a chance-constrained data envelopment analysis approach to the measurement of vietnam's bank...

A chance-constrained data envelopment analysis approach to the measurement of Vietnam’s bank efficiency and productivity from 2006 to 2010.

2012

Nguyen Hoang Thao, MDE Class 18 i

National Economics University

Vietnam-Netherlands master's program in development economics

THESIS SUMMARY

A chance-constrained data envelopment analysis approach to the measurement of Vietnam’s bank efficiency and productivity from

2006 to 2010.

A thesis presented

by

Nguyen Hoang Thao

Supervisor

Prof. Nguyen Khac Minh

Hanoi, 2013

A chance-constrained data envelopment analysis approach to the measurement of Vietnam’s bank efficiency and productivity from 2006 to 2010.

2012

Nguyen Hoang Thao, MDE Class 18 ii

TABLE OF CONTENTS

CHAPTER 1. INTRODUCTION ............................................................................................................... 3

CHAPTER 2. LITERATURE REVIEW ................................................................................................... 4

CHAPTER 3. METHODOLOGY ............................................................................................................... 6

CHAPTER 4. EMPIRICAL STUDY ........................................................................................................ 11

CHAPTER 5. CONCLUSIONS ................................................................................................................ 17

Nguyen Hoang Thao, MDE Class 18 3

CHAPTER 1. INTRODUCTION

1.1. Problem Statement

The measurement of productivity and efficiency as well as the decomposition of the sources of change in productivity always play an indispensable part in managerial issues of a banking system. This is because being able to comprehend those important information will certainly assist people in making better decisions like how to improve output targets such as profit while consuming the same old amount of inputs like total asset, labor and interest expense especially under the fiercely competitive markets these days. As the development of international economic integration and the liberalization of trade, the increasing pressure of competition has been put on the financial market, particularly commercial banks. Commercial banks which mainly function as an intermediary between saving sectors and investment ones are currently faced with intensive competition of other financial intermediaries from abroad. Banks without competitiveness will be replaced by other successful banks. The increasing pressure of globalization, therefore, forces commercial banks to improve its productivity.

In 1990, Vietnam carried out a rehabilitation of bank structure, which divided the system into two-tier one. From then on, the banking industry has experienced a rapid increase in the quantity of commercial banks as well as the quality of management skills (see section 2 below). Besides, it is undoubted that the banking system in terms of a financial intermediary has greatly contributed to the economic development of a country, and Vietnam is not an exception. Consequently, understanding the production efficiency of banks can help to encourage more contribution of banks to the development of the economy. Additionally, when detecting an inefficiency source, a change in technology and technical efficiency, suitable policies can be formulated to improve the performance of the system.

For those reasons mentioned above, this paper is denominated “A chance-constrained data envelopment analysis approach to the measurement of Vietnam’s bank efficiency and productivity from 2006 to 2010”. It will utilize both deterministic DEA and chance-constrained DEA to measure efficiency of Vietnamese banking system. After that by evaluating which factors affect productivity of the system, this paper will develop some deliberate policies to improve the system’s performance.

1.2. Objectives and scope of thesis

Objective: The objective of this paper is to employ chance-constrained DEA to measure the performance of Vietnamese banking system. Additionally, the paper also conducts deterministic approach to make a comprehensive comparison between the two approaches. After that, by utilizing the Malmquist total factor productivity indices (TFP) integrated with the results of those models, the paper will make a decomposition of the TFP in banking industry in Vietnam into separated sources of change: technological change and technical efficiency change including pure efficiency change and scale one. Consequently, some periods where technological change and efficiency change happened will be made clear. Additionally, this paper also uses Tobit regression to detect which factors affect Vietnamese banking system’s efficiency.

Scope of study: This paper employs the dataset from annual reports of 33 domestic banks in Vietnam from 2006 to 2010. This study only focuses on domestic banks in Vietnam. There are only two types of banks in this study, including 4 state-owned commercial banks and 29 joint stock commercial banks.


The scope of study employed in this paper is due to three main reasons. Firstly, this period marked the event that Vietnam officially became member of WTO, which indicated the greater international integration of Vietnamese economy into world’s economy. Secondly, this period also witnessed the occurrence of the global financial crisis during 2007 – 2008. The two reasons implied that Vietnamese banking system had to continue its improvement in productivity, management and competitiveness to prepare for the financial liberalization and the financial crisis. Thirdly, the data source in this period is verified to be sufficient, accurate and synchronized since it is taken from official annual reports during the period.

CHAPTER 2. LITERATURE REVIEW

2.1 Methods to evaluate efficiency

For evaluating productivity between organizational units, two main different approaches: non-parametric and parametric ones can be employed. Both of them have to construct a “best-practice” frontier from observed data set in order to assess relative efficiency of each unit. This is because theoretical frontier is unknown in reality (J.C. Paradi et al, 2004). To estimate the “best-practice” frontier, the parametric has to assume restrictions on the frontier production function and the error term. Although organizational unit can be efficient without being in the frontier due to randomness of human behavior, specification error and unavailability of data, strict restrictions or assumptions on distributional form of the error term is the major drawback of this method. On the other hand, non-parametric approach or Data Envelopment Analysis (DEA) places fewer assumptions on the frontier. It, however, is deterministic and does not allow efficient unit to be out of the frontier (Kasman and Turgutlu, 2007). To address this disadvantage of deterministic DEA, in 1993, K. C. Land et al. developed a new model for DEA – chance-constrain DEA (CCDEA). The new model included uncertainty into its inputs as well as outputs. Therefore it overcomes the disadvantage of deterministic or non-stochastic DEA model. Besides, the banking industry is the complicated service-provided one which has a large number of inputs and outputs, while the parametric approach requires a specific production function. Consequently, when examining the simultaneous relation between multiple inputs and multiple outputs, parametric approach could lead to erroneous conclusions if the function is miss-specified.

