BANK SIZE AND PROFITABILITY OF COMMERCIAL BANKS. EVIDENCE IN VIETNAM45470

IN TERNATIONAL CONFERENCE ON - CIFBA 2020 BANK SIZE AND PROFITABILITY OF COMMERCIAL BANKS EVIDENCE IN VIETNAM Le Dong Duy Trung1*, Nguyen Duy Thanh2 Vietcombank Head Office & National Economics University, Economic and Management Department, Thang Long University INTRODUCTION The banking industry is considered the lifeline of every economy and plays an important role in the development of the country's economy The bank acts as a monetary trading organization with the task of storing, mobilizing and distributing money Moreover, banks are the financial intermediary between depositors and borrowers with the core activity of receiving deposits from depositors and lending for borrowers (Casu et al, 2015) Therefore, banking operations in general and profitability in particular attract the attention of researchers as well as investors in the market In Vietnam, there have also been some researchs on the profitability of the banking system, but they often not focus on specific bank groups according to their size Following to Article of Circular No 52/2018-/-TT-NHNN on the ranking of credit institutions and foreign bank branches, this study divides the commercial banking system into two groups Large-scale bank group (average total assets on a quarterly basis of over VND 100 trillion) and small bank group (average quarterly total assets less than or equal to VND 100 trillion), will provide empirical evidence about the difference in the impact of different determinants on ROA and ROE of the two groups of banks As the time of the study, Circular 52 is valid for more than months (from April 1, 2019) Therefore, this study carried out a classification approach according to groups of Circular 52 ABTRACT Most of theories and empirical researchs on the world in the past showed differences in the behavior of large banks and small banks Large banks often have the advantage of diversification and using more non-traditional activities than small banks This study focuses on analyzing the differences in impact of Bank profitability determinants according to the approach consistent with the offical classification of * Corresponding author Email address: ledongduytrung@gmail.com 430 VIETNAM NATIONAL UNIVERSITY - UNIVERSITY OF ECONOMICS AND BUSINESS banks according to the size of total assets in Vietnam The results show that large banks are less affected by external characteristics such as industry concentration or macroeconomic characteristics compared to small banks The results also support the structure conduct performance (SCP) hypothesis as a concentration of the industry increases, helping large banks strengthen ROE while negatively impacting ROE of small banks In addition, large banks will gain more benefits than small banks when boosting non-interest income, reflecting the advantage of diversification according to financial intermediary theory However, while the negative impact of liquidity risk on the profitability of large banks is clear, this impact on small banks is negative in the case of ROA and positive in the case of ROE as a secondary variable independence LITERATURE REVIEW 3.1 Dependent variable This study focuses on assessing impacts on profitability so the dependent variable is defined as the profit that the bank generates in a certain business period - usually measured by the year term Studies on business efficiency or profitability of commercial banks often use dependent variables such as ROA or ROE ROA reflects the level of efficiency in bank managers' asset management, while ROE indicates the profitability of investment in banks that shareholders are interested in Athanasoglou et al (2008) evaluated the characteristics affecting the profitability of commercial banks are often considered according to three characteristics: Group of characteristics of Bank-specific, lndustryspecific and Macroeconomic-specific determinants 3.2 Independent variables The group of commercial bank characteristics often includes elements describing the size of commercial bank assets, equity size, liquidity risk, credit risk, income structure, and operation expenses Bank size: When assessing the impact of bank size to the profitability of the bank, there are groups of different views The first view is that: when size of banks reach a threshold, they can have economies of scale (Economies of scale) Experimental studies on this point of view are Kosmidou (2008), Athanasoglou et al (2006), Drake and Hall (2003), Shehzad et al (2013), Beccalli et al (2015) The second group of points said that the bigger the scale of assets, the lower the profitability, because the higher the total asset, the more it becomes less flexible, rigid and administrative, which reduces Bank efficiency This research