Out of Sight, Out of Mind: Nearby Branch Closures and Small Business Growth

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Out of Sight, Out of Mind: Nearby Branch Closures and Small Business Growth Ben Ranish, Andrea Stella, Jeffery Zhang 2024-071 Please cite this paper as: Ranish, Ben, Andrea Stella, and Jeffery Zhang (2024). “Out of Sight, Out of Mind: Nearby Branch Closures and Small Business Growth,” Finance and Economics Discussion Series 2024-071. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2024.071. 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.

Out of Sight, Out of Mind: Nearby Branch Closures and Small Business Growth* Ben Ranish Andrea Stella Jeffery Zhang Federal Reserve Board Federal Reserve Board University of Michigan July 29, 2024 Abstract Since2010,thetotalnumberofcommercialbankbranchesintheUnitedStateshasfallen byabout20%. Dobranchclosuresmeaningfullyaffecteconomicactivity? Weinvestigatethe impactofbranchclosuresonsmallbusinesses,whosecreditaccessmaybefacilitatedthrough local relationships with bankers. We use exogenous variation in branch closures related to mergers and acquisitions to show that closures of nearby branches decrease small business employmentgrowthandentry. Ourresultsarerobusttovariationsinourmeasureofemployment, proximity, and construction of the instrument. Altogether, our analysis highlights the importanceoflocalbankbranchestosmallbusinesses. JELclassifications: E32,E44. D22,D53,G21,J23,L25. Keywords: Creditaccess;smallbusinesses;firmgrowth;branchclosures;bankmergers;LongitudinalBusinessDatabase. *We thank Nicola Cetorelli, Ce´dric Huylebroek, Don Morgan, Dulani Seneviratne, and seminar participants at theBankforInternationalSettlements,the2023FSRDCResearchConference,IAAE2024,FEBS2024,theFederal ReserveBoard,andtheWhartonSchoolforhelpfulcommentsandsuggestions. Theviewsexpressedhereinareour ownanddonotnecessarilyreflecttheviewsoftheBoardofGovernorsoftheFederalReserveSystemortheFederal ReserveSystem.Moreover,anyviewsexpressedarethoseoftheauthorsandnotthoseoftheU.S.CensusBureau.The CensusBureauhasreviewedthisdataproducttoensureappropriateaccess,use,anddisclosureavoidanceprotection of the confidential source data used to produce this product. This research was performed at a Federal Statistical ResearchDataCenterunderFSRDCProjectNumber2427(CBDRB-FY24-P2427-R11431). 1

1 Introduction According to the U.S. Census Bureau’s Business Dynamic Statistics, firms with under 20 employees account for over 20 million jobs in the United States. These small firms rely heavily on relationship lending, which is facilitated by bank branch access (Berger and Udell (2002)). However, about one third of U.S. commercial bank branches have closed since 2010, with the total number of branches declining by about 20%. The pace of closures was particularly high during and after the Covid-19 pandemic. Could these branch closures have significant consequences for smallbusinessesandtheU.S.economyinthecomingyears? As an anecdote, consider a bakery with ten employees that is looking to expand to a larger kitchen and acquire a delivery van. The local branch’s bankers are familiar with the bakery’s cash flow as well as its role and contribution to the local economy. This relationship-based knowledge maygivethelocalbankerstheconfidencerequiredtomaketheloanthebakeryneedstoexpand. If thatbranchweretoclose,itmaytakethebakerytimetoestablishadequaterapportwithbankersat whicheverbranchthebakerytakesitsbankingbusinesstonext. Consequently,thebakerymaysee reducedcreditavailability,whichcouldcurtailitsgrowth. Ontheotherhand,thebakerymayhave sufficientlytransparentoperationsandriskstoenableanynearbybank—orevenadistantFinTech lender—to evaluate its creditworthiness. Moreover, even if the nearby branch closes, perhaps the loan officer working with the bakery can find employment at another bank and preserve the “soft information”requiredtocontinuethelendingrelationship. Inshort,it’snotobviousthattheclosing ofanearbybranchshouldhaveanimpactonthebakery. The existing academic literature also lacks a clear answer. Branch closures could hurt existinglendingrelationshipsbecauseofswitchingcostsandthereforeconstrainfirms’accesstocredit (Bonfim, Nogueira and Ongena (2021)). On the other hand, branch closures could reduce local competitionandithasbeenarguedthatbankingmarketpowermightbeneededforbankstoestablishlendingrelationshipswithriskyfirms(PetersenandRajan(1995)). We contribute to this literature by empirically assessing the impact of local branch closures using an instrumental variable identification strategy as well as data from the U.S. Census Bureau 1

