Journal of Productivity Analysis, 18, 59–77, 2002


C 2002 Kluwer Academic Publishers. Manufactured in The Netherlands.

An Efficiency Comparison of European Banking

Systems Operating under Different
Environmental Conditions
ANA LOZANO-VIVAS avivas@uma.es
Departamento de Teorı́a Económica, Facultad CC.EE., Universidad de Málaga, Plza. El Ejido s/n,
29013 Málaga, Spain

JESÚS T. PASTOR jtpastor@umh.es

Departamento de Estadı́stica y Matemática Aplicada, Universidad Miguel Hernández, Elche, Spain

JOSÉ M. PASTOR jose.m.pastor@uv.es

Departament d’Anàlisi Econòmica, Facultad d’Economia. Campus dels Tarongers, Avda. dels Tarongers,
s/n. Edifici Departamental Oriental, 46022 Valencia, España

Abstract

The paper investigates the operating efficiency differences of a sample of commercial banks
across 10 European countries. First, the paper analyzes the technical efficiency of each coun-
try sample following the “basic” Data Envelopment Analysis (DEA) model incorporating
only banking variables. Then, a “complete” DEA model is introduced, incorporating envi-
ronmental factors together with the banking variables of the basic model. The comparison
between the two models shows that country-specific environmental conditions exercise a
strong influence over the behavior of each country’s banking industry.
JEL classification: D2, D24, G21

Keywords: efficiency, DEA, environmental conditions, European banking

1. Introduction

In anticipation of the expected lowering of barriers to competition among financial institu-

tions within the European Monetary Union (EMU), changes in the regulation of financial
markets have been adopted by many members of the EMU in the last few years. These
changes were designed to liberalize the provision of services and to increase competition.
This way, banks will adjust better to the needs of their customers when they set up branches
in any other country—subject only to the regulations of their home country. As domestic
markets become more competitive, current differences in performance among the banking
industries of the EMU members will largely determine each country’s banking structure and
future competitive viability. Therefore, in the increasingly harmonized European market for
banking services, it is important to know, as well as possible, the differences or similarities
60 LOZANO-VIVAS, PASTOR AND PASTOR

between countries in their current banking performance, the better to predict and/or prepare
for an expected increase in cross-border competition.
Several international comparisons based on analyses of banking efficiency scores have
been performed. Berg et al. (1993, hereafter BFHS), used Data Envelopment Analysis
(DEA) in order to capture the differences in banking efficiency among Norway, Sweden,
and Finland. Berg, Bukh and Førsund (1995, hereafter BBF) followed up the study by
adding Denmark to the previous sample. The same four countries were investigated in
Bergendahl (1995) using mixed optimal strategy. Fecher and Pestieau (1993) and Pastor,
Perez and Quesada (1997) applied the distribution free approach (DFA) and DEA analysis to
11 OECD countries and 8 developed countries, respectively. The former study found results
opposite to those obtained by BFHS and BBF with regard to the efficiency levels of the same
set of countries. Allen and Rai (1996) used DFA and the stochastic frontier approach (SFA)
in order to undertake a systematic comparison of efficiency measures across 15 developed
countries under different regulatory environments.1 Finally, Maudos et al. (2001) used DFA
to compare efficiency measures across 11 European countries and show that nation-wide
efficiency frontiers understate cost and profit efficiency in comparison to specific frontiers
for each specialization.
All these international studies build a common frontier pooling the cross-country banks
and measuring the banking efficiency differences between countries without considering
environmental conditions. In other words, in existing studies that estimate the efficiency of
banks in a cross-national scenario, the standard approach is to construct a common efficient
frontier for all firms, regardless of their home country. However, this standard approach is
unable to compare the different banking systems on an equal footing, because it does not
account for cross-country differences in regulation, economic and demographic conditions,
which are beyond the control of bank managers.2
However, there are several studies of U.S. banking efficiency that control for environ-
mental conditions across states because, in the recent past, the economic and regulatory
conditions faced by U.S. banks have varied greatly depending on the state in which they op-
erated. Evanoff and Israilevich (1991) showed that bank efficiency could vary significantly
with state regulatory environments. Berger and Humphrey (1991) devised the thick cost
frontier method to control for differences in environmental conditions (output demands and
input prices). Mester (1997) showed that nation-wide efficiency frontiers understate cost
efficiency in comparison to separate regional cost frontiers. DeYoung (1998) incorporated
both regulatory and economic variables directly into the cost function in order to control for
differences in environmental conditions across states. Recently, Dietsch and Lozano-Vivas
(2000) have incorporated regulatory and economic variables for a cost efficiency compar-
ison of the French and Spanish banking industries, using a distribution-free parametric
approach.
This paper expands and improves the existing methodology used in international banking
efficiency studies by introducing a new method based on DEA models. Overall, our goal
is to improve the existing methodology used in international banking efficiency studies as
well as to extend the existing literature that controls for environmental conditions into the
areas of DEA methodology and European Banking. This is the first paper to undertake a
systematic comparison of average efficiency measures across countries taking into account
environmental variables and using a non-parametric approach. As is well known, DEA has
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 61

a major advantage and a major drawback. On the one hand, DEA does not need to specify
any functional form for the frontier. On the other, DEA is unable to measure any error term.
Therefore, DEA inefficiency estimates are unable to disentangle the noisy component from
technical inefficiency. The alternative approach is to use a parametric frontier approach.
This requires the specification of a functional form for the frontier, as well as certain
distributional assumptions regarding the error term and the residual inefficiency component.
Parametric frontier approach is therefore able to measure the noisy component of the error
term and technical inefficiency separately but at the expense of establishing a functional
form beforehand. Since neither approach strictly predominates over the other, we have also
considered Stochastic Frontier Approach (SFA) in order to check the consistency of the
efficiency results obtained in our primal analysis.3
The paper is structured as follows. Section 2 depicts a DEA model appropriate for evalu-
ating cross-country technical efficiency, which incorporates country-specific environmental
variables as well as the procedure for detecting the influence of environmental variables in
a DEA framework. The data and the specifications of inputs, outputs and environmental
variables are described in Section 3. Section 4 presents the empirical results based on the
three DEA models considered, and, Section 5 provides conclusions.

