Journal of Banking & Finance 53 (2015) 249265

Contents lists available at ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf

Credit rationing and relationship lending. Does rm size matter?

Stefano Cenni a,e, Stefano Monferr b, Valentina Salotti c,, Marco Sangiorgi d, Giuseppe Torluccio a,e
a
Department of Management, University of Bologna, Via Capo di Lucca 34, 40126 Bologna, Italy
b
Department of Management, University of Naples Parthenope, Via A. Ferdinando Acton 38, 80133 Naples, Italy
c
Department of Finance, Iowa State University, 3345 Gerdin Business Bldg., Ames, IA 50014, USA
d
EN.E.R. Trading S.p.A. and Department of Management, University of Bologna, Via Capo di Lucca 34, 40126 Bologna, Italy
e
CEFIN Centro Studi Banca e Finanza, University of Modena and Reggio Emilia, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Using survey based data, we investigate factors inuencing credit rationing within a bank-based nancial
Received 7 July 2011 system. We show that rationing depends on various dimensions of the rm-bank relationships and that
Accepted 16 December 2014 the effects of relationship lending on rationing are not identical for different rm size groups. Multiple-
Available online 29 December 2014
banking increases the probability of rationing for small and large rms. Debt concentration with the main
bank affects positively smaller rms, while the opposite is true for large companies. The length of the
JEL classication: relationship with the main bank decreases the probability of rationing for both groups, but more so
G20
for large rms endowed with more bargaining power. Finally, the rm-bank spatial proximity, measured
G21
G32
by the headquarters vicinity, does not affect the rms access to credit.
L23 2014 Elsevier B.V. All rights reserved.

Keywords:
Credit rationing
Relationship banking
SME
Firm nancing
Probit sample selection models

1. Introduction
Credit rationing has been the subject of academic debate for
more than fty years. Its existence, a possible result of loan market
imperfections or asymmetric information, has been investigated at
length by research in nancial economics.
Within nancial systems where bank nancing is prevalent, i.e.
bank-oriented systems, the nature of the rm-bank relationship
is logically understood as the primary determinant for the imple-
mentation of and extent of credit rationing. Studies of relationship
lending (RL) utilize the depth and duration of the relationship
between the bank and the borrower as a measure of information
production which is a way to overcome information asymmetries
and also ease the rms access to credit (Petersen and Rajan, 1994,
1995; Berger and Udell, 1995, Degryse and Van Cayseele, 2000;
Ongena and Smith, 2001).
We show empirically that the effect of RL on small and large
borrowers is more nuanced than predicted by prior studies. In par-
ticular, we note that RL has non-linear effects on the rms ability
to get credit and is signicantly different across size groups. Bene-
ts of RL are more predictable on smaller borrowers, which are
generally characterized by more information opacity as well as
less bargaining power with lenders than larger rms.
Our empirical analysis relies on survey-based information
about rm-bank relationships and rm access to credit for a com-
prehensive sample of Italian enterprises. Italy is the ideal setting to
investigate the effect of RL on credit rationing. Its nancial system
is essentially dominated by banks with strong ties to mostly pri-
vately-held rms and the availability of alternative sources of
funding, such as stock and bond markets, is limited to a relatively
narrow group of larger companies. Therefore, since alternatives to
traditional borrowing are either scarce or unavailable for the aver-
age Italian rm, a denial of bank credit will likely result in a limited
ability to pursue positive NPV projects and, ultimately, curb the
rms growth potential.1
Our empirical methodology offers several advantages with
respect to prior studies on credit rationing and RL. First, we use
survey data to identify rms that are subject to credit rationing

Corresponding author. Tel.: +1 (515) 2944201. 1

Herrera and Minetti (2007) and Minetti and Zhu (2011) offer a good description of
E-mail address: vsalotti@iastate.edu (V. Salotti). the bank-oriented nature of the Italian nancial system.

