4. Results
4.1. Linear regression
First, a univariate regression of leverage is conducted to see the own R² of each variable and the results are presented in Table 6. It can be seen that profitability has the highest explanatory power of the variables included with R² of 0.081 and standardized coefficient of -0.284. Although profitability was also found to have the highest
explanatory power in the research by Chang et al. (2014), the results are significantly different, as in their study profitability was found to have extremely high R² of 0.256 with a coefficient of -1.09. The effects of profitability on leverage found in this study are much
Variable Standardized
Coefficient
R² t-statistic
Profitability -0.284 0.081 -52.086
Asset Tangibility 0.265 0.070 48.239
Firm Size 0.213 0.045 38.374
Ind. Med. Leverage 0.107 0.011 18.952
DRSoCG -0.080 0.006 -14.076
Quarterly Exports -0.078 0.006 -13.842
Money Supply -0.076 0.006 -13.364
FDI -0.058 0.003 -10.179
PMI 0.040 0.002 7.108
Largest Shareholder 0.014 0.000 2.476
Asset Growth 0.010 0.000 1.753
SOE -0.008 0.000 -1.380
Table 6 Results of univariate regressions of leverage.
22
closer to those reported for developing countries by Fan, Titman, & Twite (2012), where the profitability coefficient equals -0.2268. The variable with second highest R² and coefficient is asset tangibility with R² of 0.07 and coefficient of 0.265 and it is followed by firm size with R² of 0.045 and coefficient of 0.213 and industry median leverage as the fourth variable with R² of 0.011 and coefficient of 0.107. These results are also quite different from the previous research focusing on China, where industry median leverage was the second most crucial variable with R² of 0.079 and coefficient of 0.716, followed by asset growth in the third place with R² of 0.007 and coefficient of 0.082. In this research, however, the R² of asset growth equals 0 and has the second smallest coefficient of 0.010, only higher than SOE. Both, SOE and the largest shareholder also have lower explanatory power than before. On the other hand, firm size seems to be much more important compared to before when its R² was 0.001 and coefficient 0.019.
The remaining variables were not used before and relatively to the top three factors seem to have relatively small explanatory power, with R² ranging from 0.002 to 0.006.
Next, a few models comprising different groups of variables are going to be
compared. The first of the models is going to use only firm-specific factors, the second one is going to use firm and industry-specific factors, the third model is going to include industry-specific and macroeconomic factors, the fourth one is going to include all of the factors and the fifth one is going to be developed by stepwise regression. The results of these linear regressions, including R² of each model and the coefficient and significance of each of the variables in each model, can be seen in the Table 7.
The first model includes profitability, firm size, asset tangibility, asset growth, state control dummy and the largest shareholding as the independent variables and has R² of 0.199, meaning the model including only these variables explains around 19.9% of the
23
variation within leverage. All of the variables are significant at the 0.01 level, but the three variables with by far the highest standardized coefficients are profitability with -.304, firm size with .267 and asset tangibility with .219, followed by largest shareholding with only -.044, asset growth with .035 and state control dummy with .-020.
The second model, which only adds a single variable to the first model, industry median leverage, has R² higher by .021 and results in R² of .221. The industry median leverage itself has coefficient of .148, but adding the variable also affected the
Model 1 Model 2 Model 3 Model 4 Model 5
R² 0.200 0.221 0.013 0.227 0.227
Coeff. Sig. Coeff. Sig. Coeff. Sig. Coeff. Sig. Coeff. Sig.
Profitability -.304 .000 -.293 .000 - - -.306 .000 -.306 .000 Firm Size .267 .000 .268 .000 - - .278 .000 .278 .000 Asset Tang. .219 .000 .249 .000 - - .243 .000 .243 .000
Asset Grwth
.035 .000 .035 .000 - - .034 .000 .034 .000
SOE -.020 .000 -.025 .000 - - -.029 .000 -.029 .000 Lrgst.
