You are on page 1of 203

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/294874383

Data Envelopment Analysis and its Applications

Book · February 2016


DOI: 10.13140/RG.2.1.4082.9202/2

CITATION READS

1 3,595

4 authors, including:

Ali Emrouznejad Mohsen Afsharian


Aston University Technische Universität Braunschweig
165 PUBLICATIONS 2,786 CITATIONS 20 PUBLICATIONS 79 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Data Envelopment Analysis Course at Aston University – Most recent run was 19 -20 June 2017. See below
for future runs View project

Network DEA View project

All content following this page was uploaded by Ali Emrouznejad on 10 March 2016.

The user has requested enhancement of the downloaded file.


PROCEEDINGS OF THE 13TH INTERNATIONAL CONFERENCE OF DEA
DEA2015, AUGUST 2015, BRAUNSCHWEIG, GERMANY

DATA ENVELOPMENT ANALYSIS


AND ITS APPLICATIONS
Proceedings of the 13th International Conference of DEA

EDITED BY:
Ali Emrouznejad, Rajiv Banker ,
Heinz Ahn and Mohsen Afsharian
I
DOI: 10.13140/RG.2.1.4082.9202

I
II
Title: Data Envelopment Analysis and its Applications

Subtitle (series): Proceedings of the 13th International Conference on Data


Envelopment Analysis

Venue: DEA2015, August 2015, Braunschweig, Germany

Edited by: Ali Emrouznejad, Rajiv Banker,


Heinz Ahn, and Mohsen Afsharian

Date of Publication: February 2016

ISBN: 978 1 85449 497 9

DOI: 10.13140/RG.2.1.4082.9202

Citation: Emrouznejad, A., R. Banker, H. Ahn and M. Afsharian (2016), Data


Envelopment Analysis and its Applications: Proceedings of the 13th International
Conference of DEA, August 2015, Braunschweig, Germany,
DOI: 10.13140/RG.2.1.4082.9202, ISBN: 978 1 85449 497 9.

III
Data Envelopment Analysis
and its Applications
Proceedings of the 13th International Conference of DEA, August 2015, Braunschweig, Germany

Ali Emrouznejad
Aston Business School
Aston University
Birmingham B4 7ET
UK

Rajiv Banker
Fox School of Business and Management
Temple University
Philadelphia, PA 19121
USA

Heinz Ahn
Technische Universität Braunschweig
Department of Business Sciences
Braunschweig
Germany

Mohsen Afsharian
Technische Universität Braunschweig
Department of Business Sciences
Braunschweig
Germany

February 2016
ISBN: 978 1 85449 497 9
DOI: 10.13140/RG.2.1.4082.9202

IV
V
PREFACE: A MESSAGE FROM THE LOCAL ORGANIZERS

On behalf of the organizing committee, we are pleased to warmly


welcome you to the 13th International Conference on Data
Envelopment Analysis (DEA2015) in Braunschweig. Being the
eldest of its kind in Germany, the Technische Universität (TU)
Braunschweig possesses a 250-year tradition of research and
teaching. It is located close to the center of Braunschweig, a
dynamic city founded in the early middle ages which lies at the foot
of the Harz, a lovely low mountain range.
The organizing committee of DEA2015 has been working
enthusiastically to bring together researchers and practitioners
interested in the development of the DEA methodology and its
applications to performance management in both public and
private sectors, with a special emphasis on regulation. Theoretical
and empirical papers on DEA and related fields in efficiency and
productivity analysis as well as performance measurement have
been invited.
As a conference participant, you will find a rich scientific program
with paper presentation sessions, keynotes and panel discussions;
also, DEA2015 offers a tutorial day including a number of pre-
conference tutorials about "Benchmarking with DEA" by top
international scholars. These academic events will provide you
with great opportunities to discuss your work and exchange your
ideas with experts from your field. You may also take full
advantage of the social program to meet your colleagues and
friends for the sake of networking and exchange of ideas in a more
relaxed environment.
Let us finally express our gratitude to each of you for joining us
and bringing a unique and valuable perspective to the conference.
Enjoy your stay in Germany and your visit to the dynamic city of
Braunschweig!
Heinz Ahn
Mohsen Afsharian
Chair of the Organizing Committee

VI
VII
TABLE OF CONTENTS
In the alphabetic order of titles

A TWO-STAGE PREDICTION MODEL FOR DEA EFFICIENCY SCORES ............................................... 1


ABDULLAH ALDAMAK ........................................................................................................................................... 1
SAEED ZOLFAGHARI .............................................................................................................................................. 1
ACADEMIC EFFICIENCY COMPARISON OF COUNTRIES VIA DEA ...................................................... 9
HASAN BAL .......................................................................................................................................................... 9
AYHAN GÖLCÜKCÜ............................................................................................................................................ 9
AN EVALUATION OF EFFICIENCIES FOR TURKEY’S CITIES WITH DATA ENVELOPMENT
ANALYSIS (DEA) IN TERMS OF CINEMAS: A SUPER EFFICIENCY APPLICATION ........................ 15
EMINE DEMET MECIT .......................................................................................................................................... 15
İHSAN ALP ........................................................................................................................................................... 15
ASSESSING PERFORMANCE OF CZECH PUBLIC TRANSPORT COMPANIES WITH TWO-STAGE
DEA ........................................................................................................................................................................ 19
MARKÉTA MATULOVÁ ........................................................................................................................................ 19
CONSTRUCTION OF UNIVERSITIES' TYPOLOGY VIA DEA .................................................................. 25
IRINA ABANKINA ................................................................................................................................................. 25
FUAD ALESKEROV ............................................................................................................................................... 25
VERONIKA BELOUSOVA ....................................................................................................................................... 25
LEONID GOKHBERG ............................................................................................................................................. 25
KIRILL ZINKOVSKY .............................................................................................................................................. 25
SOFYA KISELGOF ................................................................................................................................................. 25
VSEVOLOD PETRUSHCHENKO .............................................................................................................................. 25
SERGEY SHVYDUN ............................................................................................................................................... 25
DYNAMIC EFFICIENCY EVALUATION OF GERMAN PUBLIC MULTIDISCIPLINARY THEATRES
BY RC-DEA MODEL ........................................................................................................................................... 33
ANDREAS KLEINE ................................................................................................................................................ 33
STEFFEN HOFFMANN ........................................................................................................................................... 33
EFFICIENCY ANALYSIS OF STATE OWNED AND PRIVATE COMMERCIAL BANKS OPERATING
IN BANGLADESH: A DEA APPROACH .......................................................................................................... 39
MUHAMMAD AMIR HOSSAIN ............................................................................................................................... 39
EFFICIENCY MEASUREMENT OF REGIONAL DEVELOPMENT AGENCIES IN TURKEY ............. 47
V. ALPAGUT YAVUZ ............................................................................................................................................ 47
EFFICIENCY OF CONTAINER TERMINALS OF APEC COUNTRIES, 2000-2013: A
MEASUREMENT WITH A DYNAMIC DEA MODEL ................................................................................... 55
ODETTE V. DELFÍN-ORTEGA ............................................................................................................................... 55
CÉSAR L. NAVARRO-CHÁVEZ .............................................................................................................................. 55
EFFICIENCY OF MUNICIPAL GOVERNMENTS IN SLOVAKIA ............................................................. 61
MIROSLAV HUŽVÁR............................................................................................................................................. 61
ZUZANA RIGOVÁ ................................................................................................................................................. 61
KATARÍNA SÝKOROVÁ ........................................................................................................................................ 61

VIII
ESTIMATING RIGHT AND LEFT RETURNS TO SCALES IN THE PRESENCE OF UNDESIRABLE
FACTORS: A DEA APPROACH ........................................................................................................................ 67
MOHAMMAD KHOVEYNI ...................................................................................................................................... 67
ROBABEH ESLAMI................................................................................................................................................ 67
EVALUATION OF EFFICIENCY AND EFFECTIVENESS OF PUBLIC TRANSPORT SECTOR OF
INDIA USING DEA APPROACH ....................................................................................................................... 73
SHIVI AGARWAL .................................................................................................................................................. 73
EVALUATION ON THE EFFICIENCY OF LOGISTICS INDUSTRY IN WUHAN CITY CIRCLE........ 85
SHU LIU ............................................................................................................................................................... 85
LEI CHEN ............................................................................................................................................................. 85
EXAMINING THE EFFICIENCY OF MGNREGS IN NORTH EASTERN STATES OF INDIA USING
THE DATA ENVELOPMENT ANALYSIS APPROACH................................................................................ 91
INDRANEEL BHOWMIK ......................................................................................................................................... 91
PRITAM BOSE....................................................................................................................................................... 91
IDENTIFYING STRONG AND WEAK CONGESTION IN THE PRESENCE OF NEGATIVE DATA: A
DEA APPROACH ................................................................................................................................................. 99
ROBABEH ESLAMI................................................................................................................................................ 99
MOHAMMAD KHOVEYNI ...................................................................................................................................... 99
MEASUREMENT OF BANKING EFFICIENCY IN VISEGRAD COUNTRIES: A DEA WINDOW
ANALYSIS ........................................................................................................................................................... 105
IVETA PALEČKOVÁ ............................................................................................................................................ 105
MEASURING EFFICIENCY OF SALES TERRITORIES FOR MEDICAL DEVICES IN BRAZIL
THROUGH DATA ENVELOPMENT ANALYSIS (DEA) ............................................................................. 111
RENATO BARBOSA DA CUNHA ........................................................................................................................... 111
DENISE LINDSTROM BANDEIRA ......................................................................................................................... 111
CAMILO JOSÉ BORNIA-POULSEN........................................................................................................................ 111
MEASURING ENVIRONMENTAL EFFICIENCY OF THE 29 CHINESE PROVINCES:
APPLICATION OF THREE-STAGE DEA ..................................................................................................... 117
HAILING ZHAO................................................................................................................................................... 117
MEASURING THE MUTUAL FUND PERFORMANCE IN DATA ENVELOPMENT ANALYSIS ....... 123
HAOCHEN GUO .................................................................................................................................................. 123
EFFICIENCY IN THE UNIVERSITIES OF MEXICO 2008-2012: AN ANALYSIS THROUGH
DYNAMIC-NETWORK DEA MODEL ........................................................................................................... 131
CÉSAR L. NAVARRO-CHÁVEZ ............................................................................................................................ 131
ODETTE V. DELFÍN-ORTEGA ............................................................................................................................. 131
REGIONAL OPPORTUNITY COSTS OF ENVIRONMENTAL REGULATIONS IN TURKISH
MANUFACTURING INDUSTRY ..................................................................................................................... 139
YÜCEL ÖZKARA ................................................................................................................................................. 139
MEHMET ATAK .................................................................................................................................................. 139
SELECTING BEST SUPPLIER BY ADVANCED CROSS-EFFICIENCY (ACE) ..................................... 145
ELHAM ROSTAMIYAN ........................................................................................................................................ 145
FRANCK K. ADJOGBLE ....................................................................................................................................... 145
BEHDAD VATANI ............................................................................................................................................... 145

IX
SUPPORT OF EFFICIENT RESOURCE ALLOCATION OF TECHNOLOGICAL PROCESSES BY A
HEURISTIC SOLUTION AND AGENT TECHNOLOGY ............................................................................ 153
TIBOR DULAI ..................................................................................................................................................... 153
ÁGNES WERNER-STARK .................................................................................................................................... 153
GYULA ÁBRAHÁM ............................................................................................................................................. 153
TECHNICAL AND SCALE EFFICIENCY OF SCHEDULED BANKS OF BANGLADESH IN TERMS
OF SME FINANCE USING DATA ENVELOPMENT ANALYSIS (DEA) .................................................. 159
SHAMIM ARA ..................................................................................................................................................... 159
HAZERA AKTER ................................................................................................................................................. 159
THE DIFFERENT APPROACHES OF BANKING EFFICIENCY: A META-ANALYSIS....................... 167
LUCAS SERRÃO MACORIS .................................................................................................................................. 167
ALEXANDRE PEREIRA SALGADO JUNIOR ........................................................................................................... 167
EDUARDO FALSARELLA JÚNIOR......................................................................................................................... 167
USING DYNAMIC DEA TO EVALUATE THE EFFECTIVENESS OF MACHINE-BUILDING
ENTERPRISES OF UKRAINE ......................................................................................................................... 173
LEONID GALCHYNSKY ....................................................................................................................................... 173
CONCLUSIONS .................................................................................................................................................... 178
REFERENCES ...................................................................................................................................................... 178
X-EFFICIENCY OF INDIAN COMMERCIAL BANKS AND THEIR DETERMINANTS OF SERVICE
QUALITY: A STUDY OF POST GLOBAL FINANCIAL CRISIS ............................................................... 181
GAGANDEEP SHARMA........................................................................................................................................ 181
DIVYA SHARMA ................................................................................................................................................. 181

X
XI
A TWO-STAGE PREDICTION MODEL FOR DEA EFFICIENCY SCORES
ABDULLAH ALDAMAK
Department of Mechanical and Industrial Engineering, Ryerson University, Toronto, ON, Canada

SAEED ZOLFAGHARI
Department of Mechanical and Industrial Engineering, Ryerson University, Toronto, ON, Canada

ABSTRACT
The standard data envelopment analysis (DEA) model usually evaluates decision-making units (DMUs) using the best
relative efficiency approach. This approach is known as the CCR model or optimistic approach, where all DMUs are
evaluated according to the efficiency frontier. The pessimistic DEA model is another well-known DEA model that
ranks all DMUs according to the inefficiency frontier. Results across both models have large variation. In the
literature, it is well known that using only one model leads to biased evaluation, also it is known that both models are
solely used for data evaluation and alternative models are suggested for efficiency prediction. In this paper, we
propose a DEA model for score prediction and efficiency evaluation by combining both approaches after
incorporating regression analysis into each model. The second stage of the proposed model combines both approaches
to obtain a final predicted score.

Keywords: regression analysis; ranking; prediction; optimistic DEA; pessimistic DEA.

INTRODUCTION
To measure the efficiency of decision-making units (DMUs), Data Envelopment Analysis (DEA) is often
used. Charnes, Cooper, and Rhodes [1] introduced it more than 40 years ago when they developed their
CCR model, which remodeled the fractional linear measure of efficiency into linear programming. As a
unique ability, DEA does not need to assign prior weights to inputs and outputs when measuring the
efficiency of multiple input and output DMUs. Since its introduction, a tremendous amount of research has
been conducted on DEA, and this has generated a large body of literature.
Many empirical applications exist for DEA in many different sectors, like education [2], banking and
economics [3, 4], manufacturing [5], logistics [6], telecommunications [7], healthcare [8], and athletics [9,
10]. The versatility of DEA as a decision analysis tool comes from its ability to assign every unit in a dataset
its own production function; it does not need to discover a universal relationship between all units in a
sample. Comparing the efficiency of a single unit to other units in the dataset, DEA can evaluate the
efficiency of that single unit. After completing this process for every unit in the dataset, all units are
classified into either an efficient group, which possess 100% efficiency scores, or an inefficient group,
which possess less than 100% efficiency scores.
DEA’s classification of the dataset into two groups has its strengths and weaknesses. As a strength, DEA
can evaluate the efficiency of any dataset; however, as a weakness, it lacks the discrimination power that
creates a full ranking of all the efficient units in the dataset. For most decision makers, it is not enough to
simply classify the data into efficient or inefficient. More often, decision makers aim to achieve
distinguished scores for all the units in the data under evaluation. In order to achieve distinguished scores,
they must either modify DEA or add another method to fully rank all DMUs under assessment [11].

1
On the other hand, regression analysis (RA) is a well-known comparative efficiency technique that is widely
used to determine the efficiency of data units. RA uses least square methods to produce the average line,
while DEA uses linear programming to obtain efficiency frontier lines. Both DEA and RA focus on finding
or estimating the efficiency frontier of the data, and for that reason, they are considered equivalent
alternatives. However, DEA and RA result in different efficiency rankings, with major variations on their
respective efficiency scores.
This paper suggests a new method that predicts the score of any possible data point by combining regressed
optimistic DEA and regressed pessimistic DEA. The method is discussed in detail in the next three sections
of the paper, and illustrative example is followed. The method shows how the new model predicts the
efficiency score of any DMU within or outside the dataset in reference to that data.

OPTIMISTIC APPROACH
Optimistic DEA Model
In 1957, Farrell [12] implemented DEA’s basic efficiency measurement: the ratio of output to input in cases
that involve a single input and output. From this efficiency measurement, they developed efficiency frontier
analysis. The efficiency frontier is constructed by graphing input and output of the DMUs along two-axes.
To be defined as efficient units and receive a 100% efficiency score, units must lie on the efficiency frontier.
In contrast, inefficient units do not lie on the efficiency frontier. Their efficiency scores are calculated from
the distances of their graphed positions from the frontier line. Lastly, DEA is called Data Envelopment
Analysis because the frontier line envelops the whole dataset, as shown in [13].
As noted above, in 1978, twenty-two years after Farrell developed the basic efficiency measurement,
Charnes et al. [1] transformed envelopment analysis from its graphical form into a linear program. Their
linear program requires no restrictions on the number of inputs or outputs. Furthermore, their CCR model
assigns virtual weights to inputs and outputs, and this allows the model to measure efficiency without
assigning prior weights to the inputs and outputs. On the virtual weights, the model applies a linear program
that maximizes the efficiency ratio of DMUs. Finally, the linear program is applied to all DMUs in the
dataset one by one.
To clarify this, let us suppose there are n DMUs with s outputs and m inputs, so for every DMUk there are:

Virtual input 𝑉1 𝑋1𝑘 + … + 𝑉𝑚 𝑋𝑚𝑘 and

Virtual output𝑈1 𝑌1𝑘 + … + 𝑈𝑠 𝑌𝑠𝑘 


Where the virtual weight of the inputs are represented by Vi (i=1,..., m) and the virtual weight of the outputs
are represented by Ur (r=1, ..., s). The following function represents the DEA efficiency of any DMU:
𝑉𝑖𝑟𝑡𝑢𝑎𝑙 𝑜𝑢𝑡𝑝𝑢𝑡
𝐷𝐸𝐴 𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 = 
𝑉𝑖𝑟𝑡𝑢𝑎𝑙 𝑖𝑛𝑝𝑢𝑡
In the CCR model, the weights of each DMU are selected so as to maximize the efficiency ratio. Charnes,
Cooper, and Rhodes maximized the efficiency of each DMUk by applying linear programming and solving
the following fractional program:

2
𝑢1 𝑦1𝑘 + 𝑢2 𝑦2𝑘 + … + 𝑢𝑠 𝑦𝑠𝑘
Max 𝜃𝑘 =
𝑣1 𝑥1𝑘 + 𝑣2 𝑥2𝑘 + … + 𝑣𝑚 𝑥𝑚𝑘

Subject to:
𝑢1 𝑦1𝑘 +𝑢2 𝑦2𝑘 + …+𝑢𝑠 𝑦𝑠𝑘 (1)
0≤ ≤ 1 ; (𝑘 = 1 … , 𝑛)
𝑣1 𝑥1𝑘 +𝑣2 𝑥2𝑘 + …+𝑣𝑚 𝑥𝑚𝑘

𝑣1 , 𝑣2 , … , 𝑣𝑚 ≥ 0

𝑢1 , 𝑢2 , … , 𝑢𝑠 ≥ 0

In the above program, the solution is calculated n times for every DMU in the data, each with its own
efficiency score. The fractional model of CCR can be transformed into a linear form by normalizing the
denominator of the objective function. The model must also assume that no DMU will have a negative
efficiency. The following objective function represents the maximization of the weighted sum of the output
of the linear form:

Max 𝜃𝑘 = 𝑢1 𝑦1𝑘 + 𝑢2 𝑦2𝑘 + … + 𝑢𝑠 𝑦𝑠𝑘

Subject to:

𝑣1 𝑥1𝑘 + 𝑣2 𝑥2𝑘 + … + 𝑣𝑚 𝑥𝑚𝑘 = 1

𝑢1 𝑦1𝑘 + 𝑢2 𝑦2𝑘 + … + 𝑢𝑠 𝑦𝑠𝑘 ≤ 𝑣1 𝑥1𝑘 + 𝑣2 𝑥2𝑘 + … + 𝑣𝑚 𝑥𝑚𝑘 (2)

(𝑘 = 1, … , 𝑛)

𝑣1 , 𝑣2 , … , 𝑣𝑚 ≥ 0

𝑢1 , 𝑢2 , … , 𝑢𝑠 ≥ 0

So each DMU is classified efficient if θk obtains a value of 1 [14]. The CCR model is also known as the
optimistic model and the best relative efficiency model.
Regressed Optimistic DEA Model
The previous section discussed a standard DEA model that used an optimistic approach to rank all DMUs
based on the efficiency frontier. Along this frontier all units are considered efficient or optimistically
efficient DMUs. Furthermore, as modeled in CCR, scores are equal to 1 for all efficient DMUs.
In this project’s two-stage model, the basic DEA model will be applied on the dataset in the first stage.
Based on Arnold et al. [15], a traditional, single output regression analysis will be applied to the DEA
results in the second stage. In our model, we defined the DEA efficiency score as a final optimum output
for the regression function. The RA model of the score function was as follows:

𝜃𝑘𝑜 = 𝜆0 + ∑𝑠𝑖=1 𝛽𝑥𝑜 𝑖 𝑋𝑘𝑖 + ∑𝑚


𝑙=1 𝛽𝑦𝑜 𝑙 𝑌𝑘𝑙 ≤ 1 + 𝜀 (3)

where 𝜃𝑘𝑜 is the estimated optimistic score for DMUk, whether included in the dataset or from another
dataset; λ0 is optimistic intercept; Xki is the value of the ith output of DMUk; Ykl is the value of the lth input
of DMUk; and ε is the random error. Furthermore, we can conclude that 𝜃𝑘𝑜 ≤1+ε.

3
After applying both stages and obtaining the regressed optimistic function 𝜃𝑘𝑜 , the score of any DMU in
the data can be easily calculated. However, as Entani et al. noted [16], approaches that utilize only the
optimistic approach is biased. To eliminate this bias, our model also includes a pessimistic approach.

PESSIMISTIC APPROACH
Pessimistic DEA Model

The CCR model, an example of the optimistic approach, aims to optimize all units towards a maximum
efficiency score of 1. In contrast, the pessimistic approach evaluates all DMUs along an inefficiency
frontier. Rather than maximizing efficiency ratio, ratios are minimized in the fractional program of the
pessimistic approach. Inefficient DMUs receive a score of 1, and all constraints in the pessimistic approach
should be greater or equal to 1. The following represents the fractional pessimistic model for any DMU:
𝑢1 𝑦1𝑘 +𝑢2 𝑦2𝑘 + …+𝑢𝑠 𝑦𝑠𝑘
Min 𝜃𝑘∗ =
𝑣1 𝑥1𝑘 +𝑣2 𝑥2𝑘 + …+𝑣𝑚 𝑥𝑚𝑘

subject to:
𝑢1 𝑦1𝑘 + 𝑢2 𝑦2𝑘 + … + 𝑢𝑠 𝑦𝑠𝑘
≥1
𝑣1 𝑥1𝑘 + 𝑣2 𝑥2𝑘 + … + 𝑣𝑚 𝑥𝑚𝑘 (4)

(𝑘 = 1 … , 𝑛)
𝑣1 , 𝑣2 , … , 𝑣𝑚 ≥ 0

𝑢1 , 𝑢2 , … , 𝑢𝑠 ≥ 0

Regressed Pessimistic Approach


In our proposal, we suggest introducing regression analysis to obtain efficiency scores using a pessimistic
approach. After applying the RA, the final pessimistic efficiency scores can be calculated as follows:

𝜃𝑘𝑝 = 𝜆𝑝 + ∑𝑠𝑖=1 𝛽𝑥𝑝 𝑖 𝑋𝑘𝑖 + ∑𝑚


𝑙=1 𝛽𝑦𝑝 𝑙 𝑌𝑙 ≥ 1 − 𝜀 (5)

Where 𝜃𝑘𝑝 is the estimated pessimistic score for DMUk, whether included in the dataset or not; 𝜆𝑝 is the
regressed pessimistic intercept; 𝑋𝑘𝑖 is the value of the ith output of DMUk; Ykl is the value of the lth input
of DMUk; and ε is the random error. Furthermore, we can conclude that 𝜃𝑘𝑝 will always be ≥1- ε.

COMBINING OPTIMISTIC AND PESSIMISTIC APPROACHES


Both optimistic and pessimistic models can be used to obtain efficiency scores of a data-point or even
predict a score of a virtual data point. However, relying on one approach lacks precision and introduces
bias. Each model may lead to an inaccurate efficiency score; nevertheless, using one model over the other
is biased, as shown by Entani et al. [16].
Therefore, unbiased methods must combine pessimistic and optimistic approaches. This combination can
be achieved by calculating the geometric average of both scores. Wang et al. provided a detailed theorem
and proof of this [17]. They advocated ranking all DMUs by taking the geometric average between the

4
pessimistic and optimistic scores of each DMU. Following Wang et al., this paper proposes to combine a
regressed optimistic approach with a regressed pessimistic approach by taking their geometric average. The
final efficiency score prediction formula of our method can be calculated as follows:

𝑓𝑖𝑛𝑎𝑙 𝜆0 + ∑𝑠𝑖=1 𝛽𝑥𝑜 𝑖 𝑋𝑘𝑖 + ∑𝑚


𝑙=1 𝛽𝑦𝑜 𝑙 𝑌𝑘𝑙
𝜃𝑘 = √ 𝑠 𝑚 −1
(6)
(𝜆 𝑝 + ∑𝑖=1 𝛽𝑥𝑝 𝑖 𝑋𝑘𝑖 + ∑𝑙=1 𝛽𝑦𝑝 𝑙 𝑌𝑘𝑙 )

ILLUSTRATIVE EXAMPLE
As an example of the new model proposed by this paper, the multiple inputs and outputs from the branches
of a hypothetical bank are provided and analyzed. Table 1 shows the raw data of 12 bank branches. (For a
clarification of the details see [18].) Table 2 shows the results of the optimistic model, pessimistic model,
regressed optimistic model, regressed pessimistic model, and the final score of our proposed model.

Table 1: Raw data for bank branches evaluation


Input Output
Interest paid per Interest paid per
Branch Employee Operating cost Non-interest income
saving loan
1 20 829,326 4,449,202 4,786,608 1,000,188
2 7 342,554 1,020,605 1,686,859 307,375
3 7 262,008 861,443 1,516,144 426,604
4 11 301,114 4,022,446 6,491,851 1,152,494
5 9 244,918 400,783 654,434 407,243
6 6 326,759 3,056,784 1,994,946 1,055,240
7 7 269,277 1,634,220 2,291,636 1,083,105
8 6 288,521 1,232,645 1,788,427 1,001,151
9 4 165,573 445,955 904,764 462,190
10 6 218,150 536,914 1,036,494 545,877
11 3 132,788 229,635 387,528 160,227
12 23 924,037 4,879,496 8,471,185 470,160

Figure 1: Score comparisons of all models with the proposed final model

It can be observed that both the regressed optimistic model and regressed pessimistic model are consistent
with original optimistic or pessimistic DEA. Figure 1 shows consistency between the proposed model and

5
all other models. This allows us to rely on the model for prediction. In addition, overcoming the issue of
bias within each model, the final proposed model yields accurate results, and it can be used for efficiency
score prediction on any virtual data point that does not belong to the data but decision maker interested in
evaluating it in reference to this data. Moreover, the proposed model can also be used to rank the existing
data when discrimination among DMUs is needed.

CONCLUSIONS
Since 1978, when DEA first emerged, a large body of research has been conducted on DEA, resulting in
major developments in its methods and applications [19]. A sub-field of research, focusing on score
prediction and the ranking of DMUs, has emerged. To the best of our knowledge, the combination of
optimistic and pessimistic DEAs with regression models has not yet been explored. In our proposed model,
we suggest a two-stage model for determining and especially predicting efficiency scores for any possible
data point. The first stage of the model individually applies RA to optimistic and pessimistic DEA models
in order to obtain regressed optimistic scores and regressed pessimistic scores. In the second stage, both
scores are combined together through a geometric average technique. Combining both scores yields much
more accurate results. Moreover, even though this model is used for score prediction, it can also rank
efficient DMUs, since final scores of all DMUs tends to be much more discriminated than the scores
generated from the standard DEA model.

Table 2: Comparison results of the proposed model


Regressed Regressed Proposed Model
Pessimistic Pessimistic
Optimistic model Optimistic 𝑓𝑖𝑛𝑎𝑙
Branch model (𝜃𝑘 )
Model (𝑌𝑘𝑜 ) Model (𝑌𝑘𝑝 )
Score Rank Score Rank Score Rank Score Rank Score Rank
1 0.512 8 1.189 8 0.613 6 1.735 7 1.032 7
2 0.413 11 1.000 9 0.513 10 1.349 9 0.831 9
3 0.456 9 1.370 7 0.526 9 1.369 8 0.849 8
4 1.000 1 3.005 4 1.064 1 3.279 4 1.867 2
5 0.413 10 1.000 9 0.329 12 0.507 12 0.409 12
6 1.000 1 3.968 1 0.895 4 3.499 2 1.770 4
7 1.000 1 3.548 3 0.996 3 3.442 3 1.851 3
8 0.996 4 3.736 2 0.999 2 3.510 1 1.873 1
9 0.729 5 1.646 5 0.624 5 1.886 5 1.085 5
10 0.622 7 1.504 6 0.612 7 1.781 6 1.044 6
11 0.342 12 1.000 9 0.409 11 1.033 10 0.650 10
12 0.624 6 1.000 9 0.527 8 0.576 11 0.551 11

ACKNOWLEDGEMENT
Abdullah Aldamak wishes to thank King Abdulaziz University (KAU), Jeddah, Saudi Arabia for providing
a full scholarship to carry out this research work.

REFERENCES
[1] A. Charnes, W.W. Cooper, E. Rhodes (1978) Measuring the efficiency of decision making units, European
Journal of Operational Research, 2 (6): 429-444.

6
[2] A. Bessent, W. Bessent, J. Kennington, B. Reagan (1982) An Application of Mathematical Programming to
Assess Productivity in the Houston Independent School District, Management Science, (28): 1355-1367.
[3] E. Thanassoulis (1999) Data Envelopment Analysis and Its Use in Banking, Interfaces, (29): 1-13.
[4] A. Emrouznejad (2003) An alternative DEA measure: a case of OECD countries, Applied Economics Letters,
(10): 779-782.
[5] M.I.M. Wahab, D. Wu, C.-G. Lee (2008) A generic approach to measuring the machine flexibility of
manufacturing systems, European Journal of Operational Research, (186): 137-149.
[6] J. Xu, B. Li, D. Wu (2009) Rough data envelopment analysis and its application to supply chain performance
evaluation, International Journal of Production Economics, (122):628-638.
[7] W.W. Cooper, K.S. Park, G. Yu (2001) An Illustrative Application of IDEA (Imprecise Data Envelopment
Analysis) to a Korean Mobile Telecommunication Company, Operations Research, (49):807-820.
[8] R. Jacobs (2001) Alternative Methods to Examine Hospital Efficiency: Data Envelopment Analysis and
Stochastic Frontier Analysis, Health Care Management Science, (4):103-115.
[9] W.W. Cooper, J.L. Ruiz, I. Sirvent (2009) Selecting non-zero weights to evaluate effectiveness of basketball
players with DEA, European Journal of Operational Research, (195): 563-574.
[10] [10] T.R. Sexton, H.F. Lewis (2003) Two-Stage DEA: An Application to Major League Baseball, Journal of
Productivity Analysis, (19):227-249.
[11] N. Adler, L. Friedman, Z. Sinuany-Stern (2002) Review of ranking methods in the data envelopment analysis
context, European Journal of Operational Research, (140):249-265.
[12] M.J. Farrell (1957) The Measurement of Productivity Efficiency, Journal of the Royal Statistical Society, 120
(A):253-281.
[13] R. Ramanathan (2003) An introduction to data envelopment analysis: a tool for performance measurement
Thousand Oaks, Calif. : Sage Publications.
[14] W.W. Cooper, L.M. Seiford, J. Zhu (2004) Handbook on data envelopment analysis.
[15] V.L. Arnold, I.R. Bardhan, W.W. Cooper, S.C. Kumbhakar (1996) Chapter 12 New uses of DEA and statistical
regressions for efficiency evaluation and estimation — with an illustrative application to public secondary schools
in Texas, Annals of Operations Research, (66): 255-277.
[16] T. Entani, Y. Maeda, H. Tanaka (2002) Dual models of interval DEA and its extension to interval data, European
Journal of Operational Research, (136):32-45.
[17] Y.M. Wang, K.S. Chin, J.B. Yang (2006) Measuring the performances of decision-making units using geometric
average efficiency, Journal of the Operational Research Society, (58):929-937.
[18] L.H. Čaklović, Tihomir (2012) Measurement of DMU-efficiency by Modified Cross Efficiency approach,
Mathematical communications, (17):1331-0623.
[19] W.D. Cook, L.M. Seiford (2009) Data envelopment analysis (DEA) – Thirty years on, European Journal of
Operational Research, (192):1-17.

7
8
ACADEMIC EFFICIENCY COMPARISON OF COUNTRIES VIA DEA
HASAN BAL
Department of Statistics, Faculty of Science, Gazi University, Ankara, Turkey (Corresponding author)

AYHAN GÖLCÜKCÜ
Center of Statistical Consultancy, Applications and Research, Süleyman Demirel University, Isparta, Turkey

ABSTRACT
The number of academic studies and the H index derived from the citations made to these academic studies per year
are the indicator of academic progress not only for an academician but also for the countries, or institutions. In this
context, Data Envelopment Analysis (DEA) is a relevant tool for the academic comparison of countries. Population
which is the source of human capital and GDP per capita which is the indicator of financial support potential could
be determined as inputs while the number of studies and H index are output. In our study, 177 countries which have
available data segmented according to mean and median of their inputs and compare each other via BCC and CCR
models of DEA. Later, an efficient set is evaluated by the efficient countries of each segment and these efficient
countries compared each other.

Keywords: Data Envelopment Analysis, Academic efficiency comparison, countries, CCR, BCC

INTRODUCTION
The perspective of this study is the academic progress of countries, one of the main indicator of the wealth
of nations. That is, undeniable fact and a cycle, economic development produces academic outcomes and
academic progress is a driving force of economy. Rapid growth of countries mainly depend on
technological prosperity and logically sequence to reach economical prosperity can be given as education,
qualified staff, qualified high technology, production of high technology goods and last economic growth.
This is also an important issue in terms of the sustainability of economic development. At the center of
technological improvement research and development attempts within universities and academic activities
play a major role at this point. On the other hand this phenomenon has a cost on economies. Our study is
an effort to find out the countries which uses its human capital and economy efficiently to produce academic
outcome.
In literature despite of DEA, Hanushek studied various aspects of education and derive educational
production function. (Hanushek, 1970; Hanushek, 1979; 1986; 2003; Hanushek and Wößmann, 2007;
Hanushek, 2010) . On the other side Rugirerio applied DEA to education various times with some authors
(Ruggiero, 1996; Ruggiero, 2000; Ruggiero, 2004; 2006; Johnson and Ruggiero, 2014) in addition to these
studies there are many attempt of using DEA to evaluate the efficiency of educational activities . (Beasley,
1990; Abbott and Doucouliagos, 2003; Abramo and D'Angelo, 2009; Agasisti and Pérez-Esparrells, 2009;
Agasisti, 2011; Agasisti et al., 2011; Agha et al., 2011; Alwadood et al., 2011; Agasisti et al., 2012; Aziz
et al., 2013; Bayraktar et al., 2013; Altamirano-Corro and Peniche-Vera, 2014) These studies mainly
concerns, compares schools, universities or university departments or economical sustainability of
education. Cross country comparisons also made for this purpose and regionally limited. In this study all
of the countries which have available data included to analyze and compared each others in respect of the
academic outcome produces performance by the use of human capital and share of economic wealth. For
this purpose, DEA is applied to 177 countries as a decision making units (DMUs). Population, which is the

9
source of human capital and GDP per capita, which is the indicator of financial support potential could be
determined as inputs while the number of studies and H index are output.

METHODS
Variables and Data
In this study to compare countries according to their academic productivity in respects of their population
and average wealth share, four variables are used. Population (POP) which is the source of human capital
and GDP per capita (GDPpC) which is the indicator of financial support potential could be determined as
inputs while the number of studies (NoS) and H index (H index) are output. While the outputs of our study
are taken from internet portal of The SCImago Journal & Country Rank, the inputs are taken from World
Economic Outlook Database of IMF internet page for the available most recent year of 2013. Only 177
countries are common in both database and have available data.
Model and Application
Afterwards, to specify overall efficient countries CCR model of DEA is used (Charnes et al., 1978). Further,
to identify best performing countries on their scale BCC model of DEA is used (Banker et al., 1984). The
used models are given below as (1) and (2) respectively.

max w0   r mr yr 0 max w0   r mr yr 0  uo
s.t.; s.t.;

 i
vi xi 0  1
(1),  i
vi xi 0  1
(2)
 r
mr yrj  i vi xij  0  r
mr yrj  i vi xij  uo  0
mr ; vi   mr ; vi  
j  1,..., n; r  1,..., s; i  1,..., t j  1,..., n; r  1,..., s; i  1,..., t
The data has wide range as could be seen in Table 1. Especially in input data the range for population is
over 1 billion and similar for GDPpC. Thus, Countries segmented according to their mean and median
whether they are upper or lover than mean and median. The number inside bracket in Table 2 shows
segments and the numbers are the count of countries after segmentation

Table1: Descriptive Statistics of the Data


NoS H index GDPpC POP
N 177 177 177 177
Minimum 4 (Sao Tome and Principe) 10 (Timor-Leste) 223,398 (Malavi) 0,018 (Palau)
Maximum 563292 (USA) 1518(USA) 112472,530 (Luxembourg) 1360,763 (China)
Range 563288 1508 112249,132 1360,745
Mean 16599,79 155,41 14190,59449 39,59209
Std. Error of
4363,186 15,314 1499,020884 10,720925
Mean
Std. Deviation 58048,421 203,734 19943,175755 142,632634
Median 712,00 81,00 5881,55100 8,48400
Kurtosis 58,278 12,774 6,554 70,187
Skewness 7,081 3,032 2,381 8,099

10
Table 2: Segmentation Matrix and number of countries due to segmentation*
POP
Mean Median
under over
over world
under world mean
mean
Total world world Total
median median
Code 0 1 0 1
over world mean 0 103 (2) 42 (1) 145 41 (2) 48 (1) 89
GDPpC
under world mean 1 22 (4) 10 (3) 32 48 (4) 40 (3) 88
Total 125 52 177 89 88 177
 The number inside brackets shows the segments and outside brackets are the number of countries

Successively, after grouping data, model (1) and (2) applied for each group separately. Subsequently, for
determining a common efficient set, efficient DMUs of segments combined and analyzed with model (1)
and (2) as a second stage for mean and median groups separately.

RESULTS AND DISCUSSIONS


The number of countries found to be efficient is presented in Table 3 according to segments. According to
separation of mean, totally 29 countries are found efficient for CCR model and 47 countries are found
efficient for BCC model. Similarly according to separation of median, 28 countries are found efficient for
CCR model and 50 countries are found efficient for BCC model. In the subsequent analysis, for finding
overall efficient DMUs a new CCR model is applied to the set of CCR efficient countries for mean and
median separately, then BCC model is applied to the set of BCC efficient countries to find out technical
efficient countries. The results are shown in Table 3 and 4 respectively.

Table 3: Number of efficient DMUs in first stage (countries)


GDPpC
Mean Median
0 1 Total 0 1 Total
0 8 11 19 7 8 15
CCR
1 8 2 10 5 8 13
Total 16 13 29 12 16 28
POP
0 15 16 31 12 13 25
BCC
1 11 5 16 10 15 25
Total 26 21 47 22 28 50
0: less than; 1: Greater than

Table 4: Overall efficient countries


DMU Country Documents H index GDP per Capita Population Segment of mean Segment of Median
1 United States 563292 1518 53000.972 316.373 1 1
2 China 425677 436 6958.686 1360.763 3 1
3 UK 162574 934 39371.653 64.087 1 1
7 India 106029 341 1509.495 1243.337 3 3
11 Australia 76357 583 64578.190 23.319 2 1
14 Netherlands 50939 636 50815.753 16.802 2 1
18 Switzerland 38450 629 81276.126 8.003 2 2
21 Sweden 33033 567 58014.168 9.635 2 1
24 Denmark 21382 476 59129.127 5.591 2 2
62 Estonia 2428 148 18852.091 1.320 2 2
78 Iceland 1299 181 45415.578 0.322 2 2
106 Malawi 423 89 223.398 17.111 4 3
134 Gambia 149 85 453.056 1.876 4 4
197 Palau? 14 23 14021.839 0.018 4 2

11
Table 5: Technical efficient countries
GDP per Segment Segment of
DMU Country Documents H index Capita Population of mean Median
29 Israel 17278 456 36926.33 7.871 2 2
44 Hungary 9379 277 13387.975 9.879 4 1
65 Kenya 1955 149 1315.623 41.8 3 3
145 Grenada 110 27 7696.876 0.106 4 2
172 Central African Republic 37 36 333.604 4.611 4 4
177 Dominica 34 18 7031.173 0.071 4 2
189 Comoros 20 11 927.847 0.709 4 4
212 Tonga 6 17 4572.056 0.104 4 4
215 Sao Tome and Principe 4 15 1625.118 0.191 4 4

Table 6a: Efficiency of Some Selected countries


DATA Mean Median
D H Seg Seg
M Ra Docum inde GDP per Populatio me me
U nk Country ents x Capita n nt CCR BCC nt CCR BCC
First 5 and last 5 countries according to Gross Domestic Product per Capita (GDPpC) and their efficiencies based on their
segment
74 1 Luxembourg 1494 92 112472.530 0.537 2 0.620 0.675 2 0.620 0.629
32 2 Norway 16863 362 100578.968 5.096 2 0.694 0.695 2 0.694 0.694
69 3 Qatar 1690 60 98985.986 2.045 2 0.177 0.210 2 0.177 0.184
18 4 Switzerland 38450 629 81276.126 8.003 2 1.000 1.000 2 1.000 1.000
11 5 Australia 76357 583 64578.190 23.319 2 1.000 1.000 1 1.000 1.000
147 174 Niger 104 51 446.526 16.601 4 0.403 0.631 3 0.568 0.722
166 175 Democratic Republic Congo 45 33 388.305 76.991 3 0.527 1.000 3 0.213 0.575
172 176 Central African Republic 37 36 333.604 4.611 4 0.500 1.000 4 0.575 1.000
170 177 Burundi 38 26 303.004 8.985 4 0.328 0.979 3 0.524 1.000
106 178 Malawi 423 89 223.398 17.111 4 1.000 1.000 3 1.000 1.000
First 5 and last 5 countries according to Population (POP) and their efficiencies based on their segment
2 1 China 425677 436 6958.686 1360.763 3 1.000 1.000 1 1.000 1.000
7 2 India 106029 341 1509.495 1243.337 3 1.000 1.000 3 1.000 1.000
1 3 United States 563292 1518 53000.972 316.373 1 1.000 1.000 1 1.000 1.000
55 4 Indonesia 4175 126 3509.824 247.954 3 0.262 0.281 3 0.167 0.242
13 5 Brazil 59111 342 11172.523 201.032 3 0.715 1.000 1 0.785 0.822
212 174 Tonga 6 17 4572.056 0.104 4 0.919 1.000 4 1.000 1.000
165 175 Seychelles 46 38 14918.411 0.093 2 0.727 1.000 2 0.700 0.902
203 176 Antigua and Barbuda 10 10 13733.958 0.087 4 0.304 0.650 2 0.198 0.650
177 177 Dominica 34 18 7031.173 0.071 4 0.865 1.000 2 0.614 1.000
197 178 Palau 14 23 14021.839 0.018 4 1.000 1.000 2 1.000 1.000

All of the tables given above, namely, Table 3-6 states that neither population nor gross domestic product
per capita which can be seen as the indicator of wealth are alone determine the academic performance.
There are also less populated small economies that are using their resources efficiently for academic
purposes like Palau, Dominica, Malawi, and Central African Republic as can be seen in Table 6a. Besides
these results, Table 6a also shows that recent economic prosperity does not a sign of academic achievement.
The efficiency of Luxembourg, Norway and Qatar supports this idea.

12
Table 6b: Efficiency of Some Selected countries
DATA Mean Median
DMU Docum H GDP per Populatio Segm Segm
No Country ents index Capita n ent CCR BCC ent CCR BCC
The countries that are CCR efficient when segmented according to mean but inefficient when segmented to median*
9 Canada 88711 725 52037.148 35.105 2 1.000 1.000 1 0.925 0.925
16 Taiwan 41188 300 20924.921 23.374 2 1.000 1.000 1 0.720 0.898
19 Turkey 37446 237 10721.063 76.484 3 1.000 1.000 1 0.632 0.785
20 Poland 34933 336 13435.263 38.533 4 1.000 1.000 1 0.935 0.978
27 Czech Republic 18029 268 18871.009 10.516 2 1.000 1.000 1 0.879 1.000
50 Croatia 5928 161 13401.419 4.281 4 1.000 1.000 2 0.903 1.000
65 Kenya 1955 149 1315.623 41.800 3 1.000 1.000 3 0.684 1.000
77 Ethiopia 1365 82 517.713 88.850 3 1.000 1.000 3 0.409 0.454
85 Tanzania 993 102 719.259 46.277 3 1.000 1.000 3 0.437 0.615
145 Grenada 110 27 7696.876 0.106 4 1.000 1.000 2 0.798 1.000
The countries that are CCR efficient when segmented according to median but inefficient when segmented to mean*
47 Serbia 6992 86 5902.489 7.199 4 0.983 1.000 2 1.000 1.000
51 Tunisia 5672 97 4316.795 10.887 4 0.638 0.661 3 1.000 1.000
56 Bulgaria 3654 154 7328.488 7.238 4 0.688 0.691 2 1.000 1.000
63 Jordan 2298 92 5174.020 6.544 4 0.504 0.525 4 1.000 1.000
86 Armenia 980 116 3173.083 3.287 4 0.804 0.950 4 1.000 1.000
95 Macedonia 665 67 4931.401 2.071 4 0.672 0.681 4 1.000 1.000
105 Zimbabwe 430 81 1006.825 13.117 4 0.441 0.465 3 1.000 1.000
194 Samoa 16 22 4164.970 0.190 4 0.975 0.987 4 1.000 1.000
212 Tonga 6 17 4572.056 0.104 4 0.919 1.000 4 1.000 1.000
 Efficiencies based on the segment of the countries

Table 6b depicts the segment change according to separation criteria. In the first part of the Table the
countries which are in lower segment separated according to mean passes to an upper segment when they
are separated according to median. Canada is an example, its efficiency is evaluated with less populated
countries when it is separated according to mean. If it is separated according to median; its efficiency
evaluated with United States, Japan and Germany. On the other hand the second part of the table shows
that when some decision making units passed upper segments, the remaining countries form a new frontier
among themselves.

CONCLUSIONS
It is the fact that Academic investments depends on human capital and economic support but your
competitiveness varies and depends on whom you compete. In this perspective positioning the country
become as important as academic attempts. The above results can easily be energized to other areas such
as economics. Any firm in market positions itself according to its self-scale.

REFERENCES
[1] Abbott, M., Doucouliagos, C., 2003. The efficiency of Australian universities: A data envelopment analysis.
Econ. Educ. Rev. 22, 89-97.
[2] Abramo, G., D'Angelo, C.A., 2009. Assessing technical and cost efficiency of research activities: A case study
of the Italian university system. Res. Eval. 18, 61-70.
[3] Agasisti, T., 2011. Performances and spending efficiency in higher education: A European comparison through
non-parametric approaches. Educ. Econ. 19, 199-224.
[4] Agasisti, T., Catalano, G., Landoni, P., Verganti, R., 2012. Evaluating the performance of academic departments:
An analysis of research-related output efficiency. Res. Eval. 21, 2-14.

13
[5] Agasisti, T., Dal Bianco, A., Landoni, P., Sala, A., Salerno, M., 2011. Evaluating the Efficiency of Research in
Academic Departments: An Empirical Analysis in an Italian Region. Higer Educ. Q. 65, 267-289.
[6] Agasisti, T., Pérez-Esparrells, C., 2009. Comparing efficiency in a cross-country perspective: The case of Italian
and Spanish state universities. High. Educ. 59, 85-103.
[7] Agha, S.R., Kuhail, I., Abdelnabi, N., Salem, M., Ghanim, A., 2011. Assessment of academic departments
efficiency using data envelopment analysis. J. Ind. Eng. Manage. 4, 301-325.
[8] Altamirano-Corro, A., Peniche-Vera, R., 2014. Measuring the Institutional Efficiency Using DEA and AHP: the
Case of a Mexican University. J Appl Res Technol 12, 63-71.
[9] Alwadood, Z., Noor, N.M., Kamarudin, M.F., 2011. Performance measure of academic departments using data
envelopment analysis, 2011 IEEE Symposium on Business, Engineering and Industrial Applications, ISBEIA
2011, Langkawi, pp. 395-399.
[10] Aziz, N.A.A., Janor, R.M., Mahadi, R., 2013. Comparative Departmental Efficiency Analysis within a University:
A DEA Approach. Procedia - Social and Behavioral Sciences 90, 540-548.
[11] Banker, R.D., Charnes, A., Cooper, W.W., 1984. Some models for estimating technical and scale inefficiencies
in data envelopment analysis. Management science 30, 1078-1092.
[12] Bayraktar, E., Tatoglu, E., Zaim, S., 2013. Measuring the relative efficiency of quality management practices in
Turkish public and private universities. J Oper Res Soc 64, 1810-1830.
[13] Beasley, J.E., 1990. Comparing university departments. Omega 18, 171-183.
[14] Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision making units. European
journal of operational research 2, 429-444.
[15] Hanushek, E., 1970. The Production of Education, Teacher Quality and Efficiency.
[16] Hanushek, E.A., 1979. Conceptual and empirical issues in the estimation of educational production functions.
Journal of human Resources, 351-388.
[17] Hanushek, E.A., 1986. The economics of schooling: Production and efficiency in public schools. Journal of
economic literature, 1141-1177.
[18] Hanushek, E.A., 2003. The Failure of Input‐based Schooling Policies*. The economic journal 113, F64-F98.
[19] Hanushek, E.A., 2010. Education Production Functions: Developed Countries Evidence.
[20] Hanushek, E.A., Wößmann, L., 2007. The role of education quality for economic growth. World Bank Policy
Research Working Paper.
[21] Johnson, A.L., Ruggiero, J., 2014. Nonparametric measurement of productivity and efficiency in education.
Annals of Operations Research 221, 197-210.
[22] Ruggiero, J., 1996. Efficiency of educational production: An analysis of New York school districts. The Review
of Economics and Statistics, 499-509.
[23] Ruggiero, J., 2000. Nonparametric estimation of returns to scale in the public sector with an application to the
provision of educational services. Journal of the Operational Research Society, 906-912.
[24] Ruggiero, J., 2004. Performance evaluation in education, Handbook on data envelopment analysis, Springer, pp.
323-346.
[25] Ruggiero, J., 2006. Measurement error, education production and data envelopment analysis. Economics of
Education Review 25, 327-333.
[26] SCImago. (2007). SJR - SCImago Journal & Country Rank. Retrieved December 18, 2014, from
http://www.scimagojr.com
[27] IMF (2015). World Economic Outlook Database. Retrieved November 4, 2014 from
http://www.imf.org/external/pubs/ft/weo/2015/01/weodata/index.aspx

14
AN EVALUATION OF EFFICIENCIES FOR TURKEY’S CITIES WITH DATA
ENVELOPMENT ANALYSIS (DEA) IN TERMS OF CINEMAS: A SUPER
EFFICIENCY APPLICATION
EMINE DEMET MECIT
Graduate School of Science and Engineering, Hacettepe University, 06800, Beytepe, Ankara, Turkey
(corresponding author)

İHSAN ALP
Department of Statistics, Gazi University, 06500, Teknikokullar, Ankara, Turkey

ABSTRACT
The cinema sector has received a great contribution to the economy and culture. The efficiencies of the theaters from
the cultural activities were evaluated in the literature. But, the cinemas have not been evaluated with DEA yet.
Therefore, Turkey’s cities in terms of cinemas in 2013 were evaluated with DEA in this study. Two inputs and two
outputs were determined for 72 cities (DMUs). Inputs are the number of cinema halls for the DMU under evaluation,
seating capacity for the DMU under evaluation. Outputs are the number of films shown and the number of audiences.
Finally, the efficient DMUs were ranked by super efficiency method.

Keywords: Data Envelopment Analysis, Cinema Sector, Super Efficiency Method

INTRODUCTION
As known that classical DEA methods are Charnes, Cooper, Rhodes (CCR) and Banker, Charnes, Cooper
(BCC) models (Charnes et al. 1978, Banker et al. 1984). Until now, it has been studied in various fields
with DEA. For example: health, education, institutions, military, etc. Although cinema sector is very
important for the economy, it has not been studied by the methods in the literature. So, the cinemas of
Turkey’s for 72 cities in 2013 were evaluated in the study. To this end, CCR (output oriented) and common
weight method (output oriented) were used. Then, all of cities are ranked by super efficiency method
(Andersen Petersen method (AP)) (Andersen and Petersen 1993). Finally, ranking results are compared
with Spearman Test.

METHODS
CCR model
CCR model is based on the assumption of constant returns to scale (Charnes et al. 1978). The model can
be input or output oriented model. CCR model use relative weights for inputs and outputs. CCR output
oriented model (CCR0 ) was shown in Equation (1).
The model with common weight:
The real weights for inputs and outputs can not identified by the models of CCR and BCC. Therefore,
applying common weights was developed (Roll and Golany 1993; Thompson et al. 1990; Cook and Zhu
2007).
m
min  0   vi xi 0 (1)
i 1

15
s
St. u
r 1
r yr 0  1
s m

ur 1
r yrj   vi xij  0
i 1

vi  0, ur  0, j  1,..., n ;
i  1,..., m ; r  1,..., s

Super efficiency method:


CCR model can determine only efficient units. It cannot rank the DMUs. But super efficiency method can
rank them. Super Efficiency Method (AP) is the first ranking method, which was applied to compare and
rank the efficient DMUs (Andersen and Petersen 1993). AP model is given by Equation (2).
s
max  k   u r yrk
r 1
s m
s.t.  ur yrk
r 1
 v x
i 1
i ij  0 j  1,..., n ; j  k
m (2)
 vi xik  1
i 1

ur , vi  
i  1,..., m ; r  1,..., s

APPLICATION
Cultural activities are an important part of a nation's life. The efficiencies of the theaters from the activities
were evaluated in the literature. But the cinemas have not been evaluated with DEA yet. So, Turkey’s cities
in terms of cinemas in 2013 were evaluated with DEA in the study. Two inputs and two outputs were
determined for 72 cities (DMUs). We used Turkish Statistical Institute Data. Inputs and outputs are given
below:
Inputs: 1- The number of cinema halls (HALL), 2-Seating capacity (SEAT)
Outputs: 1- The number of films shown (FILM), 2- The number of audiences (AUD)
We used two model in the study: M1 and M2.
M1: Classical CCR0 model (output oriented)
M2: The model calculated with common weight (output oriented)

RESULTS AND DISCUSSIONS


In according Table 1; İstanbul, Yalova are efficient DMUs by M1. İstanbul is efficient DMU by M2 in
Table 1.
Efficient units are ranked by AP model:

16
1. İstanbul
2. Yalova
So, İstanbul is the super-efficient city. Yalova is the second efficient city.

Table 1. Efficiencies calculated by M1θ vs M2θ


DMU M1θ M2θ DMU M1θ M2θ DMU M1θ M2θ DMU M1θ M2θ
1.İstanbul 1 1 19.Sakarya 6.90 7.42 37.Kırşehir 13.01 15.53 55.Erzurum 12.26 13.34
2.Tekirdağ 13.24 15.00 20.Düzce 52.42 79.61 38.Kayseri 14.06 18.22 56.Erzincan 54.77 57.67
3.Edirne 6.95 7.57 21.Yalova 1 1.07 39.Sivas 18.12 19.93 57.Ağrı 54.29 75.66
4.Kırklareli 14.08 18.16 22.Ankara 3.60 4.23 40.Yozgat 22.94 32.80 58.Kars 70.33 106.40
5.Balıkesir 14.90 16.43 23.Konya 11.49 13.97 41.Zongul. 8.26 9.22 59.Malatya 14.18 17.02
6.Çanakkale 13.44 14.46 24.Karaman 38.91 45.25 42.Karabük 14.24 17.29 60.Elazığ 31.42 44.21
7.İzmir 6.97 8.11 25.Antalya 6.14 6.81 43.Bartın 24.27 28.86 61.Bingöl 34.36 41.20
8.Aydın 14.87 19.67 26.Isparta 14.43 16.81 44.Kastam. 17.82 20.05 62.Tunceli 23.49 29.52
9.Denizli 11.56 15.38 27.Burdur 15.21 19.59 45.Çankırı 22.78 25.41 63.Van 65.30 95.82
10.Muğla 3.61 4.77 28.Adana 12.24 17.59 46.Samsun 10.50 11.78 64.Muş 37.09 48.27
11.Manisa 21.99 27.47 29.Mersin 19.53 22.04 47.Tokat 11.36 14.83 65.Bitlis 48.44 63.46
12.A.karahis 16.00 21.09 30.Hatay 16.87 18.50 48.Çorum 19.92 23.50 66.Gaziantep 18.05 19.12
13.Kütahya 12.61 14.35 31.K.Maraş 17.01 20.38 49.Amasya 40.20 53.17 67.Adıyaman 130.80 175.71
14.Uşak 17.50 20.84 32.Osmaniye 24.57 34.32 50.Trabzon 5.08 5.32 68.Şanlıurfa 71.83 78.07
15.Bursa 8.55 11.04 33.Kırıkkale 21.56 25.85 51.Ordu 11.04 14.54 69.Diyarbak. 16.88 19.35
16.Eskişehir 4.37 5.43 34.Aksaray 50.55 65.18 52.Giresun 31.49 47.01 70.Mardin 21.14 29.20
17.Bilecik 25.16 31.93 35.Niğde 31.49 41.65 53.Rize 20.47 26.38 71.Batman 19.08 23.67
18.Kocaeli 10.78 13.40 36.Nevşehir 6.02 7.76 54.Artvin 8.54 11.33 72.Siirt 7.73 11.90

Ranking results by R1 and R2 are seen in Table 2. In according to ranking results in Table 3, top cities are
seen.

