«Three essays on corporate boards R. Øystein Strøm A dissertation submitted to BI Norwegian School of Management for the degree of Dr.Oecon SERIES ...»
2.4.1 The basic model Table 2.3 shows the results of estimating model (2.2). The ﬁrst column reports unstandardized (regular) coefﬁcient estimates, and the second shows the estimates based on the standardized variables. The p-value for statistical signiﬁcance in the third column is identical for both coefﬁcient types, but the standardized coefﬁcient expresses economic signiﬁcance in a more transparent way. Because the standardized variable has an expected value of zero and a standard deviation of one, its regression coefﬁcient shows by how many standard deviations performance is expected to change if the board mechanism changes by one standard deviation. Thus, the higher the absolute value of the standardized coefﬁcient, the stronger the economic signiﬁcance of the board mechanism. We only report standardized coefﬁcients in the following.
The Hansen J statistic shows that the instruments used to identify the coefﬁcients are relevant, and the overidentiﬁcation test statistic reﬂects that the instruments are uncorrelated with the error term. We limit the attention to estimated coefﬁcients with a p-value of 10% or less18.
For the alignment mechanisms, there is a positive, signiﬁcant relationship between performance and insider ownership. This is consistent with the extant governance literature, although the familiar, negative sign on the squared insider holdings is not statistically signiﬁcant. Also, the insigniﬁcant effect of outside ownership concentration is in line with several studies in the board literature, which often ﬁnd that when more board mechanisms than just ownership are included in a regression model, the 18 The assumption underlying the GMM model in the table is that all board mechanisms and control variables are strictly exogenous. That is, E (vit |Xi1,..., XiT, ci ) = 0 when t = 1,..., T. These are the moment conditions, from which instruments may be constructed to identify the coefﬁcients. We use the Amemiya and MaCurdy (1986) procedure, which involves the raw, the time-demeaned, and the squared time-demeaned explanatory variables. Furthermore, we include the Breusch et al. (1989) instruments, which are the average and standard deviation of ﬁrm-demeaned explanatory variables. Our choice of instruments illustrates the advantage of panel data that modiﬁed versions of variables included in the model can also be used as instruments. Variables not included in the model, such as CEO age and chairman tenure, are not used as instruments.
2.4. STATISTICAL TESTS 41 signiﬁcant relationship between ownership and performance becomes weaker.
The insigniﬁcant relationship between board independence and performance is consistent with the hypothesis that although more independence increases monitoring incentives, it reduces the CEO’s willingness to share information. The net effect in our sample is zero, suggesting that most boards have optimal independence, reﬂecting that most boards strike the proper balance between being a hands-off monitor and a handson management resource. This ﬁnding lends no support to the argument that value creation will improve when board independence is made mandatory by law or strongly recommended by code. The same conclusion follows from board research in other institutional regimes based on different independence proxies. These studies ﬁnd almost without exception that the relationship between independence and performance is negative or zero.
The information centrality measure, which reﬂects direct and indirect information links created when the ﬁrm’s directors meet directors on other boards, is supposed to pick up information sources for the board with beneﬁcial economic effects. According to the table, it does so in terms of a positive association between director network and performance.
Every coefﬁcient estimate under board decisiveness is negative and signiﬁcant except for age dispersion. Although the inverse relationship between board size and performance is in line with the existing literature, it is remarkable that this pattern turns up in our sample as well, which has very small boards by international standards. This result suggests that optimal board size is indeed very moderate. Adding the ﬁnding that gender diversity is inversely associated with performance, it seems that the homogeneous, small board is superior to the heterogeneous, large one.
Finally, the use of employee directors is negatively associated with performance. This ﬁnding supports the theoretical arguments and also the empirical ﬁndings from Germany and Canada that employee directors successfully defend their interests at the expense of owners and creditors. It also shows that from the capital providers’ point of view, mandating employee directors causes an over-optimal use of them.
In terms of economic signiﬁcance, the standardized coefﬁcients show that among the estimates with a p-value of 10% or less, insider ownership is the most powerful variable, followed by network, employee directors, size, and gender. To illustrate, table 2.2 shows that the average ﬁrm has a Tobin’s Q of 1.482 and directors’ holdings of 6.4%, the standard deviCHAPTER 2. ALIGNED, INFORMED, AND DECISIVE ations being 1.105 and 19%, respectively. Along with the standardized coefﬁcients from table 2.3, this implies that if insider holdings increase by one standard deviation from its mean value of 6.4% to a new level of 25.4%, expected Tobin’s Q increases from 1.482 to 1.655, i.e., by 12%. Reducing gender diversity by one standard deviation from the sample mean increases expected Q from 1.482 to 1.542, i.e., by 4%19.
We explore endogeneity explicitly by means of two different models, which we call the dynamic performance model and the integrated mechanism model, respectively.
The dynamic performance model rests on the idea that reverse causation between performance and board mechanisms (i.e., performance drives board design) can be partially captured by including lagged performance as a determinant of current performance in (2.2). The required assumption is that lagged performance and the other explanatory variables are predetermined relative to current performance (Arellano, 2003, p. 144)20.
Since this assumption allows for feedback from past performance to current board design, it means that if the dynamic model produces different coefﬁcient estimates for board mechanisms than the basic model in table 2.3, these board mechanisms are at least partially driven by performance.
19 To explore whether panel data estimation is required in our setting, we used OLS to estimate (2.2) on the pooled (i.e., un-demeaned) sample. This approach ignores both individual effects and time effects by assuming that the error term is identical across all ﬁrms and time periods. We found several noticeable differences. First, unlike our panel data model, pooled OLS reproduces the classic result in the ownership structure literature of a positive and quadratic relationship between insider holdings and corporate performance.
