International Journal of Industrial Organization
19 (2001) 1583–1602
www.elsevier.com / locate / econbase
Non-price strategic behavior: the case of bank branches
Moshe Kim a , *, Bent Vale b
a
Department of Economics, University of Haifa, Haifa 31905, Israel
b
Norges Bank ( The Central Bank of Norway) C51, Box 1179, and
Norwegian School of Management, Sentrum, N-0107 Oslo, Norway
Received 19 December 1999; received in revised form 15 February 2000; accepted 26 February 2000
Abstract
We perform an empirical study of banks’ branching decisions as a strategic non-price
variable in an oligopolistic setting. Using panel data of banks from Norway, we find clear
evidence that banks act strategically in their branching decisions, taking into consideration
the future response from rival banks. The analysis is applied to a unique data set which
covers the entire banking sector during both pre- and post-banking crisis periods, where
very different types of conduct are found in each of these periods both for banks and
borrowers. Moreover, we find that a bank specific branch-network does not confer
externality on other banks. As a result branch network affects only market shares but not
market size.  2001 Elsevier Science B.V. All rights reserved.
JEL classification: G21; L13
Keywords: Bank branch networks; Oligopolistic setting; Borrowers; Norway
1. Introduction
This article is concerned with two important non-price aspects of conduct
among firms in an oligopolistic setting: (i) the existence and effects of informational externalities on conduct, and (ii) the impact of changes in sectoral
* Corresponding author. Fax: 1972-4-824-0059.
E-mail address: [email protected] (M. Kim).
0167-7187 / 01 / $ – see front matter  2001 Elsevier Science B.V. All rights reserved.
PII: S0167-7187( 00 )00064-3
1584
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
fundamentals on conduct. These issues are addressed in the context of the
branch-banking firm in the provision of loans.
To date, existing empirical research pertaining to the economic performance of
oligopolistic banking markets has increasingly focused on the role of banks’
behavior in determining market outcomes. These empirical studies of bank
conduct have relied on either price or quantity (Nathan and Neave, 1989; Shaffer,
1993; Berg and Kim, 1994, 1998). Some recent theoretical banking studies,
however, discuss the importance of non-price (the branch network) competition
and its effects on banking markets. See for example Degryse (1996) and Matutes
and Padilla (1994). In oligopolistic or oligopolistically-competitive markets, nonprice considerations may be the most important tool by which firms differentiate
themselves and extract market power.
Recently, several empirical studies have dealt with the issues concerning banks’
optimal branch-network. Two of these studies model branches as a strategic
variable in the market for deposits, in order to study the impact of deregulation in
some European countries (Cabral and Majure, 1994; Cerasi et al., 1997). Barros
(1995) studies the growth of bank branches in Portugal. However, he follows a
partial equilibrium approach where banks are assumed to make independent
(branching) decisions. In a later study Barros (1999) analyzes pricing decisions
(only) and investigates aspects of product differentiation induced by (exogenous
branch) location in local markets. These studies focus on the deposits market.
In the present article we depart from the above studies in two respects. Firstly,
we consider the role of the branch-network in the provision of loans. This is based
on the view that information collection and processing lie at the heart of banks’
operation and is reflected in an increasing number of theoretical contributions. See
for example Diamond (1984), the articles in Mayer and Vives (1993) and Freixas
and Rochet (1997).1 Secondly, we estimate a model of branching decision where
banks explicitly take account of both their own existing network and their
expectation of rival’s choices.
One of the main features of the institutional structure that facilitates the creation
and processing of information may be attributed to the branch network. By
introducing branches in a certain geographical area banks can better obtain and
process borrower-specific local information, and thus maintain the quality of their
loan portfolio. In fact, as has been recently documented by Jayaratne and Strahan
(1996), relaxation of the US branching regulation has been an important source of
increase in the rate of real per capita growth in income and output. This growth is
shown there to have emanated from improved loan monitoring and screening
which was the result of the branch network proliferation. An earlier paper by
1
Other contributions are those of Leland and Pyle (1977), Allen (1985) and Chan et al. (1986). An
earlier empirical paper by Mester (1992) estimates a cost function based on information-theoretic
considerations, and where outputs are categorized according to their required monitoring intensity,
though no strategic aspects are present in that study.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1585
Evanoff (1988) documents support, though not entirely conclusive, for the
contention that branches are used for the purpose of preemptive behavior.
The first issue we focus on in this paper relates to the role of the non-price
strategic variable — branches — in the creation of informational externality.
Specifically, we test whether there exist external informational spillovers among
banks due to branch network proliferation. If it does, one might expect that the
size of the branch-network (as a strategic variable) will have a market size effect
(as in Cerasi (1995)), whereas if it does not, it will only have a market share
(distributional) effect (as in Salop (1979)). If in fact, market size effect is not
present, cooperative outcomes are very unlikely (Slade, 1995).
The second issue we focus on is the impact of the general sectoral conditions on
firms’ conduct. As has been documented elsewhere, price-cost margins may be
sensitive to changes in sectoral fundamentals. In fact, as is documented in
Domowitz et al. (1986), cross sectional estimates of concentration-margins
relationship are likely to be biased and misleading. Thus, since strategic behavior
is the focus of the present research, an additional important question is whether
banks’ strategic behavior is affected by changes in sectoral fundamentals. More
specifically, we study the behavior of banks before and after the occurrence of
system-wide banking problems.
The main objective of this paper is then to set up a non-price oligopolistic
model of bank behavior in the market for loans.2 This is done in order to
empirically analyze the role of the branch network in banks’ behavior, and to test
the oligopolistic conduct in this sector. The model posed allows banks to consider
its rivals’ future reaction to its own introduction of new branches, and the effects
of such on market shares and on total market size. We address the question of
whether banks are interdependent in their branching decisions and whether the size
of the branch network affects the total market size for loans. This is carried out by
deriving and estimating a simultaneous equation system of bank-level optimization
rule for branching choice and bank-level loan demand (share) equations. We thus
hope to uncover the behavioral regime underlying the observed data.