2. Methodology

First, we evaluate the technical efficiency of the banking industry of different countries
by means of a DEA model. Initially, we will consider “n” basic banking inputs and “m”
basic banking outputs for each bank and apply the BCC (Banker et al., 1984) input-oriented
model.4 The mathematical formulation of this model is

Min θ subject to Y τ ≥ Y0
θ,τ
X τ ≤ θ X0
(1)
eT τ = 1
τ ≥ 0n

where Y is the matrix of output vectors; X is the matrix of input vectors; (X 0 , Y0 ) is the unit
being rated; e T denotes a row-vector of 1’s; τ is the vector of intensity variables; and θ is the
so-called efficiency score—a quantity between 0 and 1. If θ is lower than 1, a proportional
reduction of all inputs is needed in order to reach the efficient frontier. This reduction is
given by (1 − θ )X 0 , which means that the projected unit given by (θ X 0 , Y0 ) is efficient
in Debreu-Farrell terminology or weakly efficient in DEA terminology. No further radial
reduction of all inputs is possible given the present amount of outputs. It is possible that,
in order to be Koopmans—or DEA—efficient, further individual reduction in some inputs
and/or increase in some outputs is needed. To evaluate these mix-inefficiencies we need to
resort to a more complex BCC model, in which a non-Archimedean element5 in the target
function is multiplied by the sum of the slack variables. However, if the slack variables are
not important, we do not need to pursue this model further.6 The model considered in this
first exercise is called the “basic” model.
62 LOZANO-VIVAS, PASTOR AND PASTOR

Next, we add a set of environmental variables to our DEA “basic” model in order to
standardize the environmental conditions. We hypothesize that the banks of the countries
with bad environmental conditions would get better efficiency scores if they were performing
in a more favorable environment.
There are several ways to evaluate the influence of environmental variables in a DEA
framework (see Rouse, 1996). Here, we propose the simplest method of considering envi-
ronmental factors in DEA. That is, we incorporate these variables directly into the “basic”
DEA model. As is well known, adding variables to the DEA model raises the efficiency
scores; our method of adding each environmental factor guarantees that only the efficiency
scores of banks from countries with bad environmental conditions can change. This approach
has a pre-requisite: we must know in advance the type of influence of each environmental
variable on the efficiency scores. In other words, each uncontrolled factor must have an
influence of known orientation.
To consider the environmental variables as inputs or outputs of our model we just reverse
their condition: for example, if a given environmental variable is an input-type variable
(“less means better”) we consider it as an output in our model (see Cooper and Pastor,
1996). We assume that each unit is compared with a positive linear combination of units for
which the value of each environmental factor is not better than the corresponding value of
the unit. Moreover, all the environmental variables are treated as non-discretionary variables
(Banker and Morey, 1986).7
We consider a second DEA model, which is the extension of the “m + n” variable basic
model obtained by adding to it only those environmental variables that significantly influ-
ence the efficiency scores of the basic model. Following Pastor et al. (2001), we employ a
forward procedure for incorporating variables into a given DEA model. Basically we use a
stepwise procedure—in each step it is decided whether or not it is necessary to incorporate
a new variable into the model. In the first step, we compare the efficiency scores of the basic
model with the efficiency scores of each of the extended models obtained by adding one of
the environmental variables to the basic model. For instance, if we start with a set of “q”
environmental variables, then we have to make “q” comparisons. Each time we compute
the ratio of the efficiency scores of the basic model to the efficiency scores of the extended
model, we obtain a new set of scores, denoted by ρ. We fix a tolerance limit for ρ and we
consider that the proportion of units with ρ lower than the tolerance limit must be lower than
a certain percentage of units in order to decide that the added variable is non-influential.
We perform a non-parametric statistical test, based on the binomial distribution, to assess
whether the statistical evidence justifies adding the variable in question to the model. Let
T denote the number of units with ρ lower than the tolerance limit; the corresponding
p-value will then be given by [1 − F(T − 1)], where F is the binomial distribution function
corresponding to B(N , p), where N denotes the number of units that can have a ρ lower
than the tolerance limit, and p denotes the fixed percentage of units. Pastor et al. (1999)
have shown that N is equal to the number of units in the sample minus the number of
units with θ greater than, or equal to, the tolerance limit. Here θ represents the efficiency
score of the basic model. If the p-value is zero or close to zero we have to reject the null
hypothesis and admit that the added variable is influential. Otherwise, we stay with the basic
model. Once a variable has been added in the first step we have to perform the second step
in a similar manner. After performing a finite number of steps the procedure will stop.8
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 63

We call the final model the “complete” model. The mathematical formulation of this
model is:
Min θ subject to Y τ ≥ Y0
θ,τ
Zτ ≥ Z0
Xτ ≤ θ X0 (2)
eT τ = 1
τ ≥ 0n
where Z denotes the matrix of selected environmental outputs, and Z 0 is the correspond-
ing vector of the unit being rated. Note that we consider all the environmental variables
on the output side. This is because any non-discretionary input can be transformed into a
non-discretionary output just by reversing its sign and translating it.9 In this way we are
sure that all the environmental factors are treated as non-discretionary.

3. Data and Variables.

Data. In our empirical study we use 1993 data for 10 European banking industries for
the selection of the banking variables. These data were obtained from the BankScope
International Bank Database.10 The need to establish domestic as well as international
comparisons with the labor employment data available imposed certain restrictions on the
ability to obtain a domestically and internationally homogeneous sample of banks in terms of
specialization. In our comparison we consider a sample of commercial banks in each country.
We analyze the banking industry of Belgium, Denmark, France, Germany, Italy,
Luxembourg, Netherlands, Portugal, Spain and the United Kingdom (UK). Our study is
limited to the above countries due to the limitations of accurate data for a number of
countries and for certain samples of banks.11 As a result, after carefully checking the data
for consistency, we have usable data for 612 commercial banks belonging to 10 European
countries, with the following distribution: 24 Belgian, 29 Danish, 150 French, 203 German,
26 Italian, 68 Luxemburgian, 22 Dutch, 17 Portuguese, 28 Spanish and 45 British
institutions.
The data for the definition of the environmental variables were gathered from Bank Prof-
itability and Main Economic Indicators (OECD), Eurostat (Money and Finance), Anuario
Estadı́stico del Instituto Nacional de Estadı́stica (INE) and Boletı́n del Banco de España.
All 1993 variables, presented in terms of local currencies, were converted into a common
currency (U.S. dollar) using the purchasing power parity hypothesis.

Input and output variables In the banking literature there has been considerable disagree-
ment regarding the “proper” definition of inputs and outputs. We adopted the value-added
approach (Berger and Humphrey, 1992) to identify banking outputs. In the value-added
approach, all items on both sides of the balance sheet are considered to have some output
characteristics depending on their contribution to the bank’s generation of added value.
In a value-added context, deposits typically account for more than half of the total cap-
ital and labor expenses of banks and so, in this sense, output services are clearly being
produced. Accordingly, in this study, we specify three outputs:12 y1 = loans, y2 = deposits,13
64 LOZANO-VIVAS, PASTOR AND PASTOR

and y3 = other earning assets;14 and two inputs: x1 = personnel expenses and x2 = non-
interest expenses, which are all operating expenses minus personnel expenses.15 As usual,
we consider labor input in terms of personnel costs rather than of the number of employees,
on which there is a lack of data. The two inputs considered allow us to describe technical
efficiency as operating efficiency.
Table 1 reports average values by countries, in a common currency ($US), of bank outputs
and inputs for 1993. We observe enormous differences among countries in the average values
of portfolio loans and deposits. However, values are higher for deposits than for portfolio
loans. A comparison of the differences between the maximum and minimum average value
of deposits with the differences between the maximum and minimum average value of loan
portfolios shows that the former is almost 40% higher than the latter. Such differences may
be attributed to the different strategies of specialization followed by each banking industry
following the deregulation of interest rates.

