Abstract: Directors are not one-dimensional. We characterize their skill sets by exploiting Regulation S-K's 2009 requirement that U.S. firms must disclose the experience, qualifications, attributes, or skills that led the nominating committee to choose an individual as a director. We then examine how skills cluster on and across boards. Factor analysis indicates that the main dimension along which boards vary is in the diversity of skills of their directors. We find that firm performance increases when director skill sets exhibit more commonality.
4. Skill diversity and firm performance
Our factor analysis indicates that the diversity of skills on a board is the primary
dimension among which boards of directors vary. Organizational research emphasizes that
diversity of skills might be beneficial in decision-making as it brings greater resources to
problem-solving and could lead to a more complete analysis of an issue (Milliken and
Martins, 1996; O’Reilly and Williams, 1998). However, different personal and professional
backgrounds may lead to different ways in which team members interpret information and to
multiple representations of a problem (Beers et al., 2006; Hambrick, 2007).
Misunderstandings and disagreement can then threaten effective decision-making processes
within multidisciplinary teams. For example, Garlappi, Giammarino, and Lazrak (2017) show
that when directors have heterogeneous priors, boards may underinvest in multi-stage
projects because they anticipate future disagreement. In their model, security issuance can help alleviate the underinvestment problem. Changing board composition may also work.
Murray (1989), Knight et al. (1999), Pelled, Eisenhardt, and Xin (1999), and Simons, Pelled,
and Smith (1999) argue that having common ground among group members can overcome
some of the problems of heterogeneous teams.
Since there may be advantages and disadvantages to having more diversity of skills
on a team, it is an empirical question how director skill diversity relates to performance on
average.
4.1. The relationship between the factors and firm performance
We examine the relation between firm performance and the first factor from both our
ML and IPF factor analysis in Table 5. We regress our proxy for Tobin’s Q on our factors
and a set of controls that are common to governance performance regressions (e.g., Yermack,
1996; Adams and Ferreira, 2009; Faleye, Hoitash, and Hoitash, 2018). As governance
controls we include variables that plausibly relate to both performance and skills. For
example, we expect the number of skills to be positively related to board size and board
independence. As the number of committees increases, firms might also add more directors
with relevant skills to their board. 6 As the diversity literature argues (e.g., Milliken and
Martins, 1996), skill diversity may affect communication, so we include the logarithm of the
number of board meetings.
As firm-level controls, we include the logarithm of assets as a proxy for firm size, the
number of segments as a proxy for diversification, capital expenditures, ROA, volatility, and
the natural logarithm of firm age. We provide the exact definitions of the control variables in
Appendix A. All models include two-digit SIC code industry effects and year fixed effects and the standard errors are corrected for potential heteroskedasticity and clustering at the firm
level.
[ please insert Table 5 here ]
Column 1 of Table 5 shows that the ML diversity of skills factor is negatively related
to the firm’s Tobin’s Q. This relation is robust to controlling for other firm characteristics, as
can be seen in Column 2, and to the use of the IPF factor method, as can be seen in Columns
3 and 4. The coefficients on the firm-level controls are generally consistent with previous
literature. The negative coefficient on board meetings is consistent with Vafeas (1999), for
example.
4.2. Measuring the diversity of skills
Factor analysis is sometimes unappealing because it is difficult to assess the economic
magnitudes of coefficients on factors. It is also difficult to make the arguments necessary for
instrument validity in an instrumental variable (IV) analysis when the endogenous variable is
a factor. Thus, we examine whether the factor has a more intuitive counterpart in the data. An
obvious choice is to simply count the number of skills that are represented on a board. The
typical firm has ten different skills on the board in a given year. In unreported results, we
show that the correlations between the number of skills and the ML and IPF factors are 0.921
and 0.967, respectively. Columns 5 and 6 of Table 5 confirm our finding from the factor
analysis that the number of skills and Tobin’s Q are negatively related. Thus, the number of
skills seems to capture the essential meaning of the factor.7 4.3. Potential reverse causality
While the results from Table 5 suggest that there is a negative correlation between
skill diversity and firm performance, we cannot immediately give this relationship a causal
interpretation because of potential endogeneity problems due to reverse causality. It is
plausible, for example, that underperforming firms look for more skill diversity on their
boards to get different advice. Another potential concern is that underperforming firms
engage in window dressing by making their directors appear more talented than they really
are. These arguments would predict a negative relationship between performance and skills.
On the other hand, it is also possible that poorly performing firms have other concerns and
pay less attention to the new regulation as a result. This argument would predict a positive
relationship between performance and skills. Without a better understanding of how directors
match to firms, it is difficult to sign the bias in the ordinary least squares (OLS) results. We
attempt to formally address this concern in our set-up using an instrumental variable analysis.
