Document Sample
301 Powered By Docstoc
					                         What’s Driving Small Borrower-Lender Distance?

                                             Robert DeYoung*
                                    Federal Deposit Insurance Corporation

                                                W. Scott Frame*
                                        Federal Reserve Bank of Atlanta

                                               Dennis Glennon*
                                   Office of the Comptroller of the Currency

                                                   Peter Nigro
                                                Bryant University

Abstract: We examine the drivers of increased lending distance using a large random sample of small
business loans originated by U.S. commercial banks between 1984 and 2001 under the Small Business
Administration’s (SBA) flagship 7(a) guaranteed loan program. We provide support for the findings of
the extant literature (e.g., Petersen and Rajan, 2002) that small business borrower-lender distance
increased during the 1980s and 1990s. We extend this literature by demonstrating that this increase has
accelerated into the present decade, and that the largest and most recent increases in borrower-lender
distance are associated with banks that use small business credit scoring techniques. In addition, we find
that young firms, relatively small firms, firms with disproportionately large credit needs, and urban firms
tend to be further from their lenders, while large banks, banks with extensive branching networks, and
banks with external accreditation as high-quality lenders tend to lend to more distant small businesses.
We find weak evidence linking borrower-lender distance with high local banking market concentration,
which implies that monopoly banking profits are attracting out-of-market lenders. Finally, and perhaps
surprisingly, we find a negative relationship between borrower-lender distance and the size of government
loan guarantees: this result suggests that small borrowers may be using the promise of large loan
guarantees to secure credit from a nearby bank. After accounting for all these phenomena, we find a
strong residual trend in increasing borrower-lender distance, suggesting that general improvements in
technology and regulation have contributed to borrower-lender distance. These findings are preliminary,
and may change as we explore different regression specifications and statistical techniques.

JEL Classification Numbers: G21

Keywords: small business loans, borrower-lender distance, credit scoring.

* The views expressed here are those of the authors and do not necessarily reflect the views of: the Federal Deposit
Insurance Corporation, the Federal Reserve Bank of Atlanta, the Federal Reserve System, the Office of the
Comptroller of the Currency, or the U.S. Treasury Department. The authors are especially grateful to Dan McMillen
for sharing his time and geographic mapping.
                        What’s Driving Small Borrower-Lender Distance?

1. Introduction

        The United States has more credit-granting institutions per capita than any other developed

economy. 1 Most of these lending institutions are small: for example, of the approximately 7,500

commercial banks operating in the U.S., nearly half (46 percent) hold less than $100 million in total

assets. The standard explanation for this high degree of industry fragmentation combines geography,

economics, and politics. Economic activity in the U.S. occurs in hundreds of noncontiguous metropolitan

areas often separated by large geographic spaces; a disproportionate amount of U.S. economic activity is

generated by small and medium size enterprises; and American households and small businesspeople

have long been suspicious of concentrated financial power. Given these initial structural conditions—as

well as the state and federal restrictions on geographic expansion to which these conditions gave rise—

locally focused lenders have traditionally maintained a competitive advantage over their larger brethren in

providing relationship -based finance to small local firms (e.g., DeYoung, Hunter, and Udell 2004; Berger,

Miller, Petersen, Rajan, and Stein 2005). But there is growing evidence that this tight geographic bond

between lenders and their small business borrowers is deteriorating—and with it, the possibility that the

core competitive advantages of thousands of small lenders may be eroding.

        Petersen and Rajan (PR, 2002) were the first to document an increase in the distance between

U.S. small business borrowers and their lenders, finding a gradual secular increase in borrower-lender

distances between 1973 and 1993. 2 Using data on existing lender-small business borrower relationships in

1993, the authors estimated that the median lender-borrower distance grew from just 2 miles for

relationships that began during the 1970s, to 5 miles for relationships that began during the early 1990s.

  For example, see Berger, DeYoung, Genay, and Udell (2000), page 112, Table A-2. As of 1997, there were 11,997
persons per credit-granting institution in the U.S., compared to an unweighted average of 63,594 persons per
institution in the other G-10 nations.
  Petersen and Rajan (2002) made these backward-looking inferences based on data contained in the 1993 National
Survey of Small Business Finance (NSSBF). Elliehausen and Wolken (1990) and Kwast, Starr-McCluer and
Wolken (1997) also report (unconditional) distance estimates for various individual loan products using the 1988
and 1993 NSSBF’s respectively.

Subsequently, studies using different data sources and research methods confirmed that distances were

increasing between small U.S. firms and their bank lenders (Hannan 2003; DeYoung, Glennon, and Nigro

2006). Figure 1 compares the median borrower-lender distances reported by DeYoung, Glennon, and

Nigro (DGN 2006) for Small Business Administration (SBA) loans to extremely credit-constrained firms,

to the borrower-lender distances constructed by PR (2002) using data from the National Survey of Small

Business Finance (NSSBF). The two times series are quite sim ilar in the years (1984-1993) for which

these two studies overlap—relatively short distances that gradually increase over time—lending support

for PR’s backward-looking methodology.

                                          [Insert Figure 1 here]

        [Note: Although the median distances reported by PR (2002) and DGN (2006) are similar, there

is a systematic and stable gap of 1½ to 2 miles between the two time series. This gap may indicate a

survivorship bias in PR’s backward-looking methodology, i.e., if small businesses located closer to their

lenders were easier to evaluate and monitor, these lending relationships would be more likely to survive

until 1993. The gap may also be caused by the partial SBA loan guarantee, which reduces (but does not

eliminate) lender loss exposures, possibly inducing banks to lend to firms at longer distances. Regardless,

the gap appears to be stable over the 1984-1993 time period, which implies that the environmental

changes that were driving increased borrower-lender distances in the 1980s and early-1990s manifested

themselves similarly for both subsidized and non-subsidized small business loans.]

        With increasing borrower-lender distances established as a stylized fact, researchers began

linking lending distance to a variety of phenomena, such as loan pricing (e.g., Degryse and Ongena 2005;

Agarwal and Hauswald 2006) and loan performance (e.g., DGN 2006). But there remains little hard

evidence on the fundamental forces driving these lending distances, and whether these forces are likely to

continue pushing banks and their small business borrowers further apart. Understanding the determinants

of borrower-lender distance may hold important keys for projecting the future structure, competitiveness,

and risk profiles of the banking industry. All else equal, increased borrower-lender distances imply

greater credit availability for local businesses, because it gives these small firms access to lenders from

outside the local market. But this implication depends crucially on the drivers of increased lending

distance: if industry consolidation is the catalyst, then lending distance is increasing because there are

fewer banks located in any given geographic area, and is symptomatic of reductions in competition; if

technological advance is the catalyst, then lending distance is increasing due to entry from outside the

local geographic market, and is symptomatic of increased competition. Furthermore, technology-driven

increases in lending distance may not benefit all borrowers equally—for example, a switch to lending

technologies that exploit hard information (like credit scoring) is more likely to benefit distant small

businesses whose creditworthiness can be captured and communicated in quantifiable terms, while small

businesses that are less informationally transparent may end up less well-served and increasing reliant on

shrinking number of local lenders. 3

        For their part, PR (2002) concluded that the modest increases in borrower-lender distances

between 1973 and 1993 were largely due to lenders’ greater use of information technology, including but

not limited to credit scoring methods, and were not driven by other factors such as banking industry

consolidation or changes in the distribution of borrower location over time. They drew these inferences

based on the two following facts: borrower-lender distance continued to increase over time in their

regressions even after conditioning on other factors, and bank employment as a share of bank lending

declined between 1973 and 1993, consistent with substitution of technology inputs for labor inputs.

While we find these inferences to be very plausible—especially given that the bulk of banking industry

consolidation has occurred since 1993—we do not believe that they shed much light on the impact of

credit scoring on small business lending distance. For example, Akhavein, Frame, and White (2005)

  Suggestive evidence is offered by DGN (2006), whose findings imply that credit scoring is a relatively more
efficient lending technology for more distant borrowers and lenders; by Frame, Padhi, and Woosley (2004), who
find that credit scoring banks make relatively more loans outside their local markets, and relatively fewer loans
inside their local markets, than do non-scoring banks; and by Brevoort and Hannan (2004), who find that borrower-
lender distances within nine U.S. metropolitan areas actually decreased between 1997 and 2001.

report that bank lenders did not adopt small business credit scoring models until 1991, and by year-end

1993 only a small handful of banks were using these models.

