«Three essays on corporate boards R. Øystein Strøm A dissertation submitted to BI Norwegian School of Management for the degree of Dr.Oecon SERIES ...»
The theory of corporate governance cannot yet offer a comprehensive system of well-speciﬁed board design equations. Therefore, we analyze endogeneity by ﬁrst studying what happens to the relationship between current performance and current board mechanisms when we include past performance as an additional determinant of current performance. Next, letting insider holdings, board independence, director network, gender mix, and board size be alternative dependent variables in addition to contemporaneous performance, we relate these board mechanisms to each other and to contemporaneous performance. Since our panel data gives repeated observations of the relationship between ﬁrm performance and governance mechanisms, utilizing the panel structure increases the possibility of revealing stable relationships. Also, since our panel data allows for estimation with ﬁxed effects, we need no instruments to control for unobserved ﬁrm characteristics that are stable over time.
2.3 Descriptive statistics
Our sample is all non-ﬁnancial ﬁrms listed on the Oslo Stock Exchange (OSE) at year-end at least once over the period 1989–200211. To reduce censoring bias in the tenure measures, we start collecting director data in
1986. The ownership structure data covers every equity holding by every investor in every sample ﬁrm.12 Table 2.2 summarizes key properties of the frequency distributions for each board design mechanism. It shows that ofﬁcers as a group hold on average 6.4% of the equity, and the CEO owns 3.6%. These ﬁgures show 11 The OSE had an aggregate market capitalization of 68 bill. USD equivalents by yearend 2002, ranking the OSE sixteenth among the twenty–two European stock exchanges for which comparable data is available. During our sample period, the number of ﬁrms listed increased from 129 to 203, market capitalization grew by 8% per annum, and market liquidity, measured as transaction value over market value, increased from 52% in 1989 to 72% in 2002 (sources: www.ose.no and www.fibv.com).
12 The public securities register (VPS) provided the ownership data, accounting and share price data is from the OSE, and board data was collected manually from Kierulf’s Håndbok and a public electronic register.
CHAPTER 2. ALIGNED, INFORMED, AND DECISIVEthat powerful owners are mostly absent as inside monitors13. The three largest owners as a group have on average simple majority. The average largest outside owner has less than one third of the equity, which means he cannot alone block a charter amendment. This pattern reﬂects that the ownership concentration of Norwegian ﬁrms is low by European standards14. The key implication in our setting is that the resulting separation between ownership and control makes the board a potentially important vehicle for reducing agency costs.
The average value of the independence proxy as deﬁned in expression (2.1) is -0.301, reﬂecting that the average CEO has slightly longer tenure than the ﬁrm’s average director. This ﬁgure also follows from the difference between the separate tenure ﬁgures reported for these two insider types, which are 2.2 and 1.9 years, respectively. Still, the large difference between the extreme values of the tenure variables and the high standard deviation of the independence proxy reﬂect considerable cross-sectional variation in (2.1), which is necessary to validly test the independence hypothesis. For instance, the average director took ofﬁce almost 13 years before the CEO in the strongest independence case and more than 10 years after in the strongest dependence case.
As for the board’s information function, the CEO is not a director in the ﬁrm in 70% of the cases. Every third CEO sits on another listed ﬁrm’s board (exported CEO), but the median CEO has no outside directorships.
13 Although not shown in the table, more than 40% of the CEOs do not own shares in the ﬁrm they run. The average holdings when the CEO (the directors) does (do) own is 6% (13%). Neither the directors nor the CEO holds equity in 36% of the ﬁrms, whereas both do in 44% of the cases, when their average aggregate holding is 20%. Because inside ownership increases the directors’ incentives to monitor the CEO, it also reduces outside owners’ need to monitor the board. Thus, unlike what would be expected from an agency logic, the observed pattern suggests that the two insider ownership characteristics are used as complementary rather than substitute ways of reducing agency costs. This may reﬂect a tendency to either overinvest or underinvest in these two alignment mechanisms.
14 Norwegian ﬁrms have a less concentrated ownership structure than in any other European country except the UK. For example, the average largest owner holds close to 50% of voting equity in a continental-European listed ﬁrm, and 15% in the UK. The corresponding US ﬁgure is 3% (Barca and Becht, 2001). Norway has a civil law regime, which is generally considered less investor–protective than common law. Nevertheless, La Porta et al.
(2000) ﬁnd that Norway’s regulatory environment provides better protection of shareholder rights than the average common law country. According to their theory of institutionally determined ownership structures, the strong investor protection is a major reason why Norway’s ownership concentration is so low.
2.3. DESCRIPTIVE STATISTICS 37 Although not reported in the table, it turns out that a CEO sits considerably more often on other ﬁrms’ boards when he is also a director in the ﬁrm he runs (31%) than otherwise (21%). Thus, a potentially problematic principal-agent relationship inside the ﬁrm (the agent monitors himself) may make the CEO create the same problem in other ﬁrms (one agent by profession monitors another agent by profession).
The director network measure reﬂects that more direct and indirect links to other boards makes the ﬁrm better connected to key parts of the information network. For instance, we ﬁnd that 66% of the sample ﬁrms in 1997 had at least one direct link to another ﬁrm through overlapping directorships. The mean score on the network variable in table 2.2 is 0.184, varying between 0.069 and 0.320.
