Banking Consolidation and Small Firm Financing for Research and Development

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Banking Consolidation and Small Firm Financing for Research and Development Andrew C. Chang 2016-029 Please cite this paper as: Chang, Andrew C. (2016). “Banking Consolidation and Small Firm Financing for Research andDevelopment,”FinanceandEconomicsDiscussionSeries2016-029. Washington: Board of Governors of the Federal Reserve System, http://dx.doi.org/10.17016/FEDS.2016.029. 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.

Banking Consolidation and Small Firm Financing for Research and Development Andrew C. Chang∗ April 8, 2016 Abstract This paper examines the effect of increased market concentration of the banking industry caused by the Riegle-Neal Interstate Banking and Branching Efficiency Act (IBBEA) on the availabilityoffinanceforsmallfirmsengagedinresearchanddevelopment(R&D).Imeasure thefinancingdecisionsofthesesmallfirmsusingabalancedpanelofSmallBusinessInnovation Research (SBIR) applications. Using difference-in-differences, I find IBBEA decreased the supply of finance for small R&D firms. This effect is larger for late adopters of IBBEA, whichtendedtobestateswithstrongersmallbankingsectorspre-IBBEA. JELClassificationCodes: G21,G28,G39,O30 Keywords: BankingDeregulation;IBBEA;InterstateBankBranchingDeregulation;MarketConcentration;Riegle-Neal;ResearchandDevelopment;R&D;SmallBusinessInnovation Research;SBIR ∗BoardofGovernorsoftheFederalReserveSystem. 20thSt. NWandConstitutionAve.,WashingtonDC20551 USA. +1 (657) 464-3286. a.christopher.chang@gmail.com. https://sites.google.com/site/andrewchristopherchang/. ThisresearchwassupportedbyseparategrantsfromtheDepartmentofEconomicsandtheGraduateDivisionatthe University of California - Irvine. However, the opinions expressed in this paper are mine and do not necessarily reflectthoseoftheuniversityortheBoardofGovernorsoftheFederalReserveSystem. Ithankanonymousreferees, StephanieAaronson,VolodymyrBilotkach,MarianneP.Bitler,NickBloom,WilliamA.Branch,JiaweiChen,Linda R.Cohen,AmihaiGlazer,IvanJeliazkov,DavidJenkins,JakubKastl,DavidNeumark,GaryRichardson,GeorgeC. Saioc, Matthew Shum, and Don Walls for helpful comments. I also thank Clara C. Asmail, Nicholas Cormier, Kay Etzler,andLeslieA.JensenforhelpwithSBIRdata;DanielJ.Wilsonforgraciouslyprovidingstate-leveldataonthe usercostofR&D;andTylerJ.Hansonforvaluableresearchassistance. Anyerrorsaremine. 1

1 Introduction This paper examines how the deregulation and subsequent consolidation of the U.S. banking industry affected financing for small firms engaged in research and development (R&D). Because technological development drives economic growth (Solow, 1957), and because small firms are a significant driver of innovation (Acs and Audretsch, 1987), understanding the link between small firmfinanceandtheirR&Dexpendituresisanimportantissue. I analyze the effect of the Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994 (IBBEA) on the propensity for small firms to apply for external R&D funding through the U.S. government’s Small Business Innovation Research (SBIR) program, an award program for R&D projects at both private and public small firms.1 Prior to IBBEA, there were geographic restrictions on interstate bank branching, which dated back to the beginning of the U.S. banking system. Implemented on a state-by-state basis, IBBEA removed geographic restrictions on interstate bank branching and consolidated the banking industry. Figure 1, which plots the Herfindahl Index of the U.S. banking sector from fiscal year (FY) 1987 to FY 2000, shows an increase in marketconcentrationafterIBBEApassed. With a negative binomial model and difference-in-differences, I estimate the effect of IBBEA on SBIR applications with a balanced panel of state-year SBIR application counts. I find IBBEA increased SBIR applications between 10 to 50 percent. This effect is present for all states, but I find larger effects for states that adopted IBBEA later. Later adopters of IBBEA had stronger smallbankingsectors,morepotentialtargetsfor interstatebankmergers,andthepotentialtohave a larger change in banking structure due to the deregulation caused by IBBEA (Kroszner and Strahan,1999). The key identification assumption for relating changes in SBIR applications to implications about small R&D firm finance is that SBIR awards and bank finance are substitutes. Under this assumption, if IBBEA made it more difficult for firms to secure funding for R&D projects, then 1For a review of U.S. banking deregulation, including IBBEA, see Johnson and Rice (2008). For a summary of SBIR,seeLerner(1999). 2

firms should have sought an alternative source of funding in SBIR awards and we should observe anincreaseinSBIRapplicationsduetoIBBEA. EvidenceonthefinancingdecisionsofsmallR&Dfirmssupportstheideathatthesefirmsview SBIRfinanceandallothersourcesoffinanceassubstitutes. CombiningSBIRgrantswithfunding from other sources imposes costs on the firm and restricts the firm’s use of its inventions. The imposed costs include SBIR application time and annual reporting requirements.2 In addition, as a condition of funding research with an SBIR grant, the firm must disclose all invention(s) that come from the grant to the government and provide the government a license to use the disclosed invention(s). IfthefirmpatentsorlicensestechnologyfundedbySBIR,thenthefirmmustsubmit an annual utilization report on the technology to the government. The government may also force the firm to grant licenses to other firms when the government deems licensing to be in the public interest (National Institutes of Health, 1995). For SBIR award winning firms, Wallsten (2000) estimatesanelasticityofsubstitutionbetweenSBIRawarddollarsandallothersourcesoffunding as negative 0.82, indicating nearly dollar-for-dollar substitution between SBIR awards and other sourcesoffinanceforR&D. Assuming bank finance and SBIR are substitutes, my estimated 10 to 50 percent increase in SBIR applications implies that IBBEA diminished bank finance for small R&D firms. Because I measure changes in finance with SBIR applications, as the balance sheet data on small R&D firms are unavailable, I cannot say exactly how much the supply of finance changed because of banking consolidation. However, with some assumptions, a rough calculation suggests that IBBEA decreased the supply of finance for small R&D firms in FY 1999 between 138 - 508 million dollars.3 In FY 1999 the budget for SBIR was $1.25 billion. The total value of SBIR finance to the IBBEA-induced applicants should be 82 percent of the value of lost bank fiance, using the elasticity of substitution from Wallsten (2000) of negative 0.82 between SBIR and bank finance, as well as making the following assumptions: (1) firms did not consider the 2InformalconversationswithSBIRadministratorssuggestthattheinitialapplicationforSBIRtakesapproximately 200hourstocomplete. 3Dollar figures in this paragraph are in 2005 dollars, inflated with the Bureau of Economic Analysis’ (BEA’s) implicitpricedeflator(BEA,2011). 3

fact that their award chances changed because of banking consolidation and (2) firms that were induced into applying for SBIR subsequent to IBBEA had an equal expected value of an SBIR application as a firm that would have applied without IBBEA. Because the estimates indicate that SBIR applications increased 10 to 50 percent, the lower bound on the change in finance should be (1− 1 )∗1.25billion=0.82∗bankfinancelowerbound, or 1 ∗(1− 1 )∗1.25billion= 1+0.1 0.82 1+0.1 138million=bankfinancelowerbound. A similar calculation for the upper bound is 1 ∗(1− 0.82 1 )∗1.25billion=508million=bankfinanceupperbound. 1+0.5 My panel of state-year SBIR application counts is a data set that offers three advantages for this study. First, because the data set comes from administrative records, it is free of the selfreportingbiaspresentinsurveydata.4 Second,thedatasetisabalancedpanelofSBIRapplication counts that is free of the survivorship bias usually present in bank or firm-level data. Third, the panel of SBIR application counts represents both public and private companies as opposed to just public companies (for example, companies in Compustat). This last advantage allows an analysis of small private companies, which are important to the conduct of R&D and for which little data are available. Previous research on banking deregulation looks at the effect of deregulation on the average small firm or the effect on lending for small firms at the state level, finding mixed results on how deregulation affected the supply of credit (Jayaratne and Wolken, 1999; Vera and Onji, 2010). Tomyknowledge,thispaperisthefirstthatexaminestheeffectsofIBBEAspecificallyon small,privateR&Dfirms. 2 Institutional Details of IBBEA and SBIR ThissectionreviewsIBBEAandSBIR.IdiscusshowIBBEAledtotheconsolidationofthebankingindustry. ForSBIR,Idescribehowtheprogramisstructuredandpresentsummarystatisticsof smallR&DfirmsthatareeligibleforSBIR. 4See,forexample,BergerandUdell(1995)andPeekandRosengren(1996)forevidenceoferrorsinsurveydata. 4

