<%BANNER%>

Audits As Credence Goods

Permanent Link: http://ufdc.ufl.edu/UFE0024833/00001

Material Information

Title: Audits As Credence Goods What Do Auditors Know and How Do They Use Their Information
Physical Description: 1 online resource (182 p.)
Language: english
Creator: Causholli, Monika
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: audit, credence
Accounting -- Dissertations, Academic -- UF
Genre: Business Administration thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This study examines whether the credence attribute of auditing impacts the audit profitability and the audit production process. The credence attribute of auditing refers to the client s uncertainty about the audit effort needed (ex-ante uncertainty) and the audit effort provided by the auditor (ex-post uncertainty). That is, the auditor as both the expert, the party with the audit expertise who advises the client about the level of assurance needed, and the provider (seller) of the service possesses an information advantage about the production of the audit relative to the client and may find it beneficial to engage in strategic behavior and extract rents. This issue is important because strategic actions may result in adjustments to the audit production process that may undermine audit quality and efficiency. Initially, this study argues that private information gained through an auditor s repeated interaction with the client, enhances the auditor s incentives to extract rents. However, clients switching costs mitigate the auditor s incentives over time. As a result, it is predicted that rents initially rise with tenure, but eventually decline as switching auditors becomes less costly for the client. Moreover, clients own knowledge about the audit process may limit rents. Using proprietary audit engagement data from one of the international auditing firms, the results suggest that rents are highest for mid-tenure clients relative to the short and the long-tenured ones. Further, rents are lower for clients with superior knowledge about the audit process. Second, this study explores the factors associated with inefficient audits. Using the theory of credence goods as the basis, the study finds that inefficient audits are less likely to occur when competition is high and when clients possess superior knowledge about the audit process, but are more likely to occur in clients that are costly to audit. Moreover, inefficient audits are positively related to the auditor s private information about the client. The purpose of this study is to understand how audits are conducted by using the theory of credence goods as the basis for explaining auditors incentives to act strategically. The results presented are consistent with audits being credence goods.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Monika Causholli.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Knechel, W. Robert.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024833:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024833/00001

Material Information

Title: Audits As Credence Goods What Do Auditors Know and How Do They Use Their Information
Physical Description: 1 online resource (182 p.)
Language: english
Creator: Causholli, Monika
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: audit, credence
Accounting -- Dissertations, Academic -- UF
Genre: Business Administration thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This study examines whether the credence attribute of auditing impacts the audit profitability and the audit production process. The credence attribute of auditing refers to the client s uncertainty about the audit effort needed (ex-ante uncertainty) and the audit effort provided by the auditor (ex-post uncertainty). That is, the auditor as both the expert, the party with the audit expertise who advises the client about the level of assurance needed, and the provider (seller) of the service possesses an information advantage about the production of the audit relative to the client and may find it beneficial to engage in strategic behavior and extract rents. This issue is important because strategic actions may result in adjustments to the audit production process that may undermine audit quality and efficiency. Initially, this study argues that private information gained through an auditor s repeated interaction with the client, enhances the auditor s incentives to extract rents. However, clients switching costs mitigate the auditor s incentives over time. As a result, it is predicted that rents initially rise with tenure, but eventually decline as switching auditors becomes less costly for the client. Moreover, clients own knowledge about the audit process may limit rents. Using proprietary audit engagement data from one of the international auditing firms, the results suggest that rents are highest for mid-tenure clients relative to the short and the long-tenured ones. Further, rents are lower for clients with superior knowledge about the audit process. Second, this study explores the factors associated with inefficient audits. Using the theory of credence goods as the basis, the study finds that inefficient audits are less likely to occur when competition is high and when clients possess superior knowledge about the audit process, but are more likely to occur in clients that are costly to audit. Moreover, inefficient audits are positively related to the auditor s private information about the client. The purpose of this study is to understand how audits are conducted by using the theory of credence goods as the basis for explaining auditors incentives to act strategically. The results presented are consistent with audits being credence goods.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Monika Causholli.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Knechel, W. Robert.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024833:00001


This item has the following downloads:


Full Text

PAGE 1

1 AUDITS AS CREDENCE G OODS: WHAT DO AUDITORS KNOW AND HOW DO THEY USE THEIR INFORMATIO N By MONIKA CAUSHOLLI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Monika Causholli

PAGE 3

3 To my family

PAGE 4

4 ACKNOWLEDGMENTS I wish to acknowledge the support and friendship of a few individuals who were instrumental in my intellectual and personal development during the PhD program. First and foremost, I would like to express my deepest gratitude to my chair and advisor Dr. Robert Kne chel, f o r his endless enc ouragement during this endeavor He patien tly listened to me talk, made me think through all the research problems, and tirelessly read innumerable versions of this dissertation document I would like to thank the other members of my committee Dr. Steph en Asare, Dr. Victoria Dickinson and Dr. David Sappington, who asked hard questions and provided insightful comments on my work. A special thank you is reserved for Dr. Joel Demski, whose wor k teachings and support throughout all my time in the PhD program have inspired me in many, many ways. To fellow PhD students whom I befriended during this journey Ri chard Lu, Carlos Jimenez, Jason McGregor, Liang Fu, Mike Donohoe, Stephen Brown, Alice Bonaime, and Nicole Zein, thank you for all your support and camaraderie. I learned a great d eal from you all I wish t o acknowledge the help and teachings of a wonde rful person, Dr. Sarah Hamersma who was always there when I needed her help and advice. Fin ally, I thank the Fisher School of Accounting and especially Dr. Gary McGill for the financial support throughout the program.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................................... 4 LIST OF TABLES ................................................................................................................................ 8 LIST OF FIGURES ............................................................................................................................ 10 ABSTRACT ........................................................................................................................................ 11 CHAPTE R 1 INTRODUCTION ....................................................................................................................... 13 2 AUDITS AS CREDENCE GOODS: A THEORETICAL DISCUSSION OF AUDITORS DECISION -MAKING PROCESS ...................................................................... 19 Credence Goods: Theory and Evidence ..................................................................................... 19 Auditing under the Credence Lens ............................................................................................. 22 The Credence Attribute of Auditing ................................................................................... 22 Placing the Credence Attribute in Context ......................................................................... 23 A generic m odel of an auditors decision-making process ........................................ 24 Auditors decision-making process: search, experience, and credence frameworks ............................................................................................................... 25 Auditors Strategic Space ........................................................................................................... 30 Underauditing ...................................................................................................................... 32 Overauditing......................................................................................................................... 32 Overcharging ........................................................................................................................ 32 Disciplining Mechanisms in the Market for Auditing Services ............................................... 34 Countervailing Mechanisms to Underauditing .................................................................. 34 Legal liability and regulation ....................................................................................... 34 Audit firm reputation and size ..................................................................................... 35 Countervailing Mechanisms to Overauditing and Overcharging ..................................... 36 Clients knowledge ....................................................................................................... 36 Competi tion .................................................................................................................. 36 Auditors Rational Behavior ....................................................................................................... 37 Case 1: H Type Client ......................................................................................................... 38 Case 2: L Type Client ......................................................................................................... 39 Discussion .................................................................................................................................... 39 3 DATA AND GENERAL DESCRIPTIVE STATISTICS ........................................................ 51

PAGE 6

6 4 THE IMPACT OF THE AUDIT CREDENCE ATTRIBUTE ON PROFITABILITY AND PRODUCTION OF AUDITS IN A REPEATED SETTING ........................................ 65 Introduction ................................................................................................................................. 65 Auditor Client Relationship, Auditor Knowledge and Clients Cost of Switching ................ 66 Probing the Audit Production Process ....................................................................................... 70 Research Design .......................................................................................................................... 72 Engagement Profitability ..................................................................................................... 72 T he Independent Variables of Interest: Tenure and Clients Information ....................... 74 The Dependent Variable: Audit Profitability ..................................................................... 75 Control Variables ................................................................................................................. 76 Probing the Audit Production Process ................................................................................ 77 Empirical Results ........................................................................................................................ 80 Descriptive Statistics ........................................................................................................... 80 Tests of Hypotheses ............................................................................................................. 82 Main results .................................................................................................................. 82 Robustness tests ............................................................................................................ 85 Probing the Audit Production Process ................................................................................ 88 Conclusion and Discussion ......................................................................................................... 91 5 EXPLAINING AUDIT PRODUCTION INEFFICIENCIES ................................................ 113 Introductio n ............................................................................................................................... 113 Development of Hypotheses ..................................................................................................... 118 Research Design ........................................................................................................................ 122 Empirical Model ................................................................................................................ 122 The Dependent Variable: Overauditing ........................................................................... 124 The benchmark audit .................................................................................................. 124 Ex ante overauditing .................................................................................................. 124 Ex -post overauditing .................................................................................................. 125 Factors Associated with Overaudit ing .............................................................................. 126 Competition ................................................................................................................ 127 Cost of providing an audit ......................................................................................... 127 Auditors private information .................................................................................... 129 Clients own information ........................................................................................... 132 Clients bargaining power .......................................................................................... 132 Other control variables ............................................................................................... 133 Empirical Results ...................................................................................................................... 133 Descriptive Statistics ......................................................................................................... 133 Probit Results ..................................................................................................................... 138 Ex ante overauditing .................................................................................................. 138 Ex -post overauditing .................................................................................................. 140 Robustness Tests ................................................................................................................ 143 Conclusion and Discussion ....................................................................................................... 144

PAGE 7

7 6 CONLUDING REMARKS ...................................................................................................... 159 APPENDIX A AUDITORS RATIONAL BEHAVIOR ................................................................................. 163 Case 1: H Type Client .............................................................................................................. 163 Auditors Decision: Choose First between S1 and S2 ...................................................... 163 Case 1a: auditor chooses between S2 and S3 ............................................................ 163 Case 1b: auditor chooses between S1 and S3 ............................................................ 163 Case 2: L T ype Client ............................................................................................................... 164 Auditors Decision: Choose First between S4 and S5 ...................................................... 164 Case 2a: assume first that overcharging is not possible ........................................... 164 Case 2b: assume that overcharging is possible ........................................................ 164 B THE BEN CHMARK AUDIT .................................................................................................. 165 C DEFINITIONS OF VARIABLES ........................................................................................... 168 D THE CALCULATION OF UNEXPECTED FEES (UFEE) .................................................. 171 E AUDIT EFFICIENCY USING DATA ENVELOPMENT ANALYSIS .............................. 174 LIST OF REFERENCES ................................................................................................................. 176 BIOGRAPHICAL SKETCH ........................................................................................................... 182

PAGE 8

8 LIST OF TABLES Table page 2 1 Auditor strategies, auditor payoff and client utility ............................................................. 50 3 1 Sample composition ............................................................................................................... 57 3 2 Client characteristics .............................................................................................................. 58 3 3 Engagement characteristics ................................................................................................... 59 3 4 Frequency distribution of dichotomous variables ................................................................ 60 3 5 Distribution of client characteristics by whether clients are publicly traded ..................... 61 3 6 Distribution of engagement characteristics by whether clients are publicly traded ........... 62 3 7 Client characteristics by industry .......................................................................................... 63 3 8 Engagement characteristics by industry ............................................................................... 64 4 1 Abnormal effort and fee under each strategy ....................................................................... 95 4 2 Descriptive statistics of continuous variables ....................................................................... 96 4 3 Frequency distribution of engagement variabl es ................................................................. 97 4 4 Correlations related to variables in equations (4 1) and (4 2) ............................................. 98 4 5 Audit engagement characteristics by tenure if MEDTEN cut -off is 7 years ...................... 99 4 6 Audit engagement characteristics by tenure if MEDTEN cut -off is 8 years .................... 100 4 7 Analysis of mean engagement profitability by tenure ....................................................... 101 4 8 Analysis of median engagement profitability by tenure .................................................... 102 4 9 Engagement profitability by INTAUD and LPREP .......................................................... 103 4 10 Engagement profitability by MEDTEN, INTAUD, and LPREP ...................................... 104 4 11 Regression results of equation (4 1) when MEDTEN cut -off is 7 years .......................... 105 4 12 Regression results of equation 4 1 when MEDTEN cut -off is 8 years ............................. 106 4 13 Regression results of equation 4 2 when MEDTEN cut -off is 7 years ............................. 107 4 15 Regression results of equations 4 3, 4 4, and 45 .............................................................. 109

PAGE 9

9 4 16 Audit production characteristics by tenure ......................................................................... 110 4 17 Audit production characteristics by labor rank when MEDTEN cut -off is 7 years ......... 111 4 18 Audit production characteristics by labor rank when MEDTEN cu t -off is 8 years ......... 112 5 1 Descriptive statistics of continuous variables ..................................................................... 146 5 2 Frequency distribution of engagement characteristics ....................................................... 147 5 3 Engagement characteristics by EXAGG ............................................................................. 148 5 4 Engagement characteristics by P_EXCD ........................................................................... 149 5 5 Engagement characteristics by A_EXCD ........................................................................... 150 5 6 Engagement characteristics by U_EXCD ........................................................................... 151 5 7 Correlations related to variables in equation (5 1) ............................................................. 152 5 8 Correlations related to variables in equation (5 2) ............................................................. 153 5 9 Probit results of equation (5 1) ............................................................................................ 155 5 10 Probit results of equation (5 2) when the dependent variable is P_EXCD ...................... 156 5 11 Probit results of equation (5 2) when the dependent variable is A_EXCD ...................... 157 5 12 Probit results of equation (5 2) when the dependent variable is U_EXCD ...................... 158 B1 Regression of hours on client characteristics ..................................................................... 167

PAGE 10

10 LIST OF FIGURES Figure page 2 1 Location of goods a long the information spectrum based on their information properties ................................................................................................................................ 43 2 2 A generic model of an auditors decision-making process .................................................. 44 2 3 Audits as search goods ........................................................................................................... 45 2 4 Audits as experience goods ................................................................................................... 46 2 5 Audits as credence goods ....................................................................................................... 47 2 6 Auditors decision process under the credence lens ............................................................ 48 2 7 Auditors decision process under the credence lens. Auditor and client payoffs .............. 49 4 1 Switching costs and auditors private information over time .............................................. 94

PAGE 11

11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy AUDITS AS CREDENCE GOODS: WHAT DO AUDITORS KNOW AND HOW DO THEY USE THEIR INFORMATION By Monika Causholli August 2009 Chair: W. Robert Knechel Major: Business Administration This study examines whether the credence attribute of auditing impacts the audit profitability and the audit production process The credence attribute of auditing refers to the clients uncertainty about the audit effort needed (exante uncertainty) and the audit effort provided by the auditor (ex-post uncertainty). That is, the auditor as both the expert the party with the audit expertise who advises the client about the level of assurance needed, and the provider (seller) of the service possesses an in formation advantage about the production of t he audit relative to the client and may find it beneficial to engage in strate gic behavior and extract rents. This issue is important because strategic actions may result in adjustments to the audit production p rocess that may undermine audit quality and efficiency. Initially, this study argues that private information gained through an auditors repeated interaction with the client, enhances the auditors incentives to extract rents. Ho wever, clients sw itching costs mitigate the auditors incentives over time. As a result, it is predicted that rents initially rise with tenure, but eventually decline as switching auditors becomes less costly for the client. Moreover, clients own knowledge about the audit process may limit rents. Using proprietary audit engagement data from one of the international auditing firms, the results

PAGE 12

12 suggest that rents are highest for mid-tenure clients relative to the short and the long tenured ones Further, rents are lower for clients with superior knowledge about the audit process. Second, this study explores the factors associated with inefficient audits. Using the theory of credence goods as the basis, the study finds that inefficient audits are less likely to occur when com petition is high and when clients possess superior knowledge about the audit process, but are more likely to occur in clients that are costly to audit. Moreover, inefficient audits are positively related to the auditors private information about the clien t. The purpose of this study is to understand how audits are conducted by using the theory of credence goods as the basis for explaining auditors incentives to act strategically. The results presented are consistent with audits being credence goods.

PAGE 13

13 CHAPTER 1 INTRODUCTION This study examines how an auditors potential information advantage about the audit process impacts audit production and the engagements profitability. The investigation is grounded on the th eory of credence goods which argues and predicts that an auditors knowledge and expertise about the audit may create incentives for the auditor to act strategically. In this perspective, audits exhibit credence attributes. In general, economists refer to a good as a credence good (service) if three conditions are met: first, the good (service) is provided by a seller who is simultaneously the expert that recommends the quantity of good (service) to the buyer. Second, buyers of credence goods (services) cannot assess how much of the good (service) they need and perforce must rely on a sellers recommendation.1 Third, buyers cannot assess how well the service was performed (Darby and Karni 1973 and Dulleck and Ker s chbamer 2006). In this setting, sellers have incentives to act strategi cally. Example: A frequently cited example of a cred ence service is the auto repair The auto mechanic performs a diagnosis and prescribes a treatment and in doing so he can honestly prescribe the appropriate treatment or he can exploit his informational advantage. For example, the mechanic may replace the engine when a new muffler would simply fix the problem; or the mechanic may charge for an engine replacement even when the customer receives only a new muffler; or the mechanic m ay simply repair the engine even if the car actually needs a new engine. Audits exhibit credence attributes in that the auditor is both the expert, the party with the audit expertise who recommends a specific level of audit service to his client, as well as the provider (seller) of the service. As the expert -seller, the audito r possesses an informational advantage with respect to the audit and may find it beneficial to engage in strategic behavior. 1 More specifically, buyers need is different from buyers demand because under the credence perspective the buyer cannot assess her own underlying condition and as a result cannot determine her own demand. In this case demand is supplier driven. Only if the buyer can perfectly diagnose her condition are need and demand interchangeable.

PAGE 14

14 More specifically, according to the credence goods theory, the auditor ma y find it profit maximizing to: Underaudit: provide less auditing than the client needs. Regulators have frequently voiced concern in the past that the presence of lowballed fees or non audit services may be accompanied with underauditing. Overaudit: provide more auditing than the client needs. Overauditing has been a concern to regulators since the passage of the SOX section 404 requirements. Overcharge: while an obvious breach of professional ethics, auditors may bill for hours not actually worked, especially if an engagement requires fewer h ours than quoted to the client. Ultimately, only the auditor can say how many hours of work are actually needed for an engagement to satisfy professional standards and achieve the desired level of assurance. Consequently, all of the above strategies may occur in the audit se tting. If the credence attribute of auditing leads the aud itor to behave strategically then this has important ramifications fo r audit quality and efficiency. Initially, this study provides a theoretical analysis of the auditors decision process under th ree different information environments. In the first setting, characterized by the availability of information associated with search goods, the buyers of audit services have perfect information about how much audit they need and how much of it they actual ly receive (Nelson 1970). As a result of this perfect information, the auditor has no incentives to act strategically. In the second setting, characterized by the availability of information associated with experience goods, the buyers of audit services know how much audit they need to buy but can only know after the service is rendered whether it is consistent with their need (Nelson 1970 and Klein and Leffler 1981). In this setting, reputation arises as an effective mechanism that constrains strategic act ion. In the third setting, characterized by the information asymmetry suggested in the theory of credence goods, the buyers of audit services do not know with certainty how much audit they

PAGE 15

15 need and rely on the auditor who essentially advises the buyers on the appropriate amount to buy (Darby and Karni 1973, Fong 2005, and Dulleck and Kerschbamer 2006). Moreover, given the unobservable nature of the audit service, it is difficult for the buyers to assess whether the audit service delivered to them is either the ex ante needed level or the ex ante promised one. In this information setting, reputation is not effective by itself because it relies on full discovery of the actual level of service needed and supplied. As a result, possible inefficiencies may arise consistent with the t hree strategies listed earlier. Given that reputation may not effectively discipline auditors why would some buyers still demand audits? Beyond the obvious regulatory requirement for an audit of the financial statements of a publicly -l isted company, it is argued in this paper that countervailing mechanisms are present in the market for auditing services that facilitate the conditions for a market to operate. Examples of such countervailing forces include litigation, regulation, standardization of audits, competition, reputation (to some extent), and a clients own knowledge of the external auditors processes. That is, the clients determine their choice (and trust) by observing the auditors credentials and by knowing that the audit production process is subject to regulation. However, the effectiveness of these mechanisms relies heavily on the clients ability to fully determine whether the level of service received equals the ex ante needed one, and whether it equals the promised level. Therefore, these mechanisms are important in that they facilitate an audit exchange, however, they cannot completely resolve the information problem. It is therefore important to consider the effect of the credence attribute on audit profitability and on the audit production process. In the first set of hypotheses this study argues that the auditor gains private information about a clients audit needs through repeated interaction with the client, and may be able to use

PAGE 16

16 this information to maximize his ow n profit (Beck and Wu 2006, Fong 2005). Therefore, as private information increases with tenure, so do the engagement profits. However, a clients own learning and switching costs may mitigate strategic acts over time (Lewis and Yildirim 2002, Fong 2005, D ulleck and Kerschbamer 2006). As a result of the interaction between the auditors private information and a clients switching costs over time, the engagement profits are predicted to increase with tenure, but eventually decrease. The theory of credence g oods also argues that sellers are less able to extract rents from consumers who are well informed about the products they demand (Fong 2005, Dulleck and Kerschbamer 2006). Following a similar rational e, it is predicted in this study that a clients own kno wledge about the audit process is associated with lower engagement profits. Using proprietary audit engagement data from one of the international auditing firms, I find that engagement profits are highest for mid tenure clients relative to the short and th e longtenured ones. Further, I also find that engagement profits are lower when clients have internal audit departments and are well prepared for the external audit, where the presence of internal audit departments and the clients level of preparedness p roxy for the clients knowledge about the audit process. These results are consistent with the hypotheses generated from the credence goods theory. In addition assessment of the effort and fee variables reveals that audit production is most consistent wit h overcharging. More specifically, higher audit profitability is achieved by exerting less effort than expected and by charging a higher fee than expected, where the expected effort and fees are calculated as the predicted values of the regressions that mo del audit effort and audit fees as a function of client characteristics. Overcharging may be the natural outgrowth of fixed fee contracting where the auditor uses his superior information to negotiate an arrangement that protects him from the possibility of unforeseen cost overruns that would be difficult to recover from the client.

PAGE 17

17 In the second set of hypotheses, I explore engagement characteristics associated with the overauditing strategy. It is argued that the auditor can engage in either, ex ante over auditing, recommending more audit than necessary prior to the start of the audit, or ex -post overauditing, increasing audit hours after the start of the audit. The results show that exante overaudits are positively related to the risk of material misstate ment, number of deficiencies found and time spent understanding the clients business, and negatively related to competition, client preparedness, time spent with non accounting personnel, the number of days until earnings release, and whether the client is a subsidiary of another organization. Ex -post overauditing is more likely to occur in clients that are publicly traded, and in clients that demand multiple reports from the auditor but is less likely to occur when the auditor performs tax services and when the client is well -prepared for the external audit. Moreover, some results show that ex -post overaudits have a disproportionate amount of effort allocated to cheaper ranks of labor. This study makes an important contribution to auditing research. Tra ditional audit research relies on two main assumptions about the information environment in audit markets. First, it is often assumed that clients can determine their d emand for assurance with certainty (Simunic 1980) and second, auditor reputation and siz e ensure high quality audits (DeAngelo 1981b). These assumptions cannot accommodate strategic behavior in the production of audit services thus ruling out any potential inefficiencies. However, recent audit production research and anecdotal evidence sugges ts that audits are generally not conducted on an efficient basis so these assumptions may be unrepresentative of the actual audit environment that buyers and sellers face (Dopuch et al. 2003, Kaplan et al. 2007, and Knechel et al. 2009). Therefore, it is i mportant to analyze how audits change when these assumptions are relaxed. The analysis is grounded on the theory of credence goods.

