Verifying the State of Financing Constraints: Evidence from US Business Credit Contracts

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Verifying the State of Financing Constraints: Evidence from US Business Credit Contracts Ralf R. Meisenzahl 2011-04 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.

Verifying the State of Financing Constraints: Evidence from US Business Credit Contracts∗ Ralf R. Meisenzahl Federal Reserve Board January 3, 2011 Abstract Which of the strategies for financing constraints in economic models is the most empirically plausible? This paper tests two commonly used models of financing constraints, costly state verification (Townsend, 1979) and moral hazard (Holmstr¨om and Tirole, 1997), using a comprehensive data set of US small business credit contracts. The data include detailed information about the business, its owner, bank balance sheet information, and the terms of credit. In line with the predictions of models of financing constraints, I find that an additional dollar of net worth accounts for about 30 cents of external finance. More than twothirdsofthebusinesscreditcontractscanberationalizedbyoneperioddebtcontractswith costly state verification. The parameter values obtained in the costly state verification model imply bankruptcy costs of 28% of expected output and a rate of return ranging between 5% and 8% annually, which are consistent with studies on bankruptcy incidences and returns to entrepreneurialinvestment. Themoralhazardmodel,however,performspoorly. Thecorrelation between model implied interest rates and actual interest rates paid is close to zero because the bank variables emphasized by this model do not explain loan interest rates. ∗E-mail: ralf.r.meisenzahl@frb.gov. Mailing address: Mailstop 153, 20th and C Streets NW, Washington, DC 20551. An earlier version of this paper was circulated as “Financing Constraints and Investment: Evidence from USBusinessCreditContracts.”IwouldliketothankLarryChristiano,RochelleEdge,MartinEichenbaum,Andrea Eisfeldt, Diana Hancock, Joel Mokyr, Dale Mortensen, Skander van den Heuvel and seminar participants at CREI, ChicagoFED,Rochester,IMF,FederalReserveBoard,andNorthwesternUniversityforhelpfulcomments. Allerrors are mine. The views stated are those of the author and do not necessarily reflect those of the Board of Governors of the Federal Reserve System.

1 Introduction Economists have long recognized the importance of financial frictions to explain macroeconomic data or firm growth patterns. Commonly used macroeconomic models and firm growth models include a financing constraint induced by either costly state verification (CSV), pioneered by Townsend (1979), or moral hazard, pioneered by Holmstr¨om and Tirole (1997) (MH). While the microfoundations of other frictions, such as price or wage ridigities, have been subject to intensive empirical scrutiny, it is less clear whether these types of financing constraints are plausible vis-`a-vis actual credit contracts. In contrast with the earlier literature, I assess the empirical plausibility of particular models of financing constraints directly using micro-based evidence from debt contracts. The first model, CSV, lies at the heart of financial frictions in macroeconomic and firm growth models.1 Entrepreneurs invest in part with external finance in risky return projects. Lenders that provide external finance incur a cost to observe the realized returns. These state verification costs of bankruptcy lead to a breakdown of the Modigliani-Miller theorem. Using credit amounts, net worth, interest rates, and probabilities of bankruptcy in the Surveys of Small Business Finances (SSBF), I solve for the bankruptcy costs, rates of return, and riskiness implied by the optimal debt contract with the CSV model. More than two thirds of the credit contracts can be rationalized with CSV debt contracts. In line with the literature, the implied bankruptcy costs are 28 percent of expected output, and the implied rate of return lies between 5 percent and 8 percent annually. Thus, CSV is an empirically plausible model of financing constraints. The second model, MH, explicitly models bank capital as a scarce resource. The underlying assumption is that only banks possess a monitoring technology. Banks monitor entrepreneurs to ensure that they invest in high quality projects. Investors (depositors) are willing to invest in these projects (in a bank) as long as the bank’s incentives to monitor are high-powered. Chen (2001) and Meh and Moran (2010), for instance, include this type of financing frictions to assess the role of bank capital over the business cycle. To assess the empirical plausibility of MH, I add information about the bank’s adequacy ratio, rate of return, interest expenses, and non-interest expenses to the SSBF data. Using this information I calculate the interest rate implied by MH. The correlation between model implied 1CarlstromandFuerst(1997)andBernanke,Gertler,andGilchrist(1999)introduceCSVintodynamicstochastic general equilibrium models which is now the standard method of including financial frictions (see e.g. Christiano, Motto, and Rostagno (2007), Christensen and Dib (2008), von Heideken (2009), Gilchrist, Ortiz, and Zakrajsek (2009), and Nolan and Thoenissen (2009)). Cooley and Quadrini (2001) show that CSV financing constraints can explainfirmgrowthpattern. CovasanddenHaan(20010)useCSVtomatchthecyclicalbehaviorofdebtandequity issuance. 1

interestratesandactualinterestratesisclosetozero. Second,insteadofusingnon-interestexpense as proxy for monitoring cost, I solve for the implied monitoring cost. This yields mostly empirically implausible values. Last, I complement the structural analysis with reduced form regressions and find that in almost all instances bank variables cannot explain the interest rate. For large banks in the 2003 sample, some bank variables have a significant effect on the interest rate paid. But, sincethesmallbusinesslendingmarketisdominatedbysmallbanks, MHappearstobeempirically implausible. To be clear, while my findings do not support the MH model, bank variables may still be important for the loan supply (Bernanke and Lown, 1991; Hancock and Wilcox, 1993, 1994; Berrospide and Edge, 2010). In both models, a key deviation from perfect capital markets is that agency problems between entrepreneursandlendersrestrictexternalfinancetosomemultipleoftheentrepreneur’snetworth. Many authors have used this reduced-form collateral constraints without explicitly modeling the underlying frictions.2 Consistent with models of financing frictions an additional dollar of net worth accounts for about 30 cents of additional external finance. This finding is robust to alternative definitions of net worth and the inclusion of a large set of controls. While there are some differences between small and large banks in the 1998 sample, the point estimates are the same for both groups in the 2003 sample. Different from CSV, the interest rate paid in MH, in equilibrium, also depends on the cost of bank capital, the financing cost of a bank, and the adequacy ratio of the bank - that is, the equity share of assets. Again, contrary with the predictions of the MH model, bank variables have no significant effect. This comprehensive, micro evidence-based analysis of financing constraints has the advantage of focusing on small businesses, that are bank-dependent and have no access to bond markets. Small businesses are defined as firms with up to 500 employees, reported in the SSBF from 1998 and 2003. Only a few small firms attract additional outside equity. Therefore, the SSBF samples represent the population closest to standard models of lending; an entrepreneur without means of financing except to sign a debt contract with a lender (bank).3 While small, these businesses account for about one half of GDP, and they disproportationately contribute to labor market volatility (Haltiwanger, Jarmin, and Miranda, 2010). So, a second advantage of focusing on small business is to gain insights about the importance of financing frictions for the analysis of the labor market (Wasmer and Weill, 2004). The SSBF provide detailed information about firms and their owners, including firms’ balance sheets and private net assets. The credit contract information includes the bank’s name, the loan 2For instance, Evans and Jovanovic (1989), Cagetti and de Nardi (2006), and Buera (2009). 3For dynamic models, see Albuquerque and Hopenhayn (2004) and Clementi and Hopenhayn (2006). 2

amount, theinterestrate, andcollateralused. Thelattershowstheimportanceofprivatenetworth inthesmallbusinesscreditmarketassmallbusinessownersoftencollateralizebusinesscreditswith their private residence. To evaluate the empirical support of the CSV and MH models, I need to construct net worth of the merged firm-entrepreneur entity. To be consistent with the banks’ definition of net worth, the construction of net worth of the firm-entrepreneur entity follows the collateral requirements. Surprisingly, banks accept almost every type of asset on the firms’ balance sheets as collateral. In addition, small business owners secure credits with private residential property or personal guarantees. Since the SSBF include the net value of the private residence, I am able to construct net worth in a way that matches the banks’ definition. Using the confidential SSBF data that in many cases include unique identifier for financial institutions, I append the SSBF data with crucial bank characteristics from the Reports of Condition and Income (Call Reports), the Thrifts Financial Reports, and Call Reports for credit unions. For previously unmatched observations, I use the lender’s name to identify commercial banks, savings banks, thrifts, and credit unions. I exclude financing companies. In total I match 53 percent of the 1998 sample and 79 percent of the 2003 sample to banks and credit unions. Previous studies, starting with Fazzari, Hubbard, and Petersen (1988), inferred financing constraints using cash flow sensitivities of investment.4 However a large literature criticizes this approach. These cash flow sensitivities depend on the author’s discretion (Kaplan and Zingales, 2000), can occur in frictionless models (Gomes, 2001; Alti, 2003), or are the result of measurement error in Tobin’s q (Cummins, Hassett, and Oliner, 2006). Others employ structural estimation of entrepreneurship or household models to infer parameters of financing constraints (Evans and Jovanovic,1989;HennessyandWhited,2007)ortodistinguishbetweendifferentmodelsoffinancing constraints (Paulson, Townsend, and Karaivanov, 2006) without using debt contracts. Offering an alternative approach, Chava and Roberts (2008) analyze debt contracts of large firmsandfindthattheviolationofdebtcovenantsreducesinvestment. However,withtheexception of Levin, Natalucci, and Zakrajsek (2004) who use bond data, there is little direct, micro-based evidence from debt contracts supporting the standard modeling strategies of financing constraints in macroeconomic models or firm growth models.5 Employing a comprehensive data set of US small business credit contracts, I circumvent the problem of identifying financially constrained firms. This paper therefore provides the, thus far lacking, micro-based evidence by exploiting credit contracts of US small businesses to test specific models of financing constraints. While the debt constract with costly state verification is an 4See Hubbard (1998) and Stein (2003) for surveys. 5Most micro-based studies such as Levenson and Willard (2000) and Angelini and Generale (2008) examine the importance of credit rationing (Stiglitz and Weiss, 1981). 3

empirically plausible model, the moral hazard in model, put forward by Holmstr¨om and Tirole (1997), appears to be inconsistent with the matched small business-bank data as bank variables, emphasized by this model, do not explain loan interest rates or credit amounts. Thepaperproceedsasfollows: Section2describesandteststheempiricalplausiblityofthedebt contract with costly state verificiation. Section 3 introduces the moral hazard model of financing constraints and assess its empirical plausibility. In section 4 I test an additional implication of the both models - reduced-form collateral constraints. Differences by bank size are discussed in section 5. Section 6 concludes. 2 Costly State Verification This section first describes the debt contract with CSV. I then summarize the data used to assess the empirical plausibility of this model. Last, I present the parameter distribution for the rate of return and bankruptcy cost implied by the debt contract with CSV for each observation in the data. 2.1 The debt contract with CSV The CSV recognizes asymmetric information between the lender and the borrower. The lender can not observe the outcome of a leveraged project. In case of bankruptcy the lender incurs a cost to verify the outcome. This costly state verification ties the ability to obtain credit to the net worth of an entrepreneur (Townsend, 1979). Gale and Hellwig (1985) show that in a world with costly state verification the optimal, incentive-compatible financial contract is the standard debt contract. The debt contract with CSV is now widely used in macroeconomic models and models of firm growth.6 To derive the optimal one period debt contract, I first describe the preferences and technology of both borrowers and lenders. Risk neutral entrepreneurs own a risky return technology; entrepreneurs can finance their capital stock with equity and debt. If an entrepreneur borrows, she signs the standard one period debt contract which in the presence of costly state verification is the optimal contract. To better understand the implications of the underlying financing constraint, I abstract from aggregate uncertainty, differences in productivity and equity issuance.7 Let k be the capital stock used in production by entrepreneur i. The vast majority of credit i applications in the data, which I describe in the next section show that credit is used for working capital. Hence, I consider the case of full depreciation. The entrepreneur can only use a fraction ξ 6Seee.g. CarlstromandFuerst(1997),Bernanke,Gertler,andGilchrist(1999),andCooleyandQuadrini(2001). 7For models with debt/equity choice using the costly state verification, see Cooley and Quadrini (2001), Covas and den Haan (20010), and Hennessy and Whited (2007). 4

of her net worth n in production but can borrow d = k −ξ n to produce at scale k . She signs i i i i i a one period debt contract with a financial intermediary, the lender, specifying the credit amount and the gross interest rate Z . The fraction (1−ξ ) of net worth is pure collateral such as land i i or buildings. In line with financing constraints employed in macroeconomic models I assume a constant returns to scale technology. The payoff of production with external financing is: Π = ω Rkk +(1−ξ )n −Z d , (1) i i i i i i i i where in the first term ω is an idiosyncratic productivity shock and Rk is the rate of return. i i The productivity shock ω is unknown at the time when the debt contract is signed and i.d.d. across time. The random variable has a continuous differentiable c.d.f., F(ω) and E(ω) = 1. The ∂h(ω) distribution satisfies the following regularity condition: > 0, where h(ω) is the hazard rate. ∂ω This assumption guarantees that an interior solution is a non-rationing outcome. The property holds for, among others, the lognormal distribution. The second term (1 − ξ )n , is the pure i i collateral component of net worth. Financial intermediaries, offering loans to entrepreneurs, are risk neutral, perfectly competitive, and hold perfectly diversified portfolios. The opportunity cost of funds for financial intermediaries is the economy’s riskless gross rate of return 1+r. Financial intermediaries do not observe the realization of the idiosyncratic productivity shock. Lenders have to pay a cost to be able to observe the borrower’s productivity shock. As in Carlstrom and Fuerst (1997) monitoring costs in the case of bankruptcy are assumed to be a fraction µ of the expected output. i The entrepreneur defaults on the debt if after the realization of the idiosyncratic shock she does not have enough resources on hand to pay back the loan. Let ω¯ be the cutoff productivity for i which the entrepreneur can just repay her debts: ω¯ Rkk +(1−ξ )n = Z d . (2) i i i i i i i In case of default the financial intermediary pays the monitoring cost µ Rkk and seizes the firm’s i i i remaining assets. The default payoff to the lender is: ω Rkk +(1−ξ )n −µ Rkk . (3) i i i i i i i i Financial intermediaries are assumed to be perfectly competitive. Their zero profit condition has to hold in expectation on each contract. Using equation (2), the zero profit condition of the lender 5