2.2. Introduction to DEA model

The necessity for measuring efficiency of Decision Making Units (DMUs – organizational units which consume similar variety of inputs to produce similar variety of outputs (Igor Jemrić and Boris Vujčić, 2002)) is essential for management. It, however, remained controversial until the development of DEA. Before DEA emerged, how to rank a DMU’s performance based on outputs and inputs encountered difficulty in selecting the average weight for each input and each output. The main reason for this was that in the early studies on this field, researchers were unsuccessful in establishing a synthetic quantification from inputs and outputs to assess the performance of a DMU (Ali Mohammadi and Habibollah Ranaei, 2011). To address this problem, Charnes, Cooper, and Rhodes (1978) suggested a series of linear program known as DEA that calculate efficiency without the need of choosing average weights in advance.

DEA which is a non-parametric approach is a series of linear programming utilized to measure relative efficiency of Decision Making Units (DMU’s). This approach evaluates relative efficiency based on estimating a “best-practice” frontier. DMUs lie on the


frontier will be considered an efficient one relative to the other inefficient DMUs which is off the frontier. Fundamentally DEA is different from regression analysis approach. In regression analysis, the central tendency or the average of observations’ deportment is reflected whereas DEA approach establishes a frontier from best-performed observations and measures the divergence of the others from the frontier. Igor Jemrić and Boris Vujčić (2002) highlighted that DEA has advantage over the parametric approach in terms of less requirements imposed on the frontier production functional form. In addition, when detecting a “best-practice” frontier or a set of benchmarks of DMUs, managers can make a comparison between their DMUs and the benchmarks. Consequently, they can decide which factors (inputs) impact negatively or positively on outputs. After that, suitable adjustment to inputs such as restricting inefficient inputs and strengthening efficient ones will be took place.

After innovation of DEA, a number of studies have paid attention to this field. This paper considers the model originally developed by Charnes, Cooper, and Rhodes (1978) as Deterministic Data Envelopment Analysis model (DDEA). DDEA is the DEA model with no stochastic constrains; while Chance-constrained Data Envelopment Analysis (CCDEA) is the DEA model in which random changeability from the "best-practices" in both inputs and outputs is tolerated.

2.3. Vietnamese Researches into bank productivity measurement

In earlier studies, when evaluating productivity of Vietnamese banking system, some researchers focused on traditional method namely ratio analysis. According to Joseph C. Paradi et al (2003) ratio analysis has its own reputation for being a historic standard method to measure bank productivity. By using ratio analysis, the relation between two variables is computed (such as ROA, ROE and NIM) to provide better understanding of banks’ multifaceted activities like investment, risk management, asset quality and profitability (Joseph C. Paradi et al, 2003). Le Thi Huong (2002) – improvement to efficiency of investment activity in Vietnamese commercial banks and Le Duan (2004) – applying statistic method to analyze the Vietnamese commercial banks efficiency were two of some Vietnamese studies employing ratio analysis to assess banks’ performance. However, Ali Mohammadi and Habibollah Ranaei (2011) highlighted, based on this method, for a bank employing a number of inputs to produce numerous outputs, a synthetic indicator representing efficiency cannot be established; consequently, evaluating efficiency with this method is not adequate.

To date, Vietnamese studies into DEA, especially chance-constrained DEA, to measure efficiency of banking system has been a scarcity. N. V. Hung (2007) studied Vietnamese commercial banks’ efficiency by using DEA during the period of 2001-2003. He found that there was total factor productivity (TFP) improvement in studied period. This TFP index improvement was contributed mainly by technical efficiency and somewhat technological enhancement. In addition, in his study in 2008, both DEA and SFA are utilizing to investigate productivity of 32 commercial banks in Vietnam from 2000 to 2005. He found that the technical efficiency score results from SFA approach (0.74 under constant return to scale and 0.729 under variant return to scale respectively) are less than that in DEA model (0.791). After that, he used Tobit model to evaluate factors that affected efficiency of those commercial banks. Consequently, he made suggestion for further efficiency improvement in Vietnamese commercial banks like the need of reduction in liquidity risk, improvement of managerial skill and encouragement of high quality of existed services as well as new services (for more detail see N. V. Hung, 2008). D. T. Ngo (2010) when applying DEA into measuring efficiency of top 22 Vietnamese


commercial banks in 2008, stated that observed banks enjoyed their relatively high productivity in this year. He, nonetheless, also stated that those banks can still yield improvement in efficiency. Another study of Ngo (2012) utilized DEA window analysis to examine the productivity of 21 commercial banks in Vietnam from 1990 to 2010. He emphasized that the banking system has little to contribute to the economy, accounting for only about two-third of its capability on average. N. K. Minh et al (2012), employed DEA method to rank the performance of 145 branches of Vietnamese Agricultural bank during the period of 1997-2010. They has innovated a new method which originated from slacks-based measure of efficiency model in order to rank all inefficient DMUs. In addition to this, the new method can deal with the drawback of infeasibility. Although those studied greatly contributed to recent research on Vietnamese banks efficiency, there has been no comprehensive research into non-parametric approach in terms of both deterministic and stochastic DEA. In the studies of N. V. Hung (2007), Ngo (2010) and Ngo (2012), they only utilized deterministic DEA, which has the disadvantage of not allowing efficient DMUs to be off the frontier line. While Ngo (2010) only took 2008 banks’ data into consideration, N. V. Hung (2007) compared just three-year period from 2001 to 2003 due to lack of data. N. K. Minh et al (2012) targeted evaluating only the performance of Vietnam bank for agriculture and rural development’s branches. In addition, despite that N. V. Hung (2008) made a detailed comparison between DEA and SFA, his non-parametric approach only focused on deterministic DEA.

