Management, Auditor, and Audit Committee Influence on MD&A: Evidence from Critical

Accounting Estimate Quantitative Sensitivity Disclosures

by

Matt Glendening

Trulaske College of Business

University of Missouri—Columbia

Elaine Mauldin

Trulaske College of Business

University of Missouri—Columbia

Kenneth W. Shaw

Trulaske College of Business

University of Missouri—Columbia

September, 2014

We thank Adrienne Rhodes and workshop participants at University of Missouri and the 2014 AAA

Annual Meeting for their helpful comments.

Management, Auditor, and Audit Committee Influence on MD&A: Evidence from Critical

Accounting Estimate Quantitative Sensitivity Disclosures

Abstract: Securities and Exchange Commission (SEC) guidance calls on firms to provide

disclosures quantifying the impact on earnings of reasonably likely changes in firms’ critical

accounting estimates (CAE) in Management’s Discussion and Analysis (MD&A). The SEC

intends CAE disclosures to aid users in understanding the extent of noise and discretion within

the accrual estimation process. Consistent with equity risk incentives providing managers the

incentive to misreport, and thus the disincentive to disclose measurement leeway in accrual

estimates, we find the likelihood and number of CAE disclosures are decreasing in the sensitivity

of managers’ wealth to changes in equity risk incentives (portfolio vega). We also find the

likelihood and number of CAE disclosures are decreasing in the external auditor’s expressed

opposition to the SEC’s CAE disclosure requirement, in spite of limited required auditor

involvement in MD&A. Finally, we find the likelihood and number of CAE disclosures are

increasing in audit committee accounting expertise, consistent with more knowledgeable audit

committees constraining managers’ and auditors’ negative disclosure incentives. Overall, our

evidence suggests the discretion allowed in determining required disclosures introduces strategic

CAE disclosures involving management, auditors, and audit committees, the three primary

parties responsible for financial reporting.

Key Words: Critical accounting estimates; quantitative disclosure; management

incentives; auditor incentives; audit committee accounting expertise

1

I. Introduction

We study the determinants of firms’ decisions to provide quantitative sensitivity disclosures

about critical accounting estimates (CAE), “material” and “highly uncertain” accounts requiring

judgment (SEC 2002; SEC 2003).1 Accounting estimates comprise a large and growing

component of financial statements, making the dividing line between fact and conjecture largely

unknown to investors (Lev, Li, and Sougiannis 2010). Since capital market inefficiencies can

result if investors are led by estimates-based accounting information to misallocate resources, the

SEC mandates that firms provide quantitative CAE information when “quantitative information

is reasonably available and will provide material information for investors” (Lev et al. 2010,

SEC 2003, V.). Consistent with CAE disclosures informing investors about the reliability of

accounting estimates, CAE disclosures reduce the value relevance of reported accounting

numbers (Glendening 2014). Many firms do not provide quantitative CAE disclosures (Levine

and Smith 2011; Bauman and Shaw 2014), and the SEC remains concerned about the lack of

quantitative disclosure, frequently requesting enhancement to CAE disclosures (Cassell, Dreher,

and Myers 2013; Ernst and Young 2011).2

We suggest the discretion allowed in SEC guidance, combined with the subjective nature of

CAEs, provides firms considerable leeway in deciding whether to provide CAE disclosures.

Since discretion in mandatory reporting reduces the likelihood of voluntary disclosure and

introduces strategic disclosures, we extend prior research by demonstrating that CAE disclosure

decisions reflect strategic preferences of those responsible for financial reporting (Einhorn 2005).

1 Common accounting estimates seen in CAE disclosures include defined-benefit pension plans, sales returns,

inventory obsolescence, warranty reserves, and uncollectible accounts receivable. As an example of a CAE

disclosure, Nordstrom Inc. reports “a 10% change in our allowance for doubtful accounts would have affected net

earnings by $12 (million) for the fiscal year ended January 30, 2010” (NORDSTROM INC, 10-K, March 22, 2010). 2 Cassell et al. (2013) report critical accounting estimates appear in over 25 percent of SEC comment letters and

Ernst and Young (2011) list critical accounting estimates first in their analysis of current reporting issues.

2

The SEC addressed their initial alert concerning CAE to management, external auditors, and the

audit committee, consistent with the three-pronged corporate governance “mosaic” described in

prior literature (SEC 2001; Cohen, Krishnamoorthy, and Wright 2004). Though management has

ultimate responsibility for financial reporting, the auditor and the audit committee have

important oversight responsibilities (Cohen et al. 2004). Recognizing this shared oversight role,

the New York Stock Exchange (NYSE) and the Public Company Accounting Oversight Board

(PCAOB) require audit committee/auditor discussion of MD&A and CAE (PCAOB 2012;

Cohen, Gaynor, Krishnamoorthy, and Wright 2007). Thus, we examine how management,

external auditors, and the audit committee impact CAE disclosure decisions, after controlling for

other firm financial, CAE, and governance characteristics.

For management, we consider managers’ equity risk incentives because the sensitivity of

managers’ wealth to changes in risk (portfolio vega) unambiguously increases managers’

incentive to misreport (Armstrong, Larcker, Ormazabal, and Taylor 2013). Given that chief

financial officers (CFOs) report that accounting estimates provide the most common vehicle for

misreporting (Dichev, Graham, Harvey, and Rajgopal 2013), management may anticipate that

CAE disclosures reveal sources of accrual discretion. Consistent with this idea, managers in an

experimental market report less aggressive point estimates in a regime with mandated disclosure

of estimate ranges (Majors 2014). Extending this idea to the more discretionary CAE disclosure

regime, we first hypothesize a negative association between portfolio vega and the likelihood and

number of CAE disclosures.

For external auditors, we consider the level of opposition they expressed in comment letters

to the SEC regarding CAE disclosures. Even though auditors are only required to review the

MD&A for inconsistencies with the financial statements, CAE disclosures received substantial

3

pushback from the Big 4 audit firms. While the amount of opposition in the comment letters

varied by firm, some of the major concerns related to increased auditor involvement, higher

preparation costs, and the questionable usefulness of the CAE disclosures for investors. If

auditors influence managers’ disclosure decisions, our second hypothesis predicts a negative

association between the likelihood and number of CAE disclosures and the degree of auditor

opposition.3

For audit committees, we consider accounting expertise because CAEs are often complex

and require considerable accounting knowledge and expertise to understand and advocate for

disclosure. PCAOB inspections reveal even knowledgeable auditors have difficulty with

accounting estimates (PCAOB 2008). Since a lack of accounting expertise could constrain the

ability of the audit committee to question either the auditor or management, we hypothesize a

positive association between the likelihood and number of CAE disclosures and audit committee

accounting expertise.

Our sample includes 2,298 firm-years (317 distinct S&P 500 firms) spanning 2003-2010.4

We model the existence and number of CAE disclosures as a function of top-five management

portfolio vega, auditor opposition to CAE, and audit committee accounting expertise. To rule out

alternative explanations for CAE disclosures, we control for other equity incentives, other board

and audit committee attributes, other innate firm financial attributes, other attributes that capture

the extent of uncertainty in the accrual measurement process, litigation risk, analyst following,

and institutional ownership. We hand-collect CAE disclosure data from 10-K reports, with about

50 percent of firm-years providing CAE disclosure.

3 We use a measure from Li (2008), which captures the negative versus positive tone of the auditor’s comment letter

to the SEC on the issue of CAE disclosures. 4 Firms did not provide CAE disclosures before 2002. We sample large firms because FR-72 resulted from a

targeted review of large firms in 2001 (SEC 2003). If these firms responded by including quantitative sensitivity

analyses, it biases against our finding results.

4

We find that the likelihood of CAE disclosure is negatively associated with top-five

management portfolio vega and auditor opposition to CAE, and positively associated with audit

committee accounting expertise.5 These findings support our three hypotheses and they are also

economically significant, ranging from about a 7 to 14 percent change in the probability of CAE

disclosure as portfolio vega and auditor opposition change from the first to the third quartile or

the audit committee changes from having no accounting expertise to having accounting

expertise. We find similar results for the number of CAE disclosures. In addition, many of our

control variables are also statistically significant and in the predicted direction. Our results are

also robust to tests surrounding actual incidences of misreporting, the time trend of portfolio

vega, and deleting the year of CAE initiation.

The study makes several important contributions to the literature. First, we provide empirical

evidence on firms’ decisions to provide quantitative CAE disclosures. The SEC’s push for

increased CAE disclosure was meant to mitigate the “illusion of precision” in accounting reports,

with the idea that CAE disclosure would aid in the use and interpretation of estimates in financial

reports (SEC 2001; Glassman 2006). However, our evidence suggests the allowed discretion

under SEC guidance limits the effectiveness of MD&A regulation over CAE disclosure.

Second, we provide evidence about the influence of each of the important parties involved in

the financial reporting and disclosure processes - management, external auditors, and audit

committees - consistent with calls for further research into how these parties contribute to

MD&A quality (Cohen et al. 2007). Our findings suggest each of the financial reporting parties

5 In additional analyses, we find reductions in top-five management portfolio vega subsequent to a CAE disclosure,

consistent with firms reducing managers’ misreporting incentives and supporting our conjecture that management

could anticipate potential negative consequences of quantitative sensitivity disclosures and engage in strategic

disclosure.

5

provide input to the disclosure decision and suggest that auditors and audit committees may not

always collaboratively provide greater monitoring.

Third, we add to the existing literature on the association between management equity risk

incentives and financial statement misreporting. CAE disclosures provide an interesting and

powerful setting to test the association between managers’ equity risk incentives and disclosure

(via the incentive to misreport). Our findings suggest managers’ equity risk incentives influence

disclosure choices, especially when the disclosure decision is aligned with misreporting

incentives (Armstrong et al. 2013).

Finally, we provide evidence on the auditor’s role in MD&A disclosures. Auditors are

traditionally viewed as taking a limited role in MD&A disclosure that is confined to reviewing

the MD&A for material inconsistencies with the financial statements. However, the SEC’s call

for increased CAE disclosures increased the potential for auditor involvement. Auditors voiced

concerns about the preparation and interpretation of CAE, and our findings indicate that firms’

CAE decisions were partially conditioned on those concerns.

The remainder of this study is organized as follows. Section II describes the institutional

background pertaining to CAE disclosures and develops hypotheses. Section III describes the

sample selection process and the research design, Section IV presents empirical results and

Section V concludes the study.

II. Institutional Background and Hypotheses Development

Institutional Background

The SEC has long stressed the importance of MD&A for providing informative and

transparent disclosures that help readers understand companies (SEC 2003). Since the early

2000s the SEC has pursued increased disclosure about CAEs. The SEC defines CAE as

6

“material” and “highly uncertain” accounts requiring judgment, such as estimates of anticipated

sales returns, inventory obsolescence, warranty reserves, doubtful accounts, and pensions (SEC

2002; SEC 2003). Prior research finds that greater accrual estimation is associated with lower

accruals quality and persistence, suggesting the SEC’s desire for greater CAE transparency could

indeed help readers better interpret accruals (Chen and Li 2013).6

In 2001, the SEC issued cautionary advice FR-60 to remind registrants that, under existing

MD&A disclosure rules, MD&A should explain the firm’s critical accounting policies (CAP) in

plain English as a supplement to, and not merely duplicating, required footnote disclosure. FR-60

noted the primary purpose of CAP disclosures was to provide transparency surrounding “the

judgments and uncertainties affecting the application of those policies, and the likelihood that

materially different amounts would be reported under different conditions or using different

assumptions” (SEC 2001). While CAP disclosures moved in the direction of increasing

transparency, CAP disclosures are less precise than CAE disclosures because they do not

quantify the degree of uncertainty in the estimates.

Disclosures provided by Target and Walmart in their 2009 10-Ks, shown in Appendix A,

illustrate the difference in precision between CAP and CAE disclosures. Both firms provide a

CAP disclosure relating to their self-insurance liability, suggesting the estimate involves material

and highly uncertain judgments at each firm. However, Walmart, but not Target, also provides a

CAE disclosure quantifying the uncertainty more precisely. Here, Walmart’s CAE disclosure

stipulates that a reasonably likely change (defined as one percent increase or decrease) would

change its self-insurance accrued liability (and accrual earnings) by $26 million.

6 Accrual estimation is measured by the number of sentences in the notes to the financial statements and the CAP

section of the MD&A that portray the use of estimates, such as “we estimated receivables and purchased inventory”

(Chen and Li 2013).

