Formal and Real Authority in Organizations: An Empirical Assessment∗

Feng Li University of Michigan

Michael Minnis University of Michigan

Venky Nagar University of Michigan

Madhav Rajan Stanford University

May 2009

Analytical models such as Aghion and Tirole (1997) distinguish formal authority from real authority: a manager could be formally responsible for a decision, but in reality may acquiesce to her subordinate. Organizational theorists (Simon 1997) suggest that human communication settings are best suited to uncover real authority. We use the extent to which CEOs speak in conference calls to measure their real authority over top management. Using a large database of firm conference call transcripts, we find that our real authority measure is distinct from the CEO’s formal authority measures, and is significantly associated with organizational factors as theoretically predicted. CEOs with real authority also receive higher wages. The joint tests of our measure of real authority with real authority theories and labor market theories suggest that real authority is a distinct organizational feature that is measurable in large samples.

∗ We thank seminar participants at the University of Michigan and the University of Southern California for their comments. We also thank Paul Michaud for his significant assistance in programming and data management issues.

Formal and Real Authority in Organizations: An Empirical Assessment “Scientia potentia est (knowledge is power).” Francis Bacon, Meditationes Sacrae (1597) Heather Bellini (UBS Analyst): Hi, good morning, everybody. I just wanted to ask a question, Steve. Typically, revenue synergies in software deals have been elusive, at least that's what us and the industry would at least remember. Can you talk with us a little bit about the revenue synergies you would expect, why you would expect to get them, and over what time frame we could expect to see them, have this play out? Thank you. Steve Ballmer (Microsoft CEO): Yes. I am going to let Kevin take this.

Microsoft Corporation conference call on February 1, 2008.1

1. Introduction

A key component of organizational design that has received significant research attention is

the allocation of authority to employees (Fama and Jensen 1983; Harris and Raviv 2005; Roberts

2007). To measure authority, empirical analyses have typically relied on organizational charts (org-

charts), surveys, job titles, or statements of job responsibilities.2 Aghion and Tirole (1997) label all

these measures of delegation, decentralization, and management responsibilities as “formal

authority.” They then introduce an additional concept, “real authority”: complex management

decisions require subjective soft knowledge (Petersen and Rajan 1994; Stein 2002; Petersen 2004),

and the person with real authority is the person who possesses this knowledge. This person gains

real authority because subjective knowledge is not easily communicated and transferred through

the management chain, resulting in the individual with formal authority in many cases simply

“rubber-stamping” this person’s decision. Aghion and Tirole (1997) derive several testable

predictions about real authority and how it differs from formal authority, and then reveal a critical

1 Source: http://www.microsoft.com/presspass/press/2008/feb08/02-01Transcript.mspx. 2 See, for example, Baiman, Larcker, and Rajan (1995), Nagar (2002), Aggarwal and Samwick (2003), Moers (2006), Campbell, Datar and Sandino (2009), and Ortega (2009).

1

obstacle: “The key issue is, of course, the measurement of real authority” (p. 26, emphasis theirs).

This paper is an attempt to overcome that obstacle.

Aghion and Tirole’s (1997) concern is a substantive one: while org-charts, titles, and

formal job responsibilities are easily measured, it is difficult to extract via surveys or other means

the identities of those who possess subjective knowledge. The ambiguous, ineffable, soft nature of

subjective knowledge can cause survey responses to be unreliable and unverifiable (Bertrand and

Mullainathan 2001), and yet this knowledge is key to firm value (Li 2008; Petersen 2004; Tetlock,

Saar-Tsechansky and Macskassy 2008). Organizational theorists argue that human interactions

and communication channels such as meetings are the primary organizational mechanism by

which the management chain recognizes the presence of subjective knowledge and lets the owner

of this knowledge substantively influence the decision (Vancil 1978, Simon 1997). We therefore

focus on a publicly-available human communication setting, namely the setting of conference calls.

In addition to providing formal accounting and other firm performance reports, many

publicly traded firms host earnings conference calls each quarter during which management

describe the performance and strategy of the firm and field a dynamic question and answer session

with analysts (Kimbrough 2005). Conference calls are a significant source of information for

investors (Bushee, Matsumoto and Miller 2003, 2004), and analysts pay keen attention to softer

knowledge in these calls (e.g., Mayew and Venkatachalam 2008). We argue that conference calls

provide an excellent setting for measuring the extent to which real authority is spread across top

management.

Specifically, we propose that the extent to which the Chief Executive Officer (CEO) speaks

in these conference calls is a proxy for the extent to which he or she possesses subjective knowledge

2

about firm operations. Our assumption is that, unlike hard formal information, it is difficult to

prime a responder about subjective knowledge adequately in advance of a Q&A session. Further,

given the importance placed on the conference call by market participants, top management is

better off letting the person in the management team most comfortable with the relevant

subjective knowledge answer the analyst’s question (see the Microsoft quote above and the case

study in Appendix A). To the extent the speaker is indeed the possessor of subjective knowledge,

he by definition has real authority: another individual may have the formal authority, but has to

rely on the decision made by the person with the real authority. Knowledge thus confers power.3

We obtain machine readable texts of 17,400 conference calls from 2003-2007 and measure

the amount of communication conducted by the CEO, relative to others in top management. We

hypothesize and find that our measure of the CEO’s real authority is related to its theoretically

predicted determinants. Specifically, we predict (based on prior literature) and find that the CEO

has more real authority when the non-CEO management team has weaker monetary incentives,

the urgency of decision-making is lower, the tasks considered are more important, less technical or

innovative expertise is required, and the span of the CEO’s control is smaller. Second, we find

that these results obtain after controlling for measures of the CEO’s formal authority and CEO’s

individual characteristics such as ownership, tenure, prestige, and overconfidence. Real authority

thus appears to be a distinct organizational feature.

Our predictions above are joint tests of our assertion that conference calls can measure real

authority and of theories of real authority such as Aghion and Tirole (1997) and Baker, Gibbons

and Murphy (1999). We therefore provide additional economic evidence on the validity of our

3 Even in academic settings, for example, live seminars with interactive Q&A are the primary organizational mechanism to assess a researcher’s real contribution to a research paper.

3

measure of real authority. Under the assumption that well-functioning labor markets reward real

authority (Rosen 1982), we conduct a joint test of labor market efficiency and our measure of

authority. We show that wages are an increasing function of real authority, in that our measure of

CEO real authority is positively associated with the wages of the CEO relative to top management,

after accounting for a rich set of controls. Overall, we provide, to the best of our knowledge, the

first large sample tests of theories distinguishing real authority from formal authority.

The remainder of the paper is organized as follows. In the next section, we describe the

theory, the measure of real authority, and our hypotheses. In Section 3, we describe our data and

variable measurement, as well as provide a discussion of the descriptive statistics. In Section 4 we

discuss the results, and in Section 5 we provide the results of our robustness tests. Finally, we

conclude the paper in Section 6.

2. Hypothesis Development

2.1 Defining and Measuring Real Authority

In the course of their jobs, different employees in an organization interact with different

stakeholders, causing knowledge to be spread out across the firm (Roberts 2007). Managerial

decisions require this knowledge, leaving organizations with two design alternatives: decision rights

can be ceded to the subordinate personnel with the information, or knowledge can be transferred

from informed subordinates to those with the authority to make decisions (Jensen and Meckling

1992). The first alternative faces control costs: a self-interested subordinate may make a

suboptimal decision from the firm’s perspective. The second alternative faces knowledge transfer

costs: it is costly — both in terms of cash costs and personnel time — to transfer information.

4

If knowledge can be codified and transmitted, the organization can make assessments of

knowledge transfer costs and delegate accordingly. And in fact, organizations do so with formal

org-charts, job design, and delegation choices. However, in addition to such formal or hard

knowledge, there is also a nebulous, shifting, dynamic, subjective (or soft) knowledge that is

equally important for decision-making and firm value (Li 2008; Tetlock, Saar-Tsechansky and

Macskassy 2008). More important, this subjective knowledge is more difficult to communicate

and distribute (Petersen 2004; Stein 2002; Petersen and Rajan 1994). Most organizations

recognize and accommodate this subjective knowledge not by constantly shifting org-charts, but by

acquiescing to the decision suggested by the person with the knowledge. Petersen (2004) and

Petersen and Rajan (1994) provide several examples of subjective knowledge: a given employee may

have a good feel for the client’s prospects and needs, and thus may be best suited to build a

relationship with the client. Though the employee’s boss may be formally in charge of approving

loans to the client, he may acquiesce to the employee’s decision. This employee thus gains “real

authority” that is not evident in his or her formal org-chart position and job description.

The concept of real authority has been formalized both by game theorists (e.g., Aghion and

Tirole 1997; Baker, Gibbons and Murphy 1999) and organizational theorists.4 In his classic

treatise, Simon (1997, Ch. VII) writes at length about the overlapping roles of formal authority

and informal influence, the end result of which, as Vancil (1978, p. 62) notes, is that in many cases

it is impossible to say which manager really “made” the decision. More important, however,

organizational theorists have mapped the channels through which subjective knowledge causes

4 To illustrate real authority as an equilibrium phenomenon, Baker, Gibbons and Murphy (1999) build a reputation-based repeated principal-agent model. In this model’s equilibrium, the uninformed principal never overturns the proposal of the agent and the agent only proposes projects with high expected outcomes for the principal (i.e., does not abuse the informal authority). By approving all proposed projects, the principal has thus ceded informal (or real) authority to the agent.

5

ambiguity in the autonomy underlying org-charts. Both Simon and Vancil argue that the decision-

making process in such settings is typically one of human sociality where managers talk and

communicate in formal and informal meetings and other social intercourse settings. In the course

of such communication processes, the management chain gauges and assesses the presence of

subjective knowledge, and the person with the knowledge is ultimately able to influence the

decision of the manager with the formal authority. As a result, communication processes in the

organization are intricately tied to decision-making processes (Simon 1997, Ch. VIII).

The above arguments suggest that a natural measure of real authority is to examine a

communication setting where the researcher can observe the flow of conversation and social

intercourse among management. 5 While much of the management’s social intercourse is private,

new media and new communications technology provide a clean setting for an external researcher,

namely conference calls. In particular, top management routinely participates in these calls. We

use this setting to discern how real authority is distributed among top management on the basis of

their speaking patterns.

Conference calls are an essential form of communication and disclosure between

management and capital providers for many public firms.6 During the 1990’s and 2000’s, firms

5 The Wall Street Journal reports an interesting example of how the level of communication reflects the extent of the real authority of an individual. James Lambright was the head of the U.S. Export-Import Bank (a position that reports directly to the President of the United States) when he was asked to take the role of chief investment officer for the Troubled Asset Relief Program (a position that reports to an assistant secretary of the Treasury). While this job transition surely resulted in the loss of formal authority from an organizational chart perspective, Mr. Lambright significantly increased his real authority becoming, according to the article, “one of the most powerful men in American finance.” The article cites the extent of communication Mr. Lambright engages in during conference calls with executives as the measure of power, noting that “top executives regularly call him and his team for advice.” While Mr. Lambright is multiple layers down the org-chart from the Treasury Secretary, the executives seek to talk to the individual with real authority (Solomon 2009). 6 As another example of the link between communication and real authority is the reported need for communication between stakeholders and American International Group’s (AIG) three newest directors. When the U.S. government took a substantial stake in AIG, it appointed three new directors to the company’s Board. Since their appointment, The New York Times has reported that the three have “remained invisible to the public,” with little communication.

6

increasingly utilized conference calls as a broad, inexpensive, and timely disclosure forum (Bushee,

Matsumoto and Miller 2003).7 Further, after the Securities and Exchange Commission enacted

Regulation FD in October 2000, most companies made their conference calls openly available,

rather than limited to participants who were invited by the firm to participate (Bushee, Matsumoto

and Miller 2004).

Research examining conference calls finds that these events are not simply a regurgitation

of other corporate communications (e.g., press releases), but rather provide new and separate

information. Bowen, Davis and Matsumoto (2002) find that conference calls improve analysts’

ability to accurately forecast firms’ earnings. Frankel, Johnson and Skinner (1999) and Bushee,

Matsumoto and Miller (2003) find altered trading patterns during conference calls, suggesting that

the market receives new information during the call. Kimbrough (2005) finds that conference

calls can mitigate investor under-reaction to earnings announcements which may be partly

responsible for the well documented post-earnings announcement drift. In sum, conference calls

are a high profile form of communication upon which future capital providers or even customers

form opinions.

What kind of information is released in conference calls? While data such as firm

financial performance is discussed, recent research indicates that it is the more subjective

information (e.g., the tone of the discussion, the inflection of the executives’ voices, and how

management “feels” about achieving targets) that investors and analysts play close attention to;

after all, analysts already have the hard knowledge from the financial reports (Mayew and This lack of communication has fostered concerns about who “really is in charge” of AIG. Interestingly, the article reports that the three trustees will reportedly make “their public debut” in a forum very similar to the one in which we examine in this study—the annual shareholder’s meeting (Andrews 2009). 7 Bushee, Matsumoto and Miller (2003) note that the increase in the number of conference calls prompted several studies to examine the impact of conference calls on trading (Frankel, Johnson and Skinner 1999) and on analysts’ forecasts (Bowen, Davis and Matsumoto 2002).

7

Venkatachalam 2008).8 As a result, we argue that it is in the management’s best interest to let the

manager with the subjective knowledge or the real authority field the questions. A firm where the

CEO is very “hands on” and has more real authority is more likely to field analyst questions; firms

with a “hands-off” CEO are more likely to let other managers speak. As an example, we provide a

case study of Google in Appendix A.

One potential concern with our argument is that we could be observing “communication

delegation,” as opposed to real authority delegation. However, in the view of organizational

theorists, subjective knowledge, real authority, and communication are all intricately connected: it

is the difficulty of transferring subjective knowledge that endows the owner of the knowledge with

real authority. Communication can provide assurance to others that the speaker indeed is

comfortable with the nuances of the decision and his or her opinion should be followed. It is that

very nuance that management needs to communicate to the analysts in open-ended Q&A sessions.

Of course, the manager truly in charge can prime another manager to provide canned

information, but evidence such as Mayew and Venkatachalam (2008) and others indicates that

conference call listeners have a demand for soft information other than information easily

obtained in a press release. For example, Cisco Systems, Inc. was recently criticized for a

conference call which had “a scripted feel” (Vance 2009).

Communication patterns of course have a whole host of other underlying determinants

in addition to real authority: the business model may not require much subjective knowledge,

some managers may be self-aggrandizing and wish to “hog the mike”, etc. Our hypotheses, which

we turn to next, attempt to recognize these underlying determinants as well.

8 For example, a study by the Association for Investment Management & Research found that 95 percent of analysts and investors view the “conference call as the most important form of technology-aided communications between public companies and the investment community” (Stewart 2002).

