1996 JBF Goldberg - The Structure-Performance Relationship For European Banking
1996 JBF Goldberg - The Structure-Performance Relationship For European Banking
1996 JBF Goldberg - The Structure-Performance Relationship For European Banking
BANKING &
ELSEVIER FINANCE
Journal of Banking & Finance 20 (1996) 745-771
Abstract
The relationship between market structure and performance has been studied extensively
for American banking. In contrast, little work has been done to investigate this relationship
for European banking. Two explanations of a positive correlation between profitability and
concentration have been advanced, the traditional structure-performance hypothesis (SCP)
and the efficient-structure hypothesis. Previous empirical tests of the alternative hypotheses
have yielded mixed results but the tests were not robust because they did not incorporate
measures of efficiency directly in the model. This study applies a stochastic cost frontier as
proposed by Aigner et al. (1977) to derive measures of X-inefficiency and scale-ineffi-
ciency, under the assumption that the errors are distributed half-normal. We incorporate
these measures of inefficiencies directly into the tests as proposed by Berger and Hannan
(1993). We do not find a positive and significant relationship between concentration and
profitability for a sample of banks across 11 European countries over a four year period,
1988-91. However, we do find evidence to support one of the two versions of the
efficient-structure hypothesis for banks located in countries with low concentration of
banks. Since little support is found for either of the SCP hypotheses, a simple policy of
strict limitations on cross-border acquisitions and growth is not warranted.
1. Introduction
especially among the larger banks. On the other hand, if the alternative EFS is
supported, no application of antitrust measures is required.
The next section surveys the previous attempts to test the structure-perfor-
mance relationship in banking, including the limited research on European banks.
The Berger and Hannan (1993) model, which incorporates efficiency directly in
the tests, is discussed in Section 3. Section 4 discusses the data and defines the
explanatory variables, including the efficiency variables. Unlike Berger and Han-
nan (1993), who use deviations from the average residual over a time horizon to
estimate measures of efficiency, this paper first estimates deviations from a
stochastic cost frontier, as developed by Aigner et al. (1977), which assumes the
errors are distributed half-normal. The error terms are then used to obtain
measures of X-inefficiency for each bank, using the method proposed by Jondrow
et al. (1982). The stochastic cost frontier is also used to obtain measures of
scale-inefficiency. Section 5 presents the results of the tests using the derived
inefficiency variables. Section 6 summarizes and presents the implications of the
results for European banking in its new regulatory environment.
2. Previous literature
The bulk of the studies examining bank performance and market concentration
have been on U.S. banking markets. In contrast to the national markets used in
industrial studies, banking markets that are examined for U.S. studies are generally
local markets which service individuals and small businesses. The banking studies
have not found a positive relationship between concentration and profitability as
consistently as has been found in the inter-industry studies. Although Rhoades
(1982) concluded that most of the previous studies found a positive relationship,
Gilbert (1984), in a survey article, finds only 27 of 56 studies to have produced the
expected relationship. Osborne and Wendel (1982) also conclude that problems
with past studies are so great that there is insufficient evidence of a positive
relationship.
The main problem has been the interpretation of the positive relationship
between profitability and concentration (when it can be found), and whether it
supports the SCP or EFS hypothesis. Several methods have been proposed to
resolve this issue. For example, Smirlock (1985) models bank profitability as a
function of market share, concentration, and an interaction term between market
share and concentration (as well as several control variables). He finds that market
share is positively related to profitability while there is no relationship between
concentration and profitability, and concludes that this provides evidence in favor
of the efficient-structure hypothesis. As mentioned earlier, Shepherd (1986) ques-
tions this conclusion because it assumes that market share is a proxy for efficiency
of larger finns rather than a measure of market power. 1
Berger and Hannan (1989) try to provide a cleaner test by using price data
748 L.G. Goldberg, A. Rai /Journal of Banking & Finance 20 (1996) 745-771
rather than profit data as the dependent variable. Since the SCP hypothesis implies
that consumers will be treated less favorably in more concentrated markets, they
examine whether retail deposit rates are negatively related to market concentration.
They find evidence to support the traditional SCP hypothesis. Even though they
question the appropriateness of including market share in their model because of
its endogeneity, market share does have a positive coefficient when included in the
model. The authors say that this result may be due to firms providing different
levels of quality of service or to offering higher deposit rates, allowing banks to
increase their market share. However, the coefficient of concentration still remains
negative.
Jackson (1992) questions whether these results validate the traditional hypothe-
sis by dividing the sample into three ranges of concentration. He finds that the sign
and significance of the coefficient of concentration differ by sub-sample, suggest-
ing a non-linear relationship between concentration and price. Only for the lowest
concentration sub-sample does he find the negative relationship between deposit
rates and profitability needed to validate the traditional hypothesis. He asserts that
these results support the efficient-structure hypothesis by arguing that levels of
concentration signal an optimal structure. Jackson's (1992) results also contradict
an earlier study by McCall and Peterson (1980) who find the relationship holds in
high concentration rather than in low concentration markets. Berger and Hannan
(1992) in a reply find that some of Jackson's results are not robust, but they do
agree that the relationship varies for different concentration levels and for different
time periods. Consequently, our study of European banks examines the time
period 1988-91, and splits the banks into two sub-samples, high concentration and
low concentration.
Studies using international databases are limited with the majority supporting
the SCP hypothesis. Ruthenberg (1994), using survey data from 1984-88, finds
concentration increases profitability, especially if barriers to entry are high.
