Bank Relationships and the Geography of PPP Lending

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Bank Relationships and the Geography of PPP Lending David Glancy 2023-014 Please cite this paper as: Glancy, David (2023). “Bank Relationships and the Geography of PPP Lending,” Finance and Economics Discussion Series 2023-014. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2023.014. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Bank Relationships and the Geography of PPP Lending∗ David Glancy† January 19, 2023 Abstract I study how bank relationships affected the timing and geographic distribution of Paycheck Protection Program (PPP) lending. Half of banks’ PPP loans went to borrowers within 2 miles of a branch, mostly driven by relationship lending. Firms near less active lenders shifted to fintechs and other distant lenders, resulting in delays receiving credit but only slightly lower loan volumes. I estimate a structural model to fit the observed relationship between branch distance, bank PPP activity, and origination timing. I find that banks served relationship borrowers 5 to 9 days before other borrowers, an effect in line with reduced-form estimates using a sample of PPP borrowers with previous SBA lending relationships. ∗The analysis and conclusions in this paper are those of the author and should not be interpreted as reflecting the views of the Board of Governors or the Federal Reserve System. †Federal Reserve Board of Governors. Email: David.P.Glancy@frb.gov. 1

1 Introduction As part of the CARES Act, the Paycheck Protection Program (PPP) provided small businesses with over $500 billion in forgivable loans during the early stages of the COVID-19 pandemic. The size and speed of the rollout were unprecedented; most of the credit was extended during the first month of the program despite numerous hurdles, including lastminute policy changes, operational difficulties related to the effects of the pandemic, and the rapid exhaustion of the initial funding allocation. To manage these hurdles, banks reportedly focused on providing credit to existing customers, for whom less due diligence is required before lending. This behavior raises the question: What happened when relationship credit was less available? In this paper, I use geocoded data on the locations of bank branches and PPP borrowers to study the spatial distribution of PPP loan originations and how it relates to origination timing. As the majority of banks’ small business lending is concentrated within a short distance of their branches, the focus on preexisting relationships caused PPP lending to predominantly fall within banks’ branch footprints. Studying the location of PPP lending— and what happens when credit is less available from local lenders—is thus informative as to the role bank relationships played in supporting lending. I present three findings pertaining to the relationship between borrower proximity, bank PPP activity, and origination timing. First, PPP lending was highly localized, particularly for relationship-oriented banks.1 About half of bank loans overall, and two-thirds of loans from relationship-oriented banks, went to borrowers within 2 miles of a branch. Less relationship-orientedbanksdidmorenonlocallending. ThisresultsuggeststhatPPPlending was localized because prepandemic small business lending was. Between government guarantees removing the information-based advantages of local banks, and pandemic restrictions to on-site activities limiting the benefit of branch access, there was little reason to choose 1Relationship-oriented banks are banks where PPP lending was less than prepandemic small business lending. Banks for which this is not the case were more likely to have expanded significantly beyond relationship borrowers. 2

a nearby lender for a PPP loan absent an existing relationship. Consequently, the different geographic distribution of relationship and nonrelationship PPP lending can be exploited to examine the benefit of bank relationships. Second, there is a U-shaped relationship between a bank’s PPP activity and the average time a borrower needs to wait to receive a loan. The banks providing the shortest wait times are those whose total volume of PPP lending was comparable to the stock of prepandemic small business loans. Borrowers tended to face longer delays when their bank was less active in PPP (indicating bank-specific lending frictions) or very highly active (meaning a high share of borrowers were not existing clients). Third, borrowers who obtained PPP loans from nearby lenders received credit earlier, particularly if their lender was more active in the program. Active banks predominantly served nearby borrowers early in the program before moving on to other borrowers at later stages. Banks with more modest participation concentrated their lending around branches for the entirety of the program, suggesting these banks did not expand beyond existing clients. Ultimately, the results suggest that there is a benefit to having a relationship with an active PPP lender. I rationalize these three findings in a model where banks are heterogeneous in their PPP lending technology, borrowers face time costs to switching banks, and the spatial distribution of lending differs for relationship and nonrelationship loans. Banks with better processing technology originate loans faster and retain more of their relationship clients. However, once relationship clients are served, better technology then facilitates more nonrelationship lending, generatingtheU-shapedrelationshipbetweenPPPintensityandaverageorigination time. Nearby borrowers from active lenders receive credit earlier, on average, reflecting these loans disproportionately being relationship loans. When I estimate the model relating origination timing to the lending bank’s proximity and PPP intensity, I find that nonrelationship borrowers had to wait an extra 5 to 9 days to receive credit, on average. Finally, I present two additional pieces of evidence to validate the findings of the model. 3

First,Istudythebenefitsofrelationshipsforasampleofloanswherepreexistingrelationships are observable. Specifically, I match borrowers who previously received SBA 7(a) loans with borrowers in the PPP data. For this sample, I can observe whether borrowers took out loans from their previous SBA lenders, and can thus assess the delays that come from switching lenders. The results are generally consistent with the model-estimated switching costs, where the likelihood of being a relationship lender is inferred by geographic location instead of directly observed. Second, I show that geographic differences in average origination timing are consistent with the hypothesized benefits of relationship lending. If there is a benefit to having a relationship with an active lender, and small business relationships are overwhelmingly with banks with nearby branches, then the benefits of a bank being active in the program should belocalizedarounditsbranches. Ishowthatcensustractsnearlessactivebanksfaceddelays in receiving loans. In these tracts, firms shifted borrowing towards fintechs or other distant lenders, consistent with Erel and Liebersohn (2020). However, there is only a small effect on loan volumes, indicating that borrowers ultimately were able to receive credit, albeit with a delay. Taken together, these findings highlight the importance of bank relationships in receiving PPP loans. Most borrowers got credit from nearby banks, suggesting they borrowed from banks with whom they had an existing relationship. Borrowers faced delays when credit was less available from local lenders, pointing towards frictions in substituting away from relationship lenders. These results demonstrate an important cost to intermediating aid through the bank sector. While banks were able to distribute funds rapidly, they did so by prioritizing existing clients, thus disadvantaging firms with weaker banking ties. As such firms may have fewer alternatives for offsetting revenue shortfalls—for example credit line draws—frictions in accessing PPP loans were likely highest for the firms most in need of aid. 4

1.1 Background and Related Literature The COVID-19 pandemic caused an unprecedented increase in unemployment and temporary business closures starting in March of 2020 (Bartik et al., 2020). In response, Congress passed the Coronavirus Aid, Relief and Economic Security (CARES) Act on March 27th, a $2.2 trillion stimulus bill. One of the the largest components of the bill was the Paycheck Protection Program, which provided low-interest loans to businesses with 500 or fewer employees. Firms were able to request credit for up to 2.5 times monthly payroll, up to a maximum value of $10 million. Loans were forgivable so long as businesses maintained employment levels and payrolls. The program launched on April 3rd, a mere week after the signing of the CARES act. The grant-like nature of the program resulted in very high demand, causing the initial $349 billion allocated to the program to be exhausted by April 16th. An additional $320 billion in funding was provided, enabling the Small Business Administration (SBA) to begin taking applications again on April 27th. After another rapid period of origination activity in the following week, the pace of originations slowed. The program stopped taking applications on August 8th, at which time the program had disbursed $525 billion in credit from 5,460 different lenders.2 The size of the program and the speed with which it was rolled out caused difficulties upon the initial launch. Final guidance on the program wasn’t released until 12 hours before PPP went live, and “know your customer” requirements hindered the onboarding of new clients (Merker et al., 2020). At the same time, many banks were contending with the effects of branch closures or the shift to work-from-home, causing further operational complications. Against this backdrop, banks reportedly prioritized their own clients in extending PPP loans. For these borrowers, less additional due diligence was needed to originate a loan. Furthermore, prioritizing existing clients reduced the risk of alienating clients and losing 2The PPP reopened on January 11, 2021 after a third round of funding was provided. However, this paper focuses on lending done in the first two rounds. 5

valuable relationships (Joaquim and Netto, 2020). However, banks’ chosen allocation of credit may have had undesirable effects, as credit flowed to many borrowers less in need of aid, possibly at the expense of borrowers with weaker banking ties. Indeed, the early rush of PPP lending disproportionately went to less distressed firms (Bartik et al., 2020) and areas less affected by the pandemic (Granja et al., 2022). While bank relationships appear to be important in receiving PPP credit, studying this topic is complicated by data limitations. Little information is available about lending relationships for the small, private firms that account for the vast majority of PPP loans. To overcome this difficulty, some papers either use data from surveys (Bartik et al., 2020) or public firms (Amiram and Rabetti, 2020; Duchin et al., 2022) to analyze subsets of PPP borrowers for which information on relationships are available.3 Other papers use county or ZIP code bank branch locations to proxy for relationships, allowing authors to study the full universe of PPP lending, at the cost of using less direct relationship measures (Erel and Liebersohn, 2020; Li and Strahan, 2021; Granja et al., 2022).4 I add to this literature by studying the effects of banks’ PPP involvement at a fine spatial scale. Proxying for preexisting relationships based on branch proximity allows me to analyze the full universe of PPP lending, instead of relying on a small subsample of loans for which relationships are observable (though I also validate the estimates with such a subsample). Additionally, the fine spatial scale also allows me to more narrowly identify probable relationship lenders than other papers that analyze geographic differences in supply conditions.5 Finally, the structural model allows me to use these spatial patterns to quantify 3Bartik et al. (2020) use survey data to show that banks are more likely to approve loans to relationship borrowers. Amiram and Rabetti (2020); Duchin et al. (2022) use data from public firms in the PPP to document that relationship borrowers are more likely to get credit and get credit faster. 4ErelandLiebersohn(2020)showthatfintechlendershaveahighermarketshareinZIPcodeswithfewer branches or a larger minority share of the population. Li and Strahan (2021) show that banks allocate more PPPloanstocountieswheretheydidmoresmallbusinesslendingbeforeCOVID.Granjaetal.(2022)show that ZIP codes with more active banks had higher levels of PPP lending, though employment effects were modest. 5I estimate that lenders only a few miles away are significantly less likely to be relationship lenders than those within a mile, indicating that there is a benefit to conducting analysis at a level finer than ZIP code or county. 6

the benefits of relationship lending and magnitude of switching costs. 2 Data and Summary Statistics 2.1 Data Sources PPP loan data Loan level data on PPP originations are provided by the Small Business Administration (SBA). For each loan, the data include the name of the lender, various borrower characteristics (address, NAICS code, number of jobs reported) and various loan characteristics (loan amount, origination date). To analyze the geography of originations, I use a geocoded version from geocod.io which contains the latitude and longitude of the borrower addresses in the SBA data.6 Distance to lending bank I combine the data on borrower locations with data on bank branch locations from the FDIC’s Summary of Deposits (SOD). The SOD includes the latitude and longitude of US bank branches as well as branch-level deposits as of June 2020. A key object of interest in the loan level data is the distance between the borrower and lender. I use the following process to calculate this distance. First, I conduct a fuzzyname match between originator names in the PPP data and institutions in the National Information Center’s (NIC) Institution Directory to find the RSSD ID of the originating lender.7 I use this unique identifier to match PPP loans to the branch locations of the lending bank in the SOD and compute the distance to the closest branch of that bank.8 89% of PPP loans match to the SOD data, with nearly all of the unmatched loans coming from nonbanks. 6This data is available here: https://www.geocod.io/geocoded-ppp-loan-data/. 7Some lender names in the PPP data can match to multiple banks. I disambiguate by selecting fuzzy name matches that also match on ZIP code, then state, then finally selecting the prospective match with the highest PPP lending volume. 8Distance is calculated using the haversine formula. The minimum distance is found using the BallTree module from scikit-learn. 7

