and Liebersohn (2021) is summarized in their paper as, “We match this loan-
level data to bank identiﬁers from the Federal Financial Institutions Examination Council (FFIEC)
using the lender names provided. Most of the names are matched using automated name matching.
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... 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... Online banks, however, are regulated deposit-taking
banks but with only one administrative branch.”
Loan Size Calculation
Average monthly payroll expenses are to be based on the full 2019 calendar year for most loans.
Exceptions include seasonal businesses which may use average monthly payroll for any twelve-week
period between February 15, 2019 and February 14, 2020, and new businesses may use average
monthly payroll over the period from January 1, 2020 to February 29, 2020. Additionally, second
draw loans by businesses in the hospitality industry (NAICS starting with 72, representing 2% of
all loans) were allowed to receive 3.5 times their average monthly payroll. We account for this
when computing implied compensation, and we drop second draw loans to the hospitality industry
from the rounding analysis because their diﬀerent compensation multiplier could lead to diﬀerent
loan rounding. In determining the high implied compensation ﬂag and for the $100k discontinuity
and rounding analyses, we subtract any portion of the loan amount used to reﬁnance an EIDL. See
SBA guidance entitled “How to Calculate First Draw PPP Loan Amounts,” which was updated
over time. A list of all versions can be found [here]https://www.sba.gov/document/support-how-
calculate-ﬁrst-draw-ppp-loan-amounts.
Matching Analysis
The matching analysis reported in Table III is based on a combination of propensity score
matching and exact matching. First, we estimate a propensity score for whether a loan is originated
by a FinTech lender using a logistic regression and the following variables: 4-digit NAICS code
(industry), CBSA, business type, loan amount, jobs reported, lending draw, lending round, week
of loan approval, and whether the borrower received an EIDL Advance. Second, for each loan
originated by a FinTech lender, we identify loans made by traditional lenders such that the loans
were made to borrowers in the same industry, CBSA, business type, and year (either 2020 or 2021)
and either both received EIDL Advances or did not. Exactly matching on these characteristics
ensures that we can determine each of our ﬂags for both loans. Finally, among the loans that
1

match exactly on these features, we match the loans that have the smallest absolute diﬀerence
in propensity scores. In total, 3,614,103 of 3,969,845 FinTech loans are matched. Note that a
traditional loan may be matched to more than one FinTech loan.
Non-proﬁt Analysis
For this analysis, we compare the PPP loan amounts that non-proﬁts received to estimated
loan amounts that they are eligible for based on compensation reported in their latest Form 990
(most commonly from 2019). The main ﬁelds of interest from the Form 990 are lines 5-10 of Part
IX (“Statement of Functional Expenses”) with the sum of these lines being our measure of the
non-proﬁt’s total compensation. Note that this measure of total compensation is likely higher than
eligible compensation for the PPP because we do not cap compensation per employee at $100,000,
and the Form 990 compensation includes federal payroll taxes and employee beneﬁt costs that were
not covered by the PPP. We then use this total compensation measure to calculate the implied loan
amount that the non-proﬁt should be eligible following the SBA’s guidelines for determining the
maximum loan size (as outlined the “Loan Size Calculation” section above). Loan amount inﬂation
is deﬁned as Loan Amount−Form 990 Implied Loan Amount
Loan Amount
.
Repayment, Enforcement Action, and Canceled Loans Data
To examine round 1 and 2 loans that have been repaid, we use PPP loan-level data released
by the SBA on December 2, 2020 and data from USASpending.gov as of June 30, 2021.
The
USASpending data provide monthly updates on the PPP loans, which allows us to observe which
loans have been repaid by the borrower based on a loan balance being zeroed out without an
associated U.S. Treasury account to fund forgiveness for the loan. It is also possible that some
of these loans may have been canceled. The USASpending data do not include all of the loan
details included in the SBA data, and after being repaid or canceled, the loans are removed from
the SBA loan-level PPP data. Fortunately, the December 2020 version of the PPP loan-level data
provides the same details as the main PPP data and covers all loans that had not been repaid as
of December 2020. Many borrowers requested loans from multiple lenders to increase the odds of
receiving funds; we exclude such loans by ﬁltering out businesses with multiple loan requests where
all but one is repaid.
To examine enforcement actions by the government, we collect information from Department
of Justice criminal complaints against PPP borrowers that purportedly committed fraud based on
https://www.justice.gov/criminal-fraud/cares-act-fraud and https://www.arnoldporter.
com/en/general/cares-act-fraud-tracker. In total, we collect data on 162 complaints involv-
ing 355 loans. Of these 355 loans, 279 include enough information to be matched to the December
2020 version of the PPP loan-level data. Most of the unmatched loans were repaid before December
2020 and thus are not in the loan-level data. To examine canceled loans, we use loan-level data
released by the SBA on May 3, 2021. This data includes all PPP loans that were approved as of
May 3, 2021. We compare this data to our main loan-level data that was released on June 30, 2021
to determine which loans were canceled between May 3 and June 30, 2021.
2

Figure IA.1. Fintech Market Share
This ﬁgure shows the role of ﬁntech lenders during the PPP (expanding on Figure 1). Panel A
replicates Figure 1, Panel A based on loans from only rounds 1 and 2. Panel B shows the number
of loans originated by lender type during each week of the PPP. In both panels, red represents non-
bank ﬁntech lenders, cream represents online bank ﬁntech lenders, and grey represents traditional
lenders. Note that mid-August through December 2020 is not shown in Panel B since no PPP loans
were originated during this period.
Panel A. Rounds 1 and 2 Lenders (Top 75)
Panel B. Lender Composition, by Week
3

Figure IA.2. Florida Restaurant License Analysis
This ﬁgure shows the percentage of loans to Florida restaurants (deﬁned as NAICS code starting
with 722) structured as corporations, S-corporations, and LLC that can be matched the Florida
Department of Business and Professional Regulation’s (DBPR) restaurant/hotel license data
(available at http://www.myfloridalicense.com/DBPR/hotels-restaurants/public-records/
#1506342906681-c46dd821-30bf). Panel A shows the percentage of loans with potential matches
for three diﬀerent subsample of loans to Florida restaurants: those that are not ﬂagged by any
of our primary or secondary ﬂags, those ﬂagged as missing from the business registry, and those
ﬂagged as missing and also ﬂagged by an additional primary or secondary ﬂag. In this panel, a loan
is considered to have a potential match if there exists a license with a licensee or location name that
has at least a similarity ratio of 90% with the borrower name listed on the PPP loan and the loan
and license listed the same location (zip, city, or county). The error bars show the 95% conﬁdence
intervals. In Panel B, we show the sensitivity of the match rate for the subsample ﬂagged as being
missing from the business registry. The horizontal axis shows the similarity ratio threshold used to
determine if the loan and license matches. The blue area is the percentage of loans with a potential
match if the loan and license are required to list the same location (zip, city, or county) and the
light brown region shows the additional matches made if this location requirement is relaxed.
Panel A. Percentage with Potential Matches
Panel B. Sensitivity of Match Rate

Figure IA.3. Multiple Loans
This ﬁgure expands on Figure 2, Panel B. Panel A is based on loans at residential addresses and
Panel B decreases the number of loans at the same address in the same draw necessary for the loan
to be ﬂagged from three to two.
Panel A. Restricted to Loans Listing Residential Addresses
Panel B. More Than 2 Loans at Same Address in Given Loan Draw
5