Molyneux and Teppet (1993) examined the SCP hypothesis for 5 EFFA countries
(Sweden, Norway, Finland, Austria and Switzerland) and find support for the SCP
hypothesis. Molyneux (1993) also finds similar results for banks located in
Portugal, Spain, Sweden, UK and Turkey. Lloyd-Williams et al. (1994) also find
support for the SCP hypothesis for Spanish banks for the period 1986-88. Except
for Vander Vennet (1993), none of the above papers incorporates efficiency
measures directly in the model. Vander Vennet's (1993) results indicate that in
some European countries, Belgium, Ireland, Portugal and Spain, collusion appears
to be predominant.
This paper incorporates two measures of efficiency into the model, X-inef-
ficiency and scale-inefficiency. But unlike Berger and Hannan (1993) who use
deviations from the average cost frontier to represent measures of inefficiency, this
paper estimates X-inefficiency using a stochastic cost frontier under the assump-
tion that the error terms are distributed half-normal. As will be explained later, in
empirical testing it is more convenient to substitute estimates of inefficiency as
opposed to estimates of efficiency. Similarly, scale-inefficiency is also derived
from the parameters of the estimated stochastic cost function. An advantage of this
method is that X-inefficiency is strictly defined as the systematic deviation away
from an efficient cost frontier which is not fixed but varies for each bank.
4 Berger (1995) and Berger and Hannah (1993) also used individual price data to supplement their
results which we omit because, as will be explained later, we use net interest margin as a proxy for the
ability of banks to affect prices.
5 Berger and Halman (1993) specify the second equation as CONC m = f(EFF-Xi,EFF-Si, Zi, el) ,
with the error term multiplicative rather than additive. This is because if the Herfindahl Index is used as
a measure of concentration, then CONC,n will be the sum of squared MSi, over all i in m. We ignore
this specification because in the empirical testing, the results using CR3 and HERF are similar, making
the specification of the error term irrelevant.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 751
is strictly valid if it can be established that more efficient firms are profitable [Eq.
(1)] and more efficient firms will have larger market shares [Eqs. (2) and (3)].
On the other hand, if either of the market power hypotheses holds true, i.e. SCP
or RMP, then the expected signs of the coefficients for Eq. (1) are: CONC > 0 or
MS > 0. The CONC variable has a positive coefficient if the SCP holds and MS is
positive if the RMP hypothesis holds. Thus, the difference between the SCP and
RMP is that in the case of the latter, anti-competitive advantages due to size can
exist even in markets that are not concentrated.
Additional relationships are tested to support the market power hypothesis:
X-EFFi = f ( C O N C m, MS/, Zi) -Jr ei, (4)
S-EFFi = f ( C O N C m , MS/, Zi) + e i. (5)
Berger and Hannan (1993) refer to these conditions as testing Hicks (1935) 'quiet
life' hypothesis. This hypothesis predicts a reverse causation, that is, as firms
enjoy greater market power and concentration, inefficiency follows not because of
non-competitive pricing but more so because of a relaxed environment that
produces no incentives to minimize costs. An example would be in small
concentrated rural markets where price competition still exists because of spatial
competition but where cost efficiency is unimportant.
This paper applies the above methodology to test the structure-performance
paradigm for European banks over a four-year period. It differs from the Berger
and Hannan (1993) methodology in that it uses different measures of X-efficien-
cies and scale-efficiencies. The efficiency measures in this paper are derived from
a stochastic cost frontier as developed by Aigner et al. (1977), which assumes that
the error terms are distributed half-normal. This is the first paper to test the
structure-performance paradigm using such efficiency measures for European
banks. The results will be compared to the conclusions reached by other re-
searchers such as Molyneux (1993), Molyneux and Teppet (1993) and Vander
Vennet (1993).
The dearth of SCP studies for European banking may partially be explained by
the non-availability of data for most of the countries. This trend has changed with
the advent of several new international databases. This study uses Compustat's
Global Vantage database, which contains data for the largest banks in Europe.
Differences in accounting practices and flexibility in reporting makes comparison
a difficult task but Global Vantage has managed to standardize most of the
relevant data.
The paper only examines the large banks in each country. Since the European
banking industry is dominated by large banks with branching networks spread
752 L.G. Goldberg, A. Rai /Journal of Banking & Finance 20 (1996) 745-771
across the country, it is unlikely that individual branches have the ability to
significantly affect any prices, either loan or deposit rates. Within a country, it is
more likely for the corporate headquarters to determine the pricing and cost
strategies for the whole country, which are heavily influenced by the bank's ability
to borrow long- and short-term funds in the wholesale market. Thus, we rule out
using data of individual bank branches or very small banks. Secondly, since the
regulatory authorities are concerned with the effects of market concentration on
consumer welfare, mergers between the large banks are of primary concern to
them. With the EC having substantial power to make policies to curtail cross-coun-
try merger activities, evidence supporting the SCP will buttress antitrust sentiment
against acquisitions by large banks.
A brief description of the variables are given below while a more complete
description is provided in Appendix A. Appendix B provides a list of the names of
banks included in the 1990 data. The sample of banks differs slightly for the other
years.
