Does Board Independence Improve Firm Performance?
Outcome of a Quasi-Natural Experiment*
Marc-Oliver Fischer†
Peter L. Swan‡
Australian School of Business, University of New South Wales
First Draft: May 8, 2013
This Draft: November 18, 2013
Abstract
Since 2003 the Australian Securities Exchange Corporate Governance Council (ASX) has required
that all listed firms either adopt a majority of “independent” board members without links either to
management or to substantial shareholders (i.e., 5% or greater shareholding) or “if not, why not”.
While this close to a global standard, it is the opposite to the NYSE and NASDAQ exchanges who
are explicit in stating that significant shareholding is no barrier to independence from management.
Over a period of 9 years to 2011, 551 of our total sample of 969 firms (or 57% of our sample)
responded positively to the recommendation, with the bulk adopting in 2003 and 2004. In this paper
we analyse the effect on firm performance of this quasi-natural experiment. Due to unobservable
heterogeneity and endogeneity concerns and the likelihood that adoption of the majority
independence was not random, we apply a propensity scoring approach within a Difference-inDifference framework to match characteristics of “treated firms”, i.e., those that responded to the
ASX recommendation in 2003 and 2004, to the control group of “untreated” firms, i.e., those that
never adopted the independence majority. Additionally, we adopt a treatment-response approach to
the time-varying effect of treatment utilizing Generalized Propensity Scoring and also a dynamic
system panel GMM approach to account for dynamic endogeneity. All four methods find that
“treatment” resulted in large and statically significant falls in the two market performance measures,
Tobin’s Q and Market-to-Book ratio, and also accounting performance, ROA. Relatively poorly
performing CEOs were far less likely to be replaced. Both CEO pay and director fees increase with
the period of treatment as firm performance deteriorates. Thus it would appear that the ASX
governance recommendation to declare significant shareholders as not independent and have
“independent” directors constitute the board majority has destroyed considerable shareholder wealth.
This destruction is of no discernible benefit to other than executives and fellow board members. We
estimate these losses conservatively at about AUS $69 billion over the period 2003-2011.
JEL Classification: G34, J41, J44, L25.
Key words: Independent directors, Board monitoring, Board characteristics, Board performance.
* We wish to thank David Simmonds and SIRCA for the provision of data and the Australian Research Council (ARC)
for generous financial support. We also wish to thank Australian Security Industry Commission (ASIC) Commissioners
and other participants for comments at the ASIC seminar, September 25, 2013, David Reeb and CAFS Symposium
participants, Bangkok, 8th November, 2013, and Phong Ngo and participants at the FIRN “Art of Finance” Conference,
16 November, 2013.
†
School of Banking and Finance, ASB, UNSW, Sydney NSW 2052, Australia. Email: [email protected]
‡
School of Banking and Finance, ASB, UNSW, Sydney NSW 2052, Australia. Email: [email protected]; tel:
+612-9385-5871.
1
Electronic copy available at: http://ssrn.com/abstract=2312325
1.
INTRODUCTION
Arguably, the most important and controversial corporate governance issue is board
composition. Should corporate boards be made up of “independent” directors with no
material links to either management or substantial shareholders that could create conflict of
interest? Independent directors free of personal associations with senior executives and major
shareholders should be more dispassionate and less biased, especially when evaluating
existing business practices. In extreme form “independence” limits their interest to just their
directorship and thus excludes share ownership altogether. The two remaining board groups,
executives and “Gray” directors that retain links with shareholders or management, are both
more likely to have with interests better aligned with shareholders than are “independent”
non-executive directors that are barred from either being substantial shareholders themselves
or associated with them.1 According to the Australian Securities Exchange’s Corporate
Governance Council’s (hereafter ASX) 2010 amended rules: a director who “is a substantial
shareholder of the company or an officer of, or otherwise associated directly with, a
substantial shareholder of the company” does not qualify as “independent”. Both executive
and “Gray” directors have 24 times the pay-for-performance sensitivity of what the defines as
“independent” directors within an Australian context.2 The Australian Prudential Regulatory
Authority (APRA, 2012, Prudential Standard CPS510) has adopted the precise ASX rules on
1
Under the Australian Security Exchange Corporate Governance Council guidelines introduced in 2003 and
reaffirmed in 2010 a “non-independent”, i.e., “Gray”, non-executive director:
1) is a substantial shareholder of the company, i.e., owning 5% or more, or an officer of, or otherwise associated
directly with, a substantial shareholder of the company;
2) is employed, or has previously been employed in an executive capacity by the company or another group
member, and there has not been a period of at least three years between ceasing such employment and serving
on the board;
3) has within the last three years been a principal of a material professional adviser or a material consultant to
the company or another group member, or an employee materially associated with the service provided;
4) is a material supplier or customer of the company or other group member, or an officer of or otherwise
associated directly or indirectly with a material supplier or customer; or
5) has a material contractual relationship with the company or another group member other than as a director.
2
See Table 13 below.
2
Electronic copy available at: http://ssrn.com/abstract=2312325
“independence” for banks and financial firms without leeway in terms of an “if not, why
not” provision. “Independent” directors that are typically free to pursue their own self-interest
unburdened by concerns for management or especially substantial shareholders, have little
reason to monitor as their personal wealth is largely unaffected if the stock under-performs.3
In contrast to the ASX rules, the New York Stock Exchange (NYSE) and NASDAQ
exchanges take a contrary position to the ASX in recognizing that governance is only
effective if there is incentive alignment between directors and shareholders. For example, the
rules NYSE (2013, 303A.02 Independence Tests) state: “as the concern is independence from
management, the Exchange does not view ownership of even a significant amount of stock,
by itself, as a bar to an independence finding.” Thus neither exchange sets an upper limit on
share ownership, or lack of association with a significant shareholder, as a requirement for
“independence”. Strangely, given that the United Kingdom regards its rules as the globally
most supportive of shareholder rights, the Financial Reporting Council (2012), like the ASX,
excludes representatives of significant shareholders (3% or more) from “independence”
status, but exempts smaller companies from the requirement of a majority of independent
directors, requiring only two such directors.4 Moreover, the board can determine that a
director is “independent” even if one or more of the checklist of factors for consideration are
violated, so long as an explanation is given under the “if not, why not” provisions.
A further requirement is independence from the company in that they must be part-time
with other sources of income. In the interests of good governance and presumably to
encourage directors to act with an independence of mind and to challenge and discipline the
CEO, stock exchanges and regulatory oversight commissions have promoted such board
3
One avenue of potential incentive for non-executive board members is to be invited to serve on multiple boards
but such “busy” directors are not necessarily better monitors of management.
4
The Cadbury Report (1992, para 4.12) which is the forerunner of the FRC guidelines simply requires
“independent” directors to be “independent of management and free from any business or other relationship
which would materially interfere with the exercise of their independent judgement”. Of relevance, no mention
is made of significant shareholders.
3
Electronic copy available at: http://ssrn.com/abstract=2312325
independence globally for more than 20 years. For example, the three major United States
(US) exchanges, NYSE, NASDAQ, and AMEX, have required the board’s audit committee
to be made up entirely of independent directors since December 1999. Following a number of
spectacular bankruptcies, the Cadbury Report (1992) and Smith Report (2003) in the United
Kingdom (UK) and the European Commission (2005) led to the adoption of similar rules in
the UK and Europe. These rules were incorporated into law by the Sarbanes-Oxley Act of
2002 (SOX) in the US with a requirement that members of the firm’s audit committee be
independent of management and not accept “any consulting, advisory or other compensatory
fees”. In 2003, both the NYSE and NASDAQ announced their changing listing requirements
to have a majority independent director presence on corporate boards by 2005 and greater
independence was also required for nominating and compensation committees, in addition to
auditing committees. Likewise, many exchanges in other countries altered their listing
requirements in response to demands for a majority board presence of independent directors
for reasons of increased independence and thus objectivity and transparency.
Regulators seemed to take the view that the case for independence was self-evident and
thus did not require compelling empirical evidence that in any case was lacking. Regulatory
changes requiring board “independence” were prompted by notable corporate bankruptcies
such as the failure of the Robert Maxwell companies and the Bank of Credit and Commerce
International in the UK in 1992, Enron and WorldCom in 2001 and 2002, respectively, and
with assets of $7.8 billion, HIH Insurance Ltd. in Australia in 2001. The fact that both Enron
and WorldCom had majority board independence at the time of failure and that these
directors were not substantial shareholders reveals the paucity of this oft-quoted justification.
While three supposedly “independent” directors of HIH out of a total of eleven on the board,
including the chairperson, were former partners of the company’s auditor, Arthur Andersen,
not one was a substantial shareholder or associate of a substantial shareholder. Despite some
4
director association with the supposed independent auditor, the HIH board met the ASX
requirement of a majority of “independent” directors, as have other notable corporate
scandals, including the near failure of Centro Properties Group Ltd during the GFC period,
and the Saddam Hussein kickbacks with oil for wheat instigated by independent-director
dominated Australian Wheat Board (AWB).
How does one know that these “independent” directors actually act in the interests of the
firm, or at least its owners, namely shareholders, rather than pursue entirely private agendas?
Fama (1980) suggested that there could be “ex-post settling-up” in labor markets. Thus, if a
director gains a reputation as monitor then other lucrative board positions could open up but
potentially as well, a reputation as a weak monitor could also be valuable for certain boards.5
Pursuit of private interests seems particularly likely for independent directors as, almost by
definition, they have small or negligible shareholding or “skin in the game” (e.g., Perry
(2009)) that diminishes any intrinsic incentive to monitor that the independent director may
possess.
Independent directors also by definition have either no prior experience with the firm, or
at least no recent experience. Moreover, many are professional directors with no specific
knowledge or background in the industry and their part-time nature means that acquisition of
such information is difficult and is never likely to be comparable to that of full-time
executives.6 Ravina and Sapienza (2012) provide empirical evidence that the insider trades
of outside directors are less profitable and thus less informed than are the insider trades of
executives with this difference increasing the poorer is the firm’s governance system. Raheja
5
One avenue of potential incentive for non-executive board members is to be invited to serve on multiple boards
but such “busy” directors are not necessarily better monitors of management, especially if multiple
appointments arise from a reputation for loyal support of management. Masulis and Mobbs (2013) provides
evidence that US independent directors do care about their reputation, especially with respect to their most
prestigious board position.
6
The outgoing chief corporate regulator, Australian Securities and Investment Commission (ASIC) chairman,
Tony D'Aloiso (The Australian, March 30, 2011) stated: “Board members are advisers and are not really
involved and don't have the knowledge that management has”.
5
(2005) proposes a theory of board structure in which insiders compete to gain succession as
the CEO by providing information to outsiders. Despite being relatively uninformed about
company affairs, such information asymmetry does not always exempt non-executive
directors from responsibility for company affairs.7
Since full-time executives basically have a monopoly of firm-specific information,
boards dominated by independent directors may well find themselves subservient to
executive directors and thus ineffective as monitors (e.g., Jensen (1993), Adams and Ferriera
(2007), and Harris and Raviv (2008)). This is especially so for large companies often with
large boards consisting almost exclusively of independent directors. The larger the board size,
the less accountable are directors for board decisions. This is a classic free-rider problem.
Large market-dominant firms with large boards subservient to management, unlike more
competitive small firms, are more likely to generate rents that can be extracted by
management. Hence, it would make sense for such firms to be early adopters when regulators
propose that a majority of independents be appointed to the board. Higher pay and perquisites
for management and board members alike could well be the outcome. After all, when stock
price plummets due to poor monitoring, a director with negligible shareholding feels less pain
than does a substantial shareholder.
In 2003 the ASX took, in our view, a more commendable position than did the United
States regulatory counterparts in one respect only. The ASX required all listed firms to either
adopt a majority of independent directors or “if not, why not”8, whereas all United States
listed firms where required by the Securities and Exchange Commission (SEC) to comply.
7
In his Centro Judgement (2011, 18) Middelton J concluded that: “[A] director, whatever his or her background,
has a duty greater than that of simply representing a particular field of experience or expertise. A director is not
relieved of the duty to pay attention to the company’s affairs which might reasonably be expected to attract
inquiry, even outside the area of the director’s expertise.”
8
Under the Australian Security Exchange guidelines introduced in 2003 a “non-independent”, i.e., “Gray”, nonexecutive director:
6
Since SOX came into effect in the United States in 2002 a non-executive director
serving on the audit committee who owns 10% or more of the voting stock is no longer
deemed to be “independent” due to shareholder association. However, any director owning
10% or more would “not be deemed to be or presumed to be an affiliate” and thus lack
independence (SEC (2003). It would dependent on the situation and require investigation. It
needs to be noted that these rules as to independence apply only to members of the audit
committee, if not exempted from the 10% rule, and not generally to all “independent”
directors. The ASXs actions in 2003 followed the lead set by SOX with the difference that
the criteria for “independence” from substantial shareholders was set at the lower limit of 5%
rather than 10% and applied to all supposedly “independent” directors, not just those serving
on audit committees. This difference greatly compounds the problem of governance in
Australia.
The survey of the extensive board literature by Adams, Hermalin and Weisbach (2010)
focuses on the intrinsic endogeneity problem confronting most such studies (see also
Hermalin and Weisbach’s survey (2003)). Hermalin and Weisbach (1998) showed that in
equilibrium poorly performing firms could adopt more independent directors, reversing the
causality relationship. Unless prompted by regulators, boards are free to choose their
composition and size at will, making many findings problematic. Under these circumstances
there is far from being any consensus as to the ideal board composition with some studies
favouring independent board majorities, others the reverse or, more often than not, no
1) is a substantial shareholder of the company, i.e., owning 5% or more, or an officer of, or otherwise associated
directly with, a substantial shareholder of the company;
2) is employed, or has previously been employed in an executive capacity by the company or another group
member, and there has not been a period of at least three years between ceasing such employment and serving
on the board;
3) has within the last three years been a principal of a material professional adviser or a material consultant to
the company or another group member, or an employee materially associated with the service provided;
4) is a material supplier or customer of the company or other group member, or an officer of or otherwise
associated directly or indirectly with a material supplier or customer; or
5) has a material contractual relationship with the company or another group member other than as a director.
7
significant difference in performance. An insignificant relationship between board structure
and performance is consistent with the argument presented in Demsetz (1983) that each firm
faces a unique optimising problem with a great deal of unobservable inter-firm heterogeneity.
In the absence of an exogenous event producing a sizeable shock, such as the one analysed
here, it is difficult to separate out true impacts from unobservable heterogeneity.
One of the few studies of board composition based around a natural experiment that we
are aware of is Guo and Masulis (2013). This study is based around SOX and the subsequent
regulatory changes. It finds that US independent boards in the post-SOX environment are
more likely to force replacement of poorly performing CEOs. The evidence of Duchin,
Matsusaka and Ozbas (2010) based on the same event is more mixed with board performance
improving only when the cost of finding information is low. Chhaochharia and Grinstein
(2009) use the same experiment to argue that independent boards lowers CEO pay but
Guthrie, Sokolowsky and Wan (2012) point to a problem with outliers in the earlier study.
Consistent with our study for the ASX event, they also find that compensation committee
independence gives rise to higher rather than lower CEO pay.
In recent years a number of legal scholars have become critical of the notion of
“independent” directors and how rules have been interpreted. For example, Ringe (2013)
refer to the “dismal failure” of independent directors during the financial crisis and conclude
that they “showed serious deficits in understanding the business they were supposed to
control, and remained passive in addressing structural problems.” Le Mire and Gilligan
(2013) are also critical of the performance of independent directors.
Ours is the first study to examine board independence in Australia utilizing ASX board
structure requirements as a quasi-natural experiment. The present study utilises an array of
methodologies based around natural experiments. These include Difference-in-Difference
8
analysis, propensity scoring, a treatment-response approach to the time-varying effect of
treatment on firm performance, pay levels and incentives, dynamic system panel GMM
regressions, and forced CEO turnover, in an endeavour to overcome a variety of potential
endogeneity problems. These tests are structured around a quasi-natural experiment in which
in the first year 172 mainly large companies, or approximately 19% of the 841 firms in our
sample in that year, responded to the demand in 2003 from ASX to adopt majority board
independence. Firms that responded to the ASX demands over the period 2003-2011 are
dubbed “treated” firms.
We find statistically significant and consistent findings across all methods that majority
board independence reduces firm performance regardless of whether the criterion is
shareholder value in terms of the Tobin’s Q or Market-to-Book ratios, or accounting
performance as measured by the industry-adjusted ROA. We also find that boards once they
become dominated by independent directors are far poorer at replacing a relatively underperforming CEO, pay poorly performing CEOs significantly more, and take home
significantly higher director fees. Increasing length of treatment modelled using generalized
propensity scoring (GPS) continues to worsen performance and raise both CEO and director
pay. As far as we are aware, Campello, Ribas and Wang (2013) is the first study to introduce
GPS to the finance literature. We estimate the dollar cost of these treatment effects over the
period 2003-2011 conservatively at AUS $69 billion, making it one of the most costly and
disastrous regulatory changes ever implemented in Australia by a private regulator.
Table 1 illustrates the annual number of companies that became “treated”, i.e., adopted
a majority of independent directors, over the period 2001 to 2011 that are included in
SIRCA9’s Corporate Governance Database, the number that became “treated” and were
included for at least one year the ASX 200 Index, the number of firms included in the
9
Security Industry Research Centre of Asia-Pacific.