In general, based on the practical usage of DEA application as above and the current situation of research into this field in Vietnamese banking sector, this paper will use both deterministic DEA and chance-constrain DEA to examine the relative efficiency of domestic banks in Vietnam from 2006 to 2010. The study takes account of the fact that inputs and outputs are both stochastic. This is because customers’ behavior has the reputation for intrinsically unpredictable characteristics. For example, when evaluating performance, contemplating demand for deposit, and transaction to be deterministic variables can be misleading. Besides; the delivery of bank services is directly affected by customers’ demand, which is a production relationship. It also causes the production results to be stochastic. Consequently; in the first stage, the paper will extends the model of Chen (2005) in imposing both chance-constrained conditions on inputs and outputs to measure the productivity and efficiency of Vietnamese banks, constructing a decomposition of total factor productivity change into two separated sources of change: technological change and technical efficiency change. Besides; the paper will apply 3 methods: deterministic DEA – Model A, Chance-constrained DEA (stochastic conditions on outputs) – Model B, and chance-constrained DEA (stochastic conditions on both inputs and outputs) – Model C, to make a clearer comparison between those methods. After that, in the second stage, Tobit regression will be used to analyze the factors that determine Vietnamese banking system’s efficiency. Accordingly, economic insights for bank managers, therefore, will be suggested.

CHAPTER 3. METHODOLOGY

3.1. Production possibility set and Malmquist total factor productivity index

3.1.1. Production possibility set

Suppose that we have � = 1, 2, … , � domestic banks which consume � = 1, 2, … , � inputs ��,�

� to produce � = 1, 2, … , � outputs ��,�� at time � = 1, 2, … , �. K. Tone

(2004) introduced an integrated production possibility set with the assumptions that


�� ≥ 0, �� ≥ 0 for all � and that each DMU has at least one positive input and one positive

output:

(�, �)� = �(�, �)�� ≥ � ��

�

��

, 0 ≤ � ≤ � � ��

�

��

, � ≤ �� ≤ �, � ≥ 0 �

Where � is unit row vector in �� , � is semi-positive in �� and �, � are upper and lower bound which determine different kinds of DEA model, corresponding to CCR, BBC, IRS and DRS models respectively, (�, �) takes four values (0, ∞), (1,1), (1, ∞) and (0,1).

3.1.2. Malmquist total factor productivity index

As regards time series data, it is necessary to establish an index to measure productivity growth and decompose it into specific sources of change for a better understanding of how productivity changes. Consequently, this paper utilizes Malmquist total factor productivity (TFP) index to perform the calculation of productivity change, which is based on the application of DEA to evaluate the distance function (For more extension of decomposition technique of Malmquist TFP see C. A. Knox Lovell (2003)).

��(�� , �� ; ��, ��) = � ��

�(�� , �� )

��(��, ��)

� ��

�� (�� , �� )

�� (��, ��)

��

It is equal to:

��(�� , �� ; ��, ��) = ��

�� (�� , �� )

��(��, ��)

� × � ��

�(�� , �� )

�� (�� , �� )

� ��

�(��, ��)

�� (��, ��)

��

= �� × ��

Where EFFCH is the relative efficiency change index, or catch-up effect and TECHCH is the technological change or frontier-shift effect. Note that:

�� > 1 implies improvements in TFP of the decision making unit (DMU) from t to t+1, �� = 1 means there is no change in TFP and finally, �� < 1 indicates that TFP deteriorates.

3.2. Chance-constrained data envelopment analysis

Chance-constrained Data Envelopment Analysis (CCDEA) was developed to solve the disadvantage of non-stochastic variation of inputs and outputs in the deterministic DEA model. As mentioned in the Introduction section, the major drawback of deterministic DEA is that it does not allow efficient units to be out of the frontier. To deal with misspecification and measurement error, chance-constrained programming emerged, dating back to Charnes and Cooper (1959), C. A. Knox Lovell (1993). Chen (2002) made an comparison between chance-constrained DEA and stochastic frontier analysis to evaluate the performance of banking system in Taiwan. Chen (2005) used Chance-constrained DEA results to calculate Malmquist TFP for 46 banks in Taiwan:

� �(�� , ��

� ) = min�

Subject to:

��,�� ≤ � ��,�

� ��

�

��

� ≥ 1− ∝ � = 1, … ., �.

∝ = 5%


� ��,��

�

�

��

≤ ��,��

� = 1, … ., �

�� ≥ 0 ℎ = 1, … ., �

With assumptions

Inputs are deterministic.

Dependence between the performance of banks, ��(��,� , ��,�) for all � and ℎ ≠ � .

Variance of each stochastic output of all outputs is equal to 1, at all banks, that means: ��,� � = 1.

Observed outputs ��,� functions as an unbiased estimate of real performance, �(��,�) = ��,� .

3.3. Uncertainty in both input and output – the extension model

In reality, the process of production usually contains unsystematic factors (K. C. Land et al (1993)). There is not only the stochastic variation in outputs but also the stochastic variation of inputs. This is due to the fact that inputs are merely the results of other production process. For example, in agriculture, stochastic variation of outputs can be caused by unpredictable weather. Stochastic variations of inputs like cultivation time, amount of fertilizer and pesticide are affected by the dissimilarity of different famers’ behavior and experience. In education, it is reasonable to find out that even two schools with similar student requirement entrance and teachers’ education level have different number of students who pass university. In banking system, loan and deposit can be affected by customers’ behavior like crowd psychology which is not fully captured. Therefore, this section extend the model of Chen (2005) by taking into account both input and output as stochastic to measure the efficiency for 33 domestic banks in Vietnam. Hence, 4 new problems have to be solved.

Problem 1:

� �(�� , ��

� ) = ��

Subject to:

��,�� ≤ � ��,�

� ��

�

��

� ≥ 1− ∝ � = 1, … ., �

�� ,��

�

�

��

≤ ��,�� ≥ 1− ∝

� = 1, … ., �

�� ≥ 0 ℎ = 1, … ., �

Problem 2.

� �(�� , ��

�� ) = min �

Subject to:


��,�� ≤ � ��,�

� ��

�

��

�≥ 1− ∝ � = 1, … ., �

�� ,��

�

�

��

≤ ��,�� ≥ 1− ∝

� = 1, … ., �

�� ≥ 0 ℎ = 1, … ., �

Problem 3.