Table 1
Measures of market concentration of commercial and savings banks for 11 European countries
Commercial banks only Commercial + savings banks
Country N CR3 HERF Rank N CR3 HERF Rank
UK 68 0.60 0.15 1 135 0.50 0.12 1
Germany 121 0.31 0.06 2 214 0.26 0.04 2
Fr~mce 119 0.51 0.12 1a 151 0.49 0.11 2
Swiss 118 0.63 0.14 1 170 0.57 0.12 1
Belgium 35 0.71 0.22 l 42 0.57 0.13 I
Finland 7 0.91 0.29 1 7 0.91 0.29 1
Austria 39 0.49 0.11 2 50 0.41 0.08 2
Italy 41 0.33 0.07 2 58 0.28 0.06 2
Sweden 5 0.70 0.23 1 5 0.70 0.23 1
Denmark 26 0.85 0.30 1 31 0.72 0.2 l 1
Spain 59 0.45 0.09 2 125 0.27 0.04 2
CR3 = three-bank concentration ratio. HERF = Herfindahl Index, defined as the sum of the squares of
market shares of total deposits. 'Commercial banks only' includes foreign commercial banks. Year =
1990 (estimates for 1988, 1989 and 1991 are qualitatively similar to 1990). Source: Sheshunoff's
Information Service.
Cutoff for rankings: countries with CR3 of 0.50 and HERF of 0.12 were classified as having high
concentration of banks, else as having a low concentration.
a France has a ranking of 1 if only commercial banks are examined and a ranking of 2 if all savings
banks are included. Tests were conducted using France with both rankings and the results are
qualitatively similar.
Since banks branch throughout each country unlike in the United States and since
entry has until recently been restricted by national borders, a national market is
appropriate. Moreover, it is not possible to obtain market shares by local market
for European countries.
Since our database consists of only the largest banks in each country, the
estimation of CR3 and HERF is not possible using the Global Vantage data.
Consequently, customized data were purchased from Sheshunoff's Investment
Services which provided deposit data for all banks, including non-commercial and
foreign banks, for each country. Table 1 provides the estimates of the three-bank
concentration ratio (CR3) and the Herfindahl index (HERF) for each of the
countries in 1990. Columns 3 and 4 provide the estimates of CR3 and HERF using
only commercial banks while columns 7 and 8 provide the estimates for commer-
cial and savings banks. As Table 1 shows, some countries have a significant
number of savings banks. For example, in the case of Germany, there were data
available for 214 commercial plus savings banks of which only 121 were
classified as commercial banks (the latter includes foreign commercial banks). 6
6 Sheshunoff's Investment Services did not include any data on non-commercial banks for two
countries, Finland and Sweden.
754 L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771
However, and this will be elaborated upon next, the ranking of banks into high and
low concentrated countries (columns 5 and 9) is mostly unaffected by whether or
not savings banks are included. Consequently, subsequent analyses were per-
formed using concentration ratios based only on commercial banks.
The estimates of CR3 range from 31% to 91% when only commercial banks are
examined and range from 26% to 91% when savings banks are included in the
sample. The comparable numbers for HERF are between 0.06 to 0.30 for
commercial banks and 0.04 to 0.29 when savings banks are included. The ratios
were similar for the other three years. In order to distinguish between countries
with high and low concentrations of banks, a cutoff of 0.50 for CR3 and 0.12 for
HERF was used to rank banks. Thus, banks in countries with CR3 of 0.50 and
above or with HERF of 0.12 or above were ranked as high concentration
countries. The only border-line case is France which would be classified as a high
concentration market if the sample included only commercial banks and low
concentration market if all savings banks are included. In the results presented in
the following tables, we count UK, Switzerland, Belgium, Finland, Sweden and
Denmark as having high concentration of banks and the rest as having low
concentration. The same analysis using France as a high concentration country did
not affect the results significantly. Since the empirical results are similar employ-
ing either HERF or CR3, we only report results using HERF.
4.3.1. X-efficiency
X-efficiency provides a measure of how effectively banks are using their inputs
to produce a given level of output. Since efficient cost frontiers can vary for banks
in different countries, it is necessary to specify a stochastic cost function. This
paper uses the stochastic cost frontier proposed by Aigner et al. (1977) to
investigate cost efficiencies. This method has been used by Mester (1993) and
Cebenoyan et al. (1993) for the savings and loan industry in the U.S. The basic
model assumes that total cost deviates from the efficient cost frontier by a random
noise, vi, and an inefficiency component, u i. Thus, the efficient cost frontier is
defined as:
lntc = f ( Yi, Pi) + si
where ei = ui + vi and Yi is the output i of each bank, Pi is the cost or price of
input i, vi is statistical noise distributed normal (0,o-2), and u~ is one-sided
inefficiency measure, distributed half-normally. Here u~ represents the individual
firm's deviations from the efficient cost frontier and serves as a proxy for both
technical and allocative efficiency. The log-likelihood function is given by:
N 2 1 N N r[EiA~I
in L = _2 In --'n"_ N In o - - 20"2i=1]~ ltJJ
E"z + iE In ~O --o. (6)
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 755
where o. 2 = [o.2 + o.v2], A = o.,/o.v, and 4'(') is the standard normal density
function. Inefficiency measures are derived for each of the four years for each
bank.
In order to estimate the frontier, it is necessary to specify a cost function. This
paper uses a standard translog cost function because of its flexibility in allowing
the estimation of scale efficiencies. This paper uses two outputs, loans as the
primary output (y~) and all other earning assets as the secondary output (Y2), and
three inputs with prices defined as the price of labor (p~), capital (P2) and
borrowed funds (p3).
