AUDIT TRANSPARENCY AND AUDITORS’ REPORTING

BEHAVIOR

Penny Fanyuan Zhang

Master of Commerce (The Australian National University)

A thesis submitted for the degree of

Doctor of Philosophy

The Australian National University

28 February 2019

iii

Acknowledgements

I would like to express my sincere appreciation to my supervisor, Professor Greg Shailer,

for his continuous and energetic support, patience, motivation, supervision, suggestions, and

immense knowledge in the past four years. This thesis would have not been possible without the

strength and guidance he offered. He is not only a responsible and encouraging supervisor but also

a lifelong friend who selflessly shares his career experience. I believe that everything I have learned

from him will stand me in good stead for many years to come.

I would like to take this opportunity to express my gratitude towards my co-supervisors,

Professor Neil Fargher and Associate Professor Louise Lu. Their valuable comments, suggestions,

and kind help have enabled me to refine and improve my work significantly.

I would like to thank the Director of the Research School of Accounting (RSA), Professor

Juliana Ng, for her support and understanding. I truly appreciate Dr Alicia Jiang for her endless

academic and life advice as my lecturer and friend. I also thank Dr Sorin Daniliuc for his guidance

as my master supervisor. His kindness helped to lead me into this area of research. I would also

like to express my sincere appreciation to the administration group at the RSA.

I am indebted to my friend and colleague, Ao Li, for the wonderful times and memorable

experiences we have shared in the past few years. I would also like to thank my friend, Yuria Liu,

for helping me through all the events and difficulties of these years. Thanks to all my friends in

Australia and China for your love and encouragement.

I thank Elite Editing for the thesis editing; their editorial intervention was restricted to

Standards D and E of the Australian Standards for Editing Practice.

iv

Last, I am deeply grateful to my parents Jun Zhang and Aiqin Wang. Without their

unreserved love, support, and understanding, I could not have gone this far in my studies. This

thesis is also dedicated to my beloved husband, Cai Li, and my upcoming baby girl, Kalie Li. Both

of you have made me stronger and more fulfilled than ever.

v

Abstract

This thesis focuses on the enhanced auditor’s reports that were first introduced in the United

Kingdom (UK) from 2013, with similar auditing standards subsequently introduced in many

jurisdictions. To enhance audit transparency, national regulators adopted standards for enhanced

audit reports, requiring auditors to tailor reports that identify material audit-related matters for an

engagement in a particular year. The first three implementation years of extended auditor’s reports

(EARs) in the UK, from September 30, 2013 to September 30, 2016 are examined in this thesis, to

understand better the way auditors behaved in implementing the new reporting requirements. The

thesis addresses the following three research questions (as three separate studies) that are

collectively concerned with the content of the more tailored EARs: (1) Is auditors’ standardized

wording usage in EARs a consequence of the underlying audit quality?; (2) Are auditors’ risk

disclosures in EARs influenced by the expertise of the audit committee (AC)?; and (3) Are the

year-to-year changes in EARs disclosures associated with audit effort?

Study 1 examines whether textual similarities of an EAR to other EARs within the audit

firm’s client portfolio reflect the relative audit quality at the engagement level. Similarity scores

are found to be positively related to abnormal accruals and negatively related to abnormal audit

fees. While the relations for abnormal accruals are significant for both key audit matters (KAM)

sub-sections and full EARs, they are stronger for the KAM sub-sections.

Study 2 investigates whether auditors’ reporting decisions are influenced by the expertise

of the AC. AC influence is assessed by examining the relation between auditors’ choices of KAMs

relative to the significant accounting issues (SIFs) reported by ACs in the extended audit committee

reports (EACRs), which the UK now requires under its new reporting regime. I find that AC

vi

expertise significantly affects the auditor’s KAM disclosures, evidenced by fewer KAMs reported

in EARs and more consistency between KAM disclosures and ACs’ SIF disclosures. When

analyzing cases pertaining to non-alignment of KAMs and SIFs, I find that changes in the ACs’

accounting and industry expertise have significant effects on the auditors’ subsequent adoption of

SIFs that were not previously matched with KAMs. Further analysis of auditors’ subsequent

adoption of unmatched SIFs reveals a significant negative association with abnormal audit fees but

no significant relation with abnormal accruals, suggesting that, on average, an auditor’s initial

omission of subsequently adopted unmatched SIFs may be a consequence of insufficient audit

effort, but it is not necessarily a sign of lower audit quality.

Study 3 investigates whether auditors’ year-to-year changes in EARs disclosures are related

to the underlying audit effort, proxied by audit fees. Changes in EARs from year to year are

measured in terms of language usage and KAM selections. I find that audit fee changes are

negatively associated with language similarity scores (measured over time for a single company)

and “repeated” KAMs, and positively associated with “removed” and “added” KAMs, where the

“added” KAMs effect is further found to be driven by the “new” KAMs that are not borrowed from

the prior year’s EACR.

Overall, the results find that engagements with higher audit quality result in more tailored

EARs and greater changes in EARs disclosures reflect more audit effort. The findings indicate that

the contextual information included in EARs allows financial statement users to differentiate

among auditors and the AC plays a critical role in this new reporting regime.

vii

Table of Contents

Abstract .......................................................................................................................................... v

CHAPTER 1: Introduction .......................................................................................................... 1

CHAPTER 2: Literature Review of EARs ............................................................................... 10

CHAPTER 3: Study 1: Extended Auditor’s Reports and Audit Quality: A Textual

Analysis ........................................................................................................................................ 14

3.1 Introduction .......................................................................................................................... 14

3.2 Literature Review and Hypothesis Development ................................................................. 18

3.2.1. Prior Studies Related to EARs ...................................................................................... 18

3.2.2. Prior Studies Related to Report Transparency .............................................................. 20

3.2.3. Hypothesis Development .............................................................................................. 22

3.3 Method .................................................................................................................................. 24

3.3.1. Measurement of EAR Similarity .................................................................................. 24

3.3.2. Models for Testing the Hypothesis ............................................................................... 28

3.3.3. Sample Selection........................................................................................................... 30

3.4 Results .................................................................................................................................. 32

3.4.1. Descriptive Statistics..................................................................................................... 32

3.4.2. Regression Results ........................................................................................................ 36

3.5 Robustness Tests ................................................................................................................... 38

3.5.1. Analyses Using Alternative Measures of EAR Similarity............................................ 38

3.5.2. Analyses Using the Unadjusted Similarity Scores ....................................................... 40

3.5.3. Analyses Using Income-Increasing Abnormal Accruals .............................................. 40

3.5.4. Analyses Using Abnormal Audit Fees.......................................................................... 41

3.6 Additional Analysis .............................................................................................................. 42

3.7 Conclusions .......................................................................................................................... 45

CHAPTER 4: Study 2: The Effect of Audit Committee Expertise on External Auditors’

Disclosures of Key Audit Matters .............................................................................................. 47

4.1 Introduction .......................................................................................................................... 47

4.2 Literature Review and Hypotheses Development ................................................................ 51

viii

4.2.1. Regulatory Standards and Related Studies ................................................................... 51

4.2.2. Prior Studies Related to Audit Committee Expertise ................................................... 53

4.2.3. Hypotheses Development ............................................................................................. 56

4.3 Method .................................................................................................................................. 60

4.3.1. Models for Testing the Hypotheses .............................................................................. 60

4.3.2. Sample Selection........................................................................................................... 63

4.4 Results .................................................................................................................................. 64

4.4.1. Descriptive Statistics..................................................................................................... 64

4.4.2. Regression Results ........................................................................................................ 68

4.5 Robustness Tests ................................................................................................................... 73

4.5.1. Analyses Excluding Entities Not in Compliance with the Corporate Governance

Code .............................................................................................................................. 73

4.5.2. Analyses Using Alternative Measures of Unmatched KAMs, Unmatched SIFs,

and Count Regression ................................................................................................... 73

4.5.3. Analyses Using Alternative Measure of AC Expertise ................................................ 74

4.5.4. Analyses Using Alternative Measure of UNMATCHED............................................. 74

4.6 Additional Analyses ............................................................................................................. 75

4.6.1. Analyses on UNMATCHED_SIF Reported as KAM in the Following Year .............. 75

4.6.2. Analyses for Audit Quality on SIFxPICK Behavior .................................................... 81

4.7 Conclusions .......................................................................................................................... 83

CHAPTER 5: Study 3: Year-to-year Extended Auditor’s Report Changes and Audit

Effort ............................................................................................................................................ 86

5.1 Introduction .......................................................................................................................... 86

5.2 Literature Review and Hypothesis Development ................................................................. 90

5.2.1. Prior Studies Related to EARs ...................................................................................... 90

5.2.2. Prior Studies Related to Audit Fees .............................................................................. 91

5.2.3. Hypothesis Development .............................................................................................. 93

5.3 Method .................................................................................................................................. 95

5.3.1. Measurement of Year-to-year EAR Modifications ...................................................... 95

5.3.2. Models for Testing the Hypotheses .............................................................................. 98

5.3.3. Sample Selection........................................................................................................... 99

ix

5.4 Results ................................................................................................................................ 100

5.4.1. Descriptive Statistics................................................................................................... 100

5.4.2. Regression Results ...................................................................................................... 104

5.5 Robustness Test .................................................................................................................. 107

5.5.1. Analyses Using Cross-section Similarity Measures ................................................... 107

5.6 Additional Analysis ............................................................................................................ 108

5.6.1. Analysis on Added KAMs .......................................................................................... 108

5.7 Conclusions ........................................................................................................................ 111

CHAPTER 6: Conclusion ......................................................................................................... 112

References .................................................................................................................................. 116

Appendices ................................................................................................................................. 125

Appendix 1 Variable Definitions ............................................................................................. 125

Appendix 2 Regression Results Using Alternative Measures of EAR Similarity ................... 132

Appendix 3 Regression Results Using the Unadjusted Similarity Scores ............................... 134

Appendix 4 Regression Results Using Income-increasing Abnormal Accruals ...................... 136

Appendix 5 Regression Results Using Abnormal Audit Fees ................................................. 137

Appendix 6 Regression Results for Analyses Excluding Entities Not in Compliance with

the Corporate Governance Code ......................................................................................... 138

Appendix 7 Regression Results for Analyses Using Alternative Measures of Unmatched

KAMs and Unmatched SIFs, and Count Regression ......................................................... 139

Appendix 8 Regression Results for Analyses Using Alternative Measures of AC Expertise . 141

Appendix 9 Regression Results for Analyses Using Alternative Measures of

UNMATCHED .................................................................................................................... 142

Appendix 10 Regression Results for Analyses Using Cross-Section Similarity Measures ..... 143

x

List of Tables

Table 3.1 Sample Selection ............................................................................................................ 31

Table 3.2 Descriptive Statistics ...................................................................................................... 32

Table 3.3 Pearson Correlation Matrix ............................................................................................ 35

Table 3.4 Regression Results for Analyses .................................................................................... 37

Table 3.5 Regression Results for Pre-Post EAR Analysis ............................................................. 44

Table 4.1 Sample Selection ............................................................................................................ 64

Table 4.2 Descriptive Statistics for Hypotheses-testing Sample .................................................... 65

Table 4.3 Pearson Correlation Matrix ............................................................................................ 67

Table 4.4 Regression Results for Hypothesis 1 Test ...................................................................... 69

Table 4.5 Regression Results for Hypothesis 2 Test ...................................................................... 71

Table 4.6 Sample Selection for Further Analyses .......................................................................... 77

Table 4.7 Regression Results for Audit Fee Model ....................................................................... 78

Table 4.8 Logistic Regression Results for PICK on AC Expertise ................................................ 79

Table 4.9 Regression Results for Abnormal Accruals on SIFxPICK Behavior ............................. 83

Table 5.1 Sample Selection .......................................................................................................... 100

Table 5.2 Descriptive Statistics .................................................................................................... 101

Table 5.3 Pearson Correlation Matrix .......................................................................................... 103

Table 5.4 Regression Results ....................................................................................................... 105

Table 5.5 Regression Results for Analysis on Added KAMs ...................................................... 110

xi

List of Acronyms and Abbreviations

AC Audit Committee

AIM Alternative Investment Market

EACR Extended Audit Committee Report

EAR Extended Auditor’s Report

EC European Commission

EU European Union

FRC Financial Reporting Council

IAASB International Auditing and Assurance Standards Board

KAM Key Audit Matter

LSE London Stock Exchange

MD&A Management Discussion and Analysis

PCAOB Public Company Accounting Oversight Board

RSA Research School of Accounting

SEC Securities and Exchange Commission

UK United Kingdom

US United States

VSM Vector Space Model

1

CHAPTER 1: Introduction

The binary (pass/fail) auditor’s report has long been criticized because of concerns

regarding its form, wording, and limited communicative value (e.g., Church, Davis, and

McCracken 2008; Mock et al. 2013). Especially after the 2008 Global Financial Crisis, regulators

and investors put substantial emphasis on audit effectiveness, including whether the binary

auditor’s report provides adequate transparency to financial statements’ users regarding the audit

and the auditor’s insights about the entity (e.g., EC [European Commission] 2011; FRC [Financial

Reporting Council] 2015a; Simnett and Huggins 2014).

In May 2011, a consultation paper issued by the International Auditing and Assurance

Standards Board (IAASB) advocated that providing further information about the audit could

improve the communicative value of the auditor’s report: for example, disclosing “Key areas of

risk of material misstatement of the financial statements identified by the auditor” and “Areas of

significant auditor judgment” (IAASB 2011, paragraph 62). In June 2011, the Public Company

Accounting Oversight Board (PCAOB) issued a concept release to seek public comment on several

options for enhancing the auditor’s reporting model, including “a supplement to the auditor’s report

in which the auditor would be required to provide additional information about the audit and the

company’s financial statements” (PCAOB 2011, 2). Simultaneously, the EC proposed a regulation

in relation to enriching audit transparency through expanding the audit report, including “the key

areas of risk of material misstatements of the financial statements” (EC 2011, 7). Taken together,

all these proposed reforms aimed to make auditors provide greater transparency into the audited

entity, its financial statements, and the performed audit. Auditors, dealing with significant

information through the audit, can obtain and develop great insights regarding the entities and their

2

material misstatement risks, critical accounting estimates, and judgments applied, and the

disclosure of such information would be of much value and benefit to financial statement users.

The United Kingdom (UK) FRC is ahead of the IAASB, PCAOB, and EU with regard to

implementing regulatory changes. Under the International Standards on Auditing (ISA) (UK and

Ireland) 700, for financial periods ending on or after September 30, 2013, entities that report on

application of the UK Corporate Governance Code are required to prepare an extended auditor’s

report (EAR) (FRC 2013a), with the IAASB standard, ISA 701, becoming effective for periods

ending on or after December 15, 2016 (IAASB 2015). According to the EU (2014), public-interest

entities need to adopt a new audit regime for accounting periods ending on and after June 16, 2017.

For large accelerated filers and all other companies in the United States (US), revised audit

reporting standards will become effective for periods ending on or after June 30, 2019, and

December 15, 2020, respectively (PCAOB 2017).

Although different terminologies are used in these standards, requiring auditors to disclose

significant audit matters in the year’s financial statement audit is consistent. According to the ISA

(UK and Ireland) 700, auditors are required to “describe those assessed risks of material

misstatement that were identified by the auditor and which had the greatest effect on: the overall

audit strategy; the allocation of resources in the audit; and directing the efforts of the engagement

team” (FRC 2013a, paragraph 19A).1 Under the ISA 701, the disclosed matters are named “key

audit matters” (KAMs), defined as “those matters that, in the auditor’s professional judgment, were

1 In the UK, entities applying the UK Corporate Governance Code include those with a premium listing of equity

shares regardless of whether they are incorporated in the UK or elsewhere (FRC 2013a, 6).

3

of most significance in the audit of the financial statements of the current period” (IAASB 2015,

paragraph 8). Consistently, Regulation (EU) No. 537/2014 of the European Parliament and of the

Council states that the auditor should provide “in support of the audit opinion, a description of the

most significant assessed risks of material misstatement, including assessed risks of material

misstatement due to fraud; a summary of the auditor’s response to those risks; and where relevant,

key observations arising with respect to those risks” (EU 2014, article 10). In addition, under the

PCAOB’s standard, critical audit matters (CAMs) should be communicated in the auditor’s report,

defined as matters “communicated or required to be communicated to the audit committee (AC)

and that: (1) relate to accounts or disclosures that are material to the financial statements; and (2)

involved especially challenging, subjective, or complex auditor judgment” (PCAOB 2017, 11).

Collectively, there is a significant overlap among these reforms and the requirement of material

audit matters disclosures. Because the UK was the first jurisdiction to adopt the changes, the three

studies in this thesis are in that jurisdiction, with the FRC’s consistent regulatory improvement

making the results of this thesis generalizable to other jurisdictions.

Because the auditor’s report is the primary means by which the auditor communicates

information to financial statement users, its informativeness is of particular importance. Improving

the informativeness of audit report by enhancing their transparency in relation to the audit is the

key objective of this new reporting regime, with several suggestions to achieve this being raised

by regulators.

First, the language used to describe the auditor’s work should be non-standardized, as it is

believed that using standardized wording to discuss similar audit matters and responses across

4

entities and accounting periods leads to the EARs becoming longer, “boilerplate” reports.2 To make

the enhanced audit reports useful to investors and other financial statement users, regulators have

made a number of suggestions regarding the wording used in the standards. The ISA (UK and

Ireland) 700 advocates that communicated matters should be described “in a way that enables them

to be related directly to the specific circumstances of the audited entity and are not, therefore,

generic or abstract matters expressed in standardized language” (FRC 2013a, paragraph 19B).

According to ISA 701, when describing a KAM and discussing how this matter has been addressed,

auditors should relate “a matter directly to the specific circumstances of the entity” and avoid

“generic or standardized language,” to minimize the potential that EARs will “become overly

standardized and less useful over time” (IAASB 2015, paragraphs A44, A47). In addition, the

PCAOB advocates “communication will be tailored to the audit to avoid standardized language

and to reflect the specific circumstances of the matter” (PCAOB 2017, 32).

Second, standard setters have considered improving audit transparency through enriching

the AC disclosures; as ACs have specific responsibilities for a degree of oversight of the auditor or

aspects of the audit process, fuller disclosure of the AC activities in relation to the external audit

would benefit both the actual and perceived audit quality (IAASB 2014). In addition, the US

Securities and Exchange Commission (SEC) has considered whether ACs need to provide investors

with more external auditor-related information, such as the way that ACs perform their oversight

responsibilities (SEC 2015). It is worth noting that going beyond the US accounting standards,

some entities have provided additional voluntary disclosures (see SEC 2015, 22), including the

2 The terms “generic” “standardized,” and “boilerplate” (or “tailored,” “granular,” and “specific”) are used

interchangeably by regulators (e.g., FRC 2016a).

5

significant areas that ACs have addressed with the auditor. In this regard, the UK is ahead of other

jurisdictions; for financial periods ending on or after September 30, 2013, ACs of companies that

report on the application of the UK Corporate Governance Code are required to issue extended

audit committee reports (EACRs). These require committee members to disclose “the significant

issues that the committee considered in relation to the financial statements and how these issues

were addressed; an explanation of how it has assessed the effectiveness of the external audit process

and the approach taken to the appointment or reappointment of the external auditor, and

information on the length of tenure of the current audit firm and when a tender was last conducted;

and if the external auditor provides non-audit services, an explanation of how auditor objectivity

and independence is safeguarded” (FRC 2012, Section C3.8). The most substantial change is the

requirement to report significant issues. Together with the simultaneous change in audit reporting,

audit transparency can be further enhanced by providing investors and other financial statement

users with different but complementary perspectives on key issues in financial reporting.

While audit transparency is the core purpose of implementing the long-form auditor’s

report, several other indirect benefits are expected (FRC 2015a; IAASB 2015; PCAOB 2017). In

particular, it is argued that by providing financial statement users with more entity-specific

information, the overall information asymmetry between investors and managers with regard to a

company’s financial performance should be reduced. The more informative auditor’s report should

facilitate investors’ ability to analyze financial statements, assess financial performance, or monitor

management’s stewardship of the company. In addition, it is argued that by requiring auditors to

provide more contextual information, financial statement users can better evaluate external

auditors’ performance and differentiate between accounting firms. Moreover, the requirement for

additional disclosures, making the audit more visible, is expected to improve the behavior of

6

auditors, managers, and ACs, leading to improvements in audit quality and financial reporting

quality.

This thesis aims to gain a better understanding of auditor reporting under the new audit

transparency regime, by investigating three specific dimensions: the extent to which audit quality

differentiations can be revealed by the enhanced disclosures; the effectiveness of communications

between external auditors and ACs, with a particular focus on the influence that AC expertise has

on auditors’ disclosures; and the relation between auditors’ reporting behavior and audit effort.

The first study investigates the extent to which audit firms’ approaches to EARs can be

used to differentiate between accounting engagements in terms of audit quality. In particular, I

investigate the extent to which entity-specific information (in other words, non-standardized

language) is provided in the EARs and whether variations among the EARs’ contextual disclosures

across an audit firm’s client portfolio are associated with audit quality differentiations, and. As

noted earlier, using non-standardized wording is critical to achieving the objectives of EAR

adoption. In addition, an anticipated indirect benefit of the narrative disclosures in EARs is helping

financial statement users to differentiate between auditors’ performances. To address these issues,

I use textual analysis to examine whether the use of standardized or generic disclosures in EARs

relate to audit quality.

The second study examines the extent to which the expertise of the AC influences external

auditors’ reporting behavior. Answering this question is important because it may facilitate a better

understanding of the determinants of auditors’ risk disclosures. Examining the effect of the AC is

of particular importance because the AC is vital in monitoring external audit performance and

communication between the AC and the auditor is mandatory in determining the communicated

7

matters included in the EARs. According to the relevant auditing standards, there is a significant

overlap across jurisdictions with respect to the audit matters that should be communicated, not only

by definition but also by the approaches for determining those matters. Under the IAASB standard,

the disclosed KAMs should be determined from the matters communicated with those charged with

governance, particularly the AC (IAASB 2015). Similarly, the PCAOB standard notes that CAMs

are those “matters communicated or required to be communicated to the audit committee” (PCAOB

2017, 1). In addition, as discussed earlier, for financial periods ending on or after

September 30, 2013, the UK requires both external auditors and ACs for companies that report on

the application of the UK Corporate Governance Code, to report material matters in their respective

extended reports (FRC 2012, 2013a). Thus, the UK provides a unique setting for examining the

underlying interactions between external auditors and ACs. Although the UK reporting regulations

do not require matters disclosed in the extended reports of auditors and ACs to be entirely

consistent, a substantial degree of alignment is expected.3 By investigating the material matters

discussed in EARs and EACRs for a particular entity in a given year, the risk assessments of

auditors and ACs can be captured empirically. Because AC expertise has a significant effect on an

ACs’ accounting judgments and risk assessments of financial statements, I focus on AC expertise

to investigate the influence of ACs on EARs. Therefore, I examine the extent to which AC expertise

influences auditors’ risk disclosures in EARs, with particular focus on the differences between

matters communicated in EARs and EACRs.

3 Detailed discussions are provided in Chapter 4.

8

The third study examines whether changes in auditors’ reporting behavior under this new

reporting regime are related to changes in audit effort. Specifically, this study focuses on the year-

to-year modifications of EARs, because the lack of changes in auditors’ disclosures over time is

another concern that has been raised with regard to EAR adoption. If disclosures do not

meaningfully change over time, EARs may become less uninformative (PCAOB 2017). This topic

is worth investigating because, in any company, there are inevitable similarities across a number

of years’ EARs, and if auditors use standardized words to describe similar risks and responses over

time, EARs will become longer reports more boilerplate language (PwC [PricewaterhouseCoopers]

2015). Therefore, when describing KAMs that recur over periods, auditors should highlight aspects

specific to the entity for that financial period (IAASB 2015). In the third study, the extent to which

auditors modify the disclosures in EARs from year to year is measured using textual analysis and

communicated matters (KAMs). I further examine whether year-to-year changes are related to

audit effort, as reflected in audit fees.

The first three implementation years in the UK market, from September 30, 2013 to

September 30, 2016, are examined because, as noted earlier, the UK is one of the first jurisdictions

to implement EARs and its unique regulatory setting requires both auditors and ACs to disclose

material risks in their respective extended reports. The results found in the UK should be

informative for stakeholders in other jurisdictions, because the FRC’s approach is similar to that

of the EU and narrower than the approach of the PCAOB and IAASB (Simnett and Huggins 2014).

Under the FRC’s auditing standard, the concept of material risks during the sample period is

defined as “risks of material misstatement that had the greatest effect on the overall audit strategy;

the allocation of resources in the audit; and directing the efforts of the engagement team” (FRC

2013a, paragraph 19A). For EARs issued for accounting periods ending on or after June 18, 2017,

9

the FRC extends the concept of material risks to incorporate the EU and the IAASB requirements,

and risks are discussed under the heading “KAMs” (FRC 2016a). The FRC expects that the

incorporation of the international standards will not result in a significant change in risk

identifications, because KAMs and material risks are broadly equivalent (FRC 2016b). Therefore,

for convenience, I label material risks in EARs as KAMs in this thesis.

The remainder of the thesis is structured as follows: Chapter 2 reviews prior studies

concerned with EARs. Chapters 3, 4, and 5 present the thesis Studies 1, 2, and 3, respectively.

Chapter 6 concludes the thesis by reviewing the findings, discussing the limitations, and providing

recommendations for future research.

10

CHAPTER 2: Literature Review of EARs

Prior archival studies of EARs are focused on the UK implementation effect, using pre-post

analyses, and yield ambiguous results.

Gutierrez, Minutti-Meza, Tatum, and Vulcheva (2018) find that the implementation of

EARs had no impact on investors’ reaction, audit fees, and audit quality. Their study implements

a difference-in-difference research design that compares the mandatory adoption companies with

the London Stock Exchange (LSE) Alternative Investment Market (AIM) entities, for a four-year

period (including two years before and after the EAR effective date, September 30, 2013). Using

cumulative absolute abnormal returns and abnormal trading volume centered on the report filling

date as the proxy for investors’ reaction, Gutierrez et al. (2018) find non-significant result that the

market perceives the EAR as a more informative and useful reporting model. In addition, the study

does not find significant incremental audit costs in relation to the requirement of additional

disclosures in EARs. No supporting evidence has been found that audit quality, measured as the

absolute discretionary accruals, is improved by the adoption of EARs. Moreover, Gutierrez et al.

(2018) also report that the non-significant results are not affected by the variations of report length,

total number of KAMs, the inclusion of unique KAMs, and materiality threshold. However,

comparatively higher audit fees are found to be paid to companies with longer reports and more

KAMs included.

Reid, Carcello, Li, and Neal (2018) also investigates the implementation effect of EARs in

the UK. Similar to Gutierrez et al. (2018), it uses a difference-in-difference research design and

two years pre and post periods. By using both European and US companies as control groups, Reid

et al. (2018) documents that the adoption of EARs significantly improved financial reporting

11

quality, as measured by absolute abnormal accruals, the propensity to just meet or beat analyst

forecasts, and earnings response coefficients. In addition, the paper reports no significant difference

in the pre-post changes for those entities with more KAMs. Regarding the audit cost analysis, Reid

et al. (2018) report that the adoption of EARs does not result in a significant change on audit fees

or audit delay.

Lennox, Schmidt, and Thompson (2018) do not find any supporting evidence that investors

react to the KAM disclosures. The paper examines whether investors respond differently to entities

with more KAMs disclosed in their EARs, in which the cumulative abnormal returns, abnormal

trading volume, unsigned abnormal returns, and abnormal volatility are captured as the market

reaction proxies, during a one-year pre and post period. Lennox et al. (2018) further investigate

whether the market reacts differently to those more informative EARs and find that the non-

significant adoption effect is not changed whether companies have more KAMs that are at entity

or account-level, company or industry-specific, or expected or unexpected. Lennox et al. (2018)

categorize a KAM as an entity-level risk if it discusses doubts regarding the going-concern

assumption, management override of controls and accounting judgements, and as an account-level

risk if the discussion is associated with revenues, expenses, assets, liabilities, and the notes to the

financial statements. A KAM is also categorized as industry-specific if it has been disclosed in fifty

percent or more companies’ EARs in the relevant industry, and company-specific otherwise. The

number of unexpected KAMs is computed as the total number minus the expected number, which

is estimated by a prediction model. It find that the absence of any significant market reaction

persists after splitting the sample in relation to the strength of company’s information environment,

which is variously measured based on the firm size, the number of analysts following and large

12

shareholders, or the number of “uncertainty” words using the Loughran and McDonald (2011)

word lists.

Lennox et al. (2018) find that investors perceive the additional disclosures in EARs as

reliable, evidenced by significantly smaller coefficients on earnings per share and net assets per

share when more KAMs are included. However, similar results are obtained when using the prior

year’s data, suggesting that investors were informed about the content of the KAM disclosures

prior to the use of EARs. In addition, they do not find any significant difference for entities having

more “new” KAMs (where “new” means the KAM was not discussed in the entity’s prior annual

report, earnings announcement, or conference call transcript), more “recurring” or “nonrecurring”

KAMs (where “recurring” means the KAM is disclosed in the next year’s EAR), or containing

more “negative” words, computed as the number of positive minus negative words, scaled by total

words, using the Loughran and McDonald (2011) word lists.

Porumb, Karaibrahimoglu, Lobo, Hooghiemstra, and de Waard (2018) investigate the

usefulness of KAM disclosures for private debt holders. For a four-year period (two years pre and

post the EAR effective date), the paper reports that the implementation of EARs benefits the UK

companies through more favorable loan contracting terms in terms of lower interest spread

(premium) and higher loan maturity, compared to the US control companies. The paper also finds

that companies with more KAMs included in the EAR get more stringent lending terms, evidenced

by larger interest spread, lower loan maturity and greater covenant intensity, suggesting that banks

perceive companies with more KAMs as risker and incorporate this information into their lending

decision process. In addition, results in Porumb et al. (2018) imply that the debt market reacts to

specific types of KAMs. The spread is positively associated with the number of KAMs pertaining

to the valuation of assets (including valuation of investments, goodwill impairment, carrying value

13

of inventory, and etc.) and going concern, and the loan maturity is negatively associated with

accounting issue-related KAMs. Covenant intensity is positively associated with going concern-

related KAMs and negatively associated with asset-related KAMs.

The four EAR studies summarized above are largely concerned with user responses to the

EAR content. However, their examination of content is largely focused on the number or

characteristics of KAMs. Smith (2017), which conducts textual analysis on EARs, finds that EARs

are more readable than prior audit reports, using the Fog index as the readability measure, during

a two-year pre-post sample period. In addition, the study finds that auditors, exhibiting higher audit

quality, who are industry expertise, Big 4 auditors, and from larger offices, generate more readable

EARs compared to the others. Variations across audit firms are also found in the paper, in which

an average person needs one more year of formal education to read and comprehend the EARs

generated by Ernst & Young (EY). Using the Loughran and McDonald (2011) Negative, Positive,

and Uncertain word lists to capture tone, Smith (2017) finds that, compared to the pre-ISA reports,

higher frequency of negative and uncertain language is provided in EARs. Additionally, the paper

documents that analyst forecast dispersion is reduced after the adoption of EARs.

The pre-post analyses conducted in these prior studies are constrained by an implicit

assumption of homogeneous behavior among auditors, seemingly contrary to the regulators’

propositions that EARs will help financial statements users to distinguish between auditors with

respect to their reporting quality.