� �� (�� , ��

� ) = min �

Subject to:

��,�� ≤ � ��,�

��

�

��

� ≥ 1− ∝ � = 1, … ., �

�� ,��

�

�

��

≤ ��,�� ≥ 1− ∝

� = 1, … ., �

�� ≥ 0 ℎ = 1, … ., �

Problem 4.

� �� (�� , ��

�� ) = min�

Subject to:

��,�� ≤ � ��,�

��

�

��

� ≥ 1− ∝ � = 1, … ., �

�� ,��

�

�

��

≤ ��,�� ≥ 1− ∝

� = 1, … ., �

�� ≥ 0 ℎ = 1, … ., �

With 3 assumptions:

Dependence between the performance of banks, ��(��,� , ��,�), ��(��,�, ��,�) for

all �, � and ℎ ≠ � .

Variance of each stochastic output of all outputs is equal to 1, at all banks, that

means: ��,� � = 1, ��,�� = 1.

Observed outputs ��,� , ��,� functions as an unbiased estimate of real performance, �(��,�) = ��,�, �(��,�) = ��,�

3.5. Comparison between three Models

After estimating efficiency scores of three models, model A, model B and model C, this paper conducts two Banker’s asymptotic statistical tests (Banker, 1993) to assert that whether there is any difference between efficiency between three models.


Test1. Assumption. The three inefficiency scores 1 − �� , 1 − � � , 1 − � � which are calculated from three models A, B, C distribute exponentially with parameters �� , �� , �� respectively. The test statistics follows the F-distribution with (2��, 2�� ) degree of freedom:

Comparison between Model A and B

�0: �� = ��

�1: �� ≥ ��

�(�� ,��) = � (1 − ��) ��⁄

�

� (1 − ��) ��⁄

�

�

Comparison between Model A and C

�0: �� = ��

�1: �� ≥ ��

�(�� ,��) = � (1 − ��) ��⁄

�

� (1 − ��) ��⁄

�

�

Comparison between Model B and C

�0: �� = ��

�1: �� ≥ ��

�(��,�� ) = � (1 − ��) ��⁄

�

� (1 − ��) ��⁄

�

�

Test 2. Assumption: The three inefficiency scores 1 − �� , 1 − � � , 1 − � � follow half-normal distribution. The test statistics follows the F-distribution with (��, �� ) degree of freedom as below for each case:

�(��,��) = � (1 − ��)� ��⁄

�

� (1 − ��)� ��⁄

�

�

�(��,��) = � (1 − ��)� ��⁄

�

� (1 − ��)� ��⁄

�

�

�(��,��) = � (1 − ��)� ��⁄

�

� (1 − ��)� ��⁄

�

�

3.5. Analysis of factor affecting Vietnamese commercial banks’ efficiency

After evaluating efficiency through deterministic DEA and chance-constrained DEA, this paper conducts Tobit regression method to determine which factors influence 33 Vietnamese banks’ efficiency. The reason for the selection of Tobit model is that it allows estimation of the dependent variable (efficiency) which lies between 0 and 1.

3.5.1. Variables selection

The selection of variables is based on CAMELS – evaluation criteria which contain capital adequacy, asset quality, management quality, earnings, liquidity and sensitivity to market risk. Besides, this paper follows Fethi et al (2000), N. V. Hung (2006) paper and Chen (2011) in selecting variables as below.

�� : A dummy variable will represent the ownership status of a bank. �� = 1 implies SOCBs, while �� = 0 means the others. This variable is taken into the Tobit model to test the hypothesis that whether ownership affects banks’ efficiency or not.


�� : the proportion of business profit over business cost which reflects the ability of adjusting inputs and outputs to achieve efficiency is included into the model. A bank will gain more efficiency if it can keep this ratio as large as possible.

��: Capital adequacy ratio or the capital intensity – the ratio of equity over total asset. De Young et al (1997) highlighted that rising perilousness of loan portfolio of banks which have relatively limited capital is considered to be a moral-hazard incentive. This activity can make profit as well as efficiency for banks in short run, whereas in long run it will cause nonperforming loans to be high. Therefore, this variable is included into the model to represent the capital strength of a bank.

�� : The proportion of a bank’s total asset over total asset of all observed banks. This variable is put into the model to test whether market share affects banks’ efficiency or not.

��: The proportion of loan to total asset. Saibal Ghosh (2009) highlighted that one of the most un-safe asset of a bank is loan and an increase in this ratio means that banks will face to more risk in the balance sheet. However, if a bank can cautiously manage it own loan sufficiently, it will increase interest income, profit as well as loan market-share. As a result, this ratio reflects management ability of banks’ managers.

��: The ratio of fixed asset over total asset of a bank. It is used to analyze risk and efficiency, if this ratio is high the probability.

During the period of 2006 to 2010, there are changes in macroeconomics and banking technology, four dummies variables - �07, �08, �09 and �10 represent for 2007, 2008, 2009, 2010 respectively will be included into the model.

The weakness of this study is not including non-performing loan which can significantly impact on efficiency, into Tobit regression model. This is because bad loan is a variable which is particularly complicated to collect in Vietnam for a continuous period of years. Secondly, in Vietnam, four SOCBs dominate the market, with their total asset being at around 66% over the studied period. Therefore, the variable �� and �� are closely corrected at around 0.91. This is why this paper, in Tobit regression in empirical study, does not include �� into the model.

CHAPTER 4. EMPIRICAL STUDY

4.1. Estimation of the results

4.1.1 Analysis of efficiency estimates

Table 4 shows that on average, 33 Banks in Vietnam operated under the efficient frontier. The deterministic DEA overall efficiency score or technical efficiency score was 0.77, implying that inputs were wasted at around 23% of inputs to produce the same amount of outputs. As regards different types of ownership, although some JSBs are known as more actively than SOCBs, when taking JSBs altogether, the results represent contrarily. In the studied period, SOCBs technical efficiency score was higher than that of JSBs, with the difference being around 10.5%.

