Banking Deserts and the Paycheck Protection Program

Abstract

The Paycheck Protection Program was a highly unusual policy measure enacted to provide bridge capital to support small businesses coping with the dramatic downturn in demand due to the COVID-19 pandemic. By design, the program effectively required potential applicants to work through the bank with whom they had a relationship. Yet large swathes of the country are effectively banking deserts, which dramatically steepen the gradient for those regions’ businesses seeking Paycheck Protection Program support. This paper tests the proposition that the exogenous distribution of banks effectively discriminated against those regions where banking services were limited, while also looking at whether loans were distributed to those areas with less dense employment opportunities and higher concentrations of small businesses. The authors find that areas with fewer banking services and lower employment opportunities were systematically disadvantaged in the Paycheck Protection Program distribution, while there were no significant flows to areas with higher rates of small businesses.

Keywords

Paycheck Protection Program COVID-19 public policy banking small business

For much of 2020, 2021, and 2022 to date, the COVID-19 pandemic ravaged the globe and the United States, harming both public health and the economy. As of early June 2022, more than 85 million people in the United States have been infected and over a million have died. While the unemployment rate was down to 3.6% in May 2022, after peaking at 14.7% in April 2020, gross domestic product (GDP) shrank through the first two quarters of 2020, creating the first official downturn since the Great Recession. The labor market has improved considerably in the meanwhile, with the economy now having a deficit of under a million jobs relative to the beginning of the pandemic. Yet the effects to public health and the economy have exposed many underlying inequalities within communities, between communities—particularly across the urban-rural spectrum—and more broadly, between regions of the country.

One such existing inequality is the access to banking services. Banking services vary regionally, creating inequalities in the ability to access credit between banking clients residing in markets well-served and those under- or unserved. While many banking services can be transacted remotely through mobile banking and a rising Fintech sector, there are still public and personal benefits derived from proximity to a bank branch or headquarters. Local banks can utilize “relationship lending” where “soft” information gleaned through business networks can be used for credit decisions (Berger & Udell, 2002; Petach et al., 2021). These linkages are particularly important for rural small businesses that are informationally opaque, generating an imbalance of information to the disadvantage of the borrower based purely on geography (Akerlof, 1970; Bunten et al., 2015; Conroy et al., 2017).

The personal relationship that bankers can establish with small business owners provides them with insights into a business's managerial practices, its relationships with suppliers and customers, and its impact on the local economy, among other pieces of information that are not included on the canonical balance sheet. Through bank consolidations and closures, many Americans live in a relative banking desert—a community with no physical banks—effectively creating spatial mismatches between financial resource conduits and the business sector. Even more live in areas that are effectively banking hinterlands, with no banking headquarters and served only by branches of national chains, as the distance between headquarters and branches has grown continuously since the turn of the millennium (Petach & Weiler, 2019). Areas may become dependent on branch locations from a single national commercial bank that bases its lending on rigid policies, procedures, and credit scoring systems determined by a distant headquarters that does not incorporate the richness of soft information. The inequality in access to local banks can have significant implications on how recovery funds are distributed and thus for the trajectory of post-pandemic recovery. It also leaves a gap in the institutional leadership role that banks and bankers play in communities.

The COVID-19 pandemic caused many businesses to furlough or lay off broad swaths of employees as incoming revenues sharply declined. Businesses were forced to close by their local governments and people stayed home to prevent the spread of the virus. From February to April 2020, the number of active businesses in the United States dropped 22% (Fairlie, 2020). Amidst these losses, in March 2020, Congress created the Paycheck Protection Program (PPP) to help struggling small business owners weather these unprecedented headwinds.

The PPP allowed small businesses to obtain low-interest loans to cover payroll and other expenses. The loans were distributed through banks that were existing Small Business Administration (SBA) 7(a) lenders. For the first round of the program, $349 billion was allocated. But given the severity of economic hardships experienced by small businesses, these funds ran out in 2 weeks. The second round eased some of the constraints of the first effort, with a new tranche of $284.5 billion allocated on a first come first served basis. The second allowed nonbanks to request PPP funding, while also focusing on smaller and more diverse businesses. In neither round were there specific allocations to particular banks. The system was national, as was the potential pot of money for disbursement.

We analyze both first and second round of PPP lending patterns. Crucially, in the first round, most of the funds were allocated to businesses that had an existing relationship with a qualified bank. Given the geographical distribution of local banks, this system for loan disbursement may have created greater overall inequality for those areas already experiencing inequality of banking access. The second round opened the way for Fintech loans, which removed some of the observed discriminatory practices and reduced the reliance on a relationship with a bank for gaining fund access.

Using data on the PPP loan receipt and the locations of banking institutions, this study leverages national Federal Deposit Insurance Corporation (FDIC) and National Credit Union Administration (NCUA) data to understand how community banking density was related to the disbursement of the loans. We first examine the a priori regional distribution of banks and businesses at the commuting zone level, using combinations of counties that represent a commuting shed. We then map the distribution of PPP loans relative to those distributions. We compare the results with measures of regional economic disadvantage—in particular, the employment-to-population ratio, a particularly sensitive measure of labor market opportunity (Amior & Manning, 2018)—to see if the distribution of banking services and disbursements of PPP funds mitigated or reinforced existing patterns of regional inequality.

We evaluate banking availability by measuring the number of banks per 10,000 people within a commuting zone (CZ) using the U.S. Department of Agriculture Economic Research Service's 2000 definition. CZs, fully covering all 50 states and the District of Columbia, reflect local labor markets and more accurately capture the accessibility of banks for establishments within a given region, following the intuitive proposition that business owners shop for banking services close to where they live and/or work. Banking hinterlands are classified as CZs without a bank headquarters. The focal dependent variable is the number of loans per eligible small business establishment, although we also briefly examine the amount of the loans and jobs retained per eligible small business. The latter two variables are more likely to be programmatically tied to payrolls, while the more penetrating marginal decision by banks is the number of loans to issue to small businesses.

The empirical framework analyzes a cross-section of all CZs, geographically situating banking deserts and hinterlands. There are three primary hypotheses to investigate, which we introduce into our cross-sectional regression alongside a suite of regional control variables:
PPP loans were systematically lower in relative banking deserts and hinterlands after accounting for lower business and population concentrations.

Lower disbursements were more evident in areas that had lower employment-to-population ratios, indicating that loans were going to relatively advantaged labor markets.

Small businesses of less than 50 as well as less than 10 employees were not the major beneficiaries of the PPP, despite the political rhetoric suggesting that such establishments were primary beneficiaries of the novel funding flows.
We include controls for income inequality, per capita GDP, educational attainment, and the non-White share of the CZ's population, which may all factor into regional loan flows.

The key hypotheses of the paper hold in the empirical analysis. Regions with higher concentrations of banks received greater average numbers of loans, confirming the banking desert hypothesis. Affirming our a priori hypothesis, we show that more loans per small business were disbursed in regions that had banking headquarters, even while controlling for the number of banks in the region. The number of loans did not go systematically to smaller businesses. Finally, loans were lower in labor-market-challenged areas, as measured by employment-to-population ratios. Not only did smaller businesses not gain from the program, but both the banking desert and banking hinterland hypotheses held—loans flowed to areas with higher job concentrations—suggesting that there was a spatial mismatch in the program based on geography.

Related Literature

While the notion of banking deserts is anecdotally rich, there are remarkably few independent empirical analyses of the veracity and extent of the banking access problem. Most have found that the concept mainly applies to rural areas rather than cities (e.g., Hrushka, 2020; Kashian et al., 2015; Morgan et al., 2018). This repeated theme reinforces our work's focus on broader CZs rather than zip codes, census tracts, or counties as a preferred spatial level of analysis, given CZs’ spatial homogeneity in uniting the transportation habits of residents and workers. A focus on metropolitan areas would also miss many of the more significant banking deserts and hinterlands. For the purposes of this paper, businesses are synonymous with business establishments.

Banking deserts are just one level of the lack of bank access for businesses. The best banking relationships would develop where the borrower is in proximity of a bank headquarters rather than just a branch of a regional or national chain. Credit score sheets created by faraway headquarters are not likely to match local circumstances and promising frontier businesses, thus hampering capital-led drives for regional employment growth and diversification (Conroy et al., 2017). In contrast, areas with banking headquarters are more likely to have loan officers that can leverage soft information, improving estimates of loan viability. We therefore explore both banking deserts, CZs with no banks and banking hinterlands, those CZs with no banking headquarters, as well as the overlap of this banking geography with the geography of non-White populations.