The model is empirically tested on a unique panel of Norwegian bank-level data
from 1988 to 1995. Norway provides a good ‘laboratory’ for this kind of research
because of the existence of data for the entire sector during the periods before and
after the Norwegian banking crisis. Therefore, we are able to test whether dramatic
events like a system-wide crisis are accompanied by a significant change in bank’s
conduct. In fact, we also investigate to what extent borrower behavior was affected
by the appearance of such a crisis. Furthermore, from 1984, regulations restricting
2
¨
In a recent paper, Neven and Roller
(1999) develop a structural model to analyze margins in the
provision of loans for several European countries. In that article the authors assume that demands for
loans and deposits are unrelated thereby neglecting ‘the possibility that banks might be linking the
provision of loans to the holding of deposits’. For the treatment and discussion related to the dichotomy
and separability of loans and deposits see Barros (1999) and Chiappori et al. (1995).
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1586
the establishment of bank branches were removed for all practical purposes. Thus,
during the period covered by our data, branch setting has been completely free and
endogenously determined by the banks.3
The paper is organized as follows: Section 2 presents the theoretical model, in
Section 3 the empirical model is set up. Industry background and data are
presented in Section 4, estimation techniques and results are reported in Section 5,
while Section 6 concludes.
2. The model
The quantity of loan demand (l it ) faced by a bank in each period is a function of
the size of its own branch network (b it ), the size of its rivals’ branch network
(b iRt 5 o j ±i b jt ), the bank specific interest rate on loans (r it ), rival banks’ interest
rate (r iRt ), a vector of exogenous macro variables (z t ) and z i , a variable capturing
bank-specific factors not explicitly entering the model:
l it 5 l(b it , b iRt , r it , r iRt , z i , z t ),
(1)
(1) has the following expected signs for the partial derivatives:
≠l it
≠l it
≠l it
≠l it
≠l it
≠l it
]
. 0; ]] , 0; ] , 0; ]] . 0; ] b 0; ] b 0.
≠b it
≠b iRt
≠r it
≠r iRt
≠z t
≠z i
Furthermore, bank i has a cost function c it with loans defined as the bank’s
output: 4
c it 5 c(l it , w it )
(2)
5
and where w it is a vector of input prices.
Profits for period t are defined as:
pit 5 (r it 2 pit ) l( ? it ) 2 c( ? it )
(3)
where ( ? it ) represents the arguments presented above, and pit is the bank-specific
3
Raj et al. (1979) provide empirical evidence that in Canada for example, it may take as long as 10
years to converge to a new long-run equilibrium following legislative or regulatory changes. This may
raise the distinct possibility that the branching structure in the sample may be in short-run, but not
long-run equilibrium. In Norway however, branches are mostly leased and thus easy to adjust in a
rather short period.
4
The cost function covers the costs that are involved in setting up branches. These costs were
completely variable for the period of our analysis, including those of physical premises which are
usually leased.
5
In fact the output vector of banks could also contain an element representing the handling of
problem loans. However, this is not included in our model due to lack of sufficient data on the stock of
problem loans prior to 1992.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1587
loan loss provisions made during period t per unit of loans outstanding. That is,
profits are revenues net of loan losses and production costs.6
Bank i maximizes the discounted flow of profits:
Ob p
`
Vi 0 5
t
(4)
it
t 50
s.t. (1). bt is the discount factor. We assume that the bank uses a feedback
(Markov) strategy, that is, at period t it sets its control variable b it based on the
information available at that time, in this case the number of branches of rival
banks. A priori we will assume that in setting b it bank i is sophisticated, that is, it
takes into account the reactions from rival banks in their future setting of their
branch numbers. Banks are assumed to expect their rivals to react with a lag of one
period.7
Inserting (3), (1) and (2) into (4) and differentiating w.r.t. b it gives the
following first order condition for the number of branches:
S
D
S
D
≠Vi 0
≠c it
≠l it
≠c it 11
≠l it 11 ≠b iRt 11
]]
5 r it 2 ] 2 pit ] 1 bt r it 11 2 ]] 2 pit 11 ]] ]]
≠b it
≠l it
≠b it
≠l it 11
≠b iRt 11 ≠b it
1 Bi 5 0.
(5)
The first term in (5) captures the period t effect on a bank’s profits resulting from
a change in its number of branches. The expression in parentheses is the difference
between loan rate and marginal lending cost and loan-loss provisions per unit of
loans outstanding. ≠l it / ≠b it is the effect on quantity of loans demanded from bank i
due to the change in its number of branches. The second term captures effects
from the rival banks’ conjectured next period reaction, ≠b iRt 11 / ≠b it . The delay in
rival banks’ expected reaction stems from their need to learn about the effects of
bank i’s actions in the previous period. Note that (5) does not include terms
representing the direct effects on bank i’s profits from its own number of branches
in period t 1 1 and onwards. The reason is that in period t 1 1 the bank decides on
6
In cases where a bank wants to expand into new geographical areas, it can do so by setting up new
branches. In that way the bank may be able to better evaluate and monitor the quality of its borrowers
in that area. However, during the first period after having set up a branch in a new area, the bank may
not yet have learnt about the quality of the local borrowers, and thus may attract also low quality
borrowers. This could be referred to as the ‘lemon’ problem of a new branch. A discussion of the
information-theoretic basis for the ‘lemon’ problem can be found in Shaffer (1998) who also
documents empirical support for it in the context of bank offices in new locations. In a preliminary
version of the model we thus allowed pit in (3) to depend on the recent increase in the number of
branches, i.e. pit 5 f(Db it 21 uDb it 21 . 0). This relationship, however, turned out not to be statistically
significant.
7
What we actually compute is not a standard feedback equilibrium, which involves repercussions in
all future periods. Instead, we have used an approximation to feedback equilibrium which ignores
repercussions that occur two or more periods ahead. However, in the ensuing empirical model we do
allow for any future indirect effects arising from bank i’s period t decision on b it to be captured.
