Cost inefficiency of commercial banks in Vietnam: Stochastic frontier analysis

1019 COST INEFFICIENCY OF COMMERCIAL BANKS IN VIETNAM: STOCHASTIC FRONTIER ANALYSIS Vu Trung Hieu Department of Scientific Management, National Economics University Email: hieutrungvu@neu.edu.vn Abstract This paper measures the cost inefficiency of Vietnamese commercial banks from 2011 to 2014. Stochastic frontier analysis (SFA) is used to estimate the cost frontier. The average cost inefficiency of the whole banking system is 7.36% under intermediation approach. Findings show that state-owned banks and banks with bigger size are more efficient than joint stock or smaller ones. Negative relationships are found between efficiency and deposit-to-loan ratio as well as bank‟s age. Banks should focus on increasing their sizes and other earning assets, monitoring carefully the loan-making process to avoid non-performing loans, and investing in human resource. Key words: commercial bank cost inefficiency, Stochastic Frontier Analysis, intermediation approach. 1. Introduction The pace of globalization and trade liberalization has increased dramatically in recent years, making integration an inevitable reaction for all countries. This progress offers not only many opportunities, but also brings even more challenges, especially for developing economies like Vietnam. It is no exception for Vietnamese commercial banks, when they have to face more and more pressures and competitions from each other and from foreign banks. Therefore, being able to improve efficiency by analyzing its determinants is a must. Vietnam is a transition economy, thus its banking industry is also changing quickly since Doi Moi (renovation) in 1986. Unfortunately during this process of integration and change, Vietnam economy was affected by the financial crisis in 2007. Hence, some of the pre-crisis researches on Vietnamese bank‘s efficiency such as Nguyen (2008), and Vu & Turnell (2010) became less applicable. Some later studies included the post-crisis years in their researches, but to the authors‘ knowledge, the latest data analyzed was only until 2012 in Vu et al (2014) and Phan & Daly (2014)‘s works. Moreover, most researches focus on calculating the average efficiency of the whole Vietnamese banking system in general. Hence, the main objective of this study is to explore the efficiency measurements of Vietnamese banking system in general and commercial banks in specific and the determinants affecting these measures during the post-crisis period, utilizing intermediation approach, parametric method called Stochastic Frontier Analysis (SFA) and panel data collected from 21 Vietnamese commercial banks between 2011 and 2014. 1020 2. Literature Review The first appearance of the topic about efficiency can be dated very far back in the past to the publication day of the first modern work of economics: An Inquiry into the Nature and Causes of the Wealth of Nations (Smith, 1776). Nearly 200 years after, Koopmans (1951) and Debreu (1951) brought to light the first two thorough analytical approaches to the efficiency‘s measurement. Koopsman (1951) defined technical efficiency which reflects the ability to achieve maximal output from a given set of inputs, then Debreu (1951) initiated a measure of such technical efficiency, and Farrell (1957) first applied these ideas empirically. 2.1. The concept of efficiency Farrell (1957) idea of measuring efficiency is elaborated using a simple example of firms using 2 inputs (x1, x2) to produce 1 output (q). The idea can be illustrated in Figure 1 below. Fully efficient firms are represented by the isoquant SS‘. AA‘ is the isocost line. Point P represents the quantities of inputs used by a given firm to produce a unit of output. Thus QP – the amount where all inputs can be proportionally reduced without decreasing output – represents the technical inefficiency of that firm. Such inefficiency is usually expressed in percentage terms – QP/0P. The technical efficiency (TE) of that firm is hence given by: TE = 1 - QP/0P = 0Q/0P. TE takes value between 0 and 1. TE = 1 means the firm is fully technically efficient, for example firms using inputs at Q, i.e. producing on the efficient isoquant. When information about input price is known, the cost efficiency (CE) of that firm is also measurable. Let w represent vector of input prices; x represent the vector of inputs used at point P; x* represent the cost-minimizing input vector at point Q‘. CE is then defined as the ratio of input costs associated with input vectors at point P and Q‘: CE = w‘x* / w‘x = 0R/0P. (note: w‘ is the inverse vector of w). When information about input price ratio - represented by the slope of isocost line AA‘ - is also known, TE and allocative efficiency (AE) are also measurable. Let x~ represent the input vector used at technically efficient point Q. TE and AE is then given by: TE = w‘x~ / w‘x = 0Q/0P ; AE = w‘x* / w‘x~ = 0R/0Q. From here we have TE x AE = (0Q/0P) x (0R/0Q) = 0R/0P = CE. This is the decomposition of overall cost efficiency into 1021 technical efficiency - ability of firm to obtain maximum output from given inputs - and allocative efficiency - ability of firm to use inputs in optimal proportions given their prices. Farrell‘s research since then has become a seminal work for many studies and researches afterwards. 