Bank PPP intensity Theextentofabank’sPPPactivityismeasuredbytheratioofPPP lendingtoprepandemicsmallbusinesslendingrelationships. InthemodelinSection4, banks with superior lending technology are able to process loans more quickly and retain/attract moreborrowersrelativetotheirstockofexistingrelationships. Toreflectthis, ImeasurePPP intensity as the ratio of 2020 PPP lending for bank b to a weighted average of outstanding small business loans and outstanding C&I loans in the 2019:Q4 Call Reports.9 As the former likely excludes many eligible borrowers, while the latter includes loans to many firms that were too large to participate in PPP, my proxy for the relationship stock is an average of the two loan balance measures. I place a 90% weight on small business loans to equate the aggregate for the relationship stock proxy with total PPP lending.10 Namely, the measure of PPP activity is: PPP b PPP Intensity = b .9×SB Loans +.1×C&I Loans b,19:Q4 b,19:Q4 Tract level data The loan level analysis uses controls for tract level labor market and small business loan market conditions from Longitudinal Employer-Household Dynamics Origin-Destination Employment Statistics Workplace Area Characteristics (LODES) and Community Reinvestment Act (CRA) data, respectively. LODES provides data on employment for 2010 census blocks, which I aggregate to the tract level.11 The key variables of interest are the share of employment that is nonwhite, the earnings distribution of local employees, and total employment in firms with 500 or fewer employees (generally reflecting employment in PPP-eligible firms).12 CRA includes information on the volume of loans to 9Small business loans are C&I loans with an original balance under $1 million dollars. 10This weighting simplifies the interpretation of some results, as it centers ln(PPP Intensity) around 0. However, this decision regarding the supply measure matters little; estimates are similar when measuring supply by PPP originations to total assets, PPP originations to small business loans, or using the PPP exposure measure of Granja et al. (2022). 11I use the data on private jobs in 2017, as this is the most recent year that reports employment disaggregated by firm size. Alaska is missing for 2017, so I instead use 2016 data there. 12Earnings variables include the share of jobs in the tract earning $1250 or less per month, the share of jobs earning over $3333 per month, and the share of jobs by workers with at least a bachelor’s degree. Since PPP loan sizes are pinned to monthly payroll, these variables along with the number of employees in 8

firms with less than $1 million in revenue by census tract.13 Additionally, for tract level analysis, I aggregate ln(PPP Intensity ) to compute the b,−j average characteristics of banks within 2 miles of census tract j.14 Using the geocoded branch locations and the locations of tract centroids from 2010 Census Gazetteer files, I find the set of bank branches located within 2 miles of a tract centroid. The tract level bank supply measure is the average of ln(PPP Intensity ) for nearby branches, weighting by b,−j the deposits in those branches. As prepandemic small business lending was highly localized, this variable measures the likely availability of credit from relationship lenders in an area, and enables the analysis of how geographic differences in credit availability affected PPP outcomes at a fine spatial scale.15 2.2 Summary Statistics Table1presentssummarystatisticsofthemainvariablesofinterest, withloanlevelstatistics in Panel 1 and tract level statistics in Panel 2. The loan level data show that most lending is local, with about half of bank loans coming from lenders with a branch within 2 miles. However, there is a mean distance of 175 miles, reflecting a non-negligible share of loans being made independent of geography. Regarding the supply measure, the median PPP intensity of the lending bank was almost 1.5 (PPP lending that is 50% greater than the stock of relationship loans). However, this reflects high-intensity banks appearing as lenders more frequently; the median is slightly above 1 at the bank level. While a significant share of lending occurred in the early weeks of the program, this was due in part to larger loan sizes during that time. The median bank borrower had to wait 26 PPP-eligible firms are important controls in specifications predicting PPP lending. 13I use data from 2019 in order to reflect prepandemic small business lending relationships. The data is only reported by banks with at least $1.3 billion in assets, and thus does not reflect the extent of small business lending relationships with the smallest banks. 14PPP Intensity is the ratio of PPP lending outside of tract j to prepandemic small business lending. b,−j 15Outcomes considered include: PPP loan volumes, average days until origination, and the shares of PPP lending accounted for by fintech lenders or local lenders. 9

days after the start of the program to receive credit and only about 31% of bank borrowers got loans before the first round of funding ran out. 10

Table 1: Summary Statistics Variable Description Mean sd Percentile N 25 50 75 Panel 1: Loan level Statistics Distance Milestonearestbranchoflendingbank 174.78 519.33 0.80 2.03 8.44 4,548,796 i,b LocalBranch 1ifDistance <2miles 0.50 0.50 0.00 0.00 1.00 4,548,796 i,b i,b PPPIntensity PPP/(.9SBL +.1CI) 18.59 76.40 0.84 1.46 2.67 4,549,275 b b b b ln(PPPIntensity) LogarithmofPPPIntensity 0.64 1.54 -0.17 0.38 0.98 4,549,275 b b Top4 1iflenderisJPM,WFC,CorBAC 0.19 0.39 0.00 0.00 0.00 4,549,275 b DaystoOrigination DaysbetweenoriginationdateandApril3rd 29.32 25.65 11.00 26.00 28.00 4,549,275 i,b Round1Indicator 1iforiginatedbyApril16th 0.31 0.46 0.00 0.00 1.00 4,549,275 i,b ln(LoanAmount) LogarithmofPPPloanbalance 10.32 1.46 9.34 10.13 11.23 4,549,274 i,b SmallFirm 1ifbusinesshasfewerthan5employees 0.54 0.50 0.00 1.00 1.00 4,549,275 i,b Urban 1ifbusinesshasurbanlocation 0.80 0.40 1.00 1.00 1.00 4,549,275 i,b Panel 2: Tract level Statistics ln(PPPIntensity) PPPintensity(exlcudingj)oflocalbranches -0.10 0.49 -0.41 -0.11 0.19 54,292 j Top4Share Top4depositsharewithin2miles 0.35 0.31 0.00 0.32 0.58 54,297 j DaystoOrigination AveragedaysbetweenoriginationdateandApril3 35.36 12.38 26.90 33.28 41.18 53,889 j Round1Share PercentofloansoriginatedbyApril16 26.32 16.94 12.90 24.03 37.93 53,889 j LocalPPPShare PercentofPPPLendingbybankswithin2miles 48.87 25.83 28.03 49.13 69.62 53,145 j FintechPPPShare PercentofPPPLendingbyfintechs 8.01 12.16 1.34 3.72 9.10 50,629 j ln(TotalPPP) LogarithmofPPPlendingintract 15.08 1.36 14.26 15.16 15.98 53,889 j ln(Branches) Logarithmofbrancheswithin2miles 2.03 1.06 1.39 2.08 2.71 54,297 j ln(SBLending) Logarithmof2019smallbusinessLending 6.52 1.27 5.80 6.69 7.40 50,089 j ln(SBEmp) Logarithmofsmallbusinessemployment 6.25 1.14 5.58 6.32 7.01 53,604 j Nonwhite Nonwhiteshareofemployment 0.24 0.17 0.12 0.21 0.32 53,620 j Earnings>$3333 Sharejobsearning>$3333month 0.32 0.15 0.22 0.30 0.41 53,620 j Earnings<$1250 Sharejobsearning<$1250month 0.30 0.12 0.22 0.30 0.37 53,620 j College Shareworkersw. collegedegree 0.25 0.08 0.19 0.24 0.30 53,617 j 11

3 Distance, PPP Intensity, and Origination Timing This section documents three findings pertaining to the relationship between distance, the PPP intensity of the lending bank, and the time until origination. First, I show that PPP lending was highly localized, particularly for relationship-oriented banks. Second, I document a U-shaped relationship between PPP intensity and time-to-origination. Third, I demonstrate that borrowers received credit earlier if they are close to a branch of the lending bank, particularly if it has a high PPP intensity. 3.1 Relationships and Distance Figure 1 plots histograms of the distance between PPP borrowers and the nearest branch of the lending bank. The left panel shows results combined across all bank loans. The density falls rapidly as distance rises and starts to level off at around two or three miles. About half of loans are from banks within 2 miles of the borrower. However, there is a long right tail, with about 23% of loans being made by banks that do not have a branch within 10 miles. 39% of these distant loans were provided by three banks—Celtic Bank, Cross River Bank and WebBank—all of whom utilized fintech partnerships to provide loans. Overall, PPP loans have a similar spatial distribution to prepandemic small business lending, with a bit more mass in the right tail driven by these online lenders.16 That PPP lending remained highly localized is interesting as many of the advantages of being a local lender do not apply to PPP loans. Normally, lending is hypothesized as being local because proximity either provides an information advantage in evaluating loans (Agarwal and Hauswald, 2010) or reduces transportation costs involved in originating/monitoring loans (Degryse and Ongena, 2005). PPP loans are fully guaranteed by the U.S. government, removing advantages related to local information or monitoring. Moreover, restrictions on branchoperationsduetothepandemicweakenedthebenefitofeasyphysicalaccesstoabank 16Brevoort and Wolken (2009) document that the median distance between small firms and the bank servicing them is about 3 miles, with nearly 90% of the lenders being within 30 miles. 12