and the Federal Reserve System’s National Information Center (NIC) database. In particular, we employ U.S. Census Bureau data from the Longitudinal Business Database (LBD) on small (i.e., 20orfeweremployees)standalonefirmsintheUnitedStatesfrom1990to2020—excludingthefinance,insurance,real-estate,andgovernmentsectors. AndtheNICdatabasecontainsinformation onthelocationofallcommercialbankbranchesintheUnitedStates. Since branch closures are not randomly assigned, we use quasi-random branch closures induced by mergers and acquisitions (M&A) among large banks as an instrument. We model fiveyear changes in employment at the firm level as a function of the share of branches closing within a certain distance of the firm and a full set of fixed effects. The share of branches closing is instrumentedwiththeshareofbranchesexposedtocertainM&Aactivity. Wefindthatclosuresofbankbrancheswithin5kilometers(3.1miles)haveasignificantimpact on small business growth. At the extreme, if every branch closed within 5 kilometers, the growth insmallfirms’employmentwouldfallbyapproximately16percentagepointsoverthesubsequent five years. It is, of course, not reasonable to assume that every local branch would close. Indeed, in our sample, around 25% of local branches shut down when there are branch closures. Thus, when the typical share of nearby branches close, the growth in small firms’ employment declines byapproximately4percentagepointsoverfiveyears. In addition, we explore the impact of local branch closures on the extensive margin of small business growth. We estimate the impact of branch closures on the entry and exit rates of small businessesatthezipcodelevelandfindanegativeandsignificantimpactontheentryrate. Specifically, the entry rate is about 1.3 percentage points lower in the following year, which is approximately14%oftheaverageentryrate. Of note, our estimates may fall below the true effects of branch closures. As discussed later, ouridentificationstrategystudiesoutcomesaroundclosingbranchesthatbelongtolargebanksand are proximate to other branches. The literature has emphasized that small banks tend to be more important for small business lending and that distance matters more in relationships with small banks (Berger, Miller, Petersen, Rajan and Stein (2005)). In other words, branch closures may 2

have greater impact where they create “banking deserts,” but our estimates do not capture such outcomes. Ouranalysisengageswithtwobodiesofliterature. First,priorliteraturehasshownthatthedistancebetweenafirmanditspotentiallendersmattersforthequantity,quality,andpriceoftheloans it receives. See, for example, Petersen and Rajan (2002), Degryse and Ongena (2005), Brevoort and Hannan (2006), Agarwal and Hauswald (2010), Bellucci, Borisov and Zazzaro (2013), Hollander and Verriest (2016), Levine, Lin, Peng and Xie (2020), and Adams, Brevoort and Driscoll (2023). Nguyen(2019)showsthatbranchclosuresintheUnitedStatesleadtoapersistentdecline inlendingtolocalsmallbusinesses,andAmbergandBecker(2024)doesthesameusingSwedish data: Suchdeclinesinsmallbusinesslendingarelikelyresponsibleforthedeclineinemployment growth we estimate in this paper. Second, there is a vast literature on the real effects of shocks to bank credit. Some recent papers include Chodorow-Reich (2014) and Huber (2018) on the consequences of a large bank cutting lending during the Great Recession, Amiti and Weinstein (2018) ontheroleofbanksupplyshocksinaggregateinvestmentfluctuations,Berton,Mocetti,Presbitero and Richiardi (2018) on the elasticity of employment to credit supply shocks, Heblich and Trew (2019) on the role of banking access in the spread of the Industrial Revolution, Bottero, Lenzu and Mezzanotti (2020) on the real effects of a credit contraction, Alfaro and Moral-Benito (2021) onthepropagationofbank-lendingshocksthroughinput-outputrelationships,andfinallyBenson, Blattner,Grundl,KimandOnishi(forthcoming)ontheimpactofbankmergersofclose-proximity banksonconsumercredit. ThethreepapersclosesttooursareGreenstone,MasandNguyen(2020),whofindthatshocks to banks’ credit supply are transmitted to their small business customers, but with close-to-zero impact on small business employment; Amberg and Becker (2024), who associate shrinkage of branch networks in Sweden with declines in employment and sales and an increase in business exit; and Mann (2022), who focuses on the impact of changes in county-level bank concentration on small business lending and county-level employment. We add to this literature by looking directly at the economic consequences of bank branch closures, which can be easily measured as 3