Figure IA.4. Implied Compensation
This ﬁgure compares the implied compensation of loans to the average compensation in its industry
(NAICS) and CBSA. In Panel A, loans are split by round (rounds 1 and 2 in the left subpanel
and round 3 in the right subpanel) and lender type.
In Panel B, we focus on Round 3 loans
to sole proprietorship, self-employed, independent contractors, and single member LLCs and split
into loans approved before March 3, 20201 in the left subpanel and on or after March 3, 2021 in
the right subpanel. In both Panels, loans are sorted and binned based on their industry-CBSA
average compensation/receipts. The median annualized compensation and industry-CBSA average
compensation/receipts of each bin is shown. The solid line is a 45-degree line.
Panel A. By Round and Lender Type
Panel B. Sole Proprietorship, Self-Employed, Independent Contractors, and Single Member LLC.
Round 3, Before and After March 3, 2021
6

Figure IA.5. Relation Between Flags
This ﬁgure shows the relationship between the primary ﬂags (business registry, multiple loans, high implied compensation, or EIDL >
PPP jobs ﬂags) at the lender level. Each subpanel is a scatterplot with the percentage of loans ﬂag by one of the ﬂags on each axis.
Loans are ﬁltered to the sets for which we can determine each ﬂag (same as in Figures 2-4) for each axis separately (i.e., we do not require
that both ﬂags be able to be determined for a given loan). Lenders with at least 5,000 loans are shown; for the subpanels with the EIDL
> PPP jobs ﬂag, we additionally require that the lender have at least 1,000 loans with a matched EIDL Advance. The dashed line is
a linear ﬁt and the correlation is shown in the bottom left corner of each subpanel. Red triangles represent non-bank ﬁntech lenders,
cream squares represent online bank ﬁntech lenders, and grey circles represent traditional lenders.
7

Figure IA.6. Loan Size Inﬂation by Non-Proﬁts
This ﬁgure shows the relationship between non-proﬁts inﬂating their loan size and being ﬂagged
by our primary measures. Loan amount inﬂation is deﬁned Loan Amount−Form 990 Implied Loan Amount
Form 990 Implied Loan Amount
where Form 990 Implied Loan Amount is calculated based on the non-proﬁt’s 2019 Form 990 sub-
mission to the IRS. The left axis shows the percentage of loans with at least the given percentage
of loan amount inﬂation that are ﬂagged by our primary measures.
The right axis shows the
complementary cumulative distribution (i.e., the percentage of loans that have at least the given
percentage of loan amount inﬂation). This analysis is based on 119,397 loans.
8

Figure IA.7. Discontinuities Around $150,000 Loan Amount
This ﬁgure shows the percentage of loans that are ﬂagged by loan amount. Panel A shows this
across all loan amounts upto $400,000, while Panel B focused on around $150,000. In Panel A, loans
are binned into $1,000 wide bins. The left axis shows the percentage of ﬂagged loans in each bin and
the right axis shows the number of loans in each bin (on a log scale). In Panel B, loans are binned
into $250 wide bins, solid lines are linear ﬁts (separately for bins below and above $150,000), and
the dashed lines are 95% conﬁdence intervals. Fintech and traditional loans are shown separately.
In both panels. the purple vertical line denotes the $150,000 loan amount maximum threshold for
streamline loan forgiveness
Panel A. Overall
Panel B. Around $150,000
9

Figure IA.8. Criminal Records
This ﬁgure shows additional features (expanding on Figure 9) of our sample of criminal records for
150,000 Round 1 and 2 loans. Panel A replicates Figure 9, Panel A using felonies from 2015-2020,
Panel B replicates Figure 9, Panel B using various time cutoﬀs, Panel C replicates Figure 9, Panel
B using bankruptcies (2015-2020), and Panel D shows the percentage of felonies (2000-2020) by
lender type and across implied compensation. In Panels B and C, lenders with at least 0.2% of the
loans in the sample (300 loans) are shown. The dashed lines are linear ﬁts and correlations are in
the bottom corner. In Panel D, loans are binned into $4,000 wide bins, solid lines are third-degree
polynomial ﬁts, and the dashed lines are 95% conﬁdence intervals.
Panel A. Percentage Felony, by Lender Type and Various Features (2015 or After)
Panel B. Percentage Flagged vs. Felony, Diﬀerent Time Cutoﬀs
10

Panel C. Percentage Flagged vs. Bankruptcies
Panel D. Percentage with Felony, by Implied Compensation
11

Figure IA.9. Relationship Between Primary and Secondary Flags
This ﬁgure shows the relationship between the primary ﬂags (summarized as whether the loan
is ﬂagged by at least one of them) and the secondary ﬂags by lender. The corresponding ﬁgure
for criminal records is shown as Figure 9, Panel B. For monthly rounding, loans with one job
reported are exclude; for overrepresentation, loans to self-employed and independent contractors,
second draw loans, and loans in a industry-county pair not in the CBP data are excluded; for high
concentration, second draw loans and loans in a lender-county pair with fewer than 25 ﬁrst draw
loans are excluded. No loans are excluded for percentage ﬂagged by at least one primary ﬂag.
Lenders with at least 5,000 loans are shown. The dashed line is a linear ﬁt and the correlation is in
the bottom corner of each panel. Red triangles represent non-bank ﬁntech lenders, cream squares
represent online bank ﬁntech lenders, and grey circles represent traditional lenders.
12

Figure IA.10. Overall Misreporting Flag Rates – Additional Details
This ﬁgure shows additional variation in percentage of loans ﬂagged. Panel A replicates Figure 10,
Panel A and adds loans ﬂagged by at least one primary or secondary ﬂag as the dashed, non-shaded
portions of the bar. In Panel B, each subpanel shows a lender type and each series is the percentage
of loans ﬂagged by the given ﬂag across time. The vertical dotted lines split each subpanel into the
three PPP lending rounds. The loans used to calculate each series are ﬁltered to the sets for which
we can determine each ﬂag (same as in Figures 2-4). Note that mid-August through December
2020 is not shown in Panel B since no PPP loans were originated during this period. In both
panels, red represents non-bank ﬁntech lenders, cream represents online bank ﬁntech lenders, and
grey represents traditional lenders.
Panel A. Percentage of Loans Flagged with Broader Flags, by Lender
Panel B. Percentage of Loans Flagged Over Time, by Lender Type
13

Figure IA.11. Geography
This ﬁgure shows additional geographic variation (extending Figure 11). Panel A shows the ﬁntech
market share in each county. Panel B shows the growth in lending between rounds 1-2 and round
3 in each county. In both panels, counties are colored based on the color scheme shown in the bar
to the right of the maps and counties with fewer than 100 loans are colored grey.
Panel A. Fintech Market Share, by County
Panel B. Lending Growth, by County
14

Figure IA.12. Pairwise Lender Correlations
This ﬁgure shows pairwise lender correlations using data at the county-lender level. The lower
triangular shows correlations between the percentage of loans ﬂagged by at least one primary ﬂag
and the upper triangular shows correlations between the lenders’ market shares across counties.
Lenders are order such that those with the highest percentage of loans ﬂagged by at least one
primary ﬂag (across the entire sample) are at the top on the vertical axis and the left on the
horizontal axis. Labels are colored red for non-bank ﬁntechs, orange for online bank ﬁntechs, and
black for traditional lenders. Coloring of each square in the matrix is based pairwise correlation and
the coloring scheme is shown at the bottom with darker red representing higher positive correlation
and darker blue representing higher negative correlation. For each pairwise correlation, counties
are ﬁltered to the set that have 25 loans by both lenders. The top 75 lenders (by number of loans)
are shown.