7

Perhaps because of the lack of precision in CAP disclosures, the SEC remained concerned

about inadequate CAE disclosure and introduced Proposed Rule 33-8098 in May 2002. The

proposed rule specifically calls on firms to provide a quantitative sensitivity analysis, CAE

disclosure (SEC 2002). Though Proposed Rule 33-098 was never adopted, the SEC issued

interpretive guidance FR-72 in December 2003, again emphasizing CAE disclosures:

“Since critical accounting estimates and assumptions are based on matters that are

highly uncertain, a company should analyze their specific sensitivity to change, based

on other outcomes that are reasonably likely to occur and would have a material

effect. Companies should provide quantitative as well as qualitative disclosure when

quantitative information is reasonably available and will provide material information

for investors.” (SEC 2003, V.)

Even though FR-72 is mandatory, firms still have considerable discretion in implementation,

and the SEC remains concerned about a lack of quantitative disclosures, as indicated by the

common issuance of comment letters on the subject (Holtzman 2007; Cassell et al. 2013).

Consistent with SEC concerns, prior research finds considerable lack of quantitative disclosure.

In a review of 5,984 10-K filings before February 2005, Levine and Smith (2011) note that only

14 percent of the sample even mention sensitivity analysis and very few of those actually

quantify the uncertainty.7 In a small sample of firms with material defined-benefit pension plans,

Bauman and Shaw (2014) find 40 percent of their sample firms do not disclose pension-related

CAEs. We extend prior research by examining the determinants of firms’ CAE disclosure

decisions, focusing on the roles of management, auditors and audit committees in CAE

disclosure.

Hypotheses Development

Management Equity-Risk Incentives and CAE Disclosures

7 In contrast, Levine and Smith (2011) find that 80 percent of their sample firms provide a qualitative discussion, of,

on average, six to seven CAPs.

8

We begin by considering manager equity-risk incentives to strategically make CAE

disclosures. Management holds the primary responsibility for preparing and disseminating

financial information. Since CAE disclosures provide greater transparency about the inherent

uncertainty of firms’ accrual accounting estimates, the disclosures not only can aid users in

interpreting accruals, but also can increase users’ insights into the extent of managers’ discretion

to change earnings during the accrual estimation process. Given that estimates provide the most

common vehicle for manipulating earnings, managers have incentives to not reveal the extent of

their discretion through CAE disclosures (Dichev et al. 2013).

Though prior literature uses a number of different proxies for equity incentives, including

equity compensation, in-the-money options, or the sensitivity of wealth to stock price changes

(portfolio delta), results are mixed.8 Armstrong et al. (2013) suggest the use of portfolio vega

instead of the other proxies because the other proxies provide more ambiguous incentives which

may drive the mixed results. For example, Armstrong et al. (2013) argue that while larger

portfolio delta increases the incentive to misreport because misreporting inflates equity values,

portfolio delta also decreases the incentive to misreport because misreporting increases stock

price risk. The opposing forces on the incentive to misreport makes the theoretical effect of delta

on the incentive to misreport ambiguous. Armstrong et al. (2013) argue that, unlike portfolio

delta, portfolio vega, measuring the increase in the value of managers’ portfolio for an increase

in firm risk, unambiguously increases the incentive to misreport. Managers with greater portfolio

vega are incentivized to inflate equity values through misreporting, but also benefit from the

higher equity risk that results from misreporting. Armstrong et al. (2013) provide robust

8 Studies documenting a positive relation include Cheng and Warfield (2005), Bergstresser and Philippon (2006),

Burns and Kedia (2006), Efendi, Srivastava, and Swanson (2007), Cheng and Farber (2008), Cornett, Marcus, and

Tehranian (2008). Studies finding no relation include Erickson, Hanlon, Maydew (2006), Larcker, Richardson, and

Tuna (2007), Hribar and Nichols (2007), Armstrong, Jagolinzer, and Larcker (2010).

9

evidence of a positive relation between portfolio vega and financial misreporting (using

discretionary accruals, restatements, and SEC enforcement actions). Furthermore, portfolio vega

subsumes the effect of delta on misreporting.9 Given that portfolio vega unambiguously

increases managers’ incentive to misreport, we predict a negative association between the

likelihood and number of CAE disclosures and portfolio vega, formally stated as follows:

H1: The likelihood and number of CAE disclosures are negatively associated with

portfolio vega.

Auditor Opposition and CAE Disclosures

Auditors’ responsibility for MD&A disclosures is arguably limited to reviewing disclosures

for consistency with the financial statements and for material misstatements or omissions that

could render MD&A misleading (AICPA 1975, Statement on Auditing Standards [SAS] No. 8).

In spite of this limited role, the SEC advises auditors to bring particular focus to CAE, including

disclosure (SEC 2001). Further, the PCAOB cautions auditors about consistent problems with

estimates in the audit found during the inspection process (PCAOB 2008).

Auditors’ comment letters to the SEC about increasing CAE disclosure reveal a variety of

negative concerns. The major negative concerns voiced by auditors involved their own legal

exposure and preparation costs, whether they would be forced to become more involved in

MD&A disclosures, the practicality of the disclosures for registrants, and the usefulness of CAE

disclosures to investors. Psychology theory and research suggest affective responses, positive or

negative feelings toward a stimulus, occur automatically and impact decision-making (Slovic,

Finucane, Peters, and MacGregor 2007). Specific to auditors, research finds both negative mood

in general and negative interpersonal affect towards the client impacts auditors’ judgments about

9 Since equity compensation and options provide both delta and vega neither provide unambiguous incentives to

take risk, and hence, vega continues to be significant even after controlling for these alternative measures of equity

incentives (Armstrong et al. 2013).

10

inventory obsolescence estimates (Bhattacharjee, Moreno, and Riley 2012; Chung, Cohen, and

Monroe 2008). In a similar manner, we suggest auditor negative affect towards CAE disclosure

regulation could negatively influence client’s CAE disclosure decisions.

Auditors work in close proximity to management throughout the year and audit

methodologies rely extensively on auditor/client interactions (Knechel 2007; Hellman 2011).

Thus, auditors have many opportunities to make their views known to management. Further,

audit partners report a desire to be in the role of expert advisor, and seek to provide advice to

management (McCracken, Salterio, and Gibbons 2008). Therefore, we expect that auditors who

expressed more opposition towards CAE disclosures are less likely to consider lack of CAE

disclosure as materially misleading to investors and are less likely to advise clients to provide

CAE disclosure, formally stated as follows:

H2: The likelihood and number of CAE disclosures are negatively associated with

auditor opposition to CAE disclosures.

Audit Committee Accounting Expertise and CAE Disclosures

The SEC cautions audit committees to carefully review and proactively discuss CAE and

CAE disclosures with both management and auditors (SEC 2001). The NYSE requires the audit

committee to discuss MD&A with the auditor (Cohen et al. 2007). In addition, the PCAOB

requires the auditor to communicate information about CAEs with the audit committee (PCAOB

2012, AS NO. 16). Thus, CAE disclosure decisions should appear on audit committee agendas.

We expect audit committees advocate for complying with the spirit of CAE disclosures required

by FR-72 because of committee members’ incentives to avoid legal liability, protect shareholder

interests, and protect their own reputation capital (Fama 1980; Gilson 1990; Sahlman 1990).

However, as indicated in PCAOB inspection reports, CAEs are often complex, and by

definition highly uncertain. Though audit committee incentives provide support for greater CAE

11

disclosure, committee members must have accounting-specific expertise to actually achieve

greater CAE disclosure. Grasping the nature of estimation uncertainty that underlies accrual-

based accounting, and proactively advocating for CAE disclosure, requires solid understanding

of the GAAP-based accounting principles and firm-based accounting policies that drive

accounting estimates. Prior research highlights the importance of audit committee accounting

expertise for monitoring accounting estimates, as evidenced by the association between

discretionary accruals and accounting expertise (Cohen, Hoitash, Krishnamoorthy, and Wright

2014; Dhaliwal, Naiker, and Navissi 2010). Thus, we predict audit committees with accounting

expertise more likely advocate for CAE disclosures as they discuss the disclosures with

management and the auditor, formally stated as follows:

H3: The likelihood and number of CAE disclosure are positively associated with audit

committee accounting expertise.

III. Sample Selection and Research Design

Sample

Table 1, Panel A describes our sample selection. We begin with the 460 firms in the S&P

500 that appear in ExecuComp in 2004. We use 2004 because it was the first year after the

SEC’s guidance on CAE disclosures became effective on December 29, 2003. Concentrating on

the S&P 500 makes the CAE data hand-collection process more manageable while also

providing for a comprehensive set of the largest firms of interest to the SEC. We exclude firms in

the financial services industry (four-digit SIC code: 6000-6999), utilities industry (four-digit SIC

code: 4900-4949), and non-classifiable firms (four-digit SIC code: 9900-9999) to allow for more

commonality in both firms’ accrual estimates and the information contained in firms’ CAE

12

disclosures. We exclude acquired firms to provide a stable sample for data collection purposes.10

We obtain audit committee and other governance data from Morningstar’s Executive

Compensation Database provided by Audit Analytics and exclude observations missing data. We

also exclude firms with a non-Big 4 auditor. Finally, we remove observations without necessary

Execucomp, Compustat, and I/B/E/S data to calculate the additional test and control variables

used in our analyses. The final sample consists of 2,298 firm-years (317 distinct firms) from

2003 to 2010.

We hand-collect CAE disclosure data for these firms. We first examine the 10-Ks for each

sample firm in 2004 (i.e. the year following the SEC’s guidance) and 2010 (i.e. the final year in

our data collection sample period) to identify which firms provide a CAE disclosure in either

year. If neither the 2004 10-K nor the 2010 10-K have a CAE disclosure, we assume the firm did

not provide a CAE disclosure in any year of our sample period.11

If a firm provides a CAE

disclosure in either 2004 or 2010, then we examine the firm’s 10-Ks from 2003 to 2010 and

identify for each year (1) the presence of a CAE disclosure and (2) the number of CAEs

disclosed.12

Panel B of Table 1 reports the average CAE disclosure rate for our sample during

2003-2010 is 50.96 percent; that is, about half of the firm-year observations disclose CAE and

half do not. Panel B also reports the annual CAE disclosure frequency. CAE disclosure rates

monotonically increase from 32.19 percent in 2003 to 58.14 percent in 2010. The largest annual

increases in disclosure rates occur between 2003, after the SEC issued its proposed rule, and

2004, after the SEC issued FR-72. After 2004, CAE disclosure rates continue to increase, but at a

10

When collecting CAE data, we first check for disclosures in the boundary years of the sample, thus we require

firms to exist during the entire sample period. 11

136 sample firms do not provide a CAE disclosure in either 2004 or 2010. To assess the validity of the assumption

that the 952 unexamined firm-years have no CAE disclosure, we randomly select 20 firm-years out of the 952 firm-

years. After examining the 10-Ks for these 20 firm-years, we find none of the 20 firm-years have a CAE disclosure. 12

Appendix B provides examples of CAE disclosures.

13

slower rate. Panel C of Table 1 reports information about the number of CAEs disclosed by

disclosing firms during 2003-2010. On average, CAE-disclosing firms disclose between two and

three critical accounting estimates, and the number of CAEs disclosed remains steady throughout

the sample period.

Research Design

Empirical Model and Test Variables

To test our hypotheses, we estimate the following logistic regression:

Pr(CAEit) = α0 + α1VEGAit + α2AUDITOROPPOSITIONit + α3ACCT_EXPERTit +

α4DELTAit + α5EQUITYPAYit + α6LEGAL_EXPERTit + α7AC_SIZEit +

α8AC_TENUREit + α9AC_MEETit + α10BD_SIZEit + α11BD_INDit +

α12DUALit + α13ROAit + α14BTMit + α15SIZEit + α16LEVERAGEit +

α17ACCRUALSit + α18ICWit + α19SALESVOLit + α20OPERVOLit +

α21LOSSit + α22COMPLEXITYit + α23LITIGATIONit + α24COVERAGEit

+ α25INSTOWNit + α26PENSIONit + εit

(1)

The dependent variable, CAE, is an indicator variable equal to 1 for firm-years providing a CAE

disclosure, and zero otherwise.13

H1 predicts that the likelihood of a CAE disclosure is

negatively associated with managers’ portfolio vega (1 < 0). To test H1, we include VEGA,

which equals natural logarithm of one plus the average dollar change (in $000s) in the top-five

executives' wealth associated with a 1% change in the standard deviation of the firm’s returns

(see Core and Guay [2002] and Coles, Daniel, and Naveen [2006[).14

To test H2, we include AUDITOROPPOSITION, which equals the negative versus positive

tone of the auditor's comment letter to the SEC on the issue of CAE disclosures. We follow Li

(2008) when measuring negative versus positive tone using Linguistic Inquiry and Word Count

13

Appendix C provides definitions of all variables used in this study. 14

We obtain data on VEGA and DELTA from the following website: http://astro.temple.edu/~lnaveen/data.html. We

thank Jeffrey Coles, Naveen Daniel, and Lalitha Naveen for making their data available.