PAGE 18

18 There is also a more s ubtle contribution of this study on our understanding of audits. The theory of credence goods argues t hat an auditors strategic behavior is a function of the auditors knowledge about each specific client, therefore it predicts that strategic behavior and therefore audit production quality and efficiency are client -specific. This reasoning provides a basi s for quality differentiation at the engagement -specific level. In contrast to this perspective, traditional audit research argues that audit firm size is an effective mechanism at preserving quality (DeAngelo 1981b), an argument that provides a basis for quality differentiation across audit firms, but not within an audit firm The results in this study show that strategic behavior is primarily driven by variables that proxy for the auditors client -specific information, consistent with theorys prediction.

PAGE 19

19 CHAPTER 2 AUDITS AS CREDENCE GOODS: A THEORETICAL DISCUSSION OF AUDITORS DECISION MAKING PROCESS Credence Goods: Theory and Evidence In general, the economic literature suggests grouping product attributes into three categories according to the costs associated with quality detection (i.e., information asymmetry): search, experience, and credence attributes (Nelson 1970, Darby and Karni 1973). All products and services exhibit these three attributes but do so at varying degrees. The search a ttribute indicates little or no information asymmetry between buyers and sellers. That is, after some searching the buyer can discover the desired attribute prior to purchase (Nelson 1970). An example of a search attribute would be the color of a clothing item. The color attribute of a product can be easily searched and identified by a potential customer with little effort or cost. Experience attributes are characterized by ex ante information asymmetry whereby the buyer of the product is uncertain about the quality of the attribute prior to purchase, but the uncertainty is fully resolved after the buyer experiences the product (Nelson 1970, Klein and Leffler 1981). An example of an experience attribute would be the taste of a hamburger. A buyer has to exper ience the hamburger bef ore s he can determine the taste. The credence attribute is characterized by a two pronged information asymmetry: first, the buyer is not sure about how much of the service s he needs and relies on the expert advice of the seller, and second, the buyer is not sure about the quality of the service s he received, even after the service is consumed (Darby and Karni 1973, Dulleck and Kerschbamer 2006). The first aspect of the information asymmetry associated with credence attributes arises b ecause the buyer is not able to self diagnose her condition and therefore cannot determine the level of service needed. The decision of how much to buy is thus delegated to the seller, who as the expert

PAGE 20

20 recommends to the buyer what he thinks is an appropri ate level of service.1 The second aspect, uncertainty about the level of service received, can be further broken down into: (1) uncertainty as to whether the service received equals the level that the seller initially promised and (2) uncertainty as to whe ther the service received equals the level that the buyer needed. The uncertainty about the service received may persist indefinitely (i.e., it is very costly for the buyer to determine the quality of ser vice). It is important to distinguish the informational properties of the various attributes because they affect the sellers behavior. The greater the information asymmetry, the greater the incentives for the seller to act in his self -interest. For example, in a search setting, the seller possesses little or no information advantage over the buyer and therefore has no incentives to act strategically In the experience setting, ex ante information asymmetry creates incentives for the seller to act strategically If the seller cheats however, the buyer punish es him by switching to a different seller. Therefore, for a seller who cares about his reputation, the disincentives to act strategically (i.e., reputation loss) clearly outweigh the incentives. In a credence setting however, there is rarely an ex -post sig nal that can reliably inform buyers about sellers behavior and, as a result, a countervailing mechanism like the loss of reputation may not be effective. In this setting, the seller always possesses the information advantage and the incentives to act stra tegically remain. More specifically, the theory of credence goods predicts that in equilibrium the seller is able to choose among three available strategies: undertreatment, overtreatment and overcharging (Dulleck and Kerschbamer 2006): 1 In the credence market, the seller is able to diagnose the buyers condition with more precision than the buyer.

PAGE 21

21 Undertreatment occurs when the seller of the service provides less service than the level that the buyer needs. Overtreatment occurs when the seller provides more service than the level that the buyer needs. In this case, the additional benefits to the consumer from having received a high quality service are less than the associated costs thus creating inefficiencies. Overcharging occurs when the seller charges for a higher level of service than actually supplied.2 Numerous products and services exhibit cr edence features including car repairs, medical treatment, legal representation, financial advice and other types of services in which the provider of the service also serves as the expert who recommends a specific level and type of service to a potential c ustomer. Empirical evidence is consistent with the argument that sellers of credence goods act strategically Emons (1997) reports a Swiss study which showed that the average population had 33 percent more surgical interventions than physicians and their f amilies. Wolinsky (1993) reports a Department of Transportation study which found that 53 percent of auto repairs represented unnecessary repairs. Both illustrate examples of overtreatment. Bartels et al. (2006) study the consumer selection of water heater s when the plumber is the party who advises consumers on water heaters and also the party who installs them. They find that the plumbers advice deviates from maximizing consumers utility in the direction that maximizes plumbers profit margins. Hubbard ( 1998) investigates the probability that motor vehicles fail emission inspections in a private firm that also carries out the repairs relative to state inspectors who simply provide the inspection but do not carry out the repairs. He finds differences acros s the two settings such that the probability of failure is higher in private firm settings consistent with the notion that when the seller of a service is also the expert there is a tendency to provide an inappropriate amount of service that benefits the s ellers because it generates a higher demand 2 In later sections of the paper I explain these strategies in detail as they may apply in the audit setting.

PAGE 22

22 for their services. Bluethgen et al. (2008) argue that financial advice is another type of credence good whose quality is very costly to determine. In such a setting, they argue, financial advisors have incentive s to provide advice that maximizes their own profits but not necessarily those of the clients. Moreover, compensation structure can mitigate or deteriorate such incentives. They find that the quality of financial advice to investors is higher when advisors receive a lesser proportion of their compensat ion in the form of commissions. Finally, Rubin (2004) argues that legal services also exhibit credence characteristics which arise primarily due to the complexity of the law and make it hard for the client to determine the quality of service they receive. Further exacerbating the information asymmetry is the fact that each legal case is idiosyncratic and requires judgment on the part of the expert, judgment that is difficult to assess by both experts and non -ex perts. Auditing u nder the Credence Lens The Credence Attribute of Auditing Audits are another example of a good that exhibits credence features. In the market for auditing services the client demands an audit and hires an auditor to perform the audit. In a ddition to being the service provider, the auditor is also the expert, in that he advises the client about the level of audit service the client needs. For example, when bidding for a new client, the audit fee embedded in the bid informs the client about w hat the auditor thinks is an appropriate level of audit service. Moreover, even after the initial agreement on the audit fee, the auditor may decide to expand the audit scope if he deems it necessary to do so. Thus, as the expert, the auditor uses professi onal judgment to tailor the audit to the specific client needs and determines the audit effort necessary to achieve the desired assurance level (OKeefe et al. 1994). The use of judgment exacerbates the credence attribute of the audit because it becomes ve ry difficult ex post to assess the appropriateness of the audit, even when one may be able to determine whether

PAGE 23

23 minimum standards were met.3 Second, the client is uncertain about the level of assurance received because the assurance level is not observable (Francis 2004, Barton 2005). More specifically, the client is not sure whether the level of assurance actually received equals the level of assurance that the auditor initially promised, or whether it equals the level of assurance the client needed.4 Ther efore, as the expert -seller, the auditor possesses an information advantage over the client about the production of the audit and may find it beneficial to engage in strategic behavior. The nature of strategic behavior is discussed in detail below. Placing the Credence Attribute in Context In order to understand the implication of the credence attribute for the audit production process, it is important to first understand the information environment in the credence setting and how it differs from other sett ings. Figure 2 1 depicts the location of search, experience, and credence goods along the information spectrum. Note that credence goods are located at the end of the spectrum denoting the highest level of information asymmetry between buyers and sellers. Search goods are located at the opposite end indicating little informational problems because quality can be determined prior to the purchase. Standing between search and credence goods, are experience goods whose attributes are fully discovered only aft er consumption. As depicted in F igure 2 1, the difference between search and experience goods is the timing of the resolution of information asymmetry, exante vs. ex post, respectively. However, the difference between experience and credence goods is whethe r a resolution occurs. In the case of credence goods information asymm etry may persist indefinitely. 3 Another example of the auditor acting as an expert is when he advises the client on other matters like potential improvements in the accounting information system and internal controls, or other types of advisory services. 4 Audit failures are the only events in which the true level of assurance is publicly revealed. However, audit failures are rare and may under represent the actual audit failure rate (Francis 2004). Moreover, given that audits are not guarantees, audit fail ures could simply be normal outcomes of the audit process.

PAGE 24

24 In order to understand the implication of the credence attribute for audit production it is important to distinguish between the informational properties of search, experience, and credence goods and to carefully lay out how these properties affect an auditors decision making process. The next section begins with a general discussion of the auditors decision making process exclusive of any informational as sumptions. Information asymmetry is gradually introduced and the auditors decisions are examined under three different information environments: (1) the search setting, (2) the experience setting and (3) the credence setting. A generic model of an auditors decisionmaking process Figure 2 2 shows an auditors decision making process with respect to a new engagement. Assume initially that nature randomly determines a clients need for audit denoted by Q*. Lets assume that there are only two client types (based on their audit need): those that need high audit effort (hours), denoted by H, and those that need low effort, denoted by L.5 H and L represent a clients true demand for assurance assuming no information asymmetry. In his first decision node, the auditor decides how much audit he recommends, denoted by QR. In this case, the auditor decides between recommending a high level of audit by placing a high bid (FH) or a low level of audit by placing a low bid (FL), where FH>FL. The bid is the total f ee that the auditor expects to receive from the potential client. Assuming that P is the price per unit of effort, the total fees are defined as: FH = P*H and FL = P*L.6 A bid equal to FH (FL) implies that the auditor is promising to provide quantity QR = H (L), respectively. After receiving the bids, the client decides whether to accept (A) or reject (R). The client only accepts a high bid with probability, 5 The effort differences simply reflect risk differences of clients and a low effort audit is not an indication of a substandard audit. 6 It is thus assumed that the price per unit of labor is constant regardless of whether the auditor is dealing with an H or L client (and assuming an equal labor mix). This assumption reflects the more likely scenario that billing rate is not private information and clients can easily find out the rate charged by different partners. Thus, the differences in the fees arise primarily due to differences in the effort level or effort allocation.

PAGE 25

25 but s he always accepts a low bid. This stylized decision -making framework is consistent with the aud it fee low -balling phenomenon in the auditing environment. Once the client accepts the bid, the auditor chooses the amount of audit effort to supply denoted by QS. He may decide to expend high effort (H) or low effort (L). If the auditor expends high effor t (H) he incurs a cost equal to CH, whereas if he expends low effort (L), he incurs a cost equal to CL, where CH>CL. The auditor may appropriately recommend and provide a level of audit effort that is commensurate with the clients underlying condition or he may deviate. If the auditor decides to promise and provide a level of audit effort that is commensurate with the clients underlying demand then the effort recommended, supplied, and needed are equal (QR = QS = Q* ), otherwise these three quantities cou ld differ.7 In a later section, the paper explains exactly how the auditor can behave strategically in this setting. Auditors decisionmaking process: search, experience, and credence frameworks Figure s 2 3, 2 4, and 2 5, present the auditors decision ma king process under the assumption that: (a) audits are search goods (b) audits are experience goods and (c) audits are credence goods, respectively. Lets first analyze the situation when audits are assumed to be search goods. If audits are search goods, t hen by definition there is no information asymmetry between the client and the auditor at the time of hiring (Nelson 1970). Any information asymmetry that may have existed is resolved ex ante. A search environment assumes that 1) the client can determine with certainty her underlying condition Q* and demands it (thus Q* = QR) and 2) the client can tell quality (QS) ex ante. As a result of this full information, the client hires only auditors who provide the needed quantity (QS = Q*). The auditor is not abl e to deviate 7 To clarify this point even further, there are three quantities of audit effort to keep track of in the credence setting : first, the true quantity of audit that the client needs ( Q* = optimal level); second, the quantity of audit recommended/promised by the auditor ( QR = embedded in the audit fee); and third, the quantity of audit actually supplied by the auditor (QS)

PAGE 26

26 because if he does he will not be hired. This anticipated punishment from the client forces the auditor to always provide the audit effort that the client demands/needs (Q* = QR = QS). Figure 2 3, which presents the auditors decision making process under the assumption that audits are search goods, includes the auditors (top line) and the clients (bottom line) net payoffs. The auditors net payoff is total fees minus the cost of production, whereas the clients net payoff equals the gross ut ility s he gets from receiving the needed quantity, denoted by u, minus the audit fee paid to the auditor. It is also assumed that auditor and client payoffs are zero in case of a rejection. Note that because it is a ssumed that the client knows her true demand, certain decision nodes are not possible and are shown in dashed lines. For example, it is not possible for the auditor to provide quantity L to a client who demands quantity H. Likewise, a client always rejects a high bid (FH) if her underlying demand is L. In the case of an H client, the dominating strategy for the auditor is to bid FH, expend the costly effort CH, and realize profit FHCH.8 In the case of an L client, the dominating strategy is to bid FL, expend effort CL and realize profit FL-CL.9 T he main result of this analysis is that in the search setting the optimal strategy for the auditor is to promise and provide a level of audit that equals the clients underlying demand for that service. Lets now assume that auditing is a pure experience good (see Klein and Leffler 1981, Shapiro 1983). Under t he experience setting shown in F igure 2 4 it is assumed that 1) the client is still able to determine with certainty the quantity of audit that s he needs (Q*) and 2) the client is not able to determin e the delivered effort (QS) at the time of hiring (ex ante) but s he is able to observe it ex -post. The first assumption is identical to the search setting and as a result, the client always rejects a high bid (FH) when her true underlying condition simply requires a low effort 8 Note that the auditor can also bid FL, expend a high effort CH and realize profit FLCH. However, this strategy is dominated by the other strategy where the auditor bids FH and expends the costly effort CH, because FHCH>FLCH. 9 Similar to footnote 8 the audi tor can also bid FL and expend the high effort, but again this strategy is dominated.

PAGE 27

27 (L). The difference between the experience setting and the search setting arises from the second assumption. In experience settings, the full resolution of information asymmetry is achieved only after the service is delivered (ex post ). The perfect ex -post information is reflected in the clients utility term. If the client receives an audit effort that equals the true underlying demand, clients gross utility is u, however, if the client receives less audit effort than needed (i.e., r eceive L when s he needs H), s he incurs a disutility and her gross utility is now v, where v
PAGE 28

28 one time payoff he receives for providing L. Assuming that interest rate is r, the auditor therefore provides H only if (FH-CH)/ r>(FHCL). This inequality is satisfied if the auditor is rewarded for exerting high effort through a fee premium. Thus, in equilibrium the auditor always provides H and receives a premium for the high effort (Klein and Leffler 1981, Shapiro 1983). Under t his framework of price ensures quality the auditor has an incentive to provide high audit effort whenever the present value of all future rents earned on a client exceeds the short term value of simply shirking, collecting the first years rent and exiti ng the market. Because the existence of a premium does not fit well with competitive markets it is assumed that fee premiums dissipate through their re -investment into a brand name or reputation, the latter being the only observable signal to potential cli ents regarding auditor quality. Reputation is costly to establish and thus serves as the collateral for auditor quality (Watts and Zimmerman 1986). The main point of this discussion is that similar to the search setting, under the experience setting the op timal strategy for the auditor is to promise and provide a level of audit effort that equals that of the clients underlying demand (Q* = QR = QS). However, different from the search setting, the auditor is disciplined through the reputation mechanism and is compensated for the higher effort through a price premium. The difference in the result is driven by the differences in the informational environments of the two settings. Finally, if an audit is a credence good then it is assumed that 1) the client is not able to determine his true underlying condition (Q*) and 2) the client cannot determine the quality of the service (QS) even after the service is consumed/received (Dulleck and Kerschbamer 2006). Assume for a moment that the information asymmetry is so extreme that the client learns nothing about the level of assurance ex -post. In this environment, the client always believes that s he

PAGE 29

29 received the needed level of assurance (even when s he did not).11 The result is that the client cannot punish the auditor if the auditor deviates. Thus, in equilibrium, an auditor always deviates because it is profit -maximizing for him to do so at no cost. More specifically, an auditors preferred course of action in this case would be to promise H but to deliver L (thus Q*, QR, and QS differ), regardless of whether he is facing an H or an L client and realize profit FHCL. The information environment presented above is extreme and perhaps unrealistic because in most cases, there is no perfect information asymmetry. Most crede nce goods are not pure in the sense that some information with respect to quality is revealed ex -post. This situation is presented in F igure 2 5 Here, the auditor can be punished if he is caught deviating. However, due to the uncertainty of the informatio n, the punishment mechanism is not as effective as in the experience setting. The punishment in the credence setting is equal to the expected value of losses that auditor incurs in case he deviates, denoted by E(M). The expected value of losses equals the 12 It is obvious that the auditor acts strategically only if the expected value of losses, E(M), is less than the gross margin advantage a ssociated with the strategic actions. The possibility that some information about the delivered level of assurance is revealed ex post is also reflected in the clients utility terms. Any time audit effort is below th e expected effort, the client incurs a disutility denoted by e which is the expected value that the auditor is caught deviating times the amount of disutility that the client incurs. Depending on the extent of information asymmetry, the magnitude of e ran ges from zero to u v. 11 As will be explained later, this belief is driven by the existence of other mechanisms (like re gulation and standardization of audits) that serve to discipline auditors. 12

PAGE 30

30 Note that the experience and credence settings are related through term E(M). In a pure experience setting, there is no uncertainty with respect to the delivered level of assurance. Here, the auditor reputation serves as a perfect mon itor and the auditor has no incentives to deviate and hand, in a pure credence good setting where there is no information about the level of the supplied assurance ineffective and the auditor has strong incentives to deviate. The difference in the auditors equilibrium behavior in the experience and credence settings arises from the differences in the then audits are more like credence goods and reputation is ineffe ctive. Auditors Strategic Space The previous section concludes that the credence aspect adds an important dimension to the audit production process. More specifically, the credence attribute exacerbates the auditors information advantages and provides opportunities to the auditor to act in his self -interest. Using the theory of credence goods as the base, in this section, I discuss the following: (a) the possible strategic choices available to the auditor under the as sumption that audits are credence goods; (b) disciplining mechanisms present in the market for auditing services which constrain but do not completely eliminate an auditors incentives to act strategically ; and (c) the auditor s rational behavior given the above incentives and constraint s. I begin this section with a definition of audit q uality in the credence context. Audit quality as discussed here is the degree of mapping between the true demand for audit assurance and the amount prescribed/serviced by the auditor. The true demand is the clients demand under the full information case (i.e., no information asymmetry is present

PAGE 31

31 between auditor and client and client is able to self -diagnose and verify the auditors provision of the service with certainty) Thus assurance level in this context has to do with the whole process that produced it (quality of production process) rather than the quality of the outcome itself. Any deviation between the true needed assurance and the supplied assurance, from below ( or above), qualifies as a low -quality (inefficient) audit. By analogy to the theory of credence goods auditors can strategize by either: (a) underauditing (undertreatment) (b) overauditing (overtreatment) and (c) overcharging (Dulleck and Kerschbamer 200 5, Dulleck and Kerschbamer 2006, Dulleck and Kerschbamer 2009). Underauditing occurs when the auditor provides less audit than needed; Overauditing occurs when the auditor provides more audit than needed; Overcharging occurs when the client pays for services not supplied (the supplied amount is less than promised). For a formal discussion of the strategies refer to F igure 2 6 which replicates the decision tree presented earlier but also includes the three strategies available to the auditor at the bottom (s ee also Dulleck and Kerschbamer 2006). The discussion i n this sub -section focuses on reviewing the details of credence goods theory as applied to auditing by considering the specifics of the market for auditing services. As a reminder, once we assume that the audit has credence features, the following assumptions hold: the client is not able to self -diagnose her true condition and the client is not able to verify the quantity of service that the auditor supplies. As before, the clients true condition is such that s he requires either high or low audit effort, thus Q* = H or L. The auditor chooses t o bid FH or FL, where each bid reflects the recommended effort, QR = H or L respectively. Further, the client accepts (A) or rejects (R) the bid, and the auditor decides on how much effort to supply, thus QS = CH or CL. The auditor may expend an effort that equals the effort he promised (the level implied in the bid) or he may decide to provide a level of service

PAGE 32

32 that differs from the promised level. In a similar v ein, the auditor may decide to provide a level of service that equals the level that the client needs or may provide a level entirely different. Lets now discuss each strategy. Underauditing As a reminder, underauditing occurs when the auditor provides less audit than needed (Dulleck and Ker s chbamer 2005). Using the notati on this means that QSQ*. In F igure 2 6 overauditing amounts to strategy (L, FH, CH).14 Under this strategy the client needs a low level of effort (L), however the auditor is bidding a high fee FH, (indi cating overselling of services) and also provides a high effort commensurate with the fee (CH). In this case, the supplied effort is greater than the needed level. Overcharging Overcharging occurs when the auditor bids a high fee (FH) but only provides the low -cost effort (CL) (Pitchik and Schoetter 1987 and 1993, Wolinsky 1993 and 1995, Fong 2005 and 13 This scenario also amounts to overcharging. 14 Although strategy (L, FL, CH) implies more effort than needed, it is not the overauditing strategy in the sense described here. If the auditor is willing to put more effort than he is compensated for (as this choice indicates), he is being professionally conservative, not overauditing.

PAGE 33

33 Dulleck and Ker s chbamer 2006). Using prior notation this means that QS
PAGE 34

34 Disciplining Mechanisms in the Market for Auditing Services In a credence service transaction, the buyers are not able to judge the intrinsic characteristics of the service and make their choice based on their faith in the producer (hence the credence label). The existence of a market for credence goods is made possible by either the reputation of the seller or a quality guarantee by a third party, often in the form of regulation. However, these mechanisms only provide the buyers with a substitute for the information or trust they lack. That is, even when these mechanisms are present, information asymmetry may exist. Analogous to othe r credence services, due to the hidden nature of the audit assurance level, it is costly for the client to verify the assurance level received. However, clients may be comforted by the presence of several mechanisms that discipline auditors. For example, legal liability, regulation, and reputation concerns mitigate the incentives to underaudit. Furthermore, a clients own knowledge and the level of competition can mitigate the incentives to overcharge and/or overaudit. Disciplining mechanisms specific to aud it markets are discussed below. Countervailing Mechanisms to U nderauditing Legal liability and regulation Underauditing has important ramifications for audit quality and therefore powerful mechanisms exist to counter the incentives to underaudit. Litiga tion costs are one such mechanism. Under common law, auditors are liable to their clients and to other third parties for negligence and breach of contract if they do not exercise due care in the audit engagement. Auditors are also held liable to third part ies under the statutory law (Smieliauskas 1996). The expected losses from litigation motivate the auditor to exert a high audit effort in order to mitigate the potential litigation (Simunic 1980, Dye 1993, Willenborg 1999, Khurana and Raman 2004).