2.1 IBBEA Passed on September 29, 1994, IBBEA set a default opt-in trigger date of June 1, 1997. However, statescouldoptintoIBBEAearlyoroptoutofIBBEAentirelybythedefaultopt-indate. Approximatelyone-thirdofstateswaiteduntilthetriggerdatetooptintoIBBEA’sprovisions,andseveral states had active debates on opting out of IBBEA (Kane, 1996; Johnson and Rice, 2008). Texas andMontanainitiallyoptedout,althoughbothstateslateroptedin. Table1showstheinitialopt-in dates for each state. Subject to certain conditions, IBBEA allowed several new types of banking activity: (1) interstate bank acquisitions, (2) interstate agency operations, (3) interstate branching, and(4)denovobranching(JohnsonandRice,2008). IBBEA changed the structure of the banking industry. Prior to IBBEA, in 1994 there were only62out-of-statebankbranches—lessthan1percentofthetotalnumberofbranches. Withthe removal of the restrictions on interstate branching, by 1999 there were more than 10,000 out-ofstate bank branches — approximately 20 percent of the total number of bank branches (Johnson andRice,2008). In addition to allowing interstate bank branches, IBBEA increased interstate bank mergers, which contributed to the consolidation of the banking industry documented in Figure 1. Figure 2 plots interstate bank mergers during the 1990s and shows a significant increase in interstate mergerssubsequenttoIBBEA. IBBEA fueled research on the relationship between banking consolidation, deregulation, and finance.5 This research may have been encouraged by suspicions that IBBEA was detrimental for small firm finance, which was also a chief concern in Congress when IBBEA was debated (U.S. Congress,1993). 5ExamplesincludeBerger,Saunders,Scalise,andUdell(1998);ColeandWalraven(1998);PeekandRosengren (1998);StrahanandWeston(1998);JayaratneandWolken(1999);CraigandHardee(2007);RiceandStrahan(2010); VeraandOnji(2010);andCornaggia,Mao,Tian,andWolfe(2015). 5

2.2 SBIR Congress created the SBIR program in 1982 in part to combat market failures associated with R&D.6 Public Laws 97-219, 99-443, and 102-564 require each federal agency with an extramural research program greater than $100 million to set aside a fixed percentage of the agency’s extramural research budget for SBIR. Certain characteristics of the SBIR program are mandatory (for example, the set-aside percentage), and the Small Business Administration (SBA) oversees the general SBIR program. However, each agency administers its own SBIR program separately, whichgivestheindividualagenciessomeflexibilitytomeetSBIR’scongressionalmandates.7 Qualified businesses that can receive an SBIR award are located in the U.S., are for-profit, are at least 51 percent owned by U.S. citizens or permanent residents, and employ a maximum of 500 employees. Financial data on firms that applied for SBIR are not available. However, the 1993 National Survey of Small Business Finances (NSSBF) contains financial and other demographic data on small firms that employed R&D workers, which are the types of firms that would have applied for SBIR grants (Board of Governors of the Federal Reserve System, 1993).8 Most of the surveydescribescharacteristicsoffirmsin1993,althoughsomequestionsreferenceeither1992or 1994. Table 2 displays characteristics of firms that employed R&D workers from the 1993 NSSBF. Column (1) describes all firms that employed R&D workers in 1992. Column (2) restricts the sampletoonlyfirmswithR&Dworkersin1992thatalsoappliedforventurecapitalfrom1992to 1994. Thetableinflatesalldollarfiguresto2005dollarswiththeBEA’simplicitpricedeflator. Column (1) shows that in 1992, the average small R&D firm had 12 workers, almost a quarter million dollars in payroll, and more than $1.6 million in sales. Importantly for this paper, 38.9 6Formarketfailures,inadditiontotheliquidityconstraintproblemarisingfromuncertaintyandasymmetricinformation(Arrow,1962;HallandLerner,2009),R&Dalsosuffersfromanappropriationissue(Griliches,1992). 7As of FY 1999, the participating agencies were the Departments of Agriculture, Commerce, Defense, Education, Energy, Health and Human Services, and Transportation; the Environmental Protection Agency; the National AeronauticsandSpaceAdministration;andtheNationalScienceFoundation. 8TheNSSBFdefinesasmallfirmasonewithfewerthan500employees,whichisthesamedefinitionthatSBIR uses. The NSSBF has data available for 1987, 1993, 1998, and 2003. However, the 1998 edition did not collect informationonR&Demployees,andthe1987and2003editionsareoutsideofthispaper’ssampleperiod. 6

percent of these firms applied for a loan from 1993 to 1994 and, on average, were approved for morethanhalfamilliondollars. Thesedatashowthatloansareanimportantsourceoffinancefor these firms. Firms that also applied for venture capital were larger, on average, than those that did not. Inaddition,firmsthatappliedforventurecapitalalsosecuredloansasasourceoffinance. AgenciesdividetheirSBIRawardsintoeithertwoorthreephases. APhaseIawardisforafirm to explore the technical and commercial feasibility of the R&D project. If the results of the Phase I project are promising, the firm may be invited to apply for a Phase II award to further develop and commercialize the idea. Firms cannot undertake a Phase II project without first completing Phase I. Some agencies also have a Phase III program, which involves partnering the firm with a collaborator;thisphasedoesnotprovideadditionalgovernmentSBIRmoney. Fortheempiricalanalysis,IusethetotalstatebyFYSBIRPhaseIapplicationsfortheagencies with the largest SBIR budgets: the Departments of Defense, Energy, and Heath and Human Services; the National Aviation and Space Administration (NASA); and the National Science Foundation (NSF). These five agencies compose more than 96 percent of the SBIR budget in each FY (NationalScienceBoard,2008). IusePhaseIapplicationsasthedependentvariablebecausethese give the strongest indicator of the effort small R&D firms expend to seek external finance.9 Phase II and Phase III applications represent a mixture of firm effort and agency politics, as they are conditional on good progress in earlier SBIR phase(s) and can require an invitation by the SBIR agencytoevenapply. 3 Model and Data 3.1 Model Two features of IBBEA’s deregulation are important for this study. First, the removal of banking restrictions consolidated the banking industry and potentially affected the cost of credit (Cole and Walraven, 1998; Peek and Rosengren, 1998; Strahan and Weston, 1998; Jayaratne and Wolken, 9Unfortunately,Idonotobservethetotaldollaramountappliedfor,onlytheapplicationcount. 7

