pike - 2003 - studies on audit quality [aq]

The two links not addressed by independence are the use o f costly signaling by audit firms and the incentives for investing in forming a reputation for high quality audits.. Audit Quali

Studies on Audit Quality

Joel E Pike

A dissertation submitted in partial fulfillment of

the requirements for the degree of Doctor o f Philosophy

(Business)

at theUNIVERSITY OF WISCONSIN - MADISON

2003

UMI Number: 3101327

UMI

UMI M icroform 3101327 Copyright 2003 by ProQuest Information and Learning Company All rights reserved This m icroform edition is protected against unauthorized copying under Title 17, United States Code.

ProQuest Information and Learning Com pany

300 North Zeeb Road P.O Box 1346 Ann Arbor, Ml 48106-1346

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

STUDIES ON AUDIT QUALITY

submitted to the Graduate School of the University of Wisconsin-Madison

in partial fulfillment of the requirements for the

degree of Doctor of Philosophy

by

Joel E Pike

Date of Final Ora! Examination: June 12,2003

M o n th & Y e a r D e g r e e to b e a w a rd e d : December May August 2003

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Approval Signatures of Dissertation Committee

Signature, Dean of Graduate School

For my wife Nadine Mercil and my father Gilmour J Pike

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

I would like to thank my dissertation advisor, Ella Mae Matsumura, for her support and

guidance over the last six years I also want to express my deep gratitude to my co-chair, Brian

Mayhew, for his generous sharing of his time and expertise I hope that I can be to others in the

future the colleagues they have been to me I would also like to thank the remaining members of

my dissertation committee, M ark Covaleski, John Eichenseher, and Robert B M iller for their

time and suggestions

Several other members of the accounting community at the University o f W isconsin -

Madison also deserve special mention Terry W arfield has been particularly generous with his

time and willingness to offer help and suggestions Kathy Hurtt has also offered insightful

comments on sections of this dissertation My fellow Ph.D students have also profoundly

affected me, making this time one of intellectual exploration and joy I would like to especially

recognize Changling Chen and Helen Brown for their helpful comments and shared interests, and

Qiang Cheng for his sense of humor and inspiration In addition, I would like to thank M att

Magilke for many stimulating discussions, the University of Wisconsin - Madison Department

of Accounting and the School of Business for their financial support, and my wife Nadine for her

support and her hours o f assistance with data collection

3 Levels of Analysis, Methodological Issues in Prior Literature,

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

2.1 Game Description and Player Choices

2.2 Experimental Design

2.3 Subject Pool

3 Results

3.1 Reputation Formation; Sensitivity to Cost and Expected Life

4 Discussion and Conclusions

Chapter 4 The Value of a Reputation for Audit Quality: Experimental

Evidence from Buying and Selling Reputations

1 Buying and Selling Reputations

4.1 Buying and Selling Reputations

4.1.1 Competent Auditors Buying Preferences

4.1.2 Competent Auditors Selling Preferences

4.1.3 Inept Auditors Selling Preferences

4.1.4 Other Analysis - Purchase Price Effects

4.1.1a Effects on Buyer-Seller Combination

4.1.1b Effects of Treatments on Purchase Prices

4.2 Summary 57

Chapter 5 Audit Quality and the Provision of Non-Audit Services:

Evidence from the Property-Casualty Insurance Industry: Overview 66

Chapter 6 Audit Quality and the Provision of Non-Audit Services: Evidence

Chapter 7 Audit Quality and the Provision of Non-Audit Services: Evidence

from the Property-Casualty Insurance Industry:

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Constraint Functions and an Excerpt from Mailath

10 Background Information Request Form

References

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

TABLES Table 3-1 Table 3-2 Table 3-3 Table 3-4 Table 3-5 Table 4-1 Table 4-2 Table 4-3

Table 4-4

Table 4-5 Table 4-6

Table 6-1 Table 6-2

Table 6-3

Figures and Tables

Role of Information Asymmetry in Perceived Auditor

Industry Concentration in Actuarial Services Provided by

Estimated Coefficients, Z-Statistics, and p-Values from probit of Buyer Type on Consumer Bid and (Consumer Bid)2

Estimated Coefficients, t-Statistics, and p-Values from Regressing Buyer and Seller Type Combinations on

Estimated Coefficients, t-Statistics, and p-Values from Regressing Purchase Price on Treatment, Consumer Bid,

Descriptive Statistics for 1,836 Property-Casualty Insurers

Estimated Coefficients, t-Statistics, and p-Values from Regressing the Absolute Value of Loss Reserve Errors scaled by Total Admitted Assets (Materiality Units) on Auditor Type Interactions and Control Variables:

Estimated Coefficients, t-Statistics, and p-Values from Regressing the Transformed Absolute Value of Loss Reserve Errors, scaled by Total Admitted Assets on

Estimated Coefficients, t-Statistics, and p-Values from Regressing the Transformed Absolute Value of Loss Reserve Errors, scaled by Total Admitted Assets on Industry Specialist Auditor Type, Actuary Type and

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Chapter 1

1

Introduction

1 Independence and Audit Quality

Concern over the independence of auditors and the quality of audits has grown in recent

years The SEC, the AICPA, academic researchers and practitioners have all emphasized the

importance of independent auditors and high quality audits to the proper functioning of the

capital market system During the 1990’s, the relative importance of non-audit service revenue

to audit firms grew, litigation reform reduced audit firm liability exposure, and the ability to

organize as limited liability entities also decreased auditor liability, and some o f the largest audit

firms merged All of these factors contributed to increasing concern over the auditor

independence and audit quality This concern is evident in comments by SEC regulators