PAGE 35

35 In addit ion to litigation, self regulatory mechanisms like Generally Accepted Auditing Standards (GAAS), the AICPA Code of Professional Conduct, Quality Control Standards, peer reviews, and more recently, PCAOB inspection reports regulate and standardize the audit process. The regulation creates boundaries to what an auditor can and cannot do and therefore may serve as a vehicle that reduces underauditing. (Watts and Zimmerman 1983, Watts and Zimmerman 1986, Knechel et al. 2007, Arens et al. 2001). The countervaili ng mechanisms of litigation and regulation, while important, are not absolute. For example, the broad nature of regulation provides the auditors some degree of latitude in compliance (Defond and Francis 2005). Therefore, the boundaries of what constitutes an underaudit are fuzzy and can be exploited by the auditor. Further, legal -based punishment relies heavily on the prospect that underauditing is caught. However, given that the litigation rate is very low (Carcello and Palmrose 1994), it is not clear that litigation costs are fully capable of countering the incentives to underaudit. Audit firm reputation and size The reputation framework posits that the client can punish the incumbent auditor by switching to a different auditor, if it is discovered ex post that the incumbent auditor did not provide the expected level of assurance (Klein and Leffler 1981, Watts and Zimmerman 1986). In the reputation equilibrium, auditors earn quasi rents which are returns for the nonsalvageable-type investments (i.e., reput ation) that signal a credible commitment to quality. DeAngelo (1981b ) takes this argument a step further and posits that large audit firms (i.e., those with more clients) have more incentives to provide a high assurance audit because they suffer greater lo sses (i.e., lose more clients) if their reputation is somehow damaged. Critical to obtaining this result however, is the assumption that the auditor is caught underauditing (i.e.,

PAGE 36

36 discovery of lower than -expected assurance level). However, as previously di scussed, the actual assurance level may not be discovered and as a result the client may have no basis for switching. Countervailing Mechanisms to O verauditing and O vercharging Cl ients knowledge Incentives to overcharge and overaudit can be reduced by cli ents own knowledge and by the level of competition present in the market for auditing services (Fong 2005, Dulleck and Kerschbamer 2006). The credence goods theory argues that buyers vary in how much they know about the services they purchase. A more informed client mitigates the incentives to overcharge and/or overaudit because essentially both these strategies require that the auditor be able to convince the client about how much service to buy. An informed client is less likely to accept advice that exa ggerates the needed level of service and can more easily detect when the auditor provides a level of service that deviated from the promised level. This argument suggests that the auditors incentives to act opportunistically are client -specific and depend on the clie nts knowledge about the audit. Competition Competition among auditors also makes overauditing less likely. This is because when audit firms compete for business they are more likely to low -ball their fees (DeAngelo 1981 a ), a situation that aut omatically rules out overauditing.15 In this aspect, the client can respond to the overauditing strategy by switching the incumbent auditor and soliciting new bids. The threat from a potential switch may dissuade the auditor from the lure of overauditing. O verall, the disciplining mechanisms in the market for audit services create the conditions for a market to exist. That is, the clients determine their choice (and trust) by observing the 15 Recall that overauditing arises when both effort and fees are higher than expected. If any of these parameters is lower than expected, overauditing is automatically ruled out.

PAGE 37

37 auditors credentials and by knowing that the audit production proces s is subject to regulation. However, the effectiveness of these mechanisms relies heavily on the ability to fully determine whether the level of service received equals the ex ante needed one, and whether it equals the promised level. Therefore, these mech anisms are im portant in that they facilitate the operation of audit markets however, they c annot completely resolve the information problem. It is therefore important to consider the effect of the credence attribute on audit profitability and on the audit production process. Auditors Rational Behavior This section incorporates the effect of potential benefits and costs associated with each strategy on the auditors strategic choice. In order to facilitate the discussion refer to F igure 2 7 which replicat es F igure 2 5 with some adjustments to the auditors payoff. Recall from F igure 2 5 that the auditor incurs a penalty equal to E(M) for acting strategically It is now time to be more specific as to the nature of this penalty. If the auditor faces a client that requires H effort he can choose between underauditing and overcharging (refer to F igure 2 6 ). If the auditor underaudits he runs the risk of incurring a cost that may take the form of regulatory/legal/reputation punishm ent. To reflect this cost, in F igure 2 7 auditors payoff is reduced by amount E( m ) which is equal to the probability that the underaudit is discovered, itude of penalty, denoted by m If the auditor is facing a client that requires low effort (L), the audi tor can choose between overauditing and overcharging. In this case, auditors payoff is reduced by amount E( d ), which captures the expected value of the losses that auditor incurs in case the client detects this type of strategic behavior and decides to hire a new auditor. E( d ) is composed of two components: (1) denoted by d The magnitude of the punishment is basically the present value of future profits

PAGE 38

38 that are lost if a switch takes place. The higher the probability of a switch and the higher the potential punishment the more costly is for the auditor to either overaudit and/or overcharge. As a result, in the all nodes that are associ ated with either an overcharge or an overaudit the auditors payoff is reduced by the amount E( d ).16 Table 2 1 lists all the strategies along with the auditors net profit for each strategy.17 Case 1: H -Type C lient Assume initially that the auditor faces a client that requires high effort (H). From F igures 2 6 and 2 7 the auditor decides between the following strategies: S1. (H, FH, CL) Overcharging/underauditing, auditors net payoff = FHCL-E( m )-E( d ) S2. (H, FL, CL), underauditing, auditors net payoff = FL-CL-E( m ) S3. (H, FH, CH), promising and providing the needed audit,18 auditors net payoff = FHCH The choice between S1 and S2 depends on the magnitude of E( d ), which is the additional expected cost associated with S1. If E( d ) is large enough to wipe out the gross margin benefit of S1, then S2 dominates. If this is the case, the auditors next step is to compare the net benefit between the two remaining options, S2 and S3. The decision here is based on the magni tude of E( m ), which is the expected cost associated with S2. The decision is such that if E( m ) is too large then auditor chooses S3 and thus does not deviate from the expected behavior (i.e., auditor promises and provides the needed effort); otherwise unde rauditing (S2) dominates. On the other hand, if E( d ) is low enough then S1 dominates S2. If this is the case, auditor then decides between S1 and S3. The decision is based on the magnitude of E( m ) + E(d ) such that if this magnitude is large enough to wipe out the additional benefit associated with S1 then 16 Note that this setup assumes that the cost of overcharging and overauditing is the same. This can easily be changed by assuming that the probability of detection and the cost to the auditor is different depending on whether he pursues overauditing or overcharging. However, this would only make the setting more complicated wi thout adding any new insights. 17 The payoffs shown in T able 21 are only for the case when the client accepts. The payoffs to both auditor and client are assumed to be zero in case of a reject and are thus removed from the table. 18 Strategy (H, FL, CH) i s dominated and is thus not considered.

PAGE 39

39 auditor pursues S3, (promise and provide the needed effort). Appendix A provides a more detailed discussion of the auditors choice of the most preferred strategy. Case 2: L -Type C lient If the auditor fac es a client that needs low effort (L) then the auditor chooses between the following strategies: S4. (L, FH, CH), overauditing, auditors net payoff = FHCH-E( d ) S5. (L, FH, CL), overcharging, auditors net payoff = FHCL-E( d ) S6. (L, FL, CL), promising and providing the needed effort, auditors net payoff = FL-CL It is clear that in choosing between S4 and S5, the latter dominates because FHCL>FH-CH. 19 The auditor chooses between the remaining strategies, S5 and S6, a choice that depends on the magnitude of E( d ). If E(d ) is low enough auditor chooses overcharging (S5). See Appendix A for a more detailed discu ssion of each strategic choice. Overall, the choice between strategic actions is a function of the expected costs associated with each strategy. Th ese costs can range from the strictness of the litigation environment, the auditors client -specific knowledge, and the clients own ability to verify auditors audit production process. The choice of acting opportunistically depends on the expected benefi t that auditor realizes, which in turn depends on the expected cost, Discussion Traditional audit research relies on two powerful assumptions with regards to the information environment in the market for auditing services. Fi rst, it assumes that clients can self -diagnose and thus are able to completely assess their own demand for a ssurance. For example, Simunic (1980) assumes that clients determine how much audit they need to buy based on their trade of f analysis between inter nal and external monitoring. This view however, 19 Note that this result is by design because it was earlier assumed that the expected cost associated with overcharging (S5) equals that from overauditing (S4). In order for overauditing to arise as a possible strate gy, it must be assumed that overcharging is irrelevant

PAGE 40

40 invalidates the auditors function as an expert who recommends the appropriate audit amount to a client, in addition to providing it. In contrast, under the credence perspective this assumption is relaxed and the auditors role as the expert is restored. Second, traditional research also assumes that the actual assurance provided can be observed and auditors punished if they deliver less assurance than promised. For example, DeAngelo (1981b ) posits that large audit firms (i.e., those with more clients) have more incentives to provide a high assurance audit because they suffer greater losses (i.e., lose more clients) if their reputation is somehow damaged. Embedded in this result however, is the assumption that the auditor is caught whenever overauditing, overcharging, or underauditing takes place (i.e., discovery of true assurance level). However, as previously discussed, the actual assurance level may not be discovered and as a result the client may have no basis for punishing the auditor. Under this restrictive setting audits are essentially experience goods and as such no inefficiencies are expected. More specifically, in this setting the needed (Q*), the recommended (QR) and the delivered audit efforts (QS), all collapse into one. On the other hand, the credence perspective simply deals with investigating audit producti on choices if these two assumptions are relaxed. In this less restrictive environment strategic behavior on the part of the auditor is possible. The empirical evidence to support this view is at this point, sparse. A few examples include Brozovksy and Rich ardson (1998) who find that when audit quality is publicly known (as in the experience setting) the reputable auditors are able to charge a higher price (thus price ensures quality mechanism is effective ). However when the underlying audit quality is not l earned (as in the credence setting and in the real audit markets), theres no price premium for reputation. Although this result tells us nothing about the strategic behavior of the auditor, it is nevertheless a very important result because it shows that in conditions of unobservability

PAGE 41

41 reputation is not effective and therefore no price premiums are necessary. Similarly, Kaplan et al. (2007) argue that Andersens investments in reputation did not prevent the firm from providing some low quality audits. In an experimental study, Mayhew (2001) finds that in general reputation ensures high quality audits. However, if reputational rewards/punishments are delayed (i.e., auditors are not immediately punished for providing bad audits) then in some cases auditors c hoose to provide low quality audits. This result is very interesting because it shows that softening reputation as a disciplining mechanism by introducing uncertainty leads to strategic behavior. The author emphasizes that this result was not expected give n the assumptions initially imposed (i.e., that audits are experience goods and as such reputation will always discipline auditors). 20 Similarly, Mayhew et al. (2001) find that when there are no accounting uncertainties, auditors are concerned about their reputation and thus report information objectively. However, introducing accounting uncertainties leads to a different equilibrium behavior. More specifically, in this environment auditors do not seem to care about the reputation as they are willing to co mpromise their independence thus resulting in low quality audits. Further, Mayhew et al. (2001) also conclude that reputation concerns alone do not deter objectivity impairment under conditions of uncertainty Doyle et al. (2007) find the surprising resul t that material weaknesses associated with section 302 of SOX are more strongly associated with the accrual quality than the ones under section 404. They interpret this finding as auditors being more careful regarding the 404 20 In fact, Klein and Leffler (1981) allude to something similar in their footnote 6: If we modify the assumptions of our model to make interconsumer communication less perfect and allow inflows of new ignorant consumers over time and permit individuals to forget, the potential short run cheating gain by firms would be increased. Therefore, the quality assuring price premium would be higher. That is, in the framework of experience goods, an increase in the length of time required for the full resolution of information asymmetry simply leads to an increase in the premium. As long as the truth is learned (regardless of the timing), reputation works fine. If information asymmetry persists indefinitely su ch as the case of credence goods, then such premium would be prohibitively high. A similar argument is made in Shapiro (1983). In his model, if quality is never learned the premium increases to infinity (i.e., premium is no longer effective to ensure quali ty).

PAGE 42

42 requirement on internal contro ls (i.e., overauditing) by applying tougher standards and thus classifying some minor problems as material weaknesses. This may increase the number of material weaknesses that have no implication on the quality of internal controls and hence no effect on a ccrual quality. Moreover, Ogneva et al. (2007) find that many of disclosures related to material weaknesses do not lead to a higher cost of capital suggesting that material weaknesses may have no economic implication. OKeefe et al. (1994) state that aud itors frequently speak of value -billing their services which suggests profit maximization is the relevant objective. The paper further suggests that auditor s may i gnore the quality of controls when determining the level of effort to expend on an engagemen t. Moreover, contrary to the expectation that auditor tenure should increase learning the paper finds that audit hours do not decrease over time. Taken together this evidence suggests instances of overauditing. OKeefe et al. (1994 ) provide an alternative explanation to their unexpected finding of no learning over time. They argue that the lack of learning could be because the firm may have not provided the target level of assurance in the first year and is making up for it in the later years. The latter explanation means that the auditor in effect is underauditing during the first year and then overauditing during the later years This is also consistent with St. Pierre and Anderson (1984) who find that a large number of lawsuits occur during the first few years of auditor association with the client again pointing to the existence of underauditing during the early years The evidence cited above may not fit well with the current highly restrictive auditing framework. In contrast, the credence goods theory c an accommodate and explain these richer stories. It is thus important to formally consider how the credence attribute of auditing affects the audit production process.

PAGE 43

43 Figure 2 1 Location of goods along the information spectrum based on their information properties

PAGE 44

44 Figure 2 2 A g eneric m odel of an a uditors d ecision -m aking p rocess Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = A uditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of providing the audit to the client. CH (CL) is the cost of providing the high (low) effort, where CH>CL.

PAGE 45

45 Figure 2 3. Audits as search goods Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = Auditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of providing the audit to the client. CH (CL) is the cost of providing the high (low) effort where CH>CL. u = C lients gross utility when s he receives at least the needed quantity

PAGE 46

46 Figure 2 4. Audits as experience goods Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = Auditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of providing the audit to the client. CH (CL) is the cost of providing the high (low) effort, where CH>CL. M = Loss that auditor incurs when acting strategically v = C lients gross utility if s he receives less effort than needed u = C lients gross utility when s he receives at least the needed quantity

PAGE 47

47 Figure 2 5. Audits as credence goods Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = Auditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of providing the audit to the client. CH (CL) is the cost of providing the high (low) effort, where CH>CL. E(M) = E xpected losses that auditor incurs when acting strategically the level of the supplied assurance is revealed and M the magnitude of punishment e = Expected disutility to the client if auditor provides less effort ( assurance) than needed u = C lients gross utility when s he receives at least the needed quantity

PAGE 48

48 Figure 2 6. Auditors d ecision p rocess under the c redence l ens Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = Auditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of providing the audit to the client. CH (CL) is the cost of providing the high (low) effort, where CH>CL.

PAGE 49

49 Figure 2 7. Audit ors d ecision p rocess under the c redence l ens A uditor and c lient p ayoffs Note: H, L = Q uantity of audit effort that client needs. H (L) means that the client needs a high (low) level of audit quantity. FH, FL = Auditors bid. The auditor decides to bid either a high (FH) or a low (FL) fee. FH = P*H; FL = P*L, where P is the price per unit of effort. A, R = C lients decision after auditor bids. Client decides to accept (A) or reject (R) the bid. CH, CL = A uditors cost of prov iding the audit to the client. CH (CL) is the cost of providing the high (low) effort, where CH>CL. E( m ) = E xpected value of losses due to litigation. E( m m m is the magnitude of losses. E( d ) = E xpected value of losses due to clients switching threat. E( d d and d is the present value of foregone profits if client switches. e = Expected disutility to the client if auditor provides less effort than needed u = C lients gross utility when s he receives at least the needed quantity

PAGE 50

50 Table 2 1. Auditor strategies, auditor payoff and client utility Client's condition Auditor's choice of action Explanation of strategy Auditor's net payoff Client's net utility H F H C H No deviation F H C H u F H H F H C L Underauditing and overcharging F H C L E( m ) E( d ) u F H e H F L C H Discount F L C H u F L H F L C L Underauditing F L C L E( m ) u F L e L F H C H Overauditing F H C H E( d ) u F H L F H C L Overcharging F H C L E( d ) u F H L F L C H Conservative F L C H u F L L F L C L No deviation F L C L u F L

PAGE 51

51 CHAPTER 3 DATA AND GENERAL DES CRIPTIVE STATISTICS The data used in this study are obtained from U.S. audit engagements performed by one of the big international accounting firms. The data pertains to the most recent annual audits completed prior to the effective date of the audit provisions of the Sarbanes -Oxley Act of 2002 (SOX) The audit firm collected this data as part of their annual internal quality control reviews. Because firm policy mandates compliance with the annual quality review questionnaires, the response rate to this review is 100 percent. The audit firm used a stratifie d sampling approach that resulted in over -sampling of engagements with higher perceived auditor business risk. The initial sample consisted of 307 audit engagements in four primary industries: financial services, health care and the public sector, informat ion communication and entertainment, and consumer and industrial products. As T able 3 1 shows, firms with missing information were deleted from the sample, a procedure that reduced the sample to 294 firms. Moreover, firms in the financial services sector and the health and public sectors were also eliminated due to the different auditing procedures associated with these industries. This elimination process resulted in a final sample of 171 observations. This sample is used for the remainder of the analyses in this study. Table 3 2 describes the characteristics of clients in the final sample. The table shows that clients vary widely in their size, measured by assets or revenue with a mean value of total assets (revenue) of about $1,694 ($2,004) million and a median value of $186 ($209) million. Moreover, size percentiles also portray a high variation in client size, where the 10th percentile is $31 million in assets whereas the 90th percentile reaches up to $2,616 million. In terms of asset composition, the data shows that on average inventory and receivables make up about 29 percent of total assets. However, the value of inventory and receivables ranges from the 10th percentile value of 6 percent to its 90th percentile value of 58 percent. Closer examinatio n reveals that

PAGE 52

52 inventory seems to exhibit the highest variation, with a mean (medi an) value of $197 ($9) million and $0 and $199 million for the 10th and 90th percentile, respectively. Receivables also exhibit variation but to a lesser degree. Statistics o n capital structure reveal that firms in the sample have different capital structure s The mean (median) value of leverage, measured as the ratio of long -term debt to assets, is about 28 (22) percent, but this varies widely between 0.4 percent and 58 perce nt for the 10th and 90th percentile, respectively. In terms of the operational performance, the data suggests that on average firms in the sample are not performing very well as shown by the mean value of the return on assets equaling negative 2.8 percent. However, the median value of the return on assets, measured as the ratio of net income to total assets, is 1.1 percent indicating that the lower mean value could be driven by a few bad performers. In fact, closer examination shows that firms in the 10th p ercentile have very low performance of about negative 21 percent. Finally, the table shows that many firms in the sample have international operations. The mean value of the percent of foreign assets located internationally equals 9.4 percent, ranging from zero to 35 percent for firms in the 10th and 90th percentile, respectively. Table 3 3 shows the distribution of engagement -specific variables including data on audit hours, audit and non audit fees, auditor -client tenure, and the number of reports issued by the auditor. The data shows that clients paid on average $382 thousand dollars for their audits. The median value of audit fees is $209 thousand, whereas the values for the 10th and 90th percentiles are $69 thousand and $647 thousand, respectively. The data also reveals that the engagement teams provided other nonaudit services to their client s including advisory, merger and acquisition advising, services relating to offerings of equity, tax and other unspecified types of services. The mean value of th e total fees received from non audit services including advisory,

PAGE 53

53 M&A, offer, and other fees is $91 thousand, but the median value is zero, indicating that only a few clients purchase non audit services. Closer examination of the data shows that on average the majority of non audit fees are from the category of other fees followed by offer fees. However, the data shows that of all non audit services, tax fees generate the greatest revenue with the mean (median) value of $134 ($30) thousand. The mean (med ian) value of audit hours expended on engagements is 2,252 (1,346) hours. This can range from 453 to 4,612 hours, each pertaining to the 10th and 90th percentile. The mean (median) value of the number of reports prepared for the client is 5.3 (2) and it ra nges between 1 and 10 repor ts. Finally, the mean (median) value of tenure is 4.73 (0) years indicating that the distribution of auditor client tenure is highly skew ed. Indeed Table 3 4 shows that a great proportion of observations are first year engagemen ts. Table 3 4 also shows the frequency distribution of engagements by whether clients are public or private firms, whether engagements are first year engagements and by industry membership. The data shows that 112 firms, constituting about 66 percent of t he sample, are public. Moreover, as suggested in the previous paragraph, about 51 percent of the firms are new engagements. Finally, 58 firms (34 percent) are in the information, communication and entertainment industry (ICE) while the rest in the consume r and industrial business (CIB). Table 3 5 shows the distribution of client characteristics according to whether clients are publicly traded companies. First, note that the mean value of assets is greater for private firms ($2,211 million) relative to public firms ($1,422 million), indicating that on average the private firms in this sample are l arger than public firms, but the mean difference s are not statistically significant based on the results of t tests Moreover, if comparison is based on the median, the 10th and 90th percentiles of assets, it seems that public firms are larger than private firms.

PAGE 54

54 Similarly, when the comparison is based on the revenue, it shows that public firms are larger, but even here the mean differences are not significant. The data al so shows that in terms of asset composition, the mean (median) value of inventory and receivables as a percent of total assets is 27 (25) percent in public firms, and 35 (32) percent in private firms. Comparison of means also shows that in private firms, inventory and receivables are a significantly larger percent of total assets relative t o public firms. Another aspect in which public and private firms differ is the capital structure. Note that leverage is significantly higher for private firms relative to public firms, wit h mean (median) values of 34 (27 ) percent and 2 5 (19) percent respectively for each group. In terms of performance, it is shown that while the return on assets is smaller for public firms, this difference is not statistically significant. Also, pub lic firms seem to have a significant larger percent of their assets lo cated overseas relative to private firms. More specifically, on average about 12 percent of public firms operations are international, relative to only 5 percent for the private firms. Table 3 6 shows the distribution of engagement -specific variables. Fir st, note that the mean (median) value of audi t fees that public firms pay is $468 (238) thousand, while for private firms these values are $217 (120) thousand respectively for the mean and median. Interestingly, the mean value of audit fees is significant ly higher for public firms relative to private firms even though this difference is insignificant. In terms of the total nonaudi t service fees, the table shows that public firms buy significantly more such services from their auditor. More specifically, p ublic firms pay on average $121 thousand relative to only $34 thousand that private firms pay. Closer examination of nonaudit services reveals that public and private firms buy similar levels of advisory and M&A services, however they differ in the amount of other services and services related to equity off erings. Moreover, the data shows that public firms also

PAGE 55

55 buy a significantly larger amount of tax services (mean of $171 thousand) than their private counterparts (me an of $66 thousand). There are no significant differences across public and private firms when it comes to tenure and the number of reports issued, although it is worth mentionin g that private firms seem to have longer tenures. Finally, audit hours are significantly larger for public firms, re lative to the hours expended for private firms. More specifically, on average the auditor spends about 2,674 hours auditing public firms, and only 1,450 auditing private firms. This difference could explain the audit fee differences noted earlier. Table 3 7 show s the distribution of client characteristics by industry membership. As noted earlier, the sample includes only two major industries: information, communication and entertainment (ICE) industry, and the consumer and industrial business (CIB) First, note that firms in the CIB industry are larger although the difference seems insignificant (when based on assets) or moderately significant (when based on sales). However, there are significant industry differences when it comes to asset composition and le verage. For example firms in the CIB industry are significantly more le vered (mean leverage is 32 percent) than firms in the ICE industry (mean leverage is 20 percent). Moreover, CIB firms have a greater percentage of assets in inventory and receivables t han ICE firms, indicating that ICE firms are more service oriented than CIB firms. There are also significant differences between the two industries in terms of performance. For example, CIB firms are more successful as indicated by the mean value of th e return on assets which equals 0.3 percent, relative to negative 9 percent for the ICE firms. Finally, it is shown that ICE firms are more international than CIB firms as a greater percentage of their assets is located overseas. Table 3 8 shows the distribut ion of engagement -specific variables by industry. The data shows that there are no significant differences between the two industries when it comes to audit

PAGE 56

56 fees, non audit fees, the number of reports issued, tenure, and audit hours. The only significant d ifference is in the amount of tax services whereby ICE firms purchase a significantly larger amount of tax services (mean value is $203 thousand) than CIB firms (mean value of $99 thousand).