group received the support of Mitchell and Onvural (1996) with the sample of banks in the US from 1986-1990, Subhash (2007) studying banks in India from 1997-2003, Tabak and partner (2011) studied banks in Brazil in the period of 2003-2009 Capital: The ratio of equity to total assets indicates the percentage of total assets constituted by the money contributed by shareholders Berger (1995a) argues that when banks improve this ratio will negatively impact (negative) profitability, because 431 IN TERNATIONAL CONFERENCE ON - CIFBA 2020 then, the bank's overall risk decreases, causing its expected profitability also reduced according to the risk-profit trade-off theory Goddard et al (2010); Akbas (2012) also gives similar conclusions and implies that equity is an expensive source of financing, assuming a high equity ratio reduces the leverage impact, thus increasing the cost of financing for Bank In contrast, Molyneux (1993) argues that raising capital helps increase credit ratings and helps banks reduce capital costs Berger (1995b) also argues that when banks increase their capital, they will increase their tolerance (when risks - especially credit risks - occur), thus boosting credit growth to gain profits Higher margin Some views of some bank executives believe that the increase in equity in the short term increases the cost of capital, which large banks often have lower rates than other banks (Beccalli and colleagues) , 2015; Rose, 1999), or is often referred to as "Too big-Too fail", so there will be formal or informal sponsorship from the government (Kaufman, 2014), hence the profitability will be higher Liquidity risk: Usually measured by the ratio of credit outstanding on customer deposit balance (LDR) or on total assets An increase in this ratio indicates that there is a need for a greater amount of capital or total assets to lend This increases liquidity risk in case customers withdraw money on a large scale Bourke (1989) argues that the relationship between this ratio and profitability is positive because increasing the liquidity risk means that the bank boosts lending to collect interest income, thereby increasing profits However, many international studies show that the relationship between liquidity risk and business performance is negative Diversification of income (non-interest income): The degree of income diversification is often measured by the ratio of non-interest income to total operating income or total assets This indicator shows characteristics in the business model of commercial banks Rose (1999) said that similarly sized banks often trade similar products/ services Demsetz and Strahan (1997) argue that big banks have an advantage in diversifying income because they often take advantage of income diversification to operate In contrast, Demirguc-Kunt and Huizinga (1999) argue that the relationship between the proportion of non-interest income and the profitability of the bank is negative, given that the level of competition in the non-traditional market (as risk or investment bank) higher than traditional banking market, thus reducing profitability if banks rely on non-traditional services Operation expenses: Jiang et al (2003) argue that operating expenses ratios have a negative impact on a bank's profitability because an efficient bank will cut costs and improve profits from there In contrast, Molyneux and Thornton (1992), Guru et al (2002), and Ben Naceur (2003) argue that increasing operation expenses have a positive effect on profitability These studies suggest that a large part of operating expenses ratios are for paying salaries and employee benefits and paying higher salaries to experienced employees helps them to be motivated to work much and more effectively, this view supports efficient wage theory (Efficiency wage theory) Net Interest margin (NIM): Reflects the difference between interest income and 432 VIETNAM NATIONAL UNIVERSITY - UNIVERSITY OF ECONOMICS AND BUSINESS interest expense per asset unit Interest income plays an important role in the profitability of commercial banks in Vietnam, so, visually, research on the positive impact of NIM on bank profitability Market concentration: This indicator is often measured by the focus group (CR _k), which often measures the concentration of four commercial banks with the largest share of the region Therefore, the author selected biggest commercial banks in Vietnam, namely VCB, Vietinbank, BIDV and Agribank to analyze the scale Hypothesis Structure-Conduct-Performance, CR_k represents the level of competition in the market If this index is large, the level of competition decreases, large banks will achieve higher profits and vice versa Control variables: This study incorporates macroeconomic variables such as M2 money supply growth rate, real GDP