14

(LIWC). Appendix D provides a detailed explanation of our auditor opposition measure.

Consistent with H2, we expect a negative coefficient on AUDITOROPPOSITION (2 < 0).

H3 predicts that the likelihood of a CAE disclosure is positively associated with audit

committee accounting expertise. To test H3, we include ACCT_EXPERT, an indicator variable

equal to 1 for firm-years where the audit committee includes at least one accounting expert.15

Consistent with H3, we expect a positive coefficient on ACCT_EXPERT (3 > 0). Although the

SEC requirements for audit committee financial expertise includes supervisory or finance

expertise in addition to accounting expertise, we consider only accounting expertise due to the

complex nature of accounting estimates and prior research support for accounting expertise over

the SEC’s broader definition (e.g., Krishnan and Visvanathan 2008). ACCT_EXPERT includes

audit committee members with experience as public accountants, chief financial officers,

comptrollers, or other principal accounting officers.

Control Variables

When estimating Equation (1), we first include controls for other equity-based incentives that

are potentially correlated with VEGA and also may influence the incentive to misreport, and thus

the propensity to disclose CAEs. Specifically, we control for DELTA and EQUITYPAY. DELTA

is the natural logarithm of one plus the average dollar change (in $000s) in top-five executives'

wealth associated with a 1% change in the firm’s stock price. EQUITYPAY is the natural

logarithm of one plus the average dollar value (in $000s) of annual equity-based compensation to

the top-five executives, where equity-based compensation equals the fair value of option grants

plus the fair value of restricted stock awards. DELTA and EQUITYPAY are expected to decrease

the likelihood of a CAE disclosure if DELTA and EQUITYPAY increase the incentive to

15

In untabulated sensitivity tests, we define ACCT_EXPERT as the proportion of accounting experts on the audit

committee and find qualitatively similar results to those reported in Tables 4 and 6.

15

misreport. However, if managers with higher values of DELTA and EQUITYPAY are more

averse to the equity risk resulting from misreporting (Armstrong et al. 2013), then the likelihood

of a CAE disclosure is expected to be positively related to DELTA and EQUITYPAY. Hence, we

do not predict the sign of the coefficients on DELTA and EQUITYPAY.

Next, we include controls for other governance characteristics because prior research

indicates higher quality disclosures result from better corporate governance mechanisms (Eng

and Mak 2003; Laksmana 2008; Beyer, Cohen, Lys, and Walther 2010). Though we expect audit

committee accounting expertise is the most important governance characteristic related to CAE

disclosure decisions, we also control for other board and audit committee characteristics that

might support more transparent disclosure practices. We include BD_SIZE (AC_SIZE) to control

for the resources, or size, of the board (audit committee), where BD_SIZE (AC_SIZE) equals the

number of board of director (audit committee) members. We include LEGAL_EXPERT to

control for the legal competency of the audit committee, where LEGAL_EXPERT equals one for

firm-years where the audit committee includes at least one legal expert, zero otherwise. To

control for the experience and effort of the audit committee, we include variables for the number

of years the audit committee members have served as directors (AC_TENURE) and the number

of meetings held by the audit committee (AC_MEET). We control for the independence of the

board by including BD_IND, which equals the percent of outside board members. We control for

the CEO’s power over the board by including DUAL, which is an indicator variable equal to 1

for firm-years with a CEO that also serves as the chairperson of the board of directors, zero

otherwise. If stronger corporate governance leads to more transparent disclosure practices, we

predict positive coefficients on LEGAL_EXPERT, AC_SIZE, AC_TENURE, AC_MEET,

BD_SIZE, and BD_IND and a negative coefficient on DUAL.

16

Equation (1) also includes controls for innate firm attributes that might influence the CAE

disclosure decision. We control for return on assets (ROA), the book-to-market ratio (BTM), the

natural logarithm of total assets (SIZE), and the ratio of long-term debt to total assets

(LEVERAGE). Based on Lang and Lundholm (1993), we predict positive coefficients on ROA

and SIZE. Following Guay (2008), we predict a positive coefficient on LEVERAGE. We do not

form a prediction on BTM.

In addition to the above firm traits, we include controls to capture the extent of uncertainty in

the accrual measurement process. Since CAE disclosures capture uncertainty in accrual

estimates, and uncertain accrual estimates are more likely in firms with greater levels of accruals,

we include ACCRUALS as a control variable. ACCRUALS is measured as earnings before

extraordinary items less operating cash flows taken directly from the statement of cash flows,

scaled by total assets. We include ICW to control for the possibility that CAE disclosures are

related to internal control weaknesses, but we do not make prediction on ICW. ICW equals 1 for

firm-years with a material weakness under SOX sections 302 or 404 and zero otherwise. Because

we expect firms operating in volatile business environments to experience more uncertainty

when making accrual estimates, we control for SALESVOL and OPERVOL. SALESVOL is the

standard deviation of sales over the previous three years. OPERVOL is the standard deviation of

operating cash flows over the previous three years. We include LOSS to control for the relation

between CAE disclosures and negative earnings, without a predicted sign. CAE disclosures may

be more likely during loss years if losses are recognized with less precision or CAE disclosures

may be less likely during loss years if firms with losses are more financially weak and less

willing to disclose a CAE. We control for operating complexity (COMPLEXITY) because

managers of complex businesses likely face more uncertainty when making accrual estimates.

17

COMPLEXITY is defined as the natural logarithm of a firm's total geographic and business

segments.

Previous studies investigating the link between ex ante litigation risk and voluntary

disclosure provide evidence that managers are less likely to make voluntary disclosures when the

level of ex ante litigation risk is high (Johnson, Kasznik, and Nelson 2001; Baginski, Hassell,

and Kimbrough 2002; Rogers and Van Buskirk 2009). Other research finds that ex ante litigation

risk increases the likelihood of providing CAP disclosures (Levine and Smith 2011). Since CAE

disclosures are stated as reasonably likely changes within the current financial report, they may

not qualify for the same safe harbor protections granted by the 1995 Private Securities Litigation

Reform Act (PSLRA) to voluntary disclosures of forward-looking statements, such as earnings

guidance. Considering that investors might interpret the disclosures as managers’

acknowledgment of inaccuracies in currently reported accounting numbers, managers may

believe that CAE disclosures increase the risk of litigation. Insofar as managers are able to

exercise their discretion when providing CAE disclosures, managers of high litigation risk firms

may exhibit greater aversion to disclosing a quantitative sensitivity analysis of accounting

estimates. Following, we include LITIGATION, which equals the probability of litigation

estimated using the coefficients from the litigation risk model from Kim and Skinner (2012).16

We also control for other potential channels of information in the market, including analyst

following and institutional ownership (Einhorn 2005). COVERAGE is defined as the natural

logarithm of the number of analysts issuing a forecast for the firm prior to the fourth quarter

earnings announcement for each firm-year.17

INSTOWN equals percentage ownership by

16

See Table 7, Model (2) of Kim and Skinner (2012). 17

Analyst data are obtained from the IBES Summary database.

18

institutional investors at the fourth quarter of each firm-year.18

On one hand, we expect a positive

association between the likelihood of a CAE disclosure and analyst following and institutional

ownership. This is consistent with previous studies that document a positive relation between

disclosure levels and analyst following (Lang and Lundholm 1996) and institutional ownership

(Healy, Hutton, and Palepu 1999). However, COVERAGE and INSTOWN may negatively impact

the likelihood of a CAE disclosure. The importance of meeting analysts’ earnings expectations

increases with analyst coverage (Graham, Harvey, and Rajgopal 2005), thus the incentive to

misreport and the incentive not to disclose CAEs may also increase with analyst coverage. Also,

institutional investors are considered to be sophisticated users of financial statements (e.g., Hand

1990; Walther 1997; Collins, Gong, and Hribar 2003), suggesting the demand for information

conveyed in CAE disclosures is lower in firms with greater institutional holdings. Thus, we do

not make predictions for COVERAGE and INSTOWN.

Since the SEC explicitly mentioned that critical accounting estimates potentially include

pension-related estimates, we expect firms with defined-benefit pension plans to exhibit a higher

CAE disclosure rate. Thus, we include PENSION, an indicator variable equal to one for firm-

years with a defined-benefit pension liability (projected benefit obligation), and zero otherwise.

Finally, because CAE disclosure rates are expected to vary by year and industry the model

includes (untabulated) year and industry (at the 2-digit industry level) fixed effects.

IV. Results

18

Institutional ownership data are from Thomson Financial’s CDA/Spectrum database, which contains institutions’

quarterly shareholding data based on their 13-F filings to the SEC.

19

Descriptive Statistics

Panel A of Table 2 reports the descriptive statistics for the variables used in estimating

Equation 1. The mean (median) of VEGA is 4.403 (4.597). AUDITOROPPOSITION has a mean

of -0.256, indicating auditors’ comment letters on CAE disclosures were more positive than

negative, on average. Panel A also reports that 65.1 percent of firm-years have an audit

committee with an accounting expert.

For the equity incentive control variables, Panel A reports the mean (median) of DELTA and

EQUITYPAY is 5.591 (5.543) and 6.107 (6.903), respectively. For the controls for other board

attributes, 47.2 percent of firm-years had a CEO that also chaired the board of directors and the

percent of independent members on the board averaged 88 percent. Sample firms’ boards (audit

committees) include about ten (four) members. Panel A also reports descriptive for other firm

characteristics. Sample firms are profitable on average and exhibit a mean leverage ratio of 21.4

percent. About 68 percent of sample firms have a defined-benefit pension plan.

Panel B of Table 2 reports the differences in the mean and median values of variables

between disclosing firm-years (1,171 observations) and non-disclosing firm-years (1,127

observations). Inconsistent with H1, the mean (median) value of VEGA for disclosing firm-years

is insignificantly (significantly) greater compared to non-disclosing firm-years. In support of H2,

the CAE disclosing sub-sample has statistically lower mean and median values for

AUDITOROPPOSTION. Consistent with H3, the mean and median values for ACCT_EXPERT

are statistically greater for firm-years providing a CAE disclosure. Panel B also reveals that

firms that do and do not disclose CAE differ significantly on many of the dimensions controlled

for in our multivariate analyses.

20

Table 3 reports Pearson (below the diagonal) and Spearman (above the diagonal)

correlations. Similar to the results in Panel B of Table 2, we find that CAE exhibits a significant

positive Spearman correlation with VEGA (p-value = 0.046), but the Pearson correlation between

CAE and VEGA is statistically insignificant (p-value = 0.487). Table 3 reports significantly

negative Spearman and Pearson correlations between CAE and AUDITOROPPOSTION (p-

value<0.0001). We also find CAE exhibits significant positive correlations with ACCT_EXPERT

(p-value<0.001). While Panel B of Table 2 and Table 3 provide some univariate support for H2

and H3, we rely on our multivariate analysis for making inferences.

Determinants of CAE Disclosures

Table 4 reports the results from estimating Equation (1). Consistent with H1, the coefficient

on VEGA is negative and significant (coefficient = -0.198, p-value = 0.045), suggesting that CAE

disclosures are less likely when the incentive to misreport is high, perhaps due to the increased

transparency CAE disclosures provide about the amount of discretion in accounting estimates.

We also find a significantly negative coefficient on AUDITOROPPOSITION (coefficient = -

1.160, p-value = 0.015), consistent with H2 and suggesting that auditors’ opposition to CAE

disclosures reduces the likelihood their clients disclose CAE in the MD&A, an area of financial

reporting typically viewed as having a limited auditor role.

In support of H3, Table 4 reports a statistically positive coefficient on ACCT_EXPERT

(coefficient = 0.553, p-value = 0.012), suggesting audit committee accounting expertise increases

the likelihood of a CAE disclosure, consistent with more knowledgeable audit committees

providing constraint on managers’ and auditors’ negative disclosure incentives. We also find

audit committee size, number of audit committee meetings, and board size provide significant

incremental support for CAE disclosure.

21

For our other control variables, we find several firm attributes influence the propensity to

disclose a CAE. Table 4 reports the likelihood of a CAE disclosure is increasing in firm size,

leverage, and whether the firm has a defined-benefit pension plan and decreasing in the book-to-

market ratio and analyst coverage.