Environmental variables. The environmental variables used in this paper are variables
explaining the particular features of each country’s banking industry. These features include
macroeconomic and regulatory conditions as well as accessibility of banking services. The
environmental variables are measured on a per country basis (i.e., these variables take the
same value for each bank in each country).16 Theory offers few guidelines as to which
determinants are important for explaining the particular features of each country’s banking
industry. We therefore rely on previous empirical studies for the selection of variables when
theoretical studies do not exist. The set of environmental variables selected is presented in
Table 2. The first environmental variable is the per capita income, IC, of a country, a predeter-
mined variable defined as the ratio of the Gross National Product (in 1993 US$) to the number
of inhabitants. IC affects numerous factors related to the demand and supply of banking ser-
vices (mainly deposits and loans). Countries with a higher IC are assumed to have a banking
system that operates in a mature environment and results in more competitive interest rates
and profit margins. At the same time, the banking system may also exert more activity.17
The second environmental factor is the salary per capita, SC, (expressed in 1993 US$),
defined as the ratio of the total salary volume to the number of working inhabitants. SC is
an indicator of each country’s economic performance. It is reasonable to hypothesize that
high levels of SC indicate high quality and potential high productivity of the labor force,
which should improve bank efficiency. The population density, PD, is measured by the ratio
of inhabitants per square kilometer. We assume that high levels of PD should make retail
distribution of banking services less costly, which should improve bank efficiency. The
density of demand, DD, measured by the ratio of total value of deposits (measured in 1993
US$) per square kilometer is assumed to be a relevant feature in determining efficiency.
Banks that operate in markets with a lower density of demand incur higher expenses due to
the fact that the lower DD may impose a ceiling on the efficiency level attainable by their
branches. The above four environmental variables reflect the main economic conditions
under which banks operate.18
Income per branch, IB, and total value of deposits per branch, DB, both expressed in
1993 US$, are considered usual measures of the relative efficiency of banking industries.
Countries with a higher rate of national income per branch are assumed to have banking
systems with greater opportunities to do business. We assume that the higher the IB or the
Table 1. Summary statistic for banking outputs and inputs (1993).

Belgium Denmark France Germany Italy Luxembourg Netherlands Portugal Spain U.K.

Outputs
y1 = Loans 4588392.3 1964163.5 5256541.3 3539048.8 16967317.9 750671.1 5096565.8 2265049.4 6401894.6 1859216.7
y2 = Deposits 11747008.6 3320735.6 9082831.4 2860965.8 25327940.3 2907620.6 6846185.1 4919924.3 13208616.7 3115273.9
y3 = Other earning assets 7800801.6 1757720.8 5274525.1 2048662.5 12810940.3 2419358.9 2741402.5 2480336.5 7531540.3 1594990.9

Inputs
x1 = Personnel expenses 84994.4 39376.7 100692.7 40386.7 344562.3 10121.3 93797.1 86072.1 165255.2 55453.6
x2 = Non-interest expenses 141244.3 58311.6 131495.7 64924.3 491656.1 10769.5 100096.4 89420.4 245397.2 58934.9
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS
65
66 LOZANO-VIVAS, PASTOR AND PASTOR

Table 2. Summary statistic for environmental variables by country (1993).

IC SC PD DD IB DB BC BD EOTA ROE

Belgium 17130 30759 329.3138 5.1403 0.0088 9.5625 0.0017 0.5703 0.0398 0.0954
Denmark 16812 20617 120.4847 1.3945 0.0320 25.6712 0.0005 0.0543 0.0593 0.1061
France 17646 26964 105.9995 0.9689 0.0328 20.0462 0.0005 0.0483 0.0427 0.0287
Germany 16777 24197 227.5326 3.2902 0.0256 61.9711 0.0006 0.1259 0.0430 0.1379
Italy 16497 27038 189.4358 1.5369 0.0317 17.9591 0.0005 0.0856 0.0944 0.1111
Luxembourg 20538 28925 154.6551 55.149 0.0324 466.142 0.0008 0.1183 0.0326 0.1993
Netherlands 16061 27588 373.9067 5.6392 0.0296 32.3861 0.0005 0.1741 0.0428 0.1407
Portugal 10532 13425 107.1891 0.5611 0.0329 16.4165 0.0003 0.0342 0.1437 0.0661
Spain 12121 23022 77.4195 0.6538 0.0120 9.5020 0.0009 0.0688 0.0962 0.0376
U.K. 15422 23107 237.3547 1.9059 0.0689 40.6012 0.0002 0.0469 0.0395 0.1902
IC = Income per Capita; SC = Salary per Capita; PD = Population Density; DD = Density of Demand; IB =
Income per Branch; DB = Deposit per Branch; BC = Branches per Capita; BD = Branch Density; EOTA = Equity
over Total Assets; ROE = Return over Equity.

DB, the higher banking efficiency levels will be. Branches per capita, BC, are an indicator
of banking services. High levels of BC imply high costs of providing banking services,
which should reduce bank efficiency. The above three environmental variables are standard
measures of bank performance, used in previous banking performance studies, for instance
in DeYoung and Hasan (1998). Additionally, the variable called branch density, BD, is
defined as the number of branches per square kilometer, and is an indicator of the space
dimension for each national market.19 High levels of BD indicate overdimension of the
banking network and high bank operating costs, which should reduce banking efficiency.
These variables refer to the accessibility of banking services for customers.
Additionally, we use the average capital and profitability ratios as indicators of the
regulatory and competitive conditions, respectively, of a country’s banking industry.20 The
average capital ratio is used as a proxy for regulatory conditions and is measured by equity
over total assets, EOTA. Although there is certainly a relationship between capital ratio and
efficiency, there are theoretical arguments to support both a negative and a positive influ-
ence on efficiency levels, so we do not know a priori the direction of the influence of this
environmental variable. Berger and DeYoung (1997), assert that the higher the solvency and
prudence (capital ratio) of the banks, the lower the bad loan levels, it being less necessary to
incur additional expenses to recover these loans, and they will therefore appear more effi-
cient, i.e., banks with higher capital ratios will show higher efficiency levels. However, there
is also a theoretical argument to support the opposite idea. A low capital ratio can cause moral
hazard behavior. Thus banks with solvency problems may undertake risky business, invest-
ing in very profitable activities, and will therefore appear efficient in the short term, although
they will probably pay the consequences of their risky behavior in the long term. So we will
not make any a priori assumption about the sign of the influence of the capital ratio (EOTA).
Finally, the profitability ratio is defined as average return over equity, ROE, and is used
as an indicator of the competitiveness in each banking industry. The predicted relationship
between ROE and efficiency is positive in a competitive scenario, i.e., the higher the profits,
the higher the efficiency. This variable is also examined with the same orientation by Berger
et al. (1993), Mester (1993) and Allen and Rai (1996).
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 67