           Additional insight is provided in Figure 1, which also displays post-1993 borrower-lender

distance data from DGN (2006). On average, these data show a continuation of the gradual secular

increase in borrower-lender distance found in both PR (2002) and DGN (2006) prior to 1993. But the

figure also shows clearly that different lending technologies have different relationships with lending

distances. The increase in borrower-lender distance accelerates markedly for SBA loans issued by

banking companies that used credit-scoring models to evaluate small business loan applications (as

identified by Frame, Srininvasan, and Woosley 2001), but continues its gradual, secular pace for other

banking companies. The implication is that a specific technology—credit scoring—is responsible for a

large share of the increase in borrower-lender distance for small business loans, while the more gradual

remainder of the increase in lending distance is attributable to a combination of general technological

change (e.g., improved communications and information), ongoing industry consolidation, and other


           In this paper, we examine the drivers of increased lending distance using a large random sample

of small business loans originated between 1984 and 2001 and guaranteed by the U.S. Small Business

Administration (SBA). As we describe below, this database has many advantages for empirical work.

However, a natural concern is whether results derived from subsidized loan data may not generalize to

non-subsidized lending. We argue that any bias introduced by loan guarantees will be one of degree only,

because SBA loan subsidies only partially shield lenders from default risk; subsidies do not alter and are

independent from real distance-related expenses; and subsidies merely substitute for other potential

(though costly) forms of risk hedging. Furthermore, all of our empirical results are derived after

controlling for the magnitude of the SBA loan guarantee, which varies across loans and across time.

           Our multiple regression tests confirm the univariate images in Figure 1; namely, that small

business borrower-lender distance increased steadily during the 1980s and 1990s and has accelerated into

the present decade, and that small business credit scoring has been the key driver of this change in recent

years. Credit-scoring banks reach substantially further to make small business loans than do non-scoring

banks. Our results also suggest that increases in borrower-lender distance over time are substantially

attributable to general trends in innovation, deregulation, and other secular changes not explicit ly

specif ied in our tests.

         On average, young firms, relatively small firms, firms with disproportionately large credit needs,

and firms located in urban markets have more distant lenders, while l rge banks, banks with extensive

branching networks, and banks with external accreditation as high-quality lenders lend to more distant

small businesses. The marginal impact of credit-scoring on borrower-lender distance is greatest for small

banks, where automated lending processes depart from the ethos of locally focused, relationship lending.

Still, we find that small banks pull back their geographic lending footprint during recessions—possibly to

divert credit to their close-by, core relationship borrowers—while large banks do not need to do so. The

marginal impact of scoring also dissipates as loan size increases, consistent with how banks are currently

implementing scoring models.

         We find weak evidence linking borrower-lender distance with high local banking market

concentration, which implies that monopoly banking profits are attracting out-of-market lenders. Finally,

and perhaps surprisingly, we find a negative relationship between borrower-lender distance and the size

of government loan guarantees: this result suggests that small borrowers may be using the promise of

large loan guarantees to secure credit from a nearby bank—an outcome beneficial to both sides of the

transaction. We stress that all of these findings are preliminary, and may change as we explore different

regression specifications and statistical techniques.

         The remainder of the paper is structured as follows. Section II describes our database in detail,

provides some background on the SBA’s flagship 7(a) loan program, and reviews some of the literature

most relevant to our study. Section III presents our preliminary empirical model, and outlines various

hypotheses regarding the drivers of small borrower-lender distance. Section IV analyzes our initial

empirical results. Section V provides some preliminary conclusions.

II. Data sources

        The Small Business Administration’s 7(a) loan program is the U.S. government’s primary policy

tool for increasing small business access to credit. Under this program, the SBA provides loan guarantees

to eligible businesses through qualified financial institutions (mainly, but not exclusively, commercial

banks) that select the firms to receive loans, initiate SBA involvement, underwrite the loans within SBA

program guidelines, and monitor and report back to the SBA the progress of these loans. To qualify for an

SBA loan guarantee, a small business borrower must have been denied access to credit in “normal” loan

markets. The SBA guarantee absorbs only a portion of any loss should the loan default, and the lending

bank is exposed to the balance of the loss. We randomly sampled 32,423 loans to small businesses

originated between 1984 and 2001 by 5,535 U.S. commercial banks under the SBA 7(a) loan program.

Our random sample contains 20 percent of the SBA 7(a) loans made by commercial banks each year

between 1984 and 2001; due to the growth in the program over time, our sample is weighted toward more

recent years. 4 Summary statistics for these data are displayed in Table 1.

                                              [Insert Table 1 here]

        Most previous studies of small business lending in the U.S. have used data from the NSSBF (e.g.,

Elliehausen and Wolken 1990; Kwast, Starr-McCluer and Wolken 1997; Berger and Udell 1998; PR

2002). The NSSBF is a careful, cross-sectional representation of the population of small firms in the U.S.,

but the Survey has some limitations. For example, neither the borrowing firm nor the lending bank are

identified in the publicly accessible data; there is limited information on borrowing firms and loan terms;

the Survey is updated only every five years or so, leaving multi-year gaps in the data; and the borrowing

firms are not tracked over time, and change each time the Survey is conducted. While our random sample

of SBA loans is less representative of the small firm population, it has a number of advantages relative to

  Our long-run, historical sample only includes loans issued by commercial banks. In recent years, approximately 15
to 20 percent of SBA loans have been made by non-banks such as finance companies, thrifts, and credit unions.

the NSSBF. For example, it covers a longer time period and is updated annually; it tracks the performance

of the loans over time; it includes a variety of borrower characteristics and loan terms; and each loan can

be linked to outside databases containing detailed information about the lender, the local market, and

borrower-lender distance. Furthermore, borrower-lender distance is likely to vary substantially for these

loans—credit-constrained borrowers will often need to search further for credit—which provides a good

laboratory for testing the impact of information-based innovations like credit scoring.

        The SBA data includes basic information that allows us to identify and characterize the loan

originator, such as the name of the lender, the physical address of the office that wrote the loan, and the

SBA lender certification type.5 Using the lender name and address, we can link the SBA data with bank-

level data from the Call Reports. Specifically, we collect bank size, number of bank branches,

organizational structure (e.g., affiliation in a multi-bank holding company), and a variety of bank financial

performance ratios. 6 The SBA data also includes detailed borrower-specific information, such as the

physical location of the borrower, industry classification (SIC) code, corporate structure, number of

employees, and the age of the firm. Using the borrower location, we can link the SBA data with market-

level data from a variety of databases, including the Summary of Deposits. This allows us to construct

variables describing the borrower’s market conditions, including economic density (MSA or rural),

banking competition, and state regulatory restrictions. Finally, the SBA data includes loan-specific

information such as the size and maturity of the loan, the SBA guarantee percentage, and whether the loan

was originated under the SBA’s low documentation (low-doc) program.

        With both borrower location and lender location in-hand, we calculated the straight-line (as the

crow flies) geographic distance between the borrower and the lender. The distribution of borrower-lender

distance is highly skewed by the especially large lending distances late in our sample period; the mean

  Experienced SBA lenders with good track records can attain either preferred loan provider (PLP) or certified loan
provider (CLP) status, which reduces their administrative burden. PLP lenders have the least restrictive SBA
documentation requirements; in exchange for these reduced administrative costs, however, their loan guarantee
percentages are capped at a lower amount.
  We have not yet employed bank performance measures—such as overall profitability, apparent risk preferences,
business mix, or operational efficiency—in our regression tests. We plan to do so in the next version of the paper.

borrower-lender distance in our sample i 63 miles, while the median borrower-lender distance is only 9

miles. The dependent variable in our regressions is lnDISTANCE, which equals the natural log of

(DISTANCE + 1), where DISTANCE is measured in miles. This specification recognizes the fact that the

cost of traveling between two geographic points includes a fixed component, and as a result increases at a

decreasing rate with distance (Berger and DeYoung, 2001).