2.2. Measuring Efficiency: Frontier estimation methods To compute the above efficiency measures, i.e. to evaluate TE, AE, and CE, we first need to estimate the isoquant SS‘, which is essentially a best-practice frontier. This frontier can be estimated by either non-parametric or parametric approach. Non-parametric approaches include Data Envelopment Analysis (DEA) and Free Disposal Hull (FDH), while parametric ones consist of SFA, Distribution Free Analysis (DFA) and Thick Frontier Analysis (TFA). According to Berger & Humphrey (1997), these techniques differ primarily in what extent the restrictive form is imposed on the efficiency frontier and in the assumptions of the probability distributions of random error and inefficiency. Rangan et al (1988) are among the first to measure efficiency in the financial sector. Using DEA, they found that the average inefficiency of US banks to be 30%. They also concluded that this inefficiency is mainly due to pure technical inefficiency (wasting inputs) rather than scale inefficiency (non-constant returns-to-scale). Also using DEA, Elyasiani & Mehdian (1990a) found the average inefficiency of US banks to be much lower than above, of only 10%. In the same year, these two continued to apply SFA to find such value increased to 12% (Elyasiani & Mehdian, 1990b). Ferrier & Lovell (1990) are the first to compare the results between alternative methods (DEA vs SFA). Their average cost efficiency of US banks in 1984 from DEA and SFA were 83% and 79% respectively. They argued that such results did not meet the expectation that cost efficiency relative to non-parametric frontier is lower than that of parametric frontier (because noise is reported as inefficiency in non-parametric approach), so it means DEA frontier is more flexible to envelop data more closely than SFA frontier. However, Eisenbeis, Ferrier & Kwan (1996) did meet such expectation when SFA‘s efficiency is higher than DEA‘s. Berger & Humphrey (1991) used TFA to measure efficiency of all US banks in 1984 to be 81%. Such result is higher than SFA‘s and lower than DEA‘s value from Ferrier & Lovell (1990) above. Instead of explaining for such difference results between methods, Berger & Humphrey (1992a) continued to measure efficiency at more than one point in time, thus becoming one of the first to do so. Many later studies tried to use different methods on different points in time to measure banks‘ efficiency. Examples include Bauer, Berger & Humphrey (1993) and Resti (1996). 2.3. Inputs/Outputs specification: Intermediation vs Production approach There are 2 main types of approaches in choosing outputs, inputs prices and costs: Production approach views banks as producers which only use physical inputs of labor and capital to provide services like transactions and processing documents. Thus, 1022 output is best measured by the number and type of transactions or documents processed for a given period like in Sherman & Gold (1985)‘s and Kuussaari & Vesala (1995)‘s studies. However, such detailed transaction data is difficult to obtain, so many researches use number of accounts for output instead, such as Berger, Hanweck & Humphrey (1987), Ferrier & Lovell (1990), Ferrier et al (1993). Rangan et al (1988) also measured output in terms of the total balances of deposits and loans. Only physical inputs are needed to provide services so only their prices (labor rates and capital rates) and their costs (operating cost) are included in the model (Berger, Leusner & Mingo, 1996). Intermediation approach views banks as intermediaries which transfer funds between lenders and borrowers. Thus output should be measured as number of dollars intermediated in form of deposits, loans and other assets according to Berger & Humphrey (1991). They further stressed the dual characteristics of deposits: it is also an input, therefore its price (interest rate paid on deposits) and its cost (interest cost) will also be included in the model in addition to physical inputs prices/costs, thus the appropriate cost is total costs. Nguyen (2008) stated in his study concerning 32 Vietnamese commercial banks from 2001-2005 to follow intermediation approach, but he only included inputs (labor, capital and deposits) in his model. Deposits is also an output, but surprisingly Nguyen specifically classified it as input, leaving no output as independent variables. Vu