can be written as: (cid:90) ∞ (cid:16) (cid:17) (cid:90) ω¯i (cid:16) (cid:17) ω¯ Rkk dF(ω)+ ω Rkk −µ Rkk dF(ω)+(1−ξ )n = (1+r)(k −ξ n ) (4) i i i i i i i i i i i i i i ωi 0 and the entrepreneur’s expected income is: ∞ (cid:90) ω Rkk dF(ω)−(1−F(ω¯ ))ω¯ k . (5) i i i i i i ω¯i The loan contracts, in general, specify a pair of (k −ξ n ,Z ). Here it is convenient to use the i i i i definition of the cutoff productivity ω¯ and to rewrite the problem as a choice over (k ,ω¯ ). The i i i solution to the optimal contract maximizes over these two variables the expected payoff to the borrower subject to the zero profit condition on the lender. maxΓ(ω¯ )Rkk (6) i i i ki,ω¯i subject to G(ω¯ )Rkk +(1−ξ )n = (1+r)(k −ξ n ), (7) i i i i i i i i ∞ (cid:82) where Γ(ω¯ ) = ω dF(ω)−(1−F(ω¯ ))ω¯ is the firm’s share of output and G(ω¯ ) = (1−F(ω¯ ))ω¯ + i i i i i i i ω¯i (cid:82) ω¯i ωdF(ω)−µ F(ω¯ ) is the the lender’s share of output. The combined shares of the lender and the i i 0 borroweraretogetherlessthan1: Γ(ω¯ )+G(ω¯ ) = 1−µ F(ω¯ ). Thisillustratesthedistortioncaused i i i i byasymmetricinformationabouttheidiosyncraticshock. Anincreaseinthecutoffproductivityω¯ , i equivalent with more bankruptcies, or an increase in the monitoring cost lead to larger distortions. Let λ be the multiplier on the lender’s zero profit constraint. The problem can be scaled by n and i solved for the optimal leverage, the capital to net worth ratio, L = k /n i i i for all entrepreneurs.8 The external finance premium is defined as the rate of return over the risk free rate: Rk/(1+r). Combing the first order condition for ω¯ and k gives: i i i Rk 1 i = (8) (1+r) G(ω¯ )−Γ(ω¯ ) G(cid:48)(ω¯i) i i Γ(cid:48)(ω¯i) 8In other words, if entrepreneurs’ projects are identical, net worth matters for the total amount of credit, but in equilibrium all entrepreneurs have the same leverage and the same cutoff productivity ω¯. This feature simplifies including this type of financing constraints in general equilibrium models. 6

NotethatΓ(cid:48)(ω¯ ) = −(1−F(ω¯ ))andG(cid:48)(ω¯ ) = −Γ(cid:48)(ω¯ )−µ F(cid:48)(ω¯ ). InaninterioroptimumG(cid:48)(ω¯ ) > 0 i i i i i i i otherwise the bank could increase profits by lowering ω¯ .9 By inverting the function, Bernanke, i Rk Gertler, andGilchrist(1999)showthatω¯ isincreasingin i . Rewritingthezeroprofitcondition i (1+r) of the lender yields: 1+rξ i L = . (9) i (cid:16) Rk (cid:17) (1+r) 1−G(ω¯ ) i i (1+r) The zero profit condition of the lender implies that the amount of credit covered by collateral, (1−ξ )n , needs to earn the risk free rate. This introduces an additional wedge in the choice of i i leverage.10 The optimal contract trades off the return on an additional unit external finance against a rise in bankruptcy states, i.e. higher agency costs. The equilibrium contract, defined as a pair of Rk (L ,ω¯ ), depends on four parameters: 1) the external finance premium i , 2) the variance of the i i (1+r) idiosyncratic shock σ which measures risk, 3) the bankruptcy cost µ , and 4) the fraction of net i i worth used in production ξ . i 2.2 Credit Contracts and Net Worth Data I now describe the data used to assess the empirical plausibility of the debt contract with CSV. In particular, I summarize the construction of the crucial variables: the net worth n , the probability i of bankruptcy F(ω¯), the credit amount d , the share of collateral ξ and the interest rate Z . i i i i TheSurveyofSmallBusinessFinances(until1998NationalSurveyforSmallBusinessFinances) was conducted by the Federal Reserve Board for the fiscal years ending in 1987, 1993, 1998 and 2003. The surveys aim to assess credit availability, to provide financial statement data of small firms, and to study the effects of changes through technology and mergers in the banking sector. Within the surveys, small businesses are defined as firms with fewer than 500 employees. The 9ThevalueoftheLagrangemultiplieronthebank’szeroprofitconditiondependsonlyonthecutoffproductivity: Γ(cid:48)(ω¯ ) λ=− i ≥1. G(cid:48)(ω¯ ) i Intuitively, without the presence of bankruptcy cost, µ=0, the credit would not be risky and the Modigliani-Miller theorem holds. The marginal unit of capital can be financed with either net worth or credit without changing the payoff of the lender and λ = 1. If the entrepreneur has to take out a risky loan, then the multiplier reflects the shadow value of agency cost which are increasing in ω¯ . i 10Any additional cost of repossessing collateral reduces leverage further. To see this, assume that in case of bankruptcy the bank only receives (1−µ )(1−ξ )n . The optimal leverage now is L = 1+rξi−µiF(ω¯i)(1−ξi) which i i i i (1+r) (cid:32) 1−G(ω¯i) (1 R + i k r) (cid:33) is lower than without additional repossession cost. Hennessy and Whited (2007) estimate the cost of repossessing to be 10 percent of the capital stock. However, their point estimate is not significant. One interpretation of this repossession cost is the homestead exemption in bankruptcy laws. Berkowitz and White (2004) find that higher exemptions reduce access to credit significantly. 7

representative samples were drawn from the Dun and Bradstreet database.11 In addition to the standard firm characteristics such as number of employees, sales, region, and industry, the SSBF also provide additional information on the most recent credit (amount, interest rate) and measures of riskiness of a credit: the Dun and Bradstreet credit score and indicators whether the firm had declared bankruptcy or has delinquent obligations or judgments. For the business owner, the same indicators are available except for the personal credit score. Personalnetworth, separatedbynethomevalueandotherassets, isthecrucialvariableusedto construct net worth of a merged firm-household entity and is only available for the 1998 and 2003 surveys only.12 I therefore restrict my analysis to the last two cross sections. The SSBF provides a final, representative sample of 3,561 firms in 1998 and 4,240 firms in 2003. The sample weights imply that the 1998 sample represents 5,291,245 firms and the 2003 sample represents 6,298,087 firms.Table 1 summarizes the the standard firm characteristics for the SSBF data employed here.13 ToevaluatethedebtcontractwithCSV,IusetheSSBFdatatoconstructnetworthofthefirmhousehold entity (n ) as follows. The literature defines net worth as maximum (collateralizable) i assets, which is the sum of liquid and collateralizable illiquid assets less all obligations.14 For small businesses, the owner may collateralize credit with private net worth. To construct a measure of net worth of the firm-household entity that is consistent with a bank’s definition of net worth, I examine the application process and the collateral accepted by banks in detail. Giambona and Schwienbacher (2008) argue that only “hard” tangible assets, namely land and buildings, can be collateralized. Yet almost 45 percent used equipment as collateral, 23 percent their business real estate, and 24 percent used their private residence. Accounts receivable and inventory was accepted as collateral in 23 percent of the cases.15 The main difference between 11Adetaileddescriptionofthesamplingprocedurecanbefoundinthemethodologyreports. Firmswereaskedto use tax data and previously sent worksheets to answer the questions. The public data and methodology reports to all surveys can be downloaded at http://www.federalreserve.gov/Pubs/Oss/Oss3/nssbftoc.htm 12For a description of earlier surveys, see Berger and Udell (1998). 13The composition of small businesses in the 1998 and 2003 surveys are comparable. The average number of employees was 8.6 and the median number of employees was 3 in both samples. Firms in the manufacturing sector (SIC 20-39) have on average more than twice as many employees as the firms in the service sector (SIC 70-89) and insurance and real estate (SIC 64-65) sectors. Average firm age increases from 13.3 years in the SSBF 1998 to 14.3 years in the SSBF 2003 after falling from 14.5 years in the SSBF 1993. The median also increased by one year from 11 years to 12 years. The vast majority of the firms has only few employees and is owner-managed (94 percent) and lessthan30percentofthefirmsreportedannualsalesofmorethan$500,000. Thirty-fivepercentofsmallbusinesses were located in the South, 24 percent in the West, 21 percent in the Midwest, and 20 percent in the Northeast. Roughly 4 out of 5 firms had theirheadquarters or main office in urban areas, and 44.5 percent (1998: 49.4 percent) of firms were proprietorships, 8.7 percent (7 percent) partnerships, 31 percent (23.9 percent) C-Corporations and 15.8percent(19.8percent)S-Corporations. Formoredetaileddescriptivestatistics,seeAppendix. Bitler,Robb,and Wolken (2001) and Mach and Wolken (2006) provide descriptive statistics, summarizing the financial services used by small businesses for the SSBF 1998 and 2003, respectively. 14See e.g. Bernanke and Gertler (1989). 15In the credit application process banks require a business financial statement and a personal financial statement. The business financial statement includes balance sheet information, information on accounts receivable, 8

the two surveys is the shift from equipment (1998: 55 percent, 2003: 39 percent) to the private residence (1998: 15 percent, 2003: 29 percent) as the source of collateral. The increase in the house prices between the two samples may explain this shift. While in the 1998 sample the average net value of the private residence is $128,915 (median: $75,000), the average net value of the private residence increases to $232,632 (median: $120,000) in the 2003 sample. Only about 6 percent use “other” personal assets beyond the private residence to collateralize their credit. Alternativelyownerscanprovideapersonalguarantee-thatis,theownerdeclaresanindividual pledge to repay the business loan. Depending on the credit contract, the owner can be held responsible for the business loan even if the business is protected by limited liability laws or gets dissolved. In the pooled 1998 and 2003 SSBF 52 percent of the firms reported that collateral was required to receive credit, 54 percent gave personal guarantees, and 30 percent of business owners provided both. This is in line with Sullivan, Warren, and Westbrook (1989) who, using bankruptcy data, show that entrepreneurs are more likely to file personal bankruptcy as firm credits are collateralized or guaranteed by the owner’s private net worth. Thepatternsofcollateralstronglysuggestawiderdefinitionofnetworththantheoneadvocated by Giambona and Schwienbacher (2008). I construct net worth from the data as follows: Using balance sheet information, I subtract all liabilities from total assets to calculate firm equity. Following the patterns of collateral, I add the net value of the private residence for firm owners who own a house. Table 2 displays the equity share of total assets - total equity divided by total assets (firm+residence), net worth, which is the inverse of leverage.16 The equity share of assets of capital a firm-household entity with negative equity is 0. Table 2 provides the summary statistics for the full sample and for firms that received credit. Not that not all firms applied for and received credit. Younger firms have less net worth and a lower equity share of assets than older firms. Similarly, firms that received credit have more net worth than the average firm in the total sample in each age group. However, the equity shares of firms that received credit are indistinguishable from the total sample. On average the equity shares translate into a leverage (the capital to net worth ratio) of 1.5 or an equity to debt ratio of 2, and lie in the range of 1.3 to 2 documented by Masulis (1988). It is worth noting that in the 2003 SSBF more than 63 percent of the entrepreneurs used credit for working capital while about 20 percent indicated investment activity (9.9 percent vehicles loans and 10.2 percent equipment loans) and another 9.4 percent took out mortages. In the 1998 SSBF the share of credit used for investment activity is roughly 35 percent. This is mainly because renewals of credit lines were not included in the 1998 questionaire. Only 6-8 percent of the samples organizational form, and other business characteristics. The personal financial statement asks for a detailed description of assets and liabilities, including the net value of the private residence, total net worth, and previous bankruptcies. For detailed application forms, see Cavalluzzo and Wolken (2005). About 25 percent of the firms provided more than one type of collateral. 16Some of the firm-household entities have no positive equity. Their equity share is zero. 9