Figure IA.13. PPP Liquidity Facility
This ﬁgure shows the the total amount of PPP Liquidity Facility (PPPLF) advance received by
each of the top 100 PPP lenders (by number of loans). The horizontal axis shows the total original
outstanding advance amount for each lender and the vertical axis shows the percentage of PPP
loans ﬂagged. Lenders that did not receive a PPPLF advance are included with their PPPLF total
advance amount set to zero. The circle size represents the total dollars lent by each lender. The
dotted line is a line of best ﬁt and the correlation is shown in the bottom left corner. Red repre-
sents non-bank ﬁntech lenders, cream represents online bank ﬁntech lenders, and grey represents
traditional lenders. Lenders that received at least $500 million in PPPLF advances are labeled.
16

Figure IA.14. Lender Network
This ﬁgure replicates Panel B of Figure 12 using all loans where the borrower received a ﬁrst and
second draw loan. Node size is proportional to the number of ﬁrst draw loans (which also received
either a second draw from the same or diﬀerent lender) and second draw originated by the lender.
Edges are directed and have a width proportional to the number of loans moving clockwise from
the ﬁrst draw lender to the second draw lender. Red nodes are ﬁntech lenders and grey nodes are
traditional lenders. Pure red edges are between two ﬁntech lenders, pure grey edges are between
two traditional lenders, and darker red edges are between a ﬁntech and traditional lender. Top 100
lenders (by the same measure used for node size) are shown and the remainder are combined into
the “Other” nodes (one for other ﬁntech lenders and one for other traditional lenders).
17

Figure IA.15. Likelihood of Receiving a Second Draw Loan
This ﬁgure examines whether lenders were more/less likely to provide a second draw loan to a
borrower who’s ﬁrst draw loan is ﬂagged by at least one of our primary ﬂags. Similar to Table VII,
we estimate a OLS regression with a dummy for whether the same lender provided the ﬁrst and
second draw loans as the dependent variable and a dummy for whether the ﬁrst draw loan was
ﬂagged by at least one of the primary ﬂags interacted with an indicator for each lender as the
independent variables. The regression control for loan size and jobs and include zip code, business
type, and NAICS × CBSA ﬁxed eﬀects.
For Panel A, if a borrower did not receive a second
draw loan, the dependent variable is set to 0, and in Panel B, only borrowers that received both a
ﬁrst and second draw loans are included in the sample. In both panels, the hollow dots show the
point estimates from the regression and the error bars show 95% conﬁdence intervals corrected for
multiple comparisons using a Bonferroni correction. Lenders with at least 10,000 loans in rounds
1-2 are shown and are sorted such that those with the highest percentage of ﬂagged loans (in rounds
1-2) are on the left. Red error bars and labels represents non-bank ﬁntech lenders, cream represents
online bank ﬁntech lenders, and grey represents traditional lenders.
Panel A. Unconditional of Receiving Second Draw
Panel B. Conditional on Receiving Second Draw
18

Figure IA.16. Persistence and Growth Across Rounds
This ﬁgure shows the persistence and growth of ﬂagged loans across lending rounds. Panel A shows
this by lender using all loans and Panel B by lender using ﬁrst draw loans only. Panel C and D
shows this by county and state (by zip code is shown as Figure 13, Panel B). For all panels, the
percentage of loans ﬂagged in rounds 1 and 2 are shown on the horizontal axis and round 3 on the
vertical axis. For Panel A, lenders with at least 1,000 loans in rounds 1 and 2 combined and in
round 3 are shown. For Panel B, lenders with at least 1,000 loans in rounds 1 and 2 combined and
250 ﬁrst draw loans in round 3 are shown. In Panel C, counties with at least 100 loans in round
1 and 2 combined and in round 3 are shown. In all panels, the circle size corresponds to their
total number of loans (across all rounds), the black line is a 45-degree line, and the correlation is
presented in the bottom of each panel. For Panels A and B, the percentage of the circle that is
shaded represents the proportion of loans that each lender provided in round 3 relative to in round
1 and 2. For Panel C and D, the circles are colored based lending growth in the county/state with
the color scheme shown in the bar to the right of each panel.
Panel A. By Lender
Panel B. By Lender, First Draw Loans Only
19

Panel C. By County
Panel D. By State
20

Figure IA.17. Lender Level, Excluding Undisbursed Loans
This ﬁgure replicates Figure 10 after excluding undisbursed loans. Undisbursed loans are those
listed as “Active Un-Disbursed” in the loan-level data.
Panel A. Percentage of Loans Flagged, by Lender Type and Rounds
Panel B. Percentage of Loans Flagged, by Lender
21

Figure IA.18. Repayment, Enforcement Actions, and Canceled Loans
This ﬁgure shows the percentage of loans that have been repaid between December 1, 2020 and
June 30, 2021, part of a DOJ enforcement action, and canceled between May 3, 2021 and June
30, 2021. Loans are ﬁltered based on the criteria at the bottom of bar. In total, 16,930 loans
that were repaid between December 1, 2020 and June 30, 2021, 279 loans that are part of DOJ
enforcement actions, and 95,526 loans have been canceled between May 3, 2021 and June 30, 2021.
The percentage of repayment and enforcement actions is based on the loan-level data released by
the SBA on December 1, 2020, and the percentage of canceled loans is based on the loan-level data
released by the SBA on May 3, 2021 and .
22

Table IA.I. Cross Veriﬁcation of Flags
In this table, we examine the relationships between our four main ﬂags. We estimate OLS regres-
sions with dummies for each of our main ﬂags as dependent variables and dummies for the other
three ﬂags as independent variables. Panel A shows the relationships without lender ﬁxed eﬀects
and Panel B shows the relationship within lenders by adding lender ﬁxed eﬀects. In both panels,
speciﬁcation (1) uses business registry as the dependent variable, (2) uses multiple loans, (3) uses
high implied compensation, and (4) uses EIDL > PPP jobs. Loans are ﬁltered to the sets for which
we can determine the ﬂag being used as the dependent variable (same as in Figures!2-4). Fixed
eﬀects and control variables are as indicated at bottom of each column. Robust standard errors
are double clustered by zip code and lender.
Panel A. Without Lender Fixed Eﬀect
(1)
(2)
(3)
(4)
Dep. Variable:
Business
Multiple
High Implied
EIDL > PPP
Registry
Loans
Comp.
Jobs
Business Registry
0.00396***
0.0259***
0.00327**
(5.79)
(8.53)
(2.19)
Multiple Loans
0.0226***
0.0365***
0.0616***
(7.57)
(6.09)
(13.39)
High Implied Comp.
0.0415***
0.00724***
0.194***
(7.21)
(6.28)
(28.62)
EIDL > PPP Jobs
-0.00698***
0.0122***
0.0654***
(-4.37)
(14.40)
(13.42)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip Code FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
1(EIDL Adv. Matched)
Yes
Yes
Yes
No
Lender FE
No
No
No
No
Observations
5,305,966
11,086,014
3,410,487
2,653,048
Num. Lenders
4,722
4,873
4,765
4,732
R2
0.100
0.047
0.592
0.282
Mean of Dep. Variable
0.0513
0.0182
0.252
0.0928
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
23