(Wallman 1996; Saul 1996; Schuetze 1994; Levitt 1998; Turner 1999, 2000) and continuing

efforts o f the profession to allay fears o f increasingly impaired independence and resulting

decreased audit quality Unfortunately, it is difficult to disentangle independence and quality

since if the auditor is not independent, the incentive to do a high quality audit is weakened, as

misstatements will not be reported even if found

Regulatory bodies such as the AICPA and SEC generally define independence as a state of

mind that results in unbiased and objective judgm ent about financial reporting matters

Unfortunately, research on the psychology o f judgm ent and decision-making suggests that a lack

of bias is impossible for boundedly rational human beings (Bazerman et al 1997) A behavioral

definition that is consistent with economic incentives defines independence as the probability of

disclosing a material misstatement given that it is discovered (DeAngelo 1981a) This

conditional probability also takes into account the role of audit quality, which determines the

probability of discovering the breach Audit quality is a function of several variables, one of

which is independence Theory suggests that characteristics of both the auditor and the client as

well as o f the relationship between them are expected to influence both auditor independence and

audit quality

One major problem with this conditional probability definition is the difficulty of observing

independence Because we only observe breaches that are disclosed, we are unable to determine

what the conditional probability might be Another approach might be to start with something

that can in certain circumstances be observed - such as audit quality - and use that to make

inferences If we observe high quality in the presence of incentives to impair independence, we

can infer that the auditors were likely independent, and other incentives, perhaps auditor

investment in a reputation for quality, are dominant Because audit quality is not observable in

many circumstances, the role of a reputation for quality is also important These reputations

develop based upon signals of the firms’ commitment to quality (Ippolito 1990, Gallouj 1997)

and upon the relatively few instances where quality is observable, such as litigation, audit failure,

or regulatory filings which report on the precision of prior periods’ estimates The relative

scarcity o f these observations of quality or the lack of quality makes them even more important

This dissertation extends prior literature on the conditions and characteristics that affect the

likelihood of high-quality audit outcomes by using two complementary methodologies: 1)

experimental economics to test a new analytical model of reputation formation and transfer and

2) archival data analysis to directly examine audit quality Together these provide evidence of

associations with quality at two levels of analysis - individuals (experimental) and firms

(archival) The explicit use of the individual as the unit of analysis is also an important

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

contribution, as much o f the prior literature is not at all clear whether they are referring to an

audit firm or to an individual auditor when they discuss auditor behavior and decision-making

By using individuals as the unit of analysis, I begin to develop insight about how individual

decisions may aggregate into observed firm behavior or corporate culture (Kreps 1996)

The remainder o f this chapter is organized as follows: section 2 discusses the development

of the demand for auditing and the demand for independence and quality, section 3 briefly covers

some methodological issues and contains a short literature review, and section 4 summarizes and

links the remaining chapters

2 The Demand for Auditing, Independence, and Quality

The history o f the development of auditing (DeAngelo 1981c; Watts and Zimmerman

1983; Gaa 1994) in the English-based market systems (Canada, the U.K and the U.S.) supports

the view, articulated in much of the principal-agent literature, that the demand for auditing arises

from information asymmetries, as shown in Figure 1-1

These asymmetries were the result of the increasing separation of ownership and control

Auditor independence and audit quality were not an issue as the early auditors were not

specialists, but were a subset or representative o f the owners The principal purpose of the audits

was to report on managements’ stewardship of assets owned by others

W hen auditors became specialized, the additional information asymmetry between auditors

and owners began to cause problems Independence was an early solution to the perceived

problem To the extent that auditors were independent of management, it was assumed that they

would provide an objective assessment of management’s reported numbers The quality of the

audit and the effort expended by the auditor, both generally unobservable, were unfortunately not

affected by requiring independence

Information asymm etry

Demand for monitoring

r

Audits as monitor of managers' stewardship and effort.

Fig u re 1-1

Figure 1-2 illustrates the role o f the information asymmetry between auditors and investors

(owners) The two links not addressed by independence are the use o f costly signaling by audit

firms and the incentives for investing in forming a reputation for high quality audits Investing

in a reputation for high quality could also be considered as one method o f trying to signal both a

com m itment to providing high quality audits and to using high effort in performing the audit It

is the formation of a reputation and the relation of a proxy for audit quality with firms considered

to have a high reputation that I address in this dissertation

3 Levels of Analysis, Methodological Issues in Prior Literature, and Proposed Resolutions

3.1 Levels of Analysis

Issues of independence and quality can be examined at three levels: the profession as a

whole, audit firms, and individual auditors In much of the prior literature, the difference

between examining audit firms and individual auditors is ambiguous For example, several

models (Magee and Tseng 1990; M atsumura et al 1997; Antle 1982, 1984; DeAngelo 1981a,

1981b) use individual auditor and manager interaction, with possible implications for firm

behavior

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Audit Quality and Auditor Effort Unobservable

Figure 1-2

This dissertation focuses explicitly on behavior and implications regarding individual

auditors in Chapter 2, 3, and 4, and on audit firms in Chapter 5, 6, and 7 As discussed further

below, there appear to be few incentives for firms to impair their independence, and many

incentives for them to maintain it, while the opposite is more often the case for individual

auditors Although firms provide oversight, audit results are based on the individual and group

decisions of a subset of unobservable individuals whose incentives may differ from those of the

firm overall The effectiveness of firm oversight and the resulting reputation for quality may

differ by industry specialization as well as other characteristics, such as size W hile there is no

specific entity that is a “firm” and can be said to have made a decision or acted, there is a role for

using firms as the level of analysis

Particularly in audit firms, observed firm behavior regarding audit quality is actually an

aggregation of individual actions This individual behavior is influenced by: 1) other members

of the firm, 2) the firm s’ culture and norms, and 3) the systems of controls that are established

and followed to enforce the explicit organization norms and provide oversight When I examine

what firms “do” I am using observations about many unseen individuals who have decision­

making ability within the audit partnership to make inferences about the internal corporate

culture, norms, and controls In this dissertation I examine decisions at both this aggregate firm

level and explicitly at the individual level

3.2 Methodological Issues in the Prior Literature, and Proposed Resolutions

3.2.1 Archival Studies

One stream o f empirical research dealing directly with independence and quality has used

the appearance or perception of independence, as that can be observed (Dykxhoom and Sinning