Besides, since technical efficiency is the multiplication of scale efficiency and pure efficiency, those values reflect how efficient Vietnamese banks were. From Table 4, average pure efficiency during 2006-2010 was less than scale efficiency (0.836 and 0.922 respectively); therefore, scale efficiency had greater effect on overall efficiency than pure efficiency. In other words, pure efficiency was the main source of inefficiency in the Vietnamese banking system. The results also exhibits that scales efficiency was


the main contributor to JSBs’ overall efficiency (0.923), it, however, had an adverse effect on SOCBs’ efficiency score. The situation was inversed with the pure efficiency. SOCBs demonstrated that their management of allocating resources is better than that of JSBs, with the pure efficiency score of SOCBs and JSBs being at 0.940 and 0.822 respectively.

Table 4. Average TE, PE and SE of 33 Vietnamese banks – Model A

Years Banks Criteria Mean Std. dev Min Max Obs.

20

06

– 2

01

0 SO

CB

s TE 0.863 0.140 0.583 1.000 4

PE 0.940 0.094 0.742 1.000 4

SE 0.917 0.107 0.602 1.000 4

JSB

s TE 0.757 0.193 0.370 1.000 29

PE 0.822 0.180 0.412 1.000 29

SE 0.923 0.112 0.399 1.000 29

To

tal TE 0.770 0.190 0.370 1.000 33

PE 0.836 0.176 0.412 1.000 33

SE 0.922 0.111 0.399 1.000 33

Source: authors’ calculation

As regards scale efficiency, 33 Vietnamese commercial banks are divided into three groups from 2006 to 2010. The criteria depends on the total asset of these banks – 11 banks with lowest total asset, 11 banks with medium total asset and 11 banks with largest total asset. Firstly, Table 5 shows that there was a trend of operating at decreasing return to scale for the largest asset-group. The percentage of banks which operated at decreasing return to scale was 100%, 90.9%, 72.7%, 63.6% and 72.7% in 2006, 2007, 2008, 2009 and 2010 respectively. This means that to achieve scale efficiency, 11 banks whose own largest total assets should decrease their size and concentrate on improve new sophisticated and high quality services. Secondly, it is noticeable that the middle asset group, if excluding the financial crisis year 2008, was continuing their increasing return to scale. The percentage of banks which had increasing return to scale was increasing over time, with the number being at around 36.4%, 45.5%, 72.7% and 54.5% in 2006, 2007, 2009 and 2010 respectively. Particularly, in 2010, in this group, there were 36.4% of banks operating at efficient scale. Consequently, some banks in this group could expand their sizes in order to gain scale efficiency. Thirdly, the pattern of the lowest asset group was unstable during studied period. This is due to the fact that these banks with low total asset were the most vulnerable ones to macroeconomic changes and SBV’s policies. SVB’s policy – interest rate cap, for example, which can cause these banks to be faced with liquidity shortage and puts pressure on raising capital for them.


Table 1. Asset-bank groups with operating scales – model A

2006 2007 2008 2009 2010

Lowest asset group

TRUSTB irs TRUSTB --- MDB irs MDB --- BANVIETB irs

MDB --- MDB --- KIENLONGB --- BANVIETB irs DAIAB irs

BANVIETB --- DAIAB --- TRUSTB irs DAIAB drs KIENLONGB irs

KIENLONGB drs BANVIETB --- DAIAB irs KIENLONGB drs NAMAB irs

OCEANB irs KIENLONGB drs BANVIETB irs TRUSTB --- PGB irs

NVB irs PGB irs NAMAB irs PGB irs SGB ---

PGB --- NAMAB drs PGB irs NAMAB drs MDB irs

SHB irs VIETAB irs HDB --- SGB drs OCB irs

DAIAB irs NVB drs OCB --- OCB drs TRUSTB irs

ABB drs SGB drs VIETAB drs VIETAB --- NVB irs

NAMAB irs OCB drs NVB drs NVB drs VIETAB ---

Middle asset group

HDB irs SHB drs SGB drs HDB irs HDB irs

VIETAB irs OCEANB irs ABB drs ABB irs HBB ---

SGB --- HDB --- OCEANB drs SHB irs ABB irs

OCB drs SOUTHERNB irs SHB --- VPB irs SHB irs

MSB irs ABB irs VPB --- HBB irs MHB drs

SOUTHERNB drs MSB irs SOUTHERNB drs SEAB irs OCEANB irs

VPB drs VPB --- SEAB drs OCEANB irs SEAB irs

SEAB irs HBB drs HBB irs SOUTHERNB irs DAB ---

SCB --- SCB --- MSB irs MHB drs VPB irs

HBB drs SEAB irs DAB drs DAB --- SCB ---

DAB drs MHB drs VIB drs SCB --- SOUTHERNB ---

Highest Asset group

MB drs DAB drs MHB drs VIB drs VIB drs

VIB drs MB drs SCB --- MSB irs MB drs

TCB drs EXIMB drs MB drs EXIMB drs MSB ---

EXIMB drs VIB drs EXIMB drs MB --- EXIMB drs

MHB drs TCB drs TCB drs TCB --- TCB drs

SACOMB drs SACOMB drs SACOMB drs SACOMB drs SACOMB drs

ACB drs ACB drs ACB drs ACB drs ACB drs

CTG drs CTG drs CTG drs CTG drs VCB drs

BIDV drs VCB drs VCB drs VCB drs BIDV ---

VCB drs BIDV drs BIDV --- BIDV drs CTG drs

VBARD drs VBARD --- VBARD --- VBARD --- VBARD ---

Source: author’s calculation

4.3.2 Analysis of productivity enhancement

The result in Table 6 exhibits that there was a small deterioration of 0.5% in Malmquist TFP index, with the number being around at 0.995. Although there had a major contribution of scale efficiency growth (102%) and pure efficiency growth (103.9%) to the TFP, technological regress (4.9%) deteriorated it during the studied period. This means that bank’s technology was becoming the inefficient factor to banks’ productivity


enhancement. The reason for this was that Vietnamese bank’s technology was underdeveloped compared to the development of sophisticated quality services like Online banking, Automatic transfer, Mobile banking, SMS banking, Call center and Master card pay-pass which require high implementation of new technology. Other reasons were that technological advancement was not fully employed during the studied period and that Vietnamese banks still used technology requiring a lot of labors.