2 3 2 2
lntc = n 0 + Y'~ai In(Yi) + ]~_,¢ijIn(pj) + 1 / 2 ~ Y'~ c~ik ln(Yi) ln(y~)
i=1 j=l i=lk=l
3 3 2 3
+ 1/2Y'. ~ / 3 j h ln(pj) ln(ph) + Y'~ ]~_,6ijln(Yi) ln(pj) +e (8)
j=lh= 1 i~lj=l
where tc is total operating and interest costs, y~ is total loans, Y2 is all other
earning assets, pt is the price of labor, defined as staff expenses divided by the
number of employees, P2 is the price of fixed capital, defined as capital and
occupancy expenses divided by fixed assets, and P3 is the price of borrowed
funds, defined as total interest expenses divided by interest bearing liabilities.
Unlike in the estimation of traditional cost functions, the share equations are
not included when estimating stochastic cost functions. 7 In addition, the linear
homogeneity conditions are imposed by normalizing total costs (TC), price of
labor (p~) and price of capital (p2) by the price of deposits (p3) (see Cebenoyan
et al. (1993) for details). The usual symmetry restrictions are imposed, i.e.
O~ik= Olki and [~jh = [~hj" Maximum likelihood estimation techniques are used to
estimate the coefficients.
The empirical tests substitute inefficiency variables (X-INEFF) for efficiency
variables (X-EFF), defined in Eq. (1). The estimate of X-INEFF, derived from the
stochastic cost frontier, represents an inefficiency measure for each bank. Conse-
quently, the coefficient of X-INEFF in the regression of the efficient-structure
hypothesis will have the opposite sign as X-EFF specified in Eqs. (1), (2) and (3).
Therefore, the predicted sign of the coefficient of X-INEFF is negative when
7 See B a u e r (1990) for a discussion of the p r o b l e m s in relating the inefficient cost measure with the
share equations.
756 L.G. Goldberg, A. Rai /Journal of Banking & Finance 20 (1996) 745-771
either ROE or ROA is the dependent variable, i.e. the lower the inefficiency, the
larger the profits. When NIM is the dependent variable, the relationship is
positive, i.e. the greater the inefficiency the larger the net interest margin.
SCALE measures are estimated for each bank at its respective output levels, yj
and Y2- If SCALE < I then banks are operating below optimal scale levels and
have the ability to lower costs by increasing output further, while if SCALE > 1
then banks are required to downsize in order to achieve optimal input combina-
tions. Since SCALE > 1 and < 1 both imply inefficiencies, a measure of ineffi-
ciency, S-INEFF, is used in the actual regressions, i.e. 8
S-INEFF=SCALE- 1 if SCALE> 1
and
S - I N E F F = 1 - SCALE if S C A L E < 1.
As in the case of X-INEFF, the sign of the coefficient will be opposite to that
predicted in Eq. (1). For example, the predicted relationship between profitability
and S-INEFF is negative, i.e. the further a bank is from efficient scale, the lower
the profitability. When NIM is the dependent variable, the relationship is positive,
i.e. the larger the S-INEFF, the larger the net interest margin.
The four control variables in Z i used in this study were selected from those
used in previous studies and are detailed in Appendix A. PCI or per capita income
of a country affects numerous factors related to the supply and demand for loans
and deposits. In this paper it is hypothesized that the coefficient will be negative
because countries with higher PCI are assumed to have a banking system that
8 Berger and Hannan(1989), for example, used the two measures separately, S-EFF e if SCALE < 1
and S-EFF d if SCALE > 1. However, this paper assumes that as banks deviate from efficient scale, the
impact of inefficiency is similar, i.e. banks profitability increases monotonically whether they are
downsizing or increasing in size towards the optimal scale.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 757
9 The log of assets is used instead of assets in order to reduce the scale effect.
~o See Vander Vennet (1993), p. 11.
~ It should be noted that a number of banks had to be dropped from the original sample because of
the specific data required to estimate the efficiency measures. For example, Deutsche Bank did not
provide the labor costs for the estimation of the translog function. Similarly, banks in the Netherlands
do not report interest expense separately as is required for the translog, but rather only report net
interest margin.
oo
Table 2
Summary statistics of data for 1990
Country N Total Total deposit Total country X-INEFF Mean Per LA/TA Mean wage .~
assets ($ MIL) deposits S-INEFF net capita ($)
($ MIL} (N) marg. income
e~
($ lOO0)
United Kingdom 8 972932.7 791 845.0 855849.1 (68) 0.076 0.264 0.023 16248 0.949 33707 :~
Germany 7 647934.3 361 707.6 1210035 (121) 0.039 0.275 0.017 22632 0.963 56563 ~.
France 6 765388.5 514 180.2 1 164741 (119) 0.068 0.367 0.032 19535 0.952 62501 "~
Switz. 12 542735.8 392 149.2 457 177.3 (118) 0.081 0.230 0.014 34472 0.925 69925
Sweden 4 160579.0 111 470.2 122336.1 (5) 0.082 0.243 0.022 23987 0.982 61 599
Denmark 4 148202.9 107 861.8 (53052.4) a 60681.5(26) 0.0360.237 0.026 23364 0.946 43305
Belgium 1 73 144.9 54477.1 (32598.6) a 150505.2 (35) 0.033 0.222 0.015 19083 0.973 59210
Finland 5 124833.8 68905.5 (28901.4) a 37301.3 (7) 0.048 0.221 0.019 23757 0.941 48835
Austria 5 91 375.0 68660.1 154664.3 (39) 0.058 0.187 0.020 19060 0.948 49352 ~
Italy 15 454876.3 343931.1 (190243.6) a 731 262.1 (41) 0.082 0.335 0.030 17500 .940 73632
Spain 12 330 784. I 268 307.9 (185 783.1) a 275 698.0 (59) 0.038 0.267 0.048 I 1 861 .939 46 246 ~"~
t5
MIL = millions. 'Total assets' and 'Total deposits' refer to the total of only the subset of banks used in the regression analysis. 'Total country deposits' refers
to the total deposits of all commercial banks in a country as reported by Sheshunoff's Information Services Inc. N = number of banks used in estimating total
deposits of banks (and for estimating three-bank concentration ratio (CR3) and Herfindahl Index (HERF) in Table 1). X-INEFF and S-INEFF refer to ,~
X-inefficiency and scale-inefficiency, respectively. Mean net marg. = average net interest margin defined as net interest income over total assets, as reported
by the banks, of all the banks in a country. Mean L A / T A = average total liabilities over total assets of banks in the country. Mean wage = average total
wages and salaries divided by total employees, as reported by the banks. I
Sources: Compustat's Global Vantage, Sheshunoff's Information Service Inc. and International Financial Statistics. All data converted to U.S. dollars using
year-end rates reported in the Wall Street Journal.