Figure 1: Distribution of Lender Distances 0.40 0.30 0.20 0.10 0.00 ytisneD 0.50 0.40 0.30 0.20 0.10 0.00 1 3 5 7 9 10+ Distance (miles) (a) Distribution of Distances ytisneD Low Intensity Bank High Intensity Bank 1 3 5 7 9 10+ Distance (miles) (b) Distribution by Bank PPP Intensity Notes: This figure plots histograms of the distance between a PPP borrower and the address of the nearest branch of the lending bank. The area to the right of the dotted line gives the mass of loans made by lenders more than 10 miles away. The left panel pools all bank PPP loans, while the right separates banks with a high and low PPP intensity (the blue and red bars, respectively). branch. That PPP lending remained strongly tied to banks’ branches despite the advantages of proximity evaporating indicates that relationship lending drove PPP originations. That is, banks continued to lend locally because that is where existing clients were, even though the reasons for favoring nearby firms were no longer relevant. The right panel of Figure 1 provides more direct evidence for local lending being relationship driven. While the relationship status of a given loan is hard to ascertain, if a bank has a high ratio of PPP lending to prepandemic small business lending, then their PPP portfolio likely contains a higher share of nonrelationship loans. The histogram plots the distance distribution separately for loans from banks with a PPP intensity under 1 (which presumably predominantly served existing clients), and those with an intensity above 1 (whose portfolios would contain more nonrelationship loans). The results are consistent with relationship loans being more localized. Low-intensity bankshaveamuchhighershareofloansgoingtoborrowerswithintwomiles(68%, compared to only 40% for high-intensity banks). High-intensity banks are slightly more likely to originate loans that are between 3 and 10 miles away, and much more likely to originate 13

loans that are over 10 miles away. These results are not due to regional differences, for instance due to density, as results are similar using the residual distance after accounting for loan controls and census tract fixed effects (see Appendix Figure A1). 3.2 PPP Intensity and Origination Timing How does PPP intensity relate to origination timing? There are several mechanisms at play, working in different directions. On one hand, as discussed in the previous subsection, high-intensity banks likely did a higher share of their lending to nonrelationship customers. As relationship loans are quicker to originate—search costs have been paid and initial due diligence already conducted—banks that just focused on existing relationships may be able to process loans faster. On the other hand, banks with superior processing technology may originate loans faster and thus attract more clients, resulting in wait times that decrease in PPP intensity. Figure 2 presents the results of binscatter regressions of time-to-origination on the logarithm of the lending bank’s PPP intensity.17 The left panel is unweighted, while the right weights by the inverse of the number of loans from the lending bank in order to reduce the influence the largest lenders have on the estimates. The figures show a U-shaped relationship between a bank’s PPP intensity and the time borrowers have to wait to receive a loan. Banks with an ln(PPP Intensity) slightly above 0—meaning PPP lending just exceeded prepandemic small business lending—generally provided the shortest average wait times. Banks with PPP lending well above or below this level tended to have much longer average wait times. This pattern is consistent with the aforementioned competing mechanisms. Banks who did less PPP lending than prepandemic small business lending likely predominantly served relationship clients, meaning that marginal differences in PPP activity provides little in- 17BinscattersareconstructedusingtheBinsregcommandinStata(Cattaneoetal.,2019),residualizingon loansize,asmallfirmindicator,censustractcharacteristics(anurbanindicator,theshareofemployeesthat are nonwhite, and the logarithms of prepandemic small business employment and small business lending), and county fixed effects. 14

Figure 2: Wait Times by PPP Intensity 50 45 40 35 30 25 20 noitanigirO ot emiT 30 28 26 24 22 20 -2 0 2 4 6 ln(PPP intensity) (a) Unweighted noitanigirO ot emiT -3 -2 -1 0 1 2 3 ln(PPP intensity) (b) Inverse Count Weighted Notes: This figure plots binscatter regressions of time-to-origination on the log PPP intensity of the lending bankfor2020bankPPPloanoriginations. Controlsincludethesizeoftheloan,anindicatorforwhetherthe firmhasfewerthan5employees,censustractcharacteristics(anurbanindicator,theshareofemployeesthat are nonwhite, and the logarithms of prepandemic small business employment and small business lending), and county fixed effects. The left panel is unweighted, while the right panel weights by the inverse of the number of PPP loans from the lending bank. formation about relationship status. If most loans are relationship loans, we would expect a downward sloping relationship between PPP intensity and time-to-origination, as faster lenders are able to retain more of their clients. However, for banks with a PPP intensity above 1, increases in PPP intensity indicate more nonrelationship loans, and thus delays in theaveragetimeofloanreceipt. Thisexplanationfor the U-shapedrelationshipisformalized in the model in Section 4. 3.3 Distance and Origination Timing The final evidence motivating the model pertains to the relationship between distance and origination timing. Section 3.1 showed that relationship-oriented PPP lenders concentrated lending within two miles of their branches. Section 3.2 showed that the relationship between PPP intensity and origination timing was non-monotonic, likely due to a higher share of nonrelationship loans at the largest PPP lenders. Now I analyze the interaction of distance and proximity; if the delays receiving loans at high-intensity banks were driven by nonre- 15

lationship loans, such effects should be moderated by proximity (as such nearby loans are more likely to be relationship loans.) Figure 3 repeats the binscatter regressions from Figure 2, but disaggregating by whether loans are to borrowers within 2 miles of a lender. The figures demonstrate that the upward sloping relationship between wait times and PPP intensity are generally driven by non-local lending. Wait times are essentially monotonically decreasing in PPP intensity for nearby borrowers, but U-shaped for more distant borrowers. For banks with a low PPP intensity, there is not much of a difference in origination timing for local and non-local borrowers. This result is consistent with timing differences across originators reflecting relationship lending; if a banks seems to not have expanded lending beyond relationship clients, proximity is less informative about relationship status. However, banks with PPP lending similar to or exceeding estimates for their relationship stock are more likely to have a notable share of nonrelationship loans, making proximity more informative. For banks with a log PPP intensityaround1(meaningPPPlendingthatisalmosttripletherelationshipstock), nearby borrowers receive loans about 4 days before distant borrowers. These results are also apparent in regressions. Appendix Table A1 presents estimates from regressing wait times on ln(PPP intensity), an indicator for whether a borrower is within 2 miles, and interactions of ln(PPP intensity) with itself and the local dummy.18 Overall, borrowers from nearby banks get credit about a day earlier than other borrowers, with the benefit of using a local bank increasing in the PPP intensity of that bank. The quadratic term corroborates that there is a U-shaped relationship between PPP intensity and origination timing. Expected wait times are minimized at a PPP intensity around 2.3 for distant borrowers and 4.0 for nearby borrowers (1.4 and 4.8 in unweighted regressions). Patterns are similar if origination timing is measured by an indicator for whether the loan was originated before the first round of funding was exhausted (Appendix Table A2). 18All specifications include county and 3-digit NAICS fixed effects, and control loan size, whether the businesshasfewerthan5employees,andtractcharacteristics(anurbanindicator,theshareofemployeesthat are nonwhite, and the logarithms of prepandemic small business employment and small business lending). Regressions are either unweighted, or weight by the inverse of the number of PPP originations at the bank. 16

Figure 3: Wait Times by Bank Proximity and PPP Intensity 50 45 40 35 30 25 20 noitanigirO ot emiT Distant Bank 30 Local Bank 28 26 24 22 20 -2 0 2 4 6 ln(PPP intensity) (a) Unweighted noitanigirO ot emiT Distant Bank Local Bank -3 -2 -1 0 1 2 3 ln(PPP intensity) (b) Inverse Count Weighted Notes: This figure plots binscatter regressions of time-to-origination on the log PPP intensity of the lending bankfor2020bankPPPloanoriginations. Controlsincludethesizeoftheloan,anindicatorforwhetherthe firmhasfewerthan5employees,censustractcharacteristics(anurbanindicator,theshareofemployeesthat are nonwhite, and the logarithms of prepandemic small business employment and small business lending), and county fixed effects. The left panel is unweighted, while the right panel weights by the inverse of the number of PPP loans from the lending bank. Red dots pertain to loans to borrowers within 2 miles of a branch of the lending bank, while hollow blue dots pertain to loans to more distant borrowers. 3.4 Discussion To summarize the results so far: (1) nonrelationship PPP lending was more geographical dispersed than relationship lending, making proximity a useful proxy for a loan’s relationship status; (2) banks with a very high amount of PPP lending relative to prepandemic small business lending had longer wait times, consistent with there being delays associated with switching lenders; and (3) wait times decrease in PPP intensity for nearby borrowers, suggesting that there is a benefit to having a relationship with an active lender, even if there isn’t a clear benefit from simply borrowing from one. This set of results is consistent with banks prioritizing existing relationships first, and the more active PPP lenders then moving on to new clients later. Figure 4 presents evidence that this is the case. The figure plots the share of loans made to local borrowers by bank PPP intensity (defined by whether or not PPP intensity is above 1), as well as the market share of the low-intensity banks, over time. Banks that were less active in PPP (in red) predominantly served nearby clients throughout the program, with the share of loans to 17

Figure 4: Share of Local Lending over Time 0.80 0.60 0.40 0.20 0.00 sreworroB lacoL ot snaoL fo erahS 01apr2020 01may2020 01jun2020 01jul2020 01aug2020 Date High PPP Intensity Low PPP Intensity Market share of Low Intensity Banks Notes: This figure shows the share of loans by date that is made by banks with branches within 2 miles of theborrower. Redcrossesplotresultsforlow-intensitybanks(PPPintensityunder1),whilebluecirclesplot results for high-intensity banks. The grey area shows the share of loans made by low-intensity banks. Loans originated within 2 days of the date listed on the x-axis are pooled in order to reduces noise, for example, due to low volumes during weekends. borrowers within 2 miles typically staying in the 60 to 80 percent range. However, these banks were comparatively slower to extend credit; their market share remained below 20 percent for most of the first funding round, and picked-up later. Lending by high-intensity banks (in blue) started out being fairly localized, with about half of loans going to nearby borrowers, however this local share dropped in the early stages of the second round and typically remained around 25 percent during the later stages of the program. Overall this figure paints a similar picture to other results. Namely, existing clients of major PPP lenders receive credit earlier, as evidenced by these lenders’ high market share and focus on nearby borrowers early in the program. Those borrowing from less active banks faced delays, as evidenced by the low market share of such banks in the first round. Finally, those borrowing from nonrelationship banks faced delays, as evidenced by the higher rates of nonlocal lending later in the program. 18