9
database each year and the proportion that are treated. By 2011, a total of 551 firms were
treated as a result of the 2003 ASX directive and with a total of 969 distinct firms included in
the database, ultimately 56.9% are “treated” as a result of ASX intervention and 59.2%
inclusive of firms meeting the requirement prior to 2003. Firms could remain listed on the
ASX without majority board independence since they had the option of explaining why they
did not meet the governance recommendation. Unlike the corresponding SOX law that came
into effect one year prior to the ASX requirement, Australian firms could “opt out”. The fact
that about 40% of the Australian firms in our sample have opted out so far makes it easier for
us to obtain equivalent “untreated” firms, unlike the USA where most firms already had a
majority of independent directors prior to SOX, and most likely due to exchange listing
requirements, making it difficult to obtain suitable control firms in the post-SOX
environment.
<< Insert Table 1 about here>>
1.1. Board Ownership, Executive Director Ownership and Performance
Despite the strong regulatory support for increased board independence, empirical
studies examining the relationship between executive directors and independent directors on
overall firm performance have until present produced rather mixed results and conflicting
evidence. An early study by Pfeffer (1972) suggests a negative relationship between board
proportion constituted by outsiders and firm performance. More recent studies, on the other
hand, do find an inverse relation between the proportion of independent directors and firm
performance (Bhagat and Bolton (2008)) or no robust correlation between a greater
proportion of independent directors and firm performance despite controls for endogeneity
(Hermalin and Weisbach (1991) and Bhagat and Black (2001)). Fracassi and Tate (2012)
provide evidence that powerful CEOs appoint external directors with strong network links to
10
the CEO. Another explanation for the conflicting evidence with respect to the effects of
independent directors on firm performance is that existing studies typically treat these
directors as a homogenous group. However, individual directors almost certainly have
different characteristics in relation to individual experience, professional connections and
expertise (Masulis, Ruzzier, Xiao, and Zhao (2012); Knyazeva, Knyazeva and Raheja (2009);
Guner, Malmendier, and Tate (2008); Kor and Fredrickson (2008); and McDonald, Westphal,
and Graebner (2008)). In an effort to depart from conventional methods to treating
independent directors homogenously as a proxy for board independence, Masulis, Ruzzier,
Xiao and Zhao (2012) introduce heterogeneous independent director characteristics to their
analysis and find evidence in support of directors with relevant industry experience to be
significantly correlated with earnings restatements, cash holdings, CEO pay-for-performance
sensitivity and stock market reactions to directorship appointments. Additionally, Knyazeva,
Knyazeva and Masulis (2013) uses the geographic supply of independent board members as
an instrument to conclude that independence is positive for firm value, operating
performance, CEO turnover, and the proportion of equity-based pay for US companies.
The remainder of this paper is organised as follows. Section 2 describes the data and
methodology. Section 3 presents our main results while Section 4 concludes.
2.
2.1
DATA AND METHODOLOGY
Data
This paper aims at investigating the effects of board characteristics (Board Size and
Composition) on market-based firm performance measures (Tobin’s Q and Market-to-Book)
and accounting performance (Industry-Adjusted ROA) for an extensive set of ASX-listed
Australian companies between 2001 and 2011. The reason for including Market-to-Book as a
performance variable in addition to Tobin’s Q is because ASX-listed stocks are dominated by
11
financial stocks (largely banks) and resource stocks. The exceedingly high leverage of
financials means that there is not a huge variation in Tobin’s Q performance around 1,
whereas Market-to-Book varies a great deal more with respect to firm performance.
The core data in this study is sourced from SIRCAs10 newly released Corporate
Governance Database, in which board data has been reported based on information disclosure
in annual reports of the largest 500 ASX-listed companies between 2001 and 2011. The data
set consists of the largest 500 companies based on market capitalization with June-year-end
financial years and has been back-filled to the base financial year of 2001 when new
companies entered the list in subsequent financial years. The base year of 2001 was selected
as SIRCA’s starting point due to the implementation of new disclosure standards on company
boards as a result of the Corporations Act Section 300, 300A and newly introduced
accounting standards. Overall, the data set population covers corporate board and executive
variables for 9,837 firm-year observations and 969 distinct firms.
To control for company-level characteristics in the subsequent analyses, we obtained
relevant accounting data from the Aspect Huntley database.11 Additionally, we collected
market price data from the AGSM UNSW CRIF (Centre for Research in Finance).
Ownership data was collected from SIRCA and if absent sourced from Morningstar. For
computing risk and daily price performance measures, we used Datastream to collect daily
prices of the given companies.
Table 2 exhibits the descriptive statistics and Table 3, a pairwise correlation matrix of
the board structure, firm characteristics, and firm performance variables included in the
analyses. The average board size is about six members, the overall proportion of independent
directors is approximately 32% and the proportion of “Gray” directors who do not meet the
10
Securities Industry Research Centre of Asia-Pacific located in Sydney.
Relevant accounting data includes cash, ordinary dividends paid, CAPEX, total assets, total liabilities, and
shares outstanding at balance sheet dates.
11
12
ASX independence criterion is 39%. Table 3 indicates a negative association between
Tobins’ Q and board size, the proportion of independent directors and the overall nonexecutive director proportion. Larger firms are associated with larger boards and both larger
boards and the presence of more independent board membership are associated with lower
idiosyncratic risk.
<<Insert Tables 2 and 3 about here>>
2.2
Methodology
The current state of empirical corporate finance research that aims at inferring cause
and effect relationships between proxies of board monitoring performance and firm
performance is faced with the issue of endogeneity. Often it is difficult to find exogenous
factors or natural random experiments in observational data with which to identify the
relationships being examined. In the underlying study, we use an extensive set of six distinct
analytical approaches aiming to address the potential issue of endogeneity and to investigate
the relationship between corporate board characteristics and firm performance using an
extensive data set of the Australian corporate landscape.
First, we investigate the effects of the introduction of the “majority independence
board” criterion as part of the “Corporate Governance Principles and Recommendations”
guidelines released in 2003 by the ASX Corporate Governance Council. In this context, we
employ a Difference-in-Differences (DID) framework combined with propensity score
matching to match treated and untreated ASX listed firms and assess the firm performance
impacts based on Tobin’s Q, Market-to-Book, and Industry-Adjusted ROA as a result of the
“majority board independence” policy introduction in 2003. To recognize a distinct event, the
172 firms that responded in that year and never revert in the period to 2011 are “treated”
whereas firms that have never adopted the recommendation are “untreated”. Similarly, we
13
investigate the 107 firms that delayed treatment until 2004. Second, using a time-varying
“treatment – dosage” effect regression framework, we address the issue of how firms respond
to treatment depending on the precise year in which they enter treatment and the length of the
treatment when matched with hypothetical identical firms that never entered treatment. Third,
in an effort to address endogeneity issues as a result of simultaneity with board characteristics
altering dynamically, we examine the relationship between board characteristics and firm
performance measures by using a dynamic panel GMM estimation approach. Fourth, we
adopt a Probit-based marginal likelihood effects model of the proportion of Independent
Directors on CEO Turnover in cases of annual below industry standard firm-level firm
performance measures. Fifth, we examine mean values and mean value differences between
the treated and control firms for incentive and pay level changes brought about by adopting
the board independence majority rule. Finally, using an instrumental variable (IV) regression
framework, we address the issue of endogeneity as a result of unobserved heterogeneity in
the investigated board structure – firm performance relationship. In order to carry out the IV
analysis we test for the strength of a set of potential instruments that could be included in the
underlying regression framework. While the IV findings support all the previous findings, the
results are not reported for space reasons and are available on request.
2.2.1 Quasi-Natural Experiment – A Difference-in-Differences Approach
The purpose of this study is to examine whether certain measures of firm performance
of a broad set of ASX-listed companies are materially affected by the introduction of the
“majority independence board” criterion which has been formulated as part of the “Corporate
Governance Principles and Recommendations” guidelines released in 2003 by the ASX
Corporate Governance Council.
14
On the basis of simple observational and univariate diagnostics as illustrated in Figure
1, the underlying data set on the proportion of independent directors (Panel A) is evidently
characterized by a strong shift to board majority independence (50% +) following the
introduction of the ASX guidelines in the financial year of 2003. It is important to note that in
particular the largest 200 ASX listed companies, which are presumably more publicly
scrutinized by market participants and regulatory authorities alike and have the largest rents
to be extracted by management, are more likely to adopt a majority independent board by the
end of the financial year 2003, whereas smaller scale companies are significantly more
hesitant to adopt majority independence due to little or less public pressure. Evidently, the
strong spike to majority board independence in the financial year 2003 is neither
contemporaneously compensated by a significant reduction in the average proportion of
executive directors (Panel D), nor by a significant increase in average board size (Panel C).
This implies that non-executive directors were more likely to become even more
disincentivized following the introduction of the new ASX guidelines in order to obtain
“independence” status as classified by the ASX Corporate Governance Council.
<< Insert Figure 1 about here >>
However, it is not evident from the data that this significant upward shift in the
proportion of independent directors leads to a contemporaneous increase in the average
Tobin’s Q value. Instead, the average Tobin’s Q value experiences a visible downward spike
for both major and minor ASX listed companies in the financial year 2003 following the
introduction of the new ASX guidelines (Panel B). Additionally, we are able to observe in the
follow-up period that the average Tobin’s Q value is likely to increase more strongly for
smaller-scale companies that are supposedly more hesitant to adopt majority board
independence. These combined observational characteristics may suggest a negative relation
15
between the adoption of majority independent boards of ASX listed companies and their
subsequent firm performance. We further believe that such a negative relation may be based
in systematically disincentivizing non-executive directors which, ceteris paribus, impedes
effective internal corporate governance eventually translating into a lower firm performance.
Further univariate evidence is provided by Figure 2. Panel A shows that the pay-forperformance sensitivity of independent directors fell dramatically by two-thirds, from 1.5%
to 0.5%, when the ASX rule came into place in 2003 and remained that way, whereas in
Panel B for “Gray” directors that do not meet the independence criterion sensitivity has risen
from 3% to 7%, more than a doubling, over the 11 years of the database.
<< Insert Figure 2 about here >>
Figure 3 sets out the mean performance trends of treated and untreated firms in the raw
data. Untreated firms outstrip treated firms over the period 2005-2007 and then continue to
outperform over the GFC period to 2010 utilizing Tobin’s Q in Panel A of Figure 3. Utilizing
the Market-to-Book ratio in Panel B does not show overall untreated outperformance except
in the GFC period. Figure 3 shows that the performance of firms dominated by independent
directors was particularly disastrous during the GFC period, a finding that is supported in
numerous studies from all over the world that are cited in Ringe (2013).
<< Insert Figure 3 about here >>
Figure 4 shows Tobin’s Q performance split between large (ASX 200) companies and
the remainder (ASX 200 to 500), once again in the raw data. Both large (Panel A) and
smaller companies (Panel B) generally perform better untreated rather than treated. Panel C
provides an industry-adjusted comparison of Tobin’s Q means for large firms (ASX 200).
The apparent considerable rise in treated Tobin’s Q in 2003 appears to be consistent with the
better performing firms choosing to adopt treatment. However, when we performed a
16
statistical analysis, we found no difference in initial performance between firms that accepted
treatment and those that did not. Only a negligible number of firms reported a majority of
independent directors prior to the ASX announcement in 2003 and it is likely that few of
these met the subsequent ASX requirement of independence from significant shareholders.
Hence, one must be wary of assuming there was any economically significant difference
between firms that were subsequently treated and those that were not in the pre-period. The
huge relative out-performance of untreated firms during the period from the GFC onwards
suggests that treated firms perform particularly poorly when the corporate environment is
difficult, as noted by Ringe (2013). Panel D provides a similar industry-adjusted comparison
for smaller companies. Once again, treated firms underperform in 2008 and 2009 during the
GFC.
<< Insert Figure 4 about here >>
Given such clear sample dynamics and data characteristics that strongly resemble that
of a quasi-natural experiment, in this analytical section we propose a DID estimation
framework to evaluate the effects of the ASX policy introduction in 2003 on subsequent firm
performance, as proxied by the Tobin’s Q ratio, the Market-to-Book ratio, and IndustryAdjusted ROA. This framework is characterized by two separated time periods, D1  0,1 ,
where 0 indicates the time period before the introduction of the particular ASX guidelines in
the financial year 2003, i.e., the pre-treatment baseline period, and 1 indicates the time period
after the introduction of the particular ASX guidelines in the financial year 2003, i.e., the
post-treatment follow-up period. Within both treatment periods, we observe two distinct
groups of individual firms indexed by their binary treatment status, D2  0,1 . “0” indicates
the group of firms that are not exposed to treatment (“Control Firms”), i.e., not adopting
majority independent board in any treatment period, and “1” indicates the group of individual
17
firms that receive treatment (‘Treated Firms”), i.e., do adopt majority board independence in
the year 2003 following the ASX policy introduction and maintain a majority independent
board throughout the post-treatment follow-up period. Every individual firm observation,
i  1,..., N , is observed in both pre-treatment and post-treatment periods and the outcome
variable, yi , is defined as firm performance proxy for firm i observed in each financial year,
t  2001,..., 2011, throughout both treatment periods.
We first aim to estimate the following model to assess the effects of majority board
independence on firm performance in a simple policy evaluation setting in which we do not
specifically control for pre-treatment firm-level covariates or firm-level matching:
y  0  1D1  2 D2  3 D1D2   .
(1)
In equation (1), D1 is a binary indicator variable for the post-treatment follow-up
period and captures aggregate factors that would cause changes in firm performance
outcomes even in the absence of the ASX policy introduction. D2 is a binary treatment
indicator that captures possible differences in firm performance outcomes between treated
and control firms prior to the ASX policy introduction in 2003. The coefficient of the
interaction term D1D2 illustrates the DID estimate of the impact variable and can be
formulated as follows:
E Y1  Y0 D  1  E Y1 D  1  E Y0 D  0 ,
(2)
where E Y1 D  1  E Y0 D  0 represents the difference between the treated variable and
control following the policy change. The DID impact coefficient ˆ3 is thus aimed at capturing
the difference in firm performance proxies across the treatment states and time as a result of
the ASX policy introduction in 2003. In addition to the simple DID framework, we continue
18
to implement a full set of firm-level control variables in order to isolate the effects of the
implementation of majority independent boards on firm performance differences.
2.2.2
Difference-in-Difference with Propensity Score Matching
To improve the validity of cross treated/untreated firm comparisons, we further employ
propensity score matching estimators (Rosenbaum and Rubin (1983))12 in our empirical
investigation to estimate the treatment effects of majority board-independence adoption on
subsequent firm performance, conditional on a set of pre-treatment firm covariates. One
major advantage of this method is that it addresses the possible occurrence of a selection bias
in which subsets of firms have a higher probability of adopting a majority independent board
to achieve certain target outcomes. For example, Panel A in Figure 1 clearly illustrates the
tendency of the largest ASX-listed firms to adopt majority independent boards due to greater
public attention, regulatory scrutiny and opportunity for managers to extract rents.
The aim of propensity score matching is to have a large group of firms that do
participate in the “majority independence criterion” following the guidelines released in 2003
by the ASX Corporate Governance Council that are similar to participating firms in a wellchosen set of pre-treatment characteristics X. Statistical post-treatment differences in some
firm performance variable, y , between participating firms and adequate participating control
firms can be then more clearly attributed to the guidelines released in 2003 by the ASX
Corporate Governance Council. Since conditioning on all relevant covariates is limited in the
case of a high dimensional vector X, in their seminal paper, Rosenbaum and Rubin (1983)
12
Similar to the terminology used by Abadie and Imbens (2012), “matching estimators” refers to estimators that
match each unit of some sample subset to a small number of units with similar pre-treatment characteristics in
the opposite treatment party. In this paper, “matching estimators” do not refer to regression imputation methods
that employ large number of matches per unit and non-parametric smoothing techniques to estimate unit-level
regression values in the opposite treatment party.
19
propose to use balancing scores
, i.e., functions of the relevant observed covariates X
such that the conditional distribution of X given
is independent of assignment into
treatment.13 In this study, we employ the Probit-model based propensity score as a balancing
score, which is in essence the probability of participating in some program given the
observed pre-treatment characteristics X. The selection of a suitable matching algorithm is
clearly case sensitive and largely depends on the data structure at hand (Zhao (2000)). Our
data diagnostics reveal that about 19%, or 172 out of 887 firms, adopted a majority
independent board in the financial year of 2003, which implies a great potential for a
multitude of comparable untreated individual firm counterparts. Thus, in order to obtain a
greater matching precision and to reduce the variance due to the inclusion of more
information, we depart from conventional k-nearest neighbor matching algorithms. Instead,
we employ a non-parametric Kernel-based matching algorithm that uses weighted averages
of all individual observations in the comparison group to construct the counterfactual
outcome of a treated individual firm observation. As noted by Smith and Todd (2005), one
can view kernel-matching methods as weighted regressions of the counterfactual outcome on
an intercept with the weights drawn from kernel-density-based weights. Such weights depend
on the relative distance between each individual firm from the control group and the treated
observation for which one has estimated the counterfactual. Using kernel-based propensity
scores, we focus on matching treated firms with non-participating control firm counterparts
using fundamental pre-treatment firm characteristics of firm size (Logarithm of Total Assets)
and two-digit industry sector affiliation (SIC) for each given pre-treatment financial year
(Year). Using firm matching and a full set of control variables within a DID framework, we
the aim is to estimate the following model to evaluate majority board-independence effects
on subsequent firm performance in a complete evaluation context:
13
See Caliendo and Kopeinig (2005).