Table 2. Ranking for DMUs (R1: Ranking for M1, R2: Ranking for M2)
DMU R1 R2 DMU R1 R2 DMU R1 R2 DMU R1 R2
1.İstanbul 1.5 1 19.Sakarya 9 10 37.Kırşehir 25 26 55.Erzurum 23 17
2.Tekirdağ 26 24 20.Düzce 66 65 38.Kayseri 28 33 56.Erzincan 68 64
3.Edirne 10 9 21.Yalova 1.5 2 39.Sivas 43 39 57.Ağrı 67 67
4.Kırklareli 29 32 22.Ankara 3 3 40.Yozgat 52 55 58.Kars 70 71
5.Balıkesir 34 27 23.Konya 20 18 41.Zongul. 13 12 59.Malatya 30 29
6.Çanakkale 27 21 24.Karaman 62 63 42.Karabük 31 30 60.Elazığ 57 59
7.İzmir 11 11 25.Antalya 8 8 43.Bartın 54 51 61.Bingöl 60 57
8.Aydın 33 38 26.Isparta 32 40 44.Kastam. 41 40 62.Tunceli 53 53
9.Denizli 21 25 27.Burdur 35 31 45.Çankırı 51 47 63.Van 69 70
10.Muğla 4 4 28.Adana 22 17 46.Samsun 16 15 64.Muş 61 62
11.Manisa 50 50 29.Mersin 45 50 47.Tokat 19 23 65.Bitlis 64 65
12.A.karahis 36 43 30.Hatay 37 34 48.Çorum 46 45 66.Gaziantep 42 35
13.Kütahya 24 20 31.K.Maraş 39 37 49.Amasya 63 63 67.Adıyaman 72 72
14.Uşak 40 42 32.Osmaniye 55 56 50.Trabzon 6 5 68.Şanlıurfa 71 68
15.Bursa 15 13 33.Kırıkkale 49 49 51.Ordu 18 22 69.Diyarbakır 38 36
16.Eskişehir 5 6 34.Aksaray 65 68 52.Giresun 58.5 61 70.Mardin 48 52
17.Bilecik 56 54 35.Niğde 58.5 57 53.Rize 47 49 71.Batman 44 46
18.Kocaeli 17 18 36.Nevşehir 7 5 54.Artvin 14 14 72.Siirt 12 16

17
Table 3. Top cities by ranking results in Table 2.
The number of ranking City
1 İstanbul
2 Yalova
3 Ankara
4 Muğla
5 Eskişehir
6 Trabzon
7 Nevşehir
8 Antalya
9 Sakarya
10 Edirne

Table 4. Spearman’s rank correlation test


Model M1 M2
M1 1 0.989
Sig. (2-tailed) 0.000 (0.000)

CONCLUSIONS
In according to Table 4, Spearman’s correlation test results are seen. M1 and M2 models have got similar
results for ranking of the DMUs’ efficiencies. In this study, the cities were ranked in the terms of cinema
efficiencies. İstanbul and Yalova are the efficient cities. İstanbul is the super-efficient city by the ranking.
In this case, it considered to be the influence of the big city of Istanbul in Turkey.

REFERENCES
[1] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[2] Banker R. D., Charnes A., Cooper W.W. (1984) Some models for estimating technical and scale inefficiencies in
data envelopment analysis, Management Science 30 (9): 1078-1092.
[3] Andersen P., Petersen N.C. (1993) A procedure for ranking efficient units in data envelopment analysis,
Management Science 39 (10): 1261-1294.
[4] Roll Y., Golany B. (1993) Alternate methods of treating factor weights in DEA, Omega 21 (1):99-109.
[5] Thompson R.G., Langemeier L.N., Lee C.T., Lee E., Thrall R.M. (1990) The role of multiplier bounds in
efficiency analysis with application to Kansas farming, J. Econometr. 46: 93-108.
[6] Cook W.D., Zhu J. (2007) Within-group common weights in DEA: an analysis of power plant efficiency, J.
Oper.Res. 178:207-216.

18
ASSESSING PERFORMANCE OF CZECH PUBLIC TRANSPORT COMPANIES
WITH TWO-STAGE DEA
MARKÉTA MATULOVÁ
Masaryk University, Faculty of Economics and Administration, Lipová 41a, Brno, Czech Republic

ABSTRACT
We analyze the performance of public transport companies in 19 Czech cities by Network Data Envelopment Analysis
using two-stage approach. Network DEA models are often used for unstorable commodities, where technical efficiency
is evaluated in the first stage (transformation of inputs into production) and service effectiveness is evaluated in the
second stage (transformation of production into consumption). The unstorable feature of transportation service is
obvious, the unoccupied seats in the vehicle's past journey cannot be sold in the future. So in order to capture the
overall performance of the transit company it is necessary to treat the commonly used performance indicators such
as vehicle kilometres or seat kilometres as an intermediate measure only. By means of two-stage model we identified
one company (Public Transport Company of Liberec) as overall effective. We also derived recommendations for other
companies using frontier projections.

Keywords: Efficiency; frontier projections; network DEA; urban public transport

INTRODUCTION
Public transportation in the Czech cities is composed of the Prague metro, seven tram operators, thirteen
trolley operators and many bus operations. Transport companies were state-owned during most of the
second half of the 20th century and that influenced urban transport in the Czech Republic. State supported
mainly bus service and cancelled some tram and trolleybus networks. Nowadays there is only one operator
for all means of public transport in each city (the city itself is usually sole or decisive shareholder of the
public transport company in larger cities). The operators are subsidized by the cities, revenues from fares
cover around quarter to half of the costs. In many cities they are still looking for optimal plans for the
reorganization of public transport with focus on the regional integration of transport and tariff and departure
control systems.
The need to measure transit system performance along with its different dimensions has led to the
development of a large number of quantitative performance indicators. Various researchers have defined
alternative goals and developed quantitative measures to evaluate the extent to which transit systems
achieved these goals. Unfortunately the individual performance measures yield widely inconsistent results,
so it is agreed that a small number of indicators (or a single indicator) is needed to describe transit system
performance. This can be achieved by using data envelopment analysis, see Karalaftis (2004). Network
DEA models are often used for unstorable commodities, transportation service being one of them, because
one can never store the surplus service during periods of low demand (off-peak hours) for use during periods
of high demand (peak hours). Chiou et al. (2009) proposed DEA model based on the concept firstly
introduced by Fielding (1987) to measure performance of a public transit system by defining technical
efficiency as the ratio of production to factors and service effectiveness as the ratio of consumption to
production as depicted in Fig. 1.

19
Figure 1: Performance measurements for transport service (Chiou, Lan, Yen 2009)

Our objective was to find a model that would be able to evaluate performance of Czech public transport
companies according to the scheme of Fig. 1 and to derive a suitable method for projecting the units on
efficient frontier.

METHODS
As in the standard notation of DEA models, there are considered n Decision Making Units and for j-the
DMU (j = 1, … , n) the vector of inputs (outputs) is denoted by X j ( Yj ) with corresponding weight vectors
u, v. In the network model, we also use symbols Zj , w for the intermediate measures and weight vector
assigned to them. In the DEA literature there are mentioned many different ways of computing the overall
efficiency score and the scores for the individual stages in network DEA. In our study we followed the
composition approach suggested by Despotis and Koromikos (2014). Its advantage is that it gives bias-free
and unique efficiency scores for the two stages (as opposed to decomposition models of Kao and Hwang
(2008) and Chen et al. (2010)). The model can be stated as follows:

min(𝑣𝑋𝑗0 − 𝑢𝑌𝑗0 )

s. t. 𝑤𝑍𝑗0 = 1,

𝑤𝑍𝑗 − 𝑣𝑋𝑗 ≤ 0, 𝑗 = 1, … , 𝑛, (1)

𝑢𝑌𝑗 − 𝑤𝑍𝑗 ≤ 0, 𝑗 = 1, … , 𝑛,

𝑣 ≥ 0, 𝑤 ≥ 0, 𝑢 ≥ 0.

The efficiency scores of the unit j0 in both stages are computed ase1j0 = 1/(v ∗ X j0 ), e2j0 =
u∗ Yj0 for optimal weights v ∗ , u∗ . Unlike the ordinary DEA model it is not so straightforward to derive

20
frontier projections under the network structure. Some authors (e. g. Chen et al. 2013) propose projecting
methods that require either input or output orientation. Unfortunately such projections often lead to
unrealistic target values for the inputs and outputs of the decision making units, so they cannot be used as
a basis for recommendation in practice. More suitable projecting method of Despotis et al. (2014) is based
on the solution of the optimization problem:

𝑚𝑎𝑥(𝑠 − − 𝑠 + ), s. t.

𝑋𝜆 + 𝑠 − = 𝑋j0

𝑌𝜇 − 𝑠 + = 𝑌j0
(2)
𝑍𝜆 ≥ 𝑍̃j0

𝑍𝜇 ≤ 𝑍̃j0

𝜆 ≥ 0, 𝜇 ≥ 0, 𝑠 + ≥ 0, 𝑠 − ≥ 0, 𝑍̃j0 free in sign.

The DMUs included in our study are the members of Association of Transport Companies (SDP ČR), which
associates all operators of trolley and tram systems in the Czech Republic (member cities are depicted in
the Figure 1). SDP ČR provides a good platform for sharing information and best practices in the field of
urban public transport. It guarantees unified methodology of data collection. However, there are many
regional differences between the companies that must be taken into account, such as different size of the
serviced area and population, different types of vehicles in the rolling stock, etc.

Figure 2: Public transport companies associated in the SDP ČR

21
Two inputs (number of employees and costs), two intermediate measures (vehicle kilometres and seat
kilometres) and two outputs (number of passengers and sales) were used to compute the efficiency scores.
The data are given in the Table 1.

Table 1: Data (source: annual report of SDP ČR, 2014)


X Z Y
Employees Costs Vehicle-km Seat-km Passengers Sales
(103 CZK) (103) (103) (103) (103CZK)
Brno 2716 2853 37125 4066832 353940 975011
ČeskéBudějovice 398 404 5651 564776 38541 123462
Děčín 183 141 3691 265752 8248 48643
HradecKrálové 385 300 5950 475524 34106 118345
Chomutov-Jirkov 239 110 1825 172550 5102 47432
Jihlava 175 122 3032 212255 13777 47867
KarlovyVary 258 131 2584 225242 9587 58868
Liberec 390 518 7755 680462 42045 195909
MariánskéLázně 31 28 486 34032 3705 10343
Most-Litvínov 446 321 4512 419405 27420 103496
Olomouc 438 364 6196 651153 52193 146783
Opava 184 132 3018 248749 10397 47917
Ostrava 1923 1585 31820 3301825 91000 511306
Pardubice 404 337 5730 557912 25919 117220
Plzeň 821 1308 15077 1437851 101115 292247
Praha 10667 17745 158321 20783067 1329745 4654617
Teplice 263 231 5658 299597 14843 95164
ÚstínadLabem 484 562 7266 733999 43162 186476
Zlín-Otrokovice 331 247 4824 451512 31866 117620

Table 2: Two-stage DEA model results (2014)


𝒆𝟏 𝒆𝟐
Brno 0.804 1
ČeskéBudějovice 0.812 0.828
Děčín 1 0.6
HradecKrálové 0.815 0.847
Chomutov-Jirkov 0.771 0.974
Jihlava 0.931 0.738
KarlovyVary 0.863 0.906
Liberec 1 1
MariánskéLázně 0.754 1
Most-Litvínov 0.645 0.874
Olomouc 0.861 0.884
Opava 0.961 0.658
Ostrava 1 0.576
Pardubice 0.823 0.764
Plzeň 0.962 0.786
Praha 0.922 0.976
Teplice 0.902 0.861
Ústí nad Labem 0.843 0.941
Zlín-Otrokovice 0.9 0.925

22
RESULTS AND DISCUSSION
The Table 2 contains DEA scores of technical efficiency and service effectiveness (computed from the
model (1) as e1j0 = 1/(v ∗ Xj0 ), e2j0 = u∗ Yj0 for optimal weights v ∗ , u∗) for the 19 Czech public transport
companies. Three companies were identified as technically efficient (Děčín, Liberec, Ostrava) and three
companies achieved full score in service effectiveness (Brno, Liberec and Mariánské Lázně). Public
transport company of Liberec is overall effective.

In the Table 3 there are projections resulting from the model (2), given by 𝑋̃j0 = 𝑋j0 −𝑠 − , 𝑍̃j0 and 𝑌̃j0 =
𝑌j0 + 𝑠 + .

Table 3: Frontier projections, the model (2)


̃
𝑿 ̃
𝒁 ̃
𝒀
Employees Costs Vehicle-km Seat-km Passengers Sales
(103 CZK) (103) (103) (103) (103CZK)
Brno 2716 2093 54780 3944166 422844 1179462
ČeskéBudějovice 398 307 8027 577974 61963 172837
Děčín 183 141 3691 265752 28491 79470
HradecKrálové 385 297 7765 559096 59939 167192
Chomutov-Jirkov 143 110 2880 207324 22227 61998
Jihlava 158 122 3194 229941 24651 68762
KarlovyVary 170 131 3429 246904 26470 73834
Liberec 390 518 7755 680462 42045 195909
MariánskéLázně 31 24 625 45018 4826 13462
Most-Litvínov 417 321 8403 605010 64862 180922
Olomouc 438 337 8834 636062 68191 190208
Opava 171 132 3455 248789 26672 74398
Ostrava 1923 1482 38786 2792574 299385 835091
Pardubice 404 311 8148 586687 62897 175443
Plzeň 821 633 16559 1192254 127818 356531
Praha 10667 8592 214956 15686496 1669368 4654617
Teplice 263 203 5305 381928 40946 114212
ÚstínadLabem 484 373 9762 702863 75352 210184
Zlín-Otrokovice 321 247 6466 465537 49909 139214

CONCLUSIONS
The network DEA model was used to evaluate technical efficiency and service effectiveness of the 19
Czech public transport operators. Only Public transport company of Liberec was identified as overall
effective. Other companies could reach the frontier by adjusting their inputs, outputs and intermediate
measures as can be seen in the Table 3. As opposed to inputs (which are desirable to reduce) and outputs
(which are desirable to increase) the intermediate measures can be changed in both directions. The company
would achieve an increase of vehicle kilometres and reduction of seat kilometres at the same time e. g. by
using shorter buses (this is the case of Brno, but for example Teplice should use longer vehicles on the
contrary). However one must interpret the recommendations cautiously regarding the aforementioned
regional differences (otherwise some adjustments would have to be too large). In the future work we would
like to improve the projecting method, e. g. by imposing some weights on the slacks in the objective

23
function of (2) in order to put emphasis on the reduction of inputs (which are more controllable by the
company management than outputs).

ACKNOWLEDGEMENT
The paper was supported by the Mobility program for academic staff on FEA MU.

REFERENCES
[1] Annual report of the Association of the Transport Companies of the Czech Republic 2014. Available:
http://www.sdp-cr.cz/o-nas/vyrocni-zpravy/ [July 2015]
[2] Chen Y., Cook W.D., Zhu J. (2010) Deriving the DEA frontier for two-stage processes, European Journal of
Operational Research (202): 138-142
[3] Chen Y., Cook W.D., Kao C., Zhu J. (2013) Network DEA pitfalls: Divisional efficiency and frontier projection
under general network structures. Eur J Oper Res (226): 507-515.
[4] Chiou Y., Lan L., Yen B. (2010) Decision Support: A joint measurement of efficiency and effectiveness for non-
storable commodities: Integrated data envelopment analysis approaches, European Journal of Operational
Research (201): 477-489
[5] Despotis D. K., Koronakos G. (2014) Efficiency Assessment in Two-stage Processes: A Novel Network DEA
Approach, Procedia Computer Science, (31): 299-307
[6] Despotis D. K., Koronakos G., Sotiros D. (2014) Composition versus decomposition in two-stage network DEA:
a reverse approach‚ Journal of Productivity Analysis: 1–17
[7] Karlaftis M.G. (2004) A DEA approach for evaluating the efficiency and effectiveness of urban transit systems,
European Journal of Operational Research (152): 354-364
[8] Kao, C., Hwang, S. (2008) O.R. Applications: Efficiency decomposition in two-stage data envelopment analysis:
An application to non-life insurance companies in Taiwan, European Journal of Operational Research, ( 185):
418-429.

24
CONSTRUCTION OF UNIVERSITIES' TYPOLOGY VIA DEA
IRINA ABANKINA
Institute for Educational Studies (IES), NRU HSE

FUAD ALESKEROV
Department of Mathematics, Faculty of Economic Sciences, and, International Laboratory of Decision Choice and
Analysis (ILDCA), NRU HSE

VERONIKA BELOUSOVA
Department for Methodology of Budget Planning, Institute for Statistical Studies and Economics of Knowledge
(ISSEK), NRU HSE

LEONID GOKHBERG
Institute for Statistical Studies and Economics of Knowledge (ISSEK), NRU HSE

KIRILL ZINKOVSKY
Institute for Educational Studies (IES), NRU HSE

SOFYA KISELGOF
Department of Mathematics, Faculty of Economic Sciences, NRU HSE

VSEVOLOD PETRUSHCHENKO
International Laboratory of Decision Choice and Analysis (ILDCA), NRU HSE

SERGEY SHVYDUN
International Laboratory of Decision Choice and Analysis (ILDCA), NRU HSE

ABSTRACT
In recent decades, increased economic pressure and growing societal expectations have led to the introduction of
performance-based funding models for universities. In this respect, a great scholarly attention has paid to how to
evaluate correctly universities performance. The widely spread method used for that purpose is DEA. This paper
provides a review of different approaches how to take into account universities heterogeneity when applying DEA to
construct the typologies of university by showing statistically their similarities and differences. The authors apply the
modified DEA proposed by Aleskerov & Petrushchenko (2013) to evaluate performance scores of Russian technical
universities. This typology divides universities into specific groups with a description taking into account their
heterogeneity.

Keywords: higher education institutions (HEIs), research and education activities of HEIs, data envelopment analysis,
efficiency, performance.

INTRODUCTION
To take into account heterogeneity of universities, one often divides the sample into homogeneous clusters
and computes efficiency within these clusters. In (Johnes, 2006), likewise in (Charnes et al., 1981), authors
tried to catch the difference between three main categories of British universities. The first one consisted
of 47 HEIs which obtained the status of university before Further and Higher Education Act came into force
in 1992. The second category contained 34 HEIs which had the status of university after 1992, the last
category had 28 colleges specializing in arts. Later, in (Johnes & Johnes, 2009), the sample was divided
into five clusters depending not only on the foundation date but also on the prestige of universities. In
(Warning, 2004), German universities were distinguished in dependence with direction of activity (research

25
or education) and specialization of alumni (social or natural sciences). In turn, Russian universities typology
in (Abankina et al., 2013) defined 79 technical and 59 classic universities. The authors showed that
efficiency scores for these two groups were comparable. Abbott & Doucouliagos (2003) also used the idea
of dividing the sample into parts. In this case, the criterion was the ratio of received grants to the number
of students. According to this model, most of universities obtained high efficiency scores.
Contrarily, Athanassopoulos & Shale (1997) implemented the stage-by-stage evaluation of universities cost
efficiency taking into account expended resources. The underlying idea was first presented in (Banker &
Morey, 1986). Analysis of the data revealed that universities, had chosen research activity as a main priority,
had higher costs than those that focused on the training of specialists in applied areas. As a result, three
groups of universities were allocated: 1) universities with research activity as a main development strategy,
2) universities with balanced strategy, 3) universities with focus on applied areas. The final efficiency score
of university in the group 𝑘 = 1,2,3 was obtained from the evaluation of sub-sample consisting of
universities contained in groups with ordering number less than or equal to k.
Another approach [Pastor et al., 2002] to test the consistency of efficiency scores in homogeneous clusters
of universities is to check the impact of individual factors on the final score for both complete and truncated
set of input-output parameters. This method also works with the inclusion of value judgments as in
(Beasley, 1995). As a result, all groups of universities were found to show high productivity (around 90%).
In recent years, special modifications of DEA for universities were developed. For example, in (Abbott &
Doucouliagos, 2003), the authors built a typology of universities taking into account the geographical
location based on the cluster method (K-averages, the nearest-neighbor method). However, they found that
the assessment of HEIs efficiency scores within the clusters produced results similar to the procedure of
splitting universities into groups based on the ratio of the grants received to the number of students.
Another example is the paper (Aleskerov & Petrushchenko, 2013) in which authors used the method of
sequential exclusion of alternatives to obtain efficiency scores of 28 Russian universities. Moreover, they
found out that the rankings produced by this modification and standard DEA were quite similar. Multi-
activity DEA model, developed in (Beasley, 1995), presented another approach to solve the heterogeneity
problem of the sample. The model took into account that a university had both education and research
development strategies. The author evaluated 52 English departments of physics and chemistry. In turn,
Cinar (2013) evaluated 45 Turkish universities by applying these two models. Following (Beasley, 1995),
he suggested, that universities had just two goals, i.e. teaching and running research. Using the model of
sequential exclusion of alternatives, the author concluded that the data set was quite heterogeneous.
The aim of this paper is to provide a comparative study of standard DEA and modified one via the
heterogeneity of universities when bearing in mind similarities and differences of Russian universities.

METHODS
To illustrate the method of sequential exclusion of alternatives, let us consider the case with single input-
output parameters (generalization to arbitrary number of inputs and outputs is given in (Aleskerov &
Petrushchenko, 2013)).

26
The graphic interpretation of the standard DEA is given on Fig. 1.

Fig. 1. Interpretation of DEA in case of single input-output economy


𝑂𝐵𝑥𝐶 𝑂𝐵𝑥𝑉
The efficiency scores of the firm B according to CRS and VRS versions are 𝑂𝐵𝑥
and 𝑂𝐵𝑥
, respectively. It
can happen that some considerable part of the sample is located far from efficiency frontier, this means that
the sample is heterogeneous. The general idea of the model is to move the frontier towards the barycenter
of the sample. This allows us to reduce the influence of heterogeneity on efficiency scores.
On the figure below, the initial sample consists of the firms 𝐹1 , … , 𝐹6 and according to standard CRS model,
𝐹1 is the efficient firm. According to the algorithm, we calculate the barycenter (point B) and construct the
new frontier via generating phantom firm G, lying on the segment 𝐵𝐹1 . Clearly, 𝐹3 , … , 𝐹6 should be
benchmarked against the firm G. Still, 𝐹1 and 𝐹2 remain unevaluated, to compute their efficiency scores,
we should repeat the same algorithm excluding the firms 𝐹3 , … , 𝐹6 from consideration. The algorithm stops
when all universities are evaluated.

Fig. 2. The algorithm of sequential exclusion of alternatives in case of single input-output.

RESULTS AND DISCUSSIONS


For our typology, the data was gathered from the standardized written surveys, conducted by HSE’s
Institute for Statistical Studies and Economics of Knowledge, in the framework of the Economics of

27
Science Monitoring Project. It covered 400 state universities in Russia in the terms of research and
development, innovation and cooperation with companies
Following Bonaccorsi & Daraio (2009) and Abankina et al. (2013a, 2015), we incorporate data on discipline
differentiation when providing a full-scale classification of HEIs. In our sample, after excluding 8 outliers
via using indicators of research and education (for more details1 see Abankina et al. (2013 b, 2015)), there
are 70 technical universities. Then, we conduct a comparative study of their educational (Model 1) and
academic (Model 2) potential. The specifications of both models are in line with Abankina et al. (2013a).
In Model 1, the educational potential of Russian HEIs is tested.
The following input parameters are used: unit public funding per student provided in the relevant year
(UPF), share of faculty members with degrees (PDS), total number of faculty members (PPS). Output
parameters included the first year student average Unified State Exam grade (USE) and total number of
students (STUD). Model 2 measures the academic potential of HEIs. The input parameters are similar to
Model 1 with the addition of the first year students’ average USE. The last parameter is used as an indicator
of the academic potential of entrants, who can participate in university research projects. Output parameters
include the total number of students (as in Model 1) and the share of young faculty members (YPPS) and
publication activity of faculty members (PUB). Descriptive statistics are given in Table 1.

Table 1. Descriptive statistics of HEIs input and output parameters


Indicator Min value Max value Mean Standard deviation
YPPS 0.02 0.35 0.17 0.07
PUB 2.47 2856.60 135.45 353.36
STUD 1360 19724 8061.78 4416.29
UPF 57.71 252.46 107.99 43.99
PDS 18.21 80.35 63.55 11.04
PPS 39 2914 632.31 519.31
USE 50.10 90.4 60.95 6.32

Our results for standard DEA are provided in Table 2. In addition, all clusters of universities are separated
into three subgroups with low (less than 0.6), average (from 0.6 but less than 0.8) and high (from 0.8 to 1)
efficiency scores. Such scale for grouping HEI’s by their scores divides all the analysed HEIs to near equal
size subgroups and used just to simplify the picture. For example, many well-known Russian technical
HEIs are expected to have high efficiency scores, but they are classified on average according to their
efficiency scores. The subgroup with high efficiency scores includes several strong regional HEIs. These
HEIs fully use their potential. While having less resources and public funding than some well-known HEIs,
they obtain better results.

1
In [Abankina et al., 2013b, 2015], hierarchical algorithms depending on the potential of HEIs in the field of research
and development, innovation were applied to the sample of 219 universities. This helped to divide this sample into six
clusters or 5 economically valid categories: niche Universities (two combined clusters), universities with uncertain
position, potential and actual research and education leaders, market leaders, universities of good standing.

28
Table 2. Descriptive statistics for homogeneous groups of HEIs
Model 1 Model 2
Average score (%) 71.43 66.91
Standard deviation (%) 16.29 21.05

Furthermore, we found out that publication activity matters in distinguishing the technical universities. It
changes the distribution of efficiency measurements: if it is disregarded, the result is a histogram rising with
productivity growth and with a certain reduction in the number of universities at the marginal value of
performance. However, if it is taken into account, universities are clearly divided into groups with different
performance values.
Finally, the highest efficiency scores are demonstrated by universities belonging to the "market leaders"
cluster. At the same time, the most part of universities is characterized by limited capacity for academic
activity, this has an impact on efficiency scores. The detailed analysis reveals that, for the practical
implementation of DEA in the Russian context, one needs a model, which helps to set the thresholds of
output variables that characterize the level of publication activity, because this parameter serves as a main
cause of inefficiency.
For a robustness check, we take into account the heterogeneity of the HEIs as Aleskerov and Petrushchenko
(2013) did. Three different values of 𝜇 (0.2, 0.5 and 0.8) are introduced. The rankings of universities
obtained by this method when 𝜇 equals 0.2 or 0.5 are in most cases comparable and statistically consistent
with the DEA obtained above (for more details, see Table 3 and Table 4).

Table 3. Pearson correlation matrix


Model 1 Model 2
DEA m = 0.2 m= 0.5 m=0.8 DEA m = 0.2 m= 0.5 m=0.8
DEA 1 0.9453 0.7199 0.5569* DEA 1 0.9857 0.9004 0.8224*
m = 0.2 1 0.8323 0.6127 m = 0.2 1 0.9462 0.8572
m= 0.5 1 0.6096 m= 0.5 1 0.9169
m=0.8 1 m=0.8 1
Note: * not statistically significant on 10% level of confidence,
** not statistically significant on 5% level of confidence

Table 4. Spearman correlation matrix


Model Model 2
DEA m = 0.2 m= 0.5 m=0.8 DEA m = 0.2 m= 0.5 m=0.8
DEA 1 0.9538 0.6920 0.5145 DEA 1 0.9943 0.9278 0.8230
m = 0.2 1 0.7488 0.5420 m = 0.2 1 0.9392 0.8155
m= 0.5 1 0.4550* m= 0.5 1 0.7896*
m=0.8 1 m=0.8 1
Note: * not statistically significant on 10% level of confidence,
** not statistically significant on 5% level of confidence

29
CONCLUSION
Thus, we test performance models with different input and output parameters for Russian technical
universities and get comparable results for standard DEA and its modification known as the sequential
exclusion of alternatives method. In addition, we show that bearing in mind the heterogeneity of universities
plays a crucial role for policy purposes in setting threshold values for efficiency scores. Moreover, we find
that that the publication activity rate used as an output factor affects significantly the measurement of
university performance. Furthermore, we tend to support the evidence that well-known Russian technical
HEIs do not fully use their potential and resources. These results mean that all three issues are worth
considering when performance-based funding schemes are set up and policy initiatives are targeted at a
relevant group of universities.

ACKNOWLEDGEMENT
The financial support from the Government of the Russian Federation within the framework of the Basic
Research Program in 2014 at the National Research University Higher School of Economics and within the
framework of the 5-100 Programme Roadmap of the National Research University Higher School of
Economics is acknowledged.

REFERENCES
[1] Abankina I., Aleskerov F., Belousova V., Bonch-Osmolovskaya A., Petruschenko V., Ogorodniychuk D., Yakuba
V., Zin'kovsky K. (2012) University efficiency evaluation with using its reputational component / Proceedings of
the 4th International Conference on Applied Operational Research. Bangkok, P. 244–253
[2] Abankina I., Aleskerov F., Belousova V., Gokhberg L., Zinkovsky K., Kiselgof S., Shvydun S. (2013b)
Tipologiya i analiz nauchno-obrazovatel'noy rezul'tativnosti rossiyskikh vuzov [A Typology and Analysis of
Russian Universities’ Performance in Education and Research]. Foresight-Russia, vol. 7, no 3, pp. 48-63 (in
Russian)
[3] Abankina I., Aleskerov F., Belousova V., Zinkovsky K., Petrushchenko S. (2013a) Otsenka rezul'tativnosti
universitetov s pomoshch'yu obolochechnogo analiza dannykh [Evaluating Performance of Universities Using
Data Envelopment Analysis]. Voprosy obrazovaniya, no2, pp. 15-48 (in Russian)
[4] Abankina I.V., Aleskerov F.T., Belousova V.Y., Gokhberg L.M., Zinkovsky K.V., Kiselgof S.G., Petrushchenko
V.V., Shvydun S.V. Performance-Based Typology Of Universities: Evidence From Russia // Working papers by
NRU Higher School of Economics. Series WP BRP "Science, Technology and Innovation". – 2015. – №WP
BRP 33/STI/2015
[5] Abbott M., Doucouliagos C. The efficiency of Australian universities: a data envelopment analysis // Economics
of Education Review. – 2003. -- №22, P. 89 – 97.
[6] Aleskerov F.T., Petrushchenko V.V. DEA by sequential exclusion of alternatives // WP7/2013/02. Moscow:
Publishing House of the Higher School of Economics, 2013.
[7] Athanassopoulos A. D., Shale E. Assessing the Comparative Efficiency of Higher Education Institutions in the
UK by the Means of Data Envelopment Analysis // Education Economics. –1997. -- №5, P. 117 – 134.
[8] Banker R.D., Morey R.C. Use of Categorical Variables in Data Envelopment Analysis // Management Science. –
1986, 32: 1613-1627.
[9] Beasley J. Determining teaching and research efficiencies // Journal of the Operational Research Society. – 1995.
– №46, P. 441 – 452.
[10] Bonaccorsi A., Daraio C. (2009) Characterizing the European university system: a preliminary classification
using census microdata, Science and Public Policy №36 (10), pp. 763-775

30
[11] Charnes A., Cooper W., Rhodes E. Evaluating Program and Managerial Efficiency: An Application of Data
Envelopment Analysis to Program Follow Through // Management Science. – 1981. -- №27, P. 668 – 697
[12] Cinar Y., Research and Teaching Efficiencies of Turkish Universities with Heterogeneity Considerations:
Application of «Multi-Activity DEA» and «DEA by Sequential Exclusion of Alternatives' Models». – 2013, HSE
WP7, 04.
[13] Johnes G., Johnes J. Higher education institutions' costs and efficiency: taking the decomposition a further step //
Economics of Education Review. – 2009, -- №28. P. 107 – 113.
[14] Johnes J. Data envelopment analysis and its application to the measurement of efficiency in higher education //
Economics of Education Review. – 2006. -- №25, P. 273 – 288
[15] Pastor T., Ruiz L., Sirvent I. A statistical test for nested radial DEA models // Operations Research. – 2002. –
№50, P. 728 – 735.
[16] Warning S. Performance Differences in German Higher Education: Empirical Analysis of Strategic Groups,
Review of Industrial Organization. – 2004. - №24, P. 393 – 408.

31
32
DYNAMIC EFFICIENCY EVALUATION OF GERMAN PUBLIC
MULTIDISCIPLINARY THEATRES BY RC-DEA MODEL
ANDREAS KLEINE
FernUniversität Hagen, Department of Business Administration and Economics, Chair of Quantitative Methods,
Hagen, Germany

STEFFEN HOFFMANN
FernUniversität Hagen, Department of Business Administration and Economics, Chair of Quantitative Methods,
Hagen, Germany

ABSTRACT
Due to the strategic objectives, multidisciplinary theatres offer drama as well as music and dance events. If we
measure efficiency of these theatres by DEA, it is inappropriate to neglect a discipline when efficiency bases on it.
Hence, the efficiency structure has to be observed. Moreover, some of these theatres operate under specific conditions
which require a categorization. As decision makers are interested in efficiency changes over time we introduce a so
called dynamic RC-DEA model that incorporates all these requirements. The RC-DEA is based on an application-
oriented implementation of weight restrictions in the multiplier model. A hierarchically categorical DEA is embedded
in combination with a Window Analysis. In contrast to this well-known dynamic approach we recommend a modified
measure for a window. The use of the RC-DEA model is illustrated by an empirical example of 31 German public
multidisciplinary theatres.

Keywords: Efficiency structure; Weight restrictions; Categorical DEA; Window Analysis; Multidisciplinary theatres

INTRODUCTION
Within the season 2011/2012 145 German public theatres counted 21 million visitors in about 75,000
performances and over 5,400 productions. In the accounting year of 2011 the governmental grant for public
theatres amounts to 2.25 billion Euro paid at 99 percent by communes and federal states (Theaterstatistik,
2013). In times of deficit spending and excessive national debt, it is important to allocate resources in an
efficient and effective manner.
The structure of the German public theatre landscape is to some extend unique in the world. Most of these
theatres are characterized by multidisciplinarity. This means that many theatres have more than one
division, and often they are separated into three parts: drama, dance and music.
In order to support the decision making process of the respective political authorities, the above mentioned
uniqueness should be reflected in any used performance evaluation method. Therefore, we apply and adjust
one of the classical Data Envelopment Analysis (DEA) programs to achieve this goal.
The contribution is organized as follows: The next section develops the basics of DEA step by step with
respect to German public theatres. In the following section we introduce the new problem which we call
dynamic RC-DEA. Hereafter we present and discuss the results of our application. Finally, the last section
provides a brief summary of our findings.

METHODS
Literature review

33
The “cost desease” of performing arts has already been discussed in the nineteen-sixties by Baumol and
Bowen (1966). In addition to this general discussion, some papers analyse the efficiencies of theatres
empirically. A comprehensive application of DEA to German theatres was made by Tobias (2003). Marco-
Serrano (2006) has also used DEA but in another context – namely Spanish theatres. Whereas Last and
Wetzel (2010) applied a Stochastic Frontier Analysis to German theatres. However, none of these works is
sufficient for incorporating disciplines into the production process of multidisciplinary theatres: RC-DEA
aims to close this gap. Regarding the differentiation of Ossadnik (1987), we evaluate by this model the
quantitative efficiency – which the author called secondary efficiency – instead of measuring the qualitative
efficiency – named primary efficiency.
Inputs and outputs
For the set of inputs we distinguish between a financial and a structural configuration. The structural input-
configuration consists of staff (full-time equivalents) and number of seats, representing labour and capital.
By contrast, the latter configuration considers staff costs and material costs. The set of outputs stays the
same for both configurations and comprises the number of performances in music (PMU), dance (PDA)
and drama (PDR) as well as other performances (POT), the number of new productions within a season
(PRD) and the number of visitors (VIS) of that season. On the whole, 31 German multidisciplinary theatres
are examined from two perspectives. These theatres have been extracted by an expert consultation. The
inputs and outputs are available for 7 seasons from 2004/05 to 2010/11.
Categorical DMUs
As we mentioned before, the 31 theatres have one important feature in common: their multidisciplinarity.
But despite this fact, some theatres may have a structural advantage over their competitors. In our case it is
a high level of guest performances related to the total amount of all performances. In order to represent this
diversity in the subsequently proposed model, we split the group of theatres into two categories: theatres
with a percentage of guest performances greater than 15 % belong to the advantaged category 1, the
remaining theatres belong to the disadvantaged category 2. Regarding the concept of Categorical DMUs by
Tone (1997), we compare members of category 1 with those of category 1 and 2 but theatres of category 2
only with members of this category.
Weight restrictions
Multidisciplinarity as a strategic objective requires a minimum effort in all services offered by the
respective theatres. Thus it is not appropriate when efficiency is driven by only some, but not all, disciplines.
According to the global assurance region model of Wong and Beasley (1990) a lower bound is introduced:
this bound leads to an extended model which incorporates the weighted disciplines at a minimum
percentage. For our purpose we attain the threshold value by taking the minimum relative activity level
over all theatres and disciplines, and, in doing so, we obtain the minimum level  of 8.6 %.

Efficiency structure
Our weight restriction approach relies on an interpretation of the multiplier model which we name the
efficiency structure of an observed theatre. We know that for the input oriented model all weighted inputs

34
sum up to 1, whereas the weighted outputs have to be divided by the efficiency score *o of the observed
theatre o to sum up to 1. As highlighted above, every discipline should have impact on the overall
efficiency at a given percentage. Therefore, let  ro
*
be the optimal weight for the r -th output yro among
s outputs of the considered theatre o , then we call

1
 ro*   yro  ro* (1)
o
*

the efficiency impact of the r -th output.


Multi-period analysis
The multi-period analysis is done by applying a Window Analysis. By shifting the time-window with p
seasons along the data panel, we get a set of efficiencies for each window q as illustrated in Table 1 for
the theatre of Coburg (CO) with p  5 and q  1, 2,3 .

Table 1: Results of Window Analysis


Season
theatre window q 04/05 05/06 06/07 07/08 08/09 09/10 10/11
CO 1 1.0000 1.0000 1.0000 1.0000 1.0000
CO 2 1.0000 1.0000 1.0000 1.0000 0.9926
CO 3 1.0000 1.0000 1.0000 1.0000 0.9652
… … … … … … … …
bold values mark the defined window efficiency

The overall efficiency per window of a theatre is now defined as the last efficiency value of the
corresponding window; this approach is highlighted in Table 1. The idea bases on the following
comprehension: First, the last efficiency value is the most recent one of the window and has not been
included in an evaluation before. Second, an aggregation of the efficiency values as usually done mixes
window technologies.

THE RC-DEA MODEL


In the restricted categorical model – the RC-DEA model – yrjt (r  1, , s) represents the r -th output in
period t and r (r  1, , s) the corresponding multiplier weight, analogously for the i-th input
xijt (i  1, , m) and the respective multiplier  i (i  1, , m) ; w is the total number of evaluated time
windows. If we plug all the previously described modifications in a classical CCR-model
(Charnes/Cooper/Rhodes 1978), the following RC-DEA model has to be solved for each theatre
o  1, , n and window q  1, , w :

s
max (oq  p 1)*   r yroq  p 1 (2)
 ,
r 1

35
m
s.t.  x
i 1
q  p 1
i io 1 (3)

s m

 r yrtj   i xijt  0  j  No , t  q,
r 1 i 1
, q  p 1 (4)

  rˆ yroqˆ  p 1  r yroq  p 1  0 r  S  {PDA, PMU, PDR} (5)


rˆS

𝜈𝑖 , 𝜇𝑟 ≥ 𝜖 > 0 (6)

The index set N o in constraint (4) consists of all theatres which the observed theatre o has to be compared
with (see Categorical DMUs). Constraint (5) incorporates the weight restriction as mentioned in the
previous section, thus the set S is introduced containing the three disciplines dance, music and drama.
When rearranging (5) we get

(7)

∀𝑟 ∈ 𝑆 = {𝑃𝐷𝐴, 𝑃𝑀𝑈, 𝑃𝐷𝑅}


The relative importance of each discipline in S is obviously driven by (7).

EMPIRICAL RESULTS
The difference between a standard CCR- and the extended RC-DEA model is illustrated by the empirical
example with 31 theatres, 2 categories and 7 seasons. We choose a window width of 5 periods, thus the
efficiency values of 3 windows are calculated. Table 2 merely indicates the efficiency results for the
structural input-configuration. Each theatre is labelled by its German license number tag.

Table 2: Comparative results


season 2008/2009 2009/2010 2010/2011
theatre category CCR RC-DEA Δ CCR RC-DEA Δ CCR RC-DEA Δ
BI 2 0.8375 0.8346 - 0.9056 0.8998 - 0.9504 0,9382 -
C 2 0.7575 0.7542 - 0.8529 0.8017 - 0.8249 0,8038 -
CB 2 0.5887 0.6314 + 0.7878 0.7878 = 1.0000 1,0000 =
CO 2 1.0000 1.0000 = 0.9926 0.9894 - 0.9652 0,9684 +
… … … … … … … … … … …
S 2 0.7154 0.7117 - 0.5771 0.5771 = 0.4929 0,4906 -
SB 2 0.7316 0.7316 = 0.7572 0.7600 + 0.7354 0,7496 +
SN 2 1.0000 1.0000 = 0.9477 0.9458 - 1.0000 1,0000 =
WI 2 0.9731 0.9726 - 0.9897 0.9980 + 0.9247 0,9260 +
WU 2 0.8578 0.8776 + 0.9275 0.9958 + 0.9614 1,0000 +
Mean average 0,8614 0.8662 + 0.8631 0.8649 + 0.8716 0.8715 -
CRS efficient 7 8 5 5 8 9

36
The efficiencies of RC-DEA for category 1 theatres are always less than or equal to those of the CCR-
model due to the additional weight restrictions. It is worthwhile to note that this is not the case for theatres
of category 2 as they are not compared to category 1 theatres. In consequence we have two effects for these
theatres, they gain from the reduced number of DMUs but suffer from weight restrictions. The
predominating effect on the efficiency score depends on the data.
Table 3 depicts the comparative efficiency structure for the theatre WU as well as its peer groups. In the
third line you find the window efficiency scores followed by the efficiency structure for the outputs.

Table 3: Comparative Efficiency structure of theatre WU


2008/2009 2009/2010 2010/2011
CCR RC-DEA CCR RC-DEA CCR RC-DEA
WU
*
0,8578
0,8776 0,9275 0,9958 0,9614 1,0000
PDA 20,66%
39,67% 51,35% 66,03% 69,45% 75,26%
PDR 0,00%
4,24% 19,66% 6,86% 30,55% 7,82%
PMU 0,00%
5,41% 10,96% 6,86% 0,00% 7,82%
POT 0,00%
0,00% 0,00% 0,00% 0,00% 9,10%
PRD 6,44%
6,51% 0,00% 0,00% 0,00% 0,00%
VIS 72,90%
44,17% 18,03% 20,25% 0,00% 0,00%
peer group HZ 08/09
GI 06/07 GI 06/07 GI 09/10 HD 10/11 WU 10/11
MZ 08/09
MZ 08/09 HZ 08/09 MZ 08/09 HZ 08/09
PF 07/08
PF 07/08 MZ 05/06 WU 07/08
PF 08/09
PF 08/09 PF 07/08
WU 07/08
PDA: dance performances, PDR: spectacle performances, PMU: music performances,
POT: other performances incl. guest-performances, PRD: productions, VIS: visitors

Here we observe three effects: (1) In classical CCR not every discipline is part of the efficiency
configuration, in RC-DEA it is. (2) Taking a look at the peer group of WU we can see that WU discriminates
itself – an effect which is a notable feature of the Window Analysis. (3) Furthermore, the disadvantaged –
category 2 – theatre WU gains efficiency in the last window being no more discriminated by the advantaged
– category 1 – theatre HZ.
number of theatres

0
-18 -14 -8 -7 -5 -4 -3 -2 -1 +0 +1 +2 +4 +5 +7 +8 +9 +10 +11 +13

number of ranks

Figure 1: variation in ranks between structural and financial configuration of inputs

At last, we compare the structual and the financial perspective. After ranking the theatres for both scenarios
by the Measure of Inefficiency Dominance (Bardhan et al., 1996), Figure 1 shows the distribution of the

37
rank variation. However, the efficiency results differ a lot, even though the input numbers are highly
correlated. Hence it is very crucial that both perspectives offer different findings: in a structural perspective
a theatre may be efficient in transforming structural potential while being inefficient in transforming
financial budget. We note that this change of perspective may be very important for the cultural scene.

RESULTS AND CONCLUSIONS


The RC-DEA model may increase the reliability of an efficiency evaluation of multidisciplinary theatres
due to incorporating different categories and weight restrictions for all disciplines. The Window Analysis
allows for a multi-period analysis over time. Therefore, the window efficiency of a theatre has been defined
as the last efficiency value of a time window. Additionally, different settings (structural, financial) show
different results and should be considered as different perspectives onto the objects of investigation. At
least we propose the application of our model for decision support within the theatre sector as well as its
discussion by an interdisciplinary group.

REFERENCES
[1] Bardhan I., Bowlin W. F., Cooper W. W., Sueyoshi T. (1996) Models and Measures for Efficiency Dominance
in DEA Part I: Additive Models and MED Measure, Journal of the Operations Research Society of Japan 39 (3):
322-332.
[2] Baumol W. J., Bowen, W. G. (1966) Performing arts – The economic dilemma. A study of problems common to
theater, opera, music and dance. New York: The Twentieth Century Fund.
[3] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[4] Last A.-K., Wetzel H. (2010) The Efficiency of German Public Theaters: A Stochastic Frontier Analysis
Approach, Journal of Cultural Economics 34 (9): 89-110.
[5] Marco-Serrano F. (2006) Monitoring managerial efficiency in the performing arts: A regional theatres network
perspective, Annals of Operations Research 145 (1): 167-181.
[6] Ossadnik W. (1987) Rahmenbedingungen und Effizienzprobleme öffentlicher Theater, Betriebswirtschaftliche
Forschung und Praxis 39 (3): 275287.
[7] Theaterstatistik 2011/2012 (2013) Theaterstatistik 2011/2012: Heft 47, Köln: Deutscher Bühnenverein
Bundesverband der Theater und Orchester.
[8] Tobias S. (2003) Kosteneffizientes Theater? Deutsche Bühnen im DEA-Vergleich. Dissertation. Dortmund:
Universtiät Dortmund.
[9] Tone K. (1997) DEA with controllable category level, Proceedings of the 1997 spring national conference of the
operation research society of Japan: 126-127.
[10] Wong Y.-H. B., Beasley J. E. (1990) Restricting Weight Flexibility in Data Envelopment Analysis, The Journal
of the Operational Research Society 41 (9): 829-835.

38
EFFICIENCY ANALYSIS OF STATE OWNED AND PRIVATE COMMERCIAL
BANKS OPERATING IN BANGLADESH: A DEA APPROACH

MUHAMMAD AMIR HOSSAIN


Statistics Department, Bangladesh Bank (Central Bank of Bangladesh), Bangladesh

ABSTRACT
There are mainly three types of deposit money banks (DMBs) operating in Bangladesh, such as; state owned banks
(SOBs), private commercial banks (PCBs) and foreign commercial banks (FCBs). PCBs are two types, namely;
Islamic private banks (IPBs) and conventional private banks (CPBs). Nature of FCBs are somewhat different from
SOBs and PCBs. We have considered SOBs, IPBs and CPBs for efficiency analysis. In the present study twenty banks
have been selected to compare their performances. The main objective of this study is to find out most efficient banks
and how much inefficient the other banks are compared to most efficient banks. In the present study Data Envelopment
Analysis (DEA) technique has been applied which is nonparametric in nature. Input oriented three models namely,
constant returns to scale (CRS), variable returns to scale (VRS) and cost efficiency DEA have been applied. In each
of the models attempts have been made to work out technical efficiency (TE), scale efficiency (SE), allocative efficiency
(AE) and cost efficiency (CE) of each bank separately. After applying these measures Eastern bank has been found as
the most efficient bank and Sonali bank has been found as the least efficient bank during the period of study (2009-
2013).

Key Words: Technical Efficiency, Scale Efficiency, Allocative Efficiency, Nonparametric, Data Envelopment etc.

INTRODUCTION
The importance of commercial banks to the socioeconomic development of a country cannot be ignored,
from both developed and developing countries banks have shown a significant role in the development and
growth of economy by insuring prudent allocation of resources as well as their efficient utilization (Raphael,
G. 2013).There are different types of banks operating in Bangladesh such as; State Owned Commercial
Banks (SOBs), Specialized Banks (SPBs), Private Commercial Banks (PCBs) and Foreign Commercial
Banks (FCBs). Although specialized banks have profit oriented activities, these banks are operating in
special areas and purposes. On the other hand functions of FCBs are mostly related to international trade
and these are somewhat different from other commercial banks. As the SOBs and PCBs are functioning in
a competitive environment and hence the present research is attempted to analyze efficiency performance
of state owned commercial banks and private commercial banks. SOBs and PCBs have some social
responsibilities but profit earning motive cannot be avoided at all. In a comparative study among different
types of banks it has been shown that performanceof FCBs is better than SOBs and PCBs (Hossain, M. A.
2010). Therefore it is essential to increase efficiencies of SOBs and PCBs to run their business in a
competitive environment like Bangladesh. Before taking any policy decision regarding performance of
banks, it is necessary to study their individual efficiencies and comparative efficiencies among them. In
view of that sense, SOBs and PCBs have been chosen for study in the present research.
The initial DEA model, as originally presented in Charnes, Cooper and Rhodes (1978), built on the earlier
work of Farrell (1957). According to their name this model is also known as CCR model. The resulting
measure which is referred to as the "Farrell measure of efficiency," was regarded by Farrell as restricted to
meaning "technical efficiency" or the amount of "waste" that can be eliminated without worsening any

39
input or output. This was then distinguished by Farrell from "allocative" and "scale" efficiencies as adapted
from the literature of economics.
Charnes et al. (1978) developed and extended Farrell’s idea into a non-parametric methodology known as
Data Envelopment Analysis. Boles (1966) and Afriat (1972) suggested mathematical programming
methods which could achieve the task although the method did not receive wide attention. There are only
a few papers written on the cost efficiency of banks in the developing countries using the DEA method,
such as Bhattacharya, Lovell and Sahay (1997) for India, Taylor, Thomsom, Thrall and Dharmapala (1997)
for Mexico, Al-Faraj, Alidi and Bu-Bshait (1993) for Saudi Arabia. Raphael, G. (2012) examined the
relative efficiency of selected 20 commercial banks in Tanzania from 2008 to 2011. The findings were
categorized based on two groups of commercial banks i.e., small and large groups. He observed that through
make use of underutilized resources and reduce operating expenses most commercial banks will remain to
be relative efficient in the productive frontier; In an efficiency analysis of commercial banks in East Africa,
Raphael, G. (2013) concluded that inefficient utilization of input resources could be one of the reasons for
the inefficiency of commercial banks in East Africa. There was a very limited study of efficiency analysis
related to Bangladeshi commercial banks.
This article is divided into four sections. The first section includes the introduction and brief literature
review. Second section explores the objectives and methodologies. The third section concentrates on
empirical results. Finally, summary and conclusions have been incorporated in the fourth section.

OBJECTIVES AND METHODOLOGIES


The broad objective of this study is to make a comparative analysis of performances of state owned
commercial banks (SOBs) and private commercial banks (PCBs) operating in Bangladesh. Specific
objectives of this study are to work out different types of efficiency measures along with some policy
recommendations. There are many input and output oriented measures of DEA methodology. In the present
study five types of input oriented efficiency measures have been applied to test comparative performances
of commercial banks. The measures are 1) Technical efficiency under constant returns to scale (TECRS),
2) technical efficiency under variable returns to scale (TEVRS), 3) scale efficiency (SE), 4) allocative
efficiency and 5) cost efficiency (CE).
Twenty commercial banks have been selected randomly to study comparative performances among them.
Among 20 commercial banks, there are 4 state owned banks (SOBs), 5 are Islamic private banks (IPBs)
and remaining 11 are conventional private banks (CPBs). The period of study has been covered from 2009
to 2013. The efficiency analysis has been performed among the banks for each individual period. In this
study four inputs have been used to produce a single output for each bank i.e., DMUs. The inputs in our
consideration are, deposits (DEP), advances (ADV), expenses (EXP) and manpower (MNP). On the other
hand the one output in our consideration is net profit (PFT). The price information of selected input
variables have also been considered to work out allocative efficiencies.
According to Farrell (1957) there are two types of broad measures of DEA, such as; input-oriented
measures and output-oriented measures. The present study is attempted to apply input oriented measures.
Let us consider there are M inputs and K outputs on each of N decision making units (DMU’s). In the

40
present research DMUs will be considered as banks. If i-th DMU’s inputs and outputs represented by xi and
yi respectively, then X represents the input matrix of order MxN and Y represents the output matrix of order
KxN.
The efficiency measurement (𝑒𝑜 ) of a particular bank can be expressed as:

u r yr 0
M ax e0  r 1
m

v xi 1
i i0

Subject to the constraints (1)


k

u r y rj
r 1
m
 1; j  1,2,..., n
v x
i 1
i ij

The model measures the relative performances of banks. There are n DMUs which are j = 1, 2, …, n. In the
model, 𝑦𝑟𝑗 > 0 represent the rth output of jth bank and 𝑥𝑖𝑗 >0 represent the observed amount of ith input of
jth bank; 𝑢𝑟 and 𝑣𝑖 represent the weights of rth output and ith input, respectively; ε is a constant smaller
than any positive valued real number for 𝑒𝑜 .
If a given firm uses quantities of inputs, defined by the point P to produce a unit of output, the technical
inefficiency of that firm could be represented by the distance QP, which is the amount by which all inputs
could be proportionally reduced without a reduction in output. This is usually expressed in percentage terms
QP
by the ratio 0P
, which represents the percentage by which all inputs could be reduced. The technical
efficiency (TE) of a firm is most commonly measured by the ratio

0Q
𝑇𝐸 = (2)
0P

TE will take a value between 0 and 1, hence it provides an indicator of the degree of technical inefficiency
of the firm. A value of 1 indicates the firm is fully technically efficient. For example, the point Q is
technically efficient because it lies on the efficient isoquant.
If the input price ratio, represented by the line AA' in figure 1 is known, allocative efficiency may also be
calculated. The allocative efficiency (AE) of the firm operating at P is defined to be the ratio

0R
𝐴𝐸 = (3)
0Q

Since the distance RQ represents the reduction in production costs that would occur if production were to
occur at the allocatively (and technically) efficient point Q', instead of at the technically efficient, but
allocatively inefficient point Q.

41
Figure 1: Technical and Allocative Efficiencies

EMPIRICAL RESULTS
Technical Efficiency under CRS
It is an input-oriented constant returns to scale (CRS) methodology based on cross-section data. In this
research four inputs have been used to produce a single output for each type of banks i.e., DMUs. The
inputs in our considerations are, deposits (DEP), advances (ADV), expenses (EXP) and manpower (MNP).
On the other hand the one output in our consideration is net profit (PFT). Inputs and output data for 5 years
(2009-2013) have been analyzed separately.

Table 1: Mean Technical Efficiency under CRS


DMUs 2009 2010 2011 2012 2013 Mean
SOBs 0.549 0.563 0.703 0.306 0.424 0.509
CPBs 0.801 0.851 0.313 0.380 0.669 0.616
IPBs 0.865 0.652 0.683 0.518 0.523 0.635
Mean 0.786 0.743 0.754 0.400 0.583 0.653

The table 1 shows that mean efficiency of Islamic private banks (63.5%) is better than conventional private
banks (61.6%). However, efficiency of state owned banks (50.9%) has been found as least efficient
compared to private banks. Bank-wise data shows that Eastern Bank is the most efficient bank (1.000) and
Sonali bank is the least efficient bank (0.119) for the year 2013 among banks under consideration. National
bank and Dhaka bank are the comparatively less efficient than other private banks. Among IPBs, Islamic
bank is the most efficient bank whereas Agrani bank is the most efficient among SOBs.
Using the DEA, it is observed that mean of the mean technical efficiency under CRS over the period under
study is 65.3% (table 1). This implies that there is sufficient room for improvements in technical efficiency
of banks. Efficiency of CPBs is showing increasing trend from 2011 and onwards. Other groups of banks
are also showing increasing trend although efficiencies of SOBs and IPBs declined from 2011 to 2012.