growth rate and inflation rate, in order to make assessments of indicators affecting the profitability of commercial banks RESEARCH METHODOLOGY This study divides commercial banks into two groups: a large-scale commercial bank (the total average quarterly asset value in the year is ranked at over VND 100 trillion); and Commercial banks are small in size (the total average asset value in a quarter is equal to or lower than VND 100 trillion) according to the provisions of Circular 52/2018 / TT-NHNN Research using quantitative research method with table data (Panel data), including estimation methods (econometrics method) include: OLS regression method (Pooled OLS), fixed impact method ( Fixed Effect, the method of random effect (Random Effect) with robust standard errors with the support of Stata software 15.1 Data: Include business data of 30 commercial banks collected from the Financial Statements and Annual Report of commercial banks in Vietnam from 2009 to 2017 Besides, the data about macroeconomics collected from ADB Key Indicator Report of Asian Development Bank (ADB) Model: Based on the theory and empirical researches on the determinants affecting performance of commercial banks, the study proposes an experimental model as follows: 11 Yint = b0 n + b1Sint + b2CAint + b3 LDRint + b4 NIM + b5 DIAint + b6 LPCLR int + b7OEARint + � bk X kt + mi + eit (1) k =8 Where Xkt: Vector includes industry structure variables (CON) and macroeconomic variables (MSG, GDPG, INF) Indicators i: Only cross-units (commercial banks), n: groups that commercial banks belong to (large or small banks) at year t µi is an unobserved component reflecting inherent characteristics in each bank, can be correlated or not correlated with other independent variables in the model and ε it is the characteristic error factor (idiosyncratic error term) of the model 433 IN TERNATIONAL CONFERENCE ON - CIFBA 2020 Table Variables in the model and impact hypothesis Variables Describe Hypothesis impact Data source Dependent variables (Y) ROA Return on assets Annual report of commercial banks ROE Return on equity Annual report of commercial banks Independent variables S Natural logarithm Total assets (VND billion) +/- Annual report of commercial banks CA Equity / Total assets +/- Annual report of commercial banks LDR Loan Deposit Ratio (Net Loan/Customer Deposit - Annual report of commercial banks NIM Net Interest Margin + Annual report of commercial banks DIA Diversification (Non Interest Income/ Total Assets) + Annual report of commercial banks LPCLR Loan Loss Provision Cost to Loan Ratio (Loan Loss Provision Cost /Net Loan) - Annual report of commercial banks OEAR Operation Expenses Ratio (Operation Expenses/Total Assets) - Annual report of commercial banks CON Concentration ratio (CR_4) MSG Growth rate of Money Supply (M2) + ADB Indicator GDPG Growth rate of real GDP + ADB Indicator INF Inflation rate + ADB Indicator +/- The author calculates Estimation methodology The study used estimation techniques for table data including Pooled OLS, Fixed Effect Method, Random Effect Method First, the study will test whether high-level multicollinearity occurs in the model If serious multicollinearity does not occur, the study will continue to test heterosdasticity and autocorrelation, and estimate with Robust Standard Error to overcome these phenomena Finally, the study uses Breusch and Pagan LM Test and Hausman tests to select the most appropriate estimates and analyze the results This study separates the sample into two separate groups according to the size of their total assets without using dummy variables In this case, this method has advantages over the dummy method If using the N-binary dummy variable representing two banking groups (Intercept dummy), to estimate the difference in effect of all the independent variables between the two groups, it is necessary to include the interaction 434 VIETNAM NATIONAL UNIVERSITY - UNIVERSITY OF ECONOMICS AND BUSINESS components of N and all Slope dummy Adding the dummy Slope needs to add dummy variables N, which are both present in the model, otherwise the difference in the original coefficient due to the dummy Slope will have to load the difference due to the lack of dummy Intercept presence and yes can make estimates deviate In fact, having both of these variables at the same time is easy to cause serious linear multicollinearity that can alter the estimation of the estimation, especially in the case of non-large samples (like this study) To avoid having to choose between high-level multicollinearity and the violation of the principle of marginality, the study selected the above approach RESEARCH RESULT Discriptive statistics Table Variables descriptive statistic in the model (1) Variables Obs Mean Std Dev Min Max ROA 270 0.01051 