opposed to banking supply shocks that need to be estimated (often as a residual), and using data on the whole population of small businesses in the United States instead of focusing on a single bankorregion. 2 Data Description The Center for Economic Studies at the U.S. Census Bureau created and maintains the Longitudinal Business Database (LBD), a longitudinal establishment-level database that covers private establishments with at least one employee in the United States. The LBD provides the number of employees,zipcode,firmID,andsectoralaffiliation(NAICScode)ofeachestablishment.1 2 An LBD establishment is defined as a single physical location where business is conducted. Note that this definition is not equivalent to the IRS Establishment Identification Number (EIN), which might be composed of more than one LBD establishment. The LBD establishment is also not equivalent to a firm, as a firm may entail multiple establishments. Since we focus on small firms operating from a single location, we include in our sample all stand-alone firms with only one establishment and 20 or fewer employees from 1990 to 2020. We exclude from the analysis thefinance,insurance,real-estate,andgovernmentsectors. The banking data employed in this project come from the public Federal Reserve System’s National Information Center (NIC) database. NIC contains information from many different regulatory reports, and includes panel data on bank branch locations, ownership, corporate activity, and banking activity and balance sheets. The data we use are merged with the LBD at the zip code level. To assess proximity between branches and establishments or other branches, we use Euclidean distances measured between zip code centroids from the NBER’s ZIP Code Distance Database. 1For more information on the U.S. Census Bureau’s Longitudinal Business Database see Jarmin and Miranda (2002)andChow,Fort,Goetz,Goldschlag,Lawrence,Perlman,StinsonandWhite(2021). 2ThezipcodeinformationintheLBDisnotcleanedtobelongitudinallyconsistent.Asaconsequence,theremight bespuriousswitchesinzipcodescausingmeasurementerrorinourdata. 4

3 Empirical Model We model changes in employment at the firm level as a linear function of the share of branches closing within 5 kilometers. We believe that branch closures mainly impact small business employment by disrupting lending relationships. Previous studies show that it is costly for small businesses to establish new lending relationships. When a small business loses its lender due to a branch closure, it may suffer even if new branches open nearby. Therefore, we focus on the total number of branch closures rather than the net change in branches. Our regression equation is as follows: g =ξ +κ +γ +βShareClose +δX +ε (1) i,s,l,t,t+h i st lt ilt islt islt where g is the growth in employment for firm i in four-digit NAICS sector s in county i,s,l,t,t+h l from year t to year t+h winsorized at the 1st and 99th percentiles, ShareClose is the share ilt of bank branches closing between year t−1 and t within a distance of 5 kilometers from firm i, ξ represents firm fixed effects, κ are four-digit NAICS sector by year fixed effects to control for i st sectoraltrends,γ arecountybyyearfixedeffectstocontrolforregionaltrends,andX isavector lt islt of controls for recent neighborhood trends and firm life-cycle effects. These controls include the average employment growth rate in the zip code in the previous three years, the average entry rate in the zip code in the previous three years, the average exit rate in the zip code in the previous three years, and a dummy equal to 1 if the firm is 10-year old or younger.3 For reasons discussed furtherbelow,ourcontrolsalsoincludethreelagsoftheshareofbrancheswithin5kilometersthat areinvolvedinanyM&Aevent. Asiscommonlydoneinstudyingfirmdynamics,weestimateall of our regressions using employment at yeart as a weight. Thus, results represent the perspective of the representative small firms’ employee and are not as heavily driven by volatility associated withtheverysmallestfirms. Finally,weestimatethemodelfollowingCorreia(2016). Sincebranchclosuresarenotrandom,weusemergersandacquisitions(“M&A”)oftheparent 3Theaveragezip-code-levelemploymentgrowthrate,entryrateandexitrateinthepreviousthreeyearsarecomputedusingallestablishments,notonlysingle-unitsmallfirms. 5