1588
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
the number of branches it wants in that period irrespective of what number of
branches it had in period t. The final term Bi is a fixed effect, varying across banks
but not through time. It controls for any future indirect effects beyond period t 1 1
arising from bank i’s period t decision on b it .8
3. An empirical model of branch competition
The specification of the empirical model used to examine oligopolistic rivalry
focuses first on banks’ product demand, followed by banks’ production costs, and
by the first order condition for the choice of number of branches.
3.1. Demand for loans
To be able to distinguish between aggregate market-size effects and market
share (distributional) effects it is useful to rewrite (1) as the product of aggregate
loan demand Lt ( ? ) (market size) and market share s it ( ? ):
r it
l it 5 L(b t , r t , z t ) ? s b it , b iRt , ], z i
(6)
rt
S
D
where b t 5 o i b it is the aggregate number of branches and r t is the market average
interest rate on loans.9 The partial derivatives of s( ? ) have the following expected
signs:
≠s it
≠s it
≠s it
]
. 0; ]] , 0; ]]] , 0.
≠b it
≠b iRt
≠(r it /r t )
Note that the number of branches variable appears in the aggregate loan demand
function L(b t , r t , z t ). This allows the total market demand to change with the total
number of branches (market size effect). In general, there may be two competing
hypotheses regarding the inclusion or non-inclusion of the total number of
branches in the aggregate loan demand. One hypothesis is that the effect which
branches may have on total market demand for loans would come about from
some externality (like in the case of advertising). However branches in an
economy with a well-developed banking sector, as demonstrated by Jayaratne and
Strahan (1996), may just increase the quality of loans through better monitoring
and hence reduced adverse selection and moral hazard problems. The information
8
In order to facilitate the model and to focus on the non-price issue, we treat the choice of optimal
branch network for any given set of interest rates and loss provisions per loans outstanding. Since our
empirical first order condition focuses on the optimal branch network, it does not require to derive first
order conditions for interest rates, which are treated as endogenous in the empirical model.
9
The specification of loan interest rates in (6) ensures homogeneity of degrees zero for the market
share equations in loan interest rates.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1589
necessary for better monitoring (which additional branches may confer) is private
for a single bank and hence, no externality is conferred on the sector as a whole.
Consequently one hypothesis is that aggregate loans do not increase (and may even
decrease) as a result of branch network proliferation.10
On the other hand, since borrower information is private to the bank, borrowers
may worry about being ‘informationally locked-in’ (see Sharpe (1990) and
Klemperer (1995)). The stronger lock-in effect they expect to occur in the future,
the lower quantity of loans they would demand in the present period. A denser
branch network reduces the scope for lock-in as there are more banks with a
branch next to the borrower. These banks may have local information about the
borrower which they might not have had without their local branches. Hence,
competing banks will have more borrower-specific information, thus reducing the
costs to a borrower of being informationally locked-in. Less fear of lock-in by the
borrowers will increase the over-all loan demand in the present period, as the loan
contracts in the future are expected to be more favorable. Thus, an alternative
hypothesis is that a denser branch network increases aggregate loan demand. These
two competing hypotheses are tested in the ensuing empirical application.
We proceed by specifying the following market share (s it ) function with
constant elasticities.11
S D
r it
ln s it 5 fb ln b it 1 fbR ln b iRt 1 fr ln ] 1 z i
rt
(7)
z i is a fixed effect accounting for bank specific heterogeneity. From this
specification and (6) the derivatives of the loan demand for bank i to be used in
the first order condition are:
≠l it
≠Lt
≠s it
Lt
s it
]
5 ] s 1 ] L 5 ub ] s it 1 fb ] Lt
≠b it ≠b it it ≠b it t
bt
b it
≠l it
≠Lt
≠s it
Lt
s it
]]
5 ]] s it 1 ]] Lt 5 ub ] s it 1 fbR ] Lt
≠b iRt ≠b iRt
≠b iRt
bt
b iRt
(8)
where ub is the elasticity of total market demand (size) w.r.t. the total number of
branches. It is clear from (8) that in order to estimate the market share effects we
do not need a complete specification of the form of the market demand functions.
10
Note the possibility that aggregate loans may even decrease as a result of branch network
proliferation. This may occur if the improved screening facilitated by consolidation leads to the
rejection of a higher proportion of applicants from high-risk borrowers. Consistent with this possibility,
and as is documented in Section 5, the estimated elasticity of total market demand (size) w.r.t. the total
number of branches is negative. It is however, not statistically significant.
11
b it and b iRt are both measured at the beginning of period t, whereas l it is measured at the end of
period t. Thus, the model accommodates for the lags of lending and branching, in the sense that there is
a lag of 1 year between branching and lending.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1590
3.2. Production costs
Banks’ choice of branching is affected by their marginal costs of lending,
≠c it / ≠l it (cf. (5)). We estimate these marginal costs using the following translog
cost function (2):
O g ln w 1 g ln l
1
1 ] SO O g ln w ln w
2
1O g ln w ln l
ln c it 5 g0 1
h
hit
l
it
h
h
hl
m
hm
hit
hit
mit
D
1 gll (ln l it )2
(9)
it
h
A system of factor demand (share) equations is derived according to Shephard’s
lemma as:
O
≠ ln c it
]]]
; m hit 5 gh 1 ghm ln w mit 1 ghl ln l it
≠ ln w hit
m
(10)
where m hit is the cost share of factor h for bank i in period t.
To ensure homogeneity of degree one in factor prices and symmetry, the
following restrictions are imposed:
ghm 5 gmh
O g 51
O g 50
O g 50
;m,h
h
h
m
;h
hm
hl
h
Estimates of marginal lending costs ≠c it / ≠l it are constructed according to
c it
mcl it 5 ] gl 1 gll ln l it 1
l it
S
Og
hl
h
D
ln w hit .
(11)
The cost function (9) along with the h 2 1 share equations (10) after the
imposition of the above restrictions, comprise the cost system to be estimated.