a The data for 'Total country deposits' does not include inter-bank deposits.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 759
measures for each country were estimated using this larger set of banks provided
by Sheshunoff's Information Services. For three of the countries, Denmark,
Belgium and Finland, total deposits of the banks used in the study (column 4)
exceed the total banking deposits of the country (column 5). This is because
Sheshunoff's database does not include inter-bank deposits in their total deposits.
To provide meaningful comparisons, the total deposits without inter-bank deposits
are listed in parentheses in colunm 4 for all 5 of the countries for which these data
were available. ~2
Column 6 provides the average measures of X-inefficiency and scale-inef-
ficiency of banks within each country. Banks in Germany, Denmark, Belgium and
Spain are operating with the smallest deviation from the efficient cost frontier
(X-efficient), while banks in Italy and France are operating the furthest from the
optimal scale (scale-inefficient). Mean net interest margins in column 7 show that
the average spread is high in Spain and France while it is very competitive in
Germany, Switzerland and Belgium.
5. Empirical results
~2 It should be noted that column 5 is used strictly to calculate concentration ratios and market
shares. Sheshunoffs database, although not complete, provides a much larger set of banks, which is
important for the measurementof HERF,
Table 3
Regression results of return on equity (ROE) and net interest margin (NIM) on the Herfindahl index (HERF), market share (MS), X-inefficiency (X-INEFF)
scale-inefficiency (S-INEFF) and other control variables
ALL HC LC ALL HC LC
N 303 133 170 303 133 170
INT 0.623 (3.10) a 0.237 (0.49) 0.493 (2.06) a 0.249 (10.1) a 0.082 (2.43) a 0.182 (4.28) a
HERF - 0 . 1 1 8 ( - 1.54) - 0 . 4 2 4 ( - 2-80) a - 0 . 2 3 0 ( - 1.08) - 0 . 0 1 6 ( - 1.69) - 0 . 0 1 1 ( - 1.04) 0.028 (0.73)
MS 0"187 (3"19) a 0"288 (3"18) a - 0 " 2 2 0 ( - 2"1 I) a 0.018 (2.58) a 0.008(1.19) - 0.037 ( - 2.02) a e,
X-INEFF 0.007 (0.15) 0.004 (0.03) - 0.013 ( - 0.33) 0.033 (5.73) a 0.024 (2.67) a 0.033 (4.48) a
S-INEFF - 0 . 0 1 6 ( - 0.93) - 0.195 ( - 0.93) 0.079 (0.58) 0.011 (0.80) 0.004 (0.28) 0.084 (3.48) a
WAGE -0.000006(-0.02) 0.0002(0.30) -0.0004(-1.03) - 0.0002 ( - 4.07) a _ 0.0003 ( _ 4.24) a _0.0003(_3.71) a
LTA - 0.005 ( - 0.96) - 0.022 ( - 1.98) a 0.02 (3.87) a - 0.002 ( - 4.03) a 0.0006 ( - 0.80) 0.0009 (0.95)
RISK - 0 . 4 1 6 ( - 1.96) a 0.382(0.94) - 0 " 5 5 8 ( - 2"24) a -0'171 (-6.59) a -0.036(-1.21) -0.141 (-3.17) a
PCI -0'003(-3-09) a -0.003(-1.00) -0.011 (-7.32) a _0.001 ( _ 10.9) a _0.0004(_2.06) a _ 0.002 ( _ 8.55) a
YR89 -0.011 (-0.97) -0.041 (-2.15) a 0.00007 (0.007) -0.003 (-2.04) a -0.003 (-2.20) a -0.004(- 1.95) a
YR90 - 0.011 ( - 0.92) - 0.046 ( - 2.04) a 0.028 (2.49) a 0.005 (3.34) a 0.003 (1.75) b 0.007 (3.34) a
YR91 - 0.006 ( - 0.52) - 0.032 ( - 1.39) 0.033 (2.70) a 0.004 (2.70) a 0.003 ( 1.73) b 0.007 (3.42) a
R2 0.08 0.104 0.346 0.58 0.52 0.64
F 3.48 a 2.39 a 9.14 a 39.1 a 14.0 a 27.8 a
t,o
ALL = all banks in the sample; HC = banks located in countries with high market concentration (Rank 1 in Table 1); LC = banks located in low market
concentrations (Rank 2 in Table I). INT = intercept, MS = market share, X-INEFF = X-inefficiency, S-INEFF = scale-inefficiency, W A G E = average
wages and salary, LTA = natural log of total assets, RISK = total liabilities over total assets, PCI = per capita income, YR89, YR90 and YR91 = dummies
for 1989, 1990 and 1991 with 1988 serving as the base year. Dep var = dependent variable. Source: Compustat's Global Vantage and International Financial
Statistics. f
"-4
a, b Significant at the 5% and 10% levels, respectively.