20
y  0  1 D1   2 D2  3 D1 D2   4 Ln TA   5 D1 Ln TA   6 D2 Ln TA   7 D1 D2 Ln TA  
8 Leverage  9 D1Leverage  10 D2 Leverage  11D1D2 Leverage  12CAPAX 
13 D1CAPAX  14 D2CAPAX  15 D1 D2CAPAX  16Cash  17 D1Cash  18 D2Cash 
19 D1 D2Cash   20 Financial   21 D1 Financial   22 D2 Financial   23 D1D2 Financial 
 24 Mining   25 D1Mining   26 D2 Mining   27 D1 D2 Mining   .
(3)
This general specification allows the controls to have differential effects on the control and
treatment groups pre- and post-event.
2.2.3
Investigating Time-Varying Treatment Effects
So far we employed a binary Propensity Score Matching based Difference-in-
Difference estimation framework for analysing treatment effects as a result of early treatment
adoption in the most proximate financial year ends of 2003 and 2004, immediately after the
ASX’s introduction of the new Corporate Governance guidelines and principles in March
2003. Although the bulk of treatment conversion of S&P/ASX All Ordinaries 500 firms
happened within these years, Table 1 and Figure 5 clearly illustrates a tendency for a
diminishing treatment conversion process over time up until the fiscal year end of 2011.
<< Insert Figure 5 about here >>
This time-varying conversion phenomenon is rooted in the ASX recommendations
guided “if not, why not?” approach of corporate governance practices disclosure, in which
ASX-listed firms have the sovereign flexibility to not integrate specific recommendations
made by the ASX Corporate Governance Council, conditioned by the requirement to publicly
disclose explanations for such decisions. As a result and focusing on majority independent
board recommendations, firms may potentially uphold their “real option” to not early convert
their board to become majority independent and rather engage in a learning process in which
firm expectations on the costs and benefits of independent boards dynamically change due to
the ability to observe and assess the effects of treatment conversion on comparable firms over
21
time. This tendency may especially be true for smaller-sized firms due to being less exposed
to market and regulatory scrutiny as opposed to the larger S&P/ASX 200 firms and more
likely to be subject to product market competition that limits the ability to extract rents.
Consequently, our sample exhibits a time-varying heterogeneous composition of different
treated firm groups which are characterized by different time exposures (“changing treatment
dosages”) to the new ASX reform initiated in 2003.
Assuming that treatment adoption effects on firm performance is not constant over
time, any causal inference based on our prior DID framework clearly depends on the group of
the observed treated firms, which we have limited to the group of 2003 and 2004 fiscal year
early converters. Using these early converting groups specifically allowed us, first, to
maximize the sample of treated firms used in the DID framework and, second, to use a
sample of treated firms that are isolated from any confounding effects due to holding the
“real option” to choose the timing of conversion based on updated expectations. Since our
complete sample features changing treatment groups over time up until the fiscal year of
2011, it may become problematic to analyse causal effects in a simple binary treateduntreated state combination. Firms may react heterogeneously to observable confounders and
potentially self-select into treatment conversion in a multiple period treatment exposure
setting based on derived updated expected outcomes.
In this section, we do not only address the issue of endogeneity which potentially arises
from self-selection into treatment, but we aim to take advantage of the staggered nature of the
time-varying treatment conversion schedule of firms responding to the ASX board
recommendations of 2003. Here, we aim to depart from the classical binary treated-untreated
state nature and assign a treatment value according to the fiscal year when a firm converts
into the treated state. Accordingly, treated firms are effectively sub-grouped into treatment
22
lengths with respect to the date when a firm adopts a majority independent board and with
respect to the date when the particular outcome is being investigated. We approach the
computation of average treatment effects for each sub-grouped treatment indicator by
constructing and estimating a dosage-response function which maps the individual treatment
lengths into particular potential outcomes. Under the general approach, the difference
between two points along the
dose-response function would represent the differential
expected outcome effects of treated firms converting in period t against treated firms
converting in another period s , where t  s .
Table 4 illustrates the notion of potential outcomes in a time-varying treatment setting.
In the table, d and t represent the treatment values (i.e., the period when a firm converts into
the treated state) and the real time horizon, respectively. The vector ytd maps a given
treatment value  d  to a potential outcome
 y
in a time period t. Each cell in the matrix
table then corresponds to a potential outcome for a given firm with a particular treatment
exposure in a specific time period. For illustration purposes, if firm A and firm B convert into
the treated state at period t A  2 and tB  T , respectively, then, at period T
 column t  T  ,
yTd 2  yTd T , represents the differential outcome effect of being majority board independent
with the lapse of T  2 fiscal years with respect to converting into majority board
independence in fiscal year T . The matrix also illustrates that comparisons between treated
and untreated firms can only be made for a limited period, because as we advance in the
treatment window, more firms convert into the treated state over time and thereby reduce the
number of comparable firms in the untreated state. On the other hand, the number of treatedstate firm groups with different “treatment dosages” increase as we advance through the
treatment window.
23
<< Insert Table 4 about here >>
Let Zi ,t  max  0, t  Di  be the time of exposure to treatment in the multi-valued
treatment setting, then we are interested in,   Zi ,t  , the dose-response function of Zi ,t on Yi ,t .
With this intuition in mind, we first aim to model the conditional expectation of Yi given
treatment exposure, Di , and the Generalized Propensity Score, Ri , using a quadratic
parametric approximation 14:
E Y Di , Ri  X 0   0  1Di   2 Di2  3 Ri   4 Ri2  5 Di  Ri .
(4)
We then estimate the parameters using OLS with the estimated conditional GPS
function, Ri  rˆ  d , X 0  in order to estimate the average potential outcome at treatment level
Di  d for every level of treatment level we are interested in to obtain an estimate of the
entire dose-response function ˆ  Zi ,t  evaluated at time t:
E Yˆ  d   
1 N 
2
ˆ0  ˆ 0 d  ˆ 0 d 2  ˆ 0 rˆ  d , X 0   ˆ 0 rˆ  d , X 0   ˆ 0 d  rˆ  d , X 0   .



N i 1
(5)
From any estimate for the dose-response function, we are then able to compute an
estimate for the average time-varying treatment effect in the following way, where d
represents a ‘treatment cut’ indicator for the fiscal year ends between 2003 and 2011 and q
represents the particular treatment indicator for being exposed in the treated state for t  d
fiscal years:
t ,d ,d   ˆ  t  d   ˆ  t  d   ˆ  q   ˆ  q .
14
(6)
Methodology by Imbens and Hirano (2004) updated for categorical treatment regimes by Imai and van Dyk
(2004).
24
Although this framework would allow us to compare average treatment effects between
all possible combinations of different treatment states on a time-varying basis, for the
purposes of our empirical study, we aim to isolate our analysis by comparing average
treatment effects of being majority board independent for q  1, 2,...,9 fiscal years with the
counterfactual case of no treatment, q  0 . For the above regressions, Hirano and Imbens
(2004) demonstrate the estimator for the dose-response function to be root-N consistent and
asymptotically normal such that asymptotic standard errors can be computed using
expansions based on the estimating equations. In practice, however, it is more convenient to
employ bootstrapping methods to calculate standard errors and confidence intervals.
In the previous section on the DID framework, we discussed the problem of the
possible occurrence of a selection bias in which subsets of firms may have a higher
probability of adopting a majority independent board to achieve certain target outcomes. This
tendency is further backed by the observation that the treatment conversion schedule over
time is dominated by the largest firms listed on the ASX (see Table 1). To address this issue,
we employed the propensity score matching estimator developed by Rosenbaum and Rubin
(1983) to match treated with untreated firms conditional on a fundamental set of pretreatment firm covariates aimed at making assignment into treatment a conditionally random
event. In this time-varying multiple treatment value setting, we further need to control for
time-varying heterogeneity in outcome dynamics which may influence the timing of
treatment conversion and thus observed outcomes. This notion of a non-random treatment
assignment over time means that firms with different treatment levels may systematically
differ in characteristics other than the observed timing into treatment which may exhibit
complex correlations with the outcome variable evaluated at a particular point in time.
25
In our setting, it means that the heterogeneity in the outcome variation,  i ,t   i ,t 1 , needs
to be independent of treatment assignment over time in order to produce consistent estimates.
Since treatment assignment in our time-varying treatment setting takes on multiple treatment
values, as opposed to the earlier binary treatment case, we aim to use a set of covariates under
the GPS developed by Imai and van Dyk (2004) and Imbens (2000) for applications on
arbitrary (in our case categorical) treatment regimes. Using the notation of Campello, Ribas
and Wang (2013), the GPS function, r  D, X 0  , can be defined as the conditional probability
of receiving treatment, D, based on a set of pre-determined covariates, X 0 :
R  X 0   r  D, X 0   Pr  D  d X 0  .
(7)
For arbitrary non-binary treatment regimes, Hirano and Imbens (2004) developed a
generalization to the propensity score model by Rosenbaum and Rubin (1983) and show that
the GPS can be adjusted to identify the dose-response function under the following two weak
conditions:
Condition 1:
1D  d  Yt d X 0  , for all d  D and t 0,..., T 
Condition 2 : X0  1  D  d  r  D, X 0  , for all d  D .
(8)
(9)
In this context, condition 1 states that assignment into treatment D is independent from
the potential outcome Yt d  D , conditional on a pre-determined set of covariates contained in
X0 .
Condition 2 states that the set of pre-determined covariates is independent from
assignment into treatment, conditional on controlling for the GPS function (i.e., weakly
balanced). For our purposes, we model and estimate the GPS using a similar methodology
described by Hirano and Imbens (2004) updated for the categorical treatment case by Imai
and van Dyk (2004). Since we are now able to utilize our entire sample of treated firms over
26
the time period between 2003 and 2011, we are in a far better position to implement a full set
of pre-treatment firm specific covariates into the estimation of the GPS. These controls
include industry sector affiliation (2-digit based SIC), firm size (log of Total Assets), firm
performance (Tobin’s Q, Market to Book Ratio and ROA), firm idiosyncratic risk, firm
leverage, CAPEX ratio, cash ratio and board characteristics (board size and proportion of
independent directors) on an annual time-varying pre-treatment basis.
2.2.4
Dynamic System Panel GMM Regression
In this analytical section of our study, we aim to strengthen the results of our primary
analysis in the prior section and assess the effects of board size and board independence on
market-based firm performance measures in which we specifically address well-known
dynamic endogeneity concerns that arise due the inclusion of board structure as a choice
variable. In the following model specification, X, Z, and y represent board structure, firmcontrol characteristics, and market-based firm performance measures, respectively, and i
represents unobserved firm effects:
yi ,t     yi ,t 1   X i ,t   Zi ,t  i   i ,t .
(10)
The model specification in equation (10) suffers from unobserved heterogeneity as a source
of endogeneity if there are factors in the data generating process not explicitly controlled for
in the model that influence both the choice of board structure and firm performance in a given
time period. The existence of such unaccounted factors would introduce bias in OLS or fixedeffects estimation approaches by violating the zero conditional mean assumption
E i |X i ,t , Zi ,t   0 in equation (10). Both theory and logic suggests that this is indeed the
case.
27
Three examples may support this view. First, “skin in the game”, i.e., executive
incentive compensation, may be considered a factor which strongly depends on a given board
structure and as a mechanism of incentivizing senior management, almost certainly have an
influential role on firm performance. Second, managerial ability remains largely unobserved,
but does have a substantial impact on firm performance. As Hermalin and Weisbach (1998)
point out, firms with high-ability CEOs have a lower incentive to monitor, which should also
be reflected in those firms’ board structures. Third, product market competition, the threat of
acquisition with the possibility of subsequent replacement of incumbent management, and
informed “swing” traders (see Gallagher, Gardner and Swan (2013)) reduce the benefits of
internal monitors such as board structure and also influence firm performance.
Similar to Wintoki, Linck, and Netter (2012), in the next step we further assume the
existence of a dynamic element to the determinants of board size and the proportion of
independent directors on a board that is likely to introduce an endogeneity bias into the static
estimation approach of firm performance with board structure variables. Drawing on prior
empirical work motivated by the theories of Hermalin and Weisbach (1998), Raheja (2005),
and Harris and Raviv (2008), it can be argued that there may be at least two channels in
which past firm performance can affect the current choice of board structure. First, as
conceptualized by Hermalin and Weisbach (1998), board independence can be viewed as the
outcome of a bargaining process between the CEO and the board in which the CEO’s
bargaining power derives from his perceived skills compared to alternative CEOs in the
executive labor market. The authors’ argue that the monitoring intensity of a board on the
CEO: first, decreases with the board’s prior estimated perception of CEO ability, second,
decreases with the precision of its prior estimated perception and, third, increases with the
precision of its privately acquired signal of the CEO’s ability. Assuming board independence
to relate positively to monitoring intensity within any particular time-period, one implication
28
would be a negative relationship between board independence and the skills of a firm’s CEO
that, in turn, suggests that board composition itself would be directly determined by past firm
performance.
Another mechanism in which past firm performance can be argued to be an indirect
determinant of current board structure was suggested by Raheja (2005). If board structure is
determined by firm-specific characteristics that are, in turn, related to past levels of firm
performance, then current board structure can be endogenous. Using arguments proposed by
Boone, Field, Karpoff, and Raheja (2007), Coles, Daniel, and Naveen (2008), and Linck,
Netter, and Yang (2006), it follows that more complex and larger firms require larger boards
due to operating in a more sophisticated informational environment. Since firm size and
complexity almost certainly affect firm performance, the choice of board composition may be
indirectly but systematically related to past firm performance.
As suggested by theory, firms would vary their board structure (i.e., both size and
independence composition) at any given time period in anticipation of achieving a targeted
performance outcome in that given period; a relationship which suggests a two-way relation
between board structure and firm performance over multiple time periods. Econometrically
speaking, this issue of simultaneity would violate the zero conditional mean assumption in
any static model and would introduce estimation bias in OLS or fixed-effects panel
regressions as E  i ,t |X i ,t , Zi ,t   0 in equation (10).
In an effort to address the issues of simultaneity bias in the board structure – firm
performance relationship, we propose a dynamic model in which one is able to estimate the
effects of board structure on firm performance using a system of equations. In one equation,
firm performance is allowed to depend on board structure and other included control
variables, while in other equations, board structure is allowed to depend on firm performance
29
and other included control variables. Similar to theoretical models developed by Arellano and
Bover (1995) and Blundell and Bond (1998) that, in this context, has been applied by
Wintoki, Linck and Netter (2012), we propose a dynamic GMM panel estimator to exploit the
dynamic relationship between board structure and firm performance measures. We propose
the following model specification, where X, Z, and y represent board structure, firm-control
characteristics and market based firm performance measures, respectively, and i represents
unobserved firm effects:
yi , t     yi , t 1   X i , t   Zi , t i   i , t .
(11)
To eliminate any potential bias resulting from time invariant unobserved heterogeneity
, we first-difference the above dynamic model and estimate the parameters with GMM
using first-lagged values of the right hand side variables as instruments for current righthand-side variables:
yi ,t    yi ,t 1  X i ,t  Zi ,t   i ,t .
(12)
Using the first-lagged values of explanatory variables in the model as instruments for
current values of explanatory variables is quite a significant aspect of the dynamic panel
GMM estimation approach. However, two criteria must be met in order for these instruments
to valid. First, these lagged variables must provide a source of variation for current board
structure variables. In addition to the previously established theoretical motivation, Table 5
formally shows how lagged values in the performance and explanatory variables feed into the
choice of current values of chosen board structure variables. In particular, if last year’s
Tobin’s Q performance was poor there appears to be an association with a higher board size
and proportion of non-executives in the current year with a lower proportion of executives. A
large board size, proportion of independent directors and non-executive director proportion in
the current year is associated with low idiosyncratic risk in the previous year.
30
<<Insert Table 5 about here>>
The second criterion is that lagged values of the explanatory variables must provide an
exogenous source of variation for current values of board structure variables. In other words,
lagged values of firm performance and firm characteristics must be uncorrelated with the
error term in equation (3). As discussed by Wintoki, Linck, and Netter (2012), under the
assumption of weak rational expectations, current shocks to firm performance must have
been completely unanticipated in the prior time period in which board structure was chosen,
given that the choice of board structure is a full reflection of anticipated benefits and costs
between varying board structures. For simplicity, we hereby specifically assume that one lag
of firm performance is sufficient to capture the influence of a firm’s history on the present
and any information prior to the first lag does not directly, but only indirectly, affects current
firm performance through the channels of variant board structure and firm characteristics.
Thus, we treat firm performance beyond its first lag as exogenous, sufficient to ensure
dynamic completeness. We could then estimate equation (4) using the GMM approach.
However, as pointed out by Wintoki, Linck, and Netter (2012), the GMM estimation of
the first-differenced model specification alone suffers from at least three econometric
shortcomings. First, given that the original model specification is in levels, differencing may
reduce the variation in explanatory variables and hence reduce the power of tests (Beck et al.
(2000)). Second, explanatory variables in levels may be only weak instruments for firstdifferenced model specifications (see Arellano and Bover (1995)). Third, Griliches and
Hausman (1986) pointed out that first differencing could worsen the impact of measurement
errors on the dependent variable. In order to address such econometric drawbacks and to
validate the estimation approach of the dynamic GMM estimator, Arellano and Bover (1995),
and Blundell and Bond (1998) suggest the simultaneous estimation of both level and first-
31
differenced model specifications in a “stacked” system of equations, known as dynamic panel
system GMM estimation, in the following form:
 yi ,t 
 yi ,t 1 
 X i ,t 
 Zi ,t 
 y       y     X     Z    i ,t .
 i ,t 
 i ,t 1 
 i ,t 
 i ,t 
(13)
Since the level equation of the system of equations in (5) still includes unobserved
heterogeneity, we assume a constant correlation between board structure variables, firm
characteristics and unobserved firm effects over time, which is a reasonable assumption in
our context for short time periods.