companiesasasourceofexogenousvariationinbranchclosures.4 M&Aactivityamongbanksdisproportionatelyleadstotheclosureofbankbranchesthatarelocatedclosetoeachotherandwere competing before the parent companies merged. Benson et al. (forthcoming) find that banks involvedinaclose-proximityM&Aeventare40percentagepointsmorelikelytoshutdownbranches and that total branches in the affected markets are 17 percentage points more likely to decrease. We consider only M&A activity between banks in the top percentile by number of branches to remove the possibility that the M&A activity is associated with hyper-local economic conditions. Even if M&A activity among large banks can be plausibly considered exogenous to the local economicconditionsfacedbythesmallfirmsinourdata,thechoiceofwhichbranchesareshutdown after an M&A event is not random. Therefore, our identification relies on the exposure of local branches to M&A activity among large banks. To be more precise, our instrument is the share of bank branches within 5 kilometers of a firm that are exposed to the possibility of M&A induced branch consolidation. We consider a branch of bank A to be exposed to possible consolidation if it is within 5 kilometers of a branch of bank B, and banks A and B are both large and involved in thesameM&Aevent. Wetaketwofurtherstepsinourapproachtobolsterthevalidityoftheinstrument. First,M&A activity may be associated with significant changes in management practices, including lending decisions. For example, if M&A activity is more common with economically stressed banks, we might expect that credit availability near any branch of the merging banks would be reduced. To address this concern, we include in all regressions as a control three lags of the share of branches within 5 kilometers whose parent bank is involved in any type of M&A activity. Second, by construction, exposure to possible M&A induced branch closure is only possible where branches of at least two large banks are near each other. Areas with fewer large banks are lower density, andcouldhavedifferentfirmdynamics. Therefore,werestrictoursampletozipcodesthatsatisfy twoconditions: (a)thezipcodehadatleastonebranchofalargebankwithin5kilometersand(b) that branch was within 5 kilometers of another branch of a large bank betweent−3 andt. These 4WedonotincludeinternalreorganizationsinourdefinitionofM&Aactivity. 6

excludedzipcodesaccountforabout20%ofbankbranchesandsmallfirmsinrecentyears. Withallofthatinmind,ourfirst-stageregressionis: ShareClose =ξ FS+κ FS+γ FS+δ FSX +θZ +ν (2) islt i st sl islt ilt islt where the superscript FS denotes the first-stage estimates of the parameters and fixed effects included in the second-stage equation, and Z is the vector of instruments. These include the share ilt ofbranchesthatareexposedtothepossibilityofM&Ainducedbranchclosureineachofthethree previous years. More precisely, we first determine for each firm which branches are within 5 kilometers and belong to a large bank undergoing M&A activity between t−1 and t. Among those branches,wecountwhicharewithin5kilometersofabranchthatbelongstoadifferentlargebank that is party to the same M&A activity. We then divide this count by the total number of branches inthezipcode. Werepeatfortheperiodsbetweent−2andt−1,andbetweent−3andt−2,for atotalofthreeinstrumentalvariables. The regression model in (1) estimates the effect of branch closures on the intensive margin of employment growth but is silent on the firm entrance and exit, that is the extensive margin of growth. To investigate the latter, we aggregate the data to the zip code level and model the entry rateandexitrateaslinearfunctionsoftheshareofbranchesclosingwithin5kilometersofthezip code: y =γ y +βShareClose y +δ yX +ε y (3) zip,t,t+1 lt zipt zipt zipt where y is either the entry rate or the exit rate in zip code zip from year t to year t +1, zip,t,t+1 ShareClose istheshareofbankbranchesclosingbetweenyeart−1andt withinadistanceof5 zipt kilometers from zip code zip, γ are county by year fixed effects, and X is a vector of controls, lt zipt which includes the same zip code level controls from (1) as well as the share of small firms in the zip code that are 10 years old or younger, and the share of small firm employment in the zip code in each 2-digit NAICS sector. Entry and exit rates are computed as in the Business Dynamic 7