1999; Craig and Hardee, 2007; Rice and Strahan, 2010; Vera and Onji, 2010). Second, because there is between-state variation in deregulation dates, I can identify the effect of IBBEA on small R&Dfirmfinanceinatreatment-controlsetup. Because the dependent variable, SBIR applications, is a count variable, I estimate a negative binomialmodel(CameronandTrivedi,1998,2005). Thenegativebinomialmodelis y exp(−λ )λ i,t i,t i,t f(y |λ )= , λ =µ ν (1) i,t i,t i,t i,t y ! i,t where µ =exp(X(cid:48) β) (2) i,t i,t iid ν ∼g(ν|α) (3) E(y |µ,α)=µ (4) i,t i,t Var(y |µ ,α)=µ +αµ2 (5) i,t i,t i,t i,t In equations (1) to (5), i is a state, t is the FY, exp(•) is the exponential function, E(•) is the expectations operator,Var(•) is the variance, X is a matrix of covariates, and α, β and δ are parameterstobeestimated. In addition to the fact that the negative binomial model only predicts non-negative outcomes, the negative binomial model’s estimated marginal effects account for heterogeneous state sizes. Differentiating the conditional mean in equation (4) with respect to a single covariate x , the exj pectedmarginaleffectforywithrespecttox is j dE(y|X) =β ×exp(X(cid:48) β) (6) j dx j 8

which depends on the parameter for covariate x , β , the entire matrix of covariates X, and their j j associatedparametersβ throughthetermexp(X(cid:48)β).10 3.2 Policy Variable A standard policy variable is an indicator for post-deregulation that assumes a uniform effect of deregulation over time. I instead construct the policy variable to allow for time-varying effects of deregulation. I divide states into three cohorts, one for each FY from the passage of IBBEA to the IBBEA triggerdate: (1)deregulatorsbyFY1995,(2)deregulatorsinFY1996,and(3)deregulatorsinFY 1997. For each cohort, I model the policy implementation as a series of time dummies beginning intheyearimmediatelyafterthecohortpassesIBBEA.11 Forexample,fortheearliestderegulation cohort (by FY 1995), there are four time dummies: FY 1996, FY 1997, FY 1998, and FY 1999. For the FY 1996 deregulators, there are three dummies: FY 1997, FY 1998, and FY 1999. The samepatternholdsforthelastcohort. Thistypeofpolicyvariableallowsheterogeneouseffectsof IBBEAbyderegulationcohortaswellasthroughtime. Formally, let D be a year dummy for FYt and D be a dummy for state i if it deregulated in t i,τ FYτ. TheconditionalmeanforstateiinFYt withthepolicyvariableis: 1999 1999 1999 µ =exp( ∑ ζ D D + ∑ η D D + ∑ θ D D +X(cid:48) β) (7) i,t t t i,1995 t t i,1996 t t i,1997 i,t t=1996 t=1997 t=1998 In equation (7), ζ represents the effect of IBBEA on SBIR applications in FYt for the group t of states that deregulated by FY 1995, η represents the effect of IBBEA on SBIR applications in t FYt forthegroupofstatesthatderegulatedinFY1996,andθ representstheeffectofIBBEAon t SBIRapplicationsinFYt forthegroupofstatesthatderegulatedinFY1997. I model the policy variable using equation (7) instead of the standard policy indicator variable to be completely flexible for allowing time-varying effects of IBBEA by deregulation cohort. 10TestsforoverdispersionconsistentlyrejectH : α =0. vs. H : α (cid:54)=0. 0 A 11Section5considersalternatetimingsofthepolicyvariablethatproducesimilarresults. 9

There are at least three reasons to expect that IBBEA had different effects both over time and by deregulation cohort. One reason is that when states deregulated, it affected the banking industry in states that had already deregulated. For example, when states deregulated in 1995, banks could conduct interstate mergers but only between banks in the deregulated states. When the next wave of states deregulated in 1996, the banks in these states could merge with other banks in the newly deregulatedstatesandalsowithbanksinstatesthatwerealreadyderegulated. Similarly,statesthat werealreadyderegulatedhadanewinfluxofbankswithwhichtheycouldmerge. Therefore,each new wave of deregulation affected the banking industry in both the newly deregulated states and the states that had already deregulated, which implies that IBBEA had time-varying effects and makesthestandardindicatorpolicyvariableunsatisfactory. A second reason to expect different effects of IBBEA by deregulation cohort is that later adopters of IBBEA had stronger small banking sectors than early adopters (Kroszner and Strahan, 1999). Therefore, for later adopters there was a potential for a greater amount of change post-IBBEA. A third reason is a timing and learning story. Suppose that when the first wave of deregulation passed in 1995, banks were unfamiliar with the procedures needed to instigate the now-legal mergers. Therefore, some banks in the states that deregulated in 1995 may have delayed merging. However,by1997,banksmayhavebeenfamiliarwiththeseproceduralhurdlesandcouldexecute mergersmorequicklythanwhenIBBEAwasfirstpassed. Inthisscenario,wecanexpecttheeffect of deregulation on market concentration to vary over time. I model IBBEA’s time-varying effect astheflexibleforminequation(7)tobecompletelyagnosticonthemechanismbehindchangesin bankingconcentration.12 12An even more flexible policy variable would be able to take into account the degree of deregulation, as states had some latitude to restrict interstate branching (Johnson and Rice, 2008). However, there is not a clear way to parametrize the dimensions to which states were allowed to deregulate. As a robustness check, I create an alternate policy variable that takes into account the restrictions on interstate branching based on the indicator from Rice and Strahan (2010) with the same time-series form as equation (7). This alternate policy variable indicates that SBIR applications increased between 7 to 15 percent by FY 1999, calculated for the mean deregulator, but the estimates are less precise (significant at the 10 percent level to insignificant), which suggests that the dimensions by which stateswereallowedtorestrictinterstatebranchingmightnotbeimportantfactorsindeterminingtheeffectofbanking consolidationonSBIRapplications. 10

3.3 Dependent Variable and Controls The dependent variable is state-FY SBIR Phase I applications that come from a balanced panel from FY 1990 to FY 1999. SBIR programs at participating federal agencies are independently operated and funded. I aggregate SBIR applications from the five largest SBIR agencies to create the panel: the Departments of Defense, Energy, and Health and Human Services - National Institutes of Health; NASA; and the NSF. The SBIR programs for these five agencies compose more than 96 percent of the budget for SBIR in each FY (National Science Board, 2008). For NASA, the data are available on NASA’s website (National Aviation and Space Administration, 2010). For the remaining agencies, I query the relevant SBIR officials to obtain the data sets.13 Except for the applications from Hawaii in FY 1993 and North Dakota in FY 1994 to the Department of Defense,thedatasetcontainsdataforallfiveagenciesineachstateandFY.Theresultsarerobust toexcludingHawaiiandNorthDakota. The key identification assumption relating changes in SBIR applications caused by IBBEA to IBBEA’s effect on small R&D firm finance is that SBIR applications and bank finance are substitutes (Wallsten, 2000). Therefore, if we observe an increase in SBIR applications, then the implication is that IBBEA decreased the supply of bank finance. In this situation, firms switched from bank finance to trying to receive an SBIR award, which increased SBIR applications. The oppositeholdsforadecreaseinSBIRapplications. ToidentifytheeffectofIBBEAonSBIRapplicationrates,Iuseavarietyofadditionalcovariates that control for other factors that can influence a state’s SBIR applications. I use state fixed effectstoremovetime-invariantcharacteristicsofstatesthatcouldaffectSBIRapplications. Ialso include state-specific time trends and time dummies to control for the trend of SBIR applications priortoIBBEA. I remove the effects of the business cycle on SBIR applications using gross state product from the BEA (BEA, 2011). R&D expenditures are correlated with the business cycle, which implies 13Specifically,IusetheFreedomofInformationAct(DepartmentofDefense,2010;DepartmentofEnergy,2010; NationalScienceFoundation,2010a;NationalInstitutesofHealth,2010). 11