Table 4
Regression results o f market share (MS) and market concentration ( H E R F ) on X-inefficiency (X-INEFF), scale-inefficiency ( S - I N E F F ) and other control
variables
Dep var = MS Dep vat = HERF
ALL HC LC ALL HC LC
A L L = all banks in the sample; HC = b a n k s located in countries with high market concentration ( R a n k l in Table 1); L C = b a n k s located in low m a r k e t
e~
concentrations ( R a n k 2 in Table 1). INT = intercept, W A G E = average w a g e s and salary, L T A = natural log o f total assets, R I S K = total liabilities o v e r
total assets, PCI = per capita income, YR89, Y R 9 0 and YR91 = d u m m i e s for 1989, 1990 a n d 1991 with 1988 serving as the base year. Years = 1 9 8 8 - 9 1 . ~
Source: C o m p u s t a t ' s Global Vantage and International Financial Statistics.
~,b Significant at the 5 % and 10% levels, respectively.
b~
Table 5
Regression results o f X-inefficiency (X-INEFF) and scale-inefficiency (S-INEFF) on m a r k e t share (MS), m a r k e t c o n c e n t r a t i o n ( H E R F ) a n d other control
variables
Dep var = X - I N E F F Dep var = S - I N E F F
ALL HC LC ALL HC LC
A L L = all banks in the sample; H C = banks located in countries with high market concentration ( R a n k 1 in Table l); L C = b a n k s located in low m a r k e t '~
concentrations ( R a n k 2 in Table 1). INT = intercept, H E R F = Herfindahl index, MS = market share, W A G E = a v e r a g e w a g e s a n d salary, L T A = natural
log of total assets, R I S K = total liabilities over total assets, PCI = per capita income, YR89, Y R 9 0 and YR91 = d u m m i e s for 1989, 1990 a n d 1991 with
1988 serving as the base year. Y e a r s 1 9 8 8 - 9 1 . Source: C o m p u s t a t ' s Global V a n t a g e a n d International Financial Statistics.
a,b Significant at the 5% and 10% levels, respectively.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 763
Table 6
Regression results of non-interest margin (NIR), defmed as (1 + R O A ) / ( I + NIM) on the Herfindahl
index (HERF), market share (MS), X-inefficiency (X-INEFF), scale-inefficiency (S-INEFF) and other
control variables
Dep var = NIR
ALL HC LC
N 303 133 170
INT 0.879 (41.8) 1.005 (29.6) a 0.944 (27.1) a
HERF 0.007 (0.96) 0.006 ( - 0.56) - 0 . 0 5 9 ( - 1.89) b
IrIS - 0.012 ( - 1.91) b 0.0001 (0.02) 0.028 (1.87) b
X-INEFF - 0.029 ( - 5.89) a - 0.021 ( - 2.31) a - 0.028 ( - 4.63) a
S-INEFF - 0.017 ( - 1.42) - 0.009 ( - 0.56) - 0.081 ( - 4.13) a
WAGE 0.0002 (4.04) a 0.0003 (4.26) a 0.0002 (3.32) a
LTA 0.002 (3.79) a - 0.0001 ( - 0.13) - 0.0003 ( - 0.36)
RISK 0.056 (2.52) a - 0.031 ( - 1.04) 0.019 (0.53)
PCI 0.001 (10.6) a 0.0002 (1.30) 0.002 (7.38) a
YR89 0.002 (1.77) a 0.0009 (0.68) 0.003 (2.06) a
ALL = all banks in the sample; HC = banks located in countries with high market concentration
(Rank 1 in Table 1); LC = banks located in low market concentrations (Rank 2 in Table 1).
INT = intercept, MS = market share, X-INEFF = X-inefficiency, S-INEFF = scale-inefficiency,
WAGE = average wages and salary, LTA = natural log of total assets, RISK = total liabilities over
total assets, PCI = per capita income, YR89, YR90 and YR91 = dummies for 1989, 1990 and 1991
with 1988 serving as the base year. Dep var = dependent variable. Source: Compustat's Global
Vantage and International Financial Statistics.
~l,b Significant at the 5% and 10% levels, respectively.
The results of Eqs. (la) and (lb) are not qualitatively different from those of Eq.
(1) (results not shown). Two of the coefficients changed in statistical significance.
First, the HERF coefficient is positive and significant at the 10% level, for the LC
banks with NIM as the dependent variable (t = 1.80 instead of 0.73). Second, the
MS coefficient with NIM as the dependent variable for ALL banks is no longer
statistically significant (t = 1.61 instead of 2.58).
The coefficients of the control variables are also mixed, with the results
depending on whether ROE or NIM is used. When ROE is the dependent variable,
WAGE rates are insignificant for all three groups. They are negatively correlated
for all three groups when NIM is the dependent variable, consistent with the SCP
and EFS hypotheses. The coefficients are negative for the size factor (LTA, or the
natural log of total assets) for ALL banks, but statistically significant only when
NIM is the dependent variable. When the sample is partitioned, the coefficient is
negative for the HC banks (diversification effect) and positive for the LC banks
(scale effect) when ROE is the dependent variable. Since the coefficients are
sensitive to the dependent variable, the results cannot be generalized with confi-
dence.
