While, from a glance at the data, the trend of the average branching ... and sunk costs, it is hard to identify in the set of EEC directives the origin of changes in the ..... be recovered if the bank decides to exit the market or close down branches.
Sunk costs and competitiveness of European banks after deregulation Vittoria Cerasi Universita' degli Studi di Milano and FMG, London School of Economics Barbara Chizzolini Universita' Bocconi, Milano Marc Ivaldi GREMAQ-EHESS and Institut D’Economie Industrielle, Toulouse
April 1998 Abstract This paper investigates the determinants of the structure of the banking industry by fitting a monopolistic competition model to a sample of banks drawn from nine EEC countries over 1989-1993. In the theoretical model, banks decide strategically both entry and the branching size of their network. The estimation then measures the branching costs and an upper bound for the entry costs. It also assesses how these costs evolve over time and to what extent various European directives, aiming at deregulating the banking industry, influence them.
JEL classification: G21; L13; L59 Keywords: Banking; Market structure and oligopoly; Deregulation
Corresponding author: Vittoria Cerasi - Dipartimento di Economia e Politica Aziendale - Via Conservatorio, 7 - 20122 Milano - Italy. E-mail: vittoria.cerasi@unimi.it Acknowledgments: We would like to thank Bruno Biais, Giovanni Ferri, Jean-Charles Rochet, John Sutton and Jean Tirole for helpful comments and seminar participants at International Conference in Game Theory and Applications in Milan, EEA’97 in Toulouse, ASSET in Lisbon, Bank of England, IDEI and Ecole de Gestion in Toulouse, CIDE in Bologna. We thank Andrea Balzarini and Elena Carletti for extremely valuable research assistance. Financial support from Ricerca di base and Centro Baffi, Universita’ Bocconi (Milan), and Institut D’Economie Industrielle, University of Toulouse, is gratefully acknowledged. This paper is part of the TMR research project “The Industrial Organisation of Banking and Financial Markets in Europe" (IOBFME).
1. Introduction Since the eighties, the European banking industries, before heavily regulated, have experienced a phase of deregulation. On the one hand, at the country level, regulators have left more and more market forces to determine the structure of the market. Today, in almost all EEC countries, banks are free to compete for their clients. On the other hand, at the EEC level, each country has committed to remove barriers across financial markets, in order to move towards a complete integration of markets, both on the demand and the supply side. These two aspects have important implications for entry and competition in the banking industry, as they imply that, since the process of deregulation has started, the structure of the banking industry may be better explained by the strategic interactions of banks than by the effect of regulation. The European banking industry provides an interesting challenge for a test of a model of monopolistic banking competition. First, the set of regulations does not fully explain the evolution of the industry structure; second, the changes in the institutional setting allow us to test the impact of regulation on the conduct of banks and on the structure of the banking industry. The process of deregulation has strong implications for competition among banks and for bank decision to enter or exit markets. By deregulation, in fact, we mean the changes in rules, implemented at national level, which, having received their input mainly by EEC directives, influence either the conduct or the structure of banking markets. In particular, on the one hand, in these years almost all EEC members have implemented policies oriented to increase competition among banks, by lifting controls on interest rates or fighting more fiercely cartels among the largest banks. The effect was an increase in the ability to compete on interest rates by the overall banking system. On the other hand, policies towards larger integration of financial markets have been pursued on the demand side, by lifting capital controls at the country level, and on the supply side, by reducing the cost of entry into new banking markets. These policies may have affected the decision of entering and opening of new branches across countries and indirectly the amount of competition that banks envisage to face in the future. An important objective of deregulation is to increase the degree of competitiveness in the banking industry in order to reduce the price of financial services for customers in this industry.
1
However, as already pointed out by some papers,1 deregulation has increased the degree of concentration in the European banking industries, as a result of a consolidation process, mainly inside borders, that has occurred in the past years. This may limit the expected benefits of deregulation on allocative efficiency, if monopoly power of each bank has not decreased. Furthermore, interest rate deregulation, by increasing the toughness of price competition may have reduced the incentives to set larger branching network. This may have occurred for two reasons.2 First, as profits shrink as a result of tougher interest rate competition, banks may find less profitable to maintain larger branching networks. Second, because interest rate competition has for long time been impeded, branching networks have been the only way to capture new clients, giving rise to overinvestment in branches. The aim of this paper is to measure the degree of competitiveness acquired by the banking industry as a result of deregulation and to evaluate the costs of entry and branching for each banking system. More specifically, we estimate entry and branching costs, building our econometric analysis on a model of monopolistic competition, where banks compete for deposits both on interest rates and branching networks. The main result of this paper is that, after deregulation, the degree of competition in interest rates in the EEC banking industries has risen and that the average size of branching networks has decreased. Our econometric analysis helps in understanding the impact of deregulation on the average branching size of banks. While, from a glance at the data, the trend of the average branching size for all countries of our sample seems increasing, we show that deregulation has negatively affected the likelihood of opening new branches. Furthermore we find that, when evaluating the impact of deregulation on competitiveness and sunk costs, it is hard to identify in the set of EEC directives the origin of changes in the industry structure, even though considering the actual implementation date in each country. First of all, the degree of competitiveness of each single banking industry is very sensitive to country specific factors. Moreover, in all European banking industries, there is a trend towards tougher price competition and smaller branching networks, which can not be attributed to the actual implementation of each specific EEC directive. We propose to interpret this result as
1
See Gual and Neven (1992), Economic Research Europe (ERE) Summary Report (1996) and Cerasi (1996) for evidence that the degree of concentration in the European banking industry has increased since the process of deregulation has started. 2 See for example Neven (1989) and Cerasi (1995) for implications of deregulation on the size of the branching networks.