PAGE 57

57 Table 3 1. Sample composition Initial sample of audit engagements pertaining to one big international auditing firm 307 Less: Firms with missing information 14 Firms in the financial services and public sector industries 122 Final sample of audit engagements used for hypotheses testing 171

PAGE 58

58 Table 3 2 Client characteristics Variables Mean Standard Deviation P10 Median P90 Assets ($000s) 1,694,110 8,652,888 31,260 185,669 2,615,513 Sales ($000s) 2,004,120 7,910,376 21,000 209,211 2,276,560 Receivables ($000s) 98,019 214,552 3,157 23,846 189,216 Inventory ($000s) 196,679 860,349 0 9,318 199,140 Inv. & rec. (% of assets) 0.295 0.204 0.056 0.267 0.581 Leverage 0.278 0.242 0.004 0.224 0.581 Return on assets 0.028 0.339 0.212 0.011 0.115 Percent of foreign assets 9.4 17.3 0.0 0.9 35.0

PAGE 59

59 Table 3 3. Engagement characteristics Variables Mean Standard Deviation P10 Median P90 Audit fees ($) 381,670 633,012 69,000 209,000 647,000 Advisory fees ($) 22,403 122,510 0 0 6,000 M&A fees ($) 13,924 89,320 0 0 0 Offer fees ($) 25,541 81,752 0 0 66,000 Other fees ($) 29,128 87,729 0 0 80,000 Total non audit fees ($) 90,996 213,891 0 0 269,000 Tax fees ($) 134,602 321,807 0 30,000 311,000 Number of reports 5.3 10 .0 1 .0 2 .0 10 .0 Tenure (years) 4.73 9.45 0 .00 0 .00 15.00 Total audit hours 2,252 3,035 453 1,346 4,612 Number of observations 171

PAGE 60

60 Table 3 4. Frequency distribution of dichotomous variables Variables N Percent (%) Public 112 65% New engagements 87 51% ICE industry 58 34% CIB industry 113 66%

PAGE 61

61 Table 3 5. Distribution of client characteristics by whether clients are publicly traded Public Private Variables Mean P10 Median P90 Mean P10 Median P90 Mean Differences Assets ($000s) 1,421,856 35,585 226,411 3,061,500 2,210,931 20,825 133,888 1,085,166 789 075 Sales ($000s) 2,129,915 21,816 214,762 2,454,000 1,765,325 19,114 177,429 2,276,560 364 590 Receivables ($000s) 116,717 3,592 28,298 272,992 62,526 2,496 20,801 90,781 54, 191 Inventory ($000s) 241,201 0 7,056 291,484 112,163 0 20,385 179,586 129 038 Inv. & rec. (% of assets) 0.266 0.036 0.246 0.521 0.349 0.083 0.318 0.667 0.083*** Leverage 0.246 0.002 0.189 0.543 0.337 0.004 0.27 0.653 0.091** Return on assets 0.037 0.333 0.001 0.109 0.011 0.092 0.019 0.149 0.026 Percent of foreign assets 11.8 0 .0 1.7 44.0 4.8 0 .0 0 .0 15.0 7.0 **

PAGE 62

62 Table 3 6. Distribution of engagement characteristics by whether clients are publicly traded Public Private Variables Mean P10 Median P90 Mean P10 Median P90 Mean Differences Audit fees ($) 468,176 96,000 238,000 838,000 217,455 40,980 120,000 403,000 250,721*** Advisory fees ($) 22,079 0 0 16,000 23,017 0 0 0 938 M&A fees ($) 18,348 0 0 15,000 5,525 0 0 0 12,823 Offer fees ($) 38,549 0 0 107,000 847 0 0 0 37,702*** Other fees ($) 42,266 0 0 110,000 4,189 0 0 12,000 38,077*** Total non audit fees ($) 121,243 0 22,500 335,000 33,579 0 0 61,200 87,664*** Tax fees ($) 170,714 0 41,500 414,000 66,051 0 13,000 200,000 104,663** Number of reports 4.93 1 .00 2 .00 7 .00 6.01 1 .00 2 .00 20.00 1.08 Tenure (years) 4.36 0 .00 0 .00 13 .00 5.42 0 .00 1 .00 15 .00 1.06 Total audit hours 2,674 708 1,639 4,915 1,450 300 796 2,929 1,224*** Number of firms 112 59

PAGE 63

63 Table 3 7. Client characteristics by industry Information, Communication & Entertainment Consumer and Industrial Business Variables Mean P10 Median P90 Mean P10 Median P90 Mean Differences Assets ($000s) 793,136 24,047 160,347 2,438,261 2,156,557 34,263 212,252 2,725,495 1,363,421 Sales ($000s) 651,742 16,259 131,963 1,407,000 2,698,261 35,523 247,778 3,056,036 2,046,519* Receivables ($000s) 92,419 2,695 22,217 175,102 100,894 3,592 24,124 304,693 8,475 Inventory ($000s) 30,121 0 1,598 44,787 282,170 58 20,391 364,000 252,049* Inv. & rec. (% of assets) 0.226 0.031 0.181 0.492 0.330 0.062 0.315 0.594 0.104*** Leverage 0.201 0 .000 0.107 0.548 0.317 0.008 0.27 0 0.635 0.116*** Return on assets 0.089 0.433 0.015 0.076 0.003 0.164 0.021 0.12 0 0.092* Percent of foreign assets 12.9 0 .0 4 .0 46 .0 7.6 0 .0 0 .0 28 .0 5.3*

PAGE 64

64 Table 3 8. Engagement characteristics by industry Information, Communication & Entertainment Consumer and Industrial Business Variables Mean P10 Median P90 Mean P10 Median P90 Mean Differences Audit fees ($) 374,362 90,000 225,000 659,000 385,421 66,000 194,000 630,000 11,059 Advisory fees ($) 28,188 0 0 6,000 19,433 0 0 15,000 8,755 M&A fees ($) 3,517 0 0 0 19,265 0 0 0 15,748 Offer fees ($) 47,931 0 0 197,000 14,049 0 0 21,000 33,882*** Other fees ($) 32,189 0 0 110,000 27,557 0 0 78,000 4,632 Total non audit fees ($) 111,826 0 5,450 335,000 80,305 0 0 200,000 31,521 Tax fees ($) 203,543 0 28,000 585,000 99,217 0 30,000 250,000 104,326** Number of reports 4.53 1 .00 2 .00 6 .00 5.69 1 .00 2 .00 15 .00 1.16 Tenure (years) 3.29 0 .00 0 .00 10 .00 5.47 0 .00 0 .00 15 .00 2.18 Total audit hours 2,245 524 1,496 3,634 2,256 453 1,277 4,766 11 Number of firms 58 113

PAGE 65

65 CHAPTER 4 THE IMPACT OF THE AUDIT CREDENCE ATTRIBUTE ON PROFITABILITY AND PRODUCTION OF AUDITS IN A REPEATED SETTING Introduction Audits exhibit credence attributes in that the auditor is both, the expert the party with the audit expertise who advises the client about the level of assurance needed, as well as the provider (seller) of the service.1 Moreover, the level of assurance su pplied is only known to the auditor As a result, the auditor possesses an information advantage about the audit, vis -vis the client, and may use it strategically to extract profits More specifically, credence goods theory predicts that the auditor may (1) underaudit (provide less audit than the client needs), (2) overaudit (provide more audit than the cl ient needs), or (3) overcharge (provide less audit than promised and charged for). However, the auditor must trade -off any benefits arising from strateg ic acts against potential losses, including litigation and reputation costs. Therefore, the auditors course of action depends on the interaction between the potential profits and costs associated with each strategy. In this paper, I argue that the auditor gains private information about a clients assurance needs through his repeated interaction with the client, and may be able to use this information to maximize his own profit (Beck and Wu 2006, Fong 2005). However, a clients own learning and switching c osts may mitigate the auditors ability to extract profits over time (Lewis and Yildirim 2002, Fong 2005, Dulleck and Kerschbamer 2006). As a result of the interaction between the auditors private information and a clients switching costs over time, the engagements profits are predicted to initially increase with time, but eventually decrease. In addition, the credence goods theory predicts that consumers own knowledge about the product 1 For example, the audit fee and the total hours quoted to the client represent the auditors advice with respect to the audit needed in order to achieve the desired assurance.

PAGE 66

66 they purchase may mitigate a sellers ability to extract rents (Dul leck and Kerschbamer 2006, Emons 1997). It is therefore predicted that engagement profits are smaller in clients that are relatively more knowledgeable and better prepared for the external audit. Using proprietary audit engagement data from one of the inte rnational auditing firms, I find that engagement profits increase with tenure but then eventually decrease. In addition, I also find evidence that a clients own knowledge about the audit process is associated with lower profits. Both results are consisten t with the credence view of the audit markets in which a seller possesses an information advantage about the product and may use this information strategically to extract rents, but the ability to extract profits can be mitigated by consumers knowledge. T he result s remain robust under different sensitivity checks. Further, preliminary assessment of the effort and fee variables reveals that the audit production is most consistent with the overcharging strategy. More specifically, the higher audit profitability is achieved by exerting less effort than expected and by charging a higher fee than expected, where the expected effort and fees are calculated as the predicted values of the regressions that model audit effort and audit fee as a funct ion of client cha racteristics. Auditor -Client Relationship, Auditor Knowledge and Clients Cost of Switching It was argued earlier that the credence attribute of auditing creates incentives for the auditor to extract rents from an audit engagement. The incentives arise bec ause the auditor knows more about the production of the audit than the client. In many markets, including the market for auditing services, suppliers of the service experience learning by doing. That is, suppliers learn through experience how to provide the service at a lower cost (Lewis and Yildirim 2002, Beck and Wu 2006). Experience with a client allows the auditor to accumulate client -specific knowledge (i.e., the auditor learns), which, given the credence nature of audits, may facilitate an auditors a bility to adjust the audit production process and earn rents. This is of

PAGE 67

67 importance because adjustments to the audit production could ultimately affect audit profitability, quality, and efficiency. In this paper, I specifically examine the effect of the au ditors repeat interaction with the client (or tenure) on the profitability and the production of audit s. At first contact, the auditor has little information about the underlying condition of the client and does not know with certainty the audit effort ne eded to achieve a specific assurance level. At this stage, the auditor faces a significant level of litigation risk (St. Pierre and Anderson 1984, Stice 1991) and experiences a steep learning curve by expending considerable resources to accumulate knowledg e about the clients industry, strategy, and operations (Johnson et al. 2002, DeAngelo 1981a, Ghosh and Moon 2005, Carey and Simnett 2006, Beck and Wu 2006, Davis et al. 2007). As a result of high audit effort, the auditor begins to accumulate client -speci fic knowledge. Client -specific knowledge is important because it facilitates an auditors ability to extract rents when the opportunity arises (Fong 2005). Due to experience with the client, the auditors private information and thus audit profits are expe cted to ri se over time. From the clients perspective, switching costs (i.e., competition) and the clients own knowledge play an important role in mitigating an auditors strategic behavior. During the first few years of auditor tenure, the client incurs extra indirect costs in the form of human resources and time spent acquainting the auditor with the clients accounting information system (Butterworth and Houghton 1995). The client is willing to expend these costs with the expectation that they diminish over time. In addition, switching auditors too frequently is costly because the investors may view the switch as bad news and react accordingly (Hackenbrack and Hogan 2002, Kim et al. 2008). Therefore, switching costs are high during the early years of the

PAGE 68

68 auditor -client relationship. High switching costs keep competition away and hence may exacerbate rent extraction. On the other hand, switching costs decline with tenure. More specifically, switching auditors after a long tenure may be beneficial to the cl ient. First, research argues that over time audit programs become less rigorous, less innovative, and increasingly inefficient (McLaren 1958, Hoyle 1978, Shockley 1981, Arrunada and Paz -Ares 1997). A switch to a new auditor may therefore improve audit effi ciency and quality. Second, prior research suggests that audit fees are frequently low -balled by new auditors thus adding another benefit to switching (DeAngelo 1981a). Third, regulators have argued that long incumbencies may negatively impact auditor inde pendence (AICPA 1978, GAO 2003). Because a long tenure may convey a sense of impaired independence, a switch can provide relief to investors that the auditor is not economically tied to the client. Consistent with this argument, Boone et al. (2008) find th at while the equity risk premium initially declines with tenure, it eventually rises again as tenure increases, indicating that investors price long auditor tenures. Taken together, this evidence suggests that an auditor switch late in the incumbency is le ss costly to the client, and may even be beneficial. Therefore, long incumbencies are characterized by low switching costs potentially indicating that the threat of a switch is now credible, and could potentially serve as a mechanism that mitigates rent ex traction. In addition to declining switching costs, a clients own knowledge about the audit can mitigate rent extraction (Lewis and Yildirim 2002, Dulleck and Kerschbamer 2006).2 T he client 2 More specifically, Lewis and Yildirim 2002 assume a dynamic setting in which not only sellers, but also buyers can experience their own learning. Here, suppliers of the service gain information rents due to private information produced as a result of the repeated interaction with the buyer. The buyer may be able to disrupt these gains by switching to a different supplier. However, switching is costly to the buyer because by switching providers the buyer delays the efficiency gains that arise from learning. Similarly, Dulleck and Kerschbamer 2006 argue that in a car repair serv ice, clients can mitigate their information disadvantage by asking for the replaced parts, if the mechanic claims to have changed those parts.

PAGE 69

69 may be able to form her own knowledge base about the audit through her own repeated interactions with the auditor. The knowledge can increase the clients ability to detect rent extraction and respond (for example, by switching to a different auditor). The switch, or the threat of a switch, d isrupts information rents to the auditor. The interaction between the auditors private information, a clients own knowledge, and switching costs conditional on the auditor tenure is depicted in F igure 4 1.3 It can be seen from this figure that at the out set of the auditor -client relationship, switching costs are high whereas the auditors private information is low. High switching costs facilitate, while the lack of clientspecific knowledge and high litigation risk, mitigate the ability to extract rents. Therefore, rent extraction may be difficult to implement during the early years of tenure. Continuity on an audit increases the auditors client -specific knowledge. After many repeat engagements the auditor possesses private information about the audit pr oduction process which may facilitate rent extraction. However, low switching costs associated with long incumbencies and the clients own knowledge about the audit act as deterrents. Therefore, with long tenures, it becomes costly to maintain high profits The result of the interaction between the auditors private information and a clients switching costs is that the auditors costs associated with the strategic actions are high during the initial and later years of the incumbency, but are low at some po int in between.4 Thus, there exists a window of opportunity throughout the auditor -client tenure during which the opportunity to extract rents is maximized (i.e., the information advantage to the auditor net of the effect of any mitigating mechanisms reach es a maximum). Therefore, it is expected that engagement 3 It can be seen from the figure that switching costs subsume the clients own knowledge. This is so because as the clients kn owledge increases over time, her ability to detect audit production adjustments and the associated profits also increases. Thus, high client knowledge translates into lower switching costs. 4 Note that Figure 4 1 depicts the costs associated with strategic actions. Cost minimization corresponds to profit maximization.

PAGE 70

70 profits initially increase with time, but then decrease. The first hypothesis belo w formalizes this statement: H1: Engagement rents are more likely to increase initially, but then decrease as au ditor client tenure increases. Prior research on credence goods argues that sellers ability to behave strategically is diminished when faced with consumers that are educated about the services they purchase (Dulleck and Kerschbamer 2006). For example, Emons (1997) reports a Swiss study which showed that the average population had 33 percent more surgical interventions than physicians and their families. This result indicates that relatively more educated and knowledgeable consumers (i.e., physicians) are less likely to experience unnecessary medical procedures. A similar line of reasoning can be used in the audit markets. Some firms may have more information about the extent of audit they need to purchase, perhaps because their CEOs or CFOs may have prior audi ting experiences. This information affords a stronger bargaining power to the clients and may reduce engagement profits. The hypothesis below formalizes this idea: H2: Clients information about the audit process is associated with lower engagement profits Probing the Audit Production Process The first hypothesis predicts that during the incumbency, the auditor is presented with a window of opportunity characterized by high private information and high switching costs both of which facilitate rent extraction. The hypothesis makes no prediction with respect to the audit strategy that contributes to the higher profitability. Nevertheless, it is interesting to examine whether the production process is more consistent with any of the three strateg ic act ions predicted by the credence goods theory: underauditing, overauditing, or overcharging. Because these strategies are defined in terms of two parameters, effort and fee, the analysis focuses on

PAGE 71

71 both factors. In terms of audit effort, prior research studi es the effect of client characteristics on two audit production elements: the total labor expended on the engagement and the labor allocation among different labor ranks (OKeefe et al. 1994, Bell et al. 2001, Hackenbrack and Knechel 1997, Bell et al. 2008). In order to gain insights about the specific strategy chosen, this section focuses on the following: (1) the extent of deviation between the actual effort/fee and the expected effort/fee5 and (2) the allocation of the total effort among different ranks of labor. To facilitate the discussion refer to T able 4 1. First, the auditor may strategize by providing a level of effort that deviates from the clients need. Recall that if the auditor underaudits then the abnormal effort and fee are expected to be neg ative because the auditor bids a lower fee and provides less effort than expected (QL-Q*<0, FL-F*<0), where QL represents the supplied level of service in case of an underaudit This expectation is denoted by a negative sign in the last two rows in T able 4 1 in the column underauditing. In overauditing, the abnormal fee and the abnormal effort are expected to be positive because the auditor bids a higher fee and provides more effort than expected (QHQ*>0, FH-F*>0), where QH represents the supplied level of service This expectation is represented by a positive sign in the last two rows of T able 4 1 in the column overauditing. Finally, in overcharging, the abnormal effort is expected to be negative or zero and the abnormal fee is expected to be positive because the auditor provides less effort than expected but charges a higher fee than expected (QL-Q*<0, F* -FL >0 or Q*-Q*=0, FHF*>0), where F* is the audit fee associated with Q*. Second, I examine effort allocation among various staff. A self -interested auditor may allocate more labor than expected towards a certain rank and away from another rank depending on his objective. For example, higher profits can be achieved by structuring the audit in such a way that more labor than expected is allocated toward s the cheaper sources of labor. In order to 5 Recall that the expected audit effort equals the needed effort.

PAGE 72

72 gain insights with respect to labor allocation, I compare expected versus actual effort expended by each labor rank. Research Design Engagement Profitability The first hypothesis argues that the engagements pro fitability is a function of the auditor client tenure. During the initial years, the auditor expends a large amount of resources to gain information about the client. The auditor can use the information to earn information rents. However, rents diminish as eventually a clients own learning and switching costs mitigate the incentives to extract rents during the later years of the audit. The second hypothesis argues that a clients own information about the audit is associated with lower profits. The empiric al specifications (4 1) and (4 2) below are used to test the hypotheses: PROFi = a0 + a1(MEDTENi) + a2(IAUDi)+ a3 (LPREPi) + a4(IAUD*MEDTENi) + a5(LPREP*MEDTENi)+ a6(LASSETSi) + a7(NREPTSi) + a8(FRGNi) + a9(COMPLXi) + a10(ROAi) + a11(LEVi) + a12(PUBLICi) + a13(HRELYi) + a14(MASi) + a15(ABRi) + a16(COVIOLi) + a17(ICE) + ei (4 1 ) AFHi = b0 + b1(MEDTENi) + b2(IAUDi)+b3 (LPREPi) +b4(IAUD*MEDTENi) + b5(LPREP*MEDTENi)+ b6(LASSETSi) + b7(NREPTSi) + b8(FRGNi) + b9(COMPLXi) + b10 (ROAi) + b11(LEVi) + b12(PUBLICi) + b13(HRELYi) + b14(MASi) + b15(ABRi) + b16(COVIOLi) + b17(P TIMEi) + b18(M TIMEi) + b19(ITIMEi) + b20(ICE) + ui (4 2 ) Where: MEDTEN = is the variable used to test H1 and equals 1 for engagements with tenure 37 years (or 3 8 years) ; 0 otherwise IAUD = is the first variable used to test H2 and equals 1 for clients that have an internal audit department; 0 otherwise

PAGE 73

73 LPREP = is the second variable used to test H2 and equals 1 if the client is very well -prepared for the external audit as assessed by the auditor; zero otherwise6 PROF = actual audit fees divided by standard fee s AFH = actual audit fees divided by audit hours LASSETS = the natural log of the total assets NREPTS = the total number of audit reports prepared for the client FRGN = the percent of a client s total assets located outside the US COMPLX = clients operational complexity as assessed by the auditor. This variable ranges from 1=very simple to 5=very complex ROA= the ratio of net income to total assets LEV = the ratio of long term debt to total assets PUBLIC = equals 1 for engagements of public companies; 0 otherwise HRELY = equals 1 if the auditor places a high reliance on clients internal controls; 0 o therwise MAS = the ratio of fees related to management advisory services to audit fees ABR = equals 1 if the auditor business risk as assessed by the auditor is high; 0 otherwise7 COVIOL = equals 1 for clients that violated debt covenants during the fiscal year subject to this audit; 0 otherwise ICE = equals 1 for engagements in the information, communications and entertainment industry; 0 otherwise PTIME = hours expended by the partner divided by total audit hours MTIME = hours expended by the manager divided by total audit hours ITIME = hours expended by the in -charge divided by total hours8 6 In addition, equations (41) and (42) include the interaction terms IAUD*MEDTEN and LPREP*MEDTEN to test H2. 7 In additional model specifications, I replace ABR with other risk measures including auditor assessed risk of material misstatement, litigation risk and whether the client is bound by debt covenants. Results remain robust. 8 All variables are defined in Appendix C.

PAGE 74

74 The I ndependent V ariables of I nterest: Tenure and C lients I nformation The first hypothesis suggests that engagement profits are highest during a specific stretch of time over the entire client audi tor tenure. Prior research provides no insight into the effect of tenure on the production process and the engagements profitability. As a result, it is not known at which point during the incumbency rents are maximized. Research on the effect of tenure o n audit quality can offer some insights relating to the timing aspect of the auditors private information and a clients switching costs. This line of research suggests that it takes the auditor about two years to get familiar with the clients environmen t, operations, and accounting information system (McLaren 1958, Carey and Simnett 2006, Davis et al. 2007). As a resul t, SHRTEN equals 1 for those engagements with tenures equal or less than two years and zero otherwise. Long tenures are characterized by a familiarity threat which results in the audit becoming less rigorous and in efficient. Moreover, regulators have been concerned that long auditor tenure can erode an auditors independence and thus negatively impact audit quality. As a way to mitigate this risk, regulators have historically required that the auditor be rotated periodically. For example, until recently, the AICPA mandated audit partner rotation after 7 years.9 Following regulators recommendation and prior research (see Carey and Simnett 200 6), LNGTEN equals 1 for those engagements with tenures greater than 7 years (LNGTEN) and zero otherwise. This cut off however is somewhat arbitrary and therefore I also provide the results by using a cut -off equal to 8 years. By default, the period of time during which engagement profits are maximized is identified by the variable MEDTEN which equals 1 for those engagements with tenures between 3 7 (or 3 8 ) years and zero otherwise. MEDTEN is the primary variable of interest in testing the first hypothesis. 9 The Sarbanes Oxley Act requires that audit partner be rotated after 5 years.