Overall, the results in Table 4 are consistent with our three hypotheses, and suggest

management equity risk incentives, auditor opposition to CAE disclosures, and audit committee

accounting expertise play determinative roles in whether a firm provides a CAE disclosure. To

get a sense of the economic effects of our results, Table 5 presents the change in probability of a

CAE disclosure for selected changes in our independent variables. For each variable, Table 5

reports the change in the probability of CAE disclosure resulting from moving non-dichotomous

variables from the first to the third quartile and dichotomous variables from zero to one, holding

other variables at their mean values.19

Moving from the first to the third quartile of VEGA and

AUDITOROPPOSITION decreases the probability of a CAE disclosure by 6.77% and 13.84%,

respectively. Moving from zero to one for ACCT_EXPERT increases the probability of a CAE

disclosure by 13.73%. Considering that the mean CAE disclosure rate is about 51%, these effects

are economically significant.

Number of CAE Disclosures

While the above analysis examines the likelihood of a CAE disclosure, we also examine the

determinants of the number of CAEs disclosed. We estimate the following ordered logistic

regression to test our predictions regarding the number of CAEs disclosed:

19

We calculate the change in the probability using the following expression: eβ’X

/ (1 + eβ’X

), where β refers to the

vector of coefficients from the model in Table 4 and X refers to the vector of independent variables.

22

Pr(CAE#it) = α0 + α1VEGAit + α2AUDITOROPPOSITIONit + α3ACCT_EXPERTit +

α4DELTAit + α5EQUITYPAYit + α6LEGAL_EXPERTit + α7AC_SIZEit +

α8AC_TENUREit + α9AC_MEETit + α10BD_SIZEit + α11BD_INDit +

α12DUALit + α13ROAit + α14BTMit + α15SIZEit + α16LEVERAGEit +

α17ACCRUALSit + α18ICWit + α19SALESVOLit + α20OPERVOLit +

α21LOSSit + α22COMPLEXITYit + α23LITIGATIONit + α24COVERAGEit

+ α25INSTOWNit + α26PENSIONit + εit

(2)

In Equation (2), we replace the binary dependent variable in Equation 1 (CAE) with an ordinal

variable (CAE#) that equals the number of CAE disclosures provided. CAE# ranges from 0 to 10.

Our purpose in estimating Equation (2) is to examine whether the hypothesized determinants of

the existence of a CAE disclosure are also significant determinants of the breadth of CAE

disclosures.

Table 6 reports the results from estimating Equation 2, providing additional support for our

hypotheses. Consistent with H1, H2, and H3, we find a significantly negative coefficient on

VEGA (coefficient = -0.160, p-value = 0.034), a significantly negative coefficient on

AUDITOROPPOSITION (coefficient = -0.815, p-value = 0.037), and a significantly positive

coefficient on ACCT_EXPERT (coefficient = 0.528, p-value = 0.006), respectively. These results

suggest the number of CAEs disclosed is negatively related to executives’ equity risk incentives

and auditor opposition and positively related to audit committee accounting expertise.

With a few exceptions, the results for our control variables are similar to the results reported

in Table 4. One of the differences in the results for our control variables is that the coefficient on

COVERAGE is no longer significant when examining the number of CAEs disclosed. We also

find that the number of CAEs disclosed is negatively associated with litigation risk, whereas this

variable is not a significant determinant in Table 4.

Overall, the findings reported in Tables 4-6 indicate that management equity-risk incentives,

auditor incentives, and audit committee accounting expertise influence both the likelihood of

23

disclosing a CAE and the number of CAEs disclosed. Thus, our evidence is consistent with

discretion in CAE disclosures introducing strategic incentives among the parties involved in the

three-pronged governance “mosaic,” management, auditors, and audit committees. The results

are robust to a large number of controls for other aspects of corporate governance and firm-

characteristics.

Firm Responses to CAE Disclosures

Our tests of H1 indicate that equity risk incentives, by increasing the incentive to misreport,

discourage managers from disclosing CAEs. Managers with greater misreporting incentives

appear to be less willing to reveal information about the extent of discretion in the accrual

estimation process, and hence, their ability to misreport. However, some managers do indeed

disclose CAEs, and it remains unclear as to how internal monitors (i.e., boards) react to the more

transparent information conveyed in CAE disclosures. Given that CAE disclosures potentially

provide new information about the ability to misreport, it is plausible that boards take actions to

reduce managers’ misreporting incentives.

Though the nature of a CAE does not change with its disclosure, new quantitative disclosures

may require boards to think more carefully about the extent to which accounting reports rely on

estimates that are inherently uncertain. The potential for CAE disclosures to provide new

information is evidenced in the 2011 business roundtable discussion hosted by the SEC that

focused on measurement uncertainty in financial reporting. When speaking about the topic of

CAE disclosures, one director stated (SEC 2011):

“And all of a sudden the most important numbers on the balance sheet were

estimates. Except for the cash and investments that we had, the most numbers [sic]

that were on the balance sheet were estimates. And all of a sudden I started paying

attention to those estimates in a very different way than I did before.”

24

The quote illustrates that CAE disclosures potentially aid boards in understanding the scope

and magnitude of accrual measurement uncertainty. Moreover, since misreporting commonly

occurs through accounting estimates, the disclosures may assist in assessing the extent to which

managers are able to use discretion in accounting estimates as an earnings management device.

To the extent CAE disclosures inform boards about managers’ ability to misreport, the expected

cost of managers’ equity risk incentives (i.e., portfolio vega) increases after the disclosure of a

CAE because managers’ equity risk incentives increase managers’ incentive to misreport

(Armstrong et al. 2013). Following, we expect firms to respond to the disclosure of a CAE by

decreasing managers’ portfolio vega, thereby reducing the incentive to misreport. To test how

portfolio vega responds to the disclosure of a CAE, we estimate the following model:

VEGAit = β0 + β1CAEit-1 + β2DELTAit + β3CASHCOMPit + β4SALEit + β5BTMit +

β6LEVERAGEit + β7R&Dit + β8CAPEXit + β9RETURNVOLit +

β10POSTSOXit + β11POST123(R)it + βFirm + εit

(3)

The dependent variable, VEGA, remains as previously defined. To assess whether disclosing

firms reduce managers’ portfolio vega after the disclosure of a CAE, Equation 3 includes the

lagged version of our disclosure indicator variable, CAE. We predict a negative coefficient on

CAE. To examine how CAE disclosures influence within-firm variation of portfolio vega, we

include firm fixed-effects. By using a firm fixed-effects approach, the coefficient on CAE

conveys the average change in managers’ portfolio vega between disclosing firms’ pre-CAE and

post-CAE periods. Because we are interested in the VEGA responses of disclosing firms, the

sample used to estimate Equation 3 includes only those firms that provide at least one CAE

disclosure. The sample includes years 1995-2010 for these firms to allow for an adequate

comparison of vega practices from before to after the disclosure of a CAE. This process results

in a sample of 2,838 firm-years when estimating Equation 3.

25

Equation 3 includes controls for other known determinants of vega documented in previous

research (e.g., Guay 1999; Cohen, Hall, and Viceira 2000; Rajgopal and Shevlin 2002; Coles et

al. 2006).20

DELTA, BTM, and LEVERAGE remain as previously defined. CASHCOMP is the

natural logarithm of one plus the average cash compensation, salary plus bonus, (in $000s) for

the top-five executives. SALE is the natural logarithm of annual sales revenue. R&D is the ratio

of research and development expenditures to total assets, where firm-years with missing values

for research and development expenditures have R&D set to zero. CAPEX is the ratio of capital

expenditures to total assets. RETURNVOL equals the standard deviation of monthly returns over

the fiscal year. We also include POSTSOX and POST123(R) to control for the influence of the

Sarbanes-Oxley Act (SOX) of 2002 and SFAS No. 123(R) on equity risk incentives.21

We

measure the POSTSOX period as those firm-years ending in 2002 and afterward and the

POST123(R) period as those firm-years ending after June 15, 2006.

Table 7 reports the results from estimating Equation 3. Consistent with our prediction, the

coefficient on CAE in the first and second model specifications equals -0.147 (p-value = 0.026)

and -0.175 (p-value = 0.003), respectively. This result suggests firms reduced vega after the

disclosure of a CAE. In terms of an economic effect, we estimate that firms reduced equity risk

incentives by approximately 4 percent relative to the mean value of VEGA for the sample used in

Table 7. The evidence reported in Table 7 is consistent with the disclosure of a CAE reducing the

appeal of equity risk incentives as a form of compensation. As CAE disclosures potentially

20

The control variables used in Equation 3 are based on the vega model in Table 3 of Coles et al. (2006). 21

Because SOX may have increased the skepticism of financial reporting (Cohen, Dey, and Lys 2008; Zhang 2012),

it is possible that vega decreased after SOX to mitigate the incentive to misreport. However, SOX also constrained

risk-taking behavior (Bargeron, Lehn, and Zutter 2010), so vega may increase after SOX in order to encourage

managers to undertake new investment projects in the post-SOX period. As a result, we do not make a prediction for

the coefficient on POSTSOX. Hayes, Lemmon, and Qiu (2012) show firms reduced vega after the adoption of SFAS

No. 123(R) due to the increased accounting costs associated with stock option compensation. Following, we predict

a negative coefficient on POST123(R).

26

inform boards about managers’ ability to misreport, compensation practices respond to the

disclosure of a CAE in a manner that reduces the incentive to misreport.

For our control variables, as shown in Table 7, we find mangers’ portfolio vega is positively

associated with managers’ portfolio delta, firm size, book-to-market, leverage, R&D, and return

volatility. We note that the positive coefficient on BTM in Table 7 is inconsistent with our

prediction and potentially the result of our sample selection.22

We also find SOX (SFAS No.

123(R)) exhibits a positive (negative) influence on vega.

Additional Analyses

Incidence of Alleged GAAP Violations

CAE disclosures provide information about measurement discretion in accrual estimates.

Since managers can use this estimate flexibility to engage in aggressive financial reporting, H1

predicts that managers will have the incentive to not make CAE disclosures and, thus, reveal

their discretion, when the incentive to misreport is high. Consistent with this prediction, we find

CAE disclosures are decreasing with managers’ portfolio vega. The discretion highlighted by

CAE disclosures could be used to engage in all types of aggressive reporting (i.e., within-GAAP,

gray-GAAP, or non-GAAP). Vega captures only the incentive to engage in aggressive financial

reporting, not the actual incidence of misreporting. However, to ensure that vega does not merely

capture the incidence of non-GAAP earnings management, we re-estimate Equation (1) after

including an indicator variable, AAER, which captures whether the firm-year is associated with

22

The positive association between VEGA and BTM could also reflect the effect of growth expectations on

managers’ misreporting incentives. Managers have a reduced incentive to misreport as the book-to-market ratio

increases because there is less pressure to satisfy capital market expectations and sustain high valuations (Skinner

and Sloan 2002; Dechow, Ge, Larson, Sloan 2011). Since vega unambiguously increases the incentive to misreport

(Armstrong et al. 2013), firms may be more willing to increase managers’ portfolio vega when other misreporting

incentives (e.g. investor optimism) are low.

27

an alleged GAAP violation that resulted in an Accounting and Auditing Enforcement Release

(AAER).23

We find a significantly negative coefficient on AAER (coefficient = -1.371, p-value = 0.067),

suggesting managers are less likely to provide a CAE disclosure during the years in which they

engage in egregious misreporting. More importantly, we continue to find support for our

hypotheses. The coefficients on VEGA, AUDITOROPPOSITION, and ACCT_EXPERT are

significant and in the predicted direction. The results from this test indicate the role of

misreporting incentives provided by vega in the CAE disclosure decision are not subsumed by

the existence of egregious, non-GAAP accounting choices.

Time Trends of CAE Disclosures and Portfolio Vega

Portfolio vega tends to decrease throughout the 2000s, the time frame spanning our sample

period. We also report that CAE disclosure frequencies steadily increase throughout our sample

period. A potential issue with the opposing time trend patterns of portfolio vega and CAE

disclosures is that it could lead to spurious inferences when interpreting our results for H1. While

this issue is alleviated by our inclusion of year fixed effects in Equation (1), in an additional test

we more directly control for the passage of time. Specifically, we re-estimate Equation (1)

without year fixed effects and include TREND, which equals the current year minus 2002, the

year in which the SEC proposed its rule on CAE disclosure. Consistent with CAE disclosures

increasing throughout our sample period, we find a significantly positive coefficient on TREND

(coefficient = 0.174, p-value < 0.0001). We also continue to find support for our hypotheses. The

coefficients on our test variables remain significant and in the predicted direction.

