Who Supplies PPP Loans (and Does It Matter)? Banks, Relationships and the COVID Crisis (Working Paper 28286)
Summary
NBER Working Paper No. 28286, Who Supplies PPP Loans (And Does it Matter)? Banks, Relationships and the COVID Crisis, by Lei Li and Philip Strahan, dated December 2020. The abstract states that the authors analyze bank supply of credit under the Paycheck Protection Program and find that PPP supply reflects traditional measures of relationship lending: decreasing in bank size, and increasing in prior experience, in commitment lending, and in core deposits. It states that bank PPP supply, based on the structure of the local banking sector, alleviates increases in unemployment. The introduction states that small and medium sized banks, those with assets under $50 billion, provide about two thirds of PPP loans, $310 billion out of $494 billion by all banks, or 63%.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
- Issuer
- Source documents
- Document type
- Who Supplies Ppp Loans And Does It Matter Banks Relationships And The
- Case
- Who Supplies Ppp Loans And Does It Matter Banks Relationships And The
Cite this document
“Who Supplies PPP Loans (and Does It Matter)? Banks, Relationships and the COVID Crisis (Working Paper 28286).”
Original at www.nber.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/who-supplies-ppp-loans-and-does-it-matter-banks-relationships-and-the/
, SHA-256 99ff8da4230ae15a6753d8eaa3773ddc8b818bb7e1c4342df1d3bf4b0fe0e09d.Full text
NBER WORKING PAPER SERIES
WHO SUPPLIES PPP LOANS (AND DOES IT MATTER)?
BANKS, RELATIONSHIPS AND THE COVID CRISIS
Lei Li
Philip Strahan
Working Paper 28286
http://www.nber.org/papers/w28286
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
December 2020
The opinions in the paper do not represent those of the Federal Reserve Board of Governors, any
other affiliate of the Federal Reserve System, or the National Bureau of Economic Research. We
thank seminar participants at Boston College, the Federal Reserve Board of Governors,
University of Minnesota and Southern Methodist University, as well as participants and our
discussant Lawrence Schmidt at the TAU Conference on Financial Intermediation during the
COVID-19 Crisis.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2020 by Lei Li and Philip Strahan. All rights reserved. Short sections of text, not to exceed two
paragraphs, may be quoted without explicit permission provided that full credit, including ©
notice, is given to the source.
Who Supplies PPP Loans (And Does it Matter)? Banks, Relationships and the COVID Crisis
Lei Li and Philip Strahan
NBER Working Paper No. 28286
December 2020
JEL No. G2
ABSTRACT
We analyze bank supply of credit under the Paycheck Protection Program (PPP). The literature
emphasizes relationships as a means to improve lender information, which helps banks manage
credit risk. Despite imposing no risk, however, PPP supply reflects traditional measures of
relationship lending: decreasing in bank size; increasing in prior experience, in commitment
lending, and in core deposits. Our results suggest a new benefit of bank relationships, as they help
firms access government-subsidized lending. Consistent with this benefit, we show that bank PPP
supply, based on the structure of the local banking sector, alleviates increases in unemployment.
Lei Li
Federal Reserve Board of Governors
Washington, DC, DC
lei.li@frb.gov
Philip Strahan
Carroll School of Management
324B Fulton Hall
Boston College
Chestnut Hill, MA 02467
and NBER
philip.strahan@bc.edu
I. INTRODUCTION
When governments intervene in the financial system and the economy, they often do so
by influencing or bailing out banks. For example, in 1998 the Federal Reserve effected a
private-sector bailout of the hedge fund Long-Term Capital Management through moral suasion
of their main counterparties, banks. In 2008, the housing sector, which lay at the center of the
crisis, did not receive a substantial bailout. Instead, banks – the primary provider of mortgage
credit – were the focus of policy interventions with the Troubled Asset Relief Program and the
first round of Quantitative Easing, as well as the Home Affordable Modification Program
(HAMP) and the Home Affordable Refinance Program (HARP). 1 In 2020, the Federal Reserve
created the Main Street Lending program to help middle-market firms gain access to credit, but
they did so by providing liquidity support to banks not to firms directly. Also in 2020, the
Federal Reserve created the Money Market Mutual Fund Liquidity Facility, in which it uses
banks to provide liquidity support to money funds. This paper uses micro evidence from the
Paycheck Protection Program (PPP) to provide further evidence that banks act as the main
conduit for access to government subsidies. We argue that our results provide a new (or perhaps
unrecognized) rationale for the benefit to firms of close banking relationships.
We quantify the importance of banks in general, and relationship banks in particular, in
supplying subsidized credit under the PPP program, created as part of the Coronavirus Aid,
Relief, and Economic Security (CARES) Act. The PPP aims to help small businesses avoid
laying off their workers during the peak of the COVID-19 economic crisis. We show that
relationship banks supply more of this credit to their borrowers, and that localities with more
1
For a critique of these actions, see Mian and Sufi (2014).
2
relationship banking receive PPP credit earlier. As a result, these areas experience smaller
increases in unemployment. Despite this intended effect of the PPP program, we find no
evidence that broader measures of economic output – small business revenues or total spending –
respond to the program.
We start by asking: who supplies business credit during the COVID crisis? Aggregate
figures provide a partial answer. As shown in Li et al. (2020), before the PPP program began in
April, large banks respond to widespread demand for liquidity by large firms by expanding
lending on an unprecedented scale during the last three weeks of the first quarter of 2020.
During March, liquidity in money markets and bond markets becomes constrained, leading firms
to ‘run’ to their banks and draw funds from pre-existing credit lines. This expansion in bank
lending is evident in Figure 1, which compares cumulative loan growth for large versus small
banks. After March, however, large banks experience contractions in lending as some large
borrowers, with renewed access to liquidity from the markets, pay back their loans. 2
Lending by small banks, who traditionally focus on relationship lending to small firms,
grows sharply in April, reflecting their participation in the PPP program. Small and medium
sized banks (those with assets under $50 billion) provide about two thirds of the loans under the
PPP program ($310 billion out of $494 billion by all banks, or 63%). This share exceeds their
share of lending to small business before the COVID crisis, which was just 44% at the end of
2019. In contrast to PPP loans, new lending to business outside the PPP program stagnated for
all banks during the second quarter. Figure 2 reports state-level heat maps of the quantity of PPP
credit in the first round (April 3-17) and second round (April 27-August 9) of the program, along
2
For further evidence on the effects of bond-market disruptions, see Acharya and Steffen (2020), Darmouni and
Siani (2020), Chodorow-Reich et al. (2020), Greenwald et al. (2020), and Hotchkiss et al. (2020).
3
with a similar one for the state prevalence of small banks. These figures suggest visually – and
we verify in our regressions formally - the importance of relationship-oriented banks in getting
the PPP credit to their borrowers immediately. States with more small banks received more PPP
loans in the first round (correlation = 0.65), whereas this relationship reverses sign in the second
round (correlation = -0.77).
Press accounts and anecdotal evidence suggest that firms with better access to banks
before the COVID crisis were able to get PPP funding quickly at the outset of the program,
which was overwhelmed with demand and had its first-round funds exhausted in just two weeks
(Figure 3). 3 This suggests that pre-COVID relationships became valuable by allowing firms
close to their banks to gain access to the government subsidized lending, especially early in the
implementation of the program. Because relationship banks have a long-term interest in the
survival of their borrowers, they have an incentive to help those borrowers access the PPP
program. Moreover, because relationship banks know their borrowers from prior interactions,
application and other origination costs are likely lower for them.
To test this idea comprehensively, we focus on how bank characteristics explain their role
in the PPP program. We estimate regressions based on quarterly Call Report data, which capture
overall business lending as well as lending supplied by banks under the PPP program. We
contrast lending patterns in March – which respond to the crisis in securities markets – with
those in April and subsequent months – which respond to the economic downturn and the advent
of government subsidies. Lending after March expands most at banks typically associated with
close relationships with their borrowers. In particular, lending grows faster at small banks, at
3
See for example, “PPP Money Abounded – But Some Got It Faster than Others,” Wall Street Journal, October 6,
2020.
4
banks with high levels of small business loans prior to the crisis, at banks with high levels of
unused business credit commitments before the crisis, and at banks raising more local, retail
deposits. And, these effects are strongest at the smallest banks.
We then decompose business lending during Q2, 2020 into those made under the PPP
program vs. all other business lending. Essentially all of the growth in lending during the second
quarter comes under the PPP program, and all of the connections between relationship measures
and lending growth reflects PPP lending (as opposed to other bank loans to business).
We validate the importance of relationships using two distinct empirical strategies. First,
we separate each bank’s PPP lending based on whether or not the borrower resides in one of the
bank’s core markets, defined as a county in which the bank owns at least one branch. As we
show, the measures of relationships explain lending in core markets but only weakly in
peripheral ones. We also show that the relationship variables matter most during the first round
of PPP lending, consistent with the idea that banks advantaged their relationship borrowers over
others by helping them access the PPP program first. Second, we report within-bank tests to
compare lending as a function of bank branch and lending characteristics in their core markets.
These tests show that banks lent more to PPP borrowers in their most important core markets
(those where they made more small business loans prior to COVID). We also show that PPP
lending increases with the average age of branches located in the core markets. The within-bank
tests suggest that relationships from longstanding ties with the local economy affect PPP lending.
That is, even after controlling for all cross-bank variation (with fixed effects), bank relationships
still strongly predict their PPP credit supply.