2
evidence that legislative changes at the country level are less effective than those at the community level, at least for what concerns the shape of the banking industry. Finally, our findings show that capital flow liberalization has induced banks to close down branches, while increasing average branching costs. Our explanation is that, due to the larger integration of markets, banks envisage a greater future competition, and anticipate this effect by closing down branches. One appealing feature of our analysis is that we do not need accounting data, thus avoiding the problem of comparing accounting data taken from different banking systems. The reason is that, by using a two stage game to describe banking competition, we are able to build our estimation on a reduced form for bank profits, that is based on measures of the size of the market and on the number of branches for each single bank, but without activity volumes and interest rates. Furthermore, our econometric test is derived from a theoretical model. This allows us, first of all, to interpret our results on a stronger ground than other empirical analysis, and secondly, to derive some restrictions on the type of economic models that better describe bank competitive behavior in the retail segment. Our paper fits into the empirical literature on the European banking industry. There is a growing empirical literature on several aspects of the structure of this industry, connected to the changes in regulation at the EEC level. Chiappori, Perez-Castrillo and Verdier (1995) provide empirical evidence on the impact of deregulation on the banking industry. Although their empirical evidence rests on a model of spatial competition,3 they test its implications on a reduced form, while we follow a structural approach. Molyneux, Lloyd-Williams and Thornton (1994) and Neven and Roller (1995) estimate the degree of competitiveness of the European banking industry by using aggregate data, whereas we deal with disaggregate data at the bank level. Finally, few papers study the determinants of bank branching behavior, as for example Cabral and Majure (1993) and Barros (1995). They focus on the case of Portugal, while we are primarily interested in finding the underlying behavior of banks across countries. The next section illustrates the object of our analysis, namely the retail banking industry, the deregulation process as it is captured in the analysis and the way the sample was obtained. Then the theoretical model and the test of the model is presented in section 3, while section 4 is devoted to the econometric specification and the empirical analysis. Finally, section 5 concludes the paper.
3
2. The European retail banking industry In the banking industry, as in many other industries, there are two relevant segments: the retail and the wholesale market.4 In the retail segment, banks mainly collect deposits and lend funds to households and firms. In the wholesale segment, banks may in addition supply a large variety of financial services, as for example underwriting of new issues, intermediation of different financial activities, etc. In the two segments, banks differ also with respect to their size. In particular, large banks do tend to operate in the wholesale segment, where it is unlikely to find small banks. Here we focus on the retail market where the vast majority of banks operate, irrespective of their size, and where it is easier to model strategic interactions among banks. Banks, in the retail market, compete for clients by topping their rival interest rates and by locating their branches closer to their clients. In other words, banks can exploit the fact that their clients have preferences over locations, in order to mitigate interest rate competition and earn some market power by enlarging their branching network. The decision to open branches depends on the effect that opening new branches has on its own clientele and on rivals' response to it. Opening new branches, on one hand, augments the demand for services to the bank (expansive effect), but, on the other hand, it may cause a relocation of clients either from branches of rivals, or from preexistent branches of the same bank (competitive effect). Disregarding costs, the benefits of opening new branches depend very much on the relative strength of the expansive effect and the competitive effect. In a market where the competitive effect prevails, it is likely that banks will open few branches. However, opening new branches, independently of the size of their operations, has also a fixed cost, as for example the cost of hiring personnel, the cost of renting or buying facilities in that particular region, etc. A tougher price competition in the region reduces the expected profits of the new branch and therefore, the ability to recover the fixed costs. Then the bank will likely decide not to open new branches. In several countries, the ability to compete in interest rates has in fact been for long time impeded, in particular for small institutions, and the size of branching networks has 3
On models of spatial competition for the banking industry, see also de Palma and Uctum (1992). Bröker (1989) and Kaufman (1992) provide detailed pictures of structures and relevant strategic issues of the European banking industry.
4
4
become an important strategic variable for banks. It is now believed that, in many countries, banks own too many branches and that tougher competition in interest rates will reduce the average size of networks. These issues are particularly relevant for the European retail banking industry, given that financial markets integration is far from being achieved and that banking industries are still very different from one country to the other.
2.1 Deregulation in the European banking systems The European banking systems differ mainly for historical reasons. For example, the degree of competition or the onset of the deregulation process varies across countries. According to Gual and Neven (1992), Germany is one of the speediest in deregulating conduct, while Portugal is the slowest. Since not all these differences are captured by the institutional changes during the years analyzed in our sample, we let the degree of competition to be different from one country to the other in the empirical analysis below. In the eighties, the institutional framework in which banks were operating has changed drastically, as a consequence of the deregulation process that has affected both banking competition and the structure of markets, through the relaxation of entry barriers within and across markets. A description of the legal framework in which banks operate is out of the scope of this paper.5 However, to evaluate the impact of the changes on the conduct and structure of the banking industry, it is crucial to understand how these changes are captured in the analysis. Typically banking regulation is regrouped under three different categories, according to their objective: structural, conduct and prudential regulation. For instance, a regulation that affects the degree of competition is considered to be conduct regulation, while for example, controls on entrance of foreign banks or limitations to branching are part of structural regulations. Moreover, prudential regulation, whose main objective is to protect depositors or, in general, financial markets from the consequences of bank failures, has an impact on the structure of the banking industry through capital requirements. Regardless of its main objective, any type of regulation will have an impact on either conduct or market structure of the banking industry. Our objective is to pin down the changes in banking regulation on the 5
See Gual and Neven (1992) and ERE Summary Report (1996) for extremely helpful and complete guides on banking regulatory environment.
