Entry Decisions in Local Banking Markets: Does Potential
Entrants’ Heterogeneity Matter?*
Roberto Felici and Marcello Pagnini**
This draft: June 2004
Abstract
We examine the determinants of entry into Italian local banking markets during the period 1991-2002. We
build a simple model in which the probability of branching in a new market depends on the features of both
the local market and the potential entrant. Our econometric findings show that large markets attract entry. We
also find that the size of the potential entrants and their profitability increase entries. Moreover, banks are
more likely to expand into those markets that are closer to their pre-entry locations. These results are robust
to several controls, including the lack of independence of entry decisions.
Keywords: entry, barriers to entry, local banking markets, geographical distance.
JEL Classification Numbers: G21, L22.
* A special thank goes to Luigi Buzzacchi for his encouragement and useful comments on previous drafts of
this paper. We wish also to thank Paolo Angelini, Marcello Bofondi, Vittoria Cerasi, Fabio Quintiliani,
Federico Signorini and participants at a seminar held at the Bank of Italy (June 2003) and at the second
International Industrial Organization Conference (Chicago, April 2004) for useful comments and
suggestions. Usual disclaimers apply. The views expressed in this paper are those of the authors and do not
necessarily reflect those of the Bank of Italy.
**
Bank of Italy, Economic Research Unit, Bologna branch, Piazza Cavour 6 40124- Bologna, Italy. e-mail:
[email protected], [email protected], tel. ++39-051-64030171, fax ++39-051263925.
2
1. Introduction
At the beginning of the nineties the Italian banking sector underwent a series of
changes in the regulatory regime1. Many pre-existing constraints on bank branching across
local markets were lifted. These changes spurred a rapid increase in the number of
branches throughout the country. Banks expanded both by opening new outlets in the local
markets where they were already operating and by branching in new markets. This paper
examines entry decisions of a pool of about 300 Italian banks from 1991 to 2002.
Factors determining the accessibility of markets by a pool of potential entrants are
essential to the working of the competitive process. Moreover, entry barriers are
particularly relevant for the banking sector. First, informational problems affecting bankfirm relationships in lending markets may contribute to raise the intensity of entry barriers
in this sector. Second, users of retail banking services normally exhibit a low degree of
mobility across different local markets contributing to their geographical segmentation.
The entry literature generally assumes that access to markets only depends on the
structural characteristics of the different locations. This is equivalent to assuming that each
market is confronted with an infinite number of homogenous potential entrants. In view of
the many studies showing the sharp differences between banks in terms of performance
indicators like profitability, technical efficiency and growth opportunities, this assumption
seems to be unrealistic. It is unlikely that this high level of heterogeneity across banks may
not result into different abilities of expanding into new markets.
Moreover, as noted by Geroski (1995) entries tend to come into waves. This means
that one has to study entry processes within a dynamic set-up. But structural market
variables usually change at a slow pace and therefore may not represent a complete
explanation for the entry decisions through time. On the contrary, variables at individual
banks level, like profitability, locations and many others, may be subject to sudden
changes, therefore mimicking the irregular time patterns of entry processes. In other words,
it is difficult to undertake a study of entries, either cross section or through time,
completely ignoring the issue of economic agents’ heterogeneity.
1
The process of branching deregulation was started by the Bank of Italy in the late 80s. In particular, from
march 1990, entry through a new branch could not be denied on the ground of discretionary economic
reasons.
3
Consequently, drawing on previous contributions by Berry (1992) and Scott Morton
(1999), the present paper analyses entry choices in local banking markets taken by a pool
of heterogeneous banks.
Our contribution to this literature is threefold. First, in the theoretical section we
model entrants’ heterogeneity by assuming that entry sunk costs may vary with bank size,
profitability and with the physical distance between the bank and its target market. In this
context, we discuss conditions that may guarantee the independence of entry decisions, a
topic not covered in previous contributions. Second, thanks to the availability of a unique
data set on individual banks and local markets, we carry out an extensive econometric
analysis of entry processes. In our specification the effect of physical distance on entry is
measured as a continuous variable and not as a dichotomous variable as done in other
contributions. Furthermore, our data set enables us to introduce a wider set of robustness
checks than those carried out in other papers.
Our econometric results show that the bank heterogeneity does matter for their entry
choices. Larger banks have a higher probability to enter a local market. Furthermore, banks
earning higher profits are more likely to expand into a new market. But the most important
quantitative effect is related to distance. According to our econometric findings, a
reduction by one standard deviation, approximately corresponding to 260 kilometres, of
the distance between a bank’s location and that of its target market raises the probability of
entering a market by almost three times. All these results are robust to different definitions
of the set of the potential entrants as well as to other additional controls.
Thus, microeconomic conditions featuring individual banks may have a strong
impact on their abilities to overcome entry barriers. Consequently, any factor leading to a
change in the bank size distribution or in their profitability may modify the accessibility of
local markets. Moreover, our results show that entry decisions are conditioned by
proximity. Hence, whatever factor that improves the ability of a bank to extend its control
over activities in distant markets may have a positive impact on the competitive conditions
in those areas.
The rest of the paper is organised as follows. In section 2 we survey the empirical
literature on entry in banking. Section 3 lays out a simple model of entry choices. Section 4
describes the main features of the data set and of the entry definition. The specification
used in the econometric analysis and the definitions of the explanatory variables are
4
illustrated in section 5. Section 6 comments the main econometric results. A set of
robustness checks are introduced in section 7. Section 8 contains concluding remarks and
indications for future research.
2. A survey of empirical literature on entry in banking
Recent literature on banking has investigated conditions that may foster or hamper
entry in local markets. In a very influential paper Amel and Liang (1997) estimate a model
for the joint determination of banks’ profits at local level and market entry for the USA.
They find that entries increase with local profits, market size and market growth. They also
show that entry can limit supra-normal profits. Related papers are those of Seelig and
Critchfiled (1999) and Berger et al (2000) concentrating on entry through denovo banks
and its relation with operations of mergers & acquisitions (M&A).
More recently, Gobbi and Lotti (2003) study entry patterns across 95 Italian provinces2.
The authors, building on previous theoretical contributions by Dell’Ariccia et al (1999),
argue that the frequency of entry should be lower in provincial markets where lending
activity has more weight than other banking activities. At the same time, entry barriers
should be higher in markets where publicly available and contractible information about
borrowers is scarce. In a regression analysis on pooled data for provinces and different
time intervals they find a confirmation for these theoretical predictions.
All these papers do not deal with the issue of entrants heterogeneity and implicitly
assume that each local market is confronted with an infinite number of homogenous
potential entrants.
A series of recent contributions has however examined entry decisions by assuming
differences in the demand and costs conditions featuring potential entrants. Building on
previous papers by Berry (1992) and Scott Morton (1999), Juan (2002) estimates a model
of entry probability combining very detailed geographic and individual bank data for
Spain. She aims at testing Sutton’s ‘symmetry principle’ according to which entry does
not depend on the potential entrant characteristics but only on local market conditions
(symmetry is defined with respect to the identity of an entrant firm3). The author shows
2
3
For an analysis of entry in Italian banking markets before the nineties see Conigliani and Lanciotti (1978).
See Sutton (1998).
5
that the size of entrant bank has only a modest economic impact on entry as compared to
the effects of market conditions, thereby confirming Sutton’s symmetry principle. A
potential problem with this result however is that there are too few controls for individual
bank characteristics.
In another paper on Spanish banks, Fuentelesaz and Gomez (2001) combine in a very
fruitful way two distinct lines of research on entry. The first one
is the traditional
industrial economics approach according to which differences in entry patterns depend on
the structural characteristics of the markets. The second one looks at entry as a
diversification decision taken by an individual firm. In this way entry can be linked to a
series of variables that pin down organizational factors within a bank. In the empirical
section of their paper, they concentrate on a specific bank category (savings banks). They
find that entry in local markets is positively related to resource availability, bank
profitability and the size of the initial scope of operation. They also find that savings banks
enter into markets which are closer to their pre-entry locations. Moreover, they model
entry decisions in a unique ten years time span, assuming that these decisions may be
related to explanatory variables taken at the beginning of the period. This means that entry
in the final year may depend on factors registered ten years before.