raised new equity. The usage of new equity in both surveys exhibits the same pattern. Since most of the credit is used for working capital, I assumed complete depreciation in the previous section. To fit the debt contract with CSV described in the previous section to the data, I use the net worth of the firm-household entity for n . Information on the most recent loan application and i the riskiness of the loan are used to proxy for additional parameters of the model. In particluar, the SSBF information on the firm’s most recent loan application include the bank’s name, whether the credit application was approved and if so the details of the credit contract; in particular, the loan amount (d ) and the interest rate (Z ). The share of collateral (ξ ) is proxied by the share of i i i fixed assets (land, buildings, and machinery). The Dun and Bradstreet credit score, included in the SSBF, measures the probability of bankruptcy (F(ω¯)). i 2.3 Fitting the CSV Debt Contract To assess the empirical plausibility of CSV as modeling strategy for financing constraints, two questions need be answered: i) whether parameters for the model can be derived from the actual debt contracts - that is, whether a system of equations constituting the CSV contract can be solved for each observation - and ii) if so, whether the implied parameter values for the bankruptcy cost and the rate of return are plausible when compared to studies of bankruptcies and returns to entrepreneurial capital investments. For this purpose, I use the observed credit contract characteristics - credit amount, interest rate, net worth, and credit score - to solve for the implied parameters of a CSV debt contract. The share of fixed assets, such as land and buildings, of total firm assets in SSBF is 0.33. Hence, the fraction of the net worth used in production ξ is on average 0.67. The methodology follows Levin, Natalucci, and Zakrajsek (2004). While Levin, Natalucci, and Zakrajsek (2004) use bond data to construct empirical equivalents to one period debt contracts and to analyze CSV, this study uses bank credit contracts to back out the implied parameters the agency cost µ , the standard deviation of the underlying log-normal distribution σ , the rate of i i return Rk, and the cutoff productivity ω¯ - of the CSV credit contract for each observation i in the i i SSBF. The following four equations need to be solved by the parameters: (i) the definition of the cutoff productivity, equation (2), ω¯ RkL +(1−ξ ) = Z (L −ξ ), i i i i i i i (ii) the optimality conditions, equations (8) and (19), Rk 1 i = (1+r) G(ω¯ )−Γ(ω¯ ) G(cid:48)(ω¯i) i i Γ(cid:48)(ω¯i) 10

and 1+rξ i L = , i (cid:16) Rk (cid:17) (1+r) 1−G(ω¯ ) i i (1+r) and (iii) the probability of bankruptcy F(ω¯ ) = taken from credit score. i The distribution of the idiosyncratic shock F(ω) is log-normal with E(ω) = 1. In the data firms differ in their rate of return Rk, risk σ , and bankruptcy cost µ . For instance, an increase in the i i i rate of return increases the returns to external finance and this implies that a higher leverage is therefore optimal. This also implies a higher cutoff productivity and thereby a higher interest rate paid on the loan Z. An increase in variance of the idiosyncratic shock, on the other hand, makes low productivity states more likely to occur as E(ω) = 1. With a higher variance optimal contract exhibits a lower leverage and a higher interest rate (see Figure 1). Incomplete depreciation does not change these results but scales up leverage and reduces the probability of bankruptcy. In constructing leverage L , I use the credit amount on the most recent credit and net worth. i The interest rate on the credit in the data is used to calibrate Z in equation (2). The risk free i rate r is set to the Fed Funds Rate at the time the credit was contracted. The annual risk free rate is about 5 percent for the 1998 SSBF and 1 percent for the 2003 observations depending on the respective quarter the loan was taken out. The bankruptcy rate is taken from the Dun and Bradstreet credit scores. The credit scores provided in the SSBF indicate the probability of delinquency (payments are at least 90 days late) rather than bankruptcy.17 Therefore, I assume that firms that are more like to miss payments are also more likely to file for bankruptcy. I scale the probability of delinquency incidents to match a quarterly bankruptcy rate of 1 percent as reported in Fisher (1999). This yields that firms with the highest score (6) have a quarterly failure probability of 0.19 percent while 4.44 percent of firms with the lowest credit score (1) are expected to fail in the 2003 sample. In 1998, the probability of failure for the best (worst) risks is 0.15 percent (3.52 percent). In some cases there exists no solution to the system of equations describing the credit contract, forinstancewhenthefirm-entrepreneurentityhasnonetworth. In1998,599outof660observations can be rationalized by one period debt contracts with CSV. In 2003, for 1413 out of 1598 observationstheoptimalcontractcanbesolved. Sometimesthesolutioninvolvesimplausiblevalues suchasnegativeagencycost. Inthe1998sample,114observationsareexhibitingafittedagencycost are below 0 or larger than 1. In the 2003 sample, 364 observations cross these thresholds. However, 17For a detailed discussion, see http://www.dnb.com. 11

for 73 percent of the observations in 1998 (485 out of 660) and 67 percent of the observations in 2003 (1079 out of 1598) the system of equations can be solved with plausible parameter values. Table 3 summarizes the results. In 1998, the average agency cost µ is estimated to be 0.282 (stddev 0.09) which corresponds to 28.2 percent of output. For the 2003 sample the agency costs are 0.287 (stddev 0.10). This lies in the plausible range of 0.2-0.36 reported in Carlstrom and Fuerst (1997).18 Figure 2 plots the implied risk parameters and bankruptcy cost for the truncated sample. As reflected in Table 3 the bankruptcy costs are clustered within the 0.2-0.36 range.19 The mean standard deviation of the idiosyncratic shock, σ, measuring the riskiness of projects is 0.57 (stddev0.25)in1998. In2003, theaveragestandarddeviationis0.57(stddev0.24). Incomparison, the calibration strategy of Carlstrom and Fuerst (1997) yields σ = 0.21 and the one of Bernanke, Gertler, and Gilchrist (1999) σ = 0.53. Theimpliedannualizedrateofreturnis8percentin1998and5percentin2003. Theseaverages are close to the estimates of Moskowitz and Vissing-Jørgensen (2002), who, using Flow of Funds Accounts and the National Income and Product Accounts data from 1963 to 1999, find that the average annual rate of return to entrepreneurial investment is 13 percent with a standard deviation of 7.7. Their calculation is based on equity in the firm only. Taking into account that definition of net worth employed in my exercise includes the net value of the private residence, the annual return on equity in the firm would be about 12 percent in 1998 and 7.5 percent in 2003. The parameter distributions in Table 3 contain some implausible extreme values, such as negative or high bankruptcy cost. A closer investigation of these observations through the lens of the model helps to understand the extreme values. Observations with implied negative bankruptcy cost exhibit a high leverage, on average 8.7, while paying only 34 basis points more. The implied riskparameterforthissubgroupisonly0.08whiletherateofreturnis1. Inotherwords, themodel is unable to explain high leverage and low risk premiums for the given bankruptcy rates. Figure 1 illustrates that either low risk or high rates of return lead to higher leverage in the model. Yet in the model high rates of return also lead to significantly higher interest rates. Since the interest rates in the data are not significantly higher, risk must be low. With the mean of the idiosyncratic shock normalized to 1, the lender must receive an additional reward in the case of bankruptcy to rationalize the high leverage. Observations with high bankruptcy cost (µ > 1) exhibit a low leverage, but the firms in the i subgroup pay on average a premium of 72 basis points over the full sample despite a lower than 18Whensettingtheshareofnetworthusedinproductionξ to1forallobservations,70percentofthe1998sample i and 77 percent of the 2003 sample can be fitted with plausible parameter values. However, the average agency cost and average standard deviations of the idiosyncratic shock are close to the ones reported in Table 3. 19Bernanke,Gertler,andGilchrist(1999)formulatetheagencycostintermsofrealizedoutputandargueforlower values (0.12). Lower values for the bankruptcy cost are consistent with higher bankruptcy rates in their calibration (2.72 percent per quarter). 12

average bankruptcy rate. In the context of the model, given the lower bankruptcy rate the cost of lending must be higher for this subgroup. The spread can be increased by a higher fraction of output lost in the case of bankruptcy or a higher rate of return, which in itself increases the cost of bankruptcy as bankruptcy costs are measured in terms of expected output (the fraction µ ). In i fact, for this subgroup the model also implies an average rate of return of 1.15 with lower than average risk. Solving for the implied parameter of the optimal debt contract relies on 3 assumptions: i) the idiosyncratic shock has a lognormal distribution, ii) the credit score is a good proxy for the bankruptcy rate, and iii) net worth is measured correctly. Measurement error in net worth or the bankruptcy rate may explain the extreme values of the model-implied parameter distributions. Also the assumption of a log-normal distributed idiosyncratic shock may also be too restrictive, yet it is commonly used in macroeconomic models. Despite these pitfalls, the model of financing constraints induced by costly state verification performs well vis-`a-vis the data. The model can rationalize more than two thirds of the credit contracts with plausible parameter values. The implied rate of returns are in line with results using Flow of Funds Accounts and the National Income and Product Accounts data and bankruptcy costs are consistent with studies on bankruptcy incidences. 3 The Holmstro¨m-Tirole Model This section first describes the debt contract with moral hazard as introducted by Holmstr¨om and Tirole (1997). Next, I present the additional bank level data used in this section. I plot the model implied interest rate and the interest rate observed in the data to assess the empirical plausibility of the MH model. I complement my findings with the monitoring cost implied by the debt contract with MH and reduced-form regressions. 3.1 The Debt Contract with Moral Hazard InthemoralhazardmodelbyHolmstr¨omandTirole(1997),risk-neutralentrepreneuriownsarisky return technology. She can invest in either a high quality project with a high probability of success αg or a low quality project with a lower probability of success αb but with private benefit to the i i entrepreneur b k, where k denotes working capital in production. In case of success k is converted i i i into output with constant returns to scale. In case of failure the capital is lost. The entrepreneur’s project choice is not observable. Lender (financial intermediary) j has to pay fraction of the loan as monitoring cost µ to ensure that the entrepreneur invests in the high quality project. The lender’s j capital a is scarce. The lender’s equity share in the loan to an entrepreneur has to earn a rate of j 13

return rb reflecting outside investment opportunities. Investors in this model can be thought of as j depositors who can only invest into a project through the lender. Deposits h at lender j have to j earn a return rh. The crucial assumption of the model is rb > rh. j j j As above collateral is the share of net worth (1−ξ )n that cannot be used in production. I i i assume that in case of default, all collateral is going to the lender. The results are not sensitive to this assumption, as, in equilibrium, only the combined shares output going to the lender and the depositors matter. Let R be the total rate of return to the project, Re the entrepreneur’s share of i i the total rate of return, Rb the bank’s share, and Rh the depositors share. The optimal contract i i maximizes the payoff to the entrepreneur max αgRek i i i ki,aj,hj,R i e,R i h,R i b subject to the incentive compatibility constraint of the entrepreneur: αg(Rek +(1−ξ )n ) ≥ αb(Rek +(1−ξ )n )+b k (10) i i i i i i i i i i i i the incentive compatibility constraint of the lender (monitoring): αgRbk +(1−αg)(1−ξ )n −µ (k −ξ n ) ≥ αbRbk +ρ(1−αb)(1−ξ )n , (11) i i i i i i j i i i i i i j j j the participation constraint of the lender: αgRbk +(1−αg)(1−ξ )n −µ (k −ξ n ) ≥ (1+rb)a , (12) i i i i i i j i i i j j the participation constraint of depositors: αgRdk ≥ (1+rh)h , (13) i i i j j the feasibility requirement: a +h −µ (k −ξ n ) ≥ k −ξ n . (14) j j j i i i i i i and the budget constraint: R ≥ Re+Rb+Rh. (15) i i i i First, note that all six equations hold with equality in equlibrium. Second, the interest rate on a loan is defined only implicitly. Note that in the case of sucess of the project, the joint payoff to the lender and depositors has to be equal to the pricipal and interest payment: (Rb+Rh)k = Z (k −ξ n ), (16) i i i i i i i 14

where Z is the gross interest rate paid on the loan. This additional equation allows me to test the i model. In order to derive a interest rate as a function of observable parameters and variables, I use the adequacy ratio κ - that is, lender’s equity share in a loan j a j = κ . (17) j a +h j j Scaling all equations by k and rearranging equations yields i r¯ (1+µ )(1−ξ 1 )+µ (1−ξ 1 )−(1−αg)(1−ξ ) 1 (Rb+Rd) = j j iLi j iLi i i Li, (18) i i αg i where ki = L - the leverage - and r¯ = [(1 + rb)κ + (1 + rh)(1 − κ )] - the weighted average ni i j j j j j financing cost of the loan. In the following section I describe the data, I will use for rh, rb, κ, and µ in equation (18) to derive a model implied interest rate Z using equation (16). 3.2 Banks Data The confidential SSBF include the lender’s name. I restrict the matching of firms and financial institutions to banks and credit unions for two purposes. First, models of financing constraints usually consider banks. The second reason is a practical concern. The information about the financial institution available is more detailed for banks than the information about leasing companies or car-manufacturer-owned financing companies. The latter are typically not deposittaking institutions and are therefore not subject to the same regulatory requirements for reporting and capital standards. Table 4 summarizes the bank variables employed in the assessment of the MH model. The interest cost, used to proxy for rh and other cost are taken from the banks’ income statements in the Call Reports.20 The interest cost is measured relative to liabilities minus equity to calculate the average of an additional dollar of external finance to a bank. The other (non-interest) cost to asset ratio - the average cost for one additional dollar of assets on the banks’ balance sheets - proxy for monitoring cost (µ). Return on equity, (rb), is profit divided by bank equity. The adequacy ratio (κ), measuring the composition of a banks financing, is the equity divided by total liabilities. Note that the income and cost variables in the Call Reports are reported as year-to-date and need to be converted to obtain annualized rates. Last, the loan share is defined as the share of loans of 20I pull the crucial bank information from the Call Reports, the Thrifts Financial Reports, and Call Reports for creditunions. Thesereportsarefiledbyfinancialinstitutionsonaquarterlybasisandincludethebalancesheetofa financial institution, the income statement, and the regulatory capital ratios. In some cases the bank’s name would allow for the identification of individuals in the SSBF. Since the SSBF provides detailed private information of firm owners, bank identities are available in the confidential dataset administered by the Federal Reserve Board but not in the public dataset. 15