that the financing patterns for R&D, including SBIR applications, should also be correlated with thebusinesscycleregardlessofthestateofbankingderegulation(Barlevy,2007;Chang,2013). Changes in the number of SBIR applicants may be affected by changing demand for thefunds forSBIR,asopposedtothesupply-sideeffectsthispaperinvestigates. Forexample,ifthenumber ofsmallelectronicsfirmsinastateincreases,thenthenumberofSBIRapplicationstoagenciesthat fund R&D projects in electronics should also increase. Because firms that receive SBIR awards are primarily from the North American Industry Classification System (NAICS) industry R&D in thePhysicalSciences(NAICS541710),tocontrolfortheuniverseofpotentialapplicantsIaddthe number of employees and the total establishment count of firms in R&D in the Physical Sciences into the regressions.14,15 Total employee counts by state, six-digit NAICS industry, and FY come fromtheBureauofLaborStatistics’QuarterlyCensusofEmploymentandWages(QCEW)(BLS, 2011). Ialsoestimatespecificationswithtotalemploymentandestablishmentcountsinthetop-10 six-digitNAICScodesthatgivesimilarresults.16 The propensity for a firm to seek funding may also be a function of other state-specific factors. For example, if a state adopts policies that are more friendly to innovative activities, then it could alter the SBIR application rate for that state. Alternatively, through the fertile technology hypothesis,ifalargeamountofinnovationoccursinaparticularstate,thenitcancreateadditional 14To determine the industry composition of SBIR-award-winning firms, I use data on SBIR award winners from theSBA’sTECH-Netdatabase(SBA,2010). ThedatabaserecordsdetailsonSBIRawards: characteristicsofwinning firms, the abstracts of the SBIR proposals, amount of the award, etc. I take a random sample of 1000 SBIR awards from FY 1990 to FY 2000, divided evenly over each of the five largest SBIR agencies by budget, and use TECH- Net’s information to assign each award-winning firm to either one or two NAICS codes. I match the sampled firms from TECH-Net to publicly available databases that contain information on firms and their product lines (Dun and Bradstreet,2010;FederalGovernmentBidIntelligenceCompany,2011;GaleGroup,2010)aswellascrosscheckthe information from these databases against available public reports, company websites, published articles, and patent applicationstoaccuratelyassigntheSBIRawardeestoNAICScodes. 15Forthiscontroltobevalid,theindustrydistributionoftheSBIR-award-winningfirmsneedstobesimilartothe industrydistributionofallSBIRapplicants. Althoughtherearenodataavailabletoanalyzethisrelationshipdirectly, the most common industry classification for all SBIR applicants is likely R&D in the Physical Sciences (NAICS 541710). SBIRagenciestypicallysolicitproposalsfortechnologiestofulfillaspecificagencyresearchrequirement, andthereisnoreasontosuspectapplicantsfromoutsideR&D-intensiveindustrieswouldapplyforSBIRawards. 16TheothertopNAICScodesare:325414(biologicalproduct,excludingdiagnostic,manufacturing),334413(semiconductorandrelateddevicemanufacturing),334511(search,detection,navigation,guidance,aeronautical,andnauticalsystemandinstrumentmanufacturing),339112(surgicalandmedicalinstrumentmanufacturing),541330(engineeringservices),541380(testinglaboratories),541511(customcomputerprogrammingservices),541512(computer systemdesignservices),and541690(otherscientificandtechnicalconsultingservices). 12

opportunitiesforinnovativeactivities,whichwouldaltertheSBIRapplicationratethroughachannel other than IBBEA (Kortum and Lerner, 1998). I add three controls to the model to proxy for theinnovativeclimateinastate. First, I add total state academic R&D expenditures. Academic R&D expenditures should be correlated with the degree of innovative atmosphere in a state, particularly where basic research is concerned. Data on academic R&D expenditures come from the NSF’s WebCASPAR database National Science Foundation (2010b). Second, I use Wilson (2009)’s estimate of the state-level costofR&Dcapital,whichisafunctionofastate’scorporatetaxandR&Dtaxpolicies. Achange inthecostofR&DshouldinducefirmstochangetheirR&Dprojectportfolioandthereforeaffect theirdemandforR&Dfinancing.17 Third,Iuseutilitypatentapplications,conditionaloneventual patentapproval.18 Althoughnoisy,patentcountsofferonemeasurementoftheamountofinventive activity in a state. Data on utility patents come from the National Bureau of Economic Research (NBER) patent database, documented in Hall, Jaffe, and Trajtenberg (2001) and available from NBER(2011). Firms may also decide to apply for SBIR as a function of agency-specific investment in the state. For example, suppose the Department of Defense increases its R&D funding in Alabama. Firms in Alabama would then begin to acquire knowledge of and familiarity with the Department of Defense’s technological and R&D demands. This familiarity could induce firms in the state to apply for SBIR awards, as they would have garnered additional information on the department or have revised estimates of the expected value of an SBIR award. To control for agency-specific investment, I add total R&D obligations for U.S. performers by the five SBIR agencies into the model. DataonR&DobligationscomefromtheNSF’sWebCASPARdatabase(?). 17OtherexamplesofresearchintoR&DtaxincentivesincludeChang(2014)andGuceriandLiu(2015). 18Todeterminewhichtypesofpatentstoincludeinthismeasure,IsamplethelargestSBIRagency’s(Department ofDefense)awardwinnersfromtheSBATech-NetDatabase. Ipresentresultswherethepatentcountcontrolvariable includestotalpatentapplicationsforthefollowingtwo-digitHall,Jaffe,andTrajtenberg(2001)technologycategories: gas (13), communications (21), computer hardware and software (22), computer peripherals (23), information storage (24), electrical devices (41), electrical lighting (42), measuring and testing (43), nuclear and x-rays (44), power systems (45), semiconductor devices (46), miscellaneous electronics (49), materials processing and handling (51), metalworking(52), motors(53), optics(54), transportation(55), miscellaneousmechanical(59), andheating(66). I alsochecktheresultsusingallpatentapplications,andtheresultsarenearlyidentical. 13

Table3displayssummarystatisticsofthedependentvariableandcontrols.19 4 Results Table 4 presents the estimated average marginal effects from the baseline model.20 Column (1) presents a parsimonious specification with only aggregate time dummies, state time trends, and statefixedeffects. ThemarginaleffectsreportedindicatetheaveragechangeinSBIRapplications by cohort for each FY subsequent to deregulation relative to the pre-IBBEA period. Column (2) presents the same specification as column (1) in percent changes. For example, in the first row of results, for the states that deregulated by FY 1995, column (1) estimates deregulation to have an average effect of increasing SBIR applications by 35.8 per state in the FY immediately after deregulation. From the same row in column (2), this average marginal effect translates to a 6.71 percentincreaseoverthepre-IBBEAperiod. Addingadditionalstatecontrolsincolumn(3)yields similar effects for all IBBEA cohorts. Column (4) displays the results of column (3) in percent terms.21 From Table 4, for almost all treatment periods, the model indicates that IBBEA increased SBIR applications. Under the assumption of constant substitutability between bank finance and SBIR awards, this increase in applications implies IBBEA decreased the supply of bank finance for small R&D firms. In addition, for all cohorts this decrease in finance is exacerbated over time. For the FY 1995 cohort, immediately after deregulation (FY 1996) there is a small effect (6.71 percent) of IBBEA on SBIR applications. However, the estimates for four years after deregulation (FY 1999) indicate a 25.2 percent increase in SBIR applications. A similar upward trend in applications holds for the FY 1996 and FY 1997 cohorts, implying a downward trend in the 19Ialsoexperimentwithspecificationsthatincludethestate-levelunemploymentrateandthestate-levelunemploymentrateinteractedwiththefedfundsrate,whichisacovariatethatcouldcaptureheterogeneouseffectsofmonetary policyonSBIRapplications. Thesespecificationsgivesimilarresults,andIomitthemforbrevity. 20Forstatei’sregressor j,x i,j ,theaveragemarginaleffectis N 1 ∑N i=1d d x y i, i j ,whereN isthetotalnumberofstates. 21Removingthetimeeffectsgeneratesestimatesclosetozeroforallcohortyears. Thetimedummiesandstatetime trendsaccountforthepre-IBBEAtrendinSBIRapplications. 14