Finally, we then conduct a system GMM panel estimation for which we are able to
obtain efficient estimates while addressing the issue of simultaneity and controlling for timeinvariant unobserved heterogeneity and the dynamic relationship between current values of
board structure and firm-specific variables with lagged values of firm performance. Since, the
validity and quality of such estimation heavily relies on the explicit assumptions we
previously discussed, we further provide a set of tests for each dynamic system GMM
regression approach to assess the strength of our initial assumptions. This methodology
allows board structure to evolve dynamically in response to changes in firm performance.
3. RESULTS
3.1
Simple Difference-in-Difference Model
Findings based on the most elementary of the models we consider, the simple DID
model formulated in section 2.2.1 above and equation (1), are set out in Table 6. Panel A
reports the findings based on the Tobin’s Q ratio and Panel B, Market-to-Book. The baseline
results for both models indicate, regardless of the performance measure, firms that
subsequently adopt a majority of independent board members significantly outperformed in
32
the pre-period. This can be seen visually in Panel C of Figure 4. Following treatment,
however, firms that adopted the majority recommendation significantly underperformed the
control group during the post-event period to 2011 according to the Tobin’s Q ratio criterion
but slightly outperformed according to the Market-to-Book ratio criterion. In the final
column, the Difference-in-Difference (DID) impact shows that treated firms under-performed
in the post-event period relative to the untreated group after taking the difference between the
two periods. While the magnitude of the under-performance is slightly greater utilizing the
Tobin’s Q measure and the significance level is higher at the 1% level, even utilizing the
Market-to-Book criterion DID underperformance is still significant at the 5% level.
<<Insert Tables 6 about here>>
3.2
Difference-in-Difference Model with Propensity Score Matching
This methodology is explained in detail in Section 2.2.2 above with the specification
for propensity scoring set out in Table 7 and equation (3). The results provided in Table 8 are
for the 172 firms that became treated in 2003. They are split between Panel A with Tobin’s Q
as the dependent variable, Panel B with the Market-to-Book ratio, and Panel C, IndustryAdjusted ROA. Panel A indicates no significant difference in performance between the
treated and control group in the pre-period but in the post-period the performance of treated
firms that responded to the ASX listing requirement by adopting a majority of independent
board members performed far worse with a highly economically large, and statistically
significant (at the 1% level), fall in the Tobin’s Q ratio of approximately 5. The overall
Difference-in-Difference impact is also approximately 5 and is thus huge in magnitude. The
propensity score matching has increased the measured fall in firm performance once
independent directors have replaced a sufficient number of executives and “Gray” nonexecutive directors to constitute a majority. Firms that remained “untreated” simply had to
33
state why they preferred to remain with less than a majority of independent directors. This
Difference-in-Differences is positively related to size in the form of the logarithm of total
assets and negatively to leverage. Hence, larger firms with less leverage and thus greater free
cash flow were more likely to under-perform.
<<Insert Tables 7 and 8 about here>>
The findings utilizing the Market-to-Book Ratio (MTBR) in Panel B are similar. In the
pre-period firms that are subsequently treated perform better, although the difference is not
statistically significant. In the post-period treated firms perform significantly worse with a
similar order of magnitude (approximately -5.5) with an overall Difference-in-Difference
impact of approximately -12. A larger size is the only significant factor associated with this
underperformance. The analysis is repeated in Panel C utilizing Industry-Adjusted accounting
performance (ROA) with similar results. Firms that became treated in 2003 were initially
outperforming control firms in terms of ROA but in the post-treatment period this
outperformance weakened, resulting in a significant negative Difference-in-Differences.
The DID analysis performed above is repeated in Table 9 for the 106 firms that delayed
their treatment until 2004 with very similar results for those firms that initially adopted in
2003, as set out in Table 8. The results in Panel A in Table 9 have a similarly statistically
significant DID for the Tobin’s Q ratio as in Table 8. For the Market-to-Book ratio in Panel
B, the set of treated firms have a far lower ratio in the post-period even though the overall
difference-in difference is not significant. Once again, the findings for Industry-Adjusted
ROA in Panel C are similar to the findings for the 2003-year treatment.
<<Insert Table 9 about here>>
3.3
Time-Varying Treatment Effects
34
Figure 6 displays the estimated time-varying treatment effects on CEO, executive
member, and director pay variables for the annual treatment exposure between 1 to 9 years
post-treatment conversion against the untreated scenario using the GPS methodology
described in Section 2.2.3 above. Controlling for annual untreated matched firm groups using
the GPS we can observe that the estimates on the expected differences for each of the four
outlined compensation categories between annual treatment exposure,   Z  , and the
counterfactual case of no treatment exposure,   0  , are statistically significant for every
fiscal year of treatment exposure. With respect to this, Panel A clearly shows a tendency for
treated firms to have a lower pay-for-performance sensitivity of CEO pay over time, the
longer such firms are exposed to the treated state of majority board independence. After
being exposed in the treated state for about nine years after treatment conversion, the
expected difference in sensitivity has been reduced in magnitude by about 6 percentile points
– a huge drop since untreated firms have an average sensitivity of about 10%. Examining
executives more broadly, Panel B indicates a drop of about 3.5 percentile points after about
four years into the treated state with some recovery after that point. Since independent
directors have to be divorced from significant shareholders (5% or more), their influence
seems to extend to a gradual reduction in executive pay-for-performance sensitivity when
compared against the untreated case of firms with boards not characterized as being majority
independent. Panel C reveals one mechanism by which this reduction in pay-for-performance
sensitivity occurs, with the difference in cash-based CEO pay rising by about 50% over the
first four years in treatment compared to the corresponding case of no treatment. It thus
appears that independent directors, whose interests are largely independent of shareholders,
are far more inclined to give in to the demands by CEOs for higher cash-based pay unrelated
to performance. Panel D of Figure 6 shows that, not only do independent directors appear
generous towards CEOs, but also their generosity extends to non-executive directors
35
generally. Within four years of treatment exposure, we can observe the average director fees
to rise by over 40% compared to the average director fees in firms without treatment
exposure.
<<Insert Figure 6 about here>>
Figure 7 displays the time-varying effects of treatment exposure on firm performance
as measured by the Tobin’s Q ratio. Here, Panel A examines the effect of a given initial
treatment conversion on the immediate annual change in the Tobin’s Q ratio in the fiscal
years following the ASX guidelines introduction in 2003 to including 2011. In 2003, the first
adoption year since the majority board recommendations introduction, there is an indication
of an immediate and statistically significant negative change in the Tobin’s Q ratio of 0.2
compared to untreated firms in the same year. Continuing with the firm group of early
converters in 2004, similar to the firm group of early converters from 2003, we can observe a
negative reaction in the immediate change of the Tobin’s Q ratio of about 0.13 when
compared to the untreated firm group in the same year. As the treatment conversion schedule
advances throughout the treatment window up until the fiscal year end of 2011, it appears that
the adverse impact in the immediate change in Tobin’s Q of treated firm groups against
corresponding untreated firm groups ameliorates over time. The variability of outcome
estimates is increasing over time because of the diminishing treatment conversion schedule.
One may also argue that late converting firm groups may have updated their expectations
from observing treatment conversion outcomes of early converting firm groups and thus have
learned something from their predecessors. Further, while only a small number of firms are
treated for the first time in 2010, there is no indication of an immediate decline and, in fact, in
2011 there is an indication of a small improvement. We can observe a general tendency for
converting firms to have a more negative reaction in the Tobin’s Q ratio from one fiscal year
36
to the next as compared to corresponding untreated firm groups. However, Panel A itself
lacks the dimensionality to truly reveal treatment conversion outcomes on market-based firm
performance measures over time, including the Tobin’s Q ratio and the more narrowly
defined Market-to-Book ratio. With this regard, Panel B and Panel C display the time-varying
effects of different treatment exposure groups on the Tobin’s Q ratio over time against the
corresponding firm groups of non-exposure over time. The findings here are somewhat more
concerning. Panel B displays worsening performance for every year in treatment. There is a
very concerning decline in Tobin’s Q performance of around 0.4 after nine years in treatment
when evaluated against the corresponding pool of untreated firms. In Panel C, we replace the
raw Tobin’s Q ratio by the Industry-adjusted Tobin’s Q ratio that measures performance net
of idiosyncratic ‘noise’ in industry performance. There is a more modest overall decline of
around 0.15 between treated firms with nine years of treatment, compared with the
corresponding pool of untreated firms. Measurement net of market-wide confounding
impacts on firm performance still confirms the value-destroying tendency of treatment
adoption effects over time. Figure 8 repeats the analyses of time-varying treatment effects
conditional on different treatment dosages for the Market to Book ratio as firm performance
measure. Here, we can generally confirm similar findings to those for Tobin’s Q ratio – while
the immediate response seems to have improved over time, there is still significant secular
performance decline the longer is the firm undergoing treatment.
Some relatively straight-forward calculations designed to provide a dollar estimate of
losses inflicted on shareholders by treatment are in order. Altogether, in our sample we
observe majority board independence treatment conversion in 561 ASX-listed firms between
the fiscal years of 2003 to including 2011, with the bulk of firms early converting into the
treated state in the fiscal years of 2003 and 2004 immediately after the ASX board
recommendations have been officially introduced. The average market capitalization of ASX37
listed firms in our sample is $793 million and $887 million in 2003 and 2004, respectively.
Moreover, the immediate changes in the Tobin’s Q ratio of early converting firms (against
corresponding non-treated firms) have been estimated to be negative 0.195 and negative
0.126 in 2003 and 2004, respectively, the expected accumulated immediate cost of the
majority independence adoption for shareholders by 2004 we estimate to be approximately
$38.6 billion. When we extend the immediate treatment conversion costs for shareholders to
include the entire conversion schedule up to 2011, the expected accumulated cost of
treatment conversion we estimate to be around an astonishing $69.4 billion compared to the
scenario of the non-treated state. Due to the high proportion of S&P/ASX 200 firms
converting into the treated state throughout the treatment conversion window, these expected
accumulated immediate conversion costs could be even higher than the earlier mentioned
values. Furthermore, the length of treatment exposure given by the estimation results in
Figures 7 and 8 is likely to exacerbate the costs further as firm performance continues to
worsen the longer firms remain treated.
3.4
Dynamic System Panel GMM Regression
Table 10 sets out the findings based on the methodology set out in Section 2.2.4 above
with Panel A addressing Tobin’s Q as the performance variable, Panel B, the Market-to-Book
ratio (MTBR), and Panel C, the Industry-Adjusted ROA. Both larger boards and a higher
proportion of Independent Directors, alternatively a majority of Independent Directors, is
always harmful to Tobin’s Q performance in Panel A. Commencing with model (1) in Panel
A, the Logarithm of Board size instrumented with the strong instruments is strongly
negatively associated with Tobin’s Q ratio performance, as is the proportion of Independent
Directors that has been instrumented with the strong instruments. Consistent with Yermack’s
(1996, 2004) findings for United States and also Eisenberg, Sundgren, and M. Wells (1998),
38
small boards perform significantly better than large boards. As noted above, one of
Australia’s largest corporate collapses, that of HIH Insurance, was due to negligent oversight
by a large board of eleven members, but ASX regulatory arrangements fail to advise against
large boards. An average board size of approximately 6 implies a Tobin’s Q value of 2.142
but increasing board size to 10, a number favored by some of Australia’s largest companies
such as BHP, is associated with Tobin’s Q performance that is 17.5% lower. Similarly, if the
proportion of Independent Directors is at its mean of 31.9%, an increase in the proportion to
90% that is common with large firms lowers predicted Tobin’s Q from its mean by a fall of
26.2%.
Additionally, if the proportion of Independent Directors is high in a top 200 firm the
magnitude of the adverse impact is close to quadrupled. In both Panel B with the Market-toBook Ratio and Panel C with ROA as the performance criterion, board size ceases to be
critical. However, the proportion of Independent Directors continues to have a negative
impact that is high in absolute magnitude. Control variables play an important role. Higher
idiosyncratic risk is now beneficial for both Tobin’s Q and Market-to-Book performance but
not for ROA. Both Tobin’s Q and Market-to-Book performance are increasing in firm size
and in the Cash ratio. Accounting performance is also increasing in market capitalization but
falling in leverage, the CAPEX ratio, and the cash ratio.
<< Insert Table 10 about here >>
3.5
Replacement of Underperforming CEOs
So far, we have established utilizing a variety of mechanisms to deal with endogeneity
that firms which adopt a majority of independent board members perform poorly with respect
to market valuation, both Tobin’s Q and Market-to-Book, and accounting performance,
39
Industry-Adjusted ROA. However, apart from the indication that this is due to the almost
non-existent skin in the game of “independent” directors compared with the high and rising
skin in the game of “Gray” directors, the reasons for this are not yet fully clear. Tables 11 and
12 address the issue of how the alteration from a minority of independent board members to a
majority influences CEO replacement when the firm is relatively underperforming compared
to its peers. Utilizing Probit analysis, the findings in Table 13 indicate that regardless of what
measure of underperformance is used, market or accounting-based, having an independent
majority of directors (i.e., being “treated”) or having a higher proportion of independent
directors,
significantly
reduces
underperformance include:
the
likelihood
of
CEO
turnover.
Measures
of
negative industry-adjusted returns, a negative value for the
difference between the firm’s Tobin’s Q (Market-to-Book ratio) and the comparable industry
Tobin’s Q (Market-to-Book ratio), similarly when the difference between the firm’s ROA
(ROE) and comparable industry ROA (ROE) is negative. Thus unmotivated “independent”
directors lacking skin in the game appear less able to monitor the CEO and replace an
underperforming CEO. The Probit models in Table 12 add a range of controls to the models
of Table 13 with, on the whole, similar findings. These controls include CEO ownership and
tenure, as well as return on assets.
<< Insert Tables 11 and 12 about here >>
3.6
What Alters When a Firm is “Treated”?
Apart from falls in market and accounting performance what other consequences are
there when a firm is treated? A DID analysis is applied to an extensive range of incentive,
pay-level, and firm characteristics of the entire set of treated firms, 2003-2011, and the
corresponding control firms in Table 13. The table shows that pay-for-performance
sensitivities are generally significantly lower for treated firms in both the pre- and post40
treatment periods. Perhaps the most significant finding is that despite CEO base salary, cash
bonus and total compensation being significantly higher in the pre-treatment period for
subsequently treated firms, these items together with mean top executive total compensation
are all significantly higher again post-treatment, giving rise to a positive and highly
significant DID impact. For example, CEO total compensation for treated firms nearly
doubles to approximately $2.2 million with a mean DID of $713,618. The model of Hermalin
(2005) provides a rationale for why greater monitoring intensity by boards dominated by
independent directors could raise rather than lower CEO pay. More intensive monitoring of
talented CEOs could raise the risk of dismissal, giving rise to a compensating upward
movement in CEO pay. However, here we find no evidence that independence encourages
monitoring of the CEO. The almost doubling of pay levels appears to be a consequence of the
lack of incentives for independent directors to monitor that progressively raises CEO pay the
longer is the firm in treatment.
<< Insert Table 13 about here >>
Not only are less incentivized boards dominated by independents more likely to be
accommodating to the CEO and other executives in terms of pay levels, but they are also
more likely to be accommodating in the form of director fees and non-executive director fees
both before becoming treated and especially once treated. Thus, non-executive director fees
are higher by $44,647 post-treatment with a DID of $29,934. Another distinguishing feature
of boards that become treated is that directors are not only busier (i.e., more board interlocks)
and have a higher proportion of females prior to treatment than the control group but these
same characteristics also apply after treatment. Moreover, treated companies take fewer risks
with lower idiosyncratic risk and are less inclined to keep cash reserves (i.e., a lower cash
ratio) both pre- and post-treatment.
41
Firms that become treated tend to be larger than those that do not respond and their
relative size is higher still in the post-treatment period. Treated firms prior to treatment have
lower institutional holdings than the control group with this difference remaining in the posttreatment period but the impact of treatment still relatively raises institutional holdings.
Treated firms also have far busier boards with a significantly higher number of meetings of
both the compensation and nominating committees. Treatment results in a significant DID
impact on the likelihood that an independent director will chair the compensation and
nominating committees.
4. CONCLUSION
In 2003, the Australian Securities Exchange (ASX) requested that all listed firms should
adopt a majority of “independent” board members that are characterized to neither have links
to management nor to substantial shareholders (i.e., 5% or greater shareholding). Throughout
a majority board independence conversion window of nine years up to including the year of
2011, we find that 551 of our total sample of 969 firms (or 57% of our sample) responded
positively to the ASX guided majority independence recommendation, with the bulk of firms
early converting in the years of 2003 and 2004.
In this paper, we analyzed the effects of majority board independence adoption on firm
performance in quasi-natural experimental setting. Due to issues concerning unobservable
heterogeneity and endogeneity and the likelihood that adoption of the majority independence
criterion was non-random, we began our analysis with a Propensity Score Matching (PSM)
estimation strategy within a Difference-in-Difference framework. We aimed to control for
pre-treatment covariates in firm characteristics to isolate firm performance effects of “treated
firms”, i.e., those firms that immediately responded positively to the ASX-guided majority
42
board independence recommendations by early converting into the treated state in either 2003
or 2004.