4 Theory This section presents a model to explain the patterns documented in Section 3. I estimate the model to fit observed differences in the timing and spatial distribution of PPP loan originations across banks. I then use these estimates to quantify the costs of switching away from relationship lenders. 4.1 Model Supply of PPP Loans Banks had little discretion over the terms of PPP loans; loan terms were set by the program and were essentially uniform across borrowers, besides loan size, which was mechanically determined by a firm’s 2019 payroll. Consequently, I model banks as competing on how quickly loans are processed. Suppose that bank b takes on average T (L) = τ ( L )γ days to process their first L b b R b PPP loans, where τ measures the time that it would take to process their R relationship b b clients and γ reflects congestion effects. γ = 0 means that all loans are processed in τ days b regardless of where they are in the queue, while γ = 1 corresponds to loans being processed sequentially at a rate of R b. 2τ b Banks serve existing relationship clients first, before moving on to nonrelationship clients, and do not distinguish between borrowers within these pools. As a result, relationship and nonrelationship clients will have different expected wait times, denoted TR and TN, respecb b tively. In equilibrium, since banks prioritize existing clients, any bank that does nonrelationship lending will also be able to retain all of their existing clients.19 Consequently, the amount of relationship PPP lending will be min{L ,R }, where L is a bank’s total PPP b b b lending. The lending technology then implies that the expected processing time for a relationship loan will be TR = τ (min{L b ,R b })γ. TN will be such that the average processing time b b R b b for L loans is the average for relationship and nonrelationship borrowers, weighted by their 19Since borrowers choose the nonrelationship lender with the best processing time, TN is equalized for all b banks that do nonrelationship lending. Relationship prioritization results in TR <TN, so borrowers would b b never switch from a bank that does nonrelationship lending. 19

respective shares. Defining ρ ≡ L b as the ratio of PPP loans to relationship borrowers, TN b R b b must satisfy: τ ργ = ρ−1τ +(1−ρ−1)TN if ρ > 1. (1) b b b b b b b where τ and TN are the processing times for relationship and nonrelationship loans, respecb b tively, and ρ−1 is the share of loans to relationship borrowers. b Demand for PPP Loans We now derive the number of loans made by a bank as a function of their processing times. If a firms decides to leave their relationship bank, they choose thelenderwiththelowestexpectedprocessingtime. Inequilibrium, allnonrelationshiploans will have the same expected processing time, denoted T. Demand from relationship borrowers is determined by borrowers’ decision to accept either the time offered by their relationship bank or the processing time of T available to nonrelationship borrowers. Borrowers face a heterogeneous switching cost, which creates a markup µ ≥ 1 over T in the time to receive a nonrelationship loan, reflecting switching and i on-boarding time. Let µ be pareto distributed with shape parameter α (i.e. µ has a cdf i i (cid:16) (cid:17)α 1− 1 for µ ≥ 1). If τ > T, the bank cannot process all of their relationship borrowers µ b in a timely enough manor to induce them all to stay with the bank. In this case, the volume of lending will be such that: T L = R ×Pr(TR < µ T) = R ( )α if ρ < 1. (2) b b b i b TR b b Equilibrium Equilibrium loan volumes and processing times are determined by the intersection of the supply curves (determining the time to process a given number of loans) and the demand curves (determining the number of borrowers who choose to take out a PPP loan given a particular processing time). If τ > T, the bank operates on the downward sloping portion of the demand curve, as b some relationship clients with low switching costs leave the bank. Inverting equation (2), we 20

−1 get that the time to process relationship loans is TR = ρ α T where ρ < 1. b b b If τ < T, all relationship borrowers stick with their bank and receive an expected time b to origination τ . For these banks, the number of loans is determined by how many loans the b bankiscapableofmakingwhileofferingaprocessingtimeofTfornonrelationshipborrowers. Substituting T in for TN in equation (1), we can solve for τ = TR = ρ b −1 T.20 b b b ρ1+γ−1 b These expressions all derive the expected time that it takes to receive a relationship loan as a function of ρ : b   −1  ρ b α if ρ b < 1 TR = T × (3) b    ρ1 ρ + b − γ− 1 1 if ρ b > 1 b The expected time to receive a nonrelationship loan is µ¯T, where µ¯ is the expected switching cost for borrowers that switch lenders and T is the equilibrium time to process nonrelationship loans. Distance None of the expressions for time to origination by bank PPP intensity and relationship status explicitly depend on distance. Instead, where distance comes in to play is asasignalthataparticularborrowerisarelationshipclient. MotivatedbyFigure1, Iassume thatrelationshipandnonrelationshiploanshavedifferentgeographicdistributions. LetfR(d) and fN(d) denote the distribution of distances between banks and their relationship and nonrelationship clients, which are assumed to be independent of ρ and satisfy the monotone b likelihood ratio property (MLRP) λ(cid:48)(d) > 0 for λ(d) ≡ fN(d). In words, borrowers that are fR(d) further from a bank are more likely to be nonrelationship borrowers. We are interested in solving for the probability that a loan is to a relationship borrower based on the distance to the lending bank (d ) and the share of PPP loans going to rei,b lationship borrowers (min{ρ−1,1}). Letting R denote the event that i has a relationship b i,b 20TR is indeterminate if ρ = 1, taking values in the range [ T ,T], reflecting the range of τ that is low b b 1+γ b enough to serve all relationship customers, but high enough to not attract nonrelationship customers after. Since there are no observations in the data with this intensity, it does not matter for the estimation. 21

with b, we get the following probability for R : i,b f(d |R ,ρ )P(R |ρ ) i,b i,b b i,b b P(R = 1|d ,ρ ) = i,b i,j b f(d |ρ ) i,b b (4) (cid:16) (cid:17)−1 = 1+(ρ −1)λ(d ) if ρ > 1 b i,b b P(R = 1|d ,ρ ) = 1 if ρ < 1 since the bank only does relationship loans. i,b i,j b b Combining equations (3) and (4), we get the following expression for the expected time to loan receipt as a function of a bank’s PPP intensity and the distance to the lending bank:  −1    ρ b α if ρ b < 1   E(T |d ,ρ ) = T × (cid:16) (cid:17)−1 (cid:16) ρ b −1 (cid:17) (5) i,b i,b b µ¯− 1+(ρ −1)λ(d ) µ¯− if ρ > 1  b i,b ρ1+γ −1 b     (cid:124) P(R (cid:123) | (cid:122) d ,ρ ) (cid:125) (cid:124) (cid:123) b (cid:122) (cid:125) i,b i,b b BenefitofRelationship Relationship to Empirical Results Equation(5),combinedwiththeMLRPassumption for λ(d), delivers the three empirical results presented in Section 3. First, PPP lending is morelocalizedforrelationship-orientedbanksasMLRPimpliesthatthedistancedistribution fornonrelationshipborrowers, fN(d),first-orderstochasticallydominatesthedistributionfor relationship borrowers, fR(d). This assumption thus implies more localized lending for the banks that do not extend beyond relationship clients (banks with a ρ < 1). b Second, Equation (5) produces U-shaped relationship between between a bank’s PPP intensity (ρ ) and the average timing of loan origination. As PPP intensity rises to 1, the b average time to origination falls to T. This pattern is driven by banks with better lending technology offering faster processing times, losing fewer clients to nonrelationship lenders, and thus doing more lending. However, as the intensity rises from 1, the average time rises from T and asymptotes to µ¯T. The eventual reversal reflects a greater share of 1+γ nonrelationship loans, for which origination times include switching costs. Third, there is a benefit to geographic proximity to active lenders. Differentiating equa- 22

tion (5) with respect to d . We get that if ρ > 1 then: i,b b ∂E(T |d ,ρ ) (cid:16) ρ −1 (cid:17)(cid:16) (cid:17)−2 i,b i,b b = T µ¯− b 1+(ρ −1)λ(d ) (ρ −1)λ(cid:48)(d ) > 0 ∂d ρ1+γ −1 b i,b b i,b i,b b (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) BenefitofRelationship − ∂P(Ri,b|di,b,ρb) ∂di,b Namely, the expected time to origination rises with distance because distant lenders are less likely to be relationship lenders. There is no such effect for banks with ρ < 1 because they b only serve relationship clients. 4.2 Estimation Estimation of P(R |d ,ρ ) From equation (4), the probability that loan is from a rei,b i,b b lationship lender is a function of ρ and λ(d ). ρ is the ratio of 2020 PPP lending to b i,b b prepandemic small business lending in the Call Reports for bank b (as discussed in Section 2). I estimate λ(d) in order to match the distributions of loan distance for banks with ρ < 1 b and ρ > 1. Since banks with ρ < 1 only make relationship loans, the kernel density esb b timate of the distance distribution provides an estimate of fR(d). Banks with a ρ > 1, b make both relationship and nonrelationship loans, thus their f(d|ρ > 1) is a mixture of the b relationship and nonrelationship distributions, with a weight of E (ρ−1|ρ > 1) = 0.502 on i b b fR(d). Consequently, fN(d) is estimated as: f ˆ (d|ρ > 1)−0.502f ˆR(d) f ˆN(d) = b 1−0.502 where f ˆ (d|ρ > 1) and f ˆR(d) are the kernel density estimates of distance for high- and lowb intensity lenders, respectively. Taking the ratio of f ˆN(d) and f ˆR(d) gives an estimate of λ(d). The left panel of Figure 5 plots these kernel density estimates. The red-line shows the distribution for low-intensity lenders. The density is maximized around a distance of 1 mile, but falls sharply for distances over 2 miles. The distribution for high-intensity lenders is 23