In the last part of our analysis, we link PPP credit to local real outcomes. We show that
variation in the quantity of PPP credit across counties reflects both the size and structure of local
5
banks. Specifically, two pre-COVID measures of banking structure correlate strongly with the
quantity of PPP credit across geographies, after controlling for demographic and economic
covariates. First, and most simply, areas with more branches per eligible establishment (before
COVID) receive more PPP credit. Second, areas with more local, relationship banks (based on
our bank-level predictive model) also receive more PPP credit during the first round of
allocations (the first two weeks of April). We use the relationship variable to capture local PPP
credit supply conditions. Unlike the overall size of the local banking system, this variable
reflects credit supply, not demand, because it correlates negatively with PPP credit from external
banks. We then tie local PPP lending supply to real outcomes. Areas that receive more local
PPP lending supply in the first round (because of the presence of relationship banks), we show,
experience smaller increases in unemployment.
Several studies have assessed the impact of the PPP program on economic outcomes. For
example, Granja et al. (2020) find that more PPP first-round funds flow into localities less
affected by COVID, and that the effects of the PPP program on employment were small relative
to the scale of the funds allocated. Autor et al. (2020) find evidence that employment falls less at
firms eligible for PPP loans than at otherwise similar firms. Chetty et al. (2020) also find some
evidence that PPP lending increases employment, but the effects are small relative to the cost of
the program. Barraza et al. (2020) find that areas with more offices of banks that issued Small
Business Administration (SBA) backed loans in 2019 experience smaller increases in
unemployment after the initiation of the PPP program. Faulkender et al. (2020) also find a
beneficial effect of PPP lending on unemployment, but one that is much larger than these other
studies. Consistent with most of this literature, we also find that PPP supply helps preserve local
employment. That said, our variation only compares unemployment patterns between counties
6
based on whether or not they receive PPP loans in the first round. Ultimately, most PPP
applicants did receive funds, as the program closed with over $100 billion in unallocated funds.
As such, we hesitate to use our approach to assess the effectiveness of the program itself.
Instead, we use this last test to provide further evidence of the benefits of relationship lending,
even when the government has removed all risk from credit providers.
Ours is the first paper we know of to study empirically how bank relationship
characteristics relate to their overall supply of PPP loans. However, several recent papers have
argued that bank relationships have helped firms gain access to PPP loans. Amiram and Rabetti
(2020) focus on publicly traded firms and find that firms with existing banking relationship
received larger PPP loans faster. Cororaton and Rosen (2020) also study publicly traded firms
and find that smaller public firms with more employees, fewer investment opportunities, and
COVID-19 exposure are more likely to borrow from PPP. Erel and Liebersohn (2020) find that
borrowers in areas with fewer bank branches, lower incomes, and more minority population are
more likely to access PPP via FinTech firms rather than banks. A number of studies show that
small, community banks provide an outsized share of PPP loans (Balyuk (2020), Faulkender et
al. (2020), James et al. (2020)). Bartik et al. (2020) use firm survey evidence, finding that firms
with strong bank relationship are more likely to receive PPP loans, while Joaquim and Netto
(2020) provide a theoretical analysis of bank incentives to lend under the program. 4
Our paper contributes a new dimension to the literature on relationship banking. All of
the extant banking literature emphasizes that relationships reduce information asymmetry about
4
For evidence on government guaranteed lending outside the U.S., see Core and De Marco (2020).
7
aspects of borrowers’ ability or willingness to repay. 5 Banks, the story goes, learn about their
customer’s business over time, monitor their cash flows and financial health, and lend based on a
deep understanding of the business and its future prospects. PPP loans, in contrast, have no
credit risk exposure for lending banks, yet relationships strongly predict PPP supply. Banks act
as gatekeepers for the PPP program, shepherding small businesses through the application
process. Banks prioritize their relationship borrowers over others because managing the
application process is less costly for them, and because they have an economic interest in the
long-term survival of their borrowers. 6 Our results suggest that a close bank relationship can
help firms gain access to the program and that such access has a real effect. Hence, our results
point to a new benefit to firms of close relationships with banks: they help their close customers
gain access to government subsidies. So, while existing studies document that relationships are
valuable for conventional reasons (e.g., as shown in Bolton et al. (2016), relationship banks are
more willing to stay with their borrowers during bad times), ours documents a less conventional
but important reason (i.e., relationship banks help their borrowers gain access to government
subsidies).
5
There is a long literature on bank relationship lending which we will not review here. However, Petersen and
Rajan (1994) is the seminal empirical analysis on the subject; they emphasizes the role of duration in relationship
formation. Berger et al. (2005) and Degryse and Ongena (2005) provide evidence that distance between borrowers
and banks provides another proxy for relationships, although Petersen and Rajan (2002) argue technology has
reduced the importance of physical proximity to banks. Bank size has also been associated with relationship
lending, motivated by Stein (2002), who argues that large complex organizations are less able to manage soft
information embedded in lending relationships. The advent of technology has potentially limited the importance of
these dimensions, as discussed in Berger and Black (2019). For a meta-analysis of the effects of relationships on the
terms of bank loans, see Kysucky and Norden (2016).
6
Humphries et al. (2020) use survey data to show that the smallest businesses face an information disadvantage in
accessing PPP.
8
II. BACKGROUND: THE PAYCHECK PROTECTION PROGRAM
The negative economic impact of the coronavirus pandemic became increasingly evident
in March 2020, as the spread of COVID-19 accelerated across the U.S. and states started to
implement various emergency measures, including “lock-downs”. For the week ending on
March 28, initial claims for unemployment insurance (seasonally adjusted) reached a historical
high of 6.9 million. In response to this sharp and deep economic shock, the Congress quickly
passed the Coronavirus Aid, Relief, and Economic Security (CARES) Act, which was signed
into law on March 27th. The CARES Act provides a total of $2.2 trillion in economic assistance
for individuals, healthcare providers, businesses, and state and local governments. The Paycheck
Protection Program (PPP), established under the CARES Act, aimed to help small businesses
“maintain their payroll, hire back employees who may have been laid off, and cover applicable
overhead.” The PPP program began to disburse funds on April 3. Strong initial demand
exhausted the $350 billion allocated within two weeks. An additional $320 billion of funding
was added to the PPP program by Paycheck Protection Program and Health Care Enhancement
Act on April 24. The program closed to new loan applications on August 8, 2020, having
distributed $525 billion (Small Business Administration (2020)), although the majority of funds
had been distributed by the early May (see Figure 3).
The PPP program provides loans to small businesses that are fully forgivable under
certain conditions. Since the program’s main aim is to reduce job separation, borrowers must
maintain their employee and compensation levels to be eligible for forgiveness. 7 Banks
7
PPP loan forgiveness occurs under the following conditions: a) at least 75% of the loan proceeds cover payroll
costs and the rest covers other overhead such as mortgage interest, rents, and utilities; and, b) the borrowers maintain
their employee and compensation levels. See https://home.treasury.gov/policy-issues/cares/assistance-for-small-
businesses for details.
9
distributed most of the PPP loans. Initially, only existing Small Business Administration (SBA)
lenders or federally insured depository institutions, credit unions, and Farm Credit System
institutions could make PPP loans. The set of PPP lenders gradually expanded to include more
non-bank lenders as the SBA approved their applications to participate. As of August 8, when
the PPP loan application period ended, non-bank lenders account for only 8.3% of PPP loan
count and 3.6% of PPP loan amount (Small Business Administration (2020)). 8
With a few exceptions, only businesses with fewer than 500 employees may apply for
PPP loans. 9 Potential borrowers submit applications directly to private PPP lenders, who review
the application materials and fund the loans. All PPP loans have the same terms, with an interest
rate of 1% and a maturity of two years for loans made before June 5 or five years for loans made
on or after June 5. The SBA also pays lenders processing fees up to a limit for originating PPP
loans. Once approved, the SBA guarantees repayment at no cost to the borrowers or lenders; this
guarantee ultimately has backing of the full faith and credit of the U.S. Treasury. Hence, PPP
loans carry a zero-percent risk weight under regulatory capital rules. In addition, the federal
banking regulators (OCC, FDIC, and the Federal Reserve) issued a joint interim final rule on
April 13 that effectively neutralizes the regulatory capital effects of PPP loans that are pledged
by banks to the Paycheck Protection Program Lending Facility (PPPLF). Thus, regulated banks
may originate PPP loans without any credit risk or marginal capital requirement.
8
Those non-bank PPP lenders include small business lending companies, fintechs, non-bank CDFI funds, and other
non-bank lenders.
9
For example, firms in the accommodation and food services industry (with NAICS codes beginning with 72) are
eligible for PPP loans if they employ less than 500 employees per physical location.
10
III. WHO SUPPLIES PPP LOANS?
Data
We construct data on lending at the bank level by combining information from the Call
Reports with information on the PPP program provided publicly by the U.S. Small Business
Administration (SBA). The Call Report data normally capture bank lending to businesses both
on and off the balance sheet (commercial and industrial (C&I) loans and unused loan
commitments to businesses). An additional field added to the 2020, Q2 Call Report also
separates out lending under the PPP program. Thus, we can compare PPP lending with non-PPP,
C&I lending during 2020, Q2. These data capture all loans on bank balance sheets as of June 30,
2020.
The SBA data contain firm-level records of borrowing under the PPP program, with
information on the location of the borrower, the size of the loan, and the name of the lender. We
merge these data into the Call Report data using the name of the lender. This procedure allows
us to match most of the banks exactly. Of the non-matched banks, about 400 made no PPP loans
based on the Call Report data, so we assign zero PPP loans to these banks. Overall, we identify
PPP lending for 4,333 out of 4,980 banks. Collectively, the matched banks cover 95% of the
total PPP lending in the SBA database, so we are confident that our measures accurately
represent the bulk of the program. Some of the residual PPP lending is made by banks which we
could not match, and some by non-bank financial institutions such as Credit Unions and
Community Development Financial Institutions.