Statistics, that is the number of firms that enter or exit at timet divided by the average number of firmsatt andt−1. Usingthesameinstrumentsasbefore,thefirst-stageregressionis: ShareClose =γ FS,y +δ FS,yX +θ yZ +ν y (4) zipt lt zipt zipt zipt Table1: SummaryStatisticsofBenchmarkandExcludedSamples BenchmarkSample ExcludedSample Mean SD Mean SD BankBranchCharacteristics Totalbranches 35.85 64.00 4.462 6.627 Shareofclosures 0.030 0.051 0.015 0.083 Shareexposedtoanymerger 0.342 0.233 0.226 0.318 Instrument 0.018 0.069 FirmCharacteristics Employmentgrowth -0.015 0.343 -0.020 0.357 Employment 9.373 5.310 9.035 5.267 Young 0.425 0.494 0.414 0.493 Entryrate 0.093 0.034 0.096 0.071 Exitrate 0.083 0.016 0.081 0.025 Notes: The sample period is 1990-2020. The benchmark sample includes all firm-year observations in the main regression in Table 2. The excluded sample consists of firm-year observations not included in the main regression as they do not satisfy conditions (a) and (b) in the main text. “Young” refers to a dummy that equals 1 if the firm is 10 years old or younger. The entry and exit rates are the average entry and exit rates in the zip code during the previous three years. These summarystatisticsareweightedbyfirmemployment. Seetextfordetails. Table 1 provides summary statistics for the benchmark sample and the excluded sample—that is, where there is at most only one large bank and thus no possibility of M&A induced branch closure. The unit of observation is at the firm-year level, and the summary statistics are weighted by firm employment like the regressions. Almost by definition, the bank branch network is far less extensive in the excluded sample–with an average of 4.5 nearby branches (versus 36 in the benchmarksample). Perhapsreflectingthealreadysparsernetwork,theaverageshareofbranches closing in the excluded sample is also lower at 1.5% (versus 3.0%). Despite these differences in the local bank branch network, the representative small firm employee in both samples has about 8

eight coworkers, with almost 60% working at firms that are more than ten years old. Employment growth,onaverage,isnegative. Butthisisbalancedbyanaveragefirmentryratethatexceedsthe average exit rate. In the excluded sample, the intensive margin employment growth is a bit lower, butthegapbetweenentryandexitrateisalsoabitlarger. 4 Impact of Branch Closures on Small Business Employment Growth Figure1depictstheestimatedcumulativeeffectofclosing25%ofbrancheswithin5kilometersof thesmallbusinessesinoursample.5 Theinitialimpactissmallandstatisticallyinsignificant,butit slowly grows to a statistically significant 2 percentage points after three years and to 4 percentage points after five years. The persistence emerging from Figure 1 is consistent with the finding by Nguyen (2019) that branch closures lead to persistent declines in loan originations for up to six years. Moreover,itislikelythatemploymentdecisionsbysmallbusinessesreacttobranchclosures withalagascreditneedsgraduallyariseandremainunmet. Figure1: EmploymentGrowthAfterBranchClosures 0 −2 −4 −6 1 2 3 4 5 Years since branch closures )pp( etar htworG Notes: The figure shows the cumulative impact on employment growth after 25% of nearby branches shut down. The gray shaded area corresponds to 95% confidence intervals. See text fordetails. 5Weestimatedequation(1)varyingthehorizonhfrom1to5years. 9

Table 2 shows more details regarding the effect on the cumulative 5-year intensive-margin growth rate. The first-stage estimates speak to the strength of the instruments, as the coefficients are highly statistically significant and the F statistic is above 100. The coefficients imply that the closure rate of branches exposed to possible M&A induced closure is several times (or about 5 percentage points) higher than for other branches. In comparing columns (1) and (2), we see that theimpactofbranchclosuresisbasicallynullintheOLSspecificationandnegativeintheIVspecification. However, downward bias in our OLS estimates associated with error in our measure of branch closure rates should be mitigated by the instrumental variable approach. Focusing on (2), the coefficient of -0.157 implies that, if all of the branches shut down within 5 kilometers, firms’ employment growth would decrease by almost 16 percentage points over the next five years. Our estimates,however,areidentifiedfromclosuresofbranchesoflargebanksthatarenearotherbank branches. Closures of branches that create “banking deserts” may be more detrimental. Additionally, the previous literature has emphasized that small banks tend to be more important for small business lending (Berger, Bouwman and Kim (2017)) and that distance matters more in relationships with small banks (Brevoort and Hannan (2006)). Therefore, our estimates may be a lower boundonthe“true”effectofbranchclosuresonrealactivity. Table3showstheimpactofbranchclosuresontheentryandexitratesofsmallbusinessatthe zip code level. The effect is both economically and statistically significant for the entry rate and null on the exit rate. More precisely, if 25% of branches shut down, the entry rate is about 1.3% lowerthefollowingyear,whichisapproximately14%oftheaverageentryrate,usingtherelevant statistic from Table 1.6 The LBD does not include the self employed, thus entry and exit could reflecttransitionsfromandtofirmsownedandoperatedbytheirsoleemployee. Our evidence on intensive and extensive margin small firm employment growth combined imply that branch closures are related to a decline in small business employment. While we use merger activity as an instrument, our results do not, however, directly imply that bank mergers hamper small business growth. Indeed, mergers may lead to a reallocation of branches. The NIC 6Weexploredtheeffectonentryandexitratesafterthefirstyearsincebranchclosureanddonotfindsignificant effectsinyearstwothroughfive. 10