Non-Initiation Years for Disclosure

23

The SEC issues AAERs against a company, an auditor, or an officer for alleged accounting and/or auditing

misconduct (Dechow et al. 2011). In general, AAERs capture the incidence of egregious misreporting. We obtained

AAER data from the Center for Financial Reporting and Management (CFRM).

28

Our final robustness check examines whether our hypothesized test variables are significant

determinants of CAE disclosures in the years subsequent to disclosure initiation. Because CAE

disclosure are sticky, a potential concern is that our results are driven by the year in which the

CAE disclosure was initiated. To mitigate this concern, we re-estimate Equation (1) after

excluding from our sample firm-years associated with a first-time CAE disclosure. This process

decreases our sample to 2,176 observations. The results from re-estimating Equation (1) after

excluding first-time disclosing observations show significantly negative coefficients on VEGA

and AUDITOROPPOSITION and a significantly positive coefficient on ACCT_EXPERT. These

results suggest management, auditor, and audit committee incentives continue to play a role in

CAE disclosure decisions after the disclosure practice is initiated.

V. Conclusion

In the early 2000’s firms began providing quantitative sensitivity disclosures for their critical

accounting estimates. The new disclosures outline the earnings effects of reasonably likely

changes in highly uncertain accounting estimates. While the SEC mandated the new disclosures,

MD&A reporting is largely subjective, and it is likely that considerable discretion exists in the

determination of what is material and highly uncertain as defined by the SEC guidance. Because

of the discretionary aspect of this mandatory disclosure practice, we develop and test hypotheses

on the CAE disclosure preferences of management, auditors, and audit committees.

We find the likelihood and number of CAE disclosures are negatively related to managers’

equity risk incentives (portfolio vega) and auditor opposition to the SEC’s sensitivity disclosure

requirement, and positively related to audit committee accounting expertise. Our findings

suggest an aversion to CAE disclosures induced by the preferences of managers and auditors.

However, our findings also indicate that more knowledgeable audit committees play an

29

important role in mitigating strategic CAE disclosure practices. These results shed light on how

incentives influence reporting practices within the three-pronged governance “mosaic” of

management, auditors, and audit committees.

Other findings indicate management equity risk incentives decreased after the disclosure of a

CAE, suggesting firms respond to the increased transparency about accounting estimation

uncertainty. Consistent with CAE disclosures revealing the degree of discretion in accounting

estimates, firms adjust compensation practices to decrease misreporting incentives. This result

corroborates our conjecture that managers’ CAE disclosure strategies are partially based on their

misreporting incentives.

Our results provide several contributions for academics, regulators, and firms. First, we

provide robust empirical evidence on the determinants of firms’ decisions to provide CAE

disclosures. Second, we provide additional evidence on the role of management equity-risk

incentives in the disclosure process. Third, we document a role for auditors in CAE disclosures,

even though auditors are traditionally viewed as having a very limited role in MD&A disclosure.

Our study informs the SEC on how allowed discretion in mandatory disclosures can limit the

efficacy of such disclosures. Our study also informs firms on how managers, auditors, and the

audit committee influence disclosure quality. However, the study is not without limitations. Our

sample includes only non-financial and non-utilities companies in the S&P 500, and the results

might not generalize to smaller firms or to more regulated firms.

30

References

Armstrong, C. S., A. Jagolinzer, and D. F. Larcker. 2010. Chief executive officer equity

incentives and accounting irregularities. Journal of Accounting Research 48 (2): 225–

271.

Armstrong, C. S., D. F. Larcker, G. Ormazabal, and D. J. Taylor. 2013. The relation between

equity incentives and misreporting: the role of risk-taking incentives. Journal of

Financial Economics 109 (2): 327-350.

Baginski, S. P., J. M. Hassell, and M. D. Kimbrough. 2002. The effect of legal environment on

voluntary disclosure: Evidence from management earnings forecasts issued in US and

Canadian markets. The Accounting Review 77 (1): 25-50.

Bargeron, L. L., K. M. Lehn, and C. J. Zutter. 2010. Sarbanes-Oxley and corporate risk-taking.

Journal of Accounting and Economics 49 (1): 34-52.

Bauman, M., and K. W. Shaw. 2014. An analysis of critical accounting estimate disclosures of

pension assumptions. Forthcoming, Accounting Horizons.

Bergstresser. D., and T. Philippon. 2006. CEO incentives and earnings management. Journal of

Financial Economics 80: 511-529.

Beyer, A., D. A. Cohen, T. Z. Lys, and B. R. Walther. 2010. The financial reporting

environment: Review of the recent literature. Journal of Accounting and Economics 50

(2): 296-343.

Bhattacharjee, S., K. K. Moreno, and T. Riley. 2012. The interplay of interpersonal affect and

source reliability on auditors’ inventory judgments. Contemporary Accounting Research

29 (4): 1087-1108.

Burns, N., and S. Kedia. 2006. The impact of performance-based compensation on misreporting.

Journal of Financial Economics 79 (1): 35–67.

Cassell, C., L. Dreher, and L. Myers. 2013. Reviewing the SEC’s review process: 10-K comment

letters and the cost of remediation. The Accounting Review 88 (6): 1875-1908.

Chen, J. V., and F. Li. 2013. Estimating the amount of estimation in accruals. University of

Michigan Working Paper.

Cheng, Q., and D. Farber. 2008. Earnings restatements, changes in CEO compensation, and firm

performance. The Accounting Review 83 (5): 1217–1250.

Cheng, Q., and T. D. Warfield. 2005. Equity incentives and earnings management. The

Accounting Review 80 (2): 441–476.

31

Chung, J. O. Y., J. R. Cohen, and G. S. Monroe. 2008. The effect of moods on auditors’

inventory valuation decisions. Auditing: A Journal of Practice and Theory 27 (2): 137-

159.

Cohen, D. A., A. Dey, and T. Z. Lys. 2008. Real and accrual-based earnings management in the

pre- and post-Sarbanes-Oxley Periods. The Accounting Review 83 (3): 757-787.

Cohen, J., L. M. Gaynor, G. Krishnamoorthy, and A. M. Wright. 2007. Auditor communications

with the audit committee and the board of directors: Policy recommendations and

opportunities for future research. Accounting Horizons 21 (2): 165-187.

Cohen, J. R., U. Hoistash, G. Krishnamoorthy, and A. Wright. 2014. The effect of audit

committee industry expertise on monitoring the financial reporting process. The

Accounting Review 89 (1): 243-273.

Cohen, J., G. Krishnamoorthy, and A. M. Wright. 2004. The corporate governance mosaic and

financial reporting quality. Journal of Accounting Literature 23 (1): 87-152.

Cohen, R., B. Hall, and L. Viceira. 2000. Do executive stock options encourage risk-taking?

Working Paper, Harvard Business School.

Coles, J. L., N. D. Daniel, and L. Naveen. 2006. Managerial incentives and risk-taking. Journal

of Financial Economics 79 (2): 431-468.

Collins, D. W., G. Gong, and P. Hribar. 2003. Investor sophistication and the mispricing of

accruals. Review of Accounting Studies 8 (2-3): 251-276.

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.

Cornett, M., A. Marcus, and H. Tehranian. 2008. Corporate governance and pay-for-

performance: the impact of earnings management. Journal of Financial Economics 87

(2): 357–373.

Dechow, P. M., W. Ge, C. Larson, and R. Sloan. 2011. Predicting material accounting

misstatements. Contemporary Accounting Research 28 (1): 17–82.

Dhaliwal, D., Naiker, V., and Navissi, F., 2010. The association between accruals quality and the

characteristics of accounting experts and mix of expertise on audit committees.

Contemporary Accounting Research 27 (3): 782-827.

Dichev, I., J. Graham, C. R. Harvey, and S. Rajgopal. 2013. Earnings quality: Evidence from the

field. Journal of Accounting and Economics 56 (2): 1-33.

32

Efendi, J., A. Srivastava, and E. Swanson. 2007. Why do corporate managers misstate financial

statements? The role of option compensation and other factors. Journal of Financial

Economics 85 (3): 667–708.

Einhorn, E. 2005. The nature of the interaction between mandatory and voluntary disclosures.

Journal of Accounting Research 43 (4): 593-621.

Eng, L. L., and Y. T. Mak. 2003. Corporate governance and voluntary disclosure. Journal of

Accounting and Public Policy 22 (4), 325-345.

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.

Ernst and Young. 2011. SEC Comments and Trends: An Analysis of Current Reporting Issues.

New York, NY: Ernst and Young.

Fama, E. 1980. Agency problems and the theory of the firm. Journal of Political Economy 88:

288–307.

Gilson, S. C. 1990. Bankruptcy, boards, banks, and blockholders. Journal of Financial

Economics 27 (2): 355-387.

Glassman, C. A. June 8, 2006. Speech by SEC Commissioner: “Complexity in financial

reporting and disclosure regulation.” Remarks before 25th

Annual USC Leventhal School

of Accounting SEC and Financial Reporting Institute Conference.

Glendening, M. 2014. The value relevance of critical accounting estimate sensitivity disclosures.

Working Paper, University of Missouri.

Graham, J. R., C. R. Harvey, and S. Rajgopal. 2005. The economic implications of corporate

financial reporting. Journal of Accounting and Economics 40 (1): 3-73.

Guay, W. R. 1999. The sensitivity of CEO wealth to equity risk: an analysis of the magnitude

and determinants. Journal of Financial Economics 53 (1): 43–71.

Guay, W. R. 2008. Conservative financial reporting, debt covenants, and the agency costs of

debt. Journal of Accounting and Economics 45 (2): 175-180.

Hand, J. 1990. A test of the extended functional fixation hypothesis. The Accounting Review 65

(4): 740-763.

Hayes, R. M., M. Lemmon, and M. Qiu. 2012. Stock options and managerial incentives for risk

taking: Evidence from FAS 123R. Journal of Financial Economics 105 (1): 174-190.

33

Healy, P. M., A. P. Hutton, and K. G. Palepu. 1999. Stock performance and intermediation

changes surrounding sustained increases in disclosure. Contemporary Accounting

Research 16 (3): 485-520.

Hellman, N. 2011. Chief financial officer influence on audit planning. International Journal of

Auditing 15 (3): 247-274.

Holtzman, M. P. 2007. Reporting critical accounting policies. The CPA Journal 77 (12): 42.

Hribar, P., and D. C. Nichols. 2007. The use of unsigned earnings quality measures in tests of

earnings management. Journal of Accounting Research 45 (5): 1017-1053.

Johnson, M. F., R. Kasznik, and K. K. Nelson. 2001. The impact of securities litigation reform

on the disclosure of Forward‐Looking information by high technology firms. Journal of

Accounting Research 39 (2): 297-327.

Kim, I., and D. J. Skinner. 2012. Measuring securities litigation risk. Journal of Accounting and

Economics 53: 290-310.

Knechel, W. R. 2007. The business risk audit: Origins, obstacles and opportunities. Accounting,

Organizations and Society 32 (4): 383–408.

Krishnan, G.V., and G. Visvanathan. 2008. Does the SOX definition of an accounting expert

matter? The association between audit committee directors’ accounting expertise and

accounting conservatism. Contemporary Accounting Research 25 (3), 827-857.

Laksmana, I. 2008. Corporate board governance and voluntary disclosure of executive

compensation practices. Contemporary Accounting Research 25 (4), 1147-1182.

Lang, M. H., and R. J. Lundholm. 1993. Cross-sectional determinants of analyst ratings of

corporate disclosure. Journal of Accounting Research 31 (2): 246-271.

Lang, M. H., and R. J. Lundholm. 1996. Corporate disclosure policy and analyst behavior.

Accounting Review 31 (2): 467-492.

Larcker, D., S. Richardson, and I. Tuna. 2007. Corporate governance, accounting outcomes, and

organizational performance. The Accounting Review 82 (4): 963–1008.

Lev, B., S. Li, and T. Sougiannis. 2010. The usefulness of accounting estimates for predicting

cash flows and earnings. Review of Accounting Studies 15 (4): 779-807.

Levine, C. B., and M. J. Smith. 2011. Critical Accounting Policy Disclosures. Journal of

Accounting, Auditing and Finance 26 (1): 39-75.

Li, F. 2008. Annual report readability, current earnings, and earnings persistence. Journal of

Accounting and Economics 45 (2): 221-247.

34

Majors, T. M. 2014. Communicating measurement uncertainty: An experimental study of

financial reporting implications for managers and investors. Working paper, University of

Illinois.