In taking advantage of the staggered nature of the time-varying treatment conversion
schedule inherent in our sample, we continued the analysis of treatment effects by subgrouping treated firms according to the fiscal year when a firm adopts a majority independent
board. We then approached the computation of average treatment effects for each subgrouped treatment group by constructing and estimating dosage-response functions arising
from the use of generalized propensity scoring (GPS) which map the individually assigned
treatment lengths into particular potential firm performance outcomes. Next, we strengthened
the results of our primary analyses by supplementing the employed methodologies using a
dynamic system panel GMM approach to include all firm-year observations and to account
for issues of dynamic endogeneity and, finally, we examine the difference-in difference
impact of treatment on pay-for-performance sensitivity, CEO pay, director fees, and a
number of other variables.
In this study, we generally find that the longer a firm is subject to treatment exposure,
the poorer is its performance based on both market- and accounting-based performance
measures and the higher is CEO pay and average director fees, with directors generally
suffering from weakened pay-for-performance sensitivity.
Using a comprehensive dataset on the Australian corporate landscape, all employed
methods in this paper find that “majority board independence treatment” resulted in large and
statically significant falls in two primary market performance indicators measured by the
Tobin’s Q and Market-to-Book ratio, and in accounting-based performance as measured by
the Industry-adjusted ROA. An important mechanism by which this shareholder value
destroying process occurs are weakened incentives to effectively monitor the CEO due to
“independent” board members having negligible “skin in the game” relative to “Gray”
43
directors which appear to having been displaced following treatment adoption. We further
find evidence that relatively poorly performing CEOs were less likely to be replaced, while at
the same time, were far better paid with directors themselves receiving far higher fees.
Thus, it would appear that the ASX Corporate Governance Council guidelines that
deem any director who owns 5% or more shares (or is associated with such a shareholder) as
lacking independence, has destroyed considerable shareholder wealth of the order of $69
billion or more. This is in conjunction with the requirement for conformity with the
recommendations that the majority of directors be “independent”, a recommendation that is
common to the UK and the US. The ASX has perpetuated this situation knowing that it is
directly counter to the necessity for incentive alignment between boards and shareholders.
The ASX must also be aware that US exchanges explicitly rejected the ASX rule based on
setting a limit on share ownership in the interests of encouraging necessary incentive
alignment and better monitoring of management by boards. This massive loss of wealth has
resulted in no discernible benefit to other than board members, both executives and nonexecutive directors alike.
44
REFERENCES
Adams, Renée B., and Daniel Ferreira (2007). “A Theory of Friendly Boards,” Journal of
Finance 62, 217–50.
Adams, R. B., B. E. Hermalin, and M. S. Weisbach (2010). “The Role of Boards of Directors
in Corporate Governance: A Conceptual Framework and Survey,” Journal of Economic
Literature 48, 58–107.
Arellano, M., and S. Bond (1991), “Some tests of specification for panel data: Monte Carlo
evidence and an application to employment equations,” Review of Economic Studies 58,
277-297.
Arellano, M., and O. Bover (1995), “Another look at the instrumental variable estimation of
error-component models,” Journal of Econometrics 68, 29-51.
ASX Corporate Governance Council (2003). “Principles of Good Corporate Governance and
Best Practice Recommendations.” See
http://www.asx.com.au/about/corporate_governance/principles_good_corporate_governan
ce.htm
ASX Corporate Governance Council (2010). “Corporate Governance Principles and
Recommendations with 2010 Amendments.” Second Edition.
Australian Prudential Regulation Authority (2012). “Prudential Standard CPS 510.” July.
Bhagat, S. and Black, B. S. (2002). "The Non-Correlation Between Board Independence and
Long-Term Firm Performance." Journal of Corporation Law 27, 231 - 273.
Bhagat, S., and Bolton, B. (2008). Corporate Governance and Firm Performance. Journal of
Corporate Finance 14, 264 – 269.
Boone, A. L., Field, L. C., Karpoff, J. C., and Raheja, C. G. (2007). “The Determinants of
Corporate Board Size and Composition: An Empirical Analysis.” Journal of Financial
Economics 85, 66–101.
Cadbury Report (1992). “Report of the Committee on the Financial Aspects of Corporate
Governance.” Gee and Co. Ltd, London.
Cagan, Penny (2001). “HIH Insurance.” ERisk.com, November 2001.
Campello, Murillo, Rafael P. Ribas, and Albert Wang (2013), “Is the Stock Market Just a
Side Show? Evidence from a Structural Reform,” Working Paper, Cornell University.
Centro Judgement (27 June, 2011). Australian Securities and Investments Commission v
Healey [2011] FCA 717. Judge: Middleton J.
Chhaochharia, V., and Y. Grinstein (2009). “CEO compensation and board structure,”
Journal of Finance 64, 231-261.
Coles, J. L., N. D. Daniel, and L. Naveen (2008). “Boards: Does one size fit all?” Journal of
Financial Economics 87 329-356.
Demsetz, H. (1983). “The Structure of Ownership and Theory of the Firm.” Journal of Law
and Economics 26, 375 - 390.
Dicks, David E., (2012). “Executive Compensation and the Role for Corporate Governance
Regulation,” Review of Financial Studies 25, 1971-2004.
45
Duchin, R., J. Matsusaka, and O. Ozbas (2010). “When Are Outside Directors Effective?”
Journal of Financial Economics 96, 195-214.
Eisenberg, T., S. Sundgren, and M. Wells (1998). “Larger Board Size and Decreasing Firm
Value in Small Firms,” Journal of Financial Economics 48, 35 - 54.
European Commission (2005). “Recommendation on the Role of Non-Executive or
Supervisory Directors of Listed Companies and on the Committees of the (Supervisory)
Board”. See http://europa.eu.int/eurlex/
Fama, E. (1980). “Agency problems and the theory of the firm,” Journal of Political
Economy 88, 288-307.
Financial Reporting Council (FRC) (2012). “The UK Corporate Governance Code.”
September.
Fracassi, C., and G. Tate (2012). “External Networking and Internal Firm Governance,”
Journal of Finance 67, 153 – 194.
Gallagher, David R., Peter A. Gardner, and Peter L. Swan, (2013), “Governance through
Trading: Institutional Swing Trades and Subsequent Firm Performance,” Journal of
Finance and Quantitative Analysis 48, 427–458.
Guo, Lixiong and Ronald Masulis (2013), “Board Structure and Monitoring: New Evidence
from CEO Turnovers,” ECGI Finance Working Paper N° 351/2013.
Gillette, A. B., T. H. Noe, and M. J. Rebello (2003). “Corporate Board Composition,
Protocols, and Voting Behavior,” Journal of Finance 58, 1997-2031.
Griliches, Zvi, and Jerry A. Hausman (1986). “Errors in Variables in Panel Data,” Journal of
Econometrics 31, 93-118.
Guner, A. B., U. Malmendier, and G. Tate (2008). “Financial Expertise of Directors,”
Journal of Financial Economics 88, 323 – 354.
Guthrie, K., J. Sokolowsky, and K. Wan (2012). “CEO Compensation and Board Structure
Revisited,” Journal of Finance 67, 1149-1168.
Harris, M. and Raviv, A. (2008), “A Theory of Board Control and Size,” Review of Financial
Studies 21, 1797-1832.
Heckman, James J., and Edward J. Vytlacil (2007), “Econometric evaluation of social
programs, part I: Causal models, structural models and econometric policy evaluation,”
Handbook of Econometrics, v. 6B, 4779-4874.
Heckman, James J., and Petra Todd Ichimura (1997). Matching as an econometric evaluation
estimator: Evidence from evaluating a job training program, Review of Economic Studies
64, 605-654.
Hermalin, Benjamin E., (2005). “Trends in Corporate Governance,” Journal of Finance 60,
2351–84.
Hermalin, B., and M. S. Weisbach (1991). “The Effect of Board Composition and Direct
Incentives on Firm Performance.” Financial Management 20, 101 - 112.
Hermalin, B. E., and M. S. Weisbach (1998). “Endogenously Chosen Boards of Directors and
their Monitoring of the CEO.” American Economic Review 88, 96–118.
46
Hermalin, B. E. and M. S. Weisbach (2003). “Boards of Directors as an Endogenously
Determined Institution: A Survey of the Economic Literature.” Federal Reserve Bank of
New York Economic Policy Review 9, 7–26.
Hirano, Keisuke, and Guido Imbens (2004), “The propensity score with continuous
treatments,” in A. Gelman and X.-L. Meng (eds.), Applied Bayesian Modeling and Causal
Inference from Incomplete-Data Perspectives, Wiley Series in Probability and Statistics,
73-84.
Imai, Kosuke and David A. van Dyk (2004), “Casal Inference with General Treatment
Regimes: Generalizing the Propensity Score,” Journal of the American Statistical
Association 99, 854-866.
Imbens, Guido (2000), The role of the propensity score in estimating dose-response
functions, Biometrika 87, 706-710.
Jensen, C. M. (1993). “The Modern Industrial Revolution, Exit and Failure of Internal
Control Systems.” Journal of Finance 48, 831 - 880.
Joffe, Marshall M., and Paul Rosenbaum (1999), “Invited commentary: Propensity scores,”
American Journal of Epidemiology 150, 327-333.
Knyazeva, A., D. Knyazeva, and C. Raheja (2009). “Heterogeneity in Expertise and
Incentives of Board Members,” Working paper.
Knyazeva, Anzhela, Diana Knyazeva, and Ronald Masulis (2013). “The Supply of Corporate
Directors and Board Independence,” Review of Financial Studies. Forthcoming.
Le Mire, Suzanne, and George Gilligan (2013). “Independence and Independent Company
Directors,” Journal of Corporate Law Studies 13, Number 2, 443-475.
Linck, J., J. J. Netter, and T. Yang (2006). “A Large Sample Study on Board Changes and
Determinants of Board Structure,” Journal of Financial Economics 87, 308 - 328.
Masulis, Ronald, and Shawn Mobbs (2013). “Where Do Talented Directors Spend Their
Limited Time and Energy?,” European Corporate Governance Institute, Finance Working
Paper N° 355/2013.
Masulis, R. W., C. Ruzzier, S. Xiao, and S. Zhao (2012). “Do Independent Directors
Matter?” Working paper, University of New South Wales.
Masulis, R. W., C. Wang, and F. Xie (2012). “Globalizing the Boardroom – The Effects of
Foreign Directors on Corporate Governance and Firm Performance,” Journal of
Accounting and Economics, Forthcoming.
McConnell, J. and H. Servaes (1990). “Additional Evidence on Equity Ownership and
Corporate Value” Journal of Financial Economics 27, 595 - 612.
McConnell, J. J., H. Servaes and K. V. Lins (2008). “Changes in insider ownership and
changes in the market value of the firm,” Journal of Corporate Finance 14, 92-106.
Mehran, H. (1995). “Executive Compensation Structure, Ownership, and Firm Performance,”
Journal of Financial Economics 38, 163 - 184.
Morck, R., Shleifer, H. and R. Vishny (1988). "Management Ownership and Market
Valuation: An Empirical Analysis," Journal of Financial Economics 20, 293 - 315.
NASDAQ
(2003)
Corporate
Governance
www.nasdaq.com/about/CorpGovSummary.pdf
47
Rule
Changes,
see
NYSE (2003) Listed Company Manual, s303A Corporate Governance Rules, see
www.nyse.com/pdfs/finalcorpgovrules.pdf
NYSE Listed Company Manual (2013). “303A.00 Corporate Governance Standards.” NYSE
Euronext.
OECD (2004). Principles of Corporate Governance. See www.oecd.org.
Perry, T. (1999). “Incentive compensation for outside directors and CEO turnover,” Working
Paper, Arizona State University.
Pfeffer, J. (1972). “Size and Composition of Corporate Boards of Directors: The
Organization and Environment,” Administrative Science Quarterly 17, 218 – 229.
Raheja, Charu (2005). “Determinants of board size and composition: A theory of corporate
boards,” Journal of Financial and Quantitative Analysis 40, 283-305.
Ravina, Enrichetta, and Paola Sapienza (2010). “What Do Independent Directors Know?
Evidence from Their Trading,” Review of Financial Studies 23, 962-1003.
Ringe, Wolf-Georg (2013). “Independent Directors: After the Crisis”. European Business
Organization Law Review 14, 401-424 doi:10.1017/S1566752912001206
Rosenbaum, Paul R., and Donald B. Rubin (1983). “The central role of the propensity score
in observational studies for causal effects,” Biometrika 70, 41-55.
Securities and Exchange Commission (2003), “Final Rule. Standards Relating to Listed
Company Audit Committees.” Release No. 34-47654. 17 CFR PARTS 228, 229, 240, 249
and 274.
Wintoki, M. Babajide, James S. Linck, and Jeffry M. Netter (2012), “Endogeneity and the
Dynamics of Internal Corporate Governance,” Journal of Financial Economics 105, 581606.
Yermack, D. (1996). “Higher market valuation of companies with a small board of directors.”
Journal of Financial Economics 40, 185-211.
Yermack, D. (2004). “Remuneration, Retention, and Reputation Incentives for Outside
Directors,” Journal of Finance 59, 2281-2308.
48
APPENDIX
Table 1: "Treated" Firms relative to Database Firm-Year Population
Year
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Firm's "Treated"
23
172
107
67
46
44
44
36
22
13
No. in ASX 200 "Treated"
12
110
60
27
21
24
14
19
7
4
Total Number of Firms
867
884
891
915
939
961
957
920
902
870
731
Cumul. "Treated" (% Ttl)
0.0
2.6
21.9
33.0
39.3
43.2
48.0
54.7
59.8
64.5
78.5
Table 2: Descriptive Statistics
Variable
Mean
Median
Std Deviation
5th Percentile
95th Percentile
Tobin’s Q
2.1420
1.3239
2.4735
0.6303
6.3638
Market-to-Book
2.6476
1.5432
4.1150
0.2829
8.7626
Board Size
6.0468
6
2.3506
3
10
Proportion IDs
0.3186
0.31
0.2987
0
0.8181
Proportion GDs
0.3861
0.33
0.3083
0
0.8750
Proportion NEDs
0.7047
0.75
0.1834
0.3333
1
Proportion EDs
0.2945
0.25
0.1834
0
0.6666
3,690
51.4
33,700
1.531
5,100
Total Assets ($m)
Market Cap ($m)
1,240
57
6,260
2.740
4,700
Leverage Ratio
0.4229
0.3651
0.4473
0.0210
0.9112
CAPEX Ratio
0.0921
0.0328
.3926
0
0.3420
Cash Ratio
0.1966
0.0953
0.2396
0.0029
0.7767
Idiosyncratic Risk
0.0438
0.0366
0.0312
0.0116
0.0983
49
Table 3: Partial Correlation Matrix
T’s Q
MBR
B Size
Pr IDs
Pr NED
Pr EDs
Ttl Ats
Mt Cp
Lev
CAPEX
Cash
Tobin’s Q
1
M to Bk R
0.662
1
Board Size
-0.085
-0.001
1
Prop IDs
-0.013
0.017
0.132
1
Prop NEDs
-0.034
-0.022
0.252
0.254
1
0.034
0.021
-0.259
-0.255
-0.997
1
Ln Ttl Assets
-0.323
-0.138
0.573
0.342
0.226
-0.229
1
Ln Mkt Cap
0.016
0.117
0.564
0.362
0.218
-0.222
0.879
1
Leverage
0.107
-0.016
-0.019
0.026
0.018
-0.018
-0.064
-0.019
1
CAPEX Ratio
0.182
0.137
-0.014
-0.006
-0.019
0.019
-0.097
-0.032
-0.002
1
Cash Ratio
0.311
0.209
-0.187
-0.048
-0.045
0.046
-0.419
-0.244
0.038
0.035
1
Idiosync Risk
0.111
0.006
-0.268
-0.216
-0.125
0.127
-0.539
-0.622
-0.015
0.040
0.228
Prop EDs
50
Table 4: Time Varying Treatment (d) as Matrix of Potential Outcomes (y)
t 0
d 1
t 1
y1d 1
d 2
d T 2
y0d 0
y1d 1
d  T 1
t2
t T 2
t  T 1
t T
y2d 1
yTd 21
yTd 11
yTd 1
y2d  2
yTd 22
yTd 12
yTd  2
yTd 2T  2
yTd 1T  2
yTd T  2
yTd 1T 1
yTd T 1
yTd 1T 1
yTd T
y2d  2
d T  2
T 2
y
d T
Table 5: Pairwise Correlation Current Board Structure with Past Firm Characteristics
Board Size
Prop IDs
Prop NEDs
Prop EDs
Lag1 Tobin’s Q
-0.0671***
-0.0072
-0.0262**
0.0257**
Lag1 MTB Ratio
0.0076
0.0174
-0.0217*
0.0211*
Lag1 LN Market Cap
0.5843***
0.3838***
0.2387***
-0.2422***
Lag1 Leverage
-0.0205*
0.0263**
0.0194
-0.0193
Lag1 CAPEX Ratio
-0.0244**
-0.0029
-0.0071
0.0076
Lag1 Cash Ratio
-0.1795***
-0.0623***
-0.0518***
0.0534***
Lag1 Idiosync. Risk
-0.2967***
-0.2436***
-0.1470***
0.1493***
51
Table 6: Simple Difference-In-Difference Analysis of the Introduction of ASX “Majority Board
Independence” Criterion on Firm Performance
This table illustrates the regression results of the Difference-in-Difference estimation approach of the following
model specification, where
is characterized as a dummy variable indicating 1 if the firm-year observation is
in the follow-up period and
is characterized as a dummy indicating 1 if the firm observation a treated firm
(i.e., a majority of independent board members):
1
 2
Tobin's Q  0  1D1   2 D2  3D1D2   , and
MTBR = 0  1D1   2 D2  3D1D2   .