0.008989 -0.0551175 0.052521 ROE 270 0.082232 0.080501 -0.8200213 0.268235 S 270 11.26727 1.222231 8.11071 13.99973 CA 270 0.104862 0.055986 0.0346185 0.375897 LDR 270 0.874762 0.238939 0.3632857 2.521567 NIM 270 0.026259 0.010965 -0.0064123 0.074219 DIA 270 0.005844 0.004935 -0.0058772 0.038609 LPCLR 270 0.010376 0.007879 -0.0101409 0.050291 OEAR 270 0.016018 0.00556 0.008961 0.051961 CON 270 0.525493 0.022397 0.4824446 0.572989 MSG 270 0.19882 0.064375 0.1207439 0.332977 GDPG 270 0.060463 0.005433 0.0524737 0.068122 INF 270 0.068184 0.049938 0.0063 0.1858 With the exception of CON variable (concentration level) and variable GDPG, the level of variation is quite low (the ratio of standard deviation to the average value is 4% and 9% respectively), the remaining variables have the degree of relatively high volatility, especially variables ROE (98%), ROA (86%) and DIA (84%) and LPCLR (76%) These are all variables that show the business characteristics of banks Therefore, we can see the diversity of profitability, business model (based on traditional or nontraditional products) and credit risk among banks in the sample Results of the multicollinearity test Testing the phenomenon of multicollinearity In order to detect high linear multicollinearity in the model, firstly, study the implementation of the Pearson correlation matrix According to research practice, if between two independent variables have a correlation coefficient greater than 0.8, it is a sign of high-level multicollinearity Next, the study presents the estimation of the magnification coefficient of variance VIF (Variance Inflation Factor), if the coefficient VIF of a variable explains more than 10, it can be concluded that there is a serious multi435 IN TERNATIONAL CONFERENCE ON - CIFBA 2020 linear phenomenon in model (Kennedy, 2008) Some stricter views suggest that when VIF> is capable of high-level multicollinearity in the model Table Pairwise matrix of correlation coefficient S S CA LDR CA -0.7154 *** -0.207 *** DIA 0.0734 0.2603 *** *** 0.3336 *** 0.2029 *** -0.077 0.1393 0.229 OEAR CON MSG GDPG INF 0.0455 0.0403 -0.1419 0.3911 0.2553 ** *** 0.6061 *** CON -0.0723 -0.0007 0.0312 -0.2488 0.1711 0.2062 *** 0.2111 *** -0.2151 *** *** *** -0.1865 *** ** *** -0.0396 INF LPCLR OEAR GDPG DIA 0.3658 -0.0546 MSG NIM NIM LPCLR LDR 0.0029 0.0938 *** -0.1404 0.1688 ** *** 0.2456 -0.0991 0.0031 -0.0832 0.1857 0.2276 0.1773 *** *** *** 0.4303 -0.0546 -0.2271 *** *** -0.0393 0.0748 -0.0222 *** -0.1276 ** -0.2377 0.4315 *** -0.1185 *** 0.2091 0.2156 * *** *** -0.1418 ** 0.0485 0.6413 *** 0.0404 0.2298 *** Note: ***, ** and * represent 1%, 5% and 10% significance levels, respectively The pair correlation matrix shows that all correlation values are less than 0.8 Therefore, it can be assumed that there are no high-level pluses in the model However, the study also estimated VIF coefficients for more certain results Table Estimated results of VIF coefficient Independent varibale CA CON INF S NIM OEAR LPCLR MSG DIA GDPG LDR Mean VIF VIF 3.6 3.38 3.2 3.11 2.35 1.98 1.66 1.64 1.61 1.6 1.44 2.32 The estimated VIF result table shows that all VIF values ​​are smaller than 5, the average VIF value is 2.32 quite small Thus, the results of the VIF estimation once again confirm that there is no high-level multicollinearity in the research model, and allow to continue the estimation of the model (1) The results of model estimation (1) with two banking groups in the two cases of ROA and ROE are dependent variables presented in table (5) and table (6) In both models with ROA and ROE, the dependent variable, the result of simultaneous 436 VIETNAM NATIONAL UNIVERSITY - UNIVERSITY OF ECONOMICS AND BUSINESS validation (Wald Test) rejects the hypothesis H0: independent variables have no concurrent effect on the dependent variable, and the coefficient Rsquare of regular models above 0.85 shows that the independent variables included in the model are able to explain most of the variation of the dependent variable Verification Breusch & Pagan LM Test also rejected the hypothesis H0: the model does not contain unobservable components, thereby estimating the inappropriate POLS in the models Hausman's test validates hypothesis H0: µi does not correlate with other independent variables in the model, thereby accepting REM estimates The Hausman test results show that in the case of banks with a scale of less than VND 100 trillion, the estimation of FEM is most suitable, while for banks with a scale of over VND 100 trillion, REM estimates are suitable unify To test the phenomenon of heteroscedasticity, the study uses two tests simultaneously, namely Breusch-Pagan / Cook-Weisberg test for estimating POLS and Wald calibration (Modified Wald) in school FEM estimates Two tests carry out the test for hypothesis H0: the