1981, 1982; Firth 1980; Lavin 1976; Lowe and Pany 1994; Pany and Reckers 1980; Pearson

1985; Shockley 1981; Farmer et al 1987) Although the use of investor perceptions may help

auditors more effectively signal their quality and effort choices, the investor perceptions may not

be correct, and thus these studies tell us little about actual quality and effort

Another stream of prior research uses litigation, which is also observable, to proxy for audit

failure - a lack o f quality - and examines the relationship of client and audit firm characteristics

to litigation Stice (1991), Lys and Watts (1994), Krishnan and Krishnan (1997) and Bonner et

al (1998) all take this approach

Kleinman et al (1998) summarize the literature on independence and discuss a variety of

characteristics o f the audit firm, the client, and the nature of the auditor-client relationship that

are theoretically related to auditor independence These are summarized in Figure 1-3

The major drawback to prior archival studies of auditor independence and quality, as

Bonner et al (1998) point out, is the lack o f evidence that their proxy for quality or

independence is in fact a valid construct This criticism motivates my study in Chapters 5, 6, and

7, which uses a directly observable measure o f the precision of an important management

estimate o f an account balance included in the audited financial statements as a more direct

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

CHARACTERISTICS AND RELATIONSHIPS TO INDEPENDENCE AND QUALITY Panel A: AUDIT FIRM CHARACTERISTICS AND THE RELATIONSHIP TO INDEPENDENCE AND QUALITY

A udit Q uality

Panel B: CLIENT FIRM CHARACTERISTICS AND THE RELATIONSHIP TO INDEPENDENCE AND QUALITY

[F in an cial H ealth of C lient

A uditor I n d e p e n d e n c e > -► £ ! A udit Q uality ]

[Audit Firm S iz e

A uditor I n d e p e n d e n c e U

[P ro fe s s io n a l S o c ie ty M e m b e rsh ip

P e e r R ev iew R e s u lts

measure of audit quality

3.2.2 Analytical Models

Antle (1982, 1984) develops analytical models based on game theory that demonstrate

equilibrium conditions for independence, while Magee and Tseng (1990), M atsumura and

Tucker (1995), and Dye (1991, 1995) develop models which examine the effects o f audit pricing,

second partner review, auditor replacement, and the organizational form o f the audit firm on

independence and quality Matsumura et al (1997) model the decision o f issuing an unqualified

vs going-concern report in a model which develops incentives of auditors If instead we

substitute any report that does not include some important information the auditor is aware of,

this model also develops implications for independence and quality Yost (1995) shows that by

allowing managers to hire and set auditors’ compensation, independence may actually increase

Antle (1982) attempts to determine optimal contracting arrangements between a utility

maximizing, strictly risk averse owner and a utility maximizing, strictly risk averse manager and

a utility maximizing, strictly risk averse auditor by modeling the non-cooperative strategic game

they play While this paper is still often cited, he notes in his conclusion:

modeling the auditor as a strategic player introduces two complexities First, the mathematical program formulated may yield solutions that are not reasonable This arises because the program may call for the auditor and manager to play dominated Nash equilibria in some subgames Second, the nontrivial nature of the subgames implies that randomized strategies by the auditor and manager may be of cmcial importance

Another crucial limitation is that the model is a one-period game How the outcomes and

implications would change in repeated play is not known

Antle (1984), in another often-cited paper, looks at the implications of a similar analysis on

auditor independence His results are that, while owners prefer a strongly independent auditor,

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

The problem here is not with the methodology Some important insights can be gained, but

unfortunately these models quickly become intractable An example of an extensive form game

between only a manager and auditor is shown in Figure 1-4 The choices o f m anagers’

1 M a n ag ers C h o o s e B u s in e ss S tra te g y

2 N ature D eterm in es O u tco m e S ta te (High, M edium, Low)

3 M a n ag ers O b se rv e O u tco m e, M ake R eporting D ecision ( R eport Low, A ccurate, or High)

4 Auditor O b s e rv e s R eport, C h o o s e s Audit T echnology (Quality Low, M edium, High)

5 Auditor O b s e rv e s T e stin g R e su lts a n d C h o o s e s Audit R eport (N ote: T h e re is a sto c h a stic e le m e n t h e re - n a tu re m o v e s to d e te rm in e if testin g d e te c ts m issta te m e n ts).