The use of environmental factors within DEA models requires knowledge of the influence
of the environmental variables on the efficiency scores. If we presume that the higher (lower)
the value of an environmental variable, the higher (lower) the efficiency scores for the
complete model, then we can say that the environmental variable is an output-type variable.
On the other hand, if the opposite relationship holds, we say that the environmental variable
is an input-type variable. Consequently, IC, SC, PD, DD, IB, DB and ROE are classified as
output-type variables while BC and BD are considered input-type variables. In the case of
EOTA the results prove that this variable should be classified as an output-type variable.21
Following Cooper and Pastor (1996), the first eight environmental factors must be introduced
as inputs and the last two factors as outputs.
Table 2 contains the values of the environmental variables for 1993 and at a country level.
Overall, the values of these variables suggest large differences among countries in terms
of their particular economic, banking accessibility and regulatory conditions. Particularly,
Luxembourg has the highest level of income per capita, salary per capita and density of de-
mand. Portugal and Spain have the lowest levels of these main economic conditions. Thus,
it may be harder to perform banking activities in Portugal or Spain than in Luxembourg.
In terms of the variables which refer to the accessibility of banking services for customers,
Belgian and Spanish banks appear to work under conditions requiring high levels of operat-
ing costs given the level of accessibility of banking services for customers. Finally, the mean
values of the capital and profitability ratios show important differences among countries.
Differences in capital ratio could be attributed to the particular solvency constraints imposed
by the banking authorities of each country. On the other hand, differences in the profitability
ratio could show that competitive conditions are quite different among countries.

4. Empirical Results

Table 3 depicts the bank efficiency scores of each country based on its own national frontier
estimate. The results are quite distinct; while Italy and Spain show small differences in the
efficiency of their own commercial banks, France and Germany show just the opposite.
The results in Table 3 give information about bank efficiency on the basis of the national

Table 3. Efficiency scores: Internal efficiency by country (Individual

frontiers. Only banking variables).

Average Std.

Belgium 77.43 27.48

Denmark 70.93 13.13
France 36.65 28.21
Germany 48.91 25.97
Italy 85.15 12.73
Luxembourg 53.66 26.04
Netherlands 71.03 31.40
Portugal 77.65 23.17
Spain 81.52 15.38
U.K. 55.50 32.73
68 LOZANO-VIVAS, PASTOR AND PASTOR

Table 4. Efficiency scores: Basic model (Common frontier. Only banking variables).

Average Std.

Belgium 42.20 28.47

Denmark 19.91 13.86
France 24.23 24.26
Germany 26.67 18.49
Italy 25.43 19.98
Luxembourg 49.49 25.67
Netherlands 37.38 25.32
Portugal 15.99 15.63
Spain 18.91 15.06
U.K. 22.08 20.85

frontiers. However, since our goal is to infer the bank’s productivity based on its ineffi-
ciency measurement, the inefficiency measures need to have been derived from a common
frontier.
For international comparison of banking efficiency, we first defined the common frontier
based on the traditional approach, i.e., building a common frontier by pooling the bank
data of all the countries and considering a DEA model with two banking inputs and three
banking outputs. Table 4 gives descriptive statistics of the average efficiency scores for each
country using the basic model—i.e., without taking into account the specific environmental
conditions of each country. The results show that Luxembourg gets the highest average
efficiency score, around 49.5%, and Spain and Portugal the lowest, around 18.9% and 16%,
respectively.
Overall, the results show lower average efficiency scores for each country than the results
obtained from the national frontiers. From a technical point of view this is no surprise: adding
units in a DEA analysis, without modifying the variable set, produces lower efficiency
scores. Interestingly, when we studied the country-specific conditions from the raw data,
we observed that, on average, the environmental conditions in Luxembourg were more
favorable than in Spain; this fact could explain, at least partly, the differences in efficiency
between the two countries shown in Table 4. The same theory is valid when making pairwise
comparisons among the remaining countries in our study. These results seem to confirm our
conjecture that if the country-specific variables are an important factor in explaining average
efficiency differences, then we have to consider those variables in order to get comparable
efficiency levels.
We start by specifying the complete model. To define a common frontier with environmen-
tal variables, first we have to select a suitable subset of environmental variables. Initially, we
consider the entire set of 10 environmental variables (see Table 2). When applying the tech-
nique explained in Section 2, we implemented a forward procedure in order to incorporate
the least number of influential environmental variables into our basic model. The results
of using this procedure are shown in Table 5. In the first step, we compared the efficiency
scores of the basic model with the efficiency scores of an extended model. The extended
model was obtained by adding only one of the environmental variables to the basic model,
e.g., variable IC. We computed the ratios of the efficiency scores of both the basic and the
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 69

Table 5. Selection of environmental variables.

Step 1 Step 2 Step 3 Step 4 Step 5

ρ < 0.9 ρ < 0.9 ρ < 0.9 ρ < 0.9 ρ < 0.9

IC (Income per capita) 97 4 4 66 40

SC (Salary per capita) 100 42 29 184
PD (Population density) 116 22 0 0 0
DD (Deposit density) 319
IB (Income per branch) 75 98
DB (Deposit per branch) 293 59 0 0 40
BC (Branch per 100000 inhabitant) 23 45 0 0 0
BD (Branch density) 22 47 2 2 35
EOTA (Equity over total assets) 10 97 64
ROE (Return over equity) 185 47 19 0 3

