Bank concentration and efficiency of commercial banks in Vietnam

Bank concentration and efficiency of commercial banks in Vietnam LE NGUYEN QUYNH HUONG University of Economics HCMC – quynh_huong@ueh.edu.vn NGUYEN HUU BINH University of Economics HC

Bank concentration and efficiency of

commercial banks in Vietnam

LE NGUYEN QUYNH HUONG University of Economics HCMC – quynh_huong@ueh.edu.vn

NGUYEN HUU BINH University of Economics HCMC – huubinh_ais@ueh.edu.vn

Abstract

The relationship between bank concentration and bank efficiency remains a controversial topic This paper investigates to what degree bank concentration dampens or enhances the response of bank efficiency in Vietnam and vice versa This study applies Concentration Ratio (CR) and Herfindahl - Hirschman Index (HHI) as proxies of bank concentration, while efficiency scores are calculated by stochastic frontier approach (SFA) and data envelopment analysis (DEA)

To test the Structure Conduct Performance (SCP) and Efficient Structure (ES) paradigm, the authors use Granger causality approach However, regarding the causality running from bank efficiency and bank concentration, the results are complex: we find the causality running from concentration to efficiency is weak, whereas efficiency Granger-caused negatively competition Over a relatively long time period, from 2007 to 2014, the more efficient commercial banks operated in the less concentrated market

Keywords: Vietnam; bank concentration; efficiency; structure conduct performance

1 Introduction

In the process of integration into the world economy, Vietnam's financial market is under great pressure Strong competition among commercial banks would be a great opportunity for the banking sector if Vietnam domestic banks are more adaptable and operate more efficiently, especially under the Restructuring Plan Thus, operational efficiency becomes a vital part for the survival of a bank in the increasingly competitive environment The relationship between bank concentration and bank efficiency, especially in Vietnam, is open to doubt and highly ambiguous There are numerous studies testing for this relationship Some concentrate on the Structure Conduct

Performance (SCP) paradigm (Bikker & Haaf, 2002a; Deltuvaitė, Vaškelaitis, & Pranckevičiūtė, 2015; T P T Nguyen & Nghiem, 2016), while others support the reverse relationship namely efficient structure hypothesis (ES), which considers that bank efficiency positively influence on market concentration (Punt & Van Rooij, 2003; Weill, 2004) Recently, this topic has received tremendous attention in Vietnam, and only three studies found hitherto (Chinh & Tiến, 2016; Huyền, 2016; Thơm & Thủy, 2016) Unfortunately, no study analyses simultaneously the relationship between bank concentration and efficiency by using Granger causality Thus, this is a noticeable research gap needed further investigation

The purpose of this paper is to examine the relationship between bank concentration and efficiency by using the application of Granger causality method It also tests Structure Conduct Performance and Efficient Structure hypothesis The rest of the paper is structured as follow Section 2 presents a brief overview of Vietnamese banking system Section 3 contains the previous related literature Section 4 describes the methodology and the data Section 5 contains the empirical results while section 6 gives conclusions and policy recommendations

2 Overview of Vietnamese banking system

According to the State Bank of Vietnam (SBV), the history of banking activities is divided into four stages, including 2 critical periods: 1986 - 2001 (reforming from the mono-banking system into the two-tier banking system) and after 2011 (restructuring the Vietnamese banking system) The process of restructuring the banking system and clean-

up bad debts has implemented drastically under Vietnam’s banking restructuring Scheme in 2011-2015 (Decision 254, 1/3/2012) and Non-performing debt settlement Scheme of credit institutions (Decision 843, 31/5/2013) These Schemes focus on some

central goals, including controlling the weak credit institutions, bad debts, development

of the banking system and to contribute significantly to macroeconomic stability, removing difficulties for production and business, promoting economic growth To sum

up, the process of restructuring of Vietnam's banking system consists:

• The privatisation of state-owned commercial banks

• Increasing the financial scale and capacity: raising capital, acquisitions and mergers, expanding mobilisation

• Improving asset quality, credit quality and reduce bad debt

Vietnamese commercial banking system can be classified into 4 main groups: (1) owned commercial bank, (2) Joint stock commercial bank, (3) Foreign commercial bank, and (4) Joint venture commercial bank Figure 1 shows the number of commercial banks

as well as Non-performing loans (NPLs) over the period of 8 years It is noticed that owned banks and foreign banks still remained in number, while Joint stock commercial banks decreased their number from 40 in 2008 to 30 in 2014 According to Vietnam’s

