Econometric Research and Special Studies Department
The profit-structure relationship, efficiency and mergers in the European
banking industry: an empirical assessment
L.W. Punt and M.C.J. van Rooij
Research Memorandum WO&E no. 604
December 1999
'H1HGHUODQGVFKH%DQN
THE PROFIT-STRUCTURE RELATIONSHIP, EFFICIENCY AND MERGERS IN THE
EUROPEAN BANKING INDUSTRY: an empirical assessment *
L.W. Punt ** and M.C.J. van Rooij
* The authors are grateful to Peter Vlaar and Charles van Marrewijk for helpful comments and
remarks.
** Erasmus University Rotterdam, Department apprentice from April till September 1999.
Research Memorandum WO&E no. 604/9937
December 1999
De Nederlandsche Bank NV
Econometric Research and
Special Studies Department
P.O. Box 98
1000 AB AMSTERDAM
The Netherlands
ABSTRACT
The profit-structure relationship, efficiency and mergers in the European banking industry: an
empirical assessment
L.W. Punt and M.C.J. van Rooij
An empirical investigation of the relationship between market share or concentration and return on
equity or assets provides evidence for the existence of a profit-structure relationship in the European
banking sector. Testing several market-power and efficient-structure theories reveals that X–efficiency
is the crucial factor explaining the profit-structure relationship because it stimulates both profitability
and market share. Bank mergers in recent years have been successful because, on average, Xefficiency and profitability have improved after the consolidation. Moreover, there are no indications
of unfavourable price setting behaviour as a result of increased market power.
Key words: Profit-structure relationship, bank efficiency, bank mergers
JEL codes: G14, G21, G34, L11
SAMENVATTING
De winst-marktstructuur relatie, efficiëntie en fusies in het Europese bankwezen: een empirische
evaluatie
L.W. Punt en M.C.J. van Rooij
Een empirische evaluatie van de relatie tussen marktaandeel of concentratie enerzijds en rendement op
eigen vermogen of activa anderzijds duidt op het bestaan van een verband tussen winstgevendheid en
marktstructuur
in
het
Europese
bankwezen.
Toetsen
op
verschillende
marktmacht-
en
efficiëntietheorieën tonen aan dat X-efficiënties het bestaan van deze relatie verklaren doordat deze
zowel de winstgevendheid als het marktaandeel van banken bevorderen. Bankfusies in de jaren ’90
blijken succesvol te zijn geweest in die zin dat, gemiddeld genomen, zowel X-efficiëntie als
winstgevendheid na een bankfusie zijn verbeterd. Er zijn geen aanwijzingen voor prijszettinggedrag
ten koste van consumenten als gevolg van toegenomen marktmacht.
Trefwoorden: Winst-marktstructuur relatie, efficiëntie, bankenfusies
JEL codes: G14, G21, G34, L11
-1-
1 INTRODUCTION
These days, the banking sector is characterised by an ongoing wave of mergers and acquisitions. This
urge for expansion is attended by rapid changes in the market structure of the banking industry and
raises questions about the desirability of these developments from the point of view of the customer.
Empirical studies based on American data often find a positive relationship between concentration or
market share on the one hand and profitability on the other. Several explanations for the existence of
this profit-structure relationship have been brought up in the literature. Broadly speaking an increase
in profits is attributed to efficiency improvements or to a rise of market power. The implications for
merger and antitrust policy, however, are radically different under efficient-structure hypotheses vis-àvis market-power theories. Under the former hypotheses market concentration is motivated by an
increase in efficiency – which would yield an increase in total surplus – while under the latter mergers
are motivated by increasing market power – yielding a reduction in total surplus. Mergers should be
encouraged in the first case, while in the second situation they should not be sustained.
The profit-structure relationship is a frequently explored topic within industrial as well as financial
economics. However, most studies that examine the profit-structure relationship for the banking sector
use American data. In this study the focus is on Europe and we examine the validity of different
explanations for the profit-structure relationship. Furthermore, the effects of mergers on bank
performance are assessed empirically and combined with the empirical evidence on the profit-structure
relationship to provide further insight in the economic rationale for banks to merge. The outline of this
paper is as follows. Section 2 contains a brief discussion on the existence of the profit-structure
relationship and possible explanations that have been brought up, including some results reported in
the empirical literature. Section 3 investigates the existence of a profit-structure relationship in the
European commercial banking industry. Different hypotheses explaining the existence of the profitstructure relationship are tested in section 4. To this end efficiency levels for individual banks are
estimated and discussed. Thereafter, section 5 studies the implications of bank mergers for
profitability and the economic rationale of mergers against the background of the profit-structure
relationship and its explanations. Finally, section 6 concludes with a summary.
-2-
2 THE PROFIT-STRUCTURE RELATIONSHIP AND POSSIBLE EXPLANATIONS
Within the existing literature on the profit-structure relationship, most studies find empirical support
for the assumed positive relationship between bank profitability and a measure of market structure,
such as market share or some other measure of concentration (e.g. concentration ratios or a Herfindahl
index), although this evidence is weak at times 1. Studies that do find evidence for this relationship are
for example Altunbas & Molyneux (1994), and Berger (1995), whereas e.g. Berger & Hannan (1998)
provide no empirical support. Because overall previous research, mostly based on American data, is
supportive on the existence of the profit-structure relationship, it is expected that data on the European
commercial banking industry will show proof of the existence of this relationship as well. This
proposition is investigated empirically in section 3. The empirical evidence on the profit-structure
relationship reported in previous studies has given rise to several theoretical hypotheses, which are
used to underpin the phenomena found. The present section discusses propositions about the possible
determinants of a profit-structure relationship. These hypotheses can be divided into two categories,
the market-power hypotheses and the efficient-structure hypotheses, discussed in section 2.1 and 2.2,
respectively. The empirical evidence on these propositions is reviewed in section 2.3.
2.1 Market-power hypotheses
In explaining a profit-structure relationship, the market-power (MP) hypotheses state that market
power is the main variable that causes profitability to change. In concentrated markets, market
imperfections are present. These imperfections can be the result of collusion, facilitated by high
concentration, or by (legislative) entry and exit barriers, which are often present in banking, since
regulation is strict. Because of these imperfections, firms operate in a market that deviates from
perfect competition. This enables them to exert influence on prices charged and/or paid. Through price
setting firms can achieve higher profits at the expense of their customers. Which market structure
variable is the best proxy for market power and thus for market imperfections determines the
difference between two main types of market-power hypotheses in this category: the structureconduct-performance hypothesis and the relative-market-power hypothesis.
1 For a detailed summary of the literature, see e.g. Berger & Humphrey (1997) or Berger, Hunter & Timme (1993).
-3-
The structure-conduct-performance (SCP) hypothesis assumes that market concentration is the best
proxy for market power, and that more concentrated markets show larger market imperfections, which
enables firms to set prices at levels that are less favourable to customers. Through market-wide price
setting, each individual firm is able to realise higher profits.
The relative-market-power (RMP) hypothesis asserts that only firms with large market shares have the
power to set prices and thus to earn supernormal profits. In this case there is no market-wide price
setting, but only price setting by dominant firms with large market shares. Firms with smaller market
shares are forced to operate as if under perfect competition and are unable to earn the same
supernormal profits. This implies that the firm-specific market share is the better proxy for market
power and market imperfections.
A third additional hypothesis, which is mainly used to explain the possible absence of a significant
profit-structure relationship, is the so-termed quiet life (QL) hypothesis, which actually is a special
case of the market-power hypotheses. The reasoning of this theory is that as firms have more market
power, either through market share or concentration, the management is less focussed on efficiency,
since setting prices at more favourable levels can increase revenues. The quiet life hypothesis states
that firms do increase revenues as a result of increased market power, but because of higher
inefficiencies this does not lead to higher profitability.
2.2 Efficient-structure hypotheses
The efficient-structure (ES) hypotheses explain the positive relationship between profitability and
either concentration or market share with reference to efficiency measures. Efficiency of individual
banks causes high profits and a high market share. If one controls for efficiency, the link between
profitability and market structure variables will become insignificant and thus economically
meaningless. The ES hypotheses can be divided into the efficient-structure-X-efficiency and efficientstructure-scale-efficiency hypotheses.
The efficient-structure-X-efficiency (ESX) hypothesis states that firms are able to realise higher profits
as a result of superior management. X-efficiency indicates the level of managerial efficiency. It
measures to what extent the management is successful in earning maximum profits given input and
output prices, or in minimising costs given input prices and output quantities. In addition to increasing
profits, higher X-efficiency levels also enable firms to increase their market share, at the expense of
-4-
less efficient firms, which may result in a higher level of concentration 2. The profit-structure
relationship is found because of the positive effect of X-efficiency on both profitability and market
shares, and possibly on market concentration. Because of these simultaneous linkages, a significant
but economically meaningless relationship between profitability and market structure is found, when
no correction for X-efficiency is made.
Under the efficient-structure-scale efficiency (ESS) hypothesis, the difference in profitability between
firms is not caused by differences in the quality of management, but by differences in the level of scale
efficiency at which a firm is operating. Some firms may operate below their efficient scale, given input