To date there have been limited studies of the PPP, with most focusing on optimal allocation theory (Elenev et al., 2020; Joaquim & Netto, 2020), bank performance (Granja et al., 2020; Kapinos, 2021), flows to minority communities (Fairlie & Fossen, 2021), and/or business/employment survival (Autor et al., 2020; Bartik et al., 2020). Yet the question of bank access and consequent local relationships is particularly important for this unusual business support policy, as banks were the sole conduit for securing PPP monies. Soft information is especially crucial in these circumstances, as the margins on PPP loans were very small (Marsh & Sharma, 2020). Banks thus had extra incentive to rely on soft information about the borrower to maximize the chance of the loan getting repaid. Again, the availability of such information is least likely in banking deserts, and less likely in banking hinterlands. Banks in such hinterlands are unlikely to have loan officers and are more likely to use credit score sheets developed by distant headquarters.

Previous work on PPP and bank exposure indicated that there is a disconnect between the status of the local economy and the likelihood of receiving PPP loans. Kapinos (2021) found that PPP loans did not systematically flow to counties that experienced unemployment surges in the first round of the pandemic. More generally, using both zip code and county data, Granja et al. (2020) found little relationship between loan disbursement and local economic conditions; the number of COVID-19 cases was not a better predictor, with some indication that loans were more prevalent where caseloads were lower. In general, however, even this tremendously comprehensive study was somewhat limited by its choice to rely on zip codes and counties, neither of which are natural markets. Fairlie and Fossen (2021) also used zip codes in their analysis, which we believe represents too small a microscope to properly understand PPP disbursement.

In contrast, this paper relies on CZs as its primary geographic scale of analysis, as they define the extent of commuting and intraregional cohesion (Amior & Manning, 2018). Buyers of loans are most likely to choose banks that are close to home or their workplace. In terms of the status of the local economy, we focus on the employment-to-population ratio as an indicator of the density of jobs available to the region's population following the significant downturn in March/April 2020, as well as the proportion of citizens with a higher education degree and per capita GDP. Amior and Manning underscored the employment-to-population ratio as being a particularly appropriate measure of economic opportunity, which we leverage in this work.

Our approach follows clues left by the handful of studies on PPP loan distribution. Amiram and Rabetti (2020) and Li and Strahan (2020) found that those establishments with existing banking relationships tended to get loans first and in the largest amounts. We indirectly test both propositions, through the resource flows channeled toward the smallest businesses that are less likely to have established banking relationships. Granja et al. (2020) found that those relationships tended to outweigh stated goals of the program, namely targeting the areas and businesses in greatest need of loans due to the pandemic. We empirically test the Barrios et al. (2020) finding that establishment payrolls closely predicted PPP loan receipt, indicating that there may be a positive relationship between business size and loan disbursement. Finally, Fairlie and Fossen (2021) found that early loans went mainly to nonminority applicants, while the later tranche flowed more to marginalized populations. We follow their lead in testing the CZs with a greater share of non-White who people received a fewer number of PPP loans.

As previous work has demonstrated, the implications of systematic informational asymmetries based on geography can fundamentally shift innovation and resources away from lagging regions, further entrenching their economic struggles (Weiler, 2000). These geographic information asymmetries (GIA) are most likely to be felt in business-to-business transactions built on the lender's understanding of those demanding services. Small business lending may be particularly vulnerable to GIA discrimination, given its reliance on credit scoring developed at a bank's headquarters—which may not be congruent with the realities of a rural economy—as well the past viability of similar projects. The latter will be a particularly high hurdle for innovative projects that have no track record in the focal economy. Rural areas tend to have thin informational markets due to lower establishment dynamism and thus fewer data points from which to extract the viability of loans (Bunten et al., 2015).

Statistically, the perception of an otherwise identical probability distribution of outcomes in two regions is skewed toward the market with thicker information due to their greater business experience. In contrast, the market with fewer data points will have a higher perceived variance of outcomes that heightens uncertainty and thus, risk for bankers (Weiler et al., 2006). These risks may deter bankers from making loans to companies without existing intensive relationships, leading to disproportionate flows toward advantaged regions and businesses. Such informational asymmetries may be a driving force for systematic discrimination of PPP loans toward thick-market regions that have denser labor markets as measured by the employment-to-population ratio, established banking networks, and larger establishments. Our empirical work tests these propositions.

Data and Empirical Model

Data on PPP loans come from the SBA. These data contain information on all individual loans distributed through the program's first phase, which was to end June 30, 2020, but was extended to August 8, 2020. The loan amount, business address, number of jobs reported, loan approval date, and the demographic characteristics of the business owner are included for each loan. For our work, the loan amount, number of jobs reported, and business address are used in the construction of the final data. The demographic characteristics are not used due to the large number of loans where those questions were not answered and the likelihood of introducing sample selection bias through their use.

CZs from the U.S. Department of Agriculture Economic Research Service's 2000 delineations are the primary unit of analysis. CZs offer two main advantages over other geographic delineations. First, CZs represent local economies better than other boundaries by grouping together counties with strong commuting ties. Second, CZs contiguously cover the continental United States, meaning that all businesses that received a PPP loan are retained in our sample. CZs lend themselves particularly well to studying the distribution of PPP loans. It is conceivable that a small business may have to seek banking services outside of its city or county due to a lack of access to banks in that location. Using CZs as the geographic unit more accurately captures the number of banks available to a small business. To obtain a measure of the number of loans in a CZ, we first aggregate the number of loans to the zip code level. We then map the zip codes to the counties in which they reside. For the zip codes that cross county borders, we weight the number of loans by the proportion of businesses located in each county.¹ Finally, we map the aggregate number of loans at the county level to the appropriate CZ. We repeat this process for both the amount of loans and the number of jobs reported to create CZ measures for these outcomes.

CZs are geographically diverse, do not have identical populations, nor identical economies. For this reason, we normalize our outcomes of interest by the number of small businesses in each CZ. Since the PPP was designed to provide economic relief for small businesses, our study focuses on the number of small businesses in each CZ. The SBA's definition of a small business varies by industry, so we utilize its general definition as a business having fewer than 500 employees. The number of businesses with fewer than 500 employees was taken from the most recent County Business Patterns data from the U.S. Census Bureau. We implement the same aggregation procedure to get the total number of small businesses in every CZ.

Our main source of CZ-level economic characteristics, such as population, household income, demographic characteristics, and level of education, is from the American Community Survey 5-year estimates for 2015 to 2019. This source provides data on these characteristics at the county level, which we aggregate up to the CZ level. Data on county GDP come from the U.S. Bureau of Economic Analysis Regional Economic Accounts, which we sum up to the CZ GDP. These data are for 2019 since that is the most recent data available at the county level. Data on the number of COVID-19 cases come from The New York Times, which has been tracking the number of daily COVID-19 cases by county since the beginning of the pandemic. To accurately measure the potential impact of COVID-19 cases and local lock-down measures in an area, we use the cumulative count COVID-19 cases for each county on April 3, 2020, which is when the PPP opened for small businesses. We aggregate that case count up to the CZ level. We get our data for labor market outcomes from the U.S. Bureau of Labor Statistics’ (BLS) Local Area Unemployment Statistics. Since the BLS measures employment in the first 2 weeks of the month, we use the county data on employment from April 2020 because they reflect the reality of local labor markets at the time business establishments were deciding whether to apply for a PPP loan.

We obtain our data on bank location from two sources: the Federal Deposit Insurance Corporation (FDIC) and the National Credit Union Administration (NCUA). We focus on banks and credit unions because both were authorized to provide PPP loans to small businesses. We obtain the location of all bank branches and headquarters from the FDIC's Institutions and Locations database. This provides the addresses (including counties) of all federally insured banks, the type of banking services provided, and whether the bank is the main office or a branch location. We create a measure of total banks by including all full-service banks, both brick and mortar and retail locations, as well as permanent limited-service banks that only accept deposits and payments. We include the latter type of banks to capture the effect of banking hinterlands on the distribution of PPP loans. The county data are also aggregated up to the CZ level. We use NCUA Quarterly Call Report Data for the second quarter of 2020 to get a list of all credit unions by location. The list does not include the county of residence, so we follow a zip-code-to-county procedure similar to what we used on the PPP loan data before aggregating it to the CZ level. The number of credit unions is combined with the number of banks to create a total measure of bank concentration for each CZ.

Table 1 presents summary statistics for each of the variables in our sample. The final sample consists of 706 CZs. Due to the structure of the program, we only have data for loans that were made during its first two iterations in early to mid-2020. To account for population differences across CZs, we normalize the number of banks and credit unions, bank headquarters, COVID-19 cases, and number of small businesses by CZ population per 10,000 in regional population—in the spirit of Amior and Manning (2018).

Table 1.
Summary Statistics.