PAGE 75

75 H2 suggests that a clients own information about the audit process mitigates rent extraction. To test this hypothesis, I include two variables that capture the clients knowledge : whether a client has an internal audit department (IAUD) and whether the c lient is well -prepared for the audit (LPREP). The first variable, IAUD, indicates that the client has a designated group of staff that is responsible for internal auditing processes and therefore is well educated about the audit process. The internal audit department interacts frequently with the auditor and might be able to learn through this interaction. Moreover, internal auditors may serve as monitors of the external audit. The second variable, LPREP, indicates a clients preparedness about the external audit (as assessed by the external auditor) which relates to issues like how well -organized are the clients records, significant changes in key personnel, preparation for the external audit, timely voluntary disclosure of audit relevant information and s o on. A client that is well -prepared may know more about the audit process and may be in a better position to monitor the external auditor. In addition to testing for the main effect, IAUD and LPREP are interacted with MEDTEN to check whether the effect of a clients knowledge is relatively stronger during the time when incentives to extract rents are maximized. The D ependent V ariable: A udit P rofitability Two measures are used to capture audit profitability. The first measure (PROF) is the ratio of the actual audit fee obtained from the client to the standard audit fee. The standard audit fee represents the desired fee that the auditor wants to collect for a particular engagement. PROF is a reasonable measure of audit profitability because it measures th e deviation of the actual fee from the desired fee or the extent of discount that the auditor offers the client (Simunic and Stein 1996, Dopuch et al. 2003). If the auditor offers a low discount rate, the ratio of actual to standard audit fee is high (actu al fee is closer to the standard fee) and by definition audit pro fitability (PROF) is also high.

PAGE 76

76 The second measure of audit profitability (AFH) is audit fees obtained divided by the number of total hours expended on an engagement (or the hourly rate). Variation in AFH primarily reflects risk differences between engagements and differences in the proportion of total audit hours expended by the various ranks of labor (i.e., partners, managers, in -charge, and staff). For example, a higher hourly rate resul ts if a greater percentage of total effort is incurred by the partner because partners charge more per hour than other labor ranks. Thus, in order to test the effect of tenure, equation (4 2) controls for the labor mix (Bell et al. 2001, Bedard and Johnsto ne 2004).10 Control V ariables Prior research suggests that audit profitability varies with client size, complexity, risk, provision of non audit services, and industry membership (see Francis 1984, OKeefe et al. 1994, Simunic and Stein 1996, Bell et al. 2001, Dopuch et al. 2003, Bedard and Johnstone 2004, Hackenbrack and Hogan 2005, Schelleman and Knechel 2007, and Bell et al. 2008). Equations (4 1) and (4 2) include the natural log of total assets (LASSETS) to control for greater audit effort associated w ith large clients. The models also control for factors related to client complexity including the number of reports issued by the auditor (NREPTS), the percent of total assets located outside the US (FRGN), and the complexity of operations as assessed by t he auditor (COMPLX). The models control for variables related to client specific risk including the return on assets (ROA), the ratio of longterm debt over assets (LEV), whether the client is a public company (PUBLIC), auditor business risk as assessed by the auditor (ABR), and whether the client violates debt covenants (COVIOL). Further, the models control for the variation in the internal control quality by including a dummy variable that indicates high reliance on internal 10 Dulleck, Kerschbamer and Sutter (2009) also use profitability to measure the strategic behavior of service providers in a credence setting.

PAGE 77

77 controls (HRELY). This variabl e captures the quality of a clients internal controls as assessed by the auditor. Prior research suggests that provision of non audit services may result in knowledge spillovers that affect total audit effort expended on an engagement (Simunic 1984). In o rder to control for the potential effect of non audit services on the production of the audit, the models include the ratio of management advisory fees to total audit fees (MAS). Finally, ICE controls for membership in the information, communication and entertainment industry. Prior research suggests that the hourly audit fee (AFH) is a function of the allocation of total labor among different ranks of labor (OKeefe et al. 1994, Bell et al. 2001, Bedard and Johnstone 2004, Bell et al. 2008). To control for the labor mix, equation (4 -2) includes the proportions of total effort expended by partners (PTIME), managers (MTIME), and in-charge (ITIME ). Probing the Audit Production Process Equations (4 3), (4 4) and (4 5) below are used to test the audit productio n adjustments that result in higher profits for mid tenure audits: LHRSi = c0 + c1(LASSETSi) + c2(FRGNi) + c3(COMPLXi) + c4(IAUDi) + c5(NREPTSi) + c6(LEVi) + c7(PUBLICi) + c8(FIRSTYEARi) + c9(ABRi) + c10(ROMMi) + c11(MRELYi) + c12(HRELYi) + c13(MASi) + c14(R TAXi) + c15(COVBOUNDii (4 3 ) LAFi = d0 + d1(LASSETSi) + d2(FRGNi) + d3(COMPLXi) + d4(IAUDi) + d5(NREPTSi) + d6(LEVi) + d7(PUBLICi) + d8(FIRSTYEARi) + d9(ABRi) + d10(ROMMi) + d11(MRELYi) + d12(HRELYi) + d13(MASi) + d14(R TAXi) + d15(COVBOUNDi) + wi (4 4 ) LAFi = m0 + m1(LASSETSi) + m2(FRGNi) + m3(COMPLXi) + m4(IAUDi) + m5(NREPTSi) + m6(LEVi) + m7(PUBLICi) + m8(FIRSTYEARi) + m9(ABRi) + m10(ROMMi) + m11(MRELYi) + m12(HRELYi) + m13(MASi) + m14( R TAXi) + m15(COVBOUNDi) + m16(HPARTNERi) + m17(HMANAGERi) + m18(HINCHARGEi) + m19(HSTAFFi) + zi (4 5 )

PAGE 78

78 Where: LASSETS, FRGN, COMPLX, IAUD, NREPTS, LEV, PUBLIC, ABR, HRELY, MAS were previously defined and: LHRS = the natural log of total hours (or partner, manager, in -charge and staff hours) FIRSTYEAR = equals 1 for first year engagements ; 0 otherwise ROMM = equals 1 if the risk of material misstatement as assessed by the auditor is high; 0 otherwise MRELY = equals 1 if the auditor places a moderate level of reliance on the clients internal control ; 0 otherwise RTAX = the ratio of tax fees over total audit fees COVBOUND = equals 1 i f the client is bound by significant restrictive debt covenants ; 0 otherwise HPARTNER = the natural log of partner hours HMANAGER = the natural log of manager hours HINC HARGE = the natural log of in charge hours HSTAFF= the natural log of staff hours LAF = the natural log of audit fees Equation (4 3) regresses audit hours, while equations (4 4) and (4 5) regress audit fees on various client characteristics including the natural log of total assets (LASSETS), the percent of total assets located outside the US (FRGN), a clients operationa l complexity (COMPLX), whether the client has an internal audit department (IAUD), the number of reports issued by the auditor (NREPTS), the ratio of long -term debt to total assets (LEV), whether the client is a public company (PUBLIC), whether the client is a first -year client (FIRSTYEAR), auditor business risk (ABR), the risk of material misstatement (ROMM), whether auditor places moderate (MRELY) or high reliance (HRELY) on a clients internal controls, the ratio of

PAGE 79

79 management consulting fees to total fees (MAS), the ratio of tax services fees to total fees (RTAX), and whether the client is bound by covenants (COVBOUND). In addition, equation (4 5) also controls for the natural log of audit hours expended by partners (HPARTNER), managers (HMANAGER), in -ch arge (HINCHARGE) and staff (HSTAFF) (see Simunic 1980, OKeefe et al. 1994, Bell et al. 2001, Whisenant et al. 2003, Bedard and Johnstone 2004, Hay et al. 2006, and Bell et al. 2008). The dependent variable in equation (4 3), LHRS, is the natural log of to tal hours expended. The predicted value from the regression is used as a proxy for the expected audit effort given the client characteristics. The difference between the actual and the expected effort measures the unexpected effort, denoted by UEFF. A posi tive UEFF means that the auditor exerts more effort than expected, making the audit highly inefficient (see Knechel et al. 2009). A negative UEFF indicates that the auditor exerts less effort than expected.11 If the auditor accumulates private information o ver time, this information is reflected in the efficiency term UEFF. For example, increased learning should be associated with a re duction in the value of UEFF (which represents efficiency gains). Cost savings that result from efficiency gains could be pas sed onto the client in the form of lower fees or may be appropriated by the auditor instead. Equations (4 4) and (4 5) help us gain insight into this issue. The dependent variable in equation (4 4) is the natural log of audit fees (LAF). The predicted valu e from this equation is the expected audit fee given client characteristics. The difference between the actual and the expected fee forms the unexpected fee. Two measures of the unexpected fee are generated from equation (4 4). First, RUFEE is the ratio of the unexpected fees to total assets and second, UFEE is the un-scaled unexpected fees. 11 In later regressions, UEFF for each labor rank is calculated by replacing LHRS in equation (4 3) with the natural log of hours expended by partners, managers, incharge, and staff

PAGE 80

80 Following the same procedure as in equation (4 4), the predicted value of equation (45) forms the expected audit fee. The unexpected audit fee derived from equation (4 5) and denoted by TUFEE, is the difference between the actual and the expected fee. My interest lies in the distribution of unexpected effort (UEFF) and fees ( RUFEE, UFEE and TUFEE) across the different tenures. The outcomes are then compared to the patte rns deve loped earlier and displayed in T able 4 1. Empirical Results Descriptive Statistics Table 4 2 presents the descriptive statistics of engagement related variables. The mean (median) value of hourl y audit fees (AFH) is $168 ($149)/hour, whereas the mean (median) value of engagement p rofitability (PROF) is 0.62 (0.56) percent suggesting that on average audit firms discount their fees by 38 percent. The distribution of audit hours among the different ranks of labor suggests that most labor is allocated to lower ranks. For example, partners (PTIME) and ma nagers (MTIME) work on average 6 and 17 percent of the total engagement hours, respectively, whereas in -charge ( ITIME) and staff (STIME) work 33 and 37 percent, respectively. Statistics on other non audi t services show that advisory services (MAS) make up about 4 percent of audit fees, total nonaudit services ( RNAS), which include advisory and other services make up about 28 percent of audit fees, whereas fees generated from tax services (RTAX) make up a bout 37 percent. Table 4 3 shows the frequency distribution of dichotomous variables. It can be seen that 107 engagements, composing 63 percent of the sample, have short tenures (SHRTEN), 26 engagements, or 15 percent, have medium tenures (MEDTEN), and the remaining engagements

PAGE 81

81 have long tenures (LNGTEN).12 The majority of the engagements are short -tenured. This is due to the sample selection criteria that tend to draw more heavily from high risk engagements which are usually the first year engagements.13 T ab le 4 3 also shows that about 42 firms (25 percent) have internal audit departments (IAUD) while 75 firms (44 percent) are well -prepared for the audit process (LPREP) In addition, about 30 firms (18 percent) seem to have violated their debt covenant (COVI OL), 33 firms are considered to have a high auditor business r isk (ABR), and 88 firms (51 percent ) have a hig h risk of material misstatement The auditors assessment of internal control indicates that about 11 percent (18 firms) of the sample has strong c ontrols as suggested by the high reliance variable (HRELY) Finally, about 79 firms purchase nonaudit services (NAS) and about 120 firms purchase tax services (TAX) Table 4 4 shows the correlation matrix of the variables of interest related to equations (4 1) and (4 2). Close examination of the correlation coefficients and the variance inflation factors yields no multicollinearity concerns.14 Table s 4 5 and 4 6 show the distribution of various client characteristics across the short, medium, and long tenur es. In Table 4 5 the cut off for the medium tenures is 7 years, whereas in Table 4 6 this cut -off is 8 years. Note that the mean differences show that clients have similar characteristics across the three tenures. The only significant differences seem to b e those between mid -tenure and short tenure clients. For example, short -tenured clients have a significantly larger international presence (FRGN), purchase less NAS ( RNAS), incur more audit hours, and perform worse (ROA) than mid -tenure firms. 12 This distribution is based on MEDTEN cut off of 7 years. Using a cut off of 8 years the number of clients in the medium tenure is 30. 13 The paper provides robustness tests that deal with this specific issue. 14 All variance inflation factors are less than 2. Linear dependency is present when one or more factors exceed 10.0 (Neter, Wasserman and Kutner 1985)

PAGE 82

82 Tests of Hyp otheses Main results Table s 4 7 and 4 8 show the mean and median engagement profitability, respectively, across the short, medium and long tenures. In each table the results are shown using a MEDTEN cut off equal to both 7 and 8 years. Table 4 7 shows that the mean value of audit profitability measured by both PROF and AFH is highest for medium tenures. Specifically, PROF is 68 percent (69 percent for MEDTEN cut off of 8 years) for medium tenures, relative to about 61 percent for short and long tenures, res pectively. Moreover, paired t tests show that the mean differences between median and short tenures and those between long and medium tenures are significant at 5 percent. Similarly, the mean value of the hourly rate (AFH) is about $180 per hour for medium tenures, compared to $163 and $173 for short and long tenures, respectively. However, paired t tests show that these differences are, for the most part, not significant at conventional levels. Table 4 8 shows that the median profitability (PROF) is 0.73 f or medium tenures vs. 54 percent and 56 percent for short and long tenures, respectively (when MEDTEN cut -off is 8 years) Mann Whitney U test shows that the median differences between medium and long tenures are significant at 1 percent, whereas the median differences between long and medium tenures are, for the most part, not significant. Results of the hourly fee rate (AFH) also show that engagement profits for medium tenures are $178, compared to $137 and $164 for short and long tenures, respectively. H owever, only the difference between medium and short tenures is significant based on the MannWhitney U test of median differences. Overall, these univariate statistics are consistent with the prediction in the first hypothesis and suggest that audit profi tability reaches a maximum and then drops as tenure increases.

PAGE 83

83 Table 4 9 shows the mean value of engagement profitability depending on whether the client has an internal audit department (IAUD) and whether the client is well -prepared for the audit (LPREP). Further, Table 4 10 show s the mean value of engagement profitability acr oss IAUD, LPREP and MEDTEN. The results in Table 4 9 suggest that engagement profitability is similar across clients with and those without inter nal audits departments, but it is significantly smaller in clients that are well prepared for the audit process The latter result is consistent with the prediction in the second hypothesis. Results in Table 4 10 also provide univariate support for the second hypothesis. Note that for mid tenure clients (MEDTEN =1), audit profits are smallest in the presence of int ernal audit departments and where preparation for the external audit is highest. Tak en together, the results in T able s 4 9 and 4 10 provide support for the hypothesis that clients knowledge of the audit process limits engagement profits. Table s 4 11 and 4 12 display the results of equation (4 1) using MEDTEN cut-offs equal to 7 and 8 years, respectively. Columns labeled (1) through (3) represent different variations of equation (4 1). The interpretation of the results is based on the full model presented i n column (3). First, note that the model is well -specified with an adjusted R2 of 21 percent. Many of the control variables do not seem to significantly affect audit profitability. This result is consistent with prior research (see Simunic and Stein 1996, Dopuch et al. 2003). For example, while the coefficient on the LASSETS is positive suggesting that larger firms are more profitable to the auditor, this effect is statistically insignificant. ROA is positive and statistically significant (p value = 0.03) s uggesting that engagement profits are highest for firms that generate positive returns. HRELY is negative and statistically significant (p -value = 0.006) suggesting that when the auditor places high reliance on the internal controls he realizes a lower pro fit margin. Of most interest is the sign and significance of MEDTEN which indicates the profitability of

PAGE 84

84 medium -tenured engagements. The coefficient on this variable is positive and significant at the 1 percent level (p -value = 0.01) indicating that audit profitability is greatest for medium tenures relative to the short or the long tenures. This result is consistent with the prediction in H1. The coefficient on LPREP is negative and significant at the 10 percent level (pvalue=0.08), indicating that engag ement profits are smallest in clients that are well prepared for the external audit. IAUD by itself is not significant, however, its interaction with MEDTEN is negative and significant at the 1 percent level (p -value =0.005) suggesting that the presence of an internal audit department limits rent extraction when rent extraction matters the most, that is in mid tenure engagements. These results continue to hold when using a MEDTEN cut -off equal to 8 years as presented in Table 4 12, columns (1) through (3). Overall, the results in T able s 4 11 and 4 12 present evidence consistent with the hypotheses. Table s 4 13 and 4 14 display the results of equation (4 2) using MEDTEN cut -offs equal to 7 and 8 years, respectively. Columns (1) through (3) represent different variations of equation (4 2). The interpretation of the results is based on the full model presented in column (3). The model here is also well -specified with an adjusted R2 of 26 percent. Many of the control variables do not seem to exhibit a strong sign ificance. This result is consistent with prior research (see Bell et al. 2001, Bedard and Johnstone 2004). However, client size and the proportion of hours expended by the in -charge are positively associated with the hourly fee rate. Moreover, provision of non audit services (NAS) is negatively associated with engagement profitability. The variable of interest MEDTEN is positive and significant (p -value=0.009) indicating that audit profitability is highest for medium tenures, relative to short and long tenures. Interestingly, the coefficient on IAUD is negative and significant (p -value=0.07) suggesting that the presence of internal audit departments is associated with lower profit margins

PAGE 85

85 to the auditor consistent with the second hypothesis. Moreover, the coefficient on the interaction term IAUD*MEDTEN is also negative and significant (p value=0.04) suggesting that having an internal audit department has a strong effect in limiting engagement profits especially during a time when rent extraction seems to be h igh (i.e., for mid tenure audits). This result is consistent with H2. The coefficients on LPREP and LPREP*MEDTEN ar e not statistically significant, however. The results in Tables 4 11 through 4 14 suggest that audit profitability is highest during years 3 7(8) of the incumbency. This result is consistent with the first hypothesis which argues that the auditor initially expends resources to gather private information which makes audit production very costly and less profitable. As the auditor accumulates pri vate information, he may use that information to his advantage and generate informational rents. However, eventually audit profitability drops and this may be due to several factors including the clients own learning and declining switching costs. The res ults also provide support for the second hypothesis which argues that a clients own information and preparation for the audit process can limit rents. Overall, the results are consistent with the credence perspective in that the extent of an auditors inf ormation advantage is positively related to his ability to engage in profit maximizing behavior and that this can be limited by a clients own knowledge. The next sections explore in more detail how the production of audit contributes to the profits over t ime. Robustness tests This section discusses several robustness tests. First, to adjust for the effect of the overrepresentation of short tenure engagements on potentially biasing the results, observations are re -weighted using a pre -specified weighing sch eme (see Cameron and Trivedi 2005, pp 817821). More specifically, the sampling probability of short -tenured firms is calculated by utilizing the distribution of the total observations audited by a Big Four auditor in the Audit Analytics

PAGE 86

86 database. Based on this analysis, the total number of observations in Audit Analytics is 17,642, of which, 8,365 are firms with tenures less than or equal to two years. The sampling distribution of my sample of short -tenures is thus 0.013 (calculated as the ratio of short t enured observations in my sample, over the short -tenured observations in the total population of short tenured firms in the Audit Analytics database). Based on the sampling weight calculated above, weighted regressions were performed, where the weight of t he short -tenured observations was re adjusted by the value of the probability above. The results of weighted regressions are consistent with the prediction of the first hypothesis. More specifically, in OLS regressions, the coefficients on MEDTEN are positive with p-values equ al to 0.03 and 0.01, for equation (4 -1 ) and (4 2 ), respectively. Second, I use bootstrapping to create more confidence in the results, especially considering the small sample size (Brownstone and Valletta 2001, Greene 2003, pp.924-925) The bootstrapping technique runs each OLS regression multiple times and uses the variability in the slope coefficients as an estimate of their standard error. Bootstrapping was performed using 50 and 1000 repetitions. Results remain robust. That is, the coefficient on MEDTEN is positive and significant (p -value=0.03 for both models when repetitions equal 50; pvalue=0.04 for both models when repetitions equal 1000).15 Third, MEDTEN is dropped and replaced by SHRTEN and LNGTEN which control for short and long tenures, respectively. The coefficients on these two indicator variables measure the differential profitability of short and long tenures relative to medium tenures Based on the 15 In addition, the bootstrapping technique was adjusted for the overrepresentation of short tenured firms, by simply requiring that an equal number of observations is selected from the short tenured group (02 years) and the remaining group (3+ years). In this case, the number of observations is lower because the stratified bootstrapping technique requires that the sample selected in each stratum be equal to the smallest number of observations present in each stratum. Results show that the coefficient on MEDTEN is positive, but slightly weaker (p value=0.07), however this could be due to the smaller sample size.

PAGE 87

87 first hypothesis, these coefficients are expected to be negative. In un tabulated results, the coefficient on SHRTEN is negative and significant in both equations (4 1) and (4 2), with pvalues of 0.05 and 0.01, respectively. The coefficient on LNGTEN is also negative in both equations (4 1) and (4 2) and significant at p-values of 0.09 and 0.08, respectively, using twotailed tests. These results suggest that audit profitability initially rises, but then drops as tenure increases. Fourth, in other untabulated tests, MEDTEN is re -defined by changing the second cut off first to 6 years and then again to 9 years. The coefficient on MEDTEN using a cut -off equal to 6 years is positive but not significant for PROF, and significant for AFH (p value = 0.02, one tailed). Similarly, the coefficient on MEDTEN using a cut off equal to 9 years is positive and significant for PROF (p value = 0.032, one -tailed) and for AF H (p -value =0.001, one tailed). Fifth, logit regressions were performed after the depen dent variables PROF and AFH were transformed to dichotomous variables that equal 1 for observations with profitability greater than the median and zero otherwise. The results of the logit regression show support for both hypotheses. More specifically, the results based on PROF show that the coefficient on MEDTEN is significant and positive at 5 percent, while the coefficient on LPREP is negative and significant at the 1 percent level.16 The result based on AFH show that the coefficient on MEDTEN is positive and significant at the 5 percent level, while the coefficient on IAUD is negative and significant at the 5 percent level.17 The logit regressions were also performed setting the MEDTEN cut -off equal to 8 years. The results are quantitatively similar. Sixth, diagnostic tests indicate that about 8 observations are outliers and therefore equations (4 1) and (4 2) are repeated after removing these outliers. The results are shown in 16 Similar to OLS results, the coefficient on IAUD is not significant. 17 Similar to the OLS results, the coefficient on LPREP is not significant.