3.3. Empirical first order condition for branches
The empirical specification of the first order condition for branches involves the
rearrangement of (5) and substitution of the parameters from the empirical
specification of the loan demand (8) as follows:
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1591
S D
Lt
Lt
ub
2 rmc it ] s it 5 rmc it ]] s it ]
b it
f
b
i b it
O
S D
S D
Lt 11
ub a
1 bt rmc it11 ]]] s it 11 ]
fb
i b it 11
O
Lt11
fbR a
Bi
1 bt rmc it11 ]] s it 11 ]] 1 ]
b iRt 11
fb
fb
(12)
where rmc it ; (r it 2 mcl it 2 pit ) representing the interest rate cost margin adjusted
for loan losses in period t and thus, the left-hand side of (12) represents the market
share-weighted average interest rate cost margin. The a represents the expected
rival’s reaction ≠b iRt11 / ≠b it which we refer to as a conduct parameter.12 The
one-period discount factor is bt . The Bi term is a fixed effect, varying across banks
but not through time. It controls for any future indirect effects beyond period t 1 1
arising from bank i’s period t decision on b it .
In markets where the strategic variable has an external effect, one tends to find
an accommodative or cooperative behavior, i.e. a negative a, since like in the case
of a pure public good it does not matter which of the firms invests in the strategic
variable. For instance, Roberts and Samuelson (1988) find (in the context of
advertising) negative values for the conjectures and positive effect on market size.
This implies an accommodative behavior and is consistent with a notion of
external spillovers inherent in advertising. In order to test whether branches confer
informational externality and affect market size for loans, we test for the statistical
significance of ub . If ub is not statistically different from zero, i.e. branches confer
only distributional rather than market size effects, then (12) will have the simpler
form of:
S D
Lt
Lt 11
fbR a
Bi
2 rmc it ] s it 5 bt rmc it 11 ]] s it 11 ]] 1 ].
b it
b iRt11
fb
fb
(13)
To summarize, the empirical model to be estimated consists of the behavioral
Eq. (12), the cost function (9) along with h 2 1 cost share equations (10), and the
loan market share function (7). The difference between this system and the system
without market size effect (13) is thus a restriction of ub 5 0 in (12).
4. Industry background and data
The model is applied to a panel of data for Norwegian banks from 1988 to 1995.
The total number of observations in the entire panel is 1218. This data set is
12
With N firms in the industry the conduct parameter can lie anywhere between 21 and N 2 1. In
our case N varies between 142 and 177 (see Cabral (1995)).
1592
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
unique in that it covers the prelude, the climax and the recovery period from a
system-wide banking crisis, peaking in 1991 and 1992. Thus, it allows the
detection of different strategic behavior in these different economic environments.
Until 1984 the banking industry in Norway was heavily regulated, both in terms
of interest rates charged and volume of loans. Effectively it was also protected by
the authorities. The period between the deregulation and the crisis was to a large
extent characterized by very aggressive and, as later became evident, careless
lending. Banks effectively did not take into account the ability of borrowers to pay
back their loans, which was severely hampered by the deepest recessionary macro
environment in the Norwegian economy since the second world war.13 In one
sense this era could be associated with moral hazard problems among banks and
borrowers. On the other hand neither banks nor borrowers had much experience
with operating in a deregulated credit market, and hence the resulting huge loan
losses could equally well be seen as emanating from this inexperience. The
emergence of a system-wide crisis became evident in early 1991, and peaked until
the end of 1992.14 After the crisis the industry seems to have operated with greater
management discipline.
During the period covered, banks first expanded their loans, then during the
crisis several banks experienced a period of reduced lending. During the recovery
period most banks started to increase their lending again. Aggregate loans ranged
from NOK 423 billions in 1988 to 513 billions in 1995, which amounts to an
overall increase of 21%. However the total number of bank branches has steadily
declined through most of this period. Specifically, the total number of branches has
decreased from 2166 in 1988 to 1740 in 1995. An increase in the aggregate
number of branches occurs only in the last 2 years. It should be noted, however,
that many individual banks did change the size of their branch network even
though the number of branches per bank has remained fairly constant during the
sample period. Thus, the emerging picture is quite of a dynamic nature in the
context of branching.
The structure of the banking industry is characterized by a relatively large
number of small banks and a few fairly large banks, both in terms of the size of
the loan portfolio and the number of branches. It is important to stress the fact that
on both the loans and the branch dimensions the cross-sectional standard
deviations far exceed the mean, pointing to the large variation in these variables.
Summary statistics are provided in Table 1.
In estimating the cost function, loans extended are treated as bank output.15 Four
13
The number of bankruptcies rose by 87% from 1987 to 1988, and peaked in 1992 at a level 175%
above that in 1987. From 1992 to 1994 it dropped by 37%.
14
A detailed description of the Norwegian banking crisis can be found in Norwegian Official Reports
(1992) and in Hope (1993).
15
The output aggregation problem in the context of banking is dealt with in Kim (1986).
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1593
Table 1
Summary statistics, loans and number of branches a
Year
Banks
Loans:
Sum
Mean
Median
S.D.
Branches:
Sum
Mean
Median
S.D.
1988
1989
1990
1991
1992
1993
1994
1995
177
168
155
147
145
143
142
141
422,599
2388
322
8157
455,515
2711
349
9465
472,750
3050
359
13,438
444,465
3024
383
12,495
443,155
3056
406
12,648
452,792
3166
434
11,922
471,954
3324
470
11,937
512,615
3636
525
12,897
2166
12
3
26.7
2032
12
3
25.8
1885
12
3
34.2
1796
11
3
29.5
1661
11
3
28.3
1604
11
3
27.2
1679
12
3
29.7
1740
12
3
32.3
a
Loans measured in NOK mill. current prices by the end of the year. Branch numbers are measured
at start of the year.
variable factors of production are specified: labor, materials, physical capital
(machinery and buildings) and funding. The latter includes both deposits and
borrowed money, i.e. all bank debt except subordinated debt.16
Bank- and year-specific prices of labor are computed as total labor costs per
man-hour. The price of materials is measured by a price index for material inputs
to financial institutions, collected from the national accounts statistics. Thus it
varies through time but is constant across banks. That is also the case for the price
of physical capital which is proxied by the 10-year interest rate on government
bonds.