Mean Std. dev. Min. Max. N

PPP loans per small business 0.672 0.204 0 2.17 706

PPP loan amount per small business 51750.003 14474.729 0 131018.211 706

PPP jobs retained per small business 5.857 1.41 0 15.075 706

Round 1 loans per small business 0.295 0.121 0 1.261 706

Round 1 loan amount per small business 36682.66 12374.749 0 97266.570 706

Round 1 jobs retained per small business 4.027 1.27 0 11.313 706

Banks + C.U.s per 10k people 4.492 2.908 0 37.831 706

Bank headquarters per 10k people 0.636 0.972 0 9.653 706

COVID-19 cases per 10k people 2.808 5.756 0 88.287 706

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$ 0.403 0.065 0.224 0.666 706

$\frac{M e d i a n H H . I n c .}{M e a n H H . I n c .}$ 0.754 0.042 0.598 0.904 706

GDP per capita in thousands 54.045 25.504 22.774 392.175 706

Share of pop. with at least a B.A. 0.236 0.078 0.073 0.556 706

Non-White population share 0.183 0.157 0.007 0.894 706

Small businesses per 10k people 252.649 86.404 57.157 1005.714 706

Share of small businesses $<$ 10 employees 0.748 0.05 0.607 1 706

Share of small businesses $<$ 50 employees 0.962 0.017 0.911 1 706

Abbreviations: PPP = Paycheck Protection Program; GDP = gross domestic product.

To explore the geographic distribution of PPP loans, banks, and COVID-19 cases, we map the quintiles for data for each in Figure 1. Panel (a) shows the geographic distribution of the number of PPP loans per small business, Panel (b) shows the geographic distribution of banks per 10,000 people, and Panel (c) shows the geographic distribution of the number of COVID-19 cases per 10,000 people. The dollar amount of PPP loans per small business is highest in the central United States, stretching from North Dakota to the northern portion of Texas. This region shares significant overlap with the geographic distribution of banks per 10,000 people. The Southwest also has a high number of PPP loans per small business; however, there is not a high concentration of banks in this region. COVID-19 per 10,000 people as of April 3, 2020, were concentrated in the northeast United States, Louisiana, and the Southwest.

Figure 1.
Geographic distribution of Paycheck Protection Program (PPP) loans, banks, and COVID-19 cases. Panel (a) on the upper left shows the geographic distribution of PPP loans per small business. Panel (b) on the upper right shows the geographic distribution of banks per 10,000 people. Panel (c) at the bottom shows the geographic distribution of COVID-19 cases per 10,000 people. All maps show the quintile distribution of the data.

This geography aligns with where the largest outbreaks of cases were near the beginning of the pandemic. The highest quintile for COVID-19 cases has a very large range (3.52–88.29), making some areas appear to have a comparable number of cases despite having much different values of cases per 10,000 people. Overall, these maps (Figure 1) suggest that there is significant overlap between the regions where the number of PPP loans and the number of banks is the greatest. There is some overlap between loans and COVID-19 cases, but it does not appear from these maps that the PPP loans went to places where COVID-19 cases were highest. It is possible that the heterogeneity in the intensity of the stay-at-home orders and business closures near the beginning of the pandemic did not align with where the cases were the highest at the start of the PPP.

We use a more formal analysis to further explore these and other relationships. We seek to answer whether the distribution of banks and banking hinterlands partially determined the distribution of PPP loans. Since the PPP loans were distributed through banks and there was a limited amount of initial funding, we hypothesize that regions with a greater concentration of banks and bank headquarters received more PPP loans, and that banking deserts and banking hinterlands were systematically disadvantaged in PPP allocations. We take the geographic distribution of banks as being given a priori, and thus exogenous to the analysis. A higher concentration of banks provides more opportunities for small businesses to find a bank that has not already exhausted its allotment of PPP funding. Although the program was originally designed to continue through June 30, 2020, the original amount of funding provided by the CARES Act ran out by April 16, 2020. Businesses in regions with a higher concentration of banks that were able to access the PPP funds earlier probably increased their likelihood of survival, as shown in Bartik et al. (2020) and Autor et al. (2020). We are also specifically interested in whether the loans and dollars went to the small businesses that were avowedly priorities for the program, and if the program helped job retention. To empirically test these hypotheses, we estimate the following equation:
$\begin{aligned} Y_{i} = & β_{0} + β_{1} B a n k C o n c e n t r a t i o n_{i} + β_{2} B a n k C o n c e n t r a t i o n_{i}^{2} \\ + β_{3} B a n k H Q s_{i} + β_{4} B a n k H Q s_{i}^{2} + β_{5} C a s e s_{i} \\ + β_{6} N o n W h i t e S h a r e_{i} + β_{7} S h a r e S m a l l e s t B u s i n e s s e s_{i} \\ + β_{8} {\frac{E m p .}{P o p .}}_{i} + β_{9} {\frac{H H . M e d . I n c o m e}{H H . M e a n . I n c o m e}}_{i} \\ + β_{10} p e r C a p i t a G D P_{i} + β_{11} B A P l u s S h a r e_{i} + ϵ_{i}, \end{aligned}$
where $Y_{i}$ is the outcome of interest—either the number of PPP loans per small business, amount of PPP loans per small business, number of jobs reportedly retained per small business, or their first round of funding equivalent in CZ i. $B a n k C o n c e n t r a t i o n_{i}$ is the number of banks and credit unions per 10,000 people in CZ i; $B a n k M a i n O f f i c e s_{i}$ is the number of bank main offices per 10,000 people in CZ i; $C a s e s_{i}$ is the CZ number of COVID-19 cases per 10,000 people; $N o n W h i t e S h a r e_{i}$ is the share of non-White population in CZ i; $S h a r e S m a l l e s t B u s i n e s s e s_{i}$ is the CZ-level share of all small businesses that have fewer than 50 or fewer than 10 employees; ${\frac{E m p .}{P o p .}}_{i}$ is the CZ-level employment-to-population ratio for April 2020; ${\frac{H H . M e d . I n c o m e}{H H . M e a n I n c o m e}}_{i}$ is a CZ-level measure of inequality measuring the ratio of household median income to household mean income; $p e r C a p i t a G D P_{i}$ is the per capita GDP of CZ i; $B A P l u s S h a r e_{i}$ is the share of the CZ population that has at least a bachelor's degree; $S m a l l B u s i n e s s C o n c e t r a t i o n_{i}$ is the number of businesses with less than 500 employees in CZ i; and $ϵ_{i}$ is an idiosyncratic error term.

Variance inflation factor (VIF) analyses indicate that there was no substantive multicollinearity among the explanatory variables. We fully acknowledge that the cross-sectional data set can offer only a limited view of PPP impacts, given that the data show only transacted loans. The lack of application data would tend to upwardly bias the findings by overstating the effect of local banks, making our results an upper bound of the PPP effect.

Coefficients $β_{1}$ and $β_{3}$ are our main coefficients of interest. Consistent with our banking desert and hinterland hypotheses, we expect both coefficients to be positive for all outcomes of interest, as more loans flow to those areas with more banks and bank headquarters than toward banking deserts and hinterlands. We include the square of both variables to account for the possibility that the concentration of banks and headquarters beyond a certain threshold does not further increase the outcomes of interest. We also focus on the employment-to-population ratio as a particularly important measure of labor market opportunity, and thus have a keen interest in $β_{8}$ . If funds had flowed to more distressed labor markets in the wake of the COVID-19 crisis, we would expect $β_{8}$ to be negative. However, if funds went to relatively advantaged labor markets, $β_{8}$ would be positive. In that sense, $β_{8}$ becomes an especially useful bellwether in determining whether PPP served the most economically distressed regions and acted as a control for business activity in the region.

Additionally, we are interested in three subquestions: (1) whether PPP loans went to areas that were most affected by the pandemic, (2) whether CZs with a higher proportion of non-White population received fewer loans, and, in particular, (3) whether the smallest small businesses were particularly disadvantaged in accessing PPP loans. For these questions, we are interested in coefficients $β_{5}$ , $β_{6}$ , and $β_{7,}$ respectively. If PPP funds were distributed to the CZs that were most affected, we would expect $β_{5}$ to be positive. If communities of color had a more difficult time accessing or applying for PPP funds, we would expect $β_{6}$ to be negative. And if the smallest businesses were particularly disadvantaged in accessing or applying for PPP loans then we would also expect $β_{7}$ to be negative.

Results

PPP in Total

Table 2 presents results from estimating the effect of bank concentration on the number of total first- and second-round PPP loans. Column 1 reports the results for the number of loans per small business in a CZ. Column 2 reports the results for the loan amounts per small business. Column 3 displays the results for the number of jobs retained per small business. Coefficients are standardized to impacts in terms of standard deviation changes in the dependent variables in response to a one standard deviation change in the focal explanatory variable. The Appendix reports all results in nonstandardized formats. We also apply two different measures for the share of smallest businesses: we define smallest businesses as businesses with fewer than 10 employees and as businesses with fewer than 50 employees. Results using each of these measures are largely similar, so we focus on those businesses with fewer than 10 employees and reserve the specifications with the alternative measure for the Appendix.

Table 2.
Total PPP Loan Program.