42
Technical Efficiency under VRS
In the previous model it has been found technical efficiencies by considering constant returns to scale
(CRS). An attempt has been made to work out technical efficiencies in the present section by considering
variable returns to scale (VRS). This is another input-oriented DEA analysis using same output and inputs
used in case of CRS. The VRS and CRS input-oriented DEA results for the year 2013 are listed in table 2.
It has been observed that mean technical efficiency for all DMUs is 83.2% under VRS and 58.3% under
CRS in 2013. The Agrani, Janata, Islami, Social Islami, Uttara, Eastern and Dhaka bank have been found
as most efficient bank under VRS and Eastern bank is found as full efficient bank under CRS. On the other
hand, Sonali bank has been found as the inefficient bank both in CRS and VRS in 2013.

Table 2: VRS Input Oriented DEA Results: 2013


Banks CRSTE VRSTE SE
SOBs 0.424 0.758 0.536
IPBs 0.523 0.891 0.597
CPBs 0.669 0.832 0.810
Mean 0.583 0.832 0.702

It is also observed that Eastern bank is the most scale efficient and Sonali bank is the least scale efficient.

Allocative and Cost Efficiency


This is a CRS cost efficiency analysis using the three inputs and one output along with related cost data. In
the model of discussion allocative efficiencies and cost efficiencies have been worked out for each bank in
addition of technical efficiencies. The table 3 shows that, although CPBs are the most efficient in terms
technical and cost but IPBs is the most efficient in terms of allocative efficiency. Eastern bank is the most
efficient bank in terms of AE and CE in 2013. On the other hand Sonali Bank has been found as the least
cost efficient in 2013. It is observed that allocative efficiencies are higher than technical efficiencies and
cost efficiencies for all banks.

Table 3: CRS Cost Efficiency DEA Results: 2013


DMUs TECRS AE CE
SOBs 0.424 0.684 0.295
CPBs 0.669 0.898 0.599
IPBs 0.523 0.904 0.469
Mean 0.583 0.857 0.506

Overall Performance in Last 5 Years


The table 4 shows the mean TECRS, TEVRS, SE, AE and CE of last 5 years. From the table it has been
observed that IPBs found as the most efficient in terms of all measures mentioned above and SOBs has
been found as the least efficient in terms of same measures. After analyzing measures of last 5 years it is
found that Eastern bank is the full efficient and Sonali bank has been found as the least efficient in terms
of TECRS, TEVRS, SE, AE and CE throughout the period of study. From the table 4 it is observed that the
mean efficiencies of SOBs, CPBs and IPBs are 60.0%, 60.3% and 74.3% respectively.

43
Table 4: Measure-wise Efficiencies in Last 5 years
Banks TECRS TEVRS SE AE CE Mean
SOBs 0.509 0.713 0.710 0.709 0.359 0.600
IPBs 0.635 0.894 0.747 0.910 0.530 0.743
CPBs 0.616 0.747 0.323 0.811 0.521 0.603
Mean 0.653 0.835 0.785 0.880 0.536 0.738

The average efficiency of all measures for each year has been listed in table 3.5, also shown in figure 2.
The table 5 and figure 2 shows that mean efficiency level remains maximum for IPBs and remains minimum
for SOBs throughout the period of study.

Table 5: Period-wise Efficiencies in Last 5 years


Banks 2009 2010 2011 2012 2013 Mean
SOBs 0.593 0.629 0.731 0.507 0.539 0.600
CPBs 0.873 0.866 0.328 0.579 0.762 0.603
IPBs 0.815 0.749 0.766 0.681 0.677 0.743
Mean 0.803 0.790 0.802 0.590 0.696 0.738

0.743 0.738 1.000


0.8
0.600 0.603 0.800
0.6 0.600 SOBs
CPBs
0.4 0.400
IPBs
0.200
Overall
0.2
0.000
0
SOBs CPBs IPBs Overall

Figure 2: Mean of TE, SE, AE and CE of Last 5 Figure 3: Measure and Bank-wise TE of Last 5
Years Years
The figure 3 shows that efficiencies of IPBs is better than all other group of banks. In every case allocative
efficiencies are better than all other measures.

Table 6: Significance Test of Spearman’s Correlation Coefficient among Efficiency Measures


TECRS TEVRS SE AE CE
TECRS 1.000 --- --- --- ---

TEVRS 0.986* 1.000 --- --- ---


(0.002)
SE 0.963* 0.906* 1.000 --- ---
(0.008) (0.034)
AE -0.551 -0.523 -0.517 1.000 ---
(0.336) (0.366) (0.373)
CE 0.968* 0.945* 0.957* -0.342 1.000
(0.007) (0.015) (0.011) (0.573)
*Significant at 5% level of significance

44
The table 6 shows that maximum banks are concentrated having efficiency level between 0.601 to 1.000.
The results of table 5 could be validated by significant at the 5% level (2-tailed). The null hypothesis is that
the rank correlation co-efficient between two efficient variables is zero. The empirical results of table 6
indicate that there is a satisfactory relationship among different efficiency measures. It is suggested that the
various measures of banking efficiencies are satisfactorily associated with each other.

SUMMARY AND CONCLUSIONS


Five different measures have been applied in each of SOBs, IPBs and CPBs separately. In each of the
measures it is attempted to work out efficiency performance of each bank separately. After analyzing
performances of state owned and private banks from 2009 to 2013 it is observed that Easter bank has been
found as the most efficient bank and Sonali bank has been found as least efficient bank in consideration of
TECRS, TEVRS, SE, AE and CE. Further analysis is needed to find out the reasons of inefficiencies of
Sonali bank.
It is recommended that state owned commercial banks specially, Sonali bank should minimize the use of
input resources while maintaining the same level of output compared to other banks. By improving handling
of operating expenses, advances, capital and by boosting banking investment operation, the less efficient
banks can successfully endorse resource utilization efficiency. However the results of the analysis have
important implications for management of the banks, policy makers and bank regulators in Bangladesh.

REFERENCES
[1] Annual Reports (2007-2011), Selected Scheduled Banks.
[2] Afriat, S.N. (1972), "Efficiency Estimation of Production Functions", International Economic Review, 13, pp568-
598
[3] Al-Faraj, T.N., A.S. Alidi, K.A. Bu-Bshait (1993), "Evaluation of Bank Branches by Means of Data Envelopment
Analysis", International Journal of Operations and Production Management, 13, 45-52
[4] Bhattacharya, A., C.A.K. Lovell and P. Sahay (1997), "The Impact of Liberalization on the Productive Efficiency
of Indian Commercial Banks", European Journal of Operational Research.
[5] Boles, J.N. (1966), "Efficiency Squared-Efficient Computation of Efficiency Indices", Proceedings of the 39th
Annual Meeting of the Western Farm Economic Association, pp 137-142
[6] Boshrabadi, H.M., Villano, R. and Fleming, E. (2006), "Technical Efficiency and Environmental Gaps in Wheat
Production in Kerman Province of Iran: A Meta Frontier Analysis", Working Paper, 2006-6, University of New
England, Australia
[7] Charnes, A., W.W. Cooper and E. Rhodes (1978), "Measuring the Efficiency of Decision Making Units"'
European Journal of Operations Research, 2, pp 429-444
[8] Farrell, M.J. (1957), "The Measurement of Productive Efficiency", Journal of the Royal Statistical Society,
[9] A CXX, Part 3, 253-290
[10] Hossain, M. A. (2010), "Comparative Performance Analysis of Nationalized, Specialiszed, Private and Foreign
Commercial Banks Operating in Bangladesh" PhD Thesis
[11] Raphael, G. (2012), “Commercial Banks Efficiency in Tanzania: A Non Parametric Approach”, European Journal
of Business and Management; Vol 4, No. 21, 2012
[12] Raphael, G. (2013), “Efficiency of Commercial Banks in East Africa: A Non Parametric Approach”, International
Journal of Business and Management; Vol. 8, No. 4; 2013

45
[13] Sufian, F. (2006), "The Efficiency of Non-Bank Financial Institutions: Empirical Evidence From Malaysia"
[14] International Research Journal of Finance and Economics, Euro Journals Publishing, Inc, 2006
[15] Taylor, W.M., R.G. Thompson, R.M. Thrall, and P.S. Dharmapala (1997), "DEA/AR Efficiency and Profitability
of Mexican Banks: A Total Income Model", European Journal of Operational Research.

46
EFFICIENCY MEASUREMENT OF REGIONAL DEVELOPMENT AGENCIES
IN TURKEY
V. ALPAGUT YAVUZ
Mustafa Kemal University, Hatay, Turkey

ABSTRACT
Differences in wealth, prosperity and development rates at the regional level are important issues especially in
developing countries. As a developing country, Turkey has regional disparities and different local dynamics. In 2006,
a new approach in regional development was put into practice in Turkey. Thus, Regional Development Agencies
(RDAs) for each region were established in a three-year period in order to reduce disparities and accelerate regional
development. The main objective of these RDAs is to provide various support mechanisms to institutions, associations
and individuals. Thus, RDAs require performance evaluation in order to ensure the institutional effectiveness and
success of their activities. In this study, three analyses are conducted using Data Envelopment Analysis (DEA) and
efficiencies of 26 RDAs between the years of 2011 and 2014 are analysed. Also, Malmquist total factor productivity
index is used to identify the efficiency changes over the same period.

Keywords: Development Agencies; Data Envelopment Analysis; Malmquist Total Factor Productivity Index

INTRODUCTION
In order to initiate a balanced spread of wealth and prosperity local economic development activities has
been an important subject for world economies on both national and international levels. Regional
Development Agencies (RDAs) and similar organizational structures (companies or corporations) have
been one of the preferred approaches for shaping and pursuing local economic strategies since the end of
World War II to aid the post-war reconstruction. These organizational structures vary in size and function
and they are appropriately regulated and supervised by a central authority (Mountford, 2009). For example,
in the 60s and 70s in North America, these organizations have been used to address the impact of de-
industrialization. In the 1980s and 1990s in East Asia, they were used to help plan and manage rapid
urbanization and industrialization. In the recent years in Eastern Europe, they became organizational
vehicles to promote economic development in the newly integrating economies. According to the European
Association of Regional Development Agencies (EURADA, 1999) these organizations can be defined as
“an operational structure that identifies sectoral or overall development problems, and chooses a range of
opportunities or methodologies for their solution and promotes projects that can maximize the solutions to
the problems”.
Regional disparities has been an important issue in Turkey’s economic development history. Although
some policies have been established in order to eliminate the inter-regional differences; they couldn’t create
the expected positive effect in eliminating those differences (Sevinç, 2011). Historically, Turkey has a
highly centralized system of government and it’s reflected in its institutional structures. Regional policies
and projects were managed by the State Planning Organization at the national level until a decade ago.
After the European Union’s decision of Turkey’s candidacy in 1999 the reform process was accelerated
(Tiftikcigil, 2015). One step in this process was the establishment of Regional Development Agencies. In
2006, a new regional development approach was put into practice and RDAs entered into the Turkish legal
system. In the next three years, one RDA was established for each of the 26 regions. (Figure 1) The main

47
objective of these RDAs are (Toktaş, et.al. 2013) to minimize the regional development differences; to
improve the cooperation between public and private sectors; to provide efficient usage of resources in
appropriate locations and to ensure the sustainability. Achieving these objectives in the long run requires a
close watch on the results of activities and policies developed or managed by each RDA and doing the
necessary managerial adjustments accordingly. Thus measuring performances of these agencies is an
important requirement in order to ensure institutional development, manage an effective resource allocation
and increase service quality. Due to the variety of existent regional economic development models in the
world and the differences in each country’s domestic traditions and socio economy (Young-Hyman, 2008);
performance measurement of all aspects of RDAs is a very complex and difficult process and could not be
achieved with one performance measurement method or technique. Thus measurement of RDA
effectiveness could either be based on the goals stated in legislation or defined by policy makers, or broad
indicators of economic growth (Young-Hyman, 2008). For example in England, five different approaches
were employed (an interim approach, a three-tier approach, RDA tasking framework, outcome-oriented
measurement and independent performance measurement) in terms of their purpose, coverage and criteria
and used for measuring different aspects of RDAs (OECD, 2009). Turkish RDAs’ coordination at the
national level is the responsibility of the Ministry of Development. Since 2010, the Ministry has been
measuring the RDA’s performance to decide on the amount of central government’s share of the agencies
for the following years’ budget (Şimsek, 2013). Measurements are done using seven criteria reflecting
RDAs activities such as funding, technical support, investment support, cooperation and such. However
none of these measures has been enough to evaluate the efficiencies of RDAs completely. Measuring
efficiencies of RDAs could be a valuable indicator for their capacity to coordinate regional actors and
allocate available resources toward the comprehensive realization of long term economic development
projects.

Figure 1: RDAs of Turkey


For measuring efficiency, Data Envelopment Analysis (DEA) stands out among other methods because it
can handle various type of data as being a non-parametric analysis. In the literature, very few studies
employed DEA for measuring relative efficiencies of regions with the perspective of regional development
such as China (e.g. Lu, and Lo, 2007). Only two of these studies were done for Turkish RDAs. Kirkanbeş
(2013) evaluated 25 RDAs efficiency using input oriented DEA-CCR model with 1 input and 2 outputs for

48
the year 2010. Şimsek (2013) combined DEA and Balance Score Card in order to measure performance
and success level in a comprehensive approach. In this model five input oriented DEA-CCR sub models
were used for different aspects of the performance due to data unavailability for the year 2010. Because
Turkish RDAs have not existed long enough to prove their efficiency and accountability to the regions,
these two studies provide a limited perspective on RDA’s performance in the long run. Therefore, the
purpose of this study is to analyse Turkish RDA’s efficiencies in a longer span of time using DEA and
Malmquist total factor productivity index between 2011 and 2014. For this purpose, three separate analysis
were conducted. First for efficiency of each year covering the four year period. Then efficiency changes
over the years are evaluated using Malmquist index (MI). Next analysis were done combining all year’s
data to evaluate cumulative efficiencies of the RDAs. In the last analysis non-discretionary data
incorporated in the model in order to account for uncontrollable effects of regional disparities and inherent
local dynamics on RDAs performance.

METHODS
Data Envelopment Analysis has been built on the idea of efficiency analysis put forward by the studies of
Farrell (1957). In the seminal work by Charnes, Cooper and Rhodes (1978) linear programming framework
was used to define efficient frontier. In this procedure which is also called CCR model, all decision making
units (DMUs) are assumed to operate at their optimal scale. Later, Banker, Charnes and Cooper (1984)
suggested an extension of this model that is called the BCC model to account for variable returns to scale
(VRS) situations. The CCR and BCC efficiencies are applied to evaluate the relative efficiency of Turkish
RDAs for each year between 2011 and 2014 for the first analysis. The Malmquist productivity indexes of
the RDAs are measured and decomposed into technical efficiency change and technological change for the
same period. For both analyses, computations were done using DEAP v.2.1.
In the next analysis, rankings of RDAs were done using the super-efficiency DEA model (Andersen and
Petersen, 1993) under the assumption of variable returns to scale (VRS) for the cumulative efficiency
evaluation. In the final analysis, effects of inter-regional differences on efficiency of RDAs were
investigated by incorporating two different categorical variable separately in two models as non-
discretionary variable as suggested by Banker and Morey (1986). EMS (Efficiency Measurement System)
v.1.3.0 is used for the analysis.
Input and Output Variables
In order to identify suitable input output combination, it is crucially essential to have a clear understanding
of the process being evaluated. As stated by Cook et.al. (2014) “clear specification of the function to be
studied will drive the choice of inputs and outputs to be examined.” In addition, the purpose of the
performance measurement effects not only the input output selection but also the model orientation as well.
The RDAs in Turkey use a variety of mechanisms to enable regional development. These include preparing
regional plans; providing financial and technical support for the projects in the areas specified under the
national development plans and programs; fostering cooperation between universities and industry; and
supporting research and development and innovative activities of SMEs (Tiftikcigil, 2015). Experiences in
the Central and Eastern European countries suggest that RDA efficiency is effected greatly by three factors

49
(Young-Hyman, 2008): the role and performance of central coordinating institutions; the source and
character of funding for the RDAs; and the technical capacity of the agency staff. For the case of RDA
efficiency measurement where all RDAs operate under the same central coordinating institutions as in
RDAs in Turkey, the effect of this factor can be neglected since the conditions are the same for all the
RDAs.
Fundamentally for the RDAs in Turkey, all the activities are constrained by budgetary limits and technical
capacity of the RDAs’ staff. These are the inputs that can be controlled to a certain level by the RDAs. With
this insight, total budget realization and number of personnel are chosen as the inputs for the model. In
order to represent the results of RDA’s annual activities, three outputs that could best represent support
mechanisms used by the RDAs are selected. These are; project and operations support services expenditure,
number of supported projects (subtotal of the call for proposal (CFP), direct operating support (DOS),
technical support (TS) and guided project support (GPS)), and number of participants in training programs.
All data were composed from RDA’s annual activity reports between 2011 and 2014 (Table 1). Since
fundamental purpose of their existence is to become a fostering mechanism of regional development,
expanding output is important. Thus efficiency measurement needs to be in this perspective.

Table 1: Summary statistics of inputs and outputs


Number of Project and operations
Number of Total budget
participants in support services Personnel
supported projects realization
training programs expenditure
2011 Mean 1037.58 6795150.85 110.69 35.81 12175967.17
SD 1076.73 6145072.65 52.44 6.25 7337702.30
Min 194 22777.73 6 28 3923022
2012 Mean 640.15 14762588.32 84 37.19 21094760.12
SD 553.50 7166637.42 63.22 6.93 7886181.73
Min 48 6897612.86 25 26 10968689.69
2013 Mean 548.88 13628864.21 114.77 37.96 19913726.44
SD 438.56 9245080.27 58.09 5.72 10255448.39
Min 96 2602526.35 30 28 5837435.09
2014 Mean 588.08 20070383.52 113.15 36.54 28637710.74
SD 538.49 13044234.63 31.17 6.33 13792392.02
Min 70 8090449 45 25 14209937

RESULTS AND DISCUSSIONS


In the TE analysis (Table 2) by the BCC model 11 RDAs in 2011, 10 RDAs in 2012, and 13 RDAs are
efficient in 2013 and 2014. DOGAKA is the only DA efficient in all periods. When all RDAs considered
for the entire period, technological change decreases 0.6% and TFP decreases 0.04%. Efficiency change
increases 0.6%, pure technical efficiency change increases 0.2% and scale efficiency increases 0.4%.
Productivity scores and parameter estimations of the RDAs are obtained in terms of TFP Analysis with an
output oriented MI. Efficiency and pure technical efficiency change has similar behaviour. The highest
productivity is in 2012 with an increase of 0.6%. The lowest productivity is in 2014 with a decrease of
0.5%. The TFP exhibits a decrease at the rate of 0.4%. The productivity change shows a decreasing trend.
RDAs expected results can be achieved in the long run. Since some of the activities/projects/programs
spread over the years, cumulative performance evaluation covering all years can be used for overall

50
performance of RDAs. For this analysis average personnel number is calculated over the period and for the
others each year’s input and output data are combined to account for the cumulative performance.
Cumulative performance analysis result shows that; 7 RDAs are efficient when CRS is assumed. 9 RDAs
are efficient when VRS is assumed. 8 RDAs have scale efficiency. Increasing returns to scale exist for 11
RDAs. Decreasing returns to scale exist for 7 RDAs. Compare to the analysis done using annual data
cumulative performance analysis has less efficient RDAs (12 (avg.) vs 9).

Table 2: Efficiency scores of Turkish RDAs over the period 2011-2014


2011 2012 2013 2014
DMU crste vrste scale crste vrste scale crste vrste scale crste vrste scale
AHIKA 0.703 0.841 0.836 irs 0.853 1 0.853 irs 1 1 1 - 0.944 0.986 0.957 irs
ANKARAKA 0.79 1 0.79 irs 0.779 0.816 0.954 irs 0.819 0.849 0.965 irs 1 1 1 -
BAKA 0.646 0.737 0.877 irs 0.826 0.914 0.903 irs 0.817 0.819 0.998 irs 0.833 0.844 0.986 drs
BAKKA 0.754 1 0.754 irs 0.679 1 0.679 irs 0.81 1 0.81 irs 1 1 1 -
BEBKA 0.843 0.91 0.927 irs 0.908 0.947 0.959 irs 0.684 0.697 0.982 irs 0.88 0.893 0.985 drs
CKA 1 1 1 - 0.988 0.989 0.999 irs 0.926 0.931 0.995 irs 0.963 1 0.963 irs
DAKA 0.782 1 0.782 irs 0.685 0.771 0.889 irs 1 1 1 - 1 1 1 -
DIKA 0.695 0.696 0.999 irs 0.798 0.855 0.934 irs 0.769 0.783 0.982 irs 0.792 0.858 0.923 irs
DOGAKA 1 1 1 - 1 1 1 - 1 1 1 - 1 1 1 -
DOKA 0.616 0.74 0.832 irs 0.712 0.745 0.955 irs 0.806 0.841 0.959 drs 0.671 0.686 0.978 drs
FKA 0.752 0.791 0.951 irs 0.791 0.845 0.936 irs 0.995 1 0.995 irs 0.873 0.913 0.956 irs
GEKA 0.612 1 0.612 irs 0.77 0.909 0.848 irs 0.811 0.838 0.967 irs 0.929 0.933 0.995 irs
GMKA 0.5 0.529 0.945 irs 1 1 1 - 0.806 0.815 0.989 irs 1 1 1 -
IKA 0.904 0.948 0.953 irs 0.95 1 0.95 irs 0.896 1 0.896 irs 0.977 0.98 0.996 drs
ISTKA 0.707 1 0.707 irs 0.922 1 0.922 drs 0.947 1 0.947 drs 1 1 1 -
IZKA 1 1 1 - 0.646 0.67 0.963 irs 0.683 0.692 0.987 irs 0.841 0.901 0.933 irs
KARACADAG 0.894 0.914 0.978 irs 1 1 1 - 1 1 1 - 0.912 0.922 0.99 irs
KUDAKA 0.983 0.988 0.995 drs 0.891 0.938 0.95 irs 1 1 1 - 0.706 0.741 0.953 drs
KUZKA 0.44 1 0.44 irs 0.858 0.92 0.933 irs 0.669 0.872 0.767 irs 0.976 0.988 0.988 irs
MARKA 0.658 0.817 0.805 drs 0.911 0.915 0.996 irs 1 1 1 - 1 1 1 -
MEVKA 0.982 0.983 0.999 drs 0.937 0.958 0.978 irs 1 1 1 - 0.982 1 0.982 irs
OKA 1 1 1 - 1 1 1 - 0.843 0.853 0.988 irs 1 1 1 -
ORAN 0.768 0.94 0.817 irs 0.973 1 0.973 irs 0.827 0.828 0.999 drs 1 1 1 -
SERKA 0.739 0.805 0.918 irs 0.786 0.822 0.957 irs 0.751 0.798 0.94 irs 1 1 1 -
TRAKYAKA 0.721 0.756 0.954 irs 0.72 0.844 0.853 irs 1 1 1 - 0.953 0.976 0.977 irs
ZAFER 1 1 1 - 0.952 1 0.952 irs 1 1 1 - 1 1 1 -
Mean 0.788 0.9 0.88 0.859 0.918 0.936 0.879 0.908 0.968 0.932 0.947 0.983

Table 3: Malmquist index summary of annual average efficiency results


Year Efficiency Technological Pure Technical Efficiency Scale Efficiency TFP
change change change change change
2011-2012 1.105 0.958 1.026 1.077 1.059
2012-2013 1.023 0.963 0.989 1.035 0.986
2013-2014 1.063 0.889 1.045 1.017 0.946
Mean 1.063 0.936 1.02 1.043 0.996

Last analysis was done for the uncontrollable regional disparities. According to Banker and Morey (1986)
this type of uncontrollable characterization can be used to force the peer group members to be compared
only to similar RDAs. For this purpose the Regional Competitiveness Operational Programme’s (RCOP)
classification of regions was used. In RCOP regions are grouped in two categories where income per capita
is below or above the 75% of the Turkish national average according to 2001 data. Two regional
classification levels of RCOP are treated as non-discretionary input for the analysis. Results shows that
(Table 4) the amounts and the identity of efficient RDAs are mostly the same in two models.

51
Table 4: Cumulative and non-discretionary input model efficiency results
Cumulative Model Treatment (non-discretionary input)
# DMU Score Benchmarks Score Benchmarks
1 AHIKA 1.1857 4 (0,089) 9 (0,409) 19 (0,334) 23 (0,168) 1.06621 13 (0,843) 20 (0,032) 21 (0,111) 26 (0,015)
2 ANKARAKA 1.18529 4 (0,196) 9 (0,299) 23 (0,505) 1.13415 4 (0,397) 25 (0,039) 26 (0,565)
3 BAKA 1.16814 4 (0,051) 9 (0,065) 19 (0,052) 23 (0,832) 1.06334 4 (0,223) 21 (0,419) 25 (0,358)
4 BAKKA big 8 big 8
5 BEBKA 1.08001 4 (0,558) 9 (0,205) 23 (0,236) 1.03133 4 (0,232) 25 (0,442) 26 (0,326)
6 CKA 0.94845 4 0.83067 5
7 DAKA 0.91981 0 0.91981 0
8 DIKA 1.17688 4 (0,034) 9 (0,190) 23 (0,776) 1.17688 4 (0,034) 9 (0,190) 23 (0,776)
9 DOGAKA 0.53169 13 0.53169 5
10 DOKA 1.3978 4 (0,444) 9 (0,521) 23 (0,035) 1.3978 4 (0,444) 9 (0,521) 23 (0,035)
11 FKA 1.04005 4 (0,662) 9 (0,215) 23 (0,123) 1.04005 4 (0,662) 9 (0,215) 23 (0,123)
12 GEKA 1.14329 9 (0,180) 15 (0,022) 23 (0,798) 1.07635 4 (0,579) 6 (0,167) 26 (0,254)
13 GMKA 1.06501 4 (0,143) 9 (0,421) 19 (0,436) 0.8623 1
14 IKA 0.98049 0 0.98049 0
15 ISTKA 0.6713 6 0.66198 3
16 IZKA 1.20581 6 (0,527) 15 (0,140) 23 (0,333) 1.18715 4 (0,166) 6 (0,834)
17 KARACADAG 1.00352 6 (0,761) 15 (0,003) 23 (0,236) 1.00352 6 (0,761) 15 (0,003) 23 (0,236)
18 KUDAKA 1.09424 6 (0,490) 9 (0,090) 15 (0,126) 23 (0,295) 1.09424 6 (0,490) 9 (0,090) 15 (0,126) 23 (0,295)
19 KUZKA big 3 big 0
20 MARKA 1.11033 9 (0,718) 25 (0,282) 0.67992 1
21 MEVKA 1.05623 9 (0,177) 15 (0,090) 23 (0,733) 0.97592 2
22 OKA 1.06611 6 (0,299) 9 (0,042) 15 (0,041) 23 (0,619) 1.06611 6 (0,299) 9 (0,042) 15 (0,041) 23 (0,619)
23 ORAN 0.9153 13 0.9153 6
24 SERKA 0.9988 0 0.9988 0
25 TRAKYAKA 0.88701 1 0.48377 3
26 ZAFER 0.85801 0 0.57938 4

CONCLUSIONS
There is a great need for performance evaluation in RDAs at the administrative level in order to configure
programs/policies according to the changing needs of the regions. DEA is a valuable analysis tool indicating
where to focus for improvement and finding a peer for benchmarking. It should be an integral part of an
evaluation system for RDAs. However, regional disparities need to be taken into consideration in the DEA
analysis especially for benchmarking studies. Thus, there is a need for up-to-date and consistent indexing
mechanisms for capturing current regional disparities.

REFERENCES
[1] Andersen P., Petersen, N.C. (1993), A procedure for ranking efficient units in data envelopment analysis,
Management Science, (39): 1261–1264.
[2] Banker, R.D., Charnes, A., Cooper, W.W. (1984) Some models for estimating technical and scale inefficiencies
in data envelopment analysis, Management Science, 30 (9): 1078–1092.
[3] Banker, R.D., Morey, R.C., (1986) Efficiency analysis for exogenously fixed inputs and outputs, Operations
Research 34 (4): 513–521.
[4] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[5] EURADA (1999) Creation, development and management of RDAs: does it have to be so difficult?, European
Association of Development Agencies, Brussels.
[6] Farrell M.J. (1957) The measurement of production efficiency, Journal of Royal Statistical Society (120): 253-
290.
[7] Kirankabeş, M.C. (2013) Yeni bölgesel kalkinma politikasinin yerel aktörleri olarak kalkinma ajanslarinin
etkiliğinin değerlendirilmesi: Türkiye örneği, Dumlupinar University Journal of Social Science, (35): 253-268.

52
[8] Lu, W.-M., Lo, S.-F. (2007) A benchmark-learning roadmap for regional sustainable development in China,
Journal of the Operational Research Society, 58 (7): 841–849.
[9] Mountford, D. (2009) Organising for local development: the role of local development agencies (Summary
Report). CFE/LEED, OECD.
[10] OECD (2009) Governing regional development policy, OECD Publishing.
[11] Sevinç H. (2011) Bölgesel kalkınma sorunsalı: Türkiye’de uygulanan bölgesel kalkınma politikaları, Journal of
Entrepreneurship and Development, 6 (2): 37-40.
[12] Tiftikcigil, B.Y. (2015) An assessment on activities of regional development agencies in Turkey, International
Journal of Economics and Financial Issues, 5 (2): 399–409.
[13] Toktaş, Y., Sevinç, H., Bozkurt, E. (2013) The evolution of regional development agencies: Turkey case, Annales
Universitatis Apulensis Series Oeconomica, 2 (15): 670-681.
[14] Young-Hyman, T. (2008) The potential for effective regional development agencies in turkey: a comparative
analysis. Regional & Federal Studies, 18 (4): 375–402.

53
54
EFFICIENCY OF CONTAINER TERMINALS OF APEC COUNTRIES, 2000-
2013: A MEASUREMENT WITH A DYNAMIC DEA MODEL
ODETTE V. DELFÍN-ORTEGA
Instituto de Investigaciones Económicas y Empresariales de la Universidad Michoacana de San Nicolás de Hidalgo,
Morelia Michoacán México.

CÉSAR L. NAVARRO-CHÁVEZ
Instituto de Investigaciones Económicas y Empresariales de la Universidad Michoacana de San Nicolás de Hidalgo,
Centro, Morelia Michoacán México.

ABSTRACT
This research presents efficiency measurement of 38 container terminals ports from Asia-Pacific Economic
Cooperation (APEC) countries of the years 2000 to 2013. A Dynamic DEA model is implemented, using as inputs:
quay length, employees and cranes and as outputs: port revenues and TEUS. The carry over to connect one period
with the other is the variable port revenue, so in the first period is an output and in the next period is an input. These
carry-overs play an important role in measuring the efficiency of decision making units in each term as well as over
the whole terms based on the long-term viewpoint. The result show that in 2013, was the most efficiency year in
average because they got 62%, and from 2006 to 2010 it was obtained the lowest scores of all the years. Nevertheless
Shanghai was the port who had higher levels of efficiency in 8 periods and Kaohsiung, Taiwan, was the port who was
less efficiency with 15%. The results indicated too, that the carry-over impact on the efficiency measurement, and in
this case, 27 ports show inefficiency in the use of resources and not achieving maximum optimization revenue.

Keywords: Dynamic, DEA, Ports, APEC

INTRODUCTION
Asia-Pacific Economic Cooperation (APEC) is a regional trade block created in 1989 and presently it is
integrated with 21 countries: Australia, Brunei Darussalam, Canada, Chile, China (People's Republic),
South Korea, USA, Philippines, Hong Kong, Indonesia, Japan, Malaysia, Mexico, New Zealand, Papua
New Guinea, Peru, Russia, Singapore, Thailand, Taiwan, Vietnam. It represents 40% of world population:
2,600 million people, about 54% of world GDP and about 44% of world trade. APEC represents the most
economically dynamic region in the world that has generated almost 70% of world economic growth over
the past 10 years (APEC, 2014). Commercial transactions in Asia-Pacific region, are made primarily
through shipping; that is why, the ports are a very important part in the development in a country, as they
allow more efficient transportation system.
The evaluation of the efficiency is a topic that, in recent years, has attracted considerable interest due mainly
to use efficiently of resources in order to improve profitability. One way to study the efficiency is by Data
Envelopment Analysis, known as DEA. The efficient operation of any of the activities taking place in the
port area is important for using products shipping from reaching the final consumer markets at the lowest
cost and in the shortest time possible.
In the specific case of container terminals, the market for services maneuvers container comprises various
services that are made to move a container between the vessel and land transport otherwise. In these cases,
when port infrastructure is used, the production obtained in a given period depends on the inputs used for
that period and on the levels of other inputs decided in previous periods. This situation requires considering

55
a model with a dynamic approach that acknowledges the existence of an intertemporal relationship between
the inputs used and the resulting outputs.
For that reason, the aim of this research is make a measurement of efficiency using dynamic DEA model
to see its evolution over the time, of 38 container terminals in major ports of the member countries of Asia-
Pacific Economic Cooperation APEC for the period 2000-2013.

METHODS
The dynamic DEA model was initially proposed by Sengupta (1995) who presented a dynamic DEA model
that uses linear programming techniques to measure the shadow values of the quasifixed inputs and their
optimal pattern of change. Then Färe and S. Grosskopf (1996), presented the first innovative scheme for
dealing formally with inter-connecting activities, and after authors Nemoto, J. and Goto, M. (2003) and
Sueyoshi and Sekitani K. (2005), continued extended the model with returns to scale in dynamic DEA,
Ouellette, P. and Yan, L. (2008), and Tsutsui and Tone (2010) used carry-over variables in the dynamic
DEA model to estimate the production frontier over several time periods and developed a model in the
slacks-based measure (SBM) framework.
Emrouznejad and Thanassoulis (2005) define a set of dynamic production possibilities that evaluates the
behavior of a DMU not at a specific period, but along a time window comprising several time periods. This
dynamic proposal considers the level of capital stock in a period of time as an input that contributes to
production in future periods, and considers the level of capital stock present during the last period analyzed
within the time window as yet another output to be used beyond the time window.

In the dynamic DEA the relation between the inputs and outputs is considered in the dynamic productions.
Here, we will first define time-dependent parameters and then, the dynamic DEA will be formulated. The
element that distinguishes dynamic DEA from other types of DEA is the existence of a transition that links
the periods over time. The dynamic structure in DMUs (decision making units) over T terms, each term t,
each DMU has its own inputs and outputs along with the carry-over (link) to the next term t+1. Tone and
Tsutsui (2010) classified carry-over activities, into four categories as follows:
- Desirable (good)
This indicates desirable carry-over, e.g. profit carried forward and net earned surplus carried to the next
term. In this model, desirable links are treated as outputs and link value is restricted to be not less than the
observed one. Comparative shortage of links in this category is accounted as inefficiency.
- Undesirable (bad): This belongs to undesirable carry-over, e.g. loss carried forward, bad debt and dead
stock. In this model, undesirable links are treated as inputs and its value is restricted to be not greater than
the observed one. Comparative excess in links in this category is accounted as inefficiency.
- Discretionary (free): This corresponds to carry-over that DMU can handle freely. Its value can be
increased or decreased from the observed one. The deviation from the current value is not directly reflected
in the efficiency evaluation, but the continuity condition between two terms explained in the following
period exerts an indirect effect on the efficiency score

56
- Non-discretionary (fixed): This indicates carry-over that is beyond the control of DMU. Its value is fixed
at the observed level. Similarly to free link, fixed link affects the efficiency score indirectly through the
continuity condition between two terms (Tone and Tsutsui, 2010).
The continuity of connection flows between terms t and t +1 can be guaranteed by the follow condition:

∑𝑛𝑗=1 𝑧𝑖𝑗𝑡
𝛼
𝜆𝑗𝑡 = ∑𝑛𝑗=1 𝑧𝑖𝑗𝑡
𝛼 𝑡+1
𝜆𝑗 (∀𝑖; 𝑡 = 1, 𝐾, 𝑇 − 1) (1)

Where 𝛼 stands for good, bad, free or fixed carryovers.


The process of dynamic production studied in 1 to n +1 periods and showed as an index set T = {1,…, t+1}.
For the desired time period t the observation period depicts a time-dependent model of the production
process where the output used in the period t can be used as input in the next period (t + 1).
(𝑡, 𝑡+1) (𝑡, 𝑡+1) 𝑡
𝑧𝑘𝑙 = ∑𝑛𝑗=1 𝑧𝑗𝑘 𝜆𝑗𝑘 (∀𝑘𝑙 , ∀𝑘, 𝑡 = 1, … , 𝑇 − 1) (as carry-over from t) (2)
𝑙

(𝑡, 𝑡+1) (𝑡, 𝑡+1) 𝑡+1


𝑧𝑘𝑙 = ∑𝑛𝑗=1 𝑧𝑗𝑘 𝜆𝑗𝑘 (∀𝑘𝑙 , ∀𝑘, 𝑡 = 1, … , 𝑇 − 1) (as carry-over to t+1) (3)
𝑙

∑𝑛𝑗=1 𝜆𝑗𝑘
𝑡 𝑡
= 1 (∀𝑘, ∀𝑡), 𝜆𝑗𝑘 ≥ 0 (∀𝑗, ∀𝑘, ∀𝑡),
𝑡 𝑡
Where 𝜆𝑗𝑘 = (𝜆𝑗𝑘 ) ∈ 𝑅+𝑛 is the intensity vector corresponding to division k (k=1,…K) at t (t=1,…T).

𝑡, 𝑡+1
𝑧𝑗𝑘𝑙 ∈ 𝑅+ (j=1,...,n; l=1,...,𝐿𝑘 ; k =1,...,K, t=1,...,T-1) is carry-over of 𝐷𝑀𝑈𝑗 , at division k, from period
t to period t+1, where 𝐿𝑘 is the number of items in the carry-over from division k.

MODEL DEVELOPMENT
This research presents measurement of efficiency of 38 container terminals ports of countries of the Asia-
Pacific Economic Cooperation (APEC) during the period from 2000 to 2013.

Figure 1. Dynamic Model APEC Ports 2000-2013.


Source: Own prepared based on Dynamic Structure Model (Tone & Tsutsui, 2010).
Selection criteria was ports APEC members that handled more than 1 million TEUs per year by 2013,
according to the report of the World Shipping Council.

57
In order to analyses the efficiency over the time, is implemented a dynamic model with variable returns to
scale output orientation. It is used as inputs: quay length, employees and cranes in the container terminal
and as output the number of TEUs (twenty-foot equivalent unit) moved annually. The carry over is the port
revenue so in the first period is an output and in the next period is an input (see figure 1).

RESULTS
The solution to our dynamic DEA model estimates of technical efficiency for each year 2000-2013, the
scores obtained (see table 1), showed that in general the ports were inefficient, with 52% in average, but
there were years with better developing as 2013 with an efficiency average of 62%.

Table 1. Dynamic Model APEC Ports 2000-2013.


Port 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
1 Bangkok, Thailand 0.25 0.32 0.74 0.87 0.69 0.59 0.87 0.99 0.99 0.94 0.87 0.99 0.93 0.74
2 Busan, South Korea 0.35 0.42 0.49 0.50 0.51 0.54 0.50 0.52 0.46 0.50 0.49 0.60 1.00 1.00
3 Callao, Peru 0.41 0.68 0.74 0.89 0.79 0.37 0.24 0.37 0.55 0.43 0.37 0.13 0.33 0.33
4 Dalian, China 0.09 0.18 0.32 0.20 0.24 0.28 0.23 0.04 0.23 0.26 0.25 0.40 1.00 1.00
5 Seattle, USA 0.74 0.83 0.68 0.54 0.63 0.90 0.96 0.74 0.62 0.97 1.00 1.00 0.88 0.93
6 Guangzho, China 0.54 0.47 0.34 0.30 0.28 0.26 0.19 0.19 0.15 0.12 0.13 0.22 0.31 0.30
7 Osaka, Japan 0.90 0.65 0.75 0.69 0.75 0.89 0.89 1.00 0.79 0.79 0.81 0.87 1.00 1.00
8 Ho Chi Minh, Vietna 0.20 0.48 0.66 0.75 0.85 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
9 Hong Kong, China 1.00 0.99 0.87 0.75 0.53 0.69 0.43 0.55 0.34 0.32 0.32 0.47 0.43 0.41
10 Tacoma, USA 0.13 0.09 0.12 0.25 0.37 0.29 0.22 0.27 0.24 0.21 0.22 0.25 0.28 0.26
11 Kaohsiung, Taiwan 0.01 0.09 0.12 0.09 0.18 0.16 0.16 0.13 0.13 0.15 0.15 0.26 0.25 0.26
12 Keelung, Taiwan 0.98 0.85 0.80 0.71 0.75 0.78 0.73 0.10 0.10 0.12 0.12 0.12 1.00 1.00
13 Tokyo, Japan 0.99 1.00 0.87 1.00 1.00 1.00 0.65 1.00 0.87 0.88 0.89 1.00 0.42 0.38
14 Laem Chabang, Thail 0.13 0.15 0.21 0.11 0.09 0.22 0.03 0.29 0.17 0.37 0.24 0.40 0.45 0.46
15 Lázaro Cárdenas, Mx 0.00 0.00 0.01 0.02 0.00 0.00 0.18 0.23 0.23 0.24 0.22 0.46 0.28 0.32
16 Lianyungung, China 0.11 0.13 0.21 0.35 0.28 0.33 0.34 0.34 0.29 0.29 0.29 0.27 0.32 0.31
17 Long Beach, U.S.A. 0.74 0.55 0.43 0.55 0.34 0.35 0.36 0.35 0.30 0.31 0.31 0.33 0.34 0.36
18 Los Angeles, U.S.A. 0.55 0.35 0.21 0.15 0.08 0.08 0.05 0.06 0.24 0.25 0.25 0.31 0.29 0.28
19 Manila, Philippines 0.56 0.21 0.15 0.10 0.16 0.17 0.20 0.31 0.25 0.25 0.26 0.27 0.31 0.30
20 Manzanillo, México 0.35 0.43 0.24 0.48 0.52 0.47 0.22 0.37 0.26 0.26 0.21 0.34 0.33 0.25
21 Melbourne, Australia 0.70 0.25 0.15 0.13 0.35 0.35 0.32 0.41 0.37 0.36 0.38 0.34 1.00 1.00
22 Nagoya, Japan 0.90 0.79 0.99 0.87 0.64 0.40 0.33 0.27 0.25 0.17 0.21 0.20 1.00 1.00
23 Ningbo, China 0.79 0.88 0.98 0.88 0.67 0.52 0.36 0.39 0.71 0.80 0.80 0.82 0.01 0.87
24 Okland, USA 0.55 0.62 0.74 0.96 1.00 1.00 0.18 0.29 1.00 1.00 1.00 0.76 0.55 0.57
25 Port Kelang, Malaysia 0.62 0.75 0.85 0.95 0.82 0.73 0.56 0.76 0.62 0.53 0.66 0.92 0.21 0.39
26 Qingdao, China 0.98 0.75 0.81 0.71 0.83 0.79 0.52 0.94 0.57 0.65 0.65 1.00 1.00 1.00
27 Kwangyang, Korea 0.96 0.87 0.74 0.65 1.00 1.00 0.33 0.28 1.00 1.00 1.00 1.00 1.00 1.00
28 Shanghai, China 0.59 0.55 0.62 0.85 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.74 0.72
29 Shenzhen, China 1.00 0.96 0.85 0.75 0.86 0.98 0.80 1.00 0.75 0.78 0.77 1.00 0.24 0.25
30 Sidney, Australia 0.13 0.10 0.11 0.10 0.18 0.23 0.12 0.14 0.14 0.16 0.15 0.24 1.00 1.00
31 Singapore 0.90 0.75 0.96 0.85 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.19 0.20
32 T. Pelepa, Malaysia 0.36 0.55 0.69 0.58 0.41 0.35 0.29 0.26 0.25 0.19 0.20 0.17 0.87 0.82
33 T. Perak, Indonesia 0.66 0.79 0.63 0.55 0.70 0.79 0.56 0.87 0.66 0.74 0.75 0.97 0.54 0.49
34 T. Priok, Indonesia 0.55 0.66 0.85 0.96 1.00 1.00 1.00 1.00 0.84 0.75 0.75 0.56 0.41 0.87
35 Tianjin, China 0.01 0.13 0.20 0.10 0.25 0.29 0.19 0.27 0.16 0.19 0.18 0.38 1.00 1.00
36 Vancouver, Canada 0.36 0.46 0.25 0.32 0.19 0.23 0.25 0.04 0.29 0.35 0.35 0.43 0.35 0.33
37 Xiamen, China 0.33 0.55 0.69 0.41 0.31 0.45 0.28 0.04 0.28 0.30 0.31 0.51 0.74 0.83
38 Yingkou, China 0.09 0.13 0.21 0.13 0.02 0.02 0.09 0.13 0.12 0.18 0.16 0.34 0.35 0.37

Source: Own calculations based on the results of DEA.

58
The year 2006 was the lowest level of efficiency with 43% and that was due the input -quay length- was
not used at full capacity by most of the ports and the number of containers moved and the gains that they
obtained in most ports were not the most optimal. The carry over play an important role in measuring the
efficiency because connect one term with the other, in this case was the port revenues and the slacks of this
variable showed that 27 ports don’t use in appropriately way the incomes, and this situation is reflected in
the low level of efficiency obtained global ports.
We show the graphic of mean efficiency for the dynamic model over the period, where it could appreciate
that the port of Shanghai got the highest score with 86% and in the opposite size is the port of Kaohsiung,
Taiwan with 15% of efficiency.

Shanghai,China
Tokio
Ho Chi Minh, Vietnam
Singapore, Singapore
Kwangyang
Osaka
Seattle
Qingdao, China
Tanjung Priok, Jakarta, Indonesia
Shenzhen, China
Bangkok
Okland
Tanjung Perak, Surabaya, Indonesia
Ningbo-Zhoushan, China
Port Kelang, Malaysia
Keelung
Hong Kong, China
Nagoya
Busan, South Korea
Callao
Melbourne
Xiamen, China
Tanjung Pelepas, Malaysia
Long Beach, U.S.A.
Dalian, China
Manzanillo, Mexico
Tianjin, China
Vancouver
Lianyungung, China
Sidney
Guangzhou Harbor, China
Manila
Laem Chabang, Thailand
Tacoma
Los Angeles, U.S.A.
Yingkou, China (yantian)
Lázaro Cárdenas
Kaohsiung, Taiwan, China
0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.0

Figure 2. Efficiency in Ports APEC in a Dynamic Model.


Source: Own calculations based on the results of DEA.

CONCLUSIONS
In this paper it was presented the measurement of efficiency using a dynamic model for the period 2000-
2013 of container terminals in the main ports of APEC. It was used variable return to scale model with
output orientation.
DMUs used were 38 container terminals ports of countries of the Asia-Pacific Economic Cooperation
(APEC). It is used as inputs: quay length, employees and cranes in the container terminal and as output the

59
number of TEUs (twenty-foot equivalent unit) moved annually. The carryover is the port revenue so in the
first period is an output and in the next period is an input.
The most efficient year was 2013 with an average score of 62% and the year who had the lowest value was
2006 with an average score of 43%. Shanghai was the most efficiency port with 86% average score and
Kaohsiung Port was the lowest with 15%. The low scores in port revenues and the slacks of this variable
showed that 27 ports don’t use in appropriately way the incomes, and this situation is reflected in the low
level of efficiency obtained global ports.
In the current international of APEC region where the volume of goods other waterway grows significantly,
is necessary implement strategies that lead to efficiency progress, because with this type of study can be
seen the need to develop port policies aimed at strengthening investment in infrastructure and labor policies
that contribute to a better training of employed personnel.

REFERENCES
[1] Asia-Pacific Economic Cooperation (APEC), (2014), “Key Trends and Developments relating to Trade and
Investment Measures and their Impact on the APEC Region” in APEC Policy Support Unit, No. APEC#214-SE-
01.6, Pp.27
[2] Färe R. and S. Grosskopf, (1996), “Productivity and Intermediate Products: a Frontier Approach,” Econ Lett. 50,
pp. 65-70
[3] Emrouznejad, A. and Thanassoulis, E. (2005) A mathematical model for dynamic efficiency using data
envelopment analysis. Applied Mathematics and Computation, 160, 363–378.
[4] Nemoto, J. and Goto, M. (2003) Measurement of dynamic efficiency in productions: an application of Data
Envelopment Analysis to Japanese electric utilities. Journal of Productivity Analysis, 19, 191-210
[5] Ouellette, P. and Yan, L. (2008) Investment and dynamic DEA. Journal of Productivity Analysis, 29, 235-247
[6] Sengupta, K. (1995) Dynamics of Data Envelopment Analysis: Theory of System Efficiency, Kluwer Academic
Publishers, Dordrecht, London.
[7] Sueyoshi and Sekitani K. (2005). Returns to scale in dynamic DEA, European Journal of Operational Research
161(2).
[8] Tone, Kaoru, & Tsutsui, Miki. (2010). Dynamic DEA: A slacks-based measure approach. Omega, 38, (3–4), 145–
156.
[9] World Shipping Council (2013), Top 50 World Container Ports, available in:
http://www.worldshipping.org/about-the-industry/global-trade/top-50-world-container-ports.

60
EFFICIENCY OF MUNICIPAL GOVERNMENTS IN SLOVAKIA
MIROSLAV HUŽVÁR
Department of Quantitative Methods and Information Systems, Faculty of Economics, Matej Bel University, Banská
Bystrica, Slovakia

ZUZANA RIGOVÁ
Department of Quantitative Methods and Information Systems, Faculty of Economics, Matej Bel University, Banská
Bystrica, Slovakia

KATARÍNA SÝKOROVÁ
Department of Public Economics and Regional Development, Faculty of Economics, Matej Bel University, Banská
Bystrica, Slovakia

ABSTRACT
Managing public property belongs to original functions and competencies of local governments. Effective and efficient
use of these resources is naturally expected by citizens. In our study we analyze the efficiency of municipal
governments in using their financial, tangible, and intangible resources. Empirical research is done on selected Slovak
municipalities. A triple of DEA models using slacks-based measure is applied to assess their efficiency from various
viewpoints. For this purpose, we consider non-current assets, operational costs, and staff costs as inputs. The income
and the population are considered as outputs which estimate the range of services provided by the municipality. The
results reveal considerable gaps in efficiency among municipalities and provide a basis for further research.

Keywords: Municipal government efficiency; Central Slovakia; Data Envelopment Analysis (DEA); Slacks-based
Measure (SBM)

INTRODUCTION
Self-government in Slovakia is performed at two independent levels. The regional level consists of 8
regions, the local level of 2 892 local governments in 138 cities and in 2 754 villages. A representative of
the local government is municipality which provides public administration on the local level close to
citizens with regard to their needs and demands. A municipality performs its own original functions and
competencies as well as the functions delegated from the state. One of the original responsibilities consists
in managing non-current municipal property which includes tangible, intangible, and financial assets.
Efficient administration of the resources is required by the public.
Generally, an economic efficiency of a local government is expressed as the relation between input
resources and their economic and social impacts. An expression of public administration efficiency is not
a simple task since public services are primarily based on the non-market principles and service production
inputs and outputs are mostly measured in different metric units. This is an ideal space for the application
of DEA methods which are suitable, particularly, for efficiency analysis of the non-profitable subjects.
Several empirical studies on the efficiency assessment of municipalities in different countries based on
parametric or non-parametric methods have been published. A nice survey of these studies is given by
Kutlar et al. (2012). The methodology applied in previous studies is briefly summarized in Table 1.

61
Table 1: Overview of empirical studies on municipal efficiency

Countries Belgium, Greece, Spain, Finland, Italy, Turkey


Methods for:
estimating efficiency DEA, SFA, FDH
explaining the gaps in OLS regression, Tobit regression, Nonparametric smoothing techniques, Fuzzy K-means
efficiency clustering approach
Decision variables: total current expenditures, financial expenditures, spendings per capita, expected budget
Inputs spending, number of blue and white coloured workers, number of full time workers, area
of buildings, personal, social security or capital expenses ...
Outputs total population, proportion of people older than 65, number of unemployed people, number
of people who live at the lowest life level, number of pupils, average area, length of spaces,
tourism and industrial areas, range of provided tourist, educational, social and cultural
services, ...
Explanatory variables Factors describing educational, income or unemployment level

In this study we analyze the efficiency of performance of municipal governments in managing their assets
and costs by using DEA. We apply designed approach to assess relative efficiencies within the group of
selected municipalities located in two regions of Central Slovakia. The objective is to reveal input excesses
for observed municipalities.

METHODS
In order to use DEA for the assessment of performance efficiency of municipalities, we consider their non-
current assets, operating costs, and staff costs as input indicators. Non-current assets are measured in euros
and include tangible, intangible, and financial assets. Operating costs cover material and energy
consumption, cost of repairs and maintenance, expenses related to financial assets and activities, and other
operating expenses. Staff costs include wages and salaries and social security expenses. The income along
with the population are considered as output indicators. Getting income is not the main goal of the
municipality, however, it provides additional financial resources to support its functions and activities. The
income of the municipality consists of operating income, income from financial assets, and revenues from
the sale of non-current tangible and intangible assets. Finally, we use the population of municipality to
estimate the total range of services provided to the citizens although this indicator is not under a
discretionary control of the local government.

Figure 1: NUTS-3 regions of Slovakia

62
We do an empirical research on a sample of 41 municipalities located in Central Slovakia, in Banská
Bystrica and Žilina NUTS-3 regions (see Figure 1). 24 of them are in Banská Bystrica region, and 17 in
Žilina region. All of the municipalities have a special urban statute (therefore, we refer to them as “city
municipalities”). Žilina is more industrial region, while Banská Bystrica is more agricultural region with a
significantly higher unemployment rate.
The inputs and outputs data related to the year 2013 (see Appendix) are taken from publicly available
financial statements of individual municipalities and from regional statistical yearbook. The descriptive
statistics in Table 2 show there are significant differences among the observed municipalities, namely in
their population and income per capita.

Table 2: Descriptive statistics of observed municipalities


Non-current Operating Costs per 1000 Staff Costs Income
Population
assets per capita € of Non-current assets per capita per capita
Min 1453 751.22 12.96 67.33 2.56
Max 81382 6009.54 109.29 201.80 842.47
Median 8007 1978.84 37.10 120.66 49.26

The correlations between inputs and outputs are indicated in Table 3. Surprisingly, the correlations between
the income and other variables are relatively low.

Table 3: Inputs and outputs correlations


Non-current assets Operating Costs Staff Costs Income Population
Non-current assets 1 0.94 0.74 0.35 0.85
Operating Costs 0.94 1 0.88 0.40 0.95
Staff Costs 0.74 0.88 1 0.41 0.94
Income 0.35 0.40 0.41 1 0.40
Population 0.85 0.95 0.94 0.40 1

To calculate efficiency scores, we apply SBM which reflects nonzero slacks in inputs and outputs when
they are present. Since we focus on inputs directly controlled by the municipality, we use input-oriented
model. Due to significant differences in size among observed municipalities, we assume variable returns to
scale (SBM-I-V).
We assess the efficiency of municipalities with three DEA models (see Figure 2). The income and the
population are outputs in all of them. Model 1 evaluates the efficiency in using available capital and labour
resources. Capital resources are represented by non-current assets and operational costs while labour
resources by the costs of municipality staff. Non-current assets are viewed as resources used in the
production of income and services for the population. Model 2 focuses on the efficiency of using tangible
and financial assets which are the main components of non-current assets (the share of intangible assets on
the total value of non-current assets is less than 1 % in all municipalities). Model 3 provides an assessment
of the efficiency of operating and labour costs since we exclude from our considerations non-current assets
as well as revenues from their sales.

63
Figure 2: Inputs and outputs of DEA models

RESULTS AND DISCUSSIONS


General results of designed DEA models are summarized in Table 4. One third of municipalities occur to
be efficient in using their total capital and labour resources as well as their tangible and financial non-
current assets (but only 9 municipalities are efficient by both Model 1 and Model 2). The most selective is
Model 3 which identifies only 5 efficient municipalities (all of them being efficient by previous two models,
too). But still, all three models reveal significant differences in efficiency among analysed municipalities.

Table 4: Results of DEA models


Model 1 Model 2 Model 3
No. of Efficient municipalities 14 14 5
No. of Inefficient municipalities 27 27 36
No. of Efficient municipalities (efficient by default1) 5 6 1
Average Score (All) 0.71 0.76 0.63
Average Score (Inefficient) 0.56 0.63 0.58
Max Score 1 1 1
Min Score 0.41 0.26 0.31
St Dev of Score 0.22 0.22 0.21
1
Municipalities are efficient by default when they are not dominated by another municipality, but also do not dominate
any other municipality (Geys and Moesen, 2009)

A deeper insight reveals that larger municipalities tend to be more efficient than smaller ones in all models
(e.g., 10 of 14 efficient municipalities by Model 1 have the population above the median). The average

64
efficiency scores as well as the share of efficient municipalities are higher in Žilina region, but the
differences between the regions are not significant.
Recommended projections for individual municipalities (Table 5) identify the sources of input excesses.
By Model 1, operating costs are the main reason of inefficiency, followed by non-current assets. The results
of Model 3 confirm that the influence of staff costs on inefficiency is less considerable. Also, the excesses
of tangible assets cause inefficiency in more municipalities than the excesses of financial assets.

Table 5: Projections for inefficient municipalities


Number of municipalities with:
Inputs maximum slack recommended reduction recommended reduction
in % by at least 50% by at least 25%
Non-current assets 10 7 22
Model 1 Operating costs 14 19 26
Staff costs 3 3 21
Tangible assets 24 15 24
Model 2
Financial assets 3 4 11
Operating Costs 28 20 29
Model 3
Staff costs 8 6 26

CONCLUSIONS
We analysed the efficiency of city municipalities in two regions of Central Slovakia. We focused on their
ability to efficiently use various assets to produce the income and services for the population. We applied
three DEA models to assess the efficiency from different viewpoints.
The analysis showed large gaps between efficient and inefficient municipalities. Operating costs were
identified as the main source of inefficiency while tangible assets were the most significant source of
inefficiency among non-current assets. Relevant differences between the two regions were not observed.
However, the research is still in progress and requires further study with respect to the volume of provided
services and their structure as well as to the age and social structure of the population. Identification of
determinants of efficiency gaps and of efficiency changes over time are also a challenge.