State-banking restructuring Scheme mentioned above, some weak banks (Joint-stock

commercial banks) took actively and hospitably M&A with other leading banks resulted

in the drop in the number of commercial banks from 52 in 2007 to 44 in 2014 For example, Vietnam Tin Nghia Bank together with SCB and First Bank of VN merged into SCB, Western Bank and PVFC consolidated in PVcombank, Habubank is acquired by SHB, etc Because of high NPLs in weak banks, merging with leading banks could be an efficient solution encouraged by SBV in order to strengthen and improve the competition of Vietnamese domestic banks NPLs figures shown in Figure 1 followed an upward trend, from 2% (2007) to 4.55% (2013) After reaching a peak at 4.55% in 2013, NPLs decreased significantly to 3.25% It is doubtful that some banks could “cook the book”, deliberately failed to comply with regulations on debt classification and recorded bad debts in financial

statements lower than actual However, some argue that 2014 is the first year Vietnam Asset Management Company (VAMC) bought bad debt from troubled banks and moved a

considerable amount of NPLs out of banks’ financial statements (approximately 123 thousand billion VND, according to SBV – 23/12/2014)

Figure 1 Number of Vietnamese banks and NPLs from 2007 to 2014

Source: Annual Statements of State Bank of Vietnam (SBV)

Two hypothesis in the structural approach including the traditional Conduct-Performance (SCP) hypothesis, which is originated from the traditional industrial organisation literature, and the Efficient Structure (ES) hypothesis In which, SCP hypothesis argues the direct positive link between market concentration and profitability based on the presumption that banks in a high concentrated market can collude to earn higher profits resulting in efficiency (Bain, 1951, 1956) ES hypothesis, meanwhile, assumes a reverse causality that efficient banks are more profitable and gain market shares, resulting in a concentrated market In other words, the higher efficiency

Structure-of market leads to the higher market concentration (Demsetz, 1973) The “quiet life” (QL)

2007 2008 2009 2010 2011 2012 2013 2014 State owned commercial

0 5 10 15 20 25 30 35 40 45

hypothesis developed by Hicks (1935), by contrast, supports a negative relationship between market concentration and performance Following this, firms with market concentration tend to make few efforts to maximise efficiency Because managers in these firms may have no motivation and enjoy the monopoly profit of a “quiet life”, and this may result in inefficient operation

Based on these hypotheses, there were a numerous number of studies performed in the banking sector in many parts of the world Some of the studies are summarised in Table 1

Thornton (1994) 1986-1988 Spanish

SCP Supported

ES Rejected Molyneux and Forbes (1995) 1986-1989 European banking

industry

SCP Supported

ES Rejected Goldberg and Rai (1996) 1988-1991 11 European

countries

SCP Rejected

ES Supported Coccorese and Pellecchia (2010) 1992–2007 Italy QL Supported Al-Muharrami and Matthews

SCP Supported

QL Rejected Koetter and Vins (2008) 1996-2006 Germany QL Rejected Fang, Hasan, and Marton (2011) 1998–2008 South-Eastern

Casu and Girardone (2009) 2000-2005 5 EU countries QL Rejected

ES Rejected Ferreira (2013) 1996-2008 27 EU countries SCP Supported

ES Rejected Nguyen, Stewart (2013) 1999-2009 Vietnam SCP Rejected

ES Rejected

ES Supported

As can be seen from the Table 1, there are differences in the results of empirical studies concerning the relationship between bank concentration and efficiency proposed by three hypotheses mentioned above This shows that the relationship between bank concentration and efficiency depends on the characteristics of each country and region This paper uses Granger causality to test simultaneously both SCP and ES in the case of Vietnam

4 Methodology

To test the Granger causality relationship between bank concentration and bank efficiency, this section explains the methodological framework and the data: how to measure bank concentration and bank efficiency, how to choose inputs and outputs from financial statements of commercial banks, and the Granger causality procedure

The market concentration is scaled from low to high, and in this regard, the market

is catalogued into four cases: (1) perfect competition, (2) monopolistic competition, (3) oligopoly and (4) monopoly The market which is considered as perfect competition is addressed as low concentrated, and on the opposite side of the scale - the concentration of market which tends to monopoly is evaluated as high (Boďa, 2014)

Market structures

Perfect

competition Monopolistic competition Oligopoly Monopoly

There are a number of market concentration indicators based on the calculation of market shares Among other things, two standard and popular ways to measure concentration level are Concentration Ratio (CR) and Helfindhal-Hirschman Index (HHI) The other well-known indicators of concentration ratio are the Coefficient of variation, the Hall-Tideman Index (HTI), and the Comprehensive industrial concentration index Table 2 gives a brief overview of these concentration measures except for CR and HHI However, because of general consensus, data validation and straightforwardness, this paper use CRk and HHI to measure the concentration in Vietnamese banking market Technically, both CRk and HHI do not require to rank and sort in descending order all banks based on their market shares