supply of finance.22 For the FY 1995 and FY 1996 cohorts, the effect of IBBEA on SBIR applications is statistically significant at the 5 percent level in 1999; for the FY 1997 cohort the effect is significant in 1998. Late adopters of IBBEA had stronger small banking sectors, more possible targets for interstate bank mergers, and the potential to have a larger change in banking structure subsequent to deregulation (Kroszner and Strahan, 1999). One possible explanation for the upwardtrendinSBIRapplicationsisthesuccessivewavesofconsolidationcausedbyIBBEA,which causedsteadilyincreasingmarketconcentration. OfthecontrolvariablesinTable4,patentcountsandagencyR&Dobligationsareindividually significant,andthejointF-testofallcontrolvariablesindicatesthecontrolsarejointlysignificant. In unreported specifications that include additional lags of the control variables or differences of thecontrols,theeffectofIBBEAonSBIRapplicationsissimilartothebaseline.23 The tendency for SBIR applications to rise from FY 1997 to FY 1999 corresponds with the large increase in market concentration of the commercial banking sector, suggesting the increased concentration of the banking industry decreased small firm finance for R&D. The estimation is consistent with the hypothesis that the relationship lending channel of small banks is more importantthanthegeographicdiversificationpotentialoflargebanksforsmallR&Dfirmfinancing.24 5 Robustness Checks Inthissection,Ipresentadditionalrobustnesschecksonthemainresultsfromsection4. 22Weightingstatesbygrossstateproductstillshowsanupwardtrendinapplicationsforallcohorts. Theweighted estimates are similar to the unweighted estimates for the FY 1995 and FY 1996 cohorts. For the FY 1997 cohort, the weighted estimates are about half of the unweighted estimates. For all cohorts, the coefficients for IBBEA are statisticallysignificantatstandardlevels. 23Tocheckforstationarityofcovariates,IruntheHarrisandTzavalis(1999)small-Tadjustedpanelunitroottest. ThetestindicatesthatacademicR&D,theusercostofR&D,patentcount,grossstateproduct,andtheestablishment count are non-stationary. To address the potential effect of non-stationary covariates on my results, I conduct two exercises: (1)Ireestimatethemodelwithonlythestationarycovariatesfromthebaselinespecification,and(2)Ifirstdifference the non-stationary covariates and reestimate the model with these differenced covariates (the Harris and Tzavalis(1999)testindicatesthatallofthefirst-differencedcovariatesarestationary). Inbothofthesespecifications, theestimatesoftheeffectofIBBEAonSBIRapplicationsaresimilartothebaseline,withIBBEAincreasingSBIR applicationsby22to36percent,dependingonthederegulationcohort. 24Evidence on the causes and consequences of relationship lending is mixed. See, for example, Elsas (2005), PresbiteroandZazzaro(2011),andMudd(2013). 15

5.1 Policy Variable Timing The first robustness check I consider is the timing of the policy variable. The policy variable in section 4 models IBBEA as taking effect the year after deregulation. However, banks could potentially respond to deregulation either faster or slower than with a one year lag. Therefore, I adjust the timing of the policy variable to start either in the year IBBEA was passed or two years after,asopposedtooneyearafter. Table 5 shows the results for changing the timing of the policy variable. Columns (1) and (2) show the results of modeling IBBEA as having an effect the year it was passed. Columns (3) and (4) model IBBEA’s effect as two years after it was passed. Columns (1) and (3) report the averagemarginaleffectinlevels,andcolumns(2)and(4)reporttheeffectsofcolumns(1)and(3) inpercents.25 The treatment patterns when shifting the timing of the policy variable in Table 5 are similar to the baseline model in Table 4. From columns (1) and (2) of Table 5, modeling the treatment as havinganeffecttheyearIBBEAwaspassedstillsuggeststhatIBBEAincreasedSBIRapplications and therefore decreased the supply of finance for small R&D firms. As with the baseline model, there is an upward trend in SBIR applications. From column (2) of Table 5, for the FY 1995 deregulation cohort there is a mere 2.95 percent increase in SBIR applications in FY 1995, but thisnumbergrowsmonotonicallyto41.5percentbyFY1999. ThesameupwardpatternsofSBIR applicationsholdfortheothertwoderegulationcohorts. Changing the treatment to have an effect two years after deregulation (columns 3 and 4) attenuates the estimated increases in SBIR applications, but the point estimates are generally still positive and increase with time as in the previous specifications. In FY 1999, relative to the baseline(Table4,column4),theestimatedincreaseinSBIRapplicationsduetoIBBEAdisappearsfor the FY 1995 cohort, changes from a 30.5 percent increase to a 17.4 percent increase for the FY 1996 cohort, and changes from a 37.7 percent increase to a 23.1 percent increase for the FY 1997 25Table5doesnotreportthecontrolvariablestosavespace,butthesignsandmagnitudesofthecontrolvariables aresimilartothebaselineinTable4. TheF-testforsignificanceofallcontrolsissignificantforatleastthe5percent level. 16

cohort. 5.2 Specific States AnadditionalconcerniswhethercertainstatesdrivetheresultsinTable4. Specifically,Iconsider NorthDakota(ND),Montana(MT),andTexas(TX)byindividuallyexcludingeachofthesestates from the estimation. Table 6 shows the results. Column (1) presents the baseline results (Table 4, column4)forcomparison. The Bank of North Dakota anchors North Dakota’s banking system as the only state-owned bank in the United States. State law tasks the bank with “promot[ing] agriculture, commerce, and industry in North Dakota” (Bank of North Dakota, 2011). Because of the presence of a stateowned bank, the effect of IBBEA on North Dakota could have been different from the average effect across states. Therefore, I re-estimate Table 4 without North Dakota, which yields similar resultsasshowninTable6,column(2). Next, I turn my attention to the control group, which varies over time. For example, in FY 1995, identification of the effect of IBBEA uses states that deregulated later than FY 1995 as a control group. In FY 1996, the model identifies the effect of IBBEA on deregulators in FY 1995 and FY 1996 using the control group of all states that deregulated later than FY 1996. Finally, identifyingtheeffectofIBBEAinFY1997toFY1999forderegulatorsfromFY1995toFY1997 usesthetwostatesthatoptedoutofIBBEA,MontanaandTexas,asacontrolgroup. Identification of the parameters in all regressions requires the control group to be unaffected by region-specific shocks. Because Montana and Texas are in the control group for the entire time period, and the modelidentifiestheeffectofIBBEAfromFY1997toFY1999usingthesetwostatesasacontrol group,Ichecktoseewhetherastate-specificshocktoeitherMontanaorTexasdrivestheresults. To do so, I individually exclude Montana and Texas from the estimation. If, for example, MontanaexperiencedashocktoSBIRapplicationsbutTexasdidnot,thentheestimatesusingjust Montana as a control group should be different than when using just Texas as a control group and both estimates should be different than when using both states as a control group, as in Table 4. 17