ACKNOWLEDGEMENT
The research is supported with the project Mobility - enhancing research, science and education at Matej
Bel University, ITMS code: 26110230082, co-financed by the European Social Fund.

APPENDIX: INPUTS AND OUTPUTS DATA


Inputs Outputs
Non-current Tangible Financial
Municipality Operating Staff Costs Income Populatio
assets (mil. assets in % assets in % of
Costs (mil. €) (mil. €) (mil. €) n
€) of NCA2 NCA2
Banská Bystrica 226.05 82.60 16.64 7.85 9.71 2.92 79583
Banská Štiavnica 59.17 96.36 3.64 0.77 1.05 0.38 10330
Brezno 25.42 82.61 17.32 1.51 2.14 0.87 21703
Bytča 12.74 83.58 16.42 1.39 1.84 0.13 11284
Čadca 18.62 71.69 28.25 1.74 1.88 1.57 24791
Detva 29.48 89.92 10.08 1.83 1.11 1.98 15047
Dolný Kubín 31.42 65.36 34.51 1.73 2.34 1.80 19472
Dudince 6.59 89.98 10.02 0.46 0.19 0.29 1453

65
Inputs Outputs
Non-current Tangible Financial
Municipality Operating Staff Costs Income Populatio
assets (mil. assets in % assets in % of
Costs (mil. €) (mil. €) (mil. €) n
€) of NCA2 NCA2
Fiľakovo 11.49 75.21 24.73 0.34 1.32 0.18 10799
Hnúšťa 19.23 83.36 16.63 0.64 0.97 0.21 7701
Hriňová 8.79 78.13 21.73 0.58 0.90 2.65 7793
Jelšava 16.56 95.89 4.11 0.42 0.40 0.18 3224
Kremnica 33.17 89.25 10.60 0.54 0.91 0.97 5542
Krupina 40.27 93.38 6.62 0.63 1.03 0.57 8007
Kysucké Nové Mesto 18.88 84.72 15.28 0.36 1.20 0.42 15584
Liptovský Hrádok 21.24 90.09 9.73 0.43 0.90 0.73 7625
Liptovský Mikuláš 60.66 81.68 18.31 1.28 2.34 1.57 31873
Lučenec 55.55 80.65 19.35 3.31 2.89 1.73 28413
Martin 92.58 82.11 17.81 3.73 4.35 4.82 57023
Modrý Kameň 2.92 90.41 9.59 0.10 0.12 0.03 1568
Námestovo 18.51 91.80 7.86 0.46 1.01 0.47 7933
Nová Baňa 16.39 88.34 11.43 0.23 0.55 0.34 7542
Poltár 10.79 81.40 18.53 0.50 0.70 0.37 5811
Rajec 11.60 89.29 10.67 0.55 0.89 0.39 5864
Rajecké Teplice 5.78 91.12 8.39 0.30 0.59 0.14 2909
Revúca 21.15 86.70 13.16 1.24 1.89 0.29 12766
Rimavská Sobota 26.25 73.55 26.38 1.80 1.65 1.32 24454
Ružomberok 72.10 85.45 14.40 2.68 3.52 23.71 28145
Sliač 7.20 84.93 15.06 0.50 0.66 0.24 5056
Tisovec 12.73 91.51 8.49 0.32 0.56 0.21 4270
Tornaľa 18.26 91.59 8.40 0.70 0.90 0.11 7450
Trstená 21.30 94.96 4.82 0.46 1.01 0.02 7499
Turčianske Teplice 12.06 83.02 16.98 0.74 0.82 0.28 6598
Turzovka 10.23 81.14 18.48 0.36 0.54 0.17 7733
Tvrdošín 21.87 93.96 5.99 0.65 1.23 0.47 9361
Veľký Krtíš 10.76 71.14 25.35 0.77 0.92 0.35 12756
Vrútky 10.51 84.29 15.03 0.48 1.42 0.10 7645
Zvolen 123.20 91.53 8.44 3.56 3.59 1.13 43148
Žarnovica 12.84 78.56 21.36 0.59 0.73 0.15 6451
Žiar nad Hronom 41.37 89.85 10.04 1.36 1.96 0.91 19789
Žilina 489.07 59.87 40.05 10.56 5.53 7.13 81382
2
Non-current assets

REFERENCES
[1] Geys B., Moesen W. (2009) Measuring Local Government Technical (In) efficiency: An Application and
Comparison of FDH, DEA and Econometric Approaches, Public Performance & Management Review 32 (4):
499-513.
[2] Kutlar A., Bakirci F., Yüksel F. (2012) An analysis on the economic effectiveness of municipalities in Turkey,
African Journal of Marketing Management 4 (3): 80-98.
[3] Financial statements of Slovak municipalities for year 2013 (http://www.registeruz.sk/cruz-public/home/)
[4] Regional statistical yearbook of Slovakia 2013 (http://slovak.statistics.sk)

66
ESTIMATING RIGHT AND LEFT RETURNS TO SCALES IN THE PRESENCE
OF UNDESIRABLE FACTORS: A DEA APPROACH
MOHAMMAD KHOVEYNI
Department of Applied Mathematics, College of Basic Sciences, Yadegar-e-Imam Khomeini (RAH) Shahr-e-Rey
Branch, Islamic Azad University, Tehran, Iran (corresponding Author)

ROBABEH ESLAMI
Department of Mathematics, Faculty of Technology and Engineering, South Tehran Branch, Islamic Azad University,
Tehran, Iran

ABSTRACT
This paper distinguishes right and left returns to scale (RTS) of (strongly) efficient decision making units (DMUs)
with desirable (good) and undesirable (bad) inputs and outputs by introducing a new DEA (data envelopment
analysis) method. Hereafter in DEA literature, various methods have been presented for estimating RTS that all of
them are capable of determining RTS only for desirable data, whereas in the real world, both desirable and
undesirable data may be present. Hence, in current research, a new input-output oriented linear model is first
proposed to identify (strongly) efficient DMUs in the presence of undesirable factors and second, right and left RTS
of the (strongly) efficient DMUs are estimated by introducing two new non-radial models. One of the advantages of
our proposed method is that it is capable of determining (strongly) efficient DMUs in the presence of undesirable
data. Likewise, the second one is that the proposed method is capable of estimating right and left RTS of the (strongly)
efficient DMUs in the presence of undesirable inputs and outputs whiles in this situation, the existing RTS methods
are incapable of identifying RTS, truly. Finally, an empirical application is presented, and also some future works are
suggested in this context.

Keywords: Data envelopment analysis (DEA); Right and left returns to scale (RTS); Undesirable factors; Strong (free)
disposability; Weak disposability

INTRODUCTION
One of the non-parametric technique based upon mathematical programming is data envelopment analysis
(DEA) which was first introduced by Charnes, Cooper, and Rhodes (CCR) (1978). The DEA uses to
evaluate the efficiency of decision making units (DMUs) with multiple inputs and outputs. In addition,
DEA can be used to improve the performance of inefficient DMUs via decreasing their inputs and
increasing their outputs. But in the real world, there may exist DMUs with desirable (good) and undesirable
(bad) data. For example, smoke is one of the undesirable outputs in factories which makes to decrease their
efficiencies. Besides, registered students of below average intelligence and lazy staff can be considered as
undesirable inputs at universities that makes to reduce the efficiency of universities. Hence, in order to
increase the efficiency of DMUs, undesirable inputs and outputs should be respectively increased and
decreased, likewise desirable inputs and outputs should be decreased and increased, respectively. But in the
standard DEA models, it is not allowed to increase inputs and decrease outputs, and all inputs and outputs
are only decreased and increased, respectively, that it is not suitable for improving the performance of
DMUs.
Returns to scale (RTS) is one of the economic topics which has been widely studied within the framework
of DEA. Hitherto, there are several RTS approaches in DEA literature (such as: Banker and Thrall, 1992;
Golany and Yu, 1997; Eslami and Khoveyni, 2013). Note that, the main drawback of the existing RTS DEA

67
approaches is that in these approaches, all inputs and outputs of DMUs must be as desirable factors whiles
in the real world, the inputs and outputs are not necessarily desirable and undesirable inputs and/or outputs
can be appeared in many applications. For example, the amount of waste in a waste treatment process, is
an undesirable input and also, smoke is an undesirable output in a factory (Seiford and Zhu, 2002). This
fact is another motivation for creating this current research. Hence, in order to modify this drawback, we
introduce two new non-radial DEA models for estimating right and left returns to scale of the (strongly)
efficient DMUs in the presence of undesirable factors.
It is worth stressing that, one of the advantages of our proposed method is that it is capable of determining
(strongly) efficient DMUs in the presence of undesirable (bad) inputs and outputs. The second one is that
the proposed method is able to estimate right and left RTS of the (strongly) efficient DMUs in the presence
of undesirable data, whereas the existing RTS methods cannot estimate returns to scale in the presence of
undesirable factors, truly. Hence, our proposed method is more careful than the other methods in this
context.

THE PROPOSED APPROACH

Suppose that there are n DMUs as X T


j 
, Y jT   X jg , X bj , Y jg , Y jb 
T T T T

 m s
( j  1, 2, , n) so that

   
T T
X g
j  x 1gj , , x ijg , , x mg 1 j  m1
and X b
j  x 1bj , , x ijb , , x mb 2 j  m2
respectively represent

 
T
desirable and undesirable input vectors of DMU j , and also Y j
g
 y 1gj , , y rjg , , y sg1 j  s1
and

 
T
Y j
b
 y 1bj , , y rjb , , y sb2 j  s2
indicate desirable and undesirable output vectors of DMU j ,

X g
 Y jg 
respectively. Since X  j

  0 and Y   b   0 , hence m  m1  m 2 and s  s1  s 2 . Now,
j X b j
 j  Y j 
the following production possibility set (PPS) is obtained by assuming variable RTS:

 n n
 n

  X j
g
j  X ,   j X  X ,   jY jg  Y g , 
g b
j
b



T


PPS   X g , X b , Y g , Y b
T T T

 j 1
n n
j 1 j 1 
. (1)
  jY j  Y ,   j  1 ,  j  0 ; j  1, 2, , n 
b b
 
 j 1 j 1 

Identifying (strongly) efficient DMUs in the presence of undesirable data

In this subsection, we present an input-output oriented DEA model to determine the efficiency status of a
DMU under evaluation DMU p ; p  1, 2,  , n  in the presence of undesirable input and output
factors as below:

68
Max hp   pg   pg   pb   bp
n
s.t.  x
j 1
g
j ij   pg xipg , i  1, , m1 ,

 x
j 1
b
j ij   pb xipb , i  1, , m2 ,

 y
j 1
j
g
rj   pg yrpg , r  1, , s1 , (2)

 y
j 1
j
b
rj   bp yrpb , r  1, , s2 ,


j 1
j  1,  j  0, j  1, , n,

 pg  1,  pb  1,  pg  1,  bp  1.

Definition 1. DMU p is called (strongly) efficient under model (2) if and only if the two following
conditions are satisfied:
   
(i)  pg   pb   pg   bp  1 ,
(ii) All slacks are zero on optimality.
Otherwise, if only Condition (i) is satisfied, then DMU p is called (weakly) efficient. Also, if none of the
conditions are satisfied, then DMU p is called inefficient.

Estimating right and left returns to scale of (strongly) efficient DMUs in the presence of undesirable data

Suppose that DMU p is (strongly) efficient. We present the following non-radial DEA model for
estimating right returns to scale of DMU p in the presence of undesirable inputs and outputs:

69
m1 s1 m2 s2
z p   si   sr   si   sr
g g b b
Max
i 1 r 1 i 1 r 1
n

 x  si   xipg , i  1,
g
g
s.t. j ij , m1 ,
j 1
n

 x  si   xipb , i  1,
b
b
j ij , m2 ,
j 1
n

 y  sr   yrpg , r  1,
g
g
j rj , s1 ,
j 1
(3)
n

 y  sr   yrpb , r  1,
b
b
j rj , s2 ,
j 1
n


j 1
j  1,  j  0, j  1, , n,   1,

si  0, i  1, sr  0, j  1,
g g
, m1 , , s1 ,
si  0, i  1, sr  0, j  1,
b b
, m2 , , s2 .

Theorem 1. Assume that DMU p  X pg , X bp , Ypg , Ypb  T T T T

 is (strongly) efficient. Then:


i. Increasing right RTS prevail at DMU p if and only if the model (3) is feasible and the optimal
value of its objective function is non-zero.

ii. Decreasing right RTS prevail at DMU p if and only if the model (3) is infeasible.

iii. Constant right RTS prevail at DMU p if and only if the model (3) is feasible and the optimal value
of its objective function is zero.
Proof. The proof is clear. Q.E.D.

Assume that DMU p is a (strongly) efficient DMU. To estimate left returns to scale of DMU p in the
presence of undesirable inputs and outputs, we present a non-radial DEA model as follows:

70
m1 s1 m2 s2
z p   si   sr   si   sr
g g b b
Max
i 1 r 1 i 1 r 1
n

 x  si   xipg , i  1,
g
g
s.t. j ij , m1 ,
j 1
n

 x  si   xipb , i  1,
b
b
j ij , m2 ,
j 1
n

 y  sr   yrpg , r  1,
g
g
j rj , s1 ,
j 1
(4)
n

 y  sr   yrpb , r  1,
b
b
j rj , s2 ,
j 1
n


j 1
j  1,  j  0, j  1, , n, 0    1,

si  0, i  1, sr  0, j  1,
g g
, m1 , , s1 ,
si  0, i  1, sr  0, j  1,
b b
, m2 , , s2 .

Theorem 2. Assume that DMU p  X pg , X bp , Ypg , Ypb  T T T T

 is (strongly) efficient. Then:


i. Decreasing left RTS prevail at DMU p if and only if the model (4) is feasible and the optimal
value of its objective function is non-zero.

ii. Increasing left RTS prevail at DMU p if and only if the model (4) is infeasible.

iii. Constant left RTS prevail at DMU p if and only if the model (4) is feasible and the optimal value
of its objective function is zero.

Proof. The proof is clear. Q.E.D.

Table 1: The obtained results of the proposed approach


DMU Efficiency status The right RTS The left RTS
D1 (Strongly) efficient IRS IRS
D2 (Strongly) efficient IRS DRS
D3 (Strongly) efficient DRS IRS
D4 (Strongly) efficient IRS IRS
D5 (Strongly) efficient IRS IRS
D6 (Strongly) efficient DRS IRS
D7 Inefficient ---- ----
D8 (Strongly) efficient DRS IRS
D9 Inefficient ---- ----
D 10 (Strongly) efficient DRS IRS

71
EMPIRICAL APPLICATION
We now apply the proposed approach about 10 universities with six desirable inputs, two undesirable
inputs, three desirable outputs, and two undesirable outputs. Note that, the data of inputs and outputs have
not been shown for the sake of their voluminous. The obtained results of the proposed approach have been
presented in Table 1.

CONCLUSIONS
Due to exist undesirable data in the real world, this article presents some new models and theorems to
distinguish (strongly) efficient DMUs and estimate right and left returns to scale of them in the presence of
undesirable data.
Moreover, there are different RTS DEA methods to estimate RTS of DMUs with desirable inputs and
outputs. But since in the real world, both desirable and undesirable factors may be present, hence the
existing RTS methods cannot identify the RTS of DMUs in the presence of undesirable factors, truly,
whereas the proposed method is correctly capable of estimating right and left returns to scale of DMUs with
desirable and undesirable data which is an advantage of the proposed method. The second one is that our
proposed approach is capable of identifying (strongly) efficient DMUs in the presence of undesirable inputs
and outputs. Consequently, this research opens up a new DEA approach to determine the (strongly) efficient
DMUs and estimate the right and left returns to scale of them in the presence of undesirable input and
output factors.
Note that this study can be developed for imprecise desirable and undesirable data, similarly. Likewise, the
values of the right and left returns to scale in the presence of undesirable data can be measured like Eslami
and Khoveyni’s RTS approach (Eslami and Khoveyni, 2013) and also, measuring the values of right and
left RTS can be extended for imprecise desirable and undesirable data, similarly.

REFERENCES
[1] Banker R.D., Thrall R.M. (1992) Estimating of returns to scale using data envelopment analysis, European
Journal of Operational Research 62 (1): 74-84.
[2] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of DMUs, European
Journal of Operational Research 2 (6): 429-444.
[3] Eslami R., Khoveyni M. (2013) Right and left returns to scales in data envelopment analysis: Determining type
and measuring value, Computers & Industrial Engineering 65: 500-508.
[4] Golany B., Yu G. (1997) Estimating returns to scale in DEA, European Journal of Operational Research 103 (1):
28-37.
[5] Seiford L.M., Zhu J. (2002) Modeling undesirable factors in efficiency evaluation, European
Journal of Operational Research 142: 16-20.

72
EVALUATION OF EFFICIENCY AND EFFECTIVENESS OF PUBLIC
TRANSPORT SECTOR OF INDIA USING DEA APPROACH
SHIVI AGARWAL
Department of Mathematics, BITS, Pilani, Rajasthan-333031, India

ABSTRACT
Productivity analysis is a major tool for policy making and managerial control to assess the degree of the utilization
of inputs to obtain the desired outputs. The efficiency and effectiveness are two important issues related to the
productive analysis of any public sector so this study makes an attempt to provide an overview of the general status
of the public transport sector of India in terms of their productive analysis. The paper evaluates the efficiency and
effectiveness of public transport sector of India using Data Envelopment Analysis (DEA) approach. Data have been
collected for 34 State Transport Undertakings (STUs) for the year 2012-2013. This study uses three different DEA
models with the same set of inputs but with different outputs. Fleet size, Total staff and Fuel consumption are
considered as inputs for all DEA models. Bus Utilization (BU) is considered as output for efficiency model, whereas
in effectiveness model, Passenger kilometers (PassKm) is using as a measure of output. In order to examine the
relationship between efficiency and effectiveness, third DEA model is used with both BU and PassKm as output
variables. The results reveal that efficiency and effectiveness are positively related.

Keywords: Data Envelopment Analysis; Efficiency; Effectiveness; Public Transport Sector; STUs; Transport.

INTRODUCTION
Transport sector plays a significant role in the overall development of a nation’s economy. It is a fact that
transport is one of the important barometer of economic activity. Road transport is the prime motorized
mode of transport linking the remote and hilly areas with rest of the country. Road transport system co-
exists in the public and private sector in India. Although, the private sector has an effective participation in
the passenger mobility, but its operational activities are very disaggregated and unorganized while the
operational activities of public transport sector are well regulated and organized. The State Transport
Undertakings (STUs), controlled by the respective state government, are the imperative mode of passenger
mobility in public road transport sector. The STUs have been undergoing important changes over the past
few decades, mostly due to the increasing competition, technology change and noticeable decline in market
share. In such an environment, it is important to know which STUs are performing efficiently and
effectively, by assessing their productivity growth regularly.
Since STUs are public utility service with a social objective, it is essential to regularly monitor their
performance, specifically with a view to identifying appropriate measures including proper investment and
pricing policy and to improve their output efficiency and effectiveness. In public transport sector, efficiency
and effectiveness measurements are the first step in the evaluation of individual performance of STUs. This
study is an attempt in this direction to assess the relative technical efficiency and effectiveness of STUs in
India.
Passenger road transportation is a “service business” and evaluating the performance of a service business
is a complex matter. Transport performance is often more difficult to evaluate than manufacturing business
performance, because it is difficult to determine the efficient amount of resources required to produce
various service outputs as well as effective utilization of the resources. The manufacturing standard can be

73
used to identify operating inefficiencies through classical cost accounting variance analyses. However, in
service organization like road passenger transportation system, it is difficult to identify the specific
resources required to provide a specific service output.
DEA has been successfully used by researchers to measure the efficiency of the road transport service sector
of different countries. (Chu et.al., 1992 ; Cowie and Asenova, 1999; Husain et. al., 2000; Odeck and Alkadi,
2001; Karlaftis, 2004; Odeck, 2006,; Sheth et. al., 2007; Sun et al., 2010; Barnum et. al. 2007, 2008; Lin
and Lan, 2009; Balezentis and Balezentis, 2011 etc.) By a critical examination of the available literature,
it is found that DEA-based studies dealing with the relative efficiency of public transport sector, especially
in India are extremely limited. Ramanathan (1999) assessed the productivity of 29 State Transport
Undertakings (STUs) of India, using data for the year 1993-1994, by applying DEA technique. Anjaneyulu
et al. (2006) have applied DEA on 44 STUs to measure the efficiency. Agarwal et. al. (2006) examined the
technical, pure technical and scale efficiencies of the regions of Uttar Pradesh State Road Transport
Corporation (UPSRTC) for the year 2002-2003 using DEA. Bishnoi and Sujata (2007) have measured the
technical efficiency of 20 depots of Haryana SRTC for the year 2006-2007. Nagadevara and Ramanayya
(2008) used DEA to identify the inefficient depots at the district level in Andhra Pradesh SRTC. Agarwal
et. al. (2009) examined total factor productivity (TFP) growth of 34 SRTUs using DEA based Malmquist
productivity index for the period 1989-1990 to 2000-2007. Bagavath (2009) determined the technical
efficiency of 44 STUs for the year 2000-2001 using DEA. Agarwal (2009) analysed the trends of technical
and scale efficiencies of 29 STUs during the period from 2004-05 to 2007-08 using new slack DEA model
with categorical decision making units (DMUs). Nagadevara and Ramanayya (2010) have applied DEA to
analyse inter-temporal variations in efficiency of 25 depots of Karnatka SRTC over the period 2004-05 to
2008-09. Three DEA models are used by Kumar (2011) to compute the efficiencies of 31 SRTUs for the
year 2006-2007.
The purpose of this article is to evaluate the performance of STUs by providing them with a mathematical
technique to analyse the efficiency and effectiveness with which service is rendered. The paper attempts to
estimate technical efficiency, effectiveness and combined consequence of efficiency and effectiveness of
the STUs. The paper is organized as follows: in section 2 methodology is given. Empirical Results and
discussions are given in section 3, followed by conclusions in the last.

METHODOLOGY
This paper evaluates the technical efficiency, effectiveness and the combined consequence of efficiency
and effectiveness of the STUs. The technical efficiency refers to the extent to which a STU can produce
maximum output from its chosen combination of factor inputs and effectiveness refers to degree of the
utilization of existing resources to obtain a service output.
Since public road transport is “service business” and the mathematical relationship between inputs and
outputs is not known clearly, STU performance is operationalized using Data Envelopment Analysis
(DEA). It is a non-parametric linear programming model that estimates the magnitude of departure from
efficiency frontiers for each STU. The DEA is initially proposed by Charnes, Cooper and Rhodes (Charnes
et. al (1978). DEA measures the relative technical efficiency of a group of decision-making units (DMUs)

74
by simultaneously evaluating multiple inputs and outputs common to each unit; each DMU is thus assigned
an efficiency score.
DEA is chosen over other methods because

 It handles multiple inputs and multiple outputs;

 It does not require a prior weights (as in index numbers);

 It emphasizes individual observations rather than statistical estimates (as in regression analysis);

 It is a dynamic analytical decision-making tool that not only provides a “snapshot” of the current
efficiency of the DMU compared with the group, but also indicates possibilities for improving relative
efficiency;

 It uses benchmarking approach to measure STU efficiency relative to others in their group.

 It can assist in identifying best-practice or efficient STUs and inefficient STUs within the group.

 The DEA results can allow policy makers to develop policies that can assist the relatively inefficient
STUs to improve their performance.
However, the radial CCR and BCC models suffer from one shortcoming, i.e. neglect of the slacks. To
overcome this shortcoming, performance can be evaluated using the new slack DEA model [Agarwal et.
al. (2011)].

ALGORITHM
First Step: Selection of the Homogeneous DMUs
We measure the OTE of 34 STUs using data from CIRT [4] for the year 2012-2013. A list of these 34
selected STUs is given in the Appendix A.1.
Second Step: Selection of Input and Output Variables
To evaluate the efficiency and effectiveness of the STUs, three inputs, viz., Fleet size (FS), Total Staff (TS),
and Fuel consumption (FC) are considered. Bus Utilization (BU) is considered as output for efficiency
model, whereas in effectiveness model, Passenger kilometers (PassKm) is using as a measure of output. In
order to examine the relationship between efficiency and effectiveness, both BU and PassKm are used as
output variables in third DEA model.
Inputs
1. Fleet Size (number of buses in hundred) comprises the average number of buses on road in a STU; it is
representative of the capital input.
2. Total Staff (numbers in thousand) refers to the total number of employees worked in a STU; it is
representative of the labour input.
3. Fuel Consumption refers to the fuel consumed (in ten thousand kilolitres) which is measured by dividing
total earned kilometer by fuel average; it is representative of the material input.

75
Outputs
1. Bus Utilisation (in kilometers) is defined as kilometers done per bus on road per day. It is calculated from
dividing total effective kilometers done on a day by total buses on road on that day.
2. Passenger-kilometers (in Billions) is a measure of service utilization which represents the cumulative
sum of the distances ridden by each passenger. It is normally calculated by summation of the passenger
load times the distance between individual bus stops.
The statistics of the observed data for the selected STUs of the input and output variables are shown in
Table 1. There is a perceptible variation in the inputs and the outputs across STUs. The all inputs used are
in some cases hundred times larger than that used by other STU. The variation in BU is not so high while
high variation is found in Pass-Kms. The table also shows that Kurtosis of all inputs and PassKm are
greater than i.e., The distribution of these variables is leptokurtic and the central peaks are sharper and
higher whereas tails are fatter and longer as compare to normal distribution. The results of skewness reveal
that the distribution of all inputs and PassKm are highly positive skewed. The results also show that BU
has negative Kurtosis which means that it has flatter distribution and the skewness of BU is also less than
0.5 which represents that the distribution of BU is approximately symmetrical.

Table 1: Descriptive Statistics of the STUs in India (2012-2013)


Max Min Mean Median Kurtosis Skewness SD
FS 224.02 0.26 34.85 27.045 8.33 2.66 47.05
TS 1222.87 2.88 206.86 161.45 7.74 2.58 264.47
Fuel 57.83 0.022 9.21 7.58 8.19 2.6 12.08
BU 536.6 51.6 254.68 246.9 -0.93 0.24 136.21
PassKm 10171.68 1.2 1655.88 1503.93 8.35 2.43 2043.91

Third Step: Selection of the model


In this study, new slack DEA (NSM) model (Agarwal et.al. (2011)) has been employed. Since the set of
sample STUs having three categories as Rural, Hill and Urban. So, we also combine the categorical DEA
model in the NSM-DEA model. The NSM model is given in Appendix A.2.
Fourth Step: Selecting the category of the STUs
The set of the sample STUs can be classified into three categories as follows. Category 1 consists of STUs
operated in the rural areas, category 2 in the Hill areas and category 3 in the urban areas. STUs in Category
1 are in the most advantageous situation while STUs in category 2 are in severe situation. So, we evaluate
the efficiency and effectiveness of STUs in category 2 only within the category while STUs in category 3
are evaluated with reference to category 2 and 3 and STUs in category 1 are evaluated with reference to all
STUs [Banker and Morey (1986)]. The categorization of the STUs is shown in Appendix A.1.
Fifth Step: Calculating the technical efficiency of STUs
Calculate OTE of 34 sample STUs with three inputs (FS, TS and Fuel) and single output (BU). The detailed
information of DEA results for efficiency is given in Table 2.

76
Sixth Step: Measuring the effectiveness of STUs
Measure the effectiveness of 34 STUs by considering three inputs (FS, TS and Fuel) and single output
(PassKm). The results are presented in Table 3.
Seventh Step: Estimating the combined consequence of efficiency and effectiveness of STUs
In order to examine the relationship between efficiency and effectiveness, both BU and PassKm are
considered as output variables with the same three inputs (FS, TS and Fuel). The results are given in Table
4.

Table 2: Efficiency scores of STUs by NSM DEA Model with Categorical DMUs
Efficiency Peer Slacks
STU No. Reference Set
Scores Weight FS TS Fuel BU
S1 0.01 S22 424.61 113.62 0 38.85 0
S2 0.010 S22 347.81 71.71 0 25.38 0
S3 0.030 S22 140.17 30.49 0 12.55 0
S4 0.042 S22 94.86 61.68 0 17.27 0
S5 0.060 S22 68.83 22.86 0 8.90 0
S6 0.029 S22 103.17 20.90 0 8.58 0
S7 0.030 S22 125.86 39.12 0 13.99 0
S8 0.098 S22 51.16 10.63 0 4.94 0
S9 0.048 S22 77.30 22.00 0 6.90 0
S10 0.054 S22 65.87 17.61 0 5.91 0
S11 0.068 S22 76.76 14.92 0 7.57 0
S12 0.070 S22 78.57 11.94 0 7.19 0
S13 0.075 S22 60.95 14.31 0 6.11 0
S14 0.055 S22 64.08 19.65 0 5.73 0
S15 0.113 S22 45.40 9.27 0 4.78 0
S16 0.242 S22 24.31 3.00 0 2.95 0
S17 0.148 S22 11.84 1.30 0 0.48 0
S18 0.184 S22 4.18 0.20 0 0.05 0
S19 0.287 S22 7.79 1.36 0 0.52 0
S20 0.280 S22 6.46 1.08 0 0.28 0
S21 0.715 S22 2.74 2.23 0 0.57 0
S22 1.00 S22 1.00 0 0 0 0
S23 0.869 S22 0.70 0.17 2.77 0 0
S24 0.519 S22 0.43 0.04 0 0 0
S25 0.958 S22 0.48 0.18 5.11 0 0
S26 0.016 S22 127.15 4.93 0 3.56 0
S27 0.015 S22 132.30 13.65 0 6.62 0
S28 0.037 S22 81.66 9.27 0 4.28 0
S29 0.020 S22 119.02 26.52 0 6.76 0
S30 0.073 S22 14.62 0 12.76 0.22 0
S31 0.048 S22 37.77 2.72 0 1.49 0
S32 0.077 S22 19.93 2.32 0 0.82 0
S33 0.337 S22 6.70 2.50 0 0.73 0
S34 0.186 S22 8.04 0.04 0 0.27 0
Mean 0.200 113.62 0 38.85 0
Max 1.00
Min 0.01
SD 0.28

EMPIRICAL RESULTS AND DISCUSSIONS


The efficiency, effectiveness and combined consequence of efficiency and effectiveness of 34 STUs have
been estimated for the year 2012-2013. Table 2 presents the information of efficiency scores obtained from

77
NSM model along with reference set and peer weights of the sample STUs. The DEA analysis evaluates
the set of STUs which construct the production frontier. The STUs having value of the efficiency score
equal to 1.00 are form the efficient frontier and those having the value less than 1.00 are less efficient
relative to the STUs on the frontier.

Table 3: Effectiveness of STUs by NSM DEA Model with Categorical DMUs


Slacks
Reference
STU No. Effectiveness Peer Weight FS TS Fuel PassKm
Set
S1 0.648 S8, S12 0.056, 5.368 48.934 0 0 0
S2 0.478 S12 3.825 38.315 136.063 0 0
S3 0.693 S12 1.759 10.010 5.719 0 0
S4 0.902 S8 1.854 41.973 0 8.116 0
S5 0.771 S8 1.345 8.559 0 2.259 0
S6 0.689 S12 1.233 7.800 18.058 0 0
S7 0.755 S8 2.460 12.971 0 1.846 0
S8 1.00 S8 1.000 0 0 0 0
S9 0.608 S12 0.968 10.759 3.526 0 0
S10 0.691 S12 0.828 7.936 2.323 0 0
S11 0.935 S12 0.977 3.258 0 0.547 0
S12 1.000 S12 1.000 0 0 0 0
S13 0.973 S12 0.776 5.048 0 0.538 0
S14 0.655 S12 0.804 10.295 2.685 0 0
S15 0.862 S12 0.578 2.367 0 0.624 0
S16 0.783 S12 0.288 0 4.866 0.961 0
S17 0.538 S12 0.094 1.330 12.776 0 0
S18 0.508 S12 0.022 0.585 7.112 0 0
S19 0.622 S12 0.081 0.777 4.172 0 0
S20 0.422 S12 0.053 1.034 6.541 0 0
S21 0.818 S8 0.054 1.657 0 0.311 0
S22 0.953 S23 0.600 0.050 0 0.026 0
S23 1.000 S23 1.000 0 0 0 0
S24 0.683 S23 1.327 0.006 1.631 0 0
S25 0.412 S23 0.686 0.070 3.209 0 0
S26 0.502 S28 1.165 2.462 92.236 0 0
S27 0.606 S28 1.575 0 10.509 0.035 0
S28 1.000 S28 1.000 0 0 0 0
S29 0.744 S28 1.458 13.018 0 0.518 0
S30 0.536 S28 0.111 0.429 28.857 0 0
S31 0.508 S28 0.401 0.309 14.475 0 0
S32 0.457 S28 0.215 0.938 6.786 0 0
S33 0.915 S28 0.082 1.736 0 0.380 0
S34 0.473 S28 0.070 0 6.735 0.078 0
Mean 0.710 6.926 10.832 0.485 0
Max 1.00
Min 0.412
SD 0.191

Technical Efficiency (TE)


Table 2 evinces that out of 34 STUs, only 1 STU ( S22) is relatively technical efficient (efficiency score
=1) and thus forms the efficient frontier. The remaining 33 STUs are relatively less efficient as they have
efficiency score <1. This STUs are on the best-practice frontier and thus form the “reference set” i.e.,
MEGTC (S22) can set an example of good operating practice for all other 33 inefficient STUs to emulate.
The lower the TE-score for a STU, the higher the scope for it to reduce inputs and/or increase outputs

78
relative to the best practice STUs in the reference set. The average of technical efficiency scores works out
to be 0.20, which reveals that an average STU can reduce its resources or increase outputs by 80% to
become efficient. The results also show that an average STU has to reduce its fleet size by 1136 and fuel
by 38.85 thousand liters to become efficient. APSRTC (S1) is the most technical inefficient STU. It has the
highest value of slack in FS, i.e., APSRTC is not utilizing its fleet efficiently to maximize the bus utilization.
Among the inefficient STUs, 8 STUs have the efficiency scores above the average efficiency scores.

Table 4: Combined consequence of efficiency and effectiveness of STUs by NSM DEA Model with Categorical
DMUs
STU Reference
Effectiveness Peer Weight FS TS Fuel PassKm BU
No. Set
S1 0.042 S22 0.043 0.0114 0 0.0039 0 3.4686
S2 0.041 S22 0.035 0.0072 0 0.0025 0 2.7596
S3 0.099 S22 0.014 0.003 0 0.0013 0 1.0448
S4 0.142 S22 0.010 0.0062 0 0.0017 0 0.7235
S5 0.161 S22 0.007 0.0023 0 0.0009 0 0.4656
S6 0.096 S22 0.010 0.0021 0 0.0009 0 0.7684
S7 0.104 S22 0.013 0.0039 0 0.0014 0 0.958
S8 0.224 S22 0.005 0.0011 0 0.0005 0 0.3122
S9 0.009 S22 0.008 0.0022 0 0.0007 0.7684 0
S10 0.011 S22 0.007 0.0018 0 0.0006 0.6514 0
S11 0.145 S22 0.008 0.0015 0 0.0008 0 0.4372
S12 0.183 S22 0.008 0.0012 0 0.0007 0 0.5225
S13 0.215 S22 0.006 0.0014 0 0.0006 0 0.4302
S14 0.160 S22 0.006 0.002 0 0.0006 0 0.457
S15 0.228 S22 0.005 0.0009 0 0.0005 0 0.25
S16 0.422 S22 0.002 0.0003 0 0.0003 0 0.1147
S17 0.341 S22 0.001 0.0001 0 0 0 0.0729
S18 0.297 S22 0.0004 0 0 0 0 0.0172
S19 0.137 S22 0.001 0.0001 0 0.0001 0.0694 0
S20 0.267 S22 0.001 0.0001 0 0 0.0168 0
S21 0.512 S22 0.0003 0.0002 0 0.0001 0.0217 0
S22 1.000 S22 1.000 0 0 0 0 0
S23 0.132 S22 0.705 0.1668 2.7705 0 70.8124 0
S24 0.131 S22 0.935 0.2269 5.3068 0 91.1237 0
S25 0.235 S22 0.483 0.1844 5.1083 0 47.6473 0
S26 0.003 S22 0.0007 0.0013 0 0.0003 1.4322 0
S27 0.044 S22 0.013 0.0014 0 0.0007 0 0.9222
S28 0.108 S22 0.008 0.0009 0 0.0004 0 0.5735
S29 0.064 S22 0.012 0.0027 0 0.0007 0 0.8843
S30 0.293 S22 0.002 0 0.0013 0 0 0.1183
S31 0.043 S22 0.004 0.0003 0 0.0001 0.1439 0
S32 0.143 S22 0.002 0.0002 0 0.0001 0 0.1
S33 0.310 S22 0.001 0.0002 0 0.0001 0.0194 0
S34 0.198 S22 0.001 0 0 0 0 0.0051
Mean 0.192 0.0186 0.388 0.0006 6.2561 0.4531
Max 1.00
Min 0.003
SD 0.186

Effectiveness
The effectiveness of STUs are presented in table 3. The results reveal that 4 STUs (S8, S12, S23 and S28)
are operating effectively. It is interesting to note that S23 is effective among only hill category and S28 is
effective among the two categories (hill and urban) while S8 and S12 are operating most effectively among
all the 34 STUs.

79
The average of effectiveness is worked out to be 0.710; this means that an average STU is operating 71%
effectively. 11 STUs are operating effectively above average. The table evinces that an average STU has to
reduce its staff capacity by 10.83and fleet by 6.92 to operate effectively. MZST (S25) is operating as most
ineffective STU with effectiveness score 0.412.
Combined Consequence of efficiency and effectiveness
Table 4 demonstrate the combined consequence of efficiency with effectiveness of the STUs. The results
show that only S22 is efficient and effective obtained by third DEA model. It is to be noted by earlier
observation that S22 is technical efficient but not effective with single and different outputs while efficient
and effective in the combined consequence. The average result of combined consequence is 0.192. It
indicates that an average STU may be able to decrease its inputs or increase its outputs by 81% to be
efficient as well as effective.

CONCLUSIONS
The paper evaluates the efficiency and effectiveness of public transport sector of India using Data
Envelopment Analysis (DEA) approach. The study finds that only one STU (S22; MEGTC) is relatively
technical efficient. The average of technical efficiency scores works out to be 0.20, which reveals that an
average STU can reduce its resources or increase outputs by 80% to become efficient. Among the inefficient
STUs, 8 STUs have the efficiency scores above the average efficiency scores.
The results reveal that 4 STUs (S8, S12, S23 and S28) are operating effectively. The average of
effectiveness is worked out to be 0.710; this means that an average STU is operating 71% effectively. 11
STUs are operating effectively above average. The results show that only S22 is efficient and effective
obtained by third DEA model. It is to be noted by earlier observation that S22 is technical efficient but not
effective with single and different outputs while efficient and effective in the combined consequence. The
average result of combined consequence is 0.192. It indicates that an average STU may be able to decrease
its inputs or increase its outputs by 81% to be efficient as well as effective.

APPENDIX A.1: STUS SELECTED FOR THE STUDY ARE AS FOLLOWS:


STU State of Nature of Category
STU Acronym STU Name
No. Operation Organization of STU
Andhra Pradesh State Road Andhra
S1 APSRTC Corp Rural
Transport Corporation Pradesh
Maharashtra State Road Transport
S2 MSRTC Maharashtra Corp Rural
Corporation
Gujarat State Road Transport
S3 GSRTC Gujarat Corp Rural
Corporation
Uttar Pradesh State Road Transport Uttar
S4 UPSRTC Corp Rural
Corporation Pradesh
Rajasthan State Road Transport
S5 RSRTC Rajasthan Corp Rural
Corporation
Kerla State Road Transport
S6 KSRTC Kerla Corp Rural
Corporation
Karnataka State Road Transport
S7 KnSRTC Karnataka Corp Rural
Corporation
S8 MDU Madurai Division Tamil Nadu Com Rural
North West Karnataka State Road
S9 NWKnSRTC Karnataka Corp Rural
Transport Corporation

80
S10 STHAR State Transport Haryana Haryana GD Rural
S11 KUM Kumbakonam Division Tamil Nadu Com Rural
S12 VPM Villuparam Division Tamil Nadu Com Rural
S13 CBE Coimbatore Division Tamil Nadu Com Rural
North East Karnataka State Road
S14 NEKnSRTC Karnataka Corp Rural
Transport Corporation
S15 SLM Salem Division Tamil Nadu Com Rural
Tamil Nadu State Express
S16 TN Tamil Nadu Com Rural
Transport Corporation Limited
North Bengal State Road Transport West
S17 NBSTC Corp Rural
Corporation Bengal
Bihar State Road Transport
S18 BSRTC Bihar Corp Rural
Corporation
South Bengal State Road Transport West
S19 SBSTC Corp Rural
Corporation Bengal
Kadamba Transport Corporation
S20 KDTC Goa Corp Rural
Limited
Orissa State Road Transport
S21 OSRTC Orissa Corp Rural
Corporation
S22 MEGTC Meghalaya Transport Corporation Meghalaya Corp Hill
Tripura Road Transport
S23 TRPTC Tripura Corp Hill
Corporation
S24 SKNT Sikkim Nationalized Transport Sikkim GD Hill
S25 MZST Mizoram State Transport Mizoram GD Hill
Brihan Mumbai Electric Supply & Mumbai
S26 BEST MU Urban
Transport Undertaking city
S27 DTC Delhi Transport Corporation Delhi City Corp Urban
Chennai Metropolitan Transport Chennai
S28 CNI Corp Urban
Corporation Limited city
Bangalore Metropolitian Transport Bangalore
S29 BMTC Corp Urban
Corporation city
Calcutta State Transport Kolkatta
S30 CSTC Corp Urban
Corporation city
S31 PMT Pune Municipal Transport Pune city MU Urban
Ahmedabad Municipal Transport Ahmedabad
S32 AMTS MU Urban
Service city
Chandigarh
S33 CTU Chandigarh Transport Undertaking MU Urban
city
Thane Municipal Transport
S34 TMTU Thane city MU Urban
Undertaking
Corp. stands for Corporation; Com stands for Company; GD stands for Government Department; MU stands for
Municipality Undertakings

APPENDIX A.2: NSM DEA MODEL

For NSM DEA Model, let us consider the n DMUs with m inputs (x ij for i = 1,…, m) and s outputs (y rj
for r = 1, …, s) for measuring the performance.

81
1  m sik s
srk 
Min k  k     
m  s  i 1 xik r 1 yrk 
subject to
n


j 1
jk yrj  srk  yrk r  1,..., s


j 1
jk xij  sik   k xik i  1,..., m

 jk  0 j  1,..., n
 
s , s  0 ; r  1,..., s, i  1,..., m
rk ik

where srk = slack in the r th output of the k th DMU, sik = slack in the i th input of the k th DMU,  jk ' s =
dual variables known as intensity variables.  k (scalar) is the (proportional) reduction applied to all inputs
of DMU k to improve efficiency.

REFERENCES
[1] Agarwal S., (2009), " Measuring the Efficiency of Public Transport Sector in India: An Application of Data
Envelopment Analysis", paper presented at Bi-annual Conference of International DEA Society held at
Philadelphia, PA, 10-12 July, 2009. http://astro.temple.edu/~banker/dea2009/paper/Sun1105744A.pdf
(Accessed on 23-03-2012).
[2] Agarwal S., Yadav S. P. and Singh S P., (2006), “A Data Envelopment Analysis based Efficiency Assessment of
Public Transport Sector of Uttar Pradesh in India”, Indian J. Transport Management, 30(1), 5-30.
[3] Agarwal S., Yadav S. P. and Singh S P., (2007), “Measurement of Technical and Scale Efficiencies Using DEA:
A Case of Public Transport Sector of India.”, 40th Annual Convention of Operational Research Society of India,
Delhi Chapter, held at INSA, Delhi from 04th to 06th December 2007.
[4] Agarwal S., Yadav S. P. and Singh S P., (2009), “Assessment of Total Factor Productivity Growth of the State
Road Transport Undertakings in India: A DEA Based MPI Approach”, Indian Economic Review, 44(2), 203-223.
[5] Agarwal S., Yadav S. P. and Singh S P., (2010), “DEA Based Estimation of The Technical Efficiency of State
Road Transport Undertakings in India”, OPSEARCH, 47 (3), 216-230.
[6] Agarwal S., Yadav S. P. and Singh S P., (2011), “A New Slack DEA Model to Estimate the Impact of Slacks on
the Efficiencies”, International Journal of Operations Research, 12(3), 241-256.
[7] Anjaneyulu M.V.L.R. Nagraj B.N., and Chandresekhar S. (2006), " DEA Approach for Performance Analysis of
State Road Transport Undertakings", Indian Journal of Transport Management, 30(4), 392-409.
[8] Balezentis A and Balezentis T. (2011), "Assessing the Efficiency of Lithunanian Transport Sector by applying
the methods of Multimoora and Data Envelopment Analysis", Transport, 26(3), 263-270.
[9] Banker R.D. and Morey R.C., (1986), “The use of Categorical Variables in Data Envelopment Analysis”,
Management Sciences, 23(12), 1613-1627.
[10] Barnum, D. T., McNeil, S., and Jonathon, H. (2007), “Comparing the Efficiency of Public Transportation
Subunits using Data Envelopment Analysis.” J. Public Transportation, 10(2), 1–16.
[11] Barnum, D. T., Tandon, S., and McNeil, S. (2008), “Comparing the Performance of Bus Routes after Adjusting
for the Environment using Data Envelopment Analysis”, J. Transp. Eng., 134(2), 77–85.
[12] Bhagavath V. (2009), "Technical Efficiency Measurement by Data Envelopment Analysis: An Application in
Transportation", Alliance Journal of Business Research, 1(1), 60-72.

82
[13] Bishnoi N.K. and Sujata U., (2007), "Efficiency Assessment of Haryana State Roadways: A Data Envelopment
Analysis", Indian Journal of Transport Management, 32(1), 9-23.
[14] Charnes A., Cooper W.W. and Rhodes E., “Measuring the Efficiency of Decision Making Units”, European
Journal of Operational Research, 1978, 2, 429-441.
[15] Chu X., Fielding G.J. and Lamer B.W. (1992), “Measuring Transit Performance using Data Envelopment
Analysis”, Transport Research A, 26A (3), 223-230.
[16] CIRT, Pune, “Indian Journal of Transport Management”, vol.29. Available on www.asrtu.org\scan0012.pdf
accessed on 15 March 2006.
[17] Cowie J. and Asenova D. (1999), “Organization form, Scale Effects and Efficiency in the British Bus Industry”,
Transportation, 26, 231-248.
[18] Husain N, Abdullah M. and Kumar S. (2000), "Evaluating Public Sector Efficiency with Data Envelopment
Analysis (DEA): A Case Study in Road Transport Department, Selangor, Malaysia", Total Quality Management,
11(4-6), 830-836.
[19] Karlaftis, M.G. (2004), "A DEA approach for evaluating the efficiency and effectiveness of the urban transit
system", European Journal of Operational Research, 152 (2), 354-364.
[20] Kumar S. (2011), "State Road Transport Undertakings in India: Technical Efficiency and its Determinants",
Benchmarking: An International Journal, 18(5), 616-643.
[21] Lin, E.T.J., Lan, L.W. (2009), "Accounting for accidents in the measurement of transport inefficiency: a case of
Taiwanese bus transit", International Journal of Environment and Sustainable Development, 8 (3/4), 365-385.
[22] Nagadevara V. and Ramanayya T.V. (2010), "Inter-temporal Shifts in Efficiency in a Road Transport
Organizations", Journal of Academy of Business and Economics, 10(1), 55-65.
[23] Odeck J. (2006), “Identifying Traffic Safety Best Practice: An Application of DEA and Malmquist Indices”,
Omega, 34(1), 28-40.
[24] Odeck J. and Alkadi A. (2001), “Evaluating efficiency in the Norwegian Bus Industry Using Data Envelopment
Analysis”, Transportation, 28, 211-232.
[25] Ramanathan R. (1999), “Using Data Envelopment Analysis for Assessing the Productivity of State Transport
Undertakings”, Indian Journal of Transport Management, 23(5), 301-312.
[26] Sheth C., Triantis K. and Teodorovic D., (2007), "Performance Evaluation of Bus Routes: A provider and
Passenger Perspective", Transportation Research Part E, 43, 453-478.
[27] Sun L., Rong J. and Yao L., (2010), "Measuring Transfer Efficiency of Urban Public Transportation Terminals
by Data Envelopment Analysis, Journal of Urban Planning and Development, 136(4), 314-319.

83
84
EVALUATION ON THE EFFICIENCY OF LOGISTICS INDUSTRY IN WUHAN
CITY CIRCLE
SHU LIU
Jianghan University, Economic and Technological Development zone, Wuhan, Hubei, China

LEI CHEN
Jianghan University, Economic and Technological Development zone, Wuhan, Hubei, China

ABSTRACT
Along with the rapid development of urbanization in China, there rise city circles which are
beyond the traditional administrative divisions and become the basic elements of economic re-
integration divisions. The Wuhan City Circle (WCC) is the core area in Hubei province on
population, industry, and economy. It has the most intensive economy, the most comprehensive
strength, the strongest economy in Hubei Province. In December 2007, the WCC was approved by
the central government as the national "resource-saving" and "environmentally friendly" areas
for the social construction of comprehensive reform. The economic conditions in different WCC
cities are not at the same level. Especially in logistics, the level of development and the industrial
efficiency exit large difference within the regions. Because logistics has the basic network service
features, this existing differences restrict the economic development of the WCC. By building
logistics efficiency evaluation model based on DEA method and analyzing the logistics system
from internal and external efficiency, we measure the relative level of the actual urban logistics
efficiency and effectiveness based on the presence of logistics differences, and analyze of causes
of the differences. We found that the average variable-return-to-scale technical efficiency of
logistics in the WCC region is 0.865. We provide police recommendations to promote the
development of logistics in the WCC.

Keywords: Wuhan city circle; logistics comprehensive benefit; DEA model; suggestions

INTRODUCTION
Wuhan city circle, including the Wuhan and Xiantao, Xiaogan, Qianjiang, Ezhou, Huanggang, Huangshi,
Xianning, Tianmen eight surrounding city, also known as "1 + 8". On geographical scope refers to Wuhan
as the center, with a radius of 100 km circle area, the regional city of city circle, the Wuhan city circle has
its own advantage in geographic location, it is connected to the pearl river delta, Yangtze river delta and
the western economic zone, at present in our country advances the strategy of the three has a pivotal position
in the development. It is not only the requirement of the rise of central China development in our country,
but also a requirement of the western development strategy. Wuhan city circle are the most important and
the most competitive production factors in Hubei province which is one of the most promising areas in the
Midwest. At present, the development of the Wuhan city circle has exposed many problems, one of the
most critical ones is the low efficiency of logistics. Certain extent in the development of the logistics
industry is more and more inadapted to the economic development speed. Therefore the study of Wuhan
city circle logistics level will cause more objective and scientific understanding of Wuhan city circle of

85
logistics industry development current situation, find out the shortcomings and correct it, so as to improve
the efficiency of logistics, reduce unnecessary waste of resources.
Scholars in the study of regional logistics in our country, there are Wang Bo, Yang Tian Jian, Zhao Yanbin
(2005) [5] in the comprehensive analysis on the key factors of the regional logistics development, using the
AHP analytic hierarchy process (AHP) to develop a more comprehensive and reasonable evaluation
indicators and fuzzy evaluation method in tianjin area were studied to evaluate the regional logistics
development level. Li Yumin, Li xu-hong, Mao Hai-jun, Wu Jigang (2004) [6] to build the evaluation index
system of regional logistics development, and put forward the regional logistics development based on
principal component analysis of SWOT analysis method. Liu Guo-xin (2004) [7] from the perspective of
the logistics industry to regional economic development, using the optimal separation method and neural
network to evaluate the development level of logistics industry in Wuhan. But less on the study of wuhan
city circle logistics efficiency, Gu Jun Jian (2013) [8] by building economic development level, circulation,
consumption level, transportation and information development level of the four major categories of index
system, and the integrated use of principal component analysis and cluster analysis to analyze Wuhan city
circle the logistics level, based on this, puts forward the Suggestions about the development of the logistics
of Wuhan city circle. Zhou Yewang (2012) in [9] to study the efficiency of the logistics of Wuhan City
Circle, the creative use of the DEA evaluation model for evaluation and analysis, measuring the actual level
of the city circle of logistics efficiency, but the data is relatively backward, and the future development of
Wuhan City Circle the reference significance is not big, but he about how to improve in the Wuhan City
Circle from city to city logistics efficiency is an understatement, did not put forward constructive
Suggestions. Based on this, this paper based on the latest data on the basis of the above, using the DEA
model to analyze Wuhan City Circle logistics comprehensive efficiency, finally put forward relevant
Suggestions, provide reference for the future logistics development of Wuhan city circle.
The setup of evaluation index system of logistics level in WCC
1. The principle of selecting indicators
2. Comprehensiveness
3. Measurable and comparable
4. Feasibility and exercisable
5. Simple and concise
6. Objective

THE CONTENT OF THE INDICATORS


By using DEA model to analyze the crux of the problem is to determine the input and output index. At
present, the scholars to evaluate logistics efficiency research, focusing on logistics on economic efficiency
of the single level, thus has neglected to evaluate the efficiency of logistics system itself. In recent years,
with the approval of Wuhan city circle of regional development planning, the Wuhan city circle determines
the building "three areas, a nuclear area four axis" regional development pattern of Wuhan city circle from
city to city is gradually pay attention to the logistics industry development. Development of the logistics

86
industry is bound to increase the investment in the construction of logistics infrastructure, but basically no
one ever focus on the construction of logistics infrastructure which is brought more output of logistics, if
there is a waste of resources. In this paper, based on the selection of indicators, considering the impact of
logistics on the economy at the same time, also pay attention to the efficiency of the logistics itself, to the
comprehensive evaluation of Wuhan city circle logistics efficiency.
When evaluating internal logistics system efficiency, this paper determine the input index of traffic of the
city is a fixed investment and ownership of truck, output indicators is freight volume and turnover of goods.
In evaluating the output efficiency of the logistics system to economic system, the input index is the freight
volume and freight turnover, output indicators are GDP, per capital output.
In the selection of indicators point in time, considering the practical significance of this study, is to the
Wuhan city circle logistics development put forward constructive Suggestions, give full play to the
efficiency of logistics, select the relevant data in the statistical yearbook of 2013 to carry on the empirical
analysis, the research and reference value.

THE SETUP OF RESEARCH METHOD AND MODEL


Empirical analysis
Selection, according to the established model and the index above 2012, the actual data of each city in
Wuhan city circle, first by looking for relevant statistical yearbook, and the related data, calculated, using
DEAP2.1 software to reach the logistics efficiency levels of each cities in Wuhan city circle in 2012.

Table 1. The related logistics and economy data of each city in WCC in 2012
City Traffic fixed The amounts of Freight Volume Gross regional Per capital
investment freight cars turnover of freight production GRP
WUHAN 291516 1104988 410.82 24354 8003.82 9.74
HUANGSHI 567.46 103785 97.5 6433 1040.95 3.98
XIAOGAN 652.85 116263 47.23 2894 1105.16 2.1
EZHOU 374.21 157740 16.49 1718 560.39 5.12
HUANGGANG 790.31 163288 43.88 3098 1192.88 1.59
XIANNING 557.63 100361 51.35 3036 773.2 2.6
XIANTAO 182.52 29514 14.33 1133 444.2 2.86
QIANJIANG 161.14 41987 35.56 1951 441.76 4.28
TIANMEN 134.46 31300 16.08 1383 321.22 1.92

First, through the relevant data using software to calculate the internal efficiency of logistics system, input
indicators is traffic fixed investment and ownership of truck, output indicators is the freight volume and
freight turnover, software to calculate the internal logistics system comprehensive efficiency, pure technical
efficiency and scale efficiency of three indicators. The calculation results are shown in table 2.
Then, o calculate the outer efficiency of the logistics system, use traffic volume and freight turnover as the
input indicators, gross regional production and per capital production value as the output index, to calculate
the external logistics efficiency—overall efficiency, pure technical efficiency and scale efficiency.