(1) (2) (3)

PPP loans per small business PPP loan amount per small business PPP jobs retained per small business

Banks + C.U.s per 10k people 1.002* −0.165 −0.143

(6.89) (−1.29) (−1.08)

(Banks + C.U.s per 10k people)² −0.527* 0.007 −0.013

(−5.76) (0.08) (−0.17)

Bank headquarters per 10k people 0.333 0.096 0.211

(2.46) (0.71) (1.61)

(Bank HQs per 10k people)² −0.115 −0.084 −0.099

(−1.24) (−0.51) (−0.63)

COVID-19 cases per 10k people 0.060 0.066 0.031

(2.32) (1.26) (0.61)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$ 0.203* 0.255* 0.217*

(4.08) (3.24) (2.67)

$\frac{M e d i a n H H I n c .}{M e a n H H I n c}$ −0.009 0.005 −0.066

(−0.22) (0.10) (−1.23)

GDP per capita in thousands −0.124* 0.072 −0.091

(−2.84) (0.70) (−1.25)

Share of pop. with at least a B.A. 0.072 0.085 −0.067

(2.14) (1.31) (−1.00)

Non-White population share 0.159* −0.018 −0.000

(3.40) (−0.27) (−0.00)

Share of small businesses < 10 employees −0.148*

(−3.39)

PPP loans 0.247* 0.142**

(3.02) (2.54)

Observations 706 706 706

R ² 0.570 0.224 0.061

Abbreviations: PPP = Paycheck Protection Program; GDP = gross domestic product.

Standardized coefficients reported. Robust standard errors were calculated. T-statistics in parentheses. Employment statistics are from April 2020. Small business data come from the most recent County Business Patterns (2019).

p < .1, p < .05, p < .01.

Our results suggest that larger concentrations of banks and credit unions in a CZ had a positive and statistically significant impact on the number of PPP loans per small business. CZs where the number of banks per 10,000 people was one standard deviation greater saw an increase in the number of PPP loans per small business by about one standard deviation. However, the negative coefficient on the squared term appears to indicate diminishing returns to an increase in the number of banks per 10,000 people. Once the concentration of banks was greater than 23.3 banks per 10,000 people, the number of PPP loans was negatively affected by additional banks. Even before that level, each additional increase in the concentration of banks was less effective in increasing the number of PPP loans. This result suggests that in places that had fewer banks, an additional bank was helpful for obtaining a PPP loan, but in places that already had a higher concentration of banks, additional banks reduced the number of loans.

These findings indicate that the geographical distribution of banks impacted the number of PPP loans, likely by providing more options for small businesses seeking to apply for a loan. If one local bank was no longer accepting applications, the business owners could look to another bank in their area. Small businesses in areas with fewer banks did not have this luxury. Areas with a sufficient number of banks were not helped by additional banks because they already had the capacity to handle PPP loan demand.

We find similar first-order effects for the number of banking headquarters per 10,000 people. CZs with a concentration of headquarters that was one standard deviation greater received 33.5% to 35% of a standard deviation more PPP loans per small business. However, we do not find diminishing returns to the number of bank headquarters; more bank headquarters systematically increased the number of loans. Bank headquarters are likely to have a greater ability to tailor loans by leveraging soft information. This creates an advantage for small businesses located in CZs with a bank headquarters over those located in CZs with only a branch office. Our results are stable across both our definitions of the smallest businesses.

In addition to our main hypotheses about the geographic locations of banks, we are also interested in whether the PPP funds went to areas that were the most affected by the pandemic. CZs with more COVID-19 cases per 10,000 people received more loans. Places where the cases per 10,000 people were one standard deviation above the mean decreased the amount of loans received by 5.2% of a standard deviation, all else equal. In the early stages of the pandemic, regions that had more cases were likely to have had more restrictions on the types of business allowed to remain open. Our empirical results suggest that loans were going to CZs that were in greater health distress. Regions that had a greater employment-to-population ratio of one standard deviation also received about 20% of one standard deviation greater share of PPP loans, indicating that PPP loans went to regions that generally had advantageous labor market conditions. We also ran our models using March 2020 employment data and had substantially identical findings.

It is worth noting that more economic activity in a CZ makes the demand for loans greater, so in some sense the employment-to-population ratio is more of a control for economic activity in the form of demand for PPP loans. More economic activity means more loans, ceteris paribus. However, this control also makes the banking results especially stark and compelling, in that the equation has already controlled for potential business loan demand.

While the PPP was not designed to specifically help minority-owned small businesses, we do find that regions with a greater share of non-White population received a larger number of PPP loans per small business. A one standard deviation larger share of non-White population received about 16% of a standard deviation more PPP loans per small business. This finding is consistent with the findings in Fairlie and Fossen (2021) and is important because minority-owned businesses were some of the most affected by the economic shutdown (Fairlie, 2020).

We were also interested in whether the smallest small businesses were disadvantaged by the PPP loan process. We find that CZs with a one standard deviation higher concentration of businesses with fewer than 10 employees received 14.8% of one standard deviation fewer PPP loans. Places where the concentration of small businesses with less than 50 employees was one standard deviation greater also received about 15% of a standard deviation fewer PPP loan (Appendix Table A6).

Next, we study whether the geographical distribution of banks and banking headquarters affected the amount of the loans distributed to CZs and the number of jobs that were reportedly saved with the PPP loans. Since the amount of PPP loans per small business and the number of jobs retained per small business is a function of the number of loans in a CZ, we add this as a control to Equation (1). Column 2 of Table 2 presents the results for the PPP loan amount per small business and Column 3 presents results for the number of jobs retained per small business.

We find that the geographic distribution of banks and credit unions did not have a significant impact on the amount of the PPP loan per small business. Similarly, we do not find that the concentration of banks and credit unions had a significant effect on the number of jobs retained per small business. These results make intuitive sense since banks did not determine the amount of the loans that small businesses were able to receive. Both outcomes are dependent on them first being approved for a loan, so it makes sense that neither were affected by the distribution of banks. The amounts of loans and the number of jobs saved were strictly a function of the business itself, not the PPP policy.

First Round of PPP

The PPP was originally designed to be in effect from April 3 to June 30, 2020, but it ran out of funds by April 16, and it was unclear if more funding would be approved. The magnitude of the economic impacts at the beginning of the pandemic meant that missing out on the first round of funding could impact the survival of small businesses that were subject to the government-imposed lockdown orders. For this reason, we examine whether the concentration of banks impacted the receipt of these first-round loans. We present the findings in Table 3.

Table 3.
First Round of PPP.

(1) (2) (3)

PPP Loans per small business PPP Loan amount per small business PPP Jobs retained per small business

Banks + C.U.s per 10k people 0.968* 0.021 0.051

(7.05) (0.18) (0.42)

(Banks + C.U.s per 10k people)² −0.564* −0.109 −0.139*

(−6.69) (−1.55) (−1.82)

Bank headquarters per 10k people 0.397*** 0.209* 0.364*

(2.96) (1.68) (2.91)

(Bank HQs per 10k people)² −0.272* −0.258* −0.338**

(−3.39) (−1.75) (−2.15)

COVID-19 cases per 10k people −0.035* 0.006 −0.033

(−1.80) (0.13) (−0.80)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$ 0.229* 0.318* 0.262*

(5.45) (4.38) (3.72)

$\frac{M e d i a n H H I n c}{M e a n H H I n c}$ 0.038 0.014 −0.046

(0.93) (0.28) (−0.88)

GDP per capita in thousands −0.147* 0.018 −0.127*

(−4.56) (0.24) (−2.75)

Share of pop. with at least a B.A. −0.032 0.059 −0.034

(−0.98) (0.95) (−0.56)

Non-White population share 0.050 −0.020 0.013

(1.37) (−0.34) (0.22)

Share of small businesses < 10 employees −0.155*

(−4.32)

PPP loans 0.070 −0.018

(1.28) (−0.57)

Observations 706 706 706

R ² 0.572 0.149 0.100

Abbreviations: PPP = Paycheck Protection Program; GDP = gross domestic product.

Standardized coefficients reported. Robust standard errors were calculated. T-statistics in parentheses. Employment statistics are from April 2020. Small business data come from the most recent County Business Patterns (2019).

p < .1, p < .05, p < .01.

Once again, we find that greater concentrations of banks and credit unions increased the number of PPP loans per small business. Areas with a one standard deviation positive difference in the number of banks and credit unions per 10,000 people raised the number of first-round PPP loans nearly one standard deviation. This is subject to diminishing returns as in the full sample of PPP loans, with a lower threshold at 20 banks per 10,000 people. We find that bank headquarters were also an important determinant of the number of PPP loans per small business. CZs where the concentration of bank headquarters per 10,000 people was one standard deviation greater than another CZ received 39.7% of a standard deviation more PPP loans per small business. In the first round of the program, this result is also subject to diminishing returns. This finding highlights the importance of bank access in the distribution of these loans. Small businesses in CZs with more banks were able to access funds earlier than small businesses in relative banking deserts and hinterlands.