http://dx.doi.org/10.1016/j.jbankn.2014.12.010
0378-4266/ 2014 Elsevier B.V. All rights reserved.
250 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
in place of nancial-statement-based proxies used by numerous
other empirical studies. The structure of the survey allows us to
discriminate between rms needing additional nancing and the
actual denial of the credit application by the bank. Other studies
using survey-based data (Guiso, 1998; Berkowitz and White,
2004) do not distinguish instances in which a rm merely declared
itself in need of more credit from cases in which a rm was denied
credit. These studies estimate the probability of being credit
rationed based on a one-equation approach and dene credit
rationing by looking at whether the rm declared itself to be
under-funded. This weak denition of rationing includes rms
that were discouraged from applying for additional credit and
rms that needed more credit but were not willing/able to pay
the market interest rate.
In this paper we classify a rm as credit rationed if it applied for
credit and the credit application was denied. Our denition permits
us to explicitly consider that it is in fact the banks decision to deny
credit that generates credit rationing. Because the sub-sample of
companies that applied for credit is not random, we model the prob-
ability of being denied funding based on the rms propensity to need
additional credit. A similar methodological problem is explained in
detail by Greene (1992), who models consumer loan default and
credit card expenditures. We tackle this issue by using a Heckman
sample selection model adapted to dichotomous explanatory vari-
ables as rst proposed by Van de Ven and Van Praag (1981).
Our econometric approach is similar to Atzeni and Piga (2007)
and Minetti and Zhu (2011). Atzeni and Piga (2007) test whether
R&D-intensive rms are more or less credit constrained than other
types of rms and nd that rms engaged in innovation face greater
difculties in accessing credit. Minetti and Zhu (2011) investigate
whether the probability of exporting is different for credit-rationed
enterprises compared to those that do not endure similar nancing
constraints. Neither paper considers how the rm-bank relation-
ship could affect the probability of the rm being rationed.
We investigate the rm-bank relationship across four dimen-
sions: the number of banks serving the rm, the debt share held
by the main bank, the duration of the relationship with the main
bank, and the rm-main bank geographical proximity. We empir-
ically show that size has a direct effect on the probability of credit
rationing as banks behave differently depending on both the size of
their clients (Berger and Udell, 1995) and a mediated effect
which depends on the quality of the rms relationship to banks.
Our results show that the effects of RL on the probability of credit
rationing diverge depending on the size of the borrower.
We nd that the probability of credit rationing is positively
linked to the number of banking relationships that the rm main-
tains. Small and medium enterprises (SMEs) are more strongly
rationed than large rms. At the median level, 6 banking relation-
ships, the probability of rationing is 38.8% for SME and 16.3% for
large banks. SMEs are also more often rationed when the duration
of their relationship with the main bank is short (in terms of years).
Finally, the probability of being denied credit is highest when the
main banks debt share is below 35% of the SMEs total debt and
decreases afterwards. A strong reliance on the main bank by the
SME results in an increase in private information gains which sets
the main bank apart from other lenders competing for the same
borrower; this competitive advantage improves the rms chances
of furthering its credit. The opposite is true for large rms. An
increase in the share of debt held by the main bank decreases
the probability of being constrained at levels below 40% of total
debt exposure. Higher levels of exposure to the main bank are,
on the other hand, positively correlated to rationing. The informa-
tion transparency of large rms appears to allow banks to immedi-
ately take advantage of economies of scale.
A longer banking relationship makes it easier for a rm to
obtain credit. In particular, statistically-signicant differences
between small and large rms emerge after a period of seven
years; at that point, large rms gain a greater advantage from an
extension of the relationship. Measured at the median (about
15 years), the probability of rationing is 22% higher for SMEs. Sur-
prisingly, the rm-bank geographical proximity does not appear to
be statistically signicant.
Our analysis suggests that both large rms and SMEs gain an
advantage from forming close lending relationships with the effect
being more pronounced for SMEs. Specically, SMEs should aim to
concentrate their debt with the main bank in order to reduce the
risk of credit rationing.
The remainder of this paper is structured as follows. Section 2
discusses the theoretical background. In Sections 3 and 4, we
describe the dataset used and the variables included in the econo-
metric models. The econometric methodology is presented in Sec-
tion 5. The results are reported and explained in Section 6. We
conclude in Section 7.
2. Literature
2.1. The rm-bank relationship and credit rationing
Information asymmetries are intrinsic to credit markets and cen-
tral to the literature on nancial intermediation (Jaffee and Russell,
1976; Leland and Pyle, 1977; Stiglitz and Weiss, 1981; Fama, 1985;
Boyd and Prescott, 1986). Banks are able to reduce information
asymmetries thanks to their enhanced abilities of screening and
monitoring borrowers. Through relationships established with their
customers, banks are privy to proprietary information which allows
them to originate non-marketable loans and, thus, alleviate prob-
lems arising from the asymmetric distribution of information
between the agents in the credit market (Bhattacharya and
Thakor, 1993). This relationship also assumes that banks actually
gain an advantage over other credit providers based on the private
information collected. More specically, the repeated exchange of
private information over time strengthens the rm-bank relation-
ship allowing useful information to be accumulated by the bank
and therefore, reducing the borrower-lender information gap. Rela-
tionship banking directly aims at reducing asymmetric information
with borrowers by enhancing information reusability over time and
by resolving free riding problems (Boot, 2000; Albareto et al., 2011).
In addition, technological advancements have recently allowed
banks to transform relationship-based soft information into
hard, easy-to-transfer data, leading to cost savings and improv-
ing lending to more opaque borrowers. A few recent studies have
examined the relationship between rm size and the use of screen-
ing and monitoring technologies to improve lending. Cole et al.
(2004) show that while a cookie cutter approach to lending is
preferred for large rms and favours large banks, a character
approach should be used for smaller borrowers and is easier to
implement by small, local banks. DeYoung et al. (2008) show that
cost savings associated with credit scoring allow lenders to extend
lending to riskier borrowers and to take better credit decisions in
long-distance situations. Finally, Berger and Black (2011) nd that
the comparative advantages of large banks in hard technologies are
not all increasing in rm size and, while small banks appear to
have an advantage in relationship lending, this advantage is stron-
ger for lending to larger borrowers.
Concluding, the strength of the relationship with banks is a key
determinant of the presence and intensity of rm credit rationing.
Since SMEs are typically limited to bank loans in their external
means of nancing, factors inuencing the banks credit allocation
decision have a stronger economic consequence for these borrow-
ers. Small businesses could remain bound to the same main bank
due to high switching costs, i.e. costs associated with changing
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
banks. This could result in the rms capture by its main lender
(Sharpe, 1990; Kim et al., 2003) which may limit its access to alter-
native funding sources, and result in either a constraint on the
amount of credit provided or the application of higher interest
rates.
The aim of this paper is to illustrate how variables characteriz-
ing the typical rm-bank relationship affect the probability of a
rm being denied further credit. The variables considered in this
paper to describe the rm-bank tie are (i) the duration of the rela-
tionship with the main bank, (ii) the number of bank relationships
per rm, (iii) the share of corporate debt provided by the main
bank and (iv) the distance between banks and rms.
2.2. The duration of the relationship with the main bank
Relationship duration was rst recognized in theoretical models
addressing the issue of the choice of funding (e.g., Boot and Thakor,
1994; Petersen and Rajan, 1994; von Thadden, 1995). The opportu-
nity of nalizing a series of repeated transactions over time is a
way to recoup the initial screening costs sustained by the bank
and to make the investment in information collection worthwhile
for the bank. Since the incumbent bank accumulates private infor-
mation to better assess the rms ability to repay its debts, even in
a competitive market, the value of the rm-bank relationship
grows over time (Sharpe, 1990). The incumbent bank can, there-
fore, offer to its customers the best possible lending terms as well
as greater availability of credit, reducing the attractiveness of offers
from competing banks. Moreover, the rms willingness to leave
the relationship is also limited by the fact that less private and reli-
able information is available to competing banks.
If on one hand, the value of the information capital built up over
time and the desire to keep an exclusive relationship with the bor-
rower could lead banks to favour stable banking relationships. On
the other, a bank could be less sensitive to the borrowers nancial
needs since its information advantage over its competitors limits
the borrowers nancing option. The existing empirical evidence
on the effect of relationship duration does not provide an unequiv-
ocal answer. Cole (1998) nd that, while a pre-existing relationship
with the lender increases the likelihood of receiving additional
credit, the duration, per se, is unimportant. Angelini et al. (1998),
Hernandz-Canovas and Martnez-Solano (2007), Elsas et al.
(2004) and Bharath et al. (2011) nd that lengthening the relation-
ship with the main bank reduces the probability of credit rationing,
albeit, it also increases the interest rates charged by the bank.
Finally, according to Petersen and Rajan (1994) and Degryse and
Van Cayseele (2000), longer relationships are not always associ-
ated with lower costs or improved contract terms.
2.3. Number of banks with which a rm has relationships
The existing literature provides a mixed portrayal of the effect
of multiple banking on rms access to credit. Rajan (1992) and
von Thadden (1992) nd that having relationships with several dif-
ferent institutions can mitigate the rms hold-up risk. Bolton and
Scharfstein (1996) show that multiple lending reduces the rms
liquidation value and only rms with a good credit standing tend
to borrow from more than one entity. Moreover, a rm that bor-
rows from different lenders has lower risk of having to abandon
projects considered viable when banks face internal liquidity prob-
lems. This should encourage rms to open up to nancial interme-
diaries in order to mitigate the risk of a liquidity shortage
(Detragiache et al., 2000). On the other hand, Bris and Welch
(2005) argue that high credit quality rms choose to borrow from
fewer lenders to signal their ability to avoid bankruptcy since the
concentration of funding increases the bargaining power of credi-
tors. Bhattacharya and Chiesa (1995) and von Rheinbaben and
251
Ruckes (2004), highlight that the condentiality of the relationship
between a bank and borrower attracts businesses investing in R&D,
those with a higher credit worthiness, and reduces credit rationing.
Focusing on bank relationships, Dewatripont and Maskin (1995)
and Gobbi and Sette (2014) nd that the presence of a signicant
number of creditors complicates the renancing process and
makes lending less protable which deters banks from making
additional inefcient credit provisions. On the other hand, banks
could also prefer to jointly grant loans to the same subject in order
to diversify the risk of a liquidity crisis (De Bodt et al., 2005).
Carletti et al. (2007) analyze the incentives of sharing the lending
activity with competitors. Lending within a syndicate of banks
allows them to diversify their loan portfolio, but it also involves
a duplication of the monitoring and screening effort and potential
free-riding behaviour. However, when diversication benets
dominate costs, multiple lending encourages banks to maintain a
high level of overall monitoring which becomes an instrument
for mitigating the agency problems arising between banks and
their depositors. This, in turn, allows banks to increase their prots.
2.4. Share of debt held by the main bank
A signicant part of the literature on RL assumes that a rms
debt is equally-distributed among lenders. Situations where debt
is unequally distributed or even highly concentrated among one
or a few banks, although, are anything but unusual. The theoretical
literature focuses on the effects of the concentration of rm cred-
itors on the availability of credit. Among these, Elsas et al. (2004)
and Hubert and Schfer (2002), highlight how such concentration
contributes to intensify the hold-up problem. Empirical literature
does not reach unequivocal conclusions on the causes and effects
of concentration. According to Guiso and Minetti (2010), banks
prevent troubled companies from defaulting in order to take pos-
session of their assets during the debt restructuring process. For
this reason, rms with a high intrinsic asset and reuse value seek
to borrow from banks in a very fragmented way and have a lower
probability of being credit rationed.
2.5. Geographical proximity to the rms main bank
Hard information is easily transferred through the organiza-
tional structure of a banking institution, but soft information is
much more difcult to standardize and communicate (Stein,
2002). This assumption has recently been questioned as a result
of technological innovation associated with credit scoring. The
proximity of the bank to the rm was initially intended as phys-
ical proximity between locations. Benets associated with spatial
nearness were the lower information transfer costs, useful in both
the screening and monitoring phases, as well as the increased
availability of information related to the environmental context
in which the rm operates. Small rms, very opaque from an infor-
mational point of view, tend to borrow from nancial intermediar-
ies that are geographically close to them and are also modest in
terms of size (Saunders and Allen, 2002; Agarwal and Hauswald,
2010). Knyazeva and Knyazeva (2012) and Cotugno et al. (2013)
nd that banks face higher costs in collecting information about
non-local borrowers which is reected in loan terms. Moreover,
banks are more likely to require collateral when borrowers are
located far away from them. More recently, the concept of proxim-
ity between the bank and the customer has been reinterpreted in
terms of functional distance. According to Alessandrini et al.
(2009a,b), not only the purely spatial, but also cultural and social
distance, should be taken into account. In particular, the distance
of the rm from the banks decision-making center is crucial in
the lending process.
252 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
3. Data
Our empirical analysis is based on the IX Indagine sulle Imp-
rese Manifatturiere, a survey carried out by the banking group
CapitaliaUnicredit at the beginning of 2004. The dataset consists
of a stratied sample of Italian rms with at least 11 and up to
500 employees. The stratication is based on the number of
employees, sector composition, and location, which guarantees
that smaller rms are well-represented in the sample. The Uni-
creditCapitalia survey represents the most comprehensive inves-
tigation of the Italian industrial sector. The survey is primarily used
by the Unicredit banking group for strategic decision making and
provides a tremendous amount of information on investment
trends, use of technology, import and export strategies, research
and development, as well as the relationship between Italian pri-
vate and public companies and the banking system. The survey
contains two parts: the rm response to a detailed questionnaire
and the corresponding nancial statement information for the
years 20002003. Among others, Herrera and Minetti (2007) and
Minetti and Zhu (2011) use the same survey to investigate the
effect of a rms credit constraints on its ability to innovate and
export. Detragiache et al. (2000) utilize an earlier version of the
survey to investigate the use and the determinants of debt nanc-
ing by Italian SMEs.
The survey gives ample attention to the rm-bank relationship
and to credit rationing. For each rm, we have information on (a)
the number of banks nancing the rm, (b) the length (in years)
of the relationship with the main bank, (c) the share of debt held
by the main bank, and (d) whether the main banks headquarters
is located in the same province as the rms.2 We also know
whether the rm declares itself under-funded and whether it has
been denied further bank nancing. Sections 4.14.3 explain in more
detail how the dataset is used for our testable hypotheses. Our sam-
ple consists of 4289 survey respondents, of which 2217 were small
rms with fewer than 50 employees, 1584 medium-sized rms with
fewer than 250 employees, and 488 large rms.3 Financial statement
information was cross-checked for accuracy and completeness with
the AIDA database of Bureau van Dijk.
4. Variable denition and hypothesis development
4.1. Measuring credit rationing
We identify credit rationing based on the rms answer to two
questions in the CapitaliaUnicredit survey:
(1) In 2003, had the company desired more credit at the mar-
ket interest rate? (if yes, the rm is declaring it wanted more
credit).
In case of a positive answer, the rm answers to the second
question:
(2) In 2003, did the company demand more credit without
obtaining it? (if yes, the rm is declaring that it applied for
credit and was denied it).
We consider the rms credit to be rationed if the rm
responded yes to both questions. Capitalias survey denes this
condition as strong credit rationing.4
2
Italian provinces are conceptually similar to U.S. counties.
3
EU size classication.
4
Questions 1 and 2 are not specic to the main bank, but refer to all the
relationships the rm has at the time of the survey. The questions are translated from
the Italian language. For continuity with prior research we kept our denition
consistent with Becchetti et al. (2010).
Other studies (i.e. Becchetti et al., 2010) using survey-based
data do not distinguish between instances in which the rm
declared itself in need of more credit, which would be reected
in question (1), and the actual rejection of its credit application
by the bank, which would be reected in question (2). These stud-
ies dene credit rationing based on the rm declaring an un-met
need for credit. Conceptually, the researcher classies rms as
rationed if, despite needing additional funding, they choose not
to apply for a loan.5 We do not argue against this denition, but
we believe that a more precise denition of credit rationing better
serves the scope of our investigation. As the aim of our study is to
investigate how the interplay between rms and nancial interme-
diaries is reected in the probability of credit rationing, we assume
that credit rationing should be identied as the banks decision
to deny credit to an under-funded rm.
The notation utilized in the paper is the following. Credit ration-
ing is a function of two events: (1) the rms need for additional
credit, which we identify with the dichotomous variable M, where
M is equal to 1 if the rm declared it wanted further credit, and (2)
the banks denial of the new credit application, which we identify
with the dichotomous variable D, where D is equal to 1 if the rm
declared it had applied for a certain amount of credit and was
rejected. Thus, credit rationing occurs when both M and D are
equal to 1. The analysis of D should be made conditional on the
observation that M = 1, that is, D rms should be studied within
the sub-sample of rms wanting more credit. In our study, a dis-
couraged rm is likely to have declared that they wanted more
credit (M = 1), but will not be included among the constrained
rms as they were not denied further credit they did not ask for.
Section 5.1 explains in detail the methodology adopted.
4.2. Relationship-banking variables
We use four attributes that describe the rm-bank relationship
as a means of producing information about the borrower and
developed a set of hypotheses to be tested:
(a) The presence of multiple banking relationships number of
banks is used to test whether a greater number of banking rela-
tionships makes rationing more likely (Dewatripont and Maskin,
1995; Bhattacharya and Chiesa, 1995; von Rheinbaben and
Ruckes, 2004) or whether the contrary is true: increased competi-
tion among banks forces them to adapt to the rms requirements
(Rajan, 1992; von Thadden, 1992; Padilla and Pagano, 1997;
Detragiache et al., 2000; Presbitero and Zazzaro, 2011).
s H1: The probability of a rm being credit-rationed increases with
the number of bank relationships.
(b) The importance of the main bank to the rm i.e. the main
banks debt share is used to analyse whether the dominance of
the main bank on the rms debt exposure, improves or hinders the
availability of credit to the rm. We also take into account that the
relationship might not be linear as incentives might change at dif-
ferent levels of the banks exposure to the borrower.
s H2: The higher the share of debt held by the main bank, the higher
the probability of being credit rationed.
(c) The length of relationship with the main bank relationship
duration is used to test the effect of a longer relationship on the
probability of being rationed. As the contribution of additional
information on the rm is likely to become less useful in the
long term (Berger and Udell, 1995), we use the square root of the
5
For example, see Guiso (1998) and Berkowitz and White (2004).
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
number of years over which the rm has had a relationship with
the main bank.
s H3: The longer the relationship with the main bank, the higher the
probability of being credit rationed.
(d) The proximity of the main bank to the rm proximity. The
dichotomous variable takes a value of 1 when the bank and rms
headquarters are located in the same province (Alessandrini et al.
(2009a,b)). Our models test the effect of proximity on rationing,
and we expect greater proximity to reduce rationing, as the collec-
tion of private information is less costly to the bank. Alternatively,
it might also be possible that distant lenders eager to increase
their market share may be more willing to extend credit than local
banks in order to gain access to a new geographic area.
s H4: Geographical proximity to the main bank decreases the proba-
bility of being credit rationed.
Credit rationing is expected to be more pronounced for smaller
borrowers since they are more affected by information asymmetry
problems. Thus, we test the effect of size on the probability of
rationing, and we investigate whether the effect of RL is different
across different size groups. We distinguish small and medium
enterprises (SMEs) as having less than 250 employees from large
companies and test the following hypothesis.
s H5: Credit rationing is signicantly more pronounced for SMEs.
s H6: The effect of RL on credit rationing is different for SMEs and
Large companies.
4.3. Control variables
As the banks decision to grant credit depends to some extent
on hard information, our control variables reect the companys
nancial situation and how the rm is perceived on paper based
on nancial statement data. Several control variables were consid-
ered to isolate the rms characteristics that may inuence both
the probability of needing more credit and the likelihood of facing
nancial constraints. We categorize all control variables in: rms
characteristics, bank credit market characteristics, and rms gov-
ernance. The variable construction in reported in Appendix A.
4.3.1. Firms characteristics
To represent the rms current nancial standing and its evolu-
tion over time, we include both levels and changes in nancial
statement variables. Protability is measured by the level and
growth rate of the return on assets (ROA) and by the sales growth
rate, the former representing both a static and dynamic measure
of the ability to generate income and the latter measuring growth
opportunities. We would expect to see a negative relationship
between each of these variables, the likelihood of needing more
credit (M), and the likelihood of being denied further credit (D).
Net working capital to total assets is considered (both level and
change) to account for the rms liquidity. An increase in these
variables would be expected to lessen the requirement for short-
term nancial resources making any rationing less critical for the
rm. Firm creditworthiness is evaluated through the interest cover-
age ratio and the rms leverage. The interest coverage ratio
describes the ability to repay borrowing costs through ordinary
operations regardless of the level of debt. For leverage, we also
include its change in order to account for its dynamic impact on
rationing. Tangible assets, as a percentage of total assets, act as a
proxy for the ability to provide collateral which is expected to
reduce the rationing probability. An R&D expenditure dummy var-
iable is used to represent the expectation of greater rationing in the
253
case of innovation. North indicates that the rm is headquartered
in northern Italy. This geographical distinction is relevant because
Northern Italy is more industrialized than southern Italy. Industry
are industrial sector indicators following the industry classication
la Pavitt (Pavitt, 1984).
4.3.2. Characteristics of the banking market
The concentration of bank branches in the province HHI of
bank branches is represented by the HerndahlHirschman Index
which describes the degree of competitiveness of the local banking
market.6 DellAriccia and Marquez (2004) and Berger et al. (2004)
review models that relate bank competition, information, and credit
allocation. They show that monopolistic power provides incentives
to resolve asymmetric information problems and reduce credit
rationing. A high HHI value may lead to an increase in the riskiness
of the loan portfolio held by the average bank. The theoretical liter-
ature on the relationship between bank market power and rm
nancing constraints describes two mechanisms in which the com-
petition among banks may affect credit rationing (CarboValverde
et al., 2009; Ryan et al., 2014). The market power hypothesis argues
that increased market power results in decreased loan supply and
higher lending rates, thereby increasing credit rationing. The infor-
mation hypothesis (Petersen and Rajan, 1995) argues that market
power enables banks to increase the supply of credit and establish
a lending relationship that will create an informational rent in sub-
sequent periods.7
4.3.3. Corporate governance
The rms corporate governance is measured by the Firm own-
ership which represents the degree of ownership concentration
among the top three shareholders. Atzeni and Piga (2007) nd that
a high concentration of decision-making power in the hands of a
single shareholder increases the attitude to pursue high-risk prof-
itable projects, the demand for credit, and the rationing
probability.
We also consider the rms reported level of Self-nancing
which indicates the ownership a priori attitude towards credit
and more in general external nancing. We expect that a high
use of Self-nancing will reduce the need of external funding, thus
reducing the probability of rationing.
5. Methodology
We dene a rm as credit rationed if it needed more credit
(M = 1) and, conditional to this, it applied for more credit but
was denied by the banks (D = 1). We observe D only when M = 1.
The marginal effect of the vector of independent variables in our
credit rationing equation has two components: rst, the direct
effect of our explanatory variables; second, the indirect effect of
the same explanatory variables on the probability of observing
M = 1. In other words, a change in the explanatory variables not
only changes the probability of observing D, but also the probabil-
ity that an observation is actually in the sub-sample (M = 1).
Because D is conditional to (M = 1), our credit rationing examina-
tion is based on a non-random sample.
To tackle the sample selection problem, we employ a bivariate
Probit model with selection, a variant of the Heckman Selection
Model (1989) that accommodates a dichotomous variable in the
outcome equation.8 The model has two equations: an outcome
Eq. (4.1) in which we model the probability of being denied credit
6
The HerndahlHirschman Index was provided upon our request by Bank of Italy.
7
Ratti et al. (2008) using investment data for a panel of European listed rms, nds
evidence in support of the information hypothesis.
8
This model was initially developed by Van de Ven and Van Praag (1981) and has
been widely employed in several other applications (see Greene, 2012).
254 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
(D), which is the censored sample since we only observe D if M = 1,
and a selection Eq. (4.2) in which we model the probability that
the rm needs more credit (M) in the observed sample.
Outcome equation:
D xb u; D 1 if D > 0; 0 otherwise
Selection equation:
M  zc v ; M 1 if M > 0; 0 otherwise
      