et al (2014) attempted to correct this omission by including an output (total asset) in their deterministic component. However, their dependent variable of cost is operating cost, which doesn‘t include interest cost, making it inappropriate for their intermediation approach for 30 banks from 2006-2012. Vu & Turnell (2010) dealt with this problem by using 3 outputs (customer loans, other earning assets, actual value of off-balance sheet items) and 3 input prices (real wage, other non-interest expenses/fixed assets, interest expenses/total borrowed funds) to be regressed with total costs. The mean cost efficiency of Vietnam‘s banking sector found to be relatively high of 87%. They also concluded that bank ownership has insignificant effects on efficiency. However this research only investigated the pre-crisis period 2000- 2006 of 56 banks in Vietnam, so their findings will not be accurate anymore. Their outputs also didn‘t include deposits, meaning they failed to capture the output characteristic of deposits. Similarly, Matousek, Nguyen & Stewart (2014) only saw deposits as input but not as output, thus such dual traits of deposits are not covered. One development is that they managed to capture the changes over the financial crisis, using data during 1999-2009 for 48 banks. They also found that large banks were more efficient than small and medium sized ones. Phan & Daly (2014) further contributed to this topic by using more recent data of 26 banks during 2006-2012 which also covers the crisis‘s effect. Their model included 4 outputs (total loans, other earning assets, total deposits, liquid assets) and 2 input prices (price of funds, price of capital) and dependent variable of total costs. They found that state-owned banks are more efficient than private ones. 1023 To the limited knowledge of the authors, so far, there is only Berger, Leusner & Mingo (1996) who attempted to deal with both production and intermediation approach. However, their study was about branch efficiency instead of bank efficiency. They also could not include interest rates on deposits – one of the fundamental independent variables in intermediation approach – due to collinearity. 3. Model selection and Variables explanation Theoretically, according to Berger, Leusner & Mingo (1996), for studies concerned with evaluating the entire banks, intermediation approach is preferred because it is more inclusive and it captures the bank‘s role as a financial intermediary. An optimizing bank would minimize its total cost for a given output level, even if it means reducing interest cost at the expense of increasing operating cost. Such cost substitution can only be captured effectively by intermediation approach. This paper concerns about the efficiency of the entire banks rather than their branches. Therefore, intermediation approach is more appropriate, and hence applied in this study. Battese & Coelli (1995) SFA model is adopted in the form of cost frontier. 3.1. Stochastic cost function Ln(TCit) = 0 + 1*Ln(PLit) + 2*Ln(PCit) + 3*Ln(TLit) + 4*Ln(TDit) + 5*Yearit + 6*Ln(PFit) + 7*Ln(OEA) + Vit + Uit where  are unknown parameters to be estimated, it is of the i-th bank in t-th time period, Ln is natural logarithm of, TC is total cost, PL is price of labor, PC is price of capital, PF is price of funds, TL is total loan, TD is total deposit, OEA is other earning assets, V are random variables assumed to be independent, identically distributed N(0; ), and independently distributed of U, U are non-negative random variables assumed to account for the cost of inefficiency, and assumed to be independent, identically distributed as truncations at zero of N(Zitδ; ). In intermediation approach, output should be measured as number of dollars intermediated in form of deposits, loans and other assets (aforementioned in sub-section 2.3). Thus, very simply and clearly, total loans, total deposits and other earning assets are used as output measures in the function. Producing more output will generally incur more cost, thus both TL and TD are expected to have positive signs. Other earning assets is calculated by subtracting fixed assets and non-earning assets (cash and cash deposited at SBV) from total assets. Being an output, the sign expected for OEA is also positive. 