We use two instruments whose summary statistics are provided in Appendix D. Since
both instruments are time-invariant, we conduct our IV analysis for the 2010 cross-section
only.
For our first instrument, we exploit the fact that the amendments to Regulation S-K
include a requirement in Item 407(c)(vi) for firms to disclose how they consider diversity in
the director nomination process. Item 407(c) does not specify the type of diversity the
regulation pertains to.
8
Since it was bundled with Item 401(e) concerning disclosure of
director skills, it is plausible that firms interpreted 407(c) as pressure to increase skill diversity on the board. If so, we might expect firms with more time to incorporate Regulation
S-K’s requirements to attempt to increase diversity by appointing new directors to the board.
Fig. 3 provides some evidence consistent with our expectations: the proportion of firms
appointing new directors in a given proxy month is higher the later the month occurs relative
to the passage of Regulation S-K. Thus, we define our instrument to be the number of days
between the day the 2009 amendments to Regulation S-K were passed and the filing of the
firm’s proxy statement in 2010. Based on the evidence in Fig. 3, we expect this instrument to
be correlated with the number of skills on the board.
[ please insert Figure 3 here ]
On the other hand, we believe it is unlikely that the number of days between
Regulation S-K and the proxy filing is correlated with firm performance in 2010, as long as
the proxy filing date does not change in response to poor performance. We collect proxy
filing dates for 2009 and 2010 from the SEC’s Electronic Data Gathering, Analysis, and
Retrieval system (EDGAR) and examine whether there were any changes in the dates. Fig. 4
shows the distribution of changes between the two years. As is evident from the figure, most
changes occur in the -1, 0, +1, day range, which is reasonable if annual meetings are held
close to or on the weekend and firms send their proxy statements out a fixed number of days
before the meeting.9
[ please insert Figure 4 here]
The second instrument is a dummy if a firm is within 70 miles (roughly an hour’s
travel distance away) of an airport hub—an airport that handles over 1% of annual passenger
boardings according to the Federal Aviation Authority
(http://www.faa.gov/airports/planning_capacity/passenger_allcargo_stats/categories/). The
rationale for this instrument is that firms are less constrained in choosing directors when it is
easy for them to attend board meetings and this may lead to an increase in skills on the board.
Of course, distance to the airport may be directly correlated with firm performance because it
may affect firms’ transportation networks. But we believe that to a large extent this effect
should be controlled for by other variables in our regression, for example, firm size,
diversification (i.e., the number of segments), and industry.
Column 7 of Table 5 shows the results of the second stage of the IV regression of the
specification in Column 6. We report the coefficient on the instruments from the first-stage
regression at the bottom of the table. The first-stage coefficients on our instruments have the
expected signs and are statistically significant. However, the Kleibergen-Paap Wald statistic
(7.98) is mid-way between the Stock-Yogo cutoffs for 25% (7.25) and 20% (8.75) maximal
IV size, which suggests the magnitudes of our second-stage coefficients are still biased.10
To gain confidence that the bias does not affect the sign of the coefficient on the
number of skills, we substitute the instruments for the number of skills in the Tobin’s Q
regression in Column 6 of Table 5. Under the assumption that the instruments are exogenous,
the coefficients on the instruments in this reduced form are consistent estimates of the
population coefficient on the number of skills multiplied by the coefficients on the
instruments in the first-stage regression. The coefficients on both instruments in the reduced
form are negative. Since the coefficients on the instruments in the first stage are both positive, we infer that under our assumptions the ―true‖ coefficient on the number of skills is
indeed negative.
In the second-stage IV regression, the coefficient on the number of skills is negative.
The coefficient is also more negative than in the OLS regressions. This suggests that the bias
is positive [see the expression for the OLS bias in, e.g., Adams, Almeida, and Ferreira
(2009)], i.e., poorly performing firms appear to focus on skills rather than seek out greater
skill diversity for their directors. Because the coefficients on the number of skills are negative
in both OLS and IV specifications, we interpret our results as suggestive of a negative causal
effect of skill diversity on performance.
From Column 7, a one standard deviation increase in the number of skills (2.928) is
associated with a 32.26% reduction in Tobin’s Q at the mean. This is clearly too large and
confirms our suspicion that the IV results may be consistent but not unbiased. The economic
magnitude of skills in Column 6 is -2.44%. Since the IV results are more negative than the
OLS results, one way to interpret the economic magnitudes is to take -2.44% as an upper
bound for the effect of the number of skills on performance. Since this effect is arguably
already economically significant, our results suggest that skill diversity is economically
important.
5. Common ground in director skills
We document that diversity is the main dimension along which boards vary with
respect to skill. An important question is what drives the negative relationship between skill
diversity and performance. A potential explanation for this finding is the importance of
having common ground in the boardroom, i.e., the need for directors to share skills in order to
be able to communicate effectively. We examine this potential mechanism in two ways.
Posts