        Although the SBA loans for which we measure borrower-lender distance are in some ways

atypical of small business credits in general, we do not believe that DISTANCE is an especially biased

measure for our purposes. Specifically, Figure 1 above shows that distances reported in this paper are

similar in level and trend to those reported in Petersen and Rajan (2002) for the NSSBF data, and that the

differences between the two time series are stable over time. One might also be concerned that, since

SBA loans carry government guarantees against credit loss, our results may not generalize to the (non-

government subsidized) population of small business loans. We believe this concern is exaggerated, for

three reasons. First, the SBA guarantee gives lenders a put option that covers only a portion of SBA loan

losses—so long as SBA lenders incur some non-trivial share of potential loan default costs, they face the

same qualitative tradeoffs between risk and return as do non-subsidized lenders. Second, as the SBA put

option is simply a hedge against credit risk, it is qualitatively similar to credit risk hedging techniques

used by issuers of non-subsidized loans (e.g., portfolio diversification, credit derivatives). Third,

geographic distance confers information-gathering frictions on both subsidized and non-subsidized

lenders alike (e.g., increased travel costs, less frequent in-person contact) that are independent of

borrower risk; for both types of loans, credit-scoring techniques could potentially enhance borrower-

lender distance by mitigating, or dampen borrower-lender distance by exacerbating, these information

problems. Finally, we stress that all of our empirical results are derived after conditioning lnDISTANCE

on the magnitude of the SBA loan guarantee.

           We control for the effects of general technological advance on borrower-lender distance by

including annual fixed effects variables in our regressions. 7 We measure the effect of a specific

technological advance—credit-scoring technology—on borrower-lender distance more directly, using

data from a telephone survey conducted by the Federal Reserve Bank of Atlanta in January 1998.8 The

survey queried the lead (largest) bank in each of the 200 largest U.S. bank holding companies on whether

and how they used small business credit scoring (SBCS). Overall, 62 of the 99 responding sample banks

said that they used SBCS as of January 1998 and that adoption of this technology began in 1992

(Akhavein, Frame, and White 2005, Table 1). All banks that used SBCS did so for credits under

$100,000, 74.2 percent of these banks used it for credits under $250,000, but only 21.0 percent reported

using the technology on credits between $250,000 and $1 million. We make the reasonable assumption

that this underwriting technology passed freely from the lead banks queried in the survey to other non-

lead affiliates in those bank holding company organizations. Thus, our main measure of credit scoring is

SCORER, a dummy variable equal to one for all banking affiliates of banking companies in which the

lead bank used any type of SBCS technique during the year in question. (The At lanta Fed survey also

recorded whether banks used credit scores to (a) automatically approve or reject loan applications or (b)

set credit terms. We employ an augmented form of SCORER in our tests to capture these nuances.) By

this definition, about 14 percent of the small business loans in our sample were originated by “scoring


           While the Atlanta Fed SBCS survey data is the best extant source on the dissemination of small-

business credit -scoring techniques at U.S. banking companies, these data have some obvious limitations.

First, we do not know whether “scoring banks” credit scored all, or just a portion, of their small business

loan applications. Second, we cannot identify lenders that adopted credit-scoring technology after 1998.

Third, we cannot identify credit-scoring lenders affiliated with banking companies too small to be

included in the survey. We acknowledge that these limitations are not desirable, but we do not believe

    The annual fixed effects terms will also capture variation in year-specific macroeconomic conditions.
    See Frame, Srinivasan, and Woosley (2001) for a complete discussion of the survey questions and results.

they are especially problematic. The first limitation simply constrains the form with which we state our

credit-scoring hypothesis: We test whether banks that use credit scoring models have different distance

patterns, not whether credit-scored loans have different distance patterns. We address the second

limitation by estimating our regression models for a sub-sample of pre-1999 data. The third limitation is

unlikely to be meaningful insofar as small business credit scoring was almost exclusively a large bank

activity prior to 1999.

III. Regression framework and hypotheses

        We model borrower-lender distance (lnDISTANCE) as a function of lender characteristics,

borrower characteristics, loan terms, and local market conditions. Although the SBA loan originations in

our data occurred over an 18 year time period (1984-2001), we do not follow these loans over time: we

treat the data as a cross section, observing all variables in the year in which the loan was originated. We

control for changes in general economic, technological, regulatory, and industry conditions over the 18-

year period by including a vector of annual fixed effects dummies (YEAR) on the right-hand side of the

regression equation:



                          lnLOANSIZE, MATURITY3, MATURITY7, LOWDOC, GUAR%;

                          URBAN, HHI, URBAN*HHI;

                          YEAR) + ei                                                                   (1)

where the dependent variable lnDISTANCE is the natural log of borrower-lender distance as defined

above, i indexes individual loans originated during our sample period 1984-2001 (index is suppressed for

the right-hand side variables), and e is a random error term assumed to be symmetric with zero mean.

        The first row in (1) lists lender characteristics. We measure the size of the lending bank with

lnBANKSIZE, the natural log of lender assets. Like borrow-lender distance, this distribution of bank size

is highly skewed. The mean lender had around $1.5 billion of assets, while the median lender had only

around $340 million; lenders ranged from tiny banks with barely $1 million of assets to huge banks with

over $500 billion in assets. MBHC is a dummy equal to one if the lender is an affiliate in a multi-bank

holding company; slightly over half of loans in our sample were originated by affiliates in MBHC

organizations. BRANCHES is the number of branches operated by the lending bank at the time of loan

origination. Specialized lending programs are indicated by PLP and CLP, dummy variables equal to one

for lenders that are “preferred loan providers” or “certified loan providers” as discussed above. SCORER

is a dummy equal to one for lenders affiliated with organizations in which the lead bank identified itself

as a “small business credit scorer” in the Atlanta Fed survey, as discussed above.

        We expect a positive coefficient on lnBANKSIZE, because the resources available at larger banks

should better equip them to lend at longer distances, and because small banks are more likely to be

relationship-lenders that stress local focus and personalized service. We expect a negative coefficient on

MBHC, because multi-bank organizations tend to operate in a more decentralized fashion, placing

authority in the hands of loan officers that are further from headquarters and thus potentially closer to

borrowers. Holding bank size and organizational form constant, the coefficient on BRANCHES could be

either negative (if loan officers located at branch locations have the authority to write and approve loans)

or positive (if lending authority resides only at the main office or at larger branches, and large branch

networks serve as marketing tools for selling loans and providing service ex post to approved applicants).

We expect positive coefficients on PLP and CLP, because these designations indicate the most highly

skilled underwriters in the SBA program, likely to be better able to handle the additional costs and risks

of lending at longer distances.

        We are especially interested in the coefficient on SCORER. The univariate data displayed above

in Figure 1 strongly suggests a positive relationship between credit scoring and borrower-lender distance.

A positive coefficient on SCORER in these multivariate tests (a) would lend strong support to PR’s

(2002) conjecture that technological advance was driving increased borrower-lender distance during the

1980s and early 1990s, and (b) that a specific form of lending technology not available during that time

period—credit scoring of small business loans—has extended that increasing trend throughout the 1990s

and into the early-2000s.

         The second row in (1) lists borrower characteristics. Our proxy for borrowing firm size,

lnFIRMSIZE, is the natural log of the number of full-time equivalent workers employed by the borrowing

firm; the mean (median) borrower employed about twelve (five) workers. CORPORAT ION and

PARTNERSHIP are dummies equal to one, respectively, for lenders organized as corporations and

partnerships (sole proprietorship is the omitted category). NEWFIRM is a dummy equal to one if the

borrower is less than three years old; these young firms comprise about one-third of the loans in our

sample. SIC is a vector of dummy variables indicating whether the borrower’s main line-of-business falls

within one of three especially well-populated Standard Industry Classifications.

         We expect the coefficient on lnFIRMSIZE to be negative: larger firms are more likely to be

known locally (in other words, informational opacity is decreasing in firm size) and hence will not have to

search as far for a lender. For similar reasons, we expect the coefficient on NEWFIRM to be positive:

young firms tend to be both riskier and more informationally opaque, and may have to search further for a

loan. We have no a priori expectations about the coefficient signs on CORPORATION, PARTNERSHIP,

or the elements of SIC.