5
degree of competition, on the decision to open or close branches and finally on the decision to enter the market. Therefore, we consider changes in legislation that could directly affect conduct, branching and entry in the European retail banking industry.6
2.2 Our sample Our sample consists of annual data on banks in nine EEC countries, namely Belgium, Denmark, France, Germany, Greece, Italy, Netherlands, Portugal and Spain. The economic unit on which we focus is the single bank and not its single branch, because strategic decisions about entry and size of branching network are taken at the bank level. Therefore, banks that have been operating for at least two years between 1989 and 1993 in the nine countries are retained from the IBCA database. Since the relevant market is the retail banking market, only commercial banks are selected, with the exceptions of very few merchant banks (approximately 20 of more than 650 sampled banks) that collect customer deposits as indicator of their presence in the retail segment. Our sample contains between fifteen and fifty percent of the total population of banks by country. It is representative of the overall banking system in each country as it mimics fairly well its structure according to the average branching size in years 1990-1992. (See Table 1.) The average number of branches in the sample is comparable to the OECD figures for Belgium, Germany, and France, while, for the other countries, larger banks are slightly over sampled. However, because of missing data on the years at the extremes of the sample, we were not able to construct such a representative sample for the years 1989 and 1993. Very few banks in the sample entered new markets between 1989 and 1993. Entry in the industry would happen either when a new bank is created in a country, or when a previously existing bank starts operating in another country, by opening branches or by acquiring local banks. In this sense, entry and exit did not occur except for a very few banks that have been discarded because incomplete or imprecise data for what concerns the number of branches. (See in the Appendix for our data sources.) Thus, entry or exit from the sample is random and only due to the data collection process. In other terms, data attrition, relatively to the years 1989 and 1993, is exogenous.
6
Later, in section 4, we describe the specific changes in legislation accounted for in the empirical analysis.
6
Table 1 Structure of the European banking industry - Comparison between population and sample Average number of branches per bank (population) 1989 1990 Belgium 160 160 Denmark 15 15 France 42 43 Germany 47 46 Greece 26 31 Italy 39 43 Netherlands 47 44 Portugal 60 60 Spain 155 160
1991 143 22 55 47 31 47 45 72 165
1992 136 22 56 48 32 55 43 82 162
1993 na 21 54 46 36 60 41 75 161
Source: Our computation on OECD, Banks Profitability. Figures are rounded up.
Average number of branches (sample) 1989 1990 Belgium na 163 Denmark na 47 France 1 78 Germany 29 44 Greece na 112 Italy na 82 Netherlands na 30 Portugal 87 81 Spain na 499
1991 160 48 54 48 114 100 30 97 378
1992 157 46 51 46 98 101 91 94 358
1993 na 108 672 4 na 215 199 311 311
Note: The abbreviation na means that no data are available. Figures are rounded up.
Table 2 Unit versus multi-branch bank observations by country (all years) Unit bank Multi-branch bank Country Number Frequency (%) Number Frequency (%) Belgium 47 48 51 52 Denmark 12 11 96 89 France 190 58 139 42 Germany 181 37 306 63 Greece 53 7 39 93 Italy 45 12 392 88 Netherlands 12 27 33 73 Portugal 14 22 50 78 Spain 18 9 181 91 Total 522 29 1287 71
7
Moreover, to our knowledge, only eight banks of our sample have been involved in mergers over the period. This does not mean necessarily that there have been so few mergers during these years. Actually, in recent years, there has been a wave of mergers and acquisitions, although mainly inside borders, that has led to a consolidation of banking industries.7 However, on one hand, this information is not available to us and on the other hand, this phenomenon has occurred in the last years of the sample. To sum up, in principle one should be cautious as cross-borders mergers might affect the results when evaluating entry costs. A large proportion of banks (namely 522 over 1809 observations) has only one branch, presumably the headquarters of the bank. (See Table 2) Although these unit banks are more likely to operate in the wholesale segment where branches do not give any competitive edge on rivals, we consider them to be present also in the retail segment whenever collecting retail customer deposits. Indeed, the choice of being a unit bank is a signal of higher branching costs compared to other banks that concentrate their operations in the retail segment.
3. A model and a test of the industry structure In the retail segment, banking competition involves two dimensions, interest rates and network of branches. For what concerns interest rate, note that to get a loan from a bank, an account is needed first. Therefore, competition for deposits is crucial to capture clients and branching gives an edge over competitors. Hence, competition in the retail banking industry is a two-stage game where banks decide whether to enter the industry together with the size of their networks, taking into account that, in a latter stage, profits are determined by competition on interest rates. We skip the analysis of interest rate competition8, and focus instead on the entry and branching decisions.
3.1 The model The decision to open branches results from a cost-benefits trade-off. On one hand, opening new branches attracts new clients in the market (expansive effect). On the other hand, a new
7
See Gual and Neven (1992) and ERE Summary Report (1996) for detailed figures on the number of cross-borders vs. inside borders mergers and acquisitions in the last years. 8 See Cerasi (1995) for a model where banks compete both in branching and interest rates, that delivers a profit function with the same properties as the one in this paper.
8
branch steals business to old branches (competitive effect), attracting clients either from competitors or from preexistent branches of the same bank. Not taking costs into account, the benefits of opening new branches result from the balance between expansive and competitive effect. This intuition is translated by imposing that, given the outcome of the interest rate competition, overall bank profits increase in the number of branches of the bank, but at a slower pace, as the size of the branching network increases. Further, the benefits of opening a new branch in a market crowded by competitors must be smaller. In other words, overall bank profits must decrease with the number of branches of other competitors. Denote by S k i N i the overall profits of bank i, assuming that it is a concave function in the number of branches of the bank, k i and a decreasing function of the number of branches of rivals N i . This profit function is a reduced form of the profits of the second stage of the game, as it implies equilibrium in the interest rate competition game. A convenient specification for the overall bank profits is given by S k i N i
Sk ic k i N i
Sk ic N
,
(1)
where S is the size of the market, N the overall number of branches in the market, i.e. N
¦ k , while c is a non-negative parameter that measures the elasticity of profits to i
i
branching.9 The special feature of this form is that each bank profits are proportional to its share of branches in the industry and moreover, that the impact of branching on profits can be measured by the parameter c. This form is general enough as it allows testing both for the nature of competition in branching and for the degree of toughness of interest rates competition. On one hand, for low values of c, branching competition can be represented by a model of spatial competition, where, at a given market size, the only way to capture new clients is by stealing clients to other branches, while for larger values of c, the expansive effect becomes important and branching competition may be better captured with other models of monopolistic competition10. On the other hand, low values of c imply that profits are less sensitive to changes in branching and this 9
From the profit function in equation (1), one could see that the elasticity of profits to its own branches
k i converges to c as N goes to infinity.
