Table 7 presents regression results based on the fixed-effects model for each of the four leverage measures, including regression coefficients for each explanatory variable, their corresponding t-value, the level of statistical significance of the coefficient, and within R2 value for each regression model.
All FEM models presented in Table 7 are statistically significant with Prob > F = 0.000.
Table 8 shows regression results of the random-effects model for each of the four leverage measures, including regression coefficients for each explanatory variable, their corresponding z-value, level of statistical significance of the coefficient, and overall R2 value for each regression model.
All REM models presented in Table 8 are statistically significant with Prob >
chi2 = 0.000.
Table 7: Fixed-effects regression results
FACTORS TDB TDM LDB SDB
ROA -0.00317*
(-2.09)
-0.0190***
(-15.07)
-0.00304*
(-2.38)
-0.000140 (-0.51)
SIZE 0.128***
(4.10)
0.136***
(5.29)
0.123***
(4.71)
0.00415 (0.74)
GROWTH 0.000249 (0.01 )
-0.0226 (-1.08)
-0.00291 -0.14
0.00295 (0.69 )
TANG 0.00333 (0.04)
0.276***
(3.78)
0.0226 (0.31)
-0.0202 (-1.29)
NDTS 1.170
(0.41)
3.108 (3.78)
1.356 (0.56)
-0.155 (-0.30)
RISK .000179
(0.05)
.00138 (0.48)
.000624 (0.21)
-.000477 (-0.77 ) _CONS -.6761197*
(-2.50)
-.8005491***
(-3.57)
-.6919881**
(-3.04)
.0206899 (0.42 ) R-sq – overall 0.0067 0.0687 0.0074 0.0065 Prob > F 0.0006 0.0000 0.0000 0.0000
25
***Significant at 0.01; **Significant at 0.05; *Significant at 0.1 Table 8: Random effects regression results
FACTORS TDB TDM LDB SDB
ROA -0.00217*
(-2.04)
-0.0211***
(-18.45)
-0.00218*
(-2.29)
-0.000293 (-1.44)
SIZE 0.0487***
(9.50)
0.109***
(11.26)
0.0434***
(8.83)
0.00713***
(6.65)
GROWTH 0.00824 (0.36)
-0.0267 (-1.29)
0.00142 (0.07)
0.00409 (1.00)
TANG 0.0794***
(3.42)
0.226***
(5.63)
0.0887***
(4.00)
-0.0140**
(-2.87)
NDTS -0.690
(-0.60)
-1.578 (-0.87)
0.188 (0.17)
-0.725**
(-2.95)
RISK -0.00253 (-0.82)
0.0000508 (0.02)
-0.00114 (-0.43)
-0.00107 (-1.90)
_CONS -0.0650 (-1.44)
-0.481***
(-5.79)
-0.0773 (-1.79)
0.0000920 (0.01) R-sq – overall 0.0075 0.0761 0.0088 0.0076
Prob > F 0.0006 0.0000 0.0000 0.0000
Now, the Hausman test is used to find which methods between fixed- effects model and random-effects model, works better for the data set in this study.
H0: difference in coefficients not systematic
Table 9: Hausman test results
TDB TDM LDB SDB
Chi2 15.39 24.13 16.30 10.22
Prob > Chi2 0.0174 0.0005 0.0122 0.1157 Conclusion Reject Ho Reject Ho Reject Ho Fail to reject Ho
Selected model FEM FEM FEM RAM
26 It can be seen from the table 9 of Hausman test results. For TDB, TDM, LDB model, the Prob>Chi2 are smaller than 0.05, so null hypothesis is rejected. As a result, Fixed effects model will be used. For SDB model, the Pro- Chi2 is bigger than 0.05 so the Random effects model will be used for further examination.
With the TDB model, it can be seen that the overall R-squared of TDB model is 0.0075. This means that the model with explanatory variables, including profitability, firm size, growth, tangibility, and non-debt tax shield and business risk explains 0.67 percent of the variability of capital structure measured by Total Debt to Book value of Capital.
For the TDM model, the overall R-squared is 0.0687. It indicates that the model with selected firm’s characteristics, including profitability, firm size, growth, tangibility, and non-debt tax shield and business risk explains 6.8 percent of changes in leverage, as measured by Total Debt to Market value of Capital.
For the LDB model, the overall R-squared is 0.0074. It indicates that the model with selected firm’s characteristics, including profitability, firm size, growth, tangibility, and non-debt tax shield and business risk explains 0.74 percent of changes in the portion of long-term debt using in total capital, as measured by Long-term Debt to Book value of Capital.
For the TDM model using random effects, the overall R-squared is 0.0076. It indicates that the model with selected firm’s characteristics, including profitability, firm size, growth, tangibility, and non-debt tax shield and business risk explains 0.76 changes in the portion of short-term debt using in total capital, as measured by Short-term Debt to Book value of Capital.