McCracken, S., S. Salterio, and M. Gibbins. 2008. Auditor-client management relationships and

roles in negotiating financial reporting. Accounting, Organizations and Society 33 (4):

362-383.

Public Company Accounting Oversight Board (PCAOB). 2012. PCAOB Release No. 2012-004.

Auditing Standard Number 16 - Communications with Audit Committees. Washington,

D.C.: PCAOB.

Public Company Accounting Oversight Board (PCAOB). 2008. Report on the PCAOB’s 2004,

2005, 2006, and 2007 Inspections of Domestic Annually Inspected Firms. Release No.

2008-008, December 5. Washington D.C.: PCAOB.

Rajgopal, S., and T. Shevlin. 2002. Empirical evidence on the relation between stock option

compensation and risk taking. Journal of Accounting and Economics 33 (2): 145–171.

Rogers, J. L., and A. Van Buskirk. 2009. Shareholder litigation and changes in disclosure

behavior. Journal of Accounting and Economics 47 (1): 136-156.

Sahlman, W. A. 1990. Why sane people shouldn't serve on public boards. Harvard Business

Review 68 (May-June): 28–35.

SEC. 2001. Securities Act Release Nos.: 33-8040; 34-45149; FR-60. Cautionary Advice

Regarding Disclosure About Critical Accounting Policies, December 12, 2001.

SEC. 2002. Securities Act Release Nos.: 33-8098; 34-45907. Disclosure in Management’s

Discussion and Analysis about the Application of Critical Accounting Policies, May 10,

2002.

SEC. 2003. Securities Act Release Nos.: 33-8350; 34-48960; FR-72. Commission Guidance

Regarding Management’s Discussion and Analysis of financial Condition and Results of

Operations, December 19, 2003.

SEC. 2011. Transcript for Measurement uncertainty in financial reporting: How much to

recognize and how best to communicate it. Inaugural public roundtable for financial

reporting series. November 8, 2011.

http://www.sec.gov/about/offices/oca/frs/ocafrseries110811-transcript.txt.

Skinner, D. J., and R. G. Sloan. 2002. Earnings surprises, growth expectations, and stock returns

or don’t let an earnings torpedo sink your portfolio. Review of Accounting Studies 7 (2–

3): 289–312.

35

Slovic, P., M. L. Finucane, E. Peters, and D. G. MacGregor. 2007. The affect heuristic.

European Journal of Operational Research 177 (3): 1333-1352.

Walther, B. R. 1997. Investor sophistication and market earnings expectations. Journal of

Accounting Research 35 (2): 157-192.

Zhang, A. Y. 2012. Evidence on the trade-off between real activities manipulation and accrual-

based earnings management. The Accounting Review 87 (2): 675-703.

36

Appendix A

CAP Disclosures Versus CAE Disclosures

Source: TARGET CORP, 10-K, March 12, 2010 (CAP Disclosure)

Insurance/self-insurance

We retain a substantial portion of the risk related to certain general liability, workers' compensation, property loss and

team member medical and dental claims. However, we maintain stop-loss coverage to limit the exposure related to

certain risks. Liabilities associated with these losses include estimates of both claims filed and losses incurred but not

yet reported. We estimate our ultimate cost based on an analysis of historical data and actuarial estimates. General

liability and workers' compensation liabilities are recorded at our estimate of their net present value; other liabilities

referred to above are not discounted. We believe that the amounts accrued are adequate, although actual losses may

differ from the amounts provided. Refer to Item 7A for further disclosure of the market risks associated with these

exposures.

Source: WAL MART STORES INC, 10-K, March 30, 2010 (CAE Disclosure)

Self-Insurance

We use a combination of insurance, self-insured retention and self-insurance for a number of risks, including, but not

limited to, workers’ compensation, general liability, vehicle liability, and the company’s obligation for employee-related

health care benefits. Liabilities associated with the risks that we retain are estimated by considering historical claims

experience, including frequency, severity, demographic factors and other actuarial assumptions. In calculating our

liability, we analyze our historical trends, including loss development, and apply appropriate loss development factors to

the incurred costs associated with the claims made against our self-insured program. The estimated accruals for these

liabilities could be significantly affected if future occurrences or loss development differ from these assumptions. For

example, for our workers’ compensation and general liability accrual, a 1% increase or decrease to the

assumptions for claims costs or loss development factors would increase or decrease our self-insurance

accrual by $26 million (emphasis added).

37

Appendix B

Examples of CAE Disclosures

In respect to the wireless assets, a hypothetical 10% increase or decrease in the current cost factors would have

changed the impairment charge by $17 million. Also relative to the wireless assets, a hypothetical 100 basis point

change in the discount factors related to physical deterioration, functional obsolescence and economic obsolescence

would have changed the impairment charge by $10 million (Source: QWEST COMMUNICATIONS

INTERNATIONAL INC, 10-K, February 18, 2005).

An impairment charge of $1,420 million was recorded in 2004. Had we used a discount rate of 12%, the impairment

charge would have been approximately $90 million lower. Had we used a discount rate of 13%, the impairment charge

would have been approximately $80 million higher (Source: CORNING INC /NY, 10-K, February 22, 2005).

To the extent that Microelectronics' actual useful lives differ from management’s estimates by 10 percent,

consolidated net income in 2005 would have been an estimated $48 million higher if the actual lives were longer than

the estimates and an estimated $59 million lower if the actual lives were shorter than the estimates (based upon 2005

results) (Source: INTERNATIONAL BUSINESS MACHINES CORP, 10-K, February 28, 2006).

If the estimated useful lives of all depreciable assets were increased by one year, annual depreciation expense would

decrease by approximately $43 million. If the estimated useful lives of all depreciable assets were decreased by one

year, annual depreciation expense would increase by approximately $45 million (Source: UNION PACIFIC CORP, 10-

K, February 23, 2007).

As a measure of sensitivity, for every 1% of additional inventory valuation allowance at December 31, 2009 we would

have recorded an additional cost of sales of approximately $23 million (Source: AMAZON COM INC, 10-K, January

29, 2010).

For fiscal 2009, a 100 basis point change in total vendor funds earned, including advertising allowances, with no

offsetting changes to the base price on the products purchased, would impact gross profit by 10 basis points...As of

February 28, 2009, each 25 basis point change in the estimated inventory shortages would impact the allowances for

inventory shortages by approximately $13 (Source: SUPERVALU INC, 10-K, April 28, 2009).

A five percent change in the allowance for doubtful accounts would have had a pre–tax impact of approximately $2.6

million in 2005 (Source: BAKER HUGHES INC, 10-K, March 01, 2006).

A significant estimate in the McGraw-Hill Education segment, and particularly within the Higher Education,

Professional and International Group (“HPI”), is the allowance for sales returns, which is based on the historical rate

of return and current market conditions. Should the estimate for the HPI Group vary by one percentage point, it would

have an approximate $11.3 million impact on operating profit (Source: MCGRAW-HILL COMPANIES INC, 10-K,

February 29, 2008).

A change of 5% in the estimated sell-through levels by our wholesaler customers and in the estimated wholesaler

inventory levels would have an effect on our reserve balance of approximately $11 million (Source: MYLAN INC., 10-

K, February 24, 2011).

The effect of a change in the valuation allowance is reported in the current period tax expense. A 1% point increase

(decrease) in the Company’s effective tax rate would have decreased (increased) net income by approximately $15

(Source: AIR PRODUCTS & CHEMICALS INC /DE/, 10-K, November 26, 2008).

We believe that our estimates for the uncertain tax positions and valuation allowances against the deferred tax assets

are appropriate based on current facts and circumstances. A 5 percent change in the amount of the uncertain tax

positions and the valuation allowance would result in a change in net income of approximately $78 million and $26

million, respectively (Source: LILLY ELI & CO, 10-K, February 29, 2008).

38

Appendix B continued

Examples of CAE Disclosures

A 10% change in the sales return reserve would have had a $4 impact on our net earnings for the year ended January

31, 2009 (Source: NORDSTROM INC, 10-K, March 23, 2009).

a 10% variance in the workers’ compensation and general liability reserves at year-end 2008 would have affected net

income by approximately $14 million (Source: J C PENNEY CO INC, 10-K, March 31, 2009).

To the extent that our actual systems warranty costs differed from our estimates by 5 percent, consolidated pre-tax

income would have increased/decreased by approximately $10.0 in 2006 (Source: EMC CORP, 10-K, February 27,

2007).

A one-percentage point increase in the percentage of rebates to related gross sales would decrease net sales and

operating earnings by approximately $109 million in 2005 (Source: ABBOTT LABORATORIES, 10-K, February 22,

2006).

If the environmental reserve balance were to either increase or decrease based on the factors mentioned above, the

amount of the increase or decrease would be immediately recognized in earnings. For example, if the reserve balance

were to decrease by 10 percent, Occidental would record a pre-tax gain of $42 million. If the reserve balance were to

increase by 10 percent, Occidental would record an additional remediation expense of $42 million (Source:

OCCIDENTAL PETROLEUM CORP /DE/, 10-K, March 01, 2006).

A 10% change in our closed property reserves at September 28, 2008, would have affected net income by

approximately $4.0 million for fiscal year 2008 (Source: WHOLE FOODS MARKET INC, 10-K, November 26,

2008).

In addition, if future evidence indicates that the costs of performing services under these contracts are incurred on

other than a straight-line basis, the timing of revenue recognition under these contracts could change. A 10% change

in the amount of revenue recognized in 2009 under these contracts would have affected net earnings by approximately

$9 million (Source: LOWES COMPANIES INC, 10-K, March 30, 2010).

Our pension expense is sensitive to changes in our estimate of discount rate. Holding other assumptions constant, for a

100 basis point reduction in the discount rate, annual pension expense would increase by approximately $19.4 million

before taxes. Holding other assumptions constant, for a 100 basis point increase in the discount rate, annual pension

expense would decrease by approximately $19.2 million before taxes...Our pension expense is sensitive to changes in

our estimate of expected rate of return on plan assets. Holding other assumptions constant, an increase or decrease of

100 basis points in the expected rate of return on plan assets would increase or decrease annual pension expense by

approximately $7.7 million before taxes (Source: FMC TECHNOLOGIES INC, 10-K, March 01, 2007).

39

CAE Disclosure Variables :

CAE = 1 for firm-years with a CAE disclosure, zero otherwise

CAE# = number of CAE disclosures provided

Test Variables :

VEGA = natural logarithm of one plus the average dollar change (in $000s) in top-five

executives' wealth associated with a 1% change in the standard deviation of the

firm’s returns

AUDITOROPPOSITION = negative versus positive tone of the auditor's comment letter to the SEC on the

issue of CAE disclosures (see Appendix D)

ACCT_EXPERT = 1 for firm-years where the audit committee includes at least one accounting

expert, zero otherwise

Controls for Equity Incentives :

DELTA = natural logarithm of one plus the average dollar change (in $000s) in top-five

executives' wealth associated with a 1% change in the firm’s stock price

EQUITYPAY = natural logarithm of one plus the average dollar value (in $000s) of annual equity-

based compensation to the top-five executives, where equity-based compensation

equals the fair value of option grants plus the fair value of restricted stock awards

Controls for Governance Qualities :

LEGAL_EXPERT = 1 for firm-years where the audit committee includes at least one legal expert, zero

otherwise

AC_SIZE = number of audit committee members

AC_TENURE = number of years the audit committee members have served as directors

AC_MEET = number of audit committee meetings held

BD_SIZE = number of board of director members

BD_IND = percent of independent board of director members

DUAL = 1 for firm-years with a CEO that also serves as the chairperson of the board of

directors, zero otherwise.

Controls for Firm Characteristics :

ROA = the ratio of earnings before extraordinary items to total assets at year-end

BTM = ratio of book value of equity to market value of equity at year-end

SIZE = natural logarithm of year-end total assets

LEVERAGE = ratio of long-term debt to total assets at year-end

ACCRUALS = total accruals measured as earnings before extraordinary items less operating

cash flows taken directly from the statement of cash flows (adjusted for the cash

portion of discontinued operations and extraordinary items), scaled by total assets

ICW = 1 for firm-years with a material weakness under SOX Section 302 or SOX

Section 404, zero otherwise

SALESVOL = standard deviation of SALE t, SALE t-1, SALE t-2, where SALE is sales revenue

scaled by total assets

OPERVOL standard deviation of CFO t, CFO t-1, CFO t -2, where CFO is operating cash

flows scaled by total assets

LOSS = 1 for firm-years with negative ROA , zero otherwise

COMPLEXITY = operating complexity, defined as the natural logarithm of a firm's total geographic

and business segments

LITIGATION = probability of litigation estimated using the coefficients from the litigation risk

model in Table 7, Model (2) of Kim and Skinner (2012)

COVERAGE = natural logarithm of the number of analysts issuing a forecast for the firm

INSTOWN = percentage ownership by institutional investors

PENSION = 1 for firm-years with a positive projected benefit obligation, zero otherwise

Appendix C

Variable Definitions

40

Appendix D

Defining Auditor Opposition from Comment Letters

Following Li (2008), the Linguistic Inquiry and Word Count (LIWC) is used to measure the

negative versus positive emotion (NvsP) of each auditor’s comment letter. LIWC is a text

analysis software program developed by James W. Pennebaker, Roger J. Booth, and Martha E.