6 A ssu m e th a t for e a c h testin g result, th e au d ito r c a n c h o o se : 1) Unqualified, 2) Q ualified, 3) A d v erse

or 4) D isclaim er T his g iv e s 80 p o ssib le term inal n o d e s, a n d 8 0 s e t s of payoffs M a n a g e rs a n d au d ito rs payoffs d e p e n d on th e p a ra m e te rs u s e d to m odel their in cen tiv es, risk a ttitu d es, production a n d c o s t functions, a n d th e probabilities of random o u tc o m e s (n a tu re 's m o v es).

strategies, number of outcome states, and manager’s reporting decisions are all restricted to three

discrete possibilities The result is still eight possibilities confronting the auditor, and 27

possible paths Restricting the auditor to three levels of audit technology and four possible

reports, and ignoring the stochastic nature o f the auditing process in uncovering misstatements if

they do exist still gives eighty terminal nodes and payoff pairs Since this is a game of imperfect

THREE PLAYER MULTI-PERIOD GAME EXAMPLE

Figure 1-4

information1 and also, if auditors and managers cannot observe each other’s type and

preferences, determining which of multiple possible equilibria is actually played is difficult or

impossible This turns out to be the case in many non-cooperative games

The game becomes even more complex when a three-player game among managers,

investors, and auditors is modeled If more than a two- or three-period game is modeled, again

the equilibria quickly become very difficult to solve for In an infinite-period game, which is

required for most models of reputation (Selten 1978), there are an infinite number o f mixed

strategy equilibria

Magee and Tseng (1990) analyze a multi-period dynamic programming model with one

client and many auditors to examine both the audit pricing decision in a multi-period setting and

conditions that may lead to impaired independence They define independence as reporting

contrary to the auditors’ belief about what GAAP reporting requires, in a setting where auditors

may disagree about GAAP They find that economically rational auditors will only impair their

independence under certain conditions

3.2.3 Experimental Studies

Magee and Tseng’s results have been supported in a series of experiments (Caligari,

Schatzberg and Sevcik 1998, Mayhew, Schatzberg and Sevcik 2001, Mayhew and Pike 2003)

Other models have not been supported experimentally For example, as noted above, Yost

(1995) suggests that managers’ control o f auditor hiring and firing may lead to increased auditor

independence In contrast, Mayhew and Pike (2003) show experimentally that to achieve

1 This is because the auditor only observes the m anager’s report, not her actions or the state o f nature A game o f perfect information requires that players have the ability to observe states o f nature and the past actions o f every player in the game.

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

consistently high audit quality, it is crucial to break the link between managers and auditors by

not allowing managers to hire and fire auditors

This highlights the importance o f experimental documentation o f analytical predictions

These experiments also illustrate a method for extending the results o f analytical models By

experimentally testing a model and then, if it is supported, relaxing assumptions and extending

the testing, the authors infer that any deviations in behavior are the results of the treatments

employed This technique is used in Chapters 3 and 4

4 Summary of Remaining Chapters

The remainder of the dissertation proceeds as follows: Chapter 2 reviews the literature on

reputation, including the role of reputation in auditing, general theories of reputation, and a

discussion of reputation formation and depletion Chapters 3 and 4 use an experimental

economics approach to first test, then attempt to extend the implications of a model of reputation

building developed by Mailath and Samuelson (2001) The model has the appealing properties of

a unique equilibrium prediction in an infinite horizon game, and some links to the institutional

features of audit partnerships The model incorporates individual decisions about whether to

invest in order to increase the probability of a high quality outcome Investors observe outcomes

over time and make inferences about the type o f auditor they are facing In this game, auditors

may be replaced at any time, but investors cannot observe these replacements This is similar to

the setting where within the audit firm, which does not change, the actual audit team members,

and indeed the partner in charge, do change Because the audit opinion is signed in the name of

the firm, and not the actual individual auditor, this model appears especially apt

A second feature is the predictions Mailath and Samuelson (2001) derive regarding how

firms will be exchanged (bought and sold or, in effect, rented for a period of time) among auditor

types and explicitly how the value of various levels o f reputations affects their preferences The

experiments, based on the model of Mailath and Samuelson, are designed to test the individual

auditor incentives’ link to audit quality in panel C o f Figure 1-3 Chapter 3 provides the first

empirical test of M ailath and Samuelson’s (2001) model and a design for reliably generating

reputable behavior in the laboratory I also explicitly link features of the model to features of the

audit environment Chapter 4 poses important questions on how the culture within a partnership

may change over time as changing incentives alter the types of replacement partners desired

Chapters 5 reviews the literature on audit quality, non-audit services, and independence

followed by an overview of accounting practice and issues in the property-casualty insurance

industry Chapters 6 and 7 use archival data from this setting, where results of m anagers’

estimates can be reviewed ex post and the resulting measure of precision used as a proxy for

audit quality The presence or absence of quality, and the relationship with audit firm type, size,

and incentives is explored The study specifically tests the audit firm level variables for the

provision of non-audit services, audit firm size, and audit fee links to audit quality in panel A of

Figure 1-3 Chapter 6 replicates prior studies and extends them to a larger sample and a more

precise measure o f quality by extending the “look-back” period to include seven years o f ex post

realization In addition, Chapter 7 looks at the contribution of industry specialization to

increased audit quality

Chapter 8 concludes with a short discussion and offers suggestions for future research

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Chapter 2

Reputation in Auditing

1 Introduction

Less than three years ago, the auditors of Arthur Andersen and their former consulting

arm, then known as Andersen Consulting, concluded a bitter separation fight which included

disagreements over the right to use the Andersen name Andersen Consulting, in arguing for its

right to retain the use o f the name reportedly claimed to have spent nearly $7 billion to build its

reputation (Brown 2000), but was forced to relinquish the name The owners and employees of

Accenture (the former Andersen Consulting) are likely now congratulating themselves on losing

that fight after the collapse o f Arthur Andersen in 2002

What is the value of a reputation? Who or what creates that value? How might that

value be destroyed? W hat role do incentives and corporate culture2 created by organizational

and institutional features play in the creation or destruction o f a reputation’s value? These are all

important questions to the auditing profession, as the value of the profession itself is dependent

upon the reputation of its firms and practitioners for providing high quality audits that enhance

the credibility of financial reporting

Who partners select as new partners and how they evaluate managers and staff personnel

may change as their incentives and compensation change Systematic changes in preferences for

the type o f new partner and lower level employees could cause the organization to evolve over