         The third row in (1) lists loan characteristics. lnLOANSIZE is the natural log of the initial loan

principal; loans ranged from as little as $2,000 to as much as $2.5 million, with an average loan of about

$144,000 (median = $92,000). MATURITY3 and MATURITY7 are dummy variables equal to one,

respectively, if the loan has a maturity of 3 years or 7 years (the omitted contract is the 15-year loan). 9

LOWDOC is a dummy variable equal to one for loans underwritten using the SBA’s “low

documentation” option that started in 1994 to reduce paperwork for loans less than $100,000. The SBA

  The 15-year loans are typically collateralized by real estate. The SBA also markets products with maturities other
than 3-, 7-, and 15-years, such as lines of credit, but loans with these three maturities represent the most substantial
part of the SBA portfolio.

guarantee percentage, GUAR%, is the percentage of loan default losses that the lender can put to the SBA

in the event of default; the mean loan guarantee was about 80 percent, but ranged from as low as 11

percent to as high as 90 percent. Over the 18 years of our data sample, the SBA loan guarantees have (a)

declined on average and (b) exhibited increased variation across loans (DeYoung, Glennon, and Nigro


         We expect the coefficient on lnLOANSIZE to be positive, because (holding firm size constant)

firms with larger credit needs may have to search further away for a loan. Since lenders will likely be

more willing to lend to opaque borrowers if they are close by and hence easy to monitor, we expect the

coefficient on LOWDOC to be negative. The relationship between GUAR% and lnDISTANCE could be

either positive or negative. On-the-one-hand, if lenders view the SBA put option as credit insurance, then

they may be willing to reach further (i.e., take more distance-related risk) in exchange for a larger loan

guarantee; on-the-other hand, a borrower that can qualify for a large put option may not need to search

very far to find a willing SBA lender. We have no a priori expectations about the coefficient signs on


         The fourth row in (1) lists characteristics of the local market in which the borrower is located.

URBAN is a crude measure of economic density: a dummy variable equal to one if the borrower is

located in a Metropolitan Statistical Area (MSA), and zero if the borrower is located in a rural (non-MSA)

county. In our data set, about four-out-five borrowers are located in urban areas. HHI is a Herfindahl

index that measures banking industry structure in the local market (MSA or rural county), constructed

using branch-level deposit data. It is well-known that the HHI is systematically higher in rural counties

than in urban MSAs; to better measure the impact of banking concentration in both urban and rural

markets, we also include an interaction term HHI*URBAN.

         We expect the coefficient on URBAN to be positive for two reasons. First, in rural markets both

bank lenders and their small business borrowers tend to be located in small towns, with little potential for

long borrower-lender distances. (In contrast, we would expect relatively long borrower-lender distances

for agricultural loans made by rural banks, because these borrowers are located outside of town.) Second,

SBA borrowers in urban markets are often located i low-income or moderate-income neighborhoods

(LMIs) which tend to have few bank branches and even fewer lending offices. We expect a positive

relationship between HHI and lnDISTANCE in both urban and rural markets. The structure-performance

hypothesis predicts that high levels of HHI will result in high prices and profits; i the absence of entry

barriers, this creates an incentive for out-of-market banks to make long-distance loans in the local


         The final element on the right-hand side of (1) is the vector of annual fixed effects dummies,

YEAR. (1984 is the omitted year in most of our regression tests.) These variables are included to control

for changes in general economic, technological, regulatory, and industry conditions over time. We expect

these terms to increase over time—i.e., become more positive or less negative—reflecting the distance-

increasing influences of industry consolidation (fewer banking companies) and technological change

(easier and more accurate communications and information flow) that are not already controlled for by

the other right-hand side terms discussed above.

IV. Preliminary results

         We estimate equation (1) using ordinary least squares (OLS) techniques. Regression results for

the full sample of loans (1984-2001) are displayed in Table 2. Results from various sub-sample

regressions (time periods, loan size, lender size, and scoring status) are displayed in Tables 3, 4, and 5.

We stress that all of these results are preliminary, and are subject to change as we explore different

regression specifications and estimation techniques. 11

                                                [Insert Table 2 here]

   In contrast to structure-performance theory, Peterson and Rajan (1994) argue that high levels of concentration in
small business lending markets will result in low prices and high output in the short-run, but will eventually result in
high prices as lenders exploit information-based switching costs. If this theory is accurate, then out-of-market entry
will be more difficult due to information-based entry barriers.
   A list of likely extensions includes using tobit estimation techniques, controlling for lender fixed-effects, and
exploring more carefully how differences in economic density, per capita income, competitive rivalry, and industry
consolidation are associated with differences in borrower-lender distance across local borrower markets.

                                             [Insert Table 3 here]

                                             [Insert Table 4 here]

                                             [Insert Table 5 here]

        The most salient results are the estimated coefficients on the YEAR fixed effects dummies,

which strongly resemble the borrower-lender distance patterns displayed in Figure 1. In the full sample

regressions (Table 2), these coefficients indicate increasing borrower-lender distances beginning in the

early-1990s and continuing through the end of the sample period. We find similar increasing patterns for

these coefficients in the sub-sample regressions, although subtle differences in those estimations are

instructive. For instance, the increase in these coefficients in the large bank sub-sample (Table 5, column

[10]) is nearly monotonic, while in the small bank sub-sample (Table 5, column [9]) these coefficients

take substantial dips during years associated with economic slow-downs (1992 and 2000). These patterns

suggest that small banks pulled back their geographic lending footprint during recessions—perhaps to

divert credit to their close-by, core relationship borrowers—while large banks did not need to do so.

        The YEAR coefficients indicate economically substantial increases in borrower-lender distance

over time. For example, the coefficient on YEAR1998 = 0.4966 in Table 2 translates into an approximate

15 mile increase in mean borrower-lender distance between 1984 (the omitted year) and 1998, holding all

other conditions constant. 12 Of course, this is just a partial estimate of the increase in borrower-lender

distance between 1984 and 1998, because the YEAR coefficients capture only the increase in borrower-

lender distance due to trends in economic conditions, industry consolidation, technological change, and

other general conditions that are not already captured by the other right-hand side variables. We now turn

to the estimated coefficients on those variables, to analyze specific drivers of the documented increase in

borrower-lender distance over time:

  The result is derived as follows: Mean DISTANCE in 1984 was 23.6 miles. The increase in DISTANCE by 1998
due to general conditions is ln(23.6) + 0.4966 = 3.66, which de-logged equals about 38.8 miles. The difference is
15.2 miles.

        Lender characteristics. Consistent with the data displayed on the far right-hand side of Figure 1,

our regressions indicate that credit-scoring banks reach out substantially further to make small business

loans. Based on the full-sample estimates in Table 2, where the coefficient on SCORER is 0.6499, the

average loan made by a scoring bank was about 27 miles more distant than the average loan made by a

non-scoring bank, all else held equal. 13 The estimated coefficient on SCORER is largest in the small bank

sub-sample (1.1415 in Table 5, column [9]), banks for which any automated lending process marks a

departure from the ethos of locally focused, relationship lending. The impact of SCORER shrinks as loan

size increases (Table 3), consistent with the idea that credit scoring is most appropriate for “micro-

business lending.” The coefficient on SCORER is also smaller for sub-samples of earlier data (Table 3),

because borrower-lender distances were simply smaller overall during those years.

        Borrower-lender distances tend to be longer when lenders are large (lnBANKSIZE), operate large

branch networks (BRANCHES), or have recognized expertise at making SBA loans (PLP). These

findings are consistent with our earlier conjectures, and are robust throughout the sub-sample regressions.

We note, however, that l nder size has no impact on borrower-lender distance in the SCORING=1 sub-

sample; hence, among the relatively large banks that dominate small business credit scoring in our data, it

appears that the performance-improving effects of credit scoring (better information, lower production

costs) may be substitutes for the risk-absorption benefits of bank size (diversification). We find only

sporadic evidence consistent with our conjecture that borrower-lender distance would be shorter for bank

lenders affiliated with multi-bank holding companies (MBHC).