In order to understand the role of bank relationships and the importance of local banks,
we separate lending from the SBA data into loans made by banks with branches in the
11
borrower’s county (deemed “core markets”), versus loans made by banks without local branches
(deemed “peripheral markets”). To achieve this separation, we use the location of each bank’s
branches as of June 2019 from the Summary of Deposits dataset. 10 Because the branch locations
are set before the onset of COVID, we can safely assume that the definition of core vs. peripheral
markets is exogenous. About 71% of the total lending made by the matched banks comes from
core markets, and the other 29% from peripheral markets. The mean size of loans is slightly
larger in core markets ($112,145) compared to peripheral markets ($100,321).
Table 1 reports summary statistics for bank-level measures of C&I lending growth, total
C&I credit growth (loans plus unused commitments) and PPP loans. Unlike C&I lending
growth, total credit growth does not reflect variation in credit-line takedown (or repayments).
We scale these and the other bank characteristics by total assets at the end of 2019. We also
include bank characteristics as of Q4, 2019, which we use to explain the 2020 lending. Panel A
reports lending growth during the first and second quarters of 2020; panel B reports summary
statistics for PPP lending by lender size; and, Panel C reports the 2019 pre-COVID bank
characteristics. 11
As shown in Figure 1, during 2020, Q1 C&I lending grew rapidly at the largest banks,
mainly from large increases in credit-line drawdowns during March (Li et al., 2020 and
Chodorow-Reich, et al., 2020). The average bank, however, experienced a much smaller
increase in C&I lending (about 0.23-0.25 percentage points of assets). Business lending grew
10
See https://www.fdic.gov/regulations/resources/call/sod.html.
11
We relate lending in the first and second quarters of 2020 to 2019 bank covariates, but the sample changes slightly
over the two periods. Table 1 reports the summary statistics for the Q2 sample, but the Q1 figures are nearly
identical.
12
much more rapidly in Q2, but only from the effects of the PPP program. Average PPP lending
was 5.7% of assets. In contrast, non-PPP C&I lending shrank by 0.2% of assets. When we
separate lending into core vs. peripheral markets, we see similar amounts of lending. The
average bank lends 3.4% of assets to borrowers in its core markets, while lending 3.1% to
borrowers in peripheral markets. As we will see, however, the emphasis on PPP lending to core
markets is substantially higher for banks with strong local relationships. 12 As shown in Panel B,
PPP intensity per unit of assets is lower for the largest banks (mean=2.1%), compared to
medium-sized (mean=5.5%) or small banks (mean=5.8%).
In our second set of tests, we focus on county-level real outcomes. High frequency data
are available online at the Opportunity Insights Economic Tracker website
(https://tracktherecovery.org); Chetty et al. (2020) describe the data in detail. We focus on two
of these outcomes (total revenue among small firms and total spending), as well as monthly
unemployment rates at county level from the Bureau of Labor Statistics. 13 Total revenues for
small firms come from Womply. The series represents the percent change in net revenue,
calculated each weekday as a seven-day moving average (seasonally adjusted), indexed to
January 4-31, 2020. Small firms are defined as those meeting the SBA’s threshold. 14 Total
spending at county level comes from daily aggregation of consumer spending based on debit and
credit card transactions from Affinity Solutions.
12
The sum of lending to core plus peripheral markets exceeds total PPP lending from the Call Report because the
former measures include lending through early August of 2020, while the Call Report figure only includes PPP
lending through June 30.
13
See https://www.bls.gov/lau/#tables.
14
To be specific, according the Chetty et al. (2020), “For each series, we construct daily values in exactly the same
way that we constructed the consumer spending series. We first take a seven-day moving average, then seasonally
adjust by dividing each calendar date's 2020 value by its corresponding value from 2019. Finally, we index relative
to pre-COVID-19 by dividing the series by its average value over January 4-31.”
13
Table 2 reports summary statistics for the real outcomes. Both small business revenue
and spending grow during the first months of 2020, before the advent of the COVID crisis, and
then fall sharply thereafter. The declines are sharpest in March and April when most states
initiated lockdowns. Chetty et al. (2020) emphasize that the drop in spending is initially high
across the income distribution. After passage of the CARES Act, however, spending increases
sharply in low-income areas (though still below levels before COVID), but much less so in high-
income areas. (Despite these spending patterns, unemployment has more adverse effects in low-
income areas.) As is clear in Table 2, unemployment increases sharply in April, then declines
over the subsequent months. By the end of our sample, however, unemployment still well
exceeds its level at the beginning of 2020.
Business Lending during the COVID Crisis
Tables 3-7 report regressions to explain bank lending during the 2020 COVID crisis. We
first report models of overall C&I lending; second, we subdivide the analysis into PPP vs. non-
PPP, C&I lending; third, we compare PPP lending patterns by bank size; fourth, we further
subdivide PPP lending by market type (core vs. peripheral); and fifth, we report regressions at
the bank-county level.
The Cross Section of Bank Lending
Table 3 reports the first of these tests, comparing the cross section of bank C&I lending
during the first and second quarters of 2020. As outcomes, we use the change in C&I lending on
balance sheet and the change in C&I credit – the sum of lending on balance sheet plus unused
business commitments – both scaled by Q4, 2019 assets.
14
We measure all bank characteristics from the Q4, 2019 Call Reports and June 2019
Summary of Deposits. As such, they are unaffected by the COVID crisis. We separate bank
characteristics into three tiers. First, we consider measures which the prior literature has
associated with bank-firm relationship lending: Unused commitments (undrawn loan
commitments to businesses/assets), Small business loans (C&I loans under $1 million/assets),
and Core deposits (transaction deposits plus insured time deposits/assets). For example, Petersen
and Rajan (1994) is the first study to document the importance of relationships for bank lending
to small firms. Berger and Udell (1995) show, using similar data, show that credit lines (which
generate undrawn loan commitments to business) are more associated with close bank-borrower
relationships than term lending. Norden and Weber (2010) find that credit lines usage help
banks forecast default. Berlin and Mester (1999) show that core deposits help foster relationship
lending by allowing banks to cross subsidize borrowers over the credit cycle. 15 Second, we
include three additional measures of bank balance sheets: the ratio of total C&I lending to assets
(C&I lending), cash plus securities to assets (Liquid assets), and Tier 1 leverage ratio (Tier 1
Capital). C&I lending captures variation in bank overall emphasis on lending to business,
whereas the other two measures capture variation in the financial strength of the bank. Third, we
include two additional bank characteristics: Bank size (the log of total assets) and Bank age (the
log of the deposit-weighted average age of the bank’s branches). Panel C of Table 1 provides
summary statistics for these explanatory variables. The deposit-weighted COVID death rate is
derived from county-level COVID death rates obtained from the Opportunity Insights Economic
15
The theoretical idea that bank deposits provide information to lenders goes back further, to Fama (1985). For
more recent empirical evidence, see Yang (2020).
15
Tracker website within a bank's branch network. All variables, except Bank size, are winsorized
at the 1st and 99th percentiles.
Table 3 shows sharp differences in lending patterns between the first and second quarters
of 2020, particular for the effect of the three relationship characteristics. In Q1, Unused
commitments strongly correlates with C&I lending growth (Column 1). As Li et al. (2020) show,
the largest banks drive this result; these banks face unexpected increases in credit-line takedowns
from their large borrowers who lose access to short-term funding markets and the bond market.
Credit-line takedowns have no effect on total credit originations, however, which explains why
Unused commitments has no significance in Column (2). In Q2, the effect of Unused
commitments increases relative to Q1 (by a factor of eight). In contrast to Q1, both Small
business loans and Core deposits affect lending strongly in Q2. We know that these effects
represent new credit originations, rather than credit-line takedowns, because both C&I lending
growth and C&I credit growth respond similarly. Economic magnitudes are substantial. A one-
standard deviation increase raises C&I lending by 1.7% of assets (Unused commitment), 0.6% of
assets (Core deposits), and 0.5% of assets (Small business loans). For comparison, C&I lending
grew by 5.4% of assets on average during the period. Beyond the relationship variables, we also
find that banks with higher Liquid assets increased lending less than other banks in Q2, as did
larger banks and older banks.
Table 4 focuses on the difference between PPP lending patterns with those of other
unsubsidized C&I lending (for Q2, 2020 only). PPP lending, as this comparison shows, drives
the sharp differences in lending patterns between the first and second quarters. As in Q1, most
of the correlations between bank characteristics and non-PPP, C&I lending are weak. The
exception is Unused commitments, which has a negative impact on non-PPP C&I lending (but no
16
effect on non-PPP total credit). This reflects firms, who had drawn funds during the March
financial-market meltdown, repaying those funds as bond-market access came back online (see
Chodorow-Reich et al. (2020) and Darmouni and Siani (2020)).
In Q2, the three relationship measures have very strong power to explain PPP lending
(and the sign of the coefficient on Unused commitments becomes positive). This is interesting
because PPP loans do not expose banks to credit risk. The U.S. government bears the downside
risk. As such, this finding points to a benefit of banking relationships not emphasized before in
the existing literature: firms with strong bank relationships receive better access to the PPP
credit, probably because many banks have limited capacity to help firms manage the application
process, and this limited capacity is deployed first in the service of the bank’s relationship
borrowers. 16
Table 5 reports estimates of these models separately for small (<$10 billion in assets),
medium-sized ($10 to $50 billion in assets) and large (>$50 billion in assets) banks. This split
further supports the relationship-banking interpretation. First, the largest banks, which focus
much less on small relationship borrowers, lend much less in the PPP program per unit of assets
than other bank (recall Table 1, Panel C). Second, the large banks exhibit little effect of the three
relationship measures on PPP lending. Unused commitments and Small business lending are
both insignificant in the regression. In contrast, both Unused commitments and Small business
lending are very strongly tied to PPP lending for medium-sized and small banks. Core deposits
only exhibits a strong connection to PPP for the small banks.