ASX-listed firms have been defined as “Treated Firms” if they initially did not satisfy the majority criterion and
immediately adopted the majority independent board (50% +) criterion as of the “Corporate Governance
Principles and Recommendations” guidelines released in the financial year of 2003 (Treatment Year) and
maintained that status until at least 2011. “Control” firms, by contrast are firms that never adopted a majority of
independent directors during the full course of the database, 2001-2011, inclusive. For conducting the
Difference-In-Difference regression analysis in model specification (1), we use 235 “Treated” and 1,530
“Control” firm-year observations in the Baseline period and 847 “Treated” and 5,411 “Control” firm-year
observations in the Follow-Up period. In model specification (2) utilizing the Market-to-Book ratio, we use 235
“Treated” and 1,530 “Control” firm-year observations in the Baseline period and 847 “Treated” and 5,406
“Control” firm-year observations in the Follow-Up period The Outcome Variables, Tobin’s Q and Market-toBook have been winsorized at the 1% tails to avoid extremal skewness. The superscripts ***, **, * illustrate 1%,
5%, and 10% statistical significance, respectively. Absolute t-values are shown in brackets below each
coefficient.
Baseline Period (2001 - 2002)
Variable
Follow-Up Period (2003 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
Tobin’s Q
(Winsorized)
1.750***
(27.77)
2.135***
(13.28)
0.384**
(2.23)
2.268***
(67.69)
2.041***
(24.09)
-0.227**
(2.50)
-0.613***
(3.14)
R-Squared
F-Statistic
0.0064
18.11
# of Observations
8,023
3.273***
(12.22)
1.124***
( 3.91)
2.694***
( 48.25)
3.077***
( 21.81)
0.382**
( 2.52)
-0.742**
( 2.28)
PANEL A: Outcome Variable:
PANEL B: Outcome Variable:
Market-to-Book
(Winsorized)
2.148***
(20.47)
R-Squared
F-Statistic
# of Observations
0.0043
12.67
8,018
52
Table 7: Difference-in-Difference Analysis – Propensity Score Matching
This table illustrates the regression results of the first-step propensity score matching approach for matching treated with
non-participating control firms in the DID framework based on pre-treatment firm size characteristics, Log of Total Assets
Ln TA . First-step kernel-based weights for constructing the propensity scores are computed by the following Probit
model:
Majority Board Independence Adoption Indicator  0  1Ln TA  
The first-step kernel based weights are then manually adjusted for matching treated with similar firm sized untreated
control firms conditional on similar 2-digit SIC industry codes at each given pre-treatment fiscal year. The superscripts ***,
** *
, illustrate 1%, 5%, and 10% statistical significance, respectively. Absolute t-values are shown in brackets below each
coefficient.
Variable
Majority Board Independence Adoption Indicator
0.203***
(12.00)
-4.860***
(15.19)
Ln Total Assets
Constant
# of Observations
LR Chi-Squared
Pseudo R-Squared
1,765
155.97
0.1126
53
Table 8: Difference-in-Difference Analysis utilizing Propensity Scoring of the Introduction of ASX
“Majority Board Independence” Criterion (Treatment Adoption in 2003)
This table illustrates the regression results of the Difference-in-Difference estimation approach applied to 172 firms that became
“treated” in 2003 utilizing Tobin’s Q, Market-to-Book and Industry-Adjusted ROA as the performance criteria, respectively, in Panel’s
A, B, and C. The following model specification is used, where D1 is characterized as a dummy variable indicating 1 if the firm-year
observation is in the follow-up period, D2 is characterized as a dummy indicating 1 if the firm observation a treated firm (i.e., a majority
of independent board members), and TA represents Total Assets (i.e., Firm Size):
Tobin ' s Q / Market  to  Book / ROA  1D1   2 D2  3D1D2   4 Ln(TA)  5 D1Ln(TA)  6 D2 Ln(TA)  7 D1D2 Ln(TA)  8 Leverage 
9 D1Leverage  10 D2 Leverage  11D1D2 Leverage  12CAPAX  13 D1CAPAX  14 D2CAPAX 
15 D1D2CAPAX  16Cash  17 D1Cash  18 D2Cash  19 D1D2Cash   20 Financial   21D1Financial 
 22 D2 Financial   23D1D2 Financial  24 Mining  25 D1Mining  26 D2Mining  27 D1D2 Mining   .
ASX-listed firms have been defined as ‘Treated Firms” if they immediately adopted the majority independent board (50% +) criterion as
of the “Corporate Governance Principles and Recommendations” guidelines released in the financial year of 2003 (Treatment Year).
Each “Treated Firm” observation has been matched with “Control Firms” that never adopted the criterion using a kernel-density based
propensity score algorithm conditional on the basis of pre-treatment firm size and 2-digit SIC sector affiliation similarities. The
propensity scores have been estimated using a Probit model specification. We employ kernel-based matching techniques similar to
Heckman, Ichimura, and Todd (1997). Kernel-density based propensity score matching algorithm provides non-parametric matching
estimators that use weighted averages of all control group individuals to maximize the number of constructed counterfactual outcomes.
For conducting the Difference-in-Difference regression analysis in Panels A and B we use 235 “Treated” and 1,530 “Control” firm-year
observations in the Baseline period and 847 “Treated” and 5,339 “Control” firm-year observations in the Follow-Up period. For
conducting the ROA Difference-in-Difference regression analysis in Panel C we use 218 “Treated” and 1,497 “Control” firm-year
observations in the Baseline period and 805 “Treated” and 5,408 “Control” firm-year observations in the Follow-Up period. Propensity
Score Balancing test have been conducted to ensure good matching quality. For the below regression, a full set of year dummies are used
and all standard errors are corrected for heteroskedasticity. The Outcome Variables, Tobin’s Q in Panel A, the Market-to-Book ratio in
Panel B, and Industry-Adjusted ROA in Panel C have been winsorized at the 1% tails to avoid extremal skewness. The superscripts ***,
**, * illustrate 1%, 5% and 10% statistical significance, respectively. Absolute t-values are shown in brackets below each coefficient.
Panel A: Tobin’s Q Ratio (Treatment Adoption 2003)
Baseline Period (2001 - 2002)
Variable
Follow-Up Period (2003 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
5.136***
( 7.82)
4.897***
(3.25)
-0.239
(0.15)
7.608***
(13.83)
2.552***
(4.50)
-5.056***
(6.82)
-4.817***
(2.67)
-0.240***
( 6.69)
0.840***
( 5.29)
2.145***
(4.73)
1.403***
( 3.94)
-0.064
(0.31)
-0.355***
(3.36)
0.2860
29.71
7,823
-0.190**
( 2.22)
1.320
( 1.35)
2.903
(1.49)
3.995***
( 3.13)
-0.461
(1.17)
-1.480***
(4.81)
0.050
( 0.54)
0.479
( 0.49)
0.758
(0.38)
2.591*
( 1.96)
-0.397
(0.89)
-1.124***
(3.36)
-0.403***
( 13.91)
1.995***
( 14.13)
0.714***
(3.84)
2.953***
( 12.88)
-0.278***
(2.84)
0.297***
( 3.30)
-0.053*
( 1.90)
0.035
( 0.13)
1.683***
(3.07)
3.737***
( 6.61)
-0.513***
(4.53)
-0.675***
(4.81)
0.349***
( 8.94)
-1.959***
(6.24)
0.969*
(1.68)
0.783
( 1.29)
-0.234
(1.57)
-0.972***
(5.82)
0.299***
( 2.97)
-2.439**
(2.35)
0.211
(0.10)
-1.807
(1.24)
0.162
( 0.35)
0.152
(0.41)
Outcome Variable:
Tobin’s Q (Winsorized)
Control Variables:
Ln Total Assets
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm Dummy
Mining Firm Dummy
R-Squared
F-Statistic
# of Observations
54
Table 8: Difference-in-Difference Analysis utilizing Propensity Scoring of the Introduction of
ASX “Majority Board Independence” Criterion in 2003 (Cont.)
Panel B: Market-to-Book Ratio (Treatment Adoption 2003)
Baseline Period (2001 - 2002)
Variable
Follow-Up Period (2003 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
5.500***
(3.30)
11.821**
(2.36)
6.321
(1.20)
11.386***
(5.25)
5.956***
(4.05)
-5.429**
(2.32)
-11.750**
(2.03)
-0.204**
(2.05)
0.485
( 0.62)
2.476
(1.33)
1.603*
( 1.90)
-0.707*
(1.87)
-0.698*
(1.70)
0.053
9.44
7,818
-0.642**
(2.15)
8.684***
( 2.70)
5.025
(1.13)
5.749***
(2.60)
-0.891
(1.33)
-2.766***
(4.40)
-0.437
(1.39)
8.199**
( 2.48)
2.548
(0.53)
4.145*
(1.75)
-0.183
(0.24)
-2.067***
(2.75)
-0.518***
(4.58)
-0.553
(1.45)
2.129**
(2.50)
2.132***
(3.13)
-0.686**
(2.47)
0.185
(0.58)
-0.305***
(3.98)
6.038***
(5.09)
2.260*
(1.88)
5.166***
(5.10)
-1.033***
(4.11)
-0.486
(1.10)
0.212
(1.64)
6.592***
(5.30)
0.130
(0.09)
3.034**
(2.52)
-0.346
(0.93)
-0.672
(1.22)
0.649*
(1.91)
-1.607
(0.46)
-2.418
(0.48)
-1.111
(0.42)
-0.162
(0.19)
1.394
(1.50)
Outcome Variable:
MTBR
(Winsorized)
Control Variables:
Ln Total Assets
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm
Dummy
Mining Firm
Dummy
R-Squared
F-Statistic
# of Observations
Panel C: Industry-Adjusted Return on Assets (ROA) (Treatment Adoption 2003)
Baseline Period (2001 - 2002)
Variable
Follow-Up Period (2003 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
-1.584***
(10.09)
-0.785***
(4.86)
0.799***
(4.13)
-0.900***
(10.25)
-0.646***
(3.64)
0.254
(1.31)
-0.545**
(2.01)
0.099***
(11.74)
-0.277***
(3.59)
-0.643***
(2.74)
-0.092
(1.17)
-0.087**
(2.30)
0.099**
(2.08)
0.3270
20.28
7,558
0.050***
(5.96)
-0.278***
(3.99)
0.652**
(2.25)
0.404***
(0.147)
-0.092**
(2.31)
-0.001
(0.02)
-0.049***
(-4.94)
-0.001
(0.01)
1.295***
(3.47)
0.496***
(2.99)
-0.004
(0.09)
-0.100
(1.56)
0.068***
(14.82)
-0.527***
(10.48)
-0.543***
(3.19)
-0.276***
(4.95)
-0.091***
(4.93)
0.009
(0.41)
0.045***
(5.57)
-0.270***
(4.44)
0.031
(0.40)
0.067
(0.53)
-0.176***
(5.41)
-0.013
(0.56)
-0.023**
(2.52)
0.257***
(3.49)
0.575***
(3.08)
0.343**
(2.49)
-0.084**
(2.27)
-0.022
(0.68)
0.026**
(1.97)
0.259**
(2.04)
-0.720*
(1.72)
-0.153
(0.71)
-0.079
(1.20)
0.077
(1.07)
Outcome Variable:
Industry-Adjusted
ROA (Winsorized at
1% tails)
Control Variables:
Firm Size
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm
Dummy
Mining Firm Dummy
R-Squared
F-Statistic
# of Observations
55
Table 9: Difference-in-Difference Analysis utilizing Market-Based Criteria and Propensity Scoring of the
Introduction of ASX “Majority Board Independence” Criterion (Treatment Adoption in 2004)
This table illustrates the regression results of the Difference-in-Difference estimation approach applied to the 107 firms that became
“treated” in 2004. The following model specification is adopted, where D1 is characterized as a dummy variable indicating 1 if the firmyear observation is in the follow-up period, D2 is characterized as a dummy indicating 1 if the firm observation a treated firm (i.e., a
majority of independent board members), and TA represents Total Assets (i.e., Firm Size):
Tobin ' s Q / Market  to  Book / ROA  1D1   2 D2  3D1D2   4 Ln(TA)  5 D1Ln(TA)  6 D2 Ln(TA)  7 D1D2 Ln(TA)  8 Leverage 
9 D1Leverage  10 D2 Leverage  11D1D2 Leverage  12CAPAX  13 D1CAPAX  14 D2CAPAX 
15 D1D2CAPAX  16Cash  17 D1Cash  18 D2Cash  19 D1D2Cash   20 Financial   21D1Financial 
 22 D2 Financial   23D1D2 Financial  24 Mining  25 D1Mining  26 D2Mining  27 D1D2 Mining   .
ASX-listed firms have been defined as “Treated Firms” if they adopted the majority independent board (50% +) in the financial year of
2004 (Treatment Year). Each “Treated Firm” observation has been matched with “Control Firms” that never adopted the criterion using a
kernel-density based propensity score algorithm conditional on the basis of pre-treatment firm size and 2-digit SIC sector affiliation
similarities. The propensity scores have been estimated using a Probit model specification. We employ kernel-based matching techniques
similar to Heckman, Ichimura, and Todd (1997). Kernel-density based propensity score matching algorithm provides non-parametric
matching estimators that use weighted averages of all control group individuals to maximize the number of constructed counterfactual
outcomes. For conducting the Difference-in-Difference regression analysis in Panels A and B we use 380 “Treated” and 2,217 “Control”
firm-year observations in the Baseline period and 872 “Treated” and 4,508 “Control” firm-year observations in the Follow-Up period. In
Panel C we use 365 “Treated” and 2,166 “Control” firm-year observations in the Baseline period and 847 “Treated” and 4,567 “Control”
firm-year observations in the Follow-Up period. Propensity Score Balancing test have been conducted to ensure good matching quality.
For the below regression, a full set of year dummies are used and all standard errors are corrected for heteroskedasticity. The Outcome
Variables, Tobin’s Q in Panel A, the Market-to-Book ratio in Panel B, and Industry-Adjusted ROA in Panel C have been winsorized at
the 1% tails to avoid extremal skewness. The superscripts ***, **, * illustrate 1%, 5% and 10% statistical significance, respectively.
Absolute t-values are shown in brackets below each coefficient.
Panel A: Tobin’s Q Ratio (Treatment Adoption 2004)
Baseline Period (2001 - 2003)
Variable
Follow-Up Period (2004 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
2.447***
(4.42)
6.616***
(4.80)
4.169***
(2.79)
4.304***
(8.69)
4.360***
(5.72)
0.056
(0.06)
-4.112**
(2.36)
-0.087***
(2.69)
0.845***
( 3.06)
2.877***
(2.99)
2.229***
(4.54)
-0.120
(0.99)
-0.410***
(2.79)
0.2751
18.91
7,572
-0.321***
(3.84)
1.917**
( 2.05)
3.702
(1.52)
1.233*
(1.73)
0.190
(0.47)
-1.049***
(2.98)
-0.234***
(2.60)
1.071
( 1.10)
0.825
(0.32)
-0.996
(1.15)
0.311
(0.73)
-0.639*
(1.68)
-0.220***
( 8.82)
2.098***
( 9.74)
2.647***
(3.07)
3.834***
(10.60)
-0.306***
(3.51)
-0.266**
(2.17)
-0.181***
(4.03)
1.096**
( 2.13)
1.093***
(3.67)
1.333***
(2.79)
-0.249
(1.52)
0.326
(1.53)
0.038
(0.74)
-1.002*
(1.80)
-1.553*
(1.71)
-2.501***
(4.17)
0.057
(0.31)
0.592**
(2.41)
0.272***
(2.63)
-2.073*
(1.84)
-2.378
(0.86)
-1.504
(1.43)
-0.253
(0.54)
1.232***
(2.72)
Outcome Variable:
Tobin’s Q (Winsorized)
Control Variables:
Ln Total Assets
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm Dummy
Mining Firm Dummy
R-Squared
F-Statistic
# of Observations
56
Table 9: Difference-in-Difference Analysis utilizing Propensity Scoring of the Introduction of
ASX “Majority Board Independence” Criterion, Treatment Adoption 2004 (Cont.)