Shrhld
-.044 .000 -.043 .000 - - -.044 .000 -.044 .000 IndMedLev - - .148 .000 .092 .000 .133 .000 .133 .000
DRSoCG - - - - -.119 .000 -.047 .096 -
-Money Sup. - - - - .069 .013 -.138 .000 -.138 .000
PMI - - - - -.011 .118 .001 .894 -
-Qtr. Exports - - - - .006 .676 .143 .000 .143 .000
FDI - - - - .004 .545 .007 .315 -
-Table 7 Comparison of linear regression results for different models.
24
coefficients of the firm specific factors, notably reducing the coefficient of profitability by .011 and increasing the coefficient of asset tangibility by .03 to .249.
The third model includes industry median leverage, domestic retail sales of consumer goods, money supply, PMI, quarterly exports and FDI as the independent variables and has R² of 0.013, explaining only around 1.3% of the variation within leverage. This already tells us that the firm-related factors are much more important when predicting the leverage levels, but the macroeconomic factors might still improve the predictions. In this model DRSoCG and industry median leverage are significant at the 0.01 level, money supply is significant at the 0.05 level and PMI, quarterly exports and FDI are insignificant.
DRSoCG has a coefficient of -.119, industry median leverage of .092 and money supply of .069.
Model 4 includes all of the variables and its value of R² is 0.227 and thus explains around 22.7% of the variation within the leverage. This result is much lower than the explanatory power of 36% of the variation in leverage in the model using seven core factors in research basing on the data of Chinese listed companies from 1998 to 2009.
The main reason for this is the big difference in explanatory power of profitability during that period and years 2009-2014. However, it is still higher than the R² reported in the research based on companies listed in the United States by Frank & Goyal (2009), where its value equaled 0.192. Profitability, asset tangibility and firm size are the three factors with the highest coefficient, same as in Model 1, but the coefficients are slightly different and the significance of the macroeconomic factors has changed. In this case DRSoCG, PMI and FDI are all insignificant and the rest of the factors is significant at the 0.01 level, including quarterly exports which was insignificant and money supply which was not significant on this level in Model 3. The coefficient of profitability is -.306, of firm
25
size .278 and asset tangibility .243, all slightly higher than in the first model. They are followed by three factors with coefficients of similar size: quarterly exports with .143, money supply with -.138 and industry median leverage with .133, the next highest coefficient is of DRSoCG, but it was found to be insignificant. The coefficients of the last three significant factors are -.044 for the largest shareholding, .034 for asset growth and -.029 for SOE. As it can be seen, money supply has negative coefficient, which suggests companies do not borrow more money during high inflation in order to shield their profits from tax and instead reduce the leverage levels. It looks like the potential to issue
Table 8 Models of the stepwise regression.
Model Summaryj
a. Model 1. Predictors: (Constant), Profitability
b. Model 2. Predictors: (Constant), Profitability, FirmSize
c. Model 3. Predictors: (Constant), Profitability, FirmSize, AssetTangibility
d. Model 4. Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev
e. Model 5. Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev, Largest Shareholder
f. Model 6.Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev, Largest Shareholder, Money Supply g. Model 7. Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev, Largest Shareholder, Money Supply, Quaterly Exports
h. Model 8. Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev, Largest Shareholder, Money Supply, Quaterly Exports, AsstGrwth
i. Model 9. Predictors: (Constant), Profitability, FirmSize, AssetTangibility, IndMedLev, Largest Shareholder, Money Supply, Quaterly Exports, AsstGrwth, State Control dummy
j. Dependent Variable: Leverage
26
equity at higher share price that comes with increased money supply pumped into companies’ stocks outweighs the potential benefits of bigger tax shield.
Model 5 used stepwise regression, a method in which variables are being added one by one and the regression is being rerun on every step to check if there is a significant change to the model. If the next variable added to the model does not improve the model significantly, then the variable is excluded from the model in order to avoid over-fitting. It can be seen that the results are the same as those for Model 4, but DRSoCG, FDI and PMI were not included at all in the final model, which further proves that these factors do not improve the regression model in any way, even though DRSoCG had the highest explanatory power when used in Model 3 which included only the
macroeconomic and industry factors. The results for models created at each step of the stepwise regression can be seen in Table 8. It can be easily shown that as more factors are added, the marginal contribution to R² is diminishing, as the model with three factors explains 19.7% of variation within leverage, while adding the next six factors adds only 2.3% of explanatory power.