14

CHAPTER 3: Study 1: Extended Auditor’s Reports and Audit Quality: A

Textual Analysis

3.1 Introduction

This study examines whether transparency in auditor’s reports is positively related to the

underlying audit quality. Auditor’s reports have been standardized for the past many years and

extensively criticized because of the limited information they allow auditors to provide (e.g.,

Church et al. 2008). In response to such criticism, regulators, including the FRC, IAASB, EU, and

PCAOB, have adopted enhanced audit reporting standards that require auditors to disclose the most

significant audit matters in that year’s financial statement audit, using non-standardized language

(EU 2014; FRC 2013a; IAASB 2015; PCAOB 2017). These changes in audit reporting

requirements are intended to make audits more transparent by providing financial statement users

with information pertaining to the underlying audit work (e.g., FRC 2016a; PCAOB 2017).

However, the ability or willingness of the auditor to provide more engagement-specific disclosures

for a particular engagement may be a consequence of the extent to which they have acquired

sufficient knowledge of issues or their willingness to potentially reveal audit deficiencies for that

engagement. I examine this issue by investigating whether audit report transparency is a

consequence of the underlying audit quality.

As noted earlier, for reporting periods commencing on or after October 1, 2012, auditors of

UK companies that report on the application of the UK Corporate Governance Code are required

to implement the enhanced auditor reporting standard (FRC 2013a). The ISA (UK and Ireland) 700

requires auditors to disclose information on KAMs, materiality, and the scope of the audit.

Although the FRC’s regulations differ in detail from those of the IAASB, EU, and PCAOB,

15

communicating using non-standardized language is consistent to all of them (FRC 2013a, 2015a;

IAASB 2015; PCAOB 2017; Simnett and Huggins 2014). To enhance audit report transparency,

auditors are encouraged to provide contextual information regarding their approaches, judgments,

and findings. The use of tailored (granular) language, or the avoidance of standardized (generic)

wording, when discussing material risks of the audited entity is regarded as critical to achieving

the objectives of EARs (FRC 2013a; IAASB 2015; PCAOB 2017). However, when the FRC

evaluated auditors’ disclosures for each of the first two years of implementation of EARs, they

reported that a significant proportion of EAR disclosures seemed generic, becoming more tailored

in the second year (FRC 2015a, 2016a).4 It directed auditors to reduce their use of standardized

language further. Using the contents of EARs for the first three years of the UK’s application of

the EARs, I assess whether textual similarity in the EARs is negatively associated with audit

quality, as proxied by abnormal accruals, at the engagement level.

The textual similarities of the EARs are measured using the Vector Space Model (VSM).

To ensure that the engagement-level measures of textual similarities reflect audit firms’ choices at

the engagement level, rather than differences between audit firms, the textual similarity of a

specific EAR is computed as the mean of the similarity scores of that EAR’s text relative to each

of the other EARs issued by the same audit firm in the same year. This method is applied to both

4 The FRC states that their judgment is subjective and no definition of standardized wording is given from a technical

perspective (FRC 2015a, 2016a). In addition, because the analyses were only based on 153 and 278 EARs, their

findings might not be representative of the market as a whole.

16

the KAM sections and the full EARs.5 I then investigate how EAR similarity scores vary in relation

to audit quality.

The implications of textual similarities in EARs for audit quality are not straightforward.

On one hand, standardization across reports may arise from deficiencies in audit effort or

competence. For example, generic disclosures may be chosen intentionally by auditors to reduce

transparency when they have concerns regarding the detectability of poor audit performance. From

this perspective, a negative association between EAR similarity scores and audit quality is expected

because the standardized disclosures, making “unlike things look alike”, fail to convey desirable

entity-specific information. On the other hand, standardization across reports may arise from

comparability, where “like things should look alike” (FASB [Financial Accounting Standards

Board] 2010; Lang and Stice-Lawrence 2015)6. Comparability, which is defined by FASB as a

qualitative characteristic of disclosure that enables users to identify and understand similarities and

differences among items, might be enhanced by standardized disclosures. In that context, no

relation between EAR similarity scores and audit quality is expected. In this study, I first adjust the

measure of EAR similarity by taking into account the expected effect of standardized wording

based on industry comparability effects. Entities’ disclosures that deviate from those of the

5 I test KAM sections and full EARs separately because some sections of the EAR are likely to be standardized, such

as explanations of the respective responsibilities of directors and auditors, and matters on which auditors are required

to report by exception (e.g., FRC 2015a). 6 FASB (2010) states, “For information to be comparable, like things must look alike and different things must look

different. Comparability of financial information is not enhanced by making unlike things look alike any more than it

is enhanced by making like things look different” (19–20).

17

company’s peers, operating in the same industry, are more likely to be firm specific (Kravet and

Muslu 2013).

I then examine the relations between textual similarities of EARs and audit quality. Based

on the proposition in DeFond and Zhang (2014) that “audit quality is a continuous construct that

assures financial reporting quality, with high quality auditing providing greater assurance of high

quality financial reporting” (276), and consistent with the extant accounting and auditing literature,

I adopt the absolute value of cross-sectional abnormal accruals as the proxy for audit quality.7

Overall, I find that abnormal accruals are positively associated with EAR similarity scores,

and the relations are more significant for the similarity scores of KAM sub-sections, compared

with the similarity scores of full EARs. This is further supported by evidence that high abnormal

audit fees are negatively related to similarity scores of KAM sub-sections. The results are

consistent with the proposition that when the underlying audit quality is lower, there is a higher

level of standardized language in EARs. In addition, I find that while overall audit quality appears

to have improved following the implementation of EARs in the UK, negative relations between

EAR similarity scores and audit quality persist.

This study contributes to policy development and the audit research literature. First, the

study informs standard setters of the effect of the revised auditing standards and may influence

regulators in their monitoring of audit reports and their future guidance to auditors. Although the

processes for determining communicated matters differ across jurisdictions, because the FRC’s

7 I cannot test other commonly used audit quality proxies, such as qualified audit opinions, going concern opinions,

and restatements, because they too infrequent in the UK during the sample period.

18

approach is narrower than those of the PCAOB and IAASB, the UK experience should be

informative for stakeholders in jurisdictions adopting a broader approach.

Second, by examining the relations between audit quality and textual similarities of EARs,

this study contributes to the audit reporting literature and enriches the ongoing discussion regarding

EARs. Prior studies of EARs are mainly focused on examining overall EAR adoption effects by

conducting pre-post analyses, which implicitly assume homogeneous auditor behaviors (e.g.,

Gutierrez et al. 2018; Reid et al. 2018). I believe that the analysis of differences in auditors’

reporting behaviors at the engagement level provides a more nuanced understanding of the

association between audit quality and audit reporting.

Third, the extant literature analyzing company-related disclosures is growing rapidly to

encompass multiple textual resources, including annual reports, earnings announcements, analyst

reports, media articles, and regulatory documents (Young 2015). However, only a small number

of studies apply textual analysis to audit reports. In addition to extending the application of textual

analysis to audit reports, I augment established textual similarity methods by incorporating

adjustments for expected similarities that result from industry-based comparability effects in audit

reports. This methodological contribution may prove useful in other applications.

3.2 Literature Review and Hypothesis Development

3.2.1. Prior Studies Related to EARs

As noted in Chapter 2, the modest evidence as to whether the introduction of EARs affected

audit quality is mixed. Reid et al. (2018) find that the absolute value of abnormal accruals, the

likelihood of just meeting or beating analyst forecasts, and earnings response coefficients improved

19

following the implementation of EARs, but Gutierrez et al. (2018) do not find that EAR

implementation significantly affected the absolute value of abnormal accruals. As argued by the

two studies, their conflict findings may be due to research design differences.8 Reid et al. (2018)

calculate total accruals using the balance sheet method and employ both European and US

companies as control groups, while Gutierrez et al. (2018) compute total accruals as net income

minus cash flow from operations and use the non-adoption UK entities, the LSE AIM entities, as

a control group. Different from these two studies, I examine whether the auditor’s use of

standardized disclosures is associated with lower audit quality at the engagement level, during the

first three years of the EARs.

The non-significant equity market reactions to the EAR implementation reported in prior

studies might be a consequence of the existence of standardized disclosures in EARs (Gutierrez et

al. 2018; Lennox et al. 2018). As argued by Lennox et al. (2018), the information asymmetry

between investors and auditors is unlikely to be reduced if auditors use standardized language or

provide information that is not firm-specific. However, variations among auditors’ reporting

behavior have been found in prior research. Reid, Carcello, Li, and Neal (2015) find that, on

average, abnormal trading volumes around the annual report issuance increased by 13.5 per cent

following the implementation of EARs but the effect varied across audit firms. It was noted that

KPMG (EY) clients experienced the strongest (weakest) increase in abnormal trading volume while

8 In Section 5 of Gutierrez et al. (2018) and on page 5 of Reid et al. (2018), the two papers explain in detail how their

analyses are different from the other.

20

KPMG (EY) provided the most (least) entity-specific information among the Big 4 firms (Fisher

and Deans 2014, as cited by Reid et al. 2015).9 Smith (2017), a study involving a textual analysis

of EARs, finds that, on average, auditors with more industry expertise, from Big 4 accounting firms

and larger offices, produce more readable EARs. In addition, Smith (2017) finds that one additional

year of formal education is required to read EARs generated by EY.

As noted in Chapter 2, examination of the content of EARs in prior studies is focused on

the number of KAMs (Gutierrez et al. 2018; Lennox et al. 2018; Porumb et al., 2018), readability,

and tone (Smith 2017), using pre-post analyses. I extend this literature by investigating the

differentiation among EARs in relation to auditors’ standardized wording usage.

3.2.2. Prior Studies Related to Report Transparency

Prior studies examining the quality of corporate disclosures are mostly focused on

managers’ reporting behavior and do not consider audit reports. This includes a stream of studies

examining whether managers choose less transparent disclosures to reduce the detection of their

earnings management.

Lee, Petroni, and Shen (2006) find that companies with a history of strategically selling

available-for-sale securities to meet earning benchmarks are more likely to choose a less

9 Fisher and Deans (2014), as cited by Reid et al. (2015), is not publicly available. The Fisher and Deans (2014)

definition of entity-specific information and the basis for identifying differences among audit firms regarding entity-

specific information are not known.

21

transparent format when disclosing information. Managers avoid reporting comprehensive income

in a performance statement, which make investors easier to detect their selective sale of securities.

Lobo and Zhou (2001) find a negative relation between the financial reporting quality, as

proxied by discretionary accrual, and corporate disclosure quality, measured as, the Association

for Investment Management Research (AIMR) scores, an industry-specific analysts’ assessment of

the informativeness of a company’s disclosures. Higher disclosure quality reduces information

asymmetry between managers and shareholders, which, in turn, decreases the flexibility of earnings

management. The results imply that managers prepare less transparent statements when they are

more engaged in earnings management behavior.

In addition, Cassell, Myers, and Seidel (2015) find that the disclosure transparency in

relation to activity in valuation allowance reserve accounts is negatively associated with accruals-

based earnings management. It is argued that because greater disclosure transparency facilitates

financial statement users’ detection of earnings management, managers by providing more

transparent disclosures, show their confidence in company financial performance and are less likely

to engage in earning management.10

Collectively, such studies support the proposition that managers reduce report transparency

to impede detection of lower earnings quality.

10 The link between transparency and accounting manipulation is supported by experimental evidence that increased

transparency influences managers’ earnings management strategies (Hunton, Libby, and Mazza 2006).

22

Brown and Knechel (2016) compare the similarity of companies’ narrative disclosures in

10-K filings to those of other companies in the same industry and year audited by the same Big 4

audit firms. The study examines three disclosure elements: the company business description,

management discussion and analysis, and notes to the financial statements. Of these, only the last

are audited. They find that similarity is negatively related to income-increasing accruals but

positively related to the likelihood of accounting restatements. Because Brown and Knechel (2016)

examine managers’ disclosures, they cannot form strong inferences regarding auditor behavior.11

While their paper concludes that audit quality may be associated with managers’ disclosures in the

financial statements, the role of the auditor in enabling or promoting the identified similarities is

not in evidence.12

3.2.3. Hypothesis Development

Transparency has been defined in accounting as disclosures that “reveal the events,

transactions, judgments, and estimates underlying the financial statements and their implications”

(Pownall and Schipper 1999, 262). In this regard, reports that use more standardized language and

so convey less firm-specific or fiscal-period-specific information are associated with lower

disclosure transparency (e.g., Brown and Tucker 2011; Hoberg and Maksimovic 2015; Lang and

Stice-Lawrence 2015).

11 Relatedly, Smith (2017) compares the word dictionaries generated from EARs with those generated from

management disclosures and finds that auditors’ wording usage is quite different from that of managers. 12 Brown and Knechel (2016) find that companies with lower 10-K similarity scores are more likely to switch to an

auditor whose clients exhibit higher similarity to each other, suggesting managers may seek auditors that are less

concerned with transparency.

23

I argue that standardization or similarities across an auditor’s EARs may arise directly from

deficiencies in the auditor’s effort or competence. If an auditor pays insufficient attention to, or has

insufficient understanding of, entity-specific or fiscal-period-specific issues, the auditor is less able

to provide engagement-specific information in the EAR; in addition, this suggests a negative

relation between the extent of standardization (or similarities) in an auditor’s individual EARs and

the audit quality for individual engagements.

The prediction can be further supported by the prior discussions concerning managers’ use

of standardized wording. If preparers prefer less transparency to avoid revealing the quality of the

underlying performance, then lower transparency itself may signal lower quality when report

preparers are responsible for the reported performance as well (Cassell et al. 2015; Hunton et al.

2006; Tucker 2015). Applying the arguments to EARs, auditors are responsible for both the audit

report and audit quality, and transparency regarding audit work may facilitate the detection of poor

audit performance. Therefore, auditors seeking to avoid transparency for engagements that have

lower audit quality will prefer higher levels of standardization in the EARs for those engagements.

Taken together, the prior arguments lead to the following hypothesis:

HYPOTHESIS. Textual similarity in EARs is negatively related to audit quality.

The predicted association between textual similarity and audit quality might not be

observed if auditors strategically use seemingly tailored wording when preparing EARs to avoid

signaling performance deficiencies. If auditors expect that stakeholders value non-generic

disclosures in EARs, as evidenced by the feedback received from investors and other stakeholders

(FRC 2016a) and partially supported by Reid et al. (2015), an auditor may anticipate intended

users’ responses to the differential use of standardized language. Therefore, they may manipulate

24

the wording of EARs as a deliberate differentiation or obfuscation strategy. This could confound

the testing of the hypothesis, but prior research suggests the information disclosed in EARs is

regarded as reliable by financial statement users (Lennox et al. 2018).

3.3 Method

To test the hypothesis, I must measure generic disclosures in EARs. However, base

measures of textual similarities of EARs within the audit firm portfolio may be misleading and

thus confound the hypothesis tests, where textual similarities are consequences of similarities in

client-specific engagement factors; that is, if the similarities “make like things look alike”. For

example, because companies that are in the same industry and year and are audited by the same

accounting firm have more comparable accruals and earnings structures (Francis, Pinnuck, and

Watanabe 2014), and report more similar narrative disclosures in their financial statements (Brown

and Knechel 2016), it is reasonable that similarities will arise in their audit reports.13 To address

this concern, I adjust the base raw textual similarity scores for industry-based comparability effects,

as described below.

3.3.1. Measurement of EAR Similarity

I measure the textual similarity of the EARs in an audit firm’s client portfolio for both the

KAM sections and the full EARs. Each company’s base KAM or full EAR similarity score is the

13 Although the wording used by auditors is substantially different from the word dictionaries generated from

managers’ disclosures (Smith 2017), the same clustering effects may occur in EARs in a manner that confounds testing

of the hypothesis.

25

average of its similarity scores (for the KAM section or full EAR) in relation to all the KAMs

(EARs) of other companies audited by the same audit firm in the relevant year.

The base similarity scores are computed using the VSM, which was developed to compare

strings of text or documents (Salton and Buckley 1988) and has been used to compute similarity

scores in 10-K filings (e.g., Brown and Tucker 2011; Hoberg and Maksimovic 2015; Hoberg and

Phillips 2010, 2016; Merkley 2014) and the initial public offering (IPO) prospectuses (Hanley and

Hoberg 2010, 2012). In the VSM, the selected text is represented by an m-dimensional vector,

where m is the number of unique words after removing “stop words” in the defined set of

documents.14 Specifically, if m unique words are included in an audit firm’s portfolio of KAM

sections (or EARs), the vector for the KAM section (or full EAR) of company 𝑖 is represented as:

vi = (w1, w2, …, wm-1, wm)

(Model 3.1)

where w is the frequency of (each) word in the KAM section (or full EAR) for company i.

This method is also called cosine similarity because the similarity degree between the

vectors for EARs, vi, and vj is calculated as the cosine of the angle between the two vectors:

14 Consistent with prior papers, “stop words” are common words, such as “a,” “is,” “the,” and “will” (Li 2010;

Perterson et al. 2015). Words with a common stem are treated as the same word; for example, “auditing” and “audited”

are both stemmed to “audit,” and “audit,” “auditing,” and “audited” are treated as the same word (Lang and Stice-

Lawrence 2015; Merkley 2014). Some studies use complex text terms to generate similarity measures, rather than

single-word identifiers, but it has been shown that the single-word basis is preferable (e.g., Salton and Buckley 1988).

26

Similarityij = (vi×vj) / (ǁ𝑣iǁ × ǁ𝑣jǁ)

(Model 3.2)

where vi×vj yields the scalar product of vi and vj, and ‖𝑣𝑖‖ and ‖𝑣𝑗‖ represent the vector lengths.15

Similarityij is bounded within (0, 1) and it approaches 1 when the similarity of two KAM

sections (or two EARs) for companies i and j increases. I calculate Similarityij for each of the KAM

section (KAM Similarityij) and the full EAR (EAR Similarityij) for each pair of companies (i and j)

in an audit firm’s portfolio in a given year. For company i, I compute a BaseSIMKAM (and a

BaseSIMEAR) score as the average of its relevant Similarity scores in relation to all other (n-1)

companies in the same audit firm portfolio (of n companies) in the relevant year. For the KAM

similarity score for company i, this is expressed as:

BaseSIMKAMit = ∑jKAM Similarityij / (n-1)

(Model 3.3)

where n is the number of companies in the relevant audit firm’s portfolio in year t.

The same method is used to obtain BaseSIMEARit.

The BaseSIMKAM and BaseSIMEAR scores are then adjusted for industry-based

comparability effects, for the reasons discussed in the previous section. The adjustment applied to

15 The inherent normalization makes it unnecessary to further control for EAR length (see Hoberg and Phillips 2010,

2016).

27

each company’s BaseSIMKAM or BaseSIMEAR is calculated using the same procedure as

described by Models 3.1 through 3.3, except that the defined set of documents is the disclosures

for all companies in the same industry that are audited by the same audit firm in the relevant year.

Thus, company j must be in the same industry as company i.16 Based on Model 3.3, the adjustment

for industry-based comparability effects for the KAMs section of company i is:

Ind_SIMKAMit = ∑jKAM Similarityij / (m-1)

(Model 3.4)

where m is total number of companies in the industry of company i in the relevant audit firm’s

portfolio in year t.

The same method is used to obtain Ind_SIMEARit.

The Ind_SIMKAM and Ind_SIMEAR adjustments based on Model 3.4 can be obtained for

company i only where the audit firm audits another company in the same industry as company i in

the relevant year. I then use the following two methods to adjust the similarity scores for the

expected industry effects adjustments from Model 3.4, to obtain (1) the unexpected similarity score,

and (2) the adjusted similarity score:

1. To obtain the unexpected similarity score, I regress the base similarity scores (from Model

3.3) against the industry-based scores for each year (from Model 3.4). For the KAM scores,

this can be expressed thus:

16 This is similar to the grouping method used in Brown and Knechel (2016) to compare the similarity of the narrative

content in 10-K filings.

28

BaseSIMKAMit = b1Ind_SIMKAMit + it

(Model 3.5)

and save the residual it as the unexpected similarity score for the KAMs section for company i in

year t, SIM_KAMRESit. The same method is used to obtain unexpected similarity scores for the full

EARs, SIM_EARRES.

2. To obtain the adjusted similarity score, I deduct the industry-based score Ind_SIMKAM

from the base similarity scores BaseSIMKAM to obtain SIM_KAMDIF, thus:

SIM_KAMDIFit = BaseSIMKAMit – Ind_SIMKAMit

(Model 3.6)

The same method is used to obtain adjusted similarity scores for the full EARs,

SIM_EARDIF.

3.3.2. Models for Testing the Hypothesis

To test the hypothesis that KAM (EAR) similarity is negatively related to audit quality, I

estimate Model 3.7 as follows:

SIMit = β0+ β1ABSACCit + β2SIZEit+ β3ANALYSTCOVit+ β4LIST_USit+ β5MAit+ β6SEOit+

β7MTBit+ β8LOSSit+ β9PROCOSTit+ β10EARNVOLit+ β11LNSUBit+ β12CATAit+

β13LEVERAGEit+ β14FIRSTit+ β15LONDONit+ Σβ𝑗industry+ Σβ𝑘year+ it

(Model 3.7)

where SIMit represents each of SIM_KAMRES, SIM_EARRES, SIM_KAMDIF, and SIM_EARDIF,

and the variable of interest ABSACCit is the absolute value of abnormal accruals.

29

Consistent with prior studies, I obtain ABSACCit by estimating the cross-sectional modified

Jones model (Dechow, Sloan, and Sweeney 1995) within two-digit ICB industry groups for each

year, with a minimum of 15 observations per two-digit ICB industry (Carcello and Li 2013; Reid

et al. 2018).17 I then match each firm-year observation with another firm from the same two-digit

ICB industry code and year with the closet ROA and ABSACCit is the absolute value of the

difference (Kothari, Leone, and Wasley 2005). The hypothesis predicts a significant positive

coefficient on ABSACCit.18

Model 3.7 controls for client engagement factors that potentially influence auditors’ use of

standardized language in EARs. Generally, I posit that if a factor increases (decreases) audit effort,

it will be negatively (positively) associated with similarity scores.19 I include SIZE (the log of total

assets) but I do not predict the sign of the coefficient. If large firms are more stable and thus have

relatively fewer risks (Campbell, Chen, Dhaliwal, Lu, and Steele 2014), then a positive relation is

implied, but if size-related political costs induce larger companies and their auditors to provide

higher-quality disclosures, then a negative relation is implied (Watts and Zimmerman 1990).

Consistent with the political costs argument, I predict that auditors use less-standardized language

in EARs for more visible companies. Therefore, similarity scores are expected to be lower for

companies followed by more analysts (ANALYSTCOV) or cross-listed in the US (LIST_US).

Because entity events can also increase exposure or directly induce auditors to disclose more entity-

17 The regression model is estimated as follows: ABSACC = α0TOTAL_ACC – [α1 (1/TA) +α2 (𝑆𝐴𝐿𝐸𝑆 −𝑅𝐸𝐶)

+α3PPE]. Variable definitions are provided in Appendix 1. 18 For robustness, I calculate ABSACC as well, using the performance-adjusted approach, and the results (not tabulated)

are qualitatively similar to the main results. 19 For convenience, all variables in the model are also summarized in Appendix 1.

30

specific information, I include indicator variables for companies engaged in mergers-and-

acquisitions (MA) and seasoned equity offerings (SEO) in the current year. More entity-specific

information is also expected for companies that are financially distressed (LOSS), have a higher

market-to-book ratio (MTB), significant earnings volatility (EARNVOL), more complex operations

(LNSUB), current assets-related risks (CATA), or debt-related risks (LEVERAGE). I control for

entity-specific proprietary costs using PROCOST (research and development costs divided by

beginning total assets), a proxy for the product market competition, but I do not predict the

coefficient sign. While some prior studies find that companies are more likely to provide entity-

specific information if they are in competitive industries, others report a negative relation between

competition-related proprietary costs and the specific disclosures (see Beyer, Cohen, Lys, and

Walther 2010 for a review of related literature). I include two auditor-related indicator variables

that control for cases where the EAR is the first report issued by the audit firm for that client

(FIRST) and where the EAR is signed by an auditor who is located in the London office

(LONDON), and I control for year and industry fixed effects.

3.3.3. Sample Selection

In the UK, the EAR is mandatory for that report on the application of the UK Corporate

Governance Code for fiscal years ending on or after September 30, 2013 (FRC 2013a). I obtain all

of the available annual reports of relevant companies with fiscal years ending between

September 30, 2013 and September 30, 2016, yielding 2,352 EARs that I use here to calculate base

similarity scores. Because my adjustment for industry-based comparability effects excludes cases

that are the only observation in the relevant auditor-industry-year portfolio, there are 2,273

observations with all necessary data to calculate SIM_KAMRES, SIM_EARRES, SIM_KAMDIF,

and SIM_EARDIF. The observations are then merged with the necessary financial data from

31

DataStream (losing 204 unmatched cases) and during this process, I exclude 1,036 observations in

financial-related industries (for which the abnormal accruals model is not suited) and 210

observations missing the data necessary to calculate abnormal accruals or for my control

variables.20 The final sample for the main analysis is comprised of 813 firm-year observations, as

described in Table 3.1.21

Table 3.1 Sample Selection

Firm-year

Observations

Total available EARs (used to calculate the unadjusted similarity scores) 2,352

Less: EARs with insufficient observations to calculate the industry-based

comparability effects (79)

Total available EARs to calculate the similarity scores (adjusted) 2,273

Less: Missing observations when merging with DataStream (204)

Less: Financial firm-year observations (including investment funds) (1,036)

Less: Observations with missing data to compute abnormal accruals (92)

Less: Observations with missing data to compute control variables (128)

Final Sample 813

20 Three early adopters are included in the sample and results remain unchanged after removing them from the analyses. 21 The final sample size is reasonable, compared with other EAR studies examining abnormal accruals in the UK. For

example, Gutierrez et al. (2018) and Reid et al. (2018) report 872 and 1,088 firm-year observations, respectively, for

a four-year period.

32

3.4 Results

3.4.1. Descriptive Statistics

Descriptive statistics for SIM_KAMRES, SIM_EARRES, SIM_KAMDIF, and SIM_EARDIF

for the pooled sample and by year are presented in Panel A of Table 3.2. SIM_KAMRES

(SIM_KAMDIF) necessarily exhibits more variability than SIM_EARRES (SIM_EARDIF). The

mean values of SIM_KAMRES and SIM_EARRES (obtained using the full sample of 2,273

observations) are close to zero and the mean values of SIM_KAMDIF and SIM_EARDIF are

negative.22 The latter is expected because the similarity scores estimated within industry-auditor-

year are consistently larger than the ones estimated within the auditor-year groups, supporting the

argument that narrative disclosures of an entity are likely to be more similar to those of other firms

from the same industry.

Table 3.2 Descriptive Statistics

Panel A

Descriptive statistics of similarity variables for full samples and year distribution

Total N

Year 1

2013–14

Year 2

2014–15

Year 3

2015–16

Variable N Mean SD N Mean SD N Mean SD N Mean SD

SIM_KAMRES 2273 0.005 0.075 721 0.003 0.073 780 0.006 0.075 772 0.007 0.078

SIM_EARRES 2273 0.000 0.032 721 0.000 0.030 780 0.000 0.033 772 0.001 0.034

SIM_KAMDIF 2273 -0.117 0.086 721 -0.111 0.080 780 -0.119 0.087 772 -0.120 0.092

SIM_EARDIF 2273 -0.042 0.033 721 -0.036 0.030 780 -0.045 0.033 772 -0.046 0.035

Variable definitions are provided in Appendix 1.

22 I do not tabulate audit-firm level mean similarity scores but, generally, the large audit firms have smaller mean

scores than other audit firms, seemingly consistent with the prevailing view that large firms deliver higher audit quality

than small audit firms (DeFond and Zhang 2014).

33

Panel B

Descriptive statistics for all variables

Variable N Mean SD p25 p50 p75

SIM_KAMRES 813 0.025 0.052 -0.003 0.036 0.057

SIM_EARRES 813 0.008 0.025 -0.006 0.011 0.023

SIM_KAMDIF 813 -0.089 0.057 -0.117 -0.081 -0.055

SIM_EARDIF 813 -0.034 0.026 -0.047 -0.030 -0.020

ABSACC 813 0.058 0.064 0.020 0.040 0.076

SIZE 813 13.854 1.705 12.632 13.717 14.837

ANALYSTCOV 813 11.587 7.824 5.000 10.000 17.000

LIST_US 813 0.026 0.159 0.000 0.000 0.000

MA 813 0.144 0.351 0.000 0.000 0.000

SEO 813 0.123 0.329 0.000 0.000 0.000

MTB 813 3.783 4.856 1.388 2.580 4.499

LOSS 813 0.173 0.379 0.000 0.000 0.000

PROCOST 813 0.011 0.020 0.000 0.000 0.009

EARNVOL 813 0.128 0.297 0.039 0.073 0.135

LNSUB 813 3.708 1.126 3.099 3.609 4.792

CATA 813 0.422 0.213 0.263 0.395 0.558

LEVERAGE 813 0.212 0.174 0.072 0.197 0.301

FIRST 813 0.379 0.485 0.000 0.000 1.000

LONDON 813 0.587 0.493 0.000 1.000 1.000

Variable definitions are provided in Appendix 1.

Descriptive statistics for the variables used in Model 3.7 are presented in Panel B of Table

3.2.23 Compared with the initial sample in Panel A, the mean values of the similarity variables for

the reduced sample in Panel B are larger and standard deviations are smaller, consistent with size

or survivorship bias arising from missing data, as encountered in most studies that use abnormal

accruals models. Overall, the distributions of variables appear consistent with prior studies of UK

23 To mitigate the effect of outliers, all continuous financial variables are winsorized at 0.01 (Francis and Yu 2009).

As a robustness check, I estimate the models without winsorizing as well, and do not observe any significant

differences in the regression results.

34

firms. The mean absolute abnormal accruals (ABSACC) is 0.058, consistent with those in two EAR-

related papers: 0.050 in Gutierrez et al. (2018) and 0.057 in Reid et al. (2018). The average

company is large (total assets of around £1 billion), with about 11 analysts following

(ANALYSTCOV). Around 3 per cent of cases are for entities that are cross-listed in the US

(LIST_US), 14 per cent for those with merger-and-acquisitions (MA) and 12 per cent with seasoned

equity offerings (SEO). Only 17 per cent of firm-years have a reported loss (LOSS). Other

performance-related factors (EARNVOL, LNSUB, CATA, and LEVERAGE) are similarly

unremarkable: 38 per cent of the tested EARs are the first EAR the auditor prepared for that

particular client (FIRST) and 59 per cent have a signing partner located in a London office

(LONDON). These descriptions are consistent with those papers examining the effect of EARs

(e.g., Lennox et al. 2018; Reid et al. 2018).

3.4.1.1.Correlations between the Regression Variables

Pearson correlations among all regression variables are reported in Table 3.3. For the

similarity measures, there is a very high correlation between each residual measure and the

corresponding difference measure; the correlation coefficient for SIM_KAMRES and

SIM_KAMDIF is 0.97, and the coefficient for SIM_EARRES and SIM_EARDIF is 0.98. Regarding

the correlations among the test variable and control variables, while no correlations appear

particularly large, there are sufficient significant associations to raise multicollinearity concerns;

this is assessed for each regression using variance inflation factors (VIFs), which do not reveal any

evidence of multicollinearity.