Similarly, I can rule out a state-specific shock driving the results when the estimates using either MontanaorTexasorbothMontanaandTexasareallsimilar. When sequentially excluding the control states, the results are similar to using both control states. IBBEA continues to increase SBIR applications and decrease the supply of finance for small R&D firms. Using just Texas as a control state in column (4) generates larger estimates of this effect than either using both Montana and Texas or just Montana, but all of the qualitative resultsfromthebaselinemodelcontinuetohold. 6 Conclusion Thederegulationofinterstatebankbranchingandrelaxedrestrictionsoninterstatebankmergersby IBBEA increased market concentration in the U.S. banking industry. This paper uses a balanced panel to investigate how the increase in market concentration by IBBEA affected the supply of finance for small R&D firms. The applicants to SBIR are small R&D firms, both private and public. Economictheorygivesanambiguouspredictionoftheeffectofbankingconsolidationonsmall firm financing. Large banks benefit from geographic diversification. Because large banks are involved in multiple, geographically distinct product markets, they are able to distribute risk over different regions and shield themselves against adverse regional capital or business shocks (Peek and Rosengren, 1996). The diversification potential of large banks gives them an advantage over smallbankswhenofferingfinancingtermsforsmallR&Dfirms. However, when trying to obtain a source of finance, the firm will generally have superior informationaboutthevalueofthefirmrelativetoaprospectivefinancier. Thisinformationdisparity is particularly true of small R&D firms, which have little collateral or other hard information to signaltheirworthtofinanciers. Smallbanks,byforminglong-termrelationshipswithandcollecting soft information on clients (for example, a firm owner’s work ethic), can reduce informational asymmetriesandthereforeoffersuperiorfinancingtermstolargebanks,whichrelyontransaction 18

lending (for example, credit histories) to make investment decisions (Petersen and Rajan, 1994; Stein,2002). I find that IBBEA decreased the supply of finance for small R&D firms. This result implies that the relationship lending channel of small banks, in which small banks develop long-term relationships with potential clients to overcome information asymmetries associated with finance, might be more important than the geographic diversity advantage of large banks for small R&D firmfinance(PetersenandRajan,1994;PeekandRosengren,1996). Government support for R&D is justified by the presence of market failures for R&D. These market failures stem from at least two characteristics of R&D: (1) the social return to R&D is higherthantheprivatereturntoR&D,asinnovatorsareunabletocaptureprofitsfromthepositive spilloversassociatedwiththeirinventions(Griliches,1992;Samuelson,1954),and(2)asymmetric information between firms and potential financiers complicates the financing of R&D and gives rise to market failures due to moral hazard and adverse selection problems (Arrow, 1962). This papersuggestsbankingconsolidationworsenedthesemarketfailures. References Acs, Zoltan J., and David B. Audretsch, “Innovation, Market Structure, and Firm Size,” Review of EconomicsandStatistics69:4(1987),567-574. Arrow, Kenneth J., “Economic Welfare and the Allocation of Resources for Invention” (pp. 609- 626), in Richard R. Nelson (Ed.), The Rate and Direction of Inventive Activity: Economic and SocialFactors(Princeton: PrincetonUniversityPress,1962). Bank of North Dakota, Bank of North Dakota. http://www.banknd.nd.gov/. Accessed March 10, 2011. 19

Barlevy, Gadi, “On the Cyclicality of Research and Development,” American Economic Review 97:4(2007),1131-1164. Berger, Allen N., Anthony Saunders, Joseph M. Scalise, and Gregory F. Udell, “The Effects of Bank Mergers and Acquisitions on Small Business Lending,” Journal of Financial Economics 50:2(1998),187-229. Berger, Allen N., and Gregory F. Udell, “Universal Banking and the Future of Small Business Lending,”NewYorkUniversityWorkingPaperFIN-95-009(1995). Bureau of Economic Analysis (BEA), Regional Economic Accounts. http://www.bea.gov/regional/gsp/action.cfm.AccessedFebruary7,2011. Bureau of Labor Statistics (BLS), Quarterly Census of Employment and Wages. http://www.bls.gov/cew/.AccessedFebruary12,2011. Board of Governors of the Federal Reserve System, 1993 National Survey of Small Business Finances.http://www.federalreserve.gov/pubs/oss/oss3/nssbf93/nssbf93home.html#nssbf93dat. (1993).AccessedApril30,2013. Cameron, A. Colin, and Pravin K. Trivedi, Regression Analysis of Count Data (New York: CambridgeUniversityPress,1998). Cameron, A. Colin, and Pravin K. Trivedi, Microeconometrics: Methods and Applications (New York: CambridgeUniversityPress,2005). Chang,AndrewC.,“Inter-IndustryStrategicR&DandSupplier-DemanderRelationships,”Review ofEconomicsandStatistics95:2(2013),491-499. Chang,AndrewC.,“TaxPolicyEndogeneity: EvidencefromR&DTaxCredits,”BoardofGovernorsoftheFederalReserveSystemFinanceandEconomicsDiscussionSeries2014-101(2014). Cole, Rebel A., and Nicholas Walraven, “Banking Consolidation and the Availability of Credit to SmallBusinesses,”BoardofGovernorsoftheFederalReserveSystemWorkingPaper(1998). 20

Cornaggia, Jess, Yifei Mao, Xuan Tian, and Brian Wolfe, “Does Banking Competition Affect Innovation?” JournalofFinancialEconomics115:1(2015),189-209. Craig, Steven G., and Pauline Hardee, “The Impact of Bank Consolidation on Small Business CreditAvailability,”JournalofBankingandFinance31(2007),1237-1263. DenHaan,WouterJ.,StevenW.Sumner,andGuyM.Yamashiro,“ConstructionofAggregateand RegionalBankDataUsingtheCallReports,”DataManual(2002). Department of Defense (DoD), SBIR Awards by State, Arlington, VA: Department of Defense (2010),obtainedbytheauthorthroughaFreedomofInformationActrequest. Department of Defense (DoD), DoD’s SBIR and STTR Programs. http://www.acq.osd.mil/osbp/sbir/overview/index.htm.AccessedMarch13,2011. Department of Energy (DoE), Phase I by DOE Program, Washington DC: Department of Energy (2010),obtainedbytheauthorthroughaFreedomofInformationActrequest. DunandBradstreet,Manta.http://www.manta.com/.AccessedDecember15,2010. Elsas, Ralf, “Empirical Determinants of Relationship Lending,” Journal of Financial Intermediation14:1(2005),32-57. FederalGovernmentBidIntelligenceCompany,Fedvendor.http://www.fedvendor.com/.Accessed January1,2011. Federal Reserve Bank of Chicago, Commercial Bank Data. http://www.chicagofed.org/web pages/banking/financial_institution_reports/commercial_bank_data.cfm. Accessed March 6, 2011. GaleGroup,Goliath.http://goliath.ecnext.com/.AccessedDecember15,2010. Griliches, Zvi, “The Search for R&D Spillovers,” Scandinavian Journal of Economics 94 (1992), S29-S47. 21

Guceri, Irem, and Li Liu, “Effectiveness of Fiscal Incentives for R&D: Quasi-Experimental Evidence,”OxfordCentreforBusinessTaxationWorkingPaper15/12(2015). Hall, Bronwyn H., Adam B. Jaffe, and Manuel Trajtenberg, “The NBER Patent Citations Data File: Lessons,InsightsandMethodologicalTools,”NBERWorkingPaper8498(2001). Hall, Bronwyn H., and Josh Lerner, “The Financing of R&D and Innovation,” NBER Working Paper15325(2009). Harris, Richard D. F., and Elias Tzavalis, “Inference for Unit Roots in Dynamic Panels Where the TimeDimensionisFixed,”JournalofEconometrics91:2(1999),201-226. Jayaratne, Jith, and John Wolken, “How Important Are Small Banks to Small Business Lending? New Evidence from a Survey of Small Firms,” Journal of Banking and Finance 23 (1999), 427-458. Johnson,ChristianA.,andTaraRice,“AssessingaDecadeofInterstateBankBranching,”Federal ReserveBankofChicagoWorkingPaper2007-03(2007),publishedinWashingtonandLeeLaw Review65:1(2008),73-127. Kane, Edward J., “De Jure Interstate Banking: Why Only Now?” Journal of Money, Credit and Banking28:2(1996),141-161. Kashyap,AnilK.,andJeremyC.Stein,“TheImpactofMonetaryPolicyonBankBalanceSheets,” Carnegie-RochesterConferenceSeriesonPublicPolicy42(1995),151-195. Kashyap,AnilK.,andJeremyC.Stein,“WhatDoaMillionObservationsonBanksSayAboutthe TransmissionofMonetaryPolicy?” AmericanEconomicReview90:3(2000),407-428. Kortum, Samuel, and Josh Lerner, “Stronger Protection or Technological Revolution: What Is BehindtheRecentSurgeinPatenting?” Carnegie-RochesterConferenceSeriesonPublicPolicy 48(1998),247-304. 22