total assets.21 For the 1998 sample 493 out of 924 (53 percent) credit contracts could be matched with bank data.22 The 493 credit contracts originate from 316 different institutions. Many of the unmatched observations do not include a financial institution’s name or were labeled “fake bank” by the interviewer. The second, large group received credit for car manufacturer-owned financing companies, leasing companies, or specialized investors. For firms that were denied credit the match rate is 66 percent (110 out of 166). For the 2003 sample the match rate is significantly higher. Here, 1340 out of 1693 (79 percent) credit contracts could be matched to financial statements of banks. The 1340 credit contracts involve 613 different institutions. Only few firm owners did not identify the financial institutions. Most of the unmatched observations received financing from car manufacturer-owned financing companies. Of the firms whose credit application was rejected, 72 percent (101 out of 136) were matched to banks. 3.3 Fitting the MH Contract Having concluded that CSV is an empirically plausible model of financing frictions, I now turn to the model of financing frictions based on moral hazard by Holmstr¨om and Tirole (1997). This model focuses on banks’ financing costs and balance sheet compositions. To my knowledge, this is the first paper to use credit contract data to assess this model. As pointed out in section 3.1, this model does not explicitly specify an interest rate on the loan. The implied interest rate payment can be derived from the sum of the payments to the bank (Rb) and the investors (depositors) Rh. Recall equations (16) and (18) (Rb+Rh)L = Z (L −ξ ) i i i i i i and r¯ (1+µ )(1−ξ 1 )+µ (1−ξ 1 )−(1−αg)(1−ξ ) 1 (Rb+Rh) = j j iLi j iLi i i Li i i αg i where r¯ = [(1+rb)κ +(1+rh)(1−κ )] - the weighted average financing cost of the loan. Note j j j j j that the right-hand side of the last equation contains only variables that can be constructed from the SSBF and Call Reports. As in the assessment of CSV I use the credit score to proxy for the probability of success αg and the share of fixed assets proxies for the share of collateral (ξ ). The i 21The respective variables classification in the Call Report data can be found in the Appendix. 22Theconfidential2003SSBFdataalreadycontainprecisematchesfrominterviewsformostobservations. Forthe rest,Iusethebank’snamefromthecreditcontractintheSSBFtoretrievethebank’sidentifier(RSSDID)fromthe NationalInformationCenter(NIC).NICtrackstheownershipofbanksincludingallmergersandthereforeallowsfor an accurate identification of banks. 16

construction of the rate of return on bank capital (rb), interest on deposits (rh), the adequacy ratio j j (κ ), and the monitoring cost (µ ) is explained in previous section j j Figure 3 plots the interest rates paid and the interest rates implied by equation (18). The left panel show the joint distribution of interest rates for the 1998 data and the right panel for the 2003 data. Both graphs omit observations with implied interest rates less than 0 percent or more than 20 percent. The average difference is 3.22 percentage points in the 1998 data and 2.98 percentage points in 2003. The correlation between the actual and the implied interest rate is 0.05 in both samples. Removing 25 outliers - an actual interest rate lower than 4 percent or an implied interest rates larger than 20 percent - increases the correlation to 0.1 in the 1998 sample. However, the same criteria applied to the 2003 yields 5 outliers and reduce the correlation to 0.03. The standard deviation of the implied interest rate (5.78) is substantially larger than the standard deviation in the actual interest rate (1.90) for the 1998 sample. For the 2003 sample the standard deviations are about the same, 2.35 and 2.68 respectively. In light of these these findings the MH model appears to be inconsistent with the data. There are three potential explanations for the poor performance of the MH model: 1) bank costs are mismeasured , 2) the bank balance sheets and costs, for instance due to competition, do not matter for small business credit interest rates,23 or 3) the model misspecifies how bank variables affect the interest rate on a credit. So far, this exercise assumes that balance sheet and income statement information of the bank allows me to measure banks’ financing costs, rate of returns, and monitoring costs accurately. Taking the balance sheet and income statement information from a particular quarter may not adequately reflect the medium to long run cost but within quarter special items in accounting. Asafirstrobustnesstest,Iuseanalternativemeasureofmonitoringcosttodealwiththispoint, mismeasured cost. Instead of using all “other” (non-interest) cost as a measure of monitoring cost, Iusesalariesandemployeebenefits, dataprocessingexpenses, legalfees, andinsuranceassessments only.24 Due to data availability this exercise is done for the 2003 sample only. The average cost fall to 1.4 percent compared to 2.9 percent when all other cost are considered. Employing this measure of monitoring cost reduces the difference between the model implied interest rate and the actual interestrateto40basispoints. However,thecorrelationremainslow(0.06).25 Second,thestandard deviation of the implied interest rate drops to only 1.47, far lower than the standard deviation of the interest rate in the data (2.68). Similarly, using the FDIC’s capital-ratio requirements for well- 23BergerandBlack(2010)arguethatlargebankshaveacompetitiveadvantageinprovidingcollateral-basedloans. Since almost all loans are collateralized, small bank can be thought of the competitive fringe. 24This measure therefore excludes expenses related to real estate and other fixed assets. 25Removing extreme outliers increases the correlation to 0.07 in the 2003 data. The scatter plots can be found in the Appendix. 17

capitalized banks - 10 percent (Basel I) or 7.5 percent (Basel II) - instead of actual adequacy ratios does not change the results. Next, instead of using the information from one quarter, I average the quarterly data over three years. However, I find no difference in terms of correlation or differences in the interest rates (see Appendix). Last,insteadoftakingthemonitoringcostfromtheincomestatement,Isolveforthemonitoring costµ thatequalizesthemodelimpliedinterestrateandtheactualinterestrate. Themodelimplies j medianmonitoringcostof2.14percentin1998and5.37percentin2003. While2.14percentappears to be a reasonable number when compared to the cost measure in Table 4, 5.37 percent is close to the maximum. Only 53 percent of contracts in the 1998 sample imply µ ∈ [0,0.1] - that is, monitoring cost between $0 and $100,000 on a $1 mill. loan. In the 2003 sample 34 percent of the contracts imply monitoring cost in this, arguably wide, range. Therefore, measurement error in the monitoring cost is not driving the results. To assess the other two hypotheses on importance of bank variables, I run reduced-form regressions to explain the loan interest rate. CSV and MH highlight the importance of different variables for the loan interest rate. The interest rate in the debt contract with CSV depends on firm-specific parameters and the risk-free rate. To see this, substitute equation (9) in equation (2). ω¯ Rk(1+rξ )+(1−ξ )(1+r) (cid:16) 1−G(ω¯ ) R i k (cid:17) i i i i i (1+r) Z = . (19) i (cid:16) Rk (cid:17) 1+rξ −(1+r) 1−G(ω¯ ) i i i (1+r) The interest rate on the loan in MH depends on firm-specific and on bank-specific parameters. Using equations (16) and (18) yields Z = r¯ j (1+µ j )+µ j − (1−α i g) × (1−ξ i ) L 1 i, (20) i αg αg (1−ξ 1 ) i i iLi with r¯ = [(1+rb)κ +(1+rh)(1−κ )] - the weighted average financing cost of the loan. Hence, j j j j j the MH model predicts that, in equilibrium, the interest rate is increasing in the financing cost parameters rb, rh, and κ and decreasing in the quality of the project α . j j j j The regression analysis of the interest rate confirms the hypothesis that bank balance sheets and costs do not matter for small business credit interest rates and, by extension, rejects the third hypothesis. Table 5 shows that bank variables have little explanatory power.26 By contrast, firm 26Excluding quarterly dummies yields highly significant estimates of the interest cost, -24.95 for the 1998 sample and 43.52 in the 2003 sample which may be due to difference in the federal funds rate over the sample period. The latter translates to a 43.5 basis points increase on the credit interest rate in response to a 100 basis points increase in the cost of funding for the bank. 18

specific variables are highly significant. This highlights the differences between CSV and MH. In CSV the interest rate is a function of firm variables and the risk-adjusted outside investment opportunities of the bank - conventionally proxied by the risk-free rate. MH, on the other hand, focuses on the importance of the bank’s balance sheets. On the firm’s side, size and age have a strong negative impact on the interest rate, so does the credit score. Yet net worth, measured in millions, as only a neglectible economic effect. An additional million dollars of net worth reduces the interest rate by 9-14 basis points. Rajan (1992) points out that longer lending relationships result in informational monopolies which is especially true for small businesses. Banks also may try to attract new customers with teaser rates on the first loan and raise the interest rate on subsequent loans. The positive and significant coefficient on the length of the relationship in the 2003 sample supports this view. There appears to be no relationship effect in the 1998 SSBF sample. However, this may be the result of the omission of credit renewals which biases the 1998 sample. Onepotentialsourceofbiasarisesfromdifferencesinlocaleconomicconditionsthatarearguably important in order to assess the riskiness of a small business. The regressions therefore include the fraction of late and non-accrual loans which, at least for smaller banks, is a plausible proxy for differences in local economic conditions. In fact, in most regressions the share of loans that is 90 day past due has a significant positive effect on the loans interest rate (not shown here). Using the alternative cost measures described above confirm the findings from the regression analysis (see Appendix). While it cannot be excluded that there may be additional measurement errors in the rate of return of a bank or the monitoring cost, but it is hard to imagine that such measurement errors are correlated with covariates in way that would overturn the key finding of the section: The poor performance of MH vis-`a-vis the data is the result of a lack of explanatory power of bank variables for the interest rate of the loan. 4 Reduced-Form Financing Constraints Frictionless capital markets imply an undetermined capital structure of firms, and no systematic relationship between net worth and credit emerges. One exception is the case of firms operating a decreasing economies of scale technology limiting optimal firm size. In that case, everything else equal, firms with higher net worth should take on less credit as they are closer to the optimal size. Here, when capital markets are frictionless, a negative relationship between net worth and the credit amount should be observed in the data. The defining deviation from frictionless capital markets - and one that I will verify empirically 19

- is that financing frictions imply a positive relationship between net worth and the credit amount. In particular, the credit amount is some multiple of net worth. Using the definition of debt and leverage in CSV in equation (9) yields: (cid:16) Rk (cid:17) 1−ξ + G(ω¯ ) i ξ (1+r) i i (1+r) i d = n . (21) i (cid:16) Rk (cid:17) i (1+r) 1−G(ω¯ ) i i (1+r) In the MH model, the relationship of net worth and loans, which I will assess in the next section, can be derived using equations (10), (11), (13), (14), (15), and (17). For simplicity, assume ξ = 1. (cid:16) (cid:17) µj +αg R − bi+µj αg−αb i i αg−αb d = i i i i n (22) i (cid:16) (cid:17) i (1+rh)(1−κ )(1+µ )−αg R − bi+µj j j j i i αg−αb i i ThederivationcanbefoundintheAppendix. NotethataccordingtotheMHmodel,inequilibrium, debt should be higher for projects with higher success probability (αg), for projects with higher i gross rates of return (R ), for banks with low financing cost rh, and projects with a higher bank i j share of funding (κ ). The latter is due to the fact that this model does leverage net worth of the j entrepreneur and the net equity of the financial intermediary. As for the interest rate, the models differ on the importance of bank-specific variables. Equations (21) and (22) show that debt contract models imply that for the firms that received credit the credit amount is some multiple of net worth n : i d = ϕn , i i where d is the credit amount of entrepreneur i and ϕ the net worth multiplier. Reduced-form i financing constraints of this form are also commonly used in the literature on entrepreneurship, for instance in Evans and Jovanovic (1989), Cagetti and de Nardi (2006), and Buera (2009). I test this relationship in the small business data. Clearly, not all firms in need of external finance receive credit and the SSBF provide indicators of a firm’s success in obtaining credit.27 So the regression analysis will take selection into account. Roughly one fifth of all firms in the 1998 sample and about one third of all firms in the 2003 sample obtained credit. This large difference is due to the fact that the 1998 explicitly excluded 27Thefirstmeasureiswhetherfirmsappliedforcreditwhentheywereinneedofcredit. Inparticular,bothsamples identify firms reporting that were in need of credit, but decided against applying for a loan as they feared denial. A general pattern emerges. First, young firms are less likely to apply for credit, and their applications are turned down more frequently. Second, not all credit applications were granted by financial institutions. Not being able to obtain (additional) external finance, these firms have to finance investment with internal resources only. Levenson andWillard(2000),KonandStorey(2003)andMeisenzahl(2010)discussthistypeofcreditrationinginmoredetail. 20