model has no heteroscedasticity phenomenon Breusch-Pagan / Cook-Weisberg test rejects the hypothesis H0 with the model with an average banking group of over VND 100 trillion and does not reject H0 with the school with an average group size of under 100 VND trillion with ROA is the dependent variable, the opposite of the case where ROE is the dependent variable Meanwhile, the Modified Wald test rejects H0 with two groups of bank Thus, the model is more likely to happen a phenomenon of heterosdasticity for two groups of bank Woodridge test with hypothesis H0: the model does not happen a phenomenon of first order autocorrelation that also gives conflicting results In case of ROA is dependent variable, Woodridge Inspection rejects H0 in the model with bank group with total assets of less than 100 trillion and accept H0 with bank group with total assets of over 100 trillion, and results vice versa in case of ROE is dependent variable In order to overcome the phenomenon of heteroscedasticity and the phenomenon of autocorrelation, the estimation results are done simultaneously with the robust standard error method The robust standard error method is still usable in the absence of such defects Analyzing the results of estimating the case of ROA is the dependent variable The estimated results with ROA are the dependent variables presented in Table Estimated results for banks of less than VND 100 trillion according to FEM, the estimated results for the group of banks of the scale VND 100 trillion under REM FEM is only different from REM in allowing µi to correlate with other independent variables (Woodridge, 2012) Therefore, we can still compare the magnitude of the effect of independent variables to ROA according to two banking groups when using these two methods together Similar to the case where ROE is the dependent variable The semi - elasticity coefficient of the total asset scale is not statistically significant in the case of small banks, while for large banks implies that when other conditions are constant, if the total assets of the banks This item increased by 1%, reducing ROA by 0.09 percentage points This implies that banks in large groups not have an increase in profits corresponding to their scale increase or the purpose of 437 IN TERNATIONAL CONFERENCE ON - CIFBA 2020 increasing their scale is not entirely to achieve economies of scale but also to enjoy the advantage "too big to fail" (Brewer and Jagtiani, 2013) or simply because they become more important because they are bigger The impact of equity on total assets is not statistically significant for either group Liquidity risk measured by LDR shows a more serious negative impact on large banks' ROA In return, large banks have an advantage when increasing credit interest rates to increase NIM because of the impact of NIM on their ROA is larger than small banks, which can be explained by group monopoly theory Since most of Vietnam's commercial banks' assets are assets formed from credit, a large group of banks accounts for credit market share Table shows the average value of the industry concentration index (CON), showing the proportion of assets of the largest commercial banks in the industry (Agribank, Vietinbank, BIDV, Vietcombank) is approximately 53% Large banks have a high market position in traditional products, so the level of sensitivity to interest rates on deposits and loans of large banks is lower than that of small banks Taking advantage of this allows large banks to increase profit margins with a lower level of negative impact on their product demand, so NIM has a better positive impact than small banks Credit risk (LPCLR) has a fairly equal impact on both groups of banks, despite negative impacts on small banks, although the difference is small This may be due to the lower credit market power, small banks often have to identify borrowers who are small and medium enterprises, or have a lower level of credit rating higher than the big banks On the other hand, large banks often have advantages in managing credit risk better than small banks The impact of non-traditional services (services, insurance, investment ) is positive for ROA for both banks This implies that Vietnamese banks have been transforming their business model into a more balanced model (increasing the proportion of service fees, non-interest activities) is correct However, this positive impact with smaller banks is higher than that of big banks The result of this positive impact supports some recent studies on the negative relationship between NIM and the degree of bank diversification such as James (2012) or Tu (2017) that has been done with banks in Vietnam in both directions When large banks take advantage of their advantages over (i) economies of scale by financial intermediation theory or (ii) price advantage in traditional markets by market position The field, which has a worse impact (though not much for this ROA case) to the effect that non-interest services bring to ROA The impact of enterprise operating expenses ratios on ROA is also about 19% more negative for large banks (-1,0817) compared to small banks (-0.9111) It should be noted that, although there are differences, the negative impact of corporate operating expenses ratios on ROA in both groups is the largest compared to other characteristics such as liquidity risk or credit risk in the research model Market concentration ratio (CON) shows the level of competition in the market according to the theory of competitive structure in general and more specifically the Structure-Behavior-Perfomance hypothesis (SCP) This Harvard-based hypothesis assumes that when the market concentration is high, meaning that the level of market competition decreases, banks that are on the market (big banks) can make use of 438 VIETNAM NATIONAL UNIVERSITY - UNIVERSITY OF ECONOMICS AND BUSINESS it High price range to increase profit margin However, in this case this effect is not statistically significant, so there is no evidence to support or reject this hypothesis For macroeconomic variables: Economic growth is not statistically significant in both cases with two banking groups Money supply and inflation have a positive impact on small banks, while inflation only has a positive impact on large banks The increase in money supply often causes positive inflation, so it is possible, for large banks, that inflation has shown the final impact This result is similar to some recent studies such as Pasiouras and Kosmidou (2007), Athanasoglou et al (2008), or GarciaHerrero et al (2009), these studies also show positive effects of inflation on business performance of commercial banks It should be noted that the value and statistical significance of lower inflation effects as well as the number of macroeconomic variables are less statistically significant for large banks showing that these banks bear Less impact from the external environment compared to small banks With its large scale, large banks easily manage and exploit to improve their business performance, which represents the advantage of diversifying according to financial intermediary theory In addition, Perry (1992) points out that this effect is positive only in the case of inflation being "predictable." A policy of inflation targeting policy is transparent and clear to help commercial banks benefit from the initiative to adjust credit and mobilizing interest rates and vice versa Analysis of the results of estimating the case of ROE is the dependent variable The difference when using ROE and ROA is also the impact of the non-interest income ratio (DIA) variables, the operating expenses ratio (OEAR), the concentration of the market (CON) and the business variables macroeconomics such as money supply growth rate (MSG), GDP growth rate (GDPG), inflation rate (INF) In the case of ROE, the ratio of non-interest income shows a positive effect stronger than that, thus showing the advantage of diversification of large banks On the contrary, when operating expenses ratios show that the negative impact is much stronger (44%) for small banks, it shows the conflict in business goals between bank managers and shareholders according to the theory of great people Agency theory in small banks may be more serious than large banks Market concentration has been highly significant in this case This variable has a negative impact on small banks while positively impacting large banks, thus supporting the SCP hypothesis, particularly large banks will have an advantage when the banking market becomes focused more, allowing large banks to take advantage of market position and allow their ability to impose higher prices 439 440 -0.4718*** -0.9043*** 0.0146** 0.0055*** 0.0211 LPCLR OEAR CON MSG GDPG 152 24 Number of individuals 0.9908 Rsquare Number of observations 0.000 Wald Test P-value -0.0201*** 0.9400*** DIA _cons 0.8808*** NIM 0.0190*** -0.0021*** LDR INF 0.0136*** CA 0.005 0.003 0.022 0.001 0.006 0.020 0.030 0.021 0.014 0.000 0.003 18 118 0.9655 0.000 0.0059 0.0103** -0.00001 0.0020 0.0144 -1.0527*** -0.4724*** 0.9510*** 0.9456*** -0.0046*** 0.0272*** -0.0009*** 0.009 0.005 0.028 0.002 0.012 0.044 0.036 0.037 0.035 0.001 0.009 0.000 0.0010*** S 0.000 Robust Std Err Robust Std Err Coef Variables Coef Total asset >100 trillion VND Total asset 100 trillion VND REM Total asset 100 trillion VND 0.09951 0.000 Coef Total asset 100 trillion VND Total asset 100 trillion VND Total asset 100 trillion VND REM Total asset 100 trillion VND REM Total asset 100 trillion VND 0.000 0.001 Breusch and Pagan LM Test (P-value) Coef Total asset 100 trillion VND Total asset