        Borrower characteristics.    The full-sample regressions (Table 2) indicate that young firms

(NEWFIRM) tend to be located further away from their lenders on average. This result is driven mainly

by loans made by small banks, and by banks that do not credit score small business loans (Table 5,

columns [9] and [11]). So although our results support the conventional wisdom that new businesses must

    The result is derived as follows: Mean DISTANCE for non-scoring banks is 30.1 miles. The increase in
DISTANCE due to SCORER is ln(30.1) + 0.6499 = 4.05, which de-logged equals about 57.4 miles. The difference
is 27.3 miles.

search far and wide for credit, the evidence also indicates that their search is likely to end at a small

lender that uses traditional underwriting methods.

         As expected, our regressions indicate that larger small businesses (lnFIRMSIZE) tend to secure

credit at lenders located relatively close by, although this effect is a small one: a 10 percent increase in

firms size is associated with only a 4/10ths percent decrease in distance (Table 2).14 Interestingly, legally

incorporated small businesses (CORPORATION) also find credit close b one might speculate that

incorporated status indicates a financially sophisticated firms that understands that value of close banking

relationships. Both of these results are robust throughout most of the sub-sample regressions.

         Loan characteristics. Perhaps the biggest surprise in our results is the estimated relationship

between borrower-lender distance and the size of the SBA loan guarantee. With only two exceptions, the

coefficient on GUAR% is negative, suggesting that the SBA put option helps borrowers find credit

relationships nearby. In the overall data (Table 2) the coefficient on GUAR% is -0.7875; this translates

into an approximate five-mile decrease in borrower-lender distance, for a 10 percentage point increase in

the SBA loan guarantee at the means of the data. 15 In contrast, GUAR% is positively related to borrower-

lender distance for larger loans (Table 4, column [8]) and during the first half of the sample period (Table

3, column [4]). The former result indicates that loan guarantees can influence banks to take on distance-

related risks, but only if the dollar amount of the put option is large. (Another interpretation: Perhaps the

SBA has been systematically over-guaranteeing large SBA loans.) We note without interpretation that the

latter result is from an earlier time period when there was very little variation in GUAR% across SBA

loans. In any case, the collective results for GUAR% indicate that the relationships among the SBA put

option, the incentives facing lenders and borrowers, and borrower-lender distance are not simple ones.

         The coefficient on lnLOANSIZE tends to be positive and significant, consistent with our

conjecture that borrowers seeking large amounts of credit are willing to search further. As expected, the

  Both the dependent and independent variables are in natural logs, so the coefficient is interpreted as an elasticity.
   The result is derived as follows: Mean DISTANCE is 63.8 miles. The decrease in DISTANCE due to a 10
percentage point increase in GUAR% is ln(63.8) – (0.7875*0.10) = 4.08, which de-logged equals about 58.9 miles.
The difference is 4.9 miles.

coefficient on LOWDOC tends to be negative and significant. The coefficients on the MATURITY

dummies follow no discernable pattern and are usually non-significant.

        Local market characteristics. Consistent with the predictions of the structure-performance

paradigm, the marginal effect of HHI on borrower-lender distance tends to be positive—that is, we find

longer borrower-lender distances when local lenders have market power, circumstantial evidence that

monopoly profits attract out-of-market lenders. However, the statistical strength of these results is weak.

The derivative of lnDISTANCE with respect to HHI is statistically positive in only two of the twelve

regressions, whether it is evaluated for URBAN=0 or for URBAN=1. 16 This implicit “market-entry

effect” is strongest for mid-sized loans (Table 5, column [7]), perhaps because these loans are small

enough to be passed over by the local monopolist, but large enough to attract out-of-market competitor

lenders, and in the second half of the sample period (Table 3, column [5]), perhaps because deregulation

had by then relaxed barriers on geographic expansion and entry.

        Finally, the coefficient on URBAN is positive and significant throughout. This is consistent with

our conjectures: (a) that rural development patterns naturally place small businesses and their potential

lenders close together in small towns, and (b) a disproportionate number of urban SBA borrowers are

located in under-banked low-income neighborhoods. Further research and additional data is needed to

separately identify these two hypotheses.

V. Preliminary conclusions

        One might think of the distance between small business borrowers and their lenders as a

barometer for conditions important to the health of the small business sector: the degree to which credit is

available in local markets, the ease with which small businesses can credibly demonstrate their

creditworthiness to off-site lenders, and the ability of out-of-market lenders to manage the risks associated

with long-distance lending to informationally opaque small firms. A documented by several previous

  For example, in the full sample results in Table 2, the derivative being evaluated is 0.2416 + 0.1314*URBAN. For
the average rural bank (URBAN=0), this derivative equals 0.2416 and has a p-value of 11%. For the average urban
bank (URBAN=1), this derivative equals 0.3730 and has a p-value of 16%.

studies, borrower-lender distances began increasing in the U.S. over two decades ago; the most popular

explanation—extended and argued largely without direct evidence—was that improved financial and

communications technologies had relaxed the informational barriers to evaluating creditworthiness,

monitoring behavior, and hedging the risk associated with extending credit to distant small businesses.

        In this study, we exploit a large random sample of small business loans originated by U.S.

commercial banks between 1984 and 2001 under the Small Business Administration’s (SBA) flagship

7(a) guaranteed loan program. Using these data, we corroborate findings in the extant literature that the

distance between small business borrowers and their lenders increased during the 1980s and 1990s; we

demonstrate that this increase has continued, even accelerated, into the present decade; and we provide

direct evidence that links the largest and most recent increases in borrower-lender distance to a specific

innovation, credit scoring. For example, our estimates indicate that borrower-lender distance was 27 miles

longer on average for banks that used small business credit scoring techniques, all else equal.

        While credit-scoring is the most important single driver of borrower-lender distance during our

sample period, we find substantial evidence that other phenomena are important as well. Our tests

indicate a substantial increase in borrower-lender distance over time that is unrelated to the adoption of

credit scoring methods, but rather is attributable to general (read: unspecified in our regressions) changes

in technology, deregulation, and other secular trends. We find that young firms, relatively small firms,

firms with disproportionately large credit needs, and firms located in urban markets end up with more

distant lenders, on average. And we find that l rge banks, banks with extensive branching networks, and

banks with external accreditation as high-quality lenders tend to lend to more distant small businesses.

Although we do not directly specify industry-wide deregulation or consolidation trends in our tests (an

omission that will be rectified in the next version of the paper), we do include a logical outcome of these

trends in our regressions: local banking market concentration. While the results are statistically weak,

they indicate longer borrower-lender distances when local lenders have market power, which we interpret

as circumstantial evidence that monopoly profits attract out-of-market lenders.

        Although these results are derived for loans that carry (partial) government guarantees, we argue

that they should generalize to non-subsidized borrower-lender distances. We operationalize this argument

by including the size of the pro-rated subsidy as a control in our regressions, which as a first

approximation should statistically absorb any impact that loan guarantees may have on borrower-lender

distance. Including this control variable generated an intriguing ancillary result: we find that borrowers

receiving loans with larger percentage guarantees tended to be located closer to their lenders. While one

might naturally expect that larger loan guarantees would be an incentive for lenders to take additional

distance-related risks (as well as other types of risk), this result suggests that small borrowers may be able

to use the promise of large loan guarantees to secure credit from a nearby bank—an outcome that is

beneficial to both sides of the transaction.

        We stress that our results are preliminary, and may change as we explore different regression

specifications and statistical techniques. Having made this disclaimer, we believe that our current results

are consistent with the growing belief—based on casual observation as well as research studies (e.g.,

DeYoung, Hunter, and Udell, 2004)—that the banking industry is in the process of bifurcating into two

separate strategy groups. Stated in the context of our findings for small business lending, one group is

comprised of large, credit-scoring banks that treat their small business borrowers as transactions credits or

financial commodities; for these banks, the informational costs of long-distance lending may be offset by

the economies associated with large scale lending operations (e.g., unit cost reductions, diversification

gains, potential asset securitization). The other group is comprised of small, locally focused banks that

rely largely (though perhaps not completely) on traditional underwriting techniques for small business

loans; long-distance lending is antithetical to the relationship-based production methods practiced by

these banks.


Agarwal, Sumit and Robert Hauswald. 2006. “Distance and Information Asymmetries in Lending
       Decisions”.     Working      paper,      American        University.           Available   at:

Akhavein, Jalal, Frame, W. Scott Frame, and Lawrence J. White. 2005. “The Diffusion of Financial
       Innovation: An Examination of the Adoption of Small Business Credit Scoring by Large Banking
       Organizations.” Journal of Business, 78(2), 577-96.