u  0 1 q
 x; z  N ; 4:3
v 0 q 1
x and z represent the vectors of the explanatory variables for D and
M, b and c the vectors of parameters, and u and v the error terms. D
and M are the index functions of the Probit models. If the correla-
tion q0, estimates of the D equation will be biased unless we
account for selection. In other words, the model-building process
for D is constructed from a non-random sample of rms, raising
the issue of conditionality and, thus, producing systematically
biased parameters (Greene, 2012).
In terms of probabilities, the outcome equation probability of
being denied credit (D), given that more credit (M) was needed,
is expressed as follows:
ProbDi 1jMi 1 Probxi b ui > 0jM i 1
This methodology uses a maximum likelihood estimation approach
in which the robust Huber/White estimator of the variance is used.
The log-likelihood function of the sample is given by:
X senting the above mentioned rms characteristics; Model 2 con-
ln L ln U2 xi b; zi c; q
Mi 1;Di 1
X X
ln U2 xi b; zi c; q lnUzi c
Mi 1;Di 0 M i 0
where U2 : represents the joint cumulative probability distribution
function with q Corru; v  and U: the cumulative distribution
function of a standard normal random variable. If q = 0, then the
log-likelihood for the probit model with sample selection is equal
to the sum of the probit models for D and M.9
5.1. Sample and empirical model specication
5.1.1. Descriptive statistics
Tables 1 and 2 contain descriptive statistics and t-tests for mean
difference for the independent and control variables for the full
sample, for rms that declared they wanted more credit (M = 1),
and for rms that had applied for more credit but were denied
(D = 1). Firms declaring the need for additional credit (M = 1) and
the group of credit constrained rms (D = 1) are smaller (total
assets) and more leveraged than the rest of the sample (81% and
84% compared to 73% for the full sample). The level of ROA is also
lower: 3.9% mean for M = 1 and 3.2% for D = 1 compared to 6.3% for
the full sample (Tables 1 and 2). In addition, M = 1 rms tend to
invest more in tangible assets and less in working capital are, on
average less creditworthy, are located in less competitive local
banking markets, and have stronger ties with their main bank
(Table 2). D = 1 companies are less liquid, but also show a lower
attitude toward nancing their investments with internal capital
(self-nancing) and lower sales growth.
5.1.2. Empirical model
The aim of our work is to analyse rm-bank relationships and
the effect on the credit rationing probability with a particular
9
A Wald test for independence was performed to test the null hypothesis that
q = 0.
emphasis on rm size. For this purpose, we link the relationship
banking variables to rm size via interaction terms. The model
specication for the D and M equations are as follows:
4:1 D b0 dD0 SME b14 RL14 dD14 SME  RL14
bn controlsn u 5:1
4:2
M c0 dM M
0 SME c14 RL14 d14 SME  RL14
cm controlsm v 5:2
where SME is our size indicator, taking the value of 1 for small and
medium enterprises and 0 otherwise. RL14 in the outcome equa-
tion D and in the selection equation M represent our four RL prox-
ies, and b14 and c14 are their respective coefcients. SME RL14
are the four interaction terms between the RL proxies and our size
indicator (SME), and dD14 , dM
14 are their coefcients. b0 and c0 are
intercepts. Controls is a vector of n (m) control variables.10
Since the banks decision to grant credit depends partly on
hard information, the set of control variables reects the com-
panys nancial situation and the rms perceived need for credit.
In particular, we account for protability (return on assets), asset
liquidity (net working capital to total assets), leverage (debt to
4:4 assets ratio), interest coverage (Ebit to Interest expenses), and
growth opportunities (sales growth rate).
Our empirical investigation builds on three models. Model 1
controls for both levels and growth rates of the variables repre-
siders only their levels; and Model 3 considers only the growth
rates. It is worth noting that Models 2 and 3 are nested in Model
4:5 1, or, in other words, they are restrictions of Model 1.
The use of the three models is interesting because it allows us to
check whether a bank looks at the evolution of a rms character-
istics over time, at a snapshot of the current situation, or it takes
both into account. Since Models 2 and 3 are nested in Model 1,
we can use Model 1 to test the joint signicance of the set of vari-
ables used as levels and growth rates.11 Test statistics show that
both sets of variables (levels and growth rates) present a high joint
signicance suggesting that banks take all of them into account in
their decisions to grant or deny credit. This conclusion leads us to
prefer Models 1 to 2 and 3, which better explain a rms need for
additional credit and the factors inuencing the banks decision to
deny or grant it.
We initially estimated a model with all four relationship vari-
ables interacting with the rms size (results are not reported for
brevity). The spatial proximity variable and its interaction term
were not signicantly different from zero, showing that the vari-
able and its interaction do not improve the goodness of t of our
model. We, therefore, decided to drop the spatial proximity inter-
action term. The exclusion of spatial proximity interaction term
improved the signicance of some of the other variables in the
model.12
10
In this type of model there is no obvious standard for dening the set of
regressors in Eqs. (5.1) and (5.2). Sartori (2003) suggest dropping one variable from
the outcome model while Greene (2012) indicates that the regressors in the two
equations may even coincide because there are no issues of identiability or
estimability in the bivariate Probit model. Consistent with empirical applications of
selection models in the nance literature, we drop one explanatory variable (R&D)
from the (5.1) regression after noting that it was not signicant in any of our models.
11
Model 2 is the restricted Model 1, where growth variables do not enter the model;
instead Model 3 is the restricted Model 1, where level variables do not enter the
model.
12
The improved signicance of other coefcients in the model is probably due to
some degree of multi-collinearity among the set of explanatory variables. However,
maintaining the spatial proximity variable in the model without interacting it with
size does not produce any change in our results.
Table 1
Descriptive statistics.