2 1 use corporate culture in the sense o f Kreps (1996) Kreps m odels corporate culture as the organizational

evolution resulting from repeated plays o f a multi-period gam e by partners w hose incentives and preferences for choosing the type o f fellow partner and replacement partner may change W hile I do not test corporate culture as such, I do test whether changing conditions result in different types o f partners being chosen as replacements, and hypothesize about how that would affect the evolution o f the organization over time.

time in to a degree that may not be apparent ex ante.3 If changes in incentives change the types

o f partners in authority, the types of partners that enter the firm, the types of partners selected for

exiting the firm, or the criteria by which partners and other employees are evaluated, then

studying how incentives impact the choice o f partners could help us understand some o f the

causes o f the current problems in the profession

In the audit setting, one important source o f incentives and preferences may be the

mechanism by which the audit firms “rent” their reputations to partners through the process of

selecting new partners and compensating existing partners Generally, the largest audit firms

allow new partners to buy a partnership share when they are admitted, which gives them the right

to a certain percentage of the partnership profits They may be allowed to buy additional shares

as their career continues (Trompeter 1988) W hen partners of some firms retire, they receive

their original buy-in amount, plus a fixed rate o f return, in effect renting the reputation o f the

firm for their active career as a partner Other firms may pay them the current value of their

partnership shares

Another important source of incentives and preferences may be the outside litigation

environment Under a system where all partners are potentially personally liable for the errors of

any one o f them, partners would be expected to monitor the actions of each other (Narayanan

1995) Under a system where partners are personally liable only for their own actions, they may

prefer partners who generate more revenue even if at a higher risk o f litigation If changing from

joint and several liability to proportionate liability further reduces the potential liability for the

3 D avis, Hecht, and Perkins (2003) demonstrate analytically and using agent-based simulation that, in a tax

com pliance setting, sm all incremental changes in enforcem ent levels and perceptions o f social norms result in large and sudden drops in com pliance when certain thresholds are reached Thus, it would seem possible that the culture within an organization could respond in the same non-linear manner to small incremental changes in partner preferences and perceptions o f organization norms as communicated by performance evaluations.

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

loss o f partnership assets, the shift in preference for revenue over litigation risk would be

strengthened

This suggests that even in the presence o f professional integrity and high personal ethics,

organizations may evolve and gradually adopt behavioral norms that would have been

unthinkable at the start of the process This is the basis for the hypotheses and tests in Chapter 4

A recent popular press book (Toffler and Reingold 2003) on the collapse o f Andersen lays the

blame on their corporate culture They describe a culture of growing emphasis on increasing

revenues and profits despite the risk, and where questioning authority was discouraged They

suggest the result was there was no one left to challenge questionable decisions by partners and

upper management Swartz and W atkins (2003) cite a February, 2001 review of the Enron

account by Andersen management where it was acknowledged that this was the highest risk

account in the firm, but the client was kept because o f the large fees it generated and the

potential for even larger fees in the future

The remainder of this chapter is a review o f theories of reputation Reputation in auditing

is then examined in the next two chapters as follows: Chapter 3 lays essential groundwork for

further study by demonstrating a laboratory environment where reputations reliably form I start

with an existing analytical model by Mailath and Samuelson (2001) that has clear parallels to the

auditing environment and also has the advantage of a unique equilibrium prediction I use a four

by two design to test four cost levels4 and two expected lives5, which are drawn from the

institutional setting of auditing

4 A s discussed in Chapter 3, the model is unclear about the exact cost level required.

5 A s discussed in Chapter 3, the expected life w hile operating the firm is not a factor in the m odel, but may be a factor in an auditing setting.

Chapter 4 then uses that setting to experimentally test hypotheses about the effects of

changing incentives on individual behavior and on firm behavior over time I experimentally

explore the effects o f changing the litigation environment and compensation schemes on the

preferences o f players for the type of player they choose to sell their reputations to Buying and

selling reputations is analogous to the admission o f new partners in a partnership Existing

partners essentially sell a portion of their firm ’s reputations to new partners, and correspondingly

new partners essentially purchase a piece of the firm ’s reputation My treatments map into

changes in the audit litigation environment, namely the change to limited liability entities from

general partnerships and differences in the types o f compensation schemes used

2 Theories of Reputation

Reputation, defined somewhat broadly as beliefs about how others will behave in the

future, based on their past behavior and/or other information (Fombrun and Shanley 1990), plays

a role in essentially every strategic interaction, from globally significant political negotiations to

individual decisions o f where to take a car for repairs Public accounting firm s’ reputations play

a vital role in providing credible financial statement audits The role of reputation in auditing

has received increasing attention and generated a great deal o f discussion over the last several

years The formation, maintenance, and dissipation o f reputations have played important parts of

the discussions In particular, concerns with maintaining existing reputations and the risk of

dissipating the firms and professions reputation have grown in recent years Beginning with

Arthur Levitt’s challenges to the profession over the appearance of independence (reputation)

and the provision of non-audit services, continuing with the SEC ’s actions against a major

accounting firm in early 2000 over violations of independence appearance rules, the recent

demise of Arthur Anderson over the alleged failures at Enron, Waste Management, Sunbeam and

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

others, and finally the passage of the Sarbanes-Oxley Act, auditors’ reputation for quality has

come under attack Prior research (Mayhew 2001) suggests that an audit firm ’s reputation

impacts the credibility o f their clients’ financial statements In turn, credibility of financial

statements impacts the efficient flow of capital in our capital markets (Mayhew and Pike 2003)