The third section of the table, which deals with mechanisms for inﬂuencing the board’s decisiveness, reports summary statistics for board size, gender, age, and employee directors. Because employee directors may behave differently than other directors, we measure board size both with and without employee directors (SizeAll and Size, respectively). The average board has six members, and one less if we ignore employee directors. This is a very small board by international standards15. The average fraction of women is 4.7% (GenderAll), dropping to 3.4% if we exclude employee directors (Gender). Although not shown in the table, we ﬁnd that employees elect women considerably more often than the owners (15% vs 3%, respectively). This may suggest that the fraction of women in the workforce is considerably higher than the fraction of women considered qualiﬁed for owner-elected directorships. The proportion of female directors increases with board size, and the substitution of male directors by females for given board size occurs over the whole sample period and is particularly strong after 1995. The fraction of female directors is roughly three times higher in the end of the sample period than in the middle.
Like gender, age is a potential source of board diversity. The average CEO is 47 years old and roughly three years younger than the average director. Average age per board varies between 27 and 74 years, and the standard deviation of director age per board is eight years on average,
15 Wymeersch (1998, p. 1105-1109) reports an average board size of 10.07 in the UK, 12.05
in France, 10.44 in Belgium, 12.00 in Italy, and 6.54 in the Netherlands. The average size of the German supervisory board is 13.25 (Hopt, 1998, p. 248). Carter and Lorsch (2004) ﬁnd that the average US board has about 12 directors, which is down from 16 in the 1980s.
Although the largest boards in our sample become less common over time, the average size is quite stable. For instance, the 25% largest boards have on average 8.97 members in the ﬁrst half of the sample period and 8.67 in the second.
CHAPTER 2. ALIGNED, INFORMED, AND DECISIVEvarying between zero (every director has the same age) and 22 years. As for employee directors, there is about one per board on average when we consider all boards regardless of whether or not they have employee directors. There is at least one employee director in 42% of the ﬁrms, declining from a typical value of 50% in the ﬁrst half of the sample period to less than 40% in the second. This decline may be due to a higher proportion of ﬁrms in exempted industries, a relative increase in the fraction of small ﬁrms, or a larger proportion of ﬁrms organized as holding companies. When employees are represented, they have between one and four seats.
We measure performance by Tobin’s Q and operationalize it as the market value of assets per unit book value. The market value of debt is set equal to its book value. Since we will later regress Q on board characteristics while controlling for ﬁrm size, we use sales rather than assets to measure ﬁrm size16.
Summarizing the descriptive statistics, outside and inside ownership concentration in our sample ﬁrms is low. The board’s average independence of the CEO is medium in the sense that the CEO and the average director have roughly the same tenure. The CEO is a director in the ﬁrm he manages in less than one third of the cases, and those who are sit on other listed ﬁrms’ boards more often than others. The information centrality measure shows that boards differ considerably in their information access through their directors’ links to other boards. The average board has six directors, female directors are rare, average director age per board varies by almost ﬁfty years across the sample, and there is large age heterogeneity within the board. Less than half the ﬁrms have employee directors.
16 Although many bivariate correlation coefﬁcients in table 2.2 differ signiﬁcantly from zero at the 5% level, a rule of thumb says the coefﬁcient must exceed 0.7 before multicollinearity causes problems in regressions. Moreover, Hsiao (2003, p. 3-4) argues that multicollinearity problems are unlikely in panel data settings, since this normally involves more data points and larger data variability than a cross-section. Also, our regressions will use deﬁnitions of size and gender that exclude employee directors. To illustrate, the Pearson correlation between board size and the fraction of employee directors is 0.65 when employee directors are included in the size measure (SizeAll), dropping to 0.07 when the size measure ignores employee directors (Size). This suggests multicollinearity is not a potential problem in our regressions unless employee directors are included in the size and gender proxies.
2.4. STATISTICAL TESTS 39
2.4 Statistical tests
Here, i is the ﬁrm, t is the time period, θ is a constant, β and γ are the coefﬁcient vectors for board mechanisms and controls, respectively, ci is the unobserved, time-independent and random ﬁxed effect of ﬁrm i, and vit is the idiosyncratic error, which varies randomly across ﬁrms and time periods. We observe Qit and the explanatory variables representing governance mechanisms and controls, and want to estimate β and γ while holding the unobserved individual effect ci constant. The error term vit is assumed to be uncorrelated with the explanatory variables and ci.
Since the unobserved ci is constant over time per ﬁrm, the term disappears when we time-demean the variables. Although this FE approach handles unobserved time-independent ﬁrm heterogeneity by eliminating it from the data before estimation starts, we include ﬁrm size and risk to control for observed ﬁrm heterogeneity which varies over time.
We will apply the FE approach in several settings. The basic FE model (2.3) as speciﬁed in (2.2) is estimated in section 2.4.1. Section 2.4.2 analyzes endogeneity explicitly by ﬁrst estimating a dynamic FE model where lagged performance allows for feedback from past performance to current board mechanisms. Subsequently, we estimate a FE model with six equations where the dependent variable is performance, directors’ holdings, independence, director network, gender mix, and board size, respectively.
As discussed earlier, we prefer GMM to alternative estimation methods
17 If we ignore the time-series nature of the data and instead use a pooled cross-
section approach, we would disregard possible correlation between observable and nonobservable variables in general and unobserved heterogeneity between ﬁrms in particular.
The pooled approach may still try to capture ﬁrm heterogeneity by adding control variables such as the ﬁrm’s age and industry, but this is normally insufﬁcient to control for ﬁxed effects. For instance, two shipping ﬁrms founded in the same year may still have different optimal board design mechanisms if the ﬁrms have different exogenous characteristics such as location of headquarters, age of the ﬂeet, and stage in their life-cycle. A further advantage of panel methods is that the moments needed for GMM estimation are readily available from the data structure. This property becomes particularly important when analyzing mechanism endogeneity in section 2.4.2.