The price of funding is a weighted sum of the interest rate on deposits and the
interest rate on borrowed money. The latter is measured by the 3-month money
market interest rate,17 thus it only varies across time, not across banks. The interest
rate on deposits, however, varies in both dimensions. It is calculated as banks’
interest expenditures on deposits divided by the mean of deposits at the beginning
and the end of the year.18 Bank specific interest rates on loans and the ratios of
loan loss provisions are measured likewise.
The discount factor bt applied for estimating the parameters of the first order
16
Subordinated debt is considered quasi equity, partially being counted as part of the capital base
when measuring the capital adequacy according to the BIS rules. Only a few banks in our sample have
issued subordinated debt.
17
The 3-month money market interest rate is represented by the 3-month effective NOK eurorate.
18
Whenever lagged values of physical capital, loans, deposits, debt or number of branches are used,
the lagged values are adjusted for bank mergers, i.e. the bank structure in period t is imposed on the
variables lagged to year t 2 1.
1594
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
condition (12) is calculated using the 9 to 12 month interest rate on t-bills, which
amounts to a sample average of 9.84%.19
5. Estimation and results
The model presented in Section 3 is estimated in two stages.20 In the first stage
the translog cost function (9) is estimated jointly with the three cost share
equations (10). The resulting estimated parameters are then used to form the
marginal cost of lending for each observation. In the second stage these estimates
of marginal costs are used as data input when the first order condition for the
number of branches (12) is estimated jointly with the market share Eq. (7).
The cost function and the cost share equations are estimated as a four equations
system. Because of the adding up constraints of the cost shares we drop the share
equation for costs of materials. The cost function is estimated as a second-order
Taylor series approximation around unity, i.e. all the right hand side variables in
the system are measured by the log of their relative deviations from their
respective means.21 Since the bank output, loans extended, is an endogenous
variable in the system we instrument it, using 3SLS. The instruments used are its
own 1 year lag and the lagged number of branches. As required for instrumental
variables they are highly correlated with the variable instrumented for; the
correlation coefficients for these two instruments being 0.98 and 0.70, respectively.
Due to the cross equation restrictions an iterative procedure is used, in this case the
Newton–Raphson algorithm in GAUSSX. Parameter standard error estimates are
White (1980a) heteroskedasticity adjusted. The estimated parameters are reported
in Table 2.
From these parameter estimates the marginal costs of lending are constructed.
The mean of the marginal cost across all 1218 observations is 0.099 with a
standard deviation of 0.043. In the period covered by this estimation the mean of
the interest rate banks charged borrowers corresponded to 0.126. The annual loss
provisions made corresponded to 0.013. Thus, the net interest rate was 0.113.
The estimate of the parameter gl is significantly larger than unity, which points
to increasing marginal costs in the provision of banking loans. This is a reasonable
result in light of the theory of delegated monitoring within an organization. Cerasi
19
This is a common practice in empirical dynamic oligopoly models (Roberts and Samuelson, 1988;
Karp and Perloff, 1989).
20
The model proved too complex to be estimated in one stage. However, as is shown by Pagan
(1984) theorem 3 (iii), our second stage estimators are consistent and their t-statistics are valid.
21
The estimated parameters of the translog cost function are used to predict marginal costs that enter
(the second stage of) the estimated first-order-condition equations. In the context of the translog
estimation, White (1980b) shows that as predictors the approximations have desirable properties, even
though the general validity of the translog to provide reliable information about local properties of
unknown functions may be doubtful.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1595
Table 2
Parameter estimates of the cost function and cost share equations a
Parameter
Estimate
S.E.
g0 (intercept)
gl (output)
gw (labor)
gk (capital)
gf (funding)
gll
gww
gwk
gwf
gkk
gkf
gff
gwl
gkl
gf l
12.407*
1.053*
0.145*
0.018*
0.714*
0.154*
0.043*
20.008*
20.079*
0.020*
20.013*
0.160*
0.88 c
20.45 b
0.66 b
0.090
0.021
0.001
0.001
0.002
0.031
0.004
0.002
0.003
0.003
0.002
0.004
0.47 b
0.24 b
0.83 b
Number of observations: 1218
a
White heteroskedasticity adjusted standard errors. Coefficients marked with a * are significant at
5%. Superscript b indicates the coefficient is multiplied by E 203 , and superscript c indicates the
coefficient is multiplied by E 204 .
and Daltung (1994) show that the costs of delegation increases with size of a bank,
due to an increase in the required number of monitoring levels.22
We now use the constructed marginal costs from the first stage estimation, and
estimate the first order condition (12) jointly with the market share Eq. (7). Since
(12) contains a term with a lead of one in the variables, last year’s observations
are excluded from the sample, hence t51988, . . . , 1994, leaving 1036 observations. In estimating the first order condition for branches and the market share
equation we allow for different intercepts for all banks in both equations. In that
way we control for time invariant heterogeneity among banks, for instance
differences in expected future effects beyond t 1 1 of decisions on branches in
year t. Thus, in estimating (7) and (12) we use only the within-bank variation in
the data. Since both b it and r it /r t are endogenous and appear as explanatory
variables in the market share equation (7), they are instrumented.23 Due to the
non-linearity of the parameters in (12) an iterative procedure is applied, using the
Newton–Raphson algorithm in GAUSSX. Parameter standard error estimates are
White heteroskedasticity adjusted.
The estimate of the elasticity of total market demand (size) w.r.t. the total
22
It is noted that non-constant returns to scale is also a sufficient condition for multimarket contact
effects (cf. Bernheim and Whinston (1990)), which enhances collusive behaviour.
23
As instruments we use the lagged number of branches, the price of funding, the rival number of
branches and the variable appearing in the third term of the r.h.s. of (12).
1596
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
number of branches ub is 218.47 with standard error of 12.83. Thus, ub is not
significantly different from zero at any acceptable level of significance, indicating
that the number of branches has no effect on the total market size (demand). This
result is consistent with the hypothesis that there may be no external effect of
branching on market size. Our finding is in the spirit of the Salop (1979) type of a
model in which an increase in the number of products (here branches) only crowds
the market further. This is to be contrasted with a Cerasi (1995) type of model
which shows that market size may increase with the total number of branches.24
Based on our empirical result regarding the market size effect, we proceed with
the model (13) and (7) that excludes ub . Results are reported in Table 3.
