Economic growth rate (GPD)
The economic situation is favorable, businesses are profitable, ensuring debt repayment obligations, commercial banks have good credit quality, increasing profitability. On the contrary, when the economic situation worsens, it will reduce the quality of the loan portfolio, increase credit risk provisions, thereby reducing the profitability of the bank.
Pasiouras and Kosmidou (2007), Neely and Wheelock (1997) argued that the relationship between GDP growth and Bank profitability is positive. This paper uses annual GDP growth rate to find the correlation between economic situation and bank profitability.
Inflation (CPI)
Inflation factor will directly affect the motivation of customers to deposit and borrow money. Therefore, it will directly affect the cost and income of commercial banks, thereby affecting the profitability of commercial banks. The thesis uses the annual inflation rate based on the CPI index (Athanasoglou, 2008). Detailed summary of information of variables in the table below:
Table 4.1 Variables in the research model
Variable | Measurement | Symbol | Expectation | Research before | |
Dependent variable | Rate of return | Profit after tax/Equity | ROE | Rokwaro (2013), Goddard et al. (2004) | |
Independent variable | Form of ownership | Dummy variables: 1: Commercial joint stock bank in which the state holds controlling shares 0: Joint Stock Commercial Bank | OWN | - | Cornett et al. (2008), Dietrich and Wanzenried (2014) |
Maybe you are interested!
-
Relationship Between Research Variables in the Model -
Evaluating How the Research Model Measures Variables -
Research Model of Factors Impacting Work Motivation of Tour Guide Team in Binh Dinh Province -
Preliminary Test of Reliability of Scale in Research Model -
Building a Research Model of Factors Affecting Agribank's Brand Value
Control variables
Bank size | Natural logarithm of total assets | SIZE | + | Emery (1971), Akhavein et al. (1997), Bourke (1989), Molyneux and Thornton (1992), Bikker and Hu (2002), Goddard (2004) |
Operating costs | Operating Expenses/Operating Income | OC | - | Alexiou and Sofoklis (2009), Dietrich and Wanzenried (2011 |
Credit risk | Loan Loss Provision/Loan Outstanding | CR | - | Dietrich and Wanzenried (2009), Trujiloo-Ponce (2013) |
Equity | Equity/Total Assets | CAP | - | Athanasoglou (2008) |
Asset structure | Outstanding Loans/Total Assets | L/A | + | Trujillo-Ponce (2013), Syafri (2012) |
Customer deposits | Customer deposits/Total liabilities | DEP | + | Trujillo Ponce (2013) |
Economic growth | Annual GDP growth rate | GDP | + | Pasiouras and Kosmidou (2007), Neely and Wheelock (1997) |
Inflationary | Annual CPI growth rate | CPI | + | Kunt and HuiZinga (1999), Alexious and Sofoklis (2009), |
Kasman (2010)
Source: Author's construction
4.2.2 Research methods
To determine the correlation between ownership form and profitability of Vietnam Joint Stock Commercial Banks, the thesis uses panel data. Many authors have used panel data in their models when researching the banking sector such as Srairi (2013) used to examine the risk level at Islamic National Banks, Athanasoglou (2008) also used panel data in his research to find out the factors affecting the profitability of banks. Using panel data has some major advantages: Studying the differences between cross-units, increasing the number of observations of the sample and partly overcoming the phenomenon of multicollinearity, containing more information than other data, studying the changing dynamics of cross-units over time...