87
Table 2. The results of inner efficiency of logistics system in WCC
DMU CRSTE VRSTE SCALE
WUHAN 0.69 1 0.69
HUANGSHI 1 1 1
XIAOGAN 0.432 0.47 0.92
EZHOU 0.379 0.401 0.945
HUANGGANG 0.339 0.347 0.975
XIANNING 0.545 0.575 0.947
XIANTAO 0.619 1 0.619
QIANJIANG 1 1 1
TIANMEN 0.869 1 0.869

Table 3: the results of outer efficiency of logistics system in WCC


DMU CRSTE VRSTE SCALE
WUHAN 0.838 1 0.838
HUANGSHI 0.413 0.462 0.893
XIAOGAN 0.974 1 0.974
EZHOU 1 1 1
HUANGGANG 0.982 1 0.982
XIANNING 0.65 0.666 0.975
XIANTAO 1 1 1
QIANJIANG 0.736 0.769 0.957
TIANMEN 0.628 0.891 0.704

RESULTS AND ANALYSIS


Take the results of tables 2 and 3 into consideration, 2012 in Wuhan city in both the VRS values in the table
1, shows that the internal and external logistics system is effective, that is for internal logistics, logistics
enterprises in the business activities of the management, coordination and organization is effective, and is
also a significant role in promoting logistics on the economy. But there are still larger rising space Wuhan
economic development is the premise of increasing attention and investment in transport infrastructure.
From the whole, the Wuhan city circle can want to be and three big city circle of large cities, achieve the
strategy of rise of central China, to the logistics industry development, for the swimmer, circle the city, we
should use the advanced information technology to accelerate the process of development of modern
logistics industry, to find their own shortcomings, in view of the deficiency was improved, at the same time
to use for reference the experience of the development of the three big city circle, and then improve the
efficiency of logistics level inside the circle. To reduce the logistics system of the internal and external
logistics system differences, first is to minimize the logistics system internal differences, namely developing
modern logistics industry. One aspect of the development of modern logistics industry is very important,
which is to make full use of the elements, emphasizing the use of advanced information technology at the
same time pay attention to the unity of existing logistics resources reorganization and management,
encourage the logistics enterprise logistics theory of knowledge and technology learning, realize the
logistics modernization, intelligent, maximum limit to improve resource utilization and accelerate the
healthy and sustainable development of logistics industry, improve the overall efficiency of the logistics

88
and reduce logistics cost. The development of modern logistics industry to the first, second and third
industry has great influence and promote role, can effectively promote the economic development of Wuhan
city circle, in fully realize the importance of developing modern logistics industry on the basis of the
enterprise is to use knowledge and technology to arm themselves, while the government is increasing
attention to the logistics industry, the secondary development of logistics enterprises at the same time, also
to supervise the daily activities of the logistics enterprise, for those outdated technology, logistics capability
is weak, companies with poor credit, to urge him to correct, because we are forced its exit the market. After
all, the government's power and energy is limited, so in a consumer society as well as to supervise logistics
companies to report some unreasonable behavior, so as to shape the logistics enterprise's good reputation,
so that more conducive to logistics enterprise to follow up the pace of The Times.

THE SUGGESTIONS FOR LOGISTICS DEVELOPMENT IN WCC


As a national center city, Wuhan has unique geographical advantages, is the most developed city in Hubei
province economy, has the important status and role in the whole province. Wuhan to draw lessons from
the successful experience of China's three big city circle at the same time to combine its own characteristics
and advantages, looking for accurate positioning, create conditions for their own development. To realize
the Wuhan city circle logistics industry more efficient and sustainable development, also should create
logistics need external conditions and atmosphere, strengthen the cooperation of logistics enterprises and
communication, this needs the government and the department to build a platform to realize the information
sharing among enterprises, which can greatly improve the efficiency of logistics system as a whole. But a
major premise of implementing the resources sharing of logistics is to achieve the standardization of
logistics information, and at present about information standardization of logistics and didn't have a unified
standard, which requires the relevant departments in reference to foreign experience and combined the
actual area as soon as possible to formulate a set of perfect logistics information standards. This is definitely
not dispensable things, even though there is no reflects its importance, but the informatization and
standardization is the inevitable trend of modern logistics development, the basis of the realization of "two
type" logistics strategy. To achieve logistics information, you must first establish a complete set of advanced
standard of logistics, logistics standards for logistics information is of great significance and role;
Standards, to logistics informatization development, also to the further development of the space and
possibility. Now wuhan logistics enterprises with three big city circle is relatively, the country's overall
strength is weak, the lower level of modernization in the process of logistics enterprises and logistics
activities by means of information technology and management is relatively backward, which to a certain
extent, restricted the logistics the process of modernization.

REFERENCES
[1] Zhou Jun (2006) The analysis of the influence of regional logistics to regional economic growth, Chinese Journal
of Statistics and Decision (4):109-112.
[2] Feng Hua (2009) Based on principal component analysis of the regional logistics capability study, Chinese Journal
of Business Time (10):34-36.
[3] Yan Xiu Xia, Sun Lin Yan (2006) The influence of logistics capability of regional economic development
analysis, Chinese Journal of Scientific and Technological Progress and Countermeasures (10):160-163.

89
[4] He Xiao Zhou, Zhang Lin Li, Deng Zheng Hua (2007) The influence of the logistics industry on the regional
economic structure effect analysis, logistics industry development in Chongqing as an example, Chinese Journal
of Science and Technology Management Research(6):118-119.
[5] Wang Bo, Yang Tian Jian, Zhao Yan Bing (2005) The comprehensive evaluation of regional logistics
development level, Chinese Journal of Industrial Engineering (1):83-86.
[6] Li Yu Min, Li Xu Hong, Mao Hai Jun (2004) Comprehensive evaluation and based on principal component
analysis of the regional logistics development strategy, Chinese Journal of Transportation systems engineering
and information technology (2):92.
[7] Liu Guo Xin (2004) The logistics industry development in Wuhan level comprehensive evaluation, Chinese
Journal of Industrial Technology Economy (2):101.
[8] Gu Jun Jian (2013) The empirical analysis of the difference of Wuhan city circle logistics level, Chinese Journal
of Commercial Circulation (13):55-57.
[9] Zhou Ye Wang (2012) Based on the DEA efficiency of the logistics of Wuhan city circle difference comparative
analysis, Chinese Journal of Logistics Engineering and Management (10):20-22.
[10] Ellinger.A. (2000) Improving marketing/logistics cross functional collaboration in the supply chain, Journal of
Industrial Marketing Management (1):1-12.
[11] Du Zhi Ping, Mu Dong (2005) System coordination development degree of DEA evaluation research, Journal of
Strategy and Management (1):23.
[12] Hai Feng, Wu Lan Fen, Zhang Li Li (2004)Regional logistics development to promote the regional economy,
Journal of Scientific and technological progress and countermeasures(9):71-73.
[13] Tong Meng Da (2002) The theory of evaluation index system of regional modern logistics development, Journal
of Modern Logistics (9):11-12.
[14] Tong Hui Ning (2010) Port logistics comprehensive ability evaluation system of building research, Journal of
Logistics engineering and Management (4):25-28.
[15] Hu Shao Shan, Gao Feng (2011) Wuhan city circle industry cluster development present situation and
countermeasure research, Journal of The Theoretical Study (7):14-15.
[16] Cai Hong (2005) Nanjing modern logistics development level index system, Journal of Nanjing university of
Technology: 32-35.
[17] Kuo-chung Shang, Peter B. Marlow (2005) Logistics capability and performance in Taiwan’s major
manufacturing firms, Journal of Transportation Research Part (41):217-234.

90
EXAMINING THE EFFICIENCY OF MGNREGS IN NORTH EASTERN
STATES OF INDIA USING THE DATA ENVELOPMENT ANALYSIS
APPROACH
INDRANEEL BHOWMIK
Department of Economics, Tripura University (A Central University), Suryamaninagar, India (corresponding
Author)

PRITAM BOSE
Department of Economics, Tripura University (A Central University), Suryamaninagar, India

ABSTRACT
Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGS) operational in India since 2005-06
with the dual objective of providing employment to the rural households and creation of sustainable rural assets, is
extremely important for the backward and less developed North Eastern region of India. The extent of the scheme
among the 8 constituent states of the region is not uniform and though in aggregate the region has a greater share in
the country’s total. The present paper attempts to examine the implementation efficiency of the scheme in terms of
multi-output and multi-input indicators using the DEA approach. The efficiency scores of the DMUs (districts) are
pooled to obtain the state average which indicates wide variation. For the secondary stage of analysis OLS regression
has been used for explaining the efficiency scores. Literacy rate, used as a proxy for the empowerment of the people
is found to be highly significant positive contributor to efficiency. Pro-active governance is visible in the states with
better implementation efficiency.

Keywords: MGNREGS; Employment; Efficiency

INTRODUCTION
Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGS) is the world’s largest
employment generation programme initiated by the Government of India in 2005 with the objective of
enhancing livelihood security in the rural areas by providing a guarantee of 100 days wage employment to
all the applicant rural households in a financial year irrespective of any conditionality save the willingness
of the adult members to perform unskilled manual work. The enormity of the programme is evident in the
fact that Rs 300000 crores (more than US $ 50 billion) has been spent by the Central government since its
inception. The scheme is revolutionary in providing a right based framework for wage employment and
makes the government legally bound to those who demand it (Shah, 2012) and has the potential to serve as
a ‘big push’ to those regions which are in distress and could be an important first step to a full employment
strategy for India (Bhatia and Dreze, 2006; Ambasta et. al. 2008, Hirway, 2008). The rural poor across the
country, bereft of viable income opportunities have certainly gained economic power through job creation
activities undertaken at a grass root level (Dreze and Oldiges, 2007; Sarma, 2009; Ghosh, 2009). On the
other hand the quality of works under NREGS is often been put to question by many researchers
(Shrivastava, 2006). The World Bank Report (2011) observed that many of the public works has been
washed away in the subsequent monsoon. The scheme has also been criticised for “turning out to be the
officials’ baby.....” (CSE, 2008), being inflationary (Sethi, 2011), and breeding corruption (Mandavalli,
2010).

91
The 8 North eastern (NE) states of India accounting for 8% of area and 4% of population are linked to the
mainland through the 27 km long Siliguri corridor (GoI, 2008), often known as the chicken’s neck. These
states, though heterogeneous in ethnicity, language, culture and religion and bound by the commonality of
economic backwardness. Characterised by minimal industrialization, limited gainful employment
opportunities, geo-physical isolation and infrastructure bottlenecks, these predominantly agrarian states are
part of the ‘Special Category States’ earmarked for greater Central government support. The development
pursuits have witnessed various extra-economic hurdles in the form of ethnic disturbances, insurgency and
marginalization. The landscape is mostly hilly, though intermittent valleys provide opportunities for limited
plain-land agriculture. The people, of which almost 34 % are from the scheduled castes and tribal
communities, have higher poverty levels than the national average (GoI, 2008).
Naturally with this background the significance of programmes like Mahatma Gandhi National Rural
Employment Guarantee Scheme (MGNREGS) increases manifold for the region. The promise of 100 days
of guaranteed employment per household per year would provide them not only additional income but will
also serve as a safety net for people living in this region. This is evident from the fact that in 2013-14, the
share of NE states among the MGNREGS job-card holders is 5.17%, much higher than the share of
population. Further, the share of the region in terms of employment demand is even higher, 6.85% while
the share of region in employment generation is 7.02% of the country’s total. It is also observed that Tripura
and Mizoram are on top position for creating highest average person-days per household per year from
2010 onwards, however, the extent of scheme and its implementation is not uniform across the region and
we find that the average person days generated per household in the region as a whole, 43.9, is lower than
the national average (45.86) in 2013-14. However, the region has a higher work completion rate, 31.60%,
than the national average of 14.69%. Thus, with various indicators, it becomes tough to make an overall
assessment of the performance of the various administrative units in the implementation of the scheme. A
comprehensive indicator taking into account of the various objectives of the scheme becomes warranted.
Thus it is in this background, the present study stems with the objective to undertake an efficiency analysis
of implementation of MGNREGS using the Data Envelopment Analysis approach for the North Eastern
States.

METHODS
The DEA is a linear (mathematical) programming based method first originated in the literature by Charnes,
Cooper & Rhodes (1978) as a reformulation of the Farrell’s (1957) single-output, single- input radial
measure of technical efficiency to multiple- output, multiple-input case. The originators described DEA as
a mathematical programming model applied to observational data (that) provides a new way of obtaining
empirical estimates of relations- such as the production functions and/or efficient production possibility
surfaces- that are cornerstones of modern economics. The Extended Pareto- Koopmans definition states
that full (100%) efficiency is attained by any Decision Making Unit (DMU) if and only if none of its inputs
or outputs can be improved without worsening some of its other inputs or outputs. Under this technique for
each of the n decision making units (DMU) which consume m different inputs to produce S different
outputs, technical efficiency is given by the measure- ∑ruryro/ ∑ivixio, where, yro = rth output of a particular
DMU, O; xio = ith input of that particular DMU, O; ur is the weight associated with each kind of output & vi

92
is the weight associated with each kind of input. The DEA method is applicable to identify a host of
efficiency parameters. The technical efficiency score θ CRS is called the overall technical efficiency (OTE)
and is popularly known as CCR model. However, if the DMUs are not operating at an optimal scale, it can
be decomposed into pure technical efficiency (PTE) and scale efficiency (SE). Symbolically, PTE = θ VRS
and is generally referred as the BCC model. Scale Efficiency for the i th DMU, is obtained as SEi = θiCRS /
θiVRS , where SE = 1 indicates scale efficiency or constant returns to scale and SE < 1 indicates scale
inefficiency. Further, in case of scale inefficiency, we identify increasing returns to scale when, θ VRS ≠
θNIRS; while for θVRS = θNIRS the DMUs face decreasing returns to scale.
For the present purpose, we consider the districts as the decision making unit (DMU) as the major
responsibility of implementation of the scheme (MGNREGS) falls on them. The total person-days
generated in the district, and the number of assets created is considered as output; whereas, the availability
of fund, the number of works undertaken and the total number of participating households are considered
as inputs. The efficiency analysis is done for 2013-14 considering 83 districts (3 districts of Arunachal
Pradesh was left out due to non-availability of data) spread across 8 states for an output oriented model
since better efficiency means higher man-days generation and larger number of asset creation. Secondary
data obtained from nrega.nic.in, the official website of the scheme has been used. The computed efficiency
scores of the DMUs for each state have been pooled to calculate the average efficiency of each state. The
second stage of analysis includes an OLS regression considering the PTE scores of the DMUs as the
dependent variable. Literacy rates and the proportion of SC&ST population have been considered as
independent variables being the proxy indicator of empowerment and backwardness, respectively. The
expected signs for both the coefficients are positive because empowerment of the people is likely to foster
effective governance while backwardness also warrants greater government action. The efficiency analysis
has been undertaken using the EMS software, while the OLS regression analysis has been done using MS
Excel Software.

RESULTS AND DISCUSSIONS

Table 1: Frequency Distribution & Descriptive of OTE, PTE & SE of Districts of North East India (2013-14)
Efficiency Scores OTE PTE SE
E< 0.5 35 (42.17) 30 (36.14) 01 (01.20)
0.5 ≤ E < 0.6 13 (15.66) 13 (15.66) 02 (02.41)
0.6 ≤ E < 0.7 08 (09.64) 08(09.64) 01 (01.20)
0.7 ≤ E < 0.8 04 (04.82) 07 (08.43) 04 (04.82)
0.8 ≤ E < 0.9 06 (07.23) 05 (06.02) 08 (09.64)
0.9 ≤ E < 1.0 08 (09.64) 07 (08.43) 58 (69.88)
E= 1.0 09 (10.84) 13(15.66) 09 (10.84)
Descriptive Statistics
No. Of Districts (DMUs) 83 83 83
Mean 0.6092 0.6492 0.9299
Median 0.5648 0.5837 0.9694
Standard Deviation 0.2407 0.2307 0.1150
Minimum 0.0875 0.173 0.3178
Maximum 1 1 1
Source: Computed

93
Table 1 shows that more than 42% of the DMUs (Districts) operate with OTE levels below 50%, while in
terms of PTE, the situation is marginally better. The average efficiency levels under OTE are 0.6092, while
it improves to 0.6492 with PTE. Only 9 DMUs (10.84 %) are found to be efficient according to CCR
specifications whereas with BCC model, the number of efficient DMUs increase to 13 (15.66%). Further
on the basis of SE scores, we see that only 9 DMUs obtain the 100 percent score, while almost 70% of the
DMUs have a score above 0.9 and the number of DMUs with a score less than 0.5 is only 1 (one). Among
the 74 scale inefficient DMUs, only 5 (2 each from Sikkim and Tripura and 1 from Meghalaya) are found
to exhibit decreasing returns to scale while the remaining 69 depicts increasing returns to scale. Further
from Table 2, we find that Tripura leads in terms of both average OTE and average PTE, with scores being
0.9318 and 0.9418 respectively. That Tripura emerges at the top is not unexpected as in terms of the
traditional performance indicators of the scheme- a] Average person days generated per household in a
year, b] Proportion of households provided with 100 days of work & c] work completion rate, the state
leads over the rest of its peers by a margin. Mizoram ranks second in terms of implementation efficiency
with average OTE and average PTE scores of 0.8334 and 0.8458 respectively. Mizoram, it may be noted
here also has the second highest rank in terms of the average person days per household and work
completion rate, though it could not provide 100 days of work to any of the households demanding work.
On the other hand, Arunachal Pradesh stands at the nadir in terms of both average OTE and average PTE,
with scores of 0.3382 and 0.4887 respectively, because of the lowest work completion rate and provision
of 100 days of work to households (along with Mizoram), though Manipur and Assam has lower average
person days generated per household. The least efficient DMU, East Kameng district, also belong to
Arunachal Pradesh. Expectedly, the average scores for SE increase and apart from Arunachal Pradesh, the
scores are above 0.95 for all the other states. Interestingly, in terms of ranking under the average SE, we
find that Manipur is placed at the 6th position ahead of Assam, which had been the 6th rank holder in terms
of average OTE and average PTE. Further, to be noted that the two bottom ranked states in terms of SE are
the two largest in terms of area among the North Eastern region of India. It may also be noted that, even
though Tripura leads in average scores, the state has only 1DMU as efficient in the Meta frontier against 2
each from Mizoram and Assam under OTE. The number of efficient DMUs in terms of PTE increases to 4
in Assam, as also 1 DMU each from Arunachal Pradesh and Meghalaya emerges with score of 1.

Table 2: Average Efficiency of the State


States Number of Districts (DMUs) Average OTE Average PTE Average SE
Arunachal Pradesh 13 0.3382 [0] 0.4887 [1] 0.7259 [0]
Assam 27 0.5615 [2] 0.5901 [4] 0.9547 [2]
Manipur 9 0.5049 [1] 0.5178 [1] 0.9668 [1]
Meghalaya 7 0.7573 [1] 0.7733 [2] 0.9748 [1]
Mizoram 8 0.8334 [2] 0.8458 [2] 0.9795 [2]
Nagaland 11 0.6944 [1] 0.7085 [1] 0.9729 [1]
Sikkim 4 0.7820 [1] 0.7906 [1] 0.9880 [1]
Tripura 4 0.9318 [1] 0.9418 [1] 0.9897 [1]
Notes: Figures in parentheses indicates the number of efficient DMUs in the state
Source: Computed

The second stage of the analysis using OLS regression provides us with interesting results. The value of the
‘F’ statistic, 8.034, indicates the significant specification of the model. Rural literacy rates have a significant

94
positive impact (p < 0.01) on the efficiency of implementation of the scheme. Higher literacy rates are
considered as a proxy for the empowerment of the people. Empowered people have greater capability to
articulate their demand as well as are more effective in ensuring better governance. In this context, it may
be noted that Mizoram has the second highest literacy rates in the country, while Tripura has had the highest
improvement in literacy rate during the inter Census period of 2001 to 2011, in the country (RGI, 2012).
Arunachal Pradesh has the lowest literacy rate among the North Eastern States, thus, the validity of the
value of the coefficient for literacy rate is explained. On the other hand, the positive coefficient regarding
the proxy variable for economic backwardness can be considered as welcome outcome for governance. Our
expectation was that MGNREGS will have better implementation in backward areas, and the positive
coefficient (though not significant), visible in Table 3, attest to it. Interestingly, the average PTE in districts
with more than 50% of the population belonging to the scheduled communities is 0.6717 (N=47), while
that of non-scheduled community majority districts, is 0.6198 (N=36). Interestingly, MGNREGS, though
universal and not targeted in nature and had a latent objective of creating employment opportunities for the
weaker sections of the society as part of the inclusive development strategy. In Tripura, the average person
days generated per household in tribal majority areas has been higher mostly because of the lack of
alternative opportunities (Bhowmik, 2013), however that cannot be generalised for the entire north eastern
region right now, though such a cause cannot be ruled out, particularly for some of the hill districts of the
region with shifting cultivation, the traditional livelihood strategy of the tribal people, in the wane.

Table 3: Results of OLS Regression


Coefficients Standard Error T Stat P- value
Constant 2.673 15.705 0.170 0.8653
SC & ST population (%) 0.089 00.069 1.311 0.1936
Rural Literacy rates (%) 0.807 00.209 3.870 0.0002
R Squared 16.72
Adj. R Squared 14.64
Observations 83
F (2, 80) 08.03 0.0006
Source: Computed

CONCLUSIONS
The foregoing section shows that there are wide variations in the implementation efficiency of the scheme
among the North Eastern states. States like Tripura and Mizoram have outperformed their peers by quite a
distance, though there are a few districts across the states emerging as efficient. Arunachal Pradesh, the
largest state in terms of area is by far the least efficient in the implementation of MGNREGS, which can be
explained to an extent, owing to its lower rates of rural literacy, which makes the people less empowered
to call for better governance in this issue. The exceptional scores exhibited by all the districts (DMUs) of
Tripura can also be explained by the pro-active role played by the state government in implementation of
the scheme. The Left Front government of the state has been a vociferous supporter of MGNREGS
(Bhowmik & Bose, 2015) and has utilised the scheme effectively for mobilizing rural support and
consolidating its political power in successive elections. However, the most encouraging conclusion that
emerges is that better implementation efficiency is visible in regions dominated by the scheduled
communities, which is absolute necessary for the North eastern region, which warrants greater state action
owing to limited private sector economic activity.

95
APPENDIX: OTE, PTE & SE OF DMUS (DISTRICTS) (IN %)
DMUs OTE PTE RTS DMUs OTE PTE RTS
Arunachal Pradesh Assam
Anjaw 44.33 63.72 IRS Baksa 28.59 29.20 IRS
Changlang 27.10 37.37 IRS Barpeta 47.62 49.83 IRS
East Kameng 8.75 17.30 IRS Bongaigaon 43.15 45.28 IRS
East Siang 27.77 35.78 IRS Cachar 67.31 71.16 IRS
Kurung Kumey 39.10 45.34 IRS Chirang 60.03 62.05 IRS
Lohit 52.46 62.47 IRS Darrang 40.52 42.55 IRS
Lower Dibang Valley 40.15 72.80 IRS Dhemaji 46.31 47.99 IRS
Lower Subansiri 23.44 29.83 IRS Dhubri 63.26 63.40 IRS
Papum Pare 37.28 43.14 IRS Dibrugarh 48.76 49.56 IRS
Tawang 28.53 38.03 IRS Dima Hasao 48.03 57.78 IRS
Tirap 36.14 42.42 IRS Goalpara 44.44 46.15 IRS
Upper Dibang Valley 31.78 100.00 IRS Golaghat 84.77 100.00 IRS
Upper Subansiri 42.87 47.10 IRS Hailakandi 40.32 44.46 IRS
Manipur Jorhat 56.48 56.59 IRS
Bishnupur 100.00 100.00 CRS Kamrup 50.83 51.66 IRS
Chandel 27.54 30.38 IRS Kamrup (Metro) 59.50 73.01 IRS
Churachandpur 58.17 58.37 IRS Karbi Anglong 39.82 41.90 IRS
Imphal East 49.14 49.87 IRS Karimganj 100.00 100.00 CRS
Imphal West 34.66 36.28 IRS Kokrajhar 47.04 48.02 IRS
Senapati 44.53 45.20 IRS Lakhimpur 53.06 55.55 IRS
Tamenglong 53.14 56.28 IRS Morigaon 56.79 56.94 IRS
Thoubal 29.26 31.31 IRS Nagaon 56.07 56.10 IRS
Ukhrul 57.95 58.34 IRS Nalbari 100.00 100.00 CRS
Meghalaya Sivasagar 47.73 48.27 IRS
East Garo Hills 93.41 100.00 IRS Sonitpur 93.09 100.00 IRS
East Khasi Hills 80.06 81.45 DRS Tinsukia 48.46 50.27 IRS
Jaintia Hills 55.86 56.53 IRS Udalguri 44.03 45.61 IRS
Ri Bhoi 65.73 67.80 IRS Nagaland
South Garo Hills 77.58 81.18 IRS Dimapur 95.87 96.01 IRS
West Garo Hills 100.00 100.00 CRS Kiphire 66.66 69.93 IRS
West Khasi Hills 57.45 57.86 IRS Kohima 100.00 100.00 CRS
Mizoram Longleng 34.70 38.64 IRS
Aizawl 91.05 91.34 IRS Mokokchung 76.48 76.68 IRS
Champhai 100.00 100.00 CRS Mon 57.23 58.88 IRS
Kolasib 41.35 45.25 IRS Peren 96.48 97.13 IRS
Lawngtlai 89.75 90.10 IRS Phek 45.52 48.21 IRS
Lunglei 85.69 85.78 IRS Tuensang 43.98 45.34 IRS
Mamit 100.00 100.00 CRS Wokha 61.51 63.00 IRS
Saiha 66.03 69.66 IRS Zunheboto 85.45 85.51 IRS
Serchhip 92.85 94.52 IRS
Sikkim Tripura
East 69.59 71.59 DRS Dhalai 93.97 97.59 DRS
North 100.00 100.00 CRS North Tripura 90.23 90.60 DRS
South 70.97 72.28 DRS South Tripura 100.00 100.00 CRS
West 72.24 72.37 IRS West Tripura 88.53 88.54 IRS
Notes: OTE- Overall Technical Efficiency, PTE- Pure Technical Efficiency, RTS- Returns to Scale
Source: Computed

REFERENCES
[1] Ambasta, P, Shankar, P. S. V. and M. Shah (2008), Two Years of NREGA: The Road Ahead, Economic and
Political Weekly 43(8): 41-50.
[2] Banker, R.D., A. Charnes, & W.W. Cooper, (1984), Some Models for Estimating Technical and Scale
Inefficiencies in Data Envelopment Analysis, Management Science, 30 (9): 1078-92.

96
[3] Bhatia, B. and J. Dreze (2006), ‘Employment Guarantee in Jharkhand: Ground Realities’, Economic and Political
Weekly 41(3): 198-202
[4] Bhowmik, Indraneel (2013), MGNREGS in Tripura: A Study on Efficiency and Equity, NLI Research Studies
Series, No. 102/2013
[5] Bhowmik, Indraneel & P. Bose (2015), Efficiency and Impact of MGNREGS in Tripura, Social Change and
Development, XII (1): 1-19.
[6] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[7] CSE (2008), ‘NREGA: Opportunities and Challenges’ available at http://knowledge.nrega.net/105 (accessed on
15/02/2012)
[8] Farrell, M. J. (1957), The Measurement of Productive Efficiency, Journal of the Royal Statistical Society, Series
A, Vol. 120 (3): 253-90.
[9] Ghosh, Jayati (2009), ‘Equity and Inclusion through Public Expenditure: The potential of the NREGS’. Paper
presented at the International Conference on NREGA, Ministry of Rural Development and ICAR, January.
[10] GoI (2008), NER 2020: A Vision Document, Ministry of DONER, New Delhi available at www.necouncil.nic.in
(accessed on 28/09/2015)
[11] Hirway, Indira (2008), ‘NREGA: A Component of Full Employment Strategy for India, An Assessment” Paper
presented at International Seminar on National Rural Employment Guarantee Scheme in India, September 16-17,
New Delhi.
[12] Jha, Raghabendra and Raghav Gaiha (2012), NREGS: Interpreting the Official Statistics, Economic and Political
Weekly, 47 (40): 18-22.
[13] Mandavilli, Sujoy Rao (2010), ‘NREGA: Corruption mars welfare society, India’ available at
http://www.merinews.com/article/some-measures-for-enhancing-the-efficacy-of nrega/15833850.shtml (
accessed on 1/03/2013)
[14] Sethi, Anirudh (2011), MS- A Disaster Called MNREGA at http://www. anirudhsethireport.com/tag/nrega/ (
accessed on 1/03/2013)
[15] Shah, Deepak (2012) Implementation of NREGA in Maharashtra: Experiences, Challenges and Ways Forward,
MPRA paper No P39270 ( accessed on 1/03/2013)
[16] Sharma, Amita (2011), ‘Rights-based Legal Guarantee as Development Policy: The Mahatma Gandhi National
Rural Employment Guarantee Act’; Discussion Paper available at www.undp.org
[17] Srivastava, Ravi (2006), ‘National Rural Employment Guarantee and Seasonal Migration in India’, Migration
Newsletter, 1 (3)
[18] World Bank (2011), ‘Social Protection for a Changing India’, Volume II (Washington DC: World Bank).

97
98
IDENTIFYING STRONG AND WEAK CONGESTION IN THE PRESENCE OF
NEGATIVE DATA: A DEA APPROACH
ROBABEH ESLAMI
Department of Mathematics, Faculty of Technology and Engineering, South Tehran Branch, Islamic Azad University,
Tehran, Iran (corresponding Author)

MOHAMMAD KHOVEYNI
Department of Applied Mathematics, College of Basic Sciences, Yadegar-e-Imam Khomeini (RAH) Shahr-e-Rey
Branch, Islamic Azad University, Tehran, Iran

ABSTRACT
One of the important concepts of data envelopment analysis (DEA) is congestion. A decision making unit (DMU) has
congestion if an increase (decrease) in one or more inputs of the DMU leads to a decrease (increase) in one or more
its outputs. In DEA literature, there are many approaches for recognizing congestion status of decision making units
(DMUs). All existing congestion approaches are applicable only to technologies specified by non-negative data,
whereas in the real world, it may exist negative data, too. Hence, in this research, by assuming DMUs in the presence
negative data, we first propose a DEA model to determine candidate DMUs for having congestion and then, a DEA
approach is presented to detect congestion status for these DMUs. The advantage of the proposed method is that it is
capable of exploring the congestion status for DMUs in the presence of negative data, whiles in this situation, the
existing congestion methods are incapable of identifying the congestion status. This matter is the main drawback of
the existing congestion approaches which is modified in this study. Finally, an empirical application is presented to
highlight the purpose of this research.

Keywords: Data envelopment analysis (DEA); Strong and weak congestion; Negative data in DEA; Slack variables

INTRODUCTION
Congestion is one of the important topics in the economic theory. A decision making unit (DMU) has
congestion if decreasing the inputs of DMU causes to increase its outputs. In other words, whenever the
excessive values of inputs lead to a reduction of the outputs, then the target DMU has congestion. Hence,
congestion is a wasteful stage of the production process. For this reason, the existence of congestion for
DMUs makes to increase the costs and decrease the efficiency that these issues are not acceptable for
decision makers.
It is worth stressing that the main drawback of the existing congestion DEA approaches (such as: Färe et
al., 1985, 1994; Brockett et al., 1998; Cooper et al., 2000, 2001a, 2001b, 2002; Tone and Sahoo, 2004;
Sueyoshi and Sekitani, 2009; Khoveyni et al., 2013) is that in these approaches, all inputs and outputs of
DMUs must be as non-negative data, whiles in the real world, the inputs and outputs are not necessarily
non-negative and negative inputs and/or outputs can be appeared in many applications. For instance,
effluent in “the notional effluent processing systemˮ, is as a non-positive input and also, loss when net
profit is considered as an output variable.
This fact is the motivation for creating this current research. Hence, in order to modify this drawback, we
introduce a DEA approach for exploring the congestion status of DMUs in the presence of negative inputs
and/or outputs.

99
THE PROPOSED APPROACH

Consider a set of n observed DMUs, DMU j j  1, 2, , n , with input-output combinations

DMU j   X Tj , YjT   m s
( j  1, 2, , n) in the presence of negative data, which use input vector

Xj m
to produce output vector Y j 
s
. Then, the production possibility set (PPS) under weak
disposability assumption is as follows:

 X  n n n 
PConvex      j X j  X ,   jY j  Y ,  j  1,  j  0; j  1, 2, , n  , (1)
 Y  j 1 j 1 j 1 

Definition 1 (Strong congestion) (Tone and Sahoo, 2004): A DMU k  X kT , YkT   is strongly congested if
there exists an activity X T
k , YkT   PConvex such that X k   X k (with 0    1 ) and Yk   Yk (with
  1 ).

Definition 2 (Weak congestion) (Tone and Sahoo, 2004): A DMU is (weakly) congested if it is strongly
efficient with respect to PConvex and there exist an activity in PConvex that uses less resources in one or more
inputs for making more products in one or more outputs.
Specifying candidate DMUs for having congestion in the presence of negative data
In the absence of input strong disposability assumption, PConvex is also acquired as (1) in the presence of
negative inputs and outputs. Here, we introduce the following linear programming (LP) model in evaluating
DMU k   X kT , YkT  such that xk  yk 
m s
and :

s
Max hk   qrk
r 1
n
s.t.  x
j 1
j ij  xik , i  1, , m,

 y
j 1
j rj  qrk  yrk , r  1, , s, (2)


j 1
j  1,

 j  0, j  1, , n,
qrk  0, r  1, , s.
Definition 3 ( PConvex - efficient): DMU k is called “ PConvex  efficientˮ if and only if in each optimal

solution of the model (2), qrk  0 (r  1, 2, , s) (or equivalently hk  0 ). In this case, DMU k is on the

100
strongly efficient frontier of PConvex , hence it is a candidate DMU for having congestion. Otherwise, DMU k
is called PConvex  inefficient.

Recognizing strong and weak congestion in the presence of negative data

Now, assume that DMU k  X kT , YkT   is P Convex  efficient. Then, we present the following LP model:

s
Max s
r 1

r

n
s.t.  x
j 1
j ij  si  xik , i  1, , m,

 y
j 1
j rj  sr  yrk , r  1, , s,


j 1
j  1, (3)

 j  0, j  1, , n,
si  0, i  1, , m,
sr   , r  1, , s,

where  is a non-Archimedean small positive number.


After solving model (3), we have two following cases:

Case (1): If model (3) be feasible, then DMU k has congestion. In this case, we consider the following LP
model:
m
Max s
i 1

i

n
s.t.  xj 1
j ij  si  xik , i  1, , m,

 y

j rj  yrk  sr , r  1, , s, (4)


j 1
n

 j 1
j  1,

 j  0, j  1, , n,
si   , i  1, , m,


where sr (r  1, 2, , s) are the obtained optimal solutions from model (8) which are fixed in model (4).
In addition,  is a non-Archimedean small positive number.

101
By solving model (4), we have two following subcases:

Subcase (1.1) If model (4) be feasible, then DMU k has strong congestion. Hence, DMU k
has weak congestion, too.
Subcase (1.2) If model (4) be infeasible, then there is no possible to decrease all inputs of
DMU k . Thereupon, DMU k has weak congestion.

Case (2): If model (3) be infeasible, then DMU k has no strong congestion. Therefore, we present the
following LP model:
s
Max zk   sr
r 1
n
s.t.  x
j 1
j ij  si  xik , i  1, , m,

 y
j 1
j rj  sr  yrk , r  1, , s,


j 1
j  1, (5)

 j  0, j  1, , n,
si  0, i  1, , m,
sr  0, r  1, , s.

Here, presume that  z , λ ,s



k
  
, s  is an obtained optimal solution from model (5). Then, we have two
subcases as below:

If zk  0 , then s   0 because DMU k is PConvex  efficient.

Subcase (2.1) Hence, DMU k
has weak congestion.
Subcase (2.2) If zk  0 , then DMU k has no weak congestion, too.

“Note that strong congestion implies weak congestion but not vice versa, and that, in a single input and a
single output case, there is no distinction between strong and weak congestions. Weak but not strong
congestion occurs only for the case with more than one input or one output, (Tone and Sahoo, 2004, pp.
761–762)”.

EMPIRICAL APPLICATION
In this section, in order to evaluate the effectiveness of the proposed congestion approach, we apply it into
the data set of “the notional effluent processing systemˮ. This data set has been extracted from Sharp et al.
(2006) and later, it has been also used by Emrouznejad et al. (2013) and Cheng et al. (2013). The data set
contains 13 DMUs which they have two inputs as cost (positive input) and effluent (non-positive input).
They have also three outputs as saleable (positive output) and CO2 and methane (non-positive outputs).

102
Note that, the data of inputs and outputs have not been shown for the sake of their voluminous. The obtained
results of the proposed approach have been presented in Table 1.

Table 1: The obtained results of the proposed approach


DMU Results of Definition 3 Results of the proposed approach
D1 PConvex-efficient Weak congestion
D2 PConvex-inefficient ----
D3 PConvex-efficient No congestion
D4 PConvex-efficient Weak congestion
D5 PConvex-inefficient ----
D6 PConvex-efficient Weak congestion
D7 PConvex-efficient No congestion
D8 PConvex-efficient No congestion
D9 PConvex-efficient Weak congestion
D10 PConvex-inefficient ----
D11 PConvex-efficient No congestion
D12 PConvex-inefficient ----
D13 PConvex-efficient No congestion

CONCLUSIONS
Hitherto, there are several congestion approaches in the DEA literature (such as: Färe et al., 1985, 1994;
Brockett et al., 1998; Cooper et al., 2000, 2001a, 2001b, 2002; Tone and Sahoo, 2004; Sueyoshi and
Sekitani, 2009; Khoveyni et al., 2013). In these congestion approaches, all inputs and outputs of DMUs
must be non-negative, which this matter is the main drawback of them. Because in the real world, negative
data could be appeared in many applications. This fact is the motivation for creating this current study.
Thus, in order to modify this drawback, we present a DEA congestion approach in the presence of negative
inputs and/or outputs. For this reason, a LP model is first introduced to identify candidate DMUs for having
congestion, and then we propose an analytical method to recognize congestion status of these DMUs.
Lastly, an empirical application is presented to illustrate the purpose of this study. Besides, this research
can be similarly proliferated for special data such as: interval, integer, stochastic, fuzzy, and etc. in the
presence of negative data.

REFERENCES
[1] Brockett P.L., Cooper W.W., Shin H.C., Wang Y. (1998) Inefficiency and congestion in Chinese production
before and after the 1978 economic reforms, Socio-Economic Planning Sciences 32: 1-20.
[2] Cheng G., Zervopoulos P., Qian Z. (2013) A variant of radial measure capable of dealing with negative inputs
and outputs in data envelopment analysis, European Journal of Operational Research 225: 100-105.
[3] Cooper W.W., Seiford L.M., Zhu J. (2000) A unified additive model approach for evaluating inefficiency and
congestion with associated measures in DEA, Socio-Economic Planning Sciences 34: 1-25.
[4] Cooper W.W., Deng H., Gu B.S., Li S., Thrall R.M. (2001a) Using DEA to improve the management of
congestion in Chinese industries (1981–1997), Socio-Economic Planning Sciences 35: 227-242.
[5] Cooper W.W., Gu B., Li S. (2001b) Comparisons and evaluations of alternative approaches to the treatment of
congestion in DEA, European Journal of Operational Research 132: 62-74.
[6] Cooper W.W., Deng H., Huang Z.M., Li S.X. (2002) A one-model approach to congestion in data envelopment
analysis, Socio-Economic Planning Sciences 36: 231-238.

103
[7] Emrouznejad A., Anouze A.L., Thanassoulis E. (2010) A semi-oriented radial measure for measuring the
efficiency of decision making units with negative data, using DEA, European Journal of Operational Research
200: 297-304.
[8] Färe R., Grosskopf S., Lovell C.A.K. (1985) The measurement of efficiency of production, Kluwer-Nijhoff
Publishing.
[9] Färe R., Grosskopf S., Lovell C.A.K. (1994) Production Frontiers, Cambridge University Press.
[10] Khoveyni M., Eslami R., Khodabakhshi M., Jahanshahloo G.R., Hosseinzadeh Lotfi F. (2013) Recognizing strong
and weak congestion slack based in data envelopment analysis, Computers & Industrial Engineering 64: 731-
738.
[11] Sharp J.A., Liu W.B., Meng W. (2006) A modified slacks-based measure model for data envelopment analysis
with ‘natural’ negative outputs and inputs, Journal of the Operational Research Society 57 (11): 1-6.
[12] Sueyoshi T., Sekitani K. (2009) DEA congestion and returns to scale under an occurrence of multiple optimal
projections, European Journal of Operational Research 194: 592-607.
[13] Tone K., Sahoo B.K. (2004) Degree of scale economies and congestion: A unified DEA approach, European
Journal of Operational Research 158: 755-772.

104
MEASUREMENT OF BANKING EFFICIENCY IN VISEGRAD COUNTRIES: A
DEA WINDOW ANALYSIS
IVETA PALEČKOVÁ
Silesian University, School of Business Administration, Univerzitní náměstí 1934/3, Karviná, 734 01, Czech Republic

ABSTRACT
The aim of the paper is to apply the Data Envelopment Analysis (DEA) window analysis on commercial banks of the
group of Visegrad countries (Czech Republic, Hungary, Poland and Slovakia) and to examine the banking efficiency
of Visegrad countries’ banking sectors during the period 2009-2013. We use the DEA window analysis based on an
input oriented model to measure banking efficiency. There is a lack of studies in Visegrad countries examining banking
efficiency using the Window DEA, which creates an opportunity for this research. The obtained results allow for an
analyses of trends of the overall banking sector efficiency. The technical efficiency is analysed sequentially with a
certain window width using a panel data of the commercial domestic banks. The results of the paper is that the average
efficiency was decreasing during the period 2009-2013. The Hungarian commercial banks were the most efficient and
the Slovak banking industry was the lost efficient in both models.

Keywords: efficiency; window DEA; banking sector; Visegrad countries

INTRODUCTION
The aim of the paper is to apply the Data Envelopment Analysis (DEA) window analysis on commercial
banks of the group of Visegrad countries (Czech Republic, Hungary, Poland and Slovakia) and to examine
the banking efficiency of Visegrad countries’ banking sectors during the period 2009-2013.
There is a lack of studies in Visegrad countries examining banking efficiency using the Window DEA,
which creates an opportunity for this research. Řepková (2014a) applied the DEA window analysis on the
data of the Czech banking sector and estimated that average efficiency was increasing during the period
2003-2012. Řepková (2014b) used this approach to estimated efficiency of the Slovak banks and found that
average efficiency during the period 2003-2008 the average efficiency was increasing and then during the
period 2010-2011 the average efficiency decreased as a result of financial crisis. Also Zimková (2014)
estimated technical efficiency of the Slovak banking industry using window analysis. Next, Kisielewska et
al. (2005) estimated efficiency of the largest ten Polish banks during the period 1995-2003 employing the
DEA window analysis and found that efficiency slightly increased within this period. This paper could fill
the gap following time line in the empirical literature. The contributions of this paper is the fact, that the
DEA window analysis approach will be applied on the Visegrad courtiers‘ commercial banks. The structure
of the paper is follow. Next section describes the methodology and data and selection of variables. Next
part of paper reveals the estimated results and last section concluded the paper.

METHODS AND DATA


The study of the efficient frontier began with Farrell (1957), who defined a simple measure of a firm’s
efficiency that could account for multiples inputs. DEA is a mathematical programming technique that
measures the relative efficiency of a homogeneous set of decision-making units in their use of multiple
inputs to produce multiple outputs. DEA also identifies, for inefficient DMUs, the sources and level of

105
inefficiency for each of the inputs and output (Charnes et al., 1995). It provides a means of comparing the
efficiency of DMUs with each other based on several inputs and / or outputs.
The CCR (Charnes, Cooper, Rhodes) model is the basic DEA model, as introduced by Charnes et al. (1978).
The CCR model presupposes that there is no significant relationship between the scale of operations and
efficiency by assuming constant returns to scale (CRS) and delivery of overall technical efficiency. Then it
was modified by Banker et al. (1984) and became the BCC (Banker, Charnes, Cooper) model, which
accommodates variable returns to scale. The resulting BCC model was used to assess the efficiency of
DMUs characterized by variable returns to scale (VRS).
Data Envelopment Analysis is performed in only one time period, hampering the measurement of efficiency
changes when there is more than one time period. Window analysis is one of the methods used to verify
productivity change over time. As Savić et al. (2012) showed, window analysis technique works on the
principle of moving averages (Charnes et al., 1995; Yue, 1992; Cooper et al., 2007). DEA window analysis
was proposed by Charnes et al. (1985) in order to measure efficiency in cross sectional and time varying
data. Each DMU (i.e. bank) is treated as a different bank in a different period which can increase the number
of data point.
The number of firms that can be analyzed using the DEA model is virtually unlimited. Therefore, data on
firms in different periods can be incorporated into the analysis by simply treating them as if they represent
different firms. In this way, a given firm at a given time can compare its performance at different times and
with the performance of other firms at the same and at different times. Through a sequence of such windows,
the sensitivity of a firm’s efficiency score can be derived for a particular year according to changing
conditions and a changing set of reference firms. A firm that is DEA efficient in a given year, regardless of
the window, is likely to be truly efficient relative to other firms. Conversely, a firm that is only DEA
efficient in a particular window may be efficient solely because of extraneous circumstances. In addition,
window analysis provides some evidence of the short-run evolution of efficiency for a firm over time. Of
course, comparisons of DEA efficiency scores over extended periods may be misleading (or worse) because
of significant changes in technology and the underlying economic structure (Yue, 1992).
Following Asmild et al. (2004) and Gu and Yue (2011), consider 𝑁 DMUs (𝑛 = 1,2, . . , 𝑁) observed in
𝑇 (𝑡 = 1,2, … , 𝑇) periods using 𝑟 inputs to produce 𝑠 outputs. Let 𝐷𝑀𝑈𝑛𝑡 represent an 𝐷𝑀𝑈𝑛 in period 𝑡
with a 𝑟 dimensional input vector 𝑥𝑛𝑡 = (𝑥𝑛1𝑡 , 𝑥𝑛2𝑡 , … , 𝑥𝑛𝑛 )′ and 𝑠 dimensional output vector
𝑦 = (𝑦𝑛1𝑡 , 𝑦𝑛2𝑡 , … , 𝑦𝑛𝑠𝑡 )′ . If a window starts at time 𝑘 (1 ≤ 𝑘 ≤ 𝑇) with window width𝑤 (1 ≤ 𝑤 ≤ 𝑡 − 𝑘),
then the metric of inputs is given as follows:

𝑥𝑘𝑤 = (𝑥1𝑘 , 𝑥2𝑘 , … , 𝑥𝑁𝑘 , 𝑥1𝑘+1 , 𝑥2𝑘+1 , … , 𝑥𝑁𝑘+1 , 𝑥1𝑘+𝑤 , 𝑥2𝑘+𝑤 , … , 𝑥𝑁𝑘+𝑤 ) , (1)

The metric of outputs as:



𝑦𝑘𝑤 = (𝑦1𝑘 , 𝑦2𝑘 , … , 𝑦𝑁𝑘 , 𝑦1𝑘+1 , 𝑦2𝑘+1 , … , 𝑦𝑁𝑘+1 , 𝑦1𝑘+𝑤 , 𝑦2𝑘+𝑤 , … , 𝑦𝑁𝑘+𝑤 ) , (2)

106
The CCR model of DEA window problem for 𝐷𝑀𝑈𝑡𝑘 is given by solving the following linear program:

min 𝜃, (1)
′ ′
subject to 𝜃 𝑋𝑡 − 𝜆 𝑋𝑘𝑤 ≥ 0, (2)
𝜆′ 𝑌𝑘𝑤 − 𝑌𝑡 ≥ 0, (3)
𝜆𝑛 ≥ 0 (𝑛 = 1,2, … , 𝑁 × 𝑤). (4)
BCC model formulation can be obtained by add the restriction ∑𝑛𝑛=1 𝜆𝑛 = 1 (Banker et al., 1984). The
objective value of CCR model is designated technical efficiency and the objective of BCC model is pure
technical efficiency. The BCC model is illustrated as:

min 𝜃, (5)
′ ′
subject to 𝜃 𝑋𝑡 − 𝜆 𝑋𝑘𝑤 ≥ 0,
(6)
𝜆′ 𝑌𝑘𝑤 − 𝑌𝑡 ≥ 0, (7)
𝑛

∑ 𝜆 𝑛 = 1, (8)
𝑛=1
𝜆𝑛 ≥ 0 (𝑛 = 1,2, … , 𝑁 × 𝑤). (9)
Asmild et al. (2004) point out that there are no technical changes within each of the windows because all
DMUs in each window are compared and contrast against each other and suggest a narrow window width
should be used. Charnes et al. (1995) found that w = 3 or 4 tended to yield the best balance of in
formativeness and stability of the efficiency scores. In order to be sure that the results will be credible,
a narrow window width must be used. Therefore, a three year window has been chosen in this paper.
The data set used in this paper was obtained from the database BankScope and the annual reports of
commercial banks during the period 2009–2013. All the data is reported on an unconsolidated basis. Due
to the homogenous data set we analyze only commercial banks. We use unbalanced panel data from 13
Czech commercial banks, 11 Slovak commercial banks, 23 Hungarian commercial banks and 38 Polish
commercial banks (with regard to mergers and acquisitions of banks).
In order to conduct a DEA window analysis estimation, inputs and outputs need to be defined. Four main
approaches (intermediation, production, asset and profit approach) have been developed to define the input-
output relationship in financial institution behavior. We adopted an intermediation approach which assumes
that the banks’ main aim is to transform deposits into loans. We employed three inputs (labor, fixed assets
and deposits), and two outputs (loans and net interest income). We measure labor by the total personnel
costs covering wages and all associated expenses and deposits by the sum of demand and time deposits
from customers, interbank deposits and sources obtained by bonds issued. Loans are measured by the net
value of loans to customers and other financial institutions and net interest income (NII) as the difference
between interest incomes and interest expenses. Descriptive statistics of inputs and outputs are in Table 1.

Table 1: Descriptive statistics


Variable Deposits Labor Fixed assets Loans NII
Mean 389443 6515 8211 315250 20534
Median 23978 391 193 19740 925
Minimum 0.00 2.22 0.09 0.00 0.29
Maximum 7876341 204277 261523 7047179 656202
St. Dev. 1092663 23278 31382 933918 81557

107
RESULTS AND DISCUSSIONS
We adopted DEA window analysis SBM (slack based model – non-radial) models that can evaluate the
overall efficiency of decision-making units for the whole terms as well as the term efficiencies. We used
input-oriented model to estimate efficiency under the assumptions of constant and variable returns to scale.
For empirical analysis we used MaxDEA software.

Table 2: Average efficiency of banking sectors in Visegrad countries in CCR model


2009-2011 2010-2012 2011-2013
Visegrad countries 0.4532 0.3137 0.3142
Poland 0.4558 0.3269 0.3331
Slovakia 0.4155 0.2739 0.2792
Hungary 0.4864 0.3353 0.3183
Czech Republic 0.4270 0.2752 0.2865

The results of the DEA efficiency scores under constant variable of scale are presented in Table 2. The
average efficiency in commercial banks in Visegrad countries reached the value 31-45% in CCR model.
The average efficiency was decreasing during the analysed period. We found that this decrease in efficiency
was caused as a result of financial crisis. It confirm the fact that it was found the decrease in total loans and
then net interest income in balance sheet of individual banks. Our finding confirms the study of Anayiotos
et al. (2010) who concluded that the banking efficiency decreased during the period of financial crisis. The
Hungarian commercial banks were the most efficient in CCR model. On the other hand, the Slovak banks
was the lost efficient. These results confirm the finding of Palečková (2015).

Figure 1: The range of score of efficiency of Visegrad countries’ commercial banks in CCR model

Figure 1 presents estimated value of minimum, maximum and median of banking efficiency in Visegrad
countries and also in individual countries, ie. the Polish, Slovak, Hungarian and Czech banking sector. In
Polish and Hungarian banking sector banks reach the maximum value of 100%. This fact imply that some
banks in these markets had produced its output on the efficiency frontier. But on the other hand, in
Hungarian banking sector differences between individual banks are the largest. The lowest efficient banks
was almost 90% inefficient. In Slovak banking industry the maximum value was 51%. It means that the
most efficient bank in Slovakia reach the efficiency score only 51%.

108
Table 3: Average efficiency of banking sectors in Visegrad countries in BCC model
2009-2011 2010-2012 2011-2013
Visegrad countries 0.5617 0.4169 0.4098
Poland 0.5090 0.3821 0.3701
Slovakia 0.4903 0.3669 0.3579
Hungary 0.6932 0.5408 0.5361
Czech Republic 0.5452 0.3290 0.3399

The results of average efficiency scores under variable return to scale are showed in Table 3. Average
efficiency of Visegrad banking sectors with variable return to scale was between 40 to 56%. The values of
efficiency computed by the BCC model reach higher values than efficiency computed by the CCR model
by eliminating the part of the inefficiency that is caused by a lack of size of production units. The Hungarian
banking sector was the most efficient and the lowest efficiency reached the Slovak banking industry in BCC
model.

Figure 2: The range of score of efficiency of Visegrad countries’ commercial banks in BCC model
Figure 2 shows the minimum and maximum value and median of efficiency of Visegrad countries’ banking
sectors. Similar like in CCR model, the maximum value of 100% reached banks in Visegrad countries,
Hungary and Poland. The lowest value as well as the range (the difference between maximum and minimum
values) achieved the Slovak commercial banks.

CONCLUSIONS
The aim of the paper was to apply the Data Envelopment Analysis window analysis on commercial banks
of the group of Visegrad countries (Czech Republic, Hungary, Poland and Slovakia) and to examine the
banking efficiency of Visegrad countries’ banking sectors during the period 2009-2013. We found that
average efficiency in commercial banks in Visegrad countries reached the value 31-45% in CCR model and
40-56% in BCC model. The average efficiency was decreasing during the period 2009-2013 and this
development was probably as a result of financial crisis. Total loans and then net interest income decreased
in balance sheet of individual banks. The Hungarian commercial banks were the most efficient in both
models. On the other hand, the Slovak banking industry was the most efficient unit.

109
ACKNOWLEDGEMENT
Research behind this paper was supported by the Czech Science Foundation within the project GAČR 13-
03783S ‘Banking Sector and Monetary Policy: Lessons from New EU Countries after Ten Years of
Membership’.

REFERENCES
[1] Anayiotos, G., Toroyan, H., Vamvakidis, A. (2010). The efficiency of emerging Europe’s banking sector before
and after the recent economic crisis, Financial Theory and Practice, 34 (3): 247–267.
[2] Asmild, M., Paradi, J.C., Aggarwall, V., Schaffnit, C. (2004). Combining DEA Window Analysis with the
Malmquist Index Approach in a Study of the Canadian Banking Industry, Journal of Productivity Analysis, 21
(1): 67-89.
[3] Banker, R.D., Charnes, A., Cooper, W. (1984). Some Model for Estimating Technical and Scale Inefficiencies in
Data Envelopment Analysis, Management Science, 30: 1078-1092.
[4] Charnes A., Cooper, W.W., Lewin, A.Y., Seiford L.M. (1995). Data Envelopment Analysis: Theory, Methodology
and Applications. New York: Springer-Verlag.
[5] Charnes, A., Clark, T., Cooper, W.W., Golany, B. (1985). A developmental study of data envelopment analysis
in measuring the efficiency of maintenance units in U. S. Air Forces. In: Thompson, R., Thrall, R.M. (Eds.).
Annals of Operational Research, 2: 95-112.
[6] Charnes, A., Cooper, W.W., Rhodes, E. (1978). Measuring the Efficiency of Decision Making Units, European
Journal of Operational Research, 2: 429–444.
[7] Cooper, W., Seiford, L.M., Tone, K. (2007). Data Envelopment Analysis: A Comprehensive Text with Models
Applications. New York: Springer Science.
[8] Farrell, M.J. (1957). The Measurement of Productive Efficiency. Journal of the Royal Statistical Society (Series
A), 120 (2): 253–281.
[9] Gu, H., Yue, J. (2011). The Relationship between Bank Efficiency and Stock Returns: Evidence from Chinese
Listed Banks, World Journal of Social Sciences, 1 (4): 95-106.
[10] Kisielewska, M., Guzowska, M., Nellis, J. G., Zarzecki, D. (2005). Polish banking industry efficiency: DEA
window analysis approach. International Journal of Banking and Finance, 3 (1): 15-31.
[11] Palečková, I. (2015). Efficiency Change in Banking Sectors of Visegrad Countries. In: Proceedings of the 7th
International Scientific Conference Finance and the Performance of Firms in Science, Education, and Practice,
1153-1168.
[12] Řepková, I., (2014a). Efficiency of the Czech banking sector employing the DEA window analysis approach. In:
Procedia Economics and Finance, 12: 587-596.
[13] Řepková, I. (2014b). Efficiency of the Slovak Commercial Banks Applying the DEA Window Analysis. World
Academy of Science, Engineering and Technology, 8 (5): 470-475.
[14] Savić, G., Radosavljević, M., Ilievski, D. (2012). DEA Window Analysis Approach for Measuring the Efficiency
of Serbian Banks Based on Panel Data, Management Journal for Theory and Practice Management, 65: 5-14.
[15] Yue, P. (1992). Data Envelopment Analysis and Commercial Bank Performance: A Primer with Applications to
Missouri Banks, Federal Reserve Bank of St. Louis Review, 74(1): 31-45.
[16] Zimková, E. (2014) Window Analysis of Technical Efficiency: Case of the Slovak Banking System. In: 17th
Applications of Mathematics and Statistics in Economics, International Scientific Conference, 294-300.

110
MEASURING EFFICIENCY OF SALES TERRITORIES FOR MEDICAL
DEVICES IN BRAZIL THROUGH DATA ENVELOPMENT ANALYSIS (DEA)
RENATO BARBOSA DA CUNHA
Universidade Federal do Rio Grande do Sul (UFRGS), Porto Alegre, Brazil (corresponding Author)

DENISE LINDSTROM BANDEIRA


Universidade Federal do Rio Grande do Sul (UFRGS), Porto Alegre, Brazil

CAMILO JOSÉ BORNIA-POULSEN


Universidade Federal do Rio Grande do Sul (UFRGS), Porto Alegre, Brazil

ABSTRACT
Medical devices market has almost unavailable information of customer demands, markets forecasts or estimated
market share. Through the analysis of 43 sales territories of medical devices companies in Brazil, this study uses the
data envelopment analysis (DEA) technique to create a model for comparative efficiency analysis that can assist the
decision-making processes, even with a restricted and limited database. The application of DEA technique provides
results that [1] highlight the points of improvement for inefficient territories [2] in order to homogenize the efficiency
of regions through [3] benchmarking between them, [4] to improve overall competitive level.