prices and product mix, and others may produce above efficient scale. By moving towards a fully
efficient scale, firms can realise higher profits per unit of output, higher market shares and possibly
higher levels of concentration. The effects of scale efficiency on both profitability and market
structure variables may lead a spurious profit-structure relationship.
2.3 Empirical evidence
The results documented in the existing literature with respect to the different hypotheses are, as with
the profit-structure relationship itself, not very consistent. Table 2.1 shows the evidence of a selection
of existing studies, which explicitly explored the profit-structure relationship. The table shows that all
hypotheses receive support at least once which makes it impossible, at this point, to accept one
hypothesis as the dominant explanation of the profit-structure relationship.
Table 2.1 Recent studies on hypotheses explaining the profit-structure relationship
Study
_______________________________
Berger (1995)
Berger & Hannan (1997)
Berger & Hannan (1998)
Berger & Humphrey (1997)
Molyneux & Thornton (1992)
Goldberg & Rai (1996)
SCP
RMP
QL
ESX
ESS
________ ________ ________ ________ ________
NT
+
+
−
−
+
+
−
−
−
+
NT
NT
NT
NT
−
+
+
−
−
NT
+
NT
NT
NT
NT
+
+
−
−
Note: a ‘+’ marks empirical support for the hypothesis in question, a ‘−’ marks lack of support, and ‘NT’
means the hypothesis was not tested for.
2 Whether concentration increases depends on the measure used and on the market share rank of the merging firms. The
Herfindahl index is based on the squared market shares of all firms present in a market, so this concentration measure will
always increase as a result of a merger. Concentration ratios, however, use the largest firms in a market to determine the
level of concentration. This measure will only show an increase if either one of the merging firms or both are already
included in the ratio, or if the merged firm enters the ratio with a higher market share then the firm that is pushed out of the
ratio.
-5-
3 EXISTENCE OF A PROFIT-STRUCTURE RELATIONSHIP FOR EUROPEAN BANKS
The present section provides an empirical evaluation of the existence of a profit-structure relationship
in the European banking sector. This evaluation is based on return on equity and return on assets as
measures of profitability for all individual banks in the sample. Individual market shares, the market’s
Herfindahl index and the largest-five-bank concentration ratio are used as measures of the market
structure. The data used are described in section 3.1. The existence of a profit-structure relationship in
Europe is investigated using correlation analyses (section 3.2) and regression analyses (section 3.3).
Section 3.4 concludes.
3.1 Data
The data used for this study relates to information from balance sheets and income statements,
obtained from the BankScope database, which is compiled by Bureau van Dijk in co-operation with
FitchIBCA. The sample contains annual data of commercial banks in eight European countries,
namely Belgium, France, Germany, Great Britain, Italy, Luxembourg, the Netherlands and Spain, for
the period 1992 – 1997. The reason to study only commercial banks instead of all European banks is
twofold: homogeneity and importance. Using one type of banks makes the group of observations more
homogeneous, allowing for better efficiency estimates (see e.g. Bikker (1999)). Except for Germany
and Italy, commercial banks are largest in number in each of the selected countries. To illustrate the
relative importance of different categories of banks, table 3.1 gives the total number of banks available
in the 1995 BankScope data set by country and category. Although co-operative banks are large in
number as well, their presence is very low in some countries and very high in others. Commercial
banks are much more evenly spread over the different countries, which enhances the quality of a
cross-country analysis.
Table 3.1 Number of banks by country and category in 1995
Medium & Long
Specialised
Country Commercial Co-operative Investment Term
Non-banking Mortgage Savings Governmental Total
_______ __________ __________ _________ ______________ __________ ________ __________________ ______
BE
DE
ES
FR
GB
IT
LU
NL
50
264
77
272
131
75
121
51
9
1034
6
93
0
450
3
2
4
13
2
7
116
2
6
7
18
6
1
9
0
9
1
2
6
7
5
192
42
1
9
1
2
68
1
6
64
1
1
4
19
608
37
22
5
64
3
4
1
12
0
1
0
17
1
0
109
2012
129
602
358
619
145
71
Total
1041
1597
157
46
263
147
762
32
4045
Source: BankScope
-6-
Definitions of price and market structure variables
Prices of loans, securities and deposits are calculated by dividing the income or expense flow
associated with the particular input or output by the average outstanding dollar amount. The price of
labour is based on a different construction. Because of the lack of data on numbers of employees for
most banks in the data set, the price of labour is defined as the ratio of personnel expenses to total
assets. Because input and output prices have to be approximated an outlier correction is used to
remove the most extreme prices from the sample.
The market share (MS) of a bank is defined as the ratio of total assets of the individual bank divided
by the total assets of all banking institutions in the bank’s national market. The Herfindahl index
(Herf) is a measure of concentration in a national market and is calculated by summing the squared
market shares of all banking institutions within this market. The five-bank concentration ratio (Conc5)
is defined as the sum of the market shares of the five largest banks within the national market. For the
calculation of these market structure variables, data on all banks within a country are used, because
commercial banks do not only compete with other commercial banks, but also with other types of
banks, e.g. savings banks and co-operative banks. The Herfindahl index and the five-bank
concentration ratio are both included in the present study, because they have different properties that
might result in different estimates. The Herfindahl index includes all banks present in the market, so in
principle it captures all movements of consolidation or concentration. The concentration ratio includes
the market shares of the five largest banks in the data set and, thus, only captures some of the
movements in the market. The former measure contains more information, whereas the latter is less
sensitive to measurement errors. Because of these different characteristics, both measures are used
separately for all estimations that involve measures of market concentration.
Data
The present study uses an unbalanced set of panel data on commercial banks, running from 1992 to
1997. The total number of observations equals 4175, with a minimum of 530 banks and a maximum of
797 banks for individual years. A summary of key variables is given in table 3.2. The reported values
are average values for the whole sample period. Germany and France have the highest absolute
number of commercial banks. The countries with the smallest number of commercial banks are the
Netherlands and Belgium. Based on the return on assets and equity as measures of profitability, Great
Britain and Luxembourg perform best on average and Italy and France clearly perform poorly. The
differences in average profitability are quite large, ranging from −1.15 percent to 16.5 percent for
return on equity (ROE) and from 0.08 to 0.77 percent for return on assets (ROA). Average prices show
a much smaller range, although some differences are still present. The average price of loans is highest
of all prices and the prices of securities and deposits are of about the same magnitude. Where market
-7-
structure variables are concerned, the Netherlands and Belgium clearly show the largest market
concentration whereas Germany is by far the least concentrated market.
Table 3.2 Summary of key variables
Number
of banks
____________ _______
BE
37
DE
183
ES
64
FR
166
GB
63
IT
58
LU
95
NL
30
All
696
Country
ROE
ROA PLoans PSecurities PLabor
PDeposits MS
______
0.1286
0.0802
0.1055
0.0129
0.1650
-0.0115
0.1519
0.1157
0.0823
_____
0.0039
0.0027
0.0069
0.0008
0.0077
0.0010
0.0048
0.0051
0.0037
______
0.0619
0.0897
0.0724
0.0967
0.0614
0.1021
0.0779
0.0672
0.0818
_____
0.0973
0.0895
0.1287
0.1093
0.1062
0.1198
0.1058
0.0974
0.1049
______
0.0748
0.0636
0.0970
0.0871
0.0737
0.0940
0.0760
0.0690
0.0780
_____
0.0095
0.0102
0.0159
0.0112
0.0145
0.0158
0.0031
0.0122
0.0121
_____
0.0157
0.0017
0.0078
0.0019
0.0069
0.0081
0.0079
0.0169
0.0056
Herf
____
0.094
0.006
0.056
0.034
0.040
0.038
0.028
0.159
0.037
Conc5
_____
0.6138
0.2646
0.4615
0.3328
0.3699
0.3429
0.2673
0.7884
0.3561
3.2 Correlation analyses
The existence of the profit-structure relationship in the European banking sector is evaluated on the
basis of the market structure variables (MS, Herf and Conc5) and the measures for profitability (ROE
and ROA). The null-hypothesis asserts that the profit-structure relationship does not exist in the
European commercial banking industry, implying that there is no correlation between profitability
variables and market structure variables. For acceptance of the alternative hypothesis, i.e. the existence
of the profit-structure relationship, a significant positive correlation has to be found. To test for these
hypotheses, the analysis concentrates on the significance of the Pearson correlation test statistic, which
follows a t-distribution 3.
The correlations documented in table 3.3 suggest that the profit-structure relationship exists in the
European banking sector 4. All have the right sign and are, except for one, significant at the fivepercent level or higher. The two measures of profitability do show some differences in results.
3
The Pearson correlation test statistic is defined by
tρ =
1 − ρ
2
ρ
ρ
=
correlation coefficient,
N
=
number of observations
tρ follows a t-distribution with N-2 degrees of freedom
Because the data contain some serious outliers that have a large impact on the test statistics, a truncation has been applied
to the return on equity and return on assets observations. The observations in the lowest and highest percentile have been
assigned the value of the 1% and 99% observations, respectively.
with
4
N − 2
-8-
Whereas return on equity shows the highest correlation with market share (MS), return on assets fails
to show a significant correlation with this measure. Return on assets on the other hand seems to be
more strongly correlated with the concentration measures (Herf and Conc5) than return on equity. This
suggests that the profit-structure relationship is mainly based on market shares, when return on equity
is concerned, whereas return on assets is clearly positively related to the level of market concentration.
Table 3.3 Correlations to test for the existence of the profit-structure relationship
ROE
ROA
____________
Market structure
Correlation
variable
__________________ ____________
0.0999**
0.0436**
0.0388**
MS
Herf
Conc5
________________ ____________ ________________
Pearson correlation Correlation
Pearson correlation
test statistic
test statistic
________________ ____________ ________________
6.488
2.820
2.509
0.0197
0.0648**
0.0710**
1.275
