CHAPTER 8 CONCLUSIONS, IMPLICATIONS AND LIMITATIONS . 239
7.2 T HE RELATIONSHIP BETWEEN CORPORATE GOVERNANCE STRUCTURES AND FIRM
7.2.3.2 Empirical findings from the combined dataset of both markets: The
Testing for endogeneity of the regressors
In this subsection, the endogeneity of the regressors is checked empirically through the use of the DWH test for endogeneity. The test, executed for all the regressors as a group, is under the null hypothesis that the endogenous regressors may actually be treated as exogenous variables (Baum et al., 2007). Test statistics follow a Chi-squared (Chi-sq) distribution with the degrees of freedom equal to eight, which is the number of suspected regressors (laglnq, female, indep_nonexe, dual, lnbsize, ownership, fsize, and lev).
223
Following Schultz et al. (2010), the test is conducted based on the levels equation of firm performance and corporate governance variables in which one-year lagged differences of the regressors are employed as instrumental variables. Year dummies, industry dummies and lnfage are included in the test specification and treated as exogenous variables. It is found that the null hypothesis cannot be accepted at any conventional levels of significance (Chi-sq(8) = 24.03; p = 0.000), thus suggesting that the System GMM model will be superior in terms of consistency when compared with the OLS and FE models.
The validity of the System GMM estimator
The validity of the System GMM estimator is contingent on whether the lagged instrumental variables are exogenous (Roodman, 2009b). For this reason, this subsection checks empirically the validity of the System GMM estimator through the use of the Hansen-J test of over-identification and difference-in-Hansen tests of exogeneity of instrument subsets.
As reported in the last row of Table 7.10, the Hansen-J test yields a p-value of 0.152 confirming that the instruments (as a group) used in the System GMM model are valid. Applying a similar approach to Subsection 5.2.2.2 of Chapter 5, the author also follows Roodman (2009b) and applies the difference-in-Hansen tests of exogeneity to the subsets of System GMM-type instruments and standard instruments. The tests are under the null hypothesis of joint validity of a specific instrument subset. The results reported in Table 7.9 confirm that all the subsets of instruments used in the System GMM model are econometrically exogenous.
224
Table 7.9: Difference-in-Hansen tests of exogeneity of instrument subsets Tested instrument subsets
Test statistics
Degrees of freedom
p- value
Panel A: System GMM-type instruments
Instruments for levels equation as a group 12.45 8 0.132
lnqit-2 and lnqit-3 (for transformed equation) 3.81 2 0.149
Δlnqit-1 (for levels equation) 0.29 1 0.589
Instruments for board structure variables 12.99 8 0.112 Instruments other corporate governance
and control variables 10.58 6 0.102
Panel B: Standard instruments
2009 and 2010 year dummies, and lnfage 0.63 3 0.890
Note: This table presents difference-in-Hansen tests for exogeneity of instrument subsets, under the null hypothesis of joint validity of a specific instrument subset. The variables are as defined in Table 4.6.The test statistics are asymptotically Chi-squared distribution with degrees of freedom equal to the number of suspect instrumental variables (Roodman, 2009b).
GMM instrument subset used for the equation in levels includes one-year lagged differences of firm performance variable; two-year lagged differences of board structure, ownership concentration, and other control variables. GMM instrument subset used for board structure variables includes two- year lagged differences and lag 3 in levels of board structure variables.
GMM instrument subset used for the other corporate governance and control variables includes two-year lagged differences and lag 3 in levels of these variables. The subset of standard instruments for the equation in levels includes 2009 and 2010 year dummies, and lnfage. 2008 and 2011 year dummies are dropped due to collinearity.
Given that the OLS and FE estimates of 𝛼1 (the coefficient on laglnq) tend to be biased in opposite directions when the length of panel is short (Bond, 2002;
Nickell, 1981), a reasonable estimate of 𝛼1should lie between the FE estimate (lower bound) and the OLS estimate (upper bound) (Bond, 2002). It is evident from Table 7.10 that 𝛼1 obtained from the System GMM (0.268) is higher than that obtained from the FE (–0.053), but well below the OLS estimate (0.655). This is consistent with what one would expect, thus suggesting that the System GMM is likely to produce reasonable estimates, at least better than the OLS and FE estimates.