Regarding change in TFP during 2006-2010, it is noticeable that the culprit of the slightly deterioration in TFP was the financial crisis which happened in 2008. The number of TFP index in 2008 was only 0.842 and this year was considered to be the most difficult one for Vietnamese banking system for the last 20 years. Banks had to be faced with a significant number of changes in government’s policies. For example, there were three decreases and five increases in based interest rate and the situation was similar for refinancing interest rate and re-discount interest rate. Additionally, the government’s policies on interest rate cap had caused liquidity shortage in a lot of banks and led to an interest rate race during this time.

Figure 1. TFP index of 33 banks, SOCBs and JSBs from 2006 to 2010 – Model A


Furthermore, when considering the progress of different types of Vietnamese banks: SOCBs and JSBs from Table 6, the estimated result shows that both JSBs and SOCBs did not make any improvement in TFP index during 2006-2010. SOCBs’ TFP index was around at 99% while JSBs’ TFP index was 99.6%. The TFP index of JSBs on average was slightly higher than that of SOCBs about 0.6%. This implies that JSBs’ employment of input resources was becoming better than SOCBs. Additionally, JSBs continued to improve their management of allocating resources (PECH=1.058), whereas SOCBs did not (PECH=0.974).

1.003

0.842

1.173

0.961

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

2006-2007 2007-2008 2008-2009 2009-2010

Total SOCBs JSBs


Table 2. Average EFFCH, TECHCH, PECH, SECH and TFP during 2006-2010

Banks Efficiency mean Std. dev min Max

SOC

Bs

EFFCH 1.042 0.077 0.940 1.103

TECHCH 0.950 0.043 0.911 1.001

PECH 0.974 0.033 0.932 1.000

SECH 1.072 0.072 1.004 1.162

TFP 0.990 0.108 0.858 1.122 JS

Bs

EFFCH 1.058 0.095 0.850 1.325

TECHCH 0.951 0.045 0.826 1.049

PECH 1.048 0.082 0.861 1.262

SECH 1.013 0.064 0.850 1.193

TFP 0.996 0.084 0.809 1.153

To

tal

EFFCH 1.056 0.092 0.850 1.325

TECHCH 0.951 0.044 0.826 1.049

PECH 1.039 0.081 0.861 1.262

SECH 1.020 0.067 0.850 1.193

TFP 0.995 0.086 0.809 1.153


4.3.3 Model A, Model B efficiency and productivity enhancement

Table 7 shows the average efficiency from three models – Model A, Model B and Model C. It is noticeable that there is merely a little difference between efficiency scores from three models. The efficiency scores are decreasing from Model A to Model B and then Model C. This is because the production frontier in three models are different, with the Model A frontier line is softer than that of Model B and that of Model B is softer than that of Model C. In other words, Model A and Model B allows at some probability level that an efficient DMU can lie outside the frontier line; however, Model C does not (Chen, 2011).

Table 3. Model A, Model B, Model C average efficiency

Year Banks Model A Model B Model C

20

06

-20

10

SOCBs

TE 0.86258 0.86047 0.85996

PE 0.93979 0.93977 0.93973

SE 0.91687 0.91473 0.91424

JSBs

TE 0.75709 0.75412 0.75358

PE 0.82184 0.81957 0.81885

SE 0.92260 0.92159 0.92176

Total

TE 0.76988 0.76701 0.76647

PE 0.83613 0.83414 0.83350

SE 0.92190 0.92076 0.92085


As for Mamquist productivity index, the same situation happens to the results of three models, with the malmquist TFP indices are 0.99505, 0.99653 and 0.99680.


Table 4. Model A, Model B, Model C average EFFCH, TECHCH, PECH, SECH, TFP

Year Banks Model A Model B Model C

20

06

-20

10

SOCBs

EFFCH 1.04218 1.04416 1.04463

TECHCH 0.95035 0.94911 0.94842

PECH 0.97410 0.97410 0.97413

SECH 1.07218 1.07427 1.07472

TFP 0.99014 0.99049 0.99017

JSBs

EFFCH 1.05777 1.06130 1.06176

TECHCH 0.95138 0.95050 0.95068

PECH 1.04793 1.05086 1.05151

SECH 1.01280 1.01326 1.01305

TFP 0.99572 0.99736 0.99772

Total

EFFCH 1.05588 1.05923 1.05969

TECHCH 0.95126 0.95033 0.95041

PECH 1.03898 1.04156 1.04213

SECH 1.02000 1.02066 1.02052

TFP 0.99505 0.99653 0.99680


Comparison between three Models

From section above, it is necessary to find out whether the efficiency scores from three models are different or not. Conducting Banker (1993) asymptotic test for 2006’s efficiency scores, comparison between three models A, B and C is showed in the table below.

Table 5. Hypothesis test for comparison of efficiencies between three Models

Hypothesis Test procedure

Model A vs. Model B

Model A vs. Model C

Model B vs. Model C

Critical value � = �%

H0: σ� = σ�

H1: σ� ≥ σ�

Exponential assumption

1.003 1.026 1.023 1.50

H0: σ� = σ�

H1: σ� ≥ σ�

Half-normal assumption

1.004 1.038 1.034 1.79

Note.

σ� is parameter of exponential distribution of inefficiency in first model

σ� is parameter of exponential distribution of inefficiency in second model


The null hypothesis is accepted for both exponential and half-normal distribution of efficiency; therefore, there are no difference between efficiency results in three Model A, B and C. This means that for the data of Vietnamese banks from 2006 to 2010, there are no difference in efficiency scores between deterministic DEA and chance-constrained DEA. Although the chance-constrained DEA is integrated with a flexible production


frontier and the stochastic behavior of banks as well as customers, similar structures are implemented in both chance-constrained DEA and deterministic DEA. Consequently, the results can concur with each other (Chen, 2011).