4.192
4.599
Note: Correlations estimated over the whole sample period, based on 4175 observations. Correlations
significant at the 5 % level are marked by **.
3.3 Regression analyses
To check the results based on bilateral correlations, ordinary least squares regressions are run.
Moreover, a regression analyses is more advanced than a correlation analysis because it may provide
insights into the relation underlying the co-movement of variables. The functional specification of the
regressions is given by equation (3.1).
π
i ,t
= ψ + τ 1 MS
with:
πi,t
MSi,t
Concm,t
i ,t
+ τ 2 Conc
m ,t
+ ξ 1 ∆ GDP
m ,t
+ ξ 2 D Cons
,i ,t
+ ξ 3 T t + η i ,t
(3.1)
=
=
=
profitability, either return on equity or return on assets, of bank i in period t
market share of bank i in period t
market concentration, either Herfindahl index or concentration ratio, of
country m in period t
∆GDPm,t = annual growth rate of real gross domestic product of country m in period t
=
consolidation dummy for bank i in period t
DCons,i,t
Tt
=
trend variable
ηi,t
=
error term
ψ, τ1, τ2, ξ1, ξ2, ξ3
= parameters to be estimated
The regression analysis uses return on equity and return on assets as dependent variables, and market
share and either the Herfindahl index or the concentration ratio as independent variables. By including
the national annual growth rate of real gross domestic product, a correction is made for the cyclical
stance in each country. Besides this correction, a dummy variable is included to allow for a difference
-9-
in profitability between unconsolidated banks and consolidated banks. This dummy variable takes the
value of one for consolidated banks, and zero for unconsolidated banks. It is introduced because of
possible differences between consolidated and unconsolidated banks with regard to the way the
balance sheet is constructed which may lead to systematic differences in profitability. To accept the
null-hypothesis of no profit-structure relationship, both τ1 and τ2 have to be zero. To reject the nullhypothesis and accept the alternative hypothesis, i.e. the profit-structure relationship does exist, either
τ1 or τ2 has to be significantly positive.
The empirical results of the profit-structure relationship for European commercial banks using
regression (3.1) are presented in table 3.4. The estimated linear relations indicate that over the whole
period, return on equity is positively related to market share, the Herfindahl index and the
concentration ratio. For return on assets, the positive relationship is only found with the concentration
measures. An interesting aspect of these estimates is the low value of the adjusted R2. Although some
of the market structure variables are significant a larger part of the return fluctuations remains
unexplained.
Table 3.4 OLS estimates of the profit-structure relationship
Independent variables
Dependent variable
____________________
ROE
ROE
ROA
ROA
____________ ____________ ____________ ____________
Constant
0.0001
(0.018)
-0.0103
(-1.252)
0.0015**
(3.489)
0.0002
(0.293)
MS
0.7517**
(7.58)
0.7402**
(7.429)
-0.003
(-0.708)
-0.0037
(-0.883)
Herf
0.0853**
(2.055)
0.0153**
(4.542)
0.0372**
(3.115)
Conc5
0.0051**
(5.511)
GDP
0.8845**
(7.038)
0.9054**
(7.22)
0.0189**
(2.639)
0.0222**
(3.105)
DCons
0.0002
(0.033)
-0.0007
(-0.112)
0.001**
(2.463)
0.0009**
(2.275)
T
0.0064**
(4.467)
0.0065**
(4.521)
0.0003**
(2.96)
0.0003**
(3.085)
0.0263
0.0268
0.0086
0.0101
Adj. R2
Note: Significant coefficients are marked by ** for a 5 % significance level. T-values are reported
in parentheses.
- 10 -
3.4 Conclusions
The correlation and regression analyses show that the existence of the profit-structure relationship in
the European banking sector is supported strongly for return on equity, notably in relation to market
share. The existence of a profit-structure relationship based on return on assets is less clear. The
positive relationship between return on assets and market concentration is supported strongly whereas
there is no significant positive relationship between this measure of profitability and market share.
- 11 -
4 EXPLAINING THE PROFIT-STRUCTURE RELATIONSHIP
After having established that there are strong indications pointing to a profit-structure relationship in
the European banking sector the question about the origin of this relationship becomes relevant. This
section will formulate and test some hypotheses concerning possible explanations of the profitstructure relationship. The discussion in section 2 revealed that, generally, the existence of a profitstructure relationship is attributed to market power or efficiency. To distinguish between the several
theories, bank estimates of efficiency are crucial. Because these estimates are not straightforward,
section 4.1 discusses the methodology used here. An impression of the empirical results is given in
section 4.2. Section 4.3 discusses the methodology of testing the various explanations for a profitstructure relationship. Section 4.4 describes the test results.
4.1 Measurement of efficiency
Now that the existence of the profit-structure relationship has been empirically confirmed, it is
necessary to construct a model for the empirical assessment of efficiency levels of individual banks.
These are needed to test the validity of the different hypotheses explaining the profit-structure
relationship. To estimate X-efficiency and scale efficiency, a model is constructed to estimate an
efficiency frontier. The construction of this model involves several issues, some of which concern the
estimation of both X-efficiency and scale efficiency, while others are specific for the type of efficiency
being estimated. The present section discusses the methodology of efficiency measurement 5.
Cost or profit function
This study uses a profit function to estimate efficiencies. The advantage of profit efficiency is that it
considers both the cost side and the revenue side of operations. Taking only the cost side into account
may create a bias against banks with high quality service as their competitive strategy. These banks
will show relatively high costs per unit of output as a result of high quality service, but it might very
well be that because of this service the revenue per unit of output is relatively high as well. Efficiency
based on a cost function only concentrates on the higher costs, resulting in a lower level of efficiency.
Besides the possible bias, banks are usually considered to be profit-maximising entities and, hence, an
efficiency estimate based on profits will be more consistent with actual bank behaviour. Another
argument is that previous studies on bank efficiency and mergers have shown that most inefficiencies
5 Berger & Humphrey (1997) survey 130 studies that analyze financial institutions with the use of frontier efficiency
techniques. Their paper gives a good overview of approaches used and the differences in average efficiency levels found.
Berger & Mester (1997) examine possible sources of differences in measured efficiency. Most efficiency studies concentrate
on US banks. Exceptions are Schure & Wagenvoort (1999) who examine the fifteen EU-countries and Pastor, Pérez &
Quesada (1997) who provide an international comparison between the major European countries and the United States.
- 12 -
are found on the output (revenue) side, which can be expected to be the source of potential gains from
mergers (e.g. Berger, Hancock & Humphrey (1993); Akhavein, Berger & Humphrey (1997)). Profit
efficiency takes output efficiency into account, and is therefore better suited to analyse the efficiency
benefits of bank mergers. Indeed, Berger (1998) reported for US bank mergers a relative improvement
of profit efficiency with very little change in cost efficiency.
Efficiency frontier approach
To estimate profit efficiencies, the stochastic frontier approach is employed. Although this approach
has the drawback of imposing a distribution on the error term, it is preferred over nonparametric
approaches. Nonparametric approaches do not impose a distribution on the error term, but do not
allow for random error either. As the data set is comprised of balance sheet data from banks that
operate in different countries, this is a restriction that can have serious implications for the estimated
efficiencies. Other parametric approaches than the stochastic frontier approach are not selected,
because they do not allow for annual estimates of individual bank efficiencies, which renders them
useless for the purpose of this study.
The stochastic frontier approach entails the estimation of a profit function with a composite error term
ε = v – u, with v as the random error component and u as the efficiency error component, measuring
the downward effect of inefficiency 6. This efficiency error component forms the basis of the Xefficiency estimates discussed below. The distribution of v is assumed to follow a normal distribution
with mean zero and standard deviation σv. By definition, the distribution of u is a one-sided
distribution, because inefficiency can only cause profits to be lower than the frontier profit. Often used
distributions are the half-normal distribution and the truncated normal distribution, truncated at zero.
The truncated normal distribution is applied in this study, to minimise the restrictions on the
distribution of u. The truncated normal distribution leaves the mode of the distribution to be
determined by the data, whereas the half-normal distribution restricts the mode of the distribution to
zero. Because of this larger flexibility, u is assumed to follow a truncated normal distribution with
mode µ and standard deviation σu.
Bank inputs and outputs
Labour and deposits are considered as bank inputs, with output consisting of loans and securities.
Fixed capital, off-balance-sheet items and financial equity are included as fixed netputs, i.e. as fixed
inputs or outputs. This corresponds to the intermediation approach, which, as suggested by Berger &
Humphrey (1997), is appropriate for the analysis of entire banks. The second fixed netput, off-
6 See Aigner, Lovell & Schmidt (1977).
- 13 -
balance-sheet items, is used as such for two reasons. First, previous studies have shown that this item
can have a relatively large influence on the performance of a bank. Since the early 1990’s, offbalance-sheet activities have increased strongly and, hence, have gained importance. Second, it is
virtually impossible to construct a price per unit for these off-balance-sheet items from available data.
Thus, in order to avoid estimating unreliable prices, and yet to be able to use off-balance-sheet items
for the estimation of the profit efficiency frontier, this variable is treated as a fixed output.
Functional specification of the profit function
Preference is given to a specification that does not use logarithms, since profits may be negative. The
profit efficiency frontier that we estimate for the unbalanced set of panel data of European commercial
banks is given by equation (4.1). This functional specification is based on the Fuss quadratic flexible
form function, as used by Akhavein, Berger & Humphrey (1997) and Berger, Hancock & Humprhey
(1993).
Π
i ,t
( p n ,i ,t z k ,i ,t )
=
n
∑α
r =1
r
 p r ,i ,t