8

2.2 Hypotheses

We frame our hypotheses in the context of the real authority of the CEO relative to top

management. Broadly speaking, analytical studies such as Aghion and Tirole (1997) and Baker,

Gibbons and Murphy (1999) model a manager with the formal authority as the principal (in our

case the CEO) and a subordinate with the subjective knowledge as the agent (in our case the

remaining management team). In these models, the agents propose projects which the principal

must accept or reject. The principal cannot interact with all the company’s stakeholders

personally; as a result much subjective knowledge about the projects resides with her subordinates.

The uninformed principal must then decide to invest resources in becoming informed. This is

costly, however, in that time and expense are required to gain this information, and further, the

potential for an informed principal to overrule the subordinate may reduce the subordinate’s

initiative to search for projects ex ante. Therefore, the extent to which the principal acquiesces to

the agent’s decision and expertise is the measure of the agent’s real authority.9

These theoretical analyses generate a rich set of equilibrium results, some of which are

tightly tied to specific modeling assumptions, and others (in our opinion) that are more broadly

valid. We cull these results to generate a set of hypotheses that are a) broadly consistent with these

models, and b) empirically testable. Specifically, following Aghion and Tirole’s (1997) equilibrium

9 Despite their broader similarities, there are significant differences across the model setups. For example, Aghion and Tirole (1997) argue that a principal can delegate formal authority to an agent, while Baker, Gibbons and Murphy (1999) argue that “decision rights in organizations are not contractible” and, therefore, “formal authority only resides at the top” of the organization (abstract). However, Baker, Gibbons and Murphy (1999) do suggest that there exists “delegated authority” and “informal authority” within organizations. According to Baker, Gibbons and Murphy (1999), delegated authority arises when an informed principal effectively promises not to overturn the decisions of the subordinate, whereas informal authority arises when an uninformed principal “rubber-stamps” the subordinate’s decisions. This notion of informal authority is similar in spirit to Aghion and Tirole’s (1997) real authority and results in similar predictions, which we discuss in Section 2.

9

results, we predict an association between the extent of real authority that the principal retains and

the incentives of the agents, the decision urgency of the projects, the importance of the task

considered, the technical competency required, and the span of control of the principal.10

Likewise, Baker, Gibbons and Murphy (1999) suggest similar situations in which the principal

must approve or veto proposals prior to being informed, such as projects in which the subordinate

has particular expertise, the outcome is inconsequential for the principal, or a decision is needed

quickly.11 We discuss each of these factors in turn and then state our main hypothesis.

Incentives: Aghion and Tirole (1997) note that higher monetary incentives for the agent

may have two complementary effects on the agent’s real authority: 1) it will increase the probability

that the agent will recommend a good project, and 2) it will decrease the need for monitoring by

the principal and hence the probability that the principal will overrule the project proposal. These

incentives align the agent’s interests with those of the principal, reducing control costs. If the

agent’s monetary incentives are low, the principal is likely to retain real authority.

Decision Urgency: Another factor affecting the principal’s real authority is the time it takes

for the agent to effectively communicate the information to the principal (a form of high

knowledge transfer costs). The benefits to the principal to maintain real power over the decision

may be more than offset by the crucial time lost in the communication and decision process.

Task Importance: Ceding real authority to the agent creates control costs: the agent can

make suboptimal decisions from the perspective of the principal. When the suboptimal decision

10 Aghion and Tirole (1997) also predict that a CEO’s real authority will be lower when she has a reputation for infrequently intervening in the decision process and when the agent has multiple principals. As we had difficulty identifying proxies for these predictions, we do not test these in this paper. 11 Baker, Gibbons and Murphy (1999) present these as potential scenarios in which informal authority may arise, not sufficient conditions for informal authority. In these situations in their model, if the agent’s incentive to abuse the informal authority is too great then the principal may also choose to veto all projects, which would prohibit the agent from acquiring informal authority.

10

may result in a sufficiently large negative outcome (i.e., control costs are high), the principal may

be unwilling to cede real authority. Therefore, the principal will be relatively more willing to cede

real authority on projects with less importance to the principal.

Technical Competency: The principal simply may not have the technical competence or

experience to fully understand the nature of the projects proposed by the manager. Instead, the

principal may be selected for her ability to effectively impact the marginal productivities of the

employees throughout the organization (Rosen 1982) — i.e., manage people and processes. The

principal thus may have lower real authority in firms which require a high level of technical

expertise (e.g., scientific innovation).

Span of Control: A principal with broad control over productive labor is likely

overburdened with project proposals from subordinates. Increasing numbers of project proposals

by subordinates increases the knowledge transfer costs from the subordinates to the principal.

With each marginal project proposal, the average time spent on each project proposal decreases

resulting in the CEO being less informed about each project, on average. Therefore, increases in

span of control magnify the information asymmetry (i.e., increase knowledge transfer costs)

between the principal and the agent, thus endowing the agent with more real authority.

Drawing on our measure of CEO speaking as the proxy for her real authority, we combine

the above predictions into our first hypothesis:

H1: The real authority of the CEO is lower in firms with higher monetary incentives for subordinates, higher decision-making urgency, lower shareholder communication importance, higher technical complexity, and greater span of control.

11

An immediate alternative explanation to H1 is that what we are observing are variations

not in real authority but in formal authority. A talented subordinate who wants to rise through

the ranks may not be content with verbal CEO promises to acquiesce to the subordinate’s

knowledge and authority. He may want to see definitive formal limits on CEO power, which

ultimately seeps into conference call patterns. Likewise, a “hands on” CEO may wish to sidestep

internal top management dissent and challenges to her decisions, and seek authority over top

management through formal means. As a result, in models such as Aghion and Tirole (1997),

organizations employ both formal and real authority mechanisms to deal with economic forces.

We therefore need to control for measures of formal authority in H1.

We measure the CEO’s formal authority through job title concentration, including the

CEO’s position on the board and founder status.12 In addition, CEOs derive formal authority or

power not only from formal channels but also from other sources such as ownership and prestige

(Finkelstein 1992). We consider these attributes in development of our second hypothesis to

consider the possibility that other forms of CEO authority may be associated with a CEO’s real

authority. We motivate these additional measures next.

A significant stream of literature has argued that management gains power through

ownership (e.g., Cheng, Nagar and Rajan 2005). In addition, Finkelstein (1992) suggests that

CEOs also gain power through prestige. Prestige is the “reputation in the institutional

environment and among stakeholders” (pg. 510) and bestows power upon the holder via perceived

and real reductions in uncertainty. Other managers and stakeholders view a CEO with prestige as

having power through perceived powerful connections; while prestige itself allows a CEO to make

12 We typically do not have access to specific CEO employment contracts wherein the scope of formal authority is delineated.

12

connections with outside stakeholders and gain valuable information from outside the

organization. Thus, with perceived abilities to gain information and connections to actually

acquire information, stakeholders and other managers cede power to a more prestigious CEO.

Finally, a CEO can also acquire power through psychological and sociological means.

People tend to defer to psychological traits such as confidence, and recent literature has paid much

attention to CEO overconfidence and its impact on corporate decisions (e.g., Malmendier and

Tate 2005, Ben-David, Graham and Harvey 2007, and Goel and Thakor 2008). Likewise, a CEO

with a long tenure may know the sociological innards of the organization well enough to get what

she wants. We therefore account for CEO individual characteristics such as tenure and

overconfidence.13 We now formally propose our second hypothesis:

H1a: Hypothesis H1 holds after controlling for the CEO’s formal authority metrics, CEO ownership, and CEO individual characteristics such as prestige, tenure, and overconfidence.

Hypotheses H1 and H1a are joint tests of the analytical theories of real authority and our

assertion that the communication patterns of conference calls can capture real authority. To

further establish the validity of our measure of real authority, we turn to joint tests involving the

CEO labor markets and our measure of the CEO’s real authority.

Efficient labor markets should reward authority within an organization (Rosen 1982). This

equilibrium is an outcome of a matching process of talent and responsibility. Those with more

responsibility (or authority) have an impact on productive resources below them in the hierarchical

chain and, therefore, require more talent given the multiplicative impact of their decision and

13 In a perfect labor market, CEO characteristics should match perfectly with firm characteristics and there would be little reason to expect any variation in the former after controlling for the latter. However, a significant body of research (e.g., Bertrand and Schoar 2003, Bennedsen, Perez-Gonzalez and Wolfenzon 2007, and Kaplan, Klebanov and Sorensen 2008) has concluded that CEO characteristics and abilities matter to firm performance and decisions.

13

actions. Therefore, we suggest that the CEO’s compensation relative to top management will

increase with her real authority.14 Formally stated:

H2: The compensation of the CEO is increasing in the CEO’s level of real authority.

Because hypotheses H1 and H1a rely on a principal-agent paradigm, they apply naturally to

the CEO (the principal) and top management (the subordinates). By contrast, the labor market

joint tests can be applied to multiple managerial positions. In our conference call setting, the

manager who speaks the most other than the CEO is the Chief Financial Officer (CFO). Testing

hypothesis H2 with the CFO would further establish the validity of our conference call measure.

We formally state this as:

H2a: The compensation of the CFO is increasing in the CFO’s proportion of communication on earnings conference calls.

3. Data and Variable Definitions

This section describes our data sources, variable definitions and descriptive statistics. Our

study uses data from multiple sources. We collect conference call data from transcripts compiled

by ThomsonReuters, firm financial and compensation data from Compustat’s Xpressfeed and

ExecuComp databases, respectively, stock return data from CRSP, and board of directors and

governance data from RiskMetrics (formerly IRRC).15 We discuss each of the variables below,

focusing most of our attention on our conference call data as this data is the unique contribution

of our study. Appendix B provides descriptions and definitions for all variables used in our

analysis.

14 Note that our tests measure relative compensation (though we examine the level of CEO compensation as well). We cannot test if the management team as whole is overpaid or underpaid relative to their marginal product. 15 We use company founding year data from Jovanovic and Rousseau (2001) as downloaded from Boyan Jovanovic’s website: http://www.nyu.edu/econ/user/jovanovi/.

14

3.1 CEO’s Real Authority Measure

Our measure of the CEO’s real authority is the extent to which the CEO participates in

earnings conference calls. Because the variable has not been used in previous studies, we describe

our process for measuring real authority in detail. We obtained over 129,000 conference call

transcripts compiled from January 2001 to September 2008 by ThomsonReuters. Table 1 reports

our conference call selection process. From the original sample of conference calls, we ultimately

use the data from 17,400 transcripts. As Table 1 indicates, we discard conference calls that have

foreign text or are not related to earnings.16 Earnings conference calls have the advantage in our

setting in that they generally follow a consistent format across firms: an opening dialogue by

company executives followed by a question and answer session between analysts and company

executives. Non-earnings related conference calls which we eliminate are generally presentations

made at conferences or are conference calls for special events, such as mergers and acquisitions.

We eliminate these transcripts to control for special one-time items and to ensure a consistent

format for the transcript. Further, because we require compensation data for our analysis, the most

significant single reason for eliminating conference calls from the sample is the lack of

ExecuComp data.

For each conference call, we use FORTRAN code to parse the text and identify the date of

the call, the name and ticker symbol of the firm, and whether the call related to an earnings

report.17 In addition, each time a person spoke during a conference call, the transcript reports the

16 Our FORTRAN code is not able to process transcripts which include foreign characters. These firms would likely have been removed from the sample because of missing compensation data anyway. 17 We received the transcripts in XML (Extensible Markup Language) formatting. XML provides “tags” to identify elements of the transcript. In particular, we used XML tags to identify the company name, date of the conference call,

15

name and title of the individual who spoke.18 We then determine the amount of speech (both the

number of times that a person spoke as well as the number of characters spoken) for each

individual on the conference call by title. Our measure of real authority is the amount of text

spoken by the CEO as a percentage of text spoken by all company personnel on the conference

call.19

Our coding procedure eliminated any transcripts from our sample for which we could not

identify at least one speaker during the conference call. We could not identify speakers in the

transcripts for at least three reasons: 1) the transcript explicitly stated that a speaker was

unidentified; 2) our code was unable to properly parse the name and title for at least one speaker

(or the title that was parsed was insufficient to identify the individual’s position); or, 3) analysts

were not labeled with an “Analyst” title and, therefore, we could not identify them as such.20 As

these reasons for omitting conference call observations should not be systematically related to our

analysis, our results should be unbiased, but may have reduced power.21

type of conference call (e.g., earnings, conference presentation, etc.) and company ticker symbol. The main body of the conference call text did not contain XML tags and was effectively a plain text document. 18 While the transcript reports the title of the individual speaking, there is by no means a consistent set of titles across all firms. We identified over 16,000 unique titles for individuals who spoke at least once on the conference calls. Because it is important to our study that the executive positions are identified correctly, we manually examined each of the titles and grouped the more than 16,000 unique titles into one of nine major title groupings: CEO, CFO, Operations, Functions, Investor Relations, Other, Analyst, Operator, and Unknown. As the CEO and CFO conduct a majority of the conference call speaking, we focus our reported statistics on these two roles. 19 As we describe below, in addition to the amount of text spoken, we also considered the number of times that the CEO spoke as well as a dummy variable indicating the presence of the CEO on the conference call. See Sections 4 and 5 for a discussion of the results using these measures. 20 In the conference calls which were parsed properly, analysts were identified in the transcript by their name, firm name and the title of “Analyst.” However, conference call transcripts also identified analysts by their name and firm name only, omitting the title of “Analyst.” Therefore, when the transcript was parsed, the title variable was missing and we were unable to identify the position of the individual without examining each transcript manually. 21 To provide some assurance that the conference calls that we delete in this step are not biasing our results in some manner, we compared the eliminated conference calls to the remaining conference calls two ways: 1) We compared the average total length of the deleted and retained conference calls (total number of characters spoken by anybody on the conference call). The deleted conference calls on average were about 460 characters longer than the conference calls remaining in our sample; however this is only about a 1% difference which is likely immaterial. 2) We hand collected a random sample of 20 conference calls that we deleted and processed these conference calls manually. The

16

The conference calls are typically quarterly events; however, all other variables that we use

in this study are measured on an annual basis. We therefore convert the conference call data to

annual observations by averaging across all conference calls for a firm within a fiscal year.22 After

this procedure, there are 6,854 firm-year conference call observations. We further eliminate firm-

year observations with insufficient data from all other data sources.23 Finally, the unit of

observation in this study is a firm-level basis (rather than firm-year); therefore we average the

annual observations for all variables within a firm to derive 1,372 firm observations.24 Ultimately,

we reduce the sample further to 1,142 firm observations after we require each firm to have a

minimum of 2 years of data.