PAGE 88

88 columns (4) through (6) in T ables 4 11 through 4 14. Note that results using the reduced sample are quantitatively similar to those using the full sample. Overall, the robustness tests provide more confidence to the main results. Probing the Audit Production Process This section explores in more detail the structure of the production process that contributes to the high profitability observed for medium tenures. In order to examine the production process and the associated fees, the output from regressions (4 3), (4 4), and (4 5) is presented in columns (1), (6) and (7) in T able 4 15. All the models are well -specified with coefficients and overall model fit consistent with prior literature (see Bell et al. 2001 and Bell et al. 2008). The outputs from these models are then used to calculate the unexpected effort and fee measures (UEFF, R UFEE, UFEE, and TUFEE). That is, the unexpected effort (UEFF) is the residual from regression (4 3), RUFEE and UFEE are scaled and raw residuals from regression (4 4), respectively, and TUFEE is the residual from regression (4 5). Table 4 1 6 displays the u nexpected effort (UEFF) and unexpected fees (RUFEE, UFEE, and TUFEE) across auditor tenures using MEDTEN cut -off of 7 and 8 years Based on the MEDTEN cut -off of 7 years, note that the unexpected effort is positive for short tenures, indicating that on ave rage, auditors expend considerable resources during the first few years to accumulate client -specific knowledge. However, as tenure increases from short to medium, UEFF decreases from 0.002 to 0.052. The drop in UEFF is consistent with learning such that auditors become more efficient over time and thus may exert less effort on an engagement. Second, two out of the three measures of unexpected fees (UFEE and TUFEE) are negative for short tenures, consistent with the low -balling hypothesis (DeAngelo 1981a). More interestingly, the unexpected fees (all three) rise and become positive as tenure increases from short to medium. This result suggests that, while audit efficiency (UEFF) increases with tenure,

PAGE 89

89 indicating cost reductions, the efficiency gain is not a ssociated with a fee reduction of the same rate, thus resulting in higher than expected audit fees. The less than expected effort and the higher than expected fees for medium tenures contribute to the higher profitability observed at this time. The pattern of negative unexpected effort and positive unexpected fees is most consistent with the overcharging strategy.18 T he total unexpected effort can be further broken down into unexpected effort for each labor rank to see which rank exhibits the highest level of learning. In order to gain additional insights about rank -specific efficiency gains, the dependent variable in regression (4 3) is replaced one at a time by the natural log of audit hours expended by partners, managers, incharge, and staff. The result s of these regressions are presented in columns (2) through (5), in Table 4 15. The output shows that all models are well -specified and consistent with prior literature (see Bell et al. 2008). The unexpected effort (UEFF) is measured as before, by subtract ing the predicted value of hours from the actual hours, for each rank. Table s 4 1 7 and 4 18 present the audit production characteristics by rank of labor. Table 4 17 presents the distribution of UEFF across time for each rank and the ratio of rank -specific hours to total hours for partners (PTIME), managers (MTIME), in -charge (ITIME), and staff (STIME) using MEDTEN cut -off equal to 7 years, whereas Table 4 18 presents the same information using a cut -off equal to 8 years. First, note from Table 4 17, that a s tenure increases from short to medium, the UEFF changes from positive to negative for partners, managers, and in -charge, but follows a reverse trend for staff. This may indicate that partners, managers, and in charge become more efficient, whereas staff becomes less efficient over time. In other words, the higher ranks of labor seem to be exerting less effort than expected, whereas the lower ranks 18 Recall from T able 41 that overcharging is the only strategy associated with a negative abnormal effort and a positive abnormal fee.

PAGE 90

90 (staff) seem to be exerting more effort than expected as tenure increases. Also, note that the labor mix chan ges as tenure increases from short to medium, with all ranks experiencing a decline in their share of labor with the exception of staff who exhibits a boom in their share. Taken together, the results show that labor allocation changes over time such that m ore labor than expected shifts towards staff, while less labor than expected is allocated to higher ranks. The substitution of the more expensive labor with the cheaper labor can partially explain the efficiency gains observed for medium tenures as well as the higher profits since staff represents the cheapest labor resource. Collectively, t he results in T ables 4 16 through 4 18 indicate that the high profitability achieved for medium tenures is primarily due to: 1) auditor appropriated rents from cost redu ctions associated with learning and 2) re allocation of effort towards the cheapest labor rank, staff, in turn making their effort highly inefficient, while causing partners, managers, and in -cha rge to work less than expected. It is of interest to observe the trend in the level of unexpected effort and unexpected fees when tenure increases from medium to long. Note from table 4 16 that UEF F increases from 0.052 to 0.037 as tenure changes from medium to long. This pattern indicates that the efficiency gains observed during medium tenures disappear with a further increase in tenure. This result is consistent with the argument that long incumbencies may be characterized by inefficient audit programs. It is interesting to note that the unexpected fees also decr ease for long tenures. The collective changes in the unexpected effort and fees contribute to the lower audit profitability observed for long tenures and are consistent with the hypothesis. The results are even more interesting when considering the changes in rank -specific efficiency and labor allocation as tenure increases from medium to long. Note from T able 4 1 7 that UEFF related to partners, managers, and in -charge increases indicating an efficiency loss,

PAGE 91

91 whereas UEFF related to staff experiences a drop indicating increased efficiency for this rank. Moreover, the proportion of effort incurred by partners and in -charge increases during long tenures relative to medium tenures, whereas the proportion of staff decreases. This trend shows reversals in both efficiency and labor allocation in the long relative to medium tenures. If rent extraction is less likely in long tenures, as the first hypothesis suggests, then labor allocation and efficiency trends for long tenures can be interpreted as the non-strategi c audit production benchmark. The deviation observed during the medium tenures is consistent with overcharging. This seems to indicate that the auditor strategically assigns less labor to higher ranks and more labor to lower ranks in medium tenures, a move that contributed to a higher profit margin. However, this strategy may not be sustained in de finitely, thus the auditor is forced to revert to a more sustainable audit program during the later years of the incumbency. Overall, the joint allocation of unexpected effort and unexpected fees across time can explain the profitability pattern stated in the first hypothesis. This allocation is mostly consis tent with the overcharging strategy developed from the theory of credence goods. Conclusion and D iscussion Audits exhibit credence attributes and as such they are characterized by an information asymmetry whereby the auditor knows more about the audit production than the client and may use this information opportunistically. More specifically, credence goods th eory predicts that the auditor may underaudit, overaudit, or overcharge. Prior research on audit production assumes a constant level of assurance at the audit firm level (Simunic 1980, OKeefe et al. 1994, Dopuch et al. 2003, Knechel et al. 2009). This fra mework does not allow for intra -firm audit quality differentiation. However, it is important to relax the assumption of constant assurance because the information asymmetry present in the market for audits can affect audit quality and efficiency which even tually have an impact on the supplied level of assurance The discussion and the

PAGE 92

92 findings of this paper imply that the credence attribute of audits has important implications about the audit quality at a finer, engagement -specific level. More specifically, the theory of credence goods argues that strategic behavior is selective such that each engagement presents a unique opportunity to the auditor to extract profits. The greater the information advantage accruing to the auditor, the greater the incentive to act opportunistically. The auditor gains client -specific private information through his repeated interaction with the client. As a result, it is expected that profit -maximizing behavior increase with the auditor tenure. However, a clients own learning a nd switching costs may mitigate the auditors incentives. As a result of the interaction between the auditors private information and the clients switching costs over time, the engagement profits are predicted to initially increase with time, but then de crease In addition, credence goods theory argues that sellers profits are reduced when they face educated consumers. Using a similar reasoning I argue that engagement profits are lower in clients that are well informed about the audit process. Using proprietary data from engagements completed from one of the big international auditing firms, I find that engagement profits are highest for mid -tenure clients relative to short and long -tenured ones. Further, I also find that auditing clients with internal au dit departments and those that are well prepared for the external audit is less profitable. These results are consistent with the hypotheses generated from the credence goods theory. Further, assessment of the effort and fee variables reveals that audit pr oduction is mos t consistent with overcharging. There are several limitations associated with this study. First, the decision to switch auditors and therefore the tenure variable is endogenous. This study does not resolve this issue, however, it provides a considerable amount of sensitivity tests that provide support for the first hypothesis. Second, the data used to test the hypotheses are from one audit firm. If the audit

PAGE 93

93 processes are unique to this firm then the results might not replicate to other audit firms. Nevertheless, the results are consistent with theory and therefore the issue of one data source is somewhat mitigated. Third, the first hypothesis argues that engagement profitability changes over time however the empirical analyses use one cross -sectional sample. Ideally, the study would require the use of time series data to test how profits change each year. However, in the absence of this data, we can still gather interesting results using one cross -sectional, but as done in this study, in orde r to get the effect of time, data is grouped into cohorts to get an average effect and reduce noise. Fourth, the data pertain to audits surrounding a period of heightened attention on the audit profession and therefore the results may not generalize to other time periods. Again, this argument may be countered on the basis that the hypotheses are grounded on theory.

PAGE 94

94 Figure 4 1 Switching costs and auditors private information over time

PAGE 95

95 Table 4 1. Abnormal effort and fee under each strategy Underauditing Overauditing Overcharging Client needs Q* Q* Q* Client should get (Expected) Q* Q* Q* Client should pay (Expected) F* F* F* Client gets (Actual) Q L Q H Q L or Q* Client pays (Actual) FL FH F* or FH Abnormal effort (Actual Expected) QLQ*<0 QHQ*>0 QLQ*<0 or Q* -Q*=0 Abnormal fee (Actual Expected) F L F*<0 F H F*>0 F* F L >0 or F H F*>0 Abnormal effort pattern + or 0 Abnormal fee pattern + +

PAGE 96

96 Table 4 2. Descriptive statistics of continuous variables Variables Mean Median Standard Deviation P10 P90 AFH $168 $149 $101 $100 $235 PROF 0.62 0.56 0.23 0.41 0.89 C OMPLX 2.61 3.00 0.88 2.00 4.00 Actual hours 2,252 1,34 6 3,035 453 4,612 Partner Hours 144 82 207 21 268 Manager hours 364 229 490 67 639 In Charge Hours 616 420 691 170 1,095 Staff Hours 942 498 1 560 54 1,784 PTIME 0.06 0.0 6 0.03 0.0 3 0.11 MTIME 0.17 0. 16 0.0 6 0. 09 0.24 ITIME 0.33 0. 32 0.14 0.17 0.51 STIME 0.37 0.39 0.16 0.05 0.54 MAS 0.04 0.00 0 .19 0.00 0.02 R NAS 0.28 0.00 0. 76 0.00 0.60 R TAX 0.37 0. 19 0. 57 0. 00 0.92 Number of observations 171 All variables are defined in Appendix C.

PAGE 97

97 Table 4 3 Frequency distribution of engagement variables Variables N Percent (%) MEDTEN 26 15 SHRTEN 107 63 LNGTEN 38 22 IAUD 42 25 LPREP 75 44 COVIOL 30 18 COVBOUND 102 60 ABR 33 19 ROMM 88 51 HRELY 18 11 MRELY 121 71 INTAUD 42 25 NAS 79 46 TAX 120 70 PUBLIC 112 65 ICE 58 34 All variables are defined in Appendix C.

PAGE 98

98 Table 4 4. Correlations related to variables in equations (4 1) and (4 2) PROF AFH MEDTEN IAUD LPREP LASSETS NREPTS FRGN COMPLX LEV PUBLIC HRELY MAS ABR COVIOL PTIME MTIME ITIME PROF 1.00 AFH 0.29 1.00 MEDTEN 0.13 0.08 1.00 IAUD 0.02 0.03 0.12 1.00 LPREP 0.18 0.01 0.08 0.04 1.00 LASSETS 0.08 0.19 0.12 0.60 0.04 1.00 NREPTS 0.20 0.06 0.00 0.17 0.01 0.24 1.00 FRGN 0.07 0.06 0.19 0.17 0.09 0.25 0.12 1.00 COMPLX 0.03 0.15 0.07 0.39 0.17 0.50 0.17 0.37 1.00 LEV 0.04 0.02 0.00 0.09 0.00 0.25 0.13 0.09 0.13 1.00 PUBLIC 0.11 0.08 0.01 0.10 0.05 0.16 0.05 0.19 0.08 0.17 1.00 HRELY 0.15 0.07 0.01 0.11 0.23 0.12 0.08 0.10 0.08 0.02 0.07 1.00 MAS 0.09 0.10 0.15 0.01 0.16 0.12 0.01 0.08 0.04 0.06 0.21 0.11 1.00 ABR 0.21 0.07 0.05 0.10 0.22 0.10 0.00 0.17 0.17 0.02 0.23 0.17 0.01 1.00 COVIOL 0.12 0.17 0.09 0.09 0.12 0.07 0.02 0.14 0.13 0.05 0.04 0.04 0.02 0.16 1.00 PTIME 0.09 0.27 0.03 0.08 0.02 0.01 0.03 0.11 0.09 0.10 0.29 0.17 0.07 0.07 0.02 1.00 MTIME 0.07 0.13 0.06 0.09 0.08 0.05 0.07 0.05 0.07 0.04 0.12 0.09 0.12 0.01 0.08 0.45 1.00 ITIME 0.03 0.11 0.01 0.23 0.03 0.22 0.07 0.19 0.15 0.03 0.26 0.05 0.03 0.20 0.08 0.04 0.02 1.00 All variables are defined in Appendix C.

PAGE 99

99 Table 4 5. Audit engagement characteristics by tenure if MEDTEN cut -off is 7 years Short Medium Long Difference Difference (0 2 years) (3 7 years) (8+ years) (Medium Short) (Long -Medium) N 107 26 38 ASSETS ($000s) 1,943,371 469,298 1,830,272 1,474,073 1,360,974 LEV 0.29 0.27 0.25 0.02 0.02 COMPLX 2.67 2.46 2.53 0.21 0.07 FRGN 0.12 0.02 0.08 0.1 0 *** 0.06 Audit Fees ($) 431,345 217,091 354,403 214,254 137,312 NAS Fees($) 85,704 119,961 86,079 34,257 33,882 Tax Fees ($) 135,157 70,923 176,612 64,234 105,689 RNAS 0.20 0.60 0.26 0.4 0 ** 0.34 Audit hours 2,549 1,338 2,041 1,211* 703 ROA 0.06 0.09 0.02 0.14 0.11 NREPTS 4.7 3.9 7.9 0.8 4 .0 Mean differences are based on t tests. All variables are defined in Appendix C.

PAGE 100

100 Table 4 6. Audit engagement characteristics by tenure if MEDTEN cut -off is 8 years Short Medium Long Difference Difference (0 2 years) (3 8 years) (9 + years) (Medium Short) (Long -Medium) N 107 30 34 ASSETS ($000s) 1,943,371 495,400 1,967,355 1,447,971 1,471,955 LEV 0.29 0.28 0.24 0.01 0.04 COMPLX 2.67 2.47 2.53 0.2 0 0.06 FRGN 0.12 0.02 0.08 0.1 0 *** 0.06* Audit Fees ($) 431,345 224,645 363,891 206,700 139,246 NAS Fees($) 85,704 110,633 90,323 24,929 20,310 Tax Fees 9$) 135,157 76,733 183,919 58,424 107,186 RNAS 0.20 0.53 0.28 0.33** 0.25 Audit hours 2,549 1,322 2,138 1,227 816 ROA 0.06 0.07 0.02 0.13 0.09 NREPTS 4.7 5.3 7.1 0.6 1.8 Mean differences are based on t tests. All variables are defined in Appendix C.

PAGE 101

101 Table 4 7. Analysis of mean engagement profitability by tenure MEDTEN cut off is 7 years MEDTEN cut off is 8 years Tenure (years) PROF AFH Tenure (years) PROF AFH Short (0 2 ) 0.61 163 Short (0 2 ) 0.61 164 Medium (3 7) 0.68 180 Medium (3 8) 0.69 186 Long (8+ ) 0.62 173 Long (9+) 0.61 167 Difference (Medium Short) 0.07* 17 Difference (Medium Short) 0.08** 16 Difference (Long Medium) 0.06 7 Difference (Long Medium) 0.08** 19* Mean differences are based on t tests. All variables are defined in Appendix C.

PAGE 102

102 Table 4 8. Analysis of median engagement profitability by tenure MEDTEN cut off is 7 years MEDTEN cut off is 8 years Tenure (years) PROF AFH Tenure (years) PROF AFH Short (0 2) 0.54 137 Short (0 2) 0.54 137 Medium (3 7) 0.73 172 Medium (3 8) 0.73 178 Long (8+ ) 0.59 177 Long (9+) 0.56 164 Difference (Medium Short) 0.19** 35*** Difference (Medium Short) 0.19*** 35*** Difference (Long Medium) 0.14 5 Difference (Long Medium) 0.17* 14 Median differences are based on Mann -Whitney U tests. All variables are defined in Appendix C.

PAGE 103

103 Table 4 9. Engagement profitability by INTAUD and LPREP Average profitability by INTAUD Average profitability by LPREP PROF AFH PROF AFH INTAUD Yes 0.62 173 LPREP Yes 0.57 169 No 0.62 168 No 0.66 168 Difference 0.00 5.00 Difference 0.09*** 1.00 Mean differences are based on t tests. All variables are defined in Appendix C.

PAGE 104

104 Table 4 10. Engagement profitability by MEDTEN, INTAUD, and LPREP PROF AFH PROF AFH INTAUD INTAUD LPREP LPREP No Yes No Yes No Yes No Yes MEDTEN 0 0.60 0.64 $162 177 MEDTEN 0 0.65 0.56 $164 170 1 0.71 0.45 187 128 1 0.72 0.64 194 168 All variables are defined in Appendix C.

PAGE 105

105 Table 4 11. Regression results of equation (4 1 ) when MEDTEN cut off is 7 years Full sample Reduced sample Variable (1) (2) (3) (4) (5) (6) MEDTEN 0.080** 0.084** 0.136*** 0.091*** 0.094*** 0.142*** IAUD --0.013 0.005 --0.040 0.028 LPREP --0.054** 0.054* --0.048* 0.045 IAUD*MEDTEN ----0.292*** ----0.239*** LPREP*MEDTEN ----0.035 ----0.052 LASSETS 0.012 0.016 0.015 0.014 0.021 0.020 NREPTS 0.005 0.005 0.005 0.005 0.005 0.005 FRGN 0.025 0.025 0.018 0.051 0.056 0.037 COMPLX 0.030 0.035 0.038 0.041 0.032 ROA 0.087** 0.087** 0.085** 0.085** 0.083** 0.076* LEV 0.029 0.023 0.010 0.046 0.035 0.022 PUBLIC 0.021 0.021 0.013 0.007 0.007 0.001 HRELY 0.103*** 0.091** 0.106*** 0.117*** 0.105*** 0.109*** MAS 0.031 0.025 0.008 0.013 0.009 0.006 ABR 0.084 0.075 0.073 0.053 0.044 0.045 ICE 0.058 0.049 0.041 0.069* 0.059 0.056 COVIOL 0.082 0.080 0.085 0.047 0.048 0.046 Constant 0.45*** 0.452*** 0.044*** 0.459*** 0.412*** 0.402*** N 171 171 171 163 163 163 Adjusted R 2 0.18 0.19 0.21 0.17 0.18 0.19 This table presents the regression results for equation (4 1) when MEDTEN cut off is 7 years Results for the full sample are presented in columns (1) through (3). Results for the reduced sample (after removing outliers) are presented in columns (4) through (6). Tests of the hypothesized variables are onetailed. ** *, ** and indicate coefficient significance at p-values less than 0.01, 0.05, and 0.10, respectively. All variables are defined in Appendix C.

PAGE 106

106 Table 4 12. Regression results of equation (4 1 ) when MEDTEN cut off is 8 years Full sample Reduced sample Variable (1) (2) (3) (4) (5) (6) MEDTEN 0.0809** 0.083** 0.113** 0.093*** 0.094*** 0.124*** IAUD --0.009 0.010 --0.035 0.022 LPREP --0.053** 0.057* --0.047* 0.047 IAUD*MEDTEN ----0.275*** ----0.219*** LPREP*MEDTEN ----0.004 ----0.027 LASSETS 0.012 0.015 0.014 0.014 0.020 0.019 NREPTS 0.005 0.005 0.005 0.004 0.004 0.004 FRGN 0.032 0.032 0.026 0.060 0.063 0.045 COMPLX 0.029 0.035 0.027 0.037 0.041 0.033 ROA 0.087** 0.088** 0.087** 0.084** 0.083** 0.079** LEV 0.029 0.024 0.014 0.046 0.037 0.026 PUBLIC 0.218 0.022 0.016 0.007 0.007 0.002 HRELY 0.100*** 0.089** 0.101** 0.113*** 0.102*** 0.105*** MAS 0.031 0.026 0.009 0.013 0.009 0.006 ABR 0.083 0.074 0.073 0.051 0.043 0.044 ICE 0.059 0.051 0.042 0.071* 0.061 0.058 COVIOL 0.081 0.078 0.082 0.045 0.045 0.042 Constant 0.455*** 0.455*** 0.451*** 0.454*** 0.416*** 0.409*** N 171 171 171 163 163 163 Adjusted R 2 0.18 0.19 0.21 0.17 0.18 0.19 This table presents the regression results for equation (4 1) when MEDTEN cut off is 8 years. Results for the full sample are presented in columns (1) through (3). Results for the reduced sample (after removing outliers) are presented in columns (4) through (6). Tes ts of the hypothesized variables are onetailed. ***, ** and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. All variables are defined in Appendix C

PAGE 107

107 Table 4 13. Regression results of e quation (4 2 ) when MEDTEN cut off is 7 years Full sample Reduced sample Variable (1) (2) (3) (4) (5) (6) MEDTEN 42.163*** 39.362*** 76.181*** 32.731*** 31.386*** 49.281*** IAUD --28.590** 20.248* --22.990** 19.552** LPREP --9.995 14.201 --1.114 1.115 IAUD*MEDTEN ----123.717** ----69.008*** LPREP*MEDTEN ----42.078 ----22.482 LASSETS 12.114*** 15.618*** 15.118*** 7.300** 10.590*** 10.395*** NREPTS 0.239 0.255 0.308 0.235 0.261 0.255 FRGN 15.145 17.629 19.031 19.722 20.887 15.596 COMPLX 13.093 15.354 20.153 12.089 11.696 8.685 ROA 1.871 2.599 6.584 17.203 16.533 13.575 LEV 23.768 28.415 34.590 8.103 3.286 0.926 PUBLIC 1.137 2.510 2.268 2.980 3.614 1.482 HRELY 46.047 43.431 37.860 23.017* 22.233 22.972* MAS 23.031** 23.538** 15.239* 12.792* 13.354* 12.156* ABR 3.265 6.041 5.285 19.591 19.909 20.067 ICE 13.052 9.226 6.867 16.842* 13.772 13.315 COVIOL 51.731 56.692 59.505* 12.201 15.016 14.667 PPROP 954.858 939.891 989.982 228.564 211.191 240.282 MPROP 80.218 61.451 54.171 104.159 98.916 90.035 IPROP 135.054*** 129.014*** 139.981*** 144.393*** 138.474*** 142.585*** Constant 149.249 188.501 201.866 4.298 26.101 32.140 N 171 171 171 163 163 163 Adjusted R 2 0.23 0.24 0.26 0.30 0.31 0.32 This table presents the regression results for equation (4 2 ) when MEDTEN cut off is 7 years Results for the full sample are presented in columns (1) through (3). Results for the reduced sample (after removing outliers) are presented in columns (4) through (6). Tests of the hypothesized variables are one tailed. ***, ** and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. All variables are defined in Appendix C