While not a specific aim of the program, we also find that these early loans were not distributed to regions that were most affected by the pandemic. The pandemic was the core of the crisis, so evaluating how well the program fared in combating that crisis is an interesting question. Fewer loans went to regions with a higher concentration of COVID-19 cases and to regions where the employment-to-population ratio were lower. During the first round of funding, areas with COVID-19 cases per 10,000 that were one standard deviation greater received fewer loans per small business by 3.5% of a standard deviation. CZs with an employment-to-population ratio that was one standard deviation higher received about 22% of a standard deviation more loans per small business. Regions with fewer cases per person were likely not under as severe lockdown restrictions as regions where the virus was spreading more broadly, and businesses were able to continue operating at a closer to normal capacity. Similarly, regions with a greater employment-to-population ratio (and thus deeper job markets) may not have been experiencing the same pandemic-induced layoffs but still received a greater share of the first-round loans. At the time it was unclear whether there would be additional funding for small businesses, so these results represent a serious shortcoming of the program by advantaging areas with stronger labor market conditions.

In the first round of PPP the smallest businesses were disadvantaged compared to their large counterparts. Increasing the share of small businesses with fewer than 10 employees by one standard deviation decreased the number of first-round PPP loans per small business by over 15.5% of a standard deviation. We find a similar result when we expand our definition of the smallest businesses to those with fewer than 50 employees (12.2% of a standard deviation fewer loans as shown in Appendix Table A6). This highlights that the smallest businesses, which were likely to be most in need of the additional support, had unequal access to economic relief. We also do not find that CZs with a higher proportion of non-White residents received more of the first-round funds. Combined with the results from the full set of PPP loans, this finding suggests that small businesses in regions with a higher proportion of non-White people had to wait longer to receive PPP funds, a finding consistent with the results in Fairlie and Fossen (2021), Atkins et al. (2022), and Garcia and Darity (2022).

In Table 3, we also examine whether the geographic distribution of banks affected the amount of these first-round loans and the number of jobs they helped retain (see columns 2 and 3). Like our previous results using the full set of PPP loans, we do not find many significant results for these outcomes using just the first round of loans. For the first round of PPP loans, the number of banks and credit unions did not affect the amount of PPP loans, nor the number of jobs retained per small business. The number of bank headquarters did impact the loan amount per small business and the number of jobs retained per small business for the first-round loans. CZs with a one standard deviation positive difference in the concentration of bank headquarters received loan amounts that were 20.9% of a standard deviation greater and retained 36.4% of a standard deviation more jobs. Above a certain concentration, however, these outcomes were hurt by having a greater concentration of bank headquarters. COVID-19 cases and the share of non-White population also did not impact the amounts of the first round of PPP loans, or the number of jobs retained. All these findings are likely attributable to the applications process. The amount of the loan and the number of jobs it was used to retain are not determining factors for loan approval, so once the loan is approved, these factors are no longer an influence.

We do find that a higher CZ employment-to-population ratio correlates with a greater number of first-round loan amounts and jobs retained per small business. Increasing the CZ employment-to-population ratio by one standard deviation raised the first-round loan amount by 31.8% of a standard deviation. The same difference in the employment-to-population ratio raised the number of jobs retained per small business by 26.2% of a standard deviation. CZs with more employment received more first-round loans and therefore, a greater share of the total dollars available, enabling them to retain more jobs. Our findings consistently suggest that larger loans went to areas with relatively strong labor markets, potentially increasing interregional inequality.

PPP Loans by Bank Type

Thus far we have treated all banks and credit unions as if they are identical. However, there is heterogeneity across bank type with each having a slightly different method of doing business. For example, banks and credit unions offer many of the same financial services, but credit unions are nonprofit entities and banks are for profit. Within banks there are differences as well, with community banks operating differently than national banks. To capture how these differences affect the distribution of PPP loans, we adjust our measure of banking deserts to include the concentration of credit unions and the concentration of banks separately in the second column of Table 4. In column 3, we further disaggregate the banking concentration measure into the concentration of community banks and noncommunity banks alongside the concentration of credit unions. We do this for the full sample of PPP loans as well as for the sample of first-round PPP loans in Table 5.

Table 4.
All PPP Loans per Small Business: Disaggregated Bank Types.

(1) (2) (3)

Original specification Banks & credit unions Community banks & credit unions

Banks + C.U.s per 10k people 1.002*

(6.89)

(Banks + C.U.s per 10k people)² −0.527*

(−5.76)

Banks per 10k people 1.030*

(6.57)

(Banks per 10k people)² −0.552*

(−5.93)

Credit unions per 10k people −0.019 −0.049

(−0.39) (−1.02)

(Credit unions per 10k people)² 0.044 0.049

(0.97) (1.12)

Community banks per 10k people 1.171*

(8.87)

(Community banks per 10k people)² −0.581*

(−7.95)

Non-community banks per 10k people 0.367*

(5.29)

(Non-community banks per 10k people)² −0.247*

(−3.57)

Bank headquarters per 10k people 0.333 0.305 0.183

(2.46) (2.12) (1.32)

(Bank headquarters per 10k people)² −0.115 −0.109 −0.026

(−1.24) (−1.14) (−0.27)

Observations 706 706 706

R ² 0.570 0.576 0.606

Abbreviation: PPP = Paycheck Protection Program.

Standardized coefficients reported. Full set of controls included from Equation (1) but not reported. Robust standard errors were calculated. T-statistics in parentheses. Employment statistics are from April 2020. Small business data come from the most recent County Business Patterns (2019).

p < .1, p < .05, p < .01.

Table 5.
First-Round PPP Loans per Small Business: Disaggregated Bank Types.

(1) (2) (3)

Original specification Banks & credit unions Community banks & credit unions

Banks + C.U.s per 10k people 0.968*

(7.05)

(Banks + C.U.s per 10k people)² −0.564*

(−6.69)

Banks per 10k people 0.947*

(6.32)

(Banks per 10k people)² −0.547*

(−6.07)

Credit unions per 10k people 0.147*** 0.099*

(2.69) (1.78)

(Credit unions per 10k people)² −0.114* −0.098

(−1.77) (−1.44)

Community banks per 10k people 1.147*

(7.60)

(Community banks per 10k people)² −0.621*

(−7.12)

Non-community banks per 10k people 0.181*

(2.81)

(Non-community banks per 10k people)² −0.140*

(−3.98)

Bank headquarters per 10k people 0.397* 0.385* 0.146

(2.96) (2.70) (1.01)

(Bank headquarters per 10k people)² −0.272* −0.260* −0.120

(−3.39) (−3.11) (−1.46)

Observations 706 706 706

R ² 0.572 0.574 0.612

Abbreviation: PPP = Paycheck Protection Program.

Standardized coefficients reported. Full set of controls included from Equation (1) but not reported. Robust standard errors were calculated. T-statistics in parentheses. Employment statistics are from April 2020. Small business data come from the most recent County Business Patterns (2019).

p < .1, p < .05, p < .01.

In the full sample of PPP loans, we find that it is a higher concentration of banks that affected the distribution of PPP loans and not the concentration of credit unions. Places where the concentration of banks is one standard deviation greater is correlated with places where the number of PPP loans per small business is 1.03 standard deviations higher. The marginal impact of an additional bank is lower with each additional bank. We do not find that places with a greater concentration of credit unions increase the number of PPP loans per small business. We also find that the type of bank matters for the distribution of PPP loans. Greater concentrations of community banks and noncommunity banks lead to more PPP loans per small business, but an increased concentration of community banks has a more significant impact. CZs where the concentration of community banks one standard deviation greater are associated with a 17.1% of a standard deviation increase in the number of PPP loans. Places where the concentration of noncommunity banks is one standard deviation greater are associated with a 36.7% of a standard deviation more loans per small business. Both community and noncommunity bank concentration are subject to diminishing returns of an additional bank.

We also account for the heterogeneity in bank concentration for the sample of first-round loans. Results for the first round, accounting for differences in type of bank concentration, are included in Appendix Table A7. We find similar results for the first round of PPP funding. Places with a banking concentration that is one standard deviation greater received 94.7% of a standard deviation more first-round PPP loans. During the first round, credit unions also played a significant role in the distribution of PPP loans to small businesses. CZs with a credit union concentration one standard deviation higher saw 14.7% of a standard deviation more loans.

There was much uncertainty at the beginning of the pandemic regarding lockdowns and the resulting potential economic fallout. This resulted in small businesses scrambling to apply for and obtain a PPP loan. Since funds were limited, having more banking services available—regardless of type—mattered more than in the full sample of loans. Like our other analyses, we find that each additional bank and credit union had less of an impact on PPP loan distribution. We further separate out the community banking concentration from the previous bank concentration measure and find again that an increased community bank concentration had a larger impact on loan distribution than noncommunity bank concentration. Places where the concentration of community banks is one standard deviation high saw 115% of a standard deviation more loans while places where the concentration of noncommunity banks was one standard deviation greater saw only 18.1% more loans. Greater credit union concentration is also a factor that impacted the distribution of PPP loans and is correlated with 9.9% of a standard deviation more loans. These results suggest that the existence of more financial institutions in a CZ impacted the distribution of PPP loans through banks, with community banks having had the greatest impact in the first round.