Berger, Allen N. and Robert DeYoung. 2001. “The Effects of Geographic Expansion on Bank Efficiency.”
        Journal of Financial Services Research, 19(2-3), 163-184

Berger, Allen N. Robert DeYoung, Hesna Genay, and Gregory F. Udell. 2000. “Globalization of Financial
        Institutions: Evidence from Cross-border Banking Performance,” Federal Reserve Board working
        paper, 2000-04.

Berger, Allen N., Nathan H. Miller, Mitchell A. Petersen, Raghuram G. Rajan, and Jeremy C. Stein. 2005.
        “Does Function Follow Organizational Form? Evidence from the Lending Practices of Large and
        Small Banks” Journal of Financial Economics, 76(2), 237-269

Berger, Allen N., and Gregory F. Udell. 1998. “The Economics of Small Business Finance: The Roles of
        Private Equity and Debt Markets in the Financial Growth Cycle.” Journal of Banking and Finance,
        22(6-8), 613-673.

Cole, Rebel A., Lawrence G. Goldberg, and Lawrence J. White. 2004. “Cookie Cutter versus Character: The
        Micro Structure of Small Business Lending by Large and Small Banks.” Journal of Financial and
        Quantitative Analysis, 39(2), 227-251.

Degreyse, Hans and Steven Ongena. 2005. “Distance, Lending Relationships, and Competition.” Journal of
       Finance, 60(3), 231-266.

DeYoung, Robert, Dennis Glennon, and Peter Nigro. 2006. “Borrower-Lender Distance, Credit Scoring, and
      the Performance of Small Business Loans.”

DeYoung, Robert, William C. Hunter, and Gregory F. Udell. 2004. “The Past, Present, and Probable Future
      for Community Banks.” Journal of Financial Services Research, 25(2-3), 85-133.

Elliehausen, Gregory E. and John D. Wolken. 1990. “Banking Markets and the Use of Financial Services by
        Small and Medium-Sized Businesses.” Federal Reserve Bulletin, 801-817.

Frame, W. Scott, Aruna Srinivasan, and Lynn Woosley. 2001. “The Effect of Credit Scoring on Small
       Business Lending.” Journal of Money, Credit, and Banking, 33(3), 813-825.

Frame, W. Scott, Michael Padhi, and Lynn Woolsey. 2004. “The Effect of Credit Scoring on Small Business
       Lending in Low- and Moderate Income Areas.” Financial Review, 39(1), 35-54.

Glennon, Dennis and Peter Nigro. 2005a. “Measuring the Default Risk of Small Business Loans: A
       Survival Analysis Approach.” Journal of Money, Credit, and Banking, 37(5), 923-947.

Glennon, Dennis and Peter Nigro. 2005b. “An Analysis of SBA Loan Defaults by Maturity Structure.”
       Journal of Financial Services Research, 28(1/2/3), 77-111.

Hannan, Timothy H. and Kenneth Brevoort. 2004. “Commercial Lending and Distance: Evidence from
       Community Reinvestment Act Data.” Federal Reserve Board working paper, 2004-24.

Hannan, Timothy H. 2003. “Changes in Non-Local Lending to Small Business.” Journal of Financial
       Services Research, 24, 31-46.

Kwast, Myron L., Martha Starr-McCluer, and John D. Wolken. 1997. “Market Definition and the Analysis
       of Antitrust in Banking” Antitrust Bulletin, 42(4), 973-995.

Petersen, Mitchell A., and Raghuram G. Rajan. 1994. “The Benefits of Lending Relationships: Evidence
        from Small Business Data”. Journal of Finance, 49(1), 3-37.

Petersen, Mitchell A., and Raghuram G. Rajan. 2002. “The Information Revolution and Small Business
        Lending: Does Distance Still Matter?” Journal of Finance, 57, 2533-2570.

                                               Table 1
Summary statistics for 32,423 small business loans originated by U.S. commercial banks between 1984
 and 2001 with partial guarantees under the Small Business Administration (SBA) 7(a) loan program.

            Variable           Mean         Std. Deviation     Minimum         Maximum
           DISTANCE           63.8375          284.8216         0.1000          7882.6000
         lnDISTANCE           2.2144            1.9266          -2.3025          8.9724
            SCORER            0.1471            0.3542             0                1
              PLP             0.1292            0.3355             0                1
              CLP             0.1633            0.3696             0                1
           BANKSIZE         $1,5099,233       61,072,285         1,482         607,000,000
         lnBANKSIZE           13.2490           2.4757          6.9621           20.2241
             MBHC             0.5311            0.4990             0                1
          BRANCHES           120.4132          408.8582            1               4670
              HHI             0.1902            0.1146             0                1
            URBAN             0.8259            0.3791             0                1
        CORPORATION           0.5623            0.4961             0                1
        PARTNERSHIP           0.0732            0.2606             0                1
           FIRMSIZE           12.5077          111.4224            1               9999
          lnFIRMSIZE          1.6542            1.1283             0             9.2102
           NEWFIRM            0.3277            0.4694             0                1
             SIC_A            0.0317            0.1752             0                1
             SIC_B            0.0033            0.0576             0                1
             SIC_G            0.3321            0.4709             0                1
           LOANSIZE          $144,242          165,037           2,000          2,550,000
         lnLOANSIZE           11.4261           0.9504          7.6009           14.7516
            GUAR%             0.8002            0.1011          0.1100           0.9000
           LOWDOC             0.3109            0.4628             0                1
         MATURITY3            0.1374            0.3442             0                1
         MATURITY7            0.6460            0.4782             0                1
           YEAR1985           0.0177            0.1318             0                1
           YEAR1986           0.0271            0.1625             0                1
           YEAR1987           0.0269            0.1618             0                1
           YEAR1988           0.0245            0.1548             0                1
           YEAR1989           0.0294            0.1691             0                1
           YEAR1990           0.0312            0.1739             0                1
           YEAR1991           0.0332            0.1792             0                1
           YEAR1992           0.0398            0.1955             0                1
           YEAR1993           0.0505            0.2190             0                1
           YEAR1994           0.0785            0.2689             0                1
           YEAR1995           0.1368            0.3436             0                1
           YEAR1996           0.0824            0.2749             0                1
           YEAR1997           0.0993            0.2991             0                1
           YEAR1998           0.0922            0.2894             0                1
           YEAR1999           0.0883            0.2837             0                1
           YEAR2000           0.0887            0.2844             0                1
           YEAR2001           0.0288            0.1673             0                1

               Table 2
      Main regression, full sample

    Intercept      -0.6041*** (0.2315)
    SCORER          0.6499*** (0.0371)
      PLP           0.4434*** (0.0327)
      CLP           0.0762*** (0.0290)
 lnBANKSIZE         0.0844*** (0.0064)
     MBHC           -0.0497** (0.0226)
  BRANCHES          0.0007*** (0.00003)
      HHI               0.2416 (0.1495)
 HHI*URBAN              0.1314 (0.1872)
    URBAN           0.5287*** (0.0536)
CORPORATION        -0.0885*** (0.0227)
PARTNERSHIP            -0.0217 (0.0389)
  lnFIRMSIZE       -0.0427*** (0.0104)
   NEWFIRM           0.0460** (0.0216)
     SIC_A          0.3902*** (0.0561)
     SIC_B              0.1579 (0.1660)
     SIC_G         -0.0732*** (0.0208)
 lnLOANSIZE         0.1241*** (0.0143)
    GUAR%          -0.7875*** (0.1368)
   LOWDOC          -0.1112*** (0.0309)
 MATURITY3             -0.0115 (0.0404)
 MATURITY7              0.0184 (0.0263)
   YEAR1985            -0.0049 (0.0943)
   YEAR1986           -0.1451* (0.0844)
   YEAR1987            -0.0019 (0.0846)
   YEAR1988            -0.0412 (0.0866)
   YEAR1989            -0.0533 (0.0830)
   YEAR1990            -0.0253 (0.0819)
   YEAR1991            0.1476* (0.0809)
   YEAR1992             0.0755 (0.0781)
   YEAR1993            0.1298* (0.0751)
   YEAR1994         0.2883*** (0.0712)
   YEAR1995         0.4110*** (0.0697)
   YEAR1996         0.3925*** (0.0744)
   YEAR1997         0.3686*** (0.0734)
   YEAR1998         0.4966*** (0.0741)
   YEAR1999         0.5530*** (0.0756)
   YEAR2000         0.6139*** (0.0755)
   YEAR2001         0.7993*** (0.0886)
        N               32,423
  Adjusted-R2           0.2082

                                                     Table 3
                                   Sub-sample regressions: Various time periods.