9
may be due to tougher price competition. As a matter of fact, tougher competition in interest rates reduces profits in the second stage as well as marginal benefits of a new branch. Opening new branches is a costly decision, independently of the level of activity of the new branch, as it involves fixed costs. For each bank, the cost of entering the market with a first branch is V i , while opening each additional branch costs H i . These opportunity costs of staying in the market and keeping the network operating are sunk in the sense that they cannot be recovered if the bank decides to exit the market or close down branches. Thus the sunk costs, for a given network size ki are: F ki
Vi
Hi k i 1 .
(2)
Bank i chooses its network size ki in order to maximize net profits, S k i N i F k i , taking as given other banks decision, i.e., total network size of rivals N-i. The optimal network size, which characterizes the Nash equilibrium in branches, must satisfy the inequality: w w ki
Sk ic 1 § ki · ¨c ¸ d Hi . 2N ¹ N ©
S k i N i
(3)
For a given branching cost, condition (3) states that the bank faces the following choice: (i) either to open several branches and therefore, the condition is fulfilled with the equality sign; (ii) or to enter the market with only one branch, and therefore, the condition is fulfilled with the inequality sign. In other terms, the bank chooses to operate only with its headquarters whenever the marginal benefit lies below the branching cost H i . When the latter is low enough for the condition to be fulfilled with equality, the bank opens additional branches up to the point where the marginal benefit equals the marginal branching cost. Note that condition (3) is valid when the number of branches is a continuous choice variable. The equivalent condition in a discrete setting is much more cumbersome to derive. Now, the bank enters the market whenever post-entry profits are high enough to recover sunk costs, namely: S k i N i
S k ic N
t V i Hi k i 1 .
(4)
This entry condition gives, for a given number of banks in the market, the minimum profit that allows recovering the sunk costs. 10
In Cerasi (1995) branching is modeled in a different setting with respect to spatial competition. In particular the preferences for branching replicate those for a firm that offer several products allowing
10
Note that condition (3) and (4) are based on the reduced form of profits of the second stage. Therefore, they do not depend upon interest rates and volumes. The reduced form will be directly estimated in the econometric model
3.2 The industry structure Assume that, in each period t, bank i is in equilibrium. That is, at the beginning of each period, the bank is given a new set of costs H it , V it and decides whether to stay in the market and with how many branches. In other words, in each period, for a bank, that is still present in the market, both conditions, (3) and (4), should hold. In particular, the equilibrium number of branches at time t for bank i, k it , defined as the number of branches in the previous period k i ,t 1 plus the change in the number of branches from one period to the other, namely 'k it , must satisfy ki t 'kit · St kitc 1 § ¨c , 1 ¸ d H it , 2 Nt Nt © ¹
(3’)
The non-negative profit condition can be written as: S t k itc Nt
Hit k it 1 t V it .
(4’)
Define Ait {
S it k itc 1 §
k · ¨¨ c i ,t 1 ¸¸ . 2Nt ¹ Nt ©
(5)
Considering only multi-branch banks at time t, i.e., k it ! 1, for which condition (3’) is fulfilled with the equality sign, one can distinguish between two different cases. Either Ait t Hit 'kit t 0 ,
(6a)
Ait Hit 'kit 0 .
(6b)
or
In other terms, when a bank is in the market with more than one branch and has opened (closed) branches since the previous period, it implies that the branching cost is lower (higher) than the quantity in the LHS of inequality (6a) ((6b), respectively).
for a larger benefit of opening new branches than in a model of spatial competition.
11
The model allows distinguishing between unit and multi-branch banks. For unit banks, condition (3’) holds as a strict inequality. In addition, 'k it
0 by definition, as a unit bank
owns only one branch for at least two successive periods. Therefore, unit banks satisfy condition (6b). Summing up, the equilibrium condition allows us to allocate bank i into one of two classes, according to whether it is a multi-branch bank that has not shrunk, i.e., when 'k it t 0 and k it ! 1, and thus Ait t H it , or whether it is a multi-branch bank that has shrunk, i.e., when
'k it 0 and k it ! 1, or it is a unit bank, and therefore Ait H it . Hence, define two sets E1t =^ i such that 'k it t 0 and k it ! 1 `,
(7a)
and E 2 t =^ i such that 'k it 0 and k it ! 1 , or k it with E1t E 2 t
and E1t E 2 t
1 `,
(7b)
E t and E t is the set of banks that are in the market at
time t.
3.3 The test Assume that the costs Hit and V it are randomly drawn from two independent density functions f H and f V , with cumulative functions FH and FV , respectively. For each observation in the sample, condition (6a) and (6b), corrected for the presence of unit banks, and cast in probabilistic terms, are: Pr[ E1t ]
FH Ait
(8a)
Pr[ E 2 t ]
1 FH Ait .
(8b)
The probability that the non-negative profit condition (4’) holds is:
§ S t k itc · H it (k it 1) ¸ . ¨ Nt ¸ © ¹
FV ¨
(9)
The joint probability of the network equilibrium condition and the non-negative profit condition is the product of marginal probabilities, given the assumption of independence of the two stochastic costs. At this point, we must stress that the assumption of a continuous choice for the number of branches made in the theoretical model is neutral for the econometric model, as we are only concerned by the conditional distribution of entry and branching cost on the number of branches. These distributions have to be continuous in any case.
12
Define Bit {
S t k itc Nt
and Cit {
Bt k it 1
. According to the theoretical model, when the
restrictions on the parameters hold, in particular when the profit function is concave, it follows that Ait Cit always. We can draw the relevant area for the joint probability in the space H , V .