3.1.1 Conference Call Descriptive Statistics

Table 2 reports the descriptive statistics for the 17,400 conference calls that we parsed

without issues. We present the data for all conference calls in Panel A, by year in Panel B, and by

distribution of the percentage of times that the CEO and CFO spoke during these conference calls is similar to the distribution of conference calls that remain in our sample. The one noticeable difference between the deleted and retained conference calls is that the average year that the conference calls occurred in the deleted sample is about one and a half years earlier than the average year of the retained conference calls. Through conversation with ThomsonReuters, this likely relates to a higher incidence of inconsistent transcript formatting issues in the earlier years, which would cause parsing difficulties for our FORTRAN code. Therefore, our data may be somewhat skewed toward recent years. 22 To be precise, we calculate the percentage of talking conducted by the CEO for each conference call within a year and average this value across all valid conference calls in our sample for that year (there may be from one to four usable conference calls in a given year). We assign any conference call occurring after the third month of the fiscal year through the third month of the following year to that fiscal year—i.e., we assume that the conference calls occur with up to a three month lag following the quarter close. Bowen, Davis, and Matsumoto (2002) find that 75% of the conference calls in their large sample occur in the 9 day window surrounding the quarterly earnings announcement. 23 We matched each conference call to an ExecuComp firm based on the ticker symbol as reported in the conference call transcript. This approach matched approximately 1,500 firms. We were able to match approximately 200 additional firms by hand. After the conference calls were matched to ExecuComp firms, the firms were matched to Compustat (gvkey), CRSP (CRSP-Compustat link) and RiskMetrics (Cusip) datasets. 24 By averaging all variables across time for a firm, we are effectively using the “between” estimator. We take this approach primarily because all variables have a high degree of persistence for a firm across years—i.e., these are firm characteristics that persist over time and the main variation of interest is the heterogeneity between firms (therefore, firm fixed effects regressions are not appropriate). Using annual observations and estimating clustered standard errors at the firm level is an alternative to our approach. See Section 5 for our robustness tests.

17

Fama-French 12-industry grouping in Panel C. Several statistics are worth noting. First, the

variables appear to be well-centered as the mean and median are similar in magnitude across the

variables. Figure 1 presents the distribution of the amount of text spoken by all company

personnel during the conference calls. The distribution is similar to a normal distribution with

the exception of a few outliers to the extreme right. Figure 2 presents the distribution of the

amount of speaking by the CEO as a percent of total company personnel speech. This

distribution (which by definition is distributed between 0 and 1, inclusive) has fatter tails than the

normal distribution and also has a mass point at zero. CEOs are not present (i.e., do not speak)

for approximately 9% of the conference calls in our sample. A second item worth noting is the

significant role that the CFO plays in the conference calls. In fact, while the CEO is not present

on 9% of the conference calls, the CFO is not present on only 7% of the conference calls.

In Panel B we present the conference call statistics by year. The first item to note is the

distribution of our sample across years. Our sample is skewed toward more recent years for three

potential reasons: 1) the number of conference calls has increased over this time period; 2)

ThomsonReuters collected more conference calls in more recent years; and 3) the formatting of

the transcripts was more consistent in later years, allowing our code to more effectively parse the

text without error. A second observation to note is that the length of the conference call has

increased monotonically over the time period, while the number of analyst questions has remained

fairly constant. A final item to note in Panel B is that the role of the CEO has also increased over

the years. The percentage of CEO text has increased from 46% in 2003 to 50% in 2007. Most of

this increase in the mean likely comes from a higher rate of presence by the CEOs on the

18

conference calls. In untabulated results, we find that the percentage of conference calls attended

by the CEO increased from 86% in 2003 to 93% in 2007.

Finally, Panel C reports the conference call statistics by industry. Two items are worth

noting from this table. First, there is significant variation across industry type for not only the

length of the conference call, but also the extent to which the CEO participates on average. For

example, the Utilities industries have both the least amount of text spoken by company personnel

and the least amount of participation by the CEOs. In contrast, the Health industries have the

longest conference calls on average while CEOs play the most dominant role in the Manufacturing

industries. Second, the distribution of conference calls in our sample is concentrated in five

industry classes, which is consistent with the distribution of firms in the ExecuComp dataset.

3.2 CEO’s Formal Authority Measures

Formal authority resides with the individual who has the express right to make a decision

(Milgrom and Roberts 1992, Aghion and Tirole 1997). Ideally, this measure would be constructed

from the details of the compensation contract where contingencies are expressly spelled out.

Outside researchers are typically not privy to contract details beyond financial report disclosures.

Consequently, prior studies on CEO’s formal authority (e.g., Core, Holthausen and Larcker 1999;

Adams, Almeida and Ferreira 2005; Bebchuk, Cremers and Peyer 2008) have developed their own

proxies, and we borrow three of these measures. Our first measure of formal authority is an

indicator variable for a company founder CEO. We find that just over 12% of the CEOs in our

sample also founded the company. The second measure is an indicator variable for title

concentration. A CEO who is the Chair of the Board of Directors as well as the President of the

19

company likely has more formal authority to make decisions.25 49% of the CEOs in our sample

have concentrated titles.26 Our third measure is an indicator variable for the CEO as the only

insider on the Board of Directors. Similar to the title concentration argument, if a CEO is the

only executive with formal ties to the Board, she likely has higher levels of authority over the other

executives. Table 3 reports that approximately 59% of the CEOs are the only insiders on the

Board.27

3.3 CEO Characteristic Measures

As we discussed in Section 2.2, CEOs may derive power from various sources other than

formal channels. These include ownership of the firm, tenure, prestige and overconfidence. We

discuss our proxies for each of these constructs in this section. We calculate CEO ownership as

the total number of shares owned by the CEO divided by the total number of shares outstanding

for the firm (and because prior literature has found non-linear relations, we also include the square

of CEO ownership). CEOs in our sample own approximately 1.7% of their firms on average,

while the median is only 0.3%. As to tenure, we find that the average (median) CEO tenure is

approximately 8 (6) years. 28

25 In addition, we count a CEO who is not President as having title concentration if there is no other person with the title President or Chief Operating Officer reporting to her (see Adams, Almeida and Ferreira 2005). 26 We also considered a variable for CEO duality; however, this variable is highly correlated with the CEO title concentration variable (which is a subset of CEO duality), and so we selected only one of these variables. 27 We compare our CEO formal authority summary statistics to Adams, Almeida and Ferreira (2005), Table 1 and find that our CEO Founder, CEO Title Concentration, and CEO Only Insider variables are all somewhat larger in magnitude than their study. We suggest that much of this difference is a result of differences in our samples. The Adams, Almeida and Ferreira (2005) sample is exclusively from the 1990’s and is restricted to a sub-sample of Fortune 500 companies. It is reasonable to expect that our sample includes more company founders (our sample is much broader than the Fortune 500), higher title concentration (Rajan and Wulf 2006 find the title of COO diminishing in recent years), and fewer insider directors other than the CEO (our sample is taken after the passage of the Sarbanes-Oxley Act). There may also be the possibility that our measures are slightly inflated as described in Appendix B. 28 Again, we compare our CEO characteristic summary statistics to Adams, Almeida and Ferreira (2005), Table 1 and find that our CEO Tenure and CEO Ownership variables are similar to their study.

20

Prestige is a more difficult measure to quantify. We proxy this construct using Fortune

magazine’s annual survey of the America’s Most Admired Companies. Fortune conducts this

survey on an annual basis by surveying management of Fortune 1000 companies to rank their peer

firms on eight dimensions, including quality of management.29 While the ranking is particular to

a firm, we suggest that the CEOs of the companies that are ranked highly gain prestige. We

operationalize this measure by creating a dummy variable indicating that the firm is one of the top

5 firms in its industry in the survey.30

Finally, we consider CEO overconfidence. Malmendier and Tate (2005) consider CEOs

whose wealth is overexposed to the idiosyncratic risk of their firms as being overconfident. Using

proprietary data, they create measures of overconfidence based on the extent to which CEOs wait

to exercise their options or excessively accumulate stock. Because this proprietary data does not

apply to our sample, we must proxy for the extent to which CEOs wait to exercise their options

using end of year holdings of exercisable options which remain unexercised scaled by the sum of

the value of unexercised exercisable options, unexercised unexercisable options and shares of

stock. This results in a percentage of value of the CEO’s holdings that are held in unexercised

exercisable options. On average, 29% of the total value of the CEO’s holdings is held in

unexercised exercisable options.

3.4 Determinants of the CEO’s Real Authority

29 The other dimensions include innovation, people management, use of corporate assets, social responsibility, financial soundness, long-term investment, and quality of products or services. 30 One potential issue with this measure is that Fortune only considers the largest 1,000 companies by sales and, within this group of firms, only considers firms that are among the ten largest within their industry. While our sample includes these firms, it also includes other firms that were not eligible for the survey. Therefore, this measure is biased towards larger firms.

21

As discussed in previous sections, theoretical literature (e.g., Aghion and Tirole 1997)

suggests that the level of the principal’s real authority is affected by the incentive levels of the

agents, the degree of decision-making urgency, the extent of the agent’s technical competency or

expertise, the CEO’s span of control, and the importance of the task. We discuss our proxies for

each of these factors in this section.

Agents may be provided monetary incentives via various methods. Potential proxies for

examining agents’ incentives include bonus schemes (Holthausen, Larcker and Sloan 1995),

promotion opportunities (e.g., corporate tournaments as in Bognanno 2001), and stock-based

incentives. Bonuses are an ex-post measure of incentives and typically represent a small portion of

overall incentive compensation, while the incentive effects of promotion opportunities are difficult

to measure. Therefore, we choose the third of these alternatives and measure the extent to which

the wealth of the four highest paid executives other than the CEO is sensitive to a change in the

firm’s stock price.31 The mean reported in Table 3 of 0.95 is similar in magnitude to prior studies

using this measure (e.g., Erickson, Hanlon and Maydew 2006, Table 432).

In constructing our proxy for decision-making urgency, we use the fact that managers in

highly competitive industries need to respond quickly to competitor actions. We use the degree of

product market competition as our proxy for urgent decision-making. To do so, we calculate the

31 More precisely, we measure the extent to which the executives’ firm-based wealth as reported in proxy statements is sensitive to the change in stock price, as we cannot measure other outside wealth of the executives. We calculate these sensitivities for stock and options separately. For stock, the sensitivity of the executive’s wealth to price changes is simply dependent upon the number of shares held by the executive. The sensitivity of the value of the option portfolio to a change in firm stock price is dependent upon the number of shares underlying the options and the “delta” of the options (Core and Guay 2002). See Appendix B for the method we used to estimate this variable. 32 In comparing our value of this measure to Erickson, Hanlon and Maydew (2006) we note that the value of our variable excludes the CEO, whereas the value in Erickson, Hanlon and Maydew (2006) includes the CEO. However, their sample ends in 2002 whereas ours begins in 2003. We find that the value of this measure has increased during our sample time period. Our calculation for the 2003 sample year only (and including the CEO) is similar to (and more directly comparable to) Erickson, Hanlon and Maydew (2006).

22

Herfindahl index based on firm sales for each 4-digit SIC industry within the entire Compustat

Xpressfeed universe. Because this measure may be a noisy proxy for our underlying construct and

because we only want to capture those industries that are highly competitive, we create an

indicator variable that is set to 1 only if a firm’s industry is in the top decile of the Herfindahl

index across all industries. Table 3 reveals that approximately 39% of our sample firms are in a

highly competitive industry.33

We use firm research and development (R&D) intensity to capture the notion of technical

competency. Theory suggests that a CEO has less real authority if she is less informed than the

agent (Fama and Jensen 1983, Dessein 2002, Harris and Raviv 2005). If a CEO is generally

selected based on her ability to manage the organization (Rosen 1982), then firms with a high

degree of innovation and technical complexity likely have technically-knowledgeable agents who

are more informed than the CEO regarding project selection (and possess soft information

regarding project selection criteria). We therefore proxy for the degree of technical expertise with

the level of R&D expenditure by the firm, scaled by total sales. Table 3 reports that our sample

firms spend 4% of sales on R&D activities on average.34

Our next determinant is the span of control and the overload of the principal. Span of

control may be viewed specifically as a measure of the number of executives reporting directly to

the CEO (e.g., Rajan and Wulf 2006) or more generally as control over production labor (Rosen

1982). In either perspective, as the CEO’s span of control increases, she necessarily cedes more

33 Considering that only firms within the top decile of 4-digit SIC industries received an indicator value of “1”, 39% may seem to be high number. However, by construct of the Herfindahl index variable, those industries in the top decile generally include a higher number of firms; whereas those industries in the lower deciles include relatively few firms. 34 If the R&D variable in Compustat has a missing value, we re-code the value to zero. Approximately 43% of our sample report missing values for R&D and, including the missing values, approximately 55% report zero R&D expenditures. These results are similar to the findings of Bebchuk, Cremers and Peyer (2008) and Coles, Daniel and Naveen (2008).

23

real authority because she has a finite amount of time to monitor and review projects. Empirically

we take the latter view of span of control for data availability reasons.35 We proxy for the span of

control and CEO overload using the number of employees in the firm—as the CEO’s breadth over

production labor increases, the CEO has less real authority as a result of overload. The average

firm has approximately 22,000 employees, while the median has approximately 6,000. Given the

skewness of this variable, we use the natural log.

Finally, we examine task importance. The conference call itself is an important task for top

management, for it is in important source of information for capital providers. However, the

conference call setting may be less relevant for highly regulated firms. Industries such as utilities

and telecommunications are scrutinized by federal, state and local regulators, whose substitute

monitoring and information collection role could diminish the benefit of the conference calls to

capital providers. In such settings, the CEO may leave the conference call activities to lower-level

management. Therefore, we use an indicator variable if the firm is in a Fama-French “Utilities” or

“Telecom” industry as our proxy for low task importance.

3.5 Firm Characteristics and Controls

We include a full battery of control variables that prior literature has associated with CEO

authority, compensation and firm performance. Studies such as Yermack (1996), Core,

Holthausen and Larcker (1999), and Chhaochharia and Grinstein (2009) have shown that the

structure of the Board of Directors is related to firm performance and CEO compensation.

Following this literature, we use the Board size (number of directors), percentage of insider

directors (employees, former employees or relatives of employees), and percentage of outsider

35 We also examined the ratio of employees-to-sales and our results are unchanged using this measure.

24

directors over the age of 69 as Board structure variables. We find that the average Board size is 9.3

directors, which is smaller than the finding of Core, Holthausen and Larcker (1999), Table 1, and

Bushman et al. (2004), Table 2, but consistent with Chhaochharia and Grinstein (2009) who use a

more recent sample period similar to ours.36 The percentage of insider directors is 28%, which is

also smaller than Core, Holthausen and Larcker (1999), Table 1, but is similar to Ryan and

Wiggins (2004), Table 2.