Kroszner, Randall S., and Philip E. Strahan, “What Drives Deregulation? Economics and Politics of the Relaxation of Bank Branching Restrictions,” Quarterly Journal of Economics 114:4 (1999),1437-1467. Lerner, Josh, “The Government as a Venture Capitalist: The Long-Run Impact of the SBIR Program,”JournalofBusiness72:3(1999),285-318. Mudd, Shannon, “Bank Structure, Relationship Lending and Small Firm Access to Finance: A Cross-CountryInvestigation,”JournalofFinancialServicesResearch44:2(2013),149-174. National Aviation and Space Administration (NASA), State-Based SBIR/STTR Statistics. http://sbir.gsfc.nasa.gov/sbirweb/search/stateStats.jsp.AccessedDecember5,2010. National Bureau of Economic Research (NBER), Patent Data Project. https://sites.google.com/site/patentdataproject/Home.AccessedFebruary14,2011. NationalInstitutesofHealth,“A“20-20”ViewofInventionReportingtotheNationalInstitutesof Health,”NIHGuide24:33(1995),NOT-95-003. National Institutes of Health, NIH Small Business Innovation Research (SBIR) and Small Business Technology Transfer (STTR) Grants, Bethesda, MD: National Institutes of Health (2010), obtainedbytheauthorthroughaFreedomofInformationActrequest. NationalScienceBoard,ScienceandEngineeringIndicators2008Volume2AppendixTables(Arlington,VA:NationalScienceFoundation,2008). National Science Foundation (NSF), Small Business Phase I, Arlington, VA: National Science Foundation(2010a),obtainedbytheauthorthroughaFreedomofInformationActrequest. National Science Foundation (NSF), WebCASPAR Integrated Science and Engineering Resources DataSystem.https://webcaspar.nsf.gov/.AccessedDecember14,2010b. 23

Peek, Joe, and Eric S. Rosengren, “Small Business Credit Availability: How Important Is the Size oftheLender?” (pp.628-655),inAnthonySaundersandIngoWalter(Eds.),UniversalBanking: FinancialSystemDesignReconsidered (BurrRidge: Irwin,1996). Peek,Joe,andEricS.Rosengren,“BankConsolidationandSmallBusinessLending: It’sNotJust BankSizethatMatters,”JournalofBankingandFinance22(1998),799-819. Petersen,MitchellA.,andRaghuramG.Rajan,“TheBenefitsofLendingRelationships: Evidence FromSmallBusinessData,”JournalofFinance49:1(1994),3-37. Presbitero, Andrea F., and Alberto Zazzaro, “Competition and Relationship Lending: Friends or Foes?” JournalofFinancialIntermediation20:3(2011),387-413. Rice,Tara,andPhilipE.Strahan,“DoesCreditCompetitionAffectSmall-FirmFinance?” Journal ofFinance65:3(2010),861-889. Samuelson,PaulA.,“ThePureTheoryofPublicExpenditure,”ReviewofEconomicsandStatistics 36:4(1954),387-389. Small Business Administration (SBA), SBA TECH-Net. http://web.sba.gov/technet/docrootpages/index.cfm.AccessedDecember5,2010. Solow, Robert M., “Technical Change and the Aggregate Production Function,” Review of EconomicsandStatistics39:3(1957),312-320. Stein, Jeremy C., “Information Production and Capital Allocation: Decentralized Versus HierarchicalFirms,”JournalofFinance57:5(2002),1891-1921. Strahan, Philip E., and James P. Weston, “Small Business Lending and the Changing Structure of theBankingIndustry,”JournalofBankingandFinance22(1998),821-845. United States Congress, House Committee on Banking, Finance, and Urban Affairs, Subcommittee on Financial Institutions Supervision, Regulation, and Deposit Insurance. 1993 Costs and 24

Benefits of Interstate Banking and Branching: Hearings before the Subcommittee on Financial Institutions Supervision, Regulation, and Deposit Insurance of the Committee on Banking, Finance, and Urban Affairs, House of Representatives, One Hundred Third Congress, first session, June 22, 1993 U.S. G.P.O.: For sale by the U.S. G.P.O., Supt. of Docs., Congressional SalesOffice,Washington. Vera, David, and Kazuki Onji, “Changes in the Banking System and Small Business Lending,” SmallBusinessEconomics34(2010),293-308. Wallsten, Scott J., “The Effects of Government-Industry R&D Programs on Private R&D: The Case of the Small Business Innovation Research Program,” RAND Journal of Economics 31:1 (2000),82-100. Wilson, Daniel J., “Beggar Thy Neighbor? The In-State, Out-of-State, and Aggregate Effects of R&DTaxCredits,”ReviewofEconomicsandStatistics91:2(2009),431-436. 25

Figure1: HerfindahlIndexForU.S.BankHoldingCompaniesByTotalAssets: FY1987–2000 0.04 0.035 0.03 0.025 Herfindahl Index 0.02 0.015 0.01 0.005 0 1987 1989 1991 1993 1995 1997 1999 Fiscal Year Pre-IBBEA IBBEA Phase-In Post-IBBEA This figure measures market concentration using data from the first quarter of each FY. Source: Call Reports, (Kashyap and Stein, 1995, 2000; Den Haan, Sumner, and Yamashiro, 2002; Federal ReserveBankofChicago,2011). 26

Figure2: InterstateBankMergers: 1990–1999 250 200 150 Interstate Mergers 100 50 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Year Pre-IBBEA IBBEA Phase-In Post-IBBEA Source: VeraandOnji(2010). 27

Table1: InitialIBBEAOpt-InDates State Date State(Continued) Date(Continued) Alabama 5/31/1997 Montana 10/1/2001 Alaska 1/1/1994 Nebraska 5/31/1997 Arizona 9/1/1996 Nevada 9/29/1995 Arkansas 6/1/1997 NewHampshire 6/1/1997 California 9/28/1995 NewJersey 4/17/1996 Colorado 6/1/1997 NewMexico 6/1/1996 Connecticut 6/27/1995 NewYork 6/1/1997 Delaware 9/29/1995 NorthCarolina 7/1/1995 DistrictofColumbia 6/13/1996 NorthDakota 5/31/1997 Florida 6/1/1997 Ohio 5/21/1997 Georgia 6/1/1997 Oklahoma 5/31/1997 Hawaii 6/1/1997 Oregon 2/27/1995 Idaho 9/29/1995 Pennsylvania 7/6/1995 Illinois 6/1/1997 RhodeIsland 6/20/1995 Indiana 6/1/1997 SouthCarolina 7/1/1996 Iowa 4/4/1996 SouthDakota 3/9/1996 Kansas 9/29/1995 Tennessee 6/1/1997 Kentucky 6/1/1997 Texas 9/1/1999 Louisiana 6/1/1997 Utah 6/1/1995 Maine 1/1/1997 Vermont 5/30/1996 Maryland 9/29/1995 Virginia 9/29/1995 Massachusetts 8/2/1996 Washington 6/6/1996 Michigan 11/29/1995 WestVirginia 5/31/1997 Minnesota 6/1/1997 Wisconsin 5/1/1996 Mississippi 6/1/1997 Wyoming 5/31/1997 Missouri 9/29/1995 Source: JohnsonandRice(2008). 28

Table2: MeansandStandardDeviationsforCharacteristicsofSmallR&DFirms SmallR&DFirms SmallR&DFirmsthatApplied forVentureCapital (1) (2) Totalemploymentin1992 12.10 24.23 (33.29) (40.61) Wageexpensesin1992 233,839 805,622 (1,381,430) (1,814,489) Salesin1992 1,635,006 2,675,071 (6,673,610) (6,419,629) Percentageappliedforloan 38.9 67.1 from1993to1994 Sizeofmostrecent 504,388 4,040,527 approvedloanin1993–1994 (3,060,744) (6,447,936) Means with standard deviations in parentheses. Total employment indicates FTE employment. Loansizecalculatedforfirmsthatweregrantedloansandincludesallloansappliedforfrom1993 to 1994. Dollar amounts inflated to 2005 dollars using the BEA’s implicit price deflator (BEA, 2011). Sample size is 1191 for R&D firms and 31 for R&D firms that also applied for venture capital. Source: 1993 National Survey of Small Business Finances (Board of Governors of the FederalReserveSystem,1993). 29