3.2. Inefficiency model Uit = δ0 + δ1*Ownit + δ2*Ageit + δ3*Ln(TAit) + δ4*DLRit + δ5*EARit + δ6*Yearit + Wit where δ are unknown parameters to be estimated, Own is bank ownership, Age is bank ages, TA is total assets, DLR is deposits to loans ratio, EAR is equity to assets ratio, W are random variables. Bank ownership is used in terms of a dummy variable, which takes value of 0 if a bank is JSCB – joint stock commercial bank, or of 1 if a bank is SOCB – state-owned 1024 commercial bank. State-owned enterprises, which include SOCBs, are generally believed to be less efficient than private corporations (JSCBs) because they have the governments to back them up even if they perform badly. Thus, this variable is expected to have positive sign, i.e. if a bank is SOCB (takes value of 1), it is expected to have higher inefficiency. Bank ages is the years of operation of a bank. Banks with more years of operation will have more time to adjust, improve, and thus tend to have higher efficiency than younger ones. Hence, the expected sign for this variable is negative sign (inverse relationship between bank ages and inefficiency). Total assets represent the bank size. According to economies of scale, the bigger the bank, the more efficient it is. The expected sign for this variable is thus negative. Deposits to loans ratio (DLR) is related to bank‘s efficiency by showing how well banks use input (deposits) to produce output (loans). A high value of DLR, i.e. deposits are much higher than loans, means that the bank is not able to make sufficient loans, thus reducing its profit and efficiency. Hence it is expected that this variable has positive sign. EAR is the ratio of equity to total assets. It is a type of leverage ratio that gives us ideas about how much funds are from the bank itself (equity) and how much are from debt (deposits and other liabilities). High EAR means relatively low debt, and thus banks are not under the pressure to make much loans to pay for debts expense. This would lower the chance of encountering non-performing loans, thus reducing the cost for dealing with these bad debts, in turn would increase cost efficiency. Therefore, the expected sign for EAR is negative. 5. Results Ln(TC) = 4.53 - 0.36*Ln(PL) + 0.019*Ln(PC) + 0.061*Ln(TL) + 0.024*Ln(TD) (32.7) (1.72) (1.80) (2.09) (0.74) - 0.33*Year - 0.077*Ln(PF) - 0.028*Ln(OEA) (0.93) (2.57) (1.69) U = -0.092*Own + 0.003*Age - 0.023*Ln(TA) + 0.057*DLR + 0.016*EAR +0.027*Year (2.20) (2.03) (1.45) (1.72) (1.38) (1.78) The positive coefficient of TL suggests that the larger amount of loans a bank can make, the higher the cost it incurs. This is because of whenever banks make loans, they have to bear the provision expense. The coefficient of another output TD is also positive. However, such relationship cannot be statistically proven due to the statistical insignificance of the coefficient (t-ratio = 0.74 < 1.65). The negative and statistically significant coefficient of the last output OEA suggests the idea of more assets lead to lower cost. This is because these earning assets will generate more revenue, leading to reduction in our unit cost per revenue generated. Bank ownership‘s sign is not as expected. Its negative coefficient suggests that SOCBs are more efficient than JSCBs. This may be due to the fact that SOCBs in Vietnam have very much larger sizes compared to many other small size JSCBs. Thus the bigger SOCBs, which are Vietcombank (VCB), Vietinbank (CTG), and BIDV (BID) in this study, 1025 will benefit much more from economies of scale, having higher efficiency. In fact, these 3 SOCBs are the top most efficient banks found in this study. The positive coefficient of bank ages is also not as expected. The sign suggests that more experienced banks are actually less efficient. This may due to the old-fashioned, obsolete systems and practices of many Vietnamese banks. The older a bank is, the more reluctant it is when it comes to changes and updates. This is somewhat a culture of Vietnamese, especially with the generations 6x and 7x employees. Newer and younger banks are less prone to this problem. Their workforce mainly comprise of young adults, who have more educational qualifications and are more open to change compared to their seniors. Young banks also can adopt new, advanced organizational and technological systems more easily, while older banks cannot do this with ease since the systems that have existed for so long that they became the traditions of the banks themselves. The negative sign of TA is as expected. However, this relationship cannot be statistically proven due to the statistical insignificance of the coefficient (t-ratio = 1.45 < 1.65). Similarly, the positive coefficient of EAR is not as expected but its coefficient is statistical insignificant (t-ratio = 1.38 < 1.65). The positive coefficient of DLR is as expected. Its sign indicates that the higher the ratio, the more inefficiency there is, the less efficient a bank is. A high value of DLR, i.e. deposits are much higher than loans, means that the bank is not able to make sufficient loans, thus reducing its profit and efficiency. The positive coefficient of Year suggests that the cost inefficiency tend to increase during the 4-year period. This should be due to the high inflation rate in Vietnam during this period, which made the cost of everything higher, thus lowering the efficiency. The estimated average cost inefficiency of Vietnamese banking