renewals of lines of credit. This clearly introduces a selection bias towards 1) more denials and 2) younger and smaller firms receiving credit (see Table 1). As equations (21) and (22) show, the net worth multiplier ϕ depends on the riskiness, bankruptcy cost, and the rate of return of a project, each of these parameter can be differ across projects. Second, net worth may be measured with error. This implies the following estimation equation: d = α+βn +(cid:15) (23) i i i The agency model suggests β > 0. In the regressions credit amount and net worth are measured in millions. Table 6 summarizes the results for the full sample 1998 and 2003. The credit amount and net worth exhibits a strong, positive relationship. An additional dollar of net worth account for 30 cents of additional external finance. The variation in net worth explains the variation in credit amounts well (R2 of 0.29 and 0.27). Measurement error in n may bias the estimator downwards. i The second and the fifth column show the results after controlling for firm size as a proxy for the differences in the rate of return. The point estimates for the net worth multiplier barely change. In line with the prediction of heterogeneity in the rate of return, firm size affect the credit amount positively. Not all credit applications were accepted. Hence, the OLS estimators are potentially subject to a selection bias. The selection model estimates, columns 3 and 6 in Table 6, show that selection not a concern in the analysis. The SSBF data provide a rich set of additional control variables for sources of heterogeneity: the credit score of the firm, previous bankruptcy, census region, industry, and organizational form. The credit score, judgments against the firm, and previous private and firm bankruptcies proxy for differences in risk. Regional and industry dummies control for differences in risk and rate of return in these dimensions. While there is evidence in the 1998 sample that firms with a solid financial history received more credit, no impact of credit scores can be found in the 2003 sample. Due to a lackofvariationinthe1998sample,judgments,andpreviousprivateandfirmbankruptciescanonly be used in the 2003 sample. Judgments have a strong negative effect while previous bankruptcies exhibit no significant effect. Firms with judgments already pending are more risky. The negative effect of judgments on the credit amount is in line with the prediction of lower leverage for higher risk firms. Due to differences in liability, the variation in the legal form of the firms proxies for differences in bankruptcy cost. However, in both samples the legal form has no significant effect. This is consistent with the fact that half of the sample provided personal guarantees, and therefore limited liability of the firm has little effect. Bank variables described above have no significant effect on the credit amount. AsTable7showsemployingthislargesetofcontrolsdoesnotyieldsignificantlydifferentresults for the estimated effect of net worth. Again controlling for selection does not change the results. 21

Bankregulatorycapitalratioswereusedasexclusionrestrictionsintheselectionequationasseveral studies found an effect of capital ratios on lending.28 Section 2.2 argues that in the context of small businesses a wider definition of net worth is appropriate. To ensure that the results are not solely the result of the definition of net worth, I consider three additional definitions consistent with the pattern of collateral offered. Since equipment accounts for almost half of collateral offered, we always include depreciable assets in our definitions: 1) firm assets and less firm’s liabilities, 2) firm’s depreciable assets, land, building and net value of private residence less firm’s loans, and 3) firm’s depreciable assets, land, buildings, and net value of private residence less firm’s liabilities. The narrower definitions of net worth also yieldarobustpositivecreditamount-networthrelationshipbuttheexplanatorypowerofnetworth is substantially reduced in some regressions (see Appendix). Receivinglesscreditthanappliedforisanotherpossibleconstraint. TheSSBFexplicitlyaskfor the loan amount applied for and the loan amount received. Over 90 percent of the small business owners report that the loan amount received is equal to the loan amount they applied for. While it cannot be excluded that the loan amount on the application was agreed upon informally prior to the application, there is no indication in the surveys that firms do not receive the desired loan amount. Consistent with both models of financing constraints considered here, net worth and the credit amount exhibit a robust positive relationship. In the SSBF data, an additional dollar of net worth accounts for 30 cents of additional external finance. In comparison, in their model of entrepreneurship, Evans and Jovanovic (1989), using the National Longitudinal Survey of Young Men,estimatetheeffecttobe44centsofexternalfinanceforanadditionaldollarofnetworthwhile Cagetti and de Nardi (2006) calibrate the effect to roughly 67 cents for entrepreneurs to match the US wealth distribution.29 5 Small vs Large Banks It is worthwhile to note that small businesses tend use with small banks. In the 1998 sample, there is little evidence for dominance of large banks in the small business lending market. Summing up the market shares of all banks whose market share exceeds 1 percent, I find that 6 banks account for 23 percent (19 percent using firm weights in the SSBF) of all credit contracts and 31 percent (32 percent weighted) of denials in the sample. In the 2003 sample there is more concentration, but that was to be expected since the consolidation in the banking industry continued. Here, 11 banks 28This result is reported for instance in Bernanke and Lown (1991), Hancock and Wilcox (1993, 1994), and more recently in Berrospide and Edge (2010) who also provide an comprehensive survey of the earlier literature. 29Buera (2009) estimates a broader range from 10 cents to 3 dollars. 22

have at least 1 percent of the market. Together large banks account for 30 percent (23 percent weighted) of the credit contracts by banks. They also account for 38 percent (44 percent weighted) of denials. Large banks, in this section defined as banks with more than $50 billion assets, account for about one third of all credit contracts (and also roughly one third when measured by total credit amount) and for almost 40 percent of credit denial. This section examines difference between small and large banks to ensure that the results presented in the previous section are not driven by large financial institution. The latter are more diversified and therefore their balance sheet information may be less precise in measuring the variables of interest. Table8showsthatthereareonlyminordifferencesbetweensmallbusinesseswhodecidetothat bank with small banks and those bank with large banks. The perhaps most striking difference is that small business owners who bank with the large banks tend to have more equity. First, I report the result for the credit amount regressions by bank size. Since selection has no significant effect I omit the results in Table 9. The main result, the importance of net worth, does not change. While the point estimates for the 2003 sample are the same for small and large banks, they differ significantly for the 1998. Next, I explore whether there are differences by bank size with respect to interest rates. Tables 10 and 11 summarize the results for the 1998 and 2003 sample. With the exception of the interest cost, which contrary with MH enter negatively, no significant differences arise in the 1998 sample. Firm variables, such as size and sectoral dummies, appear to be more important. In contrast, the 2003 sample in Table 11 exhibits sharp differences. Here bank balance sheet information is important in the subsample of large banks, albeit the point estimate of cost measure is negative and weakly significant. A 100 basis point difference in the financing cost of large banks increases the interest rate on a small business loan by more than 250 basis points. This result may reflect the increasing concentration in the banking industries which leaves large banks with more market power. Yet large banks still serve less than a third of the market. For small banks, however, bank variables are not significant, even though they are arguably closer to a standard model of banks than large financial conglomerates. In contrast to MH, this result supports a more competitive banking market model, for instance small banks can be seen as competitive fringe to an oligopoly. Therefore, the poor performance of MH is not driven by large banks. While the results presented here do not support MH, bank variables may still matter in the context of credit rationing (Stiglitz and Weiss, 1981). For instance, capital ratios affect the loan supply of loans (Berrospide and Edge, 2010). 23

6 Conclusion This paper exploits business credit contracts and net worth of entrepreneurs to examine the empiricalplausibilityoftwocommonlyusedmodelsoffinancingconstraints-costlystateverification andmoralhazard. Patternsofcollateral,asguidelinefortheconstructionofnetworth,suggestthat banks accept all assets on the firm’s balance sheet and private residences as collateral. Consistent with both models, one dollar of additional net worth accounts for about 30 cents of additional external finance. Costly state verification can rationalize more than two thirds of the credit contracts in the data. Bankruptcy costs, implied by small business credit contracts when fitting the CSV debt contracts, are 28 percent of expected output and the implied rate of return lies between5percentand8percentannually. Thus, CSVisanempiricallyplausiblemodeloffinancing constraints. ThemoralhazardmodelbyHolmstr¨omandTirole(1997)performspoorly. Thismodel emphasizes the importance of bank variables for loan pricing; a hypothesis that does not hold in small business loan contracts. While shedding light on the empirical plausibility of financing constraint models, this study also raises questions about the magnitude of the impact of financing constraints in macroeconomic models. First, less than one half of the sample took out credit. Second, bank variables such as financingcostandbalancesheetcompositionhavelittleexplanatorypower. Third, smallbusinesses - accounting for almost one half of economic activity in the US - are not highly leveraged.30 Another question is whether credit is, as in most macroeconomic models, constantly demanded for investment or whether credit should be thought of as an insurance policy against business cycle or liquidity shocks (Diamond and Rajan, 2001; Kashyap, Rajan, and Stein, 2002). References Albuquerque, Rui and Hugo A. Hopenhayn. 2004. “Optimal Lending Contracts and Firm Dynamics.” Review of Economic Studies 71 (2):285–315. Alti, Aydogan. 2003. “How Sensitive Is Investment to Cash Flow When Financing Is Frictionless?” Journal of Finance 58 (2):707–722. Angelini,PaoloandAndreaGenerale.2008.“OntheEvolutionofFirmSizeDistribtions.”American Economic Review 98 (1):426–438. Berger, Allen and Gregory Udell. 1998. “The Economics of Small Business Credit Finance: The 30This support the view of den Haan, Sumner, and Yamashiro (2007) and Christensen and Dib (2008) who find little amplification through financing frictions in the response to a monetary policy shock. 24

Role of Private Equity and Debt Markets in the Finanacial Growth Cycle.” Journal of Banking and Finance 22 (6-8):613–673. Berger, Allen N. and Lamont K. Black. 2010. “Bank size, lending technologies, and small business finance.” Journal of Banking and Finance forthcoming. Berkowitz, Jemery and Michelle J. White. 2004. “Bankruptcy and Small Firms’ Access to Credit.” Rand Journal of Economics 35 (1):69–84. Bernanke, Ben S. and Mark Gertler. 1989. “Agency Costs, Net Worth, and Business Fluctuations.” American Economic Review 79 (1):14–31. Bernanke, Ben S., Mark Gertler, and Simon Gilchrist. 1999. “The Financial Accelerator in a Quantitative Business Cycle Framework.” In Handbook of Macroeconomics, edited by John Taylor and Michael Woodford. Amsterdam, New York and Oxford: Elsevier Science, North- Holland, 1341–1393. Bernanke, Ben S. and Cara S. Lown. 1991. “The Credit Crunch.” Brookings Papers on Economic Activity 1991 (2):205–247. Berrospide, Jose M. and Rochelle M. Edge. 2010. “The effects of bank capital on lending: What do we know, and what does it mean?” Finance and Economics Discussion Series 2010-44. Bitler, Marianne, Alicia Robb, and John Wolken. 2001. “Financial Services Used by Small Businesses: Evidencefromthe1998SurveyofSmallBusinessFinances.” Federal Reserve Bulletin 87:183–205. Buera, Francisco J. 2009. “A Dynamic Model of Entrepreneurship with Borrowing Constraints: Theory and Evidence.” Annals of Finance 5 (3-4):443–464. Cagetti, Marco and Mariacristina de Nardi. 2006. “Entrepreneurship, Frictions, and Wealth.” Journal of Political Economy 114 (5):835–870. Carlstrom, Charles and Timothy Fuerst. 1997. “Agency Costs, Net Worth, and Business Fluctuations: A Computable General Equilibrium Analysis.” American Economic Review 87 (5):893–910. Cavalluzzo, Ken and John Wolken. 2005. “Small Business Loan Turnovers, Personal Wealth and Discrimination.” Journal of Business 78 (6):2153–2177. Chava, Sudheer and Michael R. Roberts. 2008. “How Does Financing Impact Investment? The Role of Debt Covenants.” Journal of Finance 63 (5):2085–2121. 25

Chen,Nan-Kuang.2001.“Banknetworth,assetpricesandeconomicactivity.”JournalofMonetary Economics 48 (2):415 – 436. Christensen, Ian and Ali Dib. 2008. “The Financial Accelerator in an estimated New Keynesian Model.” Review of Economic Dynamics 11 (1):155–178. Christiano, Lawrence J., Roberto Motto, and Massimo Rostagno. 2007. “Financial Factors in Business Cycles.” manuscript, Northwestern University. Clementi, Gian Luca and Hugo Hopenhayn. 2006. “A Theory of Financing Constraints and Firm Dynamics.” Quarterly Journal of Economics 121 (1):229–265. Cooley, Thomas F. and Vincenzo Quadrini. 2001. “Financial Markets and Firm Dynamics.” American Economic Review 91 (5):1286–1310. Covas, Francisco and Wouter J. den Haan. 20010. “The Role of Debt and Equity Finance over the Business Cycle.” American Economic Review forthcoming. Cummins, Jason G., Kevin A. Hassett, and Stephen D. Oliner. 2006. “Investment Behavior, Observable Expectations, and Internal Funds.” American Economic Review 96 (3):796–810. den Haan, Wouter J., Steven W. Sumner, and Guy M. Yamashiro. 2007. “Bank Loan Portfolios and the Monetary Transmission Mechanism.” Journal of Monetary Economics 54 (3):904–924. Diamond, DouglasW.andRaghuramG.Rajan.2001. “BanksandLiquidity.” American Economic Review 92 (2):422–425. Evans, David and Boyan Jovanovic. 1989. “Estimates of a Model of Entrepreneurial Choice under Liquidity Constraints.” Journal of Political Economy 97 (4):808–827. Fazzari, Steven M., R. Glenn Hubbard, and Bruce C. Petersen. 1988. “Financing Constraints and Corporate Investment.” Brookings Papers on Economic Activity 1988 (1):141–206. Fisher, Jonas D. M. 1999. “Credit Market Imperfections and the Heterogeneous Response of Firms to Monetary Shocks.” Journal of Money, Credit and Banking 31 (2):187–211. Gale, Douglas and Martin Hellwig. 1985. “Incentive-Compatible Debt Contracts: The One-Period Problem.” Review of Economic Studies 52 (4):647–663. Giambona, Erasmo and Armin Schwienbacher. 2008. “Debt Capacity of Tangible Assets: What is Collateralizable in the Debt Market?” manuscript, University of Amsterdam. 26