extended models and got what we call ρ. We fixed a tolerance limit of 0.9 for ρ and we
assumed that the proportion of units with ρ lower than the tolerance limit must be lower
than 10% in order to consider variable IC as non-influential. This level was selected on
the basis of a prior Monte-Carlo simulation (Pastor, Ruiz and Sirvent, 1999). For variable
IC, we obtained T = 97 units with ρ < 0.9. Since the basic model had 35 banks with θ
no lower than 0.9, we considered the binomial with N = 612 − 35 = 577 and p = 0.1 in
order to assess whether there was statistical evidence that fewer than 10% of the banks had
a ρ lower than 0.9. We found that the p-value associated with IC was 0, and therefore,
we had to reject the null-hypothesis. As a result, IC was considered an influential variable.
Repeating this procedure, we counted T for each of the 10 models obtained by adding one
of the 10 environmental variables to the basic model. The results are summarized in column
2 of Table 5. We can observe that in the first step the most influential variable is DD, with
T = 319. The non-parametric binomial test offered us a p-value equal to 0 for this variable,
which means that we have to reject the null hypothesis and accept that the environmental
variable DD has to be incorporated into our model.
In the second step, we started with the new six-variable models and repeated the same
process of adding one of the nine remaining environmental variables to the new fixed model.
The results are shown in column 3 of Table 5. As a consequence, we have to add variable
IB to the model since it is the most influential variable of step 2. IB showed a p-value of
0, obtained by considering the binomial with N = 612 − 58 = 554 and p = 0.1. Table 5
shows that by expanding (iteratively) our model, the next variables to be incorporated into
our model were EOTA and SC in steps 3 and 4, respectively. In step 3, the corresponding
binomial N = 612 − 75 = 537 and p = 0.1 offered a p-value for variable EOTA equal to
0.092, and, in step 4, the binomial N = 612 − 111 = 501 and p = 0.1 offered a p-value for
variable SC equal to 0.
Finally, in step 5 the process was completed because no additional variables deserved to
be included in the model. In fact, the most influential variables at this stage, IC and DB,
both have T = 40, which—resorting to the binomial distribution with N = 612 − 123 = 489
and p = 0.1—corresponds to an associated p-value equal to 1. Consequently, our complete
model had 9 variables (two basic inputs, three basic outputs and four more non-discretionary
outputs corresponding to the environmental variables DD, IB, EOTA and SC).22
70 LOZANO-VIVAS, PASTOR AND PASTOR

Table 6. Efficiency scores: Complete model (Common frontier. Banking and four
environmental variables).

Average Std.

Belgium 79.32 25.50

Denmark 75.45 15.79
France 40.98 27.85
Germany 57.87 24.11
Italy 33.10 22.20
Luxembourg 62.30 23.58
Netherlands 51.75 27.35
Portugal 79.87 21.99
Spain 82.14 14.93
U.K. 58.65 30.15

The average efficiency scores obtained by means of the complete model are shown in
Table 6. Overall, the results show that when we introduce the environmental variables
into the model, the average efficiency scores increase markedly in almost all the countries
compared to the average efficiency scores of the basic model (Table 4). The interesting point
is to quantify the intensity of the change and, in our specific case, to note that countries
with bad environmental conditions have experienced the greatest changes.
In order to check whether or not the addition of environmental factors increases the
reliability of the results, we rank the countries on the basis exclusively of the four selected
environmental variables. To do this, we first rank each country (from 1 to 10) four times
in terms of each environmental factor by examining the values of each environmental
variable in each country. If the environmental factor is an output (input) of the model and
gets the higher (lower) value, then the country gets a ranking value of 1. Once we know
the rankings for each country with regard to each environmental factor, we calculate their
mean. This average rank value for each country permits us to determine which countries
are operating under good environmental conditions (those at the top of this ranking) and
which countries are operating under bad environmental conditions (those at the bottom of
this ranking).
Comparing the environmental ranking with the average efficiency scores of the basic
and the complete model for each country we observe that the worse the country-specific
conditions the greater the rise in the average efficiency score. For example, Luxembourg, the
Netherlands and Italy take the first, second and third place at the top of the environmental
ranking. The average change in their efficiency scores is around 12% (see Tables 4 and
6). Belgium, ranked in fourth place, and Germany, France, Portugal and the UK, ranked
equally in fifth place in terms of their average environmental conditions, show a similar
average improvement in their efficiency scores, around 36%. Finally Denmark and Spain,
which are ranked at the bottom, obtain an increase of around 60% in their average efficiency
score. Therefore, these findings confirm our belief that the environmental variables are an
important factor in explaining differences in international banking efficiency.
For example, banks in Luxembourg appear to be experiencing the highest efficiency
scores (Table 4). On the other hand, Spanish banks seem to be operating with the second
lowest efficiency scores. These results could be explained by the lower salary per capita,
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 71

density of demand and income per branch that exist in Spain compared to Luxembourg.
Thus, considering the influence of these factors on the efficiency scores of Spanish banks
we observe that, on average, they become even more efficient (82.14%) than the banks of
Luxembourg (62.30%), assuming that the banks of both countries operate under the same
environmental conditions, (see Table 6). The same type of comparison could be established
for any other two countries.
The influence of the environmental variables seems, therefore, to be as expected. Introduc-
tion of environmental conditions into a DEA model implies a change in the average efficiency
scores; the intensity of the change depends on the average country-specific condition. These
results are in accordance with the results obtained by Dietsch and Lozano-Vivas (2000).
To check for robustness of our results we study the average behavior of the efficient banks
of each country. This analysis is interesting in two ways. First, because the best banks of
each country are supposed to have the best technology in each country and consequently,
to be better prepared for hypothetical cross-border competition. Second, given that we are
using samples of widely divergent sizes, it could be possible to find different proportions of
inefficient banks when comparing large and small samples. Table 7 contains the information
on the average behavior of the efficient banks in each country evaluated by means of the
basic and the complete model. Comparing the results obtained in Tables 4 and 7 for the basic
model, we observe that the group of countries with the largest samples (French, German
and Luxembourg) shows larger differences in the average efficiency scores than the remain-
ing countries. Indeed, the most efficient French and Luxembourgian banks appear beside
the Belgian banks in Table 7 as having the best banking performance in Europe. Overall,
the results from the complete model seem to suggest that the most efficient banks from any
of the European countries under investigation, except Italy and the Netherlands, possess
enough competitive viability to be able to operate in a more unified European banking
market.
Having observed the effectiveness of the environmental variables in explaining the differ-
ences in international banking efficiency, it is interesting to know which of these variables
are more relevant in explaining such efficiency differences. Given the stepwise procedure
used to select the environmental variables which gave rise to the complete model, we can
identify the particular role that each of the four selected environmental variables plays in

Table 7. Average behaviour of the efficient banks of each country.

Basic Model Complete Model

Belgium 66.61 100.00

Denmark 48.83 100.00
France 87.53 100.00
Germany 63.53 100.00
Italy 43.15 53.72
Luxembourg 90.48 100.00
Netherlands 60.57 73.01
Portugal 23.16 100.00
Spain 35.74 100.00
U.K. 43.23 100.00
72 LOZANO-VIVAS, PASTOR AND PASTOR

Table 8. Percentage contribution of the environmental variables to the changes of the average efficiency scores of
the basic model by countries.