Figure 5: Distance and Relationship Lending .4 1 Low Intensity (Relationship) Relationship Lender Probability High Intensity Histogram: Low Intensity Non-relationship (imputed) Histogram: High Intensity .8 .3 .6 .2 .4 .1 .2 0 0 .01 .05 .25 1 3 10 75 500 .01 .05 .25 1 3 10 75 500 Distance (Miles, Log Scale) Distance (Miles, Log Scale) (a) Distances Distribution By Lender Type (b) Relationship Lender Probability Notes: The left chart plots kernel density estimates of the distribution of loan distances for banks with a high- and low-PPP intensity (dashed blue and red lines, respectively). The grey line shows the implied distributionfornonrelationshiploans. Thelineintherightfigureplotstheestimatedprobabilitythataloan is from a relationship lender by distance. shifted to the right, and has notably less mass under 2 miles (the typical distance range for a small business client) and more mass in the right tail. The implied distribution for nonrelationship loans, shown in grey, is shifted further to the right, reflecting the fact that some loans made by high-intensity banks are relationship loans and assumed to follow a similar geographic distribution as those from low-intensity lenders. The right panel plots the probability that a lender is a relationship lender by distance, determined by the ratio of the estimated PDFs for nonrelationship and relationship loans. The probability hovers around 90% for a distance of up to a mile, then falls to 40% around a distance of 10 miles, and declines further to below 20% for loans over 500 miles away. The sharp decline around 2 miles makes that a useful reference point; about 88% of bank PPP loans within that distance are estimated to be relationship loans, compared to only 44% of loans beyond that distance. Parameter Estimation Equation (5) defines the expected time to origination as a function of ρ , λ(d ), and parameters (µ, T, γ, and α). These parameters are estimated by b i,b 24

nonlinear least squares: Time-to-Origination = E(T |d ,ρ )+(X(cid:48)β)+ +ε i,b i,b i,b b i i,b whereTime-to-Origination isthenumberofdaystooriginationforborroweri,E(T |d ,ρ ) i,b i,b i,b b ˆ comes from substituting ρ and λ(d ) in to equation (5), and X is a vector of controls meant b i,b i to capture borrower-specific factors that may affect origination timing beyond differences in bank processing times and borrower switching costs. As the controls are meant to account for differences in the time to application, whereas E(T |d ,ρ ) is meant to capture differi,b i,b b ences in processing times after application, I restrict X(cid:48)β to be weakly positive.21 Since i the parameters of interest reflect bank-specific origination timing, I weight by the inverse of the number of PPP loans originated by b to reduce idiosyncratic noise from prominent PPP lenders. The sample of loans excludes firms with fewer than 5 employees, as such firms were less likely to be immediately aware of the program, and thus variation in the timing of loan receipt is less likely to reflect differences in bank processing times.22 4.3 Results Estimates from this specification are in Table 2. The first column presents estimates from the baseline specification, while the next three columns, relative to the baseline specification, omit the application time control (column 2), remove the restriction that application time is weakly positive (column 3), and remove the inverse loan count weights (column 4). The primary object of interest is the benefit of borrowing from a relationship bank, defined as the difference in the average time-to-origination between a bank’s nonrelationship andrelationshipborrowers. Fromequation5, wecanseethatthisdependsonthreevariables. 21X includes loan size and the following tract controls: an urban indicator, the share of employees that are nonwhite, and the logarithms of prepandemic small business lending and employment. 22Over 20% of businesses with fewer than 5 employees were unaware of the program as of the time the initialfundingwasexhausted,whereasthevastmajorityoflargerbusinesseswereawareofitwithinacouple days of opening (Humphries et al., 2020). 25

Table 2: Parameter Estimates T 9.768 13.625 13.930 15.078 µ¯ 1.426 1.317 1.311 1.048 γ 0.088 0.046 0.045 0.694 α 3.778 4.287 4.765 4.691 Weighted X X X Application Time (X(cid:48)β)+ 0 (X(cid:48)β) (X(cid:48)β)+ i i i Relationship Benefit (days) 5.0 - 9.1 4.9 - 8.5 4.9 - 8.6 6.9 - 15.7 Notes: This table presents estimates from the equation: Time-to-Origination =E(T |d ,ρ )+(X(cid:48)β)++ε i,b i,b i,b j i i,b whereTime-to-Origination isthenumberofdaysuntiloriginationforloani,E(T |d ,ρ )istheexpected i,b i,b i,b b processing time from equation (5) given the bank-borrower distance and bank PPP intensity, and (X(cid:48)β)+ i is a control for application time, which is restricted to be weakly positive in the baseline specification. The X vector includes the log loan amount, and the following tract level controls: the nonwhite employment share, the logarithm of prepandemic tract small business employment, the logarithm of prepandemic small businesslending, andanurbanindicator. ThespecificationisweightedbytheinverseofthenumberofPPP originations at bank b. Columns 2-4 make the following alterations to this baseline specification: Column 2 omits the controls, Column 3 removes the restriction that time to application is positive, Column 4 is unweighted. The benefit of relationship borrowing scales with T, reflecting general processing times. The benefit is also increasing in µ¯, as higher average switching costs raise the cost of leaving the relationship bank, and γ, as higher congestion increases the cost of being deprioritized relative to relationship borrowers. The bottom row of Table 2 plots the range of benefits, where the lower bound is the benefit of a relationship with a bank that is on the margin of engaging in nonrelationship lending (the right limit as ρ → 1), while the upper bound is the benefit at the bank with the highest PPP intensity in the sample. While parameter estimates vary a fair amount, the ultimate estimate of the benefit of relationship borrowing is fairly stable. In the baseline specification, the estimates imply that banks serve relationship clients between 5 and 9 days earlier than nonrelationship clients. In columns 2 and 3, which exclude controls for application time or don’t restrict that time to be positive, estimates for T rise, but this is offset by lower estimated switching costs and congestion costs. Ultimately, the range of benefits from relationship borrowing are little 26

changed. In column 4, where the results are unweighted, estimated relationship benefits rise to be from 7 to 16 days. In this specification, delays are attributed more to congestion costs than switching costs, in order to better hit the slow origination times at banks with the highest PPP intensities. While the unweighted specification implies more notable relationship benefits at the top of the PPP intensity distribution, the estimated effect for the more typical PPP intensities is only a couple of days higher than those in the unweighted results. 5 Relationships for SBA Borrowers The previous results estimate the benefit of relationship borrowing by inferring the probability of a preexisting relationship based on the lending bank’s proximity and PPP intensity. To increase confidence in these estimates, I now use name-matched SBA 7(a) loan data to study the benefits of relationship lending in a sample of PPP loans for which a preexisting bank relationship is known. 5.1 Data and Name Matching In order to identify whether borrowers are using relationship banks, I use microdata from the SBA’s 7(a) program to identify PPP borrowers with previous SBA loans.23 I use data on SBA loans originated from 2010 to 2019, and name match by firm name, blocking on the firm’s ZIP code. This process identifies about 139,00 SBA borrowers in the 2020 PPP data, 99% of which have lender names that match to an originator in the PPP data, and 91% of which match a bank PPP originator. Roughly half of these successfully-matched PPP loans have the same PPP and SBA originator. These loans are identified as relationship loans. While the SBA data is useful for proving a sample of small businesses for which prepandemic relationships are known, there are a couple of limitations involved. First, it may be unrepresentativeofothersmallbusinessrelationships. SBAloansarepartiallyguaranteedby 23SBA data is available here: https://data.sba.gov/dataset/7-a-504-foia. 27

the government, possibly moderating screening/monitoring incentives, and facilitating more distant lending. For example, the largest SBA lender by volume is Live Oak, a branchless bank that specializes in remote lending (Di and Pattison, 2020). Additionally, SBA relationships may differ from other relationships as lenders necessarily have access to and familiarity with the SBA’s application portal, facilitating the provisioning of PPP loans (Granja et al., 2022). Second, firms may have multiple relationship, and while SBA relationships can be observed, other relationships cannot. Thus, “nonrelationship” loans in this analysis may still be relationship loans even if the bank differs from the one making the SBA loan. Nonetheless, this data provides a useful complement to the other analysis, as bank relationships can be directly observed rather than inferred by borrower location. 5.2 SBA Relationships and Origination Timing BeforeanalyzingdifferencesbyPPPintensity, Figure6showshowPPPoutcomesdifferedfor borrowers that previously took out an SBA loan from one of the top 4 U.S. banks by assets. These banks were less likely to approve PPP applications early in the program (Bartik et al., 2020) and generally did less PPP lending relative to their overall small business lending than other banks (Granja et al., 2022). The left panel plots the shares of borrowers who took out a PPP loan from their relationship lender. The top 4 banks retained 38% of their SBA clients, compared to 50% retention for other SBA lenders. The right panel shows that relationship borrowers from the top 4 banks had to wait longer to receive credit than other borrowers, thus providing an explanation for why clients of the top 4 banks disproportionately sought PPP loans from other lenders. Relationship loans at the top 4 banks had an average time to origination of 35 days, compared to only 21 days for other relationship lenders. Clients of the top 4 banks also faced worse delays when they switched to new lenders, although the difference was not as pronounced. Among the borrowers that switched to a PPP lender that was different from their SBA lender, clients of the top 4 banks had to wait about 30 days, while clients of other 28

Figure 6: PPP Outcomes by Previous SBA Lender Type .6 0.50 .5 .4 0.38 .3 .2 .1 0 redneL pihsnoitaleR htiw gniyatS erahS SBA Lender Top 4 Bank Other Lender 40 34.7 29.8 30 28.2 20.9 20 10 0 (a) Share of Relationships Retained noitanigirO ot syaD SBA Lender Top 4 Bank Other Lender Switchers Relationship Loans (b) Origination Time Notes: These charts plot the share of borrowers who take out a PPP loan from their previous SBA lender (left)andtheaveragetime-to-originationbywhetherornotborrowersusetheirpreviousSBAlender(right). Red bars show results for PPP borrowers that took out an SBA loan from a top 4 bank, while blue bars show results for borrowers from other SBA lenders. lenders had to wait about 28 days.24 Table 3 similarly documents the interplay between bank relationships, bank PPP intensity, and PPP origination timing, but with regressions of the type described in Section 3.3. The regressions predict the number of days until a loan is originated as a function of the lending bank’s PPP intensity, indicators for whether the PPP loan is from a nearby bank or a relationship SBA lender, and pairwise interactions of these variables. Themostconsistentresultinthetableisthatrelationshipborrowersreceivecreditearlier, particularly if their bank is an active PPP lender. The coefficient of -3.75 on Relationship i,b means that a non-local borrower from a bank with a PPP Intensity of 1 is expected to wait 3.75 days less on average if their PPP lender is their previous SBA lender. The benefit to relationship clients is more pronounced for more active lenders, as can be seen by the negative coefficient on the interaction between the relationship indicator and the PPP intensity measure. For a bank with a ln(PPP Intensity) = 2 (around the 95th percentile of lenders) b relationship clients are expected to receive credit 11.35 days earlier. 24This result is consistent with the model; clients at lower intensity banks are willing to switch lenders at higher switching costs, resulting in higher average wait times. 29