Gilchrist, Simon, Alberto Ortiz, and Egon Zakrajsek. 2009. “Credit Risk and the Macroeconomy: Evidence from an Estimated DSGE Model.” manuscript, Boston University. Gomes, Joao F. 2001. “Financing Investment.” American Economic Review 91 (5):1263–1285. Haltiwanger, John C., Ron Jarmin, and Javier Miranda. 2010. “Who Creates Jobs? Small vs. Large vs. Young.” NBER Working Paper 16300. Hancock, Diana and James A. Wilcox. 1993. “Has There Been a ‘Credit Crunch’ in Banking? The EffectsofBankLendingofRealEstateMarketConditionsandBankCapitalShortfalls.” Journal of Housing Economics 3 (1):31–50. ———. 1994. “Bank Capital and Credit Crunch: The Roles of Risk-Weighted and Unweighted Capital Regulations.” Journal of the American Real Estate and Urban Economics Association 22 (1):59–94. Hennessy, Christopher and Toni Whited. 2007. “How Costly is External Financing? Evidence from a Structural Estimation.” Journal of Finance 62 (4):1705–1745. Holmstr¨om, BengtandJeanTirole.1997. “FinancialIntermediation, LoanableFunds, andtheReal Sector.” Quarterly Journal of Economics 112 (3):663–691. Hubbard, R. Glenn. 1998. “Capital-Market Imperfections and Investment.” Journal of Economic Literature 36 (1):193–225. Kaplan, Steven N. and Luigi Zingales. 2000. “Investment-Cash Flow Sensitivities are not valid Measures of Financing Constraints.” Quarterly Journal of Economics 115 (2):707–712. Kashyap,AnilK.,RaghuramG.Rajan,andJeremcyC.Stein.2002. “BanksasLiquidityProviders: An Explanation for the Coexistence of Lending and Deposit-Taking.” Journal of Finance 57 (1):33–73. Kon, Y. and D. J. Storey. 2003. “A Theory of Discouraged Borrowers.” Small Business Economics 21 (1):37–49. Levenson, Alec and Kristen Willard. 2000. “Do Firms get what they want? Measuring Credit Rationing Experience by Small Business in the U.S.” Small Business Economics 14 (1):83–94. Levin,Andrew,FabioNatalucci,andEgonZakrajsek.2004. “TheMagnitudeandCyclicalBehavior of Financial Market Frictions.” manuscript, Federal Reserve Board. Mach, Traci and John Wolken. 2006. “Financial Services Used by Small Businesses: Evidence from the 2003 Survey of Small Business Finances.” Federal Reserve Bulletin 92:A167–A195. 27

Masulis, Ronald W. 1988. The Debt / Equity Choice. Cambridge: Ballinger Publishing Company. Meh, CesaireA.andKevinMoran.2010. “TheRoleofBankCapitalinthePropagationofShocks.” Journal of Economic Dynamics and Control 34 (3):555–576. Meisenzahl, Ralf R. 2010. “Financing Constraints and the Evolution of Firm Size Distribution: New Evidence From US Small Businesses.” manuscript, Federal Reserve Board. Moskowitz, Tobias J. and Annette Vissing-Jørgensen. 2002. “The Returns to Entrepreneurial Investment: A Private Equity Premium Puzzle?” American Economic Review 92 (4):745–778. Nolan, Charles and Christopher Thoenissen. 2009. “Financial Shocks and the US Business Cycle.” Journal of Monetary Economics 56 (4):596–604. Paulson, AnnaL., RobertM.Townsend, andAlexanderKaraivanov.2006. “DistinguishingLimited Liability from Moral Hazard in a Model of Entrepreneurship.” Journal of Political Economy 114 (1):100–144. Rajan, Raghuram G. 1992. “Insiders and Outsiders: The Choice between Informed and Arm’s- Length Debt.” Journal of Finance 47 (4):pp. 1367–1400. Stein, Jemery C. 2003. “Agency, Information, and Corporate Investment.” In Handbook of the Economics of Finance, edited by G.M. Constantinides, M. Harris, and R. M. Stulz. Amsterdam, New York and Oxford: Elsevier Science, North-Holland, 111–165. Stiglitz, Joseph and A. Weiss. 1981. “Credit Rationing in Markets with Imperfect Information.” American Economic Review 71 (3):393–410. Sullivan, Teresa A., Elizabeth Warren, and Jay Lawrence Westbrook. 1989. As We Forgive Our Debtors: Bankruptcy and Consumer Credit in America. New York: Oxford University Press. Townsend, Robert. 1979. “Optimal Contracts and Competitive Markets with Costly State Verification.” Journal of Economic Theory 21 (2):265–293. von Heideken, Virginia Queijo. 2009. “How Important are Financial Frictions in the U.S. and the Euro Area?” Scandinavian Journal of Economics 111 (3):567–596. Wasmer, Etienne and Phillipe Weill. 2004. “The Macroeconomics of Labor and Credit Market Imperfections.” American Economic Review 94 (4):944–964. 28

Appendix A - Data Construction of Bank Variables Bank Variables from the Call Reports for commerical banks (RIAD, RCON), thrifts (SVGL), and credit unions (CUSA): • bank capital a from RIAD3210, SVGL3491, and CUSA3210 - bank equity • deposits h - for commercial banks: sum of RCON6631 (non-interest bearing deposits), RCON6636 (interest bearing deposits), RCONb993 (Fed funds purchased), RCONb995 (Securities sold, repurchase), RCON3548 (Trading Liabilities), and RCON3190 (other borrowed money), for thrifts: SVGL2339+SVGL2071 (Deposits and Escrows), after 2003Q4 SVGLC503andSVGL3101(otherborrowing), andforcreditunionsCUSA4584(borrowings), CUSA2302 (non-member deposits), and CUSA3139 (shares). • bank’s outside rate of return rb - Return on Equity. RIAD4340 (Net Income) divided by RIAD3210 (bank equity), SVGL4340/SVGL3491, and CUSA4273/CUSA3210. • depoistor’s outside rate of return rh - RIAD4073, SVGL0481, CUSA4073 (Total interest rate expense) divided by h,. • monitoring cost µ - RIAD4093, sum of SVGL4141, SVGL0495, SVGL4154, SVGL0499, CUSA4093 (non-interest expense) divided by total assets (here h and equity). Other variables in regressions • share of loans 30-90 days late, over 90 days late and non-accrual. RCFD1406, RCFD1407, RCFD1403 divided by RCON1400; SVGL3936, SVGL3942, and SVGL3948 divided by SVGL1576(non-mortgage)+SVGL0446(Mortgage) and CUSA1128 (30-60 days), CUSA1125 (60-120 days) and CUSA1127 divided by CUSA1263. • Risk-Based Capital Ratio RCFD7205, SVCC7205, for unions only CUSA7276 risk based net worth ratio. 29

Appendix B - Robustness To ensure that the results in section 4 are not only driven by the definition of net worth, I provide the regression results for various other definitions of net worth. The definition employed in the regression in Table 12 follows Giambona and Schwienbacher (2008). They argue that only hard tangible assets can be used as collateral. Hence, in Table 12 net worth is defined as the sum of land, depreciable assets, and the net value of the private residence minus loans. While net worth is still significant in all regression, the explanatory power of net worth is greatly reduced relative the baseline definition. Table13displaystheresultsofadifferentdefinitionofnetworthallowinthespiritofGiambona and Schwienbacher (2008). Instead of subtracting loans only, all firm liabilities are subtracted from the sum of land, depreciable assets, and the net value of the private residence to calculate net worth. Again, the explanatory power of net worth is reduced relative to the baseline definition for the results are still significant. The last robustness check with respect to the definition of net worth uses the firm balance sheet only. In Table 14 , net worth is defined as all firm assets minus all firm liabilities. The results are comparable with the baseline definition in terms of explanatory power, yet the point estimates are slightly higher. 30

Appendix C - Derivation of Leverage in MH Assume ξ = 1. Rewriting yields equations (10) and (11) b Re = i αg −αb i i and µ (1−L ) Rb = j i i αg −αb i i Using the budget constraint R = Re+Rb+Rh, equation (13) becomes i i i i (cid:18) (cid:19) b µ (1−L ) αg R − − j i = (1+rh)h . i i αg −αb αg −αb j j i i i i Next, replacing h , using equation (14) and the definition of the adequacy ratio, yields j (cid:18) (cid:19) b µ (1−L ) αg R − − j i = (1+rh)(1−κ )(1+µ )(1−L ). i i αg −αb αg −αb j j j i i i i i Solving for 1 = ki gives Li ni µj +(1+rd)(1−κ )(1+µ ) k i αg−αb j j j = i i . (cid:16) (cid:17) n i (1+rh)(1−κ )(1+µ )−α R− b+µj j j j g αg−αb i i Hence, (cid:16) (cid:17) µj +αg R − bi+µj αg−αb i i αg−αb d = i i i i n . i (cid:16) (cid:17) i (1+rh)(1−κ )(1+µ )−αg R − bi+µj j j j i i αg−αb i i 31

Table 1: Summary Statistics - Firm Variables 1998 2003 mean std 25% 50% 75% mean std 25% 50% 75% NumberofEmployees 12.8 28.9 2.0 5.0 11.0 15.0 31.5 3.0 6.0 13.0 FirmAge 11.6 8.7 5.0 10.0 16.0 15.3 11.3 7.0 13.0 22.0 LengthofLendingRelationship 7.0 8.1 1.6 4 9.1 10.1 9.1 3.6 7.0 14.5 CreditScorea 2.9 1.1 2.0 3.0 4.0 3.7 1.5 3.0 4.0 5.0 NetWorth(inmill.) 0.39 1.3 0.05 0.12 0.34 0.72 2.1 0.11 0.27 0.62 FirmobservationsweightedwithSSBFweights. a Thecreditscorein1998goesfrom1(lowest)to5(highest)andin2003from1(lowest)to6(highest). 32

Table 2: Net Worth and Equity Share of Assets by Firm Age all firms received credit all firms received credit Number NetWorth Number NetWorth Number Equity Number Equity ofobs. (inmill.) ofobs. (inmill.) ofobs. Share ofobs. Sharea average average average average (median) (median) (median) (median) Age (1) (2) (3) (4) (5) (6) (7) (8) <3 573 0.245 113 0.222 574 63.7% 113 61.8% (0.092) (0.115) (86.6%) (86.5%) 3-6 1507 0.303 384 0.437 1508 60.1% 385 64.7% (0.106) (0.140) (76.4%) (81.4%) 7-15 2522 0.361 761 0.544 2525 65.7% 761 63.8% (0.151) (0.225) (85.0%) (80.5%) 16-25 1784 0.508 630 0.725 1785 70.7% 630 69.3% (0.225) (0.316) (85.4%) (84.3%) >25 1404 0.801 516 1.240 1406 73.8% 517 76.0% (0.275) (0.443) (92.0%) (92.1%) Total 7790 0.432 2404 0.645 7798 66.6% 2406 67.5% (0.159) (0.230) (84.7%) (84.4%) Source: SSBF1998and2003. Fortheconstructionofnetworth,seetext. a Equitysharebeforemostrecentcreditwastakenout. 33

Table 3: Distribution of Fitted Parameters 1998 Sample mean med std min 25% 75% max all obs. µ 1.08 0.29 2.62 -13.72 0.25 0.32 23.7 σ 0.54 0.59 0.25 0.01 0.39 0.67 1.21 Rk 1.04 1.02 0.05 0.90 1.01 1.03 1.40 µ ∈ [0,1] µ 0.28 0.28 0.09 0.01 0.24 0.31 0.96 σ 0.57 0.63 0.25 0.01 0.48 0.72 1.21 Rk 1.02 1.02 0.01 1.01 1.01 1.02 1.10 2003 Sample mean med std min 25% 75% max all obs. µ 1.60 0.30 3.35 -1.29 0.26 0.79 27.90 σ 0.54 0.58 0.22 0.01 0.40 0.70 1.32 Rk 1.04 1.01 0.07 0.97 1.00 1.06 1.80 µ ∈ [0,1] µ 0.28 0.28 0.10 0.01 0.25 0.31 0.99 σ 0.57 0.64 0.24 0.01 0.47 0.72 1.18 Rk 1.01 1.01 0.01 1.00 1.00 1.01 1.11 Note: Thefull1998sampleconsistsof599observations. Thetruncatedsamplehas485observations. Thefull2003sample consistsof1413observations. Thetruncatedsamplehas1079observations. Foradiscussion,seetext. 34