Overall, we find that the type of bank matters. In both the full PPP sample and the sample of first-round loans, we find that banks affected the distribution of PPP loans more than did credit unions. When we examined banks more closely and looked at the effect of community banks separately, we found that a greater concentration of community banks had the largest effect on the distribution of PPP loans. Community banks focus their services in a much smaller geographic region than noncommunity banks. This focus gives small businesses in the banks’ catchment areas a better chance of obtaining a loan since there is less competition from businesses farther away. This helps explain why community banks had the greatest impact on the number of loans in a CZ.

Second Round of the PPP

The PPP was an evolving policy. After the initial allocated funds were exhausted, the program was re-authorized and nearly $300 billion more funds were added. Along with the additional funds came some changes to how the program was implemented to make it easier for small businesses to get the help they needed. The changes in the second round of PPP funding included prioritizing the smallest businesses as well as expanding the pool of authorized lenders to include more nontraditional lenders.

To ensure that our main results are not driven by the loans made during the first round of program funding, we use a subset of PPP loans that were made after April 27, 2020, when the second round went into effect. Using this subsample of loans, we test whether banking concentration in a CZ affected the number of PPP loans distributed, the dollar amount, and the number of jobs retained. We also examine whether our secondary hypotheses were impacted by changes in the program. The first column of Table 6 presents our findings for the number of loans.

Table 6.
PPP Loans per Small Business: Second Round Loans and Fintech Lenders.

(1) (2)

Round 2 loans Fintech loans

Banks + C.U.s per 10k people 0.700* −0.634*

(4.41) (−5.92)

Bank headquarters per 10k people 0.154 0.384*

(1.07) (3.71)

(Banks + C.U.s per 10k people)² −0.317 0.221*

(−2.47) (4.61)

(Bank headquarters per 10k people)² 0.074 −0.059

(0.59) (−1.23)

COVID-19 cases per 10k people 0.130* −0.021

(3.18) (−0.32)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$ 0.103* 0.076

(1.76) (1.18)

$\frac{M e d i a n H H I n c .}{M e a n H H I n c .}$ −0.051 0.177*

(−1.15) (3.84)

GDP per capita in thousands −0.062 0.012

(−1.11) (0.28)

Share of pop. with at least a B.A. 0.150* −0.158*

(3.42) (−4.23)

Non-White population share 0.204* −0.013

(4.04) (−0.27)

Share of small businesses < 10 employees −0.096* 0.210***

(−1.88) (4.95)

Observations 706 706

R ² 0.327 0.469

Abbreviations: PPP = Paycheck Protection Program; GDP = gross domestic product.

Standardized coefficients reported. Full set of controls included from Equation (1) but not reported. Robust standard errors were calculated. T-statistics in parentheses. Employment statistics are from April 2020. Small business data come from the most recent County Business Patterns (2019).

p < .1, p < .05, p < .01.

Even after the changes to the PPP, bank and credit union concentration was still an important factor for the number of loans distributed in a CZ. Those where the number of banks per 10,000 people was one standard deviation greater increased the number of loans per small business by approximately 70% of one standard deviation. This impact is less than the impact we found when using the full sample of PPP loans or the just the first round (about one standard deviation increase in the number of PPP loans per small business). There are still diminishing returns to increasing the number of banks per 10,000 people, but in the second round of loans the impact of an additional bank was greater than in the full sample of loans. Overall, CZs with more banks received more loans than places with fewer in the second round of the PPP.

In contrast to the full sample of PPP loans and the first-round sample of loans, banking headquarters did not impact the distribution of PPP loans in the second round. In the full sample and the first round, increasing the number of banking headquarters by one standard deviation increased the number of PPP loans per small business by 33.3% to 41.4% of a standard deviation. When only the second-round loans are used, the impact of a standard deviation increase is zero. This suggests that the effect for banking headquarters is entirely driven by the first round of the program. One possible explanation for the difference in findings is that after the first round, nonheadquarter banks were able to adjust their practices and make more loans, decreasing the importance of having a higher concentration of bank headquarters.

When examining areas with more advantaged labor markets, we find that in the second round of loans, these markets were not necessarily receiving more loans than less advantaged labor markets. Places where the employment-to-population ratio is one standard deviation higher received only 10.3% of a standard deviation more PPP loans per small business. This is less by about half the magnitude of the effect in the full sample of PPP loans and less than half the effect when compared to the first-round loans. However, this result is only significant at the 10% level and is sensitive as to which measure of small business we include. This suggests that more second-round loans went to places that were not as economically advantaged and may have had more pandemic-related layoffs.

CZs that had a greater share of the smallest businesses (those with fewer than 10 or 50 employees) still received fewer loans than CZs that did not have as high of a share. Places where the share of businesses with less than 10 employees is one standard deviation greater decreased the number of loans per small business by 9.6% of a standard deviation. A similar difference in the share of businesses with less than 50 employees results in a decrease of 13.3% (Appendix Table A7) of a standard deviation. This suggests that in the second round of the program the smallest businesses were still disadvantaged compared to their larger counterparts. Even though CZs with more small businesses were more disadvantaged, they fared better than in the first round as the impact for each measure of the smallest businesses is lower than in the other samples.

When looking at our secondary hypotheses regarding the share of non-White population and the prevalence of COVID-19 cases, we also find that the changes in the program before the second round resulted in improved outcomes. In the first round of the PPP, places with a greater non-White population were not statistically different than CZs with less of a non-White population; but in the second round, places with a share of the non-White population that is one standard deviation greater than another CZ received around 20% of a standard deviation more PPP loans. The changes to the second round appear to be driving the result in the full sample of loans. Similarly, in the first round, loans went to places that were not as affected by the pandemic. Places where the COVID-19 caseload was greater received fewer loans in the initial funding of the program. When only the second round of the program is considered, places where the caseload was one standard deviation greater received about 13% of a standard deviation more PPP loans per small business.

These results suggest that the adjustments made to the program before the second round impacted the distribution of loans. The impact of a greater concentration of banks in a CZ is lower than in the full and first-round samples. The importance of bank headquarters is diminished in the second-round sample as well. These second-round loans also went to places with a greater non-White population and places that were more affected by the pandemic.

Fintech Sample

One of the changes to the second round was allowing more Fintech companies to make PPP loans. Fintech lenders primarily use technology and automation to deliver financial services. The change to allow more Fintech companies made loans accessible to more small businesses. Since Fintech lenders rely more heavily on technology and automation, permitting them to make loans negates some of the advantages that businesses have in CZs with a high bank concentration. Small businesses in relative banking deserts could apply for a PPP loan through remote services. The automated decision process employed by the Fintech companies also removes the relational lending advantage that small businesses in areas of high bank concentration enjoy. These features may make the inclusion of Fintech lenders an especially important change to the PPP.

We test whether the inclusion of Fintech companies impacted the geographic distribution of loans by limiting the main sample of PPP loans to second-round PPP loans made by fintech companies. We rely on the list from Howell et al. (2021) of Fintech companies that made PPP loans and match the lenders in the full PPP data to their list. This leaves us with a sample of 728,117 PPP loans, to which we assign CZs using the address of the small business. We then test whether the geographic distribution of banks impacted this sample of loans made by Fintech companies. Results are presented in column 2 of Table 6 for the number of loans per small business.

When loans from Fintech companies only are included, we find that places with higher concentrations of banks no longer received more PPP loans per small business compared to places with lower banking concentrations. In fact, we find that these Fintech loans were distributed to CZs with lower concentrations of banks. CZs where the number of banks per 10,000 people was one standard deviation lower received 63.4% of a standard deviation more loans. These results make intuitive sense. Small businesses in CZs with more banks had more options for obtaining a PPP loan. Small businesses in banking deserts did not have many local options or may not have had a prior relationship with their local banks, making PPP loan approval more difficult. The features of Fintech lenders, particularly the remote application process, made it easier for small businesses to apply for and obtain a loan. CZs with fewer banks per 10,000 people benefited from the existence of Fintech lenders. Where the concentration of banks exceeded 19 banks per 10,000 people, the CZ received fewer Fintech-originated PPP loans. This is likely because there are more traditional banking options and small businesses do not need to rely as much on the Fintech lenders. When examining the effect of the distribution of bank headquarters, we do not find that their geographic distribution affects the number of loans made by Fintech companies in a CZ.