Although not directly focussing on entry, a recent paper by Cerasi, Chizzolini and
Ivaldi (1998) investigates how banks’ heterogeneity may affect branching decisions. They
propose a joint estimation of the degree of competition and of branching costs in local
banking markets. Both are considered as functions of firm and market specific variables. In
particular, they find that the marginal costs of opening an additional branch decrease with
the size of the bank and increase with the distance between the location of the bank’s
headquarters and that of its new branch.
All these papers are based on the assumption that entry or branching decisions are
independent, i.e. each bank decides whether to open a new branch in a local market
regardless of its branching decisions in other local markets. The consequences of the
violation of this assumption for the results however are not discussed. Moreover, the
econometric analysis carried out in these contributions does not have controls for the
potential lack of independence of entry decisions of a bank across different markets. In
what follows we will introduce some of these controls.
6
Two recent streams of literature are related to the issue of entry in local banking
markets. A series of recent papers have examined the effects of geographical distance
between banks and their clients on the competitive process in lending and deposits
markets4. A result of this literature is that distance has a strong impact, although declining
due to technological advances, on bank-firms relationships, on pricing and more generally
on the competition in local markets.
The issue on distance is important for entry since it shows how geographical space may
contribute to the segmentation of banking markets. However, this literature considers
location choices taken by the banks as exogenous. On the opposite, these choices are
analysed by entry literature in a dynamic set-up where banks decide in which local markets
to expand. This opens a different perspective on distance. In particular, one may ask how
entry costs may vary with the distance between a bank and its target market. The question
then is how a bank can effectively export its previously accumulated know-how to distant
markets. From this point of view, our paper is closer to a recent contribution by Berger and
De Young (2001) in which they empirically assess how control of parent financial
organization may vary with distance to the affiliate.
Another emerging stream of literature deals with the issue of diversification versus
specialisation in the lending activity of the banks5. A tenet of this literature is that
advantages from diversification, emphasized by the traditional portfolio theory, have to be
traded off against its disadvantages or benefits from specialisation. In particular,
maintaining a diversified loan portfolio may entail lending to new sectors or to sectors in
which bank competition is high. A bank may lack the monitoring expertise for lending to
new sectors and this raises monitoring costs when learning is relevant. Entry in sectors
where other banks are already operating may be subject to adverse selection and ‘winner
curse’ effects (Dell’Ariccia et al, 1999). Last, diversification may increase agency costs
due to growth in size of the bank.
The decision to expand into a new market increases the degree of geographical
diversification of a bank. In this perspective, the list of disadvantages from diversification
described above can also be considered as a series of factors determining the magnitude of
4
See Petersen and Rajan (2000), Degryse and Ongena (2003) and Buch (2001) for the effects of distance on
international banking activity.
5
See Winton (1999), Acharya, Hasan and Saunders (2002) and Stomper (2003).
7
entry costs into a local market. In the next sections of the paper, we will take account of
these factors.
3. A framework for the analysis of entry decisions
In this section we describe the entry decisions of a set of potential entrants in local
banking markets. Our focus on a single sector rules out explanations of entry behaviour
based on sectoral differences in the degree of economies of scale, product substitutability
or expenditure in R&D activities.
We assume that the banking sector is segmented in many geographically defined
local markets. This assumption is line with the observed low spatial mobility of banks’
customers and with the consequent relatively limited geographical scope of the supply of
many banking services.
We identify entry with the decision of a bank to open a branch in a market at time t,
provided it owned no branches in the same market at time t-16. We follow Berry (1992)
and assume that entry decisions can be represented in a two stage game. In the first stage, a
potential entrant decides whether to enter a market7. If it does, it has then to incur fixed
sunk costs due to the opening of a branch in the new market. These costs may vary with
individual banks and markets characteristics or with a combination of the two. Differences
in entry costs are assumed to be the only source of entrants heterogeneity. The distribution
of these costs is assumed to be exogenous and known to each potential entrant.
Regarding entry sunk costs, an entry decision requires a series of activities aimed at
evaluating business opportunities arising in a market. These activities will be undertaken
before accessing the market and hence before knowing the profit opportunities that will
arise. Moreover, these entry costs may be difficult to recover should a bank decide to exit
6
In an overwhelmingly majority of cases, a bank willing to enter a market has to open an outlet in that
location. The need for a presence through branching is explained by the observed low spatial mobility of the
users of retail banking services. Thus we identify entry with the opening of a new branch in a market.
7
Our definition of entrants differs from the one adopted in Berry (1992). He includes among the entrants
those firms that were incumbents in a market at time t-1 and decide to remain in the same market at time t.
The differences between new entrants and incumbents strategic positions suggest to separate the analysis of
their choices. As observed by Toivanen and Waterson (2001), the inclusion of incumbents in the pool of
potential entrants would result at odds with the presence of entry sunk costs. Thus we restrict our analysis to
the strategies followed by a set of potential new entrants identified with the pool of banks which were not
previously operating in a specific market.
8
that market 8. Thus we can assume that banks sink these costs before knowing the returns
from their investment.
Recently, Melitz (2002) has modelled entrants heterogeneity in a different way.
Rather than assuming a predetermined ordering from most to least efficient entrants, he
starts from ex ante identical firms who are uncertain of their future productivity.
Heterogeneity is obtained ex post by drawing from a common and known distribution
function. The range of efficiency levels and average productivity are derived endogenously
within an equilibrium model. Although this way of modelling heterogeneity opens many
interesting perspectives, we maintain our assumption of a predetermined ordering of
entrants abilities. As explained in the empirical section, our entry game is restricted to
already established banks. Thus, it is unlikely that these banks as opposed to denovo banks
may be uncertain of the level of their own abilities in entering new markets.
In the second stage of the game, banks participating the market (both previous
incumbents and new entrants) decide prices and quantities and this determines post entry
profits and market equilibrium. We skip the analysis of this second stage and we focus on
the investigation of the moves of the players in the first stage.
Let Ki indicate the maximum number of potential entrants for market i (i=1,….,M).
The set of potential entrants in the first stage of the game can be ranked according to the
level of their entry costs, F, (starting from the lowest cost bank) in the following way:
Fi1 < Fi 2 < ..... < FiK i
If a bank b decides to enter, its profits will be:
Π ib = Π ( N i , X i ) − Fib
(1)
where Ni indicates the number of actual entrants (with Ni ≤ Ki), Xi is a vector of variable
varying with market characteristics, Π(Ni,Xi) are profits gross of entry costs and Fib are
fixed entry sunk costs.
8
This conclusion may extend to the case of an exit decision carried out by selling the branch to another bank.
Indeed, it is unlikely that the price paid for the goodwill may allow the seller to recover entry costs. Note also
that , due to informational problems which affect bank-firms relationships, entry costs in the banking market
can become very relevant (Dell’Ariccia et al, 1999). Moreover, a new entrant have to spend resources to
build relations with local lenders based on soft information.
9
Thus, profits obtained from the decision of bank b to enter market i, Πib can be split
into two parts: Π( Ni,Xi) are profits gross of entry costs varying only with a set of market
characteristics represented by vector Xi and with the number of entrants, given by Ni. The
second component consists of entry fixed costs Fib that, as mentioned before, may vary
with individual bank characteristics and their interaction with market features. These costs
may help explain why specific banks enter specific markets.
In this framework we can define a very simple entry decision rule. Entry in market i
will take place as long as entrants’ profits defined by (1) are non negative. Potential
entrants with negative expected profits stay out of the market and earn zero profits. Under
this rule and previous assumptions, it can be shown that a Nash equilibrium in pure
strategies exists such that, holding the strategies of the rivals constant, all entering banks
earn positive profits and all those staying out of the market have zero expected profits. This
equilibrium uniquely defines the number of potential entrants but not their specific
identities (see Berry, 1992).
Finally, we have to parameterise both profit and cost functions. We introduce the
following assumptions:
Π ( N i, X i ) = X i ⋅ β − h ( N i )
(2)
Fib = Z b ⋅ α + Wib ⋅ γ + g ( nb ) + ε ib
where β,α, and γ are vectors of unknown parameters.
Note that profits gross of entry costs are equal for all the banks that decide to enter
the market. They linearly depend on a vector of explanatory variables varying only with
market characteristics, Xi. They also depend on the number of entrants Ni . For the moment,
let us assume that the function h be increasing in Ni or that equivalently profits decline with
the number of entrants. This inverse relation is typical of models of imperfect competition
in which entry drives down prices and hence entrants’ expected profits.