This argues for the importance of theories o f reputation to help understand the strategic

interactions involved M uch o f the prior research on reputations has focused on maintenance,

beginning with an assumption o f existence and developing analytical models of reputation’s role

These theories all require information problems to exist.6

Generally, these models have incomplete information and a positive probability of

irrational play (for one period) on the part o f one or more players.7 Much of the early work on

reputation (Selten (1978), Shapiro (1982), Milgrom and Roberts (1982), Kreps and W ilson

(1982)) explored its role in strategic interactions in simple entry deterrence games These papers

developed the basic requirements for reputation as: a) two or more types o f players, b) multiple

interactions, either a single long-lived player against many sequential short-lived players, or

repeated stage games with the same players, and c) from Selten (1978), an infinite, or at least

unknown, number o f periods He showed that in a finite-period game, where all players are

assumed to act rationally in the final period, there is no role for behavior designed only to

influence the future in that final period Since play in the final period cannot be influenced by

play in the penultimate period, again there is no role for behavior designed only to influence the

6 M ost o f the theories use multi-person decision theory (game theory) although som e have used single person principal-agent m odels This chapter focuses on the game-theoretic models.

7 Complete information is when each player’s p ayoff function is com m on know ledge (Gibbons 1992) Comm on know ledge (Aumann 1976) refers to a fact that every player knows, and every player knows that every player knows

it, and every player know s that every player knows that every player know s it, etc in an infinite regression

Incom plete information occurs in gam es where, prior to the first time when players can begin to plan their m oves, som e players have private information about the game The initial private information that a player has is called the

type o f player (M yerson 1991).

future Backward induction quickly leads to the conclusion that reputation, to the extent it

involves behavior designed only to influence the future, cannot exist

The use of infinite-period games or uncertain endpoints, both o f which are

mathematically equivalent to a discount rate for future periods in a finite-period game, is

required to make reputable play rational for “almost all” periods of the game without unraveling

due to backward induction However, if the number o f periods is infinite, we must make

additional assumptions to restrict our attention to only a few o f the otherwise unlimited number

of potential equilibria in order to predict when and under what circumstances reputations will

form Other ways to achieve the same result include limiting common knowledge so that there is

o

again some uncertainty regarding other players’ payoffs or to allow imperfect information,

which also leads to a positive probability of possibly irrational actions

2.1 Reputation Formation and Depletion

Despite the pervasiveness o f reputation, theorists o f strategic interactions (game theorists)

have barely begun to advance a theory of how and why reputation develops Again, one thing

generally missing from the models examined above is how reputations form Theorists face

significant challenges because experimental studies based on these models have produced

conflicting evidence on reputation formation Experiments conducted by DeJong et al (1985),

Dopuch and King (1991) and King (1996) found inconsistent reputation formation Reputations

formed in some markets, but in other markets with the same parameters and institutional rules,

reputations did not form

8 Perfect information requires that for each decision node reached, players are aware o f their own past actions and the actions (decisions) o f the other players - in other words they know perfectly the sequence (history) which preceded the current node When they do not know this history, the gam e is one o f imperfect information (Osborne and Rubenstein 1994).

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Selten and Stoecker (1986) found, in repeated prisoner’s dilemma supergames9, that

reputation explains some behavior in approximately 75% of their markets They also

demonstrate the need for subjects to play the game to gain experience and leam about the actual

play o f the game, especially the endplay, before any model shows significant predictive ability

Camerer and W eigelt (1988) reported similar findings in a game of incomplete information,

although they also suggest the existence o f uncontrolled “homemade” prior beliefs that subjects

hold about their opponents’ likelihood o f betrayal, despite the experimental effort to induce

common knowledge preferences for reneging

While it seems clear why players would perceive value in beliefs (especially accurate

ones) about how others would behave, it is much less clear why and under what circumstances

players would incur costs to develop those beliefs in others Although important, the difficulty

in modeling reasons for its development has slowed progress in developing a theory of reputation

formation or depletion

More recent research (Mayhew (2001), Mayhew, Schatzberg and Sevcik (2001), Mayhew

and Pike (2003)) has identified market settings that reliably generate reputation formation in a

relatively small number of periods of play Mayhew (2001) suggests there are boundary

conditions for reputation formation and that nearly immediate rewards for reputable behavior are

required This result contrasts with observed “real world” behavior in which long-term

investment in reputation capital is observed Additional work by Mayhew, Schatzberg and

Sevcik (2001) indicates that introducing additional uncertainty in the form of imperfect

information severely impairs reliable reputation formation When it is uncertain whether failures

9 M yerson (1991) defines a supergame as “a repeated gam e in which there is only one possible state o f nature and the players know all o f each other’s past m oves.”

occurred through acts o f nature or lack of effort, even nearly immediate rewards do not reliably

induce reputation formation

The evidence from these experiments suggests that learning and immediate positive

feedback for adopting reputable strategies are key determinants of reputation formation when

there is no uncertainty about the cause of failures, but these studies suffer from two drawbacks

First, none of them can ex ante rule out equilibria, so it could be argued that all observed

behaviors are a part of some longer-term mixed strategy equilibrium To rule out some of the

many possible equilibria, they suggest that Pareto dominant equilibria should emerge, and rely

on eliminating weakly dominated strategies but cannot eliminate the (many) nearly Pareto

strategies Second, it is not clear why immediate feedback is important Is the outcome so

dependent on initial conditions that early “mistakes” could rule out the reputation equilibrium?