 p
 n ,i ,t
 1
+
 2

∑∑χ
 z s ,i ,t

 z
 k ,i ,t
 1
+
 2

∑ ∑θ
k
+
∑β
+
∑∑γ
s
s =1
n −1
r =1
k
s =1
+ ζ 1 D Cons
+
with:
Πi,t
pr,i,t
zs,i,t
DCons,i,t
rs
,i ,t
 p r ,i ,t

 p
 n ,i ,t
t =1
r =1
x =1
k
k
s =1
y =1
  z s ,i ,t

 z
  k ,i ,t
rx
sy
 p r ,i ,t p x ,i ,t

2

p n ,i ,t

 z s ,i ,t z y ,i ,t

 z2
k ,i ,t

STINT
4 ,t
m ,t
+
t =1




m ,t
6
∑ζ








 L 
+ ζ 2   + ζ 3 ∆ GDP
 A  i ,t
6
∑ζ
n − 1 n −1
5 ,t
LTINT
m ,t
+ ζ 6 T t + ε i ,t
(4.1)
=
=
=
=
post-tax profits of bank i in period t; t=1,..,6 and 4175 total observations
price of variable netput r of bank i in period t; r=1,..,n and n=4
quantity of fixed netput s of bank i in period t; s=1,..,k and k=3
dummy variable for bank i in period t: equals 1 if bank i,t is a
consolidated bank and zero otherwise
L/Ai,t
=
loans-to-assets ratio of bank i in period t
=
annual growth rate of real gross domestic product of country m in period t
∆GDPm,t
=
short-term interest rate of country m in period t
STINTm,t
=
long-term interest rate of country m in period t
LTINTm,t
Tt
=
trend variable
=
composite error term, εi,t = vi,t – ui,t
εi,t
αr, χrt, βs, θsy, γrs, ζ1, ζ2, ζ3, ζ4,t,ζ5,t, ζ6 = parameters to be estimated with χrt = χtr and θsy = θys
The prices of loans and securities are denoted by p1 and p2, respectively. The price of labour is given
by p3 and p4 is the price of deposits. The fixed inputs and outputs are all quantities, denominated in
thousands of dollars. The quantity of fixed assets of a bank is given by z1; z2 is the amount of off-
- 14 -
balance-sheet items and z3 denotes financial equity. In order to impose linear homogeneity in prices,
the dependent variable and all price variables are normalised by the last netput price pn, which is the
price of deposits. In addition to this normalisation, the dependent variable and all quantity variables
are scaled by the last fixed netput, which is financial equity. To get a better fit of the profit function,
some control variables are added. A dummy variable, which equals one for banks with consolidated
statements, is introduced to allow for differences in profitability between consolidated and
unconsolidated banks. The ratio of total loans over total assets gives an indication of the ability of a
bank to diversify its portfolio, so that it can reduce its risk. The number of outstanding bank loans is
limited by several factors, like its capital and the risk associated with its portfolio, as banks are always
under supervision and have to comply with certain capital restrictions. Because loans are viewed as
both more profitable as well as riskier than securities, a bank may want to maximise its loan portfolio.
It can only do so by better diversifying its portfolio, which reduces overall risk. The third control
variable is the annual growth of real gross domestic product. This variable is country specific, and it
reflects the economic situation of a country, which has its influence on the performance of economic
entities, banks being no exception. Including this variable in equation (4.1) prevents attributing
country differences to efficiency, when these differences are caused by different macroeconomic
circumstances. If not included, this could lead to larger efficiency differences between countries,
which are not realistic. The short-term interest rate and the long-term interest rate are related for the
same reasons. Finally, a time variable is included, because the profit function is estimated over the
whole sample period. Estimating the profit function for each year separately would lead to six
different profit functions, which hampers an inter-temporal comparison. As this study aims to compare
pre- and post-merger efficiencies, the comparability of efficiencies in different years is an essential
aspect. By estimating one profit function over all years, efficiencies of different years are comparable.
To allow for a shift in the profit function and possibly the level of efficiency, the time variable is
included.
Likelihood function
Equation (4.1) is estimated by maximising the likelihood function. Although the error term ε is
asymmetrically distributed, its behaviour is captured by two well-behaved distributions which enables
the construction of a log-likelihood function. Following Stevenson (1980) and Greene (1993), this
function is defined as in equation (4.2).
- 15 -
L(α r , χ rx , β s , θ sy , γ rs , ζ 1 , ζ 2 , ζ 3 , ζ 4, t , ζ 5,t , ζ 6 , λ , σ , µ ) =
2

 − µ (1 + λ2 ) 12 

1
1  εi , t 
εi,t λ µ 


− ∑∑ lnσ + ln 2π +   − ln1 − Φ
+  + ln Φ


2 σ 
2
σ
σλ
σλ
t
i 







with:
(4.2)
αr, χrx, βs, θsy, γrs, , ζ1, ζ2, ζ3, ζ4,t,ζ5,t, ζ6
=
parameters of equation (4.1)
σ
=
√(σv2 + σu2) = standard deviation of ε
λ
=
σu/σv
µ
=
mode of the distribution of u
Φ
=
cumulative distribution function of the standard normal distribution
εi,t
=
composite error from equation (4.1)
X-efficiency
After estimating equation (4.1) with log-likelihood function (4.2), the composite error term ε is known
for every individual bank and every year in the sample. Given the distributional assumptions of u and
v the efficiency error term can be extracted from the composite error term 7 using equation (4.3) 8.
This conditional expectation of the efficiency error component is taken as an estimate for ui,t, with
which a measure of bank (in)efficiency is constructed. This study employs a measure of X-efficiency,
which, to our knowledge, has not been used in combination with profit efficiency yet. Instead of using
the ratio of predicted over efficient profit, a direct measure of inefficiency is constructed, which is
fairly easy to retrieve from the estimated efficiency error component, and leads to an intuitively
interpretable value. As ui,t resembles that part of the deviation of Πi / (pn,i zk,i) from the frontier that can
be attributed to inefficiency, it is denominated in post-tax profit divided by the price of deposits and
the quantity of equity. By multiplying ui,t by the price of deposits, the deviation is measured in terms
of post-tax profit per equity; thus inefficiency (XE i,t) equals E[ui,t |εi,t ] pn,i,t. This measure can be
interpreted as the lost return per dollar equity as a result of inefficiency, and is thus by itself a measure
of X-inefficiency. Allen & Rai (1996) use a similar measure of X-inefficiency, but based on a translog
cost function. They use the estimate of ui,t as a direct estimate of inefficiency. As a result of the
logarithmic functional specification, their estimate is multiplicative, whereas in our study, the Xinefficiency measure is additive.
7
8
To limit the influence of extreme errors, the set of εi,t is truncated. This truncation is performed by assigning the values of
the 1 % and 99 % observation to the observations with errors that fall within the lowest and highest percentile.
See Stevenson (1980) and Greene (1993).
- 16 -
E [u
i ,t
|ε
i ,t
with
]=
σλ
(1 + λ
2

 ε i ,t λ
µ
+
 φ 
σλ
 σ

ε i ,t λ
) 

µ
−
 Φ  −
σ
σλ








−
ε
i ,t
λ
σ
E [ ui,t | εi,t ]
= expected value of ui,t conditional on the value of εi,t
φ
= density of the standard normal distribution
−
µ
σλ


 (4.3)



all other variables are as defined in equations (4.1)and (4.2)
Scale efficiency
To estimate bank-specific scale efficiency levels (SEi,t), equation (4.1) is transformed through
replacing (p1/pn) and (p2/pn) by (q1/zk) and (q2/zk), respectively, because scale efficiency is related to
output quantities instead of prices. The functional specification remains the same, only the parameters
and terms containing (p1/pn) and (p2/pn) change. The new equation is also estimated by the maximising
likelihood function (4.2). After this estimation, the partial derivatives of the function with respect to q1
and q2 are calculated. The derivatives are then multiplied by the price of deposits to get a meaningful
measure. This measure gives the marginal profitability of each output, which reflects the change in
return per dollar equity as a result of a one-dollar increase in output q1 and q2, respectively. To
estimate the combined scale inefficiency for a bank, the weighted average of the two marginal
profitability measures is calculated, using the output quantities as weights, such that there is no change
in the output mix. In formula, this translates to equation (4.4).
Π
Π