35

Table 3.3 Pearson Correlation Matrix

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1 SIM_KAMRES 1.0000

2 SIM_EARRES 0.6655* 1.0000

3 SIM_KAMDIF 0.9740* 0.6782* 1.0000

4 SIM_EARDIF 0.6406* 0.9815* 0.6732* 1.0000

5 ABSACC 0.0508 0.0039 0.0444 -0.0033 1.0000

6 SIZE 0.0062 0.0458 -0.0197 0.0524 -0.0865* 1.0000

7 ANALYSTCOV -0.0408 0.0024 -0.0510 0.0178 -0.0575 0.7681* 1.0000

8 LIST_US -0.0037 0.0035 -0.0115 0.0049 0.0479 0.2326* 0.1980* 1.0000

9 MA -0.0805* -0.0677 -0.0974* -0.0734* 0.0227 -0.0710* -0.1020* -0.0668 1.0000

10 SEO -0.0824* -0.0862* -0.0625 -0.0724* 0.0890* -0.0860* -0.0865* -0.0610 0.3266* 1.0000

11 MTB -0.0321 -0.0095 -0.0067 -0.0040 -0.0024 -0.1534* 0.0518 -0.0318 -0.0347 -0.0153 1.0000

12 LOSS -0.0347 0.0499 -0.0517 0.0408 0.1169* -0.0148 -0.0385 0.0483 -0.0305 0.0757* -0.0635 1.0000

13 PROCOST -0.0211 -0.2067* -0.0250 -0.2068* 0.1447* -0.1459* -0.0269 0.0050 -0.0159 0.0060 0.0309 -0.0144 1.0000

14 EARNVOL -0.0178 0.0044 -0.0037 0.0116 0.0187 -0.0882* -0.0414 -0.0408 0.0178 0.0054 0.1834* 0.0377 -0.0406 1.0000

15 LNSUB 0.0025 -0.1090* -0.0366 -0.1100* -0.0357 0.1097* 0.1332* -0.0616 -0.0696* -0.0512 -0.0661 0.0389 0.1489* 0.0195 1.0000

16 CATA -0.0782* -0.0658 -0.0528 -0.0597 0.0325 -0.3208* -0.2353* -0.0034 -0.1226* -0.0771* 0.1680* -0.0712* 0.0862* 0.1410* -0.0997* 1.0000

17 LEVERAGE 0.1047* 0.1137* 0.0804* 0.1029* -0.0206 0.2799* 0.1366* 0.0318 -0.0444 -0.0114 -0.1035* 0.1750* -0.1865* -0.0193 0.0748* -0.3589* 1.0000

18 FIRST -0.0356 0.0038 0.0317 0.1601* -0.0358 -0.0049 0.0397 0.0007 -0.0023 0.0318 0.0051 -0.0698* -0.0013 0.0102 0.0000 0.0242 -0.0340 1.0000

19 LONDON -0.1037* -0.0384 -0.1290* -0.0379 0.0212 0.3904* 0.3837* 0.0894* 0.0096 0.0177 -0.0932* 0.0876* -0.1031* -0.1220* 0.1350* -0.1350* 0.0433 -0.0036 1.0000

Variable definitions are provided in Appendix 1.

36

3.4.2. Regression Results

I estimate Model 3.7 with each of SIM_KAMRES, SIM_EARRES, SIM_KAMDIF, and

SIM_EARDIF as the dependent variables, as reported in Table 3.4. Each measure of textual

similarity is significantly positively related to abnormal accruals, supporting the hypothesis. The

test results are stronger for the regressions of the KAM similarity measures (SIM_KAMRES,

SIM_KAMDIF), for which the coefficients for ABSACC are significant at p<0.05; the coefficients

for ABSACC in relation to EAR similarity measures are significant at p<0.10.

With respect to the control variables, the negative effects for MA, SEO and LONDON

(significant across all the regressions) are consistent with the proposition that auditors are more

likely to provide more entity-specific information for companies engaged in mergers-and-

acquisitions, seasoned equity offerings in the current year, or where the audit partner is located in

a London office. SIZE is significantly positive in relation to the KAM similarities, consistent with

the argument that because large companies tend to be more stable, auditors are less likely to provide

client-specific risk-related disclosures. There are significant negative effects for PROCOST and

LNSUB in relation to the EAR similarities, and for CATA in relation to the KAM similarities,

consistent with them reflecting engagement characteristics that increase entity-specific disclosures

in the EAR.

37

Table 3.4 Regression Results for Analyses

SIM_KAMRES SIM_EARRES SIM_KAMDIF SIM_EARDIF

(1) (2) (3) (4)

ABSACC 0.093** 0.032* 0.102** 0.033*

(2.52) (1.79) (2.51) (1.81)

SIZE 0.003* 0.001 0.003* 0.001

(1.65) (1.36) (1.27) (1.54)

ANALYSTCOV -0.001 -0.000 -0.001 -0.000

(-1.48) (-1.28) (-1.30) (-1.28)

LIST_US 0.001 -0.002 0.000 -0.003

(0.12) (-0.33) (0.00) (-0.51)

MA -0.011** -0.001* -0.015** -0.002

(-2.04) (-0.54) (-2.57) (-0.77)

SEO -0.011* -0.007*** -0.008* -0.007**

(-1.93) (-2.64) (-1.33) (-2.44)

MTB 0.000 0.000 0.001 0.000

(1.13) (1.00) (1.34) (1.00)

LOSS 0.003 0.004 0.001 0.004

(0.57) (1.48) (0.14) (1.43)

PROCOST -0.003 -0.145** -0.011 -0.135**

(-0.03) (-2.48) (-0.08) (-2.30)

EARNVOL 0.003 -0.001 0.004 -0.001

(0.50) (-0.26) (0.58) (-0.23)

LNSUB 0.006 -0.003* 0.003 -0.004*

(1.51) (-1.63) (0.66) (-1.75)

CATA -0.030*** 0.003 -0.026** 0.004

(-2.88) (0.64) (-2.25) (0.69)

LEVERAGE 0.009 0.004 0.006 0.004

(0.81) (0.70) (0.49) (0.66)

FIRST -0.000 0.004 0.003 0.004

(-0.00) (0.97) (0.38) (1.01)

LONDON -0.008** -0.003* -0.011** -0.003*

(-2.03) (-1.64) (-2.44) (-1.73)

Constant -0.036 -0.001 -0.137*** -0.039***

(-1.35) (-0.10) (-4.67) (-2.94)

Industry, Year FE Included Included Included Included

N 813 813 813 813

Adjusted R2 0.103 0.115 0.103 0.148 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

38

3.5 Robustness Tests

3.5.1. Analyses Using Alternative Measures of EAR Similarity

The traditional VSM approach used in the main analysis has been widely used in research

of textual analysis but has been criticized for treating all included words equally (Peterson,

Schmardebeck, and Wilks 2015) and for focusing on individual words. To address the concern that

all words are weighted equally under VSM, I follow Brown and Tucker (2011) and Hoberg and

Phillips (2016), and modify the calculation of similarity scores in the VSM approach by weighting

words using the term frequency–inverse document frequency (TF-IDF) approach. For word m in

EAR i, the term frequency is wmi (the same as wm in Model 3.1) and inverse document frequencies

are computed as log(N/nm), where N is the number of total documents included in the sample and

n is the number of documents containing word m. Thus, the TF-IDF weighting is [wmi log(N/nm)],

which gives larger weights to words used less frequently in the collection of EARs, and lower

weights to common words, with 0 weight for a word appearing in every document where the IDF

is log(1/1). The resultant variables are BaseSIMKAM_IDF and BaseSIMEAR_IDF.

To address the concern that VSM focuses on individual words, I follow prior accounting

studies (e.g., Lang and Stice-Lawrence 2015, Nelson and Pritchard 2007) by measuring similarity

using the n-gram approach. A potential advantage of n-grams over VSM is that n-grams measure

the similarity of n-word phrases across documents. However, this means n-gram is focused on the

occurrence of a term and is influenced by document length, while VSM uses term frequencies and

is independent of document length. The n-gram approach converts each document into a set of

overlapping n-word phrases and document similarity is measured by examining the extent to which

phrases appear in both document sets. I follow Nelson and Pritchard (2007) and use trigrams, which

39

are ordered three-word phrases within a sentence.24 I convert each KAM section and full EAR into

sets of overlapping trigrams and compute the similarity score between the reports of companies i

and j by dividing the intersection of the two sets (one for each company) by the union of the two

sets of trigrams, thus:

BaseSimilarityij = |S(i)∩S(j)| / |S(i)∪S(j)|

(Model 3.8)

where 𝑆(𝑖) and 𝑆(𝑗) represent the sets of trigrams for KAM(i) and KAM(j) or EAR(i) and EAR(j),

respectively.

The resulting variables are BaseSIMKAM_Tri and BaseSIMEAR_Tri.

The TF-IDF and Trigram measures are then adjusted for the industry comparability effects

using both the residual method (yielding SIM_KAMRES_IDF, SIM_KAMRES_Tri,

SIM_EARRES_IDF, and SIM_EARRES_Tri) and the differencing method (yielding

SIM_KAMDIF_IDF, SIM_KAMDIF_Tri, SIM_EARDIF_IDF, and SIM_EARDIF_Tri). 25 The

regression results using TF-IDF and Trigram measures are reported in Panels A and B of Appendix

2, respectively. The textual similarities in KAM sections (Columns 1 and 3) and full EARs

(Columns 2 and 4) are significantly associated with higher abnormal accruals using the TF-IDF

24 There is no established benchmark selecting the number of words in a phrase. Longer phrases (such as four-word

phrases) are used in some prior studies, but they are typically applied to much longer documents, such as annual reports

(Lang and Stice-Lawrence 2015). 25 I observe that the TF-IDF weighted measures are similar to the original VSM measures in terms of scale and

variability, while the Trigram measures exhibit much more variability. I also find that the TF-IDF weighted measures

are more highly correlated with the original VSM measures than are the Trigram measures; for example, the correlation

coefficient for SIM_KAMRES and SIM_KAMRES_IDF is 0.94, while the correlation coefficient between

SIM_KAMRES and SIM_KAMRES_Tri is 0.80.

40

measures (Columns 1 and 2), but not using the Trigram measures (Columns 3 and 4). Consistent

with the main analysis, the results are stronger for the KAM-related similarity measures, where the

negative coefficient for ABSACC is significant at p<0.05, and the regressions using difference-

based adjusted measures generally exhibit more explanatory power than the regressions using

residual-based adjusted measures.

3.5.2. Analyses Using the Unadjusted Similarity Scores

I repeat all of the analyses without adjusting for the industry effects on similarity scores. I

estimate Model 3.7 using the unadjusted similarity scores as the dependent variable: BaseSIMKAM

(from Model 3.3) and its related BaseSIMKAM_IDF and BaseSIMKAM_Tri (from Model 3.8), and

BaseSIMEAR (from Model 3.3) and its related BaseSIMEAR_IDF and BaseSIMEAR_Tri (from

Model 3.8). The regression results for KAM section and full reports are reported in Panels A and

B of Appendix 3, respectively. The coefficients on ABSACC are not significant for any of the base

similarity measurements. On this basis, I conclude that failing to adjust for the industry-based

comparability effects obscures significant associations between the similarity of EARs and the

audit quality.

3.5.3. Analyses Using Income-Increasing Abnormal Accruals

I next examine whether income-increasing abnormal accruals are larger for firms with

higher EAR similarity scores. As reported in Appendix 4, I obtain significant positive coefficients

on income-increasing abnormal accruals for all of the similarity measures. Consistent with the

results of the main analyses, the results are stronger for the regressions of the KAM similarity

measures (SIM_KAMRES, SIM_KAMDIF), for which the coefficients for income-increasing

abnormal accruals are significant at p<0.01. The results are consistent with the proposition that

41

when an entity successfully engages in income-increasing earnings management, implying a lower-

quality audit, the auditor provides less transparent disclosures.

3.5.4. Analyses Using Abnormal Audit Fees

Gutierrez et al. (2018) find that auditors issuing longer EARs or identifying more KAMs

receive higher audit fees, suggesting a positive association between auditors’ disclosures and audit

effort. This is consistent with the argument because if additional audit effort improves audit quality,

I expect it will reduce EAR or KAM similarity. To investigate further whether higher audit effort

leads to more entity-specific disclosures in EARs, I next examine whether the EARs similarity

score is negatively associated with abnormal audit fees, a proxy for audit effort. Using hand-

collected audit fee data from annual reports, I calculate abnormal audit fees as the residual of the

audit fee model, which is stated as follows:

LNAFEE = β0+β1SIZE+β2ABSACC+β3ADJSALES+β4LOSS+β5BIG4+β6INVREC+β7BUSY+

β8DTRATIO+ β9LNSUB + Σβ𝑗industry +

(Model 3.9)

where LNAFEE is the natural logarithm of audit fees; ADJSALES is total sales divided by total

assets; BIG4 is a dummy variable that equals 1 if the auditor is from BIG4, and 0 otherwise;

INVREC is inventory plus accounts receivable, divided by total assets; BUSY is a dummy variable

that equals 1 if the company’s fiscal year end is between December 1 to March 31, and 0 otherwise;

and DTRATIO is total debt divided by total assets.

Other variables are as per Model 3.7 and are summarized in Appendix 1. Prior papers find

that auditors are paid more for companies with greater discretionary accruals (Gul, Chen, and Tsui

42

2003), and to avoid the potential confounding effect, I include ABSACC in the audit fee model.26 I

use the residual to construct two indicator variables: HIGH_ABAFEE equals 1 if the abnormal audit

fee is above the highest tertile (i.e., top third); and LOW_ABAFEE equals 1 if the abnormal audit

fee is below the lowest tertile (i.e., bottom third), and 0 otherwise. Model 3.7 is re-estimated using

LOW_ABAFEE and HIGH_ABAFEE in place of ABSACC, and the results are reported in Appendix

5.

The negative coefficients on HIGH_ABAFEE are significant in relation to KAM similarity

scores (SIM_KAMRES and SIM_KAMDIF) but are not significant in relation to the overall EAR

similarity scores. None of the coefficients for LOW_ABAFEE is significant. These results are

consistent with the proposition that higher audit effort, implying a higher-quality audit, results in

less-standardized disclosures – at least in the KAM sub-sections.

3.6 Additional Analysis

The results indicate that, on average, the auditor uses more standardized disclosures in the

EAR when an entity’s relevant financial statements exhibit lower audit quality. However, Reid et

al. (2018) seem to contradict the conclusion by arguing that the actual disclosures included in EARs

are not related to audit quality, based on the absence of significant pre-post differences in audit

quality in relation to the number of KAMs. I revisit the Reid et al. (2018) analysis to investigate

whether the seemingly different results can be reconciled. To benchmark the further analysis, I first

26 Including ABSACC in the audit fee model means that the residual cannot proxy for abnormal accruals and thus

ensures that the abnormal audit fee tests complement the abnormal accruals tests.

43

re-estimate the Reid et al. (2018) model, following their method (with and without TOTAL_KAM),

using the sample. I then modify the Reid et al. (2018) model by replacing the number of KAMs

with each of the measures of textual similarity for the KAMs section, (SIM_KAMRES and,

SIM_KAMDIF), denoted by SIMit, as follows:

ABSACCit = β0+ β1POSTit+ β2 [TOTAL_KAMit or SIMit]+ β3SIZEit+ β4ROAit+ β5LOSSit+ β6MBit+

β7LEVERAGEit+ β8PRIOR_ACCit+ β9CFOit+ β10SALESVOLit+ β11BIG4it+

Σβ𝑗industry + it

(Model 3.10)

where POST equals 1 if the fiscal year is the first three years following the EAR requirements, and

0 otherwise; TOTAL_KAM is the total number of KAMs disclosed in the EAR and equals 0 for the

pre-EAR period; ROA is the net income before extraordinary items divided by total assets; MTB is

the market value divided by the book value; PRIOR_ACC is the current accruals in the previous

year, measured as net income before extraordinary items plus depreciation and amortization less

operating cash flows, scaled by total assets; CFO is the cash flow from operations divided by total

assets; and SALESVOL is the standard deviation of annual sales. Other variables are as per Model

3.7 and are summarized in Appendix 1.

The results are reported in Table 3.5. The results for the original model using the sample

are consistent with Reid et al. (2018); as shown in Columns 1 (without TOTAL_KAM) and 2 (with

TOTAL_KAM), I obtain significantly negative coefficients on POST, suggesting that the

implementation of EARs leads to an overall improvement in audit quality, and no significant effect

for TOTAL_KAM. However, when I replace TOTAL_KAM with the measures of textual similarity

in KAM sections, as shown in Columns 3 and 4, I find that the coefficients are significantly positive

44

Table 3.5 Regression Results for Pre-Post EAR Analysis

ABSACC ABSACC ABSACC ABSACC

(1) (2) (3) (4)

POST -0.006*** -0.008* -0.005** -0.011***

(-2.65) (-1.75) (-2.03) (-3.23)

TOTAL_KAM - -0.001 - -

(-0.56)

SIM_KAMRES - - 0.054* -

(1.82)

SIM_KAM_DIF - - - 0.054**

(2.00)

SIZE -0.001* -0.001* -0.002** -0.002**

(-1.88) (-1.90) (-2.32) (-2.19)

ROA -0.050*** -0.057*** -0.066*** -0.050***

(-2.72) (-3.04) (-3.41) (-2.99)

LOSS 0.006 0.005 0.005 0.008*

(1.36) (1.26) (1.14) (1.94)

MB 0.000 0.001 0.001 0.002***

(0.08) (1.19) (1.24) (2.73)

LEVERAGE 0.000 -0.002 -0.002 -0.000

(0.03) (-0.29) (-0.22) (-0.07)

PRIOR_ACC 0.002 0.002 0.005 0.002

(0.14) (0.12) (0.34) (0.14)

CFO 0.006 -0.011 -0.009 -0.004

(0.39) (-0.67) (-0.54) (-0.24)

SALESVOL 0.005 0.005 0.004 0.004

(1.28) (1.15) (1.13) (1.12)

BIG4 0.005 0.005 -0.002 -0.003

(1.05) (0.99) (-0.31) (-0.35)

Constant 0.044*** 0.048*** 0.060*** 0.062***

(3.73) (4.04) (4.38) (4.35)

Industry FE Included Included Included Included

N 1,518 1,489 1,438 1,438

Adjusted R2 0.085 0.089 0.097 0.096 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

45

for both SIM_KAMRES and SIM_KAMDIF. Therefore, the results suggest that auditors’ actual

entity-specific disclosures are associated with the extent to which entities benefit from the revised

auditing standards in terms of audit quality. After the implementation of EARs, beyond the overall

improvement in audit quality, there is positive association between the standardized wording usage

in EARs and audit quality, using the Reid et al. (2018) approach, consistent with the results for the

main analysis.

3.7 Conclusions

Regulators have responded to concerns about the limited information content in the

traditional auditor’s report by developing new audit reporting standards that require disclosures of

the most significant audit matters in that year’s financial statement audit, using non-standardized

wording (EU 2014; FRC 2013a; IAASB 2015; PCAOB 2017). These EARs are intended to

increase transparency of the underlying audit work (FRC 2016a). This study investigates, whether

during the UK’s first three years of EARs implementation, the use of standardized disclosures,

implied by textual similarities, is associated with lower audit quality on individual engagement. I

measure the textual similarities in EARs for both the KAM sub-sections and full EARs within an

audit firm client portfolio, and test whether similarity scores are positively associated with

abnormal accruals at the engagement level.

I find that textual similarities in the EARs (within each auditor’s engagement portfolio in a

given year) are positively related to the absolute abnormal accruals, and the relations are stronger

for the KAM sections, compared to the full EARs. In addition, I find that textual similarities in

KAM sections are negatively related to abnormal audit fees. Overall, the results support the

46

proposition that, on average, differences in the level of standardized disclosures in EARs are

negatively associated with differences in underlying audit quality at the engagement level.

It is beyond the scope of this study to determine whether these observed effects are a

consequence of differences in auditor effort or competence on particular engagements, or auditors

intentionally reducing transparency when audit quality is lower, or yet-to-be-discovered factors. I

suggest this is an avenue for future research.

Owing to the prevailing UK circumstances, I largely rely on abnormal accruals to proxy for

audit quality. Although abnormal accruals is the most commonly used audit quality proxy, whether

these results are robust to other financial statement-based audit quality measures can be resolved

only when additional measures become available. Nonetheless, the results and method may be of

broader relevance – but with some caveats. The UK requirements are largely contained within the

broader specifications used in jurisdictions following the IAASB and the PCAOB approaches, and

the results should inform regulators in other jurisdictions considering the adoption of EARs and

better inform future policy development and monitoring in relation to audit reports. However,

jurisdictional differences in regulatory enforcement and compliance may limit the relevance of

these results for other countries.

47

CHAPTER 4: Study 2: The Effect of Audit Committee Expertise on External

Auditors’ Disclosures of Key Audit Matters

4.1 Introduction

In this study, I investigate, the AC, one of the main determinants of auditors’ choice of

disclosed KAMs in EARs. Since the UK required EARs from 2013, regulators worldwide,

including the IAASB and PCAOB, have adopted revised audit reporting standards, requiring

auditors to disclose the key/critical audit matters in the auditor’s report (IAASB 2015; PCAOB

2017). The implementation of EARs aims to improve audit transparency and provide financial

statement users with more entity-specific information, so that the information asymmetry between

investors and auditors can be reduced (PCAOB 2017). Financial statement users can better

facilitate their analyses of financial statements and assess financial performance. Supporting

empirical evidence regarding the usefulness of KAM disclosures has been found in prior studies

that the implementation of EARs can reduce financial analyst dispersion (Smith 2017) increase the

earnings response coefficient (Reid et al. 2018) and benefit firms from more favorable loan

contracting terms, evidenced as lower interest rate spread and higher loan maturity (Porumb et al.

2018).

Although the approaches taken by the FRC, IAASB, and PCAOB in determining the

matters that auditors should disclose in EARs differ in detail, it is generally required that disclosed

risks need to be selected from matters that are communicated with ACs (IAASB 2015; PCAOB

2017) and, thus, early and effective communications between auditors and ACs are seen as critical

in the new reporting regime (Deloitte 2016a). This argument is partially supported by prior

experimental research, which finds that auditors, under the KAM standard, are less willing to

48

communicate uncertain accounting estimates with the AC when committee members and not active

(Gay and Ng 2015).

The UK is not only the first jurisdiction to implement the EAR, but is also the only

jurisdiction that currently requires both auditors and ACs to disclose material matters in their

respective extended reports. For financial periods ending on or after September 30, 2013, of

companies that report on the application of the UK Corporate Governance Code, the ISA (UK and

Ireland) 700 requires auditors to “describe those assessed risks of material misstatement that were

identified by the auditor and which had the greatest effect on: the overall audit strategy; the

allocation of resources in the audit; and directing the efforts of the engagement team” (FRC 2013a,

paragraph 19A), while simultaneously, the UK Corporate Governance Code requires ACs to

disclose “the significant issues that the committee considered in relation to the financial statements

and how these issues were addressed, statements, and how these issues were addressed” (FRC

2012, section C 3.8). By comparing risk disclosures between EARs and EACRs, the UK reporting

regime provides a unique opportunity to empirically examine how AC’s risk assessments impact

auditors’ concurrent and subsequent KAM disclosures. For convenience, I label material risks

(which are largely synonymous with KAMs) in EARs as KAMs and significant issues in EACRs

as SIFs.

Earlier studies note that expertise in the AC enhances the AC’s role in monitoring the

financial reporting process and ensuring financial reporting quality (e.g., Cohen, Gaynor,

Krishnamoorthy, and Wright 2007, 2008; DeFond and Zhang 2014; SEC 2015) and reduces

internal control problems (e.g., Krishnan 2005; Zhang, J. Zhou, and N. Zhou 2007). Relatedly, AC

expertise has been associated with auditors reducing the disclosure of internal control weaknesses

(U. Hoitash, R. Hoitash, and Bedard 2009). Applying these findings to the new EAR regime, I posit

49

that expertise in the ACs may influence auditors’ judgment of matters that have been material in

relation to that year’s financial statement audit (which they are required to disclose as KAMs in

the EARs) to the extent that auditors may be influenced by the ACs’ identification of SIFs.

The association between the total number of KAMs disclosed in EARs and AC accounting

or industry expertise is examined first. Previous research notes that AC accounting and industry

expertise are associated with higher financial reporting quality (e.g., Cohen, Hoitash,

Krishnamoorthy, and Wright 2014; Dhaliwal, Naiker, and Navissi 2010; Krishnan and Visvanathan

2008) and, therefore, are predicted to be negatively associated with the total number of KAMs

disclosed by auditors.

I then investigate whether AC expertise is associated with the similarity of auditors’ KAM

disclosures and ACs’ simultaneous SIF disclosures. Although an exact alignment between KAMs

and SIFs is not a regulatory requirement, significant consistency is expected (Deloitte 2016b; FRC

2016b; KPMG 2014).27 Providing users with different but complementary perspectives on key

financial reporting issues is one of the purposes of implementing EARs and EACRs, and thus,

some complementarity is necessary to achieve this objective (FRC 2016b). The FRC asserts that

allowing financial statement users to interact information between EARs and EACRs can enhance

the credibility and value of that information significantly (FRC 2016b). Investors and auditors have

expectations regarding a significant overlap between KAMs and SIFs (Deloitte 2016b; FRC 2014,

2015b; KPMG 2014). I predict that “unmatched” KAMs and SIFs in EARs and EACRs are

negatively associated with AC accounting and industry expertise, for two reasons. First, AC

27 Detailed discussions on KAMs and SIFs are developed in Section 4.2.1 “Regulatory Standards”.

50

members with strong accounting and industry-relevant knowledge are more likely to have opinions

regarding critical accounting judgments and risk assessments that are consistent with their auditors’

opinions. Second, early and effective communications between auditors and ACs are required prior

to the completion of the reports; therefore, AC accounting and industry expertise should contribute

to more effective communications with auditors, reducing the potential for “unmatched” disclosed

KAMs and SIFs.

Overall, the results of this study are consistent with the predictions and suggest that both

AC accounting and industry expertise have significant effects on auditor’s KAM disclosures,

evidenced by fewer total KAMs and fewer “unmatched” KAMs reported in EARs. I also find that

the overall alignment between KAMs and SIFs disclosures is enhanced by both accounting and

supervisory AC expertise. Additional analysis of the extent to which AC expertise affects auditors’

subsequent reporting behavior reveals that auditors are more likely to adopt previously unmatched

SIFs when ACs’ accounting and industry expertise increase. This adoption of previously unmatched

SIFs may signal insufficient audit effort in the prior year, as proxied by abnormal audit fees,

although it is not necessarily a reflection of lower audit quality, as proxied by discretionary

accruals.

This study contributes to the audit research literature and policy development. First, to the

best of my knowledge, there has been very little prior research related to the interactions between

EARs and EACRs.28 By comparing the audit matters disclosed in each entity’s EAR and EACR,

28 Gutierrez et al. (2018) is the only known study having tests on the number of “unmatched” KAMs. A detailed

discussion of this study can be found in Section 4.2.1.

51

this study contributes to the emergent literature on the extended reporting models for both auditors

and ACs.

Second, the study expands the literature concerned with the roles and effects of ACs.

Studies pre-dating the expanded reporting regime identify significant effects of expertise in ACs

on external auditors’ performance, including audit fees and audit quality. However, there has been

little or no research investigating the AC’s role in auditors’ reporting behavior. I address this

knowledge void by revealing some aspects of how expertise in ACs affects auditors’ disclosures.

Third, the results can help inform regulators of the consequences of the revised reporting

standards for both auditors and ACs, and may influence regulators in their monitoring of EARs and

EACRs and in their future guidance. This analysis of the influence of ACs on auditors’ reporting

behavior in the UK may assist standard setters in other jurisdictions better understand auditors’

rationales for choosing which key/critical audit matters to disclose. Although the UK is currently

the only jurisdiction that requires auditors and ACs to disclose the most significant issues in relation

to financial statements simultaneously, standard setters in other jurisdictions have proposed

revisions to AC’s reporting, to enhance transparency about AC activities (e.g., IAASB 2014; SEC

2015), and the results may better inform future deliberations in this regard.

4.2 Literature Review and Hypotheses Development

4.2.1. Regulatory Standards and Related Studies

The ISA (UK and Ireland) 700 requires auditors to disclose risks of material misstatement

that have “had the greatest effect on the overall audit strategy, the allocation of resources in the

audit and directing the efforts of the engagement team” (FRC 2013a, paragraph 19A). At the same

52

time, the UK Corporate Governance Code requires ACs to disclose “the significant issues that the

committee considered in relation to the financial statements” (FRC 2012, Section C3.8). Although

the UK reporting regulations do not require the KAMs reported in EARs and the SIFs reported in

EACRs to be entirely consistent, a substantial degree of alignment is expected.

From the regulators’ perspective, providing users with different but complementary

perspectives on key financial reporting issues is one of the purposes of implementing EARs and

EACRs, and thus the key to achieve this objective is to have the same KAMs and SIFs identified

and discussed in both EACRs and EARs (FRC 2016b). In addition, standard setters expect that by

enabling financial statement users to jointly use the EAR and EACR, the credibility and value of

that information can be significantly enhanced (FRC 2016b). Financial statement users have

expressed their particular interest in this regard. In a FRC’s financial reporting lab project, investors

surveyed indicated that “there should be a high level of interaction of the reporting of key estimates

and judgements across the auditor and audit committee reports” (FRC 2014, 6). A significant

overlap between KAMs and SIFs is expected by accounting practitioners as well (Deloitte 2016b;

KPMG 2014).

I argue that the requirement for early and effective communications between the auditor

and AC prior to the completion of the two reports should encourage a reporting entity’s auditor and

AC to reach some agreement regarding critical accounting judgments and risk assessments.

Practitioners have claimed that the implementation of EARs makes auditors more willing to review

an early draft of the EACR; to understand the AC’s risk assessment better, including the rationale

for choosing which SIFs to be reported; and to explore any inconsistencies in the auditor’s planned

KAM disclosures (Deloitte 2013). Therefore, if auditors and ACs communicate effectively before

issuing their reports, there should be fewer “unmatched” KAMs and SIFs.

53

Very little prior research examines the relations between KAMs and SIFs. Gutierrez et al.

(2018), which is the most closely related known study, examine the implementation effect of the

expanded audit reporting regime. While they do not directly consider the role of the AC or the

EACR, their tests include the effects of the number of “unique audit risks,” which equate to

“unmatched” KAMs. They find that companies for which the auditor include any unmatched

KAMs experience no difference in pre-post change regarding investors’ reactions, audit fees, or

audit quality. 29 In contrast to Gutierrez et al. (2018), I investigate how AC expertise affects

auditors’ choice of KAMs, in relation to SIFs during the first three years of the expanded reporting

regime. I examine total KAMs, KAMs that are reported in EARs but not EACRs, SIFs that are

reported in EACRs but not EARs, and auditors’ subsequent choice of KAMs, and relate these to

AC expertise.

4.2.2. Prior Studies Related to Audit Committee Expertise

A broad literature review on AC can be found in DeFond and Zhang (2014) and Malik

(2014). In this thesis, I focus on two related streams most related to the current interests: those

concerned with internal control effects and those concerned with financial reporting quality.

A stream of internal control studies presents largely consistent results that internal control

quality, considered an important determinant of higher financial reporting quality, is influenced

significantly and positively by AC financial expertise. The existence of internal control problems

is found to be associated negatively with AC financial expertise, particularly accounting expertise

29 See Chapter 2 for a review of prior studies concerned with EARs.

54

(Krishnan 2005; Zhang et al. 2007). As there is a negative association between AC accounting

expertise and a lower likelihood of disclosing material weaknesses in internal control related to

accounting matters, this would appear to have audit consequences (U. Hoitash et al. 2009). The

weight of evidence suggests that AC accounting expertise is more important than supervisory

expertise in reducing the likelihood of significant internal control weaknesses, and thus it should

result in fewer audit concerns and higher accounting quality. However, Goh (2009) reports that the

timeliness of the remediation of internal control deficiencies is improved significantly by

supervisory expertise but not by accounting expertise.