Full sample (obs = 2868) M = 1 sample (obs = 390) D = 1 sample (obs = 148)
Mean SD p5 Median p95 Mean SD p5 Median p95 Mean SD p5 Median p95
Firms characteristics
Total asset (ln) 9.051 1.280 7.161 9.007 11.342 8.828 1.157 7.070 8.758 10.850 8.910 1.201 7.057 8.982 10.842
Tangible asset 0.230 0.152 0.027 0.205 0.511 0.252 0.165 0.027 0.236 0.544 0.258 0.158 0.031 0.251 0.544
Leverage 0.732 0.184 0.386 0.767 0.970 0.806 0.153 0.472 0.838 0.991 0.842 0.130 0.599 0.872 0.993
Leverage (growth rate) 0.000 0.204 0.168 0.004 0.158 0.019 0.116 0.104 0.003 0.191 0.025 0.123 0.087 0.004 0.212
Net working capital to Total asset 0.131 0.211 0.197 0.115 0.498 0.055 0.197 0.237 0.050 0.378 0.019 0.198 0.319 0.022 0.308
Net working capital to Total asset (growth rate) 7.592 324.579 1.743 0.067 3.744 48.504 876.830 3.428 0.000 3.188 12.191 150.834 5.618 0.045 2.529
Sales (growth rate) 0.116 1.353 0.269 0.014 0.371 0.048 0.356 0.298 0.011 0.512 0.011 0.281 0.263 0.017 0.354

S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

Interest coverage 268.18 6209.17 0.263 4.697 83.359 6.834 63.948 1.389 2.640 20.224 3.815 13.364 0.204 2.264 8.599
Roa 0.063 0.083 0.031 0.054 0.191 0.039 0.083 0.095 0.045 0.132 0.032 0.078 0.095 0.047 0.113
Roa (growth rate) 0.407 11.874 1.422 0.045 2.355 1.645 26.341 1.679 0.033 2.252 0.127 5.076 1.552 0.099 1.800
Investment rate in tangible assets 14.191 255.291 0.378 0.023 1.180 9.838 120.445 0.488 0.019 1.653 2.820 28.962 0.608 0.016 1.411
R&D 47% 49% 52%
Traditional industry 52% 57% 58%
Scale intensive industry 17% 14% 16%
Specialized industry 27% 23% 23%
Hi-Tech industry 4% 5% 3%
Geographic area (North = 1) 67% 60% 62%
SME 92% 95% 96%
Corporate governance
HHI of rms ownership 0.600 0.257 0.333 0.500 1.000 0.591 0.249 0.333 0.500 1.000 0.611 0.253 0.333 0.506 1.000
Self-nancing 0.041 0.081 0.000 0.006 0.189 0.025 0.052 0.000 0.000 0.126 0.019 0.039 0.000 0.000 0.093
Bank market
HHI of bank branches 0.065 0.035 0.043 0.060 0.107 0.069 0.040 0.043 0.061 0.107 0.069 0.044 0.043 0.061 0.107
Relationship banking
Number of banks 5.879 3.431 2.000 5.000 12.000 6.077 3.410 2.000 5.000 12.000 6.635 3.816 2.000 6.000 13.000
Main banks debt share 31.538 25.117 0.000 30.000 80.000 37.493 21.574 10.000 30.000 80.000 35.064 20.236 10.000 30.000 70.000
Proximity (%) 57 58 55
Relationship duration (square root) 3.942 1.410 1.732 3.873 6.325 3.865 1.393 1.732 3.873 6.325 3.717 1.453 1.732 3.317 6.325

Mean, Standard deviation (SD), 5th percentile (p5), median and 95th percentile (p95) for the full sample and the subsamples M = 1 and D = 1.

255
256 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

Table 2
Mean difference for the M sample and the D sample.

Firm declares it needed more credit (M sample Firm applied for more credit and the credit was denied
Obs = 2868) (D Sample obs = 390)
Mean Difference t-Ratio Two-sided Mean Difference t-Ratio Two-sided
(No-Yes) p-value (NoYes) p-value
Firms characteristics
Total asset (ln) 0.258 4.020*** 0.000 0.115 0.946 0.345
Tangible assets 0.026 2.917*** 0.004 0.012 0.679 0.497
Leverage 0.087 10.083*** 0.000 0.055 3.697*** 0.000
Leverage (growth rate) 0.021 2.919*** 0.004 0.009 0.745 0.457
Net working capital to Total asset 0.088 8.086*** 0.000 0.057 2.797*** 0.005
Net working capital to Total asset (growth rate) 47.351 1.066 0.287 59.269 0.814 0.416
Sales (growth rate) 0.079 2.308** 0.021 0.058 1.686* 0.093
Interest coverage 21.703 6.543** 0.024 7.140 1.716* 0087
ROA 0.028 6.141*** 0.000 0.012 1.483 0.139
ROA (growth rate) 1.434 1.068 0.286 2.855 1.304 0.193
Investment rate in tangible assets 0.282 0.968 0.333 0.143 0.258 0.796
Corporate governance
HHI of rms ownership 0.010 0.764 0.445 0.041 1.554 0.121
Selfnancing 0.019 6.001 0.000 0.009 1.828* 0.068
Bank market
HHI of bank branches 0.005 2.274*** 0.023 0.000 0.086 0.932
Relationship banking
Number of banks 0.229 1.234 0.218 0.850 2.299 0.022
Main banks debt share 6.892 5.711*** 0.000 3.406 1.544 0.123
Relationship duration (square root) 0.089 1.169 0.243 0.230 1.565 0.119

Mean difference t-tests for the M sample (selection model) and D Sample (outcome model).
Signicance levels:
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

5.1.3. Interaction effects 5.1.4. Sample selection bias

The non-linearity of our regression models does not allow us to
directly interpret the interaction effects as we would in a linear
model. As pointed out by Allison (1999), tests on the equality of
coefcients across groups can lead to incorrect conclusions since
the test confounds the magnitude of the regression coefcients
with the amount of residual variation. As explained in Ai and
Norton (2003), Wooldridge (2006), Greene (2010), and Karaca-
Mandic et al. (2012) in a non-linear model, the coefcient of the
interaction term does not represent the change in the partial effect
of the interacted variable on the expected probability. Conse-
quently, while the computation of the partial effect of the interac-
tion term is useful in linear model, it provides no information in a
non-linear model. The interaction effects may also have different
signs for different values of the covariates, therefore, the sign of
the interaction terms coefcient does not necessarily indicate
the sign of the interaction effect and the statistical signicance of
the coefcient should not be tested with a simple t-test. Long
(2009) proposes testing the equality of the predicted probability
across groups at different levels of the variable of interest. Specif-
ically, he suggests examining group differences at multiple levels
of the explanatory variable. This approach is advantageous because
it explicitly considers that different levels of an explanatory vari-
able might have different effects on the predicted probability
across groups. Greene (2010) suggests that the graphical represen-
tation of these differences in predicted probabilities is one of the
most effective ways of describing interaction effects.
In Section 6, we report the graphical representation of the rela-
tionship between our RL proxies and the predicted probability of
rationing across SMEs and large companies. In particular, we test
the difference in the predicted probability of rationing at different
To evaluate the presence of sample selection bias, we compute
the correlation coefcients of the error terms in the M (selection)
and D (outcome) equations used in Models 13 (Table 3). Models 1
and 3 present highly signicant and similar values for the error
terms correlation coefcients (0.791 and 0.844, respectively),
while Model 2 has a value of 0.578 and lower signicance. We per-
form a Hausman test to compare the differences among the correla-
tion coefcients estimated in the three models, and we nd that they
are not statistically signicant.13 Therefore, the correlation terms
seem to be different in absolute, but not in statistical, terms. The dif-
ference in absolute value can be due to multi-collinearity affecting
Model 2 more than 1 and 3, which causes a less precise estimation
of all the parameters in the model including the estimated correlation
coefcients. Based on the residual analysis, we reject the null hypoth-
esis of independence between the error terms in the D and M which, as
expected, suggests the presence of a sample selection bias.
In conclusion, the coefcient estimates in the D equation could
be biased and inconsistent if no consideration is given to the corre-
lation in the error terms. Further verication is provided by the
Hausman tests we employed to compare the estimates obtained
through the bivariate probit model with sample selection and a
single probit approach. Specically, considering the D equation,
we test the difference between the coefcients estimated through
the bivariate probit model and those obtained through the single
probit models. The same routine was performed for the M equation
and, nally, for each of the three estimated models. For Models 1
13
Hausman tests for correlation coefcient comparison:
qMod1  qMod2 0 With v2 0:40 and p-Value 0.5255

levels of our four RL proxies. The section also provides tabulated qMod1  qMod3 0 With v2 0:26 and p-Value 0.6123
qMod2  qMod3 0 With v2 0:66 and p-Value 0.4155
results for the predicted probabilities across SMEs and large rms
which are used in the graphical analysis.
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265 257

Table 3
Credit rationing and relationship banking.

Model 1 Model 2 Model 3

Level, growth Only growth Only level
D M D M D M
Outcome Selection Outcome Selection Outcome Selection
Firms characteristics
Total asset (ln) 0.096 0.114 0.081 0.133 0.056 0.093
[1.52] [3.20]*** [1.40] [3.98]*** [0.67] [2.92]***
Tangible assets 0.046 0.683 0.074 0.644 0.309 0.697
[0.09] [2.59]*** [0.22] [3.12]*** [0.39] [2.67]***
Leverage (growth rate) 1.035 0.281 0.582 0.236
[1.66]* [1.40] [1.27] [1.48]
Leverage in 2001 0.553 1.058 0.784 1.142
[0.73] [3.56]*** [0.59] [3.96]***
Net working capital to Total asset (growth rate) 0.000 0.001 0.000 0.001
[2.74]*** [1.70]* [3.45]*** [1.97]**
Net working capital to Total asset 2001 0.096 0.060 0.108 0.034
[0.22] [0.24] [0.22] [0.14]
Sales (growth rate) 0.255 0.094 0.205 0.056
[1.13] [2.47]** [1.18] [2.65]***
Interest coverage ratio 0.004 0.001 0.001 0.002 0.002 0.001
[1.08] [0.78] [0.24] [1.45] [0.56] [0.83]
Interest coverage ratio2 0.000 0.000 0.000 0.000 0.000 0.000
[1.31] [0.77] [0.94] [1.44] [0.99] [0.82]
ROA (growth rate) 0.008 0.005 0.008 0.005
[1.82]* [2.03]** [1.85]* [2.17]**
ROA 2001 0.886 1.714 0.183 1.269
[0.96] [3.00]*** [0.15] [2.61]***
Propensity to invest in tangible assets 0.001 0.000 0.001 0.000 0.001 0.000
[2.06]** [0.47] [1.93]* [0.55] [1.23] [0.66]
R&D 0.124 0.114 0.133
[1.86]* [1.79]* [1.99]**
Scale intensive 0.102 0.134 0.098 0.129 0.074 0.143
[0.65] [1.50] [0.70] [1.47] [0.38] [1.59]
Specialized suppliers 0.017 0.042 0.011 0.081 0.035 0.052
[0.13] [0.54] [0.09] [1.05] [0.22] [0.67]
Science based 0.480 0.147 0.367 0.131 0.581 0.155
[1.65]* [0.96] [1.35] [0.86] [1.84]* [1.03]
Geographic area 0.131 0.168 0.176 0.128 0.157 0.168
[1.05] [2.42]** [1.53] [1.87]* [1.04] [2.42]**
SME 0.162 0.146 0.092 0.218 0.150 0.259
[0.13] [0.29] [0.08] [0.44] [0.10] [0.52]
Corporate governance
HHI of rms ownership 0.279 0.060 0.214 0.136 0.295 0.054
[1.17] [0.47] [0.98] [1.10] [0.99] [0.43]
Self-nancing 0.000 0.000 0.000 0.000 0.000 0.000
[0.02] [2.17]** [0.090] [3.01]** [0.42] [2.32]**
Bank market
HHI of bank branches 0.615 1.146 0.556 1.129 0.318 1.072
[0.46] [1.47] [0.44] [1.45] [0.19] [1.37]
Bank relationship
Number of banks 0.283 0.010 0.241 0.007 0.341 0.019
[2.33]** [0.50] [2.33]** [0.34] [1.98]** [0.88]
Number of banks  SME 0.287 0.038 0.247 0.046 0.333 0.042
[2.53]** [1.64] [2.54]** [2.00]** [2.16]** [1.74]*
Main banks debt share 0.072 0.019 0.055 0.021 0.081 0.015
[1.70]* [1.26] [1.38] [1.42] [1.59] [1.00]
(Main banks debt share)2 0.001 0.000 0.001 0.000 0.001 0.000
[1.96]* [1.06] [1.69]* [1.18] [1.88]* [0.97]
Main banks debt share  SME 0.052 0.004 0.033 0.005 0.063 0.008
[1.17] [0.22] [0.80] [0.35] [1.09] [0.50]
(Main banks debt share)2  SME 0.001 0.000 0.001 0.000 0.001 0.000
[1.54] [0.11] [1.24] [0.01] [1.45] [0.02]
Spatial proximity 0.059 0.039 0.054 0.042 0.031 0.025
[0.55] [0.62] [0.54] [0.67] [0.25] [0.40]
Relationship duration (square root) 0.595 0.054 0.558 0.041 0.686 0.052
[1.76]* [0.71] [1.76]* [0.51] [1.46] [0.70]
Relationship duration (square root)  SME 0.561 0.050 0.526 0.058 0.645 0.046
[1.68]* [0.62] [1.71]* [0.70] [1.38] [0.60]
Constant 0.220 1.415 0.959 0.519 0.133 1.540
[0.14] [2.04]** [0.78] [0.86] [0.06] [2.33]**
Sample
Observations 2868 2871 2868
Censored observations 2478 2481 2478