DD IB EOTA SC Total

Belgium 16.14 83.86 0.00 0.00 100.00

Denmark 23.57 7.49 0.27 68.67 100.00
France 40.00 49.01 10.99 0.00 100.00
Germany 39.17 19.58 0.77 40.48 100.00
Italy 11.60 87.09 1.17 0.13 100.00
Luxembourg 71.66 23.20 5.14 0.00 100.00
Netherlands 59.85 39.74 0.21 0.21 100.00
Portugal 7.67 91.70 0.00 0.63 100.00
Spain 9.82 86.15 3.29 0.74 100.00
U.K. 16.02 65.30 18.68 0.00 100.00
Average contribution 29.57 55.30 4.05 11.08 100.00
DD = Density of Demand; IB = Income per Branch; EOTA = Equity over Total Assets; SC = Salary per Capita.

explaining the average efficiency scores of the basic model. Table 8 shows the percentage
contribution of the environmental variables to the improvement in the basic average ef-
ficiency scores of each country. Consequently, the sum of the contributions of the four
variables to the average efficiency score of each country appears as 100 in Table 8. The
absolute value of the contributions of the variables can be easily deduced by consulting
Table 6 (the complete model) and Table 4 (the basic model). We observe that each en-
vironmental variable seems to play a particular role in explaining the rise of the basic
average efficiency score of each country. In percentage terms, the variable with the high-
est contribution is IB, income per branch. In fact, the contribution of IB is the highest
in the cases of Belgium, France, Italy, Portugal, Spain and U.K. On the other hand, DD,
the density of demand, is the most influential variable in two cases (Luxembourg and The
Netherlands) and the same happens with the salary per capita, SC (Denmark and Germany).
Curiously enough, the contribution of the variable EOTA is low and is relatively impor-
tant only for the U.K. and France. Nevertheless, in the process of incorporating environ-
mental factors, EOTA acts as an umbrella for variable SC: the influence of SC remains
hidden until EOTA is incorporated into the DEA model. Interestingly, the environmen-
tal variables, which seem to play the most important role in explaining the differences in
efficiency, are related to the main economic conditions and the accessibility of banking
services of each country. This might suggest that for cross-border competition in Europe,
banks will need to draw up a strategy in order to survive, and to compete with domes-
tic banks, in the adverse environmental conditions for banking services that exist in some
countries.

5. Conclusions

We have been able to find the differences and similarities in the current banking performance
of 10 European countries. A rigorous multistep DEA analysis reveals the importance of
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 73

the environmental variables in explaining the efficiency differences among countries. Our
results show that the basic average efficiency scores of the banks of each European country
are unusually low. In order to make a fair comparison among countries we have added the
relevant environmental variables to the basic model, obtaining in this way the complete
model. Comparing the basic and the complete average efficiency scores we observe that the
worse the country-specific environmental conditions the greater the changes in the scores.
This means that the complete model is able to evaluate the different countries on an equal
footing, in contrast to the basic model.
Finally, taking into account the effects of the environmental variables, the results of Table 7
suggest that the most efficient banks from almost any of the 10 European countries, with
the exception of Italy and the Netherlands, have enough competitive viability to be able
to operate in a more unified European banking market. However, given that environmental
variables, which play an important role in explaining differences in efficiency, are related
to the accessibility of banking services and to the particular economic conditions of each
country, it is likely that for cross-border competition in Europe, banks will need to draw
up a strategy in order to survive, and to compete with domestic banks, in the adverse
environmental conditions for banking services that exist in some countries.

Appendix23

In this appendix we re-evaluate our efficiency measures on the basis of the basic and
complete models, incorporating banks’ asset quality and capitalization into our analysis as
control variables. The financial literature emphasizes that asset quality and capitalization are
important components of risk, and points out that failure to consider risk could discriminate
against banks that choose to adopt cost-intensive measures to reduce their exposure. Such
banks may appear inefficient although operating optimally given their risk preferences.
So, considering this issue, we have performed a new exercise adding bank’s asset quality
and equity capital as new variables into our primal basic and complete models. Lack of
data reduces our original sample of 612 banks to a new sample of 511 banks: 20 Belgian,
26 Danish, 141 French, 158 German, 19 Italian, 59 Luxemburg, 19 Dutch, 16 Portuguese,
25 Spanish and 28 British. The numerical results obtained from this exercise are shown
in Table A1.
Comparing the results from the New BM and CM with and without risk it is observed
that the inclusion of the risk components increases the average efficiency measures, mainly
for the complete model. However, the rank of the country efficiency level is unchanged.
The increased efficiency level once risk is accounted for is in line with what is found in
the literature. Berger and Humphrey (1997) report in their survey that the inclusion of risk
decreases the inefficiency measures. However, it is well known that since the inclusion of
those variables introduces an additional constraint in a non-parametric model, it will imply
reducing the inefficiency measures. So one way to test the influence of including risk in our
operating efficiency measures could be to pay attention to the variation of efficiency (i.e.,
the standard deviation of efficiency measures) across banks in each country jointly with the
position of that country’s banks in the efficiency rankings (this procedure is followed also
by Sheldon, 2001). It can be observed that even the efficiency ranking of average country
74 LOZANO-VIVAS, PASTOR AND PASTOR

Table A1. Efficiency scores: Basic and complete models with and without consideration of risk.

New BM New CM New Risk BM New Risk CM

Without Risk With Risk

Average Std. Average Std. Average Std. Average Std.

Belgium 43.51 29.55 77.53 28.55 43.54 29.57 80.04 27.56

Denmark 26.80 16.38 75.48 14.10 26.95 16.54 81.28 11.90
France 28.42 25.24 39.96 28.52 29.34 26.31 40.61 28.94
Germany 32.17 19.54 60.35 25.06 32.50 19.86 60.83 25.19
Italy 25.13 20.83 31.42 23.93 25.86 21.38 32.03 24.37
Luxembourg 57.51 25.02 60.37 24.94 57.60 24.98 62.60 25.45
Netherlands 41.53 24.96 50.76 28.78 43.56 27.77 51.91 30.21
Portugal 19.46 20.43 81.54 19.12 19.61 20.44 91.67 15.09
Spain 23.10 20.17 82.96 15.53 23.62 20.66 85.92 15.72
U.K. 25.43 18.57 65.23 29.01 25.56 18.58 69.17 28.78
New BM = Basic model using 2 inputs and 3 outputs as in the old basic model but using the new sample of
511 observations. New CM = Complete model using 2 inputs, 3 outputs and 4 environmental variables as in the
old complete model but using the new sample of 511 observations. New Risk BM = Basic model defined using
the suggestion of the referee, adding asset quality and equity capital as control variables to the New BM. The new
sample of 511 observations has to be used in this exercise due to lack of data. New Risk CM is the result of adding
the environmental variables to the New Risk BM. This new model is obviously run on the reduced sample of 511
observations.

banks is unchanged when risk is taken into account, the efficiency variation across banks is
reduced for some countries (e.g., Belgium, Denmark, Portugal), while for others there is no
significant change. So, the decreased efficiency variation, together with the unchanged rank,
could suggest that controlling for risk has produced a greater improvement in the operating
efficiency of the banks that were initially less efficient, implying perhaps that some of the
intensive use of input by those banks, which makes them appear more inefficient, is to
reduce risk.
However, we can conclude that irrespective of the inclusion or exclusion of risk, when
we introduce the environmental variables into the model, in almost all the countries the
average efficiency scores increase markedly compared to those of the basic model, and it
is the same country banks that generally appear to be the most (or least) frontier efficient.
This constitutes a validation of the main results contained in our paper, and could suggest
that the differences in efficiency scores between the basic model and the complete model
in each country are influenced more by the country-specific environmental conditions than
by the risk preference choices of banks.

