Regression Model Results of 6 Factors Affecting 8 Listed Joint Stock Companies

+ The COLL variable always has a high Prob value, and is always greater than 10%, so it can be assumed that the COLL variable has no impact on financial leverage for Vietnamese commercial banks.

+ The DIV variable also has a value that fluctuates around the 5% and 10% significance levels. Specifically, in the L 1 (1) and L 1 (3) models, the DIV variable both affects financial leverage at the 5% significance level.

Thus, from the synthesis results of the above models, we see that when using the FEM model, the level of explanation of the L 1 model increases significantly. And the author chooses the L 1 model (3) to use for analysis to bring about the highest efficiency in explaining the variation of financial leverage.

4.2 Results of regression model of 6 factors affecting 8 listed commercial banks

listed

To be able to test the impact of MTB and RISK factors

How to capital structure of Vietnamese commercial banks, the author reuses model (1) with only 8 listed commercial banks (ACB, CTG, EIB, MBB, NVB, SHB, STB, VCB). This sample is only taken from listed banks because only listed banks have market prices of shares and therefore can calculate market value over book value, and daily price fluctuations. However, these banks have not been listed on the market long enough, so if the author takes data by year, the sample is very small, so the author uses quarterly data for the above 8 commercial banks.

L 2 = β 0 +β 1 MTB it-1 + β 2 SIZE it-1 +β 3 PRO it-1 +β 4 COLL it-1 +β 5 DIV it +β 6 RISK it-1 (2)

4.2.1 Regression model results for financial leverage

Similar to above, before using the Ordinary Least Squares method (OLS), because the author added 2 independent variables MTB and RISK to the model, it is necessary to re-check the correlation coefficient matrix between the independent variables. From the calculation results of the EVIEWS program, we have the correlation matrix between the variables in the model.

Table 4.5: Correlation matrix between dependent variable and 6 independent variables

Correlation

Maybe you are interested!

• Factors affecting the debt repayment ability of corporate customers at Vietnam Joint Stock Commercial Bank for Investment and Development - Long An Branch - 1
• Factors affecting customer satisfaction with the quality of international money transfer services at Dong A Commercial Joint Stock Bank - 1
• Factors affecting customer loyalty to international payment services at Asia Commercial Joint Stock Bank - 12
• Factors affecting card acceptance of Kien Long Joint Stock Commercial Bank - 18
• Factors affecting the decision of customers to deposit savings at Asia Commercial Joint Stock Bank - 1

With the above results, we see that there is no correlation between pairs of independent variables in the model, the largest correlation coefficient between independent variables is 0.28. This shows that the variables in the model are suitable and there is no multicollinearity. And next, we will test the regression model of factors affecting the financial leverage of NHTMCPVN.

Table 4.6: Financial leverage regression results of 8 listed commercial banks

Looking at the above results, we see that the coefficient R 2 = 0.4421, and the two independent variables MTB, SIZE in the estimated model both have a significance level Prob < 0.01, meaning that these variables are suitable for the model and are significant at the 1% level; the variable COLL has a significance level Prob < 0.05, meaning that the variable COLL is suitable for the model and is significant at the 5% level. The remaining variables PRO, DIV, RISK are not significant in this model. At the same time, through the F test index, we have a regression coefficient F = 17.57 with a significance level much smaller than 1%, so this model is suitable. However, the important issue of the model lies in

at Durbin-Watson coefficient = 0.435 (d<1) 5 shows that there is autocorrelation phenomenon in

model above. Therefore, despite its simplicity, using the combined regression as above may distort the actual picture of the relationship between financial leverage dependence

5 Hoang Ngoc Nham et al., 2007. Econometrics Textbook. Ho Chi Minh City University of Economics, page 193.

and independent variables in 8 surveyed commercial banks. To overcome the above autocorrelation drawback, the author uses regression analysis using fixed effects. And the above section also shows that the model using fixed effects is more suitable than the model without fixed effects. Therefore, in the next section, we will test the above regression model with fixed effects. If the fixed effects model continues to have autocorrelation, the author will proceed to correct autocorrelation according to the 2 steps of Durbin - Watson.

4.2.2 Regression model results for financial leverage incorporating fixed variable effects

Using Eiews software, the author obtained the following synthesis results of the above model when adding fixed effects:

Table 4.7: Summary of results using fixed effects on the 6-variable model.