Unlike in the other PPP loan samples that we use, the loans made by Fintech companies did not go to small businesses in CZs with relatively advantaged labor markets. The effect of a greater employment-to-population ratio is statistically indistinguishable from zero. PPP loans that originated from traditional banks tended to go to areas that had higher employment-to-population ratios. This suggests that Fintech loans were not going to places that already had a labor market advantage. We also find that, unlike the other samples, the smallest small businesses, those with fewer than 10 employees, were not disadvantaged when focusing on PPP loans made by Fintech companies. When we expand our definition to include small businesses with up to 50 employees, we find that CZs with a greater share of these businesses were still at a disadvantage in receiving loans. CZs where the share of these businesses is one standard deviation greater received 17.3% of a standard deviation fewer loans. This effect is even stronger than in the other samples of loans. The difference between the samples could be in how these businesses are geographically distributed. More businesses with less than 50 employees may be in CZs with higher bank concentrations, which received fewer Fintech loans.

We also find that places with a greater share of non-White people received more loans that originated from Fintech lenders. CZs with a non-White population share that is one standard deviation higher received approximately 22% of standard deviation more Fintech PPP loans. Our finding is consistent with the results in Howell et al. (2021), which found that Black-owned businesses received more loans from Fintech companies than from other types of lenders. This effect is greater than in the samples of PPP loans that included loans from traditional banks, suggesting that this change to the program was the greatest factor affecting the distribution in CZs with a greater non-White population share.

More Fintech loans also went to CZs that had a higher prevalence of COVID-19 cases, and the effect is greater in the Fintech sample of loans than in the full sample of PPP loans. CZs where the number of COVID-19 cases per 10,000 people is one standard deviation greater received approximately 17% of a standard deviation more PPP loans. Making the same comparison between CZs using the full sample of loans, the number of PPP loans is only 5.1% to 6% of a standard deviation greater. Places that were more affected during the early stages of the pandemic were able to get more aid in the form of PPP loans, especially from Fintech lenders.

Conclusions

This paper's novel contribution is an empirical assessment of the PPP at the appropriate geographic scale to reflect a spatial comparison of loan disbursement against measures of regional banking concentration and labor market opportunity. In contrast to other work on these emergency business loans, we use CZs as the spatial unit of analysis and leverage an oft-ignored measure of labor market opportunity, the employment-to-population ratio. We also inquire whether flows of loans went to small businesses.

Our findings show that PPP loans went disproportionately toward more job-dense regions, effectively widening existing spatial labor market inequality. Furthermore, PPP loans flowed less toward those regions characterized by banking deserts, again reinforcing existing regional inequalities. When controlling for banking deserts, banking hinterlands also received fewer loans, highlighting the vulnerability of regional businesses to only branch banking, which appears to put such regions at a significant disadvantage relative to those featuring headquarters. The type of financial institution does matter for which CZs receive more PPP loans. Community banks seem to be the most important type of lending institution for distributing these loans, which fits with other recent work on these institutions (e.g., Petach et al., 2021). For CZs that are relative banking deserts or hinterlands, Fintech lenders are especially important as most of these loans were distributed to areas with fewer banks. The smallest businesses were systematically disadvantaged in loan distribution. Areas with greater non-White populations had no advantage in the first round of loan distribution (as described by Howell et al., 2021), but gained more access to loans in the following disbursals.

This paper's findings suggest that the PPP distribution of resources seems to have furthered regional inequality at both the banking and labor market levels. Advantaged regions received more loans overall—even after the program was modified. Running future programs like the PPP through existing banking structures will exacerbate existing inequalities, including banking deserts. A distribution of funds targeted to places most affected by hardship would help lessen some of these underlying inequalities. The community banks’ result suggests that such banks’ soft information and relationship lending is an important mechanism to more equitably spreading loans. The findings on Fintech lenders suggest that these institutions should be included in future policies as they help lessen the geographic inequalities that presently exist in banking.

Two more policy considerations emerge from this research. First, specific targeting of loans to badly scarred regions would provide support where it is most needed. In a similar spirit, allocating funds by banks serving these particularly hard-hit communities would ensure that even the less nimble, but possibly regionally important, businesses would get an unhindered channel of funding access. Future work could link these loan/banking patterns to the rural/urban divide, as well as political leanings of the region in question. Overall, the findings on banking deserts and hinterlands underscore that when designing policies to be implemented locally through banks, the spatial distribution of such financial institutions should be considered to ensure that the policy is not creating or deepening spatial inequality, as was apparently the case for the PPP program.

	Mean	Std. dev.	Min.	Max.	N
PPP loans per small business	0.672	0.204	0	2.17	706
PPP loan amount per small business	51750.003	14474.729	0	131018.211	706
PPP jobs retained per small business	5.857	1.41	0	15.075	706
Round 1 loans per small business	0.295	0.121	0	1.261	706
Round 1 loan amount per small business	36682.66	12374.749	0	97266.570	706
Round 1 jobs retained per small business	4.027	1.27	0	11.313	706
Banks + C.U.s per 10k people	4.492	2.908	0	37.831	706
Bank headquarters per 10k people	0.636	0.972	0	9.653	706
COVID-19 cases per 10k people	2.808	5.756	0	88.287	706
$\frac{E m p l o y m e n t}{P o p u l a t i o n}$	0.403	0.065	0.224	0.666	706
$\frac{M e d i a n H H . I n c .}{M e a n H H . I n c .}$	0.754	0.042	0.598	0.904	706
GDP per capita in thousands	54.045	25.504	22.774	392.175	706
Share of pop. with at least a B.A.	0.236	0.078	0.073	0.556	706
Non-White population share	0.183	0.157	0.007	0.894	706
Small businesses per 10k people	252.649	86.404	57.157	1005.714	706
Share of small businesses $<$ 10 employees	0.748	0.05	0.607	1	706
Share of small businesses $<$ 50 employees	0.962	0.017	0.911	1	706

	(1)	(2)	(3)
Banks + C.U.s per 10k people	1.002***	−0.165	−0.143
	(6.89)	(−1.29)	(−1.08)

(Banks + C.U.s per 10k people)²	−0.527***	0.007	−0.013
	(−5.76)	(0.08)	(−0.17)

Bank headquarters per 10k people	0.333**	0.096	0.211
	(2.46)	(0.71)	(1.61)

(Bank HQs per 10k people)²	−0.115	−0.084	−0.099
	(−1.24)	(−0.51)	(−0.63)

COVID-19 cases per 10k people	0.060**	0.066	0.031
	(2.32)	(1.26)	(0.61)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$	0.203***	0.255***	0.217***
	(4.08)	(3.24)	(2.67)

$\frac{M e d i a n H H I n c .}{M e a n H H I n c}$	−0.009	0.005	−0.066
	(−0.22)	(0.10)	(−1.23)

GDP per capita in thousands	−0.124***	0.072	−0.091
	(−2.84)	(0.70)	(−1.25)

Share of pop. with at least a B.A.	0.072**	0.085	−0.067
	(2.14)	(1.31)	(−1.00)

Non-White population share	0.159***	−0.018	−0.000
	(3.40)	(−0.27)	(−0.00)

Share of small businesses < 10 employees	−0.148***
	(−3.39)

PPP loans		0.247***	0.142**
		(3.02)	(2.54)
Observations	706	706	706
R ²	0.570	0.224	0.061

	(1)	(2)	(3)
Banks + C.U.s per 10k people	0.968***	0.021	0.051
	(7.05)	(0.18)	(0.42)

(Banks + C.U.s per 10k people)²	−0.564***	−0.109	−0.139*
	(−6.69)	(−1.55)	(−1.82)

Bank headquarters per 10k people	0.397***	0.209*	0.364***
	(2.96)	(1.68)	(2.91)

(Bank HQs per 10k people)²	−0.272***	−0.258*	−0.338**
	(−3.39)	(−1.75)	(−2.15)

COVID-19 cases per 10k people	−0.035*	0.006	−0.033
	(−1.80)	(0.13)	(−0.80)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$	0.229***	0.318***	0.262***
	(5.45)	(4.38)	(3.72)

$\frac{M e d i a n H H I n c}{M e a n H H I n c}$	0.038	0.014	−0.046
	(0.93)	(0.28)	(−0.88)

GDP per capita in thousands	−0.147***	0.018	−0.127***
	(−4.56)	(0.24)	(−2.75)

Share of pop. with at least a B.A.	−0.032	0.059	−0.034
	(−0.98)	(0.95)	(−0.56)

Non-White population share	0.050	−0.020	0.013
	(1.37)	(−0.34)	(0.22)

Share of small businesses < 10 employees	−0.155***
	(−4.32)

PPP loans		0.070	−0.018
		(1.28)	(−0.57)
Observations	706	706	706

R ²	0.572	0.149	0.100

	(1)	(2)
Banks + C.U.s per 10k people	0.700***	−0.634***
	(4.41)	(−5.92)

Bank headquarters per 10k people	0.154	0.384***
	(1.07)	(3.71)

(Banks + C.U.s per 10k people)²	−0.317**	0.221***
	(−2.47)	(4.61)

(Bank headquarters per 10k people)²	0.074	−0.059
	(0.59)	(−1.23)
COVID-19 cases per 10k people	0.130***	−0.021
	(3.18)	(−0.32)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$	0.103*	0.076
	(1.76)	(1.18)

$\frac{M e d i a n H H I n c .}{M e a n H H I n c .}$	−0.051	0.177***
	(−1.15)	(3.84)

GDP per capita in thousands	−0.062	0.012
	(−1.11)	(0.28)

Share of pop. with at least a B.A.	0.150***	−0.158***
	(3.42)	(−4.23)

Non-White population share	0.204***	−0.013
	(4.04)	(−0.27)

Share of small businesses < 10 employees	−0.096*	0.210***
	(−1.88)	(4.95)

Observations	706	706
R ²	0.327	0.469

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Kristopher Deming

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Kristopher Deming is a postdoctoral research associate in the Department of Agricultural, Food, and Resource Economics at Michigan State University. His research interests include regional economics, public economics, and the intersection of the two. He is particularly interested in how public policy affects entrepreneurship.