In panel data regression, two popular regression models are fixed-effects and random-effects regression. Endogeneity in the model sometimes makes the estimation results from fixed-effects and random-effects models not really accurate. To solve this problem, previous studies have used instrumental variable estimation (IV estimation). However, the problem that arises when using instrumental variable estimation is that it is often difficult to find suitable instrumental variables because if weak instrumental variables are chosen, the IV estimation may be biased (Mileva, 2007). In other words, using IV estimation without choosing suitable instrumental variables will not improve the problems of OLS estimation. Therefore, the GMM dynamic panel data model is proposed to be used according to the study of Arellano and Bond (1991). One of the advantages of the GMM model over the instrumental variable estimation model is that the GMM model makes it easier to choose the instrumental variables because it uses exogenous variables in
other time periods or taking the lags of variables that can be used as instrumental variables for endogenous variables at the present time. GMM has introduced many instrumental variables to easily achieve the condition of a standard instrumental variable (Overidentification of Estimators). Therefore, to solve this problem, this paper applies another estimation method, the System Generalized Methods of Moments (System GMM) estimation method of Hansen (1982), for dynamic panel data (Dynamic Panel Data Analysis) proposed by Arellano & Bover (1995), Blundell & Bond (1998), Athanasoglou (2008). Besides, Arellano and Bond (1991) proposed two key tests to check the validity of the GMM model. The first test is the Sargan test or Hansen test for the validity (Overidentification) of the model. The second test used is the Arellano-Bond test to test for autocorrelation. The paper includes the following steps:
First, the thesis will regress the model with OLS, RE, FE methods. Test the model selection between OLS, RE, FE.
Second, handle biased and unstable model estimates due to endogeneity using the SGMM method.
Third, test the suitability of the SGMM model. The tests used in the model include:
Hausman test will be used to choose the appropriate estimation method between FE and RE (Baltagi, 2008, Gujarati, 2004). The hypothesis H 0 states that there is no correlation between the characteristic error 𝜀 𝑖𝑡 between subjects and the explanatory variables X it in the model (Cov( 𝜀 𝑖𝑡 , X it )=0). Rejecting the hypothesis H 0 means that FE is more suitable than RE. On the contrary, there is not enough evidence to reject H 0, which shows that there is no correlation between the error and the explanatory variables, so FE is no longer suitable and RE should be used.
To choose between RE and Pooled OLS, Lagrange fraction (LM) method with Breusch-Pagan test is used to verify the appropriateness of the estimate.
(Baltagi, 2008). Accordingly, the hypothesis H 0 states that the error of the estimate does not include the bias between subjects (var( 𝑢 𝑖 ) = 0)
Testing the reliability of the SGMM model: The SGMM estimate assumes that
There is no second-order autocorrelation of the residuals. Therefore, we need to check for autocorrelation in the error term, as well as test the adequacy of the proxies. The AR(1) and AR(2) test procedures can directly test for first-order and second-order autocorrelation of the residuals. According to Arrelano and Bond (1991), GMM estimation requires first-order autocorrelation (AR(1) test) and no second-order autocorrelation of the residuals (AR(2) test). Therefore, the null hypothesis is that there is no first-order AR(1) or second-order AR(2) autocorrelation of the residuals. Therefore, we want to reject H0 in the AR(1) test but want to accept H0 in the AR(2) test to get a consistent model (Basu, 2008).
4.3 Research data
The thesis data is taken from the audited financial statements of Vietnamese commercial banks in the period 2007-2014. The data was collected and selected after eliminating banks that had mergers and consolidations and banks that did not disclose information or had incomplete information. The result is a research sample of 20 banks (3 state-owned commercial banks and 17 joint-stock commercial banks) with 160 observations used for the research.
For macro data, the study uses data collected from statistical reports, information published on the websites of the General Statistics Office of Vietnam and the State Bank of Vietnam.
4.4 Research results
4.4.1 Descriptive statistics
Table 4.2 Descriptive statistical results
Variable
Mean | Std. Dev. | Min | Max | |
OWN | 0.15 | 0.36 | 0.00 | 1.00 |
ROE | 0.10 | 0.06 | 0.00 | 0.28 |
CAP | 0.11 | 0.06 | 0.04 | 0.36 |
CR | 0.01 | 0.01 | 0.00 | 0.04 |
OC | 0.48 | 0.15 | 0.19 | 0.98 |
Size | 17.77 | 1.25 | 14.60 | 20.31 |
LA | 0.53 | 0.14 | 0.19 | 0.85 |
DOP | 0.66 | 0.15 | 0.04 | 0.98 |
GDP | 0.06 | 0.01 | 0.05 | 0.08 |
CPI | 0.11 | 0.06 | 0.04 | 0.23 |
Source: Results from STATA software
Table 4.2 describes the mean, standard deviation, minimum, maximum and number of samples used in the study. The mean ROE of the sample is 10.08%, the lowest is 0.25% (NCB in 2014) and the highest is 28.46% (ACB in 2008). The mean values of the variables CAP, CR, OC, SIZE, LA, DOP, GDP, CPI are 11.35%, 1.29%, 48.22%, 17.77, 53.02%, 65.55%, 6.14%, 10.72%, respectively. In which, with a fairly high mean value of CAP, it shows that the Vietnamese banking system still ensures the minimum capital adequacy ratio. The main activity of the banking system is still lending, as shown by the high average value of the LA index. However, with an average value of 53.02%, it shows that banks have also diversified their business activities to disperse risks from credit activities.