Adams, Almeida and Ferreira (2005) provide evidence that the concentration of power in

the CEO is associated with performance variability. As such, we also include the standard

deviation of return on assets and stock returns for the five year window beginning four years

before the year of interest. To control for firm governance characteristics, we use the G Index from

Gompers, Ishii and Metrick (2003). Finally, we include control variables for firm size, growth, and

profitability.

3.6 Compensation Measures

Our first measure of compensation is the total compensation of the CEO divided by the

total compensation of the CEO and four highest paid executives other than the CEO. By scaling

CEO compensation by the top five executives’ compensation, we are attempting to control for

unobservable factors of the wage function that are firm specific.37 We find that the CEO receives

38% of the compensation that is paid to the top five executives. This is similar to the amount

reported in Bebchuk, Cremers and Peyer (2008). Our second measure of compensation is the 36 The Core, Holthausen, and Larcker (1999) sample is from the years 1982-1984; the Bushman et al. (2004) sample is from 1994; and the Chhaochharia and Grinstein (2009) sample is from 2000-2005. 37 Bebchuk, Cremers, and Peyer (2008) use this compensation variable as a proxy for “CEO Centrality,” or the “relative importance of the CEO within the top executive team in terms of ability, contribution, or power” (abstract). In contrast, we simply view this variable as a scaled version of CEO compensation which controls for unobservable wage factors and can partially be explained by the extent of CEO delegation.

25

natural log of total CEO compensation. This wage-level specification assumes that we are able to

capture all factors associated with the wage function with control variables.

4. Results

4.1 Correlations

Table 4 provides the Pearson correlations for all variables. Because we have an extensive

set of variables, we present the correlations in two panels. Panel A reports the correlations

between CEO percentage text, our main measure of CEO real authority, and two other proxies:

the percentage of questions fielded by the CEO, and the presence of the CEO. The three

measures are highly correlated — in particular, there is little difference between the number of

times that the CEO speaks and the amount of text spoken. Panel A also examines our

hypothesized determinants of the CEO’s real authority. Consistent with hypothesis H1, our real

authority measure is negatively correlated with equity incentives for other management, and with

product market competition, regulation, and levels of R&D. The correlations are statistically

significant for all variables except R&D/Sales. This indicates, at least at the univariate level, that

the variation in the CEO’s real authority is consistent with the predictions of analytical models of

real authority.

Panel B of Table 4 presents the correlations between our measure of CEO’s real authority

and all other variables. The most interesting finding is that that our CEO real authority measure

is not strongly related to formal measures of CEO authority; the correlation with CEO Only Insider

is the only statistically significant association. This provides some initial evidence that we are

measuring a construct distinct from that of formal authority.

26

4.2 Multivariate Results on CEO’s Real Authority

Table 5 provides multivariate tests of hypotheses H1 and H1a. In models (1) – (5) we test

each determinant variable separately and generally find evidence consistent with our prediction

H1. The CEO’s proportion of communication is negatively associated with non-CEO incentives,

decision urgency, task irrelevance, and technical knowledge. In model (5) we find a negative

coefficient on LN(Employees), but this is not statistically significant. We then include all

determinant variables jointly in model (6) and find that all coefficients have signs in the predicted

direction and are all statistically significant except for decision urgency. In model (7) we test H1a

by also including our variables for CEO formal authority, CEO ownership, CEO tenure, prestige,

and overconfidence. Our results continue to hold. This provides significant evidence in favor of

the predictions of real or informal authority models such as Aghion and Tirole (1997) and Baker,

Gibbons and Murphy (1999).

We assess the robustness of the results in Table 5 to our CEO text measure. We find that

our results (untabulated) are not sensitive to using the number of times that the CEO speaks

during the conference call rather than the amount of text. We also confirm that using firm-years,

rather than firms, as the unit of observation does not change our inferences.38 Therefore, Table 5

provides support of our joint hypotheses H1 and H1a that we are measuring real authority and

that the real authority of the CEO is lower in environments with high knowledge transfer costs or

low control costs.

Table 5 also highlights the relation between our measure of real authority and the

measures of a CEO’s formal authority and characteristics. The coefficients on CEO Title 38 When using firm-years as the unit of observation, we estimate the standard errors by clustering at the firm level.

27

Concentration and CEO Only Insider are positive (though CEO Only Insider is not significant),

indicating that a CEO with more formal authority has more real authority. The coefficient on

CEO Founder is negative, but insignificant. This suggests that real authority may be more closely

linked to title or position based formal authority rather than a historical notion of authority. We

also note that both ownership and tenure have non-linear relations with the CEO’s real authority.

At lower levels, ownership has a negative relation with the CEO’s real authority, which would

appear to be the opposite direction as expected because presumably the task of communicating

with the capital markets would be more important for a CEO with higher levels of share

ownership. On the other hand, the CEO’s real authority is increasing (at a decreasing rate) with

the length of the CEO’s tenure. This seems to be a reasonable result assuming that the CEO

continues to build knowledge in the organization over time and, therefore, is less reliant on the

soft information of her subordinates.39 Finally, we find that prestige is negatively associated with

the CEO’s real authority. This is an interesting result which may suggest that CEOs seeking

prestige do not do so through conference calls.40 We conclude that our measure of real authority

is associated with formal measures of authority and CEO ownership, tenure, and prestige.

Because analytical papers such as Aghion and Tirole (1997) make a distinction between

real and formal authority, we further consider the nature of formal authority. We could not find

any prior empirical studies on formal authority that examined its determinants in a manner similar

to Table 5; therefore, we repeat the regressions from Table 5 using the same regressors but with

39 Interestingly, the relation between the CEO’s real authority and tenure reaches a maximum around 10 years and then has a negative relation. At that length of tenure, the CEO may be more concerned about grooming the next CEO and therefore begins speaking less, i.e., relinquishing real authority to her replacement. 40 This result may also be the outcome of our proxy choice for prestige. As we mentioned in Section 3, Fortune magazine only covers the largest 1,000 firms. Therefore, our result may also be proxying for a dimension of firm size not already modeled with the number of employees or assets.

28

formal authority proxies as dependent variables. Table 6 presents our results. Unlike our results

in Table 5 which indicate a consistently negative relation between CEO real authority and

predicted determinants, Table 6 reveals that the same does not obtain for measures of the CEO’s

formal authority. Only one coefficient is significantly negative in each specification and CEO Title

Concentration and CEO Only Insider have a determinant variable which is significantly positive.

These results suggest that our CEO real authority measure is separate from CEO formal authority,

and lend further support to our joint tests of the validity of our CEO text variable as a measure of

real authority and theories of real authority.

4.3 CEO’s Real Authority and CEO Compensation

Hypothesis H2 predicts a positive relation between CEO wages and real authority. We

regress CEO wages on our proxy of CEO real authority and various controls for the wage function.

We consider both scaled (by the total wages of the top 5 executives) and unscaled (natural log of

wages) CEO wages. We also examine total wages (including cash, stock, and option components)

and cash wages only (salary and bonus).

Table 7 presents the results of our analysis. As an initial assessment, in the first regression,

we only include our real authority measure and firm characteristic controls and find that the

coefficient on CEO real authority is strongly significantly positive; that is, compensation is an

increasing function of real authority. In model (2), as a baseline, we regress wages on measures of

the CEO’s formal authority, the CEO’s characteristics and firm controls and find results

consistent with prior literature (e.g., Core, Holthausen and Larcker 1999, Bebchuk, Cremers and

Peyer 2008, and Chhaochharia and Grinstein 2009). Specifically, we find a strong positive

29

relation between CEO compensation and the title concentration of the CEO, the CEO as the only

insider, the CEO’s tenure length, firm size, profitability, and governance characteristics. CEO

compensation is negatively associated with the percentage of ownership the CEO has in the firm,

also consistent with prior findings. Growth is the only variable that is sensitive to the specification

of CEO compensation in that it is negatively associated with the scaled version of CEO

compensation, but positively associated with unscaled CEO compensation. Based on the results of

models (1) and (2), it appears that our measure of real authority is related to wages as predicted

and that our models are consistent with prior findings as well.

Models (3) through (6) combine our measure of real authority with the formal authority

measures, CEO characteristics and the full gamut of controls. In models (3) and (4) we use total

compensation (scaled and unscaled, respectively) and in models (5) and (6) we use cash

components of compensation only. Again, we find that our measure is positively associated with

CEO compensation, as predicted, and the coefficients on the remaining variables are consistent

with prior research. With these results, we confirm our H2 hypothesis and prior theoretical and

empirical literature that wages are positively associated with authority.

Table 7 also allows us to compare the magnitude of the impact of real versus formal

authority and other firm characteristics on wages. Using model (3) we calculate the impact on the

portion of compensation paid to the CEO of a one standard deviation change in various

independent variables. A standard deviation increase in the CEO’s real authority is associated

with an increase in the relative portion of compensation paid to the CEO of 1.5 percentage points

(or about a 4 percent increase relative to the mean value of 38%). This compares very similarly to

a one standard deviation increase in firm size which has an approximate impact of 1.7 percentage

30

points. By comparison, an increase in CEO’s formal authority (which is measured using indicator

variables, so the increase here is simply the value of the coefficient) is associated with an increase in

the CEO’s compensation portion of 2.0 and 3.0 percentage points for CEO Title Concentration and

CEO Only Insider, respectively. Thus, both real and formal authority appear to have significant

wage implications.

Table 8 provides a test of hypothesis H2a that CFO compensation should increase in CFO

real authority.41 Table 8 shows that CFO compensation is consistently significantly positively

associated with the CFO’s real authority, though the economic magnitude is substantially reduced.

A one standard deviation increase in CFO real authority is associated with the CFO’s

compensation as a portion of the top five executives increasing by 0.3 percentage points.

In sum, Tables 5 through 8 provide evidence supporting the joint tests of real authority

models and our measure of real authority as well as joint tests of labor market models and our

measure of real authority. Real authority thus appears be an organizational feature distinct from

formal authority.

5. Additional Analyses and Robustness Checks

The predicted organizational determinants of CEO’s real authority in Section 2 are driven

by some aspect of information asymmetry among the firm’s employees. While we cannot measure

directly the information asymmetry inside a firm, we can do so for the firm’s shareholders.

Recognizing that external measures of information asymmetry are a poor proxy for internal

41 Note that we include variables for the CFO’s ownership and presence on the Board of Directors, but do not include other variables that are included for the CEO either because of a substantial reduction of data (e.g., Tenure because date of employment is frequently missing) or because the variable is not relevant for the CFO (e.g., Title Concentration and Only Insider).

31

measures of information asymmetry, we conjecture that in firm environments in which

information asymmetry is higher, the association between CEO real authority and its hypothesized

the determinants should be stronger. To test this comparative static of hypothesis H1, we

segregate our sample into low and high external information asymmetry groups. We create this

partition using two proxies for external information asymmetry: 1) the amount of price protection

by market makers proxied by the spread between the bid price and ask price of the firm’s stock,

and 2) the overall information environment proxied by stock return volatility.42 In untabulated

results, we find that the coefficients are generally larger and more frequently statistically significant

in the sample of firms with high information asymmetry, reinforcing our main results.

We then reconsider the effect of the CFO on our CEO text variable. In earnings

conference calls, the CFO’s portion of communication is substantial because financial results are

typically reviewed and discussed. As a result, the variation in the CEO communication patterns

driven by the CFO may not indicate differences in authority, but rather may reflect the importance

of financial communication. We therefore consider an alternative calculation of our measure of

the percentage of CEO communication in which we remove the CFO from the denominator.43

Using this measure, the CEO speaks about 70% of the time. In Table 9, column (1) we re-execute

model (7) from Table 5, and find that our inferences remain virtually unchanged — only the

coefficient on LN(Employees) loses significance relative to the results in Table 5.

Next, we examine the potential influence of outliers in the determinants of authority

delegation. The unit of observation for our study is the firm-level. Because we average the

42 See, for example, Glosten and Milgrom (1985) and Kyle (1985) for the theoretical linkage between information asymmetry and the bid-ask spread. 43 Specifically, the alternative measure is the amount of text spoken by the CEO divided by the total amount of text spoken by company personnel other than the CFO.

32

variables across all available years for a particular firm and require at least 2 years of observations,

we did not winsorize the data, allowing, instead, the averaging process to ameliorate extreme

observations. However, we identified an extreme observation which may have undue influence on

the results. Microsoft Corporation’s Non-CEO Equity Sensitivity value is 223, which is about 8

times larger than the next largest observation.44 Table 9, column (2) presents the results of Table

5, model (7) after removing the Microsoft observation. The coefficient on Non-CEO Equity

Sensitivity remains positive, but is no longer statistically significant. Our inferences of the

remaining determinant variables are unaffected.

In columns (3) and (4) of Table 9, we relax our requirement that each firm have at least 2

annual observations.45 Column (3) indicates that our inferences are unaffected. Finally, instead of

using firm assets to proxy for firm size, we use firm sales. Column (4) presents the results using the

natural log of firm sales rather than firm assets. In this regression, Non-CEO Equity Sensitivity and

Product Market Competition are now significant, while the LN(Employees) is no longer significant at

traditional levels. Our hypotheses H1 and H1a on the theoretically motivated determinants of real

authority thus appear to be empirically robust.

Finally, we examine the number of business segments reported by the firm as an additional

construct of organizational design. Generally Accepted Accounting Principles (GAAP) “requires

that disaggregated information be presented based on how management internally evaluates the

44 This result is, of course, due to the fact that Bill Gates—who retains substantial shareholdings—is no longer the CEO, but is one of the top 4 non-CEO executives. 45 When we relax the 2 year restriction, the sample includes an extreme observation for the variable R&D/Sales. Arqule, Inc. conducted significantly more R&D activities than sales activities. As a result, the firm’s R&D/Sales value is about 7 times larger than the next highest observation. In regressions (3) and (4) of Table 9, we removed this observation.

33

operating performance of its business units” (Berger and Hann 2003, pg. 164).46 We count the

number of segments each of the firms reports and use this as a proxy for the breadth of the

organization.47 We consider that the number of segments that a firm operates (controlling for

firm size) may have counteracting forces on the CEO’s real authority. First, as the number of

operating units increases, the CEO’s span of control may be increased, thus reducing his real

authority over any specific unit as discussed in Section 2. Alternatively, the CEO may split the org-

chart into many formal units to disperse formal authority across more and smaller units (recall that

the number of reported segments reflects the internal org-chart under SFAS 131). In this regard,

as the number of segments increases, each of the individual managers has less authority, while

concentrating the authority in the centralized CEO position.