This would suggest that reputation is not a very robust response Or is immediate feedback

necessary for players to learn that reputations can pay? This suggests one of several

explanations that cannot be distinguished from the results Players may have very short time

horizons (are myopic), they may have steep discount rates, or they may be unable to think

strategically in the way that game theory assumes they can (bounded rationality)

In the next chapter, I develop hypotheses and discuss the experimental design, then report

on my experiment, based on a model developed by Mailath and Samuelson (2001) that reliably

induces reputation formation in the laboratory

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Chapter 3

Reliable Reputation Formation

I Hypotheses

A recent paper by Mailath and Samuelson (2001) develops an infinite-period model of

reputation where competent producers choose high effort to signal, at a cost, that they are not

inept producers W ithout additional conditions, there is no unique equilibrium If competent

types always chose high effort, they would be identified as competent with probability one, and

thus would no longer have any incentive to signal their type at a cost, so always investing in high

effort cannot be an equilibrium The model incorporates unseen replacement, so that consumers

can never be sure that it is the same producer operating the firm This acts to bound the

probability of any particular producer being competent or inept away from zero or one, so that

there is a unique equilibrium of competent producers always exerting effort to produce high

quality and signaling their type

I apply the analytical results of M ailath and Samuelson (2001) by labeling producers as

auditors and consumers as investors Auditors are either competent or inept The competent

type can invest to increase the probability o f producing high quality audits, while an inept firm

cannot.10 The game is an infinite-period game, so reputations may form As discussed above,

without some method of creating additional information imperfections, no pure strategy

equilibrium exists, because over an infinite number of periods, investors could identify with

certainty both competent and inept auditors Once identified, competent firms would prefer to

10 Alternatively, the inept auditor can be thought o f as facing such a high cost for investing to increase the probability o f producing high quality audits that they w ould never rationally choose to do so Competent auditors,

on the other hand, have a low enough cost that they would rationally choose to do so under som e circumstances O f course, they would prefer that investors paid them for high quality when they in fact only invested enough to produce low quality.

not invest, as they would receive a higher payoff by saving the cost of high quality effort To

state it differently, once investors were convinced that an auditor was competent and would

always deliver high quality audits, the competent auditor would stop producing the high quality

audits to save the cost o f the higher effort

To prevent identification of auditor types, Mailath and Samuelson introduce hidden

replacement o f auditors, so investors can never be certain that they are dealing with the same

auditor even though they can observe the prior history o f the quality offered by a particular firm

Investors are aware that such a change can occur but cannot directly observe the actual change

Since the probability of any individual firm being competent or inept is bounded away from both

zero and one by the possibility of replacement, investors can credibly believe that competent

firms will always invest and there is a unique equilibrium where they always do so

This model is attractive for several reasons: 1) it provides a clear ex ante prediction of a

unique reputation equilibrium, 2) to achieve this equilibrium, it uses unobserved replacement of

the individual auditor acting under a given name, which corresponds to the way in which

ownership is transferred among audit partners within an audit firm, and 3) the model is extended

to include the buying and selling of reputations among auditors

Because the Mailath and Samuelson (2001) model predicts a unique equilibrium of

reputable behavior, if laboratory participants behave reputably when they are predicted to, I can

use this as a baseline to begin to change incentives and compensation in ways the model does not

address I can then feel confident in attributing any observed change in behavior to the causal

effect of the change in incentives and compensation

Using the model predictions from M ailath and Samuelson (2001) leads to the first

hypothesis:

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

choose high (more costly) effort when the cost of high effort is sufficiently low.

I qualify the hypothesis above in terms o f cost because the model Mailath and Samuelson

(2001) develop is unclear about the level o f cost both sufficient and necessary for a unique

equilibrium Using the notation from their paper, let Vc(<|>) be the value to the competent firm of

choosing high effort in every period, and Vc($',L) be the value to a competent firm to choosing

low effort once and high effort forever after For the equilibrium to hold, Vc(<|>) > V d§;L) for all

feasible <|> By rearranging their equation (4), a sufficient condition for this inequality to hold is

(£(1 A)(l 2p))p(<pg) p (0 b) > C 11 W hile sufficient, this cost may be lower than necessary

-an equilibrium may exist at a higher cost as well, since this cost is sufficient over all feasible

levels of consumer beliefs If some of these beliefs are not observed in a given sequence of

action choices by the auditor, a higher cost may be sufficient for the observed levels o f investor

beliefs Mailath and Samuelson also show that it is necessary that C < 1 -2 p Thus, in Mailath

and Samuelson’s model, as the cost of high effort increases, the existence o f a unique

equilibrium becomes dependent on what patterns o f outcomes are actually observed For some

outcome patterns, and when the cost becomes sufficiently high, there is no unique equilibrium

prediction that auditors will always choose high effort This leads to the second hypothesis:

choose high (more costly) effort when the cost to high effort is sufficiently high As

11 Variables are defined as in A ppendix A 8 is an arbitrary discount factor X is the probability the current auditor

w ill be replaced at the end o f any period, p is the probability o f a high quality outcome when high effort is put forth and, by symmetry the probability o f a low quality outcom e when low effort is put forth P(<t>g) is the investors’ b elief about the probability the audit is performed by a com petent auditor after observing a high quality outcom e, and P(<))b) is their b elief after observing a low quality outcome C is the cost o f high effort.

the cost to high effort increases, the proportion of periods when high effort is chosen

will decrease monotonically.