As indicated in the table the market share is increasing in own number of
branches as can be expected, specifically a 1% increase in own number of
branches brings about 0.5% increase in market share. This is shown in the
coefficient fb of 0.520 which is highly significant. Regarding the branch network
of all other banks, an increase in their branch numbers by 1% (corresponding to an
increase of 18 branches) would decrease the market share of the average bank by
|1%. This is indicated by the fbR parameter estimate, which is not significantly
different from unity. Notice that because of symmetry requirements a significantly
positive fb implies a significantly negative fbR . We obtained this result without
prior restrictions. The fr parameter represents the elasticity of the market share
w.r.t. the relative interest rate on loans. A 1% increase in the relative rate of
interest decreases market share by |1.2% ceteris paribus. Thus, market shares are
quite sensitive to the relative interest rates.
As to conduct, an average bank expects all other banks to respond by increasing
their total branch network by approximately two branches, if it establishes one
more branch itself. Although a is significantly greater than zero it is not
significantly larger than unity.
The positive significant estimate of a indicates that banks expect retaliation in
their non-price competition with the number of branches. Previous studies of bank
Table 3
Parameter estimates of the first order condition and market share equation a
Parameter
Estimate
S.E.
fb (elasticity of own branches)
fbR (elasticity of rival branches)
fr (relative interest rate elasticity)
a (conduct)
0.520*
20.968*
21.163*
2.082*
0.125
0.077
0.201
1.041
Number of observations: 1036
a
White heteroskedasticity adjusted standard errors. Coefficients marked with * are significant at 5%.
24
Although Cerasi’s model concerns the deposit market a similar model could also be applied to the
credit market.
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1597
behavior in the Norwegian loan market indicate oligopolistic quantity behavior by
banks taking rivals’ expected retaliations into consideration. See, for instance,
Berg and Kim (1994, 1998). In the present study we find clear evidence that these
strategies also carry over to non-price behavior like that of establishing or closing
branches. In deciding on their branch numbers banks do not behave in a naive
fashion abstracting from rival banks’ reaction. Instead they seem to recognize their
rivals future response to their own branch decisions. Our results confirm the
contention of Evanoff (1988), that banks use branches as a strategic variable. The
economic rationale behind such retaliatory behavior can be justified by the fact
that banks would lose market share if they did not retaliate to the set up of a new
branch by a rival (cf. the significantly negative sign of fbR ). Even if retaliation
would reduce the net revenue in period t 1 1, it may still be rational to retaliate to
the set up of a new branch in order to establish a reputation effect similar to the
point discussed by Kreps and Wilson (1982).25
The entries in Table 3 indicate that an average bank (with 12 branches) that
establishes one additional branch, would expect a direct increase of its market
share by 4.3%. It would expect its rival banks to respond by increasing their total
number of branches by 2.08 which corresponds to a 0.12% increase. The expected
response from the rivals would then only decrease the bank’s market share by
0.12%. Hence the expected net effect on an average bank’s market share if it
establishes one more branch is an increase in its market share of 4.2%.
The emergence of a system-wide banking crisis in Norway during 1991 is
considered a watershed phenomenon (cf. Section 4). This crisis was characterized
by several major and small banks losing their entire equity capital due to bad
loans. One might expect that after such a crisis bank and borrower behavior would
be entirely different, and in fact more sophisticated. Thus, we estimated our model
on the two separate periods 1988 to 1990 and 1991 to 1994, which correspond to
the periods before and after the appearance of the peak of the crisis, respectively.
As seen from Table 4, all the elasticity coefficients in the post-crisis period are
about doubled compared to the pre-crisis period. The pre-crisis period (1988–
1990) conduct coefficient a is negative, indicating a forbearing behavior.
However, it is not significantly different from zero at any level of significance.
This implies that banks behaved as if their rivals would not respond by changing
their future branch network size, a Cournot like behavior. For the period following
immediately after the crisis (1991–1994) the conduct parameter is significantly
larger than zero indicating an expected retaliatory behavior. Thus, this parameter is
qualitatively similar to the one estimated for the entire period, i.e. banks behaved
in a way that considered rivals’ future reaction. This different behavior during the
pre- and post-crisis periods supports the contention that the crisis made banks
25
One could also appeal to the intuition present in the ‘switching cost’ literature (see the survey of
Klemperer (1995)) where firms are aggressively competing for market share in earlier periods, that will
be valuable for them in periods after switching costs have been created (lock-in effect).
1598
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
Table 4
Parameter estimates for the pre- and post-crisis periods a
Parameter
Pre-crisis
Post-crisis
fb (elasticity of own branches)
0.750*
(0.239)
20.426*
(0.078)
20.576
(0.372)
21.390
(1.560)
1.151*
(0.468)
20.991*
(0.115)
21.166*
(0.139)
18.612*
(7.833)
468
568
fbR (elasticity of rival branches)
fR (relative interest rate elasticity)
a (conduct)
Number of observations
a
Numbers in parentheses are White heteroskedasticity adjusted standard errors. Coefficients marked
with * are significant at 5%. Pre-crisis parameters refer to the 1988–1990 years; post-crisis refers to the
1991–1994 years.
more aware of the interactive nature of their environment, at least as far as
non-price strategies are concerned.
In addition, borrowers became much more sensitive to interest rates. This is
seen by a much more interest rate elastic loan demand facing the individual banks
(fr 5 2 1.166 for the post crisis period, whereas it is 20.576 and insignificant in
the pre-crisis period). There may be several explanations for this difference. It may
be explained by a more cautious behavior by borrowers, i.e. borrowers have
become more aware of the burden of interest payments, and adjusted their
borrowing behavior accordingly. However, this result may also stem from moral
hazard among borrowers with limited liability; after the crisis when the conditions
of the macro economy improved, and the overall number of bankruptcies fell,
these borrowers may have estimated the probability of paying back the loans as
higher, and hence became more interest rate sensitive. Finally, a 1992 tax reform
reducing significantly the marginal tax rate on interest income and expenditure
could have made borrowers more interest rate sensitive.