(continued on next page)

258 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

Table 3 (continued)

Model 1 Model 2 Model 3

Level, growth Only growth Only level
D M D M D M
Outcome Selection Outcome Selection Outcome Selection
Statistics
Wald v2 87.506 43.982 92.836
p-Value <0.01*** 0.011** <0.01***
Rho 0.791 0.578 0.844
Wald test of independence 3.685 0.492 6.500
p-Value of test of independence 0.055* 0.483 0.011**
Hausmans test 102.81*** 19.340 32.240 6.830 18.02*** 19.750
Log-likelihood
Log-likelihood 1261.41 1278.07 1294.61

Bivariate probit model with sample selection. Models 13 report the maximum likelihood estimates for the Selection (M) and Outcome (D) equations. z-Scores are reported in
squared brackets. All variables are dened in Appendix A.
Signicance levels:
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

Table 4
Relationship banking: probability of rationing and multiple lending (Number of banks) for SME and large rms.

Number of banks equal to: Selection model P(M = 1) Outcome model P(D = 1|M = 1)
Predicted probability of Needing more Differences in the Predicted conditional credit rationing Differences in the
credit conditional probability probability conditional probability
of Needing more credit of credit rationing
between large rms and between large rms and
SME SME
SME (SME = 1) Large rms (SME = 0) dy/dx z-Score SME (SME = 1) Large rms (SME = 0) dy/dx z-Score
0 0.107 0.139 0.032 0.62 0.348 0.014 0.334 5.35***
1 0.112 0.137 0.025 0.53 0.355 0.021 0.334 6.01***
2 0.117 0.135 0.018 0.41 0.362 0.032 0.329 6.57***
3 0.122 0.133 0.011 0.28 0.368 0.049 0.319 6.71***
4 0.127 0.131 0.004 0.11 0.375 0.075 0.300 6.17***
5 0.132 0.129 0.003 0.09 0.382 0.112 0.269 5.10***
6 0.137 0.127 0.011 0.33 0.388 0.163 0.225 3.84***
7 0.143 0.125 0.018 0.59 0.395 0.227 0.168 2.56***
8 0.149 0.123 0.026 0.87 0.401 0.303 0.098 1.33
9 0.154 0.121 0.033 1.14 0.408 0.389 0.019 0.23
10 0.160 0.119 0.041 1.4 0.414 0.480 0.065 0.67
11 0.167 0.117 0.049 1.62 0.421 0.572 0.151 1.37
12 0.173 0.116 0.057 1.79* 0.427 0.660 0.233 1.95*
13 0.179 0.114 0.065 1.92** 0.434 0.740 0.306 2.49**
14 0.186 0.112 0.074 2.01** 0.440 0.808 0.369 3.03***
15 0.192 0.110 0.082 2.08** 0.446 0.864 0.418 3.56***
16 0.199 0.108 0.091 2.12** 0.452 0.907 0.454 4.00***
17 0.206 0.107 0.100 2.15** 0.458 0.938 0.479 4.26***
18 0.213 0.105 0.108 2.17** 0.465 0.959 0.494 4.29***
19 0.221 0.103 0.117 2.18** 0.470 0.973 0.503 4.15***
20 0.228 0.102 0.127 2.18** 0.476 0.982 0.506 3.92***
21 0.236 0.100 0.136 2.18** 0.482 0.988 0.505 3.66***
22 0.243 0.098 0.145 2.18** 0.488 0.991 0.503 3.42***
23 0.251 0.097 0.154 2.18** 0.494 0.994 0.500 3.20***
24 0.259 0.095 0.164 2.18** 0.499 0.995 0.496 3.00***
25 0.267 0.094 0.174 2.18** 0.505 0.997 0.492 2.83***

The table reports the predicted probability of Needing More Credit P(M = 1) and the predicted conditional probability of credit rationing P(D = 1|M = 1) for SMEs and large rms
at selected levels of the variable of interest Number of banks. Levels of Number of banks range from one banking relationship to 25 which are respectively the variables
minimum and the maximum for the sample. dy/dx is the difference in predicted probabilities between large rms and SMEs.
Signicance levels:
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

and 3, the test results, displayed at the bottom of Table 3, highlight 6. Results
the signicant differences in the estimates in the D equation
obtained through the two different approaches.14 These results 6.1. Relationship banking variables
reinforce the evidence of the presence of an estimation bias if a sin-
gle probit approach is used and highlights the need to account for 6.1.1. Multiple banking
sample selection. Using the methodological approach suggested by Long (2009)
and Greene (2010), we investigate the effect of Number of banks
14
See footnote no. 12. mediated by size (interaction effect) on the probability of rationing
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
a Number fo Banks and Needing More Credit
Predicted Probabilities for SME and Large Firms
,3
,25
Probability
,2
,15
,1
0 5 10 15 20
Number of Banks
SME Large Firms
Note:
Predicted Probabilities of Needing More Credit for SME and Large Firms
at different levels of the Number of Banks variable
c Number of Banks and Being Denied Credit
Predicted Probabilities for SME and Large Firms
1
,8
Probability
,6
,4
,2
0
0 5 10 15 20
Number of Banks
SME Large Firms
Note:
Predicted Probabilities of Being Denied Credit for SME and Large Firms
at different levels of the Number of Banks variable
Fig. 1. (a) and (b) Number of banks and probability of Needing More Credit Prob(M = 1). (c) and (d) Number of banks and conditional probability of Being Denied Credit
Prob(D = 1|M = 1).
by estimating the differences in the predicted probabilities
between SMEs and Large rms at different levels of Number of
banks. The empirical results in Table 3 show that the number of
banks has a positive and signicant effect on the rms probability
of being denied credit (outcome equation, Model 1). The same var-
iable does not have any effect on the probability of needing credit
(selection equation, Model 1). Since we dene a rm as credit
rationed if it declared the need of more credit and the banks
rejected its credit application, we analyze the effect of Number of
Banks, rst, on the probability of needing more credit P(M = 1),
and, second, on the conditional probability of being denied credit
P(D = 1|M = 1), i.e. credit rationing. Table 4 reports the predicted
probabilities for P(M = 1) and P(D = 1|M = 1) for SMEs and large
rms at selected levels of the variable of interest Number of banks.
Specically, levels of Number of banks range from one to 25 bank-
ing relationships which are, respectively, the variables minimum
and maximum. dy/dx is the difference in predicted probabilities
across SMEs and Large rms. The probability of needing more
credit is increasing in the number of banking relationships
(Fig. 1a), and the difference is not statistically signicant for bank-
ing relationships below 12 which corresponds to the 95th percen-
tile of the Number of banks distribution (Fig. 1b). For SME and large
rms, the probability of rationing increases with the number of
banks. At the mean level of Number of Banks (around 6 banks),
the probability of being credit rationed is around 38.8% for SME
and 16.3% for large rms.
The curve representing the credit rationing probability for large
rms crosses the curve for SMEs when the number of banks per
rm is around 9 (Fig. 1c). The difference between these two prob-
abilities, i.e. the interaction effect between SME and Number of
259
b Number of Banks and Needing More Credit
Difference in Probabilities Between SME and Large Firms
Difference in Probability
,3
,2
,1
0
-,1
0 5 10 15 20 25
Number of Banks
25
Difference Upper/Lower Confidence Interval
Note:
Difference in the predicted probabilities of needing more credit between
SME and Large Firms at different levels fo the Number of Banks
variable (95% confidence interval)
d Number of Banks and Being Denied Credit
Difference in Probabilities Between SME and Large Firms
,5
Difference in Probability
0
-,5
-,8
0 5 10 15 20 25
Number of Banks
25
Difference Upper/Lower Confidence Interval
Note:
Difference in the predicted probabilites of being credit denied between
SME and Large Firms at different levels of the Number of Banks
variable (95% confidence interval)
Banks, is signicantly different from zero (95% condence) up to
about seven banks per rm as shown in Fig. 1d. This means that,
all conditions being equal, banks are signicantly less willing to
extend credit to SMEs (compared to large rms) only when the
number of banks per rm is below 8.
For large rms, the probability of being credit rationed is statis-
tically different from zero only for rms with relationships with
over four banks. Although, we refrain from over-interpreting this
result since very few large rms have less than four bank relation-
ships. For SMEs, the probability of rationing is always far from zero.
Empirical results are in line with the expected effect of multiple-
banking on credit rationing (H1). Multiple rm-bank relationships
are implicitly weaker than a single relationship. These results are
consistent with prior studies (e.g. Petersen and Rajan, 1994;
Berger and Udell, 1995; Cole, 1998; Harhoff and Krting, 1998;
Degryse and van Cayseele, 2000) suggesting that the rms attempt
to increase the number of lenders hinders the access and cost of
external nancing. The presence of multiple creditors makes the
process of renancing more competitive and less protable for
banks, curbing the extension of additional credit.
For large enterprises, the probability of rationing grows quickly
for values between the 10th and 75th percentiles of Number of
banks, which correspond to three and ten banks, respectively.
According to Degryse et al. (2009), large Italian companies are
characterized as having multiple relationships with lending insti-
tutions and constantly rotating their creditors. If banks recognize
this rms behaviour, they will likely respond by rationing credit
rather than investing into gathering costly information over the
borrower (Dewatripont and Maskin, 1995). The interval in which
rationing increases most steeply in relation to number of banks
260 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