To investigate this relation, we re-examine our results from Table 5, model (7) after

including a variable for the number of segments that the firm reports. Table 10 presents our

results. Columns (1) and (2) indicate that there is a significantly positive relation between the

CEO’s real authority and the number of segments, indicating that perhaps the CEO gains

authority as the firm becomes more dispersed. In column (3) we also include the squared version

of this variable. The coefficient on the number of segments remains significantly positive, while

the coefficient on the squared term is significantly negative. Our inference is that the CEO may

initially gain real authority by subdividing operating units but that eventually the effects of the

46 This statement from Berger and Hann (2003) specifically relates to the requirements of SFAS 131. Prior to SFAS 131, which was passed in 1997, SFAS 14 was the ruling standard. SFAS 14 had a more strict definition of segments, requiring firms to segment their information based on the industry level rather than internal org-charts. 47 We calculate this variable using the number of segments that the firm reports as tabulated in the Compustat Segments file. For firms that report segments in more than one category (e.g., geography and business), we use the largest number reported in any one category. In untabulated results, we find that firms report a mean (median) of 3.1 (3.0) reporting segments.

34

increased span of control dominate.48 These results further support the notion of a connection

between organizational design and the CEO’s real authority.

6. Conclusion

This study uses the setting of firm earnings conference calls to investigate differences in

formal and real authority in organizations. We use the percentage of text spoken by the CEO

during these conference calls as a proxy for the real authority of the CEO over top management.

We find that our measure of real authority is separate from previously used measures of CEO

formal authority. Guided by prior theoretical literature, we predict that the CEO’s real authority

over top management will be affected by the incentive levels of the top management, the degree of

decision-making urgency, the importance of the task, the extent of the top management’s technical

competency or expertise, and the CEO’s span of control. Using our measure of the CEO’s real

authority, we find results consistent with these hypotheses. We then examine whether managers

with more real authority also receive higher wages. Such tests are also joint tests of our measure of

real authority and theories of efficient labor markets. Our results, again, are consistent with our

predictions: wages are positively associated with the amount of real authority maintained by the

executive.

Aghion and Tirole (1997) note that measurement difficulties are the reason why theory is

ahead of empirics in concepts such as real authority. Soft concepts such as real authority,

influence, subjective knowledge, etc., manifest themselves in human interpersonal interactions

(Simon 1997), an activity that has historically been nonconforming to large-scale empirical analyses

48 Our results indicate that this switch occurs between 7 and 8 reporting units. As a specification check on the validity of the nonlinear relation with the squared term, we added a cubed term. This term was insignificant.

35

by external researchers. The traditional research methodology therefore is laboratory experiments

where the subjects and their interpersonal interactions can be clearly observed. Recent advances in

media and computing technology, however, have put vast quantities of visual, audio, and textual

information on the Internet, opening up new methodologies and measurement techniques for

large-scale capture of soft information of human interpersonal interaction. This is precisely what

we do in our study to test analytical organizational theories.

36

References

Adams, R., H. Almeida, and D. Ferreira. 2005. Powerful CEOs and Their Impact on Corporate Performance. The Review of Financial Studies 18(4): 1403-1432.

Aggarwal, R. and A. Samwick. 2003. Performance Incentives within Firms: The Effect of

Managerial Responsibility. The Journal of Finance 63(4): 1613-1649. Aghion, P. and J. Tirole. 1997. Formal and Real Authority in Organizations. Journal of Political

Economy 105: 1-29. Andrews, E. 2009. 3 AIG Trustees are Quiet, Perhaps to a Fault. The New York Times. April 20,

2009: B1. Baiman, S., D. Larcker, and M. Rajan. 1995. Organizational Design for Business Units. Journal of

Accounting Research 33(2): 205-229. Baker, G., R. Gibbons and K. Murphy. 1999. Informal Authority in Organizations. Journal of Law,

Economics, and Organization 15: 56-73. Bebchuk, L., M. Cremers, and U. Peyer. 2008. CEO Centrality. Working paper. Harvard Law

School. Ben-David, I., J. Graham, and C. Harvey. 2007. Managerial Overconfidence and Corporate

Policies. Working paper. National Bureau of Economic Research. Bennedsen, M., F. Perez-Gonzalez, and D. Wolfenzon. 2007. Do CEOs Matter? Working paper.

New York University. Berger, P. and R. Hann. 2003. The Impact of SFAS No. 131 on Information and Monitoring.

Journal of Accounting Research 41(2): 163-223. Bertrand, M. and S. Mullainathan. 2001. Do People Mean What They Say? Implications for

Subjective Survey Data. The American Economic Review 91(2): 67-72. Bertrand, M. and A. Schoar. 2003. Managing with Style: The Effect of Managers on Firm Policies.

Quarterly Journal of Economics 118(4): 1169-1208. Bognanno, M. 2001. Corporate Tournaments. Journal of Labor Economics 19(2): 290-315. Bowen, R., A. Davis, and D. Matsumoto. 2002. Do Conference Calls Affect Analysts’ Forecasts?

The Accounting Review 77: 285-316.

37

Bushee, B., D. Matsumoto, and G. Miller. 2003. Open Versus Closed Conference Calls: The Determinants and Effects of Broadening Access to Disclosure. Journal of Accounting and Economics 34: 149-180.

Bushee, B., D. Matsumoto, and G. Miller. 2004. Managerial and Investor Responses to Disclosure

Regulation: The Case of Reg FD and Conference Calls. The Accounting Review 79: 617-643. Bushman, R., Q. Chen, E. Engel, and A. Smith. 2004. Financial Accounting Information,

Organizational Complexity and Corporate Governance Systems. Journal of Accounting and Economics 37: 167-201.

Campbell D., S. Datar, and T. Sandino. 2009. Organizational Design and Control Across Multiple

Markets: The Case of Franchising in the Convenience Store Industry. The Accounting Review. Forthcoming.

Cheng, S., V. Nagar, and M. Rajan. 2005. Identifying Control Motives in Managerial Ownership:

Evidence from Antitakeover Legislation. The Review of Financial Studies 18(2): 637-672. Chhaochharia, V. and Y. Grinstein. 2009. CEO Compensation and Board Structure. The Journal

of Finance, forthcoming. Coles, J., N. Daniel, and L. Naveen. 2008. Boards: Does One Size Fit All? Journal of Financial

Economics 87(2): 329-356. Core, J. and W. Guay. 2002. Estimating the Value of Employee Stock Option Portfolios and Their

Sensitivities to Price and Volatility. Journal of Accounting Research 40(3): 613-630. Core, J., R. Holthausen, and D. Larcker. 1999. Corporate Governance, Chief Executive Officer

Compensation, and Firm Performance. Journal of Financial Economics 51: 371-406. Dessein, W. 2002. Authority and Communication in Organizations. Review of Economic Studies 69:

811-838. Erickson, M., M. Hanlon, and E. Maydew. 2006. Is There a Link between Executive Equity

Incentives and Accounting Fraud? Journal of Accounting Research 44(1): 113-143. Fama, E. and M. Jensen. 1983. Separation of Ownership and Control. Journal of Law & Economics

26: 301-325. Finkelstein, S. 1992. Power in Top Management Teams: Dimensions, Measurement, and

Validation. Academy of Management Journal 35(3): 505-538. Frankel, R., M. Johnson, and D. Skinner. 1999. An Empirical Examination of Conference Calls as

a Voluntary Disclosure Medium. Journal of Accounting Research 37(1): 133-150.

38

Glosten, L., and P. Milgrom. 1985. Bid, Ask, and Transaction Prices in a Specialist Market with

Heterogeneously Informed Traders. Journal of Financial Economics 13: 71-100. Goel, A. and A. Thakor. 2008. Overconfidence, CEO Selection, and Corporate Governance. The

Journal of Finance 63(6): 2737-2784. Gompers, P., J. Ishii, and A. Metrick. 2003. Corporate Governance and Equity Prices. The

Quarterly Journal of Economics 118: 107-155. Harris, M. and A. Raviv. 2005. Allocation of Decision-making Authority. Review of Finance 9: 353-

383. Himmelberg, C., R. Hubbard, and D. Palia. 1999. Understanding the Determinants of Managerial

Ownership and the Link Between Ownership and Performance. Journal of Financial Economics 53: 353-384.

Holthausen, R., D. Larcker, and R. Sloan. 1995. Annual Bonus Schemes and the Manipulation of

Earnings. Journal of Accounting and Economics 19: 29-74. Jensen, M. and W. Meckling. 1992. Specific and General Knowledge, and Organizational

Structure. In: L. Werin, H. Wijkander (Eds.), Contract Economics. Blackwell, Oxford, UK. Jovanovic, B. and P. Rousseau. 2001. Why Wait? A Century of Life Before IPO. American Economic

Review 91: 336-341. Kaplan, S., M. Klebanov, and M. Sorensen. 2008. Which CEO Characteristics and Abilities

Matter? Working paper. University of Chicago. Kimbrough, M. 2005. The Effect of Conference Calls on Analyst and Market Underreaction to

Earnings Announcements. The Accounting Review 80: 189-219. Kyle, A. 1985. Continuous Auctions and Insider Trading. Econometrica 53: 1315-1336. Lashinsky, A. 2009. The Axman Comes to Google. http://money.cnn.com/2009/03/17/

technology/lashinsky_google.fortune/index.htm Li, F. 2008. Annual Report Readability, Current Earnings, and Earnings Persistence. Journal of

Accounting and Economics 45: 221-247. Malmendier, U. and G. Tate. 2005. CEO Overconfidence and Corporate Investment. The Journal

of Finance 60(6): 2661-2700.

39

Mayew, W. 2008. Evidence of Management Discrimination Among Analysts during Earnings Conference Calls. Journal of Accounting Research 46(3): 627-659.

Mayew, W. and M. Venkatachalam. 2008. The Power of Voice: Managerial Affective States and

Future Firm Performance. Working paper. Duke University. McConnell, J. and H. Servaes. 1990. Additional Evidence on Equity Ownership and Corporate

Value. Journal of Financial Economics 27: 595-612. Milgrom, P. and J. Roberts. 1992. Economics, Organization, and Management. Upper Saddle River,

NJ: Prentice-Hall, Inc. Moers, F. 2006. Performance Measure Properties and Delegation. The Accounting Review 81(4):

897-924. Morck, R., A. Shleifer, and R. Vishny. 1988. Management Ownership and Market Valuation.

Journal of Financial Economics 20: 293-315. Nagar, V. 2002. Delegation and Incentive Compensation. The Accounting Review 77(2): 379-395. Ortega, J. 2009. Employee Discretion and Performance Pay. The Accounting Review 84(2): 589-612. Petersen, M. 2004. Information: Hard and Soft. Working paper. Northwestern University and

National Bureau of Economic Research. Petersen, M. and R. Rajan. 1994. The Benefits of Lending Relationships: Evidence from Small

Business Data. The Journal of Finance 49(1): 3-37. Rajan, R. and J. Wulf. 2006. The Flattening Firm: Evidence from Panel Data on the Changing

Nature of Corporate Hierarchies. The Review of Economics and Statistics 88(4): 759-773. Roberts, J. 2007. The Modern Firm: Organizational Design for Performance and Growth. Oxford

University Press. Rosen, S. 1982. Authority, Control and the Distribution of Earnings. The Bell Journal of Economics

13: 311-323. Ryan, H. and R. Wiggins. 2004. Who is in Whose Pocket? Director Compensation, Board

Independence, and Barriers to Effective Monitoring. Journal of Financial Economics 73: 497-524. Simon, H. 1997. Administrative Behavior. 4th Ed. The Free Press. New York, NY. Solomon, D. 2009. Bailout Man Turns the Screws. The Wall Street Journal. April 7, 2009: A1.

40

41

Stein, J. 2002. Information Production and Capital Allocation: Decentralized versus Hierarchical Firms. The Journal of Finance 57(5): 1891-1921.

Stewart, N., Ed. 2002. IR Guides Book. Brimsdown, UK: The Magazine Printing Company. Tetlock, P., M. Saar-Tsechansky, and S. Macskassy. 2008. More Than Words: Quantifying

Language to Measure Firms’ Fundamentals. The Journal of Finance 63(3): 1437-1467. Vance, A. 2009. Cisco Preens on TelePresence. The New York Times, as quoted from:

http://bits.blogs.nytimes.com/2009/03/16/cisco-preens-on-telepresence/, March 16, 2009. Vancil, R. 1979. Decentralization: Managerial Ambiguity by Design. Financial Executives Research

Foundation. New York, NY. Yermack, D. 1996. Higher Market Valuation of Companies with a Small Board of Directors.

Journal of Financial Economics 40: 185-211. Zhou, X. 2001. Understanding the Determinants of Managerial Ownership and the Link Between

Ownership and Performance: Comment. Journal of Financial Economics 62: 559-571.

Appendix A: Google, Inc. Example

We provide a brief example to highlight the manner in which our measure may be effective in capturing the notion of real authority.∗ On August 1, 2008, Google, Inc. hired a new Chief Financial Officer from outside the company, Patrick Pichette. Prior to arriving at Google, Mr. Pichette was the top operations executive at Canada’s largest phone company. He was brought into Google to specifically ensure that the company maintained efficient operations and, according to Google’s CEO Eric Schmidt, to review internal business plans “systematically, business after business.” Lashinsky 2009 documents the following changes at Google since Mr. Pichette joined the company:

Since he started as CFO on Aug. 1, Google has shut down numerous projects, facilities, and perks, from the seemingly trivial - an unneeded gourmet cafe at its headquarters, the annual companywide ski trip - to the significant. The latter includes the termination of a major effort called Lively, a virtual-environment product that mimicked Second Life, and the shuttering of a failed acquisition, dMarc Broadcasting, through which Google had attempted to broker radio advertising. In January, Google publicized its first layoffs, the termination of 100 recruiters made redundant because the company has dramatically reined in its hiring.

While Lashinsky (2009) also indicates that the aforementioned actions at Google may have occurred with an alternative CFO, given management’s comments and Mr. Pichette’s background, it appears that he has begun to have a significant impact on Google’s investment decisions.

Mr. Pichette has certainly made an impact on the conference calls. The chart below presents the percentage of speaking by Google’s CFO.Δ The 2008 4th Quarter conference call was the first earnings conference call in which Mr. Pichette spoke. During this conference call he spoke more text (35%) and more times (66%) than any other executive. In reading the transcript of the conference call, we noted that Mr. Pichette had significant control over the dialogue, even at one point “delegating” a response to the CEO, Eric Schmidt. While this example is only anecdotal, it provides an illustration of why we suggest that there is a link between speaking on the conference calls and real decisions made within the company.

GooglePercentage of Text Spoken by the CFO

0%

5%

10%

15%

20%

25%

30%

35%

40%

2004 2005 2006 2007 2008

∗ The source of information for this illustration is from Google, Inc.’s corporate website (www.google.com/corporate) and Lashinsky (2009), an article posted to the CNN Money website http://money.cnn.com/2009/03/17/technology/lashinsky_google.fortune /index.htm.