Panel B: Market-to-Book Ratio (Treatment Adoption 2004)
Baseline Period (2001 - 2003)
Variable
Follow-Up Period (2004 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
2.294**
(2.35)
1.320
(0.70)
-0.973
(0.45)
3.787***
(3.81)
-0.620
(-0.53)
-4.408***
(2.92)
-3.434
(1.31)
-0.063
(1.07)
1.584**
(2.37)
4.060*
(1.92)
3.167***
(4.29)
-0.463**
(2.47)
-1.055***
(3.19)
0.1004
9.29
7,567
0.091
(0.96)
-2.533***
(4.39)
7.213**
(2.23)
1.080
(1.07)
-0.409
(0.65)
-1.817***
(3.51)
0.155
(1.38)
-4.118***
(4.65)
3.152
(0.82)
-2.086*
(1.67)
0.054
(0.08)
-0.762
(1.24)
-0.105**
(2.11)
-1.064***
(4.13)
4.935***
(3.13)
4.043***
(6.87)
-0.939***
(5.60)
-1.126***
(5.21)
0.124**
(2.07)
-0.515
(1.23)
1.680***
(3.80)
1.791**
(2.48)
-0.363
(1.41)
.493
( 1.27)
0.229***
(2.94)
0.548
(1.11)
-3.254**
(1.99)
-2.252**
(2.42)
0.575*
(1.87)
1.619***
( 3.65)
0.073
(0.54)
4.667***
(4.61)
-6.407
(1.53)
-0.166
(0.11)
0.521
(0.72)
2.382***
( 3.14)
Outcome Variable:
MTBR (Winsorized)
Control Variables:
Ln Total Assets
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm Dummy
Mining Firm Dummy
R-Squared
F-Statistic
# of Observations
Panel C: Industry-Adjusted ROA (Treatment Adoption 2004)
Baseline Period (2001 - 2003)
Variable
Follow-Up Period (2004 - 2011)
Control
Treated
Difference
Control
Treated
Difference
DID Impact
-1.252***
(7.54)
-0.245
(1.31)
1.007 ***
(4.02)
-1.055***
(10.62)
-0.759***
(4.13)
0.295
(1.41)
-0.711**
(2.18)
0.085***
(9.03)
-0.456***
(6.21)
-0.592***
(2.83)
-0.234**
(2.56)
-0.101***
(3.00)
0.123***
(2.92)
0.3655
22.32
7,814
0.021**
(2.21)
0.005
(0.09)
-0.771*
(1.72)
-0.079
(0.98)
-0.072
(1.38)
0.016
(0.21)
-0.064***
(4.79)
0.462***
(4.66)
-0.178
(0.36)
0.155
(1.27)
0 .029
(0.47)
-0.107
(1.20)
0.076***
(14.92)
-0.546***
(11.39)
-0.161**
(2.07)
-0.315***
(4.73)
-0.086***
(5.02)
-0.022
(1.13)
0.059***
(5.78)
-0.484***
(5.80)
-0.183***
(3.10)
-0.097
(1.15)
-0.133***
(4.43)
-0.090***
(3.15)
-0.016
(1.45)
0.061
(0.64)
-0.021
(0.22)
0.218**
(2.02)
-0.047
(1.37)
-0.068*
(1.95)
0.047***
(2.69)
-0.401***
(2.90)
0.156
(0.31)
0.063
(0.39)
-0.076
(1.07)
0.038
(0.40)
Outcome Variable:
Industry-Adjusted ROA
(Winsorized at 1% tails)
Control Variables:
Firm Size
Leverage
CAPEX Ratio
Cash Ratio
Financial Firm Dummy
Mining Firm Dummy
R-Squared
F-Statistic
# of Observations
57
Table 10: Dynamic System Generalized Methods of Moments (GMM) Estimation of
Board Characteristics
The independent variable yi, t in Panel A is the Tobin’s Q ratio, in Panel B the Market-to-Book ratio, and in
Panel C the Industry-Adjusted Return on Assets (ROA) of an ASX-listed company at a given year between
the financial year-ends of 2001 and 2011. The matrices Xi , t , Zi , t , and Di , t include company board-specific
variables of interest, a set of firm-specific control variables, and a set of instrumental variables, respectively.
For the regressions set out below, a full set of year dummies are used and all standard errors are corrected for
heteroskedasticity. The Outcome Variables, Tobin’s Q in Panel A, the Market-to-Book ratio in Panel B, and
Adjusted ROA in Panel C, have been winsorized at the 1% tails to avoid extremal skewness. The superscripts
*** ** *
, , illustrate 1%, 5%, and 10% statistical significance, respectively. Absolute t-values are shown in
brackets below each coefficient. The Arellano-Bond tests examine for first-order and second-order serial
correlation in the first-differenced residuals under the Null of no serial correlation. The Hansen
Overidentification Test tests for the validity of instruments the used under the Null that a given employed set
of instruments is valid. The Difference in Hansen Test for Exogeneity tests for exogeneity in a given set of
instruments used in the level-equation under the Null of exogenous instruments.
Panel A: Tobin’s Q Ratio
Tobin’s Q (Winsorized at 1% tails)
Variable
(1)
(2)
(3)
(4)
-0.733**
(2.41)
-0.967**
(2.50)
-0.698**
(2.34)
-0.643**
(2.22)
-0.644**
(1.96)
Variables of Interest:
Ln Board Size
Proportion IDs
Top 200

-2.197***
(2.96)
Proportion IDs
-1.475*
(1.76)
Majority Independence Dummy
Top 200

-2.730**
(2.33)
Majority Independence
Control Variables:
0 .075
(1.00)
0.923***
(5.83)
3.098***
(5.71)
4.746**
(2.04)
17.720***
(8.66)
14.850**
(2.49)
0.443
(0.03)
-2.522
(1.36)
0.077
(1.12)
0.951***
(6.43)
3.117***
(6.17)
4.284**
(2.03)
16.893***
(7.98)
16.150***
(2.87)
-4.808
(0.45)
-2.519
(1.50)
0.076
(0.95)
0.927***
(5.51)
2.920***
(5.24)
4.775*
(1.92)
16.223***
(6.93)
14.435**
(2.53)
-8.470
(0.58)
-3.905
(1.59)
0.063
(0.89)
0.945***
(6.13)
2.986***
(6.03)
3.874*
(1.85)
14.884***
(6.68)
16.741***
(2.90)
-15.646
(0.90)
-3.775
(1.46)
-18.226***
(4.35)
-17.831***
(5.53)
-15.890***
(4.49)
-15.058***
(3.54)
# of Observations
6,728
6,728
6,728
6,728
F-Statistic
10.31
11.24
10.40
11.02
Arellano-Bond test AR1 (p-value)
0.001
0.001
0.001
0.000
Arellano-Bond test AR2 (p-value)
0.016
0.016
0.023
0.022
Hansen Test for Overidentification (p-value)
0.532
0.782
0.219
0.756
Diff in Hansen Test for Exogeneity (p-value)
0.329
0.584
0.109
0.557
Lag1 Tobin’s Q
Ln Market Value Equity
Leverage
CAPEX Ratio
Cash Ratio
Idiosyncratic Risk
Financial Firm Dummy
Mining Firm Dummy
Constant
58
Table 10: Dynamic System Generalized Methods of Moments (GMM) Estimation of
Board Characteristics (Cont.)
Panel B: Market-to-Book Ratio
Variable
Market-to-Book Ratio (Winsorized at 1% tails)
(1)
(2)
(3)
(4)
-0.474
(0.97)
-1.596**
(2.32)
-0.512
(1.05)
-0.261
(0.53)
-0.341
(0.68)
Variables of Interest:
Ln Board Size
Proportion IDs
Top 200

-3.319***
(2.55)
Proportion IDs
-4.382*
(1.66)
Majority Independence Dummy
Top 200

-4.950**
(1.99)
Majority Independence
Control Variables:
-0.025
(0.26)
1.494***
(4.82)
-2.266
(0.50)
5.444*
(1.74)
18.547***
(4.75)
24.542**
(2.03)
-20.467
(0.87)
-9.461
(1.08)
-0.008
(0.09)
1.498***
(4.85)
-1.749
(0.37)
5.025*
(1.81)
17.461***
(4.32)
26.398*
(1.88)
-25.962
(0.94)
-8.577
(0.90)
-0.0033
(0.32)
1.739***
(4.16)
-4.643
(0.75)
6.306*
(1.69)
15.641***
(3.26)
28.070**
(2.02)
-48.287
(1.30)
-17.861
(1.30)
0.002
(0.03)
1.451***
(4.75)
-1.874
(0.32)
5.010*
(1.72)
15.256***
(3.19)
30.099*
(1.74)
-45.648
(1.10)
-10.931
(0.96)
-21.716***
(3.67)
-21.210***
(3.15)
16.498*
(1.73)
-15.941
(1.46)
# of Observations
6,719
6,719
6,719
6,719
F-Statistic
7.58
7.93
6.96
7.73
Arellano-Bond test AR1 (p-value)
0.000
0.000
0.000
0.000
Arellano-Bond test AR2 (p-value)
0.018
0.037
0.040
0.108
Hansen Test for Overidentification (p-value)
0.571
0.702
0.633
0.546
Diff in Hansen Test for Exogeneity (p-value)
0.571
0.702
0.633
0.546
Lag1 MTBR
Ln Market Value Equity
Leverage
CAPEX Ratio
Cash Ratio
Idiosyncratic Risk
Financial Firm Dummy
Mining Firm Dummy
Constant
59
Table 10: Dynamic System Generalized Methods of Moments (GMM) Estimation of
Board Characteristics (Cont.)
PANEL C: Industry-Adjusted Return on Assets (ROA)
Variable
Industry-Adjusted ROA
(1)
(2)
(3)
(4)
0.012
(0.31)
-0.067**
(2.52)
0.050
(1.31)
0.016
(0.40)
0.035
(0.93)
Variables of Interest:
Ln Board Size
Proportion IDs
-0.072***
(2.97)
Top 200  Proportion IDs
-0.052**
(2.25)
Majority Independence Dummy
-0.075**
(2.32)
Top 200  Majority Independence
Control Variables:
0.048
(0.90)
0.068***
(8.23)
-0.533***
(6.29)
-0.371***
(3.41)
-0.383***
(3.74)
-0.013
(0.02)
-0.004
(0.02)
0.109
(1.15)
-0.906***
(5.72)
0.053
(0.82)
0.067***
(7.14)
-0.547***
(6.21)
-0.381***
(3.48)
-0.412***
(3.81)
0.252
(0.45)
-0.109
(0.67)
0.077
(0.86)
-0.928***
(5.36)
0.048
(0.84)
0.070***
(7.96)
-0.531***
(6.25)
-0.375***
(3.51)
-0.379***
(3.65)
0.167
(0.29)
-0.023
(0.12)
0.103
(1.08)
-.948***
(5.72)
0.051
(0.80)
0.068***
(6.81)
-0.557***
(6.53)
-0.386***
(3.65)
-0.423***
(3.88)
0.320
(0.57)
-0.027
(0.16)
0.081
(0.84)
-.928***
(5.22)
# of Observations
6,635
6,635
6,635
6,635
F-Statistic
13.59
12.00
13.55
12.88
Arellano-Bond test AR1 (p-value)
0.000
0.000
0.000
0.000
Arellano-Bond test AR2 (p-value)
0.794
0.777
0.794
0.793
Hansen Test for Overidentification (p-value)
0.000
0.000
0.000
0.000
Diff in Hansen Test for Exogeneity (p-value)
0.000
0.000
0.001
0.000
Lag1 Industry-Adjusted ROA
Ln Market Value Equity
Leverage
CAPEX Ratio
Cash Ratio
Idiosyncratic Risk
Financial Firm Dummy
Mining Firm Dummy
Constant
60
Table 11: Simple Probit-based Marginal Likelihood Effects of the Proportion of Independent Directors on CEO Turnover where
Performance is Below Industry Standard
This table illustrates the regression results of a simple Probit-based regression framework to measure a possible indication of reduced CEO replacement due to Majority Independence adoption and
the proportion of Independent directors. The variable “Treated Firms” is constructed based on ASX listed firms that do adopt and maintain majority board independence throughout the entire period
between 2003 and 2011. The CEO replacement indicator variable is equal to 1 if a company experiences a CEO turnover at a given fiscal year and 0 otherwise. All performance conditions are
annualized figures at Balance Sheet date and are aimed at capturing below industry standard firm performance on an annual basis using relative industry-adjusted stock returns, Tobin’s Q, Marketto-Book, ROA, and ROE measures. The regressions are conducted for firm-year observations where a particular firm experiences below industry standard performance in order to evaluate the
marginal probabilistic association of the proportion of independent directors on incumbent CEO replacement in scenarios where firms fail to deliver adequate performance for shareholders. The
simple model specifications (not controlling for confounding effects) are as follows:
(1) Pr  CEOTurnover  1 x  i 1  0  1 Treated Firms  
1
 0  1 Treated Firms  
1  exp 
  , and
2
2


1
 β + β1 Prop IDs  
(2) Pr  CEO Turnover  1 x   Φi=1  β0 + β1 Prop IDs  = 1+ exp  0
 .
2
2


All regressions are adjusted for error term heteroskedasticity and do control for year fixed effects by including year dummies, which remain omitted from the tables for brevity purposes.
superscripts ***, **, * illustrate 1%, 5% and 10% statistical significance, respectively. Absolute t-values are shown in brackets below each coefficient.
Variable
Conditions (annualized)
Treated Firms
Proportion IDs
The
CEO Turnover (Indicator Variable)
(1)
(2)
Negative industry adjusted
returns
-0.2495***
(4.10)
-0.8216***
(7.40)
Tobin’s Q – Industry Tobin’s Q <
0
-0.1509***
(3.06)
-0.6588***
(7.41)
(3)
(4)
(5)
MTBR – Industry MTBR < 0
ROA – Industry ROA < 0
ROE – Industry ROE < 0
***
***
-0.1637
(3.27)
-0.1339**
(2.02)
-0.1944
(2.86)
-0.6789***
(7.49)
-0.7980***
(6.41)
-0.7526***
(6.32)
1.1689***
(4.14)
1.1881***
(4.24)
1.9145***
(4.46)
1.9286***
(4.47)
1.6067***
(4.74)
1.6201***
(4.76)
1.5932***
(3.31)
1.5981***
(3.31)
1.6448***
(3.48)
1.6490***
(3.49)
# of Observations
2,517
2,517
4,210
4,210
4,052
4,052
2034
2034
2,244
2244
Wald Chi Squared-Statistic
72.12
111.53
60.01
104.27
75.02
118.73
41.24
72.85
39.83
74.68
Pseudo R-Squared
0.0325
0.0471
0.0291
0.0400
0.0294
0.0407
0.0266
0.0419
0.0260
0.0413
Constant
61
Table 12: Probit-based Marginal Likelihood Effects of the Proportion of Independent Directors on
CEO Turnover in Cases of Annual Below Industry Standard Firm-level Performance Measures
This table illustrates the regression results of a Probit-based regression framework to measure a possible indication of a reduced likelihood of
CEO turnover due to the proportion of Independent directors. The CEO turnover indicator variable is equal to 1 if a company experiences a
CEO turnover at a given fiscal year and 0 otherwise. All performance conditions are annualized figures at Balance Sheet date and are aimed at
capturing below industry standard firm performance on an annual basis using relative industry-adjusted stock returns, Tobin’s Q, Market-toBook, and ROA measures. The regressions are conducted for firm-year observations where a particular firm experiences below industry
standard performance in order to evaluate the marginal probabilistic association of the proportion of independent directors on incumbent CEO
replacement in scenarios where firms fail to deliver adequate performance for shareholders. The model specification controls for a variety of
important first lagged firm and first lagged CEO specific variables and is as follows:
1
 x  
Pr  CEO Turnover  1 x  i1  x   1  exp 
 ,
2
 2 
where
x  0  1Prop IDs + 2 Prop IDs * Rel.Return  LnTA  4 ROA  5Firm Risk +
6CAPEX  7 Firm Age + 8CEOOwnership + 9CEOTenure  10 Busy CEO.
All regressions are adjusted for error term heteroskedasticity and do control for year fixed effects by including year dummies, which remain
omitted from the tables for brevity purposes. The superscripts ***, **, * illustrate 1%, 5%, and 10% statistical significance, respectively.
Absolute t-values are shown in brackets below each coefficient.
Variable
Conditions (annualized)
Proportion IDs
Ln Total Assets
Return on Assets (ROA)
Firm Risk (Idiosyncratic)
CAPEX Ratio
Firm Age
CEO Ownership
CEO Tenure
Busy CEO
Constant
# of Observations
CEO Turnover (Indicator Variable)
(1)
(2)
(3)
(4)
(5)
Negative industry
adjusted returns
Tobin’s Q – Industry
Tobin’s Q < 0
MTBR – Industry
MTBR < 0
ROA – Industry
ROA < 0
ROE – Industry
ROE < 0
-0.4771**
(2.20)
0.0605*
(1.95)
-0.1272
(0.98)
3.5028
(1.58)
-0.1319
(0.59)
-0.0037
(0.98)
-0.4791
(1.41)
-0.0203***
(2.71)
0.1756*
(1.72)
-2.4315***
(3.77)
-0.5421***
(3.20)
0.0543**
(2.21)
0.0912
(0.69)
4.6750**
(2.55)
-0.2007
(0.60)
-0.0024
(0.97)
-0.8356***
(2.76)
-0.0117**
(2.39)
0 .0966
(1.22)
-2.2202***
(4.33)
-0.5167***
(3.10)
0.0631**
(2.49)
-0.1189
(1.04)
3.2168*
(1.76)
-0.3924
(1.24)
-0.0055*
(1.88)
-0.6397**
(2.06)
-0.0134***
(2.72)
0.1218
(1.52)
-2.2417***
(4.26)
-0.3832
(1.48)
0.0414
(1.28)
-0.3724
(1.59)
0.0412
(1.30)
4.8599*
(1.89)
0.0211
(0.17)
-0.0040
(1.00)
0.3831
(0.84)
-0.0064
(0.96)
0.1763
(1.49)
-2.1706***
(3.08)
3.6151
(1.50)
0.0619
(0.40)
-0.0057
(1.47)
-0.0116
(0.03)
-0.0095
(1.33)
0.0778
(0.68)
-2.0792***
(3.06)
1,693
686
806
961
1,725
Wald Chi Squared-Statistic
32.13
51.26
56.02
16.79
21.08
Pseudo R-Squared
0.0413
0.0385
0.0419
0.0278
0.0290
62
Table 13: Effect of Treatment on Mean Values and Mean-Value Differences
The following table illustrates the mean values for the specified treatment and control subgroups for the pre- and post-treatment periods as well as the
Difference-in-Mean values and Difference-in-Difference-in-mean values with their respective simple t-tests. All firms that ever become treated, 20032011, are included. The DID-in-mean values where computed in the following DID framework, where D1 is characterized as a dummy variable indicating
1 if the firm-year observation is in the follow-up period and D2 is characterized as a dummy indicating 1 if the firm observation is a treated firm:
*** ** *
, ,
variable x  D1  D2  D1D2 . The Difference-in-Mean values are constructed as Treated Mean – Control Mean observations. The superscripts
illustrate 1%, 5%, and 10% statistical significance, respectively.