To explain better influences of factors to decisions of capital structures.
All the variables that are not statistically significant are removed from the table of the result. Table 10 shows the chosen model will be analyzed which are fixed effects model for measuring TDB, TDM, LDB models and random effects
27 model for measuring SDB. The table also only keeps statistically variables in the models.
Table 10: Regression with selected factors
FACTORS TDB TDM LDB SDB
ROA -0.00317*
(-2.09)
-0.0190***
(-15.07)
-0.00304*
(-2.38) SIZE 0.128***
(4.10)
0.136***
(5.29)
0.123***
(4.71)
0.00713***
(6.65) GROWTH
TANG 0.276***
(3.78)
-0.0140**
(-2.87)
NDTS -0.725**
(-2.95)
RISK
_CONS -.6761197*
(-2.50)
-.8005491***
(-3.57)
-.6919881**
(-3.04) R-sq –
overall
0.0067 0.0687 0.0074 0.0076
Prob > F 0.0006 0.0000 0.0000 0.0000
Test for heteroskedasticity
The modified Wald test (Greene, 2000) is the used to test for heteroskedasticity of the sample. This test helps detect whether the errors are uncorrelated and normally distributed, and that their variances are constant and do not vary with the effects modelled.
H0: sigma(i)^2 = sigma^2 for all i
Table 11: Modified Wald test results for heteroskedasticity
TDB TDC LDB LDB
chi2 (1286) 8.7e+37 5.3e+35 1.4e+38 1.7e+38
Prob > chi2 0.0000 0.0000 0.0000 0.0000
28 It can be seen from table 11 that p-value is equals 0.0000 for all four leverage models. The test indicates a strong evidence to rejects the null hypothesis. Heteroskedasticity clearly exists in the estimation. The presence of such heteroskedasticity can invalidate the significance of the statistical model.
To deal with this violation, the regression with Driscoll and Kraay standard errors will be used.
Test for autocorrelation
The Wooldridge test is used to test for autocorrelation (Wooldridge, 2002) following the regression results above. Assume that there is no correlation between a time-series with its own past and future values; this test determines if such a correlation exists.
H0: no first order autocorrelation
Table 12: Wooldridge test results for autocorrelation
TDB TDM LDB LDB
F (1,827) 209.350 2109.411 107.850 8.521
Prob > F 0.0000 0.0000 0.0000 0.0036
Since ther p-value equals 0.0000 and one p-value is small, the Wooldridge test rejects the null hypothesis of no autocorrelation. The test indicates a strong evidence to rejects the null hypothesis. Autocorrelation exists in the estimation. To deal with this violation, the regression with Driscoll and Kraay standard errors will be used.
Dealing with the violations
The Driscoll and Kraay (1998) standard errors for coefficients are calculated to avoid estimation biases caused by heteroskedasticity, autocorrelation. The following are results of regression models for the respective leverage measures.
29 Table 133: Fixed-effects regression with Driscoll-Kraay standard errors
FACTORS TDB TDM LDB SDB
ROA -0.00317
(-2.09)
-0.0190***
(-6.46)
-0.00304*
(-2.39)
-0.000140 (-0.49)
SIZE 0.128***
(4.85)
0.136***
(3.99)
0.123***
(5.48)
0.00415 (0.90)
GROWTH 0.000249 (0.05)
-0.0226**
(-3.29)
-0.00291 (-0.58)
0.00295*
(2.53)
TANG 0.00333 (0.06)
0.276***
(6.30)
0.0226 (0.47)
-0.0202 (-1.66)
NDTS 1.170**
(3.20)
3.108**
(3.09)
1.356**
(3.60)
-0.155 (-1.63)
RISK .000179
(0.07)
.00138 (0.30)
.000624 (0.32)
-.000477 (-0.73)
_CONS -0.676**
(-3.37)
-0.801*
2.8)
-0.692***
(-4.06) 0.0207 (0.55)
N 20118 19993 20082 19527
R-sq – within 0.0013 0.0206 0.0017 0.0184 Prob > F 0.0006 0.0000 0.0000 0.0000
Profitability
It can be seen from the regression results that profitability, which is measured as return on assets, has a negative relationship with all of the measures of debt including total debt ratios, long-term debt ratios, short-term debt ratios. The relationship is highly significant for Total Debt to Book value of capital and Long-term debt ratios and moderately significant for Total Debt to Market value of capital. This indicates that when firms are profitable, they tend to borrow less except short-term debt.
This result agrees with pecking-order theory, as proposed by Myers and Majluf (1984) which states that firms make profit prefer retained earnings as a source of capital to external financing. This result agrees with Rajan and Zingales (1995). They also found a negative influence of profitability on
30 leverage U.S firms during 1987-1991. Moreover, not only in the US, the study found that during this time the other G-7 countries except Germany also experienced the same relationship of profitability. Tugba, Gulnur, and Kate when studying samples of 25 emerging market countries from different regions also found an inverse impact of profitability on leverage. The negative relationship between profitability and leverage is also found in a various empirical study such as Frank and Goyal (1995), Nelson, Antonio and Elisio (2015), Titman and Wessels (1988), Joy Pathak (2010) and Han-suck Song (2005).