16
According to press accounts about the PPP, “…the program’s expenses were also high, the big banks said. Bank
of America devoted 10,000 employees to making loans at the program’s peak, Mr. Moynihan said in July, and
expects the next stage of the program — helping companies through the paperwork to have their loans forgiven, if
they qualify — to be complicated and time consuming.” See New York Times, “Despite Billions in Fees, Banks
Predict Meager Profits on P.P.P. Loans,” October 1, 2020.
17
Pinning down the Role of Relationships: Bank Lending by Market and Branch Characteristics
We have shown that bank lending during Q2, 2020 exhibits a strong link to pre-COVID
measures of bank relationships. The links shift sharply between the first (pre-PPP) and second
(post-PPP) quarters; the links in the second quarter are only evident in the PPP lending, not in
other C&I lending; and, the links are strongest for smaller banks. All of these point to an
important role of relationships for the supply of PPP credit. But all of these tests are just cross-
bank correlations. To rule out alternative explanations due to unobserved heterogeneity across
banks, we now compare PPP lending by market type (core vs. peripheral), and then by within-
bank measures of the strength of its relationships with local borrowers.
Table 6 reports regressions of PPP lending in core and peripheral markets separately,
along with the difference between them. The regressions include the same set of explanatory
variables as in Tables 3-5, but we focus only on the effects of the relationship dimensions. We
report each set of regressions first on all banks, and then split based on bank size. Each of the
three measures – Unused commitments, Small business lending, and Core deposits – affects PPP
lending in core markets more than in peripheral ones. In fact, two of the three measures have no
power to explain PPP lending in the peripheral areas. Lending in core markets most likely
reflects bank interactions with their relationship clientele, and measures of the strength of the
relationships only matter in those areas. As in the earlier split by bank size, the effects are
evident in small and medium-sized banks, but not among the largest banks.
Table 6 also reports the same set of regressions, with the dependent variable split based
on first-round versus second-round lending (Panels B and C). This split shows the same
directional effects: more PPP lending by relationship banks. But it also shows much larger
coefficient magnitudes during the first round. The importance to firms of having a close bank
18
relationship thus manifests most strongly in the first round as banks put their closest customers to
the front of then line in applying to the SBA. As we discussed earlier, the PPP funds ran out in
just two weeks, leading to additional legislation to fund the second round (Figure 3).
In Table 7, we change the structure of the data to focus on how banks supply PPP lending
across their core markets. In particular, we estimate regressions with the following structure:
(PPP Lending/Assets)i,j = αi + γj + β1(%Bank Deposits from County)i,j +
β2 (Log Branch Age)i,j + β3(%County Deposits from Bank)i,j +
β4(%Bank CRA Loans in County)i,j + β5(%County CRA Loans from Bank)i,j + ε i,j, (1)
where i represents bank and j represents county. The restructured data disaggregate each bank’s
total PPP lending in its core markets (one outcome in Table 6) into one observation for each
county in which the bank owns at least one branch. As such, summing the outcome in Equation
(1) across all counties for a given bank recovers that bank’s total PPP lending in its core markets.
The lending measures, built from bank-county data collected under the Community
Reinvestment Act (CRA), are only available for banks with more than $1 billion in assets. Thus,
we report the models with and without the two variables representing bank lending.
We remove all county-level variation in Equation (1) with a county fixed effect (γj), and
all bank-level variation with a bank effect (αi). Including the county fixed effect removes
variation related to local demand for PPP loans such as differences in exposure to COVID or
other sources of variation. Including the bank fixed effect removes variation related to bank
heterogeneity. Equation (1) compares PPP lending for the same bank operating in different
counties. This allows us to test whether banks lend more where they raise more deposits or make
more loans (β1 and β4), whether banks lend more where their branches are older (β2), and
19
whether banks lend more where their market share is higher (β3 and β5). If relationships affect
bank supply of PPP credit, we would expect all of these effects to load positively.
Table 7 offers strong support that banks supply more PPP credit in areas where they have
stronger relationships with local borrowers. That is, banks increase their supply of PPP loans in
markets where they have more deposit market share; they increase PPP loans in markets where
their branches are older; and, they increase PPP loans in markets where they lend a greater
percentage of those funds. These results suggest that PPP lending increases more in areas that
are more important to the lender, based on banks’ pre-COVID experience. The effect of branch
age provides very strong evidence for the importance of relationships, as the literature
emphasizes the importance of relationship length in fostering close bank-borrower ties (e.g.,
Petersen and Rajan, 1994). 17
IV. RELATIONSHIP LENDING, PPP LOAN SUPPLY AND REAL EFFECTS
To test how bank supply of PPP loans affects real outcomes, we first estimate the impact
of the size and structure of the local banking sector on the quantity of PPP lending at the county
level. We test how these effects break down between core-market bank lending (banks with
branches in the county) and peripheral-market bank lending, and also how they break down by
time (first v. second rounds of PPP funds). We then report regressions of county-level real
outcomes on PPP credit measures based on the pre-COVID structure of the local banking
system.
County-Level PPP lending
17
In our bank-level regressions, we find the opposite. That is, younger banks make more PPP loans. This result
represents variation across banks. As such, it suggests that new banks without relationship capital are using the PPP
program to find new customers.
20
To document how local bank structure affects overall PPP credit supply, we regress
county-level PPP lending (scaled by the number of establishments with fewer than 500
employees) on two measures of local bank structure (controlling for other county level factors
plausibly related to demand). The regression structure follows:
PPP Loans/Estab.j = β1Branches/Estab.j + β2Predicted PPP Lendingj
+ Demand Controls + εj . (2)
We then sub-divide county PPP lending into four sub-components: based on time (first round
PPP v. second round) and based on core (lenders with branches in the borrower’s county) v.
peripheral providers of credit.
Equation (2) varies across counties (j). The outcome, PPP Loans/Establishment, which
we measure in thousands of dollars per establishment, averages about $70. Branches
/Establishmentj, equal to the total number of bank branches per establishment in county j,
measures total banking capacity before COVID (June 2019), relative to a proxy for the number
of firms eligible for PPP loans. 18 Predicted PPP Lendingj equals the weighted average of each
bank’s predicted bank-level (PPP Lending/Assetsi) in their core markets, with weights equal to
bank i’s share of total deposits in county j from 2019. Using the model from Panel A of Table 6
(columns 4, 7, and 10), PPP Lending/Assetsj is a linear combination of the characteristics of the
banks operating in county j. Since our bank-level model finds strong explanatory power from
relationship characteristics, this variable will be high in areas where relationship lenders hold a
18
We do not have a count of eligible firms, so we use the number of establishments with fewer than 500 employees
as a close proxy. These data come from the 2018 County Business Patterns data, provided by the U.S. Census
Bureau.
21
high percentage of total deposits. Moreover, this variable is predetermined because the
regression in Table 6 includes only bank covariates from the end of 2019 or earlier.
Equation (2) also includes demographic characteristic of the county (log of county
population; the fraction of the population with a college degree or better; the fraction of the
population aged 20 to 44), as well as measures of the strength of the local economy (the log of
median income in the county from 2018 and the unemployment rate from 2019). These data
come from the U.S. Census and USDA. 19 In addition, we include the COVID death rate. 20 These
variables help remove variation due to demand for PPP loans.
Equation (2) achieves two objectives. First, it allows us to assess whether or not the size
of the local banking sector affects PPP credit supply (as opposed to PPP funds flowing
frictionlessly across geographies). Second, we can assess the importance of not just the size but
also the structure of local banks. As we have seen, small banks, banks focusing on small
business lending, and banks with high levels of unused business loan commitments originate
more PPP loans; Predicted PPP Lending captures these effects. To summarize, if areas with
more local bank branches receive greater supply of PPP loans, then β1 > 0. If the types of banks
matters, then β2 > 0.
We use Equation (2) as a first-stage regression to show that variation in predetermined
measures of banking structure affects the quantity of PPP credit. To document that these are
valid measures of supply (rather than demand), we split the outcome based on PPP lending from
banks locating in the county (core-market banks) v. PPP lending by banks not located in the
19
County level education, unemployment, and median household income data are from
https://www.ers.usda.gov/data-products/county-level-data-sets/download-data/ and all other demographics data from
https://www.census.gov/data/datasets/time-series/demo/popest/2010s-counties-detail.html.
20
Data are obtained from the Opportunity Insights Economic Tracker website (https://tracktherecovery.org).
22
county (peripheral-market banks). Lending from peripheral-market banks will increase with PPP
demand. Hence, if there is no positive effect of Branches/Establishment or Predicted PPP
Lending on lending by peripheral banks, then we can rule out a spurious correlation with PPP
demand.
Column (1) of Table 8 reports estimates of Equation (2). We find strong evidence that
areas with more local bank branches receive more PPP loans; i.e., β1 > 0 (significant at 1% level).
The magnitude is also substantial. A one-standard deviation increase in Branches/Establishment
(=0.02), for example, is associated with an increase in PPP lending of about $18,000 per
establishment (=0.02 x 888.6 x 1000), or a little less than 30% of the mean (=~$70,000).
Predicted PPP Lending has a positive but insignificant coefficient, however.
Columns (2)-(5) of Table 8 separate the outcome between the first and second rounds of
PPP lending (recall Figure 3) and between core v. peripheral markets. This four-way split
suggests that both the size of the local banking system (Branches/Establishment) and its structure
(Predicted PPP Lending) increase the quantity of PPP loans in the first round (Column 2).
Columns (3)-(5) help us assess the (identification) claim that the banking structure variables
capture PPP supply. In contrast to Predicted PPP Lending, Branches/Establishment covaries
positively with PPP loans from peripheral lenders in both the first and second rounds. This
suggests a critical identification concern: areas with more bank branches (pre-COVID) contain
more firms reliant on bank credit (i.e., greater demand for PPP loans). Hence, the number of
bank branches cannot be used to assess PPP credit supply. In contrast, Predicted PPP Lending
loads negatively in Columns (3)-(5), meaning that some firms, unable to borrow locally in the
first round (because their market contains too few relationship lenders), were able to get credit
from peripheral banks. Thus, Predicted PPP Lending predicts early access to the PPP program
23
(Column (2). Borrowers in areas with higher Predicted PPP Lending move to the ‘front of the
line’ and receive credit early from their relationship lender.