Acknowledgments

Instituto Valenciano de Investigaciones Económicas (IVIE) and CICYT PB94-1523, PB98-

1408 supported this research. The authors thank IVIE for the data. The authors would also
like expressing their gratitude to David Humphrey, Loretta Mester and four anonymous
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 75

referees for their advice and suggestions. Earlier version of this paper was presented at the
Georgia Productivity Workshop in November 1996. A previous version of this paper was
published as working paper WP EC97-12, IVIE.

Notes

1. See Berger and Humphrey (1997) for more details about the methodologies used and results obtained in those
studies.
2. Although Allen and Rai’s paper takes into account the regulatory environments of each country, those en-
vironmental variables are specified at bank level and not at country level. Moreover, the authors explain the
differences in efficiency by conducting an ex-post analysis.
3. We have developed two alternative parametric models using the SFA. The first one adjusts a cost frontier
considering only the basic banking variables, i.e., the inputs and outputs of our basic DEA model. The second
model considers the basic variables together with the environmental variables, i.e., the inputs and outputs of
the complete DEA model. Comparing the parametric results with the DEA results, several features are worth
noting. First, the corresponding efficiency scores of both pairs of models are highly correlated, over 0.71 in
all cases. Second, if we rank the countries in terms of the improvement in efficiency scores, the correlation
between DEA and SFA is 0.95.
4. We have to point out that a general DEA model can deal with any type of input and output variables. This
means that we can consider as input quantities, expenses or costs, and similarly with output variables.
5. The non-Archimedian element guarantees that the sum of the slacks is always maximized without altering the
value of the efficiency scores at the optimum. This condition is needed in order to ensure that the projected
unit belongs to the efficient frontier. This concept was introduced by Charnes, Cooper and Rhodes (1979).
6. This is exactly what happens with the data in this paper where the amount of any slack always value less than
5% of the corresponding variable value.
7. In the DEA literature although non-discretionary variables are either inputs or outputs of the DEA model,
they are never considered as inputs or outputs of the production process. The role of the non-discretionary
variables is only to constrain the comparison set. Moreover, any variable, which is not an input or an output
of the production process, must be considered as a non-discretionary variable.
8. Banker (1996) has recently proposed several parametric tests for the specification of a DEA model. We preferred
the proposal of Pastor et al. (1995) because it is distribution-free and is a stepwise procedure.
9. See Pastor (1996).
10. The BankScope International Bank Database is a detailed database provided by BankScope (a joint venture
between Fitch IBCA and Bureau van Dijk). This company offers, among its wide variety of products and
services, a comprehensive financial database on 10,000 international banks (including 1,500 U.S. banks) for
a period up to eight years of detailed raw data.
11. Some other institutions in France and Germany were eliminated. These institutions represented less than 3%
of the banks in their respective country samples. The sample selected can be considered as representative,
given that it includes almost all the banks classified as commercial banks in Bankscope.
12. We have to point out here that one referee has suggested that we introduce off-balance-sheet activities since
some of these activities, e.g., credit guarantees, involve operating costs and could affect the efficiency measures.
However, the lack of information does not allow us to consider off-balance-sheet services in our exercise.
13. Deposits were defined as produced deposits (the sum of demand, savings, time, interbank and other deposits).
14. Earning assets were defined as the sum of all existing deposits with banks as well as short-term and other
investments.
15. Because we are defining labor and physical capital as inputs we are calculating a production model that is
similar to an operating cost model.
16. It is necessary to point out that our goal was not to conduct a microlevel study but to determine the national
average efficiency levels, where macroeconomic explanatory variables would become relevant.
17. The extent and quality of finance have long been believed to affect economic growth, and recent empirical
work has provided evidence of this belief; e.g., King and Levine (1993), Jayaratne and Strahan (1996), Levine
(1998, 1999), Levine and Zervos (1998) among others. In this study the per capita income is taken as given,
76 LOZANO-VIVAS, PASTOR AND PASTOR

and it is assumed to affect the demand faced by banks and the variety of banking output and services the banks
can supply. In others words, the country’s per capita income level affects the amount of banking activity.
18. Goldberg and Rai (1996) and Dietsch and Lozano-Vivas (2000), among others, have analyzed the effect of
these variables on bank performance.
19. Research on this topic shows that to ignore the geographical dimension of the market in which the banks
operate might lead to an erroneous evaluation of the banking efficiency levels (Fuentelsaz and Salas 1992).
20. Even though the banks can control the equity-capital ratio, a large part of this ratio is controlled exogeneously
since it has to be adjusted to regulatory restrictions at National and European levels. Capital requirement
has been the primary tool used by National and European regulatory authorities to control risk, using it as
prudential regulation. Moreover, in order to maintain solvency of the banking sector while preventing moral
hazard, regulators have resorted to increasing capital requirements.
21. We have computed the efficiency scores of two alternative extended models, specifying EOTA as an input-
type variable and as an output-type variable. Using the methodology of Pastor, Ruiz and Sirvent (2001) (see
Section 2), we found that the variable EOTA was influential only when EOTA was treated as output. The
direction of the influence is similar to that obtained by Berger and DeYoung and Hughes and Mester (1993)
among others.
22. Since the environmental variables DD, IB and/or EOTA may not be exogenous to banks’ decisions, we have
tested whether or not they are endogenous. We have to pointed out that although these variables could be
endogenous to a particular bank, the endogeneity problem is lessened since the environmental variables are
measured on a per country basis. However it is true that this problem is not totally eliminated since a common
frontier is estimated, hence the need to test for possible endogeneity. For this purpose we use the parametric
model that we used to check the consistency of the efficiency results of our primal DEA analysis (endnote 3), and
the omitted variable version of the Hausman test. This test allows us to investigate whether or not the regressors
are correlated with the error terms, since the environmental variables will be endogenous if the regressors are
contemporaneously correlated with the error terms. As it is well known, the Hausman test is a test of equality
of the estimates produced by two estimators (estimators of the original variables vs instrumental variables). In
order to compare these two estimators we first define the instrumental variables of the environmental variables
DD, IB, and EOTA and compare the estimates produced by the original environmental variables and those of
their instrumental variables. The omitted variable version of the Hausman test rejects the endogeneity of the
environmental variables, so our complete model will not be missepecified due to this factor.
23. We thank to the referee for suggesting us to include this section.