Francis. The software categorizes dictionary words into particular domains (e.g. positive emotion

words or negative emotion words). The LIWC dictionary contains nearly 4,500 words

categorizes 406 words as positive emotion words (e.g. “love,” “nice,” and “sweet”) and 499

words as negative emotion words (e.g. “hurt,” “ugly,” and “nasty”). See

http://www.liwc.net/tryonline.php.

For the portion of the each auditor's comment letter that discusses CAE,

AUDITOROPPOSITION is calculated as follows:

AUDITOROPPOSITION ( = NvsP) = ln((1 + Negemo)/(1 + Posemo))

where Negemo is the percentage of negative emotion words and Posemo is the percentage of

positive emotion words. The following table reports the AUDITOROPPOSITION score for each

auditor:

LIWC dimension Auditor 1 Auditor 2 Auditor 3 Auditor 4

Positive emotions 1.57 1.96 3.65 2.17

Negative emotions 0.70 1.61 1.42 2.39

Word Count 572.00 1739.00 493.00 461.00

AUDITOROPPOSITION = NvsP -0.413 -0.126 -0.653 0.067

41

Panel A: Sample

Firms Firm-years

460 4,140

(72) (648)

(40) (360)

(3) (27)

(6) (54)

Less: Firm-years without necessary governance data from Morningstar (2) (470)

Less: Firm-years without an auditor comment letter (3) (26)

Less: Firm-years without necessary Execucomp, Compustat, or I/B/E/S data (17) (257)

Sample used in determinants of CAE disclosures analysis 317 2,298

Panel B: CAE Disclosure Frequency by Year

Panel C: Descriptive Statistics for the Number of CAEs Disclosed

Year Disclosing Firms Mean Q1 Median Q3

2003 75 2.40 2.00 2.00 3.00

2004 128 2.36 2.00 2.00 3.00

2005 144 2.32 2.00 2.00 3.00

2006 139 2.26 2.00 2.00 3.00

2007 163 2.29 2.00 2.00 2.00

2008 172 2.33 2.00 2.00 3.00

2009 175 2.30 2.00 2.00 3.00

2010 175 2.48 2.00 2.00 3.00

1,171 2.34 2.00 2.00 3.00

Panel A outlines the sample. The initial sample of firms consists of firms on the S&P 500 for 2004 and includes years 2003-2010

for these firms. The initial sample of firms was identified from Compustat ExecuComp as firms with an SPCODE equal to 'SP' for

2004. Panel B reports the CAE disclosure rate by year. Panel C reports descriptive statistics for the number of CAEs disclosed.

Initial sample of S&P 500 firms identified from Execucomp in 2004

Table 1

Sample Selection and CAE Disclosure Data

Less: Firms acquired during 2002-2010

Less: Non-classifiable firms (four digit SIC code: 9900-9999)

Less: Utilities firms (four-digit SIC code: 4900-4949)

Less: Financial services firms (four-digit SIC code: 6000-6999)

2010

2009

2008

2007

2006

2005

2004

2003

Year

2,298

301

300

310

305

277

290

282

233

Firms Disclosure Rate

Number of CAEs Disclosed

50.18%

49.66%

45.39%

32.19%

50.96%

58.14%

58.33%

55.48%

53.44%

42

Panel A: Descriptive Statistics

Variable N Mean Std Dev Q1 Median Q3

CAE Disclosure Variables :

CAE 2,298 0.510 0.500 0.000 1.000 1.000

CAE# 2,298 1.192 1.485 0.000 1.000 2.000

Test Variables :

VEGA 2,298 4.403 1.305 3.859 4.597 5.234

AUDITOROPPOSITION 2,298 -0.256 0.265 -0.413 -0.413 0.067

ACCT_EXPERT 2,298 0.651 0.477 0.000 1.000 1.000

Controls for Equity Incentives :

DELTA 2,298 5.591 1.193 4.875 5.543 6.271

EQUITYPAY 2,298 6.107 2.608 6.095 6.930 7.620

Controls for Board Characteristics :

LEGAL_EXPERT 2,298 0.211 0.408 0.000 0.000 0.000

AC_SIZE 2,298 4.197 1.085 3.000 4.000 5.000

AC_TENURE 2,298 30.674 16.076 19.000 28.000 39.000

AC_MEET 2,298 9.054 3.118 7.000 9.000 11.000

BD_SIZE 2,298 10.447 2.137 9.000 10.000 12.000

BD_IND 2,298 0.880 0.071 0.857 0.900 0.917

DUAL 2,298 0.472 0.499 0.000 0.000 1.000

Controls for Other Firm Attributes :

ROA 2,298 0.074 0.077 0.043 0.075 0.112

BTM 2,298 0.366 0.253 0.208 0.316 0.482

SIZE 2,298 8.925 1.263 8.029 8.809 9.717

LEVERAGE 2,298 0.214 0.157 0.101 0.200 0.301

ACCRUALS 2,298 -0.057 0.066 -0.078 -0.049 -0.026

ICW 2,298 0.022 0.147 0.000 0.000 0.000

SALESVOL 2,298 0.096 0.108 0.034 0.066 0.117

OPERVOL 2,298 0.029 0.025 0.012 0.022 0.037

LOSS 2,298 0.073 0.260 0.000 0.000 0.000

COMPLEXITY 2,298 2.725 0.730 2.303 2.833 3.219

LITIGATION 2,298 0.576 0.365 0.197 0.611 0.979

COVERAGE 2,298 2.671 0.511 2.398 2.708 3.045

INSTOWN 2,298 0.769 0.166 0.689 0.785 0.863

PENSION 2,298 0.681 0.466 0.000 1.000 1.000

Table 2

Descriptive Statistics

43

Panel B: Differences in Variable Means and Medians

Variable Mean Median Mean Median Mean Median

Test Variables :

VEGA 4.422 4.647 4.384 4.542 0.487 0.046

AUDITOROPPOSITION -0.290 -0.413 -0.219 -0.126 <.0001 <.0001

ACCT_EXPERT 0.687 1.000 0.615 1.000 0.000 <.001

Controls for Equity Incentives :

DELTA 5.486 5.454 5.701 5.646 <.0001 <.0001

EQUITYPAY 6.129 6.915 6.085 6.951 0.685 0.384

Controls for Board Characteristics :

LEGAL_EXPERT 0.239 0.000 0.183 0.000 0.001 0.001

AC_SIZE 4.494 4.000 3.887 4.000 <.0001 <.0001

AC_TENURE 31.620 30.000 29.692 27.000 0.004 <.0001

AC_MEET 9.045 9.000 9.062 9.000 0.897 0.398

BD_SIZE 11.014 11.000 9.858 10.000 <.0001 <.0001

BD_IND 0.895 0.900 0.864 0.889 <.0001 <.0001

DUAL 0.518 1.000 0.424 0.000 <.0001 <.0001

Controls for Other Firm Attributes :

ROA 0.069 0.070 0.078 0.082 0.006 <.0001

BTM 0.371 0.339 0.361 0.303 0.364 0.034

SIZE 9.275 9.165 8.560 8.518 <.0001 <.0001

LEVERAGE 0.249 0.233 0.179 0.153 <.0001 <.0001

ACCRUALS -0.052 -0.046 -0.062 -0.051 0.000 <.0001

ICW 0.022 0.000 0.022 0.000 0.997 0.997

SALESVOL 0.099 0.066 0.094 0.066 0.243 0.700

OPERVOL 0.026 0.020 0.032 0.024 <.0001 <.0001

LOSS 0.064 0.000 0.082 0.000 0.105 0.105

COMPLEXITY 2.779 2.944 2.668 2.773 0.000 <.0001

LITIGATION 0.573 0.600 0.579 0.614 0.695 0.597

COVERAGE 2.578 2.639 2.767 2.833 <.0001 <.0001

INSTOWN 0.772 0.782 0.766 0.786 0.327 0.738

PENSION 0.899 1.000 0.455 0.000 <.0001 <.0001

Table 2 Continued

Descriptive Statistics

Panel A reports the descriptive statistics for variables used in the determinants of CAE disclosures analysis. Panel B reports p-

values for tests of differences in means and medians between disclosing and non-disclosing firm-years. A t-test is used to test

differences in means and a Wilcoxon rank-sum test is used to test differences in medians. All variables are defined in Appendix

C.

P-values for Differences(n = 1,171) (n = 1,127)

CAE = 0CAE = 1

44

CA

E

CA

E#

VE

GA

AU

DIT

OR

OP

PO

SIT

ION

AC

CT

_E

XP

ER

T

DE

LT

A

EQ

UIT

YP

AY

LE

GA

L_

EX

PE

RT

AC

_S

IZE

AC

_T

EN

UR

E

AC

_M

EE

T

BD

_S

IZE

BD

_IN

D

DU

AL

RO

A

BT

M

SIZ

E

LE

VE

RA

GE

AC

CR

UA

LS

ICW

SA

LE

SV

OL

OP

ER

VO

L

LO

SS

CO

MP

LE

XIT

Y

LIT

IGA

TIO

N

CO

VE

RA

GE

INS

TO

WN

PE

NS

ION

CAE 1.000 0.934 0.042 -0.134 0.075 -0.090 -0.018 0.069 0.301 0.081 0.018 0.273 0.244 0.094 -0.117 0.044 0.273 0.270 0.086 0.000 0.008 -0.116 -0.034 0.113 -0.011 -0.194 -0.007 0.476

CAE# 0.788 1.000 0.034 -0.151 0.089 -0.103 -0.010 0.042 0.314 0.093 0.019 0.267 0.237 0.117 -0.130 0.051 0.280 0.270 0.066 0.000 0.003 -0.088 -0.027 0.074 0.001 -0.167 0.007 0.464

VEGA 0.015 0.027 1.000 -0.025 -0.057 0.596 0.616 -0.034 0.065 0.043 0.041 0.255 0.171 0.008 0.163 -0.137 0.351 -0.086 0.016 0.006 -0.101 -0.153 -0.139 0.051 0.262 0.347 -0.190 0.069

AUDITOROPPOSITION -0.134 -0.144 -0.019 1.000 -0.026 0.058 -0.014 -0.029 -0.045 0.059 0.081 0.075 -0.001 -0.017 0.027 0.008 -0.032 -0.015 -0.030 -0.040 -0.007 0.043 0.000 -0.032 0.031 0.089 -0.041 -0.085

ACCT_EXPERT 0.075 0.072 -0.038 -0.027 1.000 -0.122 -0.116 0.070 0.007 -0.109 0.074 0.001 0.048 -0.017 -0.027 0.068 -0.096 -0.008 -0.026 -0.020 0.089 0.027 0.046 0.012 0.052 -0.071 0.122 -0.085

DELTA -0.090 -0.090 0.508 0.056 -0.088 1.000 0.440 0.023 -0.068 0.016 0.000 0.099 -0.019 0.014 0.243 -0.226 0.227 -0.181 0.011 -0.016 -0.079 -0.096 -0.175 -0.036 0.105 0.357 -0.217 -0.127

EQUITYPAY 0.008 0.025 0.539 0.009 -0.097 0.246 1.000 -0.032 0.024 -0.052 0.038 0.123 0.093 -0.010 0.108 -0.118 0.214 -0.093 0.024 0.043 -0.046 -0.067 -0.063 0.036 0.148 0.276 -0.099 0.029

LEGAL_EXPERT 0.069 0.003 -0.055 -0.031 0.070 0.027 -0.057 1.000 0.158 0.139 -0.028 -0.006 -0.018 0.019 -0.081 0.062 0.011 0.012 -0.029 0.016 -0.006 0.021 0.011 -0.015 0.026 0.001 -0.001 0.016

AC_SIZE 0.280 0.252 0.052 -0.046 0.016 -0.067 0.040 0.175 1.000 0.452 -0.083 0.399 0.241 0.124 -0.095 0.072 0.364 0.194 0.017 -0.007 -0.010 -0.050 0.015 0.111 0.107 -0.051 -0.099 0.317