Fixed entry costs are assumed to depend linearly on a set of individual bank
characteristics, given by vector Zb, and on a combination of both market and individual
bank features, represented by the vector Wib. We also assume that entry costs may vary
10
with the number of entries decided by bank b, nb. The function g is assumed to be non
decreasing in nb. This possibility is not discussed by Berry (1992) who merely assumes
that entry decisions by individual firms across markets are independent. Last, εib represents
an error term.
Substituting expressions in (2) in equation (1) we get:
Π ib = X i ⋅ β − h ( N i ) − Z b ⋅ α − Wib ⋅ γ − g ( n b ) − ε ib (3)
The introduction of Ni and nb in the specification for the entrants expected profits is
equivalent to assuming that entry decisions may not be independent. On one hand, given a
specific market, a bank evaluating its entry opportunities in that market has to take entry
choices by the other competitors Ni into account. On the other hand, a specific bank
evaluating entry opportunities in different markets knows that its entry costs in each
market may vary with the total number of its own entry decisions, as represented by nb.
The lack of independence of entry decisions may entail many difficulties in the
estimation procedure9. To address this problem we follow Scott Morton (1999) and assume
that banks take their entry decisions simultaneously10. Consequently, each bank does not
know Ni and has to make some conjectures about its level. We assume that banks base their
predictions about Ni on the pre entry market size, included in the vector Xi. Thus, we use a
sort of reduced form model in which the expected number of entering rivals is predicted
using the size of a market before entry. We expect that the number of entrants will
increase with market size.
Regarding nb, it seems reasonable assuming that, if there are enough organisational
capabilities and financial resources, costs for opening a branch in a new market be
9
See Berry (1992) for a discussion.
About the simultaneous nature of banks entry decisions, note that an Italian bank which wants to open a
new branch notifies its decision to the Bank of Italy which can oppose within 60 days. No answer from the
Central Bank implies that the bank is free to open its new branch. News of the opening of new branches are
notified to the market every six months by the Central Bank. Until that moment, decisions on the new outlets
are kept secret. Thus branching decisions are taken by each bank without knowing their rivals’ choices. This
explains why branching decisions are simultaneous and not sequential.
10
11
invariant with the number of entries. After a given threshold level of entries, a bank may
find difficult to obtain resources to be used to this aim. On one hand, a bank may find
difficult recruiting specialists on the external markets, on the other hand it can face
financial constraints to its geographic expansion. Hence we assume that the function g(nb)
is equal to a constant term, or that entry decisions taken by each bank are independent
across markets, within the limit set by the availability of organisational capabilities and
financial resources. Beyond that threshold, entry costs are assumed to be increasing in nb .
Moreover, we further assume that this threshold level may be increasing with bank size.
Hence larger banks are more likely to take independent entry decisions. In the econometric
analysis, we control for this circumstance by introducing a size variable among the
regressors varying at bank level. We are aware that the threshold level beyond which entry
decisions cease to be independent may respond to factors other than banks’ size. We
control for this possibility in the section on robustness checks.
Following previous considerations, equation (3) modifies as follows:
Π ib = X i ⋅ β ′ − Z b ⋅ α ′ − Wib ⋅ γ ′ − ε ib (4)
where the new equation does not include Ni and nb and parameters are changed to take
account of the reduced form formulation in which Ni is predicted on the basis of some
elements of the vector Xi.
4. The data set and entry definition
Our data refer to geographical units, individual banks and different time periods. We
identify local banking markets with the 103 Italian provinces as defined in 199511. This
choice is partially motivated by the fact that Italian Supervision authorities use provinces
as proxies for the local markets for deposits. We have data on population, value added,
geographic co-ordinates, loans, deposits, number of bank branches, interest rates on loans
and deposits, all spanning from 1990 to 2002 and referred to each province.
12
Data on individual banks include their locations, loans, deposits, branches, total
assets, profits, a set of dummies indicating if a bank has been involved in merger activities.
All the banking statistics come from the Bank of Italy, while the remainder comes from the
Italian Central Statistical Office (ISTAT).
Entry can take place through branching by existing banks, through the creation of
denovo banks, through M&A or through the acquisition of branches from established
banks. In this paper we concentrate only on entry through branching by existing banks. The
peculiarities of denovo banks and of entry through M&A or through branch acquisitions
from other banks suggest that these cases be dealt with separately12. We have also dropped
from our data set co-operative banks. Nowadays, these banks are still subject to some
regulatory constraints in their branching activity and for this reason we prefer not to
include them in our sample13.
We define entry into a province i in year t by bank b as the event occurring when
bank b opens a new branch in province i at time t, provided it owned no branches at time
t-1 in the same province and held branches in provinces different from i.
The data set for the regression analysis is built in the following way. For each bank,
we consider the subset of provinces where it had no branches at time t-1. If entry occurs at
time t in province i according to our definition, we create an entry variable which equals 1
when the bank enters province i. If entry does not occur this variable equals 0 for the same
bank and province.
Entry is a rare event in our data set. In addition, the span of the lag between entry
decision and its occurrence can be long and erratic. For these reasons, we prefer to group
data on entry for each bank into 4 time intervals denoted by ∆p (p=1991-1993, 1994-1996,
1997-1999, 2000-2002). Our entry variable now equals 1 if bank b has a branch in
province i in the time interval ∆p and it had no branch in the same province in the initial
year of the time interval. As before the same variable takes value 0 if no entry occurs for
11
Until 1994 Italian provinces were 95, their number was brought to the present 103 in 1995. The availability
of data at municipal level however allows us to define the variables for the 103 provinces even for the years
before 1995. For interest rates, we do not have data at municipal level and hence these variables are partially
estimated before 1995.
12
As observed in Gobbi and Lotti (2003) denovo banks are usually created under the initiative of local
communities and this fact gives them a special nature.
13
These banks have been included when we have computed variables describing the structure of local
banking markets.
13
that bank in province i during period ∆p. Data for provinces, individual banks and different
time periods are pooled together.
5. Model specification and variables definition
Following equation (4) and entry rule as defined in section 4, the probability of entry
will be determined as follows:
pr(Yib∆ p = 1) = pr ( Π
ib∆ p
≥ 0) = f ( X i , Z b,Wib )
(5)
where Yib∆p is a dichotomous variable equal to 1 if bank b enters into province i at time
interval ∆p and zero otherwise, Xi, Zb and Wib are vectors of variables summarising,
respectively, the characteristics of provinces, individual banks and their interaction. To
avoid the problems that may arise with the potential endogeneity of some regressors, all
the explanatory variables are considered at the year preceding the initial year of each time
interval (for instance, entry in the period 1991-1993 depends on explanatory variables at
1990 and so on).
We have now to define the elements of the three vectors.
1) Variables at provincial level
Coeteris paribus, a larger demand in the local market may increase incumbents
profits and encourage entry. In the traditional Cournot-type oligopoly models, for instance,
the number of competing firms in equilibrium increases with the size of total demand
(Tirole, 1988). We proxy this effect on entry with two variables: the log of residential
population in a province (LPOP) and the log of per capita bank deposits (LDEPC). This
latter variable tries to pin down the intensity of demand for banking services. Thus, we
expect that both LPOP and LDEPC positively affect entry.
In the theoretical debate, there is no agreement about the way in which rivalry within
a local market affects entry. Assume that the toughness of competition may be proxied by
market concentration. According to some authors, seller concentration in a local market
may increase the likelihood of non competitive pricing and thus increase profits margins
14
for the incumbents. This circumstance may encourage entry by new competitors. Demsetz
(1973) has criticised this approach by arguing that higher profits margins of large
incumbents in highly concentrated markets may simply reflect their superior productive
efficiency. In this case, market concentration would have no effect on entry. A similar
conclusion is achieved by the Chicago School by assuming that sunk costs are negligible
(Baumol, Panzar and Willig, 1988). Finally, according to the traditional structure-conductperformance approach (SCP), highly concentrated markets may facilitate co-ordination of
incumbents strategies aimed at deterring entry (Bain, 1956 and Sylos Labini, 1962)14. This
implies that there will be an inverse relation between the degree of market concentration
and the probability of entry.