A similar picture emerges when fb (the elasticity of market share w.r.t. own
branches) and fbR (the elasticity market share w.r.t. rival branches) are estimated
from the pre- and post-crisis periods. For the pre-crisis period fb is 0.750 and
significantly different from zero, and in the post crisis period it is 1.151 and
significantly different from zero. fbR is 20.426, significantly different from zero
and significantly less than unity, indicating a negative effect from rival branches.
For the post-crisis period, this parameter estimate is 20.991, which is significantly
different from zero but not significantly different from unity. In other words, loan
demand facing a single bank seems to have become more sensitive to the size of
banks’ branch network after the emergence of the banking crisis. This change of
behavior may be one of the effects stemming from enhanced managerial discipline
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1599
after a crisis.26 As was recorded by Norwegian Official Reports (1992) before the
crisis, bank branch managers were very aggressive in expanding their loans far
outside the area where they possessed local borrower-specific information. After
the crisis when stricter managerial discipline was imposed on the banks,
management restricted branches, in general, to operate only in their local area.
Then the proximity to a bank branch may have become more important in order to
obtain a loan, hence the effective loan demand facing a bank would have been
more dependent on its number of branches.
In the post-crisis period an average bank establishing a new branch would have
a direct increase in its market share by 10.46%, which it would expect to be
reduced by 1.1% due to the response from its rival banks. Thus, the expected net
increase in market share from putting up one more branch in the post-crisis era
would be 9.4% for an average bank. Compared to the estimates based on the entire
sample (discussed above) an increase of a bank’s branching network in the
post-crisis era would cause a larger expected net increase in the market share in
spite of the stronger expected retaliation. This result is due to the higher estimated
own-branch elasticity (fb 5 0.520 for the entire sample and 1.151 for the postcrisis era). For the pre-crisis sample an average bank setting up one additional
branch would experience a 6.25% increase in its market share. However, in the
pre-crisis environment banks would not expect any reduction in their market share
stemming from retaliation, according to the conduct parameter, for that period, a
(5 21.39) which is not significantly different from zero.
The above results and discussions give support to the contention raised by
practitioners in the banking industry of fiercer rivalry, a more sophisticated
behavior and stronger managerial discipline in the post crisis environment.
6. Conclusions
The main objective of this paper has been to set up a non-price behavioral
oligopolistic model of bank behavior. This has been done in order to empirically
analyze and assess the role of the branch network as a non-price strategic variable
and to try to uncover some empirical regularities pertaining to the use of non-price
strategies in general.
The model used allows a bank to have conjectures about its rivals’ future
reaction to its own introduction of new branches or closing of existing branches,
and the effect of such on market shares and size of loans. We find banks to be
intertemporally interdependent in their branching decisions. Additional insight is
gained by looking at the pre- and post-banking crisis periods. During the precrisis
26
See Dewatripont and Tirole (1994) for an exhaustive analysis of managerial discipline and bank
failures.
1600
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
period banks behaved in a somewhat naive (Cournot like) fashion, in that, they did
not consider future reactions from rivals, whereas during the post-crisis period
banks’ behavior was much of a retaliatory nature. In fact, borrowers are also found
to be more sensitive to both interest rates and the size of the branch network after
the crisis. This could be explained by more cautious behavior by borrowers and
stronger managerial discipline in the banks’ organization.
We find that branching has a significant effect on banks’ market shares but not
on total market size (demand). This finding is consistent with the contention that
bank-specific branch network does not confer informational externality on other
banks since the informational gain stemming from its branch network is not shared
by others in the industry. Such a finding corresponds well with the retaliatory
behavior of banks in their branching decisions. This result should be contrasted
with those obtained in other studies of non-price competition, where the strategic
instrument (e.g. advertising) does confer an externality, and the behavior has been
found to be accommodative. In such cases there is a pronounced effect on market
size.
Also, based on our finding, the branch network is used as a strategic non-price
variable in banking conduct. Thus, traditional empirical studies which use the
number of branches as just another exogenous variable in their specified
technology (production, cost or profit functions) may well be misspecified.
To summarize, non-price strategic behavior is found here as elsewhere to be an
important attribute of firm conduct. Firms recognize their interdependence in the
non-price dimension of their operation. Moreover, an empirical regularity emerges
from this and other studies, in that retaliatory conduct in the non-price space is
consistent with no external effect on market demand, whereas accommodative
conduct is consistent with a positive external effect on market demand. These
differences relate to the degree of externality inherent in the non-price strategic
instrument. Lastly, it is demonstrated that the nature of conduct is importantly
affected by the industry’s crisis. It is interesting to investigate further the specific
characteristics inherent in crises which affect firms’ conduct in a formal model.
This, however, is left for future research.
Acknowledgements
The authors are grateful to Pedro Barros, Benjamin Bental, Sigbjørn Berg, Erik
Biørn, Mark Flannery, Chaim Fershtman, Steinar Holden, David Humphrey, Eilev
Jansen, Eirik Kristiansen, Loretta Mester, Tore Nilssen, Margaret Slade, Menachem Spiegel, Yossi Spiegel, Manuel Trajtenberg and participants in seminars at
Hebrew University of Jerusalem, Norges Bank, Norwegian School of Management, Tel Aviv University, University of Oslo, University of Trondheim, the
French Finance Association in Grenoble, and the EARIE conference in Leuven for
helpful discussions and comments on previous versions of this paper. An
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
1601
anonymous referee has made valuable clarifying points. Views and conclusions
expressed are the responsibility of the authors alone and cannot be attributed to
any of the persons or institutions mentioned above.
References
Allen, F., 1985. Contracts to sell information. Wharton School, University of Pennsylvania, Unpublished manuscript.
Barros, P., 1995. Post-entry expansion in banking: the case of Portugal. International Journal of
Industrial Organization 13, 593–611.