Table2: BRANCH CLOSURES AND SMALL BUSINESS GROWTH: INTENSIVE MARGIN Dependentvariable: 5-yearaheadFirmEmploymentgrowth (1) (2) OLS IV Sharebranchesclosed 0.001 -0.157 (0.003) (0.052) Numobservations 41,260,000 41,260,000 IVFirststage Dependentvariable: Sharebranchesclosed Sharebranchesexposed 0.042 t,t−1 (0.004) Sharebranchesexposed 0.056 t−1,t−2 (0.004) Sharebranchesexposed 0.032 t−2,t−3 (0.004) EffectiveF-stat 112.5 NOTES: Allregressionsincludetheshareofbranchesexposedtoanymergerduringthepastthree years; the average employment growth rate in the zip code over the previous three years; the average entry rate in the zip code in the previous three years; the average exit rate in the zip code during the previous three years; a dummy equal to one if the firm is at most 10 years old; as well as fixed effects for each firm and year by four-digit sector and county by year. Robust standard errors are clustered at the zip code level and reported in parentheses. Figures are computed using employment weights. Sample: all stand-alone firms with 20 or fewer employees in years 1990- 2020,excludingthefinance,insurance,real-estate,andgovernmentsectors. Source: LBDandNIC database. TheeffectiveF-statisfromMontielOleaandPflueger(2013). datashowthatabouthalfofthetime,themergingbanksgrowtheircombinednumberofbranches faster than that of other banks over the following three years. Thus, it might be possible that new branchescreatedbyamergermightpositivelyaffectsmallfirmformationandgrowth. Our findings contrast with Greenstone et al. (2020), who show that a negative lending shock translates to reduced lending to small businesses but close-to-zero changes in employment.7 We believe that our identification is cleaner as branch closures are easier to measure than lending shocks, and relying on the quasi-exogeneous nature of branch closures induced by M&A activity 7In results available from the authors upon request, we find that our instrumentation approach also finds a link betweenbranchclosuresandnearbysmallbusinesslending. 11

Table3: BRANCH CLOSURES AND SMALL BUSINESS GROWTH: EXTENSIVE MARGIN Dependentvariable: Entryrate Exitrate (1) (2) (3) (4) OLS IV OLS IV Sharebranchesclosed -0.001 -0.051 0.003 -0.005 (0.001) (0.021) (0.001) (0.015) Numobservations 197,000 197,000 197,000 197,000 IVFirststage Dependentvariable: Sharebranchesclosed Sharebranchesexposed 0.044 t,t−1 (0.004) Sharebranchesexposed 0.057 t−1,t−2 (0.004) Sharebranchesexposed 0.034 t−2,t−3 (0.004) EffectiveF-stat 125.0 NOTES: All regressions include the share of branches exposed to any merger in the past three years, theaverage employmentgrowth ratein the zipcode inthe previousthree years, theaverage entry rate in the zip code in the previous three years, the average exit rate in the zip code in the previous three years, the share of employment in the zip code in each two-digit NAICS sector, the share of employment in the zip code in firms that are at most 10 years old, fixed effects county by year. Robuststandarderrorareclusteredatthezip-codelevelandreportedinparentheses. Figures arecomputedusingemploymentweights. Sample: 1990-2020. Source: LBDandNICdatabase. is preferable to a shift-share identification strategy that requires independence of banks’ small businesslendingstrategyandtheirregionalpresence. Moreover,ouranalysiscoversalongertimeperiodandourunitofanalysisisatthefirmlevel. 5 Robustness Inthissection,weexploredifferentregressionspecificationstotesttherobustnessoftheresultsin Table 2 and 3. These specifications examine how effects relate to, inter alia, the size of the firms, thesizeofthemergingbanks,andourmeasureofproximitybetweenfirmsandbranches. 12