PAGE 108

108 Table 4 14. Regression results of equation (4 2 ) when MEDTEN cut off is 8 years Full sample Reduced sample Variable (1) (2) (3) (4) (5) (6) MEDTEN 42.343*** 39.192*** 67.289*** 36.395*** 34.513*** 50.776*** IAUD --26.816** 18.705 --21.109** 17.680* LPREP --10.549 14.446 --0.840 2.025 IAUD*MEDTEN ----114.391** ----68.946** LPREP*MEDTEN ----33.559 ----22.909 LASSETS 12.140*** 15.371*** 15.053*** 7.421** 10.417*** 10.350*** NREPTS 0.130 0.153 0.109 0.146 0.174 0.113 FRGN 11.554 14.284 15.052 24.658 25.261 20.218 COMPLX 13.103 15.349 19.477 12.158 11.778 8.987 ROA 1.437 2.090 4.243 16.771 16.271 13.997 LEV 23.871 28.116 34.087 8.077 3.678 0.911 PUBLIC 1.426 2.754 1.987 3.121 3.719 1.210 HRELY 47.235 44.433 39.439 22.029* 21.394 22.204* MAS 22.899** 23.463** 15.657* 12.932** 13.441** 12.187* ABR 2.533 5.454 4.731 18.393 18.844 18.991 ICE 13.556 10.111 8.371 17.211* 14.421 14.415 COVIOL 50.913 55.691 57.384 11.843 14.394 13.422 PPROP 933.047 919.378 956.553 212.529 196.763 213.675 MPROP 82.785 64.365 63.406 107.172 101.998 95.425 IPROP 130.378*** 125.345*** 131.549*** 141.281*** 135.949*** 137.597*** Constant 148.029 185.062 196.097 3.111 24.482 30.004 N 171 171 171 163 163 163 Adjusted R 2 0.23 0.24 0.26 0.31 0.32 0.34 This table presents the regr ession results for equation (4 2 ) when MEDTEN cut o ff is 8 years. Results for the full sample are presented in columns (1) through (3). Results for the reduced sample (after removing outliers) are presented in columns (4) through (6) Tests of the hypothesized variables are one tailed. ***, ** and indicate coef ficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. All variables are defined in Appendix C

PAGE 109

109 Table 4 15 Regression results of equations ( 4 3 ) ( 4 4 ) and ( 4 5 ) Hours Fees (1) (2) (3) (4) (5) (6) (7) Total Partner Manager In charge Staff Audit fees Audit fees LASSETS 0.251*** 0.298*** 0.280*** 0.229*** 0.146** 0.334*** 0.123*** FRGN 0.148 0.200 0.052 0.434* 0.011 0.002 0.122 COMPLX 0.153*** 0.046 0.096 0.159*** 0.202 0.122* 0.021 INTAUD 0.322*** 0.149 0.266** 0.194* 0.874*** 0.132 0.097 NREPTS 0.005 0.009* 0.004 0.005 0.005 0.007* 0.003 LEV 0.287** 0.149 0.308* 0.113 0.562* 0.260* 0.070 PUBLIC 0.477*** 0.718*** 0.562*** 0.301*** 0.603** 0.505*** 0.091 FIRSTYEAR 0.224** 0.220** 0.299*** 0.171* 0.267 0.078 0.117* ABR 0.129* 0.174** 0.175** 0.042 0.150 0.116 0.015 ROMM 0.086 0.071 0.169 0.005 0.020 0.048 0.106 MRELY 0.135 0.101 0.014 0.084 0.528 0.001 0.081 HRELY 0.070 0.386* 0.338* 0.267 0.488 0.248 0.045 MAS 0.089 0.196 0.114 0.245 0.203 0.008 0.117 TAX 0.122* 0.221** 0.139* 0.098 0.034 0.188** 0.085* COVBOUND 0.081 0.080 0.023 0.118 0.263 0.049 0.029 HPARTNER ------0.154* HMANAGER ------0.282*** HINCHARGE ------0.326*** HSTAFF ------0.078*** Constant 2.757*** 0.440 0.732 2.289*** 1.621*** 7.064*** 6.052*** Adjusted R 2 0.69 0.63 0.63 0.56 0.32 0.56 0.87 N 171 171 171 171 171 171 171 This table presents the regression results of equations (4 3), (4 4), and (4 5). Columns (1) through (5) present the results of equation (4 3) where the dependent variable is the natural log of total audit hours, partner hours, manager hours, in -charge hours, and staff hours, respectively. Column (6) presents the results of equa tion (4 4), whereas column (7) presents the results of equation (4 5). All tests are two tailed. ***, **, and indicate coefficient significance at p values less than 0.01, 0.05, and 0.10, respectively. Variables are defined in Appendix C.

PAGE 110

110 Table 4 16. Audit production characteristics by tenure MEDTEN cut -off is 7 years MEDTEN cut -off is 8 years Tenure (years) UEFF RUFEE UFEE TUFEE Tenure (years) UEFF RUFEE UFEE TUFEE Short (0 2) 0.002 18.407 0.005 0.010 Short (0 2) 0.002 18.407 0.005 0.010 Medium (3 7) 0.052 50.428 0.010 0.063 Medium (3 8) 0.086 37.609 0.001 0.074 Long (8+) 0.037 0.498 0.009 0.013 Long (9+) 0.078 5.936 0.019 0.031 This table presents the residuals of regressions ( 4 3), (4 4) and ( 4 5) by tenure. Variables are defined in Appendix C.

PAGE 111

111 Table 4 17. Audit production characteristics by labor rank when MEDTEN cut off is 7 years The distribution of UEFF by labor rank Allocation of audit hours by labor rank Tenure (years) Partner Manager Incharge Staff P TIME M TIME I TIME S TIME Short (0 2) 0.02 0.02 0.02 0.11 0.07 0.17 0.33 0.37 Medium (3 7) 0.14 0.07 0.13 0.39 0.06 0.15 0.32 0.41 Long (8+) 0.03 0.01 0.05 0.04 0.06 0.15 0.35 0.36 This table presents the distribution of UEFF by labor rank. UEFF by labor rank is calculated as the residuals from regression ( 4 3) by replacing one at a time the dependent variable with hours expended by partners, managers, incharge, and staff. Variables are defined in Appendix C.

PAGE 112

112 Table 4 18. Audit production characteristics by labor rank when MEDTEN cut off is 8 years The distribution of UEFF by labor rank Allocation of audit hours by labor rank Tenure (years) Partner Manager Incharge Staff P TIME M TIME I TIME S TIME Short (0 2) 0.02 0.02 0.02 0.11 0.07 0.17 0.33 0.36 Medium (3 8) 0.15 0.1 0.13 0.24 0.06 0.16 0.34 0.39 Long (9+) 0.06 0.03 0.08 0.13 0.06 0.15 0.34 0.37 This table presents the distribution of UEFF by labor rank. UEFF by labor rank is calculated as the residuals from regression ( 4 3) by replacing one at a time the dependent variable with hours expended by partners, managers, incharge, and staff. Variables are defined in Appendix C.

PAGE 113

113 CHAPTER 5 EXPLAINING AUDIT PRODUCTION INEFFICIENCIES Introduction In the market for auditing services, auditors are jointly experts who adv is e their client s on how much audit to purchase, and sellers who provide the actual service. This dual role of the auditor in the market provides him with an information advantage relative to the client, regarding the necessary amount of audit that a client needs. In this perspective audits exhibit credence attributes (Darby and Karni 1973, Dulleck and Kerschbamer 2006). The theory of credence goods argues that an expert -seller such as an auditor, has incentives to use his informational advantage strategically in order to maximize his profits by advising c lients to purchase more of a service than needed (overauditing) and/or by providing less service than initially recom mended (overcharging) or less service than the client needs (underauditing) (Dulleck and Kerschbamer 2006). This information asymmetry is further exacerbated by its persistence, in that a client who accepts an auditors recommendation may never completely know that s he paid for more of a service than needed, and by the nature of the audit service, such as its unobservability. This paper focuses specifically on the association between engagement specific characteristics and overauditing. O verauditing is defi ned as the act of selling more audit services than needed (i.e., the level of audit that the client needs is simply the amount of service the client would have purchased in the absence of the information asymmetry discussed above). Overauditing may result in an inefficient quantity of audit being exchanged in the market in that the buyer s of the audit service (i.e., clients, audit committees and investors) are compelled to purchase more than what they may consider an optimal level from their perspective. Hence, understanding what drives overauditing is an important research question.

PAGE 114

114 There is limited empirical evidence on the issue of overauditing particularly because researchers and regulators have been primarily concerned with underaudits (i.e., low qual ity audits). However, overauditing became an issue of interest primarily after the implementation of the Sarbanes Oxley Act of 2002 (hereafter SOX) and SOX section 404 requirements. For example, in the speech before the Council of Institutional Investors SEC commissioner Annette Nazareth stated that the commission had heard complaints about auditors over auditing in part because they fear that PCAOB inspectors would otherwise find their audits insufficient. The PCAOB itself has said that such over audit ing is not desirable (Nazareth 2007) The inefficiencies associated with the implementation of PCAOBs auditing standard #2, eventually led to the issuance of standard #5 whose goal was partially to reduce the incentive to overaudit. Similar concerns are expressed in a report by the US Chamb ers of Commerce which states that overauditing exists and has been the cause of serious deterioration in many auditor -client relationships. Further, the report argues that PCAOBs Auditing Standard #2 doesnt provide much guidance as to when enough is enough with respect to the auditing of internal controls (US Chambers of Commerce 2006). The recent controversy surrounding overaudits underlines the important issue at stake: buyers of the audit service may be paying for something that they might not ne ed or want. Even though interest on this issue is heightened during periods of regulatory changes such as the introduction of SOX section 404, the issue nevertheless exists as long as the auditor acts as both the advisor and the supplier of audit s For exa mple, OKeefe et al. (1994) find that the extent of auditor reliance on the clients internal controls does not affect the quantity and mix of auditing labor This finding indicates that audit production could be inefficient in that the auditors might carr y unnecessary tests even in the presence of strong internal controls. In addition, OKeefe et al. 1994 do not find evidence in support of learning

PAGE 115

115 over time further suggesting that audit production is inefficient. The absence of learning is very counterint uitive since one would expect that long auditor tenure would facilitate the audit process through learning. More recently, Doyle et al. (2007) find the surprising result that material weaknesses associated with section 302 of SOX are more strongly associat ed with the quality of accruals than the ones under section 404. They interpret this finding as auditors being more careful regarding the 404 requirement s on internal controls (i.e., overauditing) by applying tougher standards and thus classifying some minor problems as material weaknesses. This may increase the number of material weaknesses that have no implication on the quality of internal controls and hence no effect on accrual quality. Moreover, Ogneva et al. (2007) find that many of the disclosures r elated to material weaknesses do not lead to a higher cost of capital suggesting that firms do not face economic consequences for having disclosed weak controls. Finally, using the audit production data from one of the big auditing firms, Knechel et al. (2009) find that audits are produced inefficiently. Among the inefficient engagement they find that the average level of inefficiency, measured by use of Data Envelopment Analysis (DEA), is 75 percent which means that inputs (costs of staff by rank) can be r educed by 25 percent to achieve the same level of audit assura nce. Despite concerns about the inefficiencies associated with the production of audits, it is important to acknowledge however, that overauditing may lead to some positive externalities such as a higher level of assurance, even though this may not be the outcome that buyers of audits want. Similar to other credence goods, w hen an auditor recommend s how much audit a client should purchase he may use his expertise and knowledge of the audit advantageously by simply overstating the number of hours that are needed to complete an audit (Hubbard 1998, Alger and

PAGE 116

116 Salanie 2006, Bartels et al. 2006, Fong 2005, Dulleck and Kerschbamer 2006) In the market for auditing services overauditing can be achieved in two ways: either ex ante (before the start of the audit), or ex -post (during the audit process). Ex -ante overauditing is achieved when the auditor is able to contract for a larger number of audit hours than the level a client needs, as implied in the fi xed fee Ex -post overauditing occurs when more audit hours than previously agreed upon are sold to the client. For example, the auditor may declare during the audit process that more testing is needed even though this may not be the case. It is assumed in this paper that given an auditors audit expertise and knowledge of his clients, he knows exactly which strategy (exante or ex post) fits a particular client, if any at all. This assumption is built upon Fong (2005) who argues that expert -sellers are able to identify client -specific information that facilitates a sellers strategic behavior. Thus, expert -sellers target their customers selectively and only prescribe unnecessary treatment if based on the information they possess they conclude that the stra tegy is implementable and profitable. In this paper I investigate the audit engagement characteristics that are associated with overauditing. More specifically, I study the effect of competition, the cost of an audit, an auditors information about the cl ient, a clients own knowledge about the audit, and a clients bargaining power on the likelihood of both ex ante and ex post overauditing. Ex ante overauditing is measured as the difference between the planned audit hours and the hours predicted from an OLS regression which regresses actual hours worked on various client -specific characteristics. The predicted hours are used to proxy for the amount of audit that the client actually needs given the clients own specific characteristics (i.e., this is the benchmark audit or the amount the client would have purchased in the absence of the information asymmetry), whereas planned audit hours are used as a proxy for the amount of audit service that

PAGE 117

117 the au ditor recommends to the client. Ex ante overauditing is t hen defined as those instances where planned hours exceed the predicted E x -post overauditing is defined as those engagements with additional charged hours above the initial recommendation. More specifically, ex -post overaudits are identified as the profitable engagements where actual audit hours exceed planned hours. I n this case, actual audit hours are used as a proxy for the final amount of audit sold to the client, and as befo re, planned hours are used as a proxy for the initial recommended amount The requirement that these engagements be profitable is important to distinguish between cases when the additional hours are needed to gather sufficient evidence to render the opinion vs. the strategic case when the additional audit hours are charged to simply increase profits. It is assumed that strategic overaudits are more profitable. The goal of this paper is to study factors that can explain the variances between the needed (predicted), the recommended (planned) and actual audit hours. Using a samp le of 171 audit engagements from one of the big auditing firms I find the following results. First, competition is negatively associated with both ex ante and ex -post overauditing. This result is consistent with the theoretical prediction in DeAngelo (1 9 8 1 a ) who finds that competition leads to low -balling, where low -balling can be viewed as the inability to sell additional audit hours. E x ante overauditing is also less likely to occur in clients that are costly to service where the audit cost is proxied by several risk measures. This result is consistent with the theoretical prediction that expert -sellers are more likely to prescribe unnecessary services to customers whose problems are more expensive to fix (Fong 2005). In the audit setting, risk is usually associated with higher costs (Simunic 1980, Hay et al. 2006) and therefore it is expected that high risk firms experience ex ante overauditing at a higher rate.

PAGE 118

118 Ex ante overauditing is also less likely to occur in engagements where the auditor spend s a gr eat amount of time talking to non accounting personnel and when clients are under pressure to release their earnings numbers. The results also suggest that ex ante overauditing is more likely to occur in clients that are highly prepared for the external au dit where preparation indicates well -organized internal records. Similar to ex ante overauditing, expost overauditing is also more likely to occur in riskier firms. Moreover, results suggest that when the auditor performs interim work on the client he is more likely to engage in ex post overauditing. In contrast to its effect on ex -ante overauditing, the clients level of preparedness is associated with less ex -post overauditing. S ome results also indicate that ex post overaudits are associated with a disp roportionate amount of labor being allocated to the lower ranks of labor. It is interesting to note that observable client-specific characteristics like size, leverage, and complexity are not important. However, there is evidence that public companies seem to experience more ex -post overauditing than private companies Development of Hypotheses Many services inclu ding medical, legal, car repair electrical, financial expertis e, and audits exhibit credence attributes The common feature across this diverse g roup of services is that the seller of the service is also the expert who offers advice as to which service is appropriate to buy and how much of it. Therefore the expert -seller possesses an information advantage relative to the buyer and he may use this information advantageously. Dulleck and Kerschbamer (2006) list three main strategies that an expert -seller (or an auditor) can pursue in this case: overtreatment (overauditing), which amounts to r ecommending more of a service than a client needs, overchar ging, which amounts to providing less service than initially recommended, and undertreatment (underauditing), where less service than needed is provided.

PAGE 119

119 This paper focuses on the strategic action of overauditing, which is defined as the act of recommendin g and selling more audit than the level that the client may need or want. In addition to their expertise about the product they sell, expert -sellers may also be able to identify customer -specific characteristics that can further facilitate the overtreatment strategy. Fong (2005) is one of the earliest papers to point out that when expert -sellers prescribe unnecessary services, they do so selectively. That is to say that expert -sellers target only the consumers that they think would be most willi ng to pay for the exaggerated service. Fong (2005) cites anecdotal evidence that shows that women are more likely to be a target of unnecessary repairs by the car mechanics. In another case, Wade (1999) reports that it is most frequently the young and the old who are targeted by tour companies. In an article in the Money magazine, Allen Wood of the California Bureau of Consumer Affairs states that dishonest mechanics dont rip off every customer -only the ones they think are easy t o fool (Belsky 1996). The se examples suggest that expert -sellers know exactly which consumers to target based on what they know and can learn about these consumers. However, there is a big difference between receiving an unnecessary auto repair or medical treatment and receiving m ore audit hours because the additional audit hours lead to a higher level of assurance, which is a positive outcome of having to pay for more audit than necessary, even though this may not be desirable on the part of the buyers of audits. In this paper, I study engagement -specific characteristics that facilitate the overauditing strategy. T here is limited guidance from the theory of credence goods regarding an expert sellers decision to engage in strategic behavior. Consequently, this investigation can be viewed as exploratory in examining the factors that might affect the extent to which an auditor might profitably over e stimate the necessary amount of audit service for a particular client.

PAGE 120

120 The first factor to be considered is the competition in the market for audit ing services. Wolinsky (1993) demonstrates that experts are less likely to recommend unnecessary services when customers search for second opinions (i.e., when competition is high). Hubbard (1998) investigates the probability that motor vehicles f ail emission inspections in a private firm that also carries out the repairs (i.e., expert -sellers) relative to state inspectors who simply provide the inspection but do not carry out the repairs (i.e., simply sellers) He finds differences across the two settings such that the probability of failure is higher in private firm settings consistent with the notion that when the seller of a service is also the expert there is a tendency to provide more of a service tha n needed which benefits the sellers because it generates a higher demand for their services. However, this study also finds that such overtreatment is less likely in areas with close geographic competitors. In audit research, it is often argued that in highly competitive markets auditors are willin g to lowball their audit fee in order to get new clients (DeAngelo 1981a Hay et al. 2006). Lowballing makes overauditing less likely because overauditing is essentially qu oting higher fees (to cover the additional recommended audit hours). These results suggest that competition mitigates the ability to exaggerate audit hours needed to complete the audit. Hypothesis 1 below formalizes this statement H1: Overauditing is less likely to occur in highly competitive markets In a theoretical model, Fong (2005) fi nds that when experts face heterogeneous consumers who differ based on how costly it is to service them, the experts are more likely to quote unnecessary services to clients that are costlier to service. In the market for auditing services some clients are riskier, and more complex, and therefore maybe more expensive to audit (Simunic 1980, OKeefe et al. 1994). Overauditing is attractive to the auditor especially since risk can be associated with high litigation cost s and therefore the auditor can protect himself

PAGE 121

121 against heightened risk by overstating the necessary audit hours. Following Fongs prediction it is then argued that overauditing is more likely to occur in clients that are costly to audit leading to hypothesis 2 below: H2: Overauditing is more li kely to occur in clients that are costly to audit When experts interact with their customers they are able to discover customer -specific features that facilitate the ability to convince customers to purchase more of a service than is necessary (Fong 2005). For example, experts can figure out who among customers is more willing to pay for additional services. In a similar fashion, an auditors private information about a client can help him identify facts about specific clients that may make them more or les s willing to accept purchasing additional audit hours. Therefore, it is expected that an auditors private information about a client may increase or decrease the likelihood of overauditing depending on the type of information received. Hypothesis 3 below formalizes this statement: H3: Auditors private information about the client affects the likelihood of overauditing Dulleck and Kerschbamer (2006) argue that clients can mitigate overtreatment by educating themselves about their condition and therefore about how much service they need to purchase. More educated consumers have a better understanding of how much service to buy and accept from the expert. For example, Fong (2005) cites evidence that men are less likely to accept car repair recommendations f rom their mechanic presumably because they are better informed about car matters. Using a similar logic, some audit clients may have more knowledge about the level of audit that is necessary and therefore may be less likely to accept an auditors recommend ation and/or more be in a more powerful position to negotiate less hours. This statement is formalized in hypothesis 4 below: H4: Overauditing is less likely to occur in well -informed clients

PAGE 122

122 Prior audit research argues that large clients have more bargain ing power vis a vis the auditor and thus maybe in stronger position to negotiate a lower fee (and thus lower hours) than the one recommended by the auditor (Castarella et al. 2004). If this is the case, then a client s bargaining power is negatively associ ated with the degree of overstating audit hours. The hypothesis below formalizes this statement: H5: Clients bargaining power is negatively associated with overauditing Research D esign Empirical Model To test the association between engagement -specific characteristics and overauditing the following pr obit regressions are estimated. EXAGG = a0 + a1(FIRSTYEARi) + a2(EFFICi) + a3(ROMMi) + a4(SDi) + a5(ROAi) + a6(SUBi) + a7(HRELYi) + a8(NREPTSi) + a9(BUSi) + a10(INTRMi) + a11(OUTSDi) + a12(DAYSEi) + a13(LPREPi) + e (5 1) EXCD = b0 + b1(FIRSTYEARi) + b2(EFFICi) + b3(ROMMi) + b4(SDi) + b5(ROAi) + b6(SUBi) + b7(HRELYi) + b8(COMPLXi) + b9(LEVi) + b10(PUBLICi) + b11(NREPTSi) + b12(BUSi) + b13(INTRMi) + b14(OUTSDi) + b15(DAYSEi) + b16(NASi) + b17(TAXi) + b18(PTIMEi) + b19(MTIMEi) + b20(ITIMEi) + b21(LPREPi) + b22(LASSETSi) + b23(OPINi) + b24(SCOPEi) + u (5 2) Where: EXAGG = equals 1 if planned hours are greater than predicted; zero otherwise EXCD = equals to P_EXCD, A_EXCD, and U_EXCD one at a time and where P_EXCD equals 1 if actual hours are greater than pla nned and engagement profitability (PROF) is greater than the median; A_EXCD equals 1 if actual hours are greater than planned and hourly audit fees (AFH) are greater than the median; and U_EXCD equals 1 if actual hours are greater than planned and au dit fe e residuals (UFEE) are positive FIRSTYEAR = equals 1 f or first -year clients; 0 otherwise EFFIC = efficiency score from the DEA model (see Appendix E )

PAGE 123

123 ROMM = equals 1 if the risk of material misstatement as assessed by the auditor is high; 0 otherwise SD = the number of significant deficiencies discovered by the auditor during the audit ROA = net income divided by total assets SUB = equals 1 if the client is a subsidiary of another organization ; 0 otherwise HRELY = equals 1 if the auditor places a high reliance on the clients internal controls; 0 otherwise COMPLX = clients operational complexity as assessed by the auditor; ranges from 1(very simple) to 5(very complex) LEV = long -term debt divided by total assets PUBLIC = equals 1 if the client is a public company; 0 otherwise NREPTS = total number of reports prepared for the client BUS = percent of time spent on understanding a clients business INTRM = equals 1 if the auditor performed interim work; 0 otherwise OUTSD = equals 1 if percent of time spent with non accounting personnel is high; 0 otherwise DAYSE = the number of calendar days lapsed after the clients fiscal year -end up to the date of the clients issuance of the earnings press release NAS = equals 1 if the auditor provides other non audit services; 0 otherwise TAX = equals 1 if the auditor provides tax services; 0 otherwise PTIME = hours expended by the partner divided by total audit hours MTIME = hours expended by the manager divided by total audit hours ITIME = hours expended by the in -charge divided by total audit hours LPREP = equals 1 if the client is very wellprepared for the external audit as assessed by the auditor; 0 otherwise LASSETS = natural log of total assets OPIN = equals 1 if the client receives an unqu alified opinion; 0 otherwise

PAGE 124

124 SCOPE = equals 1 if there was a scope expansion; 0 otherwise1 The Dependent V ariable: O verauditing The benchmark audit In the previous section it was argued that audit hours needed to complete an audit can be strategically overstated. The overstatement can either be ex ante or ex -post. In order to identify cases of overauditing, the benchmark (non-exaggerated) amount of needed audit must be estimated. To do this I run a regression of actual audit hours as a function of various client specific characteristics. More specifically, based on prior research, actual audit hours are regressed on client size, complexity, risk, auditors reliance on the clients internal cont rols, the existence of a clients internal audit department, provision of other nonaudit services, the number of reports issued by the auditor, and whether the client is a public entity (Hay et al. 2006) (see Appendix B for more details). Using the coeff icients from this regression output, the benchmark audit (PRED) is equal to the predicted hours. Thus, the benchmark audit is the expected demand for audit hours based on client characteristics. Ex ante overauditing In the market for auditing services, one method to achieve overauditing is by prescribing a higher amount of audit service than the level needed. I label this strategy as ex-ante overauditing in that it occurs prior to the start of the audit when the client initially agrees to purchase the amoun t recommended by the auditor. In the car repair business this strategy occurs when a small problem is diagnosed as a much bigger one (Belsky 1996). As argued in this paper, the auditor might strategically deviate from the benchmark audit and overestimate the audit hours needed. If this is the case, the auditor recommends a higher than 1 All variables are defined in Appendix C.