42

Δ Mr. Pichette started at Google on August 1, 2008. Upon review of the 3rd Quarter 2008 conference call, Mr. Pichette did not speak. Because our dataset is missing a few non-4th quarter conference calls for Google during these years and to mitigate any seasonality concerns, we present 4th quarter conference calls only (which typically occur in January the following year). However, the conference calls which are not included in the graph have the same downward trend prior to Mr. Pichette’s arrival at Google.

Appendix B: Variable Definitions This appendix describes the data source and measurement of each variable used in our study. All data is for the years 2003 – 2007. We collect conference call data from earnings conference call transcripts compiled by ThomsonReuters. The data for all remaining variables is sourced from Compustat, ExecuComp, CRSP, and RiskMetrics (formerly IRRC) datasets. We average the annual observations of each variable for a given firm across all available years within our timeframe. Therefore, the unit of observation for each variable is the firm level, rather than the firm-year level. (t = year, m = month, i = firm, j = executive)

Variable Description Formula CEO’s Real Authority Percentage of CEO

Text The ratio of the number of characters spoken by the CEO during the conference call to the number of characters spoken by all company executives during the conference call. This variable was created by parsing the text of earnings conference call transcripts acquired from ThomsonReuters. See Table 1 for a description of the conference call selection process. We create an annualized version of this variable by averaging the data across all conference calls within a fiscal year (we assume that conference calls that occur up to 3 months after the fiscal year-end are associated with earnings for that year). We multiply this variable by negative 1 such that larger values indicate higher real authority at the subordinate level.

T

SpokenCharacters

SpokenCharacters

tJ

jtij

tiCEO∑∑ ⎟⎟

⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜

⎝

⎛

==1

,,

,,

CEO’s Formal Authority CEO Founder This is a dummy variable = 1 if the CEO was a company founder; and 0

otherwise. To determine if the CEO was a company founder, we first established the date in which the CEO joined the company by using the earlier of the becameceo and joined_co variables from ExecuComp. We then established the year in which the company was founded by using the firm age data used in Jovanovic and Rousseau (2001) as downloaded from Boyan Jovanovic’s website: http://www.nyu.edu/econ/user/jovanovi/. This data provides the founding year, incorporation year and exchange listing year for approximately 7,700 firms. Because the founding year data is frequently missing, to establish the first year for the firm, we first use the founding year, then the incorporation year and finally the listing year. We then compare this founding year to the year that the CEO joined the company. If the difference between these years is less than 2, we then set the CEO Founder variable = 1. As the founding year (instead the incorporation or listing year) is not used for all firms, the CEO Founder variable may be overstated. On the other hand, if a founding year was missing, then we assumed that the CEO was not a founder to avoid dropping observations. This may potentially understate this variable.

T

CEOFoundert

ti∑=

,

43

CEO Title Concentration

This is a dummy variable = 1 if the CEO is the Chair of the Board and the President (or the CEO is the Chair of the Board and no other executive is the President or COO); and 0 otherwise. To determine the titles of the executives, we use the RiskMetrics database. This database contains all company directors and identifies their position within the company if the director is also an employee. Specifically, RiskMetrics has dummy variables identifying the chairman, president, and coo. If the CEO is also identified as the Chairman and President, or is identified as the Chairman and no other director is identified as the President, then CEO Title Concentration = 1; and is 0 otherwise. Using RiskMetrics may overstate this variable somewhat in the case in which a company has a President, but the President is not also a director.

T

nncentratioCEOTitleCot

ti∑=

,

CEO Only Insider This is a dummy variable = 1 if the CEO is the only company employee on the Board of Directors; 0 otherwise. The RiskMetrics database classifies all directors as employee (E), linked (L) or independent (I). This variable is coded as 1 if the CEO is the only employee (E) listed on the Board according to RiskMetrics.

T

iderCEOOnlyInst

ti∑=

,

Determinants of CEO’s Real Authority Non-CEO Equity

Sensitivity This variable measures the sensitivity of the wealth of the top 4 non-CEO executives to a 1% change in stock price. We measure this variable as the sum of total shares held (shrown_excl_opts), total unexercised unexercisable options (opt_unex_unexer_num) and total unexercised exercisable options (opt_unex_exer_num) for all 4 top executives—excluding the CEO—times the fiscal year-end stock price divided by 100. Before summing the three variables, we multiply the options holdings by the estimated delta of the options. This is necessary because options portfolio values are less sensitive to changes in stock price than stock portfolios. We calculate the delta using the Black-Scholes model. This model requires the following inputs (our assumptions in parentheses): dividend yield (3-year average yield prior to year of observation), expected volatility (standard deviation of stock returns during prior 60 months), number of years until option maturity (assumed 5 years for all options), current stock price (prcc_f from Compustat), strike price (assumed to be the price needed to derive the intrinsic value reported in ExecuComp), and risk-free rate (assumed the risk-free rate reported in ExecuComp). Note that some prior studies have calculated separate deltas for newly granted options and all other options; however, starting with fiscal year 2006, ExecuComp no longer reports the expiration date of the newly granted options, requiring us to make a similar set of assumptions for new and old options so we aggregated all options. We recoded any missing ExecuComp

( )

T

iceoptionsdeltashares

t

jtijtitij

∑∑

⎟⎟⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜

⎝

⎛⎟⎟⎠

⎞⎜⎜⎝

⎛+

=

=

100

Pr**(4

1,,,,,

44

variables to zero if missing. Product Market Competition

This is a measure of the competitiveness within an industry, as proxied by firm concentration. For each year and 4 digit SIC in Compustat, we calculate the Herfindahl index based on the sales for each firm in the industry. The Herfindahl index is calculated as the sum of the squared market shares of each firm within the 4 digit SIC. For each year, we then create deciles of the industry Herfindahl indices. The Product Market Competition variable is a dummy variable = 1 if the industry is in the lowest (most competitive) decile; and zero otherwise.

T

dummydecileHerfindahlt

ti∑=

,__

Regulated Industry This is a dummy variable = 1 if the firm is in a Utilities industry (as defined by the Fama-French 12 industry segmentation).

T

dummyindustryUtilitiest

ti∑=

,__

R&D/Sales Research and development expenses (Compustat data item xrd) divided by sales (Compustat data item sale). We re-code missing values of R&D to zero to avoid significant observation losses.

TSales

DR

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,&

LN(Employees) The natural log of the total number of employees of the firm (Compustat data item emp).

T

EmployeesTotalLNt

ti∑=

,)_(

CEO Characteristics CEO (CFO)

Ownership This is the percentage of outstanding company shares owned by the CEO (CFO). We use the shrown_excl_opts variable in ExecuComp to identify the number of shares held by the CEO (CFO) and the variable shrsout to identify the total number of shares outstanding for the firm. This variable may contain noise because the measurement of shrsout is at the end of the fiscal year, whereas shrown_excl_opts may be measured at a date after the fiscal year and before the proxy statement date. To incorporate the potential for non-linear relations, we also use the squared version of this variable, denoted CEO Ownership2 (CFO Ownership2).

Tshrsout

optsexclshrown

t ti

tiCEO∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,,__

CEO Tenure This is the number of years that the CEO has been in office. We subtract the becameceo variable in ExecuComp from the year variable and add 1 (to count the first year in office as 1). To incorporate the potential for non-linear relations, we also use the squared version of this variable, denoted CEO Tenure2.

( )T

becameceoyeart

titi∑ +−=

1,,

Prestige This is a dummy variable = 1 if the firm is listed as one of the top 5 companies in its industry of Fortune Magazine’s annual America’s Most Admired Companies survey. T

CompanyTopt

ti∑=

,_5_

45

Overconfidence This measures the extent to which the CEO holds unexercised exercisable options. Its measured as the intrinsic value of the unexercised exercisable options divided by the value of the total holdings of the CEO, which is the sum of the value of unexercised exercisable options, unexercised unexercisable options, and shares of stock. T

holdingsofvaluevalestexerunexopt

t tiCEO

tiCEO∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,,

,,

)__(____

Firm Characteristics and Controls Board Size This is the number of directors on the Board of Directors. We count the

number of directors listed in the RiskMetrics database for the measurement of this variable. T

DirectorsofNumbert

ti∑=

,__

Percentage Insiders This is the percentage of insiders on the Board of Directors. The RiskMetrics database classifies all directors as employee (E), linked (L) and independent (I). The linked directors include former employees and family members. We count any director that is an employee (E) or linked (L) as an insider. T

directorsofTotalInsidersof

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,

___#__#

Percentage of Outsiders over Age 69

This is the number of directors on the Board of Directors who are over the age of 69 divided by the total number of directors. Director ages and the total number of directors are taken from the RiskMetrics database.

TdirectorsofTotal

AgeDirectorsof

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛ >

= ,

,

___#69__#

LN(Assets) The natural log of total company assets (Compustat Fundamentals Annual dataset item at).

T

AssetsTotalLNt

ti∑=

,)_(

Growth The year-over-year sales growth as defined by (salest – salest-1)/salest-1

TSales

SalesSales

t ti

titi∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛ −

= −

−

1,

1,,

ROA Return on Assets as measured by Income before Extraordinary Items (Compustat data item ib) divided by year-end assets (Compustat data item at).

TAssetsTotal

ItemsaryExtraordinbeforeIncome

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,

____

Returns Annual returns for the fiscal year calculated by compounding the monthly returns from the CRSP monthly data file (variable ret in the CRSP file).

T

rett m

tmi∑ ∏ ⎥⎦

⎤⎢⎣

⎡−⎟⎟

⎠

⎞⎜⎜⎝

⎛+

= =

1)1(12

1,,

ROA Volatility The standard deviation of annual ROA for the 5 year window from t-4 to t.

TAssetsTotal

ItemsaryExtraordinbeforeIncome

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,

____

δ

46

Return Volatility The standard deviation of annual returns for the 5 year window from t-4 to t.

T

rett m

tmi∑ ∏ ⎥⎦

⎤⎢⎣

⎡−⎟⎟

⎠

⎞⎜⎜⎝

⎛+

= =

1)1(12

1,,δ

Capex/Assets Capital expenditures (Compustat data item capx) divided by total assets (Compustat data item at).

TAssetsTotal

esExpenditurCapital

t ti

ti∑ ⎟⎟⎠

⎞⎜⎜⎝

⎛

= ,

,

__

G Index The Gompers, Ishii and Metrick (2003) Governance Index provided by the RiskMetrics database. This is a composite measure of 24 charter provisions. Because this measure is not updated annually, for any year of missing data, we use the previous year’s value. T

GIndext

ti∑=

,

CFO on Board of Directors

This is a dummy variable = 1 if the CFO is on the Board of Directors; 0 otherwise according to the RiskMetrics dataset.

T

dummyBoardonCFOt

ti∑=

,___

Number of Segments

We count the number of segments reported by the firm in the Compustat Segments dataset. For firms that report by more than one segment type (e.g., business, geography, product, etc.), we use the segment type with the highest number of reported segments. T

SegmentsofNumbert

ti∑=

,__

Compensation Measures CEO-to-Top 5

Compensation The ratio of total CEO compensation to the top 4 highest paid executives plus the CEO. Total compensation for the variable is the same as defined for the LN(CEO Total Compensation) variable, using tdc1 from ExecuComp.

T

tdc

tdc

t

jtij

tiCEO∑∑ ⎟⎟

⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜

⎝

⎛

==

5

1,,

,,

1

1

LN(CEO Total Compensation)

Natural log of CEO total compensation. Total compensation includes salary, bonus, other annual compensation, restricted stock grants, long-term incentive program payouts, and value of option grants (as determined by the Black-Scholes formula) as reported by ExecuComp’s tdc1 variable. 1 is added to the tdc1 value prior to taking the natural log of the variable.

T

tdcLNt

tiCEO∑ +=

)11( ,,

47

Figure 1: Distribution of Conference Call Length

050

010

0015

0020

00Fr

eque

ncy

0 20000 40000 60000 80000 100000Amount of Text Spoken by Company Personnel

This figure shows the distribution of the number of characters spoken by company personnel during the 17,400 earnings conference calls in our sample. Table 2 reports the descriptive statistics.

Figure 2: Distribution of the Percentage of CEO Text

050

010

0015

0020

00Fr

eque

ncy

0 .2 .4 .6 .8 1Percentage of Text Spoken by CEO

This figure shows the distribution of the amount of text spoken by the CEO as a percent of total company personnel speech during the 17,400 earnings conference calls in our sample. During approximately 9% of the conference calls the CEO was not present (or at least did not make any comments). Table 2 reports the descriptive statistics.

48

Number of transcripts received from ThomsonReuters 129,924 (January 2001 - September 2008)

Transcripts which contain foreign characters (23,981)

Not "Earnings" related (24,840)

Do not have both a presentation section and a Q&A section (4,185)

Not an ExecuComp firm (44,235)

Unable to identify at least one speaker on the conference call (10,705) More than 1 CEO or CFO in the conference call identified (642)

More than 1 conference call in a month (493)

No data in Compustat (26)

Keep only years 2003 - 2007 (3,417)

Eligible conference calls 17,400

Data for conference calls occurring the same fiscal year are averaged within the year resulting in the following number of firm-years 6,854

Insufficient data in Compustat, ExecuComp or RiskMetrics (2,446)

Eligible firm-years 4,408

All variables are averaged across time within each firm to derive the following number of firm observations 1,372

Require a minimum of 2 annual observations per firm 1,142

TABLE 1: Conference Call Selection

This table reports the selection of conference calls. We received 129,924 individual conference call transcript files from ThomsonReuters. We then used FORTRAN to parse these text files to determine the amount of text spoken by each individual. We only kept conference calls which indicated that they were related to earnings (as indicated in the header of the text file). Additional conference calls were eliminated because of formatting issues with the text file or lack of data availability as described in this table. The conference calls are generally held on a quarterly basis; however, because all other data used in our study is on an annual basis, we average the conference call data within a year for a given firm. We then create a firm level observation by averaging all annual values for a firm across all years of data available. Finally, we require a minimum of 2 annual observations to be included in the sample, reducing the number of firm observations to 1,142.