225
Moreover, the Wald chi-squared statistic (218.017) reported in Table 7.10 confirms the overall fit of the System GMM model. Hence, the results from the Hansen-J test, difference-in-Hansen tests, Wald chi-squared test of overall model fit, together with the reasonable estimate of 𝛼1, suggest that the System GMM model appears to be well-specified.
The results from the System GMM model
The results using the System GMM estimator with the Windmeijer (2005) finite- sample correction are reported in column 4 of Table 7.10. The board structure variables have no significant effects on firm performance after controlling for dynamic endogeneity, simultaneity, and unobserved heterogeneity. This finding, obtained from the combined dataset, is in line with recent findings by Pham et al.
(2011); Wintoki et al. (2012), among others.
However, it is important to remember that some board structure variables, as documented in Chapters 5 and 6, do have significant effects on the financial performance of companies in the Vietnamese and Singaporean markets. For example, the relationship between board gender diversity and firm performance is significantly positive for Vietnamese companies but significantly negative for their Singaporean counterparts. Naturally, these contrasting effects will disappear when the combined dataset of both markets is used. The author argues that if the opposing effects of board structure variables on firm performance obtained from the separate country datasets do exist (as reported in Chapters 5 and 6), then it would be plausible to expect that they will be neutralised when the combined dataset is employed (as reported in this chapter).
226
Table 7.10: The relationship between corporate governance structures and performance: Evidence from the combined sample of Singapore and Vietnam
Dependent variable: Tobin's Q ratio [lnq]
Explanatory variables Pooled OLS Fixed-effects GMM
b/(t) b/(t) b/(z)
(2) (3) (4)
Intercept -0.796*** 5.409*** -0.350
(-5.363) (3.947) (-0.311)
One-year lagged Tobin's Q 0.655*** -0.053 0.268***
(24.199) (-1.487) (2.643)
Percentage of female directors (%) 0.001 -0.001 0.005
(1.382) (-0.368) (0.452)
Percentage of independent and/or
non-executive directors (%) 0.001** -0.000 0.000
(2.085) (-0.316) (0.036)
Duality 0.041** 0.161** 0.371
(1.972) (2.260) (1.045)
Board size 0.183*** 0.084 -0.131
(4.464) (0.734) (-0.210)
Ownership concentration (%) 0.001** 0.002** 0.014***
(2.309) (2.383) (2.652)
Firm age -0.034** -0.291*** -0.100
(-2.270) (-4.001) (-0.902)
Firm size -0.002 -0.249*** -0.023
(-0.665) (-3.946) (-0.799)
Leverage (%) 0.001** 0.005** 0.003
(2.558) (2.536) (0.575)
Industry dummies yes no no
Firm fixed-effects no yes yes
Year dummies yes yes yes
Number of observations 1064 1064 1064
R-squared 0.614 0.346
F statistic 67.722*** 29.143***
Wald Chi-squared statistic 218.017***
Number of instruments 21
Number of clusters 363 363
Hansen-J test of over-identification (p-value) 0.152
Note: This table reports the results from estimating equation (4.3). Column 2 reports the results obtained from the OLS method with clustering at the firm level. Column 3 presents the results obtained from the FE method. The estimates gained from the System GMM approach are reported in column 4. Asterisks indicate significance at 10% (*), 5% (**), and 1% (***). The notation is as defined in Table 4.6. The t-statistics of the OLS and FE estimators are reported in parentheses and are based on cluster-robust standard errors corrected for potential heteroskedasticity and serial correlation. The z-statistics of the System GMM model are reported in parentheses and based on Windmeijer-corrected standard errors. Year dummies and industry dummies are unreported.
227
Noticeably, there is a significantly positive relationship between the concentrated ownership variable and firm performance (𝛽 = 0.014), which is consistent with the findings attained from the OLS and FE procedures. Thus, the positive relationship between ownership concentration and performance is robust across different econometric estimation techniques, providing strong support for both hypotheses HVN5 and HSG5 that ownership concentration is positively correlated with firm performance. This finding is generally in agreement with Heugens et al.
(2009); Ma et al. (2010); and Yabei and Izumida (2008), among others. This empirical evidence thus supports the agency perspective that ownership concentration appears to be an effective internal corporate governance strategy that helps to enhance performance.