Small Business Revenues, Local Spending, and Unemployment
To test whether PPP credit supply from relationship lenders has real effects, we model
three outcomes: daily Small business revenue; daily Local spending; and, monthly
Unemployment rate. We construct panel data at the county-time level, from the beginning of
2020 through July. This sample incorporates a pre-COVID period, a period of transition in
which the effects of COVID pushed financial markets into turmoil (most of March), and a period
in which policy steps from the CARES Act had gone into effect (April and the subsequent
months). Our model tests how measures of local demographics, local pre-COVID economic
variables, and PPP credit supply affect real outcomes across these shifting periods. Specifically,
we estimate the following models:
Yj,t = αt,s + γj + Σβk (Itk x Predicted PPP Lendingj)
+ Σγk (Itk x Demo. & Econ. controlsj) + εj,t. (3)
In Equation (3), j represents county and t represents time (either day or month). Yj,t represents
each of our three outcomes. We remove aggregate shocks with the state-time effect (αt,s) and all
cross-county heterogeneity with the county effect (γj). All county characteristics, including
Predicted PPP Lending, vary only in the cross section, so the county effect absorbs their direct
impact on outcomes.
We focus on how Predicted PPP Lending affects outcomes as the COVID crisis emerges,
and then as the policy actions come online. In addition, we control for other county-level
24
variables to absorb as much variation in demand for PPP credit as well as the effects of the
COVID crisis on outcomes. Hence, we interact each county characteristic with Itk, defined as
five monthly indicators. 21 The first indicator (March) equals one prior to beginning of the PPP
program but after the onset of the COVID crisis. 22 The last four indicators correspond to
calendar months during the period in which the PPP program distributes funds. The coefficients
βk test for systematic shifts in the impact of Predicted PPP Lending (and the other county
characteristics) during these one-month periods, relative to the omitted period (January and
February). If the relationship lenders help firms, then βk ought to be positive for both Small
business revenue and Local spending during April-July, and negative for the Unemployment rate.
The effects measured during the Crisis period represent “placebo” tests because the PPP program
had not yet come online then. This helps us assess the plausibility of the model. 23
Equation (3) represents a reduced form, which directly links a supply instrument
(Predicted PPP Lending) to the real outcomes. We could have structured this analysis as on
instrumental variables model, with PPP Loans/Establishment modeled as the endogenous
regressor. However, if the structure of the local banking system affects economic outcomes in
ways that go beyond the PPP program, then these instruments might fail the exclusion restriction.
While this may be true, and while additional effects cannot be fully ruled out, lending patterns
suggests otherwise. First, the PPP program dominates new credit originated during the April-
July 2020. For example, non-PPP C&I lending fell during the second quarter of 2020 (Table 1).
21
The PPP program closed to new applications on August 8, just a few days after the end of our sample, so we end
our analysis in July.
22
The World Health Organization declared as a global pandemic on March 11, 2020.
23
We double cluster by time (or state-time in the case of the unemployment rate, due to the small number of time
units) and county to construct standard errors.
25
Second, as we will see, the time series pattern of the effects of banking line up with the timing of
the PPP program.
Table 9 reports the estimates of Equation (3). We find little impact of Predicted PPP
lending on either Small business revenues or Total Spending. Sign patterns are positive,
consistent with benefits, but with minimal joint statistical significance (F-statistics around 1 for
the April-July coefficients). In contrast, we find very strong statistical evidence that
Unemployment rate is lower in markets with more relationship lenders (i.e., higher Predicted
PPP lending). Coefficients become strongly negative in May, and they increase in magnitude in
June and July, with high joint statistical significance (F-statistic = 8.13). The time patterns of
our estimates in the reduced form are consistent with our core arguments. That is, there is no
evidence of “pre-trends” in the data. The coefficient on Predicted PPP loans is small and not
statistically significant in March 2020, as they should be because the PPP program had yet to
begin. 24
While counties with higher Predicted PPP lending do experience less unemployment
after the advent of the PPP program, the economic impact is small. For example, increasing
Predicted PPP lending by one standard deviation (= 0.014) lowers the unemployment rate in
June by just 0.21 percentage points (=-0.014x15.2); this decline is dwarfed by the average
increase in unemployment of 7.6 percentage points between March and April. Moreover, we
find no broader benefits, either to small businesses themselves (in terms of sales or revenues), or
in terms of total spending. The small economic magnitude likely reflects the fact that the second
24
March 2020 represents a fairly strong placebo period because banks were important during that time in alleviating
the stresses on large firms due to the disruptions in the money markets and the bond market. Thus, if we observe no
effects on local outcomes before April, this supports the claim that the effects observed after April in fact stem from
the PPP program.
26
round of PPP funding was sufficiently generous that all firms demanding these loans eventually
received them. That said, the evidence does suggest that relationship lending, by giving firms
early access to the PPP program, did help them avoid laying off workers.
V. CONCLUSION
This paper analyzes the role of banks as the primary conduit of funds for the Paycheck
Protection Program. We find that PPP lending by banks increases with traditional measures of
relationship lending: larger for small banks, increasing in prior experience in the local market,
increasing in commitment lending, and increasing in core deposits. The traditional rationale for
bank relationships – access to soft information, which mitigates asymmetric information
problems – cannot explain our findings because banks face no credit risk in making PPP loans.
Thus, our results suggest a new benefit to firms of close ties to their banks, which are often the
key conduit for access to government subsidies. Using our model of bank-level lending, we build
a local supply measure that reflect the structure of the banking systems. We find that increases
in this predicted PPP lending, which reflects the presence of relationship-oriented banks prior to
COVID, lowers local unemployment.
Our results point to an inefficiency in the distribution of PPP funds, as firms with banking
relationships receive earlier access to credit, irrespective of merit. Distributing the $525 billion
in funds so quickly – just one month – could only be achieved using the human capital employed
by the banking system. Government interventions or bailouts have often historically worked
through the banking system. We have seen this not only with the PPP program, but in other ways
in which the Federal Reserve intervened during the COVID crisis, in its interventions during the
2008 Global Financial Crisis, and also in its actions to stem bond-market disruptions during
1998. Why do government interventions work though banks, rather than just helping whatever
27
economic entity is most distressed? The answer may reflect an unpleasant tradeoff between the
short-term benefits of interventions (e.g., ending a financial panic) vs. the longer run moral
hazard costs. Using the banking system as the conduit may be a way to limit the scope of
interventions and thus limit the associated moral hazard costs.
28
REFERENCES
Acharya, V. V., and S. Steffen. 2020, “The risk of being a fallen angel and the corporate dash for
cash in the midst of COVID,” forthcoming in the Review of Corporate Finance Studies.
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,” MIT Working
paper.
Amiram, Dan and Rabetti, Daniel, “The Relevance of Relationship Lending in Times of Crisis,”
(September 2020). Available at SSRN: https://ssrn.com/abstract=3701587.
Berger, Allen and Udell, Gregory, 1995 “Relationship lending and lines of credit in small firm
finance,” Journal of Business 68, 351-82.
Berger, Allen, Nathan Miller, Mitchell Petersen, Raghuram Rajan and Jeremy Stein, 2005, “Does
Function follow Form? Evidence from the Lending Practices of Large and Small Banks,”
Journal of Financial Economics 76(2), 237-69.
Berger, Allen N. and Lamont K. Black, 2019, “Small Business Lending: The Roles of
Technology and Regulation from Pre-Crisis to Crisis to Recovery,” in Oxford Handbook of
Banking, Third Edition, Allen N. Berger, Philip Molyneux, and John O.S. Wilson, editors,
Oxford University Press, Oxford, (2019) 431-469.
Balyuk, Tatyana, Nagpurnanand R. Prabhala & Manju Puri, 2020, “Intermediary Supply Effects:
Lessons From the Paycheck Protection Program,” NBER Working Paper no. 28114.
Bartik, Alexander, Cullen, Zoe, Glaeser, Edward L., Luca, Michael, Stanton, Christopher and
Sunderam, Aditya, 2020, “The Targeting and Impact of Paycheck Protection Program Loans to
Small Businesses,” NBER Working Paper No. w27623.
Barraza, Santiago, Rossi, Martin and Yeager, Timothy J., 2020, “The Short-Term Effect of the
Paycheck Protection Program on Unemployment,” Available at SSRN:
https://ssrn.com/abstract=3667431.
Berlin, Mitchell and Loretta J. Mester, 1999, “Deposits and Relationship Lending,” Review of
Financial Studies 12(3), 579-607.
Bolton, Patrick, Xavier Freixas, Leonardo Gambacorta, Paolo Emilio Mistrulli, 2016,
“Relationship and Transaction Lending in a Crisis,” Review of Financial Studies 29(10), 2643–
2676.
Chodorow-Reich, Gabrial, Olivier Darmouni, Stephan Luck and Matthew Plosser, 2020, “Bank
Liquidity Provision Across the Firm Size Distribution,” Available at SSRN:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3702725.
29
Core, Fabrizio and Felippo De Marco, 2020, “Public Guarantees for Small Businesses in Italy
during COVID-19,” Available at SSRN:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3604114.
Cororaton, Anna and Rosen, Samuel, 2020, “Public Firm Borrowers of the US Paycheck
Protection Program,” SMU Cox School of Business Research Paper No. 20-01, Available at
SSRN: https://ssrn.com/abstract=3590913.