The k bank concentration ratio

The k Bank Concentration ratio is the simplest and required limited data measure of concentration Nevertheless, this measure only emphasises on kth leading banks while neglecting the small banks Moreover, there is no rule for determination of the value of k,

so k can be chosen on an ad hoc basis (often, k = 3, 4, 5, 8)

The Concentration ratio of k banks is calculated as:

k represents the number of banks on the market

The value of this indicator varies from 0 (perfect competition) to 1 The market is considered as oligopoly, if k > 1 or monopoly, if k = 1

This study adopts the Concentration Ratio - CR4, which means the market share of the four largest firms In the case of Vietnam, we conventionally define four largest banks

or “big-four” Vietnamese banks as BIDV, Vietcombank, Vietinbank, and Agribank Here,

we use the percentage share of the total assets held by the four largest banks for CR4 Helfindhal-Hirschman Index (HHI)

HHI is calculated by the sum of the squares of market shares of all banks on the market This index is defined as:

HHI = S&,

-&'(

where: S&, is the square of market share of ithbank

n represents the number of banks on the market

HHI spreads widely as U.S Department of Justice has used it since the 1980s to measure potential mergers issues or antitrust concerns However, there is no convention

to classify a market into high, moderate and low concentrated catalogue This problem can be addressed by using the consensus from U.S Department of Justice (DOJ) & Federal Commission Trade (FCT) and The European Commission

According to U.S Department of Justice (DOJ) & Federal Commission Trade (FCT), Horizontal Merger Guidelines § 5.2 (2010), and The European Commission, the interpretation of HHI is as follows:

The European Commission DOJ & FCT

HHI sometimes is called full-information index as it captures features of the whole banking system For this reason, this paper chooses HHI to measure the concentration ratio of Vietnamese banking market

Table 2 summarises the key features of other concentration measures which are mentioned at the beginning of this section (Bikker & Haaf, 2002b; Boďa, 2014):

Not including the number of banks

Simple to understand (this is a standard relative measure of variation of nominal variables) No consensus at which value may be considered as high or low

4.2 Bank efficiency

Defining output, input variables in banking sector

The determination of the input - output variables in banking field is a controversy issue Berger and Humphrey (1992) determined inputs and outputs in many different perspectives (National Bureau of Economic Research - NBER study "Output Measurement

in the Service Sectors”, Chapter 7 - Measurement and efficiency issues in commercial banking) Briefly, these viewpoints include three main approaches:

Intermediation Approach: banks are financial institutions, intermediation between borrowers and lenders Therefore, outputs are probably defined as loans and other assets, while inputs will be deposits and other liabilities This method was developed by Sealey and Lindley (1977)

User cost Approach: This method determines the inputs or outputs based on the ability

to contribute to revenue for the bank If the financial returns on an asset exceed the opportunity cost of funds or if the financial costs of liability are less than the opportunity cost, then the instrument is considered to be a financial output (Berger & Humphrey, 1992)

Value-added Approach: This approach considers all asset and liability categories to have output characteristic rather than distinguish inputs from outputs in a mutually exclusive way The categories having substantial value added, as judged using an external source of operating cost allocations, are employed as the important outputs Others are treated as representing mainly either unimportant outputs, intermediate products, or inputs, depending on the specifics of the category (Berger & Humphrey, 1992)

Measuring bank efficiency

Charnes, Cooper, and Rhodes (1978) is the first team using Data Envelopment Analysis model (DEA) to measure the efficiency of decision-making units (DMUs) DEA model is a non-parametric estimation which is widely used in myriad fields since 1957 The global private banking sector, particularly, has been applied DEA model in research (Nathan & Neave, 1992) (Miller & Noulas, 1996), (Iršová & Havránek, 2010), (Luo, Yao, Chen, & Wang, 2011)

Data envelopment analysis (DEA) is a linear programming formulation for measuring the relative performance of organisational units where the presence of multiple inputs and outputs makes comparisons difficult Efficiency scores are then calculated from the frontiers generated by a sequence of linear programs (convex combinations of DMUs) Assuming there are n banks, each bank can create s output by using m different inputs The relative efficiency score of a DMU p could be assessed by solving a fractional program, which is defined by extremal optimization (maximization) of the ratio of weighted sum

of outputs to weighted multiple inputs (aka virtual output to virtual input ratio), then subject to the constraints of non-decreasing weights and efficiency measure (the earlier mentioned ratio) less than or equal to one To sum up, this involves finding the optimal weights so that efficiency measure is maximised (banks choose their input and output weights that maximise their efficiency scores)

s t ?#'(v#y#&

uAxA&

B A'(

≤ 1 ∀i, v#, uA ≥ 0 ∀k, j where: k = 1, …, s; j = 1, …, m; i = 1, …, n

yki: output k produced by bank i,

xji: input j used by bank i,

vk and uj are weights given to output k and input j

However, this research will not go too deep into the complex theoretical part of the DEA estimations but focus primarily on the empirical side of the methods that concern measuring efficiency