On Figure 1, the probability that a bank has opened new branches in the last period is
defined over the white area, while the striped area defines the probability space for a bank that has closed its branches.
Figure 1 Theoretical probabilistic distribution of the industry structure V
B
A
C
H
Note that the condition of non-negative profits must hold for all observations in the sample when no entry or exit (in an economic sense) occurs. Indeed banks are included or excluded from our sample randomly, but they presumably existed before the starting year and after the ending year of this study. So, for our sample, the marginal probability implicit in (9) is always equal to 1. Since we do not observe potential entrants for whom V did not fulfill condition (9) or firms that exited the market, we are not able to characterize the distribution of entry costs. The only possible strategy is to estimate the branching cost H , and then, using condition (9), we can derive, conditionally on H a mean upper bound for V . To clarify, an observation is now a bank in a market (country) at a given time. Based on the partition of observations obtained from the theoretical model, the likelihood function for our sample can be easily written. From (8a) and (8b) and the assumption of independence of H and
13
V both through time, across banks and across countries, conditionally on bank and market
specific variables, the likelihood function is:
° ½° 1 ln F ( A ) ln F ( A ) ® ijt ijt ¾ H H °¿ 1° i E ¯i E
¦¦ ¦ T
ln L
t 1 j
J
1 jt
¦
(10)
2 jt
where the subscripts stand for bank i, market (country) j, time t. Tables 3 provide some statistics on the partition of observations in our sample.
Table 3 Partition of observations (all countries and years)
Unit banks and branching banks with 'k 0 Of which: Unit Banks Branching banks with 'k 0 Branching banks with 'k t 0
Number of observations 640 522 118 1169
Average Average 'k k 56 -6 1 299 -30 140 +6
4. Empirical analysis We turn now to the econometric specification, the statistical significance of our model and the discussion of the empirical results.
4.1 Stochastic specification Given that we need a density function that assigns probability on a non-negative support, one solution is to assume that the density of the branching cost H is a negative exponential distribution, with a non-negative parameter OH . Another solution is the lognormal distribution, with parameters PH and T H . In this latter case, given our probit-type log-likelihood function, it is necessary to normalize one parameter, in particular TH
1.
The objective of the econometric analysis is to estimate the parameters, c and, respectively,
OH or PH , conditionally on firm specific, market specific and regulatory variables that may affect the distribution of the branching costs. In other words, we assume that the distribution of the branching costs depends upon a set of exogenous variables, specifically either
OH
exp E x ,
or
14
PH
J x,
with x defining the vector of exogenous variables affecting the levels of branching costs. Similarly, we posit c
exp G z ,
where z defines the vector of exogenous variables that could influence the degree of competition in the market.
4.2 Set of exogenous variables
i) Deregulation variables. A set of dummy variables indicating when particular regulatory rules, affecting conduct, branching and entry decisions, have been implemented in the different countries are defined in Table 4. There is often a lag between the time in which a particular regulation is passed at the EEC level and the moment of its implementation at the national level. We take into account this lag by considering, for each regulation, the year of implementation in a specific country.11 We focus on three important measures. The first one, capital flow liberalization (CFL), is the set of policies whose objective was to ease the flow of financial services across the EEC and, as a consequence, to stimulate banking business across countries. (See Art.67, EEC Treaty.) We expect an increase in the demand for banking services all over the EEC and more mobility of banks as a consequence of higher capital mobility. On one hand, this may affect the expectations on the future competition, as banks envisage entry by new competitors in the market. On the other hand, entry costs will be reduced as a consequence of the greater integration of financial markets. The second rule, interest rate deregulation (IRD), is the set of laws and policies, at the national level, aimed at increasing interest rate competition among banks. We expect to be able to see two effects. On one hand, an increase in the degree of competition, namely a reduction in the parameter c, which captures the inverse of the degree of interest rate competition; on the other hand, since bank profits reduce as a result of tougher interest rate competition, we expect a shrinking in the branching networks as it will be harder to recover the fixed costs of a large network.
11
We took this information from the ERE Summary Report (1996).
15
Table 4 Indicators of regulatory changes Capital flow liberalization, CFL (Art.67, EEC Treaty) 1989 1990 1991 1992 Belgium 0 1 1 1 Denmark 1 1 1 1 France 0 1 1 1 Germany 1 1 1 1 Greece 0 0 0 0 Italy 0 1 1 1 Netherlands 1 1 1 1 Portugal 0 0 0 0 Spain 0 0 0 0
1993 1 1 1 1 0 1 1 1 1
Interest rate deregulation, IRD 1989 1990 Belgium 0 0 Denmark 1 1 France 0 1 Germany 1 1 Greece 0 0 Italy 1 1 Netherlands 1 1 Portugal 0 0 Spain 1 1
1991 1 1 1 1 0 1 1 0 1
1992 1 1 1 1 0 1 1 1 1
1993 1 1 1 1 1 1 1 1 1
Second European directive, SED (1989, n.646) 1989 1990 1991 Belgium 0 0 0 Denmark 0 0 1 France 0 0 0 Germany 0 0 0 Greece 0 0 0 Italy 0 0 0 Netherlands 0 0 0 Portugal 0 0 0 Spain 0 0 0
1992 0 1 0 1 1 0 1 1 0
1993 0 1 1 1 1 1 1 1 0
16
Finally, the third dummy refers to the second European directive 1989/n.646 (SED). This directive has introduced the principle of single license for banks, which must facilitate entrance by foreign banks across EEC markets. Furthermore this directive has shifted the capital requirement for a bank that enters a new market from each single branch to the bank level, and thus we expect to see a reduction in the cost of opening new branches across countries.