Columns 2 and 4 additionally add PPP lender fixed effects, and columns 3 and 4 weight by the inverse of the number of loans from the lender (within the matched sample). These specifications continue to predict notable benefits to relationship clients, particularly those of active PPP lenders, though the range of benefits is less expansive. Across these 3 specifications, relationship clients are expected to receive credit 3.2 to 5.9 days earlier if their bank has a ln(PPP Intensity) = 0, and 7.7 to 8.6 days earlier if ln(PPP Intensity) = 2. These b b results, in particular the weighted results in the last two columns, are quite similar to the range of benefits implied by the baseline model estimates (benefits from 5.0 to 9.3 days).25 Theestimatesindicatethatproximitystillrelatestooriginationtiming,buttheresultsare not as robust. In three of the four specifications, borrowers using a local bank receive credit slightly more than a day earlier than distant borrowers (assuming ln(PPP Intensity) = 0 b and the lender is not the SBA lender).26 Again, using a nearby bank is more beneficial if that bank is a more active PPP lender. These results are consistent with banks having multiple relationships, so even those who “switch away” from their relationship SBA lender benefit from proximity, as those banks are more likely to be another relationship lender. However, the benefit of using a local branch only accrues to non-SBA relationship borrowing. If a borrower uses their SBA lender, there is no marginal benefit to it being a local lender, consistent the effects of proximity being driven by relationships as opposed to proximity itself. 6 Tract level effects Theresultsthusfarindicatethatborrowersaretypicallyclosetotheirrelationshipbanksand get credit earlier if that relationship bank is a high-intensity PPP lender. Now I investigate 25I consider the upper range here to reflect the 95th percentile of PPP intensity, whereas in Table 2 the upperrangeisdeterminedbythehighestPPPintensityinthesample. Idothisbecausethepredictionsfrom the regression specification assume effects are linear in ln(PPP Intensity), making extrapolation to extreme values more problematic than in the model where effects are concave in the right tail. 26The positive coefficient in the unweighted specification without the lending bank fixed effect is driven by the top 4 banks, which predominantly served nearby clients and were slow to extend credit. 30

Table 3: SBA Relationships and Origination Speed Days to Origination (1) (2) (3) (4) Relationship -3.67** -3.13** -5.78** -4.76** i,b (0.78) (0.40) (0.44) (0.48) x ln(PPP Intensity ) -3.75** -2.40** -1.32** -1.10* b (0.65) (0.56) (0.46) (0.44) Local Branch 2.04** -1.13** -1.40* -1.13* i,b (0.58) (0.25) (0.63) (0.52) x ln(PPP Intensity ) -5.00** -0.22 -1.17 -0.60 b (0.51) (0.30) (0.72) (0.42) x Relationship 2.09** 1.24** 2.67** 1.63** i,b (0.66) (0.22) (0.60) (0.54) ln(PPP Intensity ) 1.77** 0.06 b (0.56) (0.34) R2 0.206 0.290 0.330 0.431 Obs. 117,487 116,950 117,487 116,950 Controls? X X X X County FE X X X X Lending Bank FE? X X Weighted X X Relationship Benefit (days) 3.7 - 11.2 3.1 - 7.9 5.8 - 8.4 4.8 - 6.9 Notes: ThistablepresentsestimatesfromregressingthenumberofdaysuntiloriginationonlogPPPintensity of the lending bank, an indicator for whether the lending bank has a branch within 2 miles of the borrower, an indicator for whether the lending bank previously provided an SBA loan to the borrower, as well as all pairwise interactions of these variables. The sample covers bank PPP loans to borrowers that previously took out an SBA loan. Columns 2 and 4 additionally include lending bank fixed effects, and columns 3 and 4 weight by the inverse of the number of loans in the sample by the lending bank. All specifications control for the size of the loan, whether the business has fewer than 5 employees and tract characteristics (an urban indicator, the share of employees that are nonwhite, and the logarithms of prepandemic small business lending and employment). Standards errors, in parentheses, are clustered by bank. +,∗,∗∗ indicate significanceat10%,5%and1%,respectively. Thebottomrowreportstherangeofthebenefitsofborrowing from a relationship bank for banks with an ln(PPP Intensity) between 0 and 2 (assuming a nonlocal bank). 31

the implications of these facts for geographic differences in credit availability. First, I show that a higher average PPP intensity for banks with branches within 2 miles of a tract is associated with a smaller delays in receiving PPP loans and fewer borrowers switching to fintechs or distant banks. Second, I show that these effects decay quickly as distance increases. 6.1 Effects of PPP Intensity Within 2 Miles Table 4 regresses tract level PPP outcomes on the average ln(PPP intensity) for the branches within2milesofacensustract,weightedbybranchdeposits. PPPintensityhereiscalculated as the ratio of a bank’s PPP lending in other tracts to prepandemic small business lending in order to avoid reverse causality from strong PPP lending in a particular tract causing local banks to have a high PPP intensity.27 I additionally include controls for prepandemic tract small business lending and the number of branches within two miles of the tract to account for the likely availability of relationship lenders. I include the logarithm of small business employment and variables pertaining to earnings for local workers to account for the amount of PPP credit firms were eligible for (which was tied to small firms’ payrolls).28 Odd columns present OLS estimates of how the PPP intensity of local banks affects lending outcomes, while even columns present results weighting by tract small business employment. All specifications include county fixed effects. In the first two columns, the dependent variable is the average number of days to origination for PPP loans in a census tract. A unit increase in the average ln(PPP intensity) of 27Additionally, AppendixTableA3presentssimilarestimates,butmeasuringlocalsupplyconditionswith theshareofdepositsinnearbybranchesthatarefromthetop4banks. Giventhegeographicextentofthese banks’ operations, there is less concern about reverse causality for this supply measure. 28The PPP program essentially provided credit for up to 2.5 times monthly payroll for businesses with 500 or fewer employees. This means that the logarithm of total PPP lending in an area will roughly be the sum of: i. ln(Small Business Employment), ii. ln(2.5 × Average Monthly Payroll of Eligible Firms) and iii. ln(ShareofEligiblePayrollFunded). Icontrolforthelogarithmofemploymentinfirmswithfewerthan500 employees (to account for i.) and the following variables pertaining to average earnings (to account for ii.): the share of jobs earning under $1250 per month, the share of jobs earning over $3333 per month, and the share of workers with a college degree. Effects of other variables can thus be more reasonably attributed to bank supply conditions. 32

local banks reduces the average time to origination by roughly 1 day. This elasticity is in the range of effects found in the loan level data.29 In columns 3 and 4, the dependent variable is the share of credit from banks with nearby branches. A unit increase in the average ln(PPP Intensity) of nearby branches increases the market share of nearby banks by roughly 4 percentage points. Roughly 15% of this effect is accounted for by a lower market share for fintech lenders, as shown by columns 5 and 6.30 The rest reflects a decline by further-off banks or other nonbank lenders (not shown). In the last two columns, the dependent variable is the logarithm of total PPP lending. While the activity of local banks significantly affects the timing of loan receipt and the composition of who does the lending, it appears to have little affect on loan volumes. A unit increase in the average ln(PPP intensity) only increases expected loan volumes by 1 or 2 percent, and is insignificant in the weighted specification. Regarding estimates for the other variables, the results also show that the racial composition of the workforce relates strongly to the speed with which PPP loans are extended. A one standard deviation increase in the share of employees that are nonwhite (0.17) results in about 4 day delay in receiving credit, on average. This result is consistent with other work presenting evidence of racial discrimination in the allocation of PPP credit (Howell et al., 2021; Chernenko and Scharfstein, 2022). However, this result could also indicative of minority-owned firms being less likely to have bank relationships (Mills and Battisto, 2020), resulting in less credit availability controlling for local supply conditions.31 Indeed, the lower market share of local banks in tracts with higher minority employment is consistent with relationship loans being less available in these tracts. However, it is unclear the extent to 29InAppendixTableA1,thecoefficientontheinteractionbetweenln(PPPIntensity) andLocalBranch b i,b is -2.5 in the unweighted specification and -0.6 in the weighted specification. 30The fintech share is the share of loans made by the following lenders: Celtic Bank, Cross River Bank, Intuit, Fundbox, Kabbage, Readycap and WebBank. This includes both nonbanks, and some prominent bankswithsignificantlendingthroughfintechpartnerships. Fintechsmakeupaboutaquarterofthedecline in market share of local banks when market shares are measured by counts rather that volumes, as they tended to originate smaller loans. 31While the SBA collected data on the race of borrowers, that field is rarely populated. Consequently it is only feasible to analyze results based on the racial composition of the local workforce instead of the demographics of the owners themselves. 33