                                                                            First Half               Second Half
                       1984-1998                 1993-1998
                                                                            1984-1992                 1993-2001
                            [2]                       [3]                      [4]                       [5]
    Intercept     1.5610*** (0.4523)        1.8002*** (0.5374)       -1.8811*** (0.5164)           0.2044 (0.2531)
    SCORER        0.3217*** (0.0615)        0.3121*** (0.0600)                                 0.5485*** (0.0383)
      PLP         0.4832*** (0.0602)        0.5539*** (0.0702)       0.4112***     (0.0848)    0.5088*** (0.0368)
      CLP             0.0641 (0.0488)          -0.0037 (0.0668)      0.1791***     (0.0453)        0.0507 (0.0387)
 lnBANKSIZE       0.0815*** (0.0129)        0.1102*** (0.0160)        0.0312**     (0.0157)    0.1043*** (0.0074)
     MBHC             0.0066 (0.0523)           0.0314 (0.0649)      -0.1089**     (0.0443)       -0.0185 (0.0265)
  BRANCHES        0.0010*** (0.00008)       0.0010*** (0.0001)           0.0004    (0.0003)    0.0007*** (0.0000)
      HHI             0.4746 (0.5417)           0.7559 (0.6691)         -0.0884    (0.2536)       0.3626* (0.1873)
 HHI*URBAN            0.0169 (0.5694)          -0.1714 (0.7049)          0.2187    (0.3288)        0.2100 (0.2317)
    URBAN         0.5042*** (0.1696)        0.6345*** (0.2050)       0.4606***     (0.1003)    0.5371*** (0.0640)
CORPORATION       -0.0936** (0.0427)       -0.1433*** (0.0507)          -0.0728    (0.0472)   -0.0987*** (0.0259)
PARTNERSHIP           0.0263 (0.0752)           0.0150 (0.0935)         -0.0422    (0.0707)       -0.0151 (0.0468)
  lnFIRMSIZE      -0.0407** (0.0189)        -0.0547** (0.0229)          -0.0121    (0.0211)   -0.0513*** (0.0120)
   NEWFIRM           -0.0027 (0.0410)          -0.0261 (0.0483)          0.0301    (0.0458)       0.0454* (0.0246)
     SIC_A        0.5184*** (0.1054)        0.6129*** (0.1239)       0.3260***     (0.1255)    0.4151*** (0.0628)
     SIC_B           -0.1408 (0.3061)           0.3022 (0.4267)          0.0674    (0.2913)        0.2246 (0.2026)
     SIC_G           -0.0537 (0.0387)          -0.0108 (0.0461)        -0.0734*    (0.0419)   -0.0723*** (0.0240)
 lnLOANSIZE           0.0324 (0.0273)           0.0176 (0.0332)      0.1745***     (0.0299)    0.1239*** (0.0166)
    GUAR%        -1.4785*** (0.2629)       -2.5513*** (0.3167)       0.7958***     (0.3084)   -1.2685*** (0.1612)
   LOWDOC            0.0992* (0.0594)        0.1513** (0.0667)                                   -0.0621* (0.0341)
 MATURITY3           -0.0704 (0.0717)          -0.0182 (0.0871)          0.0462    (0.0847)       -0.0302 (0.0466)
 MATURITY7            0.0372 (0.0492)        0.1553** (0.0628)           0.0118    (0.0473)        0.0371 (0.0320)
   YEAR1985      -0.5734*** (0.1437)                                    -0.0087    (0.0966)
   YEAR1986      -0.9755*** (0.1246)                                    -0.1143    (0.0866)
   YEAR1987      -0.7743*** (0.1262)                                     0.0411    (0.0870)
   YEAR1988      -0.7012*** (0.1295)                                     0.0136    (0.0891)
   YEAR1989      -0.6841*** (0.1152)                                     0.0299    (0.0858)
   YEAR1990      -0.5323*** (0.1145)                                     0.0404    (0.0845)
   YEAR1991      -0.4780*** (0.1097)                                 0.2209***     (0.0838)
   YEAR1992      -0.5655*** (0.1042)                                    0.1530*    (0.0810)
   YEAR1993      -0.4876*** (0.0950)                                                          -0.5720***   (0.0755)
   YEAR1994      -0.3664*** (0.0845)         -0.1361*    (0.0763)                             -0.4234***   (0.0695)
   YEAR1995      -0.2719*** (0.0762)           -0.0723   (0.0689)                             -0.3101***   (0.0657)
   YEAR1996      -0.2222*** (0.0798)           -0.1066   (0.0751)                             -0.3657***   (0.0665)
   YEAR1997      -0.2152*** (0.0706)        -0.1351**    (0.0687)                             -0.3842***   (0.0647)
   YEAR1998                                                                                   -0.2668***   (0.0646)
   YEAR1999                                                                                   -0.2325***   (0.0643)
   YEAR2000                                                                                   -0.1759***   (0.0641)
        N            9,028                      6,383                     8,240                  24,183
   Adjusted-R2      0.1840                     0.1901                    0.0357                  0.2260

                                       Table 4
                           Sub-sample regressions: Loan size.

                     Loan < $100,000       $100,000<Loan<$250,000        $250,000 < Loan
                            [6]                         [7]                    [8]
    Intercept        -0.0538 (0.3333)           -1.0787 (1.0229)    -4.2921*** (1.1178)
    SCORER        0.6859*** (0.0473)         0.5822*** (0.0746)      0.4203*** (0.1059)
      PLP         0.3899*** (0.0516)         0.4697*** (0.0558)      0.6395*** (0.0758)
      CLP          0.1000** (0.0454)         0.1321*** (0.0483)          0.0664 (0.0631)
 lnBANKSIZE       0.0825*** (0.0083)         0.1020*** (0.0128)      0.0513*** (0.0175)
     MBHC            -0.0295 (0.0295)       -0.1583*** (0.0445)          0.0227 (0.0591)
  BRANCHES        0.0007*** (0.0000)         0.0007*** (0.0001)      0.0008*** (0.0001)
      HHI             0.1836 (0.1933)           0.5634* (0.3104)        -0.1242 (0.3681)
 HHI*URBAN            0.2107 (0.2428)           -0.3307 (0.3884)         0.6368 (0.4558)
    URBAN         0.5026*** (0.0682)         0.5701*** (0.1117)      0.5250*** (0.1406)
CORPORATION      -0.0998*** (0.0280)            -0.0745 (0.0463)        -0.0653 (0.0758)
PARTNERSHIP          -0.0513 (0.0485)            0.0435 (0.0818)        -0.0799 (0.1138)
  lnFIRMSIZE      -0.0348** (0.0144)         -0.0484** (0.0197)         -0.0214 (0.0251)
   NEWFIRM            0.0319 (0.0267)            0.0593 (0.0450)         0.0797 (0.0667)
     SIC_A        0.2326*** (0.0773)         0.3463*** (0.1148)      0.7629*** (0.1300)
     SIC_B           -0.0657 (0.2530)            0.2904 (0.3187)         0.4108 (0.3080)
     SIC_G       -0.0864*** (0.0266)            -0.0266 (0.0407)       -0.1143* (0.0601)
 lnLOANSIZE       0.1221*** (0.0243)             0.0925 (0.0786)     0.3126*** (0.0735)
    GUAR%        -1.4794*** (0.1844)            -0.0310 (0.3094)     1.2249*** (0.3966)
   LOWDOC         -0.1028** (0.0416)        -0.4403*** (0.0940)
 MATURITY3            0.0452 (0.0534)           -0.0647 (0.1035)       -0.1115    (0.1678)
 MATURITY7        0.1225*** (0.0421)             0.0018 (0.0441)    -0.1439**     (0.0580)
   YEAR1985          -0.0369 (0.1344)           -0.0479 (0.1593)        0.1456    (0.2373)
   YEAR1986          -0.0411 (0.1210)           -0.1491 (0.1454)      -0.3354*    (0.2002)
   YEAR1987           0.0104 (0.1232)            0.0101 (0.1428)        0.0469    (0.2023)
   YEAR1988           0.0472 (0.1259)           -0.1080 (0.1486)       -0.0258    (0.2010)
   YEAR1989           0.0426 (0.1172)           -0.0870 (0.1447)       -0.1612    (0.2013)
   YEAR1990          -0.1013 (0.1172)           -0.0332 (0.1417)        0.2583    (0.1942)
   YEAR1991           0.1098 (0.1157)            0.1971 (0.1400)        0.2110    (0.1920)
   YEAR1992           0.0772 (0.1121)            0.1345 (0.1351)        0.1074    (0.1842)
   YEAR1993           0.1594 (0.1083)            0.1176 (0.1306)        0.1070    (0.1746)
   YEAR1994       0.3730*** (0.1007)          0.2486** (0.1269)         0.2643    (0.1719)
   YEAR1995       0.5193*** (0.0991)            0.2080* (0.1269)     0.3344**     (0.1722)
   YEAR1996       0.3997*** (0.1046)         0.4643*** (0.1385)         0.3096    (0.1919)
   YEAR1997       0.3855*** (0.1041)          0.3382** (0.1339)      0.4442**     (0.1821)
   YEAR1998       0.4307*** (0.1053)         0.5701*** (0.1331)     0.7656***     (0.1809)
   YEAR1999       0.3661*** (0.1070)         0.8288*** (0.1372)     0.9652***     (0.1832)
   YEAR2000       0.4688*** (0.1073)         0.8921*** (0.1370)     0.8438***     (0.1829)
   YEAR2001       0.6167*** (0.1219)         1.1783*** (0.1641)     0.8830***     (0.2290)
        N             19,421                       8,256                  4,746
   Adjusted-R2        0.2171                     0.2137                 0.1785