49

Panel A: All conference callsStandard

Variable Mean Median Deviation Minimum Maximum NTotal Length 32,484 32,412 9,865 1,998 100,986 17,400 Total Comments 49 46 23 1 223 17,400 Analyst Questions 42 40 21 0 203 17,400 Percentage CEO Text 48% 51% 24% 0% 100% 17,400 Percentage CFO Text 33% 32% 20% 0% 100% 17,400 Percentage CEO Comments 49% 52% 25% 0% 100% 17,400 Percentage CFO Comments 32% 30% 22% 0% 100% 17,400

Panel B: By yearTotal Total Analyst

Year Length Comments Questions CEO CFO CEO CFO N %2003 30,512 47 42 46% 33% 46% 32% 1,832 10.5%2004 31,571 49 42 47% 33% 46% 33% 3,092 17.8%2005 32,734 50 43 47% 33% 48% 33% 3,714 21.3%2006 32,977 49 41 49% 32% 50% 32% 4,102 23.6%2007 33,232 49 42 50% 32% 51% 31% 4,660 26.8%

Panel C: By industryTotal Total Analyst

Industry Length Comments Questions CEO CFO CEO CFO N %Non Durable Consumer Goods 33,669 52 45 47% 32% 47% 32% 922 5.3%Durable Consumer Goods 32,752 55 47 50% 32% 50% 32% 415 2.4%Manufacturing 31,002 53 46 53% 32% 54% 31% 2,284 13.1%Energy 30,970 52 47 50% 19% 51% 17% 683 3.9%Chemicals 33,576 54 48 51% 30% 54% 29% 480 2.8%Business Equipment 32,923 47 40 52% 33% 53% 32% 3,620 20.8%Telecom 33,799 35 28 40% 35% 40% 33% 387 2.2%Utilities 27,454 44 38 36% 42% 36% 41% 771 4.4%Shops 33,415 50 42 46% 34% 47% 34% 2,102 12.1%Health 34,326 48 40 44% 32% 45% 31% 1,203 6.9%Money 31,889 47 40 43% 34% 43% 34% 2,419 13.9%Other 33,306 51 43 50% 33% 50% 33% 2,114 12.1%

TABLE 2: Conference Call Descriptive Statistics

Text Comments

Text Comments

This table provides descriptive statistics for the 17,400 earnings conference calls that were selected as described in Table 1. Total Length = the total number of characters (letters) spoken by a company employee during the conference call; Total Comments = the total number of times that a company employee spoke during the conference call; Analyst Questions = the number of times that an analyst spoke; Percentage CEO (CFO) Text = the total number of characters spoken by the CEO (CFO) divided by the Total Length; Percentage CEO (CFO) Comments = the total number of times that the CEO (CFO) spoke divided by Total Comments.

50

Standard LagVariable Mean Median Deviation Minimum Maximum N CorrelationCEO's Real Authority

Percentage CEO Text 47.5% 49.0% 20.2% 0.0% 99.7% 1,142 0.77CEO's Formal Authority

CEO Founder 12.2% 0.0% 31.2% 0.0% 100.0% 1,142 0.92CEO Title Concentration 50.3% 50.0% 42.0% 0.0% 100.0% 1,142 0.68CEO Only Insider 59.4% 75.0% 42.6% 0.0% 100.0% 1,142 0.73

CEO CharacteristicsCEO Ownership 1.7% 0.3% 4.3% 0.0% 49.9% 1,142 0.61CEO Tenure 8.07 6.00 6.54 1 44 1,117 0.85Prestige 0.18 0.00 0.33 0.00 1.00 1,142 0.73Overconfidence 29.0% 26.3% 22.6% 0.0% 97.0% 1,142 0.70

Determinants of CEO's Real AuthorityNon-CEO Equity Sensitivity 0.95 0.31 6.80 0.01 222.88 1,132 0.90Product Market Competition 0.39 0.00 0.47 0.00 1.00 1,142 0.90Regulated Industry 0.07 0.00 0.26 0.00 1.00 1,142 1.00R&D/Sales 0.04 0.00 0.10 0.00 1.33 1,142 0.88LN(Employees) 1.88 1.81 1.57 (4.38) 7.28 1,138 0.99

Firm characteristics and ControlsBoard Size 9.41 9.20 2.28 4.67 22.60 1,142 0.89Percentage Insiders 27.7% 26.3% 12.6% 6.2% 100.0% 1,142 0.79Percentage Outsiders over Age 69 9.7% 8.0% 9.7% 0.0% 59.3% 1,142 0.74LN(Assets) 7.95 7.77 1.66 3.62 14.23 1,142 0.99Growth 13.1% 10.4% 14.2% -27.3% 195.6% 1,142 0.29ROA 4.9% 4.7% 6.8% -92.6% 39.9% 1,142 0.62Returns 18.8% 16.3% 20.5% -42.2% 145.1% 1,142 (0.01)ROA Volatility 4.3% 2.2% 7.9% 0.1% 137.4% 1,142 0.88Return Volatility 43.6% 34.3% 34.7% 4.0% 419.3% 1,129 0.78Capex/Assets 0.04 0.03 0.05 0.00 0.33 1,142 0.88G Index 9.48 9.00 2.50 2.00 18.00 1,106 0.98

Compensation MeasuresCEO-to-Top 5 Compensation 38.3% 38.4% 9.2% 0.0% 77.7% 1,142 0.40LN(CEO Total Compensation) 8.12 8.12 0.96 0.00 10.82 1,142 0.73

TABLE 3: Descriptive Statistics

This table presents the descriptive statistics for our measure of the CEO’s real authority (Percentage CEO Text) and formal authority, CEO characteristics, determinants of the CEO’s real authority, firm characteristics and controls, and compensation measures. Lag correlation is the Pearson correlation between the variable and its one year lag value for the pooled sample. See Appendix B for variable definitions.

51

Panel A: Correlation of CEO's Real Authority Measures with Determinant Variables

N = 1,128

Perc

enta

ge C

EO

Te

xt

Perc

enta

ge C

EO

N

umbe

r

CEO

Pre

sent

Non

-CE

O E

quity

Se

nsiti

vity

Prod

uct M

arke

t Co

mpe

titio

n

Util

ities

Indu

stry

R&D

/Sale

s

Percentage CEO Number 0.93

CEO Present 0.63 0.64

Non-CEO Equity Sensitivity (0.10) (0.10) (0.15)

Product Market Competition (0.12) (0.12) (0.05) 0.04

Regulated Industry (0.15) (0.15) (0.06) (0.02) 0.25

R&D/Sales (0.00) 0.01 0.03 0.03 0.13 (0.12)

LN(Employees) (0.24) (0.25) (0.18) 0.10 (0.15) (0.03) (0.27)

TABLE 4: Pearson Correlations

Panel A presents the Pearson correlations between three potential variables which may proxy for the extent of the CEO’s real authority: Percentage CEO Text (defined in Appendix B), Percentage CEO Number (the number of comments that the CEO makes divided by the total number of comments made by all company personnel), and CEO Present (a dummy variable = 1 if the CEO is present on the conference call and 0 otherwise) and the variables which proxy for the determinants of the CEO’s real authority. Panel B presents the Pearson correlations for our measure of the CEO’s real authority (Percentage CEO Text) and measures of the CEO’s formal authority, CEO characteristics, firm controls, and compensation. See Appendix B for variable definitions. Bold indicates significance at the 5% level.

52

Panel B: Correlation of All Other Variables

N = 1,077

Perc

enta

ge C

EO

Te

xt

CEO

Fou

nder

CEO

Titl

e Co

ncen

tratio

n

CEO

Onl

y In

sider

CEO

Ow

ners

hip

CEO

Ten

ure

Pres

tige

Ove

rcon

fiden

ce

Boar

d Si

ze

Perc

enta

ge In

sider

s

Perc

enta

ge O

utsid

ers

> A

ge 6

9

LN(A

sset

s)

Gro

wth

ROA

Retu

rns

ROA

Vol

atili

ty

Retu

rn V

olat

ility

Cape

x/A

sset

s

G In

dex

CEO

-to-T

op 5

Rat

io

CEO Founder (0.02)

CEO Title Concentration 0.04 0.00

CEO Only Insider 0.10 (0.18) 0.36

CEO Ownership (0.01) 0.27 0.04 (0.13)

CEO Tenure (0.05) 0.47 0.16 (0.15) 0.43

Prestige (0.22) (0.08) 0.03 (0.07) (0.09) (0.05)

Overconfidence 0.04 (0.13) 0.04 0.09 (0.31) (0.15) 0.05

Board Size (0.23) (0.11) (0.00) (0.15) (0.13) (0.15) 0.34 0.01

Percentage Insiders 0.03 0.20 (0.27) (0.47) 0.26 0.21 (0.08) (0.16) (0.12)

Percentage Outsiders > Age 69 (0.04) 0.08 (0.04) (0.03) 0.04 0.14 0.00 (0.05) (0.00) 0.01

LN(Assets) (0.31) (0.06) 0.11 (0.07) (0.14) (0.13) 0.51 0.01 0.64 (0.18) 0.06

Growth 0.02 0.04 0.03 (0.05) (0.02) 0.06 (0.03) 0.07 (0.09) 0.07 0.09 (0.04)

ROA 0.00 (0.03) (0.05) (0.06) (0.01) 0.06 0.12 0.18 (0.01) (0.03) 0.01 0.00 (0.05)

Returns 0.02 0.03 0.04 0.03 (0.02) 0.01 0.02 0.11 (0.03) (0.04) 0.02 0.01 0.27 0.17

ROA Volatility 0.05 (0.01) (0.05) 0.08 (0.03) (0.04) (0.11) (0.05) (0.20) (0.01) 0.00 (0.22) 0.10 (0.27) 0.03

Return Volatility 0.07 0.08 (0.00) 0.06 0.05 0.07 (0.18) 0.00 (0.31) 0.04 (0.02) (0.29) 0.16 (0.16) 0.21 0.44

Capex/Assets (0.03) (0.01) 0.01 (0.00) 0.01 0.05 (0.02) 0.01 (0.13) 0.03 0.06 (0.15) 0.14 0.21 0.03 (0.03) 0.01

G Index 0.02 (0.09) 0.13 0.12 (0.15) (0.12) 0.03 0.06 0.24 (0.22) (0.08) 0.13 (0.08) (0.00) 0.08 (0.10) (0.17) (0.03)

CEO-to-Top 5 Ratio 0.14 (0.08) 0.20 0.22 (0.19) (0.05) 0.04 0.15 0.07 (0.21) (0.01) 0.12 (0.06) 0.11 0.06 (0.05) (0.11) 0.02 0.15

LN(CEO Total Compensation) (0.14) (0.07) 0.14 0.04 (0.25) (0.11) 0.42 0.18 0.38 (0.23) 0.03 0.65 0.02 0.18 0.11 (0.08) (0.14) (0.06) 0.12 0.53

TABLE 4: Pearson Correlations

Panel A presents the Pearson correlations between three potential variables which may proxy for the extent of the CEO’s real authority: Percentage CEO Text (defined in Appendix B), Percentage CEO Number (the number of comments that the CEO makes divided by the total number of comments made by all company personnel), and CEO Present (a dummy variable = 1 if the CEO is present on the conference call and 0 otherwise) and the variables which proxy for the determinants of the CEO’s real authority. Panel B presents the Pearson correlations for our measure of the CEO’s real authority (Percentage CEO Text) and measures of the CEO’s formal authority, CEO characteristics, firm controls, and compensation. See Appendix B for variable definitions. Bold indicates significance at the 5% level.

53

Independent variables(1) (2) (3) (4) (5) (6) (7)

Determinants of CEO's Real AuthorityNon-CEO Equity Sensitivity (0.185) *** (0.170) *** (0.128) *** (Incentive effects) (6.83) (6.11) (5.96)Product Market Competition (2.890) ** (2.008) (2.079) (Urgency effects) (2.22) (1.38) (1.42)Regulated Industry (8.379) *** (9.580) *** (10.870) *** (Task importance effects) (3.92) (4.21) (4.61)R&D/Sales (16.630) ** (18.770) ** (19.390) ** (Expertise effects) (2.30) (2.49) (2.54)LN(Employees) (0.790) (1.612) *** (1.288) ** (Span of control effects) (1.56) (2.85) (2.28)

CEO's Formal AuthorityCEO Founder (0.092)

(0.04)CEO Title Concentration 2.997 *

1.95CEO Only Insider 2.437

1.50CEO Characteristics

CEO Ownership (0.664) *(1.89)

CEO Ownership2 0.025 ***2.67

CEO Tenure 0.494 *1.66

CEO Tenure2 (0.024) **(2.38)

Prestige (4.774) **(2.25)

Overconfidence (0.151)(0.05)

Table 5: Determinants of the CEO's Real Authority

Percentage CEO Text

(Continued)

54

Firm Characteristics and ControlsBoard Size (0.730) * (0.716) * (0.673) * (0.747) ** (0.675) * (0.687) * (0.463)

(1.93) (1.89) (1.77) (1.96) (1.78) (1.80) (1.22)Percentage Insiders (0.021) (0.031) (0.041) (0.035) (0.034) (0.044) 0.076

(0.43) (0.65) (0.86) (0.74) (0.73) (0.92) 1.47Percentage Outsiders over Age 69 (0.047) (0.047) (0.045) (0.045) (0.047) (0.059) (0.019)

(0.69) (0.68) (0.66) (0.67) (0.69) (0.87) (0.28)LN(Assets) (3.248) *** (3.217) *** (3.235) *** (3.418) *** (2.984) *** (2.207) *** (2.203) ***

(6.61) (6.45) (6.52) (6.98) (5.23) (3.49) (3.36)Growth 0.970 1.256 0.034 2.167 (0.450) (0.198) (0.688)

0.25 0.32 0.01 0.55 (0.12) (0.05) (0.18)ROA (0.013) (0.042) (0.062) (0.068) (0.002) (0.050) 0.002

(0.16) (0.49) (0.71) (0.81) (0.02) (0.61) 0.03Returns 0.022 0.021 0.029 0.015 0.020 0.012 0.012

0.69 0.67 0.92 0.48 0.64 0.38 0.39ROA Volatility (0.024) (0.030) (0.028) 0.019 (0.024) 0.031 0.026

(0.41) (0.52) (0.46) 0.35 (0.40) 0.58 0.47Return Volatility (0.017) (0.016) (0.024) (0.006) (0.018) (0.011) (0.008)

(0.98) (0.92) (1.37) (0.30) (1.01) (0.56) (0.44)G Index 0.502 ** 0.447 * 0.519 ** 0.455 * 0.543 ** 0.584 ** 0.465 *

2.03 1.82 2.09 1.84 2.21 2.39 1.90

Intercept 76.790 *** 78.270 *** 77.550 *** 79.210 *** 75.700 *** 73.380 *** 65.160 ***16.77 17.17 16.92 16.97 15.74 14.83 11.64

Industry Fixed Effects? N N N N N N NHeteroscedasticity Robust SE? Y Y Y Y Y Y Y

Adj. R2 9.8% 9.8% 10.5% 9.7% 9.7% 12.1% 14.8%N 1,092 1,100 1,100 1,100 1,096 1,088 1,067

This table presents the regression results examining the CEO’s real authority. In each model the dependent variable is our measure the CEO’s real authority, Percentage CEO Text. Models (1) through (5) examine each of the determinants of the CEO’s real authority separately with firm controls. In model (6)—our formal test of H1—we include all determinants of the CEO’s real authority. In model (7) we test H1a by including variables for the CEO’s formal authority and CEO characteristics. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