4.3.4 Factors affecting Vietnamese banks’ efficiency during 2006-2010

After estimating efficiency through three models, the CCR efficiency scores from them will be employed as a dependent variable in three Tobit models to determine which factors have effect on it. Since the efficiency scores rage from zero to one, therefore Tobit model is utilized. Below, three estimated models which take efficiency scores from model A, B and C respectively will be presented.

Firstly, the ratio of business revenue over business cost is positive and statistically significant at 1% level. As this ratio represents the ability of adjusting inputs and outputs to achieve efficiency, a Vietnamese bank can gain efficiency through efficiently managing their functional form between them. Secondly, the capital adequacy which is measure by equity over total asset has significant positive effect on efficiency. The key cause is that when acquiring sufficient capital, banks may overcome the moral-hazard incentives which can cause damage to banks’ efficiency (Berger and De Young, 2010). In their study they highlighted that thinly capitalized banks tend to make profit through investing on high risk loan portfolio, bringing them efficiency and profit in short-run. This, however, causes higher non-performing loan in long-run which has adverse effect on these banks’ efficiency. This result implies that, in short run, a bank can be more efficient if it can increase its capital adequacy ratio. Thirdly, market-share coefficient is statistical significant at 10% level. As a result, a bank which dominates the market will have higher efficiency. This is due to the fact that a bank with large market share will enjoy its reducing operation cost and making larger profit. Fourthly, loan to total asset ratio is found having significant effect on efficiency, with the level of significance being at 1%. This suggests that major profit of Vietnamese banks is still interest revenue. Therefore, increasing loan to a sufficient level will increase efficiency. Fifthly, non-interest revenue over total asset is statistically significant at 1%. This ratio represents for the diversification of banks’ operations. Basically, banks in Vietnam mainly earn profit by making loan. However, in a highly competitive financial market these days, with the foundation of a number of new banks and the emergence of foreign banks, deposit and credit market have to be divided smaller among these banks. Banks which want to attract customers have to increase quality of services as well as to expand other sophisticated products needed to satisfy their customers. Consequently, this ratio exhibits that more efficiency can be achieved through expanding other high quality services along with traditional activities such as making loan.

CHAPTER 5. CONCLUSIONS

Two approaches to data envelopment analysis – deterministic and chance-constrained DEA are employed in this paper to asset 33 Vietnamese domestic banks from 2006 to 2010. While deterministic DEA considers the multiple inputs and multiple outputs as pre-determined, chance-constrained DEA takes into account the stochastic variation of them. Therefore, it expands the deterministic DEA problem by adding the procedure that only with a probability of � = 5% or less a bank will be superior to (or do better than) the production frontier. This leads to the problem which can convert to a set of non-linear program.

Employing both methods, this paper finds that observed banks, on average, operated under efficient frontier, with the results from chance-constrained DEA in both outputs


and inputs (model A), chance-constrained DEA in outputs (model B) and deterministic DEA (model C) being at 0.777, 0.767 and 0.766 respectively. In other words, observed banks wasted 33% (Model A) of input resources to produce the same amount of outputs. Additionally, the results show that there was a decreasing return to scale to which top largest capitalized observed banks had to face each year; the situation was reversed for the middle capitalized ones which face to increasing return to scale; the pattern of the lowest asset group was unstable. Moreover, the Malmquist TFP from model A is 0.995, implying that 33 Vietnamese banks experienced a degradation in productivity at 0.5%.

As regards ownership, the SOCBs is found to be more efficient than JSBs with the efficiency scores being at 0.877 and 0.863 respectively. JSBs, however, are utilizing their resource more efficiently, with its malmquist TFP being greater than that of SOCBs at around 0.6%, whereas JSBs are enjoying their enhancement in pure efficiency, SOCBs, in contrast, are enjoying their scale efficiency enhancement.

Using Banker’s asymptotic tests are conducted to test whether these models are distinctive or not. The test results show that with Vietnamese banks’ data from 2006 to 2010, there is no difference between them. Nevertheless, the variability in banks’ efficiency has been included into chance-constrained DEA model which overcomes the disadvantage of deterministic DEA.

In addition, Tobit model is conducted to explain for the underlying causes of efficiency of Vietnamese banks, giving significant suggestions to banks’ managers. The paper finds that profitability of banks is an indicator of higher overall efficiency. Strongly capitalized banks have greater efficiency than thinly capitalized banks. A bank’s loan to total asset ratio and market share positively affects its efficiency. Also, the fact that a bank diversifies its services is another factor causing efficiency.

REFERENCE

A. Charnes and W. W. Cooper, 1959,“Chance-constrained programming,” Management Science, Vol. 6, No. 1, pp. 73-79.

A. Charnes, W. W. Cooper and E. Rhodes, 1978, "Measuring the efficiency of decision making units," European journal of operational research, Vol. 2, No. 4, pp. 429-444.

A. S. Camanho and R. G. Dyson, 1999, "Efficiency, Size, Benchmarks and Targets for Bank Branches: An Application of Data Envelopment Analysis," The Journal of the Operational Research, Society Vol. 50, No. 9, pp. 903-915.

Anne Ho and R. Ashle Baxter, 2011, “Banking reform in Vietnam,” Asia focus journal, 2011, issue June.

Anthony N. Rezitis, 2006, “Productivity growth in the Greek banking industry: A non-parametric approach,” Journal of Applied Economics.

C. A. Knox Lovell, 2003, "The Decomposition of Malmquist Productivity Indexes," Journal of Productivity Analysis, Vol. 20, pp. 437–458.

C.A. Knox Lovell, Jesús T. Pastorb, 1997, "Target setting: An application to a bank branch network," European Journal of Operational Research, Vol. 98, Issue 2, pp. 290–299.

Charnes, A., W.W. Cooper, and E. Rhodes, 1978, Measuring the efficiency of decision making units, European Journal of Operational Research 2, 429-444.