The degree of competition on provincial markets is approximated by the Herfindhal
concentration index, computed on the markets shares in terms of branches held by each
bank in the provincial market (HSPO). As explained above, this variable may have a priori
a positive, negative or a null effect on entry.
2) Variables varying with bank characteristics
Entry into a new market may require competencies and capabilities that are more
easily found in large organizations. Due to scale economies, a large bank is able to
specialise part of its internal resources to tasks tied to geographic diversification. Thus,
entry costs related to the evaluation of business opportunities in a new market or to the
recruitment of resources to be used to open a new outlet may be decreasing with the size of
the bank. Moreover, high profit margins may reflect high quality of bank managers, strong
market position in established markets, good luck or a combination of all these factors
(Cotteril and Haller, 1992). Hence, we can expect that a more profitable intermediary may
have better chances to overcome entry barriers.
But causality can also run in the opposite direction. Entry by a large or successful
organisation can threat the market position of incumbents while entry of a small or lowprofit bank could be considered as not dangerous for them. Accordingly, entry costs could
be lower for small (unprofitable) banks than for large (profitable) ones.
We expect that a bank’s decision to enter a new province be positively affected by its
size, measured by its total assets (FIT), and by its profitability, defined as net income on
14
According to the well-known Sylos postulate, incumbents may threat entrants to maintain output at the preentry level thereby determining a fall in prices should entry occur.
15
assets (ROA), although, as explained before, there may be an effect of these variables on
entry working in the opposite direction.
Last, we introduce a dummy variable MONO taking value 1 if the bank is located in
only one province and 0 otherwise. In this way, we control for the possibility that banks
located in only one market may face additional costs when they decide to increase their
geographic diversification.
3) Variables varying jointly with provinces and individual bank characteristics.
A second group of factors relating heterogeneity to entry are linked to the fact that
individual banks may find easier to enter specific markets which are less ‘distant’ from
previous bank’s experience.
In this perspective, entering a nearby target market can be an advantage for a bank
with respect to having access to more distant locations.
On the supply side, it may be easier for that bank to transfer best practices associated
with its activities in the core markets toward the newly opened outlet. Usually, bank’s
headquarters have to spend some resources to monitor managerial effort at local level. It is
likely that these agency costs may increase with the physical distance between the bank’s
prior locations and that of the new affiliate (Berger and De Young, 2002)15. On the demand
side, advantages from proximity can be related to potential benefits of reputational effects
and to a better knowledge of the customers’ tastes in the target market. Specifically, if
news about entry opportunities in the target markets are based on soft information,
proximity can help reducing the costs for gathering this information.
All the aforementioned factors referred to informational costs. Traditional transport
costs may represent an alternative explanation for the positive correlation between distance
and entry costs. Entering a distant market may entail higher travel expenses,
communication costs and whatsoever. According to some recent contributions this factor
may be even more relevant than other interpretations based on informational costs16.
Last, the disadvantages from entering a distant market can be partially compensated
from the benefits of increased diversification. This can happen when correlations of
business cycles across different local markets are expected to decline with distance. In any
15
For instance, it might be more difficult to judge whether a relatively bad performance of a local branch
can be attributed to lack of managerial effort or to a negative idiosyncratic shock affecting that local market.
16
case, it is unlikely that this effect may invert the negative correlation between distance and
entry costs.
To measure the effect of distance on entry, consider the set of provinces where the
bank is located at the beginning of period and the complementary set of provinces where
the bank has no branches. For each province belonging to the second set, we calculate the
minimum distance with respect to prior bank’s locations (DISTMIN). Distance is
computed with respect to provincial chief town geographic coordinates. We expect that
this variable should have a negative impact on the decision of the bank to enter that
province.
In other contributions17, this effect is measured by a dichotomous variable that equals
1 when the potential entrant’s previous location is adjacent to that of the target market and
0 otherwise. This is equivalent to assuming that all the advantages of proximity are
restricted to the banks adjacent to the target market. But this assumption is excessively
restrictive. It is difficult to establish a priori at which distance proximity stops producing
its effects. On the contrary, it is likely that the advantages linked to proximity decline
smoothly as distance increases. For this reason we have chosen a continous variable like
DISTMIN.
‘Economic distance’ existing between the entrant and its target market can also affect
entry decisions. Lending activity is strongly influenced by informational problems. If a
bank specialises in extending loans to some specific sectors, it may have an advantage to
enter a market with a similar sectoral specialisation18. Thus we expect that the probability
of entry may increase with similarity (proximity) between the sectoral composition of loan
portfolio of a specific potential entrant and that of the target market. Alternatively, a bank
might want to increase sectoral diversification of its loan portfolio by entering markets
with a different sectoral composition. A priori the effect of this variable on entry may be
ambiguous.
To measure this effect, we have built the variable DISTSECT. Let qsb be the share of
loans granted by bank b to sector s on bank b’s total loans. qsi denotes the share of loans
extended to sector s in province i on the total amount of loans offered to province i by the
16
17
See Buch (2001) and Degryse and Ongena (2003).
See Cotteril and Haller (1992) and Fuentelsaz and Gomez (2001).
17
whole banking system. Hence DISTSECT will be equal to
åq
sb
− q si . This variable
s
measures the distance between the sectoral composition of the bank’s loan portfolio and
that of the target market. Differences are in absolute value and not squared to prevent
extreme observations from having too much influence. As anticipated before, a priori this
variable may have a negative or a positive impact on entry, depending on the value of the
bank’ previous experience with respect to its entry opportunities.
Finally, we introduce the variable QIMPRES, measuring loans offered by bank b to
province i normalised over total loan bank’s portfolio. Notice that according to our
definition a bank b must have no branches in market i to be considered as a potential
entrant into that market. Hence our variable measures the loans offered by a bank b to
province i provided this bank had no branches in that market. In the overwhelmingly
majority of market-bank combinations QIMPRES is 0 or takes on low values. As explained
by Bofondi and Gobbi (2003), a positive value of QIMPRES should help a potential
entrant to overcome informational barriers to entry, thus we expect a positive effect on
entry.
Given the discrete nature of the dependent variable, we have estimated parameters in
(5) by using a binomial logit model. As explained before, we have pooled data referring to
provinces, individual banks and different time intervals. We have also introduced an
additional set of dummy variables. One group indicates whether a bank has been acquired
through M&A by other banks within a given time span (MP=1 for the incorporated banks
and MP=0 otherwise) or if it gains control over another bank through these operations
(MA=1 or MA=0 otherwise). Other dummies account for different time intervals (P1-P4)
and for provinces located in the southern part of Italy (SUD=1).
6. Analysis of the main econometric results
Descriptive statistics and the correlation matrix between regressors are shown in
tables 1 and 2, while table 3 summarizes our main econometric results.
18
Lending to a different sector may lower the incentives to monitoring for the bank (Winton, 1999).
Furthermore, it may increase monitoring costs and create adverse selection problems (Acharya, Hasan,
Saunders, 2002).
18
As expected, the number of users of banking services in a province, LPOP, has a
positive and significant impact on the probability of entry. Provinces with higher deposits
per capita (LDEPC) seem also to attract entry, although the coefficient of this regressor is
only weakly significant.
The degree of market concentration (HSPO) has no statistically significant effect on
entry. This result is due to the collinearity existing between variables measuring the size of
the market in absolute terms as LPOP and the degree of market concentration (the
correlation between the two is -0.44 in our data set). In other words, given that larger
markets are also less concentrated, it is difficult to disentangle the effects of the two
variables on entry. In unreported evidence, we have used population density, defined as
total provincial population divided by its surface (DENSP), to proxy for the size of
demand. We obtained estimates with a negative and statistically significant coefficient for
HSPO (correlation between density and HSPO is equal to -0.22)19.
Thus we interpret previous evidence as showing that market size may have a double
effect on entry. On one hand, the large size of demand raises expected profits for all
potential entrants. On the other hand, large markets exhibit a lower degree of
concentration, thereby encouraging entry. Disentangling these two effects is beyond the
scope of this paper.
Moving to the regressors varying with individual bank characteristics, we find that a
bank with a larger size, higher profitability and with a wider geographical scope of its
operations is more likely to enter a new market. The decision to diversify seems to require
skills and capabilities that are typical of large and successful organisations and of banks
with previous experience of geographic diversification.