Complementing these studies of internal control quality, another stream of studies provides

evidence that is more direct with regard to AC financial expertise enhancing financial reporting

quality. Early studies, without differentiating accounting and supervisory-related financial experts,

report that overall, financial expertise is associated with higher financial reporting quality. For

example, McDaniel, Martin, and Maines (2002), an experimental study, find that the framework

used by AC financial experts to evaluate the financial reporting quality is more consistent with

accounting standards and the experts are less likely to be distracted by financial items that are less

critical to reporting quality. Relatedly, archival studies report that ACs with at least one financial

expert significantly reduce the likelihood of financial restatements (Abbott, Parker, and Peters

2004); earnings management is negatively associated with having financial experts on an AC

(Bédard, Chtourou, and Courteau 2004; He and Yang 2014); and financial statement fraud firms

have fewer AC financial experts (Farber 2005).

Later studies that differentiate between accounting and supervisory expertise suggest that

the enhancement of financial reporting quality is contributed mainly by accounting experts, rather

than supervisory experts. Several studies report that having accounting experts on an AC (but not

55

supervisory experts) is positively associated with accrual quality, accounting conservatism, or more

timely restatement disclosures (e.g., Dhaliwal et al. 2010; Krishnan and Visvanathan 2008;

Schmidt and Wilkins 2013). R. Hoitash and U. Hoitash (2018) use the number of accounting items

disclosed in 10-K filings to proxy for the underlying complexity of entities’ financial reports

preparation and find a strong negative association with financial reporting quality, but this effect

is countered by having more accounting expertise in the AC. In addition, the contribution of

accounting expertise to companies’ financial reporting quality is supported by the finding that the

market reacts favorably only to the appointment of accounting experts to the AC, but not of

supervisory experts (DeFond, Hann, and Hu 2005).

Collectively, previous findings suggest that an entity’s financial reporting quality is

enhanced by AC expert members who can better monitor the company’s financial statements and

review significant financial reporting judgments, but that accounting experts are more effective in

this role. Presumably, this is because they are more familiar with accounting concepts and the

auditing process, and are better at identifying SIFs.

In recent years, the relevance of AC members’ experience in the specific entity’s industry

to financial reporting quality has received substantial attention. Cohen et al. (2014) find that AC

industry experience has a positive incremental effect on the financial reporting quality that is

additional to financial expertise alone. Specifically, they find that having more AC members with

accounting and industry expertise reduces the likelihood of financial restatements and the level of

discretionary accruals, while AC supervisory and industry expertise are negatively associated with

income-increasing discretionary accruals. Wang, Xie, and Zhu (2015) also find that AC industry

expertise is negatively associated with abnormal accruals. Experimental evidence suggests that the

influence of AC industry expertise is valued by investors; for example, Cohen, Gaynor, and

56

Krishnamoorthy (2017) find that investors assess AC members with industry experience as more

competent.30 Overall, there is persuasive evidence that AC members who have experience in the

specific entity’s industry are more effective in monitoring the financial reporting process,

presumably because this experience means they have more understanding of the industry-specific

challenges, opportunities, and financial conditions.

The pattern of prior results suggests that entities with more AC accounting expertise or

more AC industry expertise exhibit higher financial reporting quality and will present fewer audit

concerns (potentially reducing reported KAMs in EARs). However, I do not have any direct

evidence as to whether the presence of accounting experts or industry experts on an AC improves

the communications between the AC and its external auditors, or the potential for auditors and ACs

to be concerned with different significant issues, as reflected in differences in KAMs and SIFs.

4.2.3. Hypotheses Development

According to the UK auditing standard, reported KAMs should be the most significant

matters in that year’s financial statements audit, involving challenging, subjective, or complex

auditor judgment. Prior studies related to KAMs in EARs report results consistent with the

understanding of auditors’ identification of engagement risks and complexities; for example, there

are more KAMs for entities with more subsidiaries, with higher book-to-market ratios, or that are

cross-listed in the US (Lennox et al. 2018). The number of KAMs is found to be associated

30 Brazel and Schmidt (2018) report that AC chairs with prior industry experience are associated with larger

inconsistencies between companies’ reported revenue growth and related non-financial measures, which is interpreted

as higher fraud risk. This raises some interesting issues regarding the moderating effects of AC chairs on overall AC

effectiveness, but this is beyond the scope of the current study.

57

negatively with the maturity of loans and is associated positively with the number of debt covenants

included in loan contracts, which suggests that the number of KAMs is an indicator of entities’

riskiness and influences the banks’ risk assessments (Porumb et al. 2018). Similarly, consistent

with audit effort or pricing being responsive to assessed engagement risks, there is evidence that

audit fees are relatively higher for entities that have more KAMs disclosed in EARs (Gutierrez et

al. 2018).31

As discussed above, prior studies in non-EAR environments find significant positive

associations between accounting and industry expertise in the AC and financial reporting quality

(e.g., Cohen et al. 2014; Dhaliwal et al. 2010; Krishnan and Visvanathan 2008; Wang, Xie, and

Zhu 2015). It is argued that this is because AC members with accounting knowledge can better

understand and address significant accounting-specific issues, are more involved in accounting-

specific judgments, estimates, and assumptions, and thus are more effective in monitoring the

financial reporting process. AC members with relevant industry experience understand the

industry-specific complexities and risks better, may be familiar with the application of industry-

relevant accounting standards, judgments and estimates, and can thus significantly enhance the

AC’s monitoring role. Taken together, companies that have a high level of accounting and industry

expertise on their ACs are likely to have higher-quality financial statements, and it is expected that

fewer KAMs will be disclosed by auditors in their EARs. This may be because auditors directly

consider the expertise of the AC in their assessment of risks, or because of indirect effects when

auditors observe the benefits of AC expertise in relation to high-quality internal controls,

31 Gutierrez et al. (2018) do not identify any significant incremental pre-post effect of the total number of KAMs on

audit fees, implying that the disclosed KAMs do not reflect new information for auditors.

58

accounting processes, and financial reporting. A direct effect is supported by an experimental study

(D. Sharma, Boo, and V. Sharma 2008), which reports that auditors are more likely to assess

clients’ inherent and control risk as lower and conduct less-substantive audit testing if companies

have strong ACs. Based on the reasoning thus far, I predict that KAM disclosures in EARs are

affected by AC expertise, as expressed in Hypothesis 1 below:

HYPOTHESIS 1. Having accounting and industry expertise on ACs is negatively

associated with the total number of KAMs disclosed in EARs.

I next consider the extent to which AC expertise affects the implied interactions between

auditors and ACs, as signaled by differences in their reported KAMs and SIFs. I expect that AC

expertise increases the concordance of auditors and ACs’ judgments, resulting in fewer

“unmatched” KAMs and SIFs disclosed in EARs and EACRs, for two reasons.

First, the review of the literature supports the idea that AC accounting experts are more

involved with accounting judgments, estimates, and the audit process, and I argue they are more

likely to make judgments over complex accounting-related issues that are consistent with the

auditor’s judgments. Industry experts, who have superior knowledge of the specific industry’s

complexities and risks and are more capable of evaluating industry-specific issues, are more likely

than non-experts to present risk assessments that are similar to those of the auditors (Cohen et al.

2014, 2017). If AC accounting and industry expertise increases the similarity of AC and auditor

judgments regarding financial statement issues, and if the AC and the auditor have spent time

dealing with the same issue, then that risk is important enough to be disclosed in both that year’s

EAR and EACR (FRC 2013b). Therefore, a greater number of consistent (fewer unmatched)

KAMs and SIFs can be expected.

59

Second, as discussed above, the implementation of EARs and EACRs is expected to

encourage early and effective communications between ACs and auditors during the preparation

of EARs and EACRs; I argue that the effectiveness of these communications should be enhanced

by ACs having more accounting and industry expertise. Although I cannot observe the

communications between auditors and ACs, this argument is supported indirectly by evidence that

timelier financial reporting, which may signal more effective communications, are positively

associated with AC accounting expertise (Abernathy, Beyer, Masli, and Stefaniak 2014).

Abernathy, Beyer, Masli, and Stefaniak (2015) argue that AC accounting expertise reduces the

time an AC requires to discuss and evaluate significant accounting policies and unusual

transactions with auditors. It is plausible that this argument can be extended to AC industry

expertise. Hypothesis 2 expresses (inversely) the idea that more AC accounting and industry

expertise can enhance communication effectiveness between auditors and ACs and result in a

higher degree of concordance between KAMs in EARs and SIFs in EACRs:

HYPOTHESIS 2. Having accounting and industry expertise on ACs is negatively

associated with the extent of “unmatched” KAMs and SIFs in EARs

and EACRs.

60

4.3 Method

4.3.1. Models for Testing the Hypotheses

To test the extent to which AC accounting and industry expertise affects the number of

KAMs included in EARs (Hypothesis 1), I estimate Model 4.1 (which is extended from Lennox et

al. 2018) as follows32:

TOTAL_KAMit = β0 + β1%AC_AFEit + β2%AC_SFEit + β3%AC_INEit + β4SIZEit + β5MTBit+

β6CATAit + β7LOSSit + β8LEVERAGEit + β9IRISKit + β10MAit + β11LIST_USit +

β12ANALYST_COVit + β13LNSUBit + β14AC_SIZEit + β15BDSIZEit +

β16BDINDEPit + β17FIRM_CHGit+ β18AUDITOR_EXPit + Σβ𝑗industry +

Σβ𝑘year + Σβ𝑚audit firm + it

(Model 4.1)

where TOTAL_KAM is the dependent variable, which is the total number of KAMs included in

EARs; and %AC_AFE, %AC_SFE and %AC_INE are the test variables, defined as the percentage

of AC members who are accounting, supervisory, and industry experts, respectively.

The UK Corporate Governance Code (FRC 2012) does not define expertise. Consistent

with prior studies (e.g., Dhaliwal et al. 2010), I classify AC members as having accounting

expertise if their biographies indicate that they have at least one of the following qualifications or

jobs: certified public accountant, chief financial officer, auditor, chief accounting officer,

controller, treasurer, or vice president-finance. I define AC members as having supervisory

expertise if they are not classified as having accounting expertise and their biographies indicate

32 Controls for auditor size and AC independence are not included in the regression models because only 4 per cent of

the sample is audited by non-Big4 auditors and only 3 per cent has non-independent members on ACs. Removing these

observations does not change the reported results meaningfully.

61

that they have held at least one of the following positions: chief executive officer, chief operating

officer, chair of the board, or president of a company; thus, %AC_AFE and %AC_SFE are mutually

exclusive measures. AC members are classified as industry experts if they have experience in

another firm that has the same two-digit ICB code as the company that is now being served (Cohen

et al. 2014); thus, AC members can be classified as having both industry expertise and one of

accounting or supervisory expertise.

The control variables reflect the evidence that the number of KAMs is greater for riskier

and more complex entities (Lennox et al. 2018), measured by the following engagement-level

financial characteristics: the natural log of total assets (SIZE), the market-to-book ratio (MTB), the

ratio of current assets to total assets (CATA), negative net income dummy (LOSS), total debt

divided by total assets (LEVERAGE), inventory and receivables divided by total assets (IRISK),

mergers-and-acquisitions event dummy (MA), the US cross-listing dummy (LIST_US), the number

of analysts following (ANALYST_COV), and the natural log of the number of segments (LNSUB).

In addition, because AC expertise is the focus of this study, to isolate other AC- and board-related

effects, the number of AC members (AC_SIZE), the number of board members (BDSIZE), and the

percentage of independent non-executive members (BDINDEP) are controlled in the model as well.

Following Lennox et al. (2018), next year’s audit firm change dummy (FIRM_CHG) is included

because auditors are more likely to be removed after disclosing unfavorable information. I control

for the auditor industry specialization (AUDITOR_EXP) to isolate other sources of industry

experience (Cohen et al. 2014). In addition, the model controls for year, industry, and audit-firm

fixed effects. For convenience, all variable definitions are tabulated in Appendix 1.

62

To test whether inconsistencies between KAMs and SIFs disclosures are associated with

AC expertise (Hypothesis 2), I modify the previous model as follows:33

UNMATCHEDit = β0 + β1%AC_AFEit + β2%AC_SFEit + β3%AC_INEit + β4SIZEit + β5MTBit+

β6CATAit + β7LOSSit + β8LEVERAGEit + β9IRISKit + β10MAit + β11LIST_USit +

β12ANALYST_COVit + β13LNSUBit + β14AC_SIZEit + β15BDSIZEit +

β16BDINDEPit + β17FIRM_CHGit + β18AUDITOR_EXPit+ Σβ𝑗industry +

Σβ𝑘year + Σβ𝑚audit firm + it

(Model 4.2)

where the dependent variable UNMATCHED represents either %UNMATCHED_KAM or

%UNMATCHED_SIF; %UNMATCHED_KAM (%UNMATCHED_SIF) is the percentage of

KAMs (SIFs) disclosed in the EAR (EACR) but not disclosed as SIFs (KAMs) in that entity’s

EACR (EAR) for that year.

The KAM “fraud in revenue recognition” is considered the same as the SIF “revenue

recognition” in the main analyses, because it appears that some EARs (mainly but not exclusively

from PwC) use the two terms interchangeably across the sample period. 34 Test and control

variables are the same as those in Model 4.1.

33 The results remain unchanged when the model is estimated using ordered logit and the dependent variable is

measured using the number of unmatched KAMs/SIFs. A detailed discussion is provided in the robustness tests. 34 For example, the KAM “risk of fraud in revenue recognition” and the SIF “revenue recognition” discussed in the

entity “Microgen Plc” annual report for the period ended December 31, 2015 are “matched”.

63

4.3.2. Sample Selection

The sample is selected from all UK non-financial entities that reported on the application

of the UK Corporate Governance Code for the fiscal years between September 30, 2013 and

September 30, 2016.35 The sample selection process is summarized in Table 4.1. I start with all

available relevant annual reports, which yields 1,100 firm-year observations that I use to capture

the significant issues from both EARs and EACRs. These are then merged with the BoardEx

database to obtain AC-related measures; 235 observations are eliminated during this process. I next

merge the dataset with DataStream to obtain the financial measures and lose 172 observations

because of the missing data.36 The final sample for testing the two hypotheses is comprised of 693

firm-year observations.37

35 Two early adopters of EARs are included in the sample and the results are unchanged after removing them from the

analyses. 36 Entities that are not in accordance with the UK Corporate Governance Code requirements to establish their board

and AC are not excluded from the sample, because the results remain consistent after removing these observations. A

more detailed analysis is provided in the robustness tests section. 37 The final sample size is reasonable, compared to other prior EAR studies in the UK. For example, Gutierrez et al.

(2018) and Reid et al. (2018) report 872 and 1,088 firm-year observations, respectively, for four-year period pre-post

abnormal accrual analyses.

64

Table 4.1 Sample Selection

Firm-year

Observations

Total available annual reports of non-financial entities (used to capture the

significant issues from both EARs and EACRs) 1,100

Less: Missing observations when merging with BoardEx to compute AC-relevant

variables (235)

Less: Missing observations when merging with DataStream to compute control

variables (172)

Final Sample: Hypothesis 1 and Hypothesis 2 tests 693

4.4 Results

4.4.1. Descriptive Statistics

Descriptive statistics for the variables used in the analyses are presented in Table 4.2.38 In

the final sample, the average EAR identifies 3.8 KAMs (TOTAL_KAM), consistent with 4.0 in

Gutierrez et al. (2018), 3.9 in Lennox et al. (2018), 4.1 in Porumb et al. (2018), and 3.9 in Reid et

al. (2018). Approximately 31 per cent of KAMs disclosed in the EARs are not matched with any

SIFs in that entity’s EACR (%UNMATCHED_KAM), consistent with Gutierrez et al. (2018). The

average EACR identifies 4.6 SIFs. Approximately 38 per cent of SIFs are not reported in the

corresponding EAR (%UNMATCHED_SIF).

38 To mitigate the effect of outliers, all continuous variables are winsorized at 0.01 (Francis and Yu 2009). As a

robustness check, I also estimate the models without winsorizing and do not observe any meaningful differences in the

regression results.

65

Table 4.2 Descriptive Statistics for Hypotheses-testing Sample

Variable N Mean SD p25 p50 p75

TOTAL_KAM 693 3.818 1.375 3.000 4.000 5.000

%UNMATCHED_KAM 693 0.309 0.278 0.000 0.286 0.500

%UNMATCHED_SIF 693 0.379 0.283 0. 000 0.400 0.600

%AC_AFE 693 0.359 0.233 0.250 0.333 0.500

%AC_SFE 693 0.430 0.236 0.250 0.429 0.600

%AC_INE 693 0.088 0.160 0.000 0.000 0.200

SIZE 693 13.725 1.701 12.589 13.617 14.772

MTB 693 3.627 4.678 1.355 2.523 4.332

CATA 693 0.420 0.213 0.261 0.394 0.551

LOSS 693 0.179 0.384 0.000 0.000 0.000

LEVERAGE 693 0.209 0.171 0.069 0.190 0.300

IRISK 693 0.282 0.189 0.135 0.261 0.384

MA 693 0.150 0.357 0.000 0.000 0.000

LIST_US 693 0.026 0.159 0.000 0.000 0.000

ANALYST_COV 693 11.255 7.670 5.000 10.000 16.000

LNSUB 693 3.693 1.143 3.099 3.609 4.792

AC_SIZE 693 4.271 1.175 3.000 4.000 5.000

BDSIZE 693 8.260 2.251 7.000 8.000 9.000

BDINDEP 693 0.674 0.116 0.600 0.667 0.750

FIRM_CHG 693 0.051 0.219 0.000 0.000 0.000

AUDITOR_EXP 693 0.284 0.108 0.235 0.287 0.333

Variable definitions are provided in Appendix 1.

The greater mean for %UNMATCHED_SIF compared with %UNMATCHED_KAM is

expected. Under the UK auditing standard, KAMs should be communicated to the AC before their

reports are published, and if the EACR “does not appropriately address matters communicated by

the auditor to the audit committee,” the auditor is required to disclose such information in the EAR

(FRC 2013a, paragraph 22B). KAMs that the auditor discloses in the EAR are selected from those

communicated to the AC and, if a matter has been communicated by the auditor to the AC and, the

AC and the auditor has spent significant time dealing with it, the matter should be important enough

66

to be disclosed in both reports (FRC 2013b). Therefore, compared to SIFs, KAMs are more

expected to be “matched”.

On average, the AC has 1.5 members with accounting expertise for a mean %AC_AFE of

nearly 36 per cent, 1.9 members with supervisory expertise for a mean %AC_SFE of 43 per cent,

and 0.4 members with industry expertise for a mean %AC_INE of nearly 9 per cent. In five cases,

the entities are not in compliance with the UK requirement that an AC should have at least one

member with recent and relevant financial expertise.

Distributions of the financial variables are largely consistent with prior studies in the UK

(e.g., Reid et al. 2018). The average size of the AC is approximately 4.3 members (ACSIZE), the

average board size is 8.3 members (BDSIZE), and around 67 per cent of the board members are

independent non-executive members (BDINDEP). About 5 per cent of the entities change audit

firms in year t+1 (FIRM_CHG), and 28 per cent of the auditors are industry specialists

(AUDITOR_EXPERT).

4.4.1.1.Correlations between the Regression Variables

A Pearson correlation matrix is reported in Table 4.3, without raising any particular

concerns. It shows the expected large negative correlation between the mutually exclusive

%AC_AFE and %AC_SFE. To further address the multicollinearity concerns in the analyses, VIFs

are computed on the regression models. The highest VIF is 3.67, which is well below the

benchmark of 10 that is typically used to identify multicollinearity problems (Neter, Kutner,

Nachtsheim, and Wasserman 1996).

67

Table 4.3 Pearson Correlation Matrix

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

1 TOTAL_KAM 1.0000

2 %UNMATCHED_KAM 0.2660* 1.0000

3 %UNMATCHED_SIF -0.0378 0.6249* 1.0000

4 %AC_AFE -0.1199* -0.3170* -0.1925* 1.0000

5 %AC_SFE 0.0525 0.0418 0.0020 -0.5308* 1.0000

6 %AC_INE -0.1673* -0.1720* -0.0842* 0.1291* -0.0153 1.0000

7 SIZE 0.3438* 0.1334* 0.1411* -0.0052 -0.0011 -0.1810* 1.0000

8 MTB -0.0098 -0.0121 0.0954* 0.0993* -0.0450 -0.0310 0.0005 1.0000

9 CATA -0.1366* -0.0751* -0.0245 -0.0103 -0.0107 -0.0242 -0.3287* 0.1333* 1.0000

10 LOSS 0.1248* 0.0521 0.0329 -0.1019* 0.0134 -0.0417 -0.0808* -0.1215* -0.0496 1.0000

11 LEVERAGE 0.1834* 0.1097* 0.1088* -0.0121 -0.0966* -0.0931* 0.3171* 0.0188 -0.3811* 0.1267* 1.0000

12 IRISK -0.1289* -0.1452* -0.1045* 0.0546 -0.0295 -0.0071 -0.2583* 0.0671 0.8476* -0.1159* -0.3258* 1.0000

13 MA -0.0003 -0.0432 0.0014 0.0541 -0.0096 0.0296 -0.0113 -0.0270 -0.1401* -0.0064 -0.0104 -0.1192* 1.0000

14 LIST_US 0.1470* 0.1050* 0.1496* -0.0452 -0.0606 -0.0638 0.2605* -0.0983* -0.0132 0.0421 0.0297 -0.0438 -0.0686 1.0000

15 ANALYST_COV 0.2725* 0.1050* 0.1209* -0.0126 -0.0191 -0.1529* 0.7764* 0.1768* -0.2337* -0.1045* 0.1642* -0.2350* -0.0651 0.1910* 1.0000

16 LNSUB 0.1269* -0.0014 -0.0130 0.0260 0.0957* 0.0002 0.1427* -0.0513 -0.1010* -0.0271 0.0378 -0.0843* -0.0356 -0.0782* 0.1481* 1.0000

17 AC_SIZE 0.1746* 0.0370 0.0611 -0.0965 0.0673 -0.0550 0.3810* 0.0903* -0.0430 -0.0822* 0.1310* -0.0234 -0.0902* 0.1168* 0.3737* 0.0574 1.0000

18 BDSIZE 0.2734* 0.1258* 0.1022* -0.0554 -0.0251 -0.1458 0.6811* 0.1290* -0.2091* -0.0673 0.1675* -0.2170* -0.0162 0.2150* 0.6723* 0.1052* 0.3848* 1.0000

19 BDINDEP 0.1293* 0.0588 0.0967* 0.0414 0.0139 -0.0622 0.4426* 0.0451 -0.2267* -0.0009 0.2045* -0.2330* -0.0714 0.1033* 0.3970* 0.1785* 0.2721* 0.2597* 1.0000

20 FIRM_CHG -0.0031 0.0039 0.0341 -0.0551 0.0644 -0.0397 0.0931* 0.0302 -0.0150 -0.0733 0.0146 0.0160 0.0138 0.0452 0.0946* -0.0312 0.0197 0.0730 -0.0042 1.0000

21 AUDITOR_EXP 0.0082 -0.0997* -0.0525 0.0218 0.0985* 0.0101 0.1142* -0.0602 0.0156 -0.0266 -0.0350 0.0487 0.0354 0.0451 0.0961* 0.0078 0.0785* 0.0545 0.0298 0.0130 1.0000

Variable definitions are provided in Appendix 1.

68

4.4.2. Regression Results

To test Hypothesis 1, I estimate Model 4.1 with TOTAL_KAM as the dependent variable.

The results are reported in Table 4.4. Column 1 reports the results without the AC-related variables,

as a base model. Column 2 reports the results for the fully specified Model 4.1. There is a modest

increase in the adjusted R-squared, indicating that the significant test variables help to better

explain the sample variation in TOTAL_KAM. The results support Hypothesis 1 but are more

persuasive with respect to AC accounting expertise. In explaining the total number of KAMs

disclosed in EARs, AC accounting expertise (%AC_AFE) and industry expertise (%AC_INE) have

significantly negative effects (at p-values < 0.01 and 0.10, respectively). The coefficient for

supervisory expertise (%AC_SFE) is not significant.

With respect to the control variables, auditors, on average, disclose more KAMs for firms

that are larger, (SIZE), report losses (LOSS), and have more subsidiaries (LNSUB). This is

consistent with the expectation that KAMs reflect some established audit concerns regarding size,

risk, and complexity (Lennox et al. 2018).

69

Table 4.4 Regression Results for Hypothesis 1 Test

TOTAL_KAM

(1) (2)

%AC_AFE - -0.765***

(-2.91)

%AC_SFE - 0.036

(0.15)

%AC_INE - -0.553*

(-1.90)

SIZE 0.242*** 0.232***

(4.45) (4.28)

MTB 0.002 0.006

(0.19) (0.52)

CATA 0.063 -0.126

(0.14) (-0.29)

LOSS 0.515*** 0.495***

(4.07) (3.95)

LEVERAGE 0.275 0.171

(0.89) (0.55)

IRISK 0.428 0.551

(0.84) (1.09)

MA -0.034 0.011

(-0.26) (0.08)

LIST_US 0.347 0.358

(1.11) (1.16)

ANALYST_COV 0.000 0.000

(0.03) (0.02)

LNSUB 0.122*** 0.125***

(2.71) (2.81)

AC_SIZE - 0.016

(0.37)

BDSIZE 0.035 0.028

(1.17) (0.93)

BDINDEP 0.042 0.136

(0.09) (0.29)

FIRM_CHG -0.307 -0.360

(-1.46) (-1.73)

AUDITOR_EXP 0.604 0.752

(0.85) (1.06)

Constant 0.019 0.225

(0.03) (0.31)

Industry, Year, Audit firm FE Included Included

N 693 693

Adjusted R2 0.281 0.298

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

70

To test Hypothesis 2, Model 4.2 is variously estimated with %UNMATCHED_KAM and

%UNMATCHED_SIF as the dependent variables, as reported in Table 4.5. The results for

%UNMATCHED_KAM are reported in Columns 1 and 2 and the results for %UNMATCHED_SIF

are reported in Columns 3 and 4. I again first estimate Model 4.2 for each dependent variable

excluding the AC variables, as reported in Columns 1 and 3. The results for the fully specified

Model 4.2 are reported in Columns 2 and 4. For each dependent variable, the substantial increase

in the adjusted R-squared when the AC variables are added indicate that the significant AC

variables help to better explain the variation in the unmatched KAMs and SIFs. The results support

Hypothesis 2 with respect to AC accounting expertise but are ambiguous with respect to industry

expertise. While I do not hypothesis an effect for supervisory expertise, it is found to reduce the

proportions of unmatched KAMs and SIFs. For %UNMATCHED_KAM, the coefficients for each

of %AC_AFE, %AC_SFE, and %AC_INE are significantly negative (at p-values < 0.01, 0.01, and

0.10, respectively). For %UNMATCHED_SIF, the negative effects of %AC_AFE and %AC_SFE

are significant (at p-values < 0.01 and 0.05, respectively) but the effect of %AC_INE is not

significant.

71

Table 4.5 Regression Results for Hypothesis 2 Test

%UNMATCHED_KAM %UNMATCHED_SIF

(1) (2) (3) (4)

%AC_AFE - -0.451*** - -0.317***

(-8.33) (-5.27)

%AC_SFE - -0.127** - -0.118**

(-2.57) (-2.14)

%AC_INE - -0.100* - -0.009

(-1.67) (-0.13)

SIZE 0.020* 0.021* 0.022* 0.023*

(1.69) (1.91) (1.74) (1.88)

MTB -0.002 0.000 0.006** 0.007***

(-0.84) (0.03) (2.42) (3.01)

CATA 0.296*** 0.216** 0.327*** 0.275***

(3.10) (2.37) (3.20) (2.73)

LOSS 0.005 -0.002 -0.011 -0.014

(0.17) (-0.08) (-0.37) (-0.49)

LEVERAGE 0.046 -0.014 0.082 0.036

(0.69) (-0.22) (1.15) (0.51)

IRISK -0.331*** -0.268** -0.342*** -0.302***

(-3.01) (-2.58) (-2.92) (-2.62)

MA -0.027 -0.015 0.011 0.021

(-0.96) (-0.55) (0.37) (0.68)

LIST_US 0.045 0.030 0.183** 0.171**

(0.67) (0.48) (2.55) (2.41)

ANALYST_COV -0.003 -0.002 -0.001 -0.001

(-1.13) (-1.09) (-0.35) (-0.34)

LNSUB 0.008 0.013 0.007 0.011

(0.80) (1.38) (0.69) (1.12)

AC_SIZE - -0.017* - -0.005

(-1.91) (-0.53)

BDSIZE 0.007 0.006 -0.004 -0.006

(1.04) (0.91) (-0.64) (-0.88)

BDINDEP 0.007 0.106 -0.021 0.038

(0.07) (1.10) (-0.19) (0.36)

FIRM_CHG 0.004 -0.020 0.058 0.045

(0.08) (-0.46) (1.20) (0.95)

AUDITOR_EXP -0.380** -0.256* -0.123 -0.033

(-2.49) (-1.76) (-0.75) (-0.21)

Constant 0.067 0.150 0.006 0.054

(0.43) (1.01) (0.04) (0.33)

Industry, Year, Audit firm FE Included Included Included Included

N 693 693 693 693

Adjusted R2 0.179 0.268 0.100 0.134

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

72

With respect to the control variables, the proportions of both the unmatched KAMs and

unmatched SIFs are positively associated with SIZE (weakly) and CATA, and negatively associated

with IRISK. The findings suggest that the auditors and ACs of the relevant entities identify

different, but not necessarily more or fewer, significant issues; I note that neither CATA nor IRISK

is significant in relation to total KAMs (see Table 4.4). In addition, the results indicate that, on

average, the proportion of unmatched KAMs is lower for entities that are audited by auditors with

industry expertise (AUDITOR_EXP) and with larger ACs (AC_SIZE); again, I note that neither of

these two variables is significant in relation to total KAMs. The AUDITOR_EXP and AC_SIZE

effects may be associated with issues associated with entity size and industry but it is not within

the scope of this study to examine such issues in depth. The proportion of unmatched SIFs is greater

for entities with larger market-to-book ratio (MTB) and a US cross-listing (LIST_US), but these do

not have any effect for unmatched KAMs, suggesting that ACs for such entities have more concerns

regarding these aspects compared to the auditors.

Overall, the results indicate that AC expertise is an important influence in auditor’s KAM

disclosures. The accounting and industry expertise of ACs are negatively associated with both the

total number of KAMs and “unmatched” KAMs included in EARs, and the overall alignment

between KAMs and SIFs is significantly enhanced by AC accounting and supervisory expertise.

The differences in the R-squared statistics between Columns 2 and 4 in Table 4.5, and the

changes in R-squared statistics between Columns 1 and 2 compared with the change between

Columns 3 and 4 in Table 4.5, suggest that audit AC expertise has more implications for KAM

disclosures in EARs than for SIF disclosures in EACRs, despite the mean for unmatched SIFs

being higher than for unmatched KAMs (their variances are similar).

73

4.5 Robustness Tests

4.5.1. Analyses Excluding Entities Not in Compliance with the Corporate Governance Code

I repeat the analyses after removing observations for entities that are not in compliance with

the FRC’s requirements for ACs. According to the UK Corporate Governance Code, Section C.3.1,

an AC should have at least three independent non-executive directors and, under Section D.1.3,

remuneration for non-executive directors should not include share options or other performance-

related elements.39 I remove 31 observations that do not satisfy these requirements and then I re-

estimate Models 4.1 and 4.2. As reported in Appendix 6, the results are largely consistent with the

main analyses for both hypotheses. The main difference is that supervisory expertise %AC_SFE is

no longer significant in relation to unmatched SIFs. The effects for AC accounting and industry

expertise are unchanged across the three models and both hypotheses are supported by the results

in the same manner as for the main tests.