Main Bank's Debt Share and Needing More Credit Main Bank's Debt Share and Needing More Credit
a ,2 Predicted Probabilities for SME and Large Firms b Difference in Probabilities Between SME and Large Firms

,3
Difference in Probability

,2
,15
Probability

,1
0
,1

-,1
,05

0 20 40 60 80 100
Main Bank's Debt Share
0 20 40 60 80 100
Main Bank's Debt Share
Difference Upper/Lower Confidence Interval
SME Large Firms
Note:
Note: Difference in the predicted probabilities of needing more credit between
Predicted Probabilities of Needing More Credit for SME and Large Firms SME and Large Firms at different levels of the Main Bank's Debt Share
at different levels of the Main Bank's Debt Share variable variabile (95% confidence interval)

c Main Bank's Debt Share and Being Denied Credit

Predicted Probabilities for SME and Large Firms
d Main Bank's Debt Share and Being Denied Credit
Difference in Probabilties for SME and Large Firms

,5
,75

Difference in Probability
,6

0
Probability
,45

-,5
,3

-1
,15

0 20 40 60 80 100
0 20 40 60 80 100 Main Bank's Debt Share
Main Bank's Debt Share
Difference Upper/Lower Confidence Interval
SME Large Firms
Note:
Note: Difference in the predicted probabilites of being credit denied between
Predicted Probabilities of Being Denied Credit for SME and Large Firms SME and Large Firms at different levels of the Main Bank's Debt Share
at different levels of the Main Bank's Debt Share variable variable (95% confidence interval)

Fig. 2. (a) and (b) Main Banks Debt Share and probability of Needing More Credit Prob(M = 1). (c) and (d) Main Banks Debt Share and conditional probability of Being
Denied Credit Prob(D = 1|M = 1).

(i.e., below ten banks) contains 75% of the large rms and almost
all of the SMEs.
Concluding, SME and large companies benet from having a rel-
atively limited amount of credit relationships. A strong relation-
ship with the borrower creates signicant information rents for
the main bank (Sharpe, 1990), especially in a competitive market.
Therefore, it maximizes the rms chances of receiving credit when
needed.
Since the number of banking relationships and the role of the
main bank are often intertwined, we now discuss the effect of
the main banks debt share on credit rationing.
6.1.2. Relevance of main bank debt share
The extant empirical literature does not reach unambiguous
conclusions on the effects of debt concentration. As banks incen-
tives may change at different levels of skin in the game, we
investigate the possibility that the effect of debt concentration on
credit rationing may not be linear. Thus, we include in our model
the quadratic form of the Main Banks Debt Share variable.
Our empirical results show that the main banks debt share
exhibit signicant coefcients estimates in the outcome equation
of Model 1. Its square term is signicant in all estimated outcome
equations (Table 3). While Main Banks Debt Share is not a signi-
cant determinant of P(M = 1), the effect on the probability of need-
ing more credit is not monotonic for large rms and SMEs (See
Fig. 2a).
Size signicantly affects the trend and the shape of the relation-
ship between the Main Banks Debt Share and credit rationing. In
Fig. 2c, the curve representing the probability of rationing for large
rms is almost entirely below that of SMEs. Only when the main
bank provides around 75% of the debt share does the probability
of credit rationing for large rms exceed that for SMEs.15 The differ-
ence between the two probabilities is signicant when the main
banks debt share lies in the interval [15%, 60%], which approxi-
mately corresponds to the 25th and 90th percentiles in both the
SMEs and large rm sample (Table 5).
SMEs seem to be subject to a higher probability of credit ration-
ing at all level of the main banks debt share variable. The monoton-
ically decreasing curve of the SMEs rationing function is consistent
with the ability of the bank to fulll the credit needs of their cus-
tomers depending on the intensity of their relationship with the
borrower. Debt consolidation with the main bank allows the lender
to gain an information monopoly over competing lenders and to
spread transaction costs over multiple interactions with the bor-
rower, thus, gaining economies of scale to sustain the loan activity
(Petersen, 2004). Our results show that rationing probability
begins to ease immediately, both for SMEs and large rms, and
not after a given main bank share of debt. Presumably, even small
levels of consolidation beyond 1520% are sufcient to justify the
main banks costs of screening and monitoring informationally
opaque borrowers (Bongini et al., 2009) and make it worthwhile
to invest in the costly process of hardening soft information
(Petersen, 2004).
Contrary to SMEs, for large rms, the credit rationing probabil-
ity in Fig. 2c is U-shaped. Holding the other variables xed, an
increase in the debt share provided by the main bank initially low-
ers the probability that a large rm is credit constrained, but, as it
15
Although, this instance represents less than 5% of the sample of both SMEs and
large rms.
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265 261

Table 5
Relationship Banking: Probability of Rationing and share of debt held by main bank (Main banks debt share) for SME and large rms.

Main banks debt share at: (%) Selection model P(M = 1) Outcome model P(D = 1|M = 1)
Predicted probability of Needing more Differences in the Predicted credit rationing probability Differences in the
credit conditional probability conditional probability
of needing more credit of credit rationing
between large rms between large rms and
and SME SME
SME (SME = 1) Large rms (SME = 0) dy/dx z-Score SME (SME = 1) Large rms (SME = 0) dy/dx z-Score
0 0.066 0.082 0.015 0.37 0.477 0.405 0.072 0.39
5 0.081 0.095 0.015 0.38 0.464 0.331 0.134 0.96
10 0.095 0.109 0.013 0.35 0.452 0.272 0.179 1.72*
15 0.110 0.121 0.011 0.29 0.439 0.228 0.211 2.64***
20 0.125 0.132 0.007 0.19 0.427 0.196 0.230 3.59***
25 0.139 0.141 0.003 0.07 0.414 0.174 0.241 4.29***
30 0.151 0.149 0.003 0.06 0.403 0.159 0.244 4.58***
35 0.162 0.153 0.008 0.20 0.391 0.150 0.241 4.54***
40 0.171 0.156 0.015 0.34 0.380 0.147 0.234 4.28***
45 0.178 0.156 0.022 0.48 0.370 0.148 0.221 3.91***
50 0.182 0.153 0.029 0.62 0.360 0.155 0.205 3.44***
55 0.184 0.148 0.036 0.73 0.350 0.168 0.182 2.87***
60 0.183 0.141 0.042 0.8 0.341 0.188 0.153 2.20**
65 0.180 0.132 0.049 0.82 0.332 0.216 0.116 1.47
70 0.175 0.120 0.054 0.81 0.324 0.256 0.069 0.73
75 0.167 0.108 0.059 0.79 0.316 0.309 0.008 0.07
80 0.157 0.095 0.062 0.76 0.309 0.377 0.068 0.46
85 0.145 0.081 0.064 0.73 0.302 0.462 0.160 0.87
90 0.132 0.068 0.064 0.71 0.295 0.561 0.266 1.19
95 0.118 0.055 0.063 0.70 0.288 0.667 0.380 1.51
100 0.103 0.043 0.060 0.69 0.281 0.771 0.490 1.88*

The table reports the predicted probability of Needing More Credit P(M = 1) and the predicted conditional probability of credit rationing P(D = 1|M = 1) for SMEs and large rms
at selected levels of the variable of interest Main banks debt share. Levels of Main banks debt share are from 0% to 100% with 5% increments. dy/dx is the difference in predicted
probabilities between large rms and SMEs.
Signicance levels:
*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

Table 6
Relationship Banking: Probability of Rationing and Duration of the Relationship with main bank (Relationship Duration) for SME and large rms.

Selection Model P (M = 1) Outcome Model P(D = 1|M = 1)

Predicted probability of Needing more Differences in the Predicted credit rationing probability Differences in the
credit conditional conditional
probability of probability of credit
Needing more credit rationing from large
from large rms to rms to SME
SME
Relationship Duration at: SME (SME = 1) Large rms (SME = 0) dy/dx z-score SME (SME = 1) Large rms (SME = 0) dy/dx z-score
2 (p1) 0.135 0.103 0.032 0.75 0.428 0.545 0.117 0.69
3 (p5) 0.135 0.106 0.030 0.73 0.423 0.484 0.061 0.42
5 (p10) 0.136 0.111 0.025 0.68 0.415 0.393 0.022 0.21
10 (p25) 0.137 0.119 0.017 0.53 0.401 0.252 0.149 2.43
15 (p50) 0.137 0.127 0.011 0.33 0.389 0.170 0.220 3.91
24 (p75) 0.138 0.137 0.001 0.02 0.374 0.089 0.284 5.01
33 (p90) 0.139 0.147 0.008 0.16 0.361 0.051 0.309 5.63
40 (p95) 0.140 0.154 0.014 0.25 0.352 0.035 0.316 5.56
54 (p99) 0.141 0.166 0.025 0.34 0.336 0.020 0.317 4.66

The table reports the predicted probability of Needing More Credit P(M = 1) and the predicted conditional probability of credit rationing P(D = 1|M = 1) for SMEs and large rms
at selected levels of the variable of interest Relationship Duration. Levels of Relationship Duration are from 2 years to 54 years respectively 1st and 99th percentiles of the
distribution of the square root of Relationship Duration. dy/dx is the difference in predicted probabilities between large rms and SMEs.
Signicance levels:
**
p < 0.05.
***
p < 0.01.

reaches 40%, the probability of rationing begins to rise. Large rms
are typically less opaque than SMEs and it is, therefore, easier to
assess their creditworthiness. Hence, the rms access to credit is
not necessarily as dependent on strong ties between the rm and
its main bank. From our empirical analysis, pushing the share of
debt beyond the 40% threshold exposes a large rm to relatively
more intense risk of credit rationing. This is not a surprising result
if one considers that large rms need greater amounts of capital to
operate. The rms need for external nancing might be too bur-
densome to be satised by a single institution. If the credit expo-
262 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

a Relationship Duration and Needing More Credit

Predicted Probabilities for SME and Large Firms
b Relationship Duration and Needing More Credit
Difference in Probabilities Between SME and Large Firms

,2
,2

Difference in Probability

,1
,18

0
,16
Probability

-,1
,14

-,2
,12

-,3
,1

0 2 4 6 8 10
Relationship Duration
0 2 4 6 8 10
Relationship Duration
Difference Upper/Lower Confidence Interval
SME Large Firms
Note:
Note: Difference in the predicted probabilities of needing more credit between
Predicted Probabilities of Needing More Credit for SME and Large Firms SME and Large Firms at different levels of the Relationship Duration
at different levels of the Relationship Duration variable variable (95% confidence interval)

c Relationship Duration and Being Denied Credit

Predicted Probabilities for SME and Large Firms d Relationship Duration and Being Denied Credit
,8

Difference in Probability Between SME and Large Firms

,5
Difference in Probability
Probability
,6

0
,4
,2

-,8
0

0 2 4 6 8 10
0 2 4 6 8 10
Relationship Duration
Relationship Duration
SME Large Firms Duration Upper/Lower Confidence Interval
Note: Note:
Predicted Probabilities of Being Denied Credit for SME and Large Firms Difference in the predicted probabilites of being credit denied between
at different levels fo the Relationship Duration variable SME and Large Firms at different levels of the Relationship Duration
variable (95% confidence interval)