Table4: ROBUSTNESS Dependentvariable: FirmEmploymentgrowth Entryrate Exitrate Coefficient EffectiveF-stat Coefficient Coefficient EffectiveF-stat Baseline -0.157 112.5 -0.051 -0.005 125.0 (0.052) (0.021) (0.015) (1)<10employeefirms -0.123 118.7 -0.061 -0.009 131.5 (0.050) (0.023) (0.017) (2)<30employeefirms -0.134 112.2 -0.062 -0.012 124.7 (0.053) (0.022) (0.015) (3)Asset-basedbanksize -0.131 117.9 -0.045 -0.011 136.2 (0.052) (0.021) (0.015) (4)Brancheswithin10kilometers -0.267 116.9 0.008 0.031 124.1 (0.076) (0.032) (0.021) (5)NocontrolsandnofirmFE -0.161 238.1 (0.052) NOTES: “Coefficient” refers to the estimated coefficient on the share of branches closed. The regression specifications are described in the main text. The effective F-stat for the entry rate and exit rate regressions is the same and shown in the last column of the table. Robust standard errors are clustered at the zip code level and reported in parentheses. Figures are computed using employmentweights. Sample: 1990-2020. Source: LBDandNICdatabases. Starting with the firm-level intensive margin regressions, we first ask whether our conclusions about small firm growth depend on the size of the small business in question. So, we run the regression using the sample of firms with 10 or fewer employees and again with 30 or fewer employees, respectively. The new estimates in rows (1) and (2) of Table 4 suggest that the exact firm size threshold does not matter. In row (3), we present regression results when defining large banks as those above the 99th percentile in assets, as opposed to by number of branches. This definition affects the set of mergers our instrument is based on, and slightly weakens our results. In row (4), we define branches as nearby small firms if their zip codes are within 10 kilometers instead of 5 kilometers. We find a stronger impact of branch closures if we consider a wider radius. Finally, in row (5), we present results of the regression when dropping controls and firm fixedeffects. Ourresultsareessentiallyunchanged. We next apply the same methodological variations to our zip code level extensive margin regressionsinthefinalthreecolumnsofTable4. Thefirmandbanksizethresholdshavelittleimpact 13

onourestimatedfirmentranceandexitrateeffects. However,entranceandexitrateeffectsarestatistically insignificant when considering nearby branches as those up to 10 kilometers from the smallfirms. 6 Conclusion Ouranalysissuggeststhatbankbranchclosureshaveasignificanteffectonlocaleconomicactivity through small business credit access. Using data from the U.S. Census Bureau and the Federal ReserveSystem’sNationalInformationCenterdatabase,weshowthatsmallbusinessemployment grows more slowly in the five years following nearby branch closures and that the entry rate of small businesses declines. This suggests that there are limits to the extent that small businesses can substitute for broken relationship-based lending opportunities by going to another bank or a non-banklender. However, our analysis does not speak directly to the recent rise of FinTech lenders. Indeed, in recentyears,billionsofdollarshavebeeninvestedinonlinefinancialservices,whilein-branchvisits now account for only a fraction of banking transactions. This development could have reduced the role of branches. Variation in our instruments occurs mostly in the early to middle part of our panel,whichlimitsourabilitytodrawinferencesaboutrecentyears. However,westillbelieveour resultsareinformative. Adamsetal.(2023)showthat,whileaveragedistancehasincreased,banks themselves have not materially increased their lending distances. Outside of a very small subset of specialized loans, small businesses remain dependent on local banks. Thus, there is reason to believethatnearbybranchclosuresstillmatterinanageofFinTechlenders. 14

References Adams,RobertM,KennethPBrevoort,andJohnCDriscoll,“Islendingdistancereallychanging? Distancedynamicsandloancompositioninsmallbusinesslending,”JournalofBanking &Finance,2023,156. Agarwal, Sumit and Robert Hauswald, “Distance and Private Information in Lending,” Review ofFinancialStudies,2010,23(7),2757–2788. Amberg,NiklasandBoBecker,“Bankingwithoutbranches,”workingpaper2024. Amiti,MaryandDavidE.Weinstein,“HowMuchDoIdiosyncraticBankShocksAffectInvestment? Evidence from Matched Bank-Firm Loan Data,” Journal of Political Economy, 2018, 126(2),525–587. Bellucci, Andrea, Alexander Borisov, and Alberto Zazzaro, “Do banks price discriminate spatially? Evidence from small business lending in local credit markets,” Journal of Banking & Finance,2013,37(11),4183–4197. Benson,David,SamuelBlattner,SerafinGrundl,YouSukKim,andKenOnishi,“ConcentrationandGeographicProximityinAntitrustPolicy: EvidencefromBankMergers,”American EconomicJournal: Microeconomics,forthcoming. Berger, Allen N. and Gregory F. Udell, “Small Business Credit Availability and Relationship Lending: The Importance of Bank Organisational Structure,” The Economic Journal, 2002, 112(477),F32–F53. , Christa H. S. Bouwman, and Dasol Kim, “Small Bank Comparative Advantages in Alleviating Financial Constraints and Providing Liquidity Insurance over Time,” The Review of FinancialStudies,2017,30(10),3416–3454. 15