The RISK coefficient is also negative and significant for ALL banks as well as
for banks located in LC countries. This would indicate that higher leverage in
banks is associated with higher borrowing costs and not with the alternative
prediction of aggressive asset-liability management. Similarly, the coefficient is
negative for per capita income, PCI, for all groups (except for HC banks when
ROE is the dependent variable). These results support the maturity hypothesis, i.e.
banks in countries with higher PCI are likely to have more competitive environ-
ments which generate lower interest and profit margins.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 765
The dummy variables show that ROE and NIM have changed significantly over
the years, although not in the same direction. NIM declined in 1989 relative to
1988 (YR89) for all three group of banks while ROE declined only for HC banks.
Thereafter, NIM increased over the next two years for all three group of banks but
ROE only increased for LC banks in 1990 and 1991, relative to 1988. Regressions
were run separately for each year to check for yearly differences, but the results do
not change significantly.
Thus, so far the results indicate that when ROE is the dependent variable, there
is some support for the RMP hypothesis for ALL and HC banks. When NIM is the
dependent variable, there is support for both the RMP and ESX hypotheses for
ALL banks, strong support for the ESX hypothesis for HC banks, and strong
support of ESX and ESS hypotheses for LC banks. In order to establish the results
more firmly, the additional conditions specified in Eqs. (2)-(5) are examined next.
Table 4 shows the results for Eqs. (2) and (3), which set the necessary
conditions for the efficient-structure hypothesis. Estimates of market share (MS)
and market concentration (HERF) are regressed against X-INEFF and S-INEFF
and other control variables. Under the efficient-structure hypotheses, it is neces-
sary that the relationship be negative between inefficiency and concentration or
market share. For ALL banks, X-INEFF and S-INEFF are insignificant when MS
is the dependent variable, which does not support the earlier result of the ESX
version found for ALL banks. Partitioning the sample yields one negative and
statistically significant coefficient in the MS equation for X-INEFF and one
positive relationship for S-INEFF, both for the LC banks. This adds support to the
ESX result found for the LC group in Table 3 but not for the ESS result, since the
S-INEFF coefficient is positive. When HERF is the dependent variable, the
coefficient of X-INEFF is positive for ALL and LC and the coefficient of
S-INEFF is negative for ALL and HC. These mixed results do not add anything
conclusive to our analysis. Note that there is no multicollinearity problems because
the correlation between X-INEFF and S-INEFF is approximately - 0 . 0 2 . None of
the control variables contribute any more to the results. The dummies for the years
YR89, YR90 and YR91 are all insignificant except for 1989 for the LC countries
!in the HERF equation, indicating that the market shares and concentrations have
not significantly changed over the years.
Table 5 reports the results for Eqs. (4) and (5) which test Hicks 'quiet life'
hypothesis. Dependent variables X-INEFF and S-INEFF are regressed against MS,
HERF and the control variables. The signs of MS and HERF should be positive if
the RMP conclusions are to be supported, indicating that larger market shares or
concentrations should result in greater X- and scale inefficiencies. The results do
not generally support the RMP hypothesis because the coefficients are not
consistently positive. Importantly, relating to the results of Table 3, the MS
coefficient is negative and significant for ALL banks (t = - 2 . 8 9 ) and insignifi-
cant for the HC banks (t = - 0 . 6 4 ) in the X-inefficiency equation. This is in spite
of an increase in the level of inefficiencies in 1990 and 1991, as shown by the
766 L.G. Goldberg, A. Rai /Journal of Banking & Finance 20 (1996) 745-771
significant dummy coefficients YR90 and YR91 for the X-INEFF equation (except
for LC banks in 1991). These results indicate that banks are operating further from
the efficient cost frontiers relative to 1988. The coefficients for HERF are not
consistent, but this is not particularly important since we did not find any evidence
of the traditional SCP hypothesis in Table 3,
To clarify the previous results an additional test is performed. Eq. (1) is
re-estimated with NIR = (1 + ROA)/(1 + NIM) as dependent variable. 13 This
isolates the efficiency effect on ROA through non-interest factors such as better
management, high productivity, and better use of technology and capital. The
results are reported in Table 6 and as expected reveal less ambiguous efficiency
effects than before. The coefficient of X-INEFF is negative and statistically
significant for ALL, HC, and LC. The coefficient of S-INEFF is negative for each
category and is statistically significant for low concentration countries. These
results are consistent with the NIM results and indicate that ROA results, once
adjusted, are consistent. The only coefficient of the structure variables that is
significant in the expected direction is the positive coefficient of MS in the LC
equation. This formulation provides more evidence in favor of the EFS hypothesis
than the SCP hypothesis.
Thus, when all the results are integrated, they only provide conclusive support
for the ESX version of the efficient-structure hypothesis for banks located in low
concentration countries. For the rest of the banks, there is some evidence for the
EFS hypothesis. If only Eq. (1) is used (as done in other studies), the results
support the RMP hypothesis which is consistent with the results of Molyneux
(1993), Vander Vennet (1993) and others. However, the use of the additional tests
as specified by Berger and Hannah (1993) invalidates some of the results of Eq.
(I).
In addition, three other results are observed for this group of European banks:
a) concentration plays a less significant role as opposed to market share in
explaining profitability;b) the resultsappear to be very sensitiveto the measure of
performance used; and c) performances of banks vary with respect to X-efficiency
and scale-efficiency.The correlation between the two efficiency measures is
-0.02, suggesting some banks are X-inefficient and others arc scale-inefficient,
but not necessarilyboth.