Note that the banks operating in only one province (MONO=1) have a smaller size
than those operating across many markets. Despite this positive correlation, our results
show that MONO and FIT exert distinct effects on entry.
We also find that reducing the distance between the previous locations of a bank and
its target market raises the probability of entry into that market (see the coefficient of
DISTMIN in table 3). This spatial pattern of the entry processes is consistent with the idea
that proximity reduces entry costs. Thus physical distance between potential entrants and
19
We have not introduced DENSP in our main specification as it is not clear whether market size can be
better approximated by a measure in absolute terms as LPOP or in relative terms like DENSP.
19
markets location contributes to the raising of entry barriers across markets. As we have
seen, one interpretation is that these obstacles may arise from the difficulties of a potential
entrant to export some of the competitive advantages matured in its pre-entry locations to
distant markets. These difficulties regard the exploitation of reputational capital
accumulated in other markets, the knowledge of markets characteristics based on soft
information and the ability to control local managers. As show before, an alternative
explanation for the effect of DISTMIN is based on the positive correlation between
transport and entry costs.
Banks are more likely to enter markets with a sectoral composition of loans which is
different from that of their pre-entry loan portfolio. This is shown by the positive and
statistically significant coefficient of DISTSECT. This result is partially at odds with
those obtained by Scott-Morton (1999). However the positive effect of DISTSECT on
entry is obtained only after conditioning for DISTMIN. If we drop this variable from the
regression, the coefficient of DISTSECT becomes negative and continues to be significant
(estimated coefficient is equal to -.39 with a z=-2.42). This evidence is partially explained
by the fact that the two variables covariate (their correlation is equal to .19 and is
significantly different from zero). Thus, in some circumstances economic distance may be
a proxy for geographical distance. In any case, according to our evidence, once we hold
physical distance between potential entrant and the target market constant, banks seem to
prefer markets with a sectoral specialisation that is different from that of their loan
portfolios. Hence banks may use geographic diversification to change the initial sectoral
composition of their lending activity.
Finally, we find that a bank lending to a province where it has no branches has a
greater probability to enter that market. Establishing relations with customers in a province
where the bank has not yet branched gives the potential entrant some advantages to
overcome informational barriers to entry. This evidence is consistent with results obtained
by Bofondi and Gobbi (2003) showing that entrants with a positive value of QIMPRES
experience a lower post-entry default rate than other entrants.
As to the additional controls based on dummy variables, we find that the probability
of entry increases if a province is located in southern Italy (see the positive and statistically
significant coefficient of SUD=1). Moreover, a bank that during the period falls under
control of another bank through M&A operations (MP=1) has a lower probability to enter a
20
new province20, whereas a bank which gains control over another bank (MA=1) is more
likely to enter a new market. This last piece of evidence could be interpreted as the fact
that market penetration strategies based on M&A and those linked to branching in new
markets can be complementary and not alternative strategies.
To evaluate the economic impact of some of the regressors, particularly of those
varying with individual banks, we run a simulation of the model. Main results are reported
in table 6. Reducing the mean value of DISTMIN by one standard deviation raises the
predicted probability of entry by nearly 2.8 times (predicted probability of entry is equal to
0.067% at the mean values of the regressors). This is the most important effect that we
obtain in our specification. MONO has also a substantial impact on entry: banks operating
in more than one market nearly double the probability of entry with respect to banks with
branches in only one market. Moreover, a reduction by one standard deviation of LPOP
lowers the probability of entry by 44 per cent. All the remaining regressors have much
smaller effects, going from a variation of 23 per cent of the predicted probability for ROA
to a change around 10% for FIT, DISTSECT and QIMPRES.
Hence, not only variables varying with individual bank characteristics have
statistically significant effects on entry, but they also have an economic impact whose
magnitude is at least comparable to that of the structural market variables.
7. Robustness checks
We devised three different sets of controls for our econometric results. The first
group is related to different specifications of the determinants of the entry processes, the
second to alternative definitions of the potential entrant set; the third is concerned with the
issue of alternative definitions of geographical units or local markets.
7.1 Controls based on alternative specifications
In discussing the problem of lack of independence of entry choices in the same
market, we assumed that banks form their expectations about the number of potential
entrants looking at the pre entry market size. Here we conjecture that these expectations
20
Obviously this is also a consequence of fact that the majority of target banks are acquired by bidder before
the end of time intervals in which we consider entries.
21
may also depend on the observed number of past entrants. To this aim, we add to the
regression the lagged value of the number of entrants divided by the number of banks
operating in the market (LENT). A priori this variable could have a positive or a negative
effect on entry, depending on the mechanisms driving potential entrants’ expectations. In
unreported evidence, we show that LENT is never statistically significant at the usual
probability levels. Thus previous results are robust to this control.
Regarding the independence of entry decisions of each bank across different markets,
we assumed that entry costs do no vary with the number of entries up to a threshold level,
depending on a bank’s size. We introduce a further check for the independence of entry
decisions based on the following argument. Assume that our previous regressors at bank
level control also for the total growth opportunities of a bank. A bank can expand either by
opening a new outlet in its pre entry locations or by branching in a new market. Having
controlled for its total growth opportunities, the two expansion strategies may be
independent, complements or substitutes. In the latter case, we may suspect that entry
decisions across markets are not independent. To enter a new market a bank should give up
some expansion projects in its pre entry locations. Thus, it is likely that this bank can be
subject to some constraints to its expansion strategies and that these constraints can
reflect a lack of independence of entry decisions.To control for this effect, we have
introduced a variable based on the ratio between the number of pre entry locations where a
bank opened a new outlet (net of those in which it closed branches) and the total number of
its pre entry locations (LEXIN). To avoid problems with endogeneity we take the lagged
value of LEXIN. According to our unreported evidence, LEXIN has a positive and
statistically significant effect on entry (the estimated coefficient is .34 with a z = 2.25).
The introduction of this new variable into the regression does not change previous results.
The complementary between expansion strategies based on banks pre entry locations and
those based on entry into new markets may be consistent with the independence of entry
decisions across different local markets.
In unreported evidence we substituted LDEPC for the log of per capita gross
domestic product (LPILP) without affecting previous results. To measure the degree of
competition, we replaced HSPO with the difference between interest rates on loans and on
22
deposits on local credit markets (DIT)21. DIT has a negative and significant effect on entry
when market size is approximated by DENSP, whereas it has a coefficient not significantly
different from zero when LPOP has been used. Hence, results are similar to those obtained
using HSPO.
We also added a measure of borrower risk in each province, proxied by the ratio of
bad loans to total loans, to the basic specification. A priori, one can expect that banks may
not want to enter markets filled with high-risk borrowers, for instance, because of higher
ex ante informational barriers. The coefficient for this variable turns out to be not
significantly different from zero and results for the other variables do not change.
Finally we introduced a variable representing supply conditions, i.e. the density of
branches held by incumbents (total number of branches in a province divided by its
surface, DSPORT). We expected a negative impact of this variable on entry, since the
increasing ability of incumbents to cover a market area should discourage entry. The
coefficient on DSPORT has the expected sign and is also significantly different from zero
(the coefficient is -83.2 with a z= -2.53). Results related to the other regressors are left
unchanged. Despite these encouraging results, we did not introduce DSPORT into the
main specification because of collinearity with the other regressors (correlation with LPOP
and LDEPC is about 0.43)22.
Whatever controls one might think of, there is still the possibility of having omitted
relevant market and individual bank characteristics affecting entry. For this reason, we
have introduced provincial and individual bank fixed effects in our regression. Due to
collinearity, we dropped those regressors varying with either one of the two dimensions,
namely LPOP, LDEPC, HSPO and SUD, when introducing provincial fixed effects, and
ROA, FIT, MONO, MA and MP, when using individual bank fixed effects23. Results are
reported in columns 2 and 3 in table 3. Taking account of unobservable fixed effects is a
21
This variable could also reflect the degree of risk of local borrowers or the information asymmetries
between lenders and their customers across different local banking markets.
22
The sign of the coefficient for DSPORT is negative only after conditioning for LPOP. If we drop LPOP,
the density of branches has a positive effect on entry signalling that markets with a large number of branches
per square kilometre offer also better profit opportunities. We run further controls based on different
definitions of bank size and profitability. Again, our main conclusions are untouched by these changes.