4.5.2. Analyses Using Alternative Measures of Unmatched KAMs, Unmatched SIFs, and

Count Regression

I retest Hypothesis 2 by replacing the percentage of unmatched KAMs

(%UNMATCHED_KAM) with the number of unmatched KAMs (UNMATCHED_KAM), and the

percentage of unmatched SIFs (%UNMATCHED_SIF) with the number of unmatched SIFs

(UNMATCHED_SIF), as the dependent variable. Because the dependent variable is a count

variable, I estimate the models using the count regression method, but I include using ordinary least

39 For smaller companies, below the FTSE 350, the board should establish an AC of at least two independent non-

executive directors. All entities included in the sample are FTSE 350 companies.

74

squares as well. The total number of KAMs or SIFs (TOTAL_KAM or TOTAL_SIF) is added to the

respective models as an additional control variable. As reported in Appendix 7, the results are

similar to the main analyses but provided stronger support for Hypothesis 2. On average, more AC

accounting, industry, and supervisory expertise reduce the number of unmatched KAMs, and more

AC accounting and supervisory expertise (but not industry expertise) reduce the number of

unmatched SIFs.

4.5.3. Analyses Using Alternative Measure of AC Expertise

I also repeat all of the previous analyses using an alternative measure of AC expertise. I re-

estimate Models 4.1 and 4.2, using the number of AC accounting experts (numAC_AFE),

supervisory experts (numAC_SFE), and industry experts (numAC_INE) as the test variables. The

results are reported in Appendix 8 and are generally consistent with the reported results for

%AC_AFE, %AC_SFE, and %AC_INE.

4.5.4. Analyses Using Alternative Measure of UNMATCHED

In the main analyses, “fraud in revenue recognition” and “revenue recognition” are treated

as “matched,” and I re-estimate Models 4.1 and 4.2 by treating them differently

(%UNMATCHED_KAMdif and %UNMATCHED_SIFdif). As reported in Appendix 9, these results

are largely consistent with those of the main analyses.

75

4.6 Additional Analyses

4.6.1. Analyses on UNMATCHED_SIF Reported as KAM in the Following Year

The main analyses focus on the influence of AC expertise on the concurrent disclosures of

KAMs and SIFs. In this additional analysis, I explore the reporting behavior of auditors in

subsequent years, particularly in relation to SIFs that the auditor does not include as KAMs in the

current year. There are 455 observations pertaining to the second and third year of the three-year

sample period. Of these, 316 observations (70 per cent) have at least one unmatched SIF in the

prior year with no change in audit firm. Surprisingly, 59 of the 316 observations (19 per cent) have

at least one of their prior year’s unmatched SIFs disclosed as a KAM in the current year’s EAR;

for convenience, I label this change in the auditor’s reporting behavior as “PICK.”40 Given the

auditor previously “rejected” the SIF as a KAM that warranted disclosure, either the auditors’

understanding or knowledge, or engagement circumstances, are changed to cause the auditor to

consider the SIF now sufficiently important to warrant disclosure as a KAM. Given that the interest

here is the effects of the AC on auditors’ reporting decisions, I examine whether this behavior is

related to AC expertise. To do this, I consider both the expertise of the AC in the year the SIF was

unmatched (t-1) and changes in expertise in the current year (t). This “PICK” prediction model,

which is a change model based on the previous Models 4.1 and 4.2, is stated as follows, using

logistic regression:

40 In the sample, the top five most popularly “picked” KAMs are “revenue recognition,” “goodwill impairment,”

“taxation,” “exceptional items,” and “investments.” For robustness, PICK is captured as well when treating “revenue

recognition” and “fraud in revenue recognition” differently, and I exclude “revenue recognition” and/or “fraud in

revenue recognition.” Significant positive coefficients are consistently found on changes on AC accounting expertise

(d.%AC_AFE).

76

PICKit = β0 + β1d.%AC_AFEit + β2d.%AC_SFEit + β3d.%AC_INEit + β4%AC_AFEit-1 +

β5%AC_SFEit-1 + β6%AC_INEit-1 + β7UNMATCHED_SIFit-1 + β8UNMATCHED_KAMit-

1+ β9DROP_UNMATCHEDit+ β10ADD_NEWit+ β11ABAFEEit-1 + β12d.SIZEit+ β13d.MTBit

+ β14d.CATAit+ β15LOSSit + β16d.LEVERAGEit+ β17d.IRISKit+ β18MAit+ β19LIST_USit+

β20d.ANALYST_COVit+ β21LNSUBit + β22d.AC_SIZEit + β23d.BDSIZEit + β24d.BDINDEPit

+ β25d.AUDITOR_EXPit+ β26AUDITOR_EXPit-1+ Σβ𝑗industry+ Σβ𝑚audit firm + it

(Model 4.3)

where PICK equals 1 if there is at least one added KAM in the year t; EAR is the same as an

unmatched SIF disclosed in that entity’s EACR in t-1, and 0 otherwise; PICK is limited to cases

where an unmatched SIF in t-1 is reported as both a KAM and a SIF in year t; and d denotes the

change in the relevant variable form t-1 to t.

I add other variables that reflect the conjectures that the occurrence of PICKs could be

greater when there are fewer unmatched KAMs (UNMATCHED_KAM) and more unmatched SIFs

(UNMATCHED_SIF) in the prior year, and when auditors drop more unmatched KAMs from the

prior year’s EARs (DROP_UNMATCHED); and less when the auditor is adding more new KAMs

to this year’s EARs, excluding PICKs (ADD_NEW). In addition, I include the previous year’s

abnormal audit fees, ABAFEE in year t-1 to control for the possibility that the auditor’s previous

behavior was a consequence of audit effort or resource constraints. ABAFEE is computed as the

residual of the audit fee model (Reid et al. 2018), stated as follows:

LNAFEE = β0 + β1SIZE + β2MTB + β3CFO + β4IRISK + β6SALESVOL + β7LOSS + β8LEVERAGE

+ β9BUSY+ β10ROA+ β11BIG4+ Σβ𝑗industry+ Σβ𝑘year+

(Model 4.4)

where LNAFEE is the natural logarithm of audit fees; CFO is the cash flow from operations divided

by total assets at the end of year t; SALESVOL is the standard deviation of annual sales over the

77

prior seven years; BUSY is a dummy variable that equals 1 if the company’s fiscal year end from

December 1 to March 31, and 0 otherwise; ROA is the net income before extraordinary items

divided by total assets at the end of year t; and BIG4 is a dummy variable that equals 1 if the auditor

is from Big4, and 0 otherwise. Other variables are as defined earlier.

This regression is estimated using the full sample of non-financial sector firms for which

the necessary data are available.

The sample used to estimate Model 4.3 consists only of cases that have at least one

unmatched SIF in the prior year and have the same auditor in year t-1 and year t. Application of

the sample criteria yields 219 firm-year observations, as summarized in Table 4.6.

Table 4.6 Sample Selection for Further Analyses

Firm-year

Observations

Observations with unmatched SIFs in prior year’s EACR 343

Less: Observations that change auditor this year (27)

Less: Observations with missing control variable data (97)

Final sample used to estimate Model 4.3 219

The results of audit fee model are reported in Table 4.7. The model provides a good fit with

an adjusted R-squared of 0.728, in line with prior research.

78

Table 4.7 Regression Results for Audit Fee Model

LNAFEE

SIZE 0.542***

(34.63)

MTB -0.002

(-1.36)

CFO 1.036***

(3.36)

IRISK 1.026***

(6.92)

SALESVOL -0.081

(-0.34)

LOSS -0.117

(-1.52)

LEVERAGE 0.171

(1.22)

BUSY 0.277***

(5.58)

ROA -0.653**

(-2.37)

BIG4 -0.154**

(-2.05)

Constant 5.521***

(22.07)

Industry, Year FE Included N 1,660 Adjusted R2 0.728 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

The results for Model 4.3 are reported in Table 4.8, with both the estimated coefficients and

the corresponding marginal effects evaluated at the sample means of the exogenous variables

reported in Columns 1 and 2. Changes in AC accounting (d.%AC_AFE) and industry expertise

(d.%AC_INE) are significant (at p<0.10 and p<0.05 respectively), while the previous periods’ AC

expertise measures (for when the relevant unmatched SIFs occurred) are not significant. These

results imply that auditors are become more accepting of the AC propositions (as reflected in their

79

continued disclosure of previously unmatched SIFs) when the ACs’ persistence with these SIFs is

accompanied by improvements in AC accounting and industry expertise.

Table 4.8 Logistic Regression Results for PICK on AC Expertise

PICK

(1) (2)

Coefficient Marginal Effect

d.%AC_AFE 3.664* 0.263*

(1.70) (1.76)

d.%AC_SFE 2.940 0.211

(0.99) (1.00)

d.%AC_INE 6.450** 0.462**

(1.97) (2.02)

%AC_AFE (t-1) 1.436 0.103

(0.67) (0.67)

%AC_SFE (t-1) 2.635 0.189

(1.56) (1.60)

%AC_INE (t-1) 1.085 0.078

(0.40) (0.40)

UNMATCHED_SIF(t-1) 1.156*** 0.083***

(3.87) (4.48)

UNMATCHED_KAM(t-1) -0.591* -0.042*

(-1.67) (-1.71)

DROP_UNMATCHED 3.998*** 0.287***

(4.48) (5.56)

ADD_NEW -1.014*** -0.073***

(-3.07) (-3.36)

ABAFEE (t-1) -0.745* -0.053*

(-1.84) (-1.90)

d.SIZE -1.039 -0.074

(-0.46) (-0.46)

d.MTB -0.067 -0.005

(-0.70) (-0.70)

d.CATA 0.961 0.069

(0.16) (0.16)

LOSS 0.676 0.048

(0.87) (0.88)

d.LEVERAGE 0.993 0.071

(0.25) (0.25)

d.IRISK 2.142 0.154

80

(0.19) (0.19)

MA -0.279 -0.020

(-0.29) (-0.29)

LIST_US -0.813 -0.058

(-0.53) (-0.53)

d.ANALYST_COV 0.043 0.003

(0.39) (0.39)

LNSUB -0.296 -0.021

(-0.89) (-0.89)

d.AC_SIZE 0.265 0.019

(0.94) (0.95)

d.BDSIZE 0.459* 0.033*

(1.83) (1.88)

d.BDINDEP 2.628 0.188

(0.61) (0.61)

d.AUDITOR_EXP 13.716 0.983

(1.20) (1.22)

AUDITOR_EXP (t-1) 3.109 0.223 (0.49) (0.49) Industry, Audit firm FE Included

N 219

Pseudo R2 0.540

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust z-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

The results indicate that the auditors’ adoption of previously unmatched SIFs is more likely

to occur for entities with more previously unmatched SIFs (UNMATCHED_SIF), fewer previously

unmatched KAMs (UNMATCHED_KAM), and when the auditors remove more previously

unmatched KAMs (DROP_UNMATCHED) and add fewer other new KAMs in the current period

(ADD_NEW). The significant negative coefficient on lagged ABAFEE suggests that the auditors’

subsequent adoption of a previously excluded SIF can be a consequence of constraints on the prior

year’s audit effort.

81

4.6.2. Analyses for Audit Quality on SIFxPICK Behavior

The results reported in Table 4.8 show that auditors’ adoption of previously omitted SIFs

is negatively associated with lagged abnormal audit fee, which I suggest may be a consequence of

constraints on the prior year’s audit effort. This raises the prospect that unmatched SIFs may be

associated with lower audit quality. I investigate this issue by testing whether either current

unmatched SIFs or the auditors’ subsequent reporting of any of the current year’s unmatched SIFs

reflects poor audit quality in the current year. I use abnormal accruals to proxy for audit quality, as

follows:

ABSACCit = β0 + β1%UNMATCHED_SIFit + β2%UNMATCHED_SIFit X PICKit+1+ β3SIZEit+

β4ROAit + β5LOSSit + β6MTBit + β7LEVERAGEit + β8PRIOR_ACCit + β9CFOit+

β10SALESVOLit+ β11TOTAL_KAMit+ β12TOTAL_SIFit+ Σβ𝑗industry+ Σβ𝑘year +

Σβ𝑚audit firm +it

(Model 4.5)

I obtain ABSACC by estimating the cross-sectional modified Jones model (Dechow et al.

1995) within two-digit ICB industry groups for each year. A minimum of 15 observations per two-

digit ICB industry is required (Carcello and Li 2013; Reid et al. 2018). I then match each firm-year

observation with another firm from the same two-digit ICB industry code and year with the closet

ROA, and ABSACC is the absolute value of the difference (Kothari et al. 2005). The variables of

interest are %UNMATCHED_SIF (which reflects its current omission by the auditor) and its

interaction with PICK (which reflects its future adoption by the auditor). PRIOR_ACC is the

current accruals in year t-1, measured as net income before extraordinary items plus depreciation

and amortization less operating cash flows, scaled by total assets at end of year t-1. Other variables

are as per Models 4.1, 4.2, and 4.3, as summarized in Appendix 1.

82

The results for Model 4.5 are presented in Table 4.9. The coefficients obtained for

%UNMATCHED_SIF and %UNMATCHED_SIF X PICKt+1 are not significant. The results do not

suggest that unmatched SIFs, whether or not they are subsequently adopted by auditors, are

indicators of poor audit quality. Therefore, while the negative coefficient for lagged abnormal audit

fees in the PICK analysis could have signaled constraints on the prior year’s audit effort, it is not

necessarily a reflection of lower audit quality.

The result is consistent with the proposition that unmatched KAMs and SIFs are a

consequence of ineffective communication between auditors and ACs. It is possible that the

inconsistent disclosures between EARs and EACRs reflect auditors and ACs’ disagreements

regarding risk assessments. If an auditor and AC have different opinions regarding what should be

disclosed as “material risks” but the auditor changes their opinion in the next year and discloses

some previously UNMATCHED_SIF from the EACR, it seems likely that the audit in the year of

disagreement would be of poorer quality. I do not obtain any significant results in audit quality

analysis, thus lending more weight to ineffective prior communication being the more credible

explanation for auditors’ subsequent adoptions of unmatched SIFs.

83

Table 4.9 Regression Results for Abnormal Accruals on SIFxPICK Behavior

ABSACC

%UNMATCHED_SIF -0.002

(-0.18)

%UNMATCHED_SIF X PICKt+1 -0.003

(-0.25)

SIZE 0.002

(1.09)

ROA -0.085*

(1.86)

LOSS -0.027***

(-3.07)

MTB 0.001

(1.50)

LEVERAGE -0.015

(-0.87)

PRIOR_ACC -0.024

(-0.54)

CFO 0.231***

(-5.33)

SALESVOL -0.017

(-1.01)

TOTAL_KAM 0.000

(0.04)

TOTAL_SIF -0.001

(-0.45)

Industry, Year, Audit firm FE Included

N 417

Adjusted R2 0.229 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

4.7 Conclusions

Regulators in the UK have responded to concerns regarding audit and accounting

transparency by requiring external auditors and ACs to identify, according to their own judgments,

the most significant issues in relation to that year’s financial statements. The consistency of issues

84

identified in EARs and EACRs have received substantial attention from standard setters,

accounting practitioners, and investors. In this study, I investigate the influence of AC expertise on

external auditor’s KAM disclosures, and their differences in relation to SIF disclosures by ACs.

The results indicate that having higher representations of accounting and industry expertise

on ACs results in auditors reporting fewer KAMs and fewer unmatched KAMs in EARs. This is

consistent with auditors identifying fewer risks for entities with more expert ACs. In addition, this

also accords with the expectations regarding the influence of AC expertise, based on prior research

that identifies a strong relationship between KAM disclosures and other risk measures (Lennox et

al. 2018) and the substantial literature that shows AC accounting expertise is positively related to

financial reporting quality. In addition, I find that the overall alignment between KAM and SIF

disclosures is enhanced by AC financial (both accounting and supervisory) expertise, evidenced

by reporting fewer unmatched KAMs that are not included in EACRs and omitting fewer SIFs as

KAMs that are reported in EACRs.

The currently available data does not allow me to determine whether auditors and ACs were

able to communicate more effectively when ACs have more accounting and industry expertise, but

the results appear to be consistent with this proposition. I contend that AC expertise can enhance

the consistency between auditors and ACs’ risk disclosures because committee members with

accounting or industry expertise are more likely to achieve consistent opinions regarding complex

accounting-related or industry-specific issues and improve the effectiveness of communications

between ACs and auditors. However, the evidence cannot fully support the contention because the

KAMs or SIFs, including those that are unmatched, are the joints products of both auditors and

AC’s reporting behavior. The additional analysis reveals that auditors are more likely to report

previously unmatched SIFs as KAMs after the ACs acquire more accounting and industry expertise.

85

Given that this behavior is not related to changes in other engagement variables, it seems to support the

contention that auditors are more mindful of the issues identified by an AC if those ACs exhibit more

relevant expertise. While the results suggest that the SIFs that were omitted by auditors in the prior year

but are included in this year’s EAR may reflect insufficient audit effort, I do not have any evidence that

this reflects lower audit quality in the year in which the SIFs were omitted by the auditor.

This study is subject to several limitations. First, owing to data limitation, the effect of

social networks between committee members and auditors is not examined. Social ties, such as the

educational, regional, and family ties between AC members and auditors are likely to affect the

underlying communications, and their acceptance decisions. Second, this study only examines the

identification of KAMs and SIFs in EARs and EACRs, without investigating the detailed

disclosures regarding how auditors or ACs addressed the issues. Large alignment may be arisen

from the information repetition between EARs and EACRs; this could be investigated in future

research. Third, the analysis only investigates the interactions between ACs and auditors. However,

the new reporting regime also allows users to triangulate among information provided by managers,

ACs, and auditors. Future research could examine across managers’ accounting policy disclosures,

EARs and EACRs.

86

CHAPTER 5: Study 3: Year-to-year Extended Auditor’s Report Changes and

Audit Effort

5.1 Introduction

This study investigates the extent to which EAR disclosures are modified from year to year,

and the association between the year-to-year EAR changes and audit effort. The most significant

audit matters in a financial statement audit are required as a key narrative disclosure by regulators

worldwide (EU 2014; FRC 2013a; IAASB 2015; PCAOB 2017). According to auditing standards

(e.g., ISA 701; ISA (UK and Ireland) 700), auditors are required to describe each reported matter

and explain the way the particular matter was addressed in the audit. To enhance audit report

transparency and provide financial statement users with useful and valuable information, EAR

disclosures are expected to be entity-specific and fiscal-period-specific (FRC 2013a; IAASB 2015;

PCAOB 2017). However, it is up to the auditors to determine the breadth and depth of what is

discussed in the reports. The objectives for EARs are unlikely to be achieved if disclosures included

in the EARs are largely unchanged across years.

Although substantial year-to-year EAR modifications are not required under auditing

standards, regulators and investors have expressed concerns about the lack of changes in EARs

over time, especially regarding the changes in KAM identification between years (FRC 2015a,

2016b; PCAOB 2017). KAMs do change from year to year, but limited explanation is given by

auditors, and both regulators and financial statement users have stated that they expect more to be

discovered in this regard (e.g., FRC 2015a; PCAOB 2017). In addition, accounting practitioners

have expressed concerns about repetitive disclosures over time. Auditors have claimed that, while

EARs for the same entity over years inevitably have some similarities, the overuse of standardized

87

wording to describe similar risks and responses makes EARs become longer reports with more

boilerplate language (e.g., PwC 2015). In light of these concerns, it could be useful for policy

makers, regulators, and practitioners to understand the implications of the year-to-year EAR

modifications, from both the wording usage and the KAM identification perspectives.

A prior study of EARs, Smith (2017), concludes that, compared with traditional binary

audit reports, EARs are easier to read and can better capture client-specific audit risks; it bases this

conclusion on increases in readability scores and negative and uncertain tone. Moreover, the varied

reporting quality is noted by financial statement users, such as analysts and banks, as evidenced by

negative associations between readability and analyst forecast dispersion (Smith 2017), and

positive associations between the number of KAMs and loan contracting terms (Porumb et al.

2018). However, no supporting evidence has been found in equity markets (Lennox et al. 2018).

Gutierrez et al. (2018) and Reid et al. (2018) find that the implementation of EARs had no effect

on audit fees. However, the positive associations between audit fees and the EAR length and the

total number of KAMs reported in Gutierrez et al. (2018) suggest that, after the adoption of EARs,

more transparent disclosures may signal greater audit effort. I extend the literature by examining

the extent to which auditors modify their EAR disclosures from year to year, and investigate

whether any changes are a consequence of greater audit effort, as reflected in audit fees.

The year-to-year EAR modifications are captured from both wording usage and KAM

identification perspectives. In particular, textual analysis, including the VSM, the VSM with TF-

IDF weighting function, and Trigram approaches, are used to compute year-to-year document

similarity scores for an entity’s EARs. These measures capture the extent to which individual words

or three-word phrases in EARs are repetitive from period to period. KAM changes are grouped

into three categories: removed, repeated, and added. Although KAM identification has received

88

prior research attention, there is little established understanding of their use over time. It could be

that repeating KAMs from year to year means that these risks are particularly important and that

these issues attract the greatest audit effort. Conversely, they could represent a standardized

approach in auditing. However, under both of the circumstances, more “repeated” KAMs imply

relatively stable economic condition, and I therefore predict no association between repeated

KAMs and audit fee changes. In addition, both “removed” and “added” KAMs indicate greater

larger degree of EAR modifications.

Substantial changes in year-to-year EAR disclosures may arise because of changes in the

economic condition an entity faced or because of auditors seeking higher reporting quality. I argue

that if auditors provide reliable disclosures regarding the client’s risks of material misstatements,

the degree of EAR modifications is expected to be greater for firms facing substantial economic

changes. If that is the case, greater audit effort is expected as auditors cannot rely much on the prior

audit approach and need to substantially revise prior year’s audit plan, which thus leading to

increases in audit fees. Therefore, a positive association between year-to-year EAR modifications

and audit fee changes is expected.

Overall, the empirical results provide evidence that audit fees increase more when auditors

made greater modifications of EAR disclosures in terms of both wording usage and KAM

identifications. Audit fee changes are found to be negatively associated with EAR year-to-year

similarity scores and “repeated” KAMs, and positively associated with “removed” and “added”

KAMs. Further, I find that the “added” KAMs effect is driven by those “new” KAMs that are not

borrowed from the prior year’s EACR.

89

This study contributes to policy development and the audit research literature. First, it

provides timely evidence with regard to the potential effect of this new reporting regime on

auditors’ reporting behavior. I find evidence that disclosing less repetitive and more entity-specific

information means that greater effort needs to be made by auditors. The results have policy

implications and may help regulators in other jurisdictions better monitor the implementation of

EARs, owing to the narrower approach that the FRC has adopted.

Second, by examining the relationship between EAR year-to-year modifications and audit

fee, this study contributes to the audit reporting literature and enriches the ongoing discussion

regarding EARs. I add to the relevant literature by examining auditors’ narrative disclosures from

both KAM identification and language usage perspectives, and testing whether providing less

repetitive information in the EAR is a consequence of greater audit effort.

Third, the study contributes to the literature on textual analysis. Although research

concerned with narrative disclosures has been growing rapidly in recent years (Loughran and

McDonald 2016), very few studies apply textual analysis to audit reports. By analyzing auditors’

wording usage and KAM identification in EARs, this study complements the emerging research

pertaining to auditors’ contextual disclosures.

90

5.2 Literature Review and Hypothesis Development

5.2.1. Prior Studies Related to EARs

Prior studies examining the enhanced audit transparency in the new reporting regime

mainly take two perspectives: KAM identification and language usage in EARs.41 Mixed results

are found in studies focusing on KAM identification. Gutierrez et al. (2018), Reid et al. (2018),

and Lennox et al. (2018) find non-significant effects of the number of KAMs on audit fees, audit

quality and market reaction. Porumb et al. (2018), examining loan contracting terms, find a positive

association between interest rate spread and the number of KAMs, and a negative association

between loan maturity and the number of KAMs pertaining to accounting issues. Their results

suggest that the KAM-related disclosures enhance borrower transparency and this informativeness

is rewarded by banks with more favorable loan contracting terms.

Regarding language usage in EARs, Smith (2017) finds that, compared with traditional

binary audit reports, EARs’ readability score and negative and uncertain tone are larger. Auditors

associated with characteristics, such as industry expertise, Big4, and office size, are found to

provide EARs that are more readable. In addition, Smith (2017) finds that analysts respond to the

language in audit reports, evidenced by a negative association between readability and analyst

forecast dispersion. Consistent with the analysts’ apparent sensitivity to language, Porumb et al.

(2018) find that lenders’ risk perceptions are affected by auditors’ usage of an uncertain tone when

41 See Chapter 2 for a review of prior studies of EARs.

91

discussing KAMs, whereby the presence of more uncertain words weakens the negative effect of

KAMs on loan contracting terms.

I add to this emerging literature by examining the year-to-year variations in audit

transparency from the perspectives of both KAM identification and language usage, and

investigating the association with the underlying audit effort, proxied by audit fees. The combined

results of Gutierrez et al. (2018) and Reid et al. (2018) indicate that the initial adoption of EARs

does not lead to an increase on audit fees, even for entities with relatively long reports, more total

KAMs, more “unmatched” KAMs (those that were not mentioned in that entity’s EACR), and

greater materiality thresholds. However, the positive associations between audit fees and both the

length of EARs and the number of KAMs reported in Gutierrez et al. (2018) suggest that, in the

new reporting regime, more disclosures may signal more audit effort. I extend this stream of

literature by testing whether changes in auditors’ disclosure from year to year are associated with

changes in audit fees.

5.2.2. Prior Studies Related to Audit Fees

Since Simunic (1980), the audit fee studies have consistently found that audit fees are

positively associated with firm size, client complexity and engagement risk. Factors such as the

number of subsidiaries, litigation risk, and regulatory requirements, are found to determine the

amount of effort that auditors spend in an audit and thus lead to higher audit fees (e.g., Kim et al.

2012; Simunic 1980; Simunic and Stein 1996). According to Hay, Knechel, and Wong (2006),

those factors can be grouped into client attributes, auditor attributes, and engagement attributes.

Regarding client attributes, client size, entity complexity, inherent risk, profitability, and

leverage are found to affect audit fees significantly (e.g., Francis 2011; Simunic 1980). Entity size,

92

which is the most dominant determinant of audit fees, is found to be positively associated with

audit fees. The more complex an entity, the more time the audit is likely to take, resulting in higher

audit fees. Areas that are difficult to audit and are likely to contain errors, such as inventory and

accounts receivable, require specialized audit procedures and they are positively related to audit

fees. Profitability and leverage, which capture the extent to which the auditor is exposed to loss,

are also found to affect audit fees, as the worse an entity is performing, the more audit work and

therefore higher audit fees are likely to occur. Regarding auditor attributes, prior studies (e.g., Choi,

Kim, Liu, and Simunic 2008; Ireland and Lennox 2002) find that fee premiums are associated with

Big4 auditors. Choi et al. 2008) argue that because the potential legal liability cost for Big 4 auditors

is higher, they have a greater incentive to increase audit effort, leading to higher audit fees. With

respect to engagement attributes, a positive association between the busy season and audit fees is

found in the literature because of the higher likelihood of working overtime then (Hay et al. 2006).

Overall, the literature finds that the audit fee is a function of audit effort and is increased

with the complexity of the audit work. Audit fees increase when auditors are required to make

estimates that are more complex and to use greater professional judgment, and decrease when

auditors can plan and process the audit more efficiently, as audit time is saved and the decreased

audit effort is reflected in lower audit fees (e.g., Kim et al. 2012; Zhang 2018).

Prior studies related to auditors’ narrative disclosures and audit fees are very few. For

EARs, they are limited to Gutierrez et al. (2018) and Reid et al. (2018), the two studies investigating

the implementation effect of EARs on audit fees, as discussed earlier.

Yang, Yu, Liu, and Wu (2018), a relevant non-EAR study, find that audit fees are positively

associated with corporate risk disclosures. Their study uses textual analysis, specifically, a Natural

93

Language Processing technique, to capture entity-specific risk disclosures. Based on the risk

management standards of Institute of Risk Management (2002), Yang et al. (2018) examine client

self-identified risk disclosures from financial, strategic, operational, and hazard perspectives, and

find that the audit fee is positively associated with all the four individual risk disclosures and an

overall risk score that is generated using the four risk types. My study focuses on auditors’

identified risk disclosure, in contrast to Yang et al. (2018), who investigate entities’ self-disclosed

risk.

5.2.3. Hypothesis Development

Prior studies of EARs find that riskier, more complex and larger entities have comparatively

more KAMs discussed in their EARs (Lennox et al. 2018), and that banks incorporate this into

their risk assessments (Porumb et al. 2018). These findings imply that auditors’ disclosures in

EARs reliably capture client firms’ financial reporting risks. This arguments is also partially

supported by prior research of corporate disclosures. The literature concerned with managers’

disclosures suggests a positive association between risk factor disclosures and firm risks (e.g.,

Brown and Tucker 2011; Campbell et al. 2014). For example, Campbell et al. (2014), using thirteen

proxies to capture risks an entity faced, report that managers increase their disclosures in the “risk

factor” Section of the 10-K file when firms experience greater expected returns, higher leverage,

larger stock return volatility and turnover, have a Big N auditor, have greater analyst following,

experience lower profitability, lower effective tax rates, and lower stock return skewness.

Moreover, Brown and Tucker (2011) find that managers modify entities’ Management Discussion

and Analysis (MD&A) disclosures more from the prior years’ when entities are facing substantial

changes in operations, liquidity, capital resources, risk exposure, and business components.

94

Based on the aggregate findings of these prior studies, I argue that, if the EAR disclosures

reliably reflect the audit concerns, then greater modifications of EARs disclosures from year to

year should be related to greater variations in the risks faced by an entity. Identifying, evaluating,

and addressing these risks should increase audit complexity and audit time, resulting in higher audit

fees. Consistent with this argument, there is evidence that audit fees are relatively higher for entities

that have more KAMs disclosed in EARs (Gutierrez et al. 2018).

If a company experiences relatively stable underlying economic fundamentals and

operating conditions over time, then the auditor might not need to make substantial revisions to the

existing audit plan and might rely more on its prior approach to conduct the audit. This may be

reflected in more consistent risk disclosures, including fewer wording changes and more “repeated”

KAMs reported in the next year’s EARs. If, however, the auditor identifies substantial changes in

engagement risks, more audit effort is expected, as argued above. Because the auditor needs to

revise the previous audit plan to a larger extent and can place less reliance on the prior audit

approach. Consequently, I expect these engagements to have greater EAR modifications, reflected

as more wording changes and more “removed” and “added” KAMs in the following year’s EARs.

The above arguments lead to the following hypothesis:

HYPOTHESIS. Year-to-year EAR modifications are positively associated with audit fee

changes.

Testing the predicted positive association between year-to-year EAR modifications and

audit fee changes might be confounded if substantial changes in EAR disclosures are strategically

used by auditors. If auditors modify their disclosures in EARs as a deliberate differentiation or

obfuscation strategy that did not involve more audit effort, the analysis is less likely to reveal

95

significant associations. Furthermore, fewer year-to-year EARs modifications might occur because

auditors invest less effort in their reporting behavior, irrespective of their underlying audit effort.

If the extra work auditors conducted to address their concerns in relation to companies’ substantial

economic changes are not reflected in their EAR disclosures, the expected association between

year-to-year EAR modifications and changes in audit fees might not be evident.