Keywords: Data envelopment analysis; Sales performance; Medical devices; Territorial management; Sales territory.

INTRODUCTION
Healthcare market offers significant growth potential for the coming decades due to both increased human
life expectancy and access to new technologies (Dall et Al, 2013). However, the prices pressure for cost
reduction is increasing (Grennan, 2012) and the regulatory requirements (Campos 2004) are becoming more
complex and expensive. This paper focuses on the medical devices market, composed by instruments
employed to treat diseases through physical action (Baio, 2011), limited to medical devices used for the
surgical treatment of bones and their pathologies, which include a wide range of products to attend three
medical specialties that have generated a great economic impact on the healthcare system (Oliveira, 2009):
maxillofacial (MF), neurosurgery and traumatology.
Sales force papers are not as abundant as its relative importance (Zoltners et al., 2008) and few information
was available regarding the medical devices market, efficiency and sales performance. Not a surprise in an
industry worldwide known for its lack of transparency in prices (Bridy, 2009), conflicts of interest (Brennan
et al., 2006), compliance issues (Brougher et al., 2011 ), and complexity in the triangulated relation between
surgeons, hospitals and healthcare providers, three customers with different interests influencing sales
(Burns et al., 2009).
Through defining the relevant inputs and outputs for a reliable model, this paper aims to measure the
efficiency of sales territories for medical devices in Brazil in three different medical specialties based on
public information, and a two-year history of sales.

METHODS
Data envelopment analysis (DEA), the mathematical model developed by Charnes, Cooper and Rhodes
(1978) was selected to evaluate performance of sales territories. The DEA grew as a powerful tool for

111
measuring relative efficiency of different peer entities named Decision Making Units (DMU). This
methodology provides a quantitative and nonparametric analysis of relative efficiency that aim to evaluate
each DMU individually through multiple comparisons. As result, an efficiency frontier is created to
compare all DMUs, pointing out the determinant variables of each unit. In the first concept, DEA was a
management tool focused on productivity and DMUs were production units with a certain capacity to
transform resources (input) into results (output). Managers were able to plan the required resources for each
one, and define targets in order to maximize efficiency. Currently DEA has also been used as a performance
evaluation tool and best practices benchmarking. In these circumstances, DEA does not necessarily create
a production efficiency frontier, but rather a border of best practices (Cook et al., 2013).
There are two traditional models of DEA. The CCR model developed by Charnes, Cooper and Rhodes in
1978 considers Constant Returns to Scale (CRS) and allows an objective evaluation of overall efficiency.
The BCC model developed by Banker, Charnes and Cooper in 1984, considers Variable Returns to Scale
(VRS), identifying increasing or decreasing economies of scale.
An output-oriented CRS were selected once a duplication of inputs results in a proportional duplication of
outputs, optimizing the resources used (inputs) for maximum delivery (outputs). A total of 43 sales
territories representing 60% of GDP, 46% of the population and 31% of the cities in Brazil are the
equivalent to the DMUs. The DMUs were defined through a region concept molded by historical, economic
and social factors (Breitbach, 1988).
Several configurations with different inputs and outputs were tested through Frontier Analyst software© to
find a specific model for this paper. The best template was selected after an analysis and comparison of all
available data as recommended by Arenales et al. (2007).

Table 1: Input / Output validation


Model Uncontrolled inputs Controlled inputs Outputs Result
Input / Output Population Qty_salesmen by Sales by specialty and Not valid, total
Validation 1 GDP specialty by payer potential
By specialty and HDI improvement
by payer Beds by specialty and distortion
by payer
Input / Output Population Total_qty_salesmen Sales by payer Not valid, total
Validation 2 GDP potential
By payer HDI improvement
Beds by payer distortion
Input / Output Population Qty_salesmen by Sales by specialty Valid, consistent
Validation 2 GDP specialty potential
By specialty HDI improvement in each
Beds by specialty specialty

Some assumptions were made. The surgical demand is a percentage of the population in each territory.
Therefore, the healthcare needs are proportional to number of people. It is also assumed that poorer
territories require more investment from the public healthcare system, while the richer territories, measured
by Gross Domestic Product (GDP), transfer the demand from the public to the private sector. The surgery
capacity of each DMU is proportional to the sum of the beds available in each specialty. Medical clusters
receive a greater flow of patients and shall have different performance. As inputs, the following were

112
considered: population, GDP, regional average Human Development Index (HDI), number of beds, and
number of sellers and if DMU is or is not a medical cluster. Sales performance was the only output available.
First simulations in Frontier Analyst software© validated inputs and outputs as shown in Table 1. The
potential improvements were analyzed considering their viability. A split between public and private payers
caused distortions and was void. The selected model was divided by medical specialty including private
and public beds in the same input. Private and public sales were included in the same output.
A second round of simulations was performed to validate uncontrolled inputs (population, GDP, HDI) and
a medical cluster binary variable as shown in Table 2. The best rational result was selected.

Table 2: Input validation


Input tested Model 1 Model 2 Output Result
Regional potential Population Sales by specialty Selected model 1,
GDP increased efficiency
HDI on 15 small DMUs
Beds by specialty Beds by specialty
Qty_Salesmen by Qty_Salesmen by
specialty specialty
Medical_cluster Medical_cluster
Medical cluster Population Population Sales by specialty Selected model 2
GDP GDP improved efficiency
HDI HDI for 8 small DMUs
Beds by specialty Beds by specialty (not medical cluster)
Qty_Salesmen by Qty_Salesmen by
specialty specialty
Medical_cluster

RESULTS AND DISCUSSIONS


The final DEA model is presented in Table 3. The 2012 period was also studied to compare it with 2013,
assessing not only the static result of the studied year, but also verified the changes that might be noticed
from one year to another.

Table 3: Final model


General configurations Uncontrolled inputs Controlled inputs Outputs
CCR Mode Population Qty_salesmen_MF Total_sales_MF
Maximize output GDP Qty_salesmen_neuro Total_sales_neuro
All variables 100% weight HDI Qty_salesmen_trauma Total_sales_trauma
43 DMUs Total_beds_MF
02 years analysis Total_beds_neuro
Total_beds_trauma
Medical_cluster

In 2012, 31 DMUs were efficient, three DMUs were in the intermediate classification - between 81% and
99.9%, and nine were classified below 80% efficiency. In the focus year of 2013, 32 DMUs were within
the efficient frontier, five DMUs were in the intermediate classification, and six were classified below 80%.
Nine DMUs improved their relative efficiency from one year to another, while 26 DMUs remained 100%
efficient in both years. Eight DMUs worsened their classification. These results can be viewed in detail at
https://dl.dropboxusercontent.com/u/84981398/DEA/Appendice.pdf.

113
From 2012 to 2013 there was an increase in the efficient regions, which makes sense since the sales force
size was reduced by 3.33% from 2012 to 2013, while sales rose 19.32% in the same period. This decrease
in sales force size reflects a search for better performance and a growing concern about costs in the medical
industry, which can be confirmed by Buzzo (2010). Figure 1 shows the result of all inefficient DMUs in
2013.

100%
80%
60%
40%
20%
0%

Score 2012 Score 2013

Figure 1: 2013 inefficient DMUs

Each DMU was analyzed against the target of potential improvements and the differences from the previous
year of 2012. This potential and the differences within the studied years were validated with the managers
responsible for the territories.

Figure 2: 2013 Total Potential Improvements

As the DEA is a linear combination of 100% efficient departments with its inefficient pairs, the efficiency
parameters changed from year 2012 to 2013. When analyzing the total potential for improvement in Figure
2, the medical specialty with the greatest sales growth potential is neurosurgery with 34.67%, followed by
traumatology with 15.97%, and 9.80% for maxillofacial. There are targets for sales force reduction of

114
5.41% in traumatology, 5.16% and 0.95% on BFM and neurosurgery, respectively. The results were
discussed with corporate executives who confirmed the problems in inefficient territories, loss of public
bids and deficiencies in sales teams. The large potential sales improvement in neurosurgery was also
validated but impossible to achieve due to frequently unethical competition in this segment.

CONCLUSIONS
DEA application proved to be a suitable tool to compare sales force teams distributed in territories with
different sizes, development levels, health care availabilities and economic power. It points out where each
DMU can spend efforts to increase efficiency and improve sales. The market potential variables –
population, GDP and HDI – seem to have an important influence in defining, in a more accurate way, the
size of the market in a study where disparities are as great as those found in this paper. The classification
of medical clusters prevented small territories from being considered inefficient when compared with
territories with high concentration of complex surgeries and larger surgical volumes.
As a quantitative multivariate analysis, the result of DEA is directly affected by the chosen variables. The
model applied in this paper seems to be a coherent and rational option for the development of a management
support tool, but not meant to be presented as definitive. This model can be an initial step to a more detailed
performance analysis in this industry, once future researches can obtain more variables. This study also
shows the potential of DEA as a performance analysis tool for regional distribution of sales force in the
medical device market.
It would be interesting to develop more detailed evaluations in future studies, including fixed costs, selling
costs, marketing, personnel, operations and investments in infrastructure and local warehouses, which
would allow not only for a comparison of sales analysis and the number of salesmen, but also for the
efficiency of the resource allocations for each DMU. As this market is focused on service delivery, it would
also be interesting to include the support and logistics costs and instrumentation costs, which have great
influence on sales and customer service.
Other research possibility could focus only on medical clusters. These important health care centers are
those with a potential for further performance gains due to its characteristics of specialties concentration.
There is also a possibility for further studies in other medical specialties who demands a similar sales
service.

REFERENCES
[1] Arenales M.N., Armentano V., Morabito R., Yanasse, H. (2007) Pesquisa operacional para cursos de
engenharia. Rio de Janeiro: Elsevier.
[2] Baio J.M.C. (2011) A importância da despesa em dispositivos médicos nos sistemas de saúde. Coimbra, 134 f,
Dissertação (Mestrado), Curso de Economia, Faculdade de Economia da Universidade de Coimbra, Coimbra,
Portugal.
[3] Banker R.D., Charnes A., Cooper W.W. (1984) Some models for estimating technical and scale inefficiencies in
Data Envelopment Analysis, Management Science 30 (9): 1078-1092.
[4] Breitbach A.C.M. (1988) Estudo sobre o conceito de região, Porto Alegre: Fundação de Economia e Estatística
Siegfried Emanuel Heuser.

115
[5] Brennan T.A., Rothman D.J., Blank L., Blumenthal D., Chimonas S.C., Cohen, J.J., Goldman, J., Kassirer J.P.,
Kimball H., Naughton J., Smelser, N. (2006) Health industry practices that create conflicts of interest: A policy
proposal for academic medical centers, The Journal of American Medical Association 295 (4): 429-433.
[6] Bridy A. (2009) Trade secret prices & high-tech devices: How medical device manufacturers are seeking to
sustain profits by propertizing prices, Texas Intellectual Property Law Journal 17: 187-222.
[7] Brougher J., Gaba, M.M., Deem, M.E., Termini, R.B., Auchincloss, K. (2011) A practical guide to navigating the
medical device industry: Advice from experts in industry, law, intellectual property and academia, The Food and
Drug Law Institute Monograph Series 2 (4): 95.
[8] Burns L.R., Housman M.G., Booth R.E., Koenig A. (2009) Implant vendors and hospitals: Competing influences
over product choice by orthopedic surgeons, Health Care Management Review 34 (1): 2-18.
[9] Buzzo E.J. (2010) Composição da força de vendas e informação comercial: Um estudo em indústrias do setor
farmacêutico, médico e hospitalar do estado de São Paulo, Curso de Pós-graduação em Administração de
Organizações, Administração, USP, Ribeirão Preto.
[10] Campos C.C. (2004) Um estudo das relações entre operadoras de plano de assistência à saúde e prestadores de
serviço, 186 f, Dissertação (Mestrado), Curso de Engenharia, Escola de Engenharia, UFRGS, Porto Alegre.
[11] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[12] Cook W.D., Tone K., Zhu J. (2013) Data envelopment analysis: Prior to choosing a model, The International
Journal of Management Science (44): 1-4.
[13] Dall T.M., Gallo P.D., Chakrabarti R., West T., Semilla A.P., Storm M. (2013) The care span: An aging
population and growing disease burden will require a large and specialized health care workforce by 2025, Health
Aff 32 (11): 2013-2020.
[14] Grennan M. (2013) Price discrimination and bargaining: Empirical evidence from medical devices, American
Economic Review 103 (1): 145-77.
[15] Oliveira D.S. (2009) Auditoria Especializada em OPME. In: Gonçalves, V.F. et al. Fronteira da Auditoria de
Saúde, São Paulo: Farol do Forte: 179-188.
[16] Zoltners A.A., Sinha P., Lorimer S.E. (2008) Sales force effectiveness: A framework for researchers and
practitioners, Journal of Personal Selling and Sales Management 28 (2): 115–131.

116
MEASURING ENVIRONMENTAL EFFICIENCY OF THE 29 CHINESE
PROVINCES: APPLICATION OF THREE-STAGE DEA
HAILING ZHAO
Department of Economics, Pusan National University, South Korea (corresponding Author)

ABSTRACT
This study applies a three-stage DEA to measure the environmental efficiencies of the 29 Chinese provinces from 2003
to 2012. Compared with the three-stage DEA, which was introduced by Fried et al (2002) [1], the Slacks- Based
Measure with undesirable outputs is employed in the first and the third stage to measure the environmental efficiencies
of the 29 Chinese provinces, respectively. The results show that there is a significant difference between environmental
efficiency in the first stage and that in the third stage. The environmental efficiency of each province is improved after
excluding environmental influences and random shocks. Compared with the results obtained in the first stage, the
average of entire environmental efficiency increases by about 8%. Also, there was a dramatic increase in
environmental efficiency of the eastern, central, and western areas in 2012. The differences in the environmental
efficiency figures amongst each region decrease after adjusting the data. Also, the ten year study shows that the
influence factors have a remarkable effect on the slacks of the GDP and exhaust emissions.

Keywords: Chinese environmental efficiency; Three-stage DEA; Regional Difference; Undesirable outputs;

INTRODUCTION
The Chinese economic reform is a program of economic reforms named ‘Socialism with Chinese
characteristics’ and was implemented in December 1978 by Chinese government. With the successful
implementation of the economic reforms in China, the economy has been rapidly developing since then,
even exceeding the East Asian Dragon. However, with rapid economic development, environmental
problems have been becoming more and more serious in China. There are plenty of environmental issues
existed in China, severely influencing the biophysical environment and human health. According to the
World Bank in 2013[3], 16 of the world’s 20 most polluted cities are in China and China is the world’s
second-largest emitter of carbon dioxide, the main gas linked to global warming, and it is set to overtake
the United States in 2009. Faced these serious threats to economic development and residents’ health,
Chinese government has to further formulate the environmental protection regulations and enhance the
energy use efficiency. Also, Chinese environmental protection funds keep rising year after year. Expected
investment in the environmental protection funds in the Eleventh Five-year Plan was 1540 billion Yuan2
and that in the Twelfth Five-year Plan increased by 121%, attaining 3100 billion Yuan. It would be 17000
billion Yuan in the Thirteenth Five-year Plan. Although the investment in the environmental protection
funds continues to increase, deterioration of the environment has been becoming more and more serious in
China, and it constrains economic and social development.
The aim of this paper is not only to measure the environmental efficiency of the 29 Chinese provinces, but
also to measure which determinants have an impact on environmental efficiency and to what extent these
determinants influence accuracy of environmental efficiency measurement. Through adjusting statistic
data, the efficiency obtained by three-stage DEA is more close to reality.

2
Yuan is Chinese currency symbol.

117
METHODS
A three-stage DEA was introduced by Fried et al. (2002) [1], which distinguishes environmental effects
such as the state of the economy, from both types of inefficiencies and random errors. Slacks-Based
Measure (SBM), which proposed by Tone (2001) [2], is employed in the first stage and in the third stage
instead of using BCC model.
First Stage: The Initial Environmental Efficiency Evaluation
The SBM model with undesirable output model is shown below.

1 m si
1 
m i 1 xi 0
 *  min (1)
1  s1 srg s2 sr 
u
1    r 1 yu 
s1  s2  r 1 yrg0 r0 

Subject to x0  X   s 
y0g  Y g   s g
(2)
y0u  Y u   s u
s   0, s g  0, s u  0,   0

Second Stage: Using SFA to Separating Environmental Effects and Statistical Noise
According to the method was proposed by Fried et al. (2002), total Stage 1 slacks should be calculated,
which can reflect initial managerial inefficiency.

sni  f n  zi ;  n   vni  uni , n  1, , M  S1  S2


(3)

f n  zi ;  n 
are deterministic feasible slack frontiers with parameter vectors  .
n
Where the
Then, initial inputs and outputs should be adjusted from the results of the Stage 2 SFA regressions. We use
the formula, which was introduced by Tone and Tsutsui (2009), to adjust inputs and outputs instead of
applying the one shown by Fried et al. First, it employs the formula for adjustment with no recourse to max
or min as follows:

xijA  xij  z ij  i  vij


(4)

yrjA  yrj  z rj  r  vrj


(5)

xijAA yrjAA
Second, it re-adjusts them into and using the following formula.

118
xi max  xi min A A
xijAA  
xiAmax  xiAmin

xij  xi min  xi min , i  1,..., m; j  1,..., n  (6)

y y
 
yrjAA  rAmax rAmin yrjA  yrAmin  yr min ,
yr max  yr min
 r  1, , s; j  1, , n  (7)

Where

xi max  max k {xik }, xi min  min k {xik }, xiAmax  max k {xikA} and xiAmin  min k {xikA}.

yr max  max k { yrk }, yr min  min k { yrk }, yrAmax  max k { yrkA } and yrAmin  min k { yrkA }.

Third Stage: Adjusted Environmental Efficiency Evaluation


In the final stage, we feed the adjusted inputs and outputs into the same SBM model to measure efficiency
scores. The new environmental efficiency obtained in this stage can reflect the real situation without the
influence of environmental factors and statistical noise.

RESULTS AND DISCUSSIONS


There are 34 provinces in China, classified as 23 provinces, four municipalities, five autonomous regions,
and two special administrative regions. Owing to the missing data of some provinces and regions, only 29
provinces are used in this paper. Moreover, three inputs, consisting of labor, capital stock, and energy
consumption, and two outputs, consisting of GDP and exhaust emissions, are included in the model. At the
same time, the per capita GDP, proportion of secondary industry, units of GDP energy consumption,
environmental regulations, and trade openness are treated as influence factors.

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Entire Mean 0.6687 0.692 0.6975 0.7009 0.6881 0.7191 0.7223 0.7313 0.7269 0.7146
Mean of the Eastern Region 0.8255 0.8487 0.8367 0.8406 0.8126 0.834 0.8331 0.8425 0.8445 0.8057
Mean of the Central Region 0.6384 0.6785 0.7123 0.7253 0.7161 0.7601 0.7747 0.7829 0.7462 0.775
Mean of the Western Region 0.4677 0.472 0.4722 0.464 0.4699 0.5006 0.4971 0.5064 0.5286 0.51

Figure 1: Efficiency trend of each region in 1st stage


As it shows in fig 1, on the whole, entire efficiency mean of the 29 Chinese provinces (0.7061) locates at
relatively higher level. Efficiency mean of the eastern region is the best with the value of 0.8324, whereas

119
that of the western region is the worst with the value of 0.4889. Gap of overall efficiency means between
the eastern region (0.8324) and the western region (0.4889) is quite big. Moreover, the overall efficiency
trend of each region slightly fluctuates in a small range.
Table 1 indicates regression results of slack equations. Based on the results obtained in the first stage, all
of Chinese provinces selected do not have any slack of GDP, which means each Chinese province can
produce optimal GDP under the consumption of using excess inputs and of excess exhaust emission.
Consequently, this paper does not include GDP slack regression function.
As is shown in table 1, per capital GDP has a negative impact on labor slack and it is significant at the level
of 0.1%, which means that the labor slack decreases with per capital GDP increases. Environmental
regulations significantly and positively affect labor slack as well, which means increase on investment to
pollution abatement will generate more serious excess labor. Moreover, capital stock slack is only
significantly influenced by per capital GDP, and capital stock slack also declines with per capital GDP
rises. Per capital GDP, industrial structure and environmental regulations have a significant effect on energy
consumption slack. Per capital GDP has a negative influence on energy consumption.

Table 1: SFA Parameter Estimates of Slack Equations


Labor Slack Capital Stock Slack
Coefficient Std. Err. z-value Coefficient Std. Err. z-value
Per Capital GDP -0.2029*** 0.0518 -3.92 -0.1051*** 0.0261 -4.02
Industrial Structure 0.0378 0.0879 0.43 0.0497 0.0581 0.86
Unites of GDP Energy Consumption 0.082 0.0485 1.69 -0.0876 0.0452 -1.94
Environmental Regulations 0.0487* 0.0226 2.15 0.0228 0.0181 1.26
Trade Openness -0.0377 0.0228 -1.65 0.0214 0.0144 1.48
mu 2.1831 1.9337 1.13 -0.0541 0.424 -0.13
eta -0.0138 0.0126 -1.1 -0.0065 0.0094 -0.69
sigma2 0.0414 0.0094 0.1371 0.1148
gamma 0.7783 0.0529 0.9564 0.0367
Log likelihood 218.4941 273.6284
Energy Consumption Slack Exhaust Emissions Slack
Coefficient Std. Err. z-value Coefficient Std. Err. z-value
Per Capital GDP -0.1706*** 0.0279 -6.12 0.0284 0.0303 0.94
Industrial Structure 0.1476** 0.0506 2.92 -0.052 0.082 -0.63
Unites of GDP Energy Consumption 0.0374 0.0273 1.37 0.2410*** 0.0498 4.84
Environmental Regulations 0.0320** 0.0113 2.84 0.0528** 0.0212 2.49
Trade Openness -0.0013 0.0134 -0.1 -0.0143 0.0234 -0.61
mu 0.2406*** 0.2406 5.48 0.3341** 0.1209 2.76
eta 0.0058*** 0.0058 -4.91 -0.0012 0.0076 -0.16
sigma2 0.0506 0.0134 0.0882 0.0422
gamma 0.9576 0.0118 0.9138 0.0423
Log likelihood 404.2502 235.3467
Note: *, **, *** mean significance at the level of 5%, 1%, and 0.1%, respectively.

Industrial structure and environmental regulations positively affect energy consumption slack, which
illustrates that increases on the proportion of secondary industry and investment to pollution control
remarkably grow energy consumption slack, and then make provinces to develop under the unfavorable
operation circumstance. At the same time, eta in this equation is very significant with the value of 0.0058,
which indicates energy consumption slack decreases at an increasing rate with time change. Mu (0.2406)
is also significant in this equation, and it means that random error mainly contributes to energy consumption

120
slack. Exhaust emission slack is significantly and positively affected by unit of GDP energy consumption,
which mirrors that growth on both variables can augment exhaust emissions.
After adjusting the original inputs and outputs, new efficiencies of the 29 Chinese provinces, which
eliminate environmental influences and random shocks, are obtained by the same method used in first stage.
Entire efficiency average is relatively higher with the value of 0.7870. The entire efficiency average of the
eastern region is the best relative to the other regions, whereas that of the western region is still worst. In
addition, Fig 2 shows that there is no big gap of the entire efficiency averages between the eastern region
and the central region, and the entire efficiency averages of both regions are greater than that of all of
provinces. There is still a large efficiency difference between the eastern and central region and the western
region, which means technology development in the western region lags far behind the eastern and central
region. However, the entire efficiency mean of the western region suddenly and dramatically rose in 2012,
and those of three regions converged at a point. Difference at economic development level is the main
reason making the efficiency gap between the eastern region and the western region.

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Entire Mean 0.7603 0.7722 0.7752 0.7696 0.7832 0.7878 0.7703 0.7718 0.7598 0.9195
Mean in eastern region 0.8625 0.8691 0.8749 0.8371 0.8581 0.8487 0.8555 0.8596 0.8459 0.9405
Mean in central region 0.784 0.7906 0.7949 0.8056 0.8177 0.8211 0.8121 0.8148 0.7977 0.9181
Mean in western region 0.5804 0.6061 0.6034 0.6277 0.6322 0.6591 0.5955 0.5917 0.5881 0.8897

Figure 2: Efficiency trend of each region in 3rd stage

Those provinces in the eastern region has economically developed fast since the reform and opening-up of
China, and GDP produced by them accounts for approximately 50% of total China’s GDP. Nonetheless,
compared with the other two regions, the western region is the most under-developed region, and only
account for 21.24% of total population of China but covers over 50% of territory. Due to limitation on
natural environment and geographical location, economic development in the western region is at a very
slow speed and industrial structure is seriously unbalanced. Hence, difference at level of economic
development and structure imbalance are contributed to a long-term efficiency gap between each region.
However, the development of the western region in China has been operated since year 2000. Transfer
payment from the exchequer to the western region has been accumulated to RMB 1500 Billions from 2000
to 2007, which improves the whole environment. Improvement on public infrastructure, ecological

121
environment and trade policy is mainly contributed to efficiency increase of the western region. Efficiency
of the central region ranks in the middle. All of provinces in the western region are landlocked province,
and has a huge number of populations. The economic development, advanced technology import and
foreign capital investment in the central region are less than those in the eastern region but much more than
those in the western region. Because the heavy industry and agriculture are pillar industry in the central
region, amount of energy consumption and amount of exhaust emissions are also very large. However,
increase on operation cost in the eastern region and improvement on transportation network cause domestic
and oversea capital transfer from the eastern region to the central region. Therefore, the efficiency gap
between the eastern region and the central region is not as large as before.

CONCLUSIONS
This study employs three-stage DEA to measure environmental efficiencies of the 29 Chinese provinces
and influence of environmental factors on slacks of inputs and outputs during the period of 2003-2012, and
analyzes environmental efficiencies of the 29 Chinese provinces in the first stage and the third stage. On
the whole, the overall environmental efficiency of China locates at a relatively lower level with the value
of 0.7061 in the first stage. There is a small gap of the overall environmental efficiencies (about 10%)
between the eastern region and the central region, while very large gap of those (about 34%) exists between
the eastern region and the western region. In addition, there is a relatively stable trend on environmental
efficiencies of the eastern region during the sample period and the scores fluctuate between 0.8057 and
0.8487. Whereas, there are slow upward tendencies on environmental efficiencies of the central region and
the western region, respectively. In addition, according to the regression results, the per capital GDP has a
negative effect on labor slack, capital stock slack and energy consumption slack at significance level of
0.1%. Environmental regulations remarkably influence labor slack, energy consumption and exhaust
emissions, respectively. At the same time, exhaust emissions are also notably affected at the significance
level of 0.1% by unites of GDP energy consumption. In the third stage, the overall environmental efficiency
of China increases by 8% compared with the first stage, which proves that the environmentally technical
efficiency was underestimated when provinces perform under different operation circumstances. Finally,
overall environmental efficiency in the eastern region only increases by 3.28%, while that in the central
region and the western region has dramatically increased by 8.47% and 14.85%, respectively. Compared
with the overall environmental efficiency trend of each region in the first stage, that in third stage is more
stable and fluctuates in a small range. However, there is a very different point with the first stage. In the
third stage, the overall environmental efficiencies of three regions in 2012 have converged at a point.

REFERENCES
[1] H. O. Fried, C. A. K. Lovell, S. S. Schmidt, S. Yaisawarng. (2002) Accounting for environmental effects and
statistical noise in data envelopment analysis, Journal of Productivity Analysis 17: 157-174.
[2] Tone, K. (2001) A slacks-based measure of efficiency in data envelopment analysis, European Journal of
Operational Research 88(3): 559-586
[3] Air pollution grows in tandem with China’s economy, cited from
http://www.npr.org/templates/story/story.php?storyId=10221268.

122
MEASURING THE MUTUAL FUND PERFORMANCE IN DATA
ENVELOPMENT ANALYSIS
HAOCHEN GUO
VŠB-Technical University of Ostrava, Faculty of Economics, Sokolská třída 33, 701 21 Ostrava, Czech Republic

ABSTRACT
This paper presents using Data Envelopment Analysis (DEA) to evaluate the performance and efficiency of mutual
funds. The proposed DEA integrated risk assessment and ranking method for mutual fund investment. Mutual fund is
a professionally-managed type of collective investment scheme that pools money from many investors to buy stocks,
bonds, short-term money market instruments and other securities. Even there are many advantages for mutual fund
investment also exits opposite point of view. Like fees, less control over timing of recognition of fain, less predictable
income, no opportunity to customize and so on. Hence, the evaluation of the performance of mutual funds by DEA
ranking is eventful for both invest and investor of financial institutions, banking and investment institutions. DEA is
one nonparametric approach which paper applies in return risk ratios as measurement of decision making units
(DMUs) for mutual fund ranking. The example of discusses result presents the performance and efficiency of selected
mutual fund investment in financial markets.

Keywords: Data Envelopment Analysis (DEA), Efficiency, Mutual Fund, Performance

INTRODUCTION
Mutual Fund is a professionally-managed type of collective investment scheme that pools money from
many investors to buy stocks, bonds, short-term money market instruments, and or other securities.
Compared to direct investing in individual securities, advantages of mutual fund are increased
diversification, daily liquidity, and professional investment management, ability to participate in
investments that may be available only to larger investors, service and convenience, government oversight,
ease of comparison. Opposite point of view, mutual fund also have disadvantages, like fees, less control
over timing of recognition of fain, less predictable income, no opportunity to customize. That’s the reason
why fund risk management is very important. At present, mutual fund is becoming one of the primary
vehicles which invest in financial markets. Hence, the evaluation of the performance of mutual funds by
ranking is eventful for both invest and investor of financial institutions, banking and investment institutions.
This paper presents the theory of fund risk management and application of ranking method for mutual fund.
The proposed data envelopment analysis (DEA) integrated risk assessment and ranking method for mutual
fund investment. Data envelopment analysis is one nonparametric approach which applied in return-risk
ratios for mutual fund ranking. The result is discusses and show how efficient mutual fund which invest in
financial markets.
The foundation of this paper is discussion of mutual fund risk management, which including determination
of mutual fund, hedging strategies of fund risk management and measurement of mutual fund risk. Then
defines the data envelopment analysis integrated ranking method. It describes the data envelopment analysis
and return to scales of VRS model and CRS model. The aims of practical illustration is uses data
envelopment analysis on ranking mutual funds, which is based on main indicators of investment risk such
as alpha, beta, expected return, r-squared, standard deviation and Sharpe ratio.

123
DISCUSSION OF MUTUAL FUND RISK MANAGEMENT
Foundation of Mutual Fund
Determine to say that mutual fund is a company that pools money from many investors and invests the
money in stocks, bonds, short-term money-market instruments, other securities or assets, or some
combination of these investments. The combined holdings the mutual fund owns are known as its portfolio.
Each share represents an investor's proportionate ownership of the fund's holdings and the income those
holdings generate. See [U.S. Securities and Exchange Commission]
Every investment has advantages and disadvantages. But it’s depends on investors unique circumstances.
Hence, the mutual fund risk management is very important in financial markets. For example, following
Figure 1 shows until 2013, the size of US mutual funds accounted for half the world. That means mutual
funds becoming the main financial instruments, the largest institutional investors in the US financial market.
There are nearly half of households invest in mutual funds in US.

Figure 1 the size of US mutual funds


Source: International Investment Funds Association, 2014 Investment Company Fact Book

Advantages of mutual funds: Professional management – professional money managers research, select,
and monitor the performance of the securities the fund purchases; Diversification – spreading investments
across a wide range of companies and industry sectors can help lower risk if a company or sector fails;
Affordability – some mutual funds accommodate investors who don’t have a lot of money to invest by
setting relatively low dollar amounts for initial purchases, subsequent monthly purchases, or both; Liquidity
– mutual fund investors can readily redeem their shares at the current NAV plus any fees and charges
assessed on redemption at any time.
Disadvantages of mutual funds: Costs despite negative returns – investors must pay sales charges, annual
fees, and other expenses regardless of how the fund performs; Lack of control – investors typically cannot
ascertain the exact make-up of a fund’s portfolio at any given time, nor can directly influence which
securities the fund manager buys and sells or the timing of those trades; Price uncertainty – with an
individual stock, it can obtain real-time pricing information with relative ease by checking financial
websites or by calling broker.
Therefore, the advantages and disadvantages for investors that doesn’t matter what you needs.

124
Measurement of Mutual Fund Risk
According to the modern portfolio theory, the statistical measures of alpha, beta, r-squared, standard
deviation and Sharpe ratio, sortino ratio and treynor ratio are main indicators of investment risk that apply
to the analysis of stocks, bonds and mutual fund portfolios. All of these risk measurements are intended to
help investors determine the risk-reward parameters of their investments.

 Alpha – is a measure f an investment’s performance on a risk-adjusted basis.

 Beta – beta coefficient is a measure of the volatility (systematic risk) of a security or a portfolio in
comparison to the market as a whole.

 R-Squared – is a statistical measure that represents the percentage of a fund portfolio’s or security’s
movements that can be explained by movements in a benchmark index, it is the relationship between a
portfolio and its benchmark. It can be thought of as a percentage from 1 to 100.

 Standard Deviation – measures the dispersion of data from its mean. Investors use the standard
deviation of historical performance to try to predict the range of returns that are most likely for a given
fund. When a fund has a high standard deviation, the predicted range of performance is wide, implying
greater volatility.

 Sharpe Ratio – measures risk-adjusted performance was developed by Nobel Laureate William Sharpe.
It is calculated by using standard deviation and excess return to determine reward per unit of risk. The
higher the Sharpe ratio, the better the fund’s historical risk-adjusted performance.

 Sortino Ratio – a variation of the Sharpe ratio, differentiates harmful volatility from volatility in general
by using a value for downside deviation. The Sortino ratio is the excess return over the risk-rate divided
by the downside semi-variance, and so it measures the return to “bad” volatility. The Sortino ratio can
help an investor assess risk in a better manner than simply looking at excess returns to total volatility,
as such a measure does not consider how often returns are positive as opposed to how often they’re
negative.

 Treynor Ratio – measurement of efficiency utilizing the relationship between annualized risk-adjusted
return and risk. Unlike Sharpe ratio, Treynor Ratio utilizes “market” risk (beta) instead of total risk
(standard deviation). Good performance efficiency is measured by a high ratio.

DATA ENVELOPMENT ANALYSIS INTEGRATED RANKING METHOD


Data envelopment analysis (DEA) is a nonparametric analysis technique proposed by Charnes et al. (1978),
who used it to measure the performance of educational institutions. DEA is used in contexts where different
decision-making units (DMUs) can be measured by several input measures and several output measures.
To measure the efficiency score’s ratio utilized by following equation:

Output ( y0 )
Efficiency (0 )  , (1)
Input ( x0 )

125
This ratio can be interpreted as a patient per input and output and it can be found that more output amounts
and less input amounts lead to more efficiency score.
General, there are six different models for return to scale. Figure 2 shows the example for return (mean)
and risk (standard deviation) in the period of investment.

Figure 2 comparison of different returns to scale


 Red point O-H: CRS-CCR model

 Red point A-G: VRS-BCC model

 Red point O-C-D-E-F-G: CCR-BCC model

 Red point A-B-C-D-H: BCC-CCR model


DEA models for mutual funds and the assumptions needed for them and shows a non-increasing returns to
scale model is appropriate.

Define xi ( f ) as the risk measure i on fund f (i  1,, n) and yr ( f ) as the return measure r on fund f (r  1,, s)
.
To minimize the efficiency  0 , the subject to

  j f j with  j  0 ( j  1,, n) .
n
(2)
j 1


n
j 1 . (3)
j 1

Portfolio possibility set F  R .

VRS Convexity: If ( x, y), ( x , y )  , then (tx  (1  t ) x , ty  (1  t ) y )  for t  (0,1)


' ' ' '

 CRS Convexity: If ( x, y)  , then (tx, ty)  for t  0

NIRS Convexity: If ( x, y), ( x , y )  , then (tx  ux , ty  uy )  for t , u  0, t  u  1


' ' ' '

126
Initially DEA did not allow negative values for inputs and outputs. This is usually reasonable for economic
models but less so for models of funds. But negative return measures are largely unproblematic, and a NIRS
model with positive inputs and outputs unrestricted in sign is equivalent to one with positive inputs in which
any negative output value is replaced by zero. Negative or zero risk measures are more problematic (at least
in input-oriented and output-oriented models). It is also often reasonable to include a risk-free “reference
instrument” in a model of funds.

ILLUSTRATION OF DATA ENVELOPMENT ANALYSIS ON RANKING MUTUAL FUNDS


It use 3 years mean annual returns of 20 mutual funds with risk measures according to MPT statistics and
volatility measures’ data in Morningstar. Table 1 summarizes the main features of the fund returns. Table
2 shows rank results comparison between CCR model and BCC model.

Table 1 summary statistics and efficiencies scores of mutual funds


Outp
ut -
Input - Risk
Retur
n
Sta
Tre Mean
nda Sha Sort
3 R- yno Annu
TICKE rd rpe ino
Year FUND NAME CATEGORY Alpha Beta squ r al
R Dev Rati Rati
s ared Rati Retur
iatio o o
o n
n
DDDA 68.4 10.0 24.2
F1 13D Activist A Mid-Cap Growth 6.34 0.87 1.96 4.52 1.65
X 1 6 5
Jhancock Bond Intermediate-
F2 JBFRX 2.47 0.96 74.6 3.24 1.67 5.69 2.79 0.46
R4 Term Bond
Putnam
Target Date 2036- 95.5 10.4
F3 RetirementReady PRZZX 2.74 1.25 8.03 1.57 2.82 1.06
2040 8 2
2040 Y
Victory Integrity
MRVS 71.9 13.0 15.0
F4 Small-Cap Value Small Value -0.9 1.15 1.3 2.21 1.41
X 6 1 1
R
Virtus Emerging
VREM Diversified 65.0 13.5
F5 Markets -2.83 0.88 0.24 2.7 0.32 0.28
X Emerging Mkts 6 9
Opportunities R6
Fidelity Advisor
82.9
F6 Limited Term EFIPX Short-Term Bond 0.29 0.64 2.02 1.1 3.5 1.72 0.19
8
Bond I
AMG Chicago
MBEA Moderate 83.6 12.1
F7 Equtiy Patners 3.45 0.88 6.06 1.72 3.77 0.87
X Allocation 8 7
Bal Inv
Touchstone
60.0 13.9 13.4
F8 Sands Capital CISGX Large Growth -2.21 1.13 1.09 1.91 1.27
7 3 9
Inst Gr
Franklin Federal
Muni National 98.5
F9 Tax-Free Income FAFTX -0.24 1.21 4.46 1.02 3.78 1.6 0.39
Long 1
Adv
Columbia Large 83.8 10.5
F10 CGFYX Large Growth 1.69 1.01 1.63 17.9 3.22 1.44
Cap Growth Y 2 5
86.8 10.9 19.9
F11 Smead Value R2 SVFKX Large Blend 3.55 1.06 1.82 3.88 1.67
9 5 3
FGMN Intermediate 74.8
F12 Fidelity GNMA -0.12 0.82 2.75 0.87 2.89 1.42 0.2
X Government 4

127
Alpine Global
84.7 10.2
F13 Infrastructure AIFRX World Stock 7.61 0.76 1.26 17.3 2.08 1.08
8 3
Instl
Thomburg CA
Muni Single State 87.2
F14 Limited-Term LTCIX 0.7 0.49 1.92 1.38 5.44 2.51 0.23
Short 9
Muni Instl
American Funds
Target Date 2016- 96.6
F15 2020 Trgt Date FAOTX 2.03 0.99 6.31 1.56 10.2 2.82 0.83
2020 5
Retire F1
Principal Global Global Real 36.7 11.9 20.6
F16 POSPX 6.64 0.69 1.17 1.95 1.17
Real Estate Sec P Estate 2 7 4
Nuveen
Symphony 51.0
F17 NFRIX Bank Loan 5.61 0.12 2.42 2.24 2.67 5.95 0.5
Floating Rate 4
Income I
Thomburg
Diversified 75.9
F18 Developing THDIX -0.51 0.91 13 0.44 5.58 0.62 0.49
Emerging Mkts 2
World I
11.0 19.2
F19 Janus Triton I JSMGX Small Growth 2.86 0.95 67.8 1.58 3.25 1.46
1 8
Mass Mutual
22.0
F20 Premier High DLHYX High Yield Bond 7.38 0.4 6.86 4.44 1.94 3.49 0.72
8
Yield Svc

Table 2 rank results comparison between CCR and BCC


CCR BCC
No. DMU Score Rank No. DMU Score Rank
1 F1 1 1 1 F1 1 1
4 F4 1 1 4 F4 1 1
5 F5 1 1 5 F5 1 1
8 F8 1 1 6 F6 1 1
11 F11 1 1 8 F8 1 1
16 F16 1 1 11 F11 1 1
17 F17 1 1 12 F12 1 1
20 F20 1 1 14 F14 1 1
19 F19 0.9897 9 16 F16 1 1
18 F18 0.9553 10 17 F17 1 1
10 F10 0.9432 11 18 F18 1 1
13 F13 0.8893 12 19 F19 1 1
7 F7 0.848 13 20 F20 1 1
3 F3 0.8046 14 13 F13 0.9855 14
15 F15 0.7871 15 10 F10 0.985 15
2 F2 0.7756 16 2 F2 0.96 16
14 F14 0.6139 17 9 F9 0.937 17
9 F9 0.5267 18 7 F7 0.9307 18
6 F6 0.509 19 3 F3 0.924 19
12 F12 0.4247 20 15 F15 0.9069 20

Funds are listed in the rank order given by this improved efficiency estimate. The highest of the scores of
1.00 (for example, Fund 1, 4, 5, 8, 11, 16, 17, 20) imply that these funds offer the best performance relative
to risk, compared to all other funds. Further down the ranking table, funds with scores less than 1.00 (for
example, Fund 2, 3, 7, 9, 10, 13, 15) implies inefficiency. In the ranking results, the higher up to the top in
the ranking of the mutual funds, the better is the funds in term of their risk and return performance and also
the better correlated is the fund with all other mutual funds under consideration.

128
Figure 3 mutual funds efficiency in CCR Figure 4 mutual funds efficiency in BCC

CONCLUSION
Data envelopment analysis (DEA) finds the optimal weights with the maximum efficiency score for each
organization. In this paper it provide an alternative ranking method, incorporating correlations of returns to
assess directly the relationship and outcomes of a portfolio of mutual funds of a variety of structures,
geographical locations and complex trading strategies. Measures of risk would be interest to fund’s
managers whose role is to allocate the best mix of funds that not only maximize returns, but also minimize
costs for its clients, with regards to the individual funds’ performance fees. The purpose of this paper is
based on the DEA framework to estimate the potential of funds’ performance, which means of a ranking
table to show how well the ranked funds correlate with other funds under considerations.

ACKNOWLEDGMENT
The research was supported by the SGS project of VSB‐TU Ostrava under no. SP2015/15. This paper has
been also elaborated in the framework of the Operational Programme Education for Competitiveness –
Project No. CZ.1.07/2.3.00/20.0296.

REFERENCES
[1] Alexander C. (2008) Market risk analysis volume III: pricing, hedging and trading financial instruments. John
Wiley & Sons, Ltd.
[2] Basso A., Funari S. (2001) Theory and methodology a data envelopment analysis approach to measure the mutual
fund performance, European Journal of Operational Research 135: 477-492.
[3] Branda M. (2013) Diversification-consistent data envelopment analysis with general deviation measures,
European Journal of Operational Research 226: 626-635.
[4] Chen Y., Liang L., Yang F., Zhu J. (2006) Evaluation of information technology investment: a data envelopment
analysis approach. Computers & Operations Research 33: 1368 – 1379.
[5] Cooper W., Seiford L., Tone K. (2007) Data envelopment analysis: a comprehensive text with models,
applications, references and DEA-Solver software. 2ed. Springer. ISBN 978-0387-45281-4.
[6] Elton. J. E., Gruber. J. M. (1991) Modern portfolio theory and investment analysis. John Wiley & Sons, Inc.
[7] Guo H. (2012) Managing of financial funds. MEKON 2012.
[8] Guo J., Ma C., Zhou Z. (2012) Performance evaluation of investment funds with DEA and higher moments
characteristics: financial engineering perspective. Systems Engineering Procedia 3: 209-216.

129
[9] Kao C. (2014) Network data envelopment analysis: a review, European Journal of Operational Research 239: 1-
16.
[10] Lamb. J. D., Tee K. (2012) Data envelopment analysis (DEA) integrated risk assessment technique on hedge
funds investment: theory and practical application. AcademyPublish.org – Risk Assessment and Management.
[11] Lamb. J. D., Tee K. (2012) Data envelopment analysis models of investment funds, European Journal of
Operational Research 216: 687-696.
[12] Paradi. J. C., Asmild M., Simak. P. C. (2004) Using DEA and worst practice DEA in credit risk evaluation.
Journal of Productivity Analysis 21: 153-165.

130
EFFICIENCY IN THE UNIVERSITIES OF MEXICO 2008-2012: AN ANALYSIS
THROUGH DYNAMIC-NETWORK DEA MODEL
CÉSAR L. NAVARRO-CHÁVEZ
Instituto de Investigaciones Económicas y Empresariales de la Universidad Michoacana de San Nicolás de Hidalgo,
Centro, Morelia Michoacán México.

ODETTE V. DELFÍN-ORTEGA
Instituto de Investigaciones Económicas y Empresariales de la Universidad Michoacana de San Nicolás de Hidalgo,
Centro, Morelia Michoacán México.

ABSTRACT
This research aims to measure the efficiency of universities in Mexico in 2008-2012, using Dynamic Network DEA
Model. For the realization of this measurement is presented an output oriented model with Variable Returns to Scale.
In the construction of Network DEA model, it was contemplated two nodes: teaching and research. In teaching node,
the inputs were: funding and professors and as output were contemplated graduates doctors. This output is used as
an intermediate link as an input to the next node that is research and the outputs for this node are: doctors in the SNI
and ISI publications. The carry-over for the dynamic model is the enrolment of doctors as input and as output:
postgraduates, so that these graduates are inputs next period. The scores of Network DEA in 2008-2012 period, show
that teaching node is who determines the efficiency of universities, because in 2008, there were 8 universities that
were effective in this node, and for 2012, the number has increased to 13 universities. The scores have a significant
difference to research node, where only one university was effective for 2008 and 4 universities for 2012. This reflects
a deficiency in the area of research for most universities in Mexico.

Keywords: Universities, DEA, Dynamic-Network

INTRODUCTION
Education has remained a major importance since the beginning of Mexico as an independent country, and
that has always been seen as a factor of production, as one of the main strategies to address the economic
problems of the country (Rodríguez, 2014). The educational policy of the country has the aim to raise the
quality of education at different levels, implementing programs to provide them with infrastructure,
improving teaching schedules, as well as the quality of their training and performance (Mayston & Jesson,
1988), (Ruggiero, 1996).
Higher education system in Mexico is characterized by its large size and diversity, integrated by
universities, technical institutes, technical universities, polytechnic universities, pedagogical universities,
intercultural universities, research centers, normal schools and specialized education center (SEP, 2014).
However, in 2011-2012 school year, the gross coverage rate of enrolment was only 29.5% in the school
modality and 32.8% of non-school mode. This means that only three of each ten young age, are enrolled in
educational programs of higher education (ANUIES, 2013).
This paper aims to use Data Envelopment Analysis (DEA) in a dynamic–network model to evaluate the
efficiency of the public universities in Mexico in the period 2008-2012. In this sense the working hypothesis
is that "teaching node is who determines the efficiency in the universities of México in the period 2008-
2012".

131
METHODS
Data Envelope Analysis (DEA) is a non-parametric methodology that evaluates the efficiency of decision-
makers units (DMUs), taking into account only information about the inputs and outputs of the production
or processing procedures they perform. Recently, DEA has extended to Dynamic Models and network
processes (Network DEA), allowing incorporate the internal structure of the DMUs, which corresponds to
the internal processes and their interrelationships in which any organization or industry is divided and it
can evaluate processes efficiency in different years using a dynamic model (Tone and Tsutsui, 2014).
Dynamic DEA models were introduced by Färe and Grosskopf (1996); Sengupta (1992, 1997, 1999); and
later working by Nemota and Gota (1999, 2003); Tone and Tsutsui (2010), among others. Network DEA
model started with Färe, Grosskopf (2000) too and continued with Lewis and Sexton (2003, 2004); Cook
WD, Liang L, Zhu J (2010); Kao (2009), Tone and Tsutsui (2009). Finally Bogetoft, Färe, Grosskopf,
Hayes and L. Taylor (2009) and Tone and Tsutsui (2014) developed DEA model using Dynamic with
network structure.
A dynamic model has the ability to analyze the way in which a set of variables behave in different time
periods. So the first period, is connecting with the next period through the intermediate variables called
carryovers, so that an output time 0 will be used as input in time 1. Also each period has its own
technological production and through the dynamic model can be seen the way a decision made in a period
of time, subsequent term impacts. Each DMU has its own inputs and outputs that has linked to the next
term t+1, by carryovers that are held over between time periods. The Carry-over activities are classified
into four categories as follows: Desirable (good), Undesirable (bad), Discretionary (free) and Non-
discretionary (fixed) (Tone and Tsutsui, 2010).
The network DEA model was developed to take into account the internal structure of DMUs using link
variables. It consists of a finite set of sub-technologies (or activities) that are connected to form a network.
The single-stage DEA models are based on thinking about production technology as a black box that
transforms inputs into outputs, Färe, Grosskopf (2000). This system has several processes or steps with
their own inputs and outputs and intermediate flows, in such situations, some components play important
roles in producing outputs through the use of intermediate outputs obtained from their previous components.
The Dynamic Network Model (DNM)
The DNM model takes into account the internal structure of a DMU, in which divisions are vertically
connected by links (intermediate products) (Bogetoft et al. 2009), in addition, consecutive periods are
horizontally connected by carry-overs.

It is presented a network structure where are used some inputs in one step: 𝑋 𝑡 ∈ 𝑅+ , to produce intermedia
outputs 𝑍 𝑡 ∈ 𝑅+

It is considered n DMUS (j=1,…,n) with K divisions (k=1,…,K) over T time periods (t=1,…,T) and let
to 𝑚𝑘 𝑎𝑛𝑑 𝑟𝑘 to be inputs and output of division K respectively. We define the main link of the
division k to the division h by (𝑘, ℎ)𝑙 (Tone and Tsutsui, 2014).

132
𝑡
𝑋𝑖𝑗𝑘 ∈ 𝑅+ (1)

(i=1,..,𝑚𝑘 ; j=1,...,n; k=1,...,K; t=1,...,T) is input resource i to DMUj for division k in period t,

𝑡 (2)
𝑦𝑖𝑗𝑘 ∈ 𝑅+

(i=1,...,𝑟𝑘 ; j=1,...,n; k=1,...,K; t=1,...,T) is output product i from 𝐷𝑀𝑈𝑗 , division k, in period t.
𝑚𝑘 and 𝑟𝑘 are the numbers of inputs and outputs to division k, respectively.
𝑡
Link is an intermediate product, which is an output from Division k and also an input to Division h 𝑧𝑗(𝑘ℎ)𝑖 ∈
𝑅+ (j=1,...,n; l=1,...,𝐿𝑘ℎ ; t=1,...,T), is linking intermediate products of 𝐷𝑀𝑈𝑗 from division k to division h
in period t, where 𝐿𝑘ℎ is the number of items in links from k to h
𝑡, 𝑡+1
𝑧𝑗𝑘𝑙 ∈ 𝑅+ (j=1,...,n; l=1,...,𝐿𝑘 ; k =1,...,K, t=1,...,T-1) is carry-over of 𝐷𝑀𝑈𝑗 , at division k, from period t
to period t+1, where 𝐿𝑘 is the number of items in the carry-over from division k.
Production possibility set composed of n DMUs (j = 1,…, n) consisting of K divisions (k = 1,…, K) over
T time periods (t = 1,…,T) (Tone and Tsutsui, 2013):

𝑡 𝑡, 𝑡+1 (3)
P= {𝑥𝑘𝑡 , 𝑦𝑘𝑡 , 𝑧(𝑘ℎ) , 𝑧𝑖𝑘 } is defined by

𝑥𝑘𝑡 ≥ ∑𝑛𝑗=1 𝑥𝑗𝑘


𝑡 𝑡
𝜆𝑗𝑘 (∀𝑘, ∀𝑡) (4)

𝑦𝑘𝑡 ≤ ∑𝑛𝑗=1 𝑦𝑗𝑘


𝑡 𝑡
𝜆𝑗𝑘 (∀𝑘, ∀𝑡) (5)

𝑡 𝑡
𝑧(𝑘ℎ )𝑙
= ∑𝑛𝑗=1 𝑧𝑗(𝑘ℎ 𝜆𝑡 (∀𝑙, ∀(𝑘ℎ)𝑙 , ∀𝑡) (as output from k in period t)
)𝑙 𝑗𝑘
(6)

𝑡 𝑡
𝑧(𝑘ℎ )𝑙
= ∑𝑛𝑗=1 𝑧𝑗(𝑘ℎ 𝜆𝑡 (∀𝑙, ∀(𝑘ℎ)𝑙 , ∀𝑡) (as input to h in period t)
)𝑙 𝑗ℎ
(7)

(𝑡, 𝑡+1) (𝑡, 𝑡+1) 𝑡 (8)


𝑧𝑘𝑙 = ∑𝑛𝑗=1 𝑧𝑗𝑘 𝜆𝑗𝑘 (∀𝑘𝑙 , ∀𝑘, 𝑡 = 1, … , 𝑇 − 1) (as carry-over from t)
𝑙

(𝑡, 𝑡+1) (𝑡, 𝑡+1) (9)


𝑧𝑘𝑙 = ∑𝑛𝑗=1 𝑧𝑗𝑘 𝑡+1
𝜆𝑗𝑘 (∀𝑘𝑙 , ∀𝑘, 𝑡 = 1, … , 𝑇 − 1) (as carry-over to t+1)
𝑙

∑𝑛𝑗=1 𝜆𝑗𝑘
𝑡 𝑡
= 1 (∀𝑘, ∀𝑡), 𝜆𝑗𝑘 ≥ 0 (∀𝑗, ∀𝑘, ∀𝑡), (10)

𝑡
Where 𝜆𝑡𝑘 = (𝜆𝑗𝑘 ) ∈ 𝑅+𝑛 is the intensity vector corresponding to division k (k=1,…K) at t (t=1,…T).

133
MODEL DEVELOPMENT
For the realization of this research is presented an output oriented model with Variable Returns to Scale.
For the DMU selection, 32 public universities of Mexico were identified and considering those receiving
federal or state funding.
The first step was the elaboration of the nodes of the network structure, for that, it was contemplated two
nodes: teaching and research. In teaching node, the inputs were: funding, professors and doctoral students
enrolled and as output were contemplated doctorate degree professors. This output is used as an
intermediate link to the next node that is research so doctorate degree professors are the input and the
outputs for this node are: doctors in the SNI and ISI publications. For the dynamic model the years
considered are 2008 and 2012, and the carry-over (output) that is used to attach one year with another, is
PhD graduates, so that graduates are inputs next period like professor or researcher (see figure 1).

Figure 1. Dynamic-Network Model.


Source: Personal elaborated based on Tone and Tsutsui model, (2014).

RESULTS
The results show that in both periods the node more efficient was “teaching”, and in 2008 the average of
efficiency was 51%. Nevertheless there were 8 universities that were efficient in this node: El Colegio de
México, Universidad Autónoma de Chapingo, Universidad Autónoma de Nayarit, Universidad Autónoma
de Nuevo León, Universidad Autónoma de San Luis Potosí, Universidad Autónoma Metropolitana,
Universidad de Colima and Universidad Nacional Autónoma de México, in contrast is research node with
33% of average efficiency, where only was efficient Universidad Nacional Autónoma de México, and for
the year 2012, the number of efficient universities has increased to 13 universities and with an average of
58% of efficiency.

134
Universidad Nacional Autónoma de México was efficient in both years and in both nodes and in the
opposite side is Universidad Autónoma Benito Juárez de Oaxaca who showed the lowest score (see table
1).

Table 1. Network Model in México Universities 2008-2012.


2008 2012
UNIVERSITY TEACHING RESEARCH TEACHING RESEARCH
NODE NODE NODE NODE
Benemérita Universidad Autónoma de Puebla 0.991 0.253 1 0.578
El Colegio de México 1 0.514 1 1
Instituto Politécnico Nacional 0.467 0.345 0.052 0.728
Instituto Tecnológico Autónomo de México 0.981 0.355 1 0.239
Instituto Tecnológico de Sonora 0.260 0.335 0.059 0.239
Universidad Autónoma Benito Juárez de Oaxaca 0.293 0.028 0.487 0.074
Universidad Autónoma Chapingo 1 0.209 0.409 0.965
Universidad Autónoma de Aguascalientes 0.217 0.062 0.498 0.717
Universidad Autónoma de Baja California 0.071 0.322 0.412 0.340
Universidad Autónoma de Baja California Sur 0.988 0.242 1 0.400
Universidad Autónoma de Campeche 0.578 0.548 0.375 0.678
Universidad Autónoma de Chihuahua 0.053 0.056 1 0.274
Universidad Autónoma de Coahuila 0.366 0.047 1 0.088
Universidad Autónoma de Nayarit 1 0.234 1 0.405
Universidad Autónoma de Nuevo León 1 0.221 1 0.364
Universidad Autónoma de Querétaro 0.073 0.338 0.250 0.900
Universidad Autónoma de San Luis Potosí 1 0.383 0.618 0.757
Universidad Autónoma de Sinaloa 0.276 0.085 1 0.163
Universidad Autónoma de Tamaulipas 0.520 0.060 1 1
Universidad Autónoma de Zacatecas 0.077 0.213 0.967 0.282
Universidad Autónoma del Estado De Hidalgo 0.250 0.567 0.296 0.733
Universidad Autónoma del Estado De México 0.100 0.431 0.205 0.460
Universidad Autónoma Metropolitana 1 0.455 1 0.467
Universidad de Colima 1 0.215 0.176 0.917
Universidad de Guadalajara 0.237 0.259 0.070 0.474
Universidad de Guanajuato 0.046 0.676 0.394 1
Universidad de Quintana Roo 0.969 0.458 1 0.675
Universidad de Sonora 0.248 0.887 0.04 0.998
Universidad Juárez Autónoma de Tabasco 0.049 0.103 0.05 0.361
Universidad Juárez del Estado de Durango 0.254 0.259 0.36 0.167
Universidad Michoacana de San Nicolás De Hidalgo 0.115 0.440 0.11 0.903
Universidad Nacional Autónoma de México 1 1 1 1

Source: Personal compilation based on DEA results.