23
Given that our data have also a temporal dimension we could have introduced fixed effects together with
variables varying only with markets or banks. We have not opted for this possibility because we only have 4
observations for each market and bank.
23
relevant robustness check for the other regressors in the model, particularly for those
variables varying jointly with market and individual bank characteristics.
From an econometric point of view, these two specifications correspond to the so
called fixed effect conditional logit model. With this procedure , the probability of entry in
a market is conditioned on the total number of entries in a market when using market fixed
effects or on the total number of entries by a bank when individual bank fixed effects are
introduced. Observations for which a bank enters in all the remaining target markets or
never enters do not contribute to the likelihood function, hence they are dropped from the
regression (see Fuentelesaz and Gomez, 2001 for a discussion). For this reason, regression
in column 3 has a smaller number of observations.
A broad overview of results in columns 2 and 3 shows that main findings do not
change much after the introduction of fixed effects. In most cases estimated coefficients
preserve the same sign they had in the basic specification. The only exception is given by
the parameter on DISTSECT that takes on a negative sign in column 3 and is no longer
significant.
7.2 Controls based on different definitions of potential entrants
Until now our definition of the pool of potential entrants includes all existing banks
which had no branches in a specific market at the beginning of a predetermined time
interval. One can wonder whether this definition is too comprehensive. Some banks may
not consider a market as a target for their entry decisions independently from the
attractiveness of that area in terms of profits gross of entry costs. This would happen when
a bank entry costs are so high as to prevent entry in any market or when entry costs rapidly
increase with the number of entry decisions.
These special cases may drive our previous findings in many ways. First,
explicative power of the variables representing market characteristics is obviously reduced
by the presence of these special banks. Second, as far as the regressors varying with bank
characteristics are not able to catch the peculiarities of these potential entrants, we are
faced with a problem of omitted variables.
To address this problem, our estimation strategy consists of an identification of four
groups of special banks. We then drop these individuals from the regression and compare
24
the results with those of table 3 for which the broadest definition of the pool of potential
entrants is used.
The four groups of special banks include: a) banks who never enter; b) small sized
banks located at great distance from their target markets; c) banks with a low free capital;
d) banks entering a market through mergers.
A bank taking no positive decision about entry during a period may have an
exceptionally high level of fixed entry costs24.
Banks belonging to group b) are meant to represent the cases in which small banks
do not consider distant markets as actual targets for their entry decisions. Consequently, we
drop from the regression those observations for which banks have total assets below the
first quartile of FIT and are located at a distance from their target markets greater than the
third quartile of DISTMIN.
A low level of free capital may constraint entry decisions because of lack of
financing. Hence, entry costs may start increasing after a few entries. In this circumstance,
entry decisions cannot be assumed to be independent across markets. Moreover, a bank
subject to strong financial constraints can consider only a limited subset of market areas as
potential targets. In finance literature, it is usually assumed that small entities are more
likely to face financial constraints. Thus, as far as this relation holds, the size variable in
our regression should control for the influence of these constraints on entry. Yet, given the
special nature of banks, the correlation between size and financial constraints may be
looser than the corresponding relation holding for non financial firms. For this reason, we
drop banks with a ratio of free capital over total assets below the first quartile of the same
variable.
Last, we dropped observations related to banks entering a market via merger or by
acquiring branches from incumbents. In previous regressions, ENTRY for these banks
equals 0. Although we control for the special nature of these banks with a dummy variable,
they may still drive previous results. Thus we have estimated the probability of entry
conditional on those banks entering only through branching.
24
Independently from its size, profitability, previous location choices, this bank may think for instance of
having managerial resources to operate only in markets where it has already branches.
25
Note that that there is only a partial overlapping between the different groups of
banks defined above. Results are reported in table 4. The main findings are the same as
those in table 3 even after dropping bank-markets cells as defined before. In particular,
previous conclusions should not be driven by including in the set of potential entrants those
banks with exceptionally high level of entry sunk costs or those entities whose entry
decisions across markets are not independent and therefore may restrict their entry
decisions only to a narrow subset of local markets.
7.3 An alternative definition of local markets
A correct identification of market borders is essential to the definition of entry. A
wrong delimitation of market areas would imply that branching in the same market can be
considered as an entry decision or, the other way round, a true entry decision may be
identified as branching in a market where a bank has already opened an outlet. The so
called ‘border effects’ may have serious consequences for the validity of econometric
results using spatial data and therefore it is important to control for their influence.
To this aim, we replicated our estimates by identifying geographical units with
local labour systems (LLS) instead of provinces. According to the Italian Central Statistical
Office (ISTAT), in 1991 there were 784 LLS in Italy covering all the national territory.
These are self-contained clusters of municipalities, whose boundaries are defined on the
basis of daily commuting patterns in a way that the majority of workers living in a
geographical unit have their own workplace within the same area. With respect to the
classification based on provinces, LLS offer a much more detailed representation of local
markets. Moreover, they are a good starting point to identify local banking markets, given
that most banking services have a limited geographical scope. Their usage comes however
at a cost as, at this level of aggregation, we do not have the availability of data to compute
DISTSECT and QIMPRES.
Apart from the latter regressors, table 5 replicated estimates in table 3 using LLS
definition. In column 1 we reported the basic specification and in columns 2 and 3 we
introduced unobservable LLS and individual bank fixed effects. None of previous results
are changed, some conclusions are even sharper than those based on provinces.
26
To get an idea about the magnitude of the economic effects implicit in the estimated
coefficients, we simulated again the model (see table 6). As compared to the market
definition based on provinces, we get stronger effects for the variables varying with market
characteristics and for DISTMIN. These results may be due to the fact that LLS may offer
a better representation of banking markets at local level.
8. Concluding remarks
In this paper we have examined entry decisions of approximately 300 Italian banks
across different local markets and time periods. Differently from what was done in other
contributions on entry, we have stressed the importance of the heterogeneity of potential
entrants in determining accessibility to local markets. The features of a unique data set
have enabled us to carry out an extensive econometric analysis of the determinants of
entry, including a set of controls based on alternative specifications, different definitions of
the pool of potential entrants and of geographical units. It turns out that the scale of a bank
activity, its profitability and above all the structural characteristics of its geographic
network deeply influence the ability to overcome entry barriers. In particular, physical
distance between potential entrant and market location has a negative effect on entry.
Different conclusions can be drawn from these econometric findings. The
restructuring of the Italian banking sector through the nineties, largely based on M&A
operations, has increased the size of the banks, their profitability and their spreading across
different local markets. According to our results, these changes should have brought about
a reduction in entry sunk costs and a consequent increase in the contestability of local
markets. It is evident that M&A operations may have had many other effects on the
competitive conditions of local banking markets25.
As shown by our econometric findings, entry costs increase with physical distance
between a potential entrant and its target market. This implies that at least a part of entry
barriers can be explained without resorting to the strategic interaction between different
banks. Thus, the competitive process in the banking activity may be hampered by
geographical space. Recent technological advances has partially changed the way in which
clients have access to banking services, reducing the need of direct contact between the
25
For the analysis of M&A in the Italian banking sector see Focarelli et al (2002). Colombo and Turati
(2002) look at the consequences of these operations for the spatial pattern of local credit markets.
27
bank and its customers26. Although these changes may reduce the importance of branching,
it is unlikely that they can completely overshadow it. In this perspective, our results on the
effects of distance may sound as bad news for the process of integration of the European
banking markets.
Our work can be extended along different directions. As we have already shown, the
independence of entry decisions across markets is crucial for the validity of our
econometric procedure. In a previous section, we have introduced controls aiming at
identifying groups of banks for which this assumption may not hold. But the violation of
the assumption of independence of entry decisions may also be explained by other factors.
A pool of oligopolistic banks might, for instance, collude on the number of markets each
bank may enter. A violation of this agreement by a bank might generate a reaction by the
others increasing its entry costs in other markets. In more general terms, an entry process
could be designed as a two stage decisional procedure where a bank chooses in a first stage
the number of entries and in a second stage which markets to enter. These topics might be
covered in a future research project.
26
On the effects of internet banking on the contestability of banking markets, see Corvoisier and Gropp
(2001).
28
References
Acharya V. V., Hasan I. and Saunders A. (2002), Should Banks Be Diversified? Evidence
from Individual Bank Loan Portfolios, mimeo.