ii) Inputs costs. Costs may affect the choice of opening a new branch, for example the cost of labor and the interest rate on real capital, as they represent the fixed factors in operating a branch. On the one hand, the labor cost is an important component of fixed costs in branching, as personnel is needed to run a branch and must be paid independently of the level of activity of that particular branch. More specifically, the unit labor cost in a particular region (country) must affect the decision as to whether to open a new branch in that region. Therefore, we look at the way in which the real wage in the banking sector (Wage) affects the average cost of branching, expecting a positive relationship. On the other hand, opening a branch involves renting or buying facilities in a particular location, i.e., an investment in fixed capital. We consider this effect by taking the real interest rates (Rirate) as a proxy for the cost of investing in fixed capital. The average cost of branching should increase if the opportunity cost of investing in fixed capital increases. Unfortunately this proxy may pin down also other effects, as for example the costs of interbank funds, and therefore the results should be carefully interpreted. If the real interest rate were to capture the cost of funds in the interbank market, we should expect a negative relationship with the cost of branching, as a rise in the interbank rate increases the convenience of collecting funds through branches.
iii) Market and network variables. It remains to measure the market size, S, the network size, k, and the overall size of networks, N. For each country, total deposits are used as a proxy for S, while N is the total number of branches. Then, for each bank included in the sample, we have a measure of the number of branches, k. Table 5 provides the average values for the exogenous variables included in the econometric model.
17
Table 5 Average sample values on main variables S N Belgium 1.995 17245.8 Denmark 0.755 2608.2 France 7.033 25454.8 Germany 9.178 39423.2 Greece 0.329 1124.5 Italy 5.990 17453.7 Netherlands 2.248 7669.0 Portugal 0.441 2395.1 Spain 3.449 35296.0 Note: S N k Wage Rirate
= = = = =
k 160 55 79 45 107 97 130 94 359
Wage 6.18 4.45 5.55 2.67 2.85 7.34 5.33 2.45 5.43
Rirate 6.31 7.30 6.19 4.93 -0.18 6.73 5.27 5.18 6.00
National deposits, 10 millions US$, (constant prices) Total number of branches Number of branches per bank Real wage rate in the banking sector, 10 thousands of US$ Real interest rate
4.3 Goodness-of fit The maximization of both log-likelihood functions associated with the exponential and lognormal densities is performed by a non-linear optimization routine provided by the statistical software package GAUSS. In terms of the theory of Pseudo Maximum Likelihood (see Gouriéroux and Jouneau, 1993), given that both densities belong to the exponential family, parameters are consistently estimated. However, the choice of densities is arbitrary and involves a potential problem of mis-specification. Hence, correct standard errors must be obtained by computing the heteroskedastic-consistent covariance matrix. Results obtained from the two densities are comparable in terms of goodness-of-fit, with a slightly better fit for the lognormal model that produces a higher value for the likelihood function. The ability to predict of our model is gauged in Tables 6a and 6b which compares the observed and estimated partition of observations for the two distributions. The model predicts correctly the branching behavior for 80 percent of banks in the sample. Note that the incorrectly predicted observations are located outside the diagonals of Tables 6a and 6b. The average branching size of banks that have opened branches but that the model places into the group of banks that have closed branches for the lognormal case (see the southwest cell of Table 6b) is 15. On the other hand, the average branching size of banks that have closed branches and that are placed by the model into the group of banks that have opened branches (see the north-east cell of Table 6b) is 397. Hence, the model predicts that large banks are more likely to open new branches compared to small banks. Moreover, the model
18
overestimates the number of banks that have closed branches, namely more than 850 instead of 640 observed cases of this type. Thus, although the model is well behaved, it may underestimate factors explaining why banks open new branches.
Table 6a Observed and estimated frequencies (exponential density) Observation 'kt0 'k < 0 or k = 1 Estimation 881 (182) 73 (404) 'k t 0 288 (14) 567 (11) 'k < 0 or k = 1 Total 1169 640 Goodness-of-fit index = 80.04 % Table 6b Observed and estimated frequencies (log-normal density) Observation 'kt0 'k < 0 or k = 1 Estimation 882 (181) 73 (397) 'k t 0 287 (15) 566 (11) 'k < 0 or k = 1 Total 1169 640 Goodness-of-fit index = 80.00 %
Total 954 855 1809
Total 956 853 1809
Note: In each cell, the figures provide the number of banks that fall in the corresponding case. The average branching size for these banks is in brackets.
4.4 Assessment of results Parameter estimates as well as t-ratios are gathered in Table 7. Many parameters are significant at the five-percent level and the significance of most parameters is of the same order in both experiments. The estimates reported in Table 7 correspond to the best, according to a cautious selection process, combination among the variables that explain the levels of the two parameters of the model, the degree of competition and the parameter of the distribution of branching costs. The coefficients in the upper part of Table 7 measure the impact of exogenous variables on c , the elasticity of profits to branching, or, alternatively, the inverse of the degree of toughness of competition. A negative sign implies an increase in the degree of competition, or, alternatively, a reduction in the elasticity of profits to branching. We can also interpret a negative sign as inducing a lower probability that banks open new branches, namely a reduction in the average size of branching networks. The reason is that a smaller c , implies a
19
reduction in the quantity Ait , in equation (6a) and (6b), which is the upper bound of the area defining the probability of opening new branches. As for the parameter OH , it is the inverse of the mean of the branching cost for the negative exponential distribution, and note that an increase in the mean implies giving higher probability to larger values of H . Therefore, a coefficient associated with a variable affecting OH , provides the magnitude of its effect on the inverse of the average cost of branching and, as a consequence, its direct impact on the probability of opening new branches. For the lognormal density, the effect of a variable on the average cost of branching P H is directly obtained from the associated coefficient. Given this remark, note that the two sets of coefficients are perfectly comparable in terms of signs. For the sake of simplicity, only results obtained from the lognormal distribution are discussed below.