∂  i ,t
∂ i ,t


p
z
p
z
q
q1,i ,t
n ,i ,t k ,i ,t 
n ,i ,t k ,i ,t 


2 ,i ,t
SE i ,t =
p n ,i ,t +
pn ,i ,t
( q1,i ,t + q 2,i ,t )
∂q1,i ,t
( q1,i ,t + q 2 ,i ,t )
∂q 2 ,i ,t
(4.4)
When a bank is fully scale efficient, the scale inefficiency measure SE will equal zero. The bank
cannot improve its profitability by increasing or decreasing its output quantity. If SE is larger than
zero, a bank can increase its profitability by operating on a larger scale (i.e. increasing its output). In
this case, economies of scale are present. Scale inefficiencies associated with economies of scale are
denoted by SEE. If SE is smaller than zero, a bank is operating above fully efficient scale. It can
increase profitability by operating on a smaller scale (i.e. decreasing its output). This bank is operating
with diseconomies of scale, and its scale inefficiency is denoted by SED. The further away from zero,
the less scale efficient a bank is operating, so in this way, SE is in fact a measure of scale inefficiency.
Economies and diseconomies are treated separately in the analysis of the profit-structure relationship,
because they have different implications for the coefficients that are estimated to test the different
hypotheses.
- 17 -
An important aspect of this estimate of scale inefficiency is that it does not measure the total loss due
to scale inefficiencies as the measure of X-inefficiency does. The estimate gives the local marginal
profitability of a bank, which resembles the potential gain or loss in return on equity per one-dollar
increase in output. This local estimate cannot, however, be used to approximate the total loss in terms
of return on equity. This is not possible for two reasons. The first is that, being a local estimate, this
measure cannot be used to calculate the total loss in profitability, even if the amount by which output
has to increase to reach fully efficient scale were known, because the marginal profitability cannot be
assumed constant for different levels of output. Second, it is not possible to increase output without
changing other variables as well. It is impossible to increase output, i.e. loans and securities, without
increasing equity. Also, deposits are needed for an increase in output, so it is likely that a bank has to
raise its price of deposits in order to attract the necessary funding need.
Because of the lack of an estimate of the total loss in profitability due to scale inefficiencies, it is
difficult to draw conclusions on differences in magnitude between scale inefficiency and Xinefficiency. The scale inefficiency, however, does allow for a comparison of the relative scale
inefficiency levels of individual banks. Our further analysis concentrates on this feature.
4.2 Efficiency of European commercial banks: empirical results
The efficiency estimates are discussed briefly, since they are just a means in the analysis of the profitstructure relationship 9. The average annual inefficiency levels for all countries together are given in
figures 4.1 and 4.2, displaying X-inefficiency and marginal scale inefficiency, respectively. Figure 4.1
clearly shows a downward trend in the level of X-inefficiency through time. This means that within
the sample of commercial banks, X-efficiency has steadily improved over the years. The same can be
said about scale efficiency, although the trend is not as clear-cut as with X-efficiency, since a few
exceptions are present (figure 4.2). A possible explanation of this steady improvement of efficiency
can be the formation of the European internal market, which was completed in the sample period. This
development might have forced European banks into a fiercer competition with banks outside their
national markets. This increase in international competition might have led to increased pressure on
banks to improve their efficiency.
9 The estimates have been obtained with the use of the computer program ‘Frontier 4.1’, developed by T. Coelli of the
University of New England, Australia. See Coelli (1996).
- 18 -
Figures 4.1 and 4.2 seem to imply another often found feature of inefficiencies, namely that the
magnitude of X-inefficiencies is many times that of scale-inefficiencies. As explained above, though,
this is not as straightforward as it seems. X-inefficiency resembles the total loss in profitability,
whereas scale inefficiency only gives the local estimate of marginal profitability with respect to
output. The magnitude of the local scale inefficiencies, however, is of such an order, that it is plausible
to assume that the total loss in profitability due to scale inefficiencies will not be as large as the loss
due to X-inefficiency. Unfortunately, this assumption cannot be tested, because of the unknown value
of the overall marginal profitability with respect to output quantities.
Figure 4.1 Annual average X-inefficiency for total sample
0 .2
0 .1 8
0 .1 6
0 .1 4
0 .1 2
0 .1
0 .0 8
0 .0 6
0 .0 4
0 .0 2
0
1992
1993
1994
1995
1996
1997
Figure 4.2 Annual average scale inefficiency for total sample
8 .0 E - 0 9
6 .0 E - 0 9
4 .0 E - 0 9
2 .0 E - 0 9
0 .0 E + 0 0
- 2 .0 E - 0 9
- 4 .0 E - 0 9
- 6 .0 E - 0 9
1992
1993
1994
S c a le - in e f f ic ie n c y (E c o n o m ie s )
1995
1996
1997
S c a le - in e f f ic ie n c y ( D is e c o n o m ie s )
- 19 -
4.3 Tests on the explanation of the profit-structure relationship
After the different types of efficiency have been estimated, they are used, together with the market
structure variables and profitability, to determine the economic forces driving the profit-structure
relationship. To test the different hypotheses explaining this relationship, which were described in
section 2, we use a mathematical formulation of the hypotheses based on Berger & Hannan (1997).
Tests on the two categories of hypotheses are discussed separately, starting with the efficient-structure
hypotheses.
Testing the efficient-structure hypotheses
Under the efficient-structure hypotheses, differences in efficiencies are the cause of profitability
differences. Once a correction is made for these efficiency differences, the market structure variables
should have no influence on profitability. Furthermore, as the use of market power is no source of
higher profitability, banks with higher market shares or concentration should show no superior
profitability, through price setting, than banks with lower values. So, the market structure variables
should not have any significant influence on prices either. When a bank becomes more efficient, it can
choose to pass through part of the efficiency gains by changing its prices in favour of its customers.
However, the bank is not obliged to do so and prices can also remain unchanged. Higher efficiency
should therefore have either no impact or a downward (upward) effect on output (input) prices. These
expectations lead to the following functional forms for profitability and prices 10.
<0
<0
>0
0
π i , t = f 1 ( XE i , t , SE i , t , SE i , t , Conc
E
≥0
≥0
D
≤0
0
Pi , t = f 2 ( XE i , t , SE i , t , SE i , t , Conc
E
0
m , t , MS i , t , Z i , t ) + ϑ 1 i , t
(4.5)
0
D
m ,t
, MS i , t , Z i , t ) + ϑ 2 i , t
(4.6)
These and the following equations use variables and indices that are defined above, with Zi being a
vector of control variables, consisting of the same set of variables used in equation (4.1).
Besides the influence on profitability and possibly on prices, efficiencies also have an impact on the
market structure variables: market share and concentration. This is essential to the efficient-structure
hypotheses, because it is efficiency that is supposed to drive both market structure and profits. These
functions and conditions are defined as:
10 The signs in equations (4.6) and (4.6*) are based on output prices. For the price of deposits, which is an input price in the
intermediation approach, the signs should be reversed.
- 20 -
≤0
MS
i ,t
= f 3 ( XE
≤0
E
≤0
Conc
m ,t
≥0
i , t , SE i , t , SE i , t , Z i , t ) + ϑ 3 i , t
= f 4 ( XE
D
≤0
≥0
, SE i , t , SE i , t , Z i , t ) + ϑ 4 i , t
E
i ,t
D
(4.7)
(4.8)
Testing the market- power hypotheses
The market-power hypotheses claim that there is a causal relation from market structure variables to
profitability. The profitability and price equations that are used to test these hypotheses, have the same
form as with the tests of the efficient-structure hypotheses above, only the predicted signs of the
coefficients are different. As can be seen in the following equations, the market-power hypotheses do
not assume the influence of efficiencies on profitability to be absent. They are only concerned with the
direct causal relation that runs from the measures of market structure to profitability, even when a
correction for other variables has been made. So, as long as the market structure variables have an
impact on profitability, other variables are allowed to influence profitability as well.
≤0
≤0
≥0
>0
>0
π i , t = f 1* ( XE i , t , SE i , t , SE i , t , Conc m , t , MS i , t , Z i , t ) + ϑ 1*i , t
E
≥0
≥0
D
≤0
>0
>0
Pi , t = f 2* ( XE i , t , SE i , t , SE i , t , Conc m , t , MS i , t , Z i , t ) + ϑ 2*i , t
E
(4.5*)
D
(4.6*)
In addition to these relations, the most strict version of the market-power hypotheses assumes the
absence of a causal relation between efficiency (as independent variable) and market structure
variables (as dependent variables). The reverse might also be true, although it is not a necessary
condition for these hypotheses to be valid. In fact, these reversed relationships can be seen as a
manifestation of the quiet life hypothesis, which is a special case of the market-power hypotheses.
This theory claims that because of more market power, management becomes less focused on
efficiency, resulting in a negative relationship between market power and efficiency, as stated in
equations (4.7*), (4.8*) and (4.8**).
≥0
≥0
XE i ,t = f 3* (Conc m ,t , MS i ,t , Z i ,t ) + ϑ 3*i ,t
≥0
≥0
SE i ,t = f 4* (Conc m ,t , MS i ,t , Z i ,t ) + ϑ 4*i ,t
E
≤0
≤0
SE i ,t = f 4** (Conc m ,t , MS i ,t , Z i ,t ) + ϑ 4** i ,t
D
(4.7*)
(4.8*)
(4.8**)
Summary
The equations (4.5) to (4.8**) are estimated with the use of the efficiency estimates from section 4.2
and assuming linearity for the functions f1 to f4**. The estimated equations serve as a test of the
different hypotheses. If the estimated coefficients do not match the signs implied by a certain
- 21 -
hypothesis, this hypothesis has to be rejected. This way, conclusions can be drawn with regard to the
relations underlying the profit-structure relationship in the European commercial banking industry.
4.4 Empirical results on hypotheses explaining the profit-structure relationship
Equations (4.5) to (4.8**) are estimated in order to test the different efficient-structure and marketpower hypotheses. Tables 4.1 and 4.2 report the estimates 11. The estimated coefficients are analysed
horizontally; for each independent variable, the estimates are compared with the predictions of the
hypothesis regarding this independent variable. Then, if there is any reason to investigate the validity
of a hypothesis any further, the coefficients of the other independent variables are analysed to see
whether they are in line with the assumptions of the hypothesis concerned. The outcomes of these
estimations and the consequences for the validity of the different hypotheses are discussed per
hypothesis.
Testing the efficient-structure hypotheses
The first hypothesis considered is the efficient-structure-X-efficiency hypothesis (i.e. the first row of
estimated coefficients). This hypothesis predicts a negative coefficient of X-inefficiency in equation
(4.5). The estimation results in fact show a significantly negative coefficient, using either return on
equity or return on assets, so they support the hypothesis. Both output prices are significantly
positively related to the level of X-inefficiency in equation (4.6), again in line with the hypothesis. The
price of deposits should be negatively related to X-inefficiency, which is not the case. Instead, the
price of deposits decreases as efficiency improves; this finding is in conflict with the predictions of the
hypothesis. Market share is negatively related to X-inefficiency in equation (4.7). The relationship
between concentration and X-inefficiency depends on the concentration measure used. It is either
insignificant (using the Herfindahl index) or significantly negative (using the concentration ratio).
Both estimates are in support of the hypothesis, as it predicts a negative relationship or no relationship
between these variables. All in all, without regarding the coefficients of the other independent
variables, the efficient-structure-X-efficiency hypothesis is generally supported by estimates of
equations (4.5) to (4.8), although the influence of X-inefficiency on the price of deposits has the
wrong sign.
11 For clarity of exposition, tables 4.1 and 4.2 only report coefficients that are relevant for the analysis of the different