Table 4: Tract Level Origination Speed Days to Local Bank Fintech ln(PPP Lending) Origination Share (pp) Share (pp) (1) (2) (3) (4) (5) (6) (7) (8) ln(PPP Intensity) -1.17** -1.02** 4.03** 4.71** -0.70** -0.56** 0.02* 0.01 j (0.20) (0.16) (0.34) (0.36) (0.18) (0.10) (0.01) (0.01) ln(Branches) -0.81** -0.54** 14.73** 14.87** -0.46** -0.27** 0.00 0.00 j (0.12) (0.07) (0.28) (0.28) (0.10) (0.05) (0.00) (0.00) ln(SB Lending) -0.30** 0.03 -0.56** 0.13 0.22** 0.41** 0.17** 0.16** j (0.10) (0.07) (0.17) (0.20) (0.08) (0.06) (0.01) (0.01) Nonwhite 23.61** 20.75** -13.00** -9.92** 18.13** 9.83** -0.08 -0.43** j (2.41) (2.19) (2.03) (2.28) (2.87) (1.75) (0.09) (0.06) ln(SB Emp) -3.87** -3.33** 1.45** 0.43+ -4.04** -1.98** 0.87** 0.94** j (0.12) (0.09) (0.21) (0.25) (0.26) (0.12) (0.01) (0.01) Earnings >$3333 -0.02 -0.84 -9.34** -15.20** 0.52 -1.61* 0.46** 0.39** j (0.77) (0.84) (1.81) (1.89) (1.04) (0.72) (0.08) (0.06) Earnings <$1250 1.85+ 4.25** 4.44* 1.51 5.10** 4.26** -0.82** -0.89** j (1.11) (1.05) (1.92) (2.11) (1.15) (0.77) (0.10) (0.07) College -8.61** -1.50 21.61** 25.88** -4.00** 0.56 0.35** 0.03 j (1.38) (1.20) (2.37) (2.61) (1.48) (0.78) (0.11) (0.11) R2 0.645 0.678 0.407 0.519 0.397 0.344 0.822 0.909 Obs. 48,897 48,897 48,301 48,301 46,226 46,226 48,897 48,897 County FE? X X X X X X X X Employment Weighted? X X X X Notes: This table presents estimates from the equation: y =α +βln(PPP Intensity) +γ(cid:48)X +ε j c(j) j j j where y is a PPP outcome for a tract j. ln(PPP Intensity) is the average of the logarithm of the ratio of j j PPP lending in other tracts to prepandemic small business lending for banks with branches within 2 miles of the tract centroid, weighted by the deposits in those branches. X is a vector of tract-level controls, j including the the logarithms of 2017 small business employment (≤500 employees) and 2019 small business lending in the tract; the logarithm of the number of branches within 2 miles of the tract; and the shares of jobs in the tract that earn over $3333 per month, earn under $1250 per month, are held by a worker with a college degree, or are held by nonwhite employees. α is a county fixed effect. y is the average number of c(j) j days until origination for the loans in the tract in columns 1 & 2, the percent of credit provided by banks withintwomilesincolumns3&4,thepercentofcreditprovidedbyfintechlendersincolumns5&6andthe logarithm of 2020 total PPP lending in the tract in columns 7 & 8. Even-numbered specifications weight by tract small business employment. Standards errors, in parentheses, are clustered by county. +,∗,∗∗ indicate significance at 10%, 5% and 1%, respectively. 34

which this result reflects a lack of relationships, or a lack of credit from the relationships that do exist.32 Additionally, census tracts with more branches within 2 miles, more college educated employees, or more small business employment are able to receive credit sooner, have a higher share of PPP lending accounted for by nearby banks, and have a lower share of credit accounted for by fintechs. Variation in the volume of PPP lending is mostly accounted for by the amount of small business employment, and variables relating to local earnings, as would be expected given that PPP amounts were tied to business payrolls. Overall, the results show that when local banks are less active in PPP, borrowers turn to other lenders for loans. This switching appears to be costly, as borrowers receive loans later on average. Although low-intensity banks provide less credit to the tract than other banks would, the ultimate effect on PPP loan volumes is relatively modest. While it seems that firmsinthesetractswereusuallyabletogetfundingeventually,theinabilitytochannelcredit to firms in the earliest stage of the program—when unemployment was at its peak—may have limited its effectiveness (Doniger and Kay, 2020). 6.2 Effects of PPP Intensity by Distance Table 4 showed that the PPP intensity of banks within 2 miles of a tract is important in determining how quickly borrowers can access credit. Now I investigate whether this is the appropriate distance to consider, or whether the activity of more distant banks matters as well. Figure 7 presents estimates of {β } from the equation: d (cid:88) y = α + β ln(PPP Intensity)(d) +γ(cid:48)X +ε j c(j) d j j j d 32The effect of racial composition on loan volumes is unclear. One specification shows notable effects for the nonwhite share and the other notable effects for the college share. Assessing differences in the ability to getPPPloansiscomplicatedbytheracialwagegap: lowerPPPlendingcouldreflectloweraverageearnings among minority employees (reducing loan sizes) or greater frictions in accessing PPP credit (reducing loan counts). 35

where ln(PPP Intensity)(d) is the deposit weighted average ln(PPP intensity) of banks with j branches between d−2 and d miles from tract j. y is one of the previously discussed tract j level PPP outcomes: the average number of days until origination (Panel 7a), the share of loans from banks with nearby branches (Panel 7b), the share of loans from fintechs (Panel 7c) or the logarithm of PPP loan volume (Panel 7d).33 In each panel, the coefficient on ln(PPP Intensity)(2) is similar to the corresponding j estimate in Table 4, meaning that the controls for PPP intensity at more distant banks does not affect the estimated effects for banks within 2 miles. The effects then decline towards 0 as distance rises (with the sign reversing at greater distances when total loan volumes are the dependent variable). The effects of PPP intensity on the average number of days to origination are the slowest to decay with distance. However, even there, the PPP intensity of banks within 2 miles affects origination timing about 50 percent more than for banks 2–6 miles away, and about 3 times as much as for banks 6–10 miles away. For the fintech market share, the PPP intensity within 2 miles matters about twice as much as for banks 2–4 miles away. For the market share of local banks or total PPP lending, effects either go to zero or reverse sign for banks 2–4 miles away. In short, the effects of PPP intensity on the timing or composition of tract lending are predominantly driven by banks within 2 miles. Overall, these results show that variation in supply conditions have very localized effects. These effects are unlikely to be driven by bank preferences for nearby borrowers as loan guarantees remove the benefits of local information or easier monitoring. Likewise, these effects are unlikely to be driven by borrower preferences for nearby banks, as transit costs of going to a bank just a few miles away are unlikely to be prohibitive. However, most preexisting small business relationships were with banks within a few miles (Brevoort and Wolken, 2009), thus the narrow geographic scope of supply shocks is consistent with frictions in switching from relationship lenders driving the results. 33X is as in Table 4, but with additional dummy variables indicating whether there are no banks within j a particular distance band. To maintain the same sample as in the previous analysis, ln(PPP intensity) is recoded to 0 when there are no banks within a band and tracts are dropped if there are no banks within 2 miles. 36

Figure 7: Effects of PPP Intensity by Bank Distance 0.00 -0.50 -1.00 -1.50 )d(ytisnetnI PPP no tneicffeoC 6.00 4.00 2.00 0.00 -2.00 0-2 2-4 4-6 6-8 8-10 Distance (miles) (a) Days Until Origination )d(ytisnetnI PPP no tneicffeoC 0-2 2-4 4-6 6-8 8-10 Distance (miles) (b) Local Bank Share 0.00 -0.20 -0.40 -0.60 -0.80 -1.00 )d(ytisnetnI PPP no tneicffeoC 0.04 0.02 0.00 -0.02 -0.04 0-2 2-4 4-6 6-8 8-10 Distance (miles) (c) Fintech Share )d(ytisnetnI PPP no tneicffeoC 0-2 2-4 4-6 6-8 8-10 Distance (miles) (d) ln(PPP Lending) Notes: This figure plots estimates and 95% confidence intervals for {β } from the specification: d (cid:88) y =α + β ln(PPP Intensity)(d) +γ(cid:48)X +ε j c(j) d j j j d where ln(PPP Intensity)(d) is the deposit weighted average ln(PPP intensity) for branches between d−2 j and d miles from a census tract. y is a PPP outcome for tract j: the average number of days between the j start of PPP and the approval date (panel a), the share of lending by nearby banks (panel b), the share of lending by fintechs (panel c), and the natural logarithm of the value of PPP approvals (panel d). X is a j vector of tract-level controls, including the logarithms of 2017 small business employment (≤500 employees) and 2019 small business lending in the tract; the logarithm of the number of branches within 2 miles of the tract; the shares of jobs in the tract that earn over $3333 per month, earn under $1250 per month, are held by a worker with a college degree, or are held by nonwhite employees; and dummy variables indicating whether there are no banks within a distance band. ln(PPP Intensity)(d) is recoded to 0 when there are j no banks within a band so as to keep a consistent sample with Table 4. Each specification includes county fixed-effects. Standards errors are clustered by county. 37

7 Conclusion I use geocoded data on PPP borrowers and bank branches to study the spatial distribution of PPP originations. I show that about half of bank PPP loans go to borrowers within 2 miles of a branch. Borrowers receive loans sooner if they get credit from a nearby bank, especially if that bank is a more active PPP lender. Borrowers where nearby banks are less active in PPP switch to fintechs or other distant lenders, resulting in delays receiving credit. When I estimate a model to match the observed relationships between distance, PPP intensity, and origination timing, I find that relationship borrowers were able to receive credit 5 to 9 days earlier than other borrowers from the same bank. Analysis of a subsample of PPP loans to previous SBA 7(a) borrowers produces a similar range of benefits to relationship borrowing. Overall, the findings demonstrate the importance of bank relationships for prompt access tocredit. Thoughincreasedlendingfromdistantlenderscounteracteddeclinesinareaswhere local banks were less active, there were material delays involved. As such delays have been shown to be associated with worse job losses (Doniger and Kay, 2020), the results suggest that intermediation frictions likely reduced the program’s effectiveness. Furthermore, these findings highlight a cost to intermediating aid through the banking sector; if firms without banking relationships are more financially constrained, funding dispersed by banks is less likely to flow to where it is most needed. 38

References Agarwal, S. and R. Hauswald (2010). Distance and private information in lending. The Review of Financial Studies 23(7), 2757–2788. Amiram, D. and D. Rabetti (2020). The relevance of relationship lending in times of crisis. Available at SSRN 3701587. Bartik, A. W., M. Bertrand, F. Lin, J. Rothstein, and M. Unrath (2020). Measuring the labor market at the onset of the COVID-19 crisis. Technical report, National Bureau of Economic Research. Bartik, A. W., Z. B. Cullen, E. L. Glaeser, M. Luca, C. T. Stanton, and A. Sunderam (2020). The targeting and impact of Paycheck Protection Program loans to small businesses. Technical report, National Bureau of Economic Research. Brevoort, K. P. and J. D. Wolken (2009). Does distance matter in banking? In The changing geography of banking and finance, pp. 27–56. Springer. Cattaneo, M. D., R. K. Crump, M. H. Farrell, and Y. Feng (2019). On binscatter. arXiv preprint arXiv:1902.09608. Chernenko, S. and D. S. Scharfstein (2022). Racial disparities in the paycheck protection program. Technical report, National Bureau of Economic Research. Degryse, H. and S. Ongena (2005). Distance, lending relationships, and competition. The Journal of Finance 60(1), 231–266. Di, W. and N. Pattison (2020). Distant lending, specialization, and access to credit. Doniger, C. and B. Kay (2020). Ten days late and billions of dollars short: The employment effects of delays in Paycheck Protection Program financing. Available at SSRN. 39