5.3 Method

5.3.1. Measurement of Year-to-year EAR Modifications

The modification aspect of EARs is captured from both wording usage and KAM

identification perspectives. In particular, the VSM (with and without the TF-IDF weighting

function) and Trigram methods are used to measure the extent to which auditors’ disclosures in

EARs are changed from year to year, for both KAM sections and the full EARs. Reported KAMs

are grouped into three categories: repeated, removed, and added.

5.3.1.1. Textual Analysis Methods

I employ the VSM method to compute the similarity score of a particular company’s EARs

in two consecutive years for both KAM sections and the full report. The VSM has been widely

used in accounting research (see Section 3.3.1 for a detailed discussion). For example, Brown and

Tucker (2011) adopt the method to capture the extent to which firms modify the management

discussion and analysis (MD&A) section of the 10-K filing from year to year. Peterson et al. (2015)

apply the VSM to evaluate the extent to which an entity’s accounting policy disclosures are

consistent over years.

In the VSM, the vector for an EAR (or the KAM section) of company 𝑖 is represented as:

96

vi = (w1, w2, …, wm-1, wm)

(Model 5.1)

where w is the frequency of (each) word in the full EAR (or KAM section) for company i.

To address the concern that all words should be treated equally under the VSM, I follow

prior studies, such as Brown and Tucker (2011) and Hoberg and Phillips (2016), and modify the

calculation of similarity scores in the traditional VSM approach by weighting words using the TF-

IDF approach (see Section 3.5.1 for a detailed discussion). The TF-IDF weighting, expressed as

[wmilog(N/nm)], assigns greater weight to words that are used less frequently and lesser weights

to common words, with 0 weight for a word that appears in every document.

Both the traditional VSM and the VSM with TF-IDF weighting use the same model to

compute the similarity score. The similarity degree between EARs, vi and vj, is stated as:

Similarityij = (vi×vj) / (ǁ𝑣iǁ × ǁ𝑣jǁ)

(Model 5.2)

where vi×vj yields the scalar product of vi and vj, and ‖𝑣𝑖‖ and ‖𝑣𝑗‖ represent the vector lengths.

The resulting variables, SIMKAM (SIMEAR) and SIMKAM_IDF (SIMEAR_IDF) are the

test variables. They are the similarity scores between a company’s current year KAM section (full

EAR) and that for the previous year, estimated using the traditional VSM and VSM with TF-IDF

weighting function, respectively.

In addition, following prior accounting research (e.g., Lang and Stice-Lawrence 2015;

Nelson and Pritchard 2007), the Trigram approach is adopted to compute the year-to-year KAM

97

and EAR modifications, based on the similarity of three-word phrases across the two reports.

Consistent with Study 1 (see Section 3.5.1 for a detailed discussion), each KAM section and full

EAR are converted into sets of overlapping trigrams. Then the similarity score between the EARs

of companies i and j is computed by dividing the intersection of the two sets (one for each company)

by the union of the two sets of trigrams, as follows:

Similarityij = |S(i)∩S(j)| / |S(i)∪S(j)|

(Model 5.3)

where S(i) and S(j) represent the sets of trigrams for KAM(i) and KAM(j), or EAR(i) and EAR(j),

respectively.

The resulting variables are therefore denoted as SIMKAM_Tri and SIMEAR_Tri.

5.3.1.2. KAM Categories

I capture the modification aspect of EARs from year to year by focusing on specific KAM

identifications. By comparing the current year KAMs with the previous year, I group KAMs into

three categories: removed, repeated, and added. “Removed” refers to those KAMs that are included

in the prior year’s EAR but are removed in the current year’s EAR. “Repeated” refers to those

KAMs that are discussed in both the current and prior years’ EARs. “Added” refers to those KAMs

that are discussed in the current year’s EAR, but not that of the prior year. The test variables,

%REMOVE, %REPEAT, and %ADDED are measured as the number of “removed,” “repeated,”

and “added” KAMs, scaled by the total KAMs in current year’s EAR.

98

5.3.2. Models for Testing the Hypotheses

Model 5.4, which follows Reid et al. (2018), is estimated to examine the hypothesis, stated

as follows:

ΔAFEE = β0+ β1ΔEAR_CHG+ β2ΔSIZE+ β3ΔMTB+ β4ΔCFO+ β5ΔINV+ β6ΔREC+

β7ΔSALESVOL+ β8LOSS+ β9ΔLEVERAGE+ β10BUSY+ β11ΔROA+ β12AFEE(t-1) +

Σβ𝑗industry+ Σβ𝑘year

(Model 5.4)

where ΔAFEE is the dependent variable, which is the annual change in log of audit fees; and

ΔEAR_CHG, the variable of interest, represents each of SIMKAM, SIMEAR, SIMKAM_IDF,

SIMEAR_IDF, SIMKAM_Tri, SIMEAR_Tri, %REMOVE, %REPEAT, and %ADDED.

In line with the hypothesis, I expect the coefficient of ΔEAR_CHG to be negative when it

represents SIMKAM, SIMEAR, SIMKAM_IDF, SIMEAR_IDF, SIMKAM_Tri, SIMEAR_Tri, or

%REPEAT, and I expect the coefficient of ΔEAR_CHG to be positive when it represents

%REMOVE or %ADDED. In addition, following Reid et al. (2018), I control for firm size (SIZE),

market-to-book ratio (MTB), cash flow from operations (CFO), inventory (INV), receivables

(REC), sales volatility (SALESVOL), loss (LOSS), leverage (LEVERAGE), busy season (BUSY),

and return on assets (ROA).42 All the control variables, except for LOSS and BUSY, are calculated

in change form, which is the value in year t minus that of the same variable in year t-1. Since the

prior year’s audit fees can largely explain this year’s audit fees, I control for year t-1’s audit fees

as well [AFEE(t-1)] (Zhang 2018). Industry and year fixed effects are included in the model. For

42 I do not control for changes in Big4 auditors, because no company changed from using a Big4 auditor to a non-Big4

auditor (or vice versa) during the study period.

99

convenience, all variable definitions are tabulated in Appendix 1. Consistent with prior studies, I

cluster standard errors by company for all the tests (Gutierrez et al. 2018; Reid et al. 2018).

5.3.3. Sample Selection

The sample is selected from all UK non-financial entities that reported on the application

of the UK Corporate Governance Code for the fiscal years between September 30, 2013 and

September 30, 2016.43 I start with all available annual reports of relevant companies, which allows

me to capture any EAR disclosure change from the prior year, yielding 1,244 firm-year

observations. I then manually collect the observations’ audit fee data; 14 observations are

eliminated during this process. These are then merged with DataStream to compute control

variables and I lose 391 observations because of the missing data. The final sample for the

hypothesis test is comprised of 839 firm-year observations, as shown in Table 5.1.44

43 Two early adopters of EARs are included in the sample and the results remain unchanged after removing them from

the analyses. 44 The final sample size is reasonable, compared with other current EAR studies in the UK. For example, Gutierrez et

al. (2018) and Reid et al. (2018) report 872 and 1,088 firm-year observations, respectively, for four-year period pre-

post analyses.

100

Table 5.1 Sample Selection

Firm-year

Observations

Total available annual reports of relevant companies with fiscal years ending

between September 30, 2014 and September 30, 2016 (used to capture annual

change of EAR disclosures) 1,244

Less: Missing observations to compute audit fee data (14)

Less: Missing observations when merging with DataStream to compute control

variables (391)

Final Sample 839

5.4 Results

5.4.1. Descriptive Statistics

Descriptive statistics for the main variables are presented in Table 5.2.45 The average

change of audit fees in the natural log format is −0.043 and it is expected that the mean and median

of ΔAFEE would be centered around zero, as audit fees are relatively stable over time. The mean

of lagged AFEE [AFEE(t-1)] is 12.707, consistent with 13.084 in Gutierrez et al. (2018) and 13.485

Reid et al. (2018). The mean values of KAM-related scores (SIMKAM, SIMKAM_IDF, and

SIMKAM_Tri) are consistently lower than EAR-related scores (SIMEAR, SIMEAR_IDF, and

SIMEAR_Tri). This is expected because disclosures in the KAM section are expected to be the

section of greatest modifications. Similarity scores computed using the VSM (without TF-IDF)

measure (SIMKAM and SIMEAR) are similar to those using the VSM with TF-IDF weighted

measure (SIMKAM_IDF and SIMEAR_IDF) in terms of scale and variability; while the scores

45 To mitigate the effect of outliers, all continuous variables are winsorized at 0.01 (Francis and Yu 2009). As a

robustness check, I estimate the models without winsorizing and do not observe any significant differences in the

regression results.

101

using the Trigram measure (SIMKAM_Tri and SIMEAR_Tri) exhibit much more variability. In the

sample, around 18.2 per cent of year t-1’s communicated KAMs are removed in the year t’s EARs

(%REMOVE). About 84.2 per cent KAMs are discussed in EARs in two consecutive years

(%REPEAT), while only 7.0 per cent of year t’s KAMs are newly added (%ADDED).

Table 5.2 Descriptive Statistics

Variable N Mean SD p25 p50 p75

ΔAFEE 839 -0.043 0.650 -0.007 0.000 0.083

SIMKAM 839 0.838 0.116 0.761 0.870 0.930

SIMEAR 839 0.955 0.029 0.941 0.962 0.976

SIMKAM_IDF 839 0.822 0.124 0.746 0.852 0.921

SIMEAR_IDF 839 0.950 0.033 0.937 0.957 0.972

SIMKAM_Tri 839 0.391 0.239 0.176 0.369 0.570

SIMEAR_Tri 839 0.546 0.160 0.428 0.536 0.661

%REMOVE 839 0.182 0.280 0.000 0.000 0.333

%REPEAT 839 0.842 0.223 0.667 1.000 1.000

%ADDED 839 0.070 0.451 0.000 0.000 0.333

ΔSIZE 839 0.060 0.182 -0.017 0.048 0.122

ΔMTB 839 -0.140 1.980 -0.236 -0.009 0.128

ΔCFO 839 0.000 0.049 -0.016 0.000 0.014

ΔINV 839 -0.001 0.015 -0.001 0.000 0.000

ΔREC 839 -0.002 0.033 -0.008 0.000 0.006

ΔSALESVOL 839 0.003 0.062 -0.012 0.000 0.018

LOSS 839 0.184 0.387 0.000 0.000 0.000

ΔLEVERAGE 839 0.007 0.060 -0.009 0.000 0.019

BUSY 839 0.430 0.495 0.000 0.000 1.000

ΔROA 839 -0.015 0.103 -0.047 -0.002 0.024

AFEE(t-1) 839 12.707 1.899 10.758 12.899 13.998

Variable definitions are provided in Appendix 1.

102

On average, the annual increase in company size is 0.060 (SIZE), while the market-to-book

ratio (MTB) drops 0.140 annually. Changes in company cash flow from operations (CFO),

inventory (INV), receivables (REC), sales volatility (SALESVOL), leverage (LEVERAGE), and

return on assets (ROA) are quite stable, with the mean values centered on 0.

5.4.1.3.Pearson Correlations

Pearson correlations among all regression variables are reported in Table 5.3. There are

high correlations between each VSM measure (either with or without the TF-IDF weighting

function) and the corresponding Trigram measure. For example, the correlation coefficient of

SIM_KAM and SIM_KAM_IDF is 0.97, suggesting I am likely to observe similar results for the

different measures. High correlations among %REMOVE, %REPEAT, and %ADDED are expected

because, all else being equal, more KAMs should be repeated when fewer prior year’s KAMs are

dropped (-0.91) and fewer new KAMs are added (−0.67) to the current year’s EAR. Similarly,

when more prior year’s KAMs are dropped, more new KAMs are likely to be added to this year’s

EARs (0.76). Regarding the correlations among the control variables, no particular concerns are

raised by the pairwise correlation. To assess multicollinearity concerns for each regression, VIFs

are computed. The highest VIF was 2.03, far below the level of 10, which do not suggest any

multicollinearity problems (Neter et al. 1996).

103

Table 5.3 Pearson Correlation Matrix

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

1 ΔAFEE 1.0000

2 SIMKAM -0.0083 1.0000

3 SIMEAR -0.0506 0.6503* 1.0000

4 SIMKAM_IDF -0.0116 0.9689* 0.6364* 1.0000

5 SIMEAR_IDF -0.0499 0.5356* 0.9334* 0.5530* 1.0000

6 SIMKAM_Tri -0.0106 0.8702* 0.6868* 0.8588* 0.6010* 1.0000

7 SIMEAR_Tri -0.0533 0.6972* 0.8105* 0.6956* 0.7429* 0.8662* 1.0000

8 %REMOVE 0.0348 -0.5217* -0.4476* -0.5401* -0.3823* -0.4314* -0.4076* 1.0000

9 %REPEAT -0.0292 0.5560* 0.4664* 0.5764* 0.3910* 0.4762* 0.4597* -0.9055* 1.0000

10 %ADDED 0.0350 -0.2223* -0.2512* -0.2491* -0.2198* -0.1718* -0.1928* 0.7592* -0.6728* 1.0000

11 ΔSIZE 0.0613 -0.0397 -0.0583 -0.0396 -0.0692* -0.0634 -0.0528 0.0404 -0.0600 0.0337 1.0000

12 ΔMTB -0.0298 -0.0016 0.0384 -0.0018 0.0385 0.0014 0.0171 -0.0486 0.0394 -0.0432 -0.0510 1.0000

13 ΔCFO 0.0097 0.0670 0.0626 0.0603 0.0595 0.0863* 0.0799* -0.0862* 0.0979* -0.0997* -0.1315* -0.0723* 1.0000

14 ΔINV 0.0197 0.0228 0.0107 0.0205 0.0112 0.0321 0.0255 -0.0109 0.0030 -0.0387 -0.1280* 0.0344 -0.2199* 1.0000

15 ΔREC -0.0285 0.0091 0.0460 0.0212 0.0566 0.0064 0.0098 -0.0390 0.0155 -0.0440 -0.1845* 0.1189* -0.0678* 0.1153* 1.0000

16 ΔSALESVOL 0.0010 -0.0210 -0.0452 -0.0212 -0.0328 0.0094 -0.0166 0.0351 -0.0173 0.0577 0.0352 -0.0130 -0.0720* 0.0609 0.0075 1.0000

17 LOSS 0.0143 -0.0416 -0.1104* -0.0348 -0.1047* -0.0407 -0.0654 0.0656 -0.0385 0.0638 -0.3662* 0.0518 -0.0263 0.0565 0.0190 0.0537 1.0000

18 ΔLEVERAGE 0.0512 -0.0184 -0.0739* -0.0245 -0.1030* -0.0387 -0.0543 0.0517 -0.0438 0.0569 0.2308* 0.1229* -0.1869* -0.0516 -0.0739* 0.0083 0.1104* 1.0000

19 BUSY 0.0379 0.0060 -0.0504 -0.0116 -0.0848* -0.0466 -0.0829* 0.0374 -0.0146 0.0379 -0.0343 0.0868* -0.0563 0.0319 -0.0278 -0.0012 0.0419 0.1200* 1.0000

20 ΔROA -0.0118 0.0125 -0.0181 -0.0074 -0.0077 0.0120 -0.0102 -0.0048 -0.0021 -0.0132 0.2470* -0.0219 0.1029* -0.0288 -0.0201 -0.0513 -0.3669* -0.1436* -0.0026 1.0000

21 AFEE(t-1) -0.2049* 0.0055 -0.2218* -0.0365 -0.2886* -0.1340* -0.2575* 0.0978* -0.1312* 0.0810* 0.0431 -0.0520 -0.0824* -0.0620 -0.0305 -0.0060 -0.0761* 0.0921* 0.3226* 0.0945* 1.0000

Variable definitions are provided in Appendix 1.

104

5.4.2. Regression Results

The regression results examining annual changes in auditors’ wording usage and KAM

identifications are reported in Panels A and B of Table 5.4, respectively. I estimate Model 5.4 with

each of SIMKAM, SIMEAR, SIMKAM_IDF, SIMEAR_IDF, SIMKAM_Tri, and SIMEAR_Tri as the

test variables, as reported in Panel A of Table 5.4. 46 Each measure of textual similarity is

significantly negatively related to audit fee changes, supporting the hypothesis. The results suggest

that greater changes in auditors’ wording usage from the prior year, implying more entity-specific

and fiscal-period-specific information, signal higher audit effort. It is worth noting that similarity

scores captured over full EARs have stronger effects on audit fees, compared with those captured

in KAM sections only. This could be because the UK auditing standards require auditors to disclose

additional information on not only KAMs, but also on materiality and audit scope (FRC 2013a).47

Sections other than KAMs have significant variations over time as well and receive additional audit

effort.

46 The results indicate that most control variables are non-significant in relation to the changes in audit fees. To check

the relevance of these variables to audit fees for this sample, I estimate the basic audit fee model (AFEE =

β0+β1SIZE+β2MTB+β3CFO+β4INV+ β5REC+ β6SALESVOL+ β7LOSS+β8LEVERAGE+β9BUSY+β10ROA+Σβj

industry+ Σβk year), and obtain significant coefficients on SIZE, MTB, INV, REC, BUSY, and ROA, as expected. The

model provides a good fit with an adjusted R-squared of 0.829. 47 However, further investigation of materiality and audit scope effects is not practical at this stage and is left to further

research.

105

Table 5.4 Regression Results

Panel A: Regression Results (SIMKAM, SIMEAR, SIMKAM_IDF, SIMEAR_IDF,

SIMKAM_Tri, and SIMEAR_Tri )

ΔAFEE

(1) (2) (3) (4) (5) (6)

SIMKAM -0.348* - - - - -

(-1.86)

SIMEAR - -2.344*** - - - -

(-2.68)

SIMKAM_IDF - - -0.354** - - -

(-2.01)

SIMEAR_IDF - - - -2.061*** - -

(-2.65)

SIMKAM_Tri - - - - -0.209** -

(-2.08)

SIMEAR_Tri - - - - - -0.450***

(-2.85)

ΔSIZE 0.237 0.210 0.238 0.205 0.225 0.213

(1.61) (1.46) (1.62) (1.43) (1.53) (1.47)

ΔMTB -0.009 -0.009 -0.010 -0.009 -0.010 -0.010

(-1.32) (-1.27) (-1.33) (-1.32) (-1.34) (-1.36)

ΔCFO -0.005 -0.019 -0.009 -0.036 0.011 0.014

(-0.01) (-0.04) (-0.02) (-0.07) (0.02) (0.03)

ΔINV -0.033 -0.118 -0.046 -0.164 -0.017 -0.038

(-0.03) (-0.11) (-0.04) (-0.16) (-0.02) (-0.04)

ΔREC -0.192 -0.150 -0.177 -0.138 -0.210 -0.220

(-0.35) (-0.27) (-0.32) (-0.25) (-0.38) (-0.40)

ΔSALESVOL 0.437 0.404 0.433 0.417 0.460 0.432

(1.51) (1.41) (1.50) (1.45) (1.59) (1.49)

LOSS -0.021 -0.043 -0.021 -0.042 -0.024 -0.034

(-0.34) (-0.68) (-0.34) (-0.67) (-0.39) (-0.56)

ΔLEVERAGE 0.186 0.166 0.186 0.152 0.193 0.199

(0.54) (0.48) (0.54) (0.44) (0.56) (0.58)

BUSY 0.127** 0.131** 0.127** 0.131** 0.127** 0.129**

(2.35) (2.42) (2.35) (2.42) (2.35) (2.39)

ΔROA -0.343 -0.367 -0.351 -0.353 -0.335 -0.348

(-1.26) (-1.36) (-1.28) (-1.31) (-1.23) (-1.29)

AFEE(t-1) -0.251*** -0.259*** -0.252*** -0.261*** -0.255*** -0.260***

(-5.72) (-5.79) (-5.72) (-5.80) (-5.75) (-5.81)

Constant 3.264*** 5.303*** 3.272*** 5.052*** 3.099*** 3.329***

(5.52) (4.63) (5.58) (4.73) (5.70) (5.78)

Industry, Year FE Included Included Included Included Included Included

N 839 839 839 839 839 839

Adjusted R2 0.234 0.241 0.235 0.240 0.236 0.242

106

Panel B: Regression Results (%REMOVE, %REPEAT, and %ADDED)

ΔAFEE

(1) (2) (3)

%REMOVE 0.195*** - -

(3.05)

%REPEAT - -0.202** -

(-2.33)

%ADDED - - 0.122***

(2.96)

ΔSIZE 0.235 0.231 0.238

(1.60) (1.56) (1.62)

ΔMTB -0.008 -0.009 -0.008

(-1.17) (-1.25) (-1.22)

ΔCFO 0.023 0.015 0.056

(0.05) (0.03) (0.11)

ΔINV -0.051 -0.101 0.079

(-0.05) (-0.10) (0.07)

ΔREC -0.159 -0.198 -0.153

(-0.28) (-0.35) (-0.27)

ΔSALESVOL 0.423 0.437 0.407

(1.46) (1.51) (1.40)

LOSS -0.029 -0.023 -0.026

(-0.47) (-0.38) (-0.43)

ΔLEVERAGE 0.165 0.182 0.171

(0.48) (0.53) (0.49)

BUSY 0.125** 0.130** 0.125**

(2.29) (2.39) (2.29)

ΔROA -0.357 -0.345 -0.355

(-1.32) (-1.27) (-1.31)

AFEE(t-1) -0.255*** -0.254*** -0.255***

(-5.76) (-5.73) (-5.78)

Constant 2.978*** 3.171*** 3.013***

(5.68) (5.76) (5.74)

Industry, Year FE Included Included Included

N 839 839 839

Adjusted R2 0.238 0.235 0.238

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics adjusted for firm clustering effects are reported below coefficient estimates. Variable definitions are provided

in Appendix 1.

The regression results investigating KAM changes over time are presented in Panel B of

Table 5.4. Audit fee changes are significantly positively associated with %REMOVE and

%ADDED KAMs (at p-values < 0.01), and negatively related to %REPEAT KAMs (at p-values

107

< 0.05). The magnitude of increases in audit transparency, reflected as greater changes in risk

assessments and accounting judgments, is associated with additional audit effort, supporting the

hypothesis.

Taken together, the evidence supports the hypothesis that audit fees increase more when

auditors make greater modifications on EAR disclosures in terms of both wording usage and KAM

identifications. The results suggest that the inclusion of entity-specific and fiscal-period-specific

information could be a consequence of extra audit effort, or the issues pertaining to the changes

required more audit effort.

5.5 Robustness Test

5.5.1. Analyses Using Cross-section Similarity Measures

In the main analysis, to capture the year-to-year EAR modifications, I compare auditors’

wording usage over time. I further repeat the analysis using the cross-sectional measures adopted

in Study 1. Because disclosures that are provided in most EARs are likely to be generic and so

convey less firm-specific or fiscal-period-specific information, I investigate the relationship

between the changes in the EAR cross-section similarity scores and audit fees to test the extent to

which standardization across the client’s EARs is a consequence of deficiencies in the audit effort.

Consistent with the measures introduced in Chapter 3, the traditional VSM, the VSM with

TF-IDF weighting function, and the Trigram methods are used to compute the EAR similarity,

yielding BaseSIMKAM, BaseSIMEAR, BaseSIMKAM_IDF, BaseSIMEAR_IDF,

BaseSIMKAM_Tri, and BaseSIMEAR_Tri. The annual changes in these are the test variables in the

test. To address the concern that the standardized disclosures occurred because of accounting

108

comparability, I modify the “Base” similarity scores based on industry comparability effects, using

both residual and differencing methods. Following the same procedures as in Study 1, I estimate

SIM_KAMRES, SIM_KAMRES_IDF, SIM_KAMRES_Tri, SIM_EARRES, SIM_EARRES_IDF, and

SIM_EARRES_Tri using the residual method, and SIM_KAMDIF, SIM_KAMDIF_IDF,

SIM_KAMDIF_Tri, SIM_EARDIF, SIM_EARDIF_IDF, and SIM_EARDIF_Tri using the

differencing method. I then calculate the year-to-year differences for these cross-sectional

measures (within auditors’ client portfolios), and use these change variables as the test variables in

the change in audit fee model. For all these disclosure change variables, positive coefficients are

expected.

The regression results using base similarity scores, residual adjusted scores and

differencing adjusted similarity scores are reported in Panels A, B, and C of Appendix 10,

respectively. None of the coefficients for these variables are significant. The non-significant

coefficients may be resulted from the lack of variation among cross-section measures from year to

year. For example, the mean value and the standard deviation of ΔSIM_KAMRES are 0.000 and

0.035, respectively. Nonetheless, the results do not suggest that annual changes in standardization

across EARs are related to changes in audit effort.

5.6 Additional Analysis

5.6.1. Analysis on Added KAMs

The main analysis finds that audit fees increased more from the previous year’s fees when

more KAMs are added in the current year’s EAR. To gain a better understanding of the KAM

change, I further investigate whether the effect of “added” KAMs is mainly driven by the “newly

added” KAMs. It is worth noting that among those KAMs added in current year’s EARs, some of

109

them are prior year’s “unmatched” KAMs (those that were discussed in that entity’s EACR, but

not in their EARs). If newly disclosed KAMs are “borrowed” from the AC’s prior disclosures, then

the audit production cost and the amount of extra effort the auditors needed to spend is expected to

be lower. I therefore predict that the positive association found between changes in audit fees and

additional KAMs is driven by the “newly added” KAMs.

Particular attention is paid to the communications between the ACs and the auditors. This

is because, although the approaches to determine what KAMs auditors should disclose differ in

detail across jurisdictions, expecting disclosed KAMs to be selected from matters that were

communicated with ACs is consistent, and early and effective engagements between external

auditors and ACs are critical in this new reporting regime (Deloitte 2016a; IAASB 2015; PCAOB

2017). Moreover, because the UK requires both auditors and ACs to report material matters in their

respective extended reports, I am able to investigate the “borrowed” and “newly added” KAMs

separately.

I re-estimate Model 5.4 using “%BORROW” and “%ADD_NEW” as the test variables. The

results are reported in Table 5.5. Significant positive coefficients are obtained on “%ADD_NEW”

(at p-values < 0.01), but not “%BORROW.” The results imply that the “new” KAMs added in the

current period’s EARs receive more audit effort than those prior “ignored” matters that were

disclosed in AC reports (as reflected in greater audit fee increases).

110

Table 5.5 Regression Results for Analysis on Added KAMs

ΔAFEE

(1) (2)

%BORROW 0.122 -

(0.87)

%ADD_NEW - 0.124***

(2.73)

ΔSIZE 0.240 0.240

(1.63) (1.63)

ΔMTB -0.010 -0.009

(-1.39) (-1.28)

ΔCFO -0.049 0.058

(-0.10) (0.11)

ΔINV -0.141 0.086

(-0.14) (0.08)

ΔREC -0.247 -0.103

(-0.44) (-0.19)

ΔSALESVOL 0.454 0.402

(1.57) (1.38)

LOSS -0.016 -0.024

(-0.26) (-0.39)

ΔLEVERAGE 0.182 0.180

(0.52) (0.52)

BUSY 0.130** 0.124**

(2.40) (2.26)

ΔROA -0.340 -0.353

(-1.24) (-1.30)

AFEE(t-1) -0.251*** -0.256***

(-5.68) (-5.79)

Constant 2.962*** 3.023***

(5.63) (5.75)

Industry, Year FE Included Included

N 839 839

Adjusted R2 0.231 0.237 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics adjusted for firm clustering effects are reported below coefficient estimates. Variable definitions are provided

in Appendix 1.

111

5.7 Conclusions

Regulators have responded to concerns over the limited information content in the

traditional auditor’s report by developing new audit reporting standards that require disclosures of

the most significant audit matters in that year’s financial statement audit (EU 2014; FRC 2013a;

IAASB 2015; PCAOB 2017). EARs are intended to increase the transparency of the underlying

audit work (FRC 2016a). However, I observe a seemingly low level of changes in EARs for my

sample period. The generalization is that EARs may be less informative if it looks very similar to

that of the previous year.

This study focuses on auditors’ year-to-year modification in EARs during the UK’s first

three years of EARs implementation and examines whether the year-to-year variations in auditors’

disclosures are associated with changes in the underlying audit effort. Changes in EAR disclosures

are captured from both wording usage and KAM identification perspectives. Overall, I find

supporting evidence for the proposition that audit effort increases more when greater modifications

to EAR disclosures are made, associated with both wording usage and KAM identifications. Year-

to-year similarity scores are negatively related to audit fee changes and this result is robust to

measures using the VSM (with and without the TF-IDF weighting function) and Trigram methods.

Audit fee changes are negatively associated with “repeated” KAMs, and positively associated with

“removed” and “added” KAMs. Further, I find that “added” KAMs effect is driven by those “new”

KAMs identified by the auditor that are not “borrowed” from the prior year’s AC report. The results

indicate that there is more audit effort when EARs are modified to a larger extent.

112

CHAPTER 6: Conclusion

Regulators have responded to concerns regarding the limited information that traditional

auditors’ reports can provide by adopting enhanced reporting standards that require auditors to

disclose the most significant audit matters in that year’s audit (EU 2014; FRC 2013a; IAASB 2015;

PCAOB 2017). The approach that standard setters have taken to enhance audit transparency is

similar across jurisdictions. That is particularly evident in requiring auditors to provide more

information on audit-related matters, and the emphasis placed on communications with the AC

(e.g., IAASB 2017). The UK was the first jurisdiction to implement EARs and the only jurisdiction

that also requires ACs to report significant risks in their EACRs. The UK requirements thus enable

me to examine differences in EARs over time, and to compare auditors’ and ACs’ disclosures.

Because the FRC’s approach to audit reports is similar to that of the EU standard and narrower

than that of the IAASB and the PCAOB standards, and because most jurisdictions have similar

requirements for communications between auditors and ACs, I believe the results found in all of

my three studies can be generalized to other jurisdictions.

The extent of, or potential for, generic disclosures in EARs have received particular

attention in this new reporting regime, because having more engagement-specific (and less-

generic) disclosures in EARs is critical to enhancing audit report transparency. The extent to which

auditors use standardized language in communicating audit matters, and whether the variations

allow financial statement users to differentiate audit quality across audit engagements were

examined in the first study reported in this thesis. By using a variety of measures, textual

similarities (within each auditor’s engagement portfolio in a given year), in both KAM sections

and full EARs, were computed to capture the use of standardized wording. To address the concern

that the standardization captured is a desirable consequence of “comparability”, I also developed

113

two methods for adjusting the similarity scores for industry comparability effects. The results show

that, for measures based on the VSM (with and without the TF-IDF weighting function), the

adjusted similarity scores are positively related to the absolute abnormal accruals, which are used

as the proxy for audit quality, and the relations are stronger when analyzing KAM sections than

for full EARs. This is further supported by finding a negative association between high abnormal

audit fees and the similarity scores of KAM sub-sections. In addition, by revisiting a prior study’s

analysis, I found that, although the audit quality appears to have improved following the EAR

implementation, negative relations between EAR similarity scores and audit quality persist.

Overall, the findings support the proposition that the level of standardized disclosures included in

EARs is a consequence of the underlying audit quality differentiations.