Another common method of measuring efficiency, developed by Aigner, Lovell, and Schmidt (1977) and Meeusen and van Den Broeck (1977), is the Stochastic Frontier Approach (SFA) SFA method divides residuals into 2 groups: inefficiencies and noise, and using some assumptions about the inefficiencies’ distribution One part of residuals is called normal statistical noise (Vit) and the rest is noise inefficiency (Uit) Vit is assumed to

be independent of the explanatory variables and have the same distribution iid ~ N (0,

sL,) and represents the statistical noise, measurement error, and other random events (e.g., economic conditions, earthquakes, weather, strikes, luck) beyond the company's control Inefficiency Uit (aka inefficiency error term - non-negative) represents inefficiency factors and assumptions is truncated at 0 and idd ~N (µ, sL,) At the same time, Uit is assumed to be independent of Vit The canonical formulation that serves as the foundation for other variations is the model:

Y = b’X + v – u,

where Y is the observed outcome, b’X + v is the optimal, or frontier goal (i.e maximal production output or minimum cost) pursued by the individual The amount by which the observed individual fails to reach the optimum (the frontier) is u Alternatively, there

is a commonly used – the Translog function:

Yit = exp [Xit b + (Vit - Uit)] i = 1, …, K, t = 1, …, T

where: Yit: output, the output of the ith enterprise, at time t

Xit: Vector KX1 input of ith now, at time t

b: Vector Kx1 of unknown factors

Vit: “noise” error term - symmetric (i.e normal distribution)

Uit: “inefficiency error term” - non-negative (i.e half-normal distribution)

SFA has become the method commonly used because of many prominent advantages (Coelli & Perelman, 2000; Cuesta & Orea, 2002; Färe, Grosskopf, Lovell, & Yaisawarng, 1993; Grosskopf, Margaritis, & Valdmanis, 1995) Whereas SFA is more appropriate for emerging markets where measurement errors and uncertainties of the economic environmentare more likely to prevail (Zhang et al., 2013), we use both DEA and SFA for Vietnam case

Figure 2 DEA and SFA Frontier

Here, we adopt DEA input-oriented and follow the intermediation approach The intermediation approach, originally proposed by Sealey and Lindley (1977), is appropriate when banks operate as independent entities (Bos & Kool, 2006) and take into account interest expenses It seems appropriate to evaluate commercial banks in Vietnam because interest expenses present at least more than half of total costs in general (Berger & Humphrey, 1997) In particular, this study uses interest expenses and other operating expenses presenting for the banks’ inputs, and net interest revenue, other operating income for the banks’ outputs

To control multiple inputs and to allow a nonlinear relationship between the bank's total income and inputs, this paper uses Fiorentino's proposed translog function (Fiorentino, Karmann, & Koetter, 2006; Fontani & Vitali, 2014) Sharing the DEA data set, the translog function has two inputs, namely interest expense and other interest expense, as follows:

ln(Y it ) = b 0 + b 1 ln(X it1 ) + b 2 ln(X it2 ) + b 3 ln(X it1 ) ln(X it2 ) + b 4 ln(X it1 ) 2 + b 5 ln(X it2 ) 2 + (V it - U it )

Where: Yit: outputs (total revenue)

Xit1, Xit2: inputs (interest expense and other interest expense)

b: Vector Kx1 of unknown factors

Vit and Uit are assumed to have standard and semi-standard distributions, respectively

Granger causality is a statistical concept of causality that is based on the prediction Granger causality (or "G-causality") was developed in 1969 by Professor Clive Granger and has been widely used in economics since the 1960s Following Casu and Girardone (2009), we use autoregressive-distributed linear specification to disentangle the relationship between concentration and efficiency The lags (K, J) are determined by Augmented Dickey-Fuller Its mathematical formulation takes the following form:

0 If ES is hold, the coefficients for efficiency is positive and significant If SCP is hold, there are positive and significant coefficients of concentration

Data

Our data are collected from financial statements of 21 commercial banks in Vietnam from 2007 to 2014 We cannot cover financial data from the whole Vietnamese banking