Through the descriptive statistics table above, we can see that the variables ROE, CAP, CR, GDP, and CPI have relatively low standard deviations, showing the stability of large data, fluctuating around a small average value.
Table 4.3 Correlation matrix
OWN | CAP | CR | OC | Size | LA | DOP | GDP | CPI | |
OWN | 1.00 | ||||||||
CAP | -0.35 | 1.00 | |||||||
CR | 0.45 | -0.25 | 1.00 | ||||||
OC | -0.16 | -0.03 | 0.13 | 1.00 | |||||
Size | 0.64 | -0.68 | 0.49 | -0.02 | 1.00 | ||||
LA | 0.27 | 0.25 | -0.09 | -0.10 | -0.08 | 1.00 | |||
DOP | 0.09 | 0.16 | 0.16 | 0.17 | 0.10 | 0.50 | 1.00 | ||
GDP | 0.00 | 0.02 | -0.32 | -0.37 | -0.24 | -0.05 | -0.26 | 1.00 | |
CPI | 0.00 | 0.18 | -0.04 | -0.06 | -0.15 | -0.03 | -0.18 | 0.03 | 1.00 |
Source: Results from STATA software
Looking at the correlation matrix, we see that the correlation coefficients between the variables are all lower.
0.8. Therefore, the research model has a low possibility of multicollinearity.
4.4.2 Quantitative analysis
First, a general OLS regression is performed to analyze the relationship between ownership form and profitability. For the model estimated by this method, all coefficients are invariant across different subjects and do not change over time (Gujarati, 2004).
Table 4.4 OLS regression results
ROE
Coef. | Std. Err. | t | P>t | |
OWN | -0.013 | 0.013 | -0.960 | 0.340 |
CAP | -0.278 | 0.077 | -3.630 | 0.000*** |
CR | -1.246 | 0.523 | -2.380 | 0.018** |
OC | -0.208 | 0.023 | -8.920 | 0.000*** |
Size | 0.018 | 0.005 | 3,760 | 0.000*** |
LA | 0.019 | 0.029 | 0.660 | 0.508 |
DOP | 0.044 | 0.026 | 1,670 | 0.098* |
GDP | 0.755 | 0.543 | 1,390 | 0.166 |
CPI | 0.578 | 0.366 | 1,580 | 0.116 |
_cons | -0.214 | 0.114 | -1.890 | 0.061 |
Note: *, **, *** represent significance levels of 10%, 5% and 1% respectively.
Source: Results from STATA software
The OLS regression results show that ownership form has no impact on the profitability of commercial banks. However, the robustness and efficiency of the coefficients in the panel data analysis based on the general least squares regression method can be questioned because the general OLS model does not need to consider unobservable factors or individual effects, specific to each bank, while the problem of individual effects is one of the phenomena that occurs frequently in empirical studies (Baltagi, 2005). Therefore, to deal with the problem of unobserved heterogeneity, random effects (RE) and fixed effects (FE) models are used.
Table 4.5 Hausman Test
Test: Ho: difference in coefficients not systematic
chi2(8) = (bB)'[(V_b-V_B)^(-1)](bB)
= 3.60
Prob>chi2 = 0.8913 (V_b-V_B is not positive definite)
Source: Results from STATA software
According to the test results of choosing between FE and RE models in table 4.5, we have p-value = 0.8913 > 5% significance level, so we do not have enough evidence to reject H 0 , accepting H 0 means there is no correlation between the error and the explanatory variables, the model used is RE.
To choose between RE and Pooled OLS, Lagrange fraction (LM) method with Breusch-Pagan test is used to verify the appropriateness of the estimates (Baltagi, 2008).