Barros, P., 1999. Multimarket competition in banking with an example from the case of Portuguese
market. International Journal of Industrial Organization 17, 335–352.
Berg, S.A., Kim, M., 1994. Oligopolistic interdependence and the structure of production in banking:
an empirical evaluation. Journal of Money, Credit, and Banking 26, 309–322.
Berg, S.A., Kim, M., 1998. Banks as multioutput oligopolies: an empirical evaluation of the retail and
corporate banking markets. Journal of Money, Credit, and Banking 30, 135–153.
Bernheim, B.D., Whinston, M.D., 1990. Multimarket contact and collusive behaviour. RAND Journal of
Economics 21, 1–26.
Cabral, L., 1995. Conjectural variations as a reduced form. Economics Letters 49, 397–402.
Cabral, L., Majure, R., 1994. An empirical analysis of bank branching: Portugal 1989–1991. In: Neven,
D., Roller, L. (Eds.), The Empirical Analysis of Industrial Organization. CEPR, London.
Cerasi, V., 1995. A model of retail banking competition. London School of Economics, Mimeo.
Cerasi, V., Daltung, S., 1994. The optimal size of a bank: costs and benefits of diversification. London
School of Economics, FMG discussion paper.
Cerasi, V., Chizzolini, B., Ivaldi, M., 1997. The impact of deregulation on branching and entry costs in
the banking industry. Universita’ Bocconi, Milano, Working paper 117.
Chan, Y.S., Greenbaum, S.I., Thakor, A.V., 1986. Information reusability, competition and bank asset
quality. Journal of Banking and Finance 10, 243–253.
Chiappori, P.A., Perez-Castrillo, D., Verdier, D., 1995. Spatial competition in the banking system:
localization, cross subsidies and the regulation of deposit rates. European Economic Review 39,
889–918.
Degryse, H., 1996. On the interaction between vertical and horizontal product differentiation: an
application to banking. Journal of Industrial Economics 44, 169–186.
Dewatripont, M., Tirole, J., 1994. The Prudential Regulation of Banks. The Walras-Pareto Lectures.
MIT Press, Cambridge, MA.
Diamond, D., 1984. Financial intermediation and delegated monitoring. Review of Economic Studies
51, 393–414.
Domowitz, I., Hubbard, R.G., Petersen, B., 1986. Business cycles and the relationship between
concentration and price-cost margins. RAND Journal of Economics 17, 1–17.
Evanoff, D., 1988. Branch banking and service accessibility. Journal of Money, Credit, and Banking
20, 191–202.
Freixas, X., Rochet, J.C., 1997. Microeconomics of Banking. MIT Press, Cambridge, MA.
Hope, E., 1993. The Banking Crisis in Norway: Problems and Prospects. Foundation for Research in
Economics and Business Administration, Bergen.
Jayaratne, J., Strahan, P.E., 1996. The finance-growth nexus: evidence from bank branch deregulation.
Quarterly Journal of Economics 111, 639–670.
Karp, L., Perloff, J.M., 1989. Dynamic oligopoly in the rice export market. Review of Economics and
Statistics 71, 462–470.
1602
M. Kim, B. Vale / Int. J. Ind. Organ. 19 (2001) 1583 – 1602
Kim, M., 1986. Banking technology and the existence of a consistent output aggregate. Journal of
Monetary Economics 18, 181–195.
Klemperer, P., 1995. Competition when consumers have switching costs: an overview with applications
to industrial organization, macroeconomics, and international trade. Review of Economic Studies 62,
515–539.
Kreps, D., Wilson, R., 1982. Reputation and imperfect information. Journal of Economic Theory 27,
253–279.
Leland, H.E., Pyle, D.H., 1977. Informational asymmetries, financial structure, and financial intermediation. Journal of Finance 32, 371–387.
Matutes, C., Padilla, A., 1994. Shared ATM networks and banking competition. European Economic
Review 38, 1113–1138.
Mayer, C., Vives, X., 1993. Capital Markets and Financial Intermediation. Cambridge University Press,
Cambridge, UK.
Mester, L.J., 1992. Traditional and non-traditional banking: an information-theoretic approach. Journal
of Banking and Finance 16, 545–566.
Nathan, A., Neave, E.H., 1989. Competition and contestability in Canada’s financial system: empirical
results. Canadian Journal of Economics 22, 576–594.
¨
Neven, D., Roller,
L.H., 1999. An aggregate structural model of competition in the European banking
industry. International Journal of Industrial Organisation 17, 1059–1074.
Norwegian Official Reports, 1992. Report by the Commission on the Banking Crisis. Royal Ministry of
Finance and Customs, Oslo.
Pagan, A., 1984. Econometric issues in the analysis of regressions with generated regressors.
International Economic Review 25, 221–247.
Raj, B., Kirkham, P., Clarke, E., 1979. Dynamics of Canadian deposit-taking institutions: a Markovian
approach. The Journal of Industrial Economics XXVIII, 177–188.
Roberts, M.J., Samuelson, L., 1988. An empirical analysis of dynamic, nonprice competition in an
oligopolistic industry. RAND Journal of Economics 19, 200–220.
Salop, S., 1979. Monopolistic competition with outside goods. Bell Journal of Economics 10, 141–156.
Shaffer, S., 1993. A test of competition in Canadian banking. Journal of Money, Credit, and Banking
25, 49–61.
Shaffer, S., 1998. The winner’s curse in banking. Journal of Financial Intermediation 7, 359–392.
Sharpe, S.A., 1990. Asymmetric information, bank lending, and implicit contracts: a stylized model of
customer relationships. Journal of Finance 45, 1069–1087.
Slade, M.E., 1995. Product rivalry with multiple strategic weapons: an analysis of price and advertising
competition. Journal of Economics & Management Strategy 4, 445–476.
White, H., 1980a. A heteroskedasticity-consistent covariance matrix estimator and a direct test for
heteroskedasticity. Econometrica 48, 817–830.
White, H., 1980b. Using least squares to approximate unknown regression functions. International
Economic Review 21, 149–170.