In determining which audit matters should be disclosed in EARs, matters that are discussed

with ACs are paramount. In addition, ACs themselves may affect auditors’ risk assessments. The

UK provides a unique opportunity to examine the influence of ACs on auditors’ KAM disclosures,

as auditors and ACs are required to disclose material risks in their respective extended reports

simultaneously. By comparing their reported risks, the differences and similarities in auditors and

ACs’ risk assessments can be examined empirically. In the second study reported in this thesis, I

started by testing whether AC expertise affects auditors’ KAM disclosures overall and found

evidence that entities with more accounting and industry experts on their ACs received fewer

KAMs reported in their EARs. In investigating the extent to which AC expertise affects auditors

and ACs’ risk assessments, I found that having more accounting and supervisory expertise on ACs

meant the overall alignment between EAR and EACR disclosures is enhanced. This is evidenced

by reporting fewer unmatched KAMs that are not included in the EACRs and omitting fewer SIFs

as KAMs that are reported in the EACRs. In addition, unmatched KAMs decline with industry

114

expertise on ACs. Moreover, by examining the extent to which an AC’s current period risk

disclosures affected auditors’ subsequent period KAM disclosures, I found that auditors are more

likely to report previously unmatched SIFs as KAMs when the ACs’ accounting and industry

expertise increase from one year to the next. This “adoption” behavior may be associated with less

audit effort, although it is not necessarily a reflection of lower audit quality. Overall, evidence is

obtained to support the proposition that AC expertise has a significant effect on external auditors’

KAM-related disclosures, for both current and following periods.

The third study reported in this thesis examines the level of repetitive disclosures included

in EARs from year to year, from two perspectives: the extent to which auditors use standardized

wording in communicating audit matters; and the changes in KAM identifications. Positive

associations between audit fee increases and the modifications on EAR disclosures were found.

Specifically, year-to-year similarity scores, captured using the VSM (with and without the TF-IDF

weighting function) and Trigram methods, are negatively related to audit fee changes. In addition,

audit fee changes are negatively related to recurring KAMs (those repeated from the prior year’s

EAR), and positively related to the number of KAMs that were dropped from the previous EAR.

KAMs added to the current year’s EARs are positively associated with increases in audit fees and,

more importantly, this association is driven by the “new” KAMs that were not disclosed in the

prior year’s EACRs. Overall, the findings support the proposition that, when auditors expend more

effort, greater modifications of EARs disclosures can be expected with respect to both wording

usage and KAM identifications.

Taken together, three dimensions of audit transparency are examined in this thesis. Overall,

it is concluded that audit quality can be differentiated from the content of the enhanced audit

reports, particularly through the extent to which auditors standardize their disclosures, after

115

allowing for the industry-based comparability, across their EARs. It also appears that, in the new

reporting regime, auditors’ disclosures of KAMs are affected by AC expertise. The extent to which

auditors and ACs have similar accounting judgments and risk assessments is influenced by the

number of experts on ACs, as evidenced by risk disclosures that are more consistent across their

respective extended reports. It is also evident that increased audit effort is associated with less

repetitive and more entity-specific information in subsequent EARs. Overall, differences in the

nature of auditors’ disclosures reflect differences in audit quality at the engagement level, the

effectiveness of communications between auditors and ACs, and audit effort.

Taking into account the FRC’s approach, the results of the studies reported in this thesis

can help standard setters in other jurisdictions and financial statement users gain a better

understanding of auditors’ reporting behaviors using EARs, by providing timely evidence on the

associations between EAR disclosures and audit quality, audit fees, and the relevance of ACs.

These findings may inform regulators in monitoring EAR implementation, especially with regard

to auditors’ use of boilerplate or standardized disclosures and in proposing revised standards to

enhance transparency regarding AC activities.

116

References

Abbott, L. J., S. Parker, and G. F. Peters. 2004. Audit committee characteristics and restatements.

Auditing: A Journal of Practice & Theory 23(1): 69–87.

Abernathy, J. L., B. Beyer, A. Masli, and C. M. Stefaniak. 2014. The association between

characteristics of audit committee accounting experts, audit committee chairs, and

financial reporting timeliness. Advances in Accounting 30(2): 283–297.

Abernathy, J. L., B. Beyer, A. Masli, and C. M. Stefaniak. 2015. How the source of audit

committee accounting expertise influences financial reporting timeliness. Current Issues

in Auditing 9(1): 1–9.

Bédard, J., S. M. Chtourou, and L. Courteau. 2004. The effect of audit committee expertise,

independence, and activity on aggressive earnings management. Auditing: A Journal of

Practice & Theory 23(2): 13–35.

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.

Brazel, J. F., and J. J. Schmidt. 2018. Do auditors and audit committees lower fraud risk by

constraining inconsistencies between financial and nonfinancial measures? Auditing: A

Journal of Practice & Theory (forthcoming).

Brown, S. V., and W. R. Knechel. 2016. Auditor–client compatibility and audit firm selection.

Journal of Accounting Research 54(3): 725–775.

Brown, S. V., and J. W. Tucker. 2011. Large‐sample evidence on firms’ year‐over‐year MD&A

modifications. Journal of Accounting Research 49: 309–346.

Campbell, J. L., H. Chen, D. S. Dhaliwal, H. M. Lu, and L. B. Steele. 2014. The information

content of mandatory risk factor disclosures in corporate filings. Review of Accounting

Studies, 19(1), 396–455.

Carcello, J. V., and C. Li. 2013. Costs and benefits of requiring an engagement partner signature:

Recent experience in the United Kingdom. The Accounting Review 88: 1511–1546.

Cassell, C. A., L. A. Myers, and T. A. Seidel. 2015. Disclosure transparency about activity in

valuation allowance and reserve accounts and accruals-based earnings management.

Accounting, Organizations and Society 46: 23–38.

117

Choi, J. H., J. B. Kim, X. Liu, and D. A. Simunic. 2008. Audit pricing, legal liability regimes,

and Big 4 premiums: Theory and cross‐country evidence. Contemporary Accounting

Research, 25(1), 55–99.

Church, B. K., S. M. Davis, and S. A. McCracken. 2008. The auditor’s reporting model: A

literature overview and research synthesis. Accounting Horizons 22: 69–90.

Cohen, J., L. M. Gaynor, and G. Krishnamoorthy. 2017. The effects of audit committee ties and

industry expertise on investor judgments. Working paper, Boston College, University of

South Florida, and Northeastern University.

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., L. M. Gaynor, G. Krishnamoorthy, and A. M. Wright. 2008. Academic research on

communications among external auditors, the audit committee, and the board:

Implications and recommendations for practice. Current Issues in Auditing 2(1): A1–8.

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

committee industry expertise on monitoring the financial reporting process. The

Accounting Review 89 (1): 243-–273.

Dechow, P. M., R. Sloan, and A. Sweeney. 1995. Detecting earnings management. The

Accounting Review: 193–225.

DeFond, M. L., and J. Zhang. 2014. A review of archival auditing research. Journal of

Accounting and Economics 58 (2-3): 275–326.

DeFond, M. L., R. N. Hann, and X. Hu. 2005. Does the market value financial expertise on audit

committees of boards of directors? Journal of Accounting Research 43(2): 153–193.

Deloitte. 2013. Governance in Brief. Audit Reports to be More Informative. Available at:

https://www2.deloitte.com/content/dam/Deloitte/uk/Documents/audit/deloitte-uk-audit-

governance-in-brief-audit-reports-to-be-more-informative.pdf

Deloitte. 2016a. Clear, Transparent Reporting. The New Auditor’s Report. Available at:

https://www2.deloitte.com/content/dam/Deloitte/ch/Documents/audit/ch-en-audit-new-

auditors-report.pdf

118

Deloitte. 2016b. A Clear Version. Annual Report Insights. Available at:

https://www2.deloitte.com/content/dam/Deloitte/uk/Documents/audit/deloitte-uk-ari-16-

full-details.pdf

Deumes, R., C. Schelleman, H. V. Bauwhede, and A. Vanstraelen. 2012. Audit firm governance:

Do transparency reports reveal audit quality? Auditing: A Journal of Practice & Theory

31: 193–214.

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

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

Contemporary Accounting Research 27(3): 787–827.

European Commission (EC). 2011. Proposal for a Regulation of the European Parliament and of

the Council on Specific Requirements Regarding Statutory Audit of Public-Interest

Entities. Available at: https://publications.europa.eu/en/publication-detail/-

/publication/6de7d1dc-7261-4eb8-ae57-5e9afbda7463

European Union (EU). 2014. Regulation (EU) No. 537/2014 of the European Parliament and of

the Council of 16 April 2014 on Specific Requirements Regarding Statutory Audit of

Public-interest Entities and Repealing Commission Decision 2005/909/EC. Available at:

https://eur-lex.europa.eu/legal-

content/EN/TXT/PDF/?uri=CELEX:32014R0537&from=en

Farber, D. B. 2005. Restoring trust after fraud: Does corporate governance matter? The

Accounting Review 80(2): 539–561.

Financial Accounting Standards Board (FASB). 2010. Statement of Financial Accounting

Concepts No. 8 September 2010. Available at:

http://www.fasb.org/resources/ccurl/515/412/Concepts%20 Statement%20No%208.pdf

Financial Reporting Council (FRC). 2012. The UK Corporate Governance Code. Revised

September 2012. Available at: https://www.frc.org.uk/Our-Work/Codes-

Standards/Corporate-governance/UK-Corporate-Governance-Code.aspx

Financial Reporting Council (FRC). 2013a. International Standard on Auditing (UK and Ireland)

700: The Independent Auditor’s Report on Financial Statements. Available at:

https://www.frc.org.uk/Our-Work/Publications/Audit-and-Assurance-Team/ISA-700-

(UK-and-Ireland)-700-(Revised)-File.pdf

Financial Reporting Council (FRC). 2013b. Lab Project Report: Reporting of Audit Committees.

Available at: https://www.frc.org.uk/getattachment/6e79cc57-232c-41d4-93f8-

feed38cf6fa1/Lab-Project-Report-Reporting-of-Audit-Committees-FINAL.pdf

119

Financial Reporting Council (FRC). 2014. Lab Project Report: Accounting Policies and

Integration of Related Financial Information. Available at:

https://www.frc.org.uk/getattachment/a069c04d-948b-4b8a-9993-78bf8128174f/FRC-

Lab-Accounting-policies-and-integration-of-related-financial-information.pdf

Financial Reporting Council (FRC). 2015a. Extended Auditors’ Reports: A Review of Experience

in the First Year. Available at: https://frc.org.uk/Extended-auditors-reports.pdf

Financial Reporting Council (FRC). 2015b. Lab Implementation Study: Reporting of Audit

Committees – How Companies Responded to Investor Needs Identified by the Lab;

Experience of the First Year. Available at:

https://www.frc.org.uk/getattachment/909d8978-37fd-43ec-a5c6-

20cac2a77c93/Implementation-Study-Reporting-of-Audit-Committees-Final.pdf

Financial Reporting Council (FRC). 2016a. Final Draft, International Standards on Auditing

(UK and Ireland) 701, Communicating Key Audit Matters in the Independent Auditor’s

Report. Available at: https://www.frc.org.uk/Our-Work/Publications/Audit-and-

Assurance-Team/Final-Draft-ISA-(UK-and-Ireland)-701.pdf

Financial Reporting Council (FRC). 2016b. Extended Auditor’s Reports: A Further Review of

Experience. Available at: https://www.frc.org.uk/getattachment/ 76641d68-c739-45ac-

a251-cabbfd2397e0/Report-on-the-Second-Year-Experience-of-Extended-Auditors-

Reports-Jan-2016.pdf

Fisher, T., and S. Deans. 2014. New UK Auditor’s Reports Update. Citigroup Research.

Francis, J. R. 2011. A framework for understanding and researching audit quality. Auditing: A

Journal of Practice & Theory 30: 125–152.

Francis, J. R., and M. D. Yu. 2009. Big 4 office size and audit quality. The Accounting Review

84(5): 1521–1552.

Francis, J. R., M. L. Pinnuck, and O. Watanabe. 2014. Auditor style and financial statement

comparability. The Accounting Review 89(2): 605–633.

Gay, E. H. T., and T. B. P. Ng. 2015. Effects of Key Audit Matter Standard and Audit Committee

Proactiveness on Auditors’ Communication to the Audit Committee and Decisions on

Client’s Accounting Estimates. Working paper, Nanyang Technological University.

Goh, B. W. 2009. Audit committees, boards of directors, and remediation of material weaknesses

in internal control. Contemporary Accounting Research 26(2): 549–579.

120

Gul, F.A., C. J. Chen, and J. S. Tsui. 2003. Discretionary accounting accruals, managers’

incentives, and audit fees. Contemporary Accounting Research 20(3): 441–464.

Gutierrez, E., M. Minutti-Meza, K. W. Tatum, and M. Vulcheva. 2018. Consequences of

adopting an expanded auditor’s report in the United Kingdom. Review of Accounting

Studies 23: 1543–1587.

Hanley, K. W., and G. Hoberg. 2010. The information content of IPO prospectuses. Review of

Financial Studies 23: 2821–2864.

Hanley, K. W., and G. Hoberg. 2012. Litigation risk, strategic disclosure and the underpricing of

initial public offerings. Journal of Financial Economics 103: 235–254.

Hay, D. C., W. R. Knechel, and N. Wong. 2006. Audit fees: A meta‐analysis of the effect of

supply and demand attributes. Contemporary Accounting Research 23(1): 141-191.

He, L., and R. Yang. 2014. Does industry regulation matter? New evidence on audit committees

and earnings management. Journal of Business Ethics 123(4): 573–589.

Hoberg, G., and V. Maksimovic. 2015. Redefining financial constraints: A text-based analysis.

Review of Financial Studies 28(5): 1312–1352.

Hoberg, G., and G. Phillips. 2010. Product market synergies and competition in mergers and

acquisitions: A text-based analysis. Review of Financial Studies 23: 3773–3811.

Hoberg, G., and G. Phillips. 2016. Text-based network industries and endogenous product

differentiation. Journal of Political Economy 124(5): 1423–1465.

Hoitash, R., and U. Hoitash. 2018. Measuring accounting reporting complexity with XBRL. The

Accounting Review 93(1): 259–287.

Hoitash, U., R. Hoitash, and J. C. Bedard. 2009. Corporate governance and internal control over

financial reporting: A comparison of regulatory regimes. The Accounting Review 84(3):

839–867.

Hunton, J. E., R. Libby, and C. L. Mazza. 2006. Financial reporting transparency and earnings

management. The Accounting Review 81(1): 135–157.

Institute of Risk Management. (2002). A risk management standard. London: IRM. Available at:

https://www.theirm.org/media/886059/ARMS_2002_IRM.pdf

121

International Auditing and Assurance Standards Board (IAASB). 2011. Enhancing the Value of

Auditor Reporting: Exploring Options for Change. Available at:

http://www.ifac.org/sites/default/files/publications/exposure-

drafts/CP_Auditor_Reporting-Final.pdf

International Auditing and Assurance Standards Board (IAASB). 2014. A Framework for Audit

Quality: Key Elements that Create an Environment for Audit Quality. Available at:

https://www.ifac.org/system/files/publications/files/A-Framework-for-Audit-Quality-

Key-Elements-that-Create-an-Environment-for-Audit-Quality-2.pdf

International Auditing and Assurance Standards Board (IAASB). 2015. International Standard

on Auditing 701. Communicating Key Audit Matters in the Independent Auditor’s Report.

Available at: https://www.iaasb.org/ system/files/publications/files/ISA-701_2.pdf

International Auditing and Assurance Standards Board (IAASB). 2017. The New Auditor’s

Report. A Comparison Between the IAASB and the US PCAOB Standards. Available at:

https://www.iaasb.org/system/files/publications/ files/Auditor-Reporting-Comparison-

between-IAASB-Standards-and-PCAOB-Standard.pdf

Ireland, J. C., and Lennox, C. S. 2002. The large audit firm fee premium: A case of selectivity

bias? Journal of Accounting, Auditing & Finance, 17(1), 73–91.

Kim, J. B., X. Liu, and L. Zheng. 2012. The impact of mandatory IFRS adoption on audit fees:

Theory and evidence. The Accounting Review 87(6): 2061–2094.

Kothari, S. P., A. J. Leone, and C. E. Wasley. 2005. Performance matched discretionary accrual

measures. Journal of Accounting and Economics 39(1): 163–197.

KPMG. 2014. Audit Committees’ and Auditors’ Reports. A Survey of December Year-ends Under

the New Reporting. Available at: https://home.kpmg.com/content/

dam/kpmg/pdf/2015/10/survey-of-December-year-ends-under-the-new-reporting.pdf

Kravet, T. and V. Muslu. 2013. Textual risk disclosures and investors’ risk perceptions. Review

of Accounting Studies 18(4): 1088-1122.

Krishnan, J. 2005. Audit committee quality and internal control: An empirical analysis. The

Accounting Review 80(2): 649–675.

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): 827–858.

122

Lang, M., and L. Stice-Lawrence. 2015. Textual analysis and international financial reporting:

Large sample evidence. Journal of Accounting and Economics 60: 110–135.

Lee, Y. J., K. R. Petroni, and M. Shen. 2006. Cherry picking, disclosure quality, and

comprehensive income reporting choices: The case of property-liability insurers.

Contemporary Accounting Research 23 (3): 655–692.

Lennox, C. S., J. J. Schmidt, and A. M. Thompson. 2018. Is the Expanded Model of Audit

Reporting Informative to Investors? Evidence from the UK. Working paper, University of

Southern California, University of Texas at Austin, and University of Illinois at Urbana-

Champaign.

Li, F. 2010. The information content of forward‐looking statements in corporate filings: A naïve

Bayesian machine learning approach. Journal of Accounting Research 48(5): 1049–1102.

Lobo, G. J., and J. Zhou. 2001. Disclosure quality and earnings management. Asia-Pacific

Journal of Accounting and Economics 8 (1): 1–20.

Loughran, T. and B. McDonald. 2011. When is a liability not a liability? Textual analysis,

dictionaries, and 10-Ks. Journal of Finance 66: 35-65.

Loughran, T., and B. McDonald. 2016. Textual analysis in accounting and finance: A survey.

Journal of Accounting Research 54(4): 1187–1230.

Malik, M. 2014. Audit committee composition and effectiveness: a review of post-SOX

literature. Journal of Management Control 25(2): 81–117.

McDaniel, L., R. D. Martin, and L. A. Maines. 2002. Evaluating financial reporting quality: The

effects of financial expertise vs. financial literacy. The Accounting Review 77(s-1): 139–

167.

Merkley, K. J. 2014. Narrative disclosure and earnings performance: Evidence from R&D

disclosures. The Accounting Review 89(2): 725-–757.

Mock, T. J., J. Bédard, P. J. Coram, S. M. Davis, R. Espahbodi, and R. C. Warne. 2013. The

audit reporting model: Current research synthesis and implications. Auditing: A Journal of

Practice & Theory 32(sp1): 323–351.

Nelson, K. K., and A. C. Pritchard. 2007. Litigation Risk and Voluntary Disclosure: The Use of

Meaningful Cautionary Language. Working paper, Texas Christian University and

University of Michigan.

123

Neter, J., M. Kutner, C. Nachtsheim, and W. Wasserman. 1996. Applied Linear Statistical

Models. 4th ed. New York, NY: McGraw-Hill.

Peterson, K., R. Schmardebeck, and T. J. Wilks. 2015. The earnings quality and information

processing effects of accounting consistency. The Accounting Review 90(6): 2483–2514.

Porumb, V. A., Y. Z. Karaibrahimoglu, G. J. Lobo, R. Hooghiemstra, and D. de Waard. 2018. Is

More Always Better? Disclosures in the Expanded Audit Report and their Impact on Loan

Contracting. Working paper, University of Groningen, and University of Houston.

Pownall, G., and K. Schipper. 1999. Implications of accounting research for the SEC’s

consideration of International Accounting Standards for US securities offerings.

Accounting Horizons 13(3): 259–280.

PricewaterhouseCoopers (PwC). 2015. Delivering the Value of the Audit: New Insightful Audit

Reports. Available at: https://www.pwc.com/gx/en/audit-

services/publications/assets/pwc-auditing-report-new-insightful.pdf

Public Company Accounting Oversight Board (PCAOB). 2011. Concept Release on Possible

Revisions to PCAOB Standards Related to Reports on Audited Financial Statements and

Related Amendments to PCAOB Standards. Available at:

http://pcaobus.org/Rules/Rulemaking/Docket034/Concept_Release.pdf

Public Company Accounting Oversight Board (PCAOB). 2017. The Auditor’s Report on an Audit

of Financial Statements when the Auditor Expresses an Unqualified Opinion and Related

Amendments to PCAOB Standards. Available at:

https://pcaobus.org/Rulemaking/Docket034/2017-001-auditors-report-final-rule.pdf

Reid, L. C., J. V. Carcello, C. Li, and T. L. Neal. 2015. Are Auditor and Audit Committee Report

Changes Useful to Investors? Evidence from the United Kingdom. Working paper,

University of Pittsburgh, and University of Tennessee.

Reid, L. C., J. V. Carcello, C. Li, and T. L. Neal. 2018. Impact of auditor and audit committee

report changes on audit quality and costs: Evidence from the United Kingdom.

Contemporary Accounting Research (forthcoming).

Salton, G., and C. Buckley. 1988. Term-weighting approaches in automatic text retrieval.

Information Processing and Management 24: 513–523.

Schmidt, J., and M. S. Wilkins. 2013. Bringing darkness to light: The influence of auditor quality

and audit committee expertise on the timeliness of financial statement restatement

disclosures. Auditing: A Journal of Practice & Theory 32(1): 221–244.

124

Securities and Exchange Commission (SEC). 2015. Concept Release: Possible Revisions to Audit

Committee Disclosures. Available at: http://www.sec.gov/ rules/concept/2015/33-

9862.pdf

Sharma, D. S., E. F. Boo, and V. D. Sharma. 2008. The impact of non-mandatory corporate

governance on auditors’ client acceptance, risk and planning judgments. Accounting and

Business Research 38(2): 105–120.

Simnett, R., and A. Huggins. 2014. Enhancing the auditor’s report: To what extent is there

support for the IAASB’s proposed changes? Accounting Horizons 28: 719–747.

Simunic, D. 1980. The pricing of audit services: Theory and evidence. Journal of Accounting

Research 18 (1): 161–90.

Simunic, D., and M. T. Stein. 1996. Impact of litigation risk on audit pricing: A review of the

economics and the evidence. Auditing: A Journal of Practice & Theory 15 (Supplement):

119–134.

Smith, K. W. 2017. Tell me More: A Content Analysis of Expanded Auditor Reporting in the

United Kingdom. Working paper, Virginia Polytechnic Institute and State University.

Tucker, J. W. 2015. The relation between disclosure quality and reporting quality: A discussion

of Cassell, Myers, and Seidel (2015). Accounting, Organizations and Society 46: 39–43.

Wang, C., F. Xie, and M. Zhu. 2015. Industry expertise of independent directors and board

monitoring. Journal of Financial and Quantitative Analysis 50(5): 929–962.

Watts, R. L., and J. L. Zimmerman. 1990. Positive accounting theory: A ten-year perspective.

The Accounting Review 65(1): 131–156.

Yang, R., Y. Yu, M. Liu, and K. Wu. 2018. Corporate risk disclosure and audit fee: A text mining

approach. European Accounting Review 27(3): 583–594.

Young, S. 2015. Large-sample Automated Analysis of Textual Data in Accounting Research.

European Accounting Association PhD Forum.

Zhang, J. H. 2018. Accounting comparability, audit effort, and audit outcomes. Contemporary

Accounting Research 35(1): 245–276.

Zhang, Y., J. Zhou, and N. Zhou. 2007. Audit committee quality, auditor independence, and

internal control weaknesses. Journal of Accounting and Public Policy 26(3): 300–327.

125

Appendices

Appendix 1 Variable Definitions

%AC_AFE = The percentage of AC members who are accounting financial

experts: i.e., if their biography indicates that they have at least

one of the following qualifications: certified public accountant,

chief financial officer, auditor, chief accounting officer,

controller, treasurer, or vice president-finance.

%AC_INE = The percentage of AC members who are industry experts: i.e.,

if they are currently working (or have previously worked) in

another firm that has the same two-digit ICB code as the

company in which they now serve as an AC member.

%AC_SFE = The percentage of AC members who are supervisory financial

experts: i.e., if their biography indicates that they are a chief

executive officer, chief operating officer, chair of the board, or

a president of a company but not an accounting financial expert.

%ADD_NEW = The number of new KAMs that auditors disclose in current

year’s EARs, after removing those included in prior year’s AC

report, scaled by the current year’s total KAMs.

%ADDED = The number of new KAMs that auditors disclose in the current

year’s EARs, scaled by the current year’s total KAMs.

%BORROW = The number of new KAMs in the current year’s EAR that are

the same as those disclosed in that entity’s prior year’s AC

report, scaled by the current year’s total KAMs.

%REMOVE = The number of removed KAMs from the prior year’s EARs,

scaled by the prior year’s total KAMs.

%REPEAT = The number of KAMs included in both the prior year’s and

current year’s EARs, scaled by the current year’s total KAMs.

%UNMATCHED_KAM = The percentage of KAMs disclosed in the EAR that are not

included in that entity’s EACR in year t.

%UNMATCHED_KAMdif = The percentage of SIFs disclosed in the EACR that are not

included in that entity’s EAR in year t, treating revenue

recognition and fraud in revenue recognition differently.

126

%UNMATCHED_SIF = The percentage of SIFs disclosed in the EACR that are not

included in that entity’s EAR in year t.

%UNMATCHED_SIFdif = The percentage of SIFs disclosed in the EACR that are not

included in that entity’s EAR in year t, treating revenue

recognition and fraud in revenue recognition differently.

REC = The change in accounts receivables from year t-1 to year t,

scaled by total assets at year t.

SALES = The sales for year t minus sales for year t-1, scaled by total

assets at year t-1.

ABAFEE = The residual of the audit fee model: LNAFEE = β0 + β1SIZE+

β2MTB + β3CFO + β4IRISK + β6SALESVOL + β7LOSS +

β8LEVERAGE + β9BUSY+ β10ROA+ β11BIG4+ Σβ𝑗 industry+

Σβ𝑘year+ .

ABSACC = The absolute value of abnormal accruals at the end of year t,

using the Kothari’s performance-matched modified Jones

model.

AC_SIZE = The number of members of the AC in year t.

ADD_NEW = The number of new KAMs that auditors disclose in this year’s

EARs, after removing those that are consistent with the prior

year’s unmatched SIFs.

AFEE = The natural log of total audit fees.

ANALYST_COV = The number of analysts following the company at the end of

year t.

AUDITOR_EXP = The percentage of aggregate number of auditors’ clients within

the industry in year t.

BaseSIMEAR = The mean similarity score of a firm’s EAR disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the VSM.

BaseSIMEAR_IDF = The mean similarity score of a firm’s EAR disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the VSM with TF-IDF weighting function.

BaseSIMEAR_Tri = The mean similarity score of a firm’s EAR disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the Trigram method.

127

BaseSIMKAM = The mean similarity score of a firm’s KAM disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the VSM.

BaseSIMKAM_IDF = The mean similarity score of a firm’s KAM disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the VSM with TF-IDF weighting function.

BaseSIMKAM_Tri = The mean similarity score of a firm’s KAM disclosure to all of

the other firms’ disclosures, audited by the same audit firm in

year t, using the Trigram method.

BDINDEP = The percentage of independent non-executive board members

in year t.

BDSIZE = The number of board members in year t.

BIG4 = 1 if the auditor is from Big 4, otherwise 0.

BUSY = 1 if the company’s fiscal year end is during the month of

December, otherwise 0.

CATA = The ratio of current assets to total assets at the end of year t.

CFO = The cash flow from operations divided by total assets at the end

of year t.

DACC+ = The income-increasing abnormal accruals at the end of year t.

DROP_UNMATCHED = The number of unmatched KAMs removed from prior year’s

EARs.

EARNVOL = The standard deviation of the operating earnings over the prior

five years.

FIRM_CHG = 1 if the company changed audit firms in year t+1, otherwise 0.

FIRST = 1 if this is the first year for the audit firm to prepare the EAR

for that particular entity, otherwise 0.

HIGH_ABAFEE = 1 if the abnormal audit fee is above the highest tertile (i.e., top

third), otherwise 0.

Ind_SIMEAR = The mean cosine similarity score of a firm’s EAR to all of the

other firms’ EARs, audited by the same audit firm in the same

industry and in year t.

128

Ind_SIMKAM = The mean cosine similarity score of a firm’s KAM disclosure

to all of the other firms’ disclosures, audited by the same audit

firm in the same industry and in year t.

INV = Inventory scaled by total assets in year t.

IRISK = Inventory plus accounts receivables scaled by total assets in

year t.

LEVERAGE = The ratio of debt to total assets at the end of year t.

LIST_US = 1 if the company is cross-listed in the US in year t, otherwise 0.

LNSUB = The natural log of the number of geographic segments.

LONDON = 1 if the EAR is signed by a partner located in the London office,

otherwise 0.

LOSS = 1 if the company reports a negative net income, otherwise 0.

LOW_ABAFEE = 1 if the abnormal audit fee is below the lowest tertile (i.e.,

bottom third), otherwise 0.

MA = 1 if a company appears in the SDC Platinum M&A database as

an acquirer in year t, otherwise 0.

MTB = The market value of equity divided by the book value of equity

at the end of year t.

numAC_AFE = The number of AC members who are accounting financial

experts, as defined above.

numAC_INE = The number of AC members who are supervisory financial

experts, as defined above.

numAC_SFE = The number of AC members who are industry experts, as

defined above.

PICK = 1 if at least one new KAM in the year t’s EAR is the same as

the SIF disclosed in that entity’s year t-1’s EACR, otherwise 0.

POST = 1 if the fiscal year is the first three years following the EAR

requirements, otherwise 0.

PPE = The property, plant, and equipment at t, scaled by total assets at

t-1.

129

PRIOR_ACC = The current accruals in year t-1, measured as net income before

extraordinary items plus depreciation and amortization less

operating cash flows, scaled by total assets at end of year t-1.

PROCOST = The research and development expense divided by total asset at

the beginning of year t (missing data are replaced by zero).

REC = Accounts receivables scaled by total assets in year t.

ROA = The net income before extraordinary items divided by total

assets at the end of year t.

SALESVOL = The standard deviation of annual sales over the prior three

years.

SEO = 1 if a company has a common equity offering in the secondary

market according to the SDC Global New Issues database in

year t, otherwise 0.

SIM_EARDIF = The difference between industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

VSM.

SIM_EARDIF_IDF = The difference between industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

VSM with TF-IDF weighting function.

SIM_EARDIF_Tri = The difference between industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

Trigram method.

SIM_EARRES = The regression residual of industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

VSM.

SIM_EARRES_IDF = The regression residual of industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

VSM with TF-IDF weighting function.

SIM_EARRES_Tri = The regression residual of industry-based similarity scores and

BaseSIMEAR, with the similarity scores computed using the

Trigram method.

SIM_KAMDIF = The difference between industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

VSM.

130

SIM_KAMDIF_IDF = The difference between industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

VSM with TF-IDF weighting function.

SIM_KAMDIF_Tri = The difference between industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

Trigram method.

SIM_KAMRES = The regression residual of industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

VSM.

SIM_KAMRES_IDF = The regression residual of industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

VSM with TF-IDF weighting function.

SIM_KAMRES_Tri = The regression residual of industry-based scores and

BaseSIMKAM, with the similarity scores computed using the

Trigram method.

SIMEAR = The similarity score of a firm’s current year’s EAR disclosure

to that of its prior year’s disclosures, using the VSM.

SIMEAR_IDF = The similarity score of a firm’s current year’s EAR disclosure

to that of its prior year’s disclosures, using the VSM with TF-

IDF weighting function.

SIMKAM = The similarity score of a firm’s current year’s KAM disclosure

to that of its prior year’s disclosures, using the VSM;

SIMKAM_IDF = The similarity score of a firm’s current year’s KAM disclosure

to that of its prior year’s disclosures, using the VSM with TF-

IDF weighting function.

SIZE = The natural log of total assets (in £,000) at the end of year t.

TA = Total assets at the end of year t.