Table 7 Estimation results Effect on the elasticity of profits to branches, parameter c Dummy Belgium Dummy Denmark Dummy France Dummy Germany Dummy Greece Dummy Italy Dummy Netherlands Dummy Portugal Dummy Spain Time trend IRD CFL Effect on the parameter of distribution of branching cost H Constant Time trend CFL SED Wage Rirate Mean log-likelihood
Exponential distribution Parameter t-ratio
Log-normal distribution Parameter t-ratio
0.34 0.48 0.44 0.45 0.75 0.49 0.47 0.95 0.52 -0.09 0.01 0.18 Parameter
-3.8 0.2 2.9 t-ratio
-0.11 0.02 0.19 Parameter
3.0 2.7 2.9 2.8 7.0 3.4 3.1 9.0 3.6 -4.1 0.1 2.4 t-ratio
1.72 0.28 -1.45 0.13 0.08 0.04
3.2 2.7 -5.4 0.7 1.3 0.7
-1.17 -0.37 1.33 -0.19 -0.04 -0.12
-2.1 -3.8 4.8 -1.1 -0.7 -1.9
-0.409
20
4.1 4.7 4.4 4.5 6.8 4.8 4.4 8.9 5.3
0.48 0.57 0.57 0.59 0.98 0.67 0.62 1.10
0.66
-0.379
First of all, the country dummies are highly significant. Thus, the starting conditions of price competition in the various countries were different before 1989 and must be taken into consideration. Alternative specifications in which the country dummies were in OH (or in P H ) rather than in c did not provide a better model in terms of likelihood ratio. So, country specificities are more effective in capturing differences in competitive framework than in the average branching costs. Note that the time trend for the parameter c is negative, so that competition becomes fiercer, as time goes. Note that this implies that the efficient size of branching networks decreases and it cannot be seen directly from descriptive statistics since the average number of branches per bank would indicate a positive trend. (See Table 1.) The effect of increasing toughness of interest rate competition does not seem to be induced directly by the changes in regulatory rules. Indeed, the coefficient of interest rate deregulation (dummy IRD) is not significant, while capital flow liberalization (dummy CFL) has a positive sign. On one hand, the time trend probably captures the effect of changes in regulation at the country level more than at the European level. On the other hand, capital flow liberalization has certainly increased the potential size of the market and hence, expected profits and marginal benefits of branching for banks, as we will discuss below. Consider now the determinants of average values of branching costs. Again, the sign of the time trend indicates that costs have decreased over time and therefore the average branching size of banks tends to be smaller. Note that this effect cannot be seen directly from descriptive statistics since the average number of branches per bank would indicate a positive trend. (See Table 1) Further, the coefficient for the real wage (Wage) has a sign opposite to what it is expected. In fact, given that a wage rate is not available at the bank level and that we consider here the real wage for the whole banking industry in each country, this variable plays the role of a trend.12 Here again, the effects of changes in the regulatory environment are disappointing. First, the dummy indicating the effectiveness of the second European directive (SED) has the expected sign but is not significant. Second, capital flow liberalization (dummy CFL) has a counterintuitive effect on branching costs. However, this liberalization could have forced banks to increase their network size in order to be present in new markets. Finally, the coefficient of the real interest rate (Rirate) is not significant and negative, which may indicate that the real 12
Indeed, in estimation (not reported) where the time trend is removed, nothing changes except that the real wage coefficient becomes significant and keeps its sign.
21
interest rate on public bonds is more than a proxy for the cost of fixed capital, the opportunity cost of raising funds through branches. For what concerns the impact of deregulation on the structure of the banking industry, our results are can be summarized in what follows. On one hand, interest rate deregulation and the second European directive did not have a significant effect. It may be that these new rules have not had a direct effect, although have induced a new context of competition in each country which is captured by the significant effect of trends introduced in the econometric model. On the other hand, capital flow liberalization has played a strong role by increasing average branching costs. One possible explanation is that banks, in response to this new rule, have envisaged a greater future competition due to the larger integration of markets and rise in demand for banking services, and therefore, anticipated this effect by closing down branches. If, however, there is an asymmetry between closing and opening branches, in the sense that banks are slower in closing their existent branches rather than in opening new branches, they will tend to keep branches open, even if they operate at a level below efficiency. The average cost of existent branches rises as a consequence of this asymmetry. A complementary explanation is that our sample is representing mainly banks that have not directly benefited from the process of integration. By not opening new branches in foreign markets, they are not exploiting the benefit of deregulation. On the contrary, they pay the costs of deregulation by having to face fiercer future competition in their domestic markets.
4.5 Costs and competitiveness across countries We discuss now the implications of our results for the degree of competition, for the choice of branching in the different countries and for the impact of deregulation on the banking industry. Together with the values of c , figures on profits, marginal benefits of branching, estimated branching costs and upper bounds for entry costs by country are shown in Table 8. While the absolute values of these estimated variables may not be entirely realistic, the ordering of the countries in terms of degree of competition and branching behavior they imply is informative and close to observed facts. Concerning the strategic interactions among banks in the retail banking industry, note that the average estimated value for the parameter c is 1.64. So the expansive effect of branching is quite important and thus, we would overestimate the competitive effect in choosing a pure model of spatial competition to analyze the branching decision. A model of monopolistic
22
competition turns out to be a better modeling choice at least for our sample. In addition, a value of c larger than one indicates that the response of profits to branching is considerable and thus, that branching still gives banks in the retail segment an edge over competitors. This is also confirmed by the fact that, on average, the estimated average marginal benefit of opening an additional branch is 0.64 percent of the overall profits of a bank. Countries can be ordered according to the degree of competition from the estimated average values of c . Figure 2 shows that Portugal is still behind all other countries for the low level of price competition, with a value of c larger than 2. Germany, France and the Netherlands have almost similar competitive conditions. The estimated average branching costs are quite different across countries. (See Figure 3) Greece, Portugal and Spain have relatively low costs of branching, indicating that there may still be scope for increasing the network of branches in these countries, whereas for the others and specially Germany the opposite is true, with Denmark in the middle. Finally, consider the estimated values for the upper bound of entry costs in the different countries. First, there is a clear cut between southern and northern countries, between high and low entry cost countries. (See Figure 3.) Second, although these upper bounds should not be taken literally, they do represent the average estimated profitability of the banking industry. Our ranking (see Figure 4) is fairly comparable to one provided in a confidential appendix of the ERE Summary report (1996) based on average returns on assets.