Dependent Variable

Similar to the previous section, the results of the table above show that the coefficient R 2 is significantly improved compared to the original model (44%), adjusted to 72%, 82%. This shows that the level of variation in financial leverage is increasingly affected by independent variables. In addition, the significance levels Prob of the F test are almost equal to 0. To check which model is really suitable, we continue to conduct hypothesis tests on the individual regression coefficients: With the model's hypothesis being H 0 : β i =0 and H 1 β i ≠0, using the results of the model's significance level Prob, we have the following results:

+The constant C has a relatively small significance level Prob, only in the L 2 model (1)

significant at the 5% level, all other models L 2 , L 2 (2), L 2 (3) are significant at the 1% level.

+ The MTB variable is always very insignificant in all four models, so it can be assumed that the financial leverage of listed Vietnamese commercial banks is affected by the market value to book value.

+ The SIZE variable always has a very small significance level in all 4 models, much smaller than 1%, thus rejecting the hypothesis H 0 : β 1 = 0. Therefore, it can be affirmed that the SIZE variable has a meaning in explaining the financial leverage variable at a significance level of 1%.

+ The COLL variable also has an unstable Prob value, and is only 1% significant in the L 2 (1) model, so it can be assumed that the COLL variable has no impact on financial leverage for listed Vietnamese commercial banks.

+ The DIV variable always has a high Prob value. Therefore, it can be assumed that DIV has no impact on the financial leverage of listed Vietnamese commercial banks.

However, at this time, using the fixed effect model FEM still cannot overcome the autocorrelation problem, the Durbin-Watson coefficients of all 3 models are not suitable (d<1). The reason for the above model's autocorrelation can be explained by the following main reasons:

- Inertia: The most prominent feature of most time series in economics is inertia. We all know that time series such as gross domestic product, price index, unemployment are cyclical. Therefore, in the regression of time series, successive observations are likely to depend on each other. Using data from the financial statements of banks continuously from the third quarter of 2006 to the first quarter of 2013 is also a time series, and because the author takes quarterly data, calculating the cycle between years is inevitable, and may be one of the causes of autocorrelation.

- The lag phenomenon is also one of the causes of autocorrelation. For example, when studying the relationship between consumption and income, we see that consumption in the current period depends not only on current income but also on consumption in the previous period. In the model, the author uses both lagged independent variables and timely independent variables. However, the use of lagged independent variables has been tested by empirical studies in the world. So autocorrelation due to lag may not be the cause of the above model.

- Modeling errors: This is a cause of modeling. There are two types of errors that can cause autocorrelation. One is: not including enough variables in the model. Two is: the wrong functional form can cause autocorrelation. This is not the cause of autocorrelation in the above model because the way of taking data of independent variables has been verified through empirical studies in the world. Therefore, the cause of autocorrelation arising from modeling errors is unlikely to occur.

For the above reasons, the author proposes two solutions to overcome autocorrelation for the above model. One is to use the 2-step Durbin-Watson method to overcome

autocorrelation. Second, instead of quarterly statistics which can easily lead to autocorrelation, the author is forced to calculate statistics by year even though the statistical sample is not large.

First, the author used the 2-step Durbin-Watson method to overcome autocorrelation on EVIEWS software, and obtained the following results:

Table 4.8: Regression results of financial leverage of 8 listed commercial banks after adjustment using the Durbin – Watson method.

Looking at the results of the table above, we see that the autocorrelation phenomenon (d=2.091359) has been overcome, however, the coefficient of determination of the model R2 has also decreased to only 61%. And after overcoming the autocorrelation phenomenon in the above way, there is only the variable PRO

statistically significant, the remaining variables are not significant in explaining the fluctuations in financial leverage. So the above model is also not really meaningful after correcting for autocorrelation.

Then, the author uses data from the annual financial reports of commercial banks instead of the quarterly data used in the above section. When taking annual data, the MTB variable and the RISK variable have a correlation coefficient of 0.78, so the author removes the RISK variable and only uses the MTB variable.

Similar to the above, the best results were obtained when the author used fixed effects for the year data (after removing the RISK variable due to its high correlation with the MTB variable) and obtained the following results:

Table 4.9: Regression results of financial leverage of 8 listed commercial banks after adjustment using annual data.

Looking at the above results, we can see that the coefficient of determination of the model R 2 is still relatively high (88%), the F tests also have a significance level Prob close to 0.