Darmouni, Olivier and Kerry Siani, 2020, “Crowding Out Bank Loans: Liquidity-Driven Bond
Issuance,” Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3693282.
Degryse, H., Ongena, S., 2005, “Distance, lending relationships, and competition,” Journal of
Finance, 60 (1), 231-266
Erel, Isil and Liebersohn, Jack, 2020, “Does FinTech Substitute for Banks? Evidence from the
Paycheck Protection Program,” NBER Working Paper no. 27659.
Fama, Eugene, 1985, “What’s Different about Bank?” Journal of Monetary Economics 15(1),
29-39.
Faulkender, Michael, Robert Jackman and Stephen Miran, 2020, “The Job-Preservation Effects
of Paycheck Protection Program Loans,” US Treasury, Office of Economic Policy Working
Paper 2020-01.
Granja, Joao and Makridis, Christos and Yannelis, Constantine and Zwick, Eric, 2020, “Did the
Paycheck Protection Program Hit the Target?” Available at SSRN:
https://ssrn.com/abstract=3585258.
Greenwald, Daniel, John Krainer, and Pascal Paul, 2020, “The Credit Line Channel,” San
Francisco Fed Working Paper 2020-26.
Hotchkiss, Edith, Greg Nini, and David C. Smith, 2020, “Corporate Capital Raising During the
COVID Crisis,” https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3723001.
James, Christopher M. and Lu, Jing and Sun, Yangfan, 2020, “Time Is Money: Relationship
Lending and the Role of Community Banks in the Paycheck Protection Program.” Available at
SSRN: https://ssrn.com/abstract=3633339.
Joaquim, Gustavo and Netto, Felipe, 2020, “Bank Incentives and the Impact of the Paycheck
Protection Program.” Available at SSRN: https://ssrn.com/abstract=3704518.
Kysucky, V. and Lars Norden, 2016, “The Benefits of Relationship Lending in a Cross-Country
Context: A Meta-Analysis,” Management Science 62, 90–110.
Humphries, John Eric and Neilson, Christopher and Ulyssea, Gabriel, 2020, “Information
Frictions and Access to the Paycheck Protection Program,” Cowles Foundation Discussion Paper
2247, NYU Stern School of Business, Available at SSRN: https://ssrn.com/abstract=3667636.
Li, Lei, Strahan, Philip E. and Zhang, Song, 2020, “Banks as Lenders of First Resort: Evidence
from the COVID-19 Crisis,” forthcoming at the Review of Corporate Finance Studies.
Mian, Atif and Amir Sufi, 2014, House of Debt, University of Chicago Press.
30
Peterson, Mitchell, and Raghuram Rajan, 1994. “The benefits of firm-creditor relationships:
Evidence from small-business data,” Journal of Finance, Vol. 49, 3-37.
Peterson, Mitchell, and Raghuram Rajan, 2002. “Does Distance still Matter,” Journal of
Finance, Vol. 57, 2533-2570.
Stein, Jeremy, 2002, “Information Production and Capital Allocation: Hierarchical vs.
Decentralized Firms,” Journal of Finance 57, 1891-1921.
Yang, Jun, 2020, “Deposit-Lending Synergies: Evidence from Chinese Students at US
Universities,” Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3281397.
31
Figure 1: Cumulative Commercial and Industrial Lending
Growth
45%
40%
35%
30%
25%
20%
15%
10%
5%
0% 2020-01-08
-5%
2020-01-15
2020-01-22
2020-01-29
2020-02-05
2020-02-12
2020-02-19
2020-02-26
2020-03-04
2020-03-11
2020-03-18
2020-03-25
2020-04-01
2020-04-08
2020-04-15
2020-04-22
2020-04-29
2020-05-06
2020-05-13
2020-05-20
2020-05-27
2020-06-03
2020-06-10
2020-06-17
2020-06-24
2020-07-01
2020-07-08
2020-07-15
2020-07-22
2020-07-29
2020-08-05
2020-08-12
2020-08-19
2020-08-26
2020-09-02
Large Banks Small Banks
This figure plots the cumulative commercial and industrial loan growth since January 2020 at large and small banks
in the U.S. The data come from the Federal Reserve’s H.8 weekly statistical release. Large banks are the top 25
domestically chartered commercial banks by assets and small banks include the rest of the banks.
32
Panel A of Figure 2 plots the heat map of state-level total PPP loan amount (in dollars) received during the first
round of PPP (April 3-16) scaled by the number of establishments with less than 500 employees in 2018. Panel B of
Figure 2 plots the heat map of state-level total PPP loan amount (in dollars) received during the second round of PPP
(April 27-August 9) scaled by the number of establishments with less than 500 employees in 2018. Panel C of
Figure 2 plots the heat map of the share of bank branches owned by small banks (with less than $10 billion in assets)
in 2019.
Figure 2, Panel A:
33
Figure 2, Panel B
34
Figure 2, Panel C
35
Figure 3: Daily approved PPP loan amount
60
Billions
50
40
30
20
10
-
This figure plots the daily approved PPP loan amount. The data come from the Small Business Administration’s
(SBA) PPP loan level dataset.
36
Table 1: Summary Statistics for Bank Characteristics
This table reports summary statistics for bank C&I lending and PPP lending, along with other balance sheet characteristics, from Call Reports. The changes in lending are
normalized by Q4, 2019 assets. C&I credit equals the sum of C&I loans on balance sheet plus unused C&I commitments. Core markets are counties in which the bank owns at
least one branch and peripheral markets are the other counties. log(Deposit-weighted average branch age) is derived from the 2019 Summary of Deposits data. Deposit-weighted
COVID death rate is derived from county-level COVID death rates, which are the number of COVID deaths per 100,000 people, within a bank's branch network. All other
variables come from the pre-COVID period (Q4, 2019) Call Reports. All variables, except log(Assets), are winsorized at the 1st and 99th percentiles.
Panel A: 2020 Lending
Q1 N mean sd p25 p50 p75
Change C&I loans/Assets 5,031 0.002 0.012 -0.002 0.001 0.005
Change C&I credit/Assets 5,031 0.003 0.014 -0.002 0.001 0.006
Q2
Change C&I loans/Assets 4,980 0.054 0.067 0.012 0.037 0.070
Change C&I credit/Assets 4,980 0.058 0.070 0.014 0.040 0.076
Change non-PPP C&I loans/Assets 4,980 -0.002 0.016 -0.009 -0.002 0.002
Change non-PPP C&I credit/Assets 4,980 0.002 0.019 -0.006 0.000 0.007
PPP loans/Assets 4,980 0.057 0.069 0.013 0.040 0.076
Banks Matched to SBL Data
PPP loan in Core Markets/Assets 4,333 0.034 0.037 0.006 0.023 0.049
PPP loan in Peripheral Markets/Assets 4,333 0.031 0.058 0.003 0.012 0.030
Panel B: PPP Loans/Assets, by Bank Size
N mean sd p25 p50 p75
Large (>$50 billion) 43 0.021 0.023 0.002 0.014 0.034
Medium ($10 to $50 billion) 90 0.055 0.050 0.025 0.055 0.072
Small (Less than $10 billion) 4,847 0.058 0.069 0.013 0.040 0.076
Panel C: Bank 2019 Characteristics (for the Q2, 2020 sample)
N mean sd p25 p50 p75
Unused C&I Commitments/Assets 4,980 0.031 0.035 0.007 0.021 0.043
Small C&I Loans/Assets 4,980 0.047 0.037 0.021 0.041 0.065
Core Deposits/Assets 4,980 0.780 0.074 0.741 0.791 0.834
C&I loans/Assets 4,980 0.083 0.065 0.040 0.069 0.111
Liquid Assets/Assets 4,980 0.295 0.157 0.178 0.256 0.381
Tier 1 Leverage Ratio 4,980 0.118 0.035 0.097 0.109 0.128
log(Assets) 4,980 12.565 1.468 11.614 12.377 13.260
log(Deposit-weighted average branch age) 4,980 3.990 0.676 3.612 4.152 4.511
Deposit-weighted COVID death rate 4,980 23.580 34.052 2.854 9.400 27.762
37
Table 2 Summary Statistics for County-Level Economic Outcomes
This table reports summary statistics for the data used in the county-time panel regressions of real economic outcomes on measures of PPP credit. Small Business Revenue and
Total Spending come from https://tracktherecovery.org; see Chetty et al. (2020). Unemployment Rate is from the Bureau of Labor Statistics. Small Business Revenue represents
the percent change in net revenue, calculated each week day as a seven-day moving average (seasonally adjusted), indexed to January 4-31, 2020. Total Spending is the
seasonally adjusted credit/debit card spending relative to January 4-31 2020. Unemployment Rate is by calendar month.
Small Business Revenue Total Spending Unemployment Rate
Standard Standard Standard
Time Period Description Mean Deviation Mean Deviation Mean Deviation
1/1/2020 - 2/15/2020 COVID affects Asia 0.028 0.162 0.002 0.078 4.635 1.975
Pre-COVID (2/16/2020 - 3/10/2020) COVID affects Western Europe 0.054 0.192 0.007 0.105 4.356 1.895
Crisis (3/11/2020 - 4/3/2020) Global Pandemic declared; Financial Market turmoil -0.150 0.269 -0.155 0.163 4.811 2.048
April (4/4/2020-4/30/2020) Beginning of PPP Program -0.229 0.289 -0.241 0.130 12.450 5.238
May Continuation of PPP Program -0.061 0.304 -0.125 0.138 10.348 4.117
June Continuation of PPP Program -0.028 0.294 -0.066 0.137 8.455 3.312
July Continuation of PPP Program -0.061 0.304 -0.050 0.124 7.878 3.082
August PPP Program closed to new applications on August 8, 2020
38
Table 3: Bank Business Lending in the First Two Quarters of 2020
This table reports regressions of bank C&I lending in the first two quarters of 2020 on pre-COVID balance sheet characteristics, from Call
Reports. The changes in lending are normalized by Q4, 2019 assets. C&I credit equals the sum of C&I loans on balance sheet plus unused
C&I commitments. log(Deposit-weighted average branch age) is derived from the 2019 Summary of Deposits data. Deposit-weighted
COVID death rate is derived from county-level COVID death rates within a bank's branch network. All other variables come from the pre-
COVID period (Q4, 2019) Call Reports. Stardard errors are clustered by bank headquarter state. T-statistics are reported in parentheses. '*'
denotes significance at the 10% level, '**' the 5% level, and '***' the 1% level.