hypotheses.
- 22 -
Table 4.1 Estimates of equations (4.5) to (4.8**), with Herfindahl index as market concentration
Independent Dependent variable
variable
ROA
PLoans
PSecurities
PDeposits
MS
Herf
XE
SEE
SED
ROE
*
*
*
*
*
*
*
((4.5)/(4.5 ) (4.5)/(4.5 ) (4.6)/(4.6 ) (4.6)/(4.6 ) (4.6)/(4.6 ) (4.7)
(4.8)
(4.7 )
(4.8 )
(4.8**)
__________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________
-0.0028
-0.7662**
-0.0371**
0.0642**
0.0499**
0.0835**
-0.0095**
XE
(-22.344)
(-19.195)
(9.048)
(8.978)
(12.105)
(-6.579)
(-1.44)
SEE
SED
-3750.137**
(-7.945)
4653.879**
(2.432)
-87689.17** -6470.852** -65341.7** -51187.56** -75642.22** 7265.152**
(-1.984)
(-2.814)
(-6.558)
(-5.981)
(-5.162)
(3.786)
10855.88**
(2.683)
5800.86
(0.835)
62.6005
(0.085)
51363.11** 36649.28**
(15.096)
(16.085)
1849.757
(0.744)
Herf
-0.0233
(-0.746)
0.0090**
(2.846)
-0.0214
(-1.293)
0.0061
(0.525)
-0.0258**
(-2.152)
-0.0309
(-0.629)
3.93E-7**
(3.012)
MS
0.0909*
(1.813)
-0.0341**
(-7.289)
-0.1063**
(-4.762)
-0.088**
(-5.253)
0.0035
(0.205)
-0.8208**
(-8.45)
-8.57E-7** 1.30E-7**
(-8.647)
(5.172)
Adj. R2
0.6471
0.3955
0.4243
0.4551
0.4200
0.1152
0.1156
0.0731
6.15E-8**
(2.959)
0.0686
0.0447
Table 4.2 Estimates of equations (4.5) to (4.8**), with concentration ratio as market concentration
Independent Dependent variable
variable
ROA
PLoans
PSecurities
PDeposits
MS
Conc5
XE
SEE
SED
ROE
((4.5)/(4.5*) (4.5)/(4.5*) (4.6)/(4.6*) (4.6)/(4.6*) (4.6)/(4.6*) (4.7)
(4.8)
(4.7*)
(4.8*)
(4.8**)
__________ __________ __________ __________ __________ __________ __________ __________ __________ __________ __________
-0.7664**
-0.037**
0.0641**
0.0499**
0.0832**
-0.0095**
-0.027**
XE
(-22.325)
(-19.168)
(9.037)
(8.976)
(12.091)
(-6.579)
(-3.533)
SEE
SED
5867.43
(0.845)
64.474
(0.088)
51301.16** 36651.61**
(15.08)
(16.09)
2026.928
(0.817)
-3750.137**
(-7.945)
17129.4**
(2.338)
-87654.63** -6447.809** -65471.81** -51170.71** -75467.83** 7265.152** 31415.45**
(-1.984)
(-2.803)
(-6.573)
(-5.979)
(-5.155)
(3.786)
(2.456)
Conc5
-0.0093
(-1.141)
0.0024**
(2.863)
-0.0027
(-0.627)
0.0015
(0.507)
-0.016**
(-5.071)
-0.0386**
(-2.989)
9.53E-8**
(2.793)
MS
0.0933*
(1.863)
-0.0340**
(-7.243)
-0.1086**
(-4.91)
-0.0879**
(-5.263)
0.0099
(0.589)
-0.7987**
(-8.077)
-8.48E-7** 1.31E-7**
(-8.5)
(5.268)
Adj. R2
0.6471
0.3955
0.4241
0.4551
0.4219
0.1152
0.1168
0.0738
0.0683
1.52E-8**
(3.541)
0.0446
Note: Significant results are marked by * and **, indicating 10 % and 5 % significance levels, respectively. T-values
are reported in parentheses.
indicates a significant coefficient that supports the corresponding hypothesis
indicates a significant coefficient that conflicts with the corresponding hypothesis
The efficient-structure-X-efficiency hypothesis is also concerned with the coefficients of the market
structure variables in equation (4.5). As the hypothesis serves as an explanation of the profit-structure
relationship, it is required that the significant positive relationship between profitability and market
structure variables disappears when a correction is made for X-efficiency. Analysing the estimated
coefficients of these variables, there seems to be a difference between the two measures of
profitability. Using return on equity results in a negative but insignificant coefficient for the measure
of market concentration, but a significantly positive coefficient for market share. Using the return on
- 23 -
assets as the dependent variable, these coefficients change sign, with the measure of market
concentration now showing a significantly positive coefficient, and market share a significantly
negative coefficient. These results are not entirely consistent with the hypothesis, as it requires a nonpositive coefficient for both measures. It seems clear, however, that the relationship between market
structure variables and profitability has weakened considerably as a result of the correction for
efficiency. This is supported by comparing the estimated coefficients of equation (4.5) with the
estimated coefficients of equation (3.1), the direct ordinary least squares estimate of the profitstructure relationship. The magnitude of the coefficients has decreased severely due to the correction
and the signs are less consistent over the different measures than they were before. This indicates that
the relationship between market structure variables and profitability established in section 3 is mainly
caused by differences in efficiencies. Therefore, the few significantly positive coefficients in equation
(4.5) are not considered to comprise convincing evidence for the rejection of the efficient-structure-Xefficiency hypothesis.
Subsequently, the efficient-structure-scale efficiency hypothesis is taken into consideration. Again, a
negative coefficient for the inefficiency measure is required, as far as economies of scale are
concerned. For diseconomies of scale, a positive coefficient is required, as the inefficiency measure
for this category is smaller than zero. Analysing the estimates of equation (4.5), the general conclusion
is that the required negative relationship between profitability and the level of scale inefficiency is
absent. The coefficients are either insignificant or they have the wrong sign. This on itself is enough
evidence to reject the hypothesis. Although the estimates of equations (4.6), (4.7) and (4.8)
qualitatively show the same results as with the X-inefficiency measure and are thus mainly in support
of the hypothesis, the lack of a significant relationship of a the right sign between profitability and
scale efficiency leads to the rejection of the efficient-structure-scale efficiency hypothesis.
Testing the market- power hypotheses
The structure-conduct-performance hypothesis is the first to be analysed out of the set of marketpower hypotheses. As mentioned above, the estimates of equation (4.5*) give different results for
different measures of profitability. For return on equity, there is no empirical support for a positive
relationship with market concentration. For return on assets, however, there seems to be a significant
positive relationship between profitability and market concentration, even after correcting for
efficiency differences. Turning to the price relations, only the price of deposits in equation (4.6**)
seems to be significantly influenced by market concentration in the direction the hypothesis predicts.
The output prices are not significantly related to market concentration. Thus, based on equations (4.5*)
and (4.6**) there is limited support to the hypothesis if return on assets is considered. The results of
equations (4.7*), (4.8*) and (4.8**), however, are not very supportive of the hypothesis. For the
Herfindahl index, two out of three efficiency measures show a significant relationship with market
- 24 -
concentration, with one estimate contradicting the hypothesis and one supporting it. Using the
concentration ratio results in three significant coefficients, of which two are contradicting and only
one is consistent with the hypothesis. All this considered, it is concluded that there exists some but
rather weak empirical support for this hypothesis. The empirical results further indicate that market
power is only exercised through the price of deposits.
The relative-market-power hypothesis is supported by the coefficient estimate in equation (4.5*), but
only if return on equity is used as measure of profitability. This is, however, the only empirical support
it receives. All other coefficients either contradict this hypothesis or fail to support it. First, the
estimation of equation (4.5*) using return on assets as the dependent variable results in a significantly
negative coefficient, whereas a positive coefficient is necessary to support the hypothesis. Second,
none of the price relations of equation (4.6*) support the hypothesis. The output price equations show
significant coefficients with wrong signs and the price of deposits seems to undergo no significant
influence from market share. The equations (4.7*), (4.8*) and (4.8**) do show a significant coefficient
for market share, but the signs of these coefficients are all contradicting the relative-market-power
hypothesis. These findings all fail to support the predictions of the hypothesis, suggesting that this
hypothesis should be rejected.
The quiet life hypothesis, which is a special case of the market-power hypotheses, does not have to be
rejected because of the lack of a consistent positive relationship between the market structure variables
and profitability. This theory claims that gains from increased market power are offset by increases in
inefficiency. The data however, however, do not support this theory. The estimated relationships
between the different types of inefficiency and the market structure variables mainly result in
significant coefficients that have a sign contradicting the supposed positive influence of increased
market power on inefficiency. In addition, the price equations fail to convincingly support price
setting, as most of the estimated coefficients are either insignificant or have the wrong sign. Together,
these estimates show more results contradicting the hypothesis than supporting it, which leads to the
rejection of the quiet life hypothesis.
Conclusions
All in all, the tests of the different hypotheses that try to explain the profit-structure relationship
through the estimation of equations (4.5) to (4.8**) lead to the following conclusions about the validity
of the hypotheses. Allowing for two exceptions, the efficient-structure-X-efficiency hypothesis is
supported by the empirical results. Some support is also found for the structure-conduct-performance
hypothesis, although not as convincing as for the former hypothesis. For the remaining hypotheses,
empirical support is lacking or even contradicts their implications. Therefore, these hypotheses are
rejected.
- 25 -
5 EFFECTS OF BANK MERGERS AND THE PROFIT-STRUCTURE RELATIONSHIP
We now turn to the effects of mergers on bank performance through prices and efficiencies. Section
5.1 examines the effects of bank mergers whereas section 5.2 checks whether these effects comply
with the profit-structure hypotheses.
5.1 Effects of mergers on bank performance
This section assesses the effects of mergers on bank performance. Bank mergers are analysed using
data on the year prior to and the two years following the year of a merger. The year of the merger itself
is not included in the analysis, because this is often a turbulent year, which might give a
misrepresentation of the bank’s general performance 12. So, of each bank in the set, three years of
observations are used per merger year considered. As the sample period runs from 1992 to 1997, the
analysis will concentrate on mergers that took place in 1993, 1994 and 1995. Methodology, empirical
results and conclusions are discussed successively.
Methodology
To be selected as a merged bank, both the pre-merger banks as well as the post-merger bank have to
be present in the data set. Thus, only mergers are analysed that involve banks that operate in the
countries included in the data set. Mergers with one or more banks from outside these countries are not
taken into consideration. If observations are incomplete, these banks are removed from the sample.
These criteria result in two subsets per merger year, containing banks involved in merger activity in
this year and banks that are not involved in mergers. From the subset of non-merged banks, banks that
have been involved in a merger during the two years prior to or two years after the year of observation
are removed from the set in order to have a peer group that is clear of any merger effects. This results
in a separate peer group for each merger year, which is a balanced panel of non-merged banks.
The pooled observations of merged banks are used to analyse the change in performance variables for
these banks. Two absolute measures for the change in bank performance are used. The first measure is
the performance change from the year prior to the merger to the first year after the merger, and the
second is measured as the change from the year prior to the merger to the second year after the merger.
The asset-weighted average two- and three-year changes for the group of merged banks are called δ1
and δ2, respectively.
12 It is sometimes argued that it is better to analyze mergers by using years even further after the year of the merger, in order
to avoid using data containing restructuring noise. Unfortunately, this is not possible due limitations of the data set used.
- 26 -
These absolute changes in performance do, however, not differentiate between changes in
performance due to the merger and changes that are the result of other, e.g. market wide, factors. To
determine the changes that can be attributed to mergers, the relative change in performance is