Panel B. With Lender Fixed Eﬀect
(1)
(2)
(3)
(4)
Dep. Variable:
Business
Multiple
High Implied
EIDL > PPP
Registry
Loans
Comp.
Jobs
Business Registry
0.00320***
0.0131***
-0.00158
(5.96)
(5.09)
(-0.79)
Multiple Loans
0.0140***
0.0293***
0.0462***
(6.96)
(7.45)
(11.08)
High Implied Comp.
0.0294***
0.00579***
0.164***
(9.25)
(7.73)
(23.30)
EIDL > PPP Jobs
-0.00706***
0.0108***
0.0580***
(-4.57)
(12.56)
(12.55)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip Code FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
1(EIDL Adv. Matched)
Yes
Yes
Yes
No
Lender FE
Yes
Yes
Yes
Yes
Observations
5,305,887
11,085,989
3,410,400
2,652,927
Num. Lenders
4,648
4,849
4,679
4,628
R2
0.116
0.051
0.608
0.311
Mean of Dep. Variable
0.0513
0.0182
0.252
0.0928
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
24

Table IA.II. Prevalence of Flags by Lender Types
Panel A shows the full results for the unadjusted and adjusted diﬀerences presented in Table III.
Fixed eﬀects and control variables are as indicated at bottom of each column. Robust standard
errors are double clustered by zip code and lender.
Panel A. Unadjusted Percentages
(1)
(2)
(3)
(4)
(5)
(6)
Dep. Variable:
Business
Multiple
High
EIDL >
At Least
At Least
Registry
Loans
Comp.
PPP Jobs
One
Two
Fintech
0.0587***
0.0244***
0.395***
0.168***
0.161***
0.0210***
(3.09)
(8.71)
(7.83)
(4.34)
(8.12)
(7.39)
Observations
5,564,750
11,768,676
3,416,620
2,788,892
11,768,676
11,768,676
Num. Lenders
4,760
4,890
4,765
4,830
4,890
4,890
R2
0.008
0.007
0.198
0.064
0.052
0.009
Mean of Dep. Var.
0.0512
0.0183
0.252
0.0911
0.129
0.0107
Panel B. Adjusted Percentages
(1)
(2)
(3)
(4)
(5)
(6)
Dep. Variable:
Business
Multiple
High
EIDL >
At Least
At Least
Registry
Loans
Comp.
PPP Jobs
One
Two
Fintech
0.0350***
0.0106***
0.0935***
0.0651***
0.0560***
0.00686***
(3.55)
(5.19)
(7.34)
(4.13)
(8.81)
(7.00)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Yes
Yes
Observations
4,305,966
11,086,014
3,410,487
2,653,048
11,086,014
11,086,015
Num. Lenders
4,722
4,873
4,765
4,732
4,873
4,783
R2
0.097
0.048
0.597
0.276
0.310
0.081
Mean of Dep. Var.
0.0513
0.0182
0.252
0.0928
0.134
0.0113
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
25

Table IA.III. Prevalence of Flags by Lender Type, with Address Fixed Eﬀect
In this table, we examine the prevalence of our primary ﬂags by lender type while including an
address × draw ﬁxed eﬀect. Both Panels include address × draw ﬁxed eﬀects. Panel B also includes
other controls and ﬁxed eﬀects as indicated at bottom of each column. Robust standard errors are
triple clustered by zip code, lender, and address × draw.
Panel A. Only Address Fixed Eﬀect
(1)
(2)
(3)
(4)
Dep. Variable:
Business
High Implied
EIDL >
At Least One
Registry
Comp.
PPP Jobs
of Three
Fintech
0.0277***
0.151***
0.0400***
0.0387***
(3.18)
(7.34)
(4.76)
(6.02)
Address × Draw FE
Yes
Yes
Yes
Yes
Observations
135,383
167,961
73,374
934,794
Num. Lenders
3,555
2,832
2,780
4,564
R2
0.572
0.801
0.837
0.632
Mean of Dep. Var.
0.0676
0.494
0.206
0.187
Panel B. Other Fixed Eﬀects and Controls Added
(1)
(2)
(3)
(4)
Dep. Variable:
Business
High Implied
EIDL >
At Least One
Registry
Comp.
PPP Jobs
of Three
Fintech
0.0187***
0.0233***
0.00988
0.0233***
(2.80)
(2.98)
(1.50)
(5.59)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Address × Draw FE
Yes
Yes
Yes
Yes
Observations
105,211
160,990
57,636
815,136
Num. Lenders
2,984
2,621
2,023
4,369
R2
0.657
0.884
0.877
0.773
Mean of Dep. Var.
0.0690
0.505
0.231
0.208
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
26

Table IA.IV. Discontinuity at $100k
In this table, we examine the relationship between our four main ﬂags and implied compensation.
We estimate OLS regressions with dummies for each of our main ﬂags as dependent variables
and dummies for $5k wide bins (i.e., ($0k, $5k], ..., ($95k, $100k], ..., ($125k, $130k]) of implied
compensation as independent variables. The dummy variable for the ($0k, $5k] bin is used as
the baseline. Panel A shows the results for ﬁntech loans and Panel B for traditional loans. In
both panels, speciﬁcation (1) uses business registry as the dependent variable, (2) uses multiple
loans, (3) uses high implied compensation, and (4) uses EIDL > PPP jobs. Loans are ﬁltered to
corporation, S-corporation, and LLC loans for speciﬁcation (1), all loans for (2), loans for which
we can determine industry-CBSA average compensation for (3), and loans with a matched EIDL
Advance for (4). Fixed eﬀects and control variables are as indicated at bottom of each column.
Robust standard errors are double clustered by zip code and lender.
Panel A. Fintech Loans
(1)
(2)
(3)
(4)
Dep. Variable:
Business
Multiple
High Implied
EIDL > PPP
Registry
Loans
Comp.
Jobs
($0k, $5k]
—————————————— Used as Baseline ——————————————
($5k, $10k]
-0.00903** (-2.45)
0.00812*** (3.63)
-0.0191** (-2.26)
-0.0123*** (-2.99)
($10k, $15k]
-0.0185*** (-3.11)
0.0122*** (3.43)
-0.351** (-2.58)
-0.0173*** (-3.38)
($15k, $20k]
-0.0155** (-2.40)
0.0136*** (3.23)
-0.0376** (-2.13)
-0.00893 (-1.39)
($20k, $25k]
-0.0201** (-2.40)
0.0162*** (3.41)
-0.0249 (-1.24)
-0.00576 (-0.82)
($25k, $30k]
-0.0143 (-1.54)
0.0170*** (3.35)
-0.0191 (-0.87)
-0.000893 (-0.11)
($30k, $35k]
-0.0151 (-1.60)
0.0185*** (3.39)
-0.0161 (-0.68)
0.0102 (1.28)
($35k, $40k]
-0.0132 (-1.31)
0.0209*** (3.55)
-0.0153 (-0.61)
0.0236** (2.43)
($40k, $45k]
-0.00996 (-0.96)
0.0220*** (3.70)
-0.00535 (-0.20)
0.0210** (2.18)
($45k, $50k]
-0.00674 (-0.61)
0.0270*** (4.22)
-0.0168 (-0.58)
0.0501*** (3.85)
($50k, $55k]
-0.00296 (-0.26)
0.0255*** (4.02)
-0.00221 (-0.07)
0.0414*** (3.64)
($55k, $60k]
-0.00612 (-0.52)
0.0277*** (3.99)
0.00916 (0.29)
0.0537*** (4.33)
($60k, $65k]
-0.00380 (-0.32)
0.0294*** (4.25)
0.0168 (0.50)
0.0696*** (4.64)
($65k, $70k]
0.0000156 (0.00)
0.0310*** (4.28)
0.0427 (1.26)
0.0877*** (5.51)
($70k, $75k]
-0.00275 (-0.24)
0.0306*** (4.19)
0.0953*** (2.72)
0.0938*** (5.45)
($75k, $80k]
0.0122 (1.07)
0.0352*** (4.78)
0.134*** (3.88)
0.127*** (5.63)
($80k, $85k]
0.0124 (1.09)
0.0336*** (4.87)
0.179*** (5.09)
0.141*** (6.37)
($85k, $90k]
0.00848 (0.72)
0.0365*** (4.83)
0.223*** (6.41)
0.174*** (7.49)
($90k, $95k]
0.0170 (1.37)
0.0398*** (5.41)
0.267*** (7.43)
0.243*** (10.28)
($95k, $100k]
0.0394** (2.37)
0.0400*** (5.73)
0.296*** (8.26)
0.299*** (17.54)
($100k, $105k]
-0.0100 (-0.59)
0.0407*** (3.46)
0.254*** (6.78)
0.100*** (5.38)
($105k, $110k]
-0.0136 (-0.94)
0.0400*** (3.84)
0.242*** (5.85)
0.106*** (5.17)
($110k, $115k]
-0.0142 (-0.91)
0.0403*** (4.44)
0.239*** (5.81)
0.0879*** (4.89)
($115k, $120k]
-0.0197 (-1.31)
0.0356*** (3.57)
0.260*** (6.46)
0.0949*** (5.59)
($120k, $125k]
-0.0126 (-0.76)
0.0310*** (3.22)
0.281*** (6.91)
0.0724*** (3.91)
($125k, $130k]
-0.0132 (-0.67)
0.0305*** (3.58)
0.258*** (5.92)
0.0625** (2.56)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Observations
693,826
3,865,035
3,710,771
673,469
Num. Lenders
77
77
77
77
R2
0.237
0.050
0.639
0.448
Mean of Dep. Var.
0.101
0.0348
0.181
0.220
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
27