                                                    Table 5
                             Sub-sample regressions: Lender size and Scoring status.

                   Assets < $1 billion        Assets > $1 billion        SCORER = 0              SCORER = 1
                           [9]                       [10]                    [11]                    [12]
    Intercept   -0.9419*** (0.3645)            0.3744 (0.5156)      -1.9182*** (0.2740)      3.1274*** (0.7522)
    SCORER       1.1415*** (0.0881)        0.3931*** (0.0478)
      PLP        0.3112*** (0.0461)        0.6152*** (0.0495)        0.4654***    (0.0369)    0.5143***    (0.0910)
      CLP        0.1631*** (0.0367)            0.0179 (0.0494)       0.1078***    (0.0298)    -0.2929**    (0.1196)
 lnBANKSIZE      0.0662*** (0.0169)        0.0739*** (0.0164)        0.0975***    (0.0082)       -0.0201   (0.0185)
     MBHC       -0.0856*** (0.0266)            0.0005 (0.0460)      -0.0973***    (0.0234)     0.2094**    (0.1011)
  BRANCHES           0.0022 (0.0024)       0.0007*** (0.0000)        0.0008***    (0.0001)    0.0007***    (0.0000)
      HHI            0.1993 (0.1516)          -0.0610 (0.9342)           0.2254   (0.1481)        1.8678   (1.8552)
 HHI*URBAN           0.2071 (0.2196)           0.4632 (0.9480)          0.3203*   (0.1918)       -2.1041   (1.8795)
    URBAN        0.5146*** (0.0576)            0.0913 (0.3038)       0.4539***    (0.0540)    2.1585***    (0.5156)
CORPORATION      -0.0594** (0.0284)       -0.1484*** (0.0379)        -0.0496**    (0.0243)   -0.2707***    (0.0629)
PARTNERSHIP         -0.0051 (0.0473)          -0.0745 (0.0681)           0.0082   (0.0406)    -0.2586**    (0.1270)
  lnFIRMSIZE    -0.0441*** (0.0134)       -0.0372*** (0.0166)       -0.0325***    (0.0112)   -0.1034***    (0.0286)
   NEWFIRM        0.0603** (0.0273)            0.0205 (0.0354)       0.0718***    (0.0233)       -0.0997   (0.0579)
     SIC_A       0.3767*** (0.0682)        0.4357*** (0.0987)        0.3776***    (0.0596)    0.5814***    (0.1620)
     SIC_B           0.1092 (0.1873)           0.2303 (0.3570)           0.1074   (0.1678)        0.6130   (0.8082)
     SIC_G      -0.0818*** (0.0259)          -0.0583* (0.0349)      -0.0793***    (0.0221)       -0.0377   (0.0592)
 lnLOANSIZE      0.1550*** (0.0189)        0.0892*** (0.0232)        0.1661***    (0.0156)        0.0225   (0.0400)
    GUAR%       -0.5935*** (0.2112)       -0.8735*** (0.2033)           -0.0028   (0.1672)   -1.5244***    (0.3574)
   LOWDOC       -0.1135*** (0.0385)           -0.0354 (0.0537)          -0.0461   (0.0329)   -0.3345***    (0.0989)
 MATURITY3           0.0564 (0.0533)          -0.0600 (0.0635)           0.0301   (0.0437)    -0.2189**    (0.1103)
 MATURITY7           0.0361 (0.0322)           0.0036 (0.0461)           0.0006   (0.0276)        0.0735   (0.0845)
   YEAR1985          0.0073 (0.1040)          -0.0706 (0.2221)          -0.0087   (0.0931)
   YEAR1986         -0.1014 (0.0939)          -0.2732 (0.1921)          -0.1394   (0.0833)
   YEAR1987          0.0428 (0.0939)          -0.1566 (0.1946)           0.0061   (0.0835)
   YEAR1988          0.0197 (0.0962)          -0.2236 (0.1980)          -0.0269   (0.0855)
   YEAR1989         -0.0047 (0.0949)          -0.0904 (0.1777)          -0.0499   (0.0820)
   YEAR1990          0.0554 (0.0926)          -0.1848 (0.1793)          -0.0238   (0.0809)
   YEAR1991       0.2164** (0.0918)            0.0248 (0.1763)          0.1471*   (0.0799)
   YEAR1992          0.1174 (0.0887)           0.0369 (0.1710)           0.0733   (0.0772)
   YEAR1993         0.1624* (0.0862)           0.1421 (0.1629)          0.1227*   (0.0743)   -1.7716***    (0.6920)
   YEAR1994      0.2908*** (0.0812)         0.3595** (0.1580)        0.2900***    (0.0706)   -1.3981***    (0.4550)
   YEAR1995      0.4061*** (0.0797)        0.5063*** (0.1552)        0.4021***    (0.0696)   -0.9618***    (0.1738)
   YEAR1996      0.3485*** (0.0866)        0.6170*** (0.1609)        0.4544***    (0.0749)   -0.9342***    (0.1614)
   YEAR1997      0.3493*** (0.0854)        0.5851*** (0.1606)        0.4741***    (0.0752)   -0.8827***    (0.1242)
   YEAR1998      0.4339*** (0.0871)        0.7814*** (0.1604)        0.4651***    (0.0764)   -0.4580***    (0.1188)
   YEAR1999      0.5879*** (0.0900)        0.7702*** (0.1616)        0.6333***    (0.0785)   -0.5683***    (0.1145)
   YEAR2000      0.4533*** (0.0912)        0.9761*** (0.1603)        0.5716***    (0.0784)   -0.3214***    (0.1152)
   YEAR2001      0.6924*** (0.1188)        1.0714*** (0.1702)        0.7229***    (0.0967)
        N            20,479                    11,944                    27,652                   4,771
  Adjusted-R2        0.0692                    0.2444                    0.0870                  0.2342

                                                                     Figure 1

                           Borrower-Lender Distances from Various Studies
                                    Petersen and Rajan 2002 (PR)
                              DeYoung, Glennon, and Nigro 2006 (DGN)



miles (median average)





















                                             PR             DGN            DGN, scorers                DGN, non-scorers


Shared By: