Bankruptcies of Small Firms and Lending Relationship

Journal of Banking & Finance 36 (2012) 857870
Contents lists available at SciVerse ScienceDirect
Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf
Bankruptcies of small rms and lending relationship

Katsutoshi Shimizu
Department of Economics, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Aichi, Japan
a r t i c l e
i n f o
Article history:
Received 2 June 2011
Accepted 28 September 2011
Available online 4 October 2011
JEL classication:
G21
G28
G33
K23
L31
a b s t r a c t
This paper empirically investigates the role played by relatively small banks in the Japanese local credit
market. We test the hypothesis that small banks enhance the recovery rate from the nancial distress and
reduce the bankruptcy ratio of small rms. Empirical evidence suggests that small banks specialize more
in relationship loans to small rms. However, this expertise is limited to the loans to unincorporated
rms or those with a very small number of employees.
2011 Elsevier B.V. All rights reserved.
Keywords:
Lending relationship
Japanese banks
Nonperforming loans
Bankruptcy
Small business
Credit union
1. Introduction
There exist many studies addressing the problem of how the
lending relationship affects the efciency of the credit market. The
past empirical analyses have examined the effects of the lending
relationship on the availability of funds, loan interest rates, likelihood of loan approval, and other contractual terms. In particular,
lending relationship attracts much concern of some economists
and policymakers because small rms have many difculties in
obtaining funds from public capital markets.
Japanese banks nd themselves in an increasingly competitive
environment of rapidly progressing nancial globalization and
deregulation where boundaries, typically geographic ones, are
gradually becoming less signicant. There have been multiple
rounds of consolidation among the nancial institutions of the
world. Further in Japan, the total number of nancial institutions
has decreased from 890 to 595 (by 33%) during the last decade.1
Tel.: +81 52 789 2378.
E-mail address: shimizu@soec.nagoya-u.ac.jp
As is well known, there are several types of nancial institutions in Japan: the
largest type are city banks that operate all over Japan, the second-largest type are
regional banks that operate in smaller regions (typically in one or a few prefectures),
the third-largest type are shinkin banks that operate in even smaller regions (typically
in a few cities and other municipalities within a prefecture). There also exist other
cooperative institutions. From 1999 to 2009, the number of city banks decreased from
9 to 6, that of regional banks from 124 to 106, and that of shinkin banks from 386 to 272.
1
0378-4266/$ - see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jbankn.2011.09.016
It is an academic and political issue how such large waves of consolidation affect the efciency of the credit market in the Japanese
banking industry as well as in the US.
Thus, it is important for us to understand whether or not small
banks provide relationship loans, the effect of these loans on the
economic activity of small businesses, and the way in which this
effect realizes. Some of the past empirical ndings are as follows:
(i) the lending relationship increases the availability of nancing
for small businesses (Petersen and Rajan, 1994), (ii) the lenders
are more likely to nance credit-constrained rms in a more
concentrated market (Petersen and Rajan, 1995), (iii) the loan
rates decrease with the distance between the rm and the lending bank (Degryse and Ongena, 2005), (iv) the rms choose multiple banks if the probability of terminating the relationship is
higher (Detragiache et al., 2000), and (v) the large banks are less
willing to lend to informationally difcult credits (Berger et al.,
2005).2
This paper addresses two questions from the viewpoint of Boot
and Thakor (2000). The rst question deals with the sectors that
small banks specialize in, and the second deals with the issue of
whether or not lending relationship affects the bankruptcy rate
of borrowers. This paper explores the idea that small banks are
2
Boot (2000), Elyasiani and Goldberg (2004), and Berger and Udell (2006) are
useful reviews of this literature.
858
K. Shimizu / Journal of Banking & Finance 36 (2012) 857870
relationship lenders to small rms in the sense that they have

greater expertise in relationship loans than larger banks.
As Berger et al. (2005) argue, small banks might be at a comparative advantage in making loans based on soft information and
hence, larger banks are more apt to lend to larger rms that have
better accounting records. Since a rm that borrows from a larger
bank is more prone to repay its trade credit late, it is suggested that
these rms are more credit constrained. The empirical implication
of Berger et al. (2005) is that small banks may play the role of alleviating the credit constraints of small rms.
This paper goes one step further to investigate whether or not
small banks play the role of reducing the bankruptcy rate or
enhancing the recovery probability of distressed small businesses.3
The previous theoretical studies mention the possibility that the success probability of a rms investment project depends on the loan
contract type: relationship loan or transaction loan. For example,
Diamond (1991) argues that bank monitoring increases the success
probability by preventing intermediate borrowers from taking risky
projects. Sharpe (1990) argues that the success probability of a good
borrower is higher for inside banks because they have access to more
precise information than outside lenders. Holmstrom and Tirole
(1997) argue that bank monitoring mitigates the moral hazard
incentive of entrepreneurs and indirectly enhances the success
probability.
This paper emphasizes the role of the organizational structure
of small banks as analyzed in Stein (2002). Large hierarchical banks
are at a comparative disadvantage when information about an
investment project is soft. Since small banks have a decentralized
organizational structure, their loan manager has a higher research
effort incentive and as such, the net expected output increases. The
role of soft information is realistically illustrated in Berger et al.
(2005, p. 239):
For example, consider a loan ofcer trying to decide whether
or not to extend credit to a small start-up company that does not
have audited accounting statements. The best the loan ofcer
may be able to do is to spend time with the company president
in an effort to determine whether she is honest, prudent, and hardworkingi.e., the classic candidate for a character loan. However,
given that this information is soft and cannot be veriably documented in a report that the loan ofcer can pass onto his superiors,
the model predicts that his incentives to produce high-quality
information are weak when he works inside a large bank..
This paper regards shinkin banks (abbreviated hereafter as SBs)
as such small banks that have a more decentralized organizational
structure and regional banks (abbreviated hereafter as RBs) as larger banks that have a more hierarchical structure.4 SBs are not only
typically smaller than regional banks, but also their operating region
is geographically restricted to a small area and their borrowers are
legally restricted to small and medium rms.5
3
Dahiya et al. (2003) note that the recovery rate on defaulted debt is likely to be
fairly high for bank loans. See footnote 2 on page 376 of Dahiya et al. (2003). Grunert
and Weber (2009) conclude that borrowers with an intense client relationship with
the bank exhibit a higher recovery rate after default. In an extreme case, small banks
might play the role of preventing small rms in distress from bankruptcies or
postponing their bankruptcies by nancial support. A classical example in Japan is
given in Hoshi et al. (1990).
4
Of course, since regional banks are smaller than major city banks, both shinkin
banks and regional banks are often considered to be relationship lenders in Japan.
5
In addition, shinkin bank is a non-prot cooperative. It is important to take care of
generalizing this papers results to typical markets because the activities of non-prot
cooperative banks (or credit unions) are relatively small in some countries. For
example, according to Goddard et al. (2008, p. 1837), credit unions of the US have
$668 billion assets (8% of total bank assets). Hackethal (2004, p. 74) reports that
German credit cooperatives have 12% of total bank assets. As appeared later in Table
1, shinkin banks have 108 trillion yen of deposits, which is only 14% of total deposits.
We go back again later to this problem.
Small decentralized banks are at a comparative advantage when

they must rely heavily on soft information. Following Boot and
Thakor (2000), we dene a relationship loan as a loan that permits
the bank to use its expertise, which we call sector specialization, to
improve the borrowers expected payoff. We consider two-bank,
two-sector model where the relationship loans of one bank add
more value to the borrower in a specic sector. In particular, we argue that it might be possible to enhance the borrowers quality, i.e.,
the success probability of investment projects by unincorporated
rms or rms with small capital, small debt obligations, and a
small number of employees that have no veriable accounting
information.
Although Japanese banks have to disclose the total amount of
loans to bankrupt rms and non-performing loans, they are not expected to provide the detailed information: how many borrowers
went bankrupt, what is the proportion of bankrupt borrowers in
each sector, what is the proportion of small rms in the bankrupt
rms, etc. Thus, there has not been any empirical study investigating the bankruptcy ratio of specic type of borrowers because the
data on the bankruptcy ratios of specic type of borrowers by bank
is unavailable. This paper uses the bankruptcy ratios of specic
borrowers by prefecture that are publicly disclosed by the administrative agency afliated with the Small and Medium Enterprise
Agency in Japan. Since the activity of small banks varies across prefectures, it is tempting to examine how correlated the activities of
small banks are with the default rate of the specic borrowers in
each prefecture. If small banks behave differently from relatively
large banks, their performance must have signicant inuences
on the aggregate bankruptcy ratios of the specic borrowers within
a prefecture.
As is well known, Japanese small banks have maintained a higher
non-performing loans ratio than large banks ever since the non-performing loans problem emerged and the nancial crisis. However,
the ratio of loans to bankrupt rms was not much higher. I found
supportive evidence for the hypothesis that small banks reduce
the bankruptcy of small, in particular, unincorporated, rms in the
local market by maintaining a larger amount of non-performing
loans (except loans to bankrupt borrowers). The repayment of these
loans is delayed during times of distress, but the banks are more
likely to recover these loans later. This suggests that small decentralized banks might specialize more in relationship lending to small
rms than larger banks.
In the remainder of this paper, I start by explaining how one
type of banks specializes in a specic sector of the regional economy, and by predicting the relationship between the bankruptcy
ratio in this sector and the amount of non-performing loans of
banks that are at a comparative advantage in this sector. I also explain how we can test the hypothesis without knowing the bankruptcy ratio of each sector by bank.
In Section 3, after briey describing the presence of small banks
and small rms in Japan, I present the main estimation results of
the panel data model including the time invariant variables. I
found evidence that the more the aggregate non-performing loans
ratio of small banks exceeds that of the larger banks in a prefecture,
the lower is the prefectural bankruptcy ratio of unincorporated
rms becomes. I tried to nd similar inuences on the bankruptcy
ratio of other sectors: rms with relatively small capital, small debt
obligations, and a small number of employees. However, a similar
result is obtained only for rms with less than ve employees.
Section 4 provides an extended econometric analysis on the
aftermath of such behavior of small banks. Using the dynamic panel data model, we show that such behavior of small banks does
not lead to a surge in the bankruptcy ratio in the subsequent years.
An explanation about the related literature appears in Section 5.
Since no past empirical literature analyzes the inuences of the
lending relationship on the bankruptcy ratio, this paper lls the
859
gap between the past theoretical works and the empirical ndings.
Section 6 concludes the paper.
t =1
t =2
t =3
State G
Y
2. Theoretical background and econometric strategy
Following Boot and Thakor (2000), this section simply demonstrates in a setting of two banks and two sectors that a rm prefers
borrowing from a relationship lender whose expertise is greater
because the lender is able to offer a lower loan rate due to a higher
recovery rate in the equilibrium. When small banks have expertise
in relationship loans to small rms, this argument implies that the
bankruptcy ratio of small rms is lower when the recovery ratio of
the borrowers receiving relationship loans from small banks is
higher.
2.1. Model
Consider a regional economy with two types of banks and two

types of borrowing rms. There are three dates. At date t = 0, the
rm needs a loan of $1 to nance a project. One type of rms belong to sector S1 and the other type to sector S2. The banks can
choose either of relationship loans or transaction loans. Observing
the contract offered by each bank, the rm chooses one contract
specifying loan rate and the type of loan. We assume that all agents
are risk neutral and that the riskless interest rate is zero.
The rms project yields a stochastic return at t = 2 or 3, following which, the banks are repaid. This return or its distribution depends on the state, the rms quality, and the type of contract. At
t = 1, one of the two states G or B is realized. State G occurs with
probability 1 q, and state B with probability q. When transaction
loans are made and state G is realized, the return is Y with certainty
at t = 2. In state B, the realization of return is delayed till t = 3.
When state B is realized, the return is Y with probability h 2 (0, 1)
and 0 with probability 1 h at t = 3. A borrowers quality his distributed according to the cumulative distribution function F(h)
with domain h 2 h;
h.
Relationship loans enable rms to recover from distress with
probability v 2 (0, 1). In other words, after state B is realized, the return at t = 3 is Y with probability h + v [1 h] and 0 with probability 1 h v[1 h]. Fig. 1 depicts the structure of return.
The recovery rate v is dependent on bank is sector specialization. Each type i of banks differs in terms of the sector specialization and this is exogenously given. Without loss of generality, we
make the following assumption.
Assumption 1.
v
State B
v Si
> v Sj i i; j 1; 2; i j.
That is, bank i has a comparative advantage in relationship

loans to sector Si. In other words, the failure probability of a rm
in sector S1 is lower if bank 1 provides a relationship loan than
when bank 2 provides a relationship loans of or when a transaction
loan is provided by either.
The banking industry is imperfectly competitive. Each bank offers loan rate Rm
i i 1; 2; m TL; RL at t = 0, where superscript TL
denotes the transaction loans and superscript RL denotes the relationship loans. The relationship lenders cost to service the borrower during the lending period is constant C. Following Boot
and Thakor (2000), we assume that there is no asymmetric information problem. Not only the state and quality are publicly observable, but also the decision on whether or not to allocate service
capacity to transaction lending or relationship lending is publicly
observable and veriable.
The bank nances $1 from depositors and owes the expected
cost of deposit rd, which is the rate required by the depositors.
Thus, the bank is expected prot is
Fig. 1. Return structure when relationship loans are made.
TL
PTL
i 1 q qhRi r d
if bank i provides a transaction loan to the borrower, and is
Sk
PRL
i 1 q q h v i 1 h
o
RRL
i rd C
if a relationship loan is provided. The borrowers expected return is

TL
pTL
k f1 q qhgY Ri
or
Sk
pRL
k 1 q q h v i 1 h
o

Y RRL
;
i
respectively.
We assume that when the bank provides a transaction loan, the
rate RTL(h) is determined to satisfy
PTL
i K;
where K(=0) is an exogenously given parameter that represents the

reservation payoff of banks.
2.2. Choice between relationship loan or transaction loan
Consider a Bertrand type competition between the two banks
(i = 1, 2). The relationship loan rate RRL
is determined as to make
i
the borrower indifferent between relationship loans and transacRL
tion loans. Equating pTL
k pk and substituting (1) and (2), we have
Sk
TL
PRL
i Pi qv i 1 hY C:
6
S
Thus, the bank offers a relationship loan if qv i k 1 hY > C or
h<1
C
S
qv i k Y
^hik :
The banks have an incentive to provide relationship loans to high

risk (low h) borrowers. This is because the marginal value for relationship loan is higher for such borrowers.
^
hik is the cutoff probability of bank is relationship loan to sector
Sk. From the Assumption 1, we have ^
hii > ^
hji i j. In other words,
bank i offers relationship loans even to the borrowers in sector Si
with higher quality (success probability) than bank j. In this simple
setting, whether relationship loans or transaction loans are provided depends on the borrower quality h, as summarized in the
next proposition.
860
Proposition. Suppose that bank 1 has greater expertise in offering

relationship loans to sector S1 v S11 > v S21 and bank 2 in offering the

same to sector S2 v S12 < v S22 . Suppose also that the following
assumptions are satised.
S
S
Assumption 2. qv i i Y1
h < C < qv j i Y1 h
i j.
.
Assumption 3. h > rd K1qY
qY
Then, in equilibrium,
^ji i; j 1; 2; i j, both banks offer transac(i) when h > ^
hii > h
tion loans and the rm borrows from either bank at the loan
b TL ;
rate R
(ii) when ^
hii > h > ^
hji , bank i offers relationship loans to sector Si,
bank j offers transaction loans, and the rm in sector Si borb RL ;
rows from bank i at the rate R
i
(iii) when ^
hii > ^
hji > h, both banks offer relationship loans and
e RL ,
the rm in sector Si borrows from bank i at a rate R
i
where
S
qv i i 1 hY K rd
K rd
b RL

;
; R
i
S
1 q qh
1 q q h v i i 1 h

S
S
q v i i v j i 1 hY K r d C

:
S
1 q q h v i i 1 h
b TL
R
e RL
R
i
If the borrower quality is high enough (h > ^

hii ), both banks do
not have to provide relationship loans to enhance the probability
b TL satises PTL = K. When borrower
of repayment. The loan rate R
quality is intermediate ^
hii > h > ^
hji , bank i nds it protable to
provide relationship loans while bank j does not. Taking the cost
advantage, bank i is able to reduce its rate slightly from that offered
by bank j. The loan rate of bank i is obtained from the equilibrium
TL
conditions pTL pRL
= K.
i and P
When the borrower quality is low enough ^
hii > ^
hji > h, both
banks nd it protable to provide relationship loans. They compete
with each other in offering a lower loan rate. As a result of the Bertrand competition, bank i can offer the lowest rate because of
e RL satises the
greater expertise in sector Si. Bank is loan rate R
i
RL
RL
equilibrium conditions PRL
K
and
p
p
.
Bank
i obtains a payj
j
i
off larger than K even when it offers the same rate as bank j. Fig. 2
depicts the banks expected prot as a function of borrower
quality.
As such, this model predicts that the rms in sector S1 with
^
h11 > h borrow only from bank 1 that specializes in sector S1. The
rms in sector S1 with ^

h11 < h borrow from any bank providing
transaction loans. On the other hand, the rms in sector S2 with
^
h22 > h borrow only from bank 2 that specializes in sector S2. The
rms in sector S2 with ^
h22 < h borrow from any bank providing
transaction loans.
The important assumption of the proposition is v S11 > v S22 and
S2
v 2 > v S12 . The rst inequality is plausible from the perspective of
the past literatures as mentioned in the introduction while the
plausibility of the second inequality is considered somewhat
weak.6 However, as Stein (2002) argues, the large hierarchical banks
have comparative advantages in loans to large rms because information can be costlessly hardened and passed along inside the bank.
The large hierarchical banks are able to implement not only more
efcient allocation across units, but also to generate more research
incentives on the part of line managers. The rst inequality of
Assumption 2 guarantees that there are some rms that use transaction loans. The second inequality guarantees that there are some
rms that use relationship loans. Assumption 3 implies that the success return Y is large enough for the most risky rm to repay the loan
rate of transaction loans, which is the highest rate.
2.3. Nonperforming loans and econometric methodology
The following argument bridges the gap between the theoretical model given above and the empirical analyses in the next section. Now, assume that the loan status is classied into three
groups at t = 1; normal, problem, and bankrupt. In state G, the
loans are classied as normal loans because they immediately
yield the return Y > 1 with certainty. However, in state B, the loans
are classied as problem loans or bankrupt loans. Such loans,
which are called non-performing loans in practice as will be explained later in Section 3, produce no repayment at t = 2.
If transaction loans have been offered, a fraction hof the problem loans yields Y with delay at t = 3 and the rest are bankrupt
loans that yield nothing. If relationship loans have been offered, a
S
fraction (1 h) fraction of the rms recover with probability v i i .
Si
Therefore, a fraction h 1 hv i of the loans are problem loans
and the rest are bankrupt loans.
Suppose that a fraction wk of borrowers are from sector Sk. We
also suppose that bank i provides transaction loans to a fraction gi
of the borrowers when bank j also provides transaction loans. Let
aSi k be bank is ratio of bankrupt loans to sector Sk and ai be its ratio
of total bankrupt loans. These are calculated as
"
S1
1
a w1 g1
aS12 w2 g1
"
q1 hdFh
^h11
aS22 w2 g2
h
q1 hdFh;
Z
#

S1
q1 h 1 v 1 dFh ;
aS21 w1 g2
h
q1 hdFh
^h22
^h11
h
^h
22
Z
h
^h22
h
#

q1 h 1 v S22 dFh ;
9
1RL
and ai aSi 1 aSi 2 .

The aggregate bankruptcy ratio of sector S1 in the regional economy is dened as the ratio of the number of bankrupt rms in sector S1 to the number of bankrupt rms in both sectors:
TL
BR
2RL
q1 hdFh;
^h
11
21
11
aS11 aS21
:
a1 a2
10
Fig. 2. Bank prots from relationship loans and transaction loans. (Note: RL:
relationship loan, TL: transaction loan.)
The ratio of problem loans to total loans of each bank is given as

6
This point was suggested by the anonymous reviewer.
861
"
c 1 w 1 g1
c 2 w 1 g2
h
qhdFh
^h
11
Z

q h 1 hv S11 dFh w2 g1
"
h
^h
11
^h
11
qhdFh w2 g2
h
Table 1
Scale and performance of SBs and RBs.
qhdFh;
^h
22
h
qhdFh
^h
22
Z
h
^h
22
#

q h 1 hv S22 dFh :
11
We now consider the relationship between the aggregate bankruptcy ratio of sector S1 and the problem loans ratio of each bank.
Let us suppose that the recovery probabilities v S11 and v S22 vary
across regional economies. The difference in the problem loan ratios of bank 1 and bank 2 is dened as
4c c1 c2
w1 g1 g2
w1
^h11
Z
h
^h22
h
^h11
qhdFh w2 g1 g2
h
qhdFh
^h22

q h 1 hv S22 dFh
12
which is an increasing function of v S11 and a decreasing function of

v S22 . (See the appendix if included.) 4cis higher if bank 1 has more
expertise than bank 2.
The recovery rate also affects the aggregate bankruptcy ratio of sector

S1. aS11 aS21 decreases as v S11 increases and aS12 aS22 decreases as
S
increases. Thus, we have dBR=dv 11 < 0 and dBR=dv 22 > 0.

Thus, when the recovery probability (v S11 or v S22 ) varies across regional economies, BR covaries with 4c. Although we are not able to
S
observe the ratio of bankrupt loans to each sector by each bank ai k
Sk
and the corresponding recovery rate v i , we are able to investigate
whether bank i enhances the recovery rate from distress in sector
Si, if we observe the problem loans ratio of bank i and the aggregate
bankruptcy ratio of the borrowers in sector Si. If bank 1 specializes
more in the relationship loans to sector S1 than bank 2, then the ratio of S1 bankrupt loans to total bankrupt loans is a decreasing
function of the difference in the problem loans ratios of bank 1
and bank 2. That is, we have
2
RB
287
7734 [26.95]
112 [0.39]
108 [0.38]
61 [0.21]
117 [0.41]
0.15
2.55
0.78
0.50
0.54
9255 [32.25]
110
10,709 [97.35]
173 [1.57]
249 [2.26]
187 [1.70]
283 [2.57]
0.12
2.14
0.62
0.57
0.66
Note: The data are as of 2006. The gures in square brackets are average values per
bank.

q h 1 hv S11 dFh w2
v S2
Number of banks
Number of branches
Number of ofcers (thousand)
Deposit (trillion yen)
Loan (trillion yen)
Asset (trillion yen)
Deposit interest rate (%)
Loan interest rate (%)
Staff cost ratio (%)
Property cost ratio (%)
Operating prot/asset (%)
Number of members (thousand)
SB
case, since the recovery rate of bank 1 is higher than that of bank 2,
BR decreases as the relationship loans of bank 1 increase. From this
viewpoint, if we can obtain the empirical support for dBR1/d4c < 0,
then it implies that the inequality v S11 > v S21 holds.
As an alternative, we may have to use the most simplied specication to introduce our econometrics. Dene the recovery rate as
v 0Si k v Si k =1 h and assume that v 0S1 1 > v 0S2 1 and v 0S1 2 < v 0S2 2 . In
equilibrium, bank 1 provides relationship loans to all the borrowers in sector S1 and bank 2 to all borrowers in sector S2. Then,
aS11 w1 q 1 Eh v 0S1 1 ; aS22 w2 q 1 Eh v 0S2 2 ;

aS12 aS21 0;
h
c1 w1 q Eh v 0S1 1 ; c2 w2 q Eh v 0S2 2 :
14
An increase in c1 decreases aS11 or a decrease in c2 increases aS22 and

hence we have dBR/d4c < 0 as earlier. Let us consider the regression
equation in this setting. The dependent variable is transformed into
the log odds ratio of the bankruptcy ratio, y = log BR/(1 BR).7 Using
Eq. (13), y is log linearized as
ypt b0 b1 4 cpt b2 c2;
pt :
15
dBR dBR dv 11
< 0:
dDc dv S11 dDc
13
This is because only bank 1 provides relationship loans to sector S1.

Whether or not Eq. (13) holds is the main hypothesis we analyze
empirically in the following section.
The following arguments may facilitate better understanding.
First, suppose that both banks have the same expertise in relationship loans to sector Sk. That is, the recovery rates of the two banks
S
S
are the same, v 1k v 2k , unlike in our earlier assumption. Then, of
course, we have no positive reason to expect that c1 differs from
c2 and hence empirically analyzing Eq. (13) does not make sense.
Second, suppose that bank 1 has greater expertise in offering relationship loans to sector S1, but both banks have no expertise in
offering relationship loans to sector S2. That is, v S11 > v S21 > 0 and
v S12 v S22 0. We do not exclude such a case. Both banks provide
transaction loans to sector S2 while only bank 1 provides relationship loans to sector S1. 4 c might become positive and dBR/
d4c < 0.
Third, our proposition predicts that the rms of sector Si borrow
from only bank i if h < ^
hii . However, realistically, it may be the case
that the rms of sector Si are more likely to borrow from bank i if
h<^
hii . Random factors outside the model prevent some rms from
borrowing from bank i with some probability. Suppose that bank 1
provides relationship loans to sector S1 with probability / and bank
2 with 1 /. Considering this imperfect specialization case permits us to investigate the role of assumption v S11 > v S21 . Even in this
Since our BR data is prefecture-time data, subscripts p and t denote

the prefecture and year, respectively. c1pt and c2pt are aggregated
across banks in a prefecture using the banks nancial statement
data.
The coefcient of 4cpt is predicted to be negative because

b1 1 aS11 , where aS11 is aS11 at an approximated point. It is
intuitively clear that ypt decreases as 4cpt increases because the
sum of c1pt and a1ptS1 is always equal to w1q.
3. Empirical results
This section provides the empirical results of our estimation.
Before examining our empirical hypothesis, we briey look at the
presence of small banks and small business rms in the Japanese
regional economy.
3.1. Presence of small banks in the regional economy
According to Table 1 that compares the characteristics of SBs
with those of RBs, we know that the former are small banks and
that the latter are larger banks.8 In 2006, there were 287 SBs and
7
By transforming into log odds, we can alleviate the truncation problem of
distribution at a neighborhood of zero.
8
Regional banks are formally member banks of the Regional Banks Association of
Japan and the Second Association of Regional Banks.
862
Fig. 3. Aggregate non-performing loan ratio of SBs and RBs. (Note: LBB: loans to borrowers in legal bankruptcy, PL: problem loans. PL ratio is dened as risk management loan
ratio minus LBB ratio.)
110 RBs. The total number of branches were 7734 and 10,709,
respectively. The average number of branches were 26.95 and
97.35 respectively (reported in square brackets). The average number of ofcers were 390 and 1570, respectively. However, the average number of ofcers per branch did not much differ (about fteen).
SBs raised 108 trillion yen in deposits to provide 61 trillion yen
worth of loans while RBs raised 249 trillion yen in deposits to offer
187 trillion yen worth of loans. SBs occupied 30.3% of the deposits
and 24.6% of the loans in the local market.9 Comparing the gures
in these rows in the square brackets, the average amount of deposits
(loans) per SB was one sixth (one eighth) as large as that of RB. Thus,
the presence and scale of SBs are considerably small as compared to
RBs in both the local deposit and local credit markets.
The deposit interest rate of SB is higher than that of RB by 3 basis points and loan interest rate of SB is higher than that of RB by 41
basis points. The ratio of staff costs to the deposit amount is 0.78%
for SB and 0.62% for RB. In contrast, the property cost of SB is lower
than that of RB. The ratio of net operating prot to total assets is
0.54% for SB and 0.66% for RB.
Thus, the size of SB, as measured by the number of branches,
number of ofcers, deposit amount, and loan amount, is less than
that of RB. In addition, the characteristics of SBs are considered to
be more suitable for relationship loans than those of RBs from the
viewpoint of Stein (2002). The amount of loans per ofcer for SB
is 0.54 (=0.21/0.39) is just a half of that for RB. The low average
productivity of the labor force implies that the output of SB is
more labor intensive, consistently with the view that SBs input
more labor to provide relationship loans by producing and processing soft information. In this sense, the organizational form
of SB might be more decentralized while that of RB might be
more hierarchical.
Fig. 3 shows the time series movement of the ratios of aggregate non-performing loans of SBs and RBs in Japan. We use ofcially dened risk management loan as a measure for the overall
non-performing loans, which consists of loans to borrowers in legal bankruptcy (hereafter abbreviated as LBBs), past due loans in
arrears by 6 months or more, loans in arrears by more than
3 months and less than 6 months, and restructured loans. I inte-
grate the last three parts and refer to it as problem loans(hereafter

abbreviated as PL). In other words, PLs are non-performing loans
except for LBBs.
At the end of scal year 1998, the ratio of non-performing loans
to total loans of SB was 6.68% and that of RB was 5.03%. They
marked the highest record in 2001, 10.01% for SBs and 8.00% for
RB. The difference between these was 1.65% points in 1998 and remained around 2% points afterward. The ratio of LBB to total loans
for SB was 1.68% and was higher than that for RB by 0.53% points in
1998. The LBB ratios of SB were always higher than those of RB
throughout our sample period, but the difference had a decreasing
trend. PL ratio of SB was 5.00% in 1998 and that of RB was 3.88%.
On average, the former was higher than the latter by 1.89% points
in our sample period.
Why do these three graphs of SB lie above the counterparts of
RB? A simple and conventional answer to this question might be
that the borrowers of SB were riskier than those of RB. Before proceeding to the main analysis, it would be valuable to see the presence of small rms in order to conrm whether or not such
conventional understanding is correct.
9
Note that the deposits and loans of city banks and other types of depository
institutions are not included in these gures.
10
Annual Report on the Bankrupt Companies (Kigyo Tosan Chosa Nenpo in
Japanese).
3.2. Small business rms and the bankruptcy ratio

Table 2 reports the number of bankrupt rms in Japan from
1999 to 2006. The original data can be found in the annual report
of the Organization for Small & Medium Enterprises and Regional
Innovation, Japan (SMRJ).10 In 1999, about seventeen thousand
rms went bankrupt. The average number of bankruptcies during
the past eight years from 1999 to 2006 is about sixteen thousand.
These gures include unincorporated rms, small and medium
incorporated rms, and large rms.
About three thousand unincorporated rms went bankrupt
each year. The bankrupt incorporated rms are classied by the
size of net worth (book value of capital, abbreviated as NW). The
average number of bankrupt incorporated rms with NW < 10 million yen is about ve thousand, and that of incorporated rms with
10 million yen 6 NW < 50 million yen is about eight thousand.
863

Table 2
Number of bankruptcies.
Year
Unincorporated
Number of bankrupt rms

1999
3689
2000
3690
2001
3225
2002
3003
2003
2453
2004
1878
2005
2389
2006
2249
Total
22,576
Incorporated (by net worth)
Total
NW < 10
10 6 NW < 50
50 6 NW < 100
100 6 NW < 500
500 6 NW < 1000
1000 6 NW
5023
5535
5769
5467
4896
4471
4306
4550
40,017
7797
9158
9977
9422
7512
6451
6069
6158
62,544
411
485
655
635
553
441
458
461
4099
170
235
283
327
279
213
199
200
1906
11
13
25
44
38
16
15
15
177
24
50
57
51
38
28
30
20
298
17,125
19,166
19,991
18,949
15,769
13,498
13,466
13,653
131,617
16,452
Average per year
2822
5002
7818
512
238
22
37
Proportion (%)
[17.15]
[30.40]
[47.52]
[3.11]
[1.45]
[0.13]
[0.23]
Unincorporated
Incorporated (by net worth)

NW < 10
10 6 NW < 50
Total
50 6 NW < 100
100 6 NW < 1000
1000 6 NW
Number of existing rms (as of 2001)

Number
(3,242,533)
808,850
741,686
36,971
23,950
6143
Proportion
[50.00]
[45.85]
[2.29]
[1.48]
[0.38]
0.71
1.35
1.77
0.71
0.93
Bankruptcy rate
(0.10)
1,617,600
Source: Annual Report of Bankrupt Enterprises (Organization for Small and Medium Enterprises and Regional Innovation, Japan.) Note: NW = net worth (million yen).
These gures sum up to 15,642, with the share in total bankruptcies being 95.07%.
The lower panel of Table 2 reports the distribution of active
rms in Japan as of 2001. The data are from the results of the
Establishment and Enterprise Census by Statistics Bureau of Ministry of Internal Affairs and Communications. Note that the data are
not available for years except for 2001. We should note that the
number of unincorporated rms is different from the number of
individuals who operate them.11
According to the table, the number of unincorporated rms
(individuals) is about 3 million and the number of incorporated
rms is about 1.6 million. The number of incorporated rms with
NW < 10 million yen was about 0.8 million. Half of the incorporated rms consist of these small enterprises. The gure for the
second-smallest class (10 million yen < NW < 50 million yen) is
about 0.74 million. Thus, the proportion of these small incorporated rms is 95.85%.
The gures in the lowest row of the table show the bankruptcy
rate of each class of rms as of 2001. The bankruptcy rate of unincorporated rms (0.10%) and that of incorporated rms with
NW < 10 million yen (0.71%) are smaller than that of incorporated
rms with 10 million yen < NW < 50 million yen (1.35%) and the
gures for the other classes.
Thus, the presence of small rms is remarkable in Japan and is
the number of bankruptcies among these rms. However, it is
quite surprising that the bankruptcy ratio of these small rms is
less than expected. Why is this phenomenon observed in Japan?
One possible answer to this question might be that small banks
can enhance the recovery rates of small rms even if these rms
appear risky. We now report the descriptive statistics and investigate this phenomenon.
11
Unfortunately, statistical data does not classify the actual number of unincorporated rms as legal entities (individuals). Instead, it reports the number of ofces,
stores, and factories of unincorporated rms. Since the number of these entities
generally differs from the number of individuals who operate these entities, we
cannot compare this gure with the gures for incorporated enterprises.
3.3. Some preliminary results and hypotheses

Our sample period is eight years from FY 1998 to 2006. We use
the data from Financial Statements of SBs, Financial Statements of
RBs, Annual Report of Bankrupt Enterprises by the SMRJ, and other
general statistical sources. 12 Since the gures for bankrupt rms are
reported on a prefectural basis by the SMRJ, our estimation strategy
involves using year-prefecture panel data. To this end, we aggregate
each variable of each bank within the prefecture for each year.
Table 3 reports the descriptive statistics of our sample. The
number of observations is 376 (=47 prefectures 8 years). The difference in the PL ratios (Dc) is dened as that of SB minus that of
RB. We can reject the null hypothesis that the mean of the difference in the PL ratios (3.39) is equal to zero. There are 61 observations from among the 376 for which the difference in PL ratios is
negative.
The bankruptcy ratio is dened as in the previous section (Eq.
(10)). Here, the sector is classied by whether or not the rm is legally incorporated, and the scale of net worth if incorporated. We
dene several bankruptcy ratios for each classication. For example, the bankruptcy ratio of unincorporated rms is dened as
the ratio of the number of bankrupt unincorporated to the total
number of bankrupt rms, whose overall mean is 18.80%. Although
not reported in the table, the mean of BRpt is 26.14 in the subsample with a negative difference in PL ratios and 17.99 in the subsample with a positive difference. These two means of BRpt are
signicantly different from each other because the t-statistic is
3.99.
As the previous arguments suggest, one of the important control variables is the proportion of loans by SBs to the total sum of
loans, which corresponds to g1 in Section 2. Its overall mean is
20.98% and standard deviation is 13.31. All other things being
equal, as this variable increases, the bankruptcy ratio of unincorporated rms rises.
12
The nancial statement data of RBs are from the Nikkei Financial Quest Database.
The data on prefectural performance described below are from the Regional Economy
Database provided by Toyo Keizai.
864
Table 3
Summary statistics.
Variable
Mean
Std. dev.
PL ratio of SB
PL ratio of RB
Difference in PL ratios
Bankruptcy ratio: unincorporated rms
Incorporated rms (NW < 10)
Incorporated rms (10 6 NW < 50)
Incorporated rms (1000 6 NW)
Proportion of SB loans to total loans
Proportion of unincorporated rms
Number of incorporated rms
Prefectural GDP growth rate
Land price change rate
Unemployment rate
Average capital asset ratio of SB
Average liquidity ratio of SB
Average number of branches per asset of SB
Average number of borrowers per asset of SB
Average loan amount per borrower of SB
9.14
5.76
3.39
18.80
34.11
42.90
2.88
1.11
0.09
0.12
20.98
54.94
34,417
1.45
5.74
4.45
5.51
19.18
0.01
10.10
5.92
3.00
2.21
3.04
12.03
9.50
8.21
1.76
0.99
0.27
0.30
13.31
5.61
43,295
2.05
2.82
1.08
1.58
4.13
0.00
3.45
1.67
Note: Data consist of eight years (from 1999 to 2006) of 47 prefectures. Total
number of observation is 376.
Another important variable we should introduce is the proportion of unincorporated rms, which corresponds to w1 in the previous section. If the proportion of unincorporated rms is higher,
the bankruptcy ratio of these rms increases. However, since these
statistics are not surveyed annually and we use only the year 2001
statistics, this variable is not time-dependent in the regression
equation. Its mean is 54.94% and standard deviation is 5.61. We
also include the number of incorporated enterprises as another
time invariant control variable. Its mean is 34,417 and standard
deviation is 43,295. As a performance measure of the prefecture
in each year, we introduce the prefectural GDP growth rate, change
rate of land price, and unemployment rate. We will later discuss
about the bank performance measure stated in the last six rows.
Our main hypothesis is that the bankruptcy ratio of small, in
particular unincorporated, rms decreases as the PL ratio of the
SB becomes higher than that of the RB in the prefecture. Although
our hypothesis predicts a negative relationship as shown in Eq.
(13), this is not trivial. The bankruptcy of unincorporated rms
may increase with the PL ratio of the SB because both borrowers
suffer from the same distress in the regional economy.
3.4. Bankruptcy ratio of small rms and sector specialization
To test the hypothesis, we analyze several panel models of
regression equation based on the Eqs. (10) and (12). Since we need
to include the time invariant explanatory variable, the regression
model is expressed as
ypt X1pt b1 X2pt b2 Z1p d1 Z2p d2 cp upt ;
loans to small rms, then the unobserved amount of these loans to

small rms may affect endogenous variables X1pt and Z1p. The specic industrial structure of each prefecture of course may be another source of endogeniety.
Thus, the assumptions here are
EX1pt cp 0;
EX1pt upt 0;
EZ1p cp 0;
EZ1p upt 0;
EX2pt cp 0;
EX2pt upt 0;
EZ2p cp 0;
EZ2p upt 0:
In other words, all independent variables are assumed to be uncorrelated with upt, the variables in X2pt and Z2p are assumed to be
uncorrelated with cp, but the variables in X1pt and Z1p are possibly
correlated with ci. We treat three main variables, the difference in
PL ratios (Dc), PL ratio of RB (c2), and proportion of unincorporated
rms to total enterprises (w1), as endogenous variables.
The within effect estimators of this model are consistent, but
the coefcients d1 and d2 cannot be obtained as a result of the within transformation. Since these coefcients are important, we also
report Hausman-Taylor (HT) estimators. It is not consistent, but
more efcient if the equation is overidentied, that is, the number
of time-varying exogenous variables is greater than the number of
time-invariant endogenous variables. The HT estimator is a threestep instrumental variable estimator.13
Table 4 reports the main regression results. In regression Eq. (1),
(the log odds ratio of) the bankruptcy ratio of unincorporated rms
is examined as a dependent variable. First, the null hypothesis of
the pooling model is rejected since the F-test statistic is 4.33,
which is signicant at the 1% level. Second, since the v2 statistic
for the Hausman specication test is 14.08, the null hypothesis that
the within estimator and the usual random effect estimator are the
same is rejected at the 5% level.
As expected, the difference in PL ratios has a signicantly negative inuence on the bankruptcy ratio of unincorporated rms. The
within estimator of the coefcient is 0.050. Using clustering robust standard errors, the z-statistic is 1.700 and signicant at
the 10% level. If SBs have higher PL ratios than RBs by 1%, the BR
in the prefecture decreases by 0.74%.14 Thus, our evidence supports
the main hypothesis that the bankruptcy ratio of unincorporated
rms is lower if the difference in PL ratios increases.
Although the consistency of the HT estimator is not guaranteed,
the HT estimator of the coefcient of the difference in PL ratios is
not much different from the within estimator. Since the HT estimate is more efcient, the z-statistic is 2.800 and the coefcient
is signicantly different from zero at the 1% level.
As predicted, the within estimate of the coefcient of the proportion of SB loans is signicantly positive at the 10% level. When
SBs increase loans by 1%, the log odds of the bankruptcy ratio of
unincorporated rms rises by 0.077 because SB loans mainly consist of loans to small rms, in particular, the unincorporated rms.
The HT estimate of this coefcient remains positive, but becomes
insignicant.15
The HT estimate of the coefcient of the proportion of unincorporated rms is positive as expected, but not signicantly differ-
16
where X1pt is a vector of endogenous time-varying variables, X2pt is

a vector of exogenous time-varying variables, Z1p is a vector of
endogenous time-invariant variables, Z2p is a vector of exogenous
time-invariant variables, cp is an unobserved individual effects
term, and upt is the idiosyncratic error.
Possible unobserved effects come from the lending behavior of
other nancial institutions, in particular, major city banks, and the
industrial structure of the prefecture. Although large major banks
have branches and loan ofcers in the local credit market, their
numbers are not reported publicly. For example, if they have many
branches and ofcers in a prefecture and provide a huge amount of
13
In the second step IV estimation, we use all the exogenous variables as
instruments. In the third step IV estimation of the GLS transformed equation, the
instrument variables are within-mean of the time-varying exogenous variables,
within transformed variables of time-varying exogenous variables, and the timeinvariant exogenous variables. See Hausman and Taylor (1981) and Baltagi and
Khanti-Akom (1990).
14
To know the effect of an independent variable on a dependent variable, we need
to retransform the log odds ratio into the original BR. The marginal effect is calculated
as dBR/dx = bexp (y)/{1 + exp(y)2}.Since the coefcient is 0.050 and the median of BR/
(100 BR) is 0.221, we have dBR/dx = 0.05 0.221/{1.2212} = 0.74.
15
Although we do not report in the table, the estimated coefcient becomes
signicant when we include the proportion of SB loans as endogenous variables. The
coefcient is 0.058 and the z-statistic is 2.39.
865

Table 4
Regression analysis of bankruptcy ratio (unincorporated and incorporated with small net worth).
Regression equation
Dependent variable
Bankruptcy ratio of
unincorporated rms
421
4.33
14.08
Bankruptcy ratio of unincorporated and

incorporated NW < 10
423 417
5.80
22.46
Bankruptcy ratio of unincorporated and

incorporated NW < 50
Number of observation
F test (null = pooling vs. xed)
Hausman test (null = random vs.
xed) v2(8)
X1pt
Difference in PL ratio
PL ratio of RB
Within estimator
HT estimator
Within estimator
HT estimator
Within estimator
HT estimator
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
0.050
0.001
1.700
0.020
0.058
0.010
2.800
0.390
0.030
0.027
2.320
1.870
0.034
0.032
4.270
3.220
0.040
0.018
2.030
0.940
0.049
0.028
3.030
1.360
0.056
0.880
0.001
0.040
0.055
1.760
Z1p
Proportion of unincorporated
rms
X2pt
Proportion of SB loans to total
loans
Unemployment rate
0.077
1.990
0.019
1.550
0.019
1.270
0.001
0.330
0.021
0.690
0.006
1.010
0.032
0.049
0.098
2.080
1.220
1.070
0.030
0.043
0.058
1.840
2.540
0.660
0.000
0.011
0.079
0.020
1.420
1.570
0.000
0.015
0.052
0.040
2.340
1.550
0.022
0.008
0.265
1.770
0.560
3.400
0.021
0.002
0.200
1.580
0.130
3.070
0.261
0.700
0.244
2.030
0.117
0.660
0.472
0.454
0.537
0.417
0.150
Z2p
Log of number of incorporated
rms
Constant
rc
ru
q
2.05
15.51
2.246
1.127
0.669
0.739
1.221
0.675
0.766
0.330
0.409
0.423
0.259
0.727
1.130
3.129
0.354
0.257
0.655
1.430
4.237
0.518
0.541
0.478
6.78
Regression equation
Dependent variable
Hausman test (null = random vs. xed) v2(8)
Bankruptcy ratio of incorporated with NW < 10

423
17.82
15.61
Bankruptcy ratio of incorporated with 10 < NW < 50

423
5.53
21.25
X1pt
PL ratio of RB
Within estimator
HT estimator
Within estimator
HT estimator
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
0.006
0.010
0.760
1.150
0.008
0.009
1.070
1.020
0.024
0.027
2.180
2.090
0.028
0.031
3.780
3.420
0.041
1.720
0.008
0.390
0.007
0.004
0.006
0.034
1.650
0.700
1.000
1.130
0.002
0.004
0.014
0.024
0.540
0.620
2.470
0.780
0.185
3.781
0.426
0.221
0.787
1.350
1.510
0.268
3.824
0.331
0.234
0.667
2.410
1.890
Z1p
X2pt
Unemployment rate
0.021
0.004
0.006
0.035
Z2p
Log of number of incorporated rms
Constant
rc
ru
q
0.438
0.223
0.794
1.920
0.800
0.710
1.000
0.018
0.003
0.011
0.047
1.530
0.550
1.750
1.110
0.422
0.393
0.236
0.735
1.390
Note: Prefecture (p)- year (t) panel data. Dependent variable is log odds of bankruptcy ratio, y = log (BR/(1 BR)). NW: net worth. X1 is a vector of endogenous time-varying
variables, X2 is a vector of exogenous time-varying variables, Z1 is a vector of endogenous time-invariant variables, Z2 is a vector of exogenous time-invariant variables. rc is a

standard deviation of individual disturbances and ru is a standard deviation of idiosyncratic disturbances. q is a serial correlation of error dened as r2c = r2c r2u . Standard
errors of within estimator is cluster-robust.
The test statistics is signicant at 10% level.
ent from zero. The estimated coefcient of the log of the number
of incorporated rms is expectedly negative, but it is not signicant either. The coefcient of PL of RB is negative, but insignicant. Both estimates of the coefcient of prefectural GDP growth
rate and the HT estimate of the land price change rate are significantly negative.
Thus, we have the empirical evidence that the bankruptcy of
unincorporated rms is negatively inuenced by the excessive PL
866
loans of small banks. 16 Did SBs specialize only in unincorporated

rms? A natural extension of the above analysis is that SBs might
have greater expertise in both small incorporated rms and unincorporated rms.
In regression Eq. (2), the denition of small rms is extended to
the unincorporated rms and the incorporated rms with NW < 10
million yen. Going back to Table 3, the sample mean of the dependent variable is 18.80 + 34.11 = 52.91%. In regression Eq. (3), BR includes the number of bankrupt incorporated enterprises with
NW < 50 million yen as well. The sample mean of the dependent
variable is exceedingly high (95.81%).
The within estimate of the coefcient of the difference in PL ratios is 0.030 in Eq. (2) and 0.040 in Eq. (3). Both are signicant at
the 5% level, and the HT estimates are also signicantly negative.
However, since the estimate is smaller than that of Eq. (1), it is suggested that SBs specialize less in lending to these incorporated
rms than lending to the unincorporated rms.
In regression Eqs. (4) and (5), the number of unincorporated
rms is excluded from the denition of BR. The estimates of the
coefcients of the difference in PL ratios are much different from
those of Eqs. (2) and (3). They are insignicant and positive in
Eq. (4) and are also signicant in Eq. (5). This result is another evidence supporting that SBs have no expertise in lending to these
incorporated borrowers.
Comparing these two equations, the result of Eq. (4) is more
ambiguous than Eq. (5). In the within effect model, only the proportion of SB loans is signicantly negative, which is consistent
with our prediction. This is also true for the HT model. In addition,
the proportion of unincorporated rms is negatively signicant. In
Eq. (5), the coefcient of the difference in PL ratios is signicantly
positive. This result is simply interpreted as the reverse of the coin
of Eq. (1).
3.5. Alternative classications of small rms
In the above analysis, I classied small rms by the amount of
net worth. Other classications might be possible. Table 5 reports
the bankruptcy ratios of small rms with a small number of
employees and a small amount of outstanding debt. The share of
bankrupt rms with less than 4 employees is unexpectedly high
(51.0 %). Further, it becomes about 73% if we include rms with
less than 10 employees. The mean of the bankruptcy ratio of borrowers with 10 million yen of outstanding debt is very small (only
1.3%), that of borrowers with more than 10 million and less than 50
million yen is 31.4%, and that of borrowers with more than 50 million yen and less than 100 million yen is 22.5%. These three classes
comprise about 55% of the bankruptcies.
Thus, a majority of bankruptcies are of such small rms. We
should note that, as previously mentioned briey, the law restricts
the borrowers of SBs to small rms. It requires that the number of
employees be less than three hundred.17 Apart from the effectiveness of this restriction, we might be able to nd that SBs might have
greater expertise in also lending to these rms.
The bankruptcy of small rms with less than 5 employees is
analyzed in regression 6 of Table 6, that of rms with less than
10 employees in regression 7, and that of rms with more than 4
and less than 10 in regression 8.
16
In this analysis, I focus on the difference between the behavior of SBs (small
banks) and RBs (relatively large banks). However, many readers may be interested in
the inuence of continuous bank size on the efciency of relationship loans. As a
robustness check, I tried shinkin bank asset size as an independent variable. Although
not reported here, the estimated coefcient was signicantly negative, which is
consistent with the results of Table 4.
17
An incorporated rm with capital larger than 900 million yen cannot borrow from
SBs even if the number of employees is less than three hundred.
Table 5
Bankruptcy ratio classied by debt and employees.
Variable
Mean
Std. dev.
Min
Max
Bankruptcy ratio
Employees 6 4
5 6 Employees 6 9
10 6 Employees 6 29
30 6 Employees 6 99
100 6 Employees 6 299
300 6 Employees
51.0
22.0
18.9
0.6
0.1
2.7
9.5
4.7
5.1
0.6
0.2
5.6
25.7
6.3
3.9
0.0
0.0
0.0
83.7
37.5
31.3
2.8
1.1
42.1
Debt < 10
10 6 Debt < 50
50 6 Debt < 100
100 6 Debt < 1000
1000 6 Debt < 10,000
10,000 6 Debt
1.3
31.4
22.5
32.9
6.1
5.2
2.0
9.3
4.8
6.8
2.3
2.2
0.0
11.7
9.1
9.8
0.5
0.0
13.7
72.6
36.6
52.1
15.0
14.6
Note: Sample mean (%) of bankruptcy ratio by prefecture. Data consist of eight years
(from 1999 to 2006) of 47 prefectures. Total number of observation is 376. Unit of
debt amount is million yen.
The within estimates of the coefcients of the difference in PL

ratios are signicantly negative in Eqs. (6) and (7), but are not so
in Eq. (8). In other words, the difference in PL ratios has a signicant effect only on the bankruptcy of rms with less than 5
employees. This is similar to the result of regression 1 and seems
quite reasonable because it is likely that unincorporated rms have
such a small number of employees. Table 7 reports the regression
analysis when we dene a small rm as that which have less than
10 million yen or less than 50 million yen of outstanding debt. The
coefcients of the difference in PL ratios are not signicant in both
equations.
The results of Tables 6 and 7, as compared to those of Table 4,
might have implications consistent with the previous theory in
the sense that SBs have greater expertise particularly in soft
information processing. This notion ts unincorporated rms because they hardly have veriable documentation. In this sense,
the most extreme result we expected is that the difference in
PL ratios has no impact on all the equations of Tables 6 and 7.
However, it is natural that we nd a signicant effect in Eq. (6)
because we expect that the unincorporated rms have a small
number of employees.
4. Repercussion of excessive holdings of non-performing loans
The previous empirical investigation naturally raises further
questions: Did not excessive problem loans of SBs increase LBBs later? Did these borrowers whose repayments were delayed really
recover from the nancial distress? Although we supposed in the
previous argument that we can distinguish between bankrupt borrowers and recovering borrowers at t = 2 before the actual repayments are made, this is not realistic. It is more realistic that
some problem loans would become LBBs and others would not.
Thus, it is necessary for us to conrm that a higher PL ratio did
not result in a higher LBB ratio in the subsequent years.
To answer this question, we analyze the dynamic panel data
regression equation, using the bank-year panel data, instead of
the prefecture-year data. As a dependent variable, we examine
the difference between LBB ratio of each SB and that of the mean
of RBs in the same prefecture. The model is expressed as
LBBit
p
X
qj LBBi;tj Xit b Zit d ci uit ;
17
j1
where LBBi,tj is a jth lag of the dependent variable. Xit and Zit are
vectors of other regressors. ci is an unobserved individual effects
term and uit is the idiosyncratic error. These disturbances satisfy
867

Table 6
Regression analysis of bankruptcy ratio (a small number of employees).
Regression equation
Dependent variable
Bankruptcy ratio of rms with

employees 6 4
423
7.45
3.540

employees < 10
423
6.85
7.880

4 < employees < 10
423
4.22
8.870
Hausman test (null = random vs.
xed) v2(8)
Hausman Taylor estimation
X1pt
PL ratio of RB
Within estimator
HT estimator
Within estimator
HT estimator
Within estimator
HT estimator
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
0.023
0.049
1.840
3.380
0.024
0.051
2.680
4.580
0.029
0.051
2.170
3.670
0.030
0.054
3.100
4.410
0.008
0.012
0.870
0.880
0.005
0.014
0.670
1.540
0.001
0.030
0.006
0.260
0.001
0.060
0.007
1.700
0.006
1.270
0.003
1.300
0.012
0.003
0.017
1.740
0.450
0.470
0.015
0.013
0.004
2.490
2.330
0.150
0.005
0.040
0.011
0.150
0.199
0.339
0.284
0.587
0.090
1.187
0.164
0.231
0.333
0.930
Z1p
Proportion of unincorporated
rms
X2pt
Proportion of SB loans to total
loans
Unemployment rate
Z2p
Log of number of incorporated
rms
Constant
rc
ru
q
0.021
0.012
0.002
0.012
0.075
0.337
0.287
0.581
1.240
1.450
0.170
0.210
0.190
0.022
0.003
0.010
0.003
0.925
0.383
0.315
0.597
1.210
0.360
0.740
0.060
2.260
0.004
0.008
0.005
0.480
0.990
0.130
0.021
0.150
0.714
0.385
0.312
0.604
0.290
0.006
0.015
0.013
0.004
1.283
0.183
0.233
0.382
0.700
2.530
2.220
0.110
5.410


Table 7
Regression analysis of bankruptcy ratio (small size of debt).
Regression equation
10
Dependent variable
Hausman test (null = random vs. xed) v2(8)
Bankruptcy ratio of rms with debt 6 10

423
6.78
12.590
Bankruptcy ratio of rms with debt 6 50

423
8.15
13.82
Hausman Taylor estimation
X1pt
PL ratio of RB
Within estimator
HT estimator
Within estimator
HT estimator
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
Coef.
z stat.
0.013
0.030
0.960
2.130
0.006
0.036
0.700
3.170
0.004
0.032
0.350
3.540
0.002
0.037
0.270
4.070
0.014
0.680
0.013
0.770
0.002
0.029
0.038
0.025
0.600
3.910
5.130
0.680
0.002
0.016
0.026
0.023
0.490
2.720
4.420
0.750
0.134
2.526
0.305
0.290
0.526
1.180
1.230
0.101
1.226
0.272
0.233
0.577
1.030
0.690
Z1p
X2pt
Unemployment rate
Z2p
Log of number of incorporated rms
Constant
rc
ru
q
0.005
0.029
0.034
0.011
0.390
3.660
3.550
0.200
0.509
0.320
0.292
0.546
1.600
0.004
0.016
0.022
0.011
0.400
3.180
3.350
0.300
0.469
0.281
0.235
0.590
1.800

868
Table 8
Regression analysis of LBB ratios (dynamic panel data model).
Regression equation
11
12
13
14
15
16
Dependent variable yit
Difference in LBB
ratios
2223
371
43
64.116
Difference in LBB
ratios
2222
370
84
68.570
Difference in LBB
ratios
2222
370
80
74.793
Difference in LBB
ratios
2222
370
80
78.745
Difference in LBB
ratios
2222
370
80
76.016
Difference in LBB
ratios
2222
370
80
77.902
8.461
0.339
7.675
1.071
8.093
0.812
8.054
0.972
8.134
0.821
8.086
0.815
Number of groups
Number of instruments
Sargan-Hansen test (null = overidentifying
constraint) v2
Serial correlation test (null = no AR (1))
Serial correlation test AR (2)
Arellano-Bond
estimator
z stat.
Coef.
z stat.
Lag of dependent variables

yit1
0.579
yit2
0.015
9.990
0.570
0.549
0.022
9.310
0.710
0.571
0.019
9.770
0.660
0.587
0.034
9.820
1.220
0.575
0.023
9.730
0.820
0.582
0.028
9.820
0.970
PL ratio of SB
t
t1
t2
0.920
1.730
1.920
0.121
0.065
0.010
2.950
1.630
0.370
0.085
0.067
0.007
2.420
1.710
0.300
0.074
0.078
0.016
2.100
1.870
0.630
0.085
0.071
0.009
2.270
1.840
0.360
0.074
0.071
0.016
2.170
1.670
0.610
Capital asset ratio

t
t1
t2
0.269
0.006
0.360
2.300
0.040
3.060
0.219
0.058
0.313
2.080
0.480
2.910
0.183
0.074
0.253
1.540
0.570
2.300
0.192
0.022
0.319
1.820
0.180
2.730
0.186
0.015
0.278
1.620
0.120
2.600
Liquidity ratio
t
t1
t2
0.024
0.013
0.006
1.210
0.630
0.380
0.024
0.014
0.018
1.250
0.700
1.230
0.029
0.017
0.014
1.260
0.730
0.920
0.020
0.014
0.012
0.890
0.690
0.840
0.023
0.017
0.013
1.040
0.740
0.820
0.308
3.460
0.286
0.820
0.732
2.380
0.003
0.540
0.001
0.170
0.042
0.950
0.048
1.490
0.003
1.750
Loan amount per

number of
borrowers
Log of number of
borrowers
Number of branches
per asset
Number of ofcers per
branch
Loan amount per
ofcer
Coef.
0.030
0.081
0.049
0.001
0.730
Note: Bank (i)- year (t) panel data. Dependent variable is the difference in LBB ratio of SB from that of average RB in the same prefecture. Independent variables are those of
SBs. Two-step GMM estimator. Standard errors are computed by Windmeijers bias-corrected robust estimator. Sargan-Hansen test is performed under the one-step
estimator of homoskedastic variance.
the standard assumptions. We assume that the variables in Xit are

strictly exogenous (uncorrelated with uit) while the variables in Zit
may be endogenous. That is, E(zituis) 0 for s 6 t and E(zituis) = 0
for s > t. We estimate a two-step GMM estimator following Arellano
and Bond (1991).
In regression Eq. (11) of Table 8, we include one-year and twoyear lags of the dependent variable, and the current, one-year, and
two-year lags of problem loans of SB. The number of observations
is 2223, and the number of instruments is 43. Sargan-Hansen test
disappointedly rejects the null of the overidentifying restriction.
The specication of regression Eq. (12) satises the overidentifying restriction. The Sargan-Hansen test statistic is not signicant
even at the 10% level. Since the serial correlation test of the rst
differenced error of uit at order 1 is rejected while that of order 2
is not, we could conrm that the necessary assumption of no serial
correlation of uit is satised.
The coefcient of the one-year lag of the dependent variable is
0.549 and signicant at the 1% level. The two-year lag of the dependent variable has no signicant effect. Thus, if SBs allow excessive
LBB at year t 2, the excess will remain for one year, but will disappear afterwards.
As suggested by the result of Table 4, the current PL ratio has a

negative impact on the LBBit. However, the lags of PL ratio have a
positive, but no signicant impact on the dependent variable. In
other words, even if a SB has a PL ratio higher than the average
PL ratio of RBs in the same prefecture at year t 2, its LBBs would
be the same as that of average RB afterwards.
We examine two balance sheet variables and some technological structure concerning the expertise in relationship loans as
explanatory variables. One hypothesis may be that SBs can afford
to have a higher LBB ratio if they have a stronger balance sheet,
that is, a higher capital/asset ratio or a higher liquidity ratio.
Going back to Table 3, the average capital asset ratio of SBs is
5.51% and the average liquidity ratio is 19.18%. In the estimation
of regression Eq. (12) of Table 8, the current capital asset ratio has
a negative impact on the dependent variable. If a SB has a lower
capital asset ratio at t, LBB increases. The one-year lag has no signicant impact, but the two-year lag has a signicantly positive
impact. The liquidity ratio has no signicant impact on the
dependent variable.
As the technological structure concerning relationship loans, we
examine ve variables. Basically, my hypothesis is that SBs have
lower LBB ratios or higher PL ratios if they have technological

structures suitable to specialization in relationship loans.
First, the loan amount per borrower has a signicantly positive
coefcient. This might be consistent with our intuition because the
larger the scale of borrowing rm is, the more transparent is the
accounting document. SBs are at comparative disadvantage in such
loans.
The sign of the log of the number of borrowers is predicted to be
ambiguous. It might be positive because it costs more to monitor a
larger number of borrowers. However, it might be negative
because a larger number implies that the loan portfolio of SBs consists of smaller rms with whom SBs enjoy comparative advantages. The estimated coefcient is negative, but insignicant.
The number of branches per asset is predicted to has a negative
sign because it costs less to monitor the borrowers if the larger
number of branches implies the shorter distance from each borrower. The estimated coefcient is negative as expected, but insignicant. For a similar reason, the number of ofcers per branch is
predicted to have a negative sign and the loan amount per ofcer
is predicted to have a positive sign. However, the estimated coefcients of these are not signicant.
To avoid the problem of multicollinearity between these intimately connected variables, Eqs. (13)(16), respectively, examine
each technological variable individually. Among these, we found
that the log of the number of borrowers has a signicantly negative
impact on the LBB ratio.
From this table, we conclude that higher holdings of PL in the past
does not raise LBB ratio after two years. This evidence has implications that most of the problem loans whose repayments were
delayed were recovered without bankruptcy. However, disappointedly, we could not provide the evidence for the hypothesis that SBs
have technological structures suitable for relationship loans.
5. Related literature
Here, we discuss here how this paper is related to the past
empirical and theoretical literature. As already mentioned in the
introduction, among lending relationship literature, Berger et al.
(2005) is the most related article. They provide evidence that bigger banks are more apt to lend to rms that are larger or that have
better accounting records. The asset size of a rm has a positive
impact on the asset size of the bank from which the rm borrows.
Moreover, banks asset size increases if the rm has documentation
such as nancial statements or accounting records. This paper is
consistent with this observation. The empirical evidence of this paper differs in that (1) it does not analyze asset size itself, but the
type of bank of which operating area is limited to small one and
of which borrowing customer is also limited to small ones by the
law, and (2) it analyzes the bankruptcy of small rms, in particular,
the unincorporated rms that have no ofcial accounting records.
Several studies focus on how the ownership of a bank affects its
performance and behavior of banks. For example, Berger et al.
(2001) conclude that large and foreign-owned institutions may
have difculty in extending relationship loans to opaque small
rms. Their results of logit regression show that the rms with larger loans are more likely to borrow from banks with the largest asset size or a foreign-owned bank. Though we have not emphasized
much so far, it should be stressed that a SB is not a stock company,
but a mutual cooperative. The difference in the objective function
might be the rst reason why SB behaves differently from RB.
The SB might afford to hold higher problem loans because they
are not closely monitored by market investors.
The existence of specialization is empirically examined by a few
articles. Carey et al. (1998) compare corporate loans made by
banks with those of nance companies and nd that they are
869
equally likely to nance information-problematic rms in the logit

estimation. However, nance companies tend to serve observably
riskier borrowers, particularly more leveraged borrowers. Carey
et al. (1998) concludes that the main dimension of specialization
is observable risk. In a similar way, Shin and Kolari (2004) examine
credit specialization in Japan. They nd that depository institutions
do not have information advantages relative to non-depository
institutions. However, they argue that the rms with information
problems are more likely to take higher proportions of loans from
main banks than from non-main banks.
Second, theoretical studies usually assume that the success
probability of an investment project depends on the contract type
or whether or not nancial institutions monitor the borrower, as I
mentioned earlier. Chemmanur and Fulghieri (1994) argue that
banks who care for reputation devote more resources toward evaluating the project and they are able to make efcient liquidation
decisions when the rm is in nancial distress. They make inefcient liquidation decisions with lower probabilities because they
receive more precise signals, i.e. the probabilities of receiving good
signals conditioned on true good states are higher. von Thadden
(1995) also argues that the monitoring enables banks to make efcient liquidation decisions because they learn about the rms
quality in addition to the short-term outcomes of investments.
Cantillo and Wright (2000) also stress that nancial intermediaries
have better reorganizational skills. Such argument of efcient liquidation decision might be one of the reasons why SBs have higher
non-performing loans than RBs.
Third, recent works on the non-performing loans problem of Japan since the early 1990s are also related to this paper. Peek and
Rosengren (2005) nd that Japanese rms are more likely to receive additional bank credit if they are in poor nancial conditions,
because troubled Japanese banks have an incentive to allocate
credit to severely impaired borrowers in order to avoid the realization of losses on their own balance sheets. They estimated a probit
equation where the dependent variable takes the value of one if
loans increase within a year. The estimated coefcients of ROA
and the ratio of net working capital to total asset of borrowers
were signicantly negative. Although they claim that this result
supports their misallocation of credit hypothesis, it seems to me
that their evidence only supports that credits are allocated to the
rms in poor nancial conditions. We can say that credits are misallocated only when net present value of loans is negative. However, their variables would not be sufcient to estimate the NPV
of future loan repayment.18
Hoshi (2006) estimated a similar probit regression as Peek and
Rosengren (2005). Instead of whether or not receiving additional
credit, the dependent variable takes one if the rm receives nancial help from lenders. 19 Hoshi nds a reasonable evidence that the
rm is more likely to receive nancial help if its prot rate is lower
and the leverage ratio is higher.20
Since my view is that econometricians have great difculty in
evaluating whether or not an investment project is efcient and
in predicting which rm is prospective, I do not investigate this
problem in further detail. However, a few points are worth noting.
18
The standard nance textbook teaches us that the optimal decision making is as
follows: y = 1 if NPV > 0 and y = 0 otherwise, where y = 1 means providing loans.
Probably, they assume that the NPV can be well explained by only six variables:
current return on assets, current net working capital, two dummy variables
concerning past stock return, current leverage, and current change rate of sales.
19
Hoshi made an ad hoc presumption that the rm receives nancial help if the
actual payment falls short of the estimated minimum interest payment. He refers to
such rms as zombies.
20
Unlike these two, this paper does not focus on large listed rms, but on small
rms. Since unincorporated rms have relatively small debt, loss from bankruptcy
can be resolved relatively easily by the cooperation of parents, relatives, or a person
enjoying a close relationship.
870
First, according to Claessens and Klapper (2005), Japans bankruptcy rate is the eighth lowest among 35 countries throughout
the 1990s.21 This is surprising because the Japanese economy is said
to have experienced a lost decade, during which the bankruptcy rate
should have been higher than before. The ndings of this paper suggest the possibility that SBs might have played the role in maintaining the bankruptcy rate so low.
Second, although more important, the credit guarantee programs provided primarily by local governments are prevalent
and available for small and medium rms in Japan. Thanks to this
guarantee, small banks can afford to cover risks from these borrowers and postpone temporarily such bankruptcies.
Third, although average borrowing per unincorporated rm
may be relatively small, the number of such borrowers is so large
that the political pressure might be considerably large. The SBs
as mutual cooperatives might nancially assist the borrowers because the local government cannot politically tolerate the bankruptcy of a large number of borrowers.
6. Conclusion
This paper provides empirical evidence that small banks specialize more in relationship loans to unincorporated rms than relatively large banks in the local credit market of Japan. In particular,
it shed some light on how the lending relationship between small
banks and small rms inuences their bankruptcy ratio. Our nding is that the bankruptcy ratio of unincorporated rms is lower
when the non-performing loans ratio of small banks is higher than
that of relatively large banks. This empirical nding is consistent
with the theory developed by Boot and Thakor (2000) and Stein
(2002). Furthermore, we conrm that excessive non-performing
loans in small banks did not consequently lead to a higher bankrupt loan rate in the subsequent years.
Acknowledgments
The author thanks Yoshiaki Ogura for his helpful comments at
the 2010 annual meeting of Japanese Economic Association and
Shinsuke Kambe at Contract Theory Workshop. I am also grateful
to Nobuyoshi Yamori for making a special effort to provide me
with the opportunity to use the database. Comments and suggestions from an anonymous referee is greatly appreciated. This work
was supported by the Nitto Foundation and a Grant-in-Aid for Scientic Research.
References
Arellano, M., Bond, S., 1991. Some tests of specication for panel data: Monte Carlo
evidence and an application to employment equations. Review of Economic
Studies 58, 277297.
21
According to them, the bankruptcy rates of some countries are as follows: Japan
(0.22%), UK (1.85%), Germany (1.03%), France (2.62%), the US (3.65%), Canada (2.96%),
Italy (0.54%).
Baltagi, B., Khanti-Akom, S., 1990. On efcient estimation with panel data: an
empirical comparison of instrumental variables estimators. Journal of Applied
Econometrics 5, 401406.
Berger, A., Udell, G., 2006. A more complete conceptual framework for SME nance.
Journal of Banking & Finance 30, 29452966.
Berger, A., Klapper, L., Udell, G., 2001. The ability of banks to lend to informationally
opaque small businesses. Journal of Banking & Finance 25, 21272167.
Berger, A., Miller, N., Petersen, M., Rajan, R., Stein, J., 2005. Does function follow
organizational form? Evidence from the lending practices of large and small
banks. Journal of Financial Economics 76, 237269.
Boot, A., 2000. Relationship banking: what do we know? Journal of Financial
Intermediation 9, 725.
Boot, A., Thakor, A., 2000. Can relationship banking survive competition? Journal of
Finance 55, 679713.
Cantillo, M., Wright, J., 2000. How do rms choose their lenders? An empirical
investigation. Review of Financial Studies 13, 155189.
Carey, M., Post, M., Sharpe, S., 1998. Does corporate lending by banks and nance
companies differ? Evidence on specialization in private debt contracting.
Journal of Finance 53, 845878.
Chemmanur, T., Fulghieri, P., 1994. Reputation, renegotiation, and the choice
between bank loans and publicly traded debt. Review of Financial Studies 7,
475506.
Claessens, S., Klapper, L., 2005. Bankruptcy around the world: explanations of its
relative use. American Law and Economics 7, 253283.
Dahiya, S., Saunders, A., Srinivasan, A., 2003. Financial distress and bank lending
relationships. Journal of Finance 58, 375399.
Degryse, H., Ongena, S., 2005. Distance, lending relationships, and competition.
Journal of Finance 60, 231266.
Detragiache, E., Garella, P., Guiso, L., 2000. Multiple versus single banking
relationships: theory and evidence. Journal of Finance 55, 11331161.
Diamond, D., 1991. Monitoring and reputation: the choice between bank loans and
directly placed debt. Journal of Political Economy 99, 689721.
Elyasiani, E., Goldberg, L., 2004. Relationship lending: a survey of the literature.
Journal of Economics and Business 56, 315330.
Goddard, J., McKillop, D., Wilson, J., 2008. The diversication and nancial
performance of US credit unions. Journal of Banking & Finance 32, 18361849.
Grunert, J., Weber, M., 2009. Recovery rates of commercial lending: empirical
evidence for German companies. Journal of Banking & Finance 33, 505513.
Hackethal, A., 2004. German banks and banking structure. In: Krahnen, J., Schmidt,
R. (Eds.), German Financial System. Oxford University Press, Oxford.
Hausman, J., Taylor, W., 1981. Panel data and unobservable individual effects.
Econometrica 49, 13771398.
Holmstrom, B., Tirole, J., 1997. Financial intermediation, loanable funds, and the real
sector. Quarterly Journal of Economics 112, 663691.
Hoshi, T., 2006. Economics of the living dead. Japanese Economic Review 57, 3049.
Hoshi, T., Kashyap, A., Scharfstein, D., 1990. The role of banks in reducing the costs
of nancial distress in Japan. Journal of Financial Economics 27, 6788.
Peek, J., Rosengren, E., 2005. Unnatural selection: perverse incentives and the
misallocation of credit in Japan. American Economic Review 95, 11441166.
Petersen, M., Rajan, R., 1994. The benets of lending relationships: evidence from
small business data. Journal of Finance 49, 337.
Petersen, M., Rajan, R., 1995. The effect of credit market competition on lending
relationships. Quarterly Journal of Economics 110, 407443.
Sharpe, S., 1990. Asymmetric information, bank lending, and implicit contracts: a
stylized model of customer relationships. Journal of Finance 45, 10691087.
Shin, H., Kolari, J., 2004. Do some lenders have information advantages? Evidence
from Japanese credit market data. Journal of Banking & Finance 28, 23312351.
Stein, J., 2002. Information production and capital allocation: decentralized versus
hierarchical rms. Journal of Finance 57, 18911921.
von Thadden, E., 1995. Long-term contracts, short-term investment and monitoring.
Review of Economic Studies 62, 557575.

Bankruptcies of Small Firms and Lending Relationship

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bankruptcies of Small Firms and Lending Relationship

Uploaded by

Copyright:

Available Formats

Journal of Banking & Finance 36 (2012) 857870

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance

Bankruptcies of small rms and lending relationship

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

relationship lenders to small rms in the sense that they have

Small decentralized banks are at a comparative advantage when

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

2. Theoretical background and econometric strategy

Consider a regional economy with two types of banks and two

That is, bank i has a comparative advantage in relationship

Fig. 1. Return structure when relationship loans are made.

if bank i provides a transaction loan to the borrower, and is

if a relationship loan is provided. The borrowers expected return is

where K(=0) is an exogenously given parameter that represents the

Thus, the bank offers a relationship loan if qv i k 1 hY > C or

The banks have an incentive to provide relationship loans to high

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

Proposition. Suppose that bank 1 has greater expertise in offering

If the borrower quality is high enough (h > ^

rms in sector S1 with ^

and ai aSi 1 aSi 2 .

The ratio of problem loans to total loans of each bank is given as

This point was suggested by the anonymous reviewer.

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

which is an increasing function of v S11 and a decreasing function of

increases. Thus, we have dBR=dv 11 < 0 and dBR=dv 22 > 0.

aS11 w1 q 1 Eh v 0S1 1 ; aS22 w2 q 1 Eh v 0S2 2 ;

An increase in c1 decreases aS11 or a decrease in c2 increases aS22 and

ypt b0 b1 4 cpt b2 c2;

This is because only bank 1 provides relationship loans to sector S1.

Since our BR data is prefecture-time data, subscripts p and t denote

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

grate the last three parts and refer to it as problem loans(hereafter

3.2. Small business rms and the bankruptcy ratio

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

Number of bankrupt rms

Incorporated (by net worth)

100 6 NW < 500

500 6 NW < 1000

Average per year

Incorporated (by net worth)

100 6 NW < 1000

Number of existing rms (as of 2001)

3.3. Some preliminary results and hypotheses

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

ypt X1pt b1 X2pt b2 Z1p d1 Z2p d2 cp upt ;

loans to small rms, then the unobserved amount of these loans to

where X1pt is a vector of endogenous time-varying variables, X2pt is

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

Bankruptcy ratio of unincorporated and

Bankruptcy ratio of unincorporated and

Bankruptcy ratio of incorporated with NW < 10

Bankruptcy ratio of incorporated with 10 < NW < 50

The test statistics is signicant at 10% level.

The test statistics is signicant at 5% level.

The test statistics is signicant at 1% level.

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

loans of small banks. 16 Did SBs specialize only in unincorporated

The within estimates of the coefcients of the difference in PL

qj LBBi;tj Xit b Zit d ci uit ;

K. Shimizu / Journal of Banking & Finance 36 (2012) 857870

Bankruptcy ratio of rms with

Bankruptcy ratio of rms with