Table 4: Summary Statistics - Bank Variables 1998 2003 mean std 25% 50% 75% mean std 25% 50% 75% InterestCosttoLiabilityRatioa 0.035 0.012 0.031 0.035 0.038 0.013 0.005 0.010 0.012 0.017 OtherCosttoAssetRatioa 0.034 0.023 0.027 0.031 0.037 0.029 0.010 0.024 0.028 0.032 ReturnonEquitya 0.140 0.064 0.107 0.143 0.174 0.135 0.065 0.099 0.138 0.171 AdequacyRatio 0.094 0.028 0.078 0.088 0.100 0.090 0.027 0.075 0.085 0.096 LoanShare 0.641 0.123 0.579 0.660 0.727 0.638 0.145 0.551 0.665 0.739 Assetsinbn. 54.8 113.0 0.2 5.2 57.5 108.0 200.0 0.5 8.9 75.4 Bank observation weighted by frequency in the credit contract sample. Number of observation is 611 in the 1998 sample and 1441 in the 2003 sample. The composition of bank types in the two samples is similar. In 1998, 4.9 percent of the financial institutions small business owners banked with were credit unions (2003: 3.1 percent), 4.4 percent were thrifts (2003: 4.8percent),and90.7percentwerecommericialbanks(2003: 92.1percent). a Costsandratesofreturnareannualized. 35

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 Leverage )w( ytivitcudorP ffotuC Bank zero profit, Baseline Indifference curve, Baseline Bank zero profit, Double Risk Indifference curve, Double Risk Bank zero profit, Double RoR Indifference curve, Double RoR Equilibrium, Baseline Equilibrium, Double Risk Equilibrium, Double RoR Double Risk: Z/(1+r)=1.0224 Double Rate of Return: Leverage=1.12 Z/(1+r)=1.0329 Baseline: Leverage=1.77 Z/(1+r)=1.0109 Leverage= 1.5 Figure 1: Comparative Statics 36

Table 5: Regression Results: Interest Rate 1998 2003 OLS OLS Selectionb OLS OLS Selectionb (1) (2) (3) (4) (5) (6) Firm Variables ∗ Share of Firm’s 0.05 0.28 0.36 0.64 -0.37 -0.17 Liquid Assets (0.23) (0.27) (0.37) (0.26) (0.30) (0.31) † † ∗∗ ∗∗ ∗ Credit Score -0.13 -0.17 -0.18 -0.24 -0.16 -0.19 (0.08) (0.09) (0.11) (0.05) (0.05) (0.08) ∗∗ ∗∗ Log Number -0.13 -0.15 -0.25 -0.25 of Employees (0.10) (0.11) (0.08) (0.10) ∗∗ ∗∗ Log Firm Age -0.10 -0.09 -0.34 -0.42 (0.12) (0.16) (0.10) (0.15) ∗∗ ∗∗ Log Length of -0.05 -0.05 0.27 0.32 Relationship (0.05) (0.06) (0.09) (0.12) † ∗ ∗ ∗∗ Net Worth -0.13 -0.15 -0.09 -0.09 (inmill.) (0.08) (0.07) (0.04) (0.03) Bank Variables ∗ ∗ † ∗∗ Interest Cost -14.52 -20.39 -21.55 51.05 21.49 28.81 to Liabilities Ratio (7.37) (8.88) (12.35) (13.44) (18.69) (28.86) Other Cost 2.85 3.96 3.67 3.37 -8.67 -7.67 to Asset Ratio (4.14) (4.62) (5.12) (7.20) (8.12) (11.16) Return on Equity -1.01 -0.31 1.68 0.28 0.56 0.95 (1.44) (1.75) (2.50) (1.18) (1.38) (1.95) ∗ Adequacy Ratio 2.07 0.17 2.626 5.13 1.28 2.09 (3.83) (4.16) (5.11) (2.52) (2.78) (3.11) † † Log Bank Assets -0.06 -0.08 0.01 0.01 (0.04) (0.05) (0.03) (0.04) Loan Share of Assets -0.42 -0.32 -0.71 -0.66 (0.84) (1.15) (0.58) (0.85) Additional Controlsa no yes yes no yes yes N 493 493 588 1335 1263 1324 Adjusted R2 0.01 0.10 - 0.03 0.19 - Significancelevels: ∗∗1%,∗5%,and†10%. AllobservationsareweightedwiththesampleweightsintheSSBF. a Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue,shareofloans>90days pastdueanddummiesforbankstype(commericalbank,federalsavingsbank,creditunion),marketconcentration,typeof cooperation,judgementsagainstthefirm,previousprivateandfirmbankruptcy,region,sector,quarterinwhichtheloans wastakenout,urban,femalefirmowner,minorityfirmowner,andeducationoffirmowner. b Heckman selection model. Only 81 firm owners report credit denial hence the model contains only a limited set of variables. The selection equation includes the bank variables, log employment, log firm age, credit score, log length of relationship with bank, a dummy for minority firm owner and quarter in which the loans was taken out, and risk-based capital ratio. Note that the risk based capital ratio is not available for credit unions (22 obs.). Using the risk-based net worthtoproxyfortherisk-basedcapitalratiodoesnotchangetheresultsreportedincolumns(3)and(6)significantly. 37

Table 6: Regression Results Credit Amount 1998 Sample 2003 Sample OLS OLS Selectiona OLS OLS Selectiona ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ Net Worth 0.320 0.298 0.318 0.360 0.334 0.318 (0.084) (0.083) (0.077) (0.055) (0.057) (0.055) ∗ ∗∗ Ln Number 0.066 0.139 of Employees (0.028) (0.032) controls no no no no no no N 711 711 959 1693 1693 1856 R2 0.29 0.30 - 0.27 0.28 - Significancelevels: : †: 10% ∗: 5% ∗∗: 1% a Heckman selection model using information on whether the firm’s credit application was accepted. The probability of receiving credit is estimated using net worth, ln(emploment), ln(length of relationship with the lender), ln(firmage), credit score, and dummy variables for industry, region, gender, minority status, organizational form of the firm, previous bankruptciesandjudgementsagainstthefirm. 38

Table 7: Regression Results: Credit Amount 1998 Sample 2003 Sample OLS OLS Selectionc OLS OLS Selectionc (1) (2) (3) (4) (5) (6) Firm Variables ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ ∗∗ Net Worth 0.320 0.259 0.259 0.360 0.302 0.302 (inmill.) (0.084) (0.024) (0.077) (0.055) (0.018) (0.081) ∗ ∗∗ ∗∗ ∗∗ Share of Firm’s 0.168 0.168 -0.509 -0.491 Liquid Assets (0.085) (0.062) (0.155) (0.154) ∗ † ∗∗ ∗ Log Number 0.068 0.062 0.122 0.120 of Employees (0.031) (0.032) (0.040) (0.051) Log Firm Age -0.011 -0.011 -0.034 -0.055 (0.038) (0.028) (0.051) (0.045) Log Length of -0.027 -0.033 0.001 0.010 Relationship (0.017) (0.021) (0.041) (0.033) ∗∗ ∗ Credit Score 0.070 0.073 -0.023 -0.030 (0.027) (0.034) (0.028) (0.021) Bank Variables Interest Cost -0.222 2.027 -0.545 2.032 to Liabilities Ratio (2.858) (1.914) (9.438) (5.894) Other Cost -1.865 -0.710 -3.556 -4.275 to Asset Ratio (1.436) (0.599) (4.212) (3.472) Return on Equity -0.748 -1.227 0.462 0.489 (0.552) (0.848) (0.702) (0.504) Adequacy Ratio -0.937 -1.412 0.569 0.293 (1.300) (0.932) (1.413) (0.845) † Log Bank Assets 0.010 0.007 0.022 0.022 (0.011) (0.007) (0.016) (0.014) Loan Share of Assets 0.054 0.109 0.168 0.091 (0.272) (0.198) (0.298) (0.231) Additional Controlsa,b no yes yes no yes yes N 711 493 588 1693 1263 1324 Adjusted R2 0.23 0.26 - 0.27 0.25 - Significancelevels: ∗∗1%,∗5%,and†10%. a Additional controls included in columns (2) and (5) are share of nonaccrual loans, share of loans 30-90 days past due, share of loans > 90 days past due and dummies for banks type (commerical bank, federal savings bank, credit union), market concentration, type of cooperation, judgements against the firm, previous private and firm bankruptcy, region, sector, quarter in which the loans was taken out, urban, female firm owner, minority firm owner, and education of firm owner. bAdditionalcontrolsincludedincolumns(3)and(6)aredummiesforbankstype,marketconcentration,typeofcooperation, region,sector,quarterinwhichtheloanswastakenout,andurban. c Heckmanselectionmodel. Theselectionequationincludesthebankvariables,logemployment,logfirmage,creditscore, loglengthofrelationshipwithbank,adummyforminorityfirmowner,networth,quarterlydummies,sectoraldummies, and risk-based capital ratio. Note that the risk based capital ratio is not available for credit unions (22 obs.). Using the risk-basednetworthtoproxyfortherisk-basedcapitalratiodoesnotchangetheresultsreportedincolumns(3)and(6) significantly. 39

Table 8: Firm Characteristics by Bank Size Small Banks Large Banks Difference mean std mean std 1998 Sample Number of Employees 14.20 32.65 14.03 28.46 -0.17 Firm Age 12.71 8.83 11.35 10.35 -1.34 ∗∗ Length of Lending Relationship 8.22 8.55 5.43 7.15 -2.80 Credit Scorea 3.10 1.10 2.91 1.04 -0.19 † Share in Manufacturing 0.10 0.30 0.08 0.27 -0.03 Net Worth (in mill.) 0.41 1.09 0.53 1.60 0.12 N 379 144 2003 Sample Number of Employees 15.37 30.89 17.07 36.35 1.70 Firm Age 15.76 11.96 15.96 10.08 0.20 Length of Lending Relationship 10.18 9.25 11.03 9.18 0.85 Credit Scorea 3.83 1.43 3.77 1.47 -0.05 ∗∗ Share in Manufacturing 0.09 0.28 0.15 0.35 0.06 ∗ Net Worth (in mill.) 0.70 2.02 0.95 2.43 0.15 N 928 412 Significancelevels: ∗∗1%,∗5%,and†10%. Firmsthatreceivedcreditonly. a Thecreditscorein1998goesfrom1(lowest)to5(highest)andin2003from1(lowest)to6(highest). 40

Table 9: Regression Results: Credit Amount by Bank Size 1998 Sample 2003 Sample Small Banks Large Banks Small Banks Large Banks <$50bn. assets ≥$50bn. assets <$50bn. assets ≥$50bn. assets (1) (2) (3) (4) Firm Variables ∗∗ ∗ ∗∗ ∗∗ Net Worth 0.365 0.147 0.296 0.298 (inmill.) (0.028) (0.066) (0.018) (0.045) ∗∗ Share of Firm’s 0.158 0.216 -0.531 -0.572 Liquid Assets (0.088) (0.328) (0.148) (0.478) † ∗ ∗ † Log Number 0.061 0.290 0.096 0.210 of Employees (0.033) (0.118) (0.040) (0.112) † Log Firm Age 0.033 -0.314 -0.061 0.053 (0.041) (0.183) (0.048) (0.168) ∗∗ Log Length of -0.059 0.007 0.020 0.090 Relationship (0.019) (0.064) (0.044) (0.143) ∗ Credit Score 0.063 0.044 -0.036 -0.010 (0.028) (0.129) (0.029) (0.080) Bank Variables † Interest Cost 9.320 -31.337 2.797 -45.694 to Liabilities Ratio (5.452) (26.754) (8.861) (58.453) Other Cost -1.620 -7.234 -3.665 0.223 to Asset Ratio (1.322) (28.268) (3.535) (15.570) Return on Equity -0.873 -1.144 0.744 -3.082 (0.527) (6.363) (0.633) (2.875) Adequacy Ratio -0.370 5.447 0.805 -7.406 (1.296 (13.071) (1.284) (6.121) ∗ Log Bank Assets 0.014 0.105 0.040 -0.197 (0.016) (0.381) (0.020) (0.187) Loan Share of Assets -0.144 -2.169 0.219 -0.763 (0.316) (2.086) (0.315) (1.602) Additional Controlsa yes yes yes yes N 379 114 877 386 Adjusted R2 0.42 -0.28 0.32 0.12 Significancelevels: ∗∗1%,∗5%,and†10%. a Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue,shareofloans>90days pastdueanddummiesforbankstype(commericalbank,federalsavingsbank,creditunion),marketconcentration,typeof cooperation,judgementsagainstthefirm,previousprivateandfirmbankruptcy,region,sector,quarterinwhichtheloans wastakenout,urban,femalefirmowner,minorityfirmowner,andeducationoffirmowner. 41