analysed. This is done by comparison of the group of merged banks to the peer group of non-merged
banks. For each variable, the difference is taken between the weighted average two- and three-year
change of merged banks and non-merged banks. These differences are named ∆1 and ∆2. This results
into the following definitions of the different deltas:
δ1 =
∑w
a
(V a , 94 − V a , 92 ) +
N M , 93
δ2 =
∑w
a
(V a , 95 − V a , 92 ) +
N M , 93
∑w
∆1 = δ 1 −
∑w
∑w
d
(V d , 94 − Vd ,92 ) +
∑w
d
(Vd ,95 − Vd , 92 ) +
b
(Vb , 96 − Vb , 93 ) +
=
NM,year
NPG,year
windex
=
=
=
(Vc , 96 − Vc , 94 )
(5.1)
∑w
(Vc , 97 − Vc , 94 )
(5.2)
(V f , 96 − V f , 94 )
(5.3)
(V f , 97 − V f , 94 )
(5.4)
c
N M , 95
∑ w (V
e
e , 95
∑ w (V
e
N PG , 94
Vindex,year
c
N M , 95
− Ve ,93 ) +
N PG , 94
N PG , 93
with:
∑w
(Vb , 95 − Vb , 93 ) +
N M , 94
N PG , 93
∆2 = δ2 −
b
N M , 94
∑w
f
N PG , 95
e , 96
− Ve, 93 ) +
∑w
f
N PG , 95
variable used in analysis of bank index in corresponding year
index a = merged banks 1993, a = 1,…,4
index b = merged banks 1994, b = 1,…,4
index c = merged banks 1995, c = 1,…,9
index d = non-merged banks 1993, d = 1,…,421
index e = non-merged banks 1994, e = 1,…,509
index f = non-merged banks 1995, f = 1,…,413
number of merged banks in corresponding year
number of non-merged banks in corresponding year
weight of individual bank, with total assets as weight and
wa + wb + wc = 1 and wd + we + wf = 1
Empirical results
To analyse the effects of mergers both return on equity and return on assets are investigated as
measures of profitability. The efficiency measures used are the estimates of section 4.2. For scale
efficiencies, the relevant issue is whether a bank’s scale efficiency measure moves closer to zero,
either on the positive or the negative side. To analyse this, the absolute value is taken of the scale
efficiency measure, so the scale economies and diseconomies are not treated separately. The change
reflects the movement towards or away from zero, which is an improvement respectively
deterioration. Furthermore, both output prices and the price of deposits are evaluated 13. Besides these
13 The price of labor is not included in this analysis, as it is not related to a bank’s customers. Therefore, there is no
theoretical link between market power and the level of the price of labor.
- 27 -
variables, the loans-to-assets ratio is analysed. This variable is included for the same reason as it is
included in equation (4.1). As a bank becomes larger (through a merger), it may be better capable of
diversifying its portfolio. This allows the bank to issue relatively more loans, which are considered
riskier but also more profitable than securities. A change in this variable may explain possible changes
in profitability.
The effects of mergers on the performance of banks are analysed by taking the absolute as well as the
relative changes for a number of variables, using the year before and the two years after the merger.
The values of these changes are reported in table 5.1. The columns headed by δ1 and δ2 report the
weighted average absolute two- and three-year changes of merged banks and the columns headed by
∆1 and ∆2 give the weighted average relative two- and three-year changes for merged banks, measured
against a peer group of non-merged banks.
Table 5.1 Absolute and relative changes as a result of bank mergers
ROE
δ1
δ2
∆1
∆2
____________ ___________ ____________ ___________
0.3135**
0.3313**
0.3036**
0.3282**
(2.868)
(2.799)
(6.747)
(6.792)
ROA
0.0110**
(2.737)
0.0115**
(2.200)
0.0098**
(6.013)
0.0106**
(4.999)
XE
-0.1052**
(-2.390)
-0.1259**
(-2.522)
-0.0877**
(-4.878)
-0.1089**
(-5.349)
SE
-1.3496E-9
(-0.129)
-1.3993E-9
(-0.136)
-6.739E-8
(-0.14)
-1.4907E-8
(-0.031)
PLoans
-0.0024
(-0.136)
-0.01488
(-1.204)
0.0041
(0.825)
0.0018
(0.358)
PSecurities
-0.005
(-0.499)
-0.0132
(-1.445)
0.0008
(0.195)
-0.0003
(-0.109)
PDeposits
-0.0023
(-0.191)
-0.0105
(-0.747)
0.0101*
(2.065)
0.0065
(1.145)
L/A
0.5589
(0.195)
-0.5570
(-0.156)
4.4762**
(3.775)
2.9887*
(2.045)
Note: Significant differences are marked by * and ** for 10 % and 5 % significance levels, respectively.
T-values are reported in parentheses.
- 28 -
Several results on absolute changes in table 5.1 are of interest. Merged banks have significantly
improved their profitability: both the two-year change and the three-year change are significantly
positive, whether return on equity or return on assets is concerned. Besides the significant absolute
improvement in profitability, the X-efficiency of merged banks has also improved, as X-inefficiency
shows a significant decrease over the years surrounding the merger. Scale efficiency seems to improve
for merged banks, as the average value of the measure moves closer to zero, but these changes are not
statistically significant. All prices, output as well as input, seem to have decreased over the period, but
again none of the changes differ significantly from zero. The loans-to-assets ratio shows no significant
absolute change, so the merged banks do not seem to change the composition of their portfolio as a
result of the merger. This indicates that the often-suggested theory that merged banks are able to issue
more loans relative to securities because of a better-diversified portfolio and thereby improve
profitability fails to receive support from the data.
More interesting than absolute movements are the relative changes of the various variables as a result
of bank mergers. Table 5.1 shows that merged banks significantly improved profitability relative to the
peer group. The increases in both return on equity and return on assets are significantly larger for
merged banks than for non-merged banks. Another important finding is the significant difference in
the change of X-efficiency. For merged banks, the X-inefficiency decreased at a higher rate than for
the peer group. This indicates that by merging, banks succeeded in improving their X-efficiency more
than they would have done without merging. Scale efficiency does not seem to have significantly
improved relative to non-merged banks. Output prices also do not show a different pattern compared
with the peer group, which indicates that merged banks neither use increased market power, nor pass
through any gains realised by the merger to their customers through the prices of loans and securities.
The price of deposits on the other hand shows a significant relative increase for the two-year change,
which disappears over the next year as indicated by the insignificant three-year relative change.
However, the two-year change shows that merged banks lowered their deposit rates less than nonmerged banks did in the year following the merger. This represents a benefit for the suppliers of cash
funds, who receive a relatively higher award for bringing their funds to merged banks. This may point
to an aggressive strategy to hold or attract depositors after the merger. It may also indicate the need for
funds after a merger. The loans-to-assets ratio shows a relative improvement. Merged banks have
succeeded in keeping this ratio stable over the merger process, indicated by the insignificant absolute
change, while non-merged banks have suffered a decrease in the relative amount of loans in their
portfolio. Thus, relatively, merged banks have been able to improve their portfolio diversification,
although in absolute terms this is not the case.
- 29 -
Conclusions
Mergers prove to have significant effects on bank performance. They lead to higher profitability and
also to a significant improvement of a bank’s level of X-efficiency. Scale efficiency seems to be
unaffected by mergers. Prices charged and paid by merged banks in general change in the same
direction and with the same magnitude as those of the non-merged banks, with the exception of the
price of deposits, which seems to remain at a higher level for a short period after the merger. Finally,
merged banks seem better capable of keeping the loans-to-assets ratio at the same level, whereas the
peer group seems to have lowered the relative amount of its loans.
5.2 Merger effects and the profit-structure hypotheses
The previous section established that bank mergers appear to be profitable, already in the first year
after the merger. The present section assesses whether any of the hypotheses explaining the profitstructure relationship can serve as the economic rationale for bank mergers. To this end, the effects of
bank mergers on the variables in the previous section are confronted with predictions of the different
hypotheses. Methodology, empirical results and conclusions are discussed successively.
Methodology
Each hypothesis entails its own expectations about the development of key variables after a bank
merger. To be accepted as a valid explanation of bank mergers, the changes in the variables considered
should comply with the predictions of the hypothesis. First, for each of the hypotheses described in
section 3, their predictions for the qualitative effects of bank mergers are formulated. Thereafter, the
empirical results on bank mergers from the previous section are evaluated to analyse whether any of
the hypotheses is accepted as an economic motivation of mergers.
If the efficient-structure-X-efficiency hypothesis rightly describes the relations that form the
motivation for bank mergers, the following changes should be found when analysing mergers. Most
importantly, mergers should lead to an improvement of X-efficiency. This improvement is supposed to
cause an increase in profitability. Besides these changes, the hypothesis predicts possible changes in
prices. If a bank decides to pass through part of the efficiency gains of the merger to its customers,
output prices should decrease and the rate of deposits, which is regarded as an input price, should
increase. However, a bank is not obliged to pass through any gains, so these prices can also remain
unchanged, without conflicting with the theory of the efficient-structure-X-efficiency hypothesis.
The efficient-structure-scale efficiency hypothesis states that scale efficiencies are the foremost
determinants of profitability. If a merger is motivated by expected gains of this mechanism, scale
efficiency should improve as a result of a merger. This would result in higher profitability and
- 30 -
possibly in prices that are more favourable to customers. So, with regard to profitability and prices, the
expectations are equal to those of the efficient-structure-X-efficiency hypothesis, but here the changes
are caused by an improvement of scale efficiency.
If the structure-conduct-performance hypothesis applies, concentration resembles market power. More
market power enables price setting, which causes output prices to be higher and input prices (i.e.
deposit rates) to be lower. Because of these price changes, banks will become more profitable. As
mergers increase market share and thus concentration, they should lead to higher output prices, lower
input prices and higher profitability in order to accept the structure-conduct-performance hypothesis as
the motivation for mergers.
The relative-market-power hypothesis follows the same line of reasoning as the structure-conductperformance hypothesis, with the difference that market share is the explanatory variable instead of
concentration. Because the expected changes of both these hypotheses are qualitatively the same, it
will be difficult to determine which of the two is dominating the other, if these expectations are met.
However, if this is the case, it is clear that the mechanisms through which mergers improve
profitability have to be sought within the market power category of explanations.
The quiet life hypothesis shows somewhat different expectations than the main market-power
hypotheses. As this hypothesis takes market power as the stimulus of profitability through prices, the
prices are expected to show the same development as under the other market-power hypotheses.
However, as it is assumed that management becomes lax with regard to efficiency, either one or both
efficiencies are expected to deteriorate. As the development of profitability depends upon the changes
in both prices and efficiencies, it is difficult to predict whether profitability will increase, decrease or
remain unchanged.
Table 5.2 contains the qualitative predictions of the hypotheses by summarising which sign the change
in the corresponding variable should have in order to follow the hypothesis in question. It should be
noted that in the table, the predicted changes in XE and SE are the changes in the inefficiency
measure, so that for the efficient-structure hypotheses, this indicates efficiency improvements, whereas
the quiet life hypothesis predicts a deterioration of efficiency.
Empirical results
Table 5.3 presents the empirical effects of mergers on bank performance from section 5.1 in a
qualitative manner. These results can be confronted with the qualitative predictions of the efficientstructure and market-power hypotheses as discussed above. As changes due to mergers are concerned,