55

Table 5, (Continued)

CEO CEO Title CEO OnlyFounder Concentration Insider

Independent variables(1) (2) (3)

Determinants of CEO's Real AuthorityNon-CEO Equity Sensitivity (0.012) (0.390) ** (0.368) ** (Incentive effects) (0.26) (2.42) (2.22)Product Market Competition 1.533 1.098 0.157 (Urgency effects) 0.74 0.38 0.06Regulated Industry (6.687) *** 5.359 15.270 *** (Task importance effects) (2.69) 1.11 3.38R&D/Sales (10.300) (7.801) 5.507 (Expertise effects) (0.65) (0.53) 0.37LN(Employees) (0.938) 2.689 ** 1.558 (Span of control effects) (0.99) 2.46 1.41

CEO CharacteristicsCEO Ownership 0.947 (0.294) (0.702)

1.21 (0.44) (1.01)

CEO Ownership2 (0.018) 0.036 ** 0.029(0.69) 2.31 1.28

CEO Tenure 1.641 *** 3.307 *** (0.004)3.75 6.55 (0.01)

CEO Tenure2 0.008 (0.061) *** (0.015)0.55 (3.94) (1.11)

Prestige (5.038) * (2.444) (2.674)(1.65) (0.58) (0.64)

Overconfidence (3.110) 4.029 1.349(0.79) 0.65 0.25

Measures of Formal Authority

Table 6: Analysis of the CEO's Formal Authority

(Continued)

56

Table 6, (Continued) Firm Characteristics and Controls

Board Size (0.760) (2.410) *** (4.049) ***(1.35) (3.25) (6.00)

Percentage Insiders 0.218 *** (0.951) *** (1.543) ***3.06 (9.71) (15.89)

Percentage Outsiders over Age 69 0.039 (0.288) ** (0.008)0.39 (2.15) (0.07)

LN(Assets) 2.255 ** 2.820 ** (1.544)2.37 2.39 (1.33)

Growth (2.039) 17.070 * (10.670)(0.33) 1.86 (1.31)

ROA (0.227) * (0.554) *** (0.410) **(1.68) (3.23) (2.39)

Returns 0.052 0.009 0.0581.06 0.14 0.94

ROA Volatility (0.064) (0.286) ** 0.072(0.66) (1.98) 0.71

Return Volatility 0.039 0.038 0.0071.19 0.88 0.18

G Index (0.016) 1.630 *** 0.834 *(0.05) 3.21 1.83

Intercept (17.740) ** 38.050 *** 144.600 ***(2.16) 3.39 14.72

Industry Fixed Effects? N N NHeteroscedasticity Robust SE? Y Y Y

Adj. R2 23.9% 16.9% 28.4%N 1,067 1,067 1,067

This table presents the regression results examining three measures of the CEO’s formal authority. The dependent variables (CEO formal authority proxies) are listed above each of the columns. We include our measures for the determinants of the CEO’s real authority and CEO characteristics in each model. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

57

CEO-to- CEO-to- CEO-to- LN(Comp- CEO-to- LN(Comp-Top 5 Top 5 Top 5 ensation) Top 5 ensation)

Independent variables(1) (2) (3) (4) (5) (6)

Determinants of CEO's Real AuthorityPercentage CEO Text 8.654 *** 7.509 *** 3.162 ** 5.735 *** 2.185 *

5.70 4.76 2.27 4.62 1.71CEO's Formal Authority

CEO Founder (0.237) (0.288) 0.399 (1.192) (0.320)(0.22) (0.27) 0.55 (1.34) (0.48)

CEO Title Concentration 2.323 *** 2.049 *** 0.816 1.764 *** 1.276 ***3.22 2.90 1.53 3.16 2.81

CEO Only Insider 3.110 *** 2.952 *** 0.739 1.591 *** (0.087)3.98 3.84 1.35 2.66 (0.18)

CEO CharacteristicsCEO Ownership (0.463) ** (0.409) * (0.472) * (0.315) * (0.241)

(2.12) (1.88) (1.78) (1.78) (1.01)

CEO Ownership2 0.006 0.004 0.007 0.004 0.0031.14 0.75 1.21 1.14 0.57

CEO Tenure 0.260 * 0.233 0.198 ** 0.258 ** 0.151 *1.75 1.63 2.11 2.53 1.94

CEO Tenure2 (0.005) (0.003) (0.004) (0.003) (0.002)(0.94) (0.68) (1.09) (0.94) (0.67)

Prestige (2.015) * (1.580) (0.156) (1.098) (0.669)(1.95) (1.59) (0.19) (1.38) (0.85)

Overconfidence 1.137 1.073 1.824 2.223 ** 1.873 *0.86 0.81 1.64 2.11 1.86

Cash OnlyTotal Compensation

Table 7: CEO Compensation Regressions

(Continued)

58

Table 7, (Continued) Firm Characteristics and Controls

Board Size (0.256) (0.095) (0.048) 0.186 0.091 0.321 **(1.46) (0.51) (0.27) 1.29 0.67 2.51

Percentage Insiders (0.097) *** (0.014) (0.021) (0.048) *** (0.012) (0.005)(4.18) (0.56) (0.84) (2.62) (0.64) (0.31)

Percentage Outsiders over Age 69 (0.012) (0.021) (0.018) (0.006) 0.028 0.027(0.44) (0.75) (0.64) (0.29) 1.27 1.42

LN(Assets) 0.984 *** 0.788 *** 1.030 *** 4.360 *** 0.387 * 2.606 ***3.79 2.59 3.32 18.25 1.69 12.97

Growth (3.296) (3.771) * (3.728) * 1.994 (0.226) 1.020(1.52) (1.78) (1.75) 1.24 (0.12) 0.59

ROA 0.099 ** 0.131 *** 0.122 *** 0.180 *** 0.102 *** 0.101 ***2.44 3.14 2.95 5.72 3.16 3.36

Returns 0.029 ** 0.023 0.022 0.026 0.013 0.0172.02 1.47 1.49 1.59 1.05 1.06

ROA Volatility 0.045 0.057 * 0.057 * 0.076 ** 0.025 (0.005)1.42 1.74 1.75 2.48 0.99 (0.35)

Return Volatility (0.020) ** (0.022) ** (0.021) ** 0.009 (0.015) ** (0.002)(1.96) (2.25) (2.12) 1.20 (2.40) (0.45)

G Index 0.282 ** 0.237 * 0.199 0.108 0.225 ** 0.177 **2.32 1.93 1.63 1.13 2.53 2.25

Intercept 29.130 *** 27.620 *** 22.370 *** 39.500 *** 22.310 *** 41.540 ***10.57 9.53 7.01 13.10 9.03 15.87

Industry Fixed Effects? Y Y Y Y Y YHeteroscedasticity Robust SE? Y Y Y Y Y Y

Adj. R2 15.2% 17.0% 19.3% 59.8% 19.9% 48.4%N 1,100 1,077 1,077 1,077 1,077 1,077

This table presents the regression results examining CEO compensation. In models (1) through (4) we use the CEO’s total compensation as the dependent variable; while in models (5) and (6) we examine cash compensation (salary and bonus) only. In models (1), (2), (3), and (5) we scale the CEO’s compensation by the compensation of the CEO and other top four executives; while in models (4) and (6) the dependent variable is natural log of CEO’s compensation without scaling. We multiplied LN(Compensation) by a factor of 10 for easier readability of the coefficients in the tables. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

59

CFO-to- CFO-to- LN(Comp- CFO-to- LN(Comp-Top 5 Top 5 ensation) Top 5 ensation)

Independent variables(1) (2) (3) (4) (5)

Determinants of CFO's Real AuthorityPercentage CFO Text 1.933 ** 1.470 * 1.226 2.079 *** 2.371 ***

2.27 1.75 1.34 3.06 2.94CFO Characteristics

CFO Ownership 6.830 *** 6.022 *** 4.008 *** 4.041 ***4.81 3.55 3.71 3.62

CFO Ownership2 (2.112) *** (2.373) *** (1.154) * (0.916)(2.70) (2.69) (1.93) (1.29)

CFO on Board of Directors 2.709 *** 1.740 *** 2.679 *** 1.934 ***3.83 2.62 5.59 3.74

CEO's Formal AuthorityCEO Founder (0.717) (0.316) 0.160 (0.267)

(1.25) (0.55) 0.40 (0.67)CEO Title Concentration (0.189) 0.290 (0.329) 0.449 *

(0.51) 0.78 (1.33) 1.67CEO Only Insider 0.804 ** 0.346 1.114 *** 0.542 *

2.12 0.77 3.82 1.72CEO Characteristics

CEO Ownership 0.006 (0.045) (0.047) 0.0430.06 (0.49) (0.59) 0.68

CEO Ownership2 (0.001) (0.001) 0.002 (0.001)(0.46) (0.42) 1.37 (0.90)

CEO Tenure (0.072) 0.045 (0.132) (0.030)(0.90) 0.68 (1.98) (0.62)

CEO Tenure2 0.003 (0.001) 0.003 0.0011.08 (0.45) 1.33 0.50

Prestige 0.414 0.791 0.227 0.2540.88 1.54 0.65 0.62

Overconfidence 0.733 1.574 (0.338) 0.7761.05 2.27 (0.68) 1.51

Cash OnlyTotal Compensation

Table 8: CFO Compensation Regressions

(Continued)

60

Table 8 (Continued)

Firm Characteristics and ControlsBoard Size 0.102 0.095 0.176 ** (0.008) 0.135 *

1.24 1.13 1.97 (0.13) 1.84Percentage Insiders 0.021 ** 0.025 ** (0.018) 0.005 0.015

1.98 2.08 (1.31) 0.52 1.61Percentage Outsiders over Age 69 0.028 ** 0.027 * 0.017 0.018 * 0.024 **

2.01 1.94 1.08 1.79 2.24LN(Assets) (0.555) *** (0.414) *** 4.024 *** (0.253) ** 2.512 ***

(4.84) (3.09) 27.65 (2.49) 19.53Growth (0.303) (0.119) 4.218 *** (0.440) 1.134

(0.26) (0.10) 3.51 (0.56) 1.37ROA (0.018) (0.022) 0.111 *** (0.019) 0.050 ***

(0.75) (0.84) 4.44 (1.27) 2.59Returns 0.010 0.007 0.032 *** 0.009 * 0.031 ***

1.39 1.02 4.19 1.75 5.41ROA Volatility (0.037) * (0.027) 0.054 ** 0.008 0.000

(1.76) (1.39) 2.54 0.73 0.01Return Volatility 0.004 0.004 0.022 *** 0.001 0.005

0.66 0.76 3.99 0.35 1.29G Index (0.013) (0.034) 0.066 (0.011) 0.092 **

(0.22) (0.57) 1.12 (0.24) 2.04

Intercept 16.850 *** 14.970 *** 33.130 *** 16.850 *** 36.940 ***12.98 9.15 20.04 14.07 28.56

Industry Fixed Effects? Y Y Y Y YHeteroscedasticity Robust SE? Y Y Y Y Y

Adj. R2 7.3% 11.9% 69.2% 15.1% 64.6%N 1,080 1,058 1,058 1,058 1,058

This table presents the regression results examining CFO compensation. In models (1) through (3) we use the CFO’s total compensation as the dependent variable; while in models (4) and (5) we examine cash compensation (salary and bonus) only. In models (1), (2), and (4) we scale the CFO’s compensation by the compensation of the CEO, CFO and other top three executives; while in models (3) and (5) the dependent variable is natural log of CFO’s compensation without scaling. We multiplied LN(Compensation) by a factor of 10 for easier readability of the coefficients in the tables. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

61

Independent variables(1) (2) (3) (4)

Determinants of CEO's Real AuthorityNon-CEO Equity Sensitivity (0.190) *** (0.286) (0.358) (0.526) * (Incentive effects) (6.88) (0.87) (1.20) (1.79)Product Market Competition (2.644) (2.105) (1.898) (2.999) ** (Urgency effects) (1.44) (1.44) (1.40) (2.32)Regulated Industry (10.570) *** (10.890) *** (9.363) *** (9.554) *** (Task importance effects) (3.22) (4.62) (4.04) (4.14)R&D/Sales (25.840) ** (19.200) ** (16.160) ** (17.490) *** (Expertise effects) (2.49) (2.52) (2.42) (2.66)LN(Employees) (0.976) (1.280) ** (1.027) ** (0.687) (Span of control effects) (1.35) (2.26) (2.01) (1.03)

Control variable for firm size Assets Assets Assets Sales

N 1,065 1,066 1,204 1,204

Percentage CEO Text

Table 9: Robustness Tests of the Determinants of the CEO's Real Authority

Percentage CEOText w/o CFO

62

This table presents robustness tests of model (7) from Table 5. We made the following modifications to the regressions: (1) Same as model (7) in Table 5 except the dependent variable is the percentage of CEO text after removing the text

spoken by the CFO from the denominator of the calculation. (2) Same as model (7) in Table 5 except Microsoft has been removed from the sample. Bill Gates’ stock holdings create

a substantial outlier in the Non-CEO Equity Sensitivity variable. (3) Same as model (2) in this table except we do not require a minimum of 2 years of data. After we remove this

restriction, this introduces a significant outlier (Arqule, Inc.) for the variable R&D/Sales, which we remove. (4) Same as model (3) in this table except we use LN(Sales) as the firm size control variable instead of LN(Assets). All control variables from Table 5 are included in the regression but are not reported for brevity. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

Independent variables(1) (2) (3)

Number of Segments VariablesNumber of Segments 0.818 ** 0.795 ** 1.980 **

2.37 2.28 2.53Number of Segments2 (0.135) *

(1.85)Determinants of CEO's Real Authority

Non-CEO Equity Sensitivity (0.132) *** (0.131) *** (0.130) *** (Incentive effects) (6.40) (6.35) (6.30)Product Market Competition (1.032) (1.501) (1.419) (Urgency effects) (0.70) (0.99) (0.94)Regulated Industry (10.060) *** (10.670) *** (10.910) *** (Task importance effects) (4.11) (4.25) (4.31)R&D/Sales (19.820) *** (21.810) *** (22.570) *** (Expertise effects) (2.64) (2.88) (3.02)LN(Employees) (1.029) * (0.999) * (Span of control effects) (1.77) (1.71)

N 1,022 1,020 1,020

Table 10: Number of Segments as a Determinant of the CEO's Real Authority

Percentage CEO Text

This table presents the results of model (7) from Table 5 after including the number of operating segments reported by the firm. Model (1) excludes the LN(Employees), whereas Models (2) and (3) include this variable. Model (3) also includes a squared term for the number of segments to account for a potential nonlinear relation. All control variables from Table 5 are included in the regression but are not reported for brevity. See Appendix B for variable definitions. *, **, *** indicate significance at the 10%, 5% and 1% levels, respectively.

63