There are two potential contributions to testing the model of Mailath and Samuelson, if

the hypotheses are supported First, it would provide the first empirical test and confirmation of

the model they develop Second, it would provide experimenters interested in the effects of

various factors on reputation with a method of starting with reliably induced reputable behavior

2 Experiment

2.1 Game Description and Player Choices

I base my experimental design on the model developed by Mailath and Samuelson

(2001) Their game is a two-player game between auditors and investors (see Figure 3-1 for a

timeline of the game and the player m o v es).12 Investors prefer a high quality audit, but are only

able to observe the previous output o f a firm That is, investors cannot observe the quality prior

to purchasing the audit

2.2 Experimental Design

The experimental economics design incorporates three categories of players: investors,

auditors, and potential auditors who can rent or purchase an existing firm ’s name if the current

auditor exits Each market period, auditors make effort decisions that determine the quality of

the services they sell Investing to increase the probability o f generating high quality output does

not guarantee that the output will be high quality Conversely, failing to invest does not

1 ^guarantee that the output will be low quality Investors purchase the audits and have

12 Mailath and Samuelson refer to consumers and producers I interpret these players as investors and auditors, respectively, and refer to them as such throughout this chapter In the experiments, to avoid unwanted contextual effects, players were referred to as consumers and producers.

13 See Appendix A for further information on the probabilistic elem ent o f determining the quality o f the output and the parameter values chosen for the experiments.

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

Auditor chooses (unobserved) effort if it is competent.

Output produced

Auditor receives revenue of pt.

Investors assign probability (j)t to auditor being competent and (l-4>t ) to being inept

Investors bid their expected utility

Expected utility from receiving audit = pt = utility

of audit (=1) x Probability (receiving high quality outcome)

Auditor exits with probability

Adapted from Mailath & Samuelson (2001) Variables are defined in Appendix A

Investor, auditor and market observe realized value o f outcome

(g for high quality

or b for low quality), 4>t and pt are calculated (updated using Bayes rule) See Appendix A for calculations

preference for high quality audits After purchasing the audit, investors learn the actual

quality of the audit The game continues for multiple periods In any period, a new auditor

can replace an extant auditor without the knowledge o f investors although investors know that

such change is possible In the set o f experiments reported in this chapter, this process of

replacement is automated Exit is exogenously determined, and all auditors immediately re­

enter as a new firm when they are replaced In the second set of experiments, reported in

Chapter 4, players explicitly bid for the right to enter and play for an unknown number of

periods until their exogenously determined exit

Instructions for all treatments and additional experimental materials, such as a

required quiz to test participants understanding of the instruction and a request for background

demographic information, are included in Appendix B As noted above, while I interpret the

players as auditors and investors, during the actual experiments players were labeled

producers and consumers to avoid uncontrolled expectations for behavior that might be

introduced with contextual labels such as auditor

Players begin the game with an opening investor bid o f E$80, which reflects an

uninformative prior belief of a 50% chance of facing either a competent auditor or of facing

an inept auditor Player screens show them their Firm and User ID ’s (which are identical in

this phase of the experiment) their conversion rate to US$, their initial endowment and their

earnings for the current cycle from investing, from forecasting, and in total This information

is down the left hand side of the screen (see Figure 3-2 for a screen shot of a player screen)

The investor bid is given at the top left o f the screen, and below this is the investment choice,

either Bin 0 (no investment to increase quality) or Bin 1 (investing to increase quality) The

cost for each choice is shown immediately below the selection No investment carries a

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.

minimum cost o f E$30 as the value to the investor is between E$30 and E$130 Investing

costs are varied across treatments (note the screen shot shows the high cost treatment, where

high effort cost E$90)

Figure 3-2 Screenshot of Player Screen

Period

rirm ID

usbi inI

R a te

I live « Profit

Fines

NrJ

P ir

lliSPtll f ffIMIilBSillSii

Once a choice is made, the player submits their choice of bin by clicking on a button

marked “submit” As soon as the choice is submitted, the outcome is revealed Players then

make predictions about probabilities and the investor bid in the lower right portion o f the

screen They are asked first to assess the probability that investors believe they are a firm that

is capable of investing, then to assess the probability investors believe they have been

replaced since the prior period Finally, they are asked to predict the investor’s bid in the next

period, after the investor observes the outcome produced this period Probabilities are

restricted to be between 0 and 100% and the investor bid prediction must be between E$30

and E$130 Once they submit their predictions, the screen refreshes If not replaced, players’

screens are updated to show the next period number and a new investor bid along with their

investment earnings and forecast earnings from the prior periods’ actions If they are

replaced, the period is reset to 1 and the investor bid is reset to E$80 W hen the period and

bid are reset, this signifies that the player has been replaced at their old firm and is starting

over at a new firm with no prior history

Players know the number of cycles they will play, where a cycle is defined as

operating a particular firm for an unknown number of periods until replacement They do not

know the number of periods in any cycle

As I am primarily interested in the behavior of auditors, I use robot investors In

addition, in the experiments in this chapter where I test the robustness of the m odel’s

predictions and sensitivity to levels of the investment cost parameter, I use robot inept

auditors because inept player types have no effort decisions to make The only human players

are the competent auditors (audit partners) who can choose high or low effort

The robot investors calculate their bids by Bayesian updating as specified in Mailath

and Samuelson (2001) The investor bids are the calculated probability that they are dealing

with a competent firm I offset this by 30, so that investor bids run between 30 and 130,

which corresponds to a 0% and 100%, respectively, calculated probability of facing a

competent firm Table 3-1 illustrates 4 of the possible progressions of opening investor bids

over 21 periods following 20 outcomes I use 16 high quality and 4 low quality outcomes for

each since this is the expected number o f outcomes for a competent auditor given the 80%

Reproduced with permission of the copyright owner Further reproduction prohibited without permission.