the measures that are of main concern are those relative to the peer group of non-merged banks.
- 31 -
Therefore, the discussion of the hypothesis tests focuses on relative changes, whereas absolute
changes are only reviewed for completeness.
Table 5.2 Predicted effects of mergers on bank performance
ESX
ESS
SCP
RMP
Quiet life
ROE/ROA XE
________ _______
+
−
+
+
+
?
+
Pinput
SE
Poutput
_________ __________ __________
0/+
0/−
0/+
0/−
−
+
−
+
−
+
+
−
Table 5.3 Empirical results on qualitative effects of mergers on bank performance
Absolute change (δ )
Relative change (∆ )
ROE ROA XE
SE
PLoans PSecurities PDeposits
_____ _____ _____ _____ _____ ______ ______
+
+
0
0
0
0
−
+
+
−
0
0
0
+/0
Comparing table 5.3 with the hypotheses predictions summarised in table 5.2 leads to the following
conclusions regarding the validity of the profit-structure hypotheses with respect to mergers. The
results of the efficient-structure-X-efficiency hypothesis correspond quite well with the predictions.
Both profitability and X-efficiency change in the right direction, both when relative and when absolute
changes are considered.
The efficient-structure-scale efficiency hypothesis is not supported by the data, as scale efficiency fails
to improve because of a merger. Although all other variables show changes that match the predictions,
the failure to improve scale efficiency leaves the theory meaningless. So, based on these findings, the
efficient-structure-scale efficiency hypothesis is rejected as an explanation of merger activity in the
European commercial banking sector.
The two general market-power hypotheses predict the same changes. The basic assumption of these
hypotheses is that merged banks use gained market power to set prices, with the intention to increase
profitability. The prices, however, do not show any significant changes, except for the two-year
relative change of the price of deposits, which has the wrong sign. With the absence of price setting by
merged banks, the causality that is supposed to run from increased market power through price setting
- 32 -
to higher profitability, is rejected. The denial of this causality leads to the rejection of both marketpower hypotheses as possible theories explaining mergers.
The acceptance of the quiet life hypothesis, being a special case of the market-power hypothesis,
requires the existence of the same relationship between market power and price setting, only allowing
for profitability to show no improvement. Together with the signs of price setting, a deterioration of
efficiency is expected. Since neither price setting nor the deterioration of efficiencies are supported by
the data, the quiet life hypothesis is rejected as well.
Conclusions
All in all, this leads to following conclusions with regard to the economic motivation of bank mergers.
The empirical evidence corroborates the efficient-structure-X-efficiency hypothesis as a valid theory
explaining the economic rationale behind a bank merger, as bank mergers in the data set lead to
improved profitability through increased X-efficiency. The efficient-structure-scale efficiency
hypothesis is rejected because the merged banks fail to improve their scale efficiency. All marketpower hypotheses, including the quiet life hypothesis are rejected as well because the data shows no
evidence of merged banks using their gained market power to set prices at a level that is favourable to
their profitability.
- 33 -
6 SUMMARY AND CONCLUSIONS
These days, there is a strong, ongoing tendency in the banking sector towards consolidation all over
the world. In Europe this trend of restructuring is boosted by monetary unification. The rapid changes
in the financial services industry bring about questions about the consequences of these developments
for customers and the relationship between the structure of the banking industry and bank profits. This
study has addressed three related issues for the banking sector of Europe: existence of a profitstructure relationship in the European banking sector, explanations for this relationship and possible
economic rationales for bank mergers through their effects on bank performance. Our analysis is based
on a sample of commercial banks in eight European countries, using an unbalanced panel data set over
the period 1992 – 1997.
Empirical evidence suggests the existence of a profit-structure relationship in the European
commercial banking sector. This relationship is either based on individual market share or on the level
of market concentration, depending on the measure of profitability used (return on equity or return on
assets). Notably, the relationship between return on equity and individual market share shows
convincing evidence on the existence of the profit-structure relationship.
Several theories explaining the profit-structure relationship in the European banking sector are tested.
The tests are categorised into efficient-structure and market-power hypotheses. The main controversy
between these hypotheses concerns the role of efficiency. Efficient-structure hypotheses state that
efficient banks are able both to collect higher profits and to increase their market share. This then
underlies the profit-structure relationship. On the other hand, market-power hypotheses claim that a
high market share or market concentration enables banks to raise profits through price setting. The
empirical evidence documented in the present study provides strongest support for the efficientstructure-X-efficiency hypothesis. This suggests that X-efficiency, often interpreted as a measure of
management quality, is the dominant factor underlying the profit-structure relationship. In order to test
the hypotheses, X-efficiencies and scale efficiencies of individual banks have been estimated based on
a profit function approach. Estimated X-inefficiency levels seem to be quite large compared to realised
profitability ratios. For scale inefficiencies, a local estimate is used, which does not allow for such an
interpretation. It seems plausible, though, that the loss of profitability caused by scale inefficiencies is
smaller than that caused by X-inefficiencies.
The third issue of our analysis concerns the effects of mergers on bank performance. The empirical
evidence reveals that merged banks significantly improve their profitability as well as their Xefficiency, both absolutely and compared to non-merged banks. Scale efficiency does not improve and
price levels generally do not change relative to non-merged banks. Another interesting result is that
- 34 -
merged banks show a relative improvement of the share of loans in their portfolio, indicating a relative
improvement of risk diversification for merging banks. Confrontation of the qualitative effects of
mergers on bank performance with the expected effects based on the different efficient-structure and
market-power hypotheses reveals that predicted and actual effects match best for the efficientstructure-X-efficiency hypothesis, since merged banks significantly improve their level of both Xefficiency and profitability. At the same time, no evidence of improved scale efficiency nor of price
setting is found, leaving the other hypotheses unsupported. This suggests that the efficient-structure-Xefficiency hypothesis may provide an economic rationale for bank mergers.
The existence of the profit-structure relationship, its explanation and the motivation of bank mergers
also have implications for antitrust legislation. Antitrust legislation aims at protecting consumer
interests by controlling concentration movements. Increasing market shares and market concentration
enlarge the market power of banks, which they might use to increase profitability through price
setting. According to this study there is at best weak evidence of price setting by banks and there is no
empirical evidence of unfavourable price setting behaviour after bank mergers. On the other hand,
there is also no evidence of merged banks passing through part of the gains in X-efficiency they
realise. All in all, the findings suggest that antitrust legislators should be careful when considering a
merger proposal, since disapproving mergers can leave potential social gains unrealised. It seems
plausible, though, that there is a level of concentration, above which a further increase will limit
competitive forces in such a way that it becomes a threat to social welfare. It is hard to determine at
what level this might happen, especially because it does not seem to have been reached yet in any of
the countries investigated. This however is hard to determine because antitrust legislation has already
been active over the period considered. Therefore, it is possible that proposed mergers between large
banks that would have had a large impact on market concentration have been prohibited or that
existing antitrust legislation discouraged large banks from attempting to merger. Since the field of the
profit-structure relationship in the European banking sector is relatively unexplored, our empirical
findings may gain weight if the results are corroborated by other studies investigating different data
sets. This is especially the case because the availability and quality of European bank data lags behind
American bank data, as the standard regulatory demands on listed companies are weaker in Europe
than in the United States.
- 35 -
REFERENCES
Aigner, D., K. Lovell and P. Schmidt, 1977, Formulation and Estimation of Stochastic Frontier
Production Function Models, Journal of Econometrics, Vol. 6, 21–37.
Akhavein, J.D., A.N. Berger and D.B. Humphrey, 1997, The Effects of Megamergers on Efficiency
and Prices: Evidence from a bank profit function, Review of Industrial Organization, Vol. 12, 95–139.
Allen, L. and A. Rai, 1996, Operational Efficiency in Banking: An international comparison, Journal
of Banking and Finance, Vol. 29, 655–672.
Altunbas, Y. and P. Molyneux, 1994, The Concentration-Performance Relationship in European
Banking: A Note, Research Papers in Banking and Finance, RP 94/12, Institute of European Finance,
Working paper.
Berger, A.N., 1995, The Profit-Structure Relationship in Banking – Tests of Market-Power and
Efficient-Structure Hypotheses, Journal of Money, Credit and Banking, Vol. 27, 404–431.
Berger, A.N., 1998, The Efficiency Effects of Bank Mergers and Acquisition: A preliminary look at
the 1990s data, in: Amihud, Y. and Miller, G. (eds.), Bank Mergers & Acquisitions, Kluwer Academic
Publishers.
Berger, A.N., D. Hancock and D.B. Humphrey, 1993, Bank Efficiency Derived from the Profit
Function, Journal of Banking and Finance, Vol. 17, 317–347.
Berger, A.N. and T.H. Hannan, 1997, Using Efficiency Measures to Distinguish Among Alternative
Explanations of the Structure-Performance Relationship in Banking, Managerial Finance, Vol. 23,
6–31.
Berger, A.N. and T.H. Hannan, 1998, The Efficiency cost of Market Power in the Banking Industry:
A test of the ‘Quiet Life’ and related hypotheses, Review of Economics and Statistics, Vol. 80,
454–465.
Berger, A.N. and D.B. Humphrey, 1997, Efficiency of Financial Institutions: International survey
and directions for future research, European Journal of Operational Research, Vol. 98, 175–212.
Berger, A.N., W.C. Hunter and S.G. Timme, 1993, The Efficiency of Financial Institutions: A
review and preview of research past, present, and future, Journal of Banking and Finance, Vol. 17,
221–249.
Berger, A.N. and L.J. Mester, 1997, Inside the Black Box: What explains differences in the
efficiencies of financial institutions?, Journal of Banking and Finance, Vol. 21, 895–947.
Bikker, J.A. , 1999, Efficiency in the European Banking Industry: an explanatory analysis to rank
countries, DNB Staff Reports, 42.
Coelli, T., 1996, A Guide to FRONTIER Version 4.1: A Computer Program for Stochastic Frontier
Production and Cost Function Estimation, Working Paper No. 96/07, Centre for Efficiency and
Productivity Analysis.
Greene, W.H., 1993, The Econometric Approach to Efficiency Analysis, in: Fried, H.O., Lovell, K.
and Schmidt, S.S. (eds.), The Measurement of Productive Efficiency: Techniques and Applications,
Oxford University Press.
- 36 -
Goldberg, L.G. and A. Rai, 1996, The Structure-Performance Relationship for European Banking,
Journal of Banking and Finance, Vol. 20, 745–771.
Moulyneux, P. and J. Thornton, 1992, Determinants of European Bank Profitability: A Note,
Journal of Banking and Finance, Vol. 16, 1173–1178.
Pastor, J.M., F.Pérez and J. Quesada, 1997, Efficiency Analysis in Banking Firms: An international
comparison, European Journal of Operational Research, Vol. 98, 395–407.
Schure, P. and Wagenvoort, R., 1999, Who Are Europe’s Efficient Bankers?, EIB Papers, Vol. 4,
105–126.
Stevenson, R.E., 1980, Likelihood Functions for Generalized Stochastic Frontier Estimation, Journal
of Econometrics, Vol. 13, 57–66.