Cooper, William Wager, Lawrence M. Seiford, and Kaoru Tone, 2007, “Data envelopment analysis: a comprehensive text with models, applications, references and DEA-solver software”. Springer Science+ Business Media.

David A. Grigorian , Vlad Manole , Oleh Havrylyshyn , Roberto Rocha, 2002, "Determinants of Commercial Bank Performance in Transition: An Application of Data Envelopment Analysis," IMF Working Paper, No. 02/146.

Dr. Ashish Kumar, Dr. Vikas Batra, 2012,“Efficiency Analysis of Indian Banks (2006-2011): An Application of Malmquist Productivity Index,” National Conference on Emerging Challenges for Sustainable Business.

Gary D. Ferrier, C.A.Knox Lovell, 1990, "Measuring cost efficiency in banking econometric and linear programming evidence," Journal of Econometrics, Vol. 46, Issues 1–2, pp. 229–245.

Gwahula Raphael, 2013, “Bank-specific, industry-specific and macroeconomic determinants of bank efficiency in Tanzania: A two stage analysis,” European Journal of Business and Management, Vol. 5, No. 2

H. David Sherman and Franklin Gold, 1985, "Bank branch operating efficiency: Evaluation with Data Envelopment Analysis," Journal of Banking & Finance, vol. 9, issue 2, pp. 297-315.

Igor Jemric, Boris Vujcic, 2002, "Efficiency of Banks in Croatia: A Dea Approach," Comparative Economic Studies, Vol. 44, Issue 2-3, pp. 169-193, 2002

J.L. Hu, C.P. Chen, and Y.Y. Su, 2006, “Ownership reform and efficiency of nationwide banks in China’, International conference, WTO, China and the Asian Economies, IV: EconomicIntegration and Economic Development, 2006.

João Rebelo, Victor Mendes, 2000, “Malmquist indices of productivity change in Portuguese banking: The deregulation period,” International Advances in Economic Research, Vol. 6. Issue. 3.

Joseph C. Paradi , Sandra Vela and Zijiang Yang, 2004, "Assessing bank and bank branch performance Modeling Considerations and Approaches," International Series in Operations Research & Management Science. Vol. 71, Handbook on Data Envelopment Analysis, pp. 349-394.

K. C. Land, C. A. Knox Lovell and S. Thore, 1993, “Chance-constrained Data Envelopment Analysis,” Managerial and Decision Economics, Vol. 14, No. 6, pp. 541- 554. doi:10.1002/mde.4090140607

Kaoru Tone, 2004, "Malmquist Productivity Index efficiency Change over Time," International Series in Operations Research & Management Science, 1, Volume 71, Handbook on Data Envelopment Analysis, pp. 203-227. doi: 10.1007/1-4020-7798-X_8

Kasman A, Turgutlu E, 2008. “A Comparison of Chance-constrained DEA and Stochastic Frontier Analysis: An Application to the Turkish Life Insurance Industry.” Empirical Econ. Lett., 7: 579-587.

Luis R. Murillo-Zamorano, 2004, “ economic efficiency and frontier techniques, “ journal of economic surveys vol. 18, no. 1.

Mette Asmild, Joseph C. Paradi, Vanita Aggarwall and Claire Schaffnit, 2004, "Combining DEA Window Analysis with the Malmquist Index Approach in a Study of the Canadian Banking Industry," Journal of Productivity Analysis, Vol. 21, No. 1, pp. 67-89, doi: 10.1023/B:PROD.0000012453.91326.ec.


Milind Sathye, 2003, “Efficiency of banks in a developing economy: The case of India,” European Journal of Operational Research, Vol. 148, Issue 3.

Mlima, A. P. & Hjalmarsson, 2002. "Measurement of inputs and outputs in the banking industry", Tanzanet Journal 12, Vol. 3.

N. K. Minh, P. V. Khanh, P. A. Tuan, 2012, "A New Approach for Ranking Efficient Units in Data Envelopment Analysis and Application to a Sample of Vietnamese Agricultural Bank Branches," American Journal of Operations Research, Vol. 2, Issue 1, pp. 126-136. doi:10.4236/ajor.2012.21015.

N.V. Hung, 2006, “Analysis of factors which affect Vietnamese commercial banks’ efficiency”, National Economic University of Vietnam, PhD. paper.

Q. T. Linh, 2011, “Vietnam Banking Sector Report”, Vietcombank

Rajiv D. Banker, 1996, "Hypothesis tests using data envelopment analysis," Journal of Productivity Analysis, July 1996, Volume 7, Issue 2-3, pp 139-159.

Reza Siregar, 1999, "Rising to the Challenge in Asia: A Study of Financial Markets," Asian Development Bank, Vol.12.

Sathye, Milind, 2003, “Efficiency of banks in a developing economy: The case of India,” European Journal of Operational Research. Vol. 148.

Sigbjørn Atle Berg, Finn R. Førsund, Lennart Hjalmarsson, Matti Suominen, 1993 “Banking efficiency in the Nordic countries,” Journal of Banking & Finance, Volume 17, Issues 2–3, April 1993.

Tser-yieth Chen, 2002, "A Comparison of Chance-Constrained DEA and Stochastic Frontier Analysis: Bank Efficiency in Taiwan," The Journal of the Operational Research Society, Vol. 53, No. 5, pp. 492-500.

Tser-yieth Chen, 2005, “A measurement of Taiwan's bank efficiency and productivity change during the Asian financial crisis,” International Journal of Services Technology and Management, Vol. 6, No. 6. doi: 10.1504/IJSTM.2005.007510.

V. V. Podinovski and A. D. Athanassopoulos, 1998, "Assessing the Relative Efficiency of Decision Making Units Using DEA Models with Weight Restrictions," The Journal of the Operational Research Society, Vol. 49, No. 5, pp. 500-508.

W. W. Cooper, L. M. Seiford, & J. Zhu, 2011, “Handbook on data envelopment analysis,” Springer.

a chance-constrained data envelopment analysis approach to the measurement of vietnam's bank...

Documents