23
Table 8 Estimates of profit elasticity, profits, marginal benefits, branching and entry costs Average Average Upper bound Average Marginal value of H value of c for V Profit benefit Belgium Denmark France Germany Greece Italy Netherlands Portugal Spain Mean
1.41 1.51 1.55 1.58 1.89 1.73 1.52 2.22 1.36 1.64
Note: Units for H and
0.23 0.17 0.24 0.29 0.14 0.22 0.20 0.10 0.07 0.18
V are 10 millions of US$.
17.02 4.27 318.29 77.21 261.21 344.12 38.23 756.73 86.44 211.50
54.03 13.15 335.18 89.94 276.67 364.11 57.39 765.48 112.37 229.81
0.11 0.16 0.48 0.58 1.93 2.54 0.31 11.23 0.19 1.95
Figure 2: Increasing degree of competition across countries Portugal Greece Italy Germany France Netherlands Denmark Belgium Spain 0,00
0,20
0,40
0,60
Degree of competition
24
0,80
Marginal benefit (% Profits) 0.20 1.21 0.14 0.64 0.70 0.70 0.54 1.47 0.17 0.64
Figure 3: Location of European countries in the entry-cost / branching-cost space 1000
Portugal
Italy France
Entry cost
Greece Spain
100
Germany
Netherlands Belgium
10 Denmark
1 0
0,05
0,1
0,15
0,2
0,25
0,3
Branching cost Note: The scale on the vertical axis is logarithmic.
Figure 4: Profitability indexes
100
10
Note: The scale on the vertical axis is logarithmic.
25
Portugal
Italy
France
Greece
Spain
Germany
Netherlands
Belgium
1 Denmark
Profitability
1000
5. Conclusion The main feature of our simple econometric analysis is to treat branching and entry costs as latent variables. Based on this idea, a model of monopolistic competition, fitted on microeconomics data, provides a fair approximation of the conduct of European banks. The analysis also produces a series of useful measures like the degree of competition, the level of economic profits and marginal benefits of opening branches. Two results are noticeable from the previous discussion. First, deregulation has certainly increased the degree of competition within the European retail banking industry. However, it is hard to identify the actual impact of rules passed at the European level, that is, how these rules have effectively changed the working of the market, while idiosyncrasies in the deregulation process at the country level play a prominent role. Second, there are differences between the south and the north of Europe. Northern banks face lower entry costs but higher branching costs, while the reverse is true for the south. Further, profitability is higher while competitiveness is lower in the south, than in the north. In other terms, whenever entry costs are low, competition is tougher and profitability is low. Provided that data are available on mergers and acquisitions, on entry to and exit from the industry, the level rather than an upper bound of entry costs could be estimated. Together with a more dynamic setting, we might be able to improve our results in this line of investigation. These tasks are part of our research agenda.
26
Appendix Sources of data IBCA database: Name of operating banks in each year. The Bankers Almanac, several years: Number of branches for each single bank, k. OECD, Bank Profitability, last issue: Total number of branches for each country (overall banking industry), N, Labor costs in the banking sector, Wage.13 IFS Yearbook, several years: Total deposits (demand and time deposits) for each country, S, Real interest rates, Rirate. ERE Summary Report: Dummy for implementation of interest rate deregulation, IRD, Dummy for implementation of capital control lifting, CFL, Dummy for implementation of Second EEC Directive, SED.
13
Since for Germany the OECD data was not available, we used a wider definition, i.e., unit labor costs in financial services, taken from another statistical source.
27
References Barros, P.P., 1995, Post-entry expansion in banking: The case of Portugal, International Journal of Industrial Organization, 13, 593-611. Bröker, G., 1989, Competition in Banking (OECD, Paris). Cabral, L. and R. Majure, 1993, A model of branching with an application to Portuguese Banking, Working Paper WP 5-93 (Banco du Portugal, Lisbon). Cerasi, V., 1995, A model of retail banking competition, mimeo (London School of Economics). Cerasi, V., 1996, An empirical analysis of banking concentration, Quaderni del Centro Baffi, January No105 (Bocconi University, Milano). Chiappori, P.A., D. Perez-Castrillo and T. Verdier, 1995, Spatial competition in the banking system: Localization, cross subsidies and the regulation of deposit rates, European Economic Review, 39, 889-918. De Palma, A. and M. Uctum, 1992, Financial intermediation under financial integration and deregulation, NER 19 (Banque de France, Paris). Economic Research Europe Ltd., 1996, A study of the effectiveness and impact of Internal Market Integration on the banking and credit sector: A Summary Report. Gouriéroux, C. and F. Jouneau, 1993, Pseudo Likelihood Methods, in: G.S. Maddala, C.R. Rao and H.Vinod, eds., Handbook of statistics (North-Holland, Amsterdam), 335-362. Gual, J. and D. Neven, 1992, Deregulation of the European Banking Industry (1980-1991), CEPR Discussion Papers, No 703 (CEPR, London). Kaufman, G., ed., 1992, Banking Structures in Major Countries (Kluwer Academic Publishers, Boston). Molyneux, P., D.M. Lloyd-Williams and J. Thornton, 1994, Competitive conditions in European Banking, Journal of Banking and Finance, 18, 445-459. Neven, D., 1989, Structural adjustment in European retail banking. Some views from industrial organization, CEPR Discussion Papers, No 311 (CEPR, London). Neven, D. and L.H. Roller, 1995, Competition in the European Banking Industry: An Aggregate Structural Model of Competition, mimeo (Wissenschaftszentrum Berlin).
28