Q1, 2020 Q2, 2020
Change C&I Change C&I Change C&I Change C&I
loans/Assets credit/Assets loans/Assets credit/Assets
(1) (2) (3) (4)
Relationship Measures
Unused C&I comm./Assets 0.0600*** 0.00572 0.423*** 0.507***
(7.120) (0.452) (7.217) (7.956)
Small C&I loans/Assets -0.0117 6.36e-05 0.131** 0.111*
(1.019) (0.00522) (2.094) (1.696)
Core deposits/Assets 0.00261 0.00390 0.0763*** 0.0836***
(1.172) (1.521) (4.310) (4.628)
Other Balance Sheet Ratios
C&I loans/Assets 0.0179** 0.0230** 0.0505 0.0790*
(2.149) (2.389) (1.116) (1.684)
Liquid assets/Assets -2.38e-05 -0.00141 -0.0315*** -0.0331***
(0.0173) (0.912) (4.388) (4.514)
Tier 1 leverage ratio 0.0256** 0.0296** 0.0470 0.0553
(2.402) (2.357) (0.982) (1.070)
Other Control Variables
log(Assets) 0.000402** 0.000192 -0.00436*** -0.00409***
(2.559) (1.326) (3.973) (3.695)
log(Deposit-weighted average branch age) -0.000371 -0.000612 -0.0317*** -0.0324***
(1.112) (1.281) (10.06) (10.20)
Deposit-weighted COVID death rate 9.73e-05* 2.23e-05 3.51e-05 3.42e-05
(1.684) (0.462) (1.002) (0.987)
Constant -0.00917*** -0.00567 0.154*** 0.148***
(2.865) (1.460) (7.298) (6.900)
Observations 5,031 5,031 4,980 4,980
R-squared 0.074 0.020 0.235 0.264
39
Table 4: Bank Lending in the Second Quarter of 2020, PPP v. non-PPP C&I Lending
This table reports regressions comparing bank non-PPP C&I lending v. PPP lending in the second quarter of 2020
on pre-COVID balance sheet characteristics, from Call Reports. The changes in lending are normalized by Q4,
2019 assets. C&I credit equals the sum of C&I loans on balance sheet plus unused C&I commitments.
log(Deposit-weighted average branch age) is derived from the 2019 Summary of Deposits data. Deposit-weighted
COVID death rate is derived from county-level COVID death rates within a bank's branch network. All other
variables come from the pre-COVID period (Q4, 2019) Call Reports. Stardard errors are clustered by bank
headquarter state. T-statistics are reported in parentheses. '*' denotes significance at the 10% level, '**' the 5%
level, and '***' the 1% level.
Change non-PPP Change non-PPP C&I
C&I loans/Assets credit/Assets PPP loans/Assets
(1) (2) (3)
Relationship Measures
Unused C&I comm./Assets -0.0778*** 0.000111 0.494***
(8.225) (0.00721) (7.912)
Small C&I loans/Assets 0.00303 -0.0140 0.127**
(0.219) (1.024) (2.168)
Core deposits/Assets -0.00412 0.00148 0.0809***
(1.502) (0.526) (4.514)
Other Balance Sheet Ratios
C&I loans/Assets -0.0229*** 0.00507 0.0716
(3.616) (0.700) (1.573)
Liquid assets/Assets -0.000192 -0.00117 -0.0320***
(0.132) (0.936) (4.146)
Tier 1 leverage ratio 0.00222 0.00878 0.0427
(0.375) (1.016) (0.897)
Other Control Variables
log(Assets) -0.00108*** -0.000728*** -0.00309***
(5.601) (3.516) (2.685)
log(Deposit-weighted average branch age) -0.000574* -0.00128*** -0.0307***
(1.851) (3.136) (9.094)
Deposit-weighted COVID death rate 8.33e-06 6.15e-06 2.36e-05
(1.594) (1.006) (0.659)
Constant 0.0188*** 0.0125*** 0.132***
(4.736) (2.699) (5.782)
Observations 4,980 4,980 4,980
R-squared 0.123 0.006 0.262
40
Table 5: Bank PPP Lending, by Size
This table reports regressions of bank PPP lending, normalized by Q4, 2019 assets, in the second quarter of 2020 on
pre-COVID balance sheet characteristics, from Call Reports. Large, medium-sized, and small banks are those with
>$50 billion, 10-50 billion, and <$10 billions in assets, respectively. log(Deposit-weighted average branch age) is
derived from the 2019 Summary of Depositsdata. Deposit-weighted COVID death rate is derived from county-level
COVID death rates within a bank's branch network. All other variables come from the pre-COVID period (Q4, 2019)
Call Report. Stardard errors are clustered by bank headquarter state. T-statistics are reported in parentheses. '*'
denotes significance at the 10% level, '**' the 5% level, and '***' the 1% level.
Large Banks Medium Sized Banks Small Banks
Relationship Measures (1) (2) (3)
Unused C&I comm./Assets 0.0466 0.601*** 0.516***
(0.876) (3.914) (7.965)
Small C&I loans/Assets -0.0549 0.570*** 0.119**
(1.037) (2.722) (2.077)
Core deposits/Assets 0.0437* 0.0319 0.0815***
(1.884) (0.678) (4.429)
Other Balance Sheet Ratios
C&I loans/Assets 0.104** -0.212*** 0.0827*
(2.628) (2.919) (1.761)
Liquid assets/Assets -0.0391** -0.0591* -0.0246***
(2.488) (1.745) (3.103)
Tier 1 leverage ratio -0.359* -0.0998 0.0514
(2.101) (0.495) (1.023)
Other Control Variables
log(Assets) -0.00736** -0.0320** 0.000215
(2.530) (2.254) (0.163)
log(Deposit-weighted average branch age) 0.00598 -0.0204 -0.0293***
(1.053) (1.076) (8.377)
Deposit-weighted COVID death rate -3.88e-05 0.000163 1.73e-05
(1.260) (0.796) (0.494)
Constant 0.137** 0.633* 0.0813***
(2.820) (1.907) (3.173)
Observations 43 90 4,847
R-squared 0.609 0.322 0.274
41
Table 6: Bank PPP Lending, Relationship Measures in Core v. Peripheral Markets
This table reports regressions of bank PPP lending on pre-COVID balance sheet characteristics, separated into markets with and without bank branches. PPP lending data are taken from the Small Business
Administration, and bank characteristics are from Call Reports. Core Markets are counties in which the bank owns at least one branch; Peripheral Markets are counties in which the bank owns no branches. Branch
locations are determined from the 2019 Summary of Deposits data. The dependent variables are PPP lending in core and peripheral markets, normalized by Q4, 2019 assets, and their difference. Round 1 PPP
lendings include PPP loans approved before April 17 and Round 2 include loans approved after April 17. Large, medium-sized, and small banks are those with >$50 billion, 10-50 billion, and <$10 billions in assets,
respectively. Regressions include (but don't report) the other control variables from Tables 3-5. Stardard errors are clustered by bank headquarter state. T-statistics are reported in parentheses. '*' denotes significance
at the 10% level, '**' the 5% level, and '***' the 1% level.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
All banks Large banks Mid-sized banks Small banks
Peripheral Core Peripheral Peripheral Peripheral
Panel A: All PPP Core Markets Markets Difference Markets Markets Difference Core Markets Markets Difference Core Markets Markets Difference
Unused C&I comm./Assets 0.367*** 0.147*** 0.221*** 0.0227 -0.0145 0.0372 0.390*** 0.145 0.245* 0.384*** 0.157*** 0.227***
(9.696) (3.494) (3.935) (0.337) (0.835) (0.607) (5.753) (1.463) (1.789) (9.665) (3.495) (3.877)
Small C&I loans/Assets 0.143*** 0.0289 0.115** -0.0806 -0.0242 -0.0563 0.699*** 0.0122 0.687** 0.149*** 0.0163 0.133***
(3.978) (0.696) (2.484) (1.065) (0.906) (0.836) (3.759) (0.0715) (2.378) (4.017) (0.381) (2.684)
Core deposits/Assets 0.0604*** 0.00386 0.0566*** 0.0626 0.00322 0.0594 0.0222 -0.00653 0.0288 0.0592*** 0.00451 0.0547***
(6.351) (0.259) (4.127) (1.644) (0.494) (1.744) (0.732) (0.316) (0.836) (6.360) (0.291) (3.810)
Observations 4,333 4,333 4,333 40 40 40 81 81 81 4,212 4,212 4,212
R-squared 0.339 0.076 0.058 0.514 0.584 0.423 0.607 0.161 0.328 0.368 0.074 0.061
All banks Large banks Mid-sized banks Small banks
Peripheral Core Peripheral Peripheral Peripheral
Panel B: First Round Core Markets Markets Difference Markets Markets Difference Core Markets Markets Difference Core Markets Markets Difference
Unused C&I comm./Assets 0.296*** 0.121*** 0.174*** 0.0102 -0.0133 0.0234 0.317*** 0.0459 0.271*** 0.312*** 0.130*** 0.181***
(10.47) (4.735) (4.959) (0.191) (1.001) (0.488) (4.801) (1.532) (3.597) (10.45) (4.717) (4.892)
Small C&I loans/Assets 0.114*** 0.0495** 0.0642** -0.0397 -0.0389* -0.000766 0.428*** 0.0853 0.342** 0.118*** 0.0423* 0.0758**
(3.864) (2.084) (2.082) (0.571) (1.984) (0.0122) (2.896) (1.201) (2.237) (3.899) (1.795) (2.291)
Core deposits/Assets 0.0459*** -0.00107 0.0469*** 0.0493** 0.00195 0.0474** 0.0190 -0.00691 0.0259 0.0449*** -0.000949 0.0459***