Panel B. Traditional Loans
(1)
(2)
(3)
(4)
Dep. Variable:
Business
Multiple
High Implied
EIDL > PPP
Registry
Loans
Comp.
Jobs
($0k, $5k]
—————————————— Used as Baseline ——————————————
($5k, $10k]
-0.00812*** (-4.58)
0.000627 (0.68)
-0.00744** (-2.25)
-0.0107** (-2.25)
($10k, $15k]
-0.0125*** (-5.26)
0.00139 (1.10)
-0.00706 (-1.60)
-0.0116** (-2.08)
($15k, $20k]
-0.0144*** (-5.14)
0.00147 (0.96)
-0.00109 (-0.23)
-0.00649 (-1.22)
($20k, $25k]
-0.0161*** (-5.11)
0.00116 (0.66)
0.00754 (1.53)
-0.000960 (-0.17)
($25k, $30k]
-0.0158*** (-4.39)
0.00108 (0.56)
0.0166*** (3.18)
0.00454 (0.78)
($30k, $35k]
-0.0152*** (-3.98)
0.00102 (0.50)
0.0266*** (4.85)
0.0101 (1.61)
($35k, $40k]
-0.0162*** (-3.91)
0.00766 (0.35)
0.0356*** (6.22)
0.0133** (2.16)
($40k, $45k]
-0.0153*** (-3.54)
0.0000999 (0.43)
0.0463*** (7.68)
0.0187*** (2.80)
($45k, $50k]
-0.0145*** (-3.24)
0.00169 (0.69)
0.0568*** (8.97)
0.0206*** (3.01)
($50k, $55k]
-0.0141*** (-2.94)
0.00118 (0.46)
0.0690*** (10.33)
0.0251*** (3.40)
($55k, $60k]
-0.0129*** (-2.69)
0.00146 (0.56)
0.0827*** (11.79)
0.0323*** (4.04)
($60k, $65k]
-0.0138*** (-2.73)
0.00186 (0.71)
0.0958*** (12.93)
0.0381*** (4.04)
($65k, $70k]
-0.0136*** (-2.63)
0.00184 (0.66)
0.114*** (14.40)
0.0460*** (4.94)
($70k, $75k]
-0.0119** (-2.24)
0.00239 (0.83)
0.128*** (16.08)
0.0405*** (5.00)
($75k, $80k]
-0.0110** (-2.02)
0.00339 (1.15)
0.136*** (17.48)
0.0347*** (4.26)
($80k, $85k]
-0.0114** (-2.07)
0.00307 (1.03)
0.148*** (17.52)
0.0403*** (4.77)
($85k, $90k]
-0.0106* (-1.91)
0.00405 (1.33)
0.166*** (17.68)
0.0474*** (5.54)
($90k, $95k]
-0.00824 (-1.49)
0.00521 (1.69)
0.178*** (16.83)
0.0555*** (5.05)
($95k, $100k]
0.0150** (2.52)
0.00475 (1.50)
0.190*** (13.96)
0.0764*** (4.38)
($100k, $105k]
-0.0141** (-2.07)
0.00345 (1.11)
0.180*** (16.45)
0.0568*** (5.04)
($105k, $110k]
-0.0162*** (-2.66)
0.00304 (0.93)
0.186*** (18.15)
0.0557*** (6.08)
($110k, $115k]
-0.0141** (-2.23)
0.00310 (0.97)
0.203*** (19.56)
0.0654*** (6.87)
($115k, $120k]
-0.0166*** (-2.59)
0.00429 (1.21)
0.223*** (22.51)
0.0683*** (6.45)
($120k, $125k]
-0.0161** (-2.45)
0.00322 (0.98)
0.240*** (19.72)
0.0645*** (6.52)
($125k, $130k]
-0.0170** (-2.46)
0.00402 (1.17)
0.258*** (24.81)
0.0698*** (6.94)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Observations
4,512,085
7,079,739
6,052,314
1,931,998
Num. Lenders
4,643
4,794
4,787
4,655
R2
0.074
0.058
0.260
0.108
Mean of Dep. Var.
0.0433
0.00923
0.0358
0.0486
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
28

Table IA.V. Criminal Records
In this table, we examine the relationship between each of our ﬂags and criminal records for loans
within a subsample of 150,000 rounds 1-2 loans. We estimate OLS regressions with a dummy for
whether the borrower has a felony from 2000 or after on their record as the dependent variable
and dummies for whether the loan is ﬂagged by each of our ﬂags individually as the independent
variable. Panel A shows the relationship for ﬁntech loans and Panel B for traditional loans. Loans
are ﬁltered to the sets for which we can determine the ﬂag (same as in Figures 2-8). Note that
the business registry ﬂag is not included since we can only determine criminal records for loans
to individuals while the business registry ﬂag can only be determined for corporations and LLCs.
Fixed eﬀects and control variables are as indicated at bottom of each column. Robust standard
errors are double clustered by zip code and lender.
Panel A. Fintech Loans
Dep. Variable: Felony Post-2000
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Multiple Loans at
0.0141***
Address
(2.77)
High Implied
0.0256***
Comp.
(7.22)
EIDL Jobs >
0.0762***
PPP Jobs
(11.90)
$100k Implied.
0.0128***
Comp.
(4.56)
Monthly
0.00899*
Rounding
(1.83)
Overrep. in
0.0173***
County/NAICS
(3.32)
High
0.00263*
Concentration
(0.63)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
No
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Yes
No
Yes
Lender FE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
54,365
17,511
10,625
54,376
54,376
28,176
52,490
Num. Lenders
69
45
47
69
59
69
61
R2
0.127
0.113
0.183
0.126
0.126
0.008
0.122
Mean of Dep. Var.
0.0487
0.0590
0.0494
0.0481
0.0481
0.0490
0.0483
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
29