Fig. 3. (a) and (b) Relationship Duration and probability of Needing More Credit Prob(M = 1). (c) and (d) Relationship Duration and conditional probability of Being Denied
Credit Prob(D = 1|M = 1).

sure to a single company soars above a desirable level, the main
bank may be unwilling or incapable of providing further support
to the rm and is, therefore, forced to ration credit.
Alternatively, other banks may interpret the excessive debt con-
centration with the main bank as a negative signal (a soft-budget
constrain problem by the main bank) over the creditworthiness of
rm, thus, making these competing banks less willing to extend
credit (Dewatripont and Maskin, 1995).
6.1.3. The duration of the relationship with the main bank
The impact of the duration of the relationship between the rm
and its main bank is a signicant determinant in the probability of
being denied credit. While the probability of needing more credit is
not affected by the duration of the relationship with the main
bank, for large rms and SMEs, the probability of credit rationing
decreases with the duration (Table 6).
Fig. 3c shows that the probability for large rms decreases more
quickly and is almost always lower than that for SMEs. Although,
as indicated in Fig. 3d, the difference between the two curves is
statistically signicant only for a relationship period exceeding
9 years; the lower condence interval crosses the horizontal axis
at approximately 3 (the square root of 9).16 When the duration of
the relationship is 10 years, the probability of rationing is 40.1% for
SMEs and 25.2% for large rms (Table 6).
16
The distribution of rms in the sample with respect to the duration of the
relationship are [p25 = 7, p50 = 14, p75 = 24 years] for large enterprises and [p25 = 10,
p50 = 15, p75 = 24 years] for SMEs.
SMEs still bear greater likelihood of rationing, but a long rela-
tionship with the main bank loosens the restrictions on credit.
The effect is not as pronounced as for large companies for which
the collection, screening, and monitoring of information is easier.
This is in line with extant literature (Angelini et al., 1998). The
opportunity of implementing a series of repeated transactions over
time with customers strengthens the rm-bank relationship, and it
is a way to recoup the initial research and information-processing
costs. As the rm-bank relationship gradually consolidates, the
bank gathers more information, becomes more efcient, and
spreads its costs over a longer period of time (Boot and Thakor,
1994; Petersen and Rajan, 1994; von Thadden, 1995).
The accumulation of information over time does not follow a
linear path since the benets of an additional unit of information
are less than proportional over time. As shown in Fig. 3c and d,
the marginal effect of an increase in the duration of the relation-
ship on the probability of rationing decreases for large rms as
the duration increases. This means that the probability continues
to decline over time but at a slower rate.
Unfortunately, without information regarding the price of
credit, it is not possible to draw conclusions about the presence
of any hold-up behaviour in which the main bank exploits its
information monopoly to increase the interest rate on new loans.
However, the observed positive effect of bank duration on the
availability of funding leads us to assume that on average the cost
of funding did not change.
Ongena and Smith (2001) show that, as the relationship
between the bank and the rm is extended, the probability
that it will come to an end increases. Banks may be aware of this
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
behaviour and, in response, may be willing to extend credit in
order to prevent rms from going to competitors. The information
capital accumulated by the bank becomes more valuable over
time,17 and, since the process of collecting and analysing informa-
tion is costly, banks are very unlikely to ignore it, thus, granting bet-
ter contractual terms to the most reliable customers. This nding is
also in line with the decreasing, although less pronounced, trend of
rationing probability for SMEs (Fig. 3c and d).
SMEs, being more opaque and, therefore, more bound to their
main bank do not need as strong incentives as large rms to stay
in the relationship. Banks, being aware of the difculties encoun-
tered by SMEs in attracting new nancial resources, will not be
incentivized to offer them better credit terms (Sharpe, 1990). Con-
sistent with this interpretation, Gopalan et al. (2007) found that
the likelihood of not discontinuing (thus maintaining) a banking
relationship is higher for very opaque rms conrming the idea
of information rent being accumulated by the bank.
6.1.4. Firm-bank distance
The proximity of the bank to the rm presents no signicant
coefcients. The geographical distance between bank and rm
does not appear to affect rationing. However, it is worth noting
that in recent years the concept of proximity between banks and
rms has been partially reconsidered. According to Alessandrini
et al. (2009a,b), it is not necessarily the physical distance between
the customer and the bank that matters but more so the distance
between the customer and the banks decision center. In other
words, the head ofce, or wherever credit decisions are actually
nalized, is more relevant than the location of the bank branch.
Holding the distance between the rm and the bank branch
constant, the outcome could change radically depending on
whether the decision-making power is in the hands of the branch
manager or is centralized. In the second case, it is clear that spatial
proximity loses signicance. Organizational forms characterized by
multiple decision-making levels, typically tiered rather than at,
seem less conducive of soft information usage (Stein, 2002). The
analysis of spatial proximity in the context of the rm-bank rela-
tionship loses its relevance if it is not possible to account for the
organizational structure of the rms main bank. In the light of
these considerations, the lack of signicance of our variable is
somewhat expected. We believe that more specic information
on the organizational structure of banks could increase the explan-
atory power of proximity in our model.18
6.1.5. Control variables
The loadings on the control variables do not show substantial
differences between the different models conrming the stability
of the results and the expected signs (Table 3). The rms charac-
teristics, common to both the D and M equations, are substantively
stable across the various models indicating the robustness of the
results obtained.
While the RL variables have a signicant effect on the probability
of rationing (outcome equation), the impact of Firms Characteristics,
as measured by nancial statement data, is generally signicant
(with a few exceptions) only in the selection equation. The prob-
ability of needing additional credit (M = 1) is higher for smaller
rms, for rms with more tangible assets, an increasing working
capital to total asset ratio, and a relatively higher leverage. Stagnant
sales, low protability, and R&D spending are also connected with
higher probability of needing of external nancing. Finally, rms
located in Northern Italy (Geography) are less likely to need funding
17
Kim et al. (2003) found that the effect of switching costs associated with the so-
called capture of a rm is of marginally positive value to the bank, and contributes
23% of the value of the bank.
18
Unfortunately, the identity of the main banks is not disclosed in the survey.
263
compared to rms located in the South. Only a few control variables
show a signicant, albeit modest, effect on the probability of ration-
ing (outcome equation). The growth rate in net working capital
ratio moderately decreases the probability of rationing as well as
the growth rate in ROA and the propensity to invest in tangible
assets while the leverage growth rate (but not its level) increases
the probability of being denied additional credit.19
7. Conclusions and implications
The 20082010 nancial crisis renewed the attention to rela-
tionship banking as a way to avoid credit constraints in times of
systemic nancial distress (Gobbi and Sette, 2014). What emerges
from our empirical analysis is that the RL effect, albeit generally
benecial to rms, is more nuanced than previously expected. In
particular, differences in information asymmetries and bargaining
power across small and large rms do not always protect borrow-
ers from the possibility of a skewed allocation of resources and call
for different policy recommendations.
We study the relationship between credit rationing and RL
measured along four dimensions: the number of bank relation-
ships, the debt share held by the main bank, the duration of the
relationship with the main bank, and rm-bank proximity. We
are able to precisely dene credit rationing based on the results
of a survey on a large sample of Italian manufacturing rms.
Our analysis suggests that both large rms and SMEs gain an
advantage from forming close lending relationships. SMEs exploit
a more immediate benet due to their greater information opacity.
The probability of rationing is increasing and signicantly higher
for SMEs, but the difference is statistically signicant only for rms
with less than eight bank relationships. While debt concentration
with the main bank is desirable for smaller rms, the same is not
true for large companies. Overall, while fewer relationship are pre-
ferred for both size groups, large rms should avoid excessive con-
solidation of debt with the main bank and more evenly spread
their borrowing across a few selected institutions. In fact, SME
rationing becomes increasingly severe as the main banks debt
share reaches 45%, the point of maximum credit restriction, and
falls afterwards. A strong commitment to the main bank betters
small rms access to credit. For large rms, the effect of debt con-
centration is remarkably different. An increase in the share of debt
diminishes the probability of being credit constrained for values
below 40% and increases rationing thereafter. Excessive debt con-
solidation appears to be detrimental to large rms.
A longer banking relationship makes it easier for a rm to
obtain credit, especially for large ones. In particular, statistically
signicant differences emerge only after a period of seven years;
at that point, large rms gain more than SMEs from a protraction
of the relationship. Finally, the rm-bank proximity does not
appear to be signicant. Technological improvements, such as
credit scoring, allow banks to transform soft information into
hard easy-to-transfer data, reducing the advantage of localism.
Acknowledgments
We wish to thank the editors (Carol Alexander and Ike Mathur),
and the anonymous referees for helpful comments and guidance.
The authors also thank participants to the IFABS 2011 Conference
(Rome, Universit Roma Tre, July 2011) for their useful comments.
We owe a special thanks to Mark Looney, Valentina and Flor Sangi-
orgi for their precious research support and insight into our work.
19
In unreported results we calculated the effect of Firms Characteristics, Corporate
Governance proxies, Bank Market proxies of the conditional probability of rationing.
The effect of these variables on the probability of rationing is minimal, which is
expected since they show little signicance in our outcome equation.
264 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265

Appendix A. Denitions of variables

Credit rationing
M Binary variable equals 1 if a rm wanted more credit, 0 otherwise
D Binary variable equals 1 if a rm applied for credit but it was denied, 0 otherwise
Firm characteristics
Total assets Log (average(Total Asset 2001; Total Asset 2002))
h   i
Tangible assets Tang:assets01 Amort:01
Tang:assets 02 Amort:02
=2
Tot:assets01 Tot:assets02
 
Leverage Short and Longtermdebt
Lev erage Short and Longtermdebt=
Equity
Leverage (growth rate) Lev erageg:r: Lev erage02  Lev erage01 =jLev erage01 j
Net working capital to total assets Networking capital02 =Total assets02
h   i
Net working capital to total assets Network:cap:02
 Network:cap: 01
=j Network:cap: 01
j
Ta02 Ta01 Ta01
(growth rate)
Sales (growth rate) Salesg:r: Sales02  Sales01 =jSales01 j
Interest coverage Interest cov erage EBIT 02 =Interest expense02
ROA Roa EBIT=Totalassets
ROA (growth rate) Roa growth rate RoajRoa
02 Roa01
01 j

Investment rate in tangible assets Tang:assets02 Tang:assets01

Tang:assets01
R&D Binary variable equals 1 if a rm invested in R&D during 20012003
Traditional industry Binary variable equals 1 if a rm belongs to a traditional industry
Scale intensive industry Binary variable equals 1 if a rm belongs to a scale-intensive industry
Specialized industry Binary variable equals 1 if a rm belongs to a specialized industry
Hi-tech industry Binary variable equals 1 if a rm belongs to a hi-tech industry
Geographical area Binary variable equals 1 if a rm is located in the North of Italy
SME Binary variable equals 1 if a rm is a SME (less than 250 employees)
Corporate governance
Firms ownership Intensity of ownership concentration among the rst three voting share owners, rated by stake magnitude
h i
Self-nancing Total inv estments0103 %of selffinancing
Total assets01 Total assets02 Total assets03 =3
The % of self-nancing is the rms response to a survey questions asking to dene the percent of total investment nanced with
internal capital/funds.
Bank market
HHI of bank branches Intensity of branch concentration by province
Bank relationship
Number of banks Number of bank relationships at rm level
Main banks debt share Firms debt share with the main bank
Relationship duration Length of relationship between a rm and its main bank. Square root (age) is used in order to capture the decreasing marginal
contribution of the information gathered over time.
Proximity Spatial proximity of a rm to the main bank. A dummy variable equal to 1 if a rms headquarters is located in the same province
as its main bank