Amel D. F. and J. N. Liang (1997), Determinants of Entry and Profits in Local Banking
Markets, in “Review of Industrial Organization”, vol. 12, pp. 59-78.
Bain J. S. (1956), Barriers to new competition, Cambridge Mass., Harvard University
Press.
Baumol A., Panzar J. C. and R. D. Willig (1988), Contestable Markets and the Theory of
Industry Structure, Harcourt Brace Jovanovich, Orlando Florida.
Berger A. N., Bonime S. D., Goldberg L. G. and White L. J. (2000), The Dynamics of
Market Entry: the Effects of Mergers and Acquisitions on De Novo Entry and Small
Business Lending in the Banking Industry, Federal Reserve Board, mimeo.
Berger A. N. and De Young R. (2002), The Effects of Geographic Expansion on Bank
Efficiency, Board of Governors of the Federal Reserve System, Finance and
Economics Discussion Series, no.2001-03.
Berry S.T., (1992), Estimation of a Model of Entry in the Airline Industry, in
“Econometrica”, vol. 60, n. 4, pp. 889-917.
Bofondi M. and Gobbi G. (2003), Bad Loans and Entry into Local Credit Markets, Banca
d’Italia, mimeo.
Buch C. M. (2001), Distance and International Banking, Kiel Working paper n. 1043.
Cerasi V., Chizzolini B. and Ivaldi M. (2000), Branching and Competitiveness across
Regions in the Italian Banking Industry, in Polo M. (a cura di) “Industria Bancaria e
Concorrenza”, Il Mulino, Bologna.
Colombo L. and Turati G. (2002). La dimensione territoriale nei processi di
concentrazione dell’industria bancaria italiana, Quaderni dell’Istituto di Economia e
Finanza, n. 50, Università Cattolica del Sacro Cuore, Milano.
Conigliani C. e G. Lanciotti (1978), Concentrazione, concorrenza e controlli all’entrata, in
G. Carli (a cura di), La struttura del sistema creditizio italiano. Elementi per un
riesame, Bologna, Il Mulino.
Corvoisier S. and R. Gropp (2001), Contestability, Technology and Banking, mimeo.
Cotterill R. W. and L. E. Haller (1992). Barrier and Queue Effects: a Study of Leading US
Supermarket Chain Entry Patterns, in “The Journal of industrial Economics”, vol.
XL, n. 4, pp. 427-440.
Degryse H. and Ongena S. (2003), Distance, Lending Relationships and Competition,
mimeo, Tilburg Univerisity.
Dell’Ariccia G., Friedman E. and Marquez R. (1999), Adverse Selection as a Barrier to
Entry in Banking Industry, in “Rand Journal of Economics”, vol. 30, no. 3, pp. 515534.
Demsetz H. (1973) Industry Structure, Market Rivalry and Public Policy , in “Journal of
Law and Economics”, vol. 16, pp 1-9.
29
Focarelli D., Panetta F. and Salleo C. (2002), Why do banks merge? in “Journal of Money
Credit and Banking”, Vol. 34, No. 4.
Fuentelesaz L. and J. Gomez (2001), Strategic and Queue Effects in Spanish Banking , in
“Journal of Economics & Management Strategies”, vol. 10, no. 4, pp. 529-563.
Geroski P. A. (1995), What Do We Know about Entry? in “International Journal of
Industrial Organization”, Vol. 13, No. 4, pp. 421-440.
Gobbi G. and Lotti F. (2003), Entry Decisions and Adverse Selection: an Empirical
Analysis of Local Credit Markets, Banca d’Italia, mimeo.
Juan De R. (2002), The Determinants of Entry and Exit in Independent Submarkets:
Emprical Evidence from Spanish Retail Banking Market, mimeo.
Melitz M. J. (2002), The Impact of Intra-Industry Reallocations and Aggregate Industry
Productivity, NBER Working Paper, 8881.
Petersen M. and Rajan R. (2000) “Does Distance Still Matter? The Information Revolution
in Small Business Lending”, NBER Working Paper no. 5602.
Scott Morton F. (1999), Entry Decisions in the Generic Pharmaceutical Industry, in “Rand
Journal of Economics”, vol. 30, no. 3, pp. 421-440.
Seelig, S. and T. Critchfield, (1999), Determinants of Denovo Entry in Banking, Federal
Deposit Insurance Corporation, WP 99-1.
Stomper A. (2003), A Theory of Bank’s Industry Expertise, Market Power and Credit Risk,
mimeo, University of Vienna.
Sutton J. (1998), Technology and market Structure. Theory and History, Cambridge Mass,
The MIT press.
Sylos-Labini P. (1962), Oligopoly and technical progress, Cambridge Mass., Harvard
University Press.
Tirole J. (1988), The Theory of Industrial Organization, Cambridge, The MIT Press.
Toivanen O. and Waterson M. (2001), Market Structure and Entry: Where’s the Beef?,
The Warwick Economics Research Papert Series, no. 593, University of Warwick.
Winton A. (1999), Don’t Put All Your Eggs in One Basket? Diversification and
Specialization in Lending, working paper, University of Minnesota.
30
Table 1– DESCRIPTIVE STATISTICS
Variables
Nr. of Obs.
Mean
Std. Deviation
Minimum
Maximum
1st period (1990)
lpop
ldepc
hspo
103
103
103
6.01
15.58
0.16
0.70
0.40
0.07
4.52
14.58
0.03
8.23
16.24
0.37
roa (x100)
fit
302
303
0.62
2391.34
0.96
6428.78
-13.12
10.60
3.19
55276.90
29496
29116
29496
390.79
0.70
0.10
268.32
0.23
0.78
12.26
0.14
0.00
1166.76
1.92
34.58
distmin
distsect
qimpres (x100)
2nd period (1993)
lpop
ldepc
hspo
103
103
103
6.02
15.80
0.15
0.70
0.39
0.06
4.53
14.85
0.03
8.24
16.43
0.37
roa (x100)
fit
271
272
0.37
3508.93
0.87
9551.89
-8.91
14.90
2.24
71538.60
26021
25819
26021
378.93
0.67
0.10
260.39
0.21
0.68
12.26
0.14
0.00
1166.76
1.73
23.23
distmin
distsect
qimpres (x100)
3rd period (1996)
lpop
ldepc
hspo
103
103
103
6.02
15.91
0.14
0.70
0.37
0.06
4.52
15.00
0.04
8.24
16.52
0.33
roa (x100)
fit
252
257
0.27
4691.60
0.74
12310.97
-4.71
10.6
1.72
100939
24356
23949
24356
377.29
0.70
0.12
260.14
0.25
0.80
12.26
0.10
0.00
1166.76
2.00
34.64
distmin
distsect
qimpres (x100)
4th period (1999)
lpop
ldepc
hspo
103
103
103
6.02
15.82
0.15
0.71
0.37
0.08
4.52
14.98
0.04
8.24
16.80
0.65
roa (x100)
fit
251
252
0.42
5460.67
2.03
13992.05
-25.13
15.2
6.03
105642.8
23901
22688
23901
357.88
0.75
0.16
256.48
0.33
1.06
12.26
0.18
0.00
1166.76
2.00
45.13
distmin
distsect
qimpres (x100)
All periods
lpop
ldepc
hspo
roa (x100)
fit
distmin
distsect
qimpres (x100)
412
412
412
6.02
15.78
0.15
0.70
0.40
0.07
4.52
14.58
0.03
8.25
16.80
0.65
1076
1084
0.43
3930.66
1.25
10814.12
-25.13
10.60
6.03
105642.80
103774
101572
103774
377.07
0.70
0.12
261.98
0.26
0.83
12.26
0.10
0.00
1166.76
2.00
45.13
31
Table 2– CORRELATIONS BETWEEN THE EXPLANATORY VARIABLES
lpop
ldepc
lpop
1.0000
0.0760
ldepc
hspo
roa
fit
distmin
distsect
1.0000
hspo
-0.4415
-0.1619
1.0000
roa
-0.0009
-0.0482
0.0168
1.0000
fit
-0.0480
-0.0135
0.0169
-0.0033
distmin
0.0601
-0.2766
-0.0173
0.0090
-0.2138
1.0000
distsect
-0.0426
0.0479
0.0283
-0.0170
-0.0311
0.1916
1.0000
qimpres
0.1855
0.0936
-0.0688
-0.0120
-0.0076
-0.0807
0.0337
qimpres
1.0000
1.0000
32
Table 3- Logit analysis of entry: 103 provinces
Explanatory
Variables
LPOP
Provincial
fixed effects
Individual bank
Fixed effects
0.649494 ***
(11.24)
0.758748 ***
(12.16)
LDEPC
0.293839
(1.58)
0.608425 ***
(3.03)
HSPO
0.081542
(0.12)
0.544993
(0.78)
ROA
17.83475 ***
(4.14)
17.98985 ***
(4.10)
FIT
0.000017 ***
(6.03)
0.000015 ***
(5.14)
DISTMIN
-0.010661 ***
(-23.40)
-0.012436 ***
(-23.85)
DISTSECT
0.460715 ***
(3.20)
0.525154 ***
(3.45)
QIMPRES
13.76868 ***
(8.68)
12.94698 ***
(7.65)
MONO
-1.087393 ***
(-8.01)
-1.070041 ***
(-7.88)
MP (1)
-1.378255 ***
(-6.60)
-1.380394 ***
(-6.62)
MA (1)
0.817668 ***
(8.99)
0.831084 ***
(9.09)
SUD
0.599895 ***
(3.93)
P2 (2)
-0.698535 ***
(-5.49)
-0.617104 ***
(-5.09)
-0.858526 ***
(-6.40)
P3 (2)
-0.163353
(-1.33)
-0.018570
(-0.17)
-0.292218 **
(-2.23)
P4 (2)
0.524010 ***
(5.14)
CONSTANT
Log Likelihood
Pseudo R
2
Number of obs.