Berger, Allen N, Nathan H Miller, Mitchell A Petersen, Raghuram G Rajan, and Jeremy C Stein, “Does function follow organizational form? Evidence from the lending practices of largeandsmallbanks,”JournalofFinancialeconomics,2005,76(2),237–269. Berton,Fabio,SauroMocetti,AndreaFPresbitero,andMatteoRichiardi,“Banks,firms,and jobs,”TheReviewofFinancialStudies,2018,31(6),2113–2156. Bonfim,Diana,GilNogueira,andStevenOngena,““Sorry,we’reclosed”bankbranchclosures, loanpricing,andinformationasymmetries,”ReviewofFinance,2021,25(4),1211–1259. Bottero, Margherita, Simone Lenzu, and Filippo Mezzanotti, “Sovereign debt exposure and the bank lending channel: Impact on credit supply and the real economy,” Journal of InternationalEconomics,2020,126,1–26. Brevoort, Kenneth P. and Timothy H. Hannan, “Commercial Lending and Distance: Evidence from Community Reinvestment Act Data,” Journal of Money, Credit and Banking, 2006, 38 (8),1991–2012. Chodorow-Reich, Gabriel, “The employment effects of credit market disruptions: firm-level evidence from the 2008-9 financial crisis,” Quarterly Journal of Economics, 2014, 129 (1), 1–59. Chow, Melissa C., Teresa C. Fort, Christopher Goetz, Nathan Goldschlag, James Lawrence, Elisabeth Ruth Perlman, Martha Stinson, and T. Kirk White, “Redesigning the LongitudinalBusinessDatabase,”CenterforEconomicStudies(CES)WorkingPaperSeriesCES- 21-082021. Correia,Sergio,“LinearModelswithHigh-DimensionalFixedEffects: AnEfficientandFeasible Estimator,”TechnicalReport2016. WorkingPaper. Degryse, Hans and Steven Ongena, “Distance, Lending Relationships, and Competition,” The JournalofFinance,2005,60(1),231–266. 16

Greenstone, Michael, Alexandre Mas, and Hoai-Luu Nguyen, “Do Credit Market Shocks Affect the Real Economy? Quasi-experimental Evidence from the Great Recession and “Normal” Economic Times,” American Economic Journal: Economic Policy, 2020, 12 (1), 200 – 225. Heblich,StephanandAlexTrew,“Bankingandindustrialization,”JournaloftheEuropeanEconomicAssociation,2019,17(6),1753–1796. Hollander,StephanandArntVerriest,“Bridgingthegap: thedesignofbankloancontractsand distance,”JournalofFinancialEconomics,2016,119(2),399–419. Huber, Kilian, “Disentangling the effects of a banking crisis: Evidence from German firms and counties,”AmericanEconomicReview,2018,108(3),868–898. Jarmin, Ron S. and Javier Miranda, “The Longitudinal Business Database,” Center for EconomicStudies(CES)WorkingPaperSeriesCES-02-172002. Laura,ManuelGarc´ıa-SantanaAlfaroandEnriqueMoral-Benito,“Onthedirectandindirect real effects of credit supply shocks,” Journal of Financial Economics, 2021, 139 (3), 895 – 921. Levine, Ross, Chen Lin, Qilin Peng, and Wensi Xie, “Communication within banking organizations and small business lending,” The Review of Financial Studies, 2020, 33 (12), 5750– 5783. Mann, Robert, “Bank Competition, Local Labor Markets, and the Racial Employment Gap,” workingpaper2022. Nguyen, Hoai-Luu Q., “Are Credit Markets Still Local? Evidence from Bank Branch Closings,” AmericanEconomicJournal: AppliedEconomics,2019,11(1),1–32. Olea, Jose Luis Montiel and Carolin Pflueger, “A Robust Test for Weak Instruments,” Journal ofBusiness&EconomicStatistics,2013,31(3),358–369. 17

Petersen, Mitchell A and Raghuram G Rajan, “The effect of credit market competition on lendingrelationships,”TheQuarterlyJournalofEconomics,1995,110(2),407–443. and , “Does distance still matter? The information revolution in small business lending,”TheJournalofFinance,2002,57(6),2533–2570. 18