Stephan Weiler, a professor of economics and co-director of the Regional Economic Development Institute at Colorado State University, is a former Federal Reserve research officer. He studies urban, rural, and regional economic development across all their dimensions.

References

Akerlof

(1970). The market for lemons: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84(3), 488‐500. https://doi.org/10.2307/1879431

Amior

Manning

(2018). The persistence of local joblessness. American Economic Review, 108(7), 1942‐1970. https://doi.org/10.1257/aer.20160575

Amiram

Rabetti

. (2020). The relevance of relationship lending in times of crisis. SSRN Working Paper 3701587.

Atkins

Cook

L. D.

Seamans

(2022). Discrimination in lending? Evidence from the Paycheck Protection Program. Small Business Economics, 58(2), 843‐965. https://doi.org/10.1007/s11187-021-00533-1

Autor

Cho

Crane

Goldar

Lutz

Montes

Peterman

W. B.

Ratner

Villar

Yildirmaz

(2020). An evaluation of the paycheck protection program using administrative payroll microdata. 113^th Annual Conference on Taxation – National Tax Association.

Barrios

Minnis

Sijthoff

(2020). Assessing the Payroll Protection Program: a framework and preliminary results. Becker Friedman Institute – University of Chicago Working Paper 2020-63.

Bartik

A. W.

Cullen

Z. B.

Glaeser

E. L.

Luca

Stanton

C. T.

Sunderam

(2020). The targeting and impact of Paycheck Protection Program loans to small businesses. NBER Working Paper, 27623. Cambridge, MA: National Bureau of Economic Research.

Berger

Udell

(2002). Small business credit availability and relationship lending: The importance of bank organizational structure. Economic Journal, 112(477), F32‐F53. https://doi.org/10.1111/1468-0297.00682

Bunten

Weiler

Thompson

Zahran

(2015). Entrepreneurship, information, and growth. Journal of Regional Science, 55(4), 560‐584. https://doi.org/10.1111/jors.12157

10.

Conroy

Low

Weiler

(2017). Fueling job engines: Impacts of small business loans on establishment births in metropolitan and nonmetro counties. Contemporary Economic Policy, 35(3), 578‐595. https://doi.org/10.1111/coep.12214

11.

Elenev

Landvoigt

Van Nieuwerburgh

(2020). Can the COVID bailouts save the economy? NBER Working Paper, 27207. Cambridge, MA: National Bureau of Economic Research.

12.

Fairlie

(2020). The impact of COVID-19 on small business owners: Evidence of early-stage losses from the April 2020 Current Population Survey. NBER Working Paper, 27309. Cambridge, MA: National Bureau of Economic Research.

13.

Fairlie

Fossen

(2021). Did the Paycheck Protection Program and economic injury disaster loan program get disbursed to minority communities in the early stages of COVID-19? Small Business Economics, 58(2), 829‐842. https://doi.org/10.1007/s11187-021-00501-9

14.

Garcia

Darity

W. A.

(2022). Game theory of self-reporting race in small business loan applications: Evidence from PPP loans in Durham, NC. AEA Papers & Proceedings, 1-4.

15.

Granja

Makridis

Yannelis

Zwick

(2020). Did the Paycheck Protection Program hit the target? NBER Working Paper, 27095. Cambridge, MA: National Bureau of Economic Research.

16.

Howell

S. T.

Kuchler

Snitkof

Stroebel

Wong

(2021). Racial disparities in access to small business credit: evidence from the Paycheck Protection Program. National Bureau of Economic Research, 29364. Washington, DC: NBER.

17.

Hrushka

(2020). Keeping the banking desert at bay in rural America. Banking Dive. Retrieved October 10, 2021, from https://www.bankingdive.com/news/bank-deserts/572124/

18.

Joaquim

Netto

(2020). Bank incentives and the impact of the Paycheck Protection Program. SSRN Working Paper 3704518.

19.

Kapinos

(2021). Paycheck Protection Program: county level determinants and effect on unemployment. Federal Reserve Bank of Dallas Working Paper 2105.

20.

Kashian

Tao

Perez-Valdez

(2015). Banking the unbanked: bank deserts in the United States. University of Wisconsin – Whitewater. Mimeo.

21.

Strahan

P. E.

(2020). Who supplies PPP loans (and does it matter)? Banks, relationships, and the COVID crisis. NBER Working Paper, 28286. Cambridge, MA: National Bureau of Economic Research.

22.

Marsh

W. B.

Sharma

(2020). PPP raised community bank revenue but lowered profitability. Federal Reserve Bank of Kansas City Economic Bulletin. December.

23.

Morgan

Pinovsky

Perlman

(2018). The ‘banking desert’ mirage. In Liberty Street Economics. The Federal Reserve Bank of New York. January 10. https://libertystreeteconomics.newyorkfed.org/2018/01/the-banking-desert-mirage/

24.

Petach

Weiler

., & Totten, G. (2019). Geography matters: the impact of geographic expansion on bank performance during the great recession. Regional Economic Development Institute Working Paper. Fort Collins, CO: Colorado State University. Under review.

25.

Petach

Weiler

Conroy

(2021). It’s a wonderful loan: Community banking and regional economic resilience. Journal of Banking and Finance, Forthcoming. https://doi.org/10.1016/j.jbankfin.2021.106077

26.

Weiler

(2000). Pioneers and settlers in LoDo Denver: Private risk and public benefits in urban redevelopment. Urban Studies, 37(1), 167‐179. https://doi.org/10.1080/0042098002348

27.

Weiler

Hoag

Fan

(2006). Prospecting for economic returns to research: Adding informational value at the market fringe. Journal of Regional Science, 46(2), 289‐312. https://doi.org/10.1111/j.0022-4146.2006.00442.x

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.07 MB

	(1)	(2)	(3)
	Original specification	Banks & credit unions	Community banks & credit unions
Banks + C.U.s per 10k people	1.002***
	(6.89)

(Banks + C.U.s per 10k people)²	−0.527***
	(−5.76)

Banks per 10k people		1.030***
		(6.57)

(Banks per 10k people)²		−0.552***
		(−5.93)

Credit unions per 10k people		−0.019	−0.049
		(−0.39)	(−1.02)

(Credit unions per 10k people)²		0.044	0.049
		(0.97)	(1.12)

Community banks per 10k people			1.171***
			(8.87)

(Community banks per 10k people)²			−0.581***
			(−7.95)

Non-community banks per 10k people			0.367***
			(5.29)

(Non-community banks per 10k people)²			−0.247***
			(−3.57)

Bank headquarters per 10k people	0.333**	0.305**	0.183
	(2.46)	(2.12)	(1.32)

(Bank headquarters per 10k people)²	−0.115	−0.109	−0.026
	(−1.24)	(−1.14)	(−0.27)
Observations	706	706	706
R ²	0.570	0.576	0.606

	(1)	(2)
	Round 2 loans	Fintech loans
Banks + C.U.s per 10k people	0.700***	−0.634***
	(4.41)	(−5.92)

Bank headquarters per 10k people	0.154	0.384***
	(1.07)	(3.71)

(Banks + C.U.s per 10k people)²	−0.317**	0.221***
	(−2.47)	(4.61)

(Bank headquarters per 10k people)²	0.074	−0.059
	(0.59)	(−1.23)
COVID-19 cases per 10k people	0.130***	−0.021
	(3.18)	(−0.32)

$\frac{E m p l o y m e n t}{P o p u l a t i o n}$	0.103*	0.076
	(1.76)	(1.18)

$\frac{M e d i a n H H I n c .}{M e a n H H I n c .}$	−0.051	0.177***
	(−1.15)	(3.84)

GDP per capita in thousands	−0.062	0.012
	(−1.11)	(0.28)

Share of pop. with at least a B.A.	0.150***	−0.158***
	(3.42)	(−4.23)

Non-White population share	0.204***	−0.013
	(4.04)	(−0.27)

Share of small businesses < 10 employees	−0.096*	0.210***
	(−1.88)	(4.95)

Observations	706	706
R ²	0.327	0.469