Table3: DescriptiveStatisticsofControlsandSBIRCount AcademicR&Dexpenditures 492.7 (Tensofmillions) (599.7) UsercostofR&D 1.21 (0.06) Patentcount 0.74 (Thousands) (1.60) AgencyR&Dobligations 303.2 (TensofMillions) (976.4) R&Demployees 7.61 (Thousands) (12.0) Grossstateproduct 168.9 (Billions) (2.01) R&Destablishmentcount 216.5 (346.4) SBIRapplicationcount 354.6 (650.1) Meanswithstandarddeviationsinparentheses. R&DemployeesandR&Destablishmentcountare NAICS industry 541710 from the QCEW. The user cost is the implicit rental rate of R&D capital from Wilson (2009). Dollar figures in 2005 dollars deflated with the BEA’s implicit price deflator. Sources: Wilson(2009),?,BEA(2011),andNBER(2011). 30

Table4: BaselineRegressions Units Deregulation YearsSince RawCount Percent RawCount Percent Cohort Deregulation (1) (2) (3) (4) FY1995 1year 35.8 6.71 36.6 6.68 (32.2) (6.04) (33.6) (6.30) 2years 6.92 1.29 8.56 1.60 (45.5) (8.51) (47.9) (8.98) 3years 98.2 18.3 102.1 19.1 (76.1) (14.2) (76.4) (14.3) 4years 135.2 25.2 142.3 26.6 (65.2)** (12.2)** (63.5)** (11.9)** FY1996 1year -1.25 -0.32 0.60 0.15 (22.3) (5.74) (21.5) (5.53) 2years 70.3 18.1 78.2 20.1 (50.0) (12.8) (48.1)* (12.3)* 3years 106.2 27.3 118.6 30.5 (42.1)** (10.8)** (40.4)*** (10.4)*** FY1997 1year 52.6 25.3 55.4 26.6 (23.1)** (11.1)** (22.1)*** (10.5)*** 2years 75.1 36.0 78.6 37.7 (22.1)*** (10.6)*** (20.8)*** (10.0)*** AcademicR&Dexpenditures 0.03 0.01 (0.10) (0.02) UsercostofR&D 66.7 18.7 (166.1) (46.7) Patentcount -20.9 -5.88 (10.7)** (3.03)** AgencyR&Dobligations -0.004 -0.001 (0.003)* (0.0008)* R&Demployees -0.12 -0.03 (1.80) (0.50) Grossstateproduct 0.63 0.17 (0.43) (0.12) R&Destablishmentcount -0.10 -0.02 (0.13) (0.04) F-testforallcontrols 0.02** (p-value) StateFE Yes Yes Yes Yes Timeeffects Yes Yes Yes Yes dy Averagemarginaleffectsreported. Columns(1)and(3)reporttheeffectinlevels( ),andcolumns dx dy/y (2)and(4)reporttheeffectsofcolumns(1)and(3)assemielasticitiesconvertedtopercents( × dx 100%). Gross state product is in real 2005 billions, all other dollar figures are in real tens of 2005 millions, patent count and employee count are in thousands. Number of observations is 510. Standarderrorsclusteredbystateinparentheses. *,**,***: significantatthe10%,5%,1%level, respectively. 31

Table5: AlternateTimingofPolicyVariable Units Deregulation YearsSince RawCount Percent RawCount Percent Cohort Deregulation (1) (2) (3) (4) FY1995 0years 15.7 2.95 (42.3) (7.93) 1year 57.9 10.8 (50.5) (9.47) 2years 105.1 19.6 -15.3 -2.87 (76.4) (14.3) (31.0) (5.81) 3years 175.4 32.8 -47.5 -8.89 (116.6) (21.8) (45.5) (8.85) 4years 221.9 41.5 39.0 7.29 (107.0)** (20.1)** (45.5) (8.85) FY1996 0years 20.7 5.33 (23.1) (5.96) 1year 69.5 17.9 (44.8) (11.5) 2years 128.7 33.1 -12.1 -3.11 (75.1)* (19.3)* (20.3) (6.07) 3years 173.7 44.6 67.6 17.4 (67.7)*** (17.4)*** (23.6)*** (6.07)*** FY1997 0years 34.7 16.6 (17.7)** (8.53)** 1year 77.1 36.9 (33.0)** (15.8)** 2years 101.8 48.8 48.3 23.1 (30.9)*** (14.8)*** (11.2)*** (5.41)*** dy Averagemarginaleffectsreported. Columns(1)and(3)reporttheeffectinlevels( ). Columns(2) dx dy/y and (4) report the effects of columns (1) and (3) as semielasticities converted to percents ( × dx 100%). Number of observations is 510. All regressions include control variables from Table 4, aggregate time dummies, state time trends, and state time-invariant effects. Standard errors clusteredbystateinparentheses. *,**,***: significantatthe10%,5%,1%level,respectively. 32

Table6: SpecificStateRobustness Deregulation YearsSince Cohort Deregulation (1) (2) (3) (4) FY1995 1year 6.68 7.88 6.75 6.75 (6.30) (6.17) (6.34) (6.33) 2years 1.60 2.98 0.56 1.25 (8.98) (8.79) (9.05) (9.12) 3years 19.1 20.3 16.2 34.6 (14.3) (14.3) (15.5) (13.5)*** 4years 26.6 28.1 26.2 39.3 (11.9)** (11.9)*** (12.9)** (18.1)** FY1996 1year 0.15 0.90 -0.78 0.06 (5.53) (5.58) (5.47) (5.66) 2years 20.1 20.5 17.3 36.5 (12.3)* (12.4)* (13.8) (11.4)*** 3years 30.5 30.9 30.2 43.6 (10.4)*** (10.4)*** (11.9)*** (18.0)** FY1997 1year 26.6 27.1 24.1 42.9 (10.5)*** (10.6)*** (12.3)** (9.74)*** 2years 37.7 38.1 38.0 50.9 (10.0)*** (10.1)*** (11.7)*** (17.8)*** AcademicR&Dexpenditures 0.01 0.01 0.01 0.01 (0.02) (0.02) (0.02) (0.03) UsercostofR&D 18.7 20.7 19.0 15.5 (46.7) (45.7) (47.7) (46.7) Patentcount -5.88 -5.63 -6.09 -5.71 (3.03)** (3.01)* (3.06)** (3.49)* AgencyR&Dobligations -0.001 -0.001 -0.001 -0.001 (0.0008)* (0.0008)* (0.0008) (0.0008)* R&Demployees -0.03 -0.06 -0.002 0.06 (0.50) (0.50) (0.50) (0.56) Grossstateproduct 0.17 0.19 0.16 0.15 (0.12) (0.12) (0.12) (0.14) R&Destablishmentcount -0.02 -0.03 -0.03 -0.02 (0.04) (0.03) (0.03) (0.04) F-testforallcontrols 0.02** 0.02** 0.03** 0.02** (p-value) No. obs. 510 500 500 500 Excludedstate ND MT TX dy/y Average marginal effects reported as semielasticities converted to percents ( ×100%). All dx regressions include aggregate time dummies, state time trends, and state time-invariant effects. Grossstateproductisinreal2005billions,allotherdollarfiguresareinrealtensof2005millions, patent count and employee count are in thousands. For excluded states, “ND” = North Dakota, “MT”=Montana,and“TX”=Texas. Standarderrorsclusteredbystateinparentheses. *,**,***: significantatthe10%,5%,1%level,respectively. 33