(5.617) (0.107) (5.464) (2.204) (0.486) (2.276) (0.796) (0.593) (1.212) (5.622) (0.0922) (5.197)
Observations 4,206 4,206 4,206 40 40 40 80 80 80 4,086 4,086 4,086
R-squared 0.334 0.084 0.071 0.461 0.554 0.384 0.544 0.184 0.477 0.370 0.084 0.077
All banks Large banks Mid-sized banks Small banks
Peripheral Core Peripheral Peripheral Peripheral
Panel C: Second Round Core Markets Markets Difference Markets Markets Difference Core Markets Markets Difference Core Markets Markets Difference
Unused C&I comm./Assets 0.0608*** 0.0201 0.0408** 0.0129 -0.00119 0.0141 0.0693** 0.0557 0.0136 0.0605*** 0.0203 0.0402**
(5.214) (1.329) (2.245) (0.417) (0.175) (0.509) (2.055) (1.243) (0.217) (5.017) (1.262) (2.149)
Small C&I loans/Assets 0.0317*** -0.00460 0.0363** -0.0403 0.0143 -0.0545 0.247*** -0.0197 0.267** 0.0338*** -0.00909 0.0429***
(3.469) (0.280) (2.386) (0.771) (0.945) (1.277) (2.804) (0.258) (2.064) (3.657) (0.531) (2.715)
Core deposits/Assets 0.0162*** 0.00271 0.0135** 0.0134 0.00132 0.0121 0.00365 -0.000912 0.00456 0.0160*** 0.00315 0.0129**
(6.320) (0.521) (2.659) (0.780) (0.439) (0.818) (0.281) (0.0942) (0.298) (6.349) (0.578) (2.423)
Observations 4,329 4,329 4,329 40 40 40 81 81 81 4,208 4,208 4,208
R-squared 0.216 0.061 0.038 0.395 0.536 0.362 0.293 0.164 0.176 0.227 0.059 0.038
42
Table 7: PPP Lending, Within-Bank Tests
This table reports bank-county level regressions of PPP lending on bank-county characteristics. The dependent variable is log(1+PPP lending amount) by the bank in the
county. We include all banks merged to the SBA PPP data, plus all banks identified as having zero PPP lending from the bank Call Reports. For each bank, we inlude
all counties in which the bank owns at least one branch. Deposit characteristics are taken from the 2019 Summary of Deposits. CRA characteristics are taken from the
2018 CRA data. Regressions with CRA data include only banks with assets over $1 billion. Large, medium-sized, and small banks are those with >$50 billion, 10-50
billion, and <$10 billions in assets, respectively. Stardard errors are clustered by bank and county. T-statistics are reported in parentheses. '*' denotes significance at the
10% level, '**' the 5% level, and '***' the 1% level.
All banks Large banks Mid-sized banks Small banks
(1) (2) (3) (4) (5) (6) (7) (8)
%Bank Deposits from the county 0.574*** 0.0753 -1.106 -1.534* -0.0706 -0.314 0.801*** 0.0184
(5.933) (0.478) (0.906) (2.019) (0.173) (0.662) (5.226) (0.0605)
log(deposit-weighted average branch age) 0.0876*** 0.172*** 0.245*** 0.242*** 0.110** 0.0926** 0.0510** 0.152**
(4.110) (5.200) (4.599) (5.266) (2.542) (2.281) (2.120) (2.326)
%County Deposits held by the bank 5.796*** 3.888*** 5.271*** 3.343*** 5.762*** 3.313*** 5.032*** 6.394***
(19.43) (11.48) (11.23) (6.683) (9.682) (5.905) (7.353) (4.699)
%Bank CRA Loans from the county 3.832** 6.798 5.385*** 8.864***
(2.451) (1.236) (3.099) (3.077)
%County CRA Loans held by the bank 3.976*** 4.264*** 3.500*** 1.386*
(10.23) (8.947) (5.952) (1.651)
Observations 22,830 13,342 4,922 4,922 2,618 2,593 13,423 3,882
R-squared 0.966 0.954 0.932 0.935 0.985 0.985 0.972 0.961
Bank FEs YES YES YES YES YES YES YES YES
County FEs YES YES YES YES YES YES YES YES
43
Table 8: County Level PPP Lending
This table reports county level regressions of PPP lending on county-level banking characteristics and demographics. The
dependent variable is county-level PPP lending scaled by the number of establishments with fewer than 500 employees. Total PPP
lending includes all PPP loans in the SBA data; the division into core v. peripheral lenders includes only loans which we were able
to match to Call Reports lenders (about 94.8% of the total). Core lenders are banks that have at least one branch in the county and
peripheral lenders are those witout branch in the county. Round 1 PPP lendings include PPP loans approved before April 17 and
Round 2 include loans approved after April 17. Branches/Establishment is the ratio of total branches / establishments with fewer
than 500 employees as of 2019. Predicted PPP Lending equals the weighted average of each bank’s predicted bank-level PPP
Lending/Assets in their core markets, with weights equal to each bank’s share of total deposits in the county from 2019. Stardard
errors are clustered by state. T-statistics are reported in parentheses. '*' denotes significance at the 10% level, '**' the 5% level, and
'***' the 1% level.
Total PPP per Establishment
Core Lenders Peripheral Lenders
Total PPP per Second
Establishment (000s of $s) First Round Round First Round Second Round
(1) (2) (3) (4) (5)
Branches/Establishment 888.6*** 192.6*** 30.15* 497.2*** 136.6**
(177.7) (42.23) (16.94) (110.1) (54.43)
Predicted PPP Lending 118.6 291.2*** -14.01 -113.3** -45.84**
(73.51) (22.15) (13.20) (51.38) (21.30)
County COVID death rate 0.118*** 0.0136 0.0183** 0.0539*** 0.0297***
(0.0215) (0.00862) (0.00853) (0.0158) (0.00718)
Log of County Population -3.049** 3.309*** 1.496*** -5.523*** -2.011***
(1.333) (0.404) (0.213) (0.806) (0.477)
Percent with College or more -25.16** 22.80*** 7.357*** -38.29*** -15.00***
(12.20) (4.423) (1.995) (6.883) (3.754)
Percent between ages 20 and 44 65.15** 3.574 -4.102 41.36*** 18.46**
(26.35) (5.640) (4.000) (14.32) (7.832)
Log of Median County Income 42.91*** 2.104 5.318*** 24.21*** 10.22***
(8.526) (1.668) (1.435) (5.525) (2.632)
Unemployment in 2019 86.32 -36.63 68.80*** 5.089 49.18
(138.8) (26.19) (20.88) (89.99) (38.40)
Constant -416.3*** -57.28*** -68.70*** -193.8*** -86.87***
(93.69) (18.04) (16.75) (60.71) (28.48)
Observations 3,068 3,068 3,068 3,068 3,068
R-Squared 0.074 0.233 0.275 0.139 0.110
44
Table 9 County-Level Economic Outcomes and PPP Lending by Local Banks
This table reports county-time panel regressions of real economic outcomes on measures of the size and structure of
local banks. Small Business Revenue and Total Spending come from https://tracktherecovery.org; see Chetty et al.
(2020). Monthly unemployment rates by county are from the Bureau of Labor Statistics. Predicted PPP Lending
equals the weighted average of each bank’s predicted bank-level PPP Lending/Assets in their core markets, with
weights equal to each bank’s share of total deposits in the county from 2019. Stardard errors are clustered by county
and time in columns (1) and (2) and by count and state-time in column (3). T-statistics are reported in parentheses.
'*' denotes significance at the 10% level, '**' the 5% level, and '***' the 1% level. F-statistic tests that the Predicted
PPP lending, times the four time interactions from April on, equal zero.
Small Business
Revenue Total Spending Unemployment Rate
(1) (2) (3)
March x Predicted PPP Lending 0.415 0.267* -0.618
(0.331) (0.158) (1.147)
April x Predicted PPP Lending 0.179 0.258 1.837
(0.472) (0.230) (5.105)
May x Predicted PPP Lending 0.525 0.398 -10.71***
(0.502) (0.264) (3.946)
June x Predicted PPP Lending 0.577 0.0783 -15.19***
(0.522) (0.301) (3.608)
July x Predicted PPP Lending 0.223 0.0578 -15.04***
(0.535) (0.262) (3.537)
F-Test 0.78 1.08 8.13
P-value 0.540 0.370 0.001
County Fixed Effects YES YES YES
State x Time effects YES YES YES
Frequency Daily Daily Monthly
Control Variables: Calendar Interaction x Branches/Establishment
Control Variables: Calendar Interaction X Ln of Population
Control Variables: Calendar Interaction X COVID Death rate
Control Variables: Calendar Interaction x County Median Income (2018)
Control Variables: Calendar Interaction X Share College
Control Variables: Calendar Inteaction x Share 20-44 years of age
Control Variables: Calendar Interaction x Local Unemployment in 2019
Observations 430,519 347,730 21,476
R-squared 0.578 0.665 0.918
45
File and source
- File
- who-supplies-ppp-loans-and-does-it-matter-banks-relationships-and-the.pdf
- Size
- 1,441,083 bytes
- SHA-256
- 99ff8da4230ae15a6753d8eaa3773ddc8b818bb7e1c4342df1d3bf4b0fe0e09d
- Original
- www.nber.org