Pre Treatment
Variable
# of Obs.
Board Committee Structures
Compensation Committee
No. of Directors
Proportion IDs
Mean ID PPS
No. of Meetings Held
Director Attendance Ratio
ID as Chairperson
Nominations Committee
No. of Directors
Proportion IDs
Mean ID PPS
No. of Meetings Held
Director Attendance Ratio
ID as Chairperson
Pay for Performance Sensitivities
CEO PPS
Top Executives PPS
Insider PPS
EDs PPS
NEDs PPS
IDs PPS
GDs PPS
All Directors PPS
CEO Compensation Variables
CEO Base Salary ($)
CEO Cash Bonus ($)
CEO Total Comp (inc. LTIP)
($)
Top Exec Compensation
Mean Top Exec Total Comp ($)
Director Fees
Mean Director fees ($)
Company Level Characteristics
Ln Market Cap
Ln Total Assets
Leverage Ratio
Idiosyncratic Risk
Cash Ratio
CAPEX Ratio
CEO Tenure
Busy CEO
Propn Busy Directors
Propn Female Directors
Institutional Ownership
(Top 20 Inst. Share Holdings)
Institutional Ownership
(Largest Inst. Share Holding)
Post Treatment
Treated
Control
Difference
Treated
Control
Difference
DID Impact
4,316
4,316
3,592
4,005
3,999
8,509
3.310
0.899
0.0002
8.7
0.976
0.243
2.785
0.834
0.0018
6.319
0.973
0.169
0.525***
0.064***
-0.0015
2.380***
0.002
0.074
3.252
0.916
0.0001
11.261
0.975
0.458
2.829
0.869
0.0024
7.073
0.977
0.310
0.423***
0.046***
-0.0023*
4.188***
-0.002
0.147***
-0.101
-0.018
-0.0007
1.807**
-0.005
0.073**
2,230
2,230
1,822
1,979
1,975
8,509
3.104
0.870
0.0001
9.137
0.988
0.111
2.586
0.762
0.008
7.322
0.979
0.041
0.518**
0.108***
-0.008
1.815
0.008
0.070***
3.219
0.907
0.0001
9.816
0.971
0.313
2.942
0.845
.0026
7.056
0.977
0.144
0.277***
0.061***
-0.0025
2.759***
-0.006
0.169***
-0.241
-0.046
0.0055
0.943
-0.014
0.099***
5,245
6,834
7,807
6,652
7,526
5,345
6,560
7,912
0.036
0.014
0.021
0.035
0.008
0.009
0.011
0.021
0.107
0.045
0.067
0.081
0.029
0.016
0.028
0.067
-0.071***
-0.031***
-0.045***
-0.046***
-0.021***
-0.006***
-0.016***
-0.045***
0.021
0.009
0.015
0.024
0.008
0.001
0.024
0.015
0.083
0.030
0.048
0.071
0.028
0.005
0.047
0.048
-0.061***
-0.021***
-0.032***
-0.047***
-0.019***
-0.004***
-0.022***
-0.032***
0.009
0.010**
0.013*
-0.0002
0.001
0.002
-0.006
0.013**
7,609
7,609
536,103
162,722
246,599
58,940
289,503***
103,782***
834,868
506,092
335,370
140,597
499,497***
365,494***
209,994***
261,712***
7,994
1,132,422
425,015
707,406***
2,156,323
735,298
1,421,024***
713,618***
7,998
601,507
359,893
241,614**
964,262
375,344
588,918***
347,304***
7,866
60,405
45,225
15,179***
102,440
56,921
45,519***
30,340***
8,047
8,284
8,155
8,237
8,155
8,285
4.961
8,519
8,519
8,509
19.488
19.578
0.437
0.031
0.134
0.067
5.644
0.404
0.589
0.059
17.344
17.452
0.411
0.051
0.182
0.088
4.591
0.245
0.432
0.029
2.143***
2.125***
0.026
- 0.019***
- 0.047***
- 0.021
1.052
0.159
0.156***
0.029***
20.394
20.375
0.434
0.027
0.152
0.073
6.838
0.427
0.663
0.075
17.865
17.685
0.423
0.044
0.209
0.097
7.056
0.242
0.471
0.039
2.528***
2.689***
0.010
- 0.017***
- 0.057***
- 0.023
-0.217
0.185***
0.191***
0.036***
0.385***
0.563***
-0.015
0.002
-0.010
-0.002
-1.270*
0.025
0.035
0.006
8,511
0.543
0.616
-0.072***
0.584
0.624
-0.040***
0.032**
8,519
0.165
0.269
-0.104***
0.176
0.241
-0.064***
0.039***
63
Figure 1 : Simple data sample diagnostics between 2001 and 2011
The following four graphs illustrate the average time-series trends in the proportion of Independent Directors, Tobin’s Q, Board Size and the Proportion of Executive Directors of ASX listed companies
between 2001 and 2011. The solid lines represent the mean values for Top ASX 200 companies and the dashed lines represent mean values for ASX listed companies, excluding the Top 200 companies.
Extremal values have been omitted to avoid outlying observational biases.
Panel A: Mean Proportion of Independent Directors
Panel B: Median Tobin's Q
0.7
1.8
0.6
1.6
0.5
1.4
0.4
0.3
1.2
0.2
1.0
0.1
0.8
0.0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
0.6
2001
Panel C: Mean Board Size
9
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2010
2011
Panel D: Mean Proportion of Executive Directors
0.34
8
0.32
7
0.30
6
0.28
0.26
5
0.24
4
0.22
3
0.20
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
2001
64
2002
2003
2004
2005
2006
2007
2008
2009
Figure 2: Graphs of Time Series of Independent Director and Gray Director Pay-forPerformance Sensitivity (PPS)
The following four graphs illustrate the average time-series trends in the Pay-For-Performance sensitivities of Independent
Directors and Gray Directors of ASX listed companies between 2001 and 2011. The solid lines represent the mean values for
Top ASX 200 companies and the dashed lines represent mean values for ASX listed companies, excluding the Top 200
companies. Extremal values have been omitted to avoid outlying observational biases.
Panel A: Mean Independent Director PPS
0.025
Top 200 ASX Listed
Not Top 200 ASX Listed
0.020
0.015
0.010
0.005
0.000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Panel B: Mean Gray Director PPS
0.08
0.07
0.06
0.05
0.04
0.03
0.02
Top 200 ASX Listed
0.01
Not Top 200 ASX Listed
0.00
2001
2002
2003
2004
2005
2006
65
2007
2008
2009
2010
2011
Figure 3: Time-Series Mean Performance Graphs for Treated versus Untreated Firms
The following graphs illustrate the average time-series trends in Treated versus Untreated Firm Tobin’s Q values of ASX listed
companies between 2001 and 2011. The solid lines represent the Tobin’s Q mean values for Treated ASX listed companies and the
dashed lines represent Tobin’s Q mean values for Untreated ASX listed companies. Extremal values have been omitted to avoid outlier
observational biases.
Panel A: Mean Tobin's Q values
3.5
3
2.5
2
1.5
1
Treated
0.5
Untreated
0
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Panel B: Mean Market-to-Book Ratio values
4.5
4
3.5
3
2.5
2
1.5
Treated
1
Untreated
0.5
0
2001
2002
2003
2004
2005
2006
66
2007
2008
2009
2010
2011
Figure 4: Time-Series Mean Performance Graphs for Treated versus Untreated Firms (Tobin’s Q Ratio)
The following graphs illustrate the average time-series trends in Treated versus Untreated Firm Tobin’s Q values of ASX listed companies between 2001 and 2011. The solid lines represent the Tobin’s Q
mean values for Treated ASX listed companies and the dashed lines represent Tobin’s Q mean values for Untreated ASX listed companies. Extremal values have been omitted to avoid outlying observational
biases.
Panel A: ASX Top 200 Mean Tobin's Q values
Panel B: ASX Non-Top 200 Mean Tobin's Q values
3
3.5
2.5
3
2.5
2
2
1.5
1.5
1
1
Treated & Top 200
0.5
0.5
Untreated & Top 200
0
Untreated & Non-Top 200
0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Panel C: ASX Top 200 Mean Tobin's Q values
(Relative to Industry Tobin's Q)
Panel D: ASX Top 500 to 200 Mean Tobin's Q values
(Relative to Industry Tobin's Q)
0.3
0.6
0.2
0.4
0.1
0.2
0
0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
-0.1
-0.2
-0.2
-0.4
-0.3
-0.6
Treated & Top 200
-0.4
Treated & Non-Top 200
Untreated & Top 200
-0.8
67
Treated & Non-Top 200
Untreated & Non-Top 200
Figure 5 : Treatment Conversion Schedule Over Time
200
Firm's "Treated"
100%
Cumul. "Treated" (% Ttl)
180
90%
78.5%
172
160
140
80%
70%
64.5%
59.8%
120
60%
54.7%
48.0%
100
107
50%
43.2%
39.3%
80
40%
33.0%
60
30%
67
21.9%
40
46
44
2.6%
20
20%
44
36
23
10%
22
13
0
2002
2003
2004
2005
2006
2007
68
2008
2009
2010
2011
0%
Figure 6 : Time-Varying Treatment Effects on CEO, Executive Member and Director Pay Variables, µ(Z) - µ(0)
The following figures present the time-varying policy effects of majority board independence adoption on various executive and director compensation variables. Figures A and B plot the expected difference
between being treated (joining the ASX reform of a majority independent board) for Z fiscal years, µ(Z), and the counterfactual case of being untreated (not joining the ASX reform of a majority independent
board), µ(0), for annual CEO and annual Top Executives (excl. CEO) Pay for Performance Sensitivity (PPS) measures, respectively. Figures C and D illustrate the expected difference between being treated for
Z fiscal years, µ(Z), and the counterfactual case of being untreated, µ(0), between the log of annual CEO Cash Compensation and the log of mean annual Director Fees, respectively. Dashed lines illustrate the
90% confidence intervals for the estimates.
Panel A: Time-Varying Treatment Effects on CEO PPS
1
0
1
-0.02
2
3
4
5
6
7
8
9
2
3
-0.04
4
-0.06
5
-0.08
6
-0.1
7
-0.12
8
-0.14
9
Panel C: Time-Varying Treatment Effects on log of
annual CEO cash-based compensation
1
1
2
3
0.8
4
0.6
5
0.4
6
0.2
7
0
8
1
2
3
4
5
6
7
8
9
9
-0.016***
(-5.45)
-0.031***
(-5.51)
-0.041***
(-5.68)
-0.048***
(-5.60)
-0.052***
(-5.22)
-0.056***
(-4.50)
-0.061***
(-3.41)
-0.068***
(-2.70)
-0.077***
(-2.72)
Panel B: Time-Varying Treatment Effects on Mean
Executive PPS
1
0
1
2
3
4
5
6
7
8
9
2
-0.01
3
-0.02
4
5
-0.03
6
-0.04
7
-0.05
8
-0.06
9
Panel D: Time-Varying Treatment Effects on log of
mean annual Director fees
0.174***
(10.14)
0.396***
(17.80)
0.566***
(24.99)
0.666***
(19.91)
0.703***
(18.42)
0.706***
(13.10)
0.703***
(8.62)
0.712***
(8.41)
0.734***
(6.99)
0.7
2
0.6
3
0.5
4
0.4
5
0.3
6
0.2
7
0.1
8
1
69
1
2
3
4
5
6
7
8
9
9
-0.008***
(-4.55)
-0.020***
(-3.97)
-0.029***
(-3.81)
-0.033***
(-3.59)
-0.032***
(-3.72)
-0.028***
(-3.59)
-0.024***
(-3.40)
-0.020***
(-2.76)
-0.018**
(-2.25)
0.119***
(13.02)
0.268***
(21.32)
0.385***
(32.15)
0.451***
(31.60)
0.474***
(19.96)
0.473***
(15.70)
0.470***
(10.77)
0.475***
(10.12)
0.489***
(8.84)
Figure 7: Time-Varying Treatment Effects of Maj. Board Independence on Tobin’s Q, µ(Z) - µ(0)
The following figures present the time-varying policy effects of majority board independence adoption on firm performance as measured by
the Tobin’s Q ratio. Panel A illustrates the difference in the expected immediate percentage change in Tobin’s Q between treated and
counterfactual firms’ observations for each fiscal year end after and including the official policy introduction year in 2003. Panels B and C
plot the expected difference between being treated (joining the ASX reform of a majority independent board) for Z fiscal years, µ(Z), and
the counterfactual case of being untreated (not joining the ASX reform of a majority independent board), µ(0), for annual Tobin’s Q and
annual Industry-Adjusted Tobin’s Q measures, respectively. Dashed lines illustrate the 90% confidence intervals for the estimates.
Panel A: Immediate Treatment Adoption Change (%) in Tobin´s Q
0.8
2003
2004
0.6
2005
0.4
2006
2007
0.2
2008
0
2003
2004
2005
2006
2007
2008
2009
2010
2011
-0.2
2009
2010
2011
-0.4
Panel B: Time-Varying Treatment Effects on Tobin´s Q
1
0.1
2
0
-0.1
1
2
3
4
5
6
7
8
3
9
4
-0.2
5
-0.3
6
-0.4
-0.5
7
-0.6
8
-0.7
9
Panel C: Time-Varying Treatment Effects on Industry Adjusted
Tobin´s Q
0.05
1
0
-0.05
2
1
2
3
4
5
6
7
-0.1
8
9
3
4
-0.15
5
-0.2
-0.25
6
-0.3
7
-0.35
8
-0.4
9
-0.45
70
-0.195**
(-2.23)
-0.126*
(-1.69)
-0.075
(-1.15)
-0.111
(-1.12)
-0.124
(-1.68)
-0.090
(-1.24)
-0.050
(-0.51)
0.014
(0.10)
0.149
(0.56)
0.033
(1.28)
0.014
(0.40)
-0.075**
(-2.00)
-0.164***
(-3.14)
-0.240***
(-2.82)
-0.307***
(-3.34)
-0.363***
(-3.30)
-0.405***
(-3.10)
-0.435***
(-3.43)
-0.046***
(-3.03)
-0.089***
(-3.44)
-0.120***
(-3.49)
-0.138***
(-3.03)
-0.148***
(-2.62)
-0.156*
(-1.89)
-0.167*
(-1.73)
-0.183
(-1.53)
-0.202
(-1.60)
Figure 8: Time-Varying Treatment Effects of Maj. Board Independence on MTBR, µ(Z) - µ(0)
The following figures present the time-varying policy effects of majority board independence adoption on firm performance as measured by the MTBR
(Market to Book) ratio. Panel A illustrates the difference in the expected immediate percentage change in MTBR between treated and counterfactual
firms’ observations for each fiscal year end after and including the official policy introduction year in 2003. Panels B and C plot the expected difference
between being treated (joining the ASX reform of a majority independent board) for Z fiscal years, µ(Z), and the counterfactual case of being untreated
(not joining the ASX reform of a majority independent board), µ(0), for annual MTBR and annual Industry-Adjusted MTBR measures, respectively.
Dashed lines illustrate the 90% confidence intervals for the estimates.
Panel A: Immediate Treatment Adoption Change (%) in MTBR
2003
1.5
2004
2005
1
2006
0.5
2007
2008
0
2003
2004
2005
2006
2007
2008
2009
2010
2011
-0.5
2010
2011
-1
Panel B: Time-Varying Treatment Effects on MTBR
1
0.4
2
0.2
3
0
-0.2
2009
4
1
2
3
4
5
6
7
8
9
5
6
-0.4
7
-0.6
8
-0.8
9
-1
Panel C: Time-Varying Treatment Effects on Industry Adjusted MTBR
1
0.2
2
0.1
3
0
-0.1
1
2
3
4
5
6
-0.2
7
8
9
4
5
-0.3
6
-0.4
-0.5
7
-0.6
8
-0.7
9
71
-0.512***
(-2.94)
-0.296
(-1.62)
-0.121
(-0.67)
-0.210
(-0.92)
-0.295
(-1.46)
-0.356*
(-1.83)
-0.411**
(-2.02)
-0.216
(-0.78)
0.509
(1.24)
0.095**
(2.02)
0.141**
(2.13)
0.085
(1.56)
-0.017
(-0.20)
-0.143
(-1.36)
-0.271
(-1.54)
-0.380**
(-2.08)
-0.462**
(-2.05)
-0.52**
(-2.12)
0.004
(0.17)
0.026
(0.56)
0.032
(0.50)
0.010
(0.14)
-0.035
(-0.38)
-0.094
(-0.81)
-0.154
(-1.27)
-0.210
(-1.19)
-0.262
(-1.24)