Size
The result reveals that firm size of the U.S firms moves in the same direction with their debt ratios. The relationship of firm size measured as the log of total assets had significantly positive relationship with all types of leverage. All the β are significant at a level of 1 percent. In general, larger firms will use more debt, both in term of the total debt, short-term debt and long-term debt. Moreover, whether debt ratios base on book value or market value of capital, there is strong evidence of a statistically significant relationship between firm size and leverage
Growth
Growths of firms, as measured by growth in sales, is found to have an insignificant relationship with all measurement of capital structures. The relationship of growth and Total Debt to Book value of capital, Short-term Debt to Book value of capital is positive. The relationship of growth and Total Debt to Market value of capital, Long-term Debt to Book value of capital is negative.
However, all the β are insignificant. This finding does not support that growth is a determinant of capital structure.
This study has found the result different with many studies. Frank and Goyal (2009) while studying publicly traded American firms from 1950 to 2003 found that firms with high market-to-book ratio tend to have low levels of
31 leverage. Rajan and Zingales (1995) in another study of G-7 countries, also found a negative relationship when using market-to-book ratio as the proxy for growth opportunities. Titman and Wessels (1988), who also used realized value of growth as this study, while examining 469 U.S firms from 1974 to 1982 found a negative relationship between growth and leverage.
However, results of this study share the same finding that is an ambiguous relationship between growth and leverage such as Han-Suck Song (2005), Cassar and Holmes (2000). Many studies which also use a change in sales as the proxy of growth also found an insignificant relationship between growth and leverage such as Hall, Hutchinson and Michaelas (2000), Cassar and Holmes (2000), Klapper, Sarria-Allende and Zaidi (2006) and Hall, Hutchinson, and Michaelas (2004). It can be inferred that choosing different measurement of factors can lead to different results. Here, many studies using growth as changes in sales find no significant relationship between leverage and growth while others many studies measuring growth in other ways such as changes in total assets find a significant relationship.
Tangibility
Tangibility, as measured by the percentage of tangible assets on total assets, in general, moves in the same direction with leverage. It can be seen from regression table that tangibility has a positive relationship with Total debt to Book value of the capital structure, Total Debt to Market value of the capital structure, Long-term debt to Book value of capital. Especially, the relationship is significant at 1 percent between tangibility and Total Debt to Market value of the capital structure. This can be explained by trade-off theory, as firms can borrow at lower costs by using tangible assets as collaterals in debt financing – reducing the financial distress costs. This result also agrees with agency theory, since high tangibility helps reduce agency problems by preventing asset substitution.
Tangibility has a positive relationship with long-term debt but has a rather significant negative relationship (at 5 percent) with short-term debt.
These findings is similar to recent studies’ findings such as Bas et al. (2009)
32 which studied small and private firms from 25 developing countries, Koksal et al. (2013) which examined the determinant of Turkish firms and Martina Harc (2015) which investigated determinant of Croatian firms. They both found that tangibility appears to be the key determinants of long-term leverage (positive relationship) but not important for short-term leverage (negative relationship).
This finding can be explained by maturity matching principle which states that firms use long-term debt to financing long-term assets and short-term funds to finance short-term assets.
Non-debt tax shield
The impact of non-debt tax shield on leverage found in this study is ambiguous. Non-debt tax shield moves in the same direction with Total debt to Book value of the capital structure, Total Debt to Market value of the capital structure, Long-term debt to Book value of capital. However, only the relationship of Total Debt to Market value and non-debt tax shield is significant.
This finding, however, still supports DeAngelo and Masulis ‘s model which states that firms with large non-debt tax shield borrow more borrow more.
Risk
This study reveals that business risk has a positive influence on the leverage of American listed firms. The higher the business risk firms have, the more debt is used in firms’ capital structure. This finding does not follow the prediction provided by trade-off theory that firms with more risky earnings should have less debt as they face higher financial distress costs from fixed commitments to debt holders and benefit less from tax shields. Instead, the result confirms the principle of pecking order theory that firms with more risky earnings have higher levels of information asymmetry and therefore have more debt. The time from 2010 to 2015 is the period follows the financial crisis, the business risk for companies as well as the whole country is very high. In response to this, American firms have increased debt financing.
Empirically, this study disagrees with Titman and Wessel (1988), Joy Pathak (2010), Frank and Goyal (2009) about the direction of the relationship between leverage and business risk of U.S companies. These two studies
33 concluded that business risk inversely affects leverage. However, the findings of this study agree with Huang and Song (2002), and especially, Pandey (2001) on the positive relationship between leverage and business risk.