et of explanatory variables to the firm’s
board structure characteristics. In Model 2 the set of explanatory variables is related to the firm’s
ownership structure. Model 3 jointly considers the firm’s board composition and ownership structure.
All models include industry dummies and a time dummy to control for industry and time
effects. The specifications are as follows:
• Model 1:
CEO compensation = α0 +βiboard structure +φicontrol variables
+λtime dummy + μ1−4industry dummies + ε
• Model 2:
CEO compensation = α0 +γiownership structure +φicontrol variables
+λtime dummy + μ1−4industry dummies + ε
• Model 3:
CEO compensation = α0 +βiboard structure +γiownership structure
+φicontrol variables + λtime dummy
+μ1−4industry dummies + ε
Heteroskedasticity of residuals is addressed by using two robust regression methods: (1) OLS
regression with robust errors based on White’s (1980) heteroskedasticity-consistent estimators, (2)
iteratively reweighted least-squares (IRLS) regression. Unlike OLS, which assigns equal weight
to all observations, IRLS regression involves an iterative procedure that assigns higher weights
to well-behaved observations and lower weights to outliers.15
4. Empirical results
Table 3 presents the regression results of CEO compensation on board and ownership structure
using OLS regression with robust errors. The results in Table 4 are based on IRLS regression. In
each Table, Model 1 analyzes the effects of board structure determinants on CEO compensation.
Model 2 investigates how CEO compensation is related to the firm’s ownership structure. Model 3
jointly considers board composition effects and ownership structure effects. All reported p-values
are based on White’s (1980) heteroskedasticity-consistent standard errors. Coefficients for time
and industry dummies are not presented, as they are not essential to this study.
15 The best-known weighting procedure is from Huber (1981). An alternative is Beaton and Tukey (1974) in which
observations with large residuals are assigned a zero weight, and thus eliminated. Due to each of these procedures’
respective limitations, we use them in combination. Huber’s weights are used in the early iterations, while Beaton and
Tukey’s (1974) weights are used in the later iterations of the regression until convergence is achieved.D. Li et al. / Research in International Business and Finance 21 (2007) 32–49 43
Table 3
Ordinary-least-squares regressions with robust errors for CEO compensation
Dependent variable is CEO
compensation (RMB thousands)
Model 1 Model 2 Model 3
Control variables
Firm size 0.008*** (0.001) 0.010*** (0.000) 0.011*** (0.000)
Firm performance 75.266* (0.051) 59.305 (0.141) 59.631 (0.109)
CEO age −0.466 (0.311) −0.685 (0.131) −0.397 (0.367)
CEO tenure 12.573** (0.014) 9.209* (0.081) 12.265** (0.015)
Board composition
CEO duality 12.497 (0.340) 10.702 (0.386)
Board (of directors) size −1.678 (0.216) −1.387 (0.307)
Supervisory board size −4.748* (0.065) −2.053 (0.423)
Outside directors 1.204** (0.013) 1.853*** (0.002)
Ownership structure
CEO ownership 707.056*** (0.000) 761.120*** (0.000)
Legal person ownership −0.093 (0.495) −0.013 (0.927)
State ownership −0.175 (0.282) −0.090 (0.576)
B shares ownership 5.284** (0.03
本论文由英语论文网提供整理,提供论文代写,英语论文代写,代写论文,代写英语论文,代写留学生论文,代写英文论文,留学生论文代写相关核心关键词搜索。