Panel B. Traditional Loans
Dep. Variable: Felony Post-2000
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Multiple Loans at
0.00851
Address
(1.65)
High Implied
0.00797
Comp.
(1.50)
EIDL Jobs >
0.0378*
PPP Jobs
(1.87)
$100k Implied
-0.00162
Comp.
(-1.19)
Monthly
0.00215
Rounding
(0.73)
Overrep. in
0.00311**
County/NAICS
(2.64)
High
0.00524
Concentration
(1.27)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
No
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Yes
No
Yes
Lender FE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
64,251
11,298
7,067
64,251
64,251
44,551
57,913
Num. Lenders
2,466
1,036
596
2,466
2,466
2,163
2,195
R2
0.225
0.305
0.379
0.225
0.225
0.053
0.219
Mean of Dep. Var.
0.0169
0.0171
0.0150
0.0136
0.0136
0.0133
0.0133
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
30

Table IA.VI. Prevalence of Secondary Flags by Lender Types
In this table, we examine the prevalence of the secondary ﬂags by lender type. We estimate OLS
regressions with each of the secondary ﬂags as the dependent variable and a dummy for whether
the lender is a ﬁntech as the independent variable. Loans are ﬁltered to the sets for which we can
determine each ﬂag (same as in Figures 5-9). Fixed eﬀects and control variables are as indicated
at bottom of each column. Robust standard errors are double clustered by zip code and lender.
Panel A. Unadjusted Diﬀerences
(1)
(2)
(3)
(4)
(5)
Dep. Variable:
$100k
Monthly
Overrep. in
High
Felony
Comp.
Rounding
County/NAICS
Concentration
Fintech
0.199***
0.0263*
0.319***
0.404***
0.0337***
(6.09)
(1.84)
(10.80)
(4.73)
(8.56)
Observations
11,768,676
5,458.913
6,414,028
8,040,494
150,000
Num. Lenders
4,890
4,872
4,875
4,405
3,656
R2
0.069
0.001
0.087
0.216
0.010
Mean of Dep. Var.
0.150
0.0787
0.467
0.225
0.0274
Panel B. Adjusted Diﬀerences
(1)
(2)
(3)
(4)
(5)
Dep. Variable:
$100k
Monthly
Overrep. in
High
Felony
Comp.
Rounding
County/NAICS
Concentration
Fintech
0.0658***
0.0168
0.122***
0.279***
0.0130***
(4.55)
(1.23)
(9.87)
(4.62)
(7.51)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
No
Business Type FE
Yes
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
No
Yes
Yes
Observations
11,086,014
5,166,522
6,408,461
7,615,707
119,799
Num. Lenders
4,873
4,830
4,874
4,240
3,045
R2
0.313
0.040
0.228
0.527
0.209
Mean of Dep. Var.
0.148
0.0800
0.467
0.214
0.0291
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
31

Table IA.VII. Within State/County Variation
This table examines the degree to which geographic variation in ﬂagged loans can be explained
by ﬁntech market share. We estimate OLS regressions with the percentages of ﬂagged loans in
each zip code as the dependent variable and the ﬁntech markey share in each zip code as the
independent variable. Speciﬁcation (1) examines the relationship across all zip codes, (2) examines
the relationship within states, and (3) examines the relationship within counties. Zip codes with at
least 100 loans are considered. Fixed eﬀects are indicated at the bottom of each column. Robust
standard errors are clustered at the county level.
Dep. Variable: Percentage Flagged in Zip Code
(1)
(2)
(3)
Fintech Market Share
0.281***
0.321***
0.321***
(18.14)
(23.81)
(16.52)
State FE
No
Yes
No
County FE
No
No
Yes
Observations
15,835
15,835
15,011
Num. Counties
2,903
2,903
2,079
R2
0.554
0.685
0.824
Mean of Dep. Var.
0.104
0.104
0.106
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
32

Table IA.VIII. County Cultural Features
This table replicates Table V at the county-level rather than the loan-level. The variables are as
deﬁned in Table V. All variables (independent and dependent) are rescaled to have a mean of 0
and a standard deviation of 1. Counties with at least 100 loans are considered. Fixed eﬀects are
indicated at the bottom of each column. Robust standard errors are clustered at the state level.
Dep. Variable: Percentage Flagged
(1)
(2)
(3)
Public Corruption
0.150**
0.116*
0.0291
(2.28)
(1.95)
(1.06)
Religious Aﬃliation
-0.0539*
-0.0502**
0.00475
(-1.93)
(-2.18)
(0.27)
Ashley Madison Usage
0.229***
0.152***
0.00704
(8.27)
(4.75)
(0.27)
Population Density
-0.0177
-0.0528***
(-1.37)
(-5.31)
Median Income
-0.0537
-0.0696**
(-1.22)
(-2.60)
Pct. Non-White
0.447***
0.0316
(8.48)
(0.81)
College Educated
0.133***
0.140***
(5.45)
(8.43)
2019 Unemployment
-0.0652
-0.0861**
(-1.17)
(-2.50)
Pct. Fintech
0.811***
(10.47)
State FE
Yes
Yes
Yes
Observations
3,013
3,012
3,012
R2
0.417
0.531
0.693
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
33

Table IA.IX. Previous SBA Lending
In this table, we examine lender-level relationship between suspicious lending in the PPP and
previous SBA lending. ln(Num. 7(a) Loans Pre-2020 + 1) is the natural log of the number of SBA
7(a) loans originated by the lender pre-2020. Num. Year Since First 7(a) Loan is the number of
years between when the lender originated its ﬁrst SBA 7(a) loan and 2020 (we have data going
back to 1990, so this variable can take a maximum value of 30). New Lender is a dummy that
takes one a value of 1 if the lender had not originated any SBA 7(a) loans pre-2020. 1(Fintech)
and 1(Traditional) are indicator functions for whether the lenders is a ﬁntech or traditional lender,
respectively. Lenders with at least 1,000 PPP loans are considered. Robust standard errors are
used.
Dep. Variable: Percentage Flagged by at Least One Primary Flag
(1)
(2)
(3)
(4)
ln(Num. 7(a) Loans Pre-2020 + 1)
-0.00220***
(-3.22)
Num. Year Since First 7(a) Loan
-0.00131***
(-6.59)
New Lender
0.0341***
(3.82)
× 1(Fintech)
0.0378
(1.44)
× 1(Traditional)
0.0224***
(2.71)
1(Fintech)
0.0484***
(4.92)
Observations
1,141
1,141
1,141
1,141
R2
0.018
0.111
0.055
0.161
Mean of Dep. Var.
0.0736
0.0736
0.0736
0.0736
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
34