It is generally observed an increase in efficiency with the dynamic model as in 2008 had an average of 58%
efficiency and for 2012 was 69%; this increase was mainly due to the improvement that occurred in the
area of research in the majority of universities.

CONCLUSIONS
In this work is presented the efficiency of universities in Mexico in 2008-2012, using Dynamic Network
DEA model. To calculate the efficiency, it was used DEA model with variable returns to scale oriented
output and the DMU's used were 32 public universities in the country.
In the construction of Network DEA model, it was contemplated two nodes: teaching and research. In
teaching node, the inputs were: funding, professors and doctoral students enrolled and as output was

135
contemplated doctorate degree professors that was used as link variable with the next node. Research node
used as input doctorate degree professors and the outputs are: doctors in the SNI and ISI publications. The
carry-over used to attach one year with another, is the postgraduates, so that graduates are inputs next
period.
In both periods the node that was more efficient was “teaching”, and in 2008, there were 8 universities that
were efficient in this node, and for 2012, the number has increased to 13 universities. Universidad Nacional
Autónoma de México was efficient in both years and in both nodes and in the opposite side is Universidad
Autónoma Benito Juárez de Oaxaca who showed the lowest score. Research node presented less efficiency
in both years, but nevertheless there were 10 universities that obtained higher score in this node than in
teaching node. In addition it can see that in average all the university increased the score in this area, in
2008 the global average in research efficiency was 33% and in 2012 the global average was 57%.
Finally the hypothesis proposed is true because teaching node is who determines the efficiency in the
universities of México in the period 2008-2012.

REFERENCES
[1] Asociacion Nacional de Universidades e Instituciones de Educacion Superior (ANUIES) (2013), Statistical
Yearbook 2013, Retrieved from http://www.anuies.mx/informacion-y-servicios/informacion-estadistica-de-
educacion-superior/anuario-estadistico-de-educacion-superior/
[2] Bogetoft, Färe, Grosskopf, Hayes and L. Taylor (2009). Dynamic Network DEA: an illustration. Journal of the
Operations Research Society of Japan . Vol. 52, No. 2, 147-162
[3] Cook WD, Liang L, Zhu J (2010) Measuring performance of two-stage network structures by DEA: a review and
future perspective. Omega 38(6):423– 30
[4] Färe R, Grosskopf S. (1996) Intertemporal Production Frontiers: with Dynamic DEA, Kluwer.
[5] Färe, R. and S. Grosskopf (2000). “Network DEA,” Socio-Economic Planning Sciences 34: 35–49.
[6] Kao, C. and Hwang, S.N., (2009). “Efficiency measurement for network systems: IT impact on firm
performance”, Decision Support Systems, (in press) (doi: 10.1016/j.dss.2009.06.002).
[7] Lewis, H. F., and Sexton, T. R. (2004). "Network DEA: Efficiency analysis of organizations with complex
internal structure." Computers and Operations Research, 31(9), 1365
[8] Mayston, D. and Jesson, A. (1988). Developing models of educational accountability, Oxford Review of
Education, 14(3). Pp. 321–339.
[9] Nemoto and M. Goto. (1999). Dynamic data envelopment analysis: modeling intertemporal behavior of a birm in
the presence of productive inefficiencies. Economic Letters, 64, 51–56.
[10] Nemota, J., and Gota, M. (2003). "Measuring dynamic efficiency in production: An application of data
envelopment analysis to Japanese electric utilities." Journal of Productivity Analysis, 19, 191-210.
[11] Rodríguez, Roberto (2014). Higher education and political transitions in Mexico. Revista de Educación Superior.
43(171), Pp.9-36
[12] Ruggiero, J. (1996). On the measurement of technical efficiency in the public sector, European Journal of
Operations Research, 90: 553–565
[13] Secretaria de Educación Pública (SEP) 2014, Educación en México. Retrieved from
http://educacion.michoacan.gob.mx/
[14] Sengupta, J. K. (1992). "Non-parametric approaches to dynamic efficiency: a non-parametric application of
cointegration for production frontiers." Applied Economics, 24, 153-159.

136
[15] Sengupta, J. K. (1997). Persistence of dynamic efficiency in Farrell models. Applied Economics, vol. 29, p. 665–
671.
[16] Sengupta, J. K. (1999). A dynamic efficiency model using data envelopment analysis. International Journal of
Production Economics, vol. 62, p. 209-218.
[17] Sexton TR and Lewis HF (2003) “Two-Stage DEA: An application to major league baseball” Journal of
Productivity Analysis, 19, 227-249Tone and M. Tsutsui (2009). Application of Network DEA Model to Vertically
Integrated Electric Utilities, Discussion Paper: 07-03.
[18] Tone and M. Tsutsui (2010). “Dynamic DEA: a slacks-based measure approach,” Omega, vol. 38, no. 3-4, pp.
145–156.
[19] Tone and Tsutsui, (2014). “Dynamic DEA with network structure: a slacks-based measure approach, “Omega,
vol. 42, no. 1, pp. 124–131.

137
138
REGIONAL OPPORTUNITY COSTS OF ENVIRONMENTAL REGULATIONS
IN TURKISH MANUFACTURING INDUSTRY
YÜCEL ÖZKARA
Republic of Turkey, Ministry of Science, Industry and Technology, Directorate General for Productivity, Ankara,
Turkey (corresponding Author)
MEHMET ATAK
Gazi University, Faculty of Engineering, Department of Industrial Engineering, Maltepe, Ankara, Turkey

ABSTRACT
In developing countries, regional disparities cause structural development problems. Improving manufacturing
industry by applying regional strategies can help dealing with these disparities. Measurement of regional performance
is a fundamental element for designing regional policies. Therefore, performance evaluation with a total-factor DEA
model will produce special information for authorities. Environmental performance index (EPI) is commonly used in
literature for evaluating impacts of environmental regulations which may cause a loss of desirable outputs.
Opportunity costs (OC) due to these regulations can be assessed by EPI scores. In this study, two slack based DEA
models are used for regional performance measurement of Turkish manufacturing industry between the years 2003-
2012; and, EPI scores of regions and OC of environmental regulations are calculated. From the study results it is
found that OC differ among regions and ratio of total OC to annual production value reaches its highest level in 2006.
The relationship between regional development and OC is also investigated.

Keywords: data envelopment analysis; slack based model; opportunity cost; environmental regulation; Turkish
manufacturing industry

INTRODUCTION
Regional development is a critical issue in many countries, and governments attach importance to regional
policies in order to provide it. Regional disparities exist especially in developing countries and they cause
structural development problems. As a leading sector, manufacturing industry has a key role in the growth
of economy. Hence, amending manufacturing industry by regional strategies can help dealing with these
disparities. Turkey has established regional development agencies in its 26 regions. These agencies support
preferential sectors by financial and technical programs according to their individual regional policies.
Determining regional performance is a fundamental element for designing regional policies. To this end,
Data Envelopment Analysis (DEA) is an applicable method for efficiency measurement. Performance
evaluation with a total-factor DEA model (including capital stock, employment and energy as inputs;
production value and CO2 emissions as outputs) can produce special information for authorities for
designing regional and environmental policies.
There are many studies in literature about measuring efficiencies of countries, regions and sectors. Hu and
Wang (2006)’s study in which the total-factor energy efficiency index presented to the literature is a leading
study. After that, researchers have studied following the approach of Hu and Wang (2006) (Chien and Hu,
2007; Honma and Hu, 2008; Xiaoli et al., 2014). Most of the performance measurement studies using DEA
treat desirable outputs and inputs, but they ignore undesirable outputs. However, production is a joint
activity in which desirable and undesirable outputs are produced simultaneously; and in the case of
environmental regulations, undesirable outputs (emissions, solid wastes etc.) may not be freely or costlessly

139
disposable (Färe et al., 1989). Therefore, researchers prefer weak disposability assumption for performance
measurement when undesirable outputs are considered (Zaim and Taşkın, 2000; Färe and Grosskopf, 2004).
Turkey does not have environmental regulation on manufacturing industry such as emission taxes or
emission limitations yet; but in long term, the government may carry out some environmental regulations.
These regulations will affect firms and industries to be more environmental, thus some sources will be
transferred to emission control and environmental pollution prevention; and as a result, this will cause a
production loss. In this paper, the opportunity costs of regions’ manufacturing industry is calculated when
such a regulation comes into force; and in addition, in the view of the regional disparities in Turkey and
Turkish industry, the levels of these costs are analyzed among regions and the pattern between development
levels of regions and opportunity costs are investigated by utilizing two types of slacks based DEA models
under constant returns to scale assumption.

METHODS
In recent years, slacks based efficiency measures have gained popularity especially in environmental
performance measurement due to its higher discriminating power (Zhou et al., 2008a). In this study, two
slack based models are preferred for the efficiency measurement of manufacturing industries of Turkey’s
regions. Slacks based efficiency measure is consisted of the slacks of inputs and outputs; and it can
determine all the economic inefficiencies (Zhou et al., 2008b). According to Tone (2001), first slack based
model (SBM1) is given as follows:

1 m 
1 si xijo
m i 1
 SBM 1  min
1 q
1   sr yrjo
q r 1
n
(1)
s.t.   j yrj  sr  yrjo , r  1,..., q
j 1
n

 x
j 1
j ij  si  xijo , i  1,..., m

 j  0, si , sr  0 j  1,..., n

where

xij ∈ R+ : ith input quantity used by jth DMU i = 1,…, m, j = 1,…,n


yrj ∈ R+ : rth output quantity produced by jth DMU r = 1,…, q, j = 1,…,n
In order to measure environmental performance, Zhou et al (2006) proposes a slack based model which
considers undesirable outputs under weak disposability assumption according to environmental DEA
technology defined by Färe and Grosskopf (2004). Following Zhou et al. (2006), the second slacks based
model (SBM2) is given as follows:

140
1 m 
 si xijo
1
m i 1
 SBM 2  min
1 q
1   srg  yrjo
g

q r 1
n
s.t.   j yrjg  srg   yrjo
g
, r  1,..., q
j 1
(2)
n

 y
j 1
j
b
kj  ykjo
b
k  1,..., l

 x
j 1
j ij  si  xijo , i  1,..., m

 j  0, si , srg   0 j  1,..., n

where ybkj ∈ R+ : kth undesirable output quantity produced by jth DMU k = 1,…, l, j = 1,…,n. In this model,
undesirable output (yb) and desirable output (yg) are considered together, and the equality constraint reflects
the weak disposability of outputs.
Environmental performance index (EPI) is commonly used in literature for evaluating impacts of
environmental regulations. EPI is defined as the ratio of two efficiency indexes which are efficiency index
without considering undesirable outputs and efficiency index considering undesirable outputs, respectively.
Environmental regulations may cause a loss of desirable outputs; thus, the opportunity costs due to these
regulations can be assessed by EPI scores. (Färe et al., 1989; Zaim and Taşkın, 2000; Zhou et al, 2006).
According to Zhou et al. (2006), Slack based Environmental Index (SBEI) is defined as:
SBEI = ρ*SBM1 ∕ ρ*SBM2 , where
ρ*SBM1 : economic efficiency scores when undesirable output is not considered,
ρ*SBM2 : efficiency scores under environmental DEA technology
SBEI cannot take any positive value greater than 1, because ρ*SBM1 ≤ ρ*SBM2. When SBEI = 1, the shift of
the production technology from traditional DEA to environmental DEA would not affect the efficiency of
the DMU concerned. Otherwise, if SBEI is less than unity, it implies that the result of environmental
regulations are waste of inputs or loss of desirable output production; and this means an opportunity cost
occurs with these regulations. Following Zhou et al. (2006), the degree of regulatory impact for a DMU can
be easily calculated as (1-SBEI). Assuming that this impact will appear in a loss of production, opportunity
cost (OC) can be formulated as:

OC =(1-SBEI) x Desirable Output (3)

RESULTS AND DISCUSSIONS


In this study, manufacturing industries of Turkey’s 26 regions are analyzed during the period 2003-2012
using two types of slack based DEA models which are given above. As a candidate country for European
Union, Turkey (TR) has formed its own Nomenclature of Territorial Units for Statistics (NUTS). According
to this nomenclature, there are 26 regions at level 2 (Turkstat, 2015), and each region has a Development
Agency. Capital stock, employment and electricity consumption (as a proxy of energy input) of

141
manufacturing industry are preferred as inputs; production value is chosen as the only output in SBM1
model. For the analyses using SBM2 model, in addition to these inputs and output, CO2 emissions are
considered as undesirable output. Data sources of the study are Turkish Statistical Institute (Turkstat) and
Turkish Electricity Transmission Company. In Figure 1, average efficiency scores of regions during the
study period are displayed. It is clearly seen that average SBM2 scores are greater or equal than average
SBM1 scores and this difference indicates the effect of environmental regulations.

Figure 1: Average Efficiency Scores of SBM1 and SBM2 (2003-2012)

The year averages of opportunity costs in Turkish manufacturing industry are given in Table 1. It reaches
its highest in 2006 as of 21.1%; and it’s lowest in 2009 as of 1%. 2006 is the critical year, because
investments on manufacturing industry had reached its highest levels until this year as a result of Turkey’s
monetary policies of low currency-high interest rates. Hence, capital accumulation of manufacturing
industry has been grown by these high investments. However, this enlarged capital stock has not been
efficiently utilized until 2006, and this case causes the lowest efficiency performance in this year. After
2006, manufacturing industry has tended to utilize its sources more efficiently. All of these are seen directly
on opportunity costs as well. Based on 2007-2012 period, ratio of total OC to annual production value of
Turkish manufacturing industry has an average of 2.4%. Manufacturing industry of Turkey can tolerate this
opportunity costs by increasing the efficiency and productivity levels.

Table 1: Opportunity costs during the study period


Years 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Average OC per region (Million TRY) 148 505 727 2302 190 203 99 280 613 464
Total OC (Million TRY) 3840 13143 18914 59855 4940 5275 2576 7268 15942 12074
Ratio of Total OC to Annual Production Value 1.70% 5.20% 7.40% 21.10% 1.70% 1.70% 1.00% 2.30% 4.30% 3.20%

As mentioned before, opportunity cost is a result of costly production of CO2 emissions (Zhou et al, 2006).
From this view, the production loss per one tonne of CO2 can be calculated by dividing opportunity cost by
amount of emissions. Regulatory impact can be assessed with not only slack based models, but also radial
DEA models. Therefore, opportunity costs are calculated by using radial models (CCR and CCR with
undesirable output under weak disposability) additionally. The production loss per one tonne of CO 2 is
given in Figure 2 according to both types of models. While total OC has shown its lowest percent level in
2009, the production loss per one tonne of CO2 has shown its lowest level in 2007. The difference between
radial and slack based DEA models also indicates the higher discriminating power of slack based models.

142
Figure 2: Opportunity cost per ton CO2

Individual analysis (target input-outputs, improvement ratios etc.) should be also done region by region;
but in this study we want to investigate the relation between opportunity costs and regional development
levels. To do this, the regions are considered in three groups according to their gross value added per capita.
Since the data is available for 2004-2011, the comparisons are done based on average values of this period.
As seen on the Figure 3, high developed regions have the highest opportunity costs due to their large scales
of industry; and the lowest costs are appeared in low developed regions. However, when the average
production loss per one tonne of CO2 is examined (the green line); high developed regions have the lowest
unit production loss. This finding indicates that high developed regions seem to be adaptive for
environmental regulations.

Figure 3: Opportunity costs for regional development groups

CONCLUSIONS
In this study, regional performance measurement of Turkish manufacturing industry is performed using
slack based DEA models for the period of 2003-2012. According to the findings, there is an inverse
relationship between average production loss per tonne CO2 and development levels. Hence, high
developed regions seem to be more adaptive for environmental regulations. For medium and less developed
regions, regional industrial policies should focus on innovation and environmental issues. Based on 2007-
2012 period, environmental regulations would affect the production value about 2.4%. Development

143
Agencies of regions in Turkey should foster structural transformation of their region’s manufacturing
industry regarding its efficiency level and opportunity cost.

REFERENCES
[1] Chien, T., Hu, J.-L. (2007). Renewable energy and macroeconomic efficiency of OECD and non-OECD
economies. Energy Policy, 35 (7), 3606-3615.
[2] Färe, R., Grosskopf, S. (2004). Modeling undesirable factors in efficiency evaluation: Comment. European
Journal of Operational Research, 157 (1), 242–245.
[3] Färe, R., Grosskopf, S., Lovell, C. K., Pasurka, C. (1989). Multilateral productivity comparisons when some
outputs are undesirable: a nonparametric approach. The review of economics and statistics, 71 (1), 90-98.
[4] Honma, S., Hu, J.-L. (2008). Total-factor energy efficiency of regions in Japan. Energy Policy, 36 (2), 821-833.
[5] Hu, J.-L., Wang, S.-C. (2006). Total-factor energy efficiency of regions in China. Energy Policy, 34 (17), 3206–
3217.
[6] Tone, K. (2001). A slacks-based measure of efficiency in data envelopment analysis. European journal of
operational research, 130 (3), 498-509.
[7] Turkstat. (2015). Classification Server. Retrieved from Turkish Statistical Institute:
http://tuikapp.tuik.gov.tr/DIESS/SiniflamaSatirListeAction.do?surumId=250&seviye=3&detay=H&turId=7&tu
rAdi=%205.%20Geographical%20Classifications
[8] Xiaoli, Z., Rui, Y., Qian, M. (2014). China's total factor energy efficiency of provincial industrial sectors. Energy,
65, 52-61.
[9] Zaim, O., Taşkın, F. (2000). Environmental efficiency in carbon dioxide emissions in the OECD: a non-
parametric approach. Journal of Environmental Management, 58 (2), 95-107.
[10] Zhou, P., Ang, B. W., Poh, K. L. (2006). Slacks-based efficiency measures for modeling environmental
performance. Ecological Economics, 60 (1), 111-118.
[11] Zhou, P., Ang, B. W., Poh, K. L. (2008a). Measuring environmental performance under different environmental
DEA technologies. Energy Economics, 30 (1), 1-14.
[12] Zhou, P., Ang, B. W., Poh, K. L. (2008b). A survey of data envelopment analysis in energy and environmental
studies. European Journal of Operational Research, 189 (1), 1-18.

144
SELECTING BEST SUPPLIER BY ADVANCED CROSS-EFFICIENCY (ACE)
ELHAM ROSTAMIYAN
Islamic Azad University of Shahmirzad Branch, Semnan, Iran
FRANCK K. ADJOGBLE
Department of Mathematics and Informatics, University of Hagen, Germany
BEHDAD VATANI
Department of Electrical and Computer Engineering, University of North Carolina at Charlotte, Charlotte, NC

ABSTRACT
In the competitive trade world, choosing the best supply chain is a vital problem. One of the most important processes
performed in enterprises today is the evaluation, selection and continuous measurement of suppliers. In this paper we
want to use advanced cross-efficiency (ACE) to select best supplier. Recently, Jahanshahloo et al. [Jahanshahloo
G.R., Hosseinzadeh Lotfi F., Jafari Y., Maddahi R., Selecting symmetric weights as a secondary goal in DEA cross-
efficiency evaluation, Applied Mathematical Modelling 35, (2011), 544–549] proposed the use of symmetric weights
for computing the elements of cross-efficiency matrix. In spite of fact that the proposed method decreases the number
of zero weights, a large number of zero weights among input and output symmetric weights may still exist. To this end,
this paper improves the proposed secondary goal model. The improved method generates more acceptable results in
the ranking process of the decision-making units (DMUs). Finally, a numerical example is provided to illustrate the
applicability of the proposed method to find best supplier.

Keywords: DEA, symmetric weights, secondary goal, cross-efficiency

INTRODUCTION
In this competitive business world, selecting the best supplier amongst all is vitally important for firms who
want to increase procurement quality levels and decrease costs. That is the reason why many companies
are working on evaluating and selecting best suppliers. [2, 4].
Supplier selection is one of the important aspects of supply chain management. Several factors such as
transportation lead time, transportation cost, production lead time, order cost, quality level, etc. are
determinant in choosing suppliers. Supply Chain Management (SCM) is one of the most important
competitive strategies used by modern enterprises. Meanwhile, supplier selection plays an effective role in
the supply chain.

Data envelopment analysis (DEA) is a linear programming-based methodology for assessing the relative
efficiency of decision-making units (DMUs) such as bank branches, hospitals, production plants,
universities, etc. In DEA the relative efficiency of the DMUs can be calculated by the ratio of the weighted
outputs to weighted inputs. The input and output weights can be obtained by the DMUs which try to assign
the most favorable weights to their inputs and outputs and achieve the best possible relative efficiency to
the other DMUs. The relative efficiency with the most favorable weights is called the self-evaluated
efficiency. In calculating the self-evaluated efficiency it is possible that a particular DMU assigns all of its
weight to only a few variables. And according to this efficiency score, more than one DMU can be evaluated
as DEA efficient. Many methods have been proposed in DEA literatures to restrict the free selection of
input and output weights.

145
Recently, Dimitrov and Sutton [3] proposed symmetric weights assignment technique (SWAT).
In this method the DMUs are allowed to make a symmetric selection of weights i.e. balance their selection
of weights to all the variables and avoid assigning all of their weights to only a few variables.
To overcome the other problem of the free selection of weights i.e. the lack of discrimination power in
DEA, several methods have been proposed for ranking the DMUs. Some of these methods were reviewed
by Adler et al. [1]. Among the ranking methods, the cross-efficiency evaluation, which was issued by
Sexton et al. [6], is one of the most significant and effective ways. The main idea of cross-evaluation is to
use DEA in a peer-evaluation instead of a self-evaluation mode [2]. In the peer-evaluation mode each of
the DMUs achieves its efficiency using the weights of the other DMUs. There are two principal advantages
of cross-evaluation: (1) it provides a unique ordering of the DMUs and (2) it eliminates unrealistic weight
schemes without requiring the elicitation of weight restrictions from application area experts [2].

In spite of the mentioned advantages, the non-uniqueness of the DEA optimal input and output weights
decreases the benefit of the cross-efficiency evaluation. As a remedy, Sexton et al. [6] and Doyle and Green
[4] suggested the aggressive and benevolent model formulations as secondary goals to deal with the issue
of the alternative optimal solutions. The benevolent formulation is benevolent towards the cross-efficiencies
of the other DMUs and maximizes them, while the aggressive formulation minimizes the cross-efficiencies
of the other DMUs. In addition to the benevolent and aggressive formulations, many other secondary goals
and methods have been proposed by different researchers.
Recently, Jahanshahloo et al. [5] in order to overcome the problems of the non-uniqueness and free selection
of the DEA optimal input and output weights in the cross-efficiency evaluation proposed selecting
symmetric weights as a secondary goal method which emphasizes selecting symmetric weights by DMUs.
The optimal weight vector provided by this method may have several zero components between its elements
and this issue can disregard many of input and output items in the evaluation process.
So to overcome this problem we improve the proposed secondary goal model. The improved method
provides optimal symmetric DEA weights with less zero components and produces more acceptable results.

The rest of this paper is organized as follows: section 2 briefly reviews the cross-efficiency method and
selecting symmetric weights as a secondary goal which is proposed by Jahanshahloo et al. [5]. Section 3
proposes a new improved method for selecting symmetric weights. Section 4 presents a numerical example
to show the helpfulness of the proposed improved model in reducing the number of zero input and output
weights in the cross-efficiency evaluation. In section 5 the conclusion is given.

PRELIMINARIES
In this section we briefly explain the cross-efficiency evaluation and the selecting symmetric weights as a
secondary goal method.

146
In productive activities, we assume that there are n DMUs; each one transforms m different inputs into
different s outputs. Input and output vectors for DMUj , j Î J = {1, L , n } are denoted by
X j = (x1j , L , x mj )T Yj = (y1j , L , ysj )T
and . All the input and output vectors are semi-positive. That is,
X j = (x1j , L , x mj ) ³ 0, X j ¹ 0
T
Yj = (y1j , L , ysj )T ³ 0, Yj ¹ 0.
and
The cross-efficiency method consists of the following steps:

In step 1, the efficiency score of each DMUk (k Î J = {1, L , n }) is obtained by solving the following
output oriented CCR model:

 KK  min VkT Xk
s.t. UTK Yk  1
(1)
UTK Yj  VkT X j  0 j  1,..., n,
Vk  0, U k  0.

where U K and VK are the input and output weight vectors of DMUk, respectively. And  kk represents the
efficiency score of DMUk.
* *
Suppose
( U k , Vk ) be the optimal weight vectors of model (1),

*
UTk Yk
 kk  * (2)
VkT Xk

 kk in equation (2) represents the self-evaluated (CCR) efficiency of DMUk and

*
UTk Yj
 jk  * (3)
VKT X j j Î J = {1, L , n }; j ¹ k

 jk
in equation (3) shows the cross-efficiency of DMUj using the optimal weights of DMUk, and indicates
the peer-evaluation of DMUk to DMUj, j Î J = {1, L , n }; j ¹ k [7]. In step 2 after solving model (1) for
n DMUs, the CCR efficiencies and the cross-efficiencies of n DMUs are entered to a cross-efficiency matrix
(Table 1).

Table 1. Cross-efficiency matrix for n DMUs.


DMUs DMU1 DMU2 L DMUn
DMU1 q11 q12 L q1n
DMU2 q21 q22 L q2n
M M M M M
DMUn qn 1 qn 2 L qnn

147
 , k  1, , n
Finally, the average of all jk will indicate the overall efficiency score for DMUj
j Î J = {1, L , n }; and n DMUs can be ranked according to their related overall efficiency scores. Model

(1) may have alternative optimal solutions. In the presence of alternative optimal solutions of model (1) the
elements of cross-efficiency matrix are not fixed and may alter.

NEW PROPOSED MODEL


In some situations, a lot of zero elements may exist among input and output weight vectors which are
obtained by model (1). Therefore the cross-efficiency scores provided by these weights can be unreliable.
In order to assign the input and output weights symmetrically to all of the inputs and outputs as much as
possible and to decrease the zero elements between the input and output weights, we propose in this section
the following model:

min et z k e  ( w1  w2  )
s.t. U k Yk  1, (a)
Vk X k   k* , (b)
U k Y  Vk X k  0, (c ) (4)
uki yki  ukj ykj  zkij , i  1,..., s  1, j  i  1,..., s (d )
 uki yki  ukj ykj  zkij , i  1,..., s  1, j  i  1,..., s (e)
ukr ykr   r  1,..., s (g)
vki xki   i  1,..., m ( h)
U 0 , V 0 (f)
k k

where w1  0 and w2  0 are the weighting coefficients for the two objectives  and  satisfying the
condition w1  w2  1 .

As mentioned by Wang et.al (2011), in economic meaning of  and  it can be said that  indicates
the minimum relative efficiency of the s outputs for DMUk and  shows the minimum relative importance
of the m inputs for DMUk. By maximizing  and  in the objective function of model (4), each DMU
tries to choose symmetric input and output weights to make its inputs and outputs as efficient or sufficiently
important as possible. This method can cause each DMU to allocate its weights symmetrically to all
variables, and each output of DMUs can produce sufficient efficiency as an individual as well as every
input of DMUs can be sufficiently used. More ever the number of symmetric constraints (d) and (e) in
model (4) is 2𝑠 2 . However some of these constraints are redundant. The number of symmetric constraints
(d) and (e) is reduced to s(s-1) in model (4). In the next section an application example of efficiency
evaluation of seven departments in a university shows the usefulness of what were mentioned.

148
NUMERICAL EXAMPLE
In this section, we provide a numerical example to illustrate potential applications of “Advanced cross
efficiency” to find the best supplier . In this example we want to select best box suppliers among 7 suppliers.
Factors considered in order to select best supplier are as follows:
The data set for the 7 box supplier is documented in Table (2), where price (USD), distance
(kilometer) and preparation time (Hour) are treated as inputs, quality and services are viewed
as outputs. One of the inputs was a price that specified by the supplier, and varies between 32
to 36.5 USD, another input is blowover distance to the target box is delivered to the factory.
Preparation time, time interval between the request and the delivery of an order based, is
specified at the time in the table (2), quality and services are the outputs that have been determined
by experts in the number 1 to 9.

Table 2. Data for 7 box suppliers


Inputs and Outputs
Preparation Time CCR-
Price Distance Quality Services
Suppliers efficiency
1 32.5 400 48 7 6 1
2 320 750 70 6 6 1
3 320 1500 70 4 8 1
4 330 600 84 3 4 1.2199
5 330 2000 72 5 5 1
6 325 730 12 3 8 1
7 365 2350 84 1 4 1

Following Tables show the input and output weight vectors produced by the different secondary goal
models.

Table 3. Input and Output weights for DMUs produced by aggressive model 20 zero elements
DMU Input 1 Input 2 Input 3 Output 1 Output 2
1 0 0 0.0008772 0.000501 0
2 0 0.00011 0.0001232 0 0
3 0 0 0.0009174 0 0
4 0.005389 5.3 E-06 0 0 0
5 0.004287 0 0.0004016 0 0.002256
6 0.000967 0 0.0007493 0 0.001241
7 0.006579 0 0 0.001696 0

Table 4. Input and Output weights for DMUs produced by benevolent model 9 zero elements
DMU Input 1 Input 2 Input 3 Output 1 Output 2
1 0.001985 8.1E-05 0 0.000905 0.000284
2 0.002866 6.6E-05 0 0.000646 0
3 0 2.9E-05 0.0007376 0.000244 0.001042
4 0.005389 5.3 E-06 0 0 3.22E-11
5 0.002465 0.0001 0 0.001124 0.000352
6 0.002069 8.4E-05 0 0.000943 0.000296
7 0.002529 0.0001 0 0.001153 0.000361

149
Table 5 Input and output weights for DMUs produced by proposed model (4) with 4 zero elements
DMU Input 1 Input 2 Input 3 Output 1 Output 2
1 0.02386 0.00072 0.02137 0.00952 0.01961
2 0.03709 0.00039 0 0.00321 0.00329
3 0.00545 0.00015 0.00774 0.0019 0.0058
4 0.07826 0.00008 0 0 0
5 0.01753 0.00005 0.00042 0.0023 0.00256
6 0.02095 0.00041 0.00602 0.00741 0.00741
7 0.02069 0.00003 0.00013 0.0021 0.00344

Tables 2, 3, 4 and 5 show that the input and output symmetric weight vectors produced by model (5) have
the least zero elements in comparison with the other methods. These weights with less zero elements present
the ranks of DMUs by considering the most input and output items of DMUs in the evaluation process. So
the proposed ranks to the DMUs by model (4) are more reliable.
Table 6 indicates the overall efficiency scores and resulted ranks to the seven Supplier by the different
secondary goal models.

Table 6. Overall efficiency scores and resulted ranks to DMUs by different secondary goal models
DMU Aggressive Benevolent Model(4)
1 0.8081 (2) 0.9442 (2) 0.86585 (3)
2 0.7191 (4) 0.9326 (3) 0.90184(2)
3 0.7669 (3) 0.7952 (6) 0.81491(5)
4 0.3904 (7) 0.5793 (7) 0.54770 (7)
5 0.6576 (5) 0.9100 (4) 0.85304 (4)
6 0.8424 (1) 0.9929 (1) 0.98443(1)
7 0.5264 (6) 0.8963 (5) 0.77632(6)

As can be seen from Table (6) all the methods assign the highest rank to DMU6 and the lowest rank to
DMU4. The resulted ranks to the DMUs by the proposed model (4) are the same as the proposed ranks by
the natural secondary goal models [7, 8]. Model (4) by offering unique symmetric input and output weight
vector with less zero elements provides reliable ranks to DMUs.

CONCLUSION
Using the SWAT for cross-efficiency evaluation, which was proposed by Jahanshahloo et al., is an effective
method that allows the DMUs to choose the symmetric weights among their optimal weights and use them
for the cross-efficiency evaluation. However, in this method a large number of zero weights among input
and output symmetric weights may exist. To resolve this problem, this paper improved the secondary goal
model, proposed by Jahanshahloo et al. By reducing the zero elements between symmetric weights and
eliminating the redundant constraints, the improved method provides more reliable results in the ranking
process of the DMUs.

REFERENCES
[1] Alder N., Friedman L., Sinuany-Stern Z., Review of ranking methods in the data envelopment analysis context,
European Journal of Operational Research,140, (2002), 249–265.
[2] Anderson T.R., Hollingsworth K.B., Inman L.B., The fixed weighting nature of a cross evaluation model, Journal
of Productivity Analysis, 18 (1), (2002), 249–255.

150
[3] Dimitrov S., Sutton W., Promoting symmetric weight selection in data envelopment analysis: a penalty function
approach, European Journal of Operational Research, 200 (1), (2010), 281–288.
[4] Doyle J.R., Green R., Efficiency and cross-efficiency in data envelopment analysis: derivatives, meanings and
uses, Journal of the Operational Research Society. 45 (5), (1994), 567– 578.
[5] Jahanshahloo G.R., Hosseinzadeh Lotfi F., Jafari Y., Maddahi R. Selecting symmetric weights as a secondary
goal in DEA cross-efficiency evaluation. Applied Mathematical Modelling, 35, (2011), 544–549.
[6] Sexton T.R., Silkman R.H., Hogan A.J. Data envelopment analysis: critique and extensions, in: Silkman R.H.
(Ed.), Measuring Efficiency: An Assessment of Data Envelopment Analysis, vol. 32, Jossey-Bass, San Francisco,
CA, (1986), pp. 73–105.
[7] Wang, Y. M., Chin, K. S. A neutral DEA model for cross-efficiency evaluation and its extension. Expert Systems
with Applications. 37(5), (2010), 3666–3675.
[8] Wang Y. M., Chin K. S., Jiang P. Weight determination in the cross-efficiency evaluation, Computers & Industrial
Engineering 61, (2011), 497–502.

151
152
SUPPORT OF EFFICIENT RESOURCE ALLOCATION OF TECHNOLOGICAL
PROCESSES BY A HEURISTIC SOLUTION AND AGENT TECHNOLOGY
TIBOR DULAI
University of Pannonia, Department of Electrical Engineering and Information System, Hungaryhu (corresponding
Author)
ÁGNES WERNER-STARK
University of Pannonia, Department of Electrical Engineering and Information System
GYULA ÁBRAHÁM
University of Pannonia, Department of Electrical Engineering and Information System

ABSTRACT
The paper presents a heuristic solution, which can help in resource allocation of technological processes. The novelty
is provided by the definition of a possible combination of the primary and secondary scheduling criteria and a limit
value which is determined by taking into account these criteria. The data which describe the process models and their
belonging resource definitions as well as the data from concrete process running are stored in a database that we
have designed and implemented. The mined information from the event logs are used to increase the efficiency of the
method. We applied the agent technology for the development of the system and for the implementation of more
efficient operation. All functions appear in the system are represented by an own agent. Using heuristic solution and
agent technology we could achieve better result in time and costs than without the use of our method.

Keywords: process; resource; scheduling; efficiency; agent technology

INTRODUCTION
Scheduling is an excellent and widely used topic of operations research and the industry can also benefit
from optimal schedules and resource allocation. A lot of different algorithms are born for organizing
process elements on the time scale (Pinedo, 2012), related to business processes (Barba, 2010), computer
networks (Brahimi, 2006), or industrial processes (Sule, 2007).
Agent technology has been recognized as a promising paradigm for next generation manufacturing systems.
Merdan et al. (Merdan, 2013) consider the application of intelligent agent-based technology as a promising
tool to improve system performance in complex and changeable environment. A multi-agent approach can
be used to enhance system flexibility and robustness. Because of it, in our work we used this technology,
too.
The most important information related to the processing of workflows is stored in log files, therefore the
current state analysis can be carried out using these logs and in case of unexpected events the multi-agent
system can react with a new resource scheduling proposal. Van der Aalst et al. play an excellent role in area
of process mining, see e.g., (Aalst, 2007) and (Rozinat, 2009). The data from concrete process running are
stored in our specific database and the mined information from the event logs are used to increase the
efficiency of the process.
This paper presents a heuristic solution, which can help in resource allocation of technological processes.
The novelty is provided by the definition of a possible combination of the primary and secondary scheduling
criteria and a limit value which is determined by taking into account these criteria. Process models, resource
definitions and the data from concrete process running are stored in a database. The hidden information of

153
the event logs are mined and used to increase the efficiency of the processes. The system is based on the
WADE agent development environment. All functions appear in the system are represented by an own
agent. The agents control the operation of the system in cooperation (based on our protocol). The main
agent is the scheduling agent that receives the data then it performs the scheduling of resources, on the basis
of which the database and the GUI are updated. Using heuristic solution and agent technology we could
achieve better result in time and costs than without the use of our method.

NOTATIONS
Process, Task, Resource, Applicable Resource, P, and O are introduced as sets, where
- Process = {proc1, proc2, … , procn} is the set of process models,
- Task (proci) = {ti1, ti2, … , tim} is the set of tasks of process proci,
- Resource = {r1, r2, … , rk} is the set of resources,
- Applicable Resource (tij) = {rij1, rij2, … , rijl} as a subset of Resource is the set of the resources
which are applicable to do the task tij of process proci, 1 ≤ i ≤ n, 1 ≤ j ≤ m, 1 ≤ l ≤ k
- P = {p1, p2, … , ph} is the set of product types
- O (rq) = {oq1, oq2, … , oqz}, is the set of operation modes of resource rq, 1 ≤ q ≤ k.

The following functions were introduced: rap ( ), t ( ), c ( ), min ( ), max ( ) and limit ( ), where
- rap: Resource × Task × P → {0, 1}, shows whether a resource is applicable to do a task on a
product type.
- t : Resource × Task × P × O → N, returns the operation time of a task carried out by a resource
in an operation mode on a product type.
- c : Resource × Task × P × O × N → Q, returns the cost of a task carried out by a resource in an
operation mode on a product type for a given operation time.
- min: min {Applicable Resource (tij)} → N, returns the ID of the resource which carries out task tij
with the minimal required factor (time or cost).
- max: max {Applicable Resource (tij)} → N, returns the ID of the resource which carries out task
tij with the maximal required factor (time or cost).
- limit : limit{min, max, percent} → Q, returns the threshold of the required factor (time or cost) in
case of mixed scheduling, where percent denotes the maximal allowed difference between the
optimal and the yet applicable value of the chosen factor.

RESOURCE ALLOCATION METHODOLOGY


The scheduling of concrete production processes we deal may belong to one of four types we defined, and
these types were classified into two main groups, which are: single scheduling (based on time or cost), and
mixed scheduling (time-based scheduling with improvement in cost, or cost-based scheduling with
improvement in time).
In case of single scheduling the system deals with only one factor (time or cost), and intends to minimize
this factor for the whole process. It is the basis for the resource-task assignment.

154
The mixed scheduling takes into account both factors. In this case both the time and the cost may have
primary or secondary role. The steps of the mixed scheduling are as follows:
Step 1: Let proci be the process to schedule. The scheduler chooses the basic factor, which has the highest
precedence during the scheduling. Then, the remaining – the secondary – factor gets a value (expressed in
percents) by the operator, which shows how much the secondary factor influences the outcome of the
scheduler.

Step 2: The minimum (minSumi) and maximum (maxSumi) values of the primary factor are calculated (we
deal with sequential processes). The minimum value is the optimum for the primary factor, while the
maximum value (together with the minimum value) is needed for determining the threshold.
Percentage = x %, given by the operator
minSumi = ∑j=1m min (Applicable Resource (tij))
maxSumi = ∑j=1m max (Applicable Resource (tij))

Step 3: Threshold is calculated, which origins from the previously calculated sums and the percentage
value:
100  percentage
limit = minSumi + (maxSumi – minSumi) *
100

Step 4: The phase of the resource assignment and scheduling. Threshold value has a key role, as it
determines how much the system is let to worsen the primary factor by improving the secondary one. When
the value of the primary factor reaches the threshold or the secondary factor reaches the optimum, the
scheduling process finishes. The strategy of the algorithm is greedy. While improving the secondary factor,
the scheduler attempts to involve the resources which operate with the best values regarding the secondary
factor. If in case of selecting the best resources the value of the primary factor remains under the threshold,
the scheduling finishes. Otherwise, the system has to review the schedule. It means that the resources have
to be changed, during which process the threshold has to be taken into account and the relapse of the
secondary factor should be minimal.

THE DATABASE
The aim of the developed method is to recommend an execution strategy, which results in a more reliable
and more effective execution for a given process. The process models and resource definitions are loaded
into a special database, as well as the data from concrete process running, what are also stored in a database
that we have designed and implemented. (Dulai, 2015)
The Catalog Handler Agent has the biggest role in data processing of the system. Its task is to download
the data stored in the database, to organize them into different catalogs, moreover to serve the asked
information in cooperation with other system entities and the update of the central catalog.

THE FUNCTION OF THE AGENT TECHNOLOGY IN THE SCHEDULING


The benefits of applying agent technology to resource allocation management include:

155
 Automation: The inherent autonomy of software agents can fulfil activities as human substitution.
 Resource management: Agents can represent resources. Task assignments and resource allocations
are done through negotiation among these agents.
 Reactivity: Agents react to changing circumstances and have the ability to generate alternative
execution paths.
 Intelligent decision-making: We can use the learning ability of agents that is helpful in the
workflow scheduling.
Our goal was that we integrate these benefits as specialties to a support system for organizing cooperating
and rival resources.
We applied agent technology for the development of the system and for the implementation of more
efficient operation. The system is based on the WADE agent development environment. All functions
appear in the system are represented by an own agent. The agents control the operation of the system in
cooperation (based on our protocol) in which agents are able to provide services to each other. The main
agent is the scheduling agent that receives the data then it performs the scheduling of resources, on the basis
of which the database and the graphical user interface are updated.
At present there are five system-level agents in the system. The agents are in contact with each other using
a specific ACL protocol via the communication channels, that are defined in the development environment.
These agents are:
1. Synchronization Agent: Its task is to keep the information presented on the GUI continuously up-to-
date. The agent receives the updated catalog from the Catalog Handler Agent.
2. SyncReminderAgent: This agent operates as a timer. It sends a synchronization request to the Catalog
Handler Agent according to the built-in behavior (in each 5 seconds, based on the present settings). It
results in a periodic synchronization between the Catalog Handler Agent and the outer database.
3. MySQLSupervisorAgent: The task of this agent is the verification of the availability of the MySQL
database, periodically (in each 2 seconds). If the database is unavailable because of any kind of failure,
the system becomes offline, automatically. This information is also presented on the GUI.
4. CatalogHandlerAgent: This agent is responsible for the synchronization with the database, when the
alert message arrives in each 5 seconds. After downloading, the agent organizes the data (processes and
resources) into catalogs and stores them. Locally, the other agents can request the data from here.
5. Scheduler Agent: The agent’s task is to schedule the processes based on the resources and the prepared
methods.
The levels of the system architecture is illustrated in Figure 1. Graphical interface belongs to the top level.
It ensures the connection between the users and the system. The second level contains the prepared ACL
protocol and all the agents who use it. The third level includes the agents who do the scheduling, the
simulating and the process-mining tasks. The bottom part means the whole WADE framework.

156
Figure 1: Architecture

PROCESS MINING IN OUR SYSTEM


There is a log database which stores the historical data of the real executions of the processes. The hidden
information of the event logs are used to increase the efficiency of the method. The stored data of previous
executions have information about that, how the resources behaved according to the schedule in the past.
Especially the efficiency and reliability of the resources are investigated related to the execution of different
tasks. A task of the module is the extraction of information which is usable for the exact tracking of the
execution and the behavior of the process. This information can be, for example, the efficiency of the
execution for each task, the result of the execution, the answer for the questions: whether all the
preconditions were satisfied or was there any sign previously for the failure of the resource. Applying these
methods e.g. the risk of the selection of the resources with high error rate can be decreased. Instead of them,
less efficient but reliable resources will be selected.

RESULTS AND DISCUSSIONS


Table 1 shows the results of the scheduling of an analyzed process. In case of the single scheduling only
one factor is dealt with: time or cost. We highlighted the result of the scheduling by green, however in the
first two rows we also showed the values we got for the other factor. This type of scheduling has neither
pre-scheduling nor improvement phase, moreover, in this case the secondary factor is not handled. For the
case of mixed scheduling, we indicated both the primary and the secondary factors and their weights, too.
The results of pre-scheduling (when the scheduler applies a single scheduling on the primary factor) are
signed with yellow. The final results – which are illustrated by green – origin from the pre-scheduling
results by improving the secondary factor. In the first row of the mixed scheduling we can see that when
the primary factor is the time, the 32 units cost we got after pre-scheduling was reduced by 14 units during
the improvement. This 44% decrease of cost caused that the operation time increased by 6 units, that is 15%
worsening for time. The next row shows the case, when the primary factor is the cost. After pre-scheduling
the cost is optimal, however, the value of the operation time is high. Doing the improvement, the time is
decreased by 9 units (it means 23% better result regarding to time), at the same time the cost increased by
6 units (by 18.75%).

157
Table 1: Summary of the results
Type of the Primary Secondary Results of pre-scheduling Final results Improvement
scheduling factor (%) factor Time Cost Time Cost (%)
1. Single Time - - - 18 32 -
(100%)
Cost - - - 34 11 -
(100%)
2. Mixed Time Cost 18 32 24 18 14 (43.75%)
(70%)
Cost (70%) Time 34 11 25 17 9 (23.1%)

CONCLUSIONS
The work presented in this paper includes the creation of an agent-based heuristic scheduler, which applies
process mining for improving the resulted schedule. Next to the agents of the system and their logics, a
database was prepared for storing the elements of the model and the data of the process executions.
Two kinds of scheduling methods were applied and compared: single scheduling took into account only
one factor (time or cost), while mixed scheduling applied a kind of multi-objective optimization regarding
both time and cost by determining the primary factor between them. Our simulations show that compared
with single scheduling, minimal worsening of the primary factor usually results in high improvement of the
second factor.

ACKNOWLEDGEMENT
The authors wish to thank the Hans Pape Foundation for the support in this work.

REFERENCES
[1] van der Aalst W.M.P., Reijers H.A., Weijters A.J.M.M., van Dongen B.F., Alves de Medeiros A.K., Song M.,
Verbeek H.M.W. (2007) Business Process Mining: An Industrial Application, Information Systems 32 (5): 713-
732.
[2] Barba I., Del Valle C. (2010) A Job-Shop Scheduling Model of Software Development Planning for Constraint-
based Local Search, Int. Journal of Software Engineering and Its Applications 4 (4): 1-16.
[3] Brahimi B., Aubrun C., Rondeau E. (2006) Modelling and Simulation of Scheduling Policies Implemented in
Ethernet Switch by Using Coloured Petri Nets, In Emerging Technologies and Factory Automation: 667-674.
[4] Dulai, T., Werner-Stark, Á. (2015) A database-oriented workflow scheduler with historical data and resource
substitution possibilities, Proceedings of 4th International Conference on Operations Research and Enterprise
Systems: 325-330.
[5] Merdan M., Moser T., Sunindyo W., Biffl S., Vrba P. (2013) Workflow scheduling using multi-agent systems in
a dynamically changing environment, Journal of Simulation 7: 144-158.
[6] Pinedo M.L. (2012) Scheduling: Theory, Algorithms, and Systems, 4th ed., Springer.
[7] Rozinat A., Zickler S., Veloso M., van der Aalst W.M.P., McMillen C. (2009) Analyzing multi-agent activity
logs using process mining techniques, In Distributed Autonomous Robotic Systems 8: 251-260.
[8] Sule D. R. (2007) Production Planning and Industrial Scheduling: Examples, Case Studies and Applications, 2nd
ed. CRC Press.

158
TECHNICAL AND SCALE EFFICIENCY OF SCHEDULED BANKS OF
BANGLADESH IN TERMS OF SME FINANCE USING DATA ENVELOPMENT
ANALYSIS (DEA)
SHAMIM ARA
Statistics Department, Bangladesh Bank (corresponding Author)

HAZERA AKTER
Bangladesh Foreign Trade Institute

ABSTRACT
This paper has investigated the efficiency of the scheduled banks in Bangladesh in financing SMEs. Taking the banks’
quarterly data from January, 2010 to June, 2013, the nonparametric approach ‘Data Envelopment Analysis’ has been
applied to measure and compare their efficiency in SME financing. Efficiency has been measured based on the output
variables ‘SME finance in service, trading and manufacturing enterprises’ and the input variables: total deposits
assigned for SME finance, cost of that fund and salaries expenses related to each category of SME finance. The study
reveals that all the banks have overall efficiency of 58 percent on average which is still below the efficient frontier
and presents the prospects for the banks to excel their performance in terms of technical and scale efficiency. Half of
the banks are found having overall efficiency less than 50 percent where PCBs are the least efficient in SME financing.
The study has important implications such as guiding public policy oriented to progress scheduled banks’ efficiency
in SME financing. The study is also expected to benefit scheduled banks in initiating proper strategic planning to
reach efficient frontier in SME financing.

Keywords: SME finance, Technical Efficiency, Scale Efficiency and Data Envelopment Analysis (DEA)

INTRODUCTION
The role of Small and Medium Enterprises (SMEs) is indispensable for overall economic development of
the developing countries like Bangladesh. SME sector contributes to around 25 percent of GDP of the
country and generates largest employment after agriculture. SMEs are termed on the basis of their fixed
asset size, sectors and employed manpower. Bangladesh Bank, the central bank of this country, defines
SMEs as having between 25 and 150 employees and more than BDT 50,000 and less than BDT 200 million
in fixed asset other than land and building, varying in internal limits considering different sectors. To uphold
the contribution of SME in the national development, SMEs need to be supported with their fundamental
requirement such as financing base to proceed to the way of growth. In providing SMEs with their required
finance, the financial institutions of Bangladesh such as banks and non-bank financial institutions are
involved highly.
Bangladesh Bank has already introduced several schemes and programs to flourish and expand SME
enterprises. Refinance scheme funded by Bangladesh Bank, IDA and ADB has been facilitated for the
development of SME sector. SMESPD (SME and Special Programs Department) of Bangladesh Bank since
its inception at the end of December 2009 has given emphasis on cluster approach in order to ensure SME
development in the country. Under SME initiatives, Banks and Non-bank financial institutions have
disbursed BDT 85,323.25 million in December, 2013. This achievement was 116 percent of the target set
for SME financing on that year. This SME finance was distributed in manufacturing (28.15 percent), service
(5.40 percent) and trading (66.45 percent) sectors. On December 2013, different group of banks provided

159
SME finance in different percentages like State-owned Commercial Banks (SCBs) 18.69 percent,
Specialized or development Banks (SBs) 8.78 percent, Private Commercial Banks (PCBs) 49.44 percent,
Islamic Banks (IBs) 20.51 percent and Foreign Commercial Banks (FCBs) 2.58 percent. Banks provided
50.96 percent of total SME finance to small enterprises in 2013.
To advance these growth histories in SME sector, such SME financing needs to be efficient to lower the
cost of banks in financing SMEs and also to provide the cost effective financing services for SMEs. Here,
efficiency implies that how much an organization uses its inputs in optimum way to producing the outputs
(Maleki Nia et al. 2012). This study is intended to expose status of both individual and group efficiency of
the scheduled banks in providing finance to SMEs of Bangladesh.

METHODS
For this study, the target population is defined to be all the scheduled banks of Bangladesh which has license
to operate under Bank Company Act, 1991 (amended in 2003).The total of 4 SCBs, 4 SBs, 7 IBs, 23 PCBs
and 8 FCBs are studied under current research work. Only Citibank, N.A. (an FCB) has been excluded from
the study for not getting adequate data of this bank on SME financing over the study periods. Thus, in total
46 scheduled banks has made up the study population. The panel data-set consists of 14 quarters starting
from January, 2010 to June, 2013. As the central bank collects each bank’s quarterly banking transaction
record, the study considered quarterly data to make the implications easy for policy initiators. Moreover,
some information has been collected from respective banks’ annual reports. In this study, ‘deposits’ is
treated as input to SME financing, focusing the two-stage DEA procedure in efficiency measurement.
However, since the study actually concentrates on measuring efficiency of SME financing, the variables
for this study comprise input and output variables related to SME finance in Service, Trading and
Manufacturing. Such as, Input variables include: Total Deposits assigned for SME finance, Cost of fund
allocated for SME finance and Salary expenses oriented to SME finance. Output variables include SME
finance (Service, Trading and Manufacturing), Cost of fund allocated for SME Finance (Service, Trading
and Manufacturing), and Salary expenses oriented to SME finance (Service, Trading and Manufacturing).
Output variables include SME Finance (Service, Trading and Manufacturing).
Here, cost of allocated fund has been derived through multiplying the total deposits targeted for SME
finance by the cost of fund (%) of the individual banks. The salary expenses oriented to SME finance have
been calculated from the proxy of [Total salary expenses in the particular quarter *(SME loans / total
loans)]. At first, all variables have been normalized through dividing the variables for each bank by their
respective number of branches to make homogenous. Then, concentrating on the output variables: SME
finance disbursed in service, trading and manufacturing sectors, major input variables: total deposits, cost
of fund and salary expenses have been segregated for each SME sector financing based on the weights of
them in total volume of SME finance by each bank. This is because, it is important to capturing as many as
possible of the relevant inputs and outputs for a DEA analysis (Cook et al. 2014). Furthermore, this study
has utilized raw data in analyzing the value of all variables for efficiency calculation to avoid the potential
problem arisen from mixed use of ratio and raw data.
DEA is a linear programming technique initially developed by Charnes, Cooper and Rhodes (1978) to
evaluate the efficiency of public sector non-profit organizations. Sherman and Gold (1985) were the first

160
to apply DEA to banking. DEA calculates the relative efficiency scores of various Decision-Making Units
(DMUs) in the particular sample. Charnes et al. (1995) propose the use of the maximum of a ratio of
weighted outputs to weighted inputs for that unit, subject to the condition that the similar ratios for all other
DMUs be less than or equal to one. That is,

Subject to: (1)

Where, , are positive known outputs and inputs respectively of the nth DMU and are the variable
weights to be determined by solving problem (1). The DMU being measured is indicated by the index 0,
which is referred to as the baseline DMU. The maximum of the objective function e0 given by problem (1)
is the DEA efficiency score assigned to DMU0. Since every DMU can be DMU0, this optimization problem
is well-defined for every DMU. If the efficiency score e0 =1, DMU0, satisfies the necessary condition to be
DEA efficient; otherwise it is DEA inefficient. It is difficult to solve problem (1) as stated, because the
objective function is non-linear and fractional. Charnes et al. (1995), however, transformed the above
nonlinear programming problem into a linear one as follows,

Subject to_
(2)

The variables defined in problem (2) are the same as those defined in problem (1). An arbitrarily small
positive number, is introduced in problem (2) to ensure that all of the known inputs and outputs have
positive weight values and that the optimal objective function of the dual problem to problem (2) is not
affected by the values assigned to the dual slack variables in computing the DEA efficiency score for each
DMU. The condition h0 =1 ensures that the base DMU0 is DEA efficient; otherwise it is DEA inefficient,
with respect to all other DMUs in the test. A complete DEA model involves the solution of N such problems,

161
each for a base DMU, yielding N different ( weight sets. In each program, the constraints are held
constant while the ratio to be maximized is changed.

RESULTS AND DISCUSSIONS


The study focuses the efficiency of the banks arranging them into several groups considering their sources
of ownership and operational basis; SCBs, SBs, PCBs, IBs and FCBs. The findings of the analysis have
been detailed below:

Results on Descriptive Statistics


SME finance is mainly disbursed under three identified sectors like service, trading and manufacturing. The
banks’ participation in SME financing under this study has been analyzed through the study period. The
highest volume of SME financing in service is disbursed by IBs over all the periods, followed by PCBs.
The remaining three group of banks’ participation is on very lower position. Besides, in case of SME
financing in trading, IBs are providing the highest volume of finance in trading except the 8th quarter under
study, followed by again PCBs. Here SCBs are disbursing more loans to trading SME enterprises again
compared to SBs’ and FCBs’. FCBs are clearly funding the lowest in SME financing disbursement to
trading sector. Though in manufacturing sector, SCBs were found to disburse the highest amount to SME
entrepreneurs in 2010 (first 4 quarters), IBs are disbursing the highest amount through the rest of the
periods. SME financing in manufacturing by FCBs is again the lowest among other bank groups.
Considering the overall participation in SME financing, IBs are the largest contributor and FCBs are the
lowest one. The other three groups are contributing variably over the periods. The overall participation of
all the bank groups except FCBs (consistently low) is contributing to develop the SME sector of the country.