This paper tests the structure-performance hypotheses for the largest banks
located in 11 European countries. Unlike conventional tests, the Berger and
Acknowledgements
Austria Finland
Bank f'tir Tirol und Vorarlberg Affarsbkn Unitas
Bank ftir OberSsterreich Alandsbanken AB
Bk fiir Kb,rnten & Steier Kansallis-Osake-Pankki
Creditanstalt Bankver Okobank
Z-L'Anderbank Bank Aus Skopbank
Switzerland France
Baer Holding LTD Banque Nat de Paris
Banq Cant Vaudoise Cetelem
Banque Paribas Suisse SA Comp Finance de Paribas
Bsi Banca Della Svizzera Compagnie Bancaire SA
Cs Holdings Comptoir des Entrepreneurs
Gotthard Bank Societ6 G6n~rale
Gzb Genossenschaftl
Leu Holdings AG United Kingdom
Neue Aargauer Bank Abbey National Plc
Schweiz Bankverein Bank of Scotland
Schweizerische Bankges Barclays Pic/England
Schweizerische Volksbank Lloyds Bank PIc
Midland Bank Group
770 L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771
References
Aigner, D., C.A.K. Lovell and P. Schmidt, 1977, Formulation and estimation of stochastic frontier
production function models, Journal of Econometrics 6, 21-37,
Bauer, P., 1990, Recent developments in the econometric estimations of frontiers, Journal of Econo-
metrics 46, 39-56.
Berger, A.N., 1995, The profit-structure relationship in banking: Tests of market-power and efficient-
structure hypotheses, Journal of Money, Credit and Banking, Forthcoming.
Berger, A.N. and T. Hannan, 1989, The price-concentration relationship in banking, The Review of
Economics and Statistics 71 (May), 291-299.
Berger, A.N. and T. Hannan, 1992, The price-concentration relationship in banking - a reply, The
Review of Economics and Statistics 74 (February), 376-379.
Berger, A.N. and T. Hannan, 1993, Using efficiency measures to distinguish among allcmative
explanations of the structure-performance relationship in banking, Federal Reserve Board working
paper.
L.G. Goldberg, A. Rai / Journal of Banking & Finance 20 (1996) 745-771 771
Bourke, Ph., 1989, Concentration and other determinants of bank profitability in Europe, North
America and Australia, Journal of Banking and Finance 13(1), 65-80.
Cebenoyan, A.S., E. Cooperman, C. Register and S. Hudgins, 1993, The relative efficiency of stock vs.
mutual S&Ls: A stochastic cost frontier approach, Journal of Financial Services Research.
Gilbert, A., 1984, Studies of bank market structure and competition: A review and evaluation, Journal
of Money, Credit and Banking 16 (November), 617-644.
Hicks, J.R., 1935, Annual survey of economic theory: The theory of monopoly, Econometrica.
Jackson, W.E., III, 1992, The price-concentration relationship in banking - a comment, The Review
of Economics and Statistics 74 (February), 373-376.
Jondrow, J., C.A.K. Lovell, I.S. Materov and P. Schmidt, 1982, On the estimation of technical
efficiency in the stochastic frontier production function model, Journal of Econometrics 19,
233-238.
Lloyd-Williams, D.M., Ph. Molyneux and J. Thornton, 1994, Market structure and performance in
Spanish banking, Journal of Banking and Finance 18,433-443.
McCall A. and M. Peterson, 1980, A critical level of commercial bank concentration, an application of
switching regressions, Journal of Banking and Finance 4, 353-369.
Mester, L., 1993, Efficiency in the savings and loan industry, Journal of Banking and Finance 17(2-3),
267-286.
Molyneux, Ph., 1993, Market structure and profitability in European banking, Unpublished manuscript.
Molyneux, Ph. and J.L. Teppet, 1993, Structure-conduct-performance in EFq'A banking markets,
Bank- en Financiewezen 3, 133-137.
Molyneux, Ph. and J. Thornton, 1992, Determinants of European bank profitability: A note, Journal of
Banking and Finance 16(6), 1173-1178.
Osborne, D.K. and J. Wendel, 1982, A critical review of empirical research in banking competition,
1964-1979, Unpublished manuscript, Oklahoma State University.
Rhoades, S., 1982, Structure-performance studies in banking: An updated summary and evaluation,
Staff Study #119, Board of Governors of the Federal Reserve System.
Ruthenberg, D., 1994, Structure-performance and economies of scale in banking in a unified Europe,
Banking Review 4, Bank of Israel (June), 95-114.
Shepherd, W.G., 1986, Tobin's q and the structure-performance relationship: Comment, American
Economic Review 76 (December), 1205-1210.
Short, B.K., 1979, The relationship between commercial bank profit rates and banking concentration in
Canada, Western Europe and Japan, Journal of Banking and Finance 3(3), 209-219.
Smirlock, M., 1985, Evidence on the (non) relationship between concentration and profitability in
banking, Journal of Money, Credit and Banking 17 (February), 69-83.
Smirlock, M., Th. Gilligan and W. Marshall, 1986, Tobin's q and the structure-performance
relationship: Reply, American Economic Review 76 (December), 1211-1213.
Stigler, G.J., 1964, A theory of oligopoly, Journal of Political Economy 72 (February), 44-61.
Vander Vennet, R., 1993, Concentration, efficiency and entry barriers as determinants of EC bank
profitability, Working Paper, University of Ghent, July.