Also, outside ownership concentration is inversely related to performance in a signiﬁcant way. Second, the negative exported CEO effect becomes signiﬁcant, and the signiﬁcant coefﬁcient of the network effect is higher. Third, only employee directors is signiﬁcant among the decisiveness mechanisms, but its sign is reversed and its signiﬁcance weaker.
As expected, the importance of the control variables increases considerably, and the R2 is below one ﬁfth. Overall, these ﬁndings show that unless we can ignore the panel structure in our data set, the pooled model is seriously misspeciﬁed. ANOVA analysis of the the pooled OLS regression shows that the pooled model is indeed misspeciﬁed. 54% of the sum of squares in the estimated OLS error term is driven by ﬁxed ﬁrm effects, 4% is due to time effects, and 40% is random. The remaining 2% is driven by joint individual and time effects.
20 Thus, lagged performance may be correlated with the lagged error term, but not with
We specify the dynamic model as:
Qit = θ + αQi,t−1 + β(Board mechanisms)it + γ(Controls)it + ci + vit (2.4) where α is the coefﬁcient of lagged performance and β and γ are the coefﬁcient vectors of the board mechanisms and control variables, respectively.
The standardized coefﬁcient estimates of the dynamic performance model are shown in the ﬁrst column of results in table 2.4. The results strongly support the ﬁndings in the static model in table 2.3, whose estimates are repeated in the second column of results in table 2.4. The only difference is that the linear insider holding term becomes more signiﬁcant both statistically and economically and that its non-linear component becomes statistically signiﬁcant. These results show that endogeneity in terms of reverse causation from past performance matters because past performance inﬂuences the effect of current board mechanisms on current performance. However, this reverse causation from past performance is still moderate, as every result from the base-case model survives.
The second model consists of six equations and is called the integrated mechanisms model in table 2.4. The ﬁrst equation is the static model from table 2.3. Each of the ﬁve other equations have a board mechanism as dependent variable, which is Directors’ holdings, Independence, Network, Gender, and Size, respectively.
The estimates of the integrated mechanisms model reﬂect two-way contemporaneous causation between performance and four of the ﬁve mechanisms. The performance equation shows that boards with high directors’ holdings, networked directors, low gender diversity, and small size produce higher performance. As shown by the directors’ holdings, network, gender, and size equations, respectively, better performance feeds back to these four board mechanisms by producing lower insider holdings, more networked directors, less gender diversity, and reduced board size. Thus, directors with high equity stakes improve the ﬁrm’s expected performance (from the performance equation), but tend to sell off or be replaced by non-owning directors as performance improves (from the directors’ holdings equation). Busy directors improve performance (the performance equation), but are also attracted to well-performing ﬁrms (the network equation). Lower gender diversity reduces performance (the performance equation), but high-performing ﬁrms tend to establish boards with CHAPTER 2. ALIGNED, INFORMED, AND DECISIVE less gender diversity (the gender equation). Finally, smaller boards improve performance (the performance equation), but ﬁrms with high performance tend to reduce board size (the size equation). Overall, these ﬁndings show that improved performance makes boards less aligned, better informed, and more homogenous.
Turning to the internal relationship between board design mechanisms within each of the three groups (i.e., the alignment, information, and decisiveness mechanisms), ﬁve of the six signiﬁcant coefﬁcients are positive.
This means mechanisms are mostly complements rather than substitutes.
Firms with high inside ownership concentration also have high outside ownership concentration (alignment), boards with well networked directors also export their CEO to other boards (information), and boards with low gender diversity are smaller and have less age diversity (decisiveness). The relative economic signiﬁcance of the mechanisms resembles what we found for the determinants of performance: Ownership and network mechanisms are not only the strongest drivers of performance, but also of other board mechanisms.
Four of the ﬁve mechanisms modeled in the table are signiﬁcantly related to either ﬁrm size, ﬁrm risk, or both, which are the two boardexogenous determinants in our model. Large ﬁrms have better networked directors, and risky ﬁrms have lower insider holdings, better networked directors, and lower gender diversity. The negative association between risk and gender mix is consistent with Adams and Ferreira (2004), who ﬁnd that ﬁrms more often choose male directors when uncertainty increases. This is in line with the argument that higher uncertainty makes ﬁrms rely more on the trust inherent in homogenous teams than on the more noisy performance-based incentives in heterogenous teams (?).
The ﬁnding that several board design mechanisms are internally related and driven by board-external variables supports the equilibrium hypothesis of Demsetz and Lehn (1985) that governance mechanisms are optimally adjusted to each other. On the other hand, the equilibrium hypothesis is apparently falsiﬁed by our ﬁnding that several mechanisms are also signiﬁcant variables in a performance equation that controls for endogeneity. We hesitate to conclude this way, since the institutional environment of our sample ﬁrms does not allow owners to freely design the board.
Mandatory employee directors in ﬁrms with more than 200 employees is the most obvious example of such a restriction.
Summarizing, the empirical tests in this section have shown that owners on the board as well as multiple directorships are positively related to
2.5. ROBUSTNESS 45 performance. This is consistent with the hypotheses that directors with high ownership stakes have stronger monitoring and advice incentives than other directors and that well-connected directors create extra value through the network they bring along to the boardroom. In contrast, more diversity produced by larger board size, more gender mix, and more employee directors is always negatively associated with performance. This suggests heterogenous boards are less effective decision makers than homogeneous boards. All these relationships are statistically signiﬁcant, and the economic signiﬁcance is stronger for insider ownership and networked directors than for the decisiveness mechanisms.
Several board design mechanisms are endogenous, both relative to performance and to each other. For instance, directors with wide networks produce high performance and gravitate towards well-performing ﬁrms, and networking declines when gender diversity increases. Moreover, board mechanisms are much more often complements than substitutes.