Empirical Results of Data Envelopment Analysis


This section presents the results of efficiency measurement through DEA. For calculating the banking
efficiency in terms of financing, deposits and operating expenses attributed to respective financing area are
the most relevant inputs. For converting the variables into homogenous ones through dividing them by
number of branches, foreign commercial banks have recorded the highest figures of inputs and outputs
since they have least number of branches to spread enough their financing volume, except trading finance
(Exim Bank’s figure) and cost of fund for trading finance (BRAC Bank’s figure). In case of the lowest
output figures, the case has been different because of studying the quarterly data of the banks. Several banks
have no disbursement of SME finance in identified sectors in different quarters. Here, the banks were
sometimes found to disburse their SME loan, allocate deposits and entitle salaries expense only in service
or trading or manufacturing sectors in a particular quarter making zero figure.

Efficiency of the Bank Groups

162
The efficiency has been calculated devising into technical efficiency and scale efficiency for periods
January-March, 2010 to March-June, 2013 quarters. Following chart reveals technical efficiency score of
the bank groups in SME financing:

Technical Efficiency of Bank Groups


1.2
Relative Efficiency Scores

1
0.8
0.6
0.4
0.2
0
SCBs SBs PCBs Ibs FCBs

Bank Groups

Figure 1: Technical efficiency scores of bank groups (2010 Q1-2013Q2)

In technical efficiency level (Figure 1), SBs experience the highest technical efficiency (100 percent).
Though FCBs are found to disburse lowest volume of SME finance, these banks have second highest
technical efficiency (96 percent). PCBs show lowest technical efficiency (50 percent). In this study, (shown
in Appendix Table-A) one third banks (36 percent) are found to have technical efficiency less than 50
percent. All the banks have 68 percent technical efficiency on average which is below the efficient frontier.
The least technically efficient bank is ICB Islamic Bank with only 12 percent technical efficiency. Here,
nineteen banks reach technical efficient frontier. Coelli et al. (1998) showed the production on any frontier
line is technically efficient; however, all are not indicated maximum production. Consequently, the scenario
of scale and technical efficiency is almost inverse in showing the banks’ efficiency level in SME financing.
The chart below presents the scale efficiency scores:

Scale Efficiency of Bank Groups


0.9
0.88
Relative Efficiency Scores

0.86
0.84
0.82
0.8
0.78
0.76
SCBs SBs PCBs Ibs FCBs
Bank Groups

Figure 2: Scale efficiency scores of bank groups (2010 Q1-2013 Q2)

In case of scale efficiency (Figure 2), FCBs have the highest efficiency scores whereas SBs show the lowest
scores. Again in this study (shown in Appendix Table-A), only 6 percent banks have scale efficiency below

163
50 percent. Therefore, scale inefficiency at DMUs may be the lesser but it could still violent scale
economics to increase its productivity. Most banks have scale efficiency more than 50 percent where twenty
banks (almost equal number in case of TE) reach the efficient frontier. All the banks have 85.60 percent
scale efficiency on average which shows their efficiency in allocating proper weights in different scale of
operations.
In the above context, overall efficiency of bank groups has been presented in the following figure:

1
Overall Efficiency of Bank Groups
Relative Efficiency Scores

0.8

0.6

0.4

0.2

0
SCBs SBs PCBs Ibs FCBs
Bank Groups

Figure 3: Overall efficiency scores of individual banks (2010 Q1-2013 Q2)

FCBs have the highest overall efficiency scores while PCBs show the lowest scores (figure 3). Appendix
Table-A shows that half of the banks have overall efficiency is less than 50 percent. All the banks have
overall efficiency of 58 percent on average which is below the efficient frontier. The least technically
efficient bank ‘ICB Islamic Bank’ scores only 10 percent here. At last, fifteen banks reach the efficient
frontier on overall efficiency level.

CONCLUSIONS
SME financing is the basement of the SME enterprises. The scheduled banks are now instructed by the
central bank to disburse more SME finance. The more the bank will be efficient in SME financing the
greater the nation will benefit. The study exposing the banks’ efficiency level in SME financing is expected
to help related policy makers take proper policy initiatives. The quarterly efficiency evaluation is expected
to help more effective policy formulation by the central bank of Bangladesh for SME sector development.
Half of the banks are found having overall efficiency less than 50 percent. All the banks have overall
efficiency of 58 percent on average which is also below the efficient frontier. In this study, fifteen banks
reach the efficient frontier on overall efficient level. FCBs’ reach on efficiency frontier presents their
highest efficiency in both the use of technology and proper scale operations. SBs are not spreading their
scale of operations enough efficiently in spite of being capable in operating on technical frontier. SCBs,
performing at moderate level of technical and scale efficiency, are not disbursing SME finance on right
sectors as well as not reaching efficient frontier. More than half of PCBs are having decreasing returns to
scale which makes the reason behind their lowest efficiency. So in conclusion, the overall efficiency records
of the scheduled banks of Bangladesh in financing SMEs are still below the efficient frontier and present a
greater prospect yet to increase efficiency as well as take proper policy measures. That is, they should

164
diverse their scale of operations and utilize greater technology for achieving maximum output from a given
set of inputs.

APPENDIX

Table A: Frequency Distribution of Technical, Scale and Overall efficiencies of banks


Technical Efficiency Scale Efficiency Overall Efficiency
Efficiency Number of Banks Percent Number of Percent Number of Percent
Ranges Banks Banks
< 10 0 0 0 0 0 0
10-20 5 11 1 2 9 20
20-30 5 11 1 2 5 11
30-40 3 7 0 0 4 9
40-50 3 7 1 2 4 9
50-60 4 9 4 9 4 9
60-70 2 4 0 0 2 4
70-80 2 4 6 13 2 4
80-90 0 0 5 11 0 0
90-100 22 48 28 61 16 35
Total 46 100 46 100 46 100
Mean 68 85.6 58
Min 12 12 10
Max 100 100 100

REFERENCES
[1] Charnes A., Cooper, W.W. and Rhodes, E. (1978) Measuring efficiency of decision making units, European
Journal of Operations Research 6(3): 429-444.
[2] Charnes A., Cooper W.W., Lewin A.Y. and Seiford L.M. (1995) Data Envelopment Analysts: Theory,
Methodology, and Applications, London: Kluwer Academic Publishers.
[3] Coelli T., Prasada Rao D.S. and Battasse G.E. (1998) An Introduction to Efficiency and Productivity Analysis,
Boston: Kluwer Academic Publishers.
[4] Cook W.D., Tone K. and Zhu J. (2013) Data envelopment analysis: Prior to choosing a model, Omega 44: 1-4.
[5] Maleki N., Asgari A.H., Sarafraz P.A. and Ghezelbash A. (2012) A Comparative Profitability Efficiency Study
of Private and Government Banking System in Iran Applying Data Envelopment Analysis (DEA), Journal of
Basic and Applied Scientific Research 2(11): 11603-11614.
[6] Sherman D.H. and Gold F. (1985) Bank Branch Operating Efficiency: Evaluation with Data Envelopment
Analysis, Journal of Banking and Finance 9: 297.

165
166
THE DIFFERENT APPROACHES OF BANKING EFFICIENCY: A META-
ANALYSIS
LUCAS SERRÃO MACORIS
Department of Business Administration, University of São Paulo, FEARP, Monte Alegre, SP, Brazil

ALEXANDRE PEREIRA SALGADO JUNIOR


Department of Business Administration, University of São Paulo, FEARP, Monte Alegre, SP, Brazil

EDUARDO FALSARELLA JÚNIOR


Department of Business Administration, University of São Paulo, FEARP, Monte Alegre, SP, Brazil

ABSTRACT
The importance of banking efficiency has been shown to be increasingly relevant not only for researchers,
but also to decision makers. However, due to the divergence of the authors in their approaches and concepts
of measuring such efficiency, its contents still lie scattered in the literature, demanding deeper studies on
the characteristics of each of these main approaches. In this sense, this article proposes to consolidate such
a concept showing the main features of banking efficiency approaches used in other literature, using
techniques of meta-analysis and content analysis. Based on the results, it was found that despite some
disagreements, each of the main approaches for banking efficiency has distinct characteristics, in that each
is more suitable for certain specific applications, thus contributing to new empirical studies in the area.
Keywords: Banks, efficiency, efficiency approaches, frontier techniques, Meta-analysis.

INTRODUCTION
In recent decades, especially in industrialized countries, it has been observed that banks are searching for
management tools aimed at the improvement of their performance (ORAL; YOLALAN, 1990). In this way,
as they are seeking improvements, they consider the management of their operations less critical than other
factors such as profitability and some financial indicators. Furthermore, their importance has already been
observed by Sherman and Gold (1985), who observed that a decrease of 1% in operating expenses could
result in an increase of over 2% in net revenues and earnings per share.
In this way, it turns out that efficiency is an important performance indicator, and can be used as a
management tool to assist the decision-making process in various fields. Specifically in the case of banking
efficiency, this has been a horizon of research that has attracted not only academics, but also decision-
makers (LIU, 2010). Its use can be made to measure efficiency in selling services, transaction costs, service
time, risk management and the amount of funds brokered among other areas of applications relevant to
bank management. As long as there is more than one way to approach banking efficiency, this theme will
continue to be discussed in academic literature, in a way that the common characteristics of the studies
within the main approaches have yet to be.
In this sense, this study sought to fill the gap for such information, organizing and analyzing information
concerning the main approaches of banking efficiency through the use of meta-analysis. The results
generate information that provide a panorama that could assist in future studies on the proper use of banking
efficiency approaches based on their respective features.

167
METHODS AND DATA SAMPLING
The methodology chosen for the present study was meta-analysis, which is regarded as “[...] a quantitative
method that allows one to combine the results of conducted studies independently (usually extracted from
published works) and synthesize its findings or even draw new conclusions” (LUIZ, 2002, p. 407). The use
of such methodology sought to identify the theoretical-methodological settings of banking efficiency
studies that used frontier techniques.
For the objective to be achieved, the study included the following variables from an analytical perspective:
(a) approach used in the study; (b) Journals in which the article was published; (c) where the study took
place; (d) scope of analysis; (e) period of study; (f) application site; and (g) used inputs and outputs.
Regarding the data sample, forty-seven articles were considered, all classified in terms of their impact factor
(FI) being greater than 1 (one), available in the databases linked to the University of São Paulo,
characterizing the sampling as being non-probabilistic for convenience. The FI was used as a criterion for
the Qualis evaluation areas that include Administration, Accounting and Tourism of upper strata. The
collection of articles was made using the Web of Science platform, applying the following keywords: DEA,
efficiency, Data envelopment analysis, financial institutions and banks. Furthermore, other criterion for
including articles was that it must have been published after 1990 and contain at least one application of a
frontier technique to assess banking efficiency.

DATA ANALYSIS TECHNIQUE


To make it possible to analyze all the information from the articles according to the variables discussed in
the methodology, the Content Analysis (AC) technique was employed in the study. According to Granja
(1995), the technique allows one to analyze the contents of publications, speeches, and writings among
other formats through a pre-established categorization based on the material to be analyzed in the study.
Mozzato and Grzybovski (2011) stated that there is a growing interest in the use of AC as a technique in
the research fields of administration, gaining legitimacy among the qualitative analysis methods with an
emphasis on scientific rigor and research depth.
The work of Bardin (2011) was taken into account for being the largest references concerning the
technique’s use in Brazil (MOZZATO; GRZYBOVSKI, 2011). In this sense, the methodology is divided
into three phases: pre-analysis, exploration of the material and results processing. Special attention should
be given to the codification process which, according to Soriano (2013), aims to transform the data by
performing a grouping into categories so that the information can be better represented and understood. For
this purpose, a categorization by themes was made, which according to Richardson (1999, p. 233) “[...]
consists of isolating the themes of a text (data from interviews or documents) and extracting the usable
parts, according to the issue being researched, to allow comparisons with other texts that had been selected”.
The categorization was done having as basis the variables included in the meta-analytical perspective, thus
ensuring that all the needed information would be duly considered.
For each theme used in the editing of information, they were obtained through mentions in the studies, and
for the purposes of their analysis, indicators corresponding to the frequency of appearances in the articles
were made. However, in the case of the theme “approach used in the study”, not all articles contained an

168
explicit reference to their respective characteristics. In this way, the objectives of this study were also
analyzed, as well as the inputs and outputs used to perform the characterization of the information.

TREATMENT OF INPUTS AND OUTPUTS


In the encoding process, the information was, in its vast majority, easily identified in the works. However,
several difficulties were found in relation to the inputs and outputs so that these could be meta-analyzed.
First, 359 variables were accounted for, composed of 172 inputs and 187 outputs, making for a large amount
of information to be processed. In addition, many articles have proposed more than one model with different
combinations of variables, further increasing the complexity of the analysis.
Additionally, the vast majority of information relating to inputs and outputs came from data obtained in
accounting statements and/or financial indicators such as interest expenses and ROE (Return on Equity),
among others. In this way, problems such as the large heterogeneity of the information brought yet more
complexity to the analysis process.
Finally, the differences between the accounting standards of different countries where applications were
made, as well as the absence of some indicators in Brazilian accounting were factors that, along with others
mentioned above, created the need for much greater attention to detail. The solution adopted was to
categorize the inputs and outputs, creating groups of variables that, though having different names, the
meanings remained the same. Table 1 summarizes the conversion of the grouped names.

Table 1 – Conversion of variables


Variable Similar variables
Number of personnel Personnel, Total Personnel
Investments Investment
Funds Borrowed Funds, Purchased Funds, Depostis, Total deposits
Personnel Expenses Cost of labor, Total Labor Expenses, Total Personnel Expenses
Loans Total volume of loans, Total Loans
Total non interest expenses, Non interest expenses, total operational expenses,
Operating expenses
operational expenses
Operating income Non-interest income, Total non-interest income, Operational income

RESULTS AND DISCUSSION


Approach vs Journal: when reviewing the periodicals in which the articles were published, there is greater
quantity of articles from the European Journal of Operational Research, which accounts for 51%, or 24
articles. Other periodicals that stood out were Expert Systems with applications, with 11 articles, and
Omega, with 6. Furthermore, others journals include Annals of Operational Research, PlosOne, Procedia,
Journal of Econometrics and Applied Mathematics and Computation.
Application scope: the studies were divided according to the type of analysis performed: between bank
branches or between institutions. The studies between bank branches consisted of applications made in the
various branches of the same financial institution, while the applications made between banks are those
among various financial institutions. There is a greater representation in applications within banks, 60% of
the cases, while applications for bank branches are seen in 40% of the articles. It is noticeable that the
intermediation approach is most often used in cases in which analysis is made between different banks,

169
while the production approach is most used in applications between branches of the same bank of a given
institution. Such information reflects a previously mentioned fact, in that the production approach has an
advantage of better efficiency to evaluate the bank branches (BERGER; HUMPHREY, 1997).
Additionally, it is noted that the profitability approach is also the most used when applied between different
banks, with it being implemented in 70% percentage of cases. A plausible explanation for such a case is
that such an approach may be more suitable to capture the diversity of strategic responses of financial
institutions due to changes in competitive and environmental conditions. On the other hand, in the case of
bank branches, less diversity of strategic responses within a single bank is expected.
Study period of the articles: the analyzed articles were allocated into two different groups: the first, when
the study has more than one period of analysis; and the second, when the paper analyzes only a single
period. It is possible to notice that most of the works use a time series, being 27, in a total of 47 articles
analyzed, a representative percentage of 57%. The works that focused only on a single period totalled 14,
representing 29% of the articles analyzed. Additionally, six cases could not be classified due to lack of
information. The large quantity of articles that analyzed different periods can be explained by the need to
analyze differences in efficiency after changes in the environment and of the corporate strategy.
Analyzing the use of time series according to the approaches, there was a dominance of use of the
intermediation approach, with 54% of representation among the articles analyzed, the representativeness of
the approaches of production and profitability were 14% and 20%, respectively. On the other hand,
observing the studies that analyzed the efficiency of banks only in a single period, the dominance was held
given by the production approach. A possible explanation for this phenomenon may be related to the scope
of analysis. In the production approach case, which has a scope of analysis often linked to the analysis of
bank branches, it was noted that the studies prioritize short-term situations (small periods), while the
intermediation approach, more often linked to the analysis of several banks, prioritizes long-term studies.
Applications by demographic region: to report on the places with the greatest amount of applications, the
studies were implemented in 22 distinct locales. It was seen that Taiwan, with 14.9% of representativeness
among the articles, has the largest number of applications. Then, Canada, India, Brazil and Greece can be
highlighted, representing 12.8%, 10.6%, 6.4% and 6.4% of the applications, respectively. In Taiwan, a
prevalence of the intermediation approach was observed, representing 71.4% of the cases of the country.
The approach was also seen in Brazil, mentioned in three articles. On the other hand, it appears that the
production approach is the most representative on applications carried out in India, Greece and Canada, the
latter being entirely represented by this approach.
Possibly, these variations found reflect differences in the concentration of banks of each country. In
countries where there are several banks, the use of the intermediation approach is plausible, as to compare
the various forms of funding and implementation of financial resources. However, in a location where there
is a predominance of a limited number of banks, this factor can limit the effectiveness and relevance of the
analysis, directing the researcher to focus on bank branches, which can be best analyzed through the
production approach.

170
Input and output configuration: since the analyzed articles applied more than one model – and not
necessarily with the same approaches –, it should be emphasized that, by taking all of the them into
consideration, there was a total of 61 settings for inputs and outputs. As noted by Berger and Humphrey
(1997), intermediation and production have been the most widely used approaches for measuring banking
efficiency in the analyzed articles, the first having a frequency of 41% and the second 28% of inclusion in
articles. Such approaches are prevalent for defining the variables of the services used by a bank’s customers:
in the intermediation approach, financial institutions operate brokering funds with agents, while in the
production approach the banks’ focus is to produce services for account holders.
Furthermore, another approach used by the authors of several papers was the profitability approach, with
21%. Other approaches of lesser relative importance are found in the other 10%, as they appear with lower
frequency in the works reviewed here. However, that does not mean that they are unimportant, since they
are a means to obtain different aspects in the measurements of institutions’ banking efficiency (YANG,
2009).
Analyzing specifically the main approaches, the intermediation approach, composed of 79 inputs and 82
outputs, can be described as being based on a small amount of variables that had greater representativeness.
In the case of inputs, it is possible to verify that ”fixed assets“ were used 11 times in the studies, making a
representativeness of 14% within all variables used as inputs. In a more heterogeneous way, the production
approach presents a wide range of variables that were used in the analyzed studies, in both cases of inputs
as well as outputs. In total, 53 inputs and 72 outputs were verified. Finally, in terms of the profitability
approach, it was found that this is the most heterogeneous of the three approaches studied when taking the
inputs and outputs into account, showing large differences not only in matter of representativeness, but also
in the types of variables used. In all, 39 inputs and 36 outputs were analyzed, being the least amount of
repeated variables of those approaches studied.

CONCLUSIONS
It is possible to categorize studies of banking efficiency in three main approaches: intermediation,
production and profitability. During the analysis, it was possible to distinguish, among the established
criteria, the different characteristics of each of the approaches.
The intermediation approach measures the efficiency in terms of how much capital the institution managed
to broker based on the resources available and their respective costs, these being the work resources
(people), the fixed assets and the capital (funds raised).
In the case of the production approach, there has been an extension of the concept of banking intermediation
in order to change the view of banks as having only a role of intermediaries, but also of providers of products
and services to their account holders. A verified firm to represent such a function was the utilization of
operational variables at the expense of those of a financial nature.
Lastly, the profitability approach was used to obtain a distinct measure of the efficiency of banking
institutions to take into consideration the basis of both cost and expenditure, as the bank main revenue base.
As much as their features are similar to those of the intermediation approach, it would appear that the
variables used as inputs and outputs are much more heterogeneous, with respect to the use of the authors,

171
and are very different from those used by the intermediation approach. Table 2 summarizes the main
characteristics of each of the approaches based on the results found.

Table 2 – Summary of the key features by approach


Intermediation Production Profitability
Between different
Scope of analysis Between different banks Between bank branches
banks
Study period Multiple periods Single period Multiple periods
Most applied locations Taiwan, Brazil and Greece Canada, India and Greece Taiwan and India
Fixed Assets, Number of Personnel, Personnel Expenses,
Number of Personnel,
Most used inputs Personnel Expenses, Interest Non personnel
Operating Expenses
Expenses, Funds expenses
Loans, Other earning assets,
Non-interest income,
Most used outputs Investments, Saving Deposits, Short Deposits, Loans
Interest income
Term Deposits
% of inputs with more than
62% 36% 38%
one apparition
% of outputs with more than
41% 24% 17%
one apparition

Given the generated results, this study contributes with relevant information for the development of new
studies in the area of banking efficiency, in a more appropriate way, according to the specific features of
how banking efficiency approaches are adopted in their applications.

REFERENCES
[1] Oral, M.; Yolalan, R (1990) An empirical study on measuring operating efficiency and profitability of bank
branches. European Journal of Operational Research, v. 46, n. 3, p. 282-294.
[2] Liu, S (2010) Measuring and categorizing technical efficiency and productivity change of commercial banks in
Taiwan. Expert Systems with Applications, v. 37, n. 4, p. 2783-2789.
[3] Luiz, A. J. B (2002) Meta-análise: definição, aplicações e sinergia com dados espaciais. Cadernos de Ciência e
Tecnologia, v. 19, n. 3, p. 407-428.
[4] Granja, E. C. Produção científica: dissertações da IPUSP – 1980/1989 - Universidade de São Paulo.
[5] Mozzato, A. R.; Grzybovski, D (2011) Análise de conteúdo como técnica de análise de dados qualitativos no
campo da Administração: potencial e desafios. Revista de Administração Contemporânea, v. 15, n. 4, p. 731-747.
[6] Richardson, R. J. (1999) Pesquisa social: métodos e técnicas. 3. ed. São Paulo: Atlas.
[7] Sherman, H. D.; Gold, F (1985) Bank branch operating efficiency. Journal of Banking and Finance, p. 297 315.
[8] Soriano, F. F. (2013) Gestão da Armazenagem: uma análise do sistema de gestão WMS. 111 f. Masters
Dissertation. FEARP, USP. Ribeirão Preto.
[9] Berger, A. N.; Humphrey, D. B. (1997) Efficiency of financial institutions: International survey and directions
for future research. European Journal of Operational Research, v. 98, n. 2, p. 175-212.
[10] Yang, Z. (2009) Bank Branch Operating Efficiency: A DEA Approach. Proceedings of the International Multi
Conference of Engineers and Computer Scientists, Vol II IMECS.

172
USING DYNAMIC DEA TO EVALUATE THE EFFECTIVENESS OF
MACHINE-BUILDING ENTERPRISES OF UKRAINE
LEONID GALCHYNSKY
Department of Mathematical Modeling of Economic Systems, Faculty of Management and Marketing, National
Technical University of Ukraine “Kyiv Polytechnic Institute”, 02056, Kyiv, Ukraine

ABSTRACT
This study considers the problem of constructing multi-criteria evaluation of the efficiency of enterprises in machine-
building complex of Ukraine. This industry always was ranked an important place in the Ukrainian economy. But the
economic crisis of recent years adversely affected the economy of Ukraine, especially in the machine-building
complex. To overcome the crisis through radical reforms in this area should be relevant method for evaluating the
effectiveness of the effectiveness of machine-building enterprises. It is shown that the methodology of the Data
Envelopment Analysis (DEA) is one of the most relevant for evaluation of multifaceted activities of enterprises. The
comparison of efficiency scores for static and dynamic models is given. It has been established that the dynamic
version of the DEA machine-building enterprises gives a more accurate estimation of efficiency compared to the static
version.
Keywords: Efficiency of enterprises, multi-criteria; Data Envelopment Analysis; inter-temporal production
technology; machine-building complex of Ukraine.

INTRODUCTION
At the beginning of 2014 the machine-building complex of Ukraine had near 2000 companies. The share
of machine–building complex in Ukraine exceeds 15%, and GDP is about 12%. In Ukraine, the growth of
engineering in post-crisis period, according to state statistics in the rate of growth outpaced the industry as
a whole. However, the global crisis sharply hit the machine-building complex of Ukraine. As a result, a
sharp decline in investment activity in 2009 in major markets has led to a decrease in exports of machinery
by 37 % - to 6.89 billion USD and sales of engineering products decreased by 40 % after an increase of 24
% in 2008, while the decline in production was almost 45 %, and in other areas - 6-26, 7 %. Thus, despite
the fact that in the first half of 2010 due to a lower global crisis there was an increase in the index of
production machinery, which in January-July this year amounted to 131.3 % over the same period as of in
2009, a steady upward trend over the years of 2010 to 2013 was observed (State Statistics Committee of
Ukraine, 2014). It is obvious, that without significant investments the Ukrainian machine-building industry
will not be able to overcome the crisis. This in turn will require substantial upgrading of engineering
companies and a role here should be played by an objective evaluation of the effectiveness of the company.
The main goal of this work is to develop a relevant methodology for evaluating the effectiveness of
machine-building enterprises of Ukraine.

METHODS
For the purpose of this study, the DEA technique (Charnes A., Cooper W.W., Rhodes E. (1978)) has been
employed to compute evaluating the effectiveness of machine-building enterprises of Ukraine. The DEA
method involves the use of linear programming methods to construct nonparametric piecewise linear hulls
or boundaries of the observable data. The effectiveness of the objects is calculated as the distance of the
object to the envelope.

173
But basic DEA models do not take into account the relationship of the production of one period to another.
Exactly for the machine-building enterprises of their long production cycle, this assumption reduces the
reliability of an efficient estimation, for in uneven conditions of supply and demand, this significantly
growing factor impacts the prior periods on the state of the current period. We employed the dynamic model
approach that recognizes the inter-temporal relationship between the inputs and the production. Färe and
Grosskopf (1996) proposed a number of models that take into account several types of intertemporal
technological interdependencies in different periods.
The (Emrouznejad and Thanassoulis ( 2005) ) work defines the set of dynamic production capabilities,
which evaluates the behavior of enterprises not in one time period but for a certain period, which consists
of several time periods. This dynamic movement, considering the level of capital in a period of time as an
input, results in production in future periods.
In recent years, many works who publish the results of applied research in various sectors of the economy
using a dynamic model of DEA. One of the most interesting research is to evaluate the effectiveness of the
Spanish ports (Diaz-Hernandez, Martinez-Budría, Salazar-Gonzalez (2014)).
Although the topology of the structure of manufacturing machine-building company can be quite
challenging, to get evaluation the efficiency of the enterprise we can use relevant scheme was proposed
(Nemoto and Goto (2003)), because at the heart of any enterprise is incorporated production cycle. Then
we will follow this methodology. As a result, we can formulate an inter-temporal scheme production
process using quasi-fixed variables:

Figure 1 - Production technology using inter-temporal parameters.

Source: Nemoto and Goto (2003).


Where xt- is a vector of variable inputs of 11 dimensions which are used during t period.; kt-1 - quasi-fixed
vector of inputs at the beginning of period t of m  1 dimensions obtained at the end of the previous period;
yt - output vector of n x 1 dimensions, the resulting production process at the end of period t; kt - quasi-
fixed vector of inputs of l 1 dimensions, the resulting production process at the end of period t;
Goto and Nemoto found that the incorporation of intertemporal effects leads the data envelopment analysis
to dynamic programming problem with all the possible problems of computational nature. But they offered
a very ingenious process of reduction of the general problem of dynamic DEA by sequence of linear
programming as follows:

174
1. First problem solved separate static optimization for given values quasifixed inputs.
2. Next static efficiency divided into technical and allocative (resource allocation). As static DEA, technical
efficiency level is seen as a uniform radial compression of input variables.
3. The final calculated allocative effectiveness as residue after removal of technical efficiency with static.
This algorithm is a sequence of LP problems for each of these stages and for each DMU. Intertemporal cost
frontier C (k0) is used to calculate the value of empirical intertemporal boundaries using non-parametric
method DEA, which will provide data shell, which limits the set of production possibilities. General linear
programming problem, which will provide an estimate for С (k0) for each of N DMUs is:

C(k0) = min T = ∑𝑇𝑡=1 γt (𝑤𝑡 φ𝑡𝑥𝑡 + 𝑣𝑡𝑘𝑡−1 )


𝑥𝑡 |λt |𝑘𝑡 |φ𝑡⟩t=1
⟨ (1)

s.t. Xt≤ λt ,
(2)
Kt-1 λt ≤kt-1, t = 1,2…..,T
(3)
Kt λt ≥kt, t = 1,2…..,T-1
(4)
Yt λt ≥yt t = 1,2,….,T
(6)
I’λ=1, t = 1,2,..., ,T
(7)
xt≥0, λt ≥0, t= 1,2,…..,T
(8)
k0=kb, t = 1,2,….,T
(9)
Xt λt ≤ φtxt t = 1,2..…,,T

Where, γt – discount rate, wt and vt - for variable inputs and quasi-fixed inputs in period t, respectively.
The initial values k (0) of the quasi-fixed inputs k0 are represented as kb: eλt=1 - means that an assumption
arises of the changeable scale effect at the enterprise, where e-unit vector of dimension N × 1.
The overall efficiency is calculated as the ratio between the effective cost of production (C (k0)) and the
discounted amount of expenses for the period from 1 to T (Cd), namely: OE = C (k0)/Cd.
This value of this index, which is lower than 1 means having smaller units accumulated over a period of
inefficiency. Further, this measurement of overall efficiency is decomposed into three types of efficiencies:
technical static (TE), a static allocation of resources (AE) and dynamic (DE).

OE=TE AE  DE . The first two components are defined as static because they refer to the inefficient use
of variable inputs and decide upon the existing values of quasi-fixed inputs. The dynamic part of
measurements on improving the effectiveness and impact on the value means that the company does not
use the best quasi-fixed factors, indicating the possibility of reducing costs by reducing the use of quasi-
fixed inputs. In order to calculate the three components of overall efficiency, it is necessary first to estimate
the static efficiency (SE), as the product of technical efficiency and allocative efficiency of static, i.e., SE
= TEAE. Dynamic efficiency is calculated as part of the performance, which is independent of the static

175
efficiency. Impact on the costs associated with the use of sub-optimal variable inputs is measured using the
static allocative efficiency index proposed by Farrell (1957): AE =SE/TE.
The static model considers all inputs (xt and kt) as variables, in addition, it ignores quasi-fixed inputs at the
end of the period, as factors that will influence a subsequent period. In this approach, the calculation of
technical efficiency statistics (TES), allocative efficiency statistics (AES) and overall performance index
(OES), based on a static model of DEA, it is necessary to view the previous linear programming to evaluate
all the components needed for the calculation of indicators CS(k0) and CSTE .
The study was conducted on the data and on the activities of machine-building enterprises of Ukraine. For
analysis, the data was obtained from standard financial statements as reports on financial results and balance
sheets. After the analysis of the production of each of the studied companies settings were selected which
together characterize the activity of a building enterprise. As input variables were selected 3 parameters:
x1-the average number of employees, pers./quarter; x2-material costs of production, UAH/quarter; x3-staff
costs,. UAH /quarter. The value of each of the selected input parameters directly affect the production
process and its results. The value of each selected input parameters directly affects the production process
and its results. The output variables were chosen as follows: y1 - production volume. UAH/quarter; y2 -
sales, thus. UAH/quarter. The feasibility of such a choice of initial parameters is that their values are
identical to the results of the production cycle. As a quasi-fixed variables were selected: k1- amount of
working assets, thousand. UAH. / quarter; k2 -coefficient of wear of capital assets.
Feasibility choice of output parameters is that their values are identical to the results of the production cycle.
In fact, the machine-building enterprise has to produce income. This is possible only through the production
and human capital. However, the very nature of the productive process also depends on the condition of
fixed assets, including the size and condition of equipment and the use working assets - optimizing
inventories, work in progress and others.
Results and discussions
For calculations assess the effectiveness of enterprises machine-building complex Ukraine were used
sample data 9 enterprises (State Property Fund of Ukraine 2014). Calculations were carried out with the
value of the discount rate of 10%, characteristic of the study period. As a tool for calculations was developed
program «DEAM». The program includes a set of algorithms for static and dynamic DEA models. This
software has been implemented in the programming language C# using «.NET FRAMEWORK 4.0» in the
programming environment «Visual Studio 2010". To calculate the efficiency of the enterprise as for the
case without quasi-fixed outputs and taking into account the need to solve a set of linear programming for
each DMU. The choice of conditions for a particular task listed in Table 1.

Table 1: Conditions for a particular task of cost function


Name cost function Denotation Variables Restrictions Value of radial measure of static
technical efficiency
Overall cost function without quasi-fixed COEs (k0) xt, λt, kt (2)-(8) 1
outputs
Static cost at quasi-fixed outputs CSE xt, λt, (2)-(7) 1
Technical cost at quasi-fixed outputs CTE φt, λt (3)-(7), (9) Variable values

176
Calculated parameters of efficiency for every enterprises for case without quasi-fixed are outlined in Table
2.

Table 2. Efficiency indices


Enterprise Technical Static Allocative Static Static Dynamic Static Overall
efficiency technical eff-cy (AE) alloc. eff-cy eff-cy overall eff-cy
(TE) efficiency eff-cy (SE) (DE) eff-cy (OE)
(TEs) (AEs) (OEs)
AMZ 0.973 0,973 1 1 0.973 0.709 0.973 0.68
KRALMA 0.929 0.929 1 0.9 0.929 0.791 0.88 0.73
GORMAS 0.518 0.518 0.911 0.901 0.471 0.794 0.46 0.37
SWERMA 0.768 0.768 0.822 0.812 0.631 1 0.62 0.63
RUMAS 0.95 0.95 1 1 0.95 0.811 0.95 0.77
PETROMA 0.963 0.963 1 1 0.963 1 0.963 0.96
DRUMS 1 1 1 1 1 1 1 1
BERMA 0.749 0.749 0.952 0.92 0.713 0.982 0.68 0.71
DROMS 0.717 0.717 0.632 0.632 0.453 0.84 0.45 0.38
KRYROMS 0.964 0.964 1 1 0.964 0.853 0.964 0.822
AVERAGE 0.9331 0.853 0.9657 0.901 0.91 0.952 0.796 0.86

The results outline the average overall efficiency of 86.3 % during the period between 2009 and 2013. That
means that irregular usage of variables and quasi-fixed outputs led to over usage of funds of 13,7% on
average. The distribution of overall inefficiency highlights the importance of dynamic component, because
the over usage of quasi-fixed inputs leads to inefficiency on average of 7.9 %. As shown in Table 2, one
company - DRUMS is Pareto-efficient. This company is on the border of efficiency and serves a benchmark
for inefficient enterprises. Also, the results can identify obvious outsiders (GORMAS and DRUMS), values
of the efficiency of such plants significantly lower in almost all components. Effective were the company
with the largest production scale, so you can make some interim conclusion that the scale of the enterprise
directly affects its efficiency. Static version, in addition to ignoring the production of quasifixed inputs, the
radial reduction of the variable inputs affects not only the variable inputs but the quasifixed inputs as well.
Comparing the results of calculations for static model the activities of enterprises indicate that static model
overestimates the total inefficiency, which in this case averages 20.2%, or 6.6% higher than using a dynamic
model. The same thing happens with technical inefficiency and ineffectiveness of resource allocation,
which revalued by 12.8% and 1.1%, respectively. At the enterprise level analysis shows the most
important aspects of the comparison between the two models. First, all enterprises which are effective in
static models also are effective and dynamic, and vice versa. Second, lowering efficiency static model is
shown for each inefficient enterprises. Also, this result applies to all components of efficiency, which means
that not only the overall efficiency was underestimated for static models, but also a technical efficiency and
allocative efficiency.
Calculations of effect return of scale do not give a clear answer on the impact it on efficiency score. For
example, the value of efficiency return of scale one of the leaders on the overall efficiency RUMAS and
outsider GORMAS estimated at 0.98 and 0.97 respectively. Obviously, enterprise that are effective under
both models give a result for this index, which is unity. From a quantitative point of view, the average
underestimation of the overall performance index is 6.6%. But for individual enterprises analysis shows
very significant differences. This applies, for example, DROMS and KRALMA, where the difference in

177
the values of the static and dynamic performance is 39.1% and 7.1%, respectively. This result confirms
many facts of this kind, was published in the literature, for example, for Japanese energy companies
(Nemoto and Goto (2003)) and for Spanish ports (Díaz-Hernández, Martínez-Budría, Salazar-González
(2014)). Importantly, such a noticeable difference is characteristic for the allocative efficient. As a result,
the overall performance indicator was underestimated in static model. These results demonstrate a disregard
by dynamic components leads to distortions, both in technical performance and allocative efficiency.
CONCLUSIONS
Presented in this paper results in the development of methodology for assessing the efficiency of enterprises
of machine-building complex of Ukraine show that the quantification of the efficiency of enterprises can
be successfully implemented on the basis of the methodology DEA. The presence of multi-objective
evaluation allows the flexibility to decide to take adequate measures to improve the activities of business
units. Methods checked for machine-building enterprises. And the specificity of this complex production
requires the use of dynamic variant methodology DEA, it allows to take into account factors that are
intertemporal nature.
In this article we calculated the efficiency of the machine-building enterprises of Ukraine based on dynamic
DEA model using data for the period 2009 to 2013. The indicator for the overall efficiency of the presented
sample is 13,2%, while overall efficiency index, calculated on the static DEA model is 20.2%, more 8%
accurate assessment. Presented in this paper results in the development of methodology for assessing the
efficiency of enterprises show that the quantification of the efficiency of enterprises can be successfully
implemented on the basis of the methodology DEA. The presence of multi-objective evaluation allows the
flexibility to decide to take adequate measures to improve the activities of business units, including
investments in manufacturing, improve management, or even to decide on bankruptcy and liquidation. The
specifics of this complex production requires the use of dynamic variant methodology DEA, it allows to
take into account factors that are intertemporal nature Wrong calculation of inefficiency can lead not only
to wrong business decisions at the operational management of the enterprise, but to wrong strategic
decisions such as getting investment for retooling.

REFERENCES
[1] Charnes A., Cooper W.W., Rhodes E. (1978) Measuring the efficiency of decision-making units, European
Journal of Operational Research 2 (6): 429-444.
[2] Emrouznejad, A. and Thanassoulis, E. (2005) A mathematical model for dynamic efficiency using data
envelopment analysis. Applied Mathematics and Computation, 160, 363–378.
[3] Färe, R. and Grosskopf, S. (1996) Intertemporal Production Frontiers: with Dynamic DEA. Kluwer Academic
Publishers.
[4] Farrell, M.J. (1957) The Measurement of Productive Efficiency. Journal of the Royal Statistical Society, Serie A,
General, 120, Part 3, 253-81.
[5] Nemoto, J. and Goto, M. (1999) Dynamic Data Envelopment Analysis: modelling intertemporal behaviour
of a firm in the presence of productive inefficiencies. Economics Letters, 64, 51-56.
[6] Nemoto, J. and Goto, M. (2003) Measurement of dynamic efficiency in productions: an application of Data
Envelopment Analysis to Japanese electric utilities. Journal of Productivity Analysis, 19, 191-210.

178
[7] Diaz-Hernandez, Juan Jose Martinez-Budria, Eduardo Salazar-Gonzalez. Measuring Cost Efficiency in the
Presence of Quasi-fixed Inputs using Dynamic Data Envelopment Analysis: The Case of Port Infrastructure
(2014) Maritime Economics & Logistics ,16, 111-126.
[8] Access to Public Information: [Electronic resource] //State Statistics Committee of Ukraine, 2014. URL: http://
www.ukrstat.gov.ua (as of 15/02/2014)
[9] Access to Public Information: [Electronic resource] //State Property Fund of Ukraine. URL:
http://www.spfu.gov.ua/_layouts/SPFUSiteDefinition/AccessToPublicInformation.aspx (as of 15/02/2014)

179
180
X-EFFICIENCY OF INDIAN COMMERCIAL BANKS AND THEIR
DETERMINANTS OF SERVICE QUALITY: A STUDY OF POST GLOBAL
FINANCIAL CRISIS
GAGANDEEP SHARMA
Department of Economics, G.G.D.S.D College, Chandigarh, India (corresponding Author)

DIVYA SHARMA
Department of Commerce, D.A.V College, Chandigarh, India

ABSTRACT
The Indian banking system is an important component of the economy and its sustainability is essential to attain
economic growth. In the present scenario, the Indian banking system needs to be flexible and competitive. For the
survival and efficiency of the banks, having larger customer satisfaction base is now a necessary condition. In 2007
Global banking sector was severely affected by financial crisis and the economies took time to recover from this crisis.
Therefore, it was felt important to study efficiency after global financial crisis. X-efficiency is an economic term that
gives the ways by which a firm is utilizing its resources and labour to produce results. In this paper an attempt has
been made to study the X-efficiency of Indian commercial banks for the post financial crisis period i.e. 2007-14 using
DEA method and also identifying important determinants of service quality of efficient banks by applying factor
analysis. The findings show that Indian public sector banks have larger network but the percent of efficient private
sector banks is more than the public sector banks. As regards service quality dimensions, on the basis of customers’
perceptions and expectations both reliability and empathy dimensions are found to be important.

Keywords: X-efficiency, Data Envelopment Analysis, SERVQUAL, Factor Analysis

INTRODUCTION
The banking sector plays a critical role in promoting economic prosperity and stability of any economy. In
India, the banking sector has changed since 1991 when the process of opening up of economy started. The
entry of new private sector banks has given hard competition to the domestic banks. The introduction of
globalization, liberalization and privatization has uplifted the technological standards of Indian banking
industry. Indian banking sector has always been under governmental control. In fact, this helped the banks
to survive the global financial crisis of 2007 without any adverse developments. Management of a bank is
always concerned about the proper utilization of inputs to get the desired combination of outputs, through
the right channel of operations. (Saha and Ravisankar, 2000). Bank efficiency is important in achieving the
competitive edge for survival in a globalised banking industry. X-efficiency is an economic term that gives
the ways by which a firm or business is utilising its resources and labour to produce results.
In this era of intense competition, banks can be successful in maintaining their business if they operate
efficiently. While operating efficiently the quest for quality improvement is a highly desired objective.
Service quality occurs from a customer point of view when the service firm provides services to the
specifications that satisfy their needs. The banks that maintain high standards of service quality are able to
retain and enhance the number of customers. Service quality is conceptualized as the gap between
customers' expectations about a service and their subsequent perception of service performance (Gronroos,
1984; Lehtinen and Lehtinen, 1991; and Parasuraman et al., 1985, 1988).

181
The present study tries to explore the efficiency of the Indian banking sector for the post financial crisis
period and uses SERVQUAL model (Parasuraman et al., 1985, 1988) to judge the customers’ expectations
and perceptions of the services offered by efficient banks on the basis of five dimensions: Responsiveness,
Reliability, Empathy, Tangibles and Assurance.
Objectives

The objectives of the study have been:


1. To study the X-efficiency of Indian commercial banks for the post financial period i.e. 2007-14.
2. To identify important determinants of service quality of efficient banks.

METHODS
The present study considers all the public (26) and private (19) sector banks operational in India during the
post financial crisis period 2007-14. Secondary data was collected to study the efficiency of the public and
private sector banks. The secondary data was extracted from Prowess Database (CMIE) and National
Accounts Statistics published by Center for Monitoring Enterprises, Report on Trend and Progress in
Banking and RBI Bulletins-publications of Reserve Bank of India. In the study, a questionnaire (five-point
Likert scale) was administered to the customers of the banks. The questionnaire consisted of 10 statements
each to analyse customers’ expectations and perceptions respectively from various services offered by their
banks. 50 Customers (on the basis of convenience and purposive sampling) from each of the six efficient
banks were chosen as the respondents. The efficient banks have been decided on the basis of findings of
DEA and Kruskal-Wallis H test.
For the purpose of Data Envelopment Analysis (DEA), Loans and advances, Net fixed assets ratio and
Financial services expense ratio have been used as the inputs in the study. The outputs have been Deposits,
Return on assets (ROA) and Non-performing assets ratio (NPA Ratio)

RESULTS AND DISCUSSIONS


Table 1 (Appendix 1) depicts the technical efficiency scores and scale efficiency scores of all the public
and private sector banks during 2007-14. As shown in the table under public sector banks, Indian bank,
Punjab National Bank and State Bank of India are found to be technical efficient with highest scores under
the Constant Returns Scale Model (CRS). It is evident that IDBI Bank (0.4513), Vijaya Bank (0.5184) and
Bank of India (0.5309) are relatively most inefficient. In other words, these banks can increase the level of
output by 54.87 percent, 48.16 percent and 46.91 percent respectively with the same level of input. 8 out of
26 banks have been judged as technical efficient. Under private sector banks, HDFC Bank, Axis Bank and
RBL Bank are found to be technical efficient with highest scores under the Constant Returns Scale Model
(CRS). It is evident that South India Bank (0.5784), ING Vysya Bank (0.5964) and IndusInd Bank (0.6463)
are relatively most inefficient. In other words, these banks can increase the level of output by 42.16 percent,
40.16 percent and 35.37 percent respectively with the same level of input. 11 out of 19 banks have been
judged as technical efficient.

182
In public sector bank category, State Bank of Bikaner & Jaipur, Bank of Baroda and Punjab National Bank
are relatively most efficient bank and IDBI Bank, Canara Bank and Central Bank of India are most
inefficient banks on the basis of pure technical efficiency after post financial crisis period. On the other
HDFC Bank, Axis Bank and RBL Bank are relatively most efficient banks and Karur Vysya Bank, IndusInd
Bank and South Indian Bank are most inefficient banks on the basis of pure technical efficiency after post
financial crisis period in private sector banks category.
The X-efficiency of IDBI Bank (public sector bank) is 31.07 percent during 2007-14. ING Vysya Bank is
40.76 percent X-efficient during 2007-14 under private sector banks category. This means that IDBI Bank
and ING Vysya Bank had used only 31.07 percent and 40.76 percent respectively of the resources actually
employed during this period in order to produce the same level of output. It is evident that the IDBI bank
and ING Vysya had failed to enhance the level of output by 7 times and 3.5 times respectively with the
same input during this period. Thus, IDBI Bank is highly X-inefficient bank in public sector category and
ING Vysya Bank is highly X-inefficient bank in private sector category during 2007-14.
Kruskal Wallis H test has been used to study the variation in mean values of different efficiency scores of
26 public sector and 19 private sector banks during 2007-14. There has been a statistically significant
difference in efficiency score among the public sector banks. Chi-square value 2 (25) = 97.372, p = 0.000
with mean rank efficiency score of 109.50 for 8 public sector banks. Co-efficient of variation in efficiency
(18.13%) shows the level of variation because of different public sector banks. Similarly, there has been a
statistically significant difference in efficiency score among the private sector banks. Chi-square value 2
(18) = 87.434, p = 0.000 with mean rank efficiency score of 67.50 for 11 private sector banks. These banks
have been found to be the top banks on the basis of mean efficiency score. Co-efficient of variation in
efficiency (23.75%) shows the level of variation because of different private sector banks.
Further, analysis was based on the data collected from the customers of six efficient banks in Chandigarh.
The six efficient banks found by DEA have been State Bank of India, Punjab National Bank, Union Bank
of India, Axis Bank, ICICI Bank and HDFC Bank.
Out of ten statements for the five dimensions of SERVQUAL, three factors have been extracted both in
case of customers’ perceptions and expectations by using Principal Component Analysis. In case of
customers’ perceptions, the first three factors have loading values 3.233, 1.238 and 1.092 respectively. The
cumulative percent of variance explained by the first three factors is 55.630 percent. In case of customers’
expectations, the three factors account for 3.198, 1.148 and 1.043 variation respectively and the cumulative
percent of variance being 53.891.
As shown in Table 2 (Appendix 2), three factors have been extracted about the customers’ perceptions. In
the first resulting dimension ‘responsiveness’, the statement with highest loading is ‘The fee charged by the
bank is reasonable.’ The second dimension ‘reliability’ includes the statements ‘Bank takes keen interest in
solving customers’ problems’. The third dimension ‘empathy’ includes statements ‘Bank provides special
services for certain types of customers’. As regards customers’ expectations for service quality, in the first
resulting dimension ‘tangibility’, the statement with highest loading is ‘Bank should provide anywhere
anytime banking’. The second dimension ‘empathy’ includes the statement ‘The staff of the bank should

183
understand the specific needs of the customers’. ‘Reliability’ which is the third dimension includes the
statement ‘Bank should take keen interest in solving customers’ problems’.
The results of factor analysis show that customers’ perceive tangibility dimension of SERVQUAL model
as the most important dimension. On the basis of customers’ expectations responsiveness has been
considered important. In this order, the other two important dimensions are reliability and empathy in both
perceptions and expectations.

CONCLUSION
The sustainability of the banking sector of any economy plays a vital role in its growth. The economy and
specifically the banking sector are affected by things like the crisis of 2007. The paper attempted to study
the efficiency levels of the public and private sector banks after this crisis. It was highlighted that although
the public sector banks are larger in number as well as holding larger share of Indian banking sector but
due to high rate of non-performing assets and weak returns on assets private banks are performing better.
11 out of 19 private sector banks were found X-efficient i.e. more than 50 percent as compared to 30 percent
public sector banks. One of the reasons behind the weak performance of public sector bank is rural debt
ratio in India, and majority of it ultimately add-up as non-performing assets. Aggressive lending by banks
has rendered many loans non-performing, impacting the banks’ profitability. The macroeconomic situation
in India is driving private sector banks to sharpen their focus on emerging sector and rural markets to boost
growth. Efficient banks in private sector such as HDFC bank, ICICI bank and Axis bank are setting up their
branches to strengthen their rural presence. Private sector banks have a long way to go because their
contribution to rural sector is less than public sector banks. Further, the factors important in service quality
have been studied for the efficient banks. The two dimensions which the customers perceive and expect are
reliability and empathy. The customers expect these two dimensions to be improved.

APPENDICES:
Table 1. Efficiency Scores of Public and Private Sector Banks after Post Financial Crisis 2007-14

Public Sector Banks


S.No Banks TE PTE AE SE OE
1 Allahabad Bank 0.8201 0.9341 0.8748 0.878 0.7174
2 Andhra Bank 0.7335 0.972 0.7866 0.7546 0.577
3 Bank of Baroda 0.7874 1 0.807 0.7874 0.6354
4 Bank of India 0.5309 0.8498 0.7095 0.6247 0.3767
5 Bank of Maharashtra 0.5561 0.9565 0.7286 0.5814 0.4052
6 Canara Bank 0.6797 0.7947 0.7902 0.8553 0.5371
7 Central Bank of India 0.4857 0.8012 0.7015 0.6062 0.3407
8 Corporation Bank 0.6857 1 0.7726 0.6857 0.5298
9 Dena Bank 0.6731 0.9703 0.7705 0.6937 0.5186
10 IDBI Bank 0.4513 0.773 0.6884 0.5838 0.3107
11 Indian Bank 1 1 1 1 1
12 Indian Overseas Bank 0.5545 0.8483 0.7438 0.6537 0.4124
13 Oriental Bank of Commerce 1 1 1 1 1
14 Punjab & Sind Bank 0.6965 0.8892 0.7611 0.7833 0.5301
15 Punjab National Bank 1 1 1 1 1
16 State Bank of Bikaner & Jaipur 1 1 1 1 1
17 State Bank of Hyderabad 0.6316 0.9012 0.8014 0.7008 0.5062
18 State Bank of India 1 1 1 1 1

184
19 State Bank of Mysore 1 1 1 1 1
20 State Bank of Patiala 0.9065 0.9869 0.924 0.9185 0.8376
21 State Bank of Travancore 0.9037 0.98 0.9227 0.9221 0.8338
22 Syndicate Bank 0.6111 0.9627 0.7891 0.6348 0.4822
23 UCO Bank 0.6612 0.9272 0.8071 0.7131 0.5337
24 Union Bank of India 1 1 1 1 1
25 United Bank of India 1 1 1 1 1
26 Vijaya Bank 0.5184 0.8458 0.7284 0.6129 0.3776

Private Sector Banks


S.No Banks TE PTE AE SE OE
1 Axis Bank 1 1 1 1 1
2 Catholic Syrian Bank 1 1 1 1 1
3 City Union Bank 0.9857 1 0.8728 0.9857 0.8603
4 Dhanlaxmi Bank 1 1 1 1 1
5 Federal Bank 0.7317 0.8516 0.7496 0.8592 0.5485
6 HDFC Bank 1 1 1 1 1
7 ICICI Bank 1 1 1 1 1
8 ING Vysya Bank 0.5964 0.881 0.6834 0.677 0.4076
9 IndusInd Bank 0.6463 0.7685 0.6779 0.841 0.4381
10 Jammu & Kashmir Bank 0.8602 0.9416 0.7877 0.9136 0.6776
11 Karnataka Bank 0.791 0.9908 0.7423 0.7983 0.5872
12 Karur Vysya Bank 0.6575 0.7409 0.7223 0.8874 0.4749
13 Kotak Mahindra Bank 1 1 1 1 1
14 Lakshmi Vilas Bank 1 1 1 1 1
15 Nainital Bank 1 1 1 1 1
16 RBL Bank 1 1 1 1 1
17 South Indian Bank 0.5784 0.7808 0.7079 0.7408 0.4094
18 Tamilnad Mercantile Bank 1 1 1 1 1
19 Yes Bank 1 1 1 1 1

Table 2. Rotated Factor Loading for Perceptions and Expectations of Customers.

Component
Panel A- Customers’ Perceptions ( Rotated Component Matrixa) 1 2 3
P1 Bank sends account statements to the customers regularly.
P2 Bank takes keen interest in solving customers’ problems. 0.789
P3 The bank is equipped with the latest and best equipments.
P4 The bank provides anywhere anytime banking.
P5 The fee charged by the bank is reasonable. 0.748
P6 The staff gives the customers individual attention.
P7 The staff of the bank is friendly, polite and courteous.
The bank keeps the customer update in the case of the services, which takes some days for
P8
completion.
P9 The staff of the bank understand the specific needs of the customers.
P10 The bank provides special services for certain types of customers. 0.746
Panel B- Customers’ Expectation (Rotated Component Matrixa)
E1 Bank should send account statements to the customers regularly.
E2 Bank should take keen interest in solving customers’ problems. 0.774
E3 Bank should be equipped with the latest and best equipments.
E4 Bank should provide anywhere anytime banking. 0.699
E5 The fees charged by the bank should be reasonable.
E6 Bank should give individual attention to each and every customer.
E7 The staff of the bank should be friendly, polite and courteous.
E8 If the service requires long time bank should update the customers from time to time.

185
E9 The staff of the bank should understand the specific needs of the customers. 0.735
E10 Bank should provide special services for certain types of customers.
Extraction Method: Principal Component Analysis. , Rotation Method: Varimax with Kaiser Normalization.

REFERENCES
[1] Al-Tamimi H. A. H., Al-Amiri A. (2003) Analyzing Service Quality in the UAE Islamic Banks, Journal of
Financial Services Marketing (2): 119-132.
[2] Ayadi, Ines (2013) Determinants of Tunisian Bank Efficiency: A DEA Analysis, International Journal of
Financial Research 4(4): 128-139.
[3] Chansarn S. (2008) The relative efficiency of commercial banks in Thailand: DEA Approach, International
Research Journal of Finance and Economics 18: 53-68.
[4] Farrell M.J. (1957) The Measurement of Efficiency of Production, Journal of the Royal Statistical Society (Series
A) 120 (3): 253-281.
[5] Gronroos C. (1984) A Service Quality Model and its Marketing Implications, European Journal of Marketing 18
(4): 36-44.
[6] Kangis, Peter, Voukelatos, Vassilis (1997) Private and Public Banks: a Comparison of Customer Expectations
and Perceptions, International Journal of Bank Marketing 15(7): 279-287.
[7] Lau Mei Mei, Chaung Ronnie, Lam Aris Y. C., Chu Yuen Ting (2013) Measuring Service Quality in the Banking
Industry: A Hong Kong based study, Contemporary Management Research 9 (3): 263-282.
[8] Lehtinen U, Lehtinen, J. (1991) Two Approaches to Service Quality Dimensions, The Service Industries Journal
11 (3): 287-303.
[9] Parasuraman, A., Zeithaml, Valarie A., Berry, Leonard L. (1988) SERVQUAL: Multiple – item scale for
measuring consumer perceptions of service quality, Journal of Retailing: 5-6.
[10] Parasuraman, A., Zeithaml, Valarie A., Berry Leonard L. (1985) A Conceptual Model of Service Quality and its
Implications for Future Research, Journal of Marketing 49: 41-50.
[11] Raphael, Gwahula (2013) X-efficiency in Tanzanian Commercial Banks: An empirical investigation, Research
Journal of Finance and Accounting 4(3):12-22.
[12] Saha, A., Ravisankar, T. S. (2000) Rating of Indian Commercial Banks: A DEA Approach, European Journal of
Operational Research 124 (1): 187-203.
[13] Sathye, M. (2003) Efficiency of Banks in a Developing Economy: the Case of India, European Journal of
Operational Research 148 (3): 662-71.
[14] Singh Piyush Kumar, Gupta V. K. (2013) Measuring Technical Efficiency of Indian Banking Sector in Post
Subprime Crises Scenario: A Non-Parametric Frontier Based Approach European, Journal of Business and
Management 5 (5): 87-99.
[15] Sturm, J-E., Williams, B. (2004) Foreign bank entry, deregulation and bank efficiency: Lessons from the
Australian experience, Journal of Banking and Finance 28 (7): 1775-99.
[16] Verma, D.P.S., Vohra, R. (2000) Customer Perception of Banking Service Quality: A Study of State Bank of
India, The Journal of Institute of Public Enterprise 23 (3-4): 46-54.

186
187
188
SPONSERS OF DEA2015 CONFERENCE

189
th
Proceedings of the 13 International Conference
on Data Envelopment Analysis (DEA2015)

ISBN: 978 1 85449 497 9


DOI: 10.13140/RG.2.1.4082.9202/
190

View publication stats

You might also like