Table IA.X. Fees
In this table, we show the dollars value of loans originated, number of loans originated, and the
estimated fees received by the top 75 PPP lenders (by number of loans). The list is sorted in
descending order by estimated fees. The ﬁrst forty lender are on this page and the remaining 35
are on the next page.
Lender
Lender Type
Dollars Lent
Number of Loans
Estimated Fees
JPMChase
Traditional
$41,724,832,566
438,571
$1,692,343,319
BoA
Traditional
$34,318,969,946
491,037
$1,471,270,588
Prestamos CDFI
Fintech
$7,706,291,810
495,547
$1,176,524,235
Capital Plus
Fintech
$7,421,361,604
463,598
$1,101,706,143
Harvest
Fintech
$8,701,102,355
433,306
$1,064,286,118
Cross River
Fintech
$12,927,230,933
479,871
$1,037,484,722
Benworth
Fintech
$4,567,637,099
331,317
$745,155,411
Wells Fargo
Traditional
$13,894,207,311
280,694
$691,472,470
Customers Bank
Traditional
$7,137,984,962
287,470
$651,576,422
Fountainhead
Fintech
$4,150,617,279
272,705
$642,366,086
Lendistry
Fintech
$4,944,781,183
249,321
$630,539,114
PNC
Traditional
$17,390,351,788
119,354
$601,990,855
Branch
Traditional
$16,727,409,331
117,952
$590,928,473
U.S. Bank
Traditional
$10,834,749,543
174,825
$506,152,576
TD Bank
Traditional
$12,291,334,013
133,136
$495,335,337
Itria
Fintech
$5,102,367,975
177,790
$488,823,558
Huntington
Traditional
$11,284,620,628
83,188
$402,221,521
KeyBank
Traditional
$11,135,483,364
69,558
$376,725,537
Zions
Traditional
$9,873,297,016
76,534
$356,275,679
M&T
Traditional
$9,661,643,581
59,094
$335,143,327
Readycap
Fintech
$5,061,982,022
111,228
$324,957,172
Citizens Bank
Traditional
$7,185,615,514
85,678
$294,273,948
Fifth Third
Traditional
$7,397,570,873
65,997
$270,545,233
Regions Bank
Traditional
$6,423,446,492
78,544
$266,969,904
Celtic
Fintech
$4,627,792,089
167,203
$235,744,519
First Horizon
Traditional
$5,797,703,273
50,976
$218,555,438
WebBank
Fintech
$3,190,673,848
119,292
$214,816,890
Citibank
Traditional
$4,786,384,820
47,768
$191,158,479
Kabbage
Fintech
$3,317,781,390
179,820
$187,787,604
BMO Harris
Traditional
$6,117,701,222
36,237
$186,505,159
City National
Traditional
$5,954,598,740
25,543
$185,730,854
First-Citizens
Traditional
$4,432,975,119
35,578
$177,150,216
Frost
Traditional
$4,702,944,223
32,518
$171,420,904
Pinnacle
Traditional
$4,282,505,226
37,943
$170,829,371
Northeast
Traditional
$3,513,827,150
35,469
$164,925,053
Bank of the West
Traditional
$4,398,310,337
31,111
$164,558,480
Leader Bank
Fintech
$1,336,378,802
62,757
$155,573,115
Square
Fintech
$681,853,014
72,570
$145,565,241
Synovus
Traditional
$3,842,389,064
27,810
$142,270,309
Comerica Bank
Traditional
$4,959,450,533
21,027
$142,259,791
35

Lender
Lender Type
Dollars Lent
Number of Loans
Estimated Fees
BBVA
Traditional
$4,080,037,532
30,695
$140,674,018
First National PA
Traditional
$3,667,129,883
30,401
$137,338,347
People’s United
Traditional
$3,697,583,298
30,952
$135,503,885
South State
Traditional
$3,238,850,035
28,037
$131,941,339
Hancock Whitney
Traditional
$3,337,831,677
21,601
$117,588,402
Umpqua
Traditional
$2,912,581,701
26,421
$111,921,925
Valley National
Traditional
$3,239,909,935
19,772
$111,182,058
MUFG
Traditional
$3,074,469,662
21,231
$109,724,419
Banco Popular
Traditional
$1,775,879,079
47,457
$99,635,470
Newtek
Fintech
$2,107,995,767
27,758
$96,761,658
United Community
Traditional
$2,220,819,108
22,239
$90,727,018
Glacier Bank
Traditional
$2,015,613,877
24,553
$88,176,938
FirstBank
Traditional
$1,984,886,943
26,429
$87,215,605
BancorpSouth
Traditional
$1,822,888,745
25,555
$83,880,069
First Bank
Traditional
$1,866,103,529
20,153
$81,071,969
Atlantic Union
Traditional
$2,213,653,002
16,923
$80,697,177
Prosperity
Traditional
$2,009,570,006
18,719
$80,548,892
Arvest
Traditional
$1,654,425,266
27,526
$79,719,868
United Bank
Traditional
$2,054,702,073
17,953
$79,217,758
Capital One
Fintech
$1,782,975,418
24,940
$78,476,632
Blue Ridge
Traditional
$1,163,973,161
24,171
$77,397,714
Webster
Traditional
$1,987,579,454
18,483
$77,283,640
Santander
Traditional
$1,768,965,307
19,963
$73,347,246
Amur Equipment
Traditional
$651,553,095
28,005
$72,596,432
MBE
Fintech
$881,524,566
37,195
$72,475,542
Peoples Bank
Traditional
$1,636,678,317
19,321
$72,368,778
First Insterstate
Traditional
$1,637,638,148
18,854
$71,495,727
American Lending
Fintech
$608,503,695
22,542
$59,858,806
A10Capital
Traditional
$661,264,536
19,617
$57,613,469
TEC-CC
Traditional
$535,671,262
21,624
$56,921,187
First State
Traditional
$992,702,386
19,288
$55,618,703
DreamSpring
Fintech
$296,007,066
29,147
$55,483,834
Texas National
Traditional
$533,598,748
20,611
$53,464,431
Funding Circle
Fintech
$573,628,956
16,858
$43,498,786
Intuit
Fintech
$638,052,437
18,560
$30,626,291
36

Table IA.XI. Persistence and Growth by Lender-Region Pairs
In this table, we examine persistence and growth of suspicious lending at the lender-region level.
Panel A is based on data at the lender-zip code level and Panel B on lender-county level. We
estimate OLS regressions with the percentage of ﬂagged loans during rounds 1-2 within each lender-
region pair interacted with whether the lender is a ﬁntech or traditional lender as the independent
variables. In speciﬁcation (1), the dependent variable is whether the lender increased its lending
within the region (as a percentage of its overall lending) between rounds 1-2 and round 3; in
speciﬁcation (2), the dependent variable is the percentage change in the percentage of the lender’s
loans that are in the region between rounds 1-2 and round 3; in speciﬁcation (3), the dependent
variable is the percentage of ﬂagged loans during round 3 within each lender-region pair. In both
panels. lender-region pairs with at least 25 loans in rounds 1-2 (combined) are considered. Fixed
eﬀects are as indicated at bottom of each column. Robust standard errors are double clustered by
region (zip code in Panel A and county in Panel B) and lender.
Panel A. By Lender-Zip Code
(1)
(2)
(3)
Dep. Variable:
1(Lending
Lending
Pct. Flagged
Growth)
Pct. Change
in Round 3
Pct. Flagged in Rounds 1-2
× 1(Fintech)
1.419***
2.209***
0.516***
(6.10)
(6.36)
(13.80)
× 1(Traditional)
0.00999
0.0128
0.114***
(0.11)
(0.10)
(4.06)
Zip Code FE
Yes
Yes
Yes
Lender FE
Yes
Yes
Yes
Observations
37,271
37,271
37,271
Num. Lenders
1,635
1,635
1,635
R2
0.410
0.575
0.320
Mean of Dep. Var.
0.349
-0.0732
0.0925
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
37