References

Allen, L. and A. Rai. (1996). “Operational Efficiency in Banking: An International Comparison.” Journal of
Banking and Finance 20, 655–672.
Banker, R. D., A. Charnes and W. W. Cooper. (1984). “Some Models for Estimating Technical and Scale Ineffi-
ciencies in Data Envelopment Analysis.” Management Science 30(9), 1078–1092.
Banker, R. D. and R. Morey. (1986). “Efficiency Analysis for Exogenously Fixed Inputs and Outputs.” Operations
Research 34(4), 513–521.
Banker, R. D. (1996). “Hypothesis Tests using Data Envelopment Analysis.” Journal of Productivity Analysis
7(2/3), 134–160.
Berg, S. A., P. N. D. Bukh and F. R. Førsund. (1995). “Banking Efficiency in the Nordic Countries: A Four-
Country Malmquist Index Analysis.” Working Paper, University of Aarhus, Denmark (September).
Berg, S., F. R. Førsund, L. Hjalmarson and M. Suominen. (1993). “Banking Efficiency in the Nordic Countries.”
Journal of Banking and Finance 17, 371–388.
Bergendahl, G. (1995). “DEA and Benchmarks for Nordic Banks.” Working Paper, Gothenburg University,
Gothenberg, Sweden (December).
Berger, A. N. and R. DeYoung. (1997). “Problem Loans and Cost Efficiency in Commercial Banks.” Journal of
Banking and Finance 21(6), 849–870.
Berger, A. N. and D. B. Humphrey. (1997). “Efficiency of Financial Institutions: International Survey and Direc-
tions for Future Research.” European Journal of Operational Research.
Berger, A. N., D. Hancok and D. B. Humphrey. (1993). “Bank Efficiency Derived from the Profit Function.”
Journal of Banking and Finance 17(April), 317–347.
AN EFFICIENCY COMPARISON OF EUROPEAN BANKING SYSTEMS 77

Berger, A. N. and D. B. Humphrey. (1992). “Measurement and Efficiency Issues in Commercial Banking.”
In Z. Griliches (ed.), Output Measurement in the Service Sectors. Chicago: The University of Chicago Press,
Chap. 7, 245–279.
Berger, A. N. and D. B. Humphrey. (1991). “The Dominance of Inefficiencies over Scale and Product Mix
Economies in Banking.” Journal of Monetary Economics 28, 117–148.
Cooper, W. W. and J. T. Pastor. (1996). “Non-discretionary Variables in DEA: A Revision and a New Approach.”
Working Paper, Universidad de Alicante, Spain.
Charnes, A., W. W. Cooper and E. Rhodes. (1979). “Short Communication: Measuring the Efficiency of Decision
Making Units.” European Journal of Operational Research 3(4), 339–354.
DeYoung, R. (1998). “X-Inefficiency and Management Quality in Commercial Banks.” Journal of Financial
Services Research 13, 5–22.
DeYoung, R. and I. Hasan. (1998). “The Performance of Denovo Banks: A Profit Efficiency Approach.” Journal
of Banking and Finance 22, 565–587.
Dietsch, M. and A. Lozano-Vivas. (2000). “How the Environment Determines the Efficiency of Banks: A Com-
parison between French and Spanish Banking Industry.” Journal of Banking and Finance 24, 985–1004.
Evanoff, D. and P. Israilevich. (1991). “Regional Differences in Bank Efficiency and Technology.” The Annals of
Regional Science 25, 41–54.
Fecher, F. and P. Pestieau. (1993). “Efficiency and Competition in O.E.C.D. Financial Services.” In H. O. Fried,
C. A. K. Lovell and S. S. Schmidt (eds.), The Measurement of Productive Efficiency: Techniques and Applications.
U.K.: Oxford University Press, 374–385.
Fuentelsaz, L. and V. Salas. (1992). Estudios sobre banca al por menor. Fundación BBV.
Goldberg, L. G. and A. Rai. (1996). “The Structure-Performance Relationship for European Banking.” Journal of
Banking and Finance 20, 745–771.
Hughes, J. P. and L. Mester. (1993). “A Quality and Risk-Adjusted Cost Function for Banks: Evidence on the
‘Too-Big-To-Fail’ Doctrine.” Journal of Productivity Analysis 4, 293–316.
Jayaratne, J. and P. E. Strahan. (1996). “The Finance-Growth Nexus: Evidence from Bank Branch Deregulation.”
Quarterly Journal of Economics August, 641–670.
King, R. G. and R. Levine. (1993). “Finance and Growth: Schumpeter Might Be Right.” Quarterly Journal of
Economics 108, 717–737.
Levine, R. (1998). “The Legal Environment, Banks, and Long-Run Economic Growth.” Journal of Money, Credit,
and Banking 30, 596–613.
Levine, R. (1999). “Law, Finance, and Economic Growth.” Journal of Financial Intermediation 8, 8–35.
Levine, R. and S. Zervos. (1998). “Stock Markets, Banks, and Economic Growth.” American Economic Review
88, 537–558.
Maudos, J., J. M. Pastor, F. Pérez and J. Quesada. (2001). “Cost and Profit Efficiency in the European Banks.”
Journal of International Financial Markets, Institutions and Money, forthcoming.
Mester, L. (1993). “Efficiency in the Savings and Loan Industry.” Journal of Banking and Finance 17(2–3),
267–286.
Mester, L. (1997). “Measuring Efficiency at U.S. Banks.” European Journal of Operational Research 98(2),
230–242.
Pastor, J. M., F. Pérez and J. Quesada. (1997). “Efficiency Analysis in Banking Firms: An International Compar-
ison.” European Journal of Operational Research 98, 395–407.
Pastor, J. T. (1996). “Translation Invariance in Data Envelopment Analysis: A Generalization.” Annals of Opera-
tions Research 66, 93–102.
Pastor, J. T., J. L. Ruiz and I. Sirvent. (1999). “A Statistical Test for Detecting Influential Observations in DEA.”
European Journal of Operational Research 115, 542–554.
Pastor, J. T., J. L. Ruiz and I. Sirvent. (2001). “A Statistical Test for Nested Radial DEA Models.” Operations
Research, forthcoming.
Rouse, P. (1996). “Alternative Approaches to the Treatment of Environmental Factors in DEA: An Evaluation.”
Working Paper, University of Auckland, presented at the II Georgia Productivity Workshop.
Sheldon, G. (2001). “Efficiency and Scale Economies in European Banking: A Cross-Country Comparison.”
Working Paper, University of Basle.