AC_TENURE 0.060 0.045 0.047 0.079 -0.125 0.037 -0.051 0.139 0.445 1.000 -0.205 0.136 0.019 0.079 0.086 -0.009 0.136 0.006 0.029 -0.034 -0.024 -0.019 -0.051 0.081 0.025 0.009 -0.101 0.102

AC_MEET -0.003 0.010 0.046 0.078 0.071 0.050 0.035 -0.025 -0.092 -0.184 1.000 0.014 0.051 -0.107 -0.055 0.046 -0.003 -0.022 -0.048 0.063 -0.005 0.082 0.057 0.035 0.154 0.101 -0.012 -0.087

BD_SIZE 0.270 0.226 0.220 0.068 0.001 0.097 0.069 -0.001 0.351 0.119 0.006 1.000 0.376 0.000 -0.065 0.060 0.449 0.147 0.069 -0.017 -0.051 -0.164 -0.057 0.079 0.160 -0.009 -0.282 0.302

BD_IND 0.220 0.182 0.104 -0.040 0.065 -0.116 0.085 0.001 0.161 -0.035 0.059 0.195 1.000 -0.090 -0.075 0.034 0.256 0.213 0.064 0.020 -0.106 -0.096 -0.024 0.070 0.056 -0.029 -0.095 0.282

DUAL 0.094 0.110 -0.019 -0.018 -0.017 0.012 -0.013 0.019 0.106 0.071 -0.103 -0.005 -0.093 1.000 0.014 0.033 0.120 0.069 -0.006 -0.042 0.031 -0.013 -0.080 -0.056 0.011 -0.021 0.052 0.096

ROA -0.058 -0.065 0.146 -0.014 -0.026 0.243 0.053 -0.044 -0.025 0.094 -0.071 -0.058 -0.069 0.045 1.000 -0.529 -0.071 -0.339 0.206 -0.099 -0.052 0.064 -0.450 -0.020 -0.093 0.206 -0.082 -0.148

BTM 0.019 -0.001 -0.122 0.009 0.071 -0.200 -0.116 0.053 0.055 -0.031 0.028 0.036 0.027 0.041 -0.406 1.000 0.142 0.034 -0.015 -0.003 0.070 -0.072 0.153 0.065 0.143 -0.112 0.094 0.101

SIZE 0.283 0.284 0.299 -0.032 -0.087 0.224 0.068 0.012 0.315 0.125 0.019 0.453 0.189 0.112 -0.040 0.117 1.000 0.106 0.100 -0.034 0.071 -0.187 -0.073 0.075 0.482 0.243 -0.322 0.264

LEVERAGE 0.223 0.214 -0.078 -0.007 -0.017 -0.170 -0.016 -0.005 0.131 -0.012 -0.043 0.105 0.162 0.050 -0.228 -0.050 0.045 1.000 0.016 -0.031 -0.056 -0.104 0.143 -0.102 -0.100 -0.292 -0.017 0.265

ACCRUALS 0.074 0.044 0.045 -0.030 -0.024 0.077 0.040 -0.029 0.020 0.032 -0.045 0.068 0.047 0.016 0.576 -0.112 0.078 -0.028 1.000 -0.034 -0.085 -0.100 -0.303 0.094 -0.150 -0.136 -0.064 0.158

ICW 0.000 -0.006 0.022 -0.040 -0.020 -0.011 0.045 0.016 -0.014 -0.038 0.082 -0.014 0.034 -0.042 -0.070 0.000 -0.031 -0.033 -0.012 1.000 -0.044 0.042 0.083 0.008 -0.027 -0.041 0.048 -0.005

SALESVOL 0.024 0.019 -0.102 0.015 0.076 -0.066 -0.064 -0.005 0.011 -0.025 0.014 -0.005 -0.075 0.002 -0.075 0.123 0.095 -0.012 -0.050 -0.014 1.000 0.384 0.102 -0.077 0.098 -0.023 0.094 -0.032

OPERVOL -0.123 -0.078 -0.142 0.052 0.038 -0.090 -0.078 0.035 -0.017 -0.001 0.051 -0.155 -0.078 -0.046 0.023 -0.052 -0.191 0.008 -0.114 0.015 0.308 1.000 0.105 -0.043 0.087 0.020 0.119 -0.129

LOSS -0.034 -0.014 -0.124 -0.002 0.046 -0.197 -0.047 0.011 0.008 -0.052 0.062 -0.045 -0.001 -0.080 -0.607 0.176 -0.076 0.150 -0.460 0.083 0.086 0.109 1.000 0.010 0.175 -0.073 0.063 -0.010

COMPLEXITY 0.076 0.041 0.017 -0.056 0.029 -0.040 0.011 -0.023 0.095 0.036 0.032 0.059 0.047 -0.047 -0.013 0.045 0.030 -0.108 0.053 0.016 -0.057 -0.019 0.004 1.000 0.055 -0.016 -0.069 0.251

LITIGATION -0.008 0.035 0.194 0.025 0.047 0.080 0.014 0.029 0.114 0.050 0.134 0.145 0.009 0.009 -0.129 0.151 0.454 -0.068 -0.156 -0.028 0.061 0.130 0.175 0.015 1.000 0.468 -0.166 -0.129

COVERAGE -0.186 -0.092 0.316 0.108 -0.072 0.332 0.139 -0.013 -0.024 0.031 0.109 -0.006 -0.049 -0.013 0.162 -0.086 0.262 -0.236 -0.065 -0.041 -0.031 0.012 -0.077 -0.036 0.410 1.000 -0.149 -0.265

INSTOWN 0.020 0.030 -0.105 -0.062 0.105 -0.208 -0.022 0.002 -0.057 -0.139 -0.016 -0.286 -0.028 0.049 -0.040 0.069 -0.259 0.000 -0.063 0.042 0.050 0.088 0.053 -0.025 -0.103 -0.129 1.000 -0.031

PENSION 0.476 0.369 0.067 -0.084 -0.085 -0.151 0.065 0.016 0.285 0.088 -0.098 0.293 0.258 0.096 -0.085 0.049 0.254 0.215 0.120 -0.005 -0.028 -0.141 -0.010 0.208 -0.122 -0.240 0.031 1.000

This table provides the Pearson (below diagonal) and Spearman (above diagonal) correlation coefficients for variables used in the determinants of CAE disclosures analysis. Correlations in bold are significant at a level of 10%.

Table 3

Correlations

45

Prediction Estimate p-value

Intercept ? -8.268 0.004

Test Variables :

VEGA - -0.198 0.045

AUDITOROPPOSITION - -1.160 0.015

ACCT_EXPERT + 0.553 0.012

Controls for Equity Incentives :

DELTA ? 0.040 0.714

EQUITYPAY ? 0.043 0.292

Controls for Board Characteristics :

LEGAL_EXPERT + 0.279 0.192

AC_SIZE + 0.192 0.058

AC_TENURE + 0.000 0.494

AC_MEET + 0.064 0.031

BD_SIZE + 0.091 0.072

BD_IND + 1.840 0.140

DUAL - 0.121 0.300

Controls for Other Firm Attributes :

ROA + -0.578 0.388

BTM ? -0.894 0.066

SIZE + 0.418 0.009

LEVERAGE + 1.747 0.025

ACCRUALS + -0.398 0.415

ICW + 0.344 0.265

SALESVOL + 0.891 0.168

OPERVOL + -3.697 0.183

LOSS ? -0.512 0.112

COMPLEXITY + 0.214 0.137

LITIGATION - -0.476 0.163

COVERAGE ? -0.551 0.039

INSTOWN ? 0.234 0.790

PENSION + 1.908 <.0001

N

Likelihood Ratio, χ2

p-value

Table 4

Determinants of CAE Disclosures

2,298

1127.41

<.0001

This table reports the coefficients and p-values from a logistic regression where the dependent variable

is CAE . Year and industry (at the 2-digit SIC level) fixed effects are included but not tabulated. P-values

are based on one-tailed tests where a prediction is made (two-tailed otherwise) using Huber/White

robust standard errors with within-firm clustering.

46

Prediction Change in probability

Test Variables :

VEGA - -6.77%

AUDITOROPPOSITION - -13.84%

ACCT_EXPERT + 13.73%

Controls for Equity Incentives :

DELTA ? N/A

EQUITYPAY ? N/A

Controls for Board Characteristics :

LEGAL_EXPERT + N/A

AC_SIZE + 9.57%

AC_TENURE + N/A

AC_MEET + 6.37%

BD_SIZE + 6.82%

BD_IND + N/A

DUAL - N/A

Controls for Other Firm Attributes :

ROA + N/A

BTM ? -6.08%

SIZE + 17.43%

LEVERAGE + 8.68%

ACCRUALS + N/A

ICW + N/A

SALESVOL + N/A

OPERVOL + N/A

LOSS ? N/A

COMPLEXITY + N/A

LITIGATION - N/A

COVERAGE ? -8.88%

INSTOWN ? N/A

PENSION + 43.83%

Table 5

Using Table 4 results to assess changes in the probability of a CAE disclosure for

selected changes in independent variables

This table reports the change in probability of a CAE disclosure as a result of selected changes in the

value of the variable of interest holding other independent variables at their mean values. Non-

dichotomous variables are changed from the first to the third quartile. Dichotomous variables are

changed from zero to 1. We calculate the change in the probability using the following expression: eβ’X

/ (1 + eβ’X), where β refers to the vector of coefficients from Table 4 and X refers to the vector of

independent variables.

47

Prediction Estimate p-value

Test Variables :

VEGA - -0.160 0.034

AUDITOROPPOSITION - -0.815 0.037

ACCT_EXPERT + 0.528 0.006

Controls for Equity Incentives :

DELTA ? -0.015 0.871

EQUITYPAY ? 0.031 0.387

Controls for Board Characteristics :

LEGAL_EXPERT + 0.077 0.371

AC_SIZE + 0.171 0.023

AC_TENURE + 0.001 0.465

AC_MEET + 0.052 0.028

BD_SIZE + 0.070 0.076

BD_IND + 1.319 0.168

DUAL - 0.209 0.119

Controls for Other Firm Attributes :

ROA + -0.966 0.293

BTM ? -0.686 0.048

SIZE + 0.402 0.003

LEVERAGE + 2.097 0.004

ACCRUALS + 0.390 0.401

ICW + 0.177 0.329

SALESVOL + 0.477 0.239

OPERVOL + 0.009 0.499

LOSS ? -0.168 0.512

COMPLEXITY + 0.092 0.253

LITIGATION - -0.656 0.042

COVERAGE ? -0.226 0.292

INSTOWN ? 0.790 0.296

PENSION + 2.000 <.0001

N

Likelihood Ratio, χ2

p-value <.0001

This table reports the coefficients and p-values from an ordered logistic regression where the dependent

variable is CAE#. The intercepts are not tabulated. Year and industry (at the 2-digit SIC level) fixed

effects are included but not tabulated. P-values are based on one-tailed tests where a prediction is made

(two-tailed otherwise) using Huber/White robust standard errors with within-firm clustering.

Table 6

Determinants of the Number of CAEs Disclosed

2,298

1246.27

48

Prediction Estimate p-value Estimate p-value

Intercept ? -1.287 0.210 -3.620 <.0001

CAE - -0.147 0.026 -0.175 0.003

DELTA + 0.389 <.0001 0.428 <.0001

CASHCOMP ? 0.075 0.360 0.100 0.225

SALE + 0.245 0.006 0.378 <.0001

BTM - 0.459 0.008 0.530 0.005

LEVERAGE + 0.586 0.041 0.893 0.003

R&D + 2.339 0.013 2.605 0.004

CAPEX - -0.273 0.372 -1.489 0.061

RETURNVOL + -0.755 0.109 0.942 0.010

POSTSOX ? 0.573 <.0001

POST123(R) - -0.276 <.0001

Year Fixed-Effects

Firm Fixed-Effects

N

Adjusted R2

Table 7

Effect of CAE Disclosures on Vega

Included Not Included

Included Included

2,838 2,838

70.88% 68.64%

This table reports the coefficients and p-values from a firm fixed-effect regression of VEGA on the

lagged CAE disclosure variable during the pre- and post-disclosure period (1995-2010). For this

expanded sample period, the model is estimated for a sample of only those firms that provide a CAE

disclosure at some point during the post-disclosure period. P-values are based on one-tailed tests

where a prediction is made (two-tailed otherwise) using Huber/White robust standard errors with

within-firm clustering.