Panel B. By Lender-County
(1)
(2)
(3)
Dep. Variable:
1(Lending
Lending
Pct. Flagged
Growth)
Pct. Change
in Round 3
Pct. Flagged in Rounds 1-2
× 1(Fintech)
1.275***
2.106***
0.466***
(3.59)
(3.15)
(4.15)
× 1(Traditional)
0.116
0.0739
0.173***
(1.11)
(0.33)
(4.80)
County FE
Yes
Yes
Yes
Lender FE
Yes
Yes
Yes
Observations
20,047
20,047
20,047
Num. Lenders
2,347
2,347
2,347
R2
0.372
0.530
0.356
Mean of Dep. Var.
0.427
0.0976
0.0947
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
38

Table IA.XII. Repayments, Enforcement Actions, and Canceled Loans
In this table, we examine the relationship between repaid loans, enforcement actions, and canceled
loans and whether the loan is ﬂagged. We estimate OLS regressions with a dummy for whether the
loan is ﬂagged by at least one primary ﬂag as the independent variable and a dummy for whether
the loan is repaid (Panel A), part of a DOJ enforcement action (Panel B), or canceled (Panel C) as
the dependent variable. Even columns include lender ﬁxed eﬀects and odd columns do not. Panel
A is based on the loan-level data released by the SBA on May 3, 2021, and Panel B and C are
based on the loan-level data released by the SBA on December 1, 2020. Since loans originated by
MBE Capital Partners makes up 42% of the repaid loans, Panel A includes speciﬁcations across
all loans and excluding loans by MBE Capital Partners. Fixed eﬀects and control variables are as
indicated at bottom of each column. Robust standard errors are double clustered by zip code and
lender.
Panel A. Repayment
Dep. Variable: 1(Repaid)
(1)
(2)
(3)
(4)
All Loans
Ex. MBE Capital Partners
Flagged
0.00195
0.000250
0.000574**
0.000907***
(1.41)
(0.36)
(2.00)
(3.91)
ln(Jobs Reported)
Yes
Yes
Yes
Yes
ln(Loan Amount)
Yes
Yes
Yes
Yes
Zip FE
Yes
Yes
Yes
Yes
Business Type FE
Yes
Yes
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Yes
Yes
Lender FE
No
Yes
No
Yes
Observations
4,866,890
4,866,836
4,843,296
4,843,296
Num. Lenders
4,823
4,769
4,822
4,768
R2
0.031
0.159
0.027
0.049
Mean of Dep. Variable
0.00328
0.00328
0.00192
0.00192
Panel B. DOJ Enforcement
Dep. Variable: 1(DOJ Enforcement Action)
(1)
(2)
Flagged
0.0000929**
0.0000900**
(2.35)
(2.26)
ln(Jobs Reported)
Yes
Yes
ln(Loan Amount)
Yes
Yes
Zip FE
Yes
Yes
Business Type FE
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Lender FE
No
Yes
Observations
4,866,890
4,866,836
Num. Lenders
4,823
4,769
R2
0.025
0.026
Mean of Dep. Variable
0.0000541
0.0000541

Panel C. Canceled Loans
Dep. Variable: 1(Canceled)
(1)
(2)
Flagged
0.00478***
0.00131*
(3.49)
(1.93)
ln(Jobs Reported)
Yes
Yes
ln(Loan Amount)
Yes
Yes
Zip FE
Yes
Yes
Business Type FE
Yes
Yes
NAICS × CBSA FE
Yes
Yes
Lender FE
No
Yes
Observations
10,043,880
10,043,853
Num. Lenders
4,885
4,860
R2
0.030
0.070
Mean of Dep. Variable
0.00893
0.00893
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
40

Table IA.XIII. Undisbursed Loans
In this table, we examine the impact of excluding undisbursed loans. Undisbursed loans are those
listed as “Active Un-Disbursed” in the loan-level data. Panel A is based on loans originated by
any lender, Panel B by ﬁntech lenders, and Panel C by traditional lenders. Column (1) shows the
percentage of ﬂagged loans in our entire main loan-level data, Column (2) shows the percentage of
ﬂagged loans after excluding the 465,769 undisbursed loans, and Column (3) shows the percentage
of ﬂagged loans in 465,769 undisbursed loans. For rows based on individual ﬂags, loans are ﬁltered
to the sets for which we can determine each ﬂag; for rows based on sets of ﬂags, all loans are
included.
Panel A. Overall
(1)
(2)
(3)
All Loans
Ex. Undisbursed
Undisbursed
Business Registry
0.0512
0.0499
0.182
Multiple Loans
0.0183
0.0180
0.0249
High Implied Comp.
0.252
0.233
0.641
EIDL > PPP Jobs
0.0911
0.0868
0.343
At Least One Flag
0.129
0.123
0.276
Primary + Additional
0.0881
0.0817
0.243
Pct. Fintech
0.337
0.318
0.809
Panel B. Fintech
(1)
(2)
(3)
All Loans
Ex. Undisbursed
Undisbursed
Business Registry
0.102
0.0965
0.241
Multiple Loans
0.0344
0.0350
0.0286
High Implied Comp.
0.489
0.467
0.687
EIDL > PPP Jobs
0.216
0.206
0.420
At Least One Flag
0.236
0.229
0.303
Primary + Additional
0.198
0.190
0.274
Panel C. Traditional
(1)
(2)
(3)
All Loans
Ex. Undisbursed
Undisbursed
Business Registry
0.0434
0.0430
0.111
Multiple Loans
0.0100
0.0100
0.00903
High Implied Comp.
0.0942
0.0907
0.388
EIDL > PPP Jobs
0.0487
0.0480
0.151
At Least One Flag
0.0743
0.0733
0.161
Primary + Additional
0.0321
0.03126
0.111
41

Table IA.XIV. Prevalence of Flags by Lender Type, Excluding Undisbursed Loans
This table replicates Table III after excluding undisbursed loans. Undisbursed loans are those listed
as “Active Un-Disbursed” in the loan-level data.
(1)
(2)
(3)
(4)
(5)
Fintech
Traditional
Unadjusted
Adjusted
Matched
Diﬀerence
Diﬀerence
Diﬀerence
Business
0.0965
0.0430
0.0535***
0.0330***
0.0246***
Registry
N = 711,086
N = 4,801,246
(3.04)
(3.47)
(3.15)
Multiple
0.0350
0.0100
0.0249***
0.0110***
0.0157***
Loans
N = 3,593,267
N = 7,709,653
(8.93)
(5.88)
(3.97)
High Implied
0.467
0.0907
0.377***
0.0925***
0.0847***
Comp.
N = 1,232,410
N = 2,026,091
(7.65)
(7.41)
(3.96)
EIDL > PPP
0.206
0.0480
0.158***
0.0629***
0.0783***
Jobs
N = 671,895
N = 2,069,311
(4.43)
(4.13)
(4.54)
At Least One
0.229
0.0733
0.155***
0.0562***
0.0547***
Flag
N = 3,593,267
N = 7,709,653
(8.08)
(8.54)
(6.55)
At Least Two
0.0244
0.00353
0.0209***
0.00709***
0.00981***
Flags
N = 3,593,267
N = 7,709,653
(7.38)
(7.07)
(5.20)
t-statistics in parentheses
* p<0.10, ** p<0.05, *** p<0.010
42

File and source

File: Griffin-Kruger-Mahajan_did-fintech-lenders-facilitate-ppp-fraud_wp2021.pdf
Size: 4,611,504 bytes
SHA-256: eb3cf9d98e8ec5a8ee1ef7d9ca0c49c32190149c75e02118d4d263d654e9e36f
Our copy: Griffin-Kruger-Mahajan_did-fintech-lenders-facilitate-ppp-fraud_wp2021.pdf
Original: fraconference.com

tial misreporting including non-registered businesses, multiple businesses at residential

Full text

File and source

Related records