-0.009400 ***
(-19.67)
-0.498312 *
(-1.87)
25.19626 ***
(11.37)
0.510714 ***
(3.06)
0.638524 ***
(6.86)
0.509258
(4.42)
-11.95729 ***
(-4.12)
-3635.7207
-3339.4914
-2679.9791
100771
63260
0.2449
100771
(1) Dummy variable relative to banks not involved in M&A operations omitted. (2) Dummy variable
relative to first period omitted.
*** indicates significance at 1% level
** indicates significance at 5% level
* indicates significance at 10% level
33
Table 4- Logit analysis of entry: 103 provinces
Explanatory
Variables
LPOP
Whithout banks who
never entry (3)
Whithout small and
distant banks (4)
Whithout banks with low
free capital (5)
Whithout banks who
enter by merger (6)
0.678874 ***
(11.03)
0.584378 ***
(9.81)
0.673874 ***
(9.68)
0.691275 ***
(9.89)
LDEPC
0.255182
(1.31)
0.196970
(1.05)
0.353109
(1.55)
0.557403 **
(2.40)
HSPO
0.104847
(0.15)
-0.023913
(-0.03)
0.773566
(0.89)
ROA
22.10565 ***
(5.06)
28.05723 ***
(6.08)
22.81241 ***
(4.50)
21.92921 ***
(3.99)
FIT
0.000012 ***
(4.18)
0.000017 ***
(6.10)
0.000027 ***
(6.72)
0.000026 ***
(7.78)
-0.010193 ***
(-21.62)
-0.010837 ***
(-22.08)
-0.010710 ***
(-19.51 )
-0.013112 ***
(-20.52 )
DISTMIN
DISTSECT
0.232204
(1.60)
QIMPRES
30.00969 ***
(11.73)
0.108103
(0.69)
0.283795
(1.59)
-0.682933
(-0.74)
0.674227 ***
(4.29)
15.8033 ***
(9.62)
14.91849 ***
(8.14)
12.5849 ***
(7.65)
MONO
-0.277779 *
(-1.94)
-0.961649 ***
(-6.73)
-1.076699 ***
(-6.93)
-0.828731 ***
(-5.86)
MP (1)
-0.647001 ***
(-2.93)
-1.369586 ***
(-6.52)
-1.403383 ***
(-5.39)
-1.384208 ***
(-6.46)
MA (1)
0.494829 ***
(5.34)
0.788381 ***
(8.66)
0.740818 ***
(6.19)
0.287354
(1.51)
SUD
0.679065 ***
(4.29)
0.532143 ***
(3.45)
0.655169 ***
(3.50)
0.695369 ***
(3.60)
-0.648264 ***
(-5.08)
-0.590431 ***
(-3.82)
-0.717475 ***
(-4.96)
-0.119783
(-0.97)
-0.098148
(-0.66)
-0.726146 ***
(-4.67)
P2 (2)
-0.215897
(-1.62)
P3 (2)
0.281748 *
(2.19)
P4 (2)
0.579183 ***
(5.42)
0.497409 ***
(4.81)
0.658575 ***
(5.09)
-10.82979 ***
(-3.57)
-9.826662 ***
(-3.36)
-13.26932 ***
(-3.72)
CONSTANT
Log Likelihood
Pseudo R
2
Number of obs.
0.082425
(0.67)
-16.11673 ***
(-4.44)
-2893.4754
-3498.0601
-2505.4918
-2530.5431
0.2369
0.2239
0.2437
0.2591
29713
79061
78020
91188
(1) Dummy variable relative to banks not involved in M&A operations omitted. (2) Dummy variable relative to first period omitted. (3)
Excluding banks with ENTRY=0 across all markets in a period. (4) Excluding banks with total assets below the first quartile of FIT and
distance greater than the third quartile of DISTMIN.(5) Excluding banks with a ratio of free capital over total assets below the first
quartile of the same variable.(6) Dropping banks entering through mergers.
*** indicates significance at 1% level
** indicates significance at 5% level
* indicates significance at 10% level
34
Table 5- Logit analysis of entry: 784 LLS
Explanatory
Variables
LPOP
LLS
fixed effects
Individual bank
fixed effects
0.804108 ***
(37.06)
0.875778 ***
(39.27)
LDEPC
0.518675 ***
(6.12)
0.527267 ***
(6.10)
HSPO
0.250763
(1.17)
0.340240
(1.61)
ROA
15.49457 ***
(5.79)
17.61724 ***
(6.42)
FIT
0.000014 ***
(13.10)
0.000012 ***
(11.16)
DISTMIN
-0.014564 ***
(-38.33)
-0.01618 ***
(-39.05)
MONO
-1.072596 ***
(-5.58)
-0.973253 ***
(-5.04)
MP (1)
-1.003646 ***
(-9.43)
-1.018715 ***
(-9.54)
MA (1)
0.697932 ***
(13.40)
0.693700 ***
(13.26)
SUD
0.359985 ***
(4.55)
P2 (2)
-0.762508 ***
(-10.97)
-0.632836 ***
(-9.45)
-0.861292 ***
(-12.37)
P3 (2)
-0.510002 ***
(-7.43)
-0.305003 ***
(-4.84)
-0.681909 ***
(-9.93)
P4 (2)
-0.015484
(5.14)
CONSTANT
-17.23749 ***
(-22.05)
Log Likelihood
Pseudo R
2
Number of obs.
-11404.115
-0.013247 ***
(-34.87)
0.358390 ***
(4.21)
0.134950 **
(2.44)
-0.171694
(-2.84)
-9991.3139
-10303.257
568702
651250
0.2656
796363
(1) Dummy variable relative to banks not involved in M&A operations omitted. (2) Dummy variable
relative to first period omitted.
*** indicates significance at 1% level
** indicates significance at 5% level
* indicates significance at 10% level
35
Table 6- Economic effects from logit analysis
Explanatory
variables
LPOP
103 provinces
Sign of the
Change in
effect (1)
predicted
probability (2)
+ ***
44%
784 LLS
Sign of the
Change in
effect (1)
predicted
probability (2)
+ ***
92%
LDEPC
+ n.s.
12%
+ ***
30%
HSPO
+ n.s.
6%
+ n.s.
6%
ROA
+ ***
23%
+ ***
19%
FIT
+ ***
10%
+ ***
14%
MONO
- ***
91%
- ***
72%
DISTMIN
- ***
279%
- ***
381%
DISTSECT
+ ***
12%
QIMPRES
+ ***
15%
(1) *** Indicates significance at 1% level; ** indicates significance at 5%
level; * indicates significance at 10% level. (2) Obtained by moving
regressors by one standard deviation.