Duchin, R., X. Martin, R. Michaely, and H. Wang (2022). Concierge treatment from banks: Evidencefromthepaycheckprotectionprogram. Journal of Corporate Finance 72, 102124. Erel, I. and J. Liebersohn (2020). Does fintech substitute for banks? Evidence from the Paycheck Protection Program. Technical report, National Bureau of Economic Research. Granja, J., C. Makridis, C. Yannelis, and E. Zwick (2022). Did the Paycheck Protection Program hit the target? Journal of financial economics 145(3), 725–761. Howell, S. T., T. Kuchler, D. Snitkof, J. Stroebel, and J. Wong (2021). Automation and racial disparities in small business lending: Evidence from the Paycheck Protection Program. Technical report, Working paper 29364. National Bureau of Economic Research. https://doi. org . Humphries, J. E., C. A. Neilson, and G. Ulyssea (2020). Information frictions and access to the Paycheck Protection Program. Journal of public economics 190, 104244. Joaquim, G.andF.Netto(2020). BankincentivesandtheimpactofthePaycheckProtection Program. Available at SSRN 3704518. Li, L. and P. E. Strahan (2021). Who supplies PPP loans (and does it matter)? Banks, relationships,andtheCOVIDcrisis. Journal ofFinancial andQuantitative Analysis 56(7), 2411–2438. Merker, V., T. Ling, and D. Tannebaum (2020). PPP is a complience minefield for banks. American Banker. Mills, C. K. and J. Battisto (2020). Double jeopardy: COVID-19s concentrated health and wealth effects in black communities. Federal Reserve Bank of New York. 40

A Appendix A.1 Figures Figure A1: Residualized Distance by Bank PPP Intensity .4 .3 .2 .1 0 ytisneD High Intensity Bank Low Intensity Bank -10 -5 0 5 10 Residual Log Distance Notes: This chart plots kernel density estimates of the logarithm of the distance to the lending bank, residualizing on log loan amount, a small firm indicator, and tract fixed effects. The blue line shows the distribution for banks with a PPP intensity above 1, and the red line the distribution for those below 1. 41

A.2 Tables Table A1: Bank Proximity and Origination Speed Days to Origination (1) (2) (3) (4) (5) (6) ln(PPP Intensity ) -2.41** -1.62 -0.65 -1.17** -1.19** -1.00** b (0.75) (1.10) (1.19) (0.18) (0.18) (0.20) ln(PPP Intensity )2 1.30** 1.12** 1.01** 0.61** 0.60** 0.59** b (0.14) (0.13) (0.13) (0.09) (0.09) (0.09) Local Branch -0.87+ -1.11* -0.39 -1.35** -1.35** -1.29** i,b (0.47) (0.52) (0.45) (0.11) (0.11) (0.11) x ln(PPP Intensity ) -2.49** -0.63** b (0.89) (0.18) ln(Loan Amount) -4.19** -4.07** -4.04** -4.64** -4.63** -4.63** i,b (0.34) (0.36) (0.35) (0.05) (0.05) (0.05) Small Firm 4.43** 4.46** 4.47** 4.09** 4.09** 4.10** i,b (0.78) (0.78) (0.78) (0.12) (0.12) (0.12) Internet Bank 2.69 1.14 12.24** 11.60** b (4.11) (4.33) (3.64) (3.56) Top 4 4.24* 3.47+ 4.58* 4.28* b (1.90) (1.91) (1.94) (1.97) R2 0.265 0.267 0.268 0.182 0.182 0.182 Obs. 4,077,112 4,077,112 4,077,112 4,077,112 4,077,112 4,077,112 Tract Controls? X X X X X X County FE? X X X X X X 3-Digit NAICS FE? X X X X X X Weighted? X X X Notes: This table presents estimates from regressing the number of days until origination on log PPP intensityofthelendingbank,logPPPintensitysquaredandanindicatorforwhetherthebankhasabranch within 2 miles of the borrower. Specifications in columns 2 & 5 add controls for whether the bank is one of thefourlargestlendersoraninternetbank,andthespecificationsincolumns3&6addaninteractionoflog PPP intensity with the local lender dummy. All specifications control for the size of the loan, whether the businesshasfewerthan5employeesandtractcharacteristics(nonwhiteemploymentshare,logsmallbusiness employment, log prepandemic small business lending, and an urban indicator), and include county and 3digit NAICS fixed effects. Estimates for tract controls are not displayed. Standards errors, in parentheses, are clustered by county. +,∗,∗∗ indicate significance at 10%, 5% and 1%, respectively. 42

Table A2: Bank Proximity and Round 1 Funding Round 1 Indicator×100 (1) (2) (3) (4) (5) (6) ln(PPP Intensity ) 7.86** 4.33** 2.52 3.63** 3.62** 3.30** b (2.00) (1.62) (1.81) (0.29) (0.29) (0.30) ln(PPP Intensity )2 -1.82** -0.70** -0.50* -1.22** -1.21** -1.20** b (0.35) (0.21) (0.22) (0.12) (0.12) (0.12) Local Branch 1.15 2.14* 0.79 2.53** 2.56** 2.45** i,b (1.19) (0.88) (0.82) (0.23) (0.23) (0.23) x ln(PPP Intensity ) 4.64** 1.07** b (1.47) (0.30) ln(Loan Amount) 9.95** 9.17** 9.13** 10.79** 10.78** 10.79** i,b (0.52) (0.68) (0.68) (0.09) (0.09) (0.09) Small Firm -5.46** -5.63** -5.64** -10.03** -10.03** -10.04** i,b (0.91) (0.84) (0.85) (0.26) (0.26) (0.26) Internet Bank -22.82** -19.94** -9.67 -8.59 b (5.85) (5.98) (6.15) (6.36) Top 4 -21.71** -20.28** -16.32** -15.81** b (3.04) (3.14) (5.91) (5.99) R2 0.259 0.283 0.285 0.226 0.226 0.226 Obs. 4,077,112 4,077,112 4,077,112 4,077,112 4,077,112 4,077,112 Tract Controls? X X X X X X County FE? X X X X X X 3-Digit NAICS FE? X X X X X X Weighted? X X X Notes: This table presents estimates from regressing an indicator for whether a loan was originated during the first round of PPP funding on log PPP intensity of the lending bank, log PPP intensity squared and an an indicator for whether the bank has a branch within 2 miles of the borrower. The round 1 indicator is multipliedby100,soestimatesareintermsofthepercentagepointchangeintheprobabilityofbeingfunded in round 1. Specifications in columns 2 & 5 add controls for whether the bank is one of the four largest lenders or an internet bank, and the specifications in columns 3 & 6 add an interaction of log PPP intensity withthelocallenderdummy. Allspecificationscontrolforthesizeoftheloan,whetherthebusinesshasfewer than 5 employees and tract characteristics (nonwhite employment share, log small business employment, log prepandemic small business lending, and an urban indicator), and include county and 3-digit NAICS fixed effects. Estimates for tract controls are not displayed. Standards errors, in parentheses, are clustered by county. +,∗,∗∗ indicate significance at 10%, 5% and 1%, respectively. 43

Table A3: Tract Level Origination Speed Days to Local Bank Fintech ln(PPP Lending) Origination Share (pp) Share (pp) (1) (2) (3) (4) (5) (6) (7) (8) Top 4 Share 2.27** 1.86** -4.86** -7.06** 1.84** 1.11** -0.04** -0.02 j (0.39) (0.37) (0.77) (0.74) (0.39) (0.24) (0.02) (0.01) ln(Branches) -0.82** -0.53** 14.72** 14.85** -0.47** -0.27** 0.00 0.00 j (0.12) (0.07) (0.29) (0.29) (0.11) (0.05) (0.00) (0.00) ln(SB Lending) -0.30** 0.03 -0.55** 0.14 0.22** 0.41** 0.17** 0.16** j (0.10) (0.07) (0.17) (0.20) (0.08) (0.06) (0.01) (0.01) Nonwhite 23.60** 20.79** -13.26** -10.25** 18.07** 9.85** -0.08 -0.43** j (2.46) (2.18) (2.09) (2.28) (2.89) (1.75) (0.09) (0.06) ln(SB Emp) -3.86** -3.31** 1.44** 0.38 -4.03** -1.97** 0.87** 0.94** j (0.12) (0.09) (0.21) (0.25) (0.26) (0.12) (0.01) (0.01) Earnings >$3333 0.02 -0.80 -9.40** -15.36** 0.55 -1.60* 0.46** 0.39** j (0.78) (0.85) (1.84) (1.96) (1.05) (0.74) (0.08) (0.06) Earnings <$1250 1.81+ 4.21** 4.49* 1.58 5.06** 4.23** -0.82** -0.89** j (1.07) (1.02) (1.96) (2.19) (1.15) (0.77) (0.10) (0.07) College -8.69** -1.57 21.75** 26.10** -4.08** 0.52 0.35** 0.03 j (1.38) (1.20) (2.44) (2.62) (1.47) (0.78) (0.11) (0.11) R2 0.645 0.678 0.405 0.517 0.397 0.344 0.822 0.909 Obs. 48,902 48,902 48,306 48,306 46,231 46,231 48,902 48,902 County FE? X X X X X X X X Employment Weighted? X X X X Notes: This table presents estimates from the equation: y =α +βTop 4 share +γ(cid:48)X +ε j c(j) j j j where y is a PPP outcome for a tract j. Top 4 share is the share of deposits in branches within 2 miles of j j the tract that belong to one of the four largest U.S. banks. X is a vector of tract-level controls, including j the logarithms of 2017 small business employment (≤500 employees) and 2019 small business lending; the logarithm of the number of branches within 2 miles of the tract; and the shares of jobs in the tract that earn over $3333 per month, earn under $1250 per month, are held by a worker with a college degree or are heldbynonwhiteemployees. α isacountyfixedeffect. y istheaveragenumberofdaysuntilorigination c(j) j for the loans in the tract in columns 1 & 2, the percent of loans originated by banks within two miles in columns3&4,thepercentofloansoriginatedbyfintechlendersincolumns5&6,andthelogarithmof2020 PPP lending in the tract in columns 7 & 8. Even-numbered specifications weight by tract small business employment. Standards errors, in parentheses, are clustered by county. +,∗,∗∗ indicate significance at 10%, 5% and 1%, respectively. 44