References
Agarwal, S., Hauswald, R., 2010. Distance and private information in lending.
Review of Financial Studies 23 (7), 27582788.
Ai, C., Norton, E.C., 2003. Interaction terms in logit and probit models. Economics
Letters 80, 123129.
Albareto, G., Benvenuti, M., Mocetti, S., Pagnini, M., Rossi, P., 2011. The Organization
of Lending and the Use of Credit Scoring Techniques in Italian Banks. Journal of
Financial Transformation 32, 143168.
Alessandrini, P., Presbitero, A.F., Zazzaro, A., 2009a. Banks, distances and rms
nancing constraints. Review of Finance 13, 261307.
Alessandrini, P., Fratianni, M., Zazzaro, A., 2009b. The Changing Geography of
Banking and Finance. Springer, US, ISBN: 978-0-387-98077-5 (Print) 978-0-
387-98078-2 (Online).
Allison, Paul D., 1999. Comparing logit and probit coefcients across groups.
Sociological Methods and Research 28, 186208.
Angelini, P., Di Salvo, R., Ferri, G., 1998. Availability and cost of credit for small
businesses: customer relationships and credit cooperatives. Journal of Banking
& Finance 22, 925954.
Atzeni, G.E., Piga, C.A., 2007. R&D investment, credit rationing and sample selection.
Bulletin of Economic Research 59, 149178.
Becchetti, L., Castelli, A., Hasan, I., 2010. Investment-cash ow sensitivities, credit
rationing and nancing constraints in small and medium-sized rms. Small
Business Economics 35, 467497.
Berger, A.N., Black, L.K., 2011. Bank size, lending technologies, and small business
nance. Journal of Banking & Finance 35, 724735.
Berger, A.N., Udell, G.F., 1995. Relationship lending and lines of credit in small rm
nance. Journal of Business 68, 351381.
Berger, A.N., Demirgc-Kunt, A., Levine, R., Haubrich, J.G., 2004. Bank concentration
and competition: an evolution in the making. Journal of Money, Credit and
Banking 36, 433451.
Berkowitz, J., White, M.J., 2004. Bankruptcy and small rms access to credit. Rand
Journal of Economics 35, 6984.
Bharath, S.T., Dahiya, S., Saunders, A., Srinivasan, A., 2011. Lending relationships and
loan contract terms. Review of Financial Studies 24, 11411203.
Bhattacharya, S., Chiesa, G., 1995. Proprietary information, nancial intermediation,
and research incentives. Journal of Financial Intermediation 4, 328357.
Bhattacharya, S., Thakor, A.V., 1993. Contemporary banking theory. Journal of
Financial Intermediation 3, 250.
Bolton, P., Scharfstein, D.S., 1996. Optimal debt structure and the number of
creditors. The Journal of Political Economy 104, 125.
Bongini, P., Di Battista, M.L., Nieri, L., 2009. Relationship banking: an old solution for
the present crisis? Bancaria 5, 220.
Boot, A.W.A., 2000. Relationship banking: what do we know? Journal of Financial
Intermediation 9, 725.
Boot, A.W.A., Thakor, A.V., 1994. Moral hazard and secured lending in an innitely
repeated credit market game. International Economic Review 35, 899920.
Boyd, J.H., Prescott, E.C., 1986. Financial intermediary-coalitions. Journal of
Economic Theory 38, 211232.
Bris, A., Welch, I.V.O., 2005. The optimal concentration of creditors. Journal of
Finance 60, 21932212.
CarboValverde, S., Rodriguez-Fernandez, F., Udell, G.F., 2009. Bank market power
and SME nancing constraints. Review of Finance 13 (2), 309340.
Carletti, E., Cerasi, V., Daltung, S., 2007. Multiple-bank lending: diversication
and free-riding in monitoring. Journal of Financial Intermediation 16, 425451.
Cole, R.A., 1998. The importance of relationships to the availability of credit. Journal
of Banking & Finance 22, 959977.
Cole, R.A., Goldberg, L.G., White, L.J., 2004. Cookie-cutter versus character: the
micro-structure of small business lending by large and small banks. Journal of
Financial and Quantitative Analysis 39, 227251.
Cotugno, M., Monferr, S., Sampagnaro, G., 2013. Relationship lending, hierarchical
distance and credit tightening: evidence from the nancial crisis. Journal of
Banking and Finance 37 (5), 13721385.
De Bodt, E., Lobez, F., Statnik, J.C., 2005. Credit rationing, customer relationship and
the number of banks: an empirical analysis. European Financial Management
11, 195228.
Degryse, H., Van Cayseele, P., 2000. Relationship lending within a bank-based
system: evidence from European small business data. Journal of Financial
Intermediation 9, 90109.
 S. Cenni et al. / Journal of Banking & Finance 53 (2015) 249265
Degryse, H., Kim, M., Ongena, S., 2009. Microeconometrics of Banking: Methods,
Applications and Results. Oxford University Press, Oxford.
DellAriccia, G., Marquez, R., 2004. Information and bank credit allocation. Journal of
Financial Economics 72, 185214.
Detragiache, E., Garella, P., Guiso, L., 2000. Multiple versus single banking
relationships: theory and evidence. Journal of Finance 55, 11331161.
Dewatripont, M., Maskin, E., 1995. Credit and efciency in centralized and
decentralized economies. Review of Economic Studies 62, 541555.
DeYoung, R., Glennon, D., Nigro, P., 2008. Borrowerlender distance, credit scoring,
and the performance of small business loans. Journal of Financial
Intermediation 17, 113143.
Elsas, R., Heinemann, F., Tyrell, M., 2004. Multiple but asymmetric bank nancing:
The case of relationship lending. Working Paper, CESifo Group Munich.
Fama, E.F., 1985. Whats different about banks? Journal of Monetary Economics 15,
2939.
Gobbi, G., Sette, E., 2014. Do rms benet from concentrating their borrowing?
Evidence from the great recession. Review of Finance 18 (2), 527560.
Gopalan, R., Nanda, V.K., Seru, A., 2007. Afliated rms and nancial support:
evidence from indian business groups. Journal of Financial Economics 86 (3),
759795.
Greene, William H., A Statistical Model for Credit Scoring (April 8, 1992). NYU
Working Paper No. EC-92-29. Available from SSRN: http://ssrn.com/abstract=
1867088.
Greene, W., 2010. Testing hypotheses about interaction terms in nonlinear models.
Economics Letters 107, 291296.
Greene, W., 2012. Econometric Analysis, 7/E. Prentice Hall, New Jersey.
Guiso, L., 1998. High-tech rms and credit rationing. Journal of Economic Behavior
and Organization 35, 3959.
Guiso, L., Minetti, R., 2010. The structure of multiple credit relationships: Evidence
from U.S. rms. Journal of Money, Credit and Banking 46, 10371071.
Harhoff, D., Krting, T., 1998. Lending relationships in Germany empirical
evidence from survey data. Journal of Banking & Finance 22, 1317
1353.
Hernandz-Canovas, G., Martnez-Solano, P., 2007. Effect of number of banking
relationships on credit availability: evidence from panel data of Spanish small
rms. Small Business Economics 28, 3753.
Herrera, A.M., Minetti, R., 2007. Informed nance and technological change:
evidence from credit relationships. Journal of Financial Economics 83 (1),
223269.
Hubert, F., Schfer, D., 2002. Coordination failure with multiple-source lending: the
cost of protection against a powerful lender. Journal of Institutional and
Theoretical Economics 158, 256275.
Jaffee, D.M., Russell, T., 1976. Imperfect information, uncertainty, and credit
rationing. Quarterly Journal of Economics 90, 651666.
Karaca-Mandic, P., Norton, E.C., Dowd, B., 2012. Interaction terms in nonlinear
models. Health Services Research 47 (1), 255274.
Kim, M., Kliger, D., Vale, B., 2003. Estimating switching costs: the case of banking.
Journal of Financial Intermediation 12, 2556.
265
Knyazeva, A.K., Knyazeva, D., 2012. Does being your banks neighbor matter?
Journal of Banking & Finance 36, 11941209.
Leland, H.E., Pyle, D.H., 1977. Information asymmetries, nancial structure, and
nancial intermediation. Journal of Finance 32, 371387.
Long, S., 2009. Group comparisons in logit and probit using predicted probabilities.
Working Paper, Indiana University.
Minetti, R., Zhu, S.C., 2011. Credit constraints and rm export: microeconomic
evidence from Italy. Journal of International Economics 83, 109125.
Ongena, S., Smith, D.C., 2001. The duration of bank relationships. Journal of
Financial Economics 61, 449475.
Padilla, A.J., Pagano, M., 1997. Endogenous communication among lenders and
entrepreneurial incentives. Review of Financial Studies, Society for Financial
Studies 10 (1), 205236.
Pavitt, K., 1984. Sectoral patterns of technical change: towards a taxonomy and a
theory. Research Policy 13 (6), 343373.
Petersen, M.A., 2004. Information: Hard and Soft. Northwestern University, Mimeo.
Petersen, M.A., Rajan, R.G., 1994. The benets of lending relationships: evidence
from small business data. Journal of Finance 49, 337.
Petersen, M.A., Rajan, R.G., 1995. The effect of credit market competition on lending
relationships. The Quarterly Journal of Economics 110 (2), 407443.
Presbitero, A.F., Zazzaro, A., 2011. Competition and relationship lending: friends or
foes?. Journal of Financial Intermediation 20 (3), 387413.
Rajan, R.G., 1992. Insiders and outsiders: the choice between informed and arms
length debt. Journal of Finance 47, 13671400.
Ratti, R.A., Lee, S., Seol, Y., 2008. Bank concentration and nancial constraints on
rm-level investment in Europe. Journal of Banking & Finance 32, 26842694.
Ryan, R.M., OToole, C.M., McCann, F., 2014. Does bank market power affect SME
nancing constraints? Journal of Banking and Finance 49, 495505.
Sartori, A., 2003. An estimator for some binary outcome selection models without
exclusion restrictions. Political Analysis 11 (2), 111138.
Saunders, A., Allen, L., 2002. Credit Risk Measurement. Wiley, New York.
Sharpe, S.A., 1990. Asymmetric information, bank lending and implicit contracts: a
stylized model of customer relationships. Journal of Finance 45, 10691087.
Stein, J.C., 2002. Information production and capital allocation: decentralized versus
hierarchical rms. Journal of Finance 57, 18911921.
Stiglitz, J.E., Weiss, A.M., 1981. Credit rationing in markets with imperfect
information. American Economic Review 71, 393410.
Van de Ven, W.P.M.M., Van Praag, B.M.S., 1981. The demand for deductibles in
private health insurance: a probit model with sample selection. Journal of
Econometrics 17, 229252.
von Rheinbaben, J., Ruckes, M., 2004. The number and the closeness of bank
relationships. Journal of Banking & Finance 28, 15971615.
von Thadden, E., 1992. The commitment of nance, duplicated monitoring and the
investment horizon. Working Paper, CEPR.
von Thadden, E.L., 1995. Long-term contracts, short-term investment and
monitoring. Review of Economic Studies 62, 557575.
Wooldridge, J.M., 2006. Introductory Econometrics: A Modern Approach, 3/E.
Thomson/South-Western, Mason.