Table 10: Regression Results: Interest Rate by Bank Size - 1998 Sample Small Banks <$50 bn. assets Large Banks ≥$50 bn. assets OLS OLS Selectionb OLS OLS Selectionb (1) (2) (3) (4) (5) (6) Firm Variables Share of Firm’s -0.13 0.26 0.34 0.34 0.61 0.62 Liquid Assets (0.25) (0.28) (0.37) (0.57) (0.81) (0.93) ∗ † Credit Score -0.20 -0.17 -0.17 -0.03 -0.18 -0.18 (0.09) (0.09) (0.11) (0.24) (0.33) (0.25) ∗ ∗ Log Number -0.24 -0.22 -0.06 -0.07 of Employees (0.10) (0.11) (0.29) (0.30) ∗ Log Firm Age -0.13 -0.08 0.68 0.70 (0.13) (0.19) (0.41) (0.31) † Log Length of -0.05 -0.05 -0.26 -0.26 Relationship (0.06) (0.07) (0.16) (0.13) Net Worth -0.09 -0.11 -0.15 -0.15 (inmill.) (0.09) (0.09) (0.17) (0.09) Bank Variables ∗∗ ∗∗ ∗∗ Interest Cost -49.15 -72.90 -80.40 2.49 -40.12 -47.38 to Liabilities Ratio (14.38) (17.10) (23.39) (12.25) (63.39) (43.33) Other Cost 3.13 6.11 6.08 -10.78 -58.17 -60.98 to Asset Ratio (3.92) (4.29) (5.11) (46.15) (66.99) (60.11) Return on Equity -0.96 0.34 2.73 -0.43 -9.38 -9.27 (1.39) (1.68) (2.32) (9.35) (15.35) (10.75) Adequacy Ratio 0.48 -2.24 1.34 20.96 3.02 3.28 (3.70) (4.10) (5.41) (20.18) (27.66) (18.79) Log Bank Assets -0.07 -0.08 -0.48 -0.71 (0.05) (0.07) (0.68) (0.68) ∗ Loan Share of Assets -0.43 0.03 7.13 5.27 (0.96) (0.99) (5.17) (3.27) Additional Controlsa no yes yes no yes yes N 379 379 433 114 114 155 Adjusted R2 0.03 0.18 - -0.02 0.25 - Significancelevels: ∗∗1%,∗5%,and†10%. a Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue,shareofloans>90days pastdueanddummiesforbankstype(commericalbank,federalsavingsbank,creditunion),marketconcentration,typeof cooperation,judgementsagainstthefirm,previousprivateandfirmbankruptcy,region,sector,quarterinwhichtheloans wastakenout,urban,femalefirmowner,minorityfirmowner,andeducationoffirmowner. b Heckmanselectionmodel. Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue, share of loans > 90 days past due and dummies for banks type (commerical bank, federal savings bank, credit union), market concentration, type of cooperation, urban, region, sector, and quarter in which the loans was taken out. The selection equation includes the bank variables, log employment, log firm age, credit score, log length of relationship with bank, a dummy for minority firm owner, net worth, sectoral dummies, and risk-based capital ratio. Note that the risk basedcapitalratioisnotavailableforcreditunions. 42

Table 11: Regression Results: Interest Rate by Bank Size - 2003 Sample Small Banks <$50 bn. assets Large Banks ≥$50 bn. assets OLS OLS Selectionb OLS OLS Selectionb (1) (2) (3) (4) (5) (6) Firm Variables ∗∗ Share of Firm’s 0.37 -0.47 -0.35 1.49 -0.08 -0.55 Liquid Assets (0.28) (0.32) (0.45) (0.57) (0.60) (0.68) ∗∗ ∗∗ ∗ Credit Score -0.28 -0.20 -0.18 -0.09 -0.07 -0.17 (0.05) (0.06) (0.09) (0.11) (0.10) (0.16) ∗∗ ∗∗ ∗∗ Log Number -0.25 -0.19 -0.48 -0.63 of Employees (0.09) (0.14) (0.14) (0.15) ∗∗ ∗∗ Log Firm Age -0.20 -0.17 -0.59 -0.55 (0.10) (0.14) (0.21) (0.34) ∗ ∗∗ ∗∗ Log Length of 0.11 0.26 0.63 0.56 Relationship (0.10) (0.12) (0.18) (0.17) ∗ ∗ Net Worth -0.08 -0.07 -0.03 -0.04 (inmill.) (0.04) (0.04) (0.05) (0.03) Bank Variables ∗∗ ∗∗ ∗∗ Interest Cost 68.47 20.60 23.24 34.07 235.28 256.94 to Liabilities Ratio (13.97) (19.82) (32.49) (37.56) (71.95) (82.73) ∗ † Other Cost 9.60 -5.14 -9.70 -27.51 -68.15 -59.92 to Asset Ratio (7.31) (7.97) (7.19) (20.54) (28.06) (33.90) ∗ Return on Equity 1.61 1.95 2.61 -7.86 1.31 0.06 (1.19) (1.39) (1.79) (3.46) (3.73) (4.10) ∗∗ † ∗∗ Adequacy Ratio 3.86 -3.57 -0.72 26.94 14.55 19.80 (2.57) (2.82) (3.14) (6.99) (7.93) (6.96) ∗∗ ∗∗ ∗∗ ∗∗ Log Bank Assets -0.17 -0.19 0.74 0.85 (0.04) (0.06) (0.24) (0.21) ∗ ∗ Loan Share of Assets -1.71 -2.06 2.45 2.33 (0.69) (0.83) (2.06) (2.36) Additional Controlsa no yes yes no yes yes N 924 877 893 411 386 431 Adjusted R2 0.05 0.21 - 0.17 0.51 - Significancelevels: ∗∗1%,∗5%,and†10%. a Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue,shareofloans>90days pastdueanddummiesforbankstype(commericalbank,federalsavingsbank,creditunion),marketconcentration,typeof cooperation,judgementsagainstthefirm,previousprivateandfirmbankruptcy,region,sector,quarterinwhichtheloans wastakenout,urban,femalefirmowner,minorityfirmowner,andeducationoffirmowner. b Heckmanselectionmodel. Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue, share of loans > 90 days past due and dummies for banks type (commerical bank, federal savings bank, credit union), market concentration, type of cooperation, urban, region, sector, and quarter in which the loans was taken out. The selection equation includes the bank variables, log employment, log firm age, credit score, log length of relationship with bank, a dummy for minority firm owner, net worth, sectoral dummies, and risk-based capital ratio. Note that the risk basedcapitalratioisnotavailableforcreditunions(22obs.). Usingtherisk-basednetworthtoproxyfortherisk-based capitalratiodoesnotchangetheresultsreportedincolumns(3)and(6)significantly. 43

Table 12: Dependent Variable: Credit Amount 1998 Sample 2003 Sample OLS OLS Selectiona OLS OLS Selectiona Net Worthb 0.225 ∗ 0.316 ∗∗ 0.224 ∗ 0.325 ∗∗ 0.264 ∗ 0.284 ∗ (0.099) (0.118) (0.099) (0.121) (0.119) (0.117) ∗∗ ∗∗ Ln Number 0.136 0.306 of Employees (0.040) (0.043) controlsc no yes no no yes no N 542 542 876 1693 1652 1829 R2 0.16 0.16 - 0.07 0.12 - Significance levels: : † : 10% ∗ : 5% ∗∗ : 1% a Heckman selection model using information on whether the firm’s credit application was accepted. The probability of receiving credit is estimated using net worth, ln(emploment), ln(length of relationship with the lender), ln(firmage), credit score, and dummy variables for industry, region, gender, minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. b Net worth is defined as the sum of land, depreciable assets, and the net value of the private residence minus loans. cAdditionalcontrolsincludeln(firmage),creditscore,anddummyvariablesforindustry,region,gender,minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. Table 13: Dependent Variable: Credit Amount 1998 Sample 2003 Sample OLS OLS Selectiona OLS OLS Selectiona Net Worthb 0.179 † 0.325 ∗ 0.196 † 0.227 ∗ 0.168 † 0.177 † (0.117) (0.138) (0.116) (0.108) (0.119) (0.097) ∗∗ ∗∗ Ln Number 0.154 0.335 of Employees (0.039) (0.041) controlsc no yes no no yes no N 711 542 876 1693 1652 1829 R2 0.02 0.13 - 0.07 0.08 - Significance levels: : † : 10% ∗ : 5% ∗∗ : 1% a Heckman selection model using information on whether the firm’s credit application was accepted. The probability of receiving credit is estimated using net worth, ln(emploment), ln(length of relationship with the lender), ln(firmage), credit score, and dummy variables for industry, region, gender, minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. b Networthisdefinedasthesumofland,depreciableassets,andthenetvalueoftheprivateresidenceminusall liabilities. cAdditionalcontrolsincludeln(firmage),creditscore,anddummyvariablesforindustry,region,gender,minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. 44

Table 14: Dependent Variable: Credit Amount 1998 Sample 2003 Sample OLS OLS Selectiona OLS OLS Selectiona Net Worthb 0.373 ∗∗ 0.341 ∗∗ 0.372 ∗∗ 0.395 ∗∗ 0.365 ∗∗ 0.392 ∗∗ (0.078) (0.109) (0.078) (0.056) (0.056) (0.056) ∗∗ Ln Number 0.052 0.176 of Employees (0.034) (0.033) controlsc no yes no no yes no N 711 542 876 1693 1652 1829 R2 0.32 0.35 - 0.27 0.28 - Significance levels: : † : 10% ∗ : 5% ∗∗ : 1% a Heckman selection model using information on whether the firm’s credit application was accepted. The probability of receiving credit is estimated using net worth, ln(emploment), ln(length of relationship with the lender), ln(firmage), credit score, and dummy variables for industry, region, gender, minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. b Net worth is defined as the all firm assets minus all firm liabilities. cAdditionalcontrolsincludeln(firmage),creditscore,anddummyvariablesforindustry,region,gender,minority status, organizational form of the firm, previous bankruptcies and judgements against the firm. 45

Table 15: Dependent Variable: Interest Rate - Alternative Measures 2003 Sample Alternative Cost Measure Three Year Averages OLS OLS Selectionb OLS OLS Selectionb (1) (2) (3) (4) (5) (6) Firm Variables ∗ ∗∗ Share of Firm’s 0.66 -0.38 -0.19 0.69 -0.38 -0.14 Liquid Assets (0.26) (0.29) (0.37) (0.26) (0.30) (0.38) ∗∗ ∗∗ ∗ ∗∗ ∗∗ ∗ Credit Score -0.24 -0.15 -0.18 -0.23 -0.15 -0.16 (0.05) (0.05) (0.07) (0.05) (0.05) (0.07) ∗∗ ∗ ∗∗ ∗∗ Log Number -0.24 -0.23 -0.25 -0.29 of Employees (0.08) (0.09) (0.08) (0.10) ∗∗ ∗∗ ∗∗ ∗∗ Log Firm Age -0.24 -0.44 -0.33 -0.40 (0.08) (0.14) (0.10) (0.14) ∗∗ ∗∗ ∗∗ Log Length of 0.30 0.34 0.27 0.27 Relationship (0.09) (0.12) (0.09) (0.35) ∗∗ ∗∗ ∗ ∗∗ Net Worth -0.09 -0.09 -0.09 -0.09 (inmill.) (0.03) (0.03) (0.03) (0.03) Bank Variables ∗∗ † Interest Cost 51.78 22.36 30.33 25.07 33.29 32.08 to Liabilities Ratio (13.46) (17.54) (27.94) (15.31) (18.22) (17.10) ∗∗ ∗ Other Cost -5.51 -40.77 -37.61 3.41 -13.28 -10.65 to Asset Ratio (13.27) (13.77) (17.49) (7.79) (8.41) (16.88) Return on Equity 0.32 0.14 0.58 -0.19 -0.28 -0.33 (1.18) (1.31) (1.83) (1.33) (1.43) (2.69) ∗ ∗∗ Adequacy Ratio 5.36 -0.50 0.44 7.48 -1.13 -1.75 (2.51) (2.63) (3.21) (2.90) (3.24) (4.03) Log Bank Assets -0.03 -0.03 0.01 -0.00 (0.03) (0.04) (0.03) (0.04) ∗ Loan Share of Assets -1.19 -1.11 -0.36 -0.36 (0.55) (0.80) (0.61) (0.91) Additional Controlsa no yes yes no yes yes N 1335 1263 1324 1338 1266 1338 Adjusted R2 0.03 0.25 - 0.02 0.21 - Significancelevels: ∗∗1%,∗5%,and†10%. a Additionalcontrolsincludedareshareofnonaccrualloans,shareofloans30-90dayspastdue,shareofloans>90days pastdueanddummiesforbankstype(commericalbank,federalsavingsbank,creditunion),marketconcentration,typeof cooperation,judgementsagainstthefirm,previousprivateandfirmbankruptcy,region,sector,quarterinwhichtheloans wastakenout,urban,femalefirmowner,minorityfirmowner,andeducationoffirmowner. b Heckman selection model. Only 81 firm owners report credit denial hence the model contains only a limited set of variables. The selection equation includes the bank variables, log employment, log firm age, credit score, log length of relationship with bank, a dummy for minority firm owner and quarter in which the loans was taken out, and risk-based capital ratio. Note that the risk based capital ratio is not available for credit unions (23 obs.). Using the risk-based net worthtoproxyfortherisk-basedcapitalratiodoesnotchangetheresultsreportedincolumns(3)and(6)significantly. 46

1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Agency Cost Parameter (mu) )amgis( kcohS citarcnysoidI fo noitaiveD dradnatS 1998 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Agency Cost Parameter (mu) )amgis( kcohS citarcnysoidI fo noitaiveD dradnatS 2003 Figure 2: CSV Implied Agency Cost and Risk Parameters 47

Figure 3: MH Implied Interest Rates and Actual Interest Rates 48

Figure 4: MH and Actual Interest Rates, Quarterly Averages. 49

Figure 5: MH and Actual Interest Rates, Alternative Cost Measure - 2003 Sample 50