TOTAL_ACC = The change in non-cash current assets minus the change in

current liabilities excluding the current portion of long-term

debt, minus depreciation and amortization, scaled by total

assets at t-1.

TOTAL_KAM = The total number of KAMs disclosed in the EAR in year t.

TOTAL_SIF = The number of SIFs disclosed in the EAR.

131

UNMATCHED_KAM = The number of KAMs disclosed in the EAR that are not

included in that entity’s EACR in year t.

UNMATCHED_SIF = The number of SIFs disclosed in the EAR that are not included

in that entity’s EAR in year t.

132

Appendix 2 Regression Results Using Alternative Measures of EAR Similarity

Panel A: Regression results for analyses (RES_IDF and RES_Tri)

SIM_KAMRES_IDF SIM_EARRES_IDF SIM_KAMRES_Tri SIM_EARRES_Tri

(1) (2) (3) (4)

ABSACC 0.101** 0.038* 0.021 0.025

(2.37) (1.88) (1.17) (1.21)

SIZE 0.006*** 0.001 -0.000 -0.001

(2.64) (1.46) (-0.01) (-0.53)

ANALYST_COV -0.001* -0.000 -0.000 -0.000

(-1.79) (-1.39) (-1.05) (-0.23)

CROSSLIST -0.015 -0.003 0.000 0.003

(-1.06) (-0.42) (0.01) (0.48)

MA -0.009 -0.001 -0.002 0.001

(-1.51) (-0.20) (-0.66) (0.40)

SEO -0.010* -0.009*** -0.006** -0.007**

(-1.52) (-2.75) (-2.11) (-2.03)

MTB 0.000 0.000 0.000 0.000*

(0.65) (1.12) (1.50) (1.80)

LOSS 0.004 0.003 0.001 0.005

(0.65) (1.23) (0.47) (1.62)

PRO_COST -0.078* -0.126* -0.147** -0.242***

(0.56) (-1.94) (-2.52) (-3.58)

EARN_COL 0.005 -0.002 0.002 -0.001

(0.64) (-0.72) (0.64) (-0.19)

LNSUB 0.008 -0.002 0.001 -0.004*

(1.70) (-0.97) (0.26) (-1.76)

CATA -0.019* 0.004 -0.010* 0.000

(-1.56) (0.68) (-1.87) (0.06)

LEVERAGE 0.011 0.007 0.002 0.005

(0.78) (1.04) (0.31) (0.78)

FIRST -0.003 0.003 0.000 0.002

(-0.35) (0.63) (0.13) (0.41)

LONDON -0.004 -0.003 -0.000 -0.001

(-0.82) (-1.46) (-0.20) (-0.54)

Constant -0.083*** -0.006 0.027** 0.049***

(-2.70) (-0.41) (2.08) (3.28)

Industry, Year FE Included Included Included Included

N 813 813 813 813

Adjusted R2 0.111 0.103 0.065 0.111

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

133

Panel B: Regression results for analyses (DIF_IDF and DIF_Tri)

SIM_KAMDIF_IDF SIM_EARDIF_IDF SIM_KAMDIF_Tri SIM_EARDIF_Tri

(1) (2) (3) (4)

ABSACC 0.108** 0.038* 0.024 0.023

(2.33) (1.86) (1.21) (1.09)

SIZE 0.006** 0.002 0.001 0.001

(2.54) (1.80) (0.92) (0.59)

ANALYST_COV -0.001* -0.000** -0.000 -0.000

(-1.79) (-1.99) (-1.27) (-0.49)

CROSSLIST -0.016 -0.000 0.001 0.002

(-1.07) (-0.06) (0.20) (0.35)

MA -0.014** -0.001 -0.003 -0.001

(-2.02) (-0.49) (-1.02) (-0.30)

SEO -0.007* -0.007** -0.001 -0.003

(-0.93) (-2.18) (-0.34) (-1.04)

MTB 0.000 0.000 0.000 0.000*

(0.61) (1.04) (1.57) (1.96)

LOSS 0.002 0.003 0.002 0.005

(0.29) (0.94) (0.61) (1.59)

PRO_COST 0.060 -0.104 -0.130** -0.181***

(0.40) (-1.60) (-2.05) (-2.62)

EARN_COL 0.004 -0.004 0.001 -0.002

(0.53) (-1.29) (0.32) (-0.51)

LNSUB 0.006 -0.002 -0.000 -0.003

(1.09) (-1.02) (-0.05) (-1.32)

CATA -0.011 0.008 -0.002 0.004

(-0.86) (1.39) (-0.31) (0.67)

LEVERAGE 0.008 0.008 0.002 0.009

(0.54) (1.22) (0.28) (1.36)

FIRST -0.000 0.002 0.000 0.004

(-0.00) (0.58) (0.12) (0.92)

LONDON -0.009* -0.003 -0.002 -0.003

(-1.66) (-1.26) (-1.03) (-1.19)

Constant -0.192*** -0.038*** -0.024* -0.015

(-5.77) (-2.63) (-1.67) (-0.99)

Industry, Year FE Included Included Included Included

N 813 813 813 813

Adjusted R2 0.129 0.129 0.057 0.139

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

134

Appendix 3 Regression Results Using the Unadjusted Similarity Scores

Panel A: Regression results for analyses on KAM section

BaseSIMKAM BaseSIMKAM_IDF BaseSIMKAM_ Tri

(1) (2) (3)

ABSACC 0.050 0.059 0.007

(1.10) (1.02) (0.22)

SIZE 0.005** 0.004** -0.002

(2.15) (1.36) (-1.19)

ANALYST_COV -0.000 -0.000 0.000

(-1.04) (-0.76) (0.10)

CROSSLIST 0.006 -0.010 -0.005

(0.37) (-0.53) (-0.42)

MA 0.007 0.007 0.002

(1.10) (0.84) (0.33)

SEO -0.022*** -0.023** -0.016***

(-3.04) (-2.53) (-3.17)

MTB -0.000 0.000 0.000

(-0.39) (0.10) (0.51)

LOSS 0.009 0.010 -0.000

(1.49) (1.34) (-0.04)

PRO_COST 0.078 0.209 -0.144

(0.52) (1.11) (-1.34)

EARN_VOL 0.000 0.008 0.005

(0.05) (0.81) (0.84)

LNSUB 0.019 0.016 0.001

(3.77) (2.46) (0.31)

CATA -0.047*** -0.047*** -0.025***

(-3.58) (-2.88) (-2.69)

LEVERAGE 0.030** 0.027 0.006

(2.06) (1.47) (0.57)

FIRST -0.012 -0.013 0.002

(-1.25) (-1.13) (0.25)

LONDON 0.002 0.015** 0.003

(0.42) (2.29) (0.93)

Constant 0.460*** 0.418*** 0.148***

(13.90) (10.00) (6.22)

Industry, Year FE Included Included Included

N 831 831 831

Adjusted R2 0.206 0.115 0.206

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

135

Panel B: Regression results for analyses on full reports

BaseSIMEAR BaseSIMEAR_ IDF Base SIMEAR_ Tri

(1) (2) (3)

ABSACC 0.019 0.006 0.019

(0.63) (0.16) (0.40)

SIZE -0.002** -0.004** -0.008***

(-1.36) (-2.01) (-3.47)

ANALYST_COV -0.000 0.000 0.000

(-0.46) (0.23) (1.00)

CROSSLIST 0.018 0.021 0.007

(1.80) (1.67) (0.44)

MA 0.011 0.016 0.015

(2.39) (2.78) (2.13)

SEO -0.017*** -0.021*** -0.025***

(-3.67) (-3.59) (-3.48)

MTB 0.000 0.000 0.000

(0.55) (0.76) (0.20)

LOSS 0.003 0.005 0.005

(0.72) (0.93) (0.72)

PRO_COST -0.319*** -0.397*** -0.588***

(-3.31) (-3.19) (-3.83)

EARN_VOL -0.001 -0.002 0.008

(-0.19) (-0.27) (0.95)

LNSUB 0.001 -0.002 -0.011**

(0.34) (-0.37) (-1.97)

CATA -0.005 0.002 -0.020

(-0.61) (0.21) (-1.52)

LEVERAGE 0.012 0.021* -0.013

(1.34) (1.73) (-0.90)

FIRST -0.001 -0.006 -0.011

(-0.13) (-0.79) (-1.14)

LONDON 0.000 0.004 0.007

(0.05) (1.05) (1.40)

Constant 0.893*** 0.917*** 0.498***

(41.68) (33.22) (14.60)

Industry, Year FE Included Included Included

N 831 831 831

Adjusted R2 0.166 0.128 0.245

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

136

Appendix 4 Regression Results Using Income-increasing Abnormal Accruals

SIM_KAMRES SIM_EARRES SIM_KAMDIF SIM_EARDIF

(1) (2) (3) (4)

DDAC+ 0.109*** 0.038** 0.120*** 0.039**

(2.84) (1.99) (2.81) (2.02)

SIZE 0.006** 0.003** 0.006** 0.003**

(2.28) (1.98) (1.97) (2.14)

ANALYSTCOV -0.001** -0.000 -0.001** -0.000

(-1.98) (-1.60) (-2.09) (-1.62)

CROSSLIST -0.016 -0.009 -0.018 -0.010

(-1.01) (-1.15) (-1.02) (-1.24)

MA -0.007 0.004 -0.012 0.003

(-0.95) (0.98) (-1.51) (0.77)

SEO -0.008* -0.009** -0.007* -0.009**

(-1.07) (-2.33) (-0.78) (-2.25)

MTB 0.000 -0.000 0.000 0.000

(0.17) (-0.01) (0.35) (0.01)

LOSS 0.007 0.005 0.003 0.005

(0.99) (1.55) (0.43) (1.50)

PROCOST 0.077 -0.143* 0.074 -0.135

(0.47) (-1.72) (0.41) (-1.61)

EARNVOL 0.023 0.007 0.031 0.007

(1.34) (0.76) (1.60) (0.74)

LNSUB 0.011 -0.003 0.007 -0.003

(1.76) (-0.97) (1.03) (-1.04)

CATA -0.006 0.011 -0.002 0.011

(-0.42) (1.36) (-0.11) (1.39)

LEVERAGE 0.016 0.009 0.016 0.009

(0.93) (1.06) (0.81) (0.96)

FIRST 0.001 0.005 0.004 0.005

(0.14) (0.98) (0.35) (1.02)

LONDON -0.005 -0.001 -0.005 -0.001

(-0.81) (-0.45) (-0.81) (-0.47)

Constant -0.091** -0.024 -0.193*** -0.062***

(-2.49) (-1.32) (-4.73) (-3.33)

Industry, Year FE Included Included Included Included

N 385 385 385 385

Adjusted R2 0.199 0.206 0.194 0.245

*** , **, * indicate significance at 1 per cent, 5 per cent , and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

137

Appendix 5 Regression Results Using Abnormal Audit Fees

SIM_KAMRES SIM_EARRES SIM_KAMDIF SIM_EARDIF

(1) (2) (3) (4)

HIGH_ABAFEE -0.010** -0.004 -0.011** -0.003

(-2.13) (-1.60) (-2.25) (-1.44)

LOW_ABAFEE -0.007 -0.000 -0.005 0.000

(-1.57) (-0.12) (-1.05) (0.12)

SIZE -0.003 -0.002* -0.005* -0.002

(-1.17) (-1.70) (-1.73) (-1.59)

ANALYSTCOV 0.000 0.000 0.000 0.000

(0.39) (0.50) (0.74) (0.54)

CROSSLIST 0.031** 0.009 0.030* 0.008

(2.10) (1.18) (1.86) (1.04)

MA -0.017*** -0.004 -0.021*** -0.004

(-3.01) (-1.39) (-3.41) (-1.63)

SEO -0.010* -0.006** -0.008 -0.006*

(-1.68) (-2.08) (-1.18) (-1.89)

MTB 0.001 0.000 0.001 0.000

(1.28) (1.34) (1.35) (1.34)

LOSS -0.001 0.002 -0.003 0.002

(-0.24) (0.91) (-0.49) (0.87)

PROCOST 0.047 -0.126** 0.061 -0.114*

(0.39) (-2.11) (0.46) (-1.90)

EARNVOL -0.000 -0.004 0.000 -0.004

(-0.01) (-0.56) (0.01) (-0.59)

LNSUB 0.002 -0.004* -0.001 -0.005**

(0.46) (-1.94) (-0.19) (-2.04)

CATA -0.047*** -0.004 -0.045*** -0.004

(-4.11) (-0.73) (-3.60) (-0.69)

LEVERAGE 0.001 0.007 0.001 0.007

(0.08) (0.99) (0.05) (0.99)

FIRST 0.001 0.004 0.003 0.004

(0.17) (0.97) (0.33) (0.95)

LONDON -0.001 -0.001 -0.002 -0.001

(-0.15) (-0.41) (-0.46) (-0.47)

Constant 0.076** 0.044** -0.009 0.008

(2.13) (2.51) (-0.24) (0.46)

Industry, Year FE Included Included Included Included

N 601 601 601 601

Adjusted R2 0.144 0.213 0.159 0.246 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-

statistics are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

138

Appendix 6 Regression Results for Analyses Excluding Entities Not in Compliance with the

Corporate Governance Code

TOTAL_KAM %UNMATCHED_KAM %UNMATCHED_SIF

(1) (2) (3)

%AC_AFE -0.892*** -0.460*** -0.292***

(-3.28) (-8.17) (-4.69)

%AC_SFE -0.159 -0.136*** -0.088

(-0.64) (-2.60) (-1.53)

%AC_INE -0.605* -0.107* -0.017

(-1.84) (-1.55) (-0.22)

SIZE 0.228*** 0.024** 0.020

(4.13) (2.09) (1.54)

MTB 0.005 -0.001 0.007***

(0.44) (-0.26) (3.02)

CATA 0.068 0.250*** 0.238**

(0.15) (2.67) (2.30)

LOSS 0.582*** 0.009 -0.018

(4.48) (0.35) (-0.60)

LEVERAGE 0.114 -0.017 0.041

(0.36) (-0.26) (0.56)

IRISK 0.498 -0.275*** -0.293**

(0.98) (-2.60) (-2.49)

MA 0.033 -0.012 0.023

(0.24) (-0.41) (0.74)

LIST_US 0.373 0.028 0.191***

(1.21) (0.43) (2.66)

ANALYST_COV 0.003 -0.002 -0.001

(0.25) (-0.87) (-0.25)

LNSUB 0.143*** 0.016* 0.011

(3.16) (1.73) (1.01)

AC_SIZE 0.022 -0.019** -0.005

(0.49) (-2.06) (-0.50)

BDSIZE 0.016 0.003 -0.007

(0.51) (0.42) (-0.93)

BDINDEP -0.124 0.032 0.009

(-0.25) (0.31) (0.08)

FIRM_CHG -0.357* -0.024 0.046

(-1.72) (-0.54) (0.96)

AUDITOR_EXP 0.729 -0.235 -0.024

(1.02) (-1.57) (-0.14)

Constant 0.267 0.161 0.124

(0.35) (0.99) (0.69)

Industry, Year, Audit firm FE Included Included Included

N 662 662 662

Adjusted R2 0.276 0.257 0.131 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-statistics

are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

139

Appendix 7 Regression Results for Analyses Using Alternative Measures of Unmatched

KAMs and Unmatched SIFs, and Count Regression

UNMATCHED_KAM UNMATCHED_SIF

OLS Count OLS Count

Coefficient

Marginal

Effect Coefficient

Marginal

Effect

(1) (2) (3) (4) (5) (6)

%AC_AFE -1.412*** -1.329*** -1.700*** -1.230*** -0.730*** -1.453***

(-6.58) (-6.37) (-6.23) (-4.85) (-4.50) (-4.46)

%AC_SFE -0.578*** -0.392** -0.502** -0.649*** -0.350** -0.697**

(-2.96) (-2.22) (-2.22) (-2.81) (-2.34) (-2.33)

%AC_INE -0.434* -0.626** -0.801** 0.164 -0.030 -0.060

(-1.84) (-2.05) (-2.05) (0.59) (-0.15) (-0.15)

SIZE 0.067 0.047 0.060 0.013 -0.005 -0.010

(1.50) (1.14) (1.14) (0.25) (-0.15) (-0.15)

MTB -0.001 0.000 0.000 0.020** 0.007 0.014

(-0.17) (0.04) (0.04) (1.99) (1.19) (1.19)

CATA 0.642* 0.655* 0.838* 0.522 0.529** 1.054**

(1.80) (1.95) (1.94) (1.23) (2.00) (2.00)

LOSS 0.002 -0.035 -0.045 -0.155 -0.133* -0.264*

(0.02) (-0.38) (-0.38) (-1.28) (-1.74) (-1.74)

LEVERAGE -0.219 -0.135 -0.173 -0.213 -0.114 -0.227

(-0.86) (-0.59) (-0.59) (-0.71) (-0.59) (-0.59)

IRISK -0.774* -0.842** -1.078** -0.681 -0.595* -1.185*

(-1.89) (-2.13) (-2.12) (-1.41) (-1.89) (-1.89)

MA -0.062 -0.048 -0.062 0.048 0.009 0.018

(-0.58) (-0.46) (-0.46) (0.38) (0.11) (0.11)

LIST_US 0.284 -0.029 -0.037 0.671** 0.175 0.349

(1.13) (-0.15) (-0.15) (2.24) (1.21) (1.21)

ANALYST_COV -0.005 -0.006 -0.008 -0.006 -0.005 -0.010

(-0.61) (-0.77) (-0.77) (-0.55) (-0.78) (-0.78)

LNSUB 0.057 0.039 0.050 0.014 0.030 0.059

(1.57) (1.10) (1.10) (0.32) (1.00) (1.00)

AC_SIZE -0.069* -0.040 -0.051 -0.050 -0.012 -0.025

(-1.92) (-1.17) (-1.17) (-1.18) (-0.45) (-0.45)

BDSIZE 0.008 0.023 0.030 -0.017 0.005 0.010

(0.31) (1.02) (1.02) (-0.57) (0.27) (0.27)

BDINDEP 0.409 0.415 0.531 0.308 0.432 0.861

(1.08) (1.11) (1.10) (0.68) (1.44) (1.44)

FIRM_CHG -0.075 -0.219 -0.280 0.134 0.111 0.222

(-0.44) (-1.34) (-1.34) (0.67) (0.86) (0.86)

AUDITOR_EXP -0.876 -0.798 -1.021 -0.758 -0.597 -1.189

(-1.53) (-1.56) (-1.56) (-1.11) (-1.40) (-1.40)

TOTAL_KAM 0.392*** 0.280*** 0.359*** - - -

(12.35) (9.99) (9.47)

140

TOTAL_SIF - - - 0.732*** 0.263*** 0.523***

(32.02) (22.73) (19.39)

Constant -0.555 -1.377 - -0.850 -1.096 -

(-0.95) (-2.48) (-1.23) (-2.16)

Industry, Year, Audit

firm FE Included Included Included Included

N 693 693 693 693

Adjusted/Pseudo R2 0.446 0.197 0.670 0.263

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics and z-statistics are reported below coefficient estimates. Variable definitions are provided in

Appendix 1.

141

Appendix 8 Regression Results for Analyses Using Alternative Measures of AC Expertise

TOTAL_KAM %UNMATCHED_KAM %UNMATCHED_SIF

(1) (2) (3)

numAC_AFE -0.146** -0.109*** -0.080***

(-2.23) (-8.08) (-5.37)

numAC_SFE 0.005 -0.031*** -0.026*

(0.09) (-2.60) (-1.93)

numAC_INE -0.195** -0.035** -0.009

(-2.58) (-2.27) (-0.52)

SIZE 0.228*** 0.020* 0.023*

(4.20) (1.82) (1.82)

MTB 0.004 -0.000 0.007***

(0.40) (-0.09) (2.99)

CATA -0.084 0.239*** 0.290***

(-0.19) (2.63) (2.89)

LOSS 0.498*** -0.002 -0.015

(3.97) (-0.09) (-0.52)

LEVERAGE 0.154 -0.016 0.038

(0.49) (-0.25) (0.53)

IRISK 0.485 -0.290*** -0.314***

(0.96) (-2.79) (-2.73)

MA 0.010 -0.013 0.023

(0.07) (-0.49) (0.75)

LIST_US 0.348 0.026 0.169**

(1.12) (0.40) (2.39)

ANALYST_COV 0.000 -0.002 -0.001

(0.04) (-0.93) (-0.23)

LNSUB 0.125*** 0.013 0.011

(2.81) (1.37) (1.10)

AC_SIZE 0.092 0.040*** 0.036**

(1.52) (3.20) (2.58)

BDSIZE 0.025 0.004 -0.007

(0.83) (0.65) (-1.03)

BDINDEP 0.043 0.081 0.026

(0.09) (0.84) (0.24)

FIRM_CHG -0.365* -0.026 0.039

(-1.75) (-0.61) (0.82)

AUDITOR_EXP 0.705 -0.241 -0.018

(0.99) (-1.64) (-0.11)

Constant 0.056 -0.038 -0.086

(0.08) (-0.26) (-0.52)

Industry, Year, Audit firm FE Included Included Included

N 693 693 693

Adjusted R2 0.295 0.265 0.136 ***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-statistics

are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

142

Appendix 9 Regression Results for Analyses Using Alternative Measures of UNMATCHED

%UNMATCHED_KAMdif %UNMATCHED_SIFdif

(1) (2)

%AC_AFE -0.450*** -0.313***

(-8.28) (-5.20)

%AC_SFE -0.129*** -0.118**

(-2.60) (-2.14)

%AC_INE -0.105* -0.009

(-1.75) (-0.14)

SIZE 0.021* 0.021*

(1.89) (1.66)

MTB 0.000 0.007***

(0.02) (2.72)

CATA 0.212** 0.286***

(2.33) (2.84)

LOSS -0.003 -0.019

(-0.11) (-0.67)

LEVERAGE -0.013 0.043

(-0.20) (0.60)

IRISK -0.261** -0.314***

(-2.50) (-2.72)

MA -0.018 0.020

(-0.65) (0.67)

LIST_US 0.030 0.163**

(0.47) (2.30)

ANALYST_COV -0.002 -0.001

(-1.12) (-0.23)

LNSUB 0.013 0.010

(1.36) (1.02)

AC_SIZE -0.017* -0.006

(-1.84) (-0.64)

BDSIZE 0.006 -0.005

(1.01) (-0.69)

BDINDEP 0.102 0.038

(1.06) (0.35)

FIRM_CHG -0.018 0.042

(-0.42) (0.88)

AUDITOR_EXP -0.256* -0.071

(-1.75) (-0.44)

Constant 0.148 0.084

(1.00) (0.51)

Industry, Year, Audit firm FE Included Included

N 693 693

Adjusted R2 0.264 0.135

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively. Robust t-statistics

are reported below coefficient estimates. Variable definitions are provided in Appendix 1.

143

Appendix 10 Regression Results for Analyses Using Cross-Section Similarity Measures

Panel A: Regression Results Using Base Similarity Scores

ΔAFEE

(1) (2) (3) (4) (5) (6)

ΔBaseSIMKAM 0.193 - - - - -

(0.66)

ΔBaseSIMEAR - -1.117 - - - -

(-1.67)

ΔBaseSIMKAM_IDF - - 0.232 - - -

(0.81)

ΔBaseSIMEAR_IDF - - - -1.352 - -

(-2.13)

ΔBaseSIMKAM_Tri - - - - 0.081 -

(0.21)

ΔBaseSIMEAR_Tri - - - - - -0.426

(-1.31)

ΔSIZE 0.226* 0.242* 0.224* 0.241* 0.229* 0.234*

(1.72) (1.84) (1.71) (1.83) (1.73) (1.76)

ΔMTB -0.007 -0.007 -0.007 -0.007 -0.007 -0.007

(-1.28) (-1.26) (-1.27) (-1.24) (-1.30) (-1.20)

ΔCFO -0.177 -0.127 -0.181 -0.122 -0.180 -0.149

(-0.38) (-0.28) (-0.39) (-0.27) (-0.39) (-0.32)

ΔINV -0.584 -0.526 -0.592 -0.574 -0.597 -0.536

(-0.53) (-0.48) (-0.53) (-0.52) (-0.54) (-0.49)

ΔREC -0.187 -0.136 -0.179 -0.130 -0.190 -0.185

(-0.33) (-0.24) (-0.32) (-0.23) (-0.34) (-0.33)

ΔSALESVOL 0.356 0.403 0.357 0.414 0.356 0.378

(1.38) (1.55) (1.39) (1.58) (1.38) (1.48)

LOSS -0.023 -0.025 -0.023 -0.029 -0.021 -0.023

(-0.39) (-0.41) (-0.39) (-0.48) (-0.36) (-0.38)

ΔLEV 0.296 0.269 0.297 0.266 0.295 0.272

(0.92) (0.83) (0.92) (0.83) (0.92) (0.84)

BUSY 0.121** 0.124** 0.120** 0.124** 0.123** 0.123**

(2.37) (2.46) (2.36) (2.46) (2.44) (2.44)

ΔROA -0.160 -0.162 -0.157 -0.163 -0.156 -0.166

(-0.65) (-0.65) (-0.64) (-0.66) (-0.63) (-0.67)

AFEE(t-1) -0.255*** -0.256*** -0.256*** -0.256*** -0.255*** -0.257***

(-6.32) (-6.36) (-6.34) (-6.37) (-6.33) (-6.35)

Constant 3.042*** 3.019*** 3.045*** 3.017*** 3.037*** 3.041***

(6.30) (6.30) (6.31) (6.31) (6.29) (6.30)

Industry, Year FE Included Included Included Included Included Included

N 975 975 975 975 975 975

Adjusted R2 0.241 0.243 0.241 0.245 0.241 0.242

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics adjusted for firm clustering effects are reported below coefficient estimates. Variable definitions

are provided in Appendix 1.

144

Panel B: Regression Results Using Adjusted Similarity Scores (Residual Method)

ΔAFEE

(1) (2) (3) (4) (5) (6)

ΔSIM_KAMRES 0.500 - - - - -

(1.14)

ΔSIM_EARRES - 0.215 - - - -

(0.22)

ΔSIM_KAMRES_IDF - - 0.362 - - -

(0.99)

ΔSIM_EARRES_IDF - - - -0.376 - -

(-0.48)

ΔSIM_KAMRES_Tri - - - - 0.478 -

(1.05)

ΔSIM_EARRES_Tri - - - - - 0.170

(0.29)

ΔSIZE 0.208 0.208 0.207 0.207 0.208 0.208

(1.50) (1.51) (1.50) (1.51) (1.52) (1.52)

ΔMTB -0.007 -0.007 -0.007 -0.007 -0.007 -0.007

(-1.27) (-1.26) (-1.25) (-1.25) (-1.27) (-1.26)

ΔCFO -0.095 -0.115 -0.093 -0.110 -0.106 -0.111

(-0.20) (-0.24) (-0.19) (-0.23) (-0.22) (-0.23)

ΔINV -1.564 -1.620 -1.568 -1.606 -1.605 -1.612

(-1.21) (-1.24) (-1.21) (-1.23) (-1.23) (-1.23)

ΔREC -0.282 -0.295 -0.301 -0.280 -0.284 -0.293

(-0.46) (-0.48) (-0.49) (-0.46) (-0.46) (-0.48)

ΔSALESVOL 0.453 0.438 0.445 0.447 0.444 0.439

(1.63) (1.57) (1.60) (1.60) (1.60) (1.58)

LOSS -0.010 -0.011 -0.010 -0.011 -0.011 -0.011

(-0.16) (-0.18) (-0.17) (-0.18) (-0.18) (-0.18)

ΔLEV 0.267 0.263 0.274 0.261 0.266 0.263

(0.82) (0.81) (0.84) (0.81) (0.82) (0.81)

BUSY 0.116** 0.119** 0.117** 0.121** 0.119** 0.119**

(2.25) (2.32) (2.28) (2.36) (2.33) (2.33)

ΔROA -0.155 -0.157 -0.156 -0.156 -0.159 -0.158

(-0.63) (-0.64) (-0.63) (-0.63) (-0.65) (-0.64)

AFEE(t-1) -0.267*** -0.268*** -0.267*** -0.268*** -0.267*** -0.267***

(-6.40) (-6.41) (-6.40) (-6.42) (-6.41) (-6.41)

Constant 3.179*** 3.187*** 3.175*** 3.196*** 3.178*** 3.185***

(6.37) (6.37) (6.37) (6.38) (6.37) (6.37)

Industry, Year FE Included Included Included Included Included Included

N 950 950 950 950 950 950

Adjusted R2 0.251 0.250 0.250 0.250 0.250 0.250

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics adjusted for firm clustering effects are reported below coefficient estimates. Variable definitions

are provided in Appendix 1.

145

Panel C: Regression Results Using Adjusted Similarity Scores (Differencing Method)

ΔAFEE

(1) (2) (3) (4) (5) (6)

ΔSIM_KAMDIF 0.400 - - - - -

(1.04)

ΔSIM_EARDIF - 0.371 - - - -

(0.38)

ΔSIM_KAMDIF_IDF - - 0.269 - - -

(0.86)

ΔSIM_EARDIF_IDF - - - -0.264 - -

(-0.34)

ΔSIM_KAMDIF_Tri - - - - 0.362 -

(0.83)

ΔSIM_EARDIF_Tri - - - - - 0.435

(0.73)

ΔSIZE 0.210 0.208 0.209 0.207 0.210 0.209

(1.52) (1.51) (1.52) (1.51) (1.53) (1.52)

ΔMTB -0.007 -0.007 -0.007 -0.007 -0.007 -0.007

(-1.28) (-1.26) (-1.26) (-1.25) (-1.26) (-1.24)

ΔCFO -0.097 -0.118 -0.094 -0.110 -0.103 -0.107

(-0.20) (-0.24) (-0.19) (-0.23) (-0.22) (-0.22)

ΔINV -1.567 -1.629 -1.569 -1.607 -1.592 -1.618

(-1.21) (-1.25) (-1.21) (-1.23) (-1.22) (-1.24)

ΔREC -0.287 -0.296 -0.305 -0.284 -0.288 -0.291

(-0.47) (-0.48) (-0.49) (-0.46) (-0.47) (-0.47)

ΔSALESVOL 0.452 0.437 0.445 0.445 0.446 0.439

(1.63) (1.57) (1.60) (1.59) (1.60) (1.58)

LOSS -0.009 -0.011 -0.010 -0.011 -0.010 -0.011

(-0.14) (-0.18) (-0.16) (-0.18) (-0.17) (-0.18)

ΔLEV 0.265 0.263 0.272 0.262 0.263 0.262

(0.82) (0.81) (0.84) (0.81) (0.81) (0.81)

BUSY 0.117** 0.118** 0.118** 0.121** 0.119** 0.118**

(2.28) (2.31) (2.30) (2.36) (2.33) (2.31)

ΔROA -0.154 -0.158 -0.156 -0.156 -0.161 -0.162

(-0.62) (-0.64) (-0.63) (-0.63) (-0.65) (-0.66)

AFEE(t-1) -0.267*** -0.267*** -0.267*** -0.268*** -0.267*** -0.267***

(-6.41) (-6.41) (-6.41) (-6.42) (-6.40) (-6.40)

Constant 3.180*** 3.188*** 3.177*** 3.192*** 3.180*** 3.181***

(6.37) (6.37) (6.38) (6.38) (6.37) (6.37)

Industry, Year FE Included Included Included Included Included Included

N 950 950 950 950 950 950

Adjusted R2 0.251 0.250 0.250 0.250 0.250 0.250

***, **, * indicate significance at 1 per cent, 5 per cent, and 10 per cent levels in a two-tailed test, respectively.

Robust t-statistics adjusted for firm clustering effects are reported below coefficient estimates. Variable definitions

are provided in Appendix 1.