PAGE 125

125 necessary level of audit service. Planned audit hours (HPLAN) are used as a proxy for the recommended level of audit. The planned hours represent the auditors exante assessment of what the auditor thinks is appropriate for the client plus an amount that reflects the auditors strategic action. Ex ante overauditing is then measured as the positive difference between planned (HPLAN) and the benchmark hours (PRED). Exante ove raudits are identified by the variable EXAGG, which equals 1 if planned hours (HPLAN) exceed the benchmark (PRED), and zero otherwise. Ex post overauditing In addition to ex ante overauditing, the auditor can also overstate hours after the start of the aud it. For example, the auditor might convince the client that additional audit hours are needed to co llect sufficient evidence to render the opinion. Belsky (1996) refers to a similar strategy that might occur in the autorepair business as the old bait and switch strategy in that a low -price deal eventually turns into a big ticket boondoggle. In many cases, the additional increase in audit hours maybe for legitimate reason, as for example is the case when the auditor discovers conditions that truly warra nt more testing, however given that the auditor is the expert that decides what is necessary and what is not, in some cases the additional increase in audit hours may just simply reflect the auditors strategic action. It is the latter that I refer to as e x -post overauditing. Ex -post overauditing is defined as overstatement of audit hours after the start of the audit process. However, as mentioned earlier ex-post overauditing could also be non-strategic in that the auditor might have a legitimate reason to increase audit hours beyond the initial level on the grounds that more testing is truly necessary to issue an opinion. Since the interest of this paper is to investigate strategic overaudits, it is important at this point to distinguish between overaudits that are strategic vs. those that are non -strategic. I do so by first imposing the assumption that if overauditing is strategic it is more likely to result in higher profits than if it is

PAGE 126

126 non -strategic. Three different measures of profitability are used to distinguish profitable overaudits. These measures are the engagement profitability (PROF), audit fee per hour (AFH), and the unexpected audit fee (UFEE). PROF is the ratio of actual audit fees divided by the standard fee. The standard fee represents the h ighest estimate of the audit fee that can be received from a client. The ratio is thus an estimate of a discount, the higher the ratio the lower the fee discount and thus the higher the profitability. AFH measures the hourly fee rate so that higher hourly rates proxy for more profitable audits. Finally, the unexpected audit fees (UFEE) are the residuals from the audit fee regressions model that regresses actual fees paid to the auditor on client specific characteristics (see Appendix D for more details), wh ere positive residuals are used to indicate profitable engagements. In addition to being profitable, ex -post overaudits must also meet the condition that final audit hours charged to the client are greater than the initial agreement. As before, planned hou rs (HPLAN) proxy for the initial contracted audit hours, whereas the actual hours worked (HACT) proxy for the final amount of audit sold to the client. I define the following three variables as proxies for ex post overauditing: P_EXCD equals 1 for engageme nts where actual audit hours (HACT) exceed planned hours (HPLAN) and for which PROF is greater than its median value; A_EXCD equals 1 for engagements where actual audit hours exceed planned hours and for which AFH is greater than its median value; and U_EXCD equals 1 for engagements where actual audit hours exceed planned hours and the audit fee residuals (UFEE) are positive. The analysis in the next section studies the association between overaudits and engagement -specific characteristics. Factors A ssociat ed with O verauditing This section discusses the measures that are used to proxy for the following theoretical constructs: competition, the cost of an audit, the auditors private information about the client, a clients own knowledge about the external aud it, and a clients bargaining power.

PAGE 127

127 Competition I use two measures that proxy for competition: first, consistent with prior research which argues that competition in the audit markets is highest in the first year of auditor -client relationship (DeAngelo 1981a ), I include FIRSTYEAR which equals 1 for first year engagements and zero otherwise. Second, consistent with prior research which argues that production efficiency is high in competitive markets (Hay and Liu 1997), I use the audit production efficiency as proxy for competition. The efficiency of the audit (EFFIC) is measured using Data Envelopment Analysis and closely follows Dopuch et al. (2003) model (see Appendix E for details). Based on hypothesis 1, FIRSTYEAR and EFFIC are expected to be negatively associated with overauditing. Cost of providing an audit I use eight different measures to proxy for the audit cost Prior audit research finds that firms with a high risk of material misstatement have higher audit fees indicating that such firms are more costly to audit (OKeefe et al. 1994, Bell et al. 2008). Therefore, I include ROMM which equals 1 if the risk of material misstatement is high and zero otherwise. ROMM is based on the auditors judgment before audit substantive tests are carried out. Seco nd, I also use the number of significant deficiencies (SD) that the auditor finds during the audit process as a proxy of a clients risk and therefore cost. Based on hypothesis 2, ROMM and SD are predicted to be positively related to the probability of ove rauditing. A clients prof itability could also affect the likelihood of overauditing. Profitable firms may have less incentives to commit a reporting fraud (Dechow et al. 1996) a condition that might make it harder for the auditor to justify an increase in audit hours. Therefore, a clients profitability is predicted to be negatively related to the likelihood of overauditing. Clients profitability (ROA) is measured as the ratio of net income over total assets. The analysis also considers whether the client is a subsidiary (SUB) of another

PAGE 128

128 organization. Subsidiaries maybe smaller and less complex than their parent organizations and less costly to audit, therefore they may be a l esser target of overauditing (Fong 2005). Based on this discussion, it is predicted that SUB is negatively related to the probability of overauditing. The auditors extent of reliance on the clients internal controls can also affect the overstatement of a udit hours. If a clients internal controls are strong then the auditor may decide to rely on these controls and limit the extent of testing (OKeefe et al. 1994). Therefore, reliance on controls is a good indicator of a clients risk and the cost of perfo rming the audit. To capture this situation, HRELY equals 1 for engagements with high internal control reliance and zero otherwise. Based on hypothesis 2, since HRELY indicates less costly audits, it is predicted that HRELY is negatively associated with ove rauditing. Two other measure s of audit cost s are a clients complexity of operations and leverage. Prior audit research argues that a clients complexity and leverage are positively associated with audit fees (OKeefe et al. 1994, Hay et al. 2006) indicati ng high risk. Therefore, I include COMPLX which measures the complexity of operations as assessed by the auditor, and LEV, which measures the ratio of long-term debt to total assets. Finally, public firms are considered to be more risky and therefore costl ier to audit. For example, OKeefe et al. (1994) find that publicly traded firms pay higher fees than private firms. Therefore, it is expected that overaudits are more likely to occur in publicly traded firms. The variable PUBLIC equals 1 if the client is a publicly traded company and zero otherwise. Note that while some of the measures that are used to explain ex ante overauditing might not be discovered until after the start of the audit, they could still be reasonably anticipated by the partner based on his expertise and prior information. Therefore, they are expected to affect both the likelihood of exante overauditing and ex post overauditing.

PAGE 129

129 Auditors private information I use eight measures to proxy for an auditors private information about the cli ent. The first factor is the number of reports the auditor prepares for the client. Prior research finds that audit fees and effort are higher when the auditor issues multiple reports (Hay et al. 2006, Bell et al. 2008), however preparation of multiple rep orts can be used to obscure the level of the necessary amount of audit and thus facilitate the likelihood of ex ante overauditing. For example, a larger number of hours can be quoted and justified on the basis of having to prepare multiple reports. The num ber of reports can also affect the likelihood of ex-post overauditing. The rationale is that when multiple reports are issued the auditor may have to do additional audit work to satisfy the requirements for each report. As a consequence, the auditor may be able to learn additional information about the client that would have not been possible if only one report were issued and can use that information to justify an increase in audit hours The number of reports (NREPTS) is therefore included as a factor tha t is associated with the likelihood of overauditing. The a uditors knowledge about the client can also be inferred by observing certain characteristics about how audits are structured. More specifically, I include the percent of time that the auditor spends on understanding the clients business (BUS), the auditors interaction with nonaccounting personnel (OUTSD), and whether the auditor performs interim work for the client (INTRM). The logic for including BUS and OUTSD in determining the likelihood of ex ante overauditing is that when auditors spend a lot of ti me understanding the clients business or talking to non accounting staff it is probably because they lack general information about the client (Johnson et al. 2002, DeAngelo 1981a Ghosh and Moon 2005, Carey and Simnett 2006, Beck and Wu 2006, and Davis e t al. 2007) hence they lack client -specific information which is a necessary condition for any strategic action. If this is the case, then both BUS and OUTSD are predicted to be negatively associated with the likelihood of exante overauditing. Time spen t

PAGE 130

130 understanding the business or talking to nonaccounting personnel can also affect the probability of ex post overauditing. The logic here is that the more time an auditor spends on these activities the more likely he is to discover various aspects of the clients business that he may have not known before or could not learn from only performing audit tests This additional information could facilitate the strategy of ex post overauditing Therefore, I include the variable BUS which measures the percent of time spent on understanding the clients business and OUTSD which equals 1 for engagements where percent of time spent with personnel outside of the financial reporting proce ss is high, and zero otherwise. An auditor can also learn additional information about the client by performing interim audit work. This information could help to justify an increase in the audit hours for the year -end audit therefore it can affect both ex ante and ex -post overauditing. Therefore, I include the variable INTRM which eq uals 1 if the auditor performed i nterim work and zero otherwise. An auditors information on whether a client is under pressure to release earnings numbers could also affect the likelihood of overauditing. Auditors may respond to this pressure by accommoda ting a clients request for a faster audit in exchange for an increased number of audit hours. However, their ability to do so may depend on the underlying sources of the pressure. Prior research suggests that larger clients are under more pressure to not delay the release of their earnings partly in response to these firms investor base that demands more timely information (Sengupta 2004). Prior research also suggests that large firms have more bargaining power to negotiate lower fees (Castarella et al. 2004) therefore making it difficult for an auditor to overaudit in this case. On the other hand, firms with bad news also have incentives to release their earnings early as a way to mitigate litigation risk (Skinner 1994, Skinner 1997). If the pressure to release earnings arises from litigation risk then overauditing may be more likely

PAGE 131

131 because as argued before risky clients are more costly to audit and in turn more likely to experience overauditing (Fong 2005). I measure a clients pressure to release earn ings (DAYSE) as the number of calendar days lapsed after the clients fiscal year -end up to the date of the clients issuance of the annual earnings press release. Another element that could affect the overstatement of the necessary audit is whether the cl ient also purchases other non audit services from the auditor. Here too, the prediction is not straight -forward. First, the provision of non audit services can exacerbate the auditors information advantage as provision of such services can enrich auditor s knowledge about the clients operations beyond the knowledge gained simply through the audit process. The theory of credence goods argues that information asymmetry makes strategic actions more likely, thus from this perspective additional nonaudit serv ices can increase auditors ability to overstate audit hours (Dulleck and Kerschbamer 2006) However, prior research and anecdotal evidence suggest that often audits are used as loss leaders in order to gain the more lucrative consulting contracts (Asthana et al. 2004, Antle et al. 2006) This perspective predicts that high nonaudit services lead to lower audit fees (and thus hours), therefore mitigating the auditors ability to quote unnecessary audit hours. Two measures are used to proxy for nonaudit services, NAS, which is an indicator variable that indicates whether audit clients also purchase other nonaudit services from the auditor, and TAX which indicates whether the client purchases tax services. Because the information learned through th e provision of non audit services is generated after the start of the audit, it is not used as a determinant of ex ante overaudits. Finally, I also consider factors that relate to labor mix. Labor mix has to do with how labor is allocated among different r anks. Labor mix can be a good proxy for an auditors information because different ranks of labor do different tasks (Hackenbrack and Knechel 1997) and doing

PAGE 132

132 different tasks can lead to different information sets. Therefore, I include PTIME, MTIME, and ITIME which measure the ratio of hours expended by partners, managers, and incharge, respectively, over total hours. The coefficient on each rank measures the effect of that rank on the likelihood of overauditing relative to the lowest rank, which is hours w orked by staff (STIME). Similar to non audit services, labor mix proxies for information gathered ex -post and therefore is not included as a determinant of exante overauditing. Clients own information A clients preparedness about the external audit is u sed to proxy for the clients own knowledge. The level of preparedness relates to issues like how well -organized are the clients records, significant changes in key personnel, preparation for the external audit, timely voluntary disclosure of audit releva nt information and so on. A client that is well -prepared may know more about the audit hours that are necessary to complete the engagement and therefore may be less likely to accept a recommendation that exceeds her estimate. A clients own knowledge is ex pected to have a stronger effect on ex -post overauditing because this is the time when the auditor has a more direct interaction with lower level personnel like internal audit staff who based on their level of preparation may know about the audit process a nd can monitor th e auditor s work In order to control for a clients own information I include LPREP which equals 1 if the auditor assesses that the client is very well prepared for the audit i n question, and zero otherwise. Clients bargaining power Foll owing prior literature (Castarella et al. 2004), I use client size (LASSETS) measured as the natural log of total assets to proxy for a clients bargaining power.

PAGE 133

133 Other control variables Finally, I also control for factors that might be associated with an increase in audit hours. These factors are SCOPE which equals 1 if the auditor declared that a scope expansion was necessary, and OPIN that equals 1 if the client received an unqualified opinion. These two factors are only used to determine the likelihood of ex -post overauditing because the opinion is determined at the very end of the audit process, well after the ex ante overauditing decision has been implemented. Further, auditors usually declare an audit scope expansion during the audit process and not b efore the start of the audit. Therefore OPIN and SCOPE are considered irrelevant for the purposes of ex ante overauditing.2 Empirical Results Descriptive Statistics Table 5 1 presents the summary statistics of engagement variables related to equations ( 5 1) and ( 5 2). Table 5 1 shows that the mean (median) value of audit production efficiency (EFFIC) is 0.91 (0.92). This level of efficiency is considered high when compared to prior findings in the literature (Dopuch et al. 2003, Knechel et al. 2009), howev er efficiency seems to vary considerably with its 10th and 90th percentile values of approximately 0.76 and 1, respectively. Moreover, the higher efficiency could also result from the smaller sample size in this study relative to prior studies. Further, i t is also shown that on average the auditor spends about 36 percent of the total time on understanding the clients business (BUS), although this measure also varies as shown by the values of the 10th and 90th percentile which range between 15 and 50 perce nt. Information on significant deficiencies (SD) shows that the average (median) number of significant deficiencies discovered for a particular client is 8 (4). In terms of the 2 Note that equation (51) does not include COMPLX, LASSETS, LEV, and PUBLIC because by construction these factors are included on the left hand side of the equat ion.

PAGE 134

134 number of days that it takes to issue earnings releases (DAYSE), it is shown th at on average it takes about 47 days from the clients fiscal year -end up to the release of annual earnings, with a median level of 37 days. Statistics on labor mix reveal that consistent with prior research (Bell et al. 2008), higher ranks of labor are al located less time than the lower ranks. More specifically, partners (PTIME) and managers (MTIME) use about 6 and 17 percent of the total audit hours, respectively, while in -charge (ITIME) and staff (STIME) about 33 and 37 percent, respectively. Information about audit hours reveals that the mean (median) level of actual, planned, and predicted audit hours is 2,252, 2,027, and 1,920 (1,346, 1,10 0, and 1,226), respectively. This means that on average, planned hours (HPLAN) exceed predicted (PRED) by an additi onal 106 hours, whereas actual hours (HACT) exceed planned (HPLAN) by 225 hours. These simple statistics may provide some support f or the overauditing hypotheses. Table 5 2 shows the frequency distribution of dichotomous variables. A bout half of the sample is first -year clients as indicated by the frequency distribution of FIRSTYEAR which shows that 87 firms ( 51 percent of the sample) are new clients. In terms of the risk distribution, it is shown that 88 firms (51 percent) are considered to have a high risk of material misstatement (ROMM), whereas assessment of internal control quality reveals that 18 firms (11 percent) seem to have strong controls such that the auditor is willing to place a high reliance on those controls (HRELY). In terms of whether clients are public firms and whether they are subsidiaries of other organizations, it is shown that a good majority (about 66 percent) are public firms, whereas only about 18 percent of the samp le, or 31 firms, are subsidiaries of other organizations. Also, about 149 clients (87 percent) received an unqualified opinion that their financial statements fairly present the financial condition. The data also reveals that in about 89 firms (52 percent) there is a

PAGE 135

135 scope expansion, suggesting that potential problems discovered during the audit process could explain the variance in the actual versus planned hours. Clients preparation for the external audit is considered to be high in about 75 engagements (44 percent of the sample) suggesting that the majority of firms are not well -prepared for the audit process in that they maintain messy records, have a high turnover of personnel etc. In terms of the audit structure, the data shows that in a considerable number of engagements (39 percent) the auditor spends a long time conducting inquiries with personnel outside the finance and accounting functions (OUTSD). Also, interim work (INTRM) is performed in about 28 engagements (16 percent). Data on nonaudit serv ices reveals that a greater number of firms (120) purchase tax services (TAX), while only about 79 firms purchase other consulting services (NAS) Finally, industry distribution reveals that 66 percent of the sample belongs to the consumer and industrial b usiness industry (CIB), while the remainder is from the information, communication and entertainment industry (ICE). Table 5 3 shows the mean dis tribution of certain engagement variables by EXAGG, the proxy for ex ante overauditing. Note that based on the value of EXAGG, about a third of the sample, or 57 engagements, are classified to be ex ante overaudits. The table also shows mean differences of various variables across EXAGG, and the associated significance based on a t test. First, note that hours planned exceed the predicted value (HPLAN PRED) by 1,545 hours in the case of overaudits (EXAGG = 1), whereas for non-overaudits the predicted exceed planned hours by 613 hours. Moreover, the mean difference of the excess hours (2,157 hours) between ex ante overaudits and the remainder of the sample is significant at 1 percent. Further, profitability data indicates that pro fits are higher for exante o veraudits relative to the remainder of the sample when using all three measures of profitability (PROF, AFH a nd UFEE), and significantly mo re profitable under two measures of profitability (AFH and UFEE). These

PAGE 136

136 descriptive statistics provide evidence that ex ante overaudits are profitable to the auditor. Also note that the efficiency score (EFFIC), which is used to proxy for competition, is significantly smaller for ex ante overaudits providing support for the hypothesis that overauditing is less likely to occur in competitive settings. The data also reveals that in overaudited engagements, a greater percent of ti me is spent on understanding the business of the client (47 versus 30 percent), although the difference is not significant. It is interesting to note that overaudits have significantly more deficiencies than the non -overaudits, which provides univariate su pport for the theoretical prediction that overaudits are more likely to occur in clients that are costly to service (i.e., those that have significant reporting deficiencies). There seems to be no differences across the two groups in the number of reports produced and the number of days to the earnings release. Table s 5 4 5 5 and, 5 6 show the distribution of the continuous engagement variables by the three measures of ex -post overauditing, P_ EXCD, A_EXCD, and U_EXCD, respectively. The last column in the tables presents mean differences across the overaudits and non-overaudits and the associated significance based on the t -tests. First, note that the value of HACT HPLAN is positive, (or more positive) for the ex-post overaudits, relative to the nonovera udits. Mean differences of HACT HPLAN between the overaudits and nonoveraudits are significant for P_EXCD and U_EXCD. Interestingly, in all three tables the overaudits are more profitable than non -overaudits when using all three measures of profitability (PROF, AFH, and UFEE) indicating that ex -post overaudits are profitable to the auditor. This provides confidence that ex post overaudits are well -identified. Univariate mean differences between the overaud its and the non -overaudits show that they are not s ignificantly different from each other in many different aspects. The only differences are i n the number of reports (Table 5 5 ), where the number of

PAGE 137

137 reports (NREPTS) issued for the overaudit group is twice as large as that of the non -overaudit group, and i n the number of significant deficiencies (SD) which is also greater for the overaudit group (Table 5 4 ). Moreover, the two groups also differ in that the in -charge (ITIME) is allocated less labor in the overaudit group, and it seems to be replaced by the s taff (STIME). Table 5 7 shows the correlation matrix of the variables related to equation (5 1). First note th at EXAGG is negatively related to both measures of competition (FIRSTYEAR and EFFIC) further indicating that ex ante overstatement is less likely in competitive settings. Moreover, the results show that EXAGG is also negatively related to the number of days it takes for the earnings release indicating that when clients are under pressure to release earnings, there is less of a chance to overstate au dit hours. On the other hand, EXAGG is positively related to the number of significant deficiencies (SD) and the percent of time spent understanding the clients business (BUS). Table 5 8 shows the correlations for variables related to equation ( 5 2). Firs t, note that the three measures of ex -post overauditing (P_EXCD, A_EXCD, and U_EXCD) are positively correlated. This correlation is especially strong between P_EXCD and A_EXCD, where the correlation coefficient is 0.67 indicating that these two variables m easure a similar construct. The correlation between P_EXCD and U_EXCD is 0.39, whereas the correlation between A_EXCD and U_EXCD is 0.37 indicating that U_EXCD may be measuring a slightly different construct than P_EXCD and A_EXCD. Some interesting observa tions include that in general EXCD is positively correlated with PUBLIC indicating that the likelihood of ex -post audit overstatement is more likely for public firms, while it is negatively associated with LPREP suggesting that a well -prepared firm has les s of a chance to experience ex post overauditing. Finally, interim work (INTRM) is positively associated with EXCD indicating that if an auditor performs interim work

PAGE 138

138 he is more likely to ex -post overstate audit hours. This result is consistent with the information view in that interim work provides the auditor with additional information that can be used strategically later. Importantly, EXCD is also positively associated with SCOPE suggesting that for some of the engagements the overstatement of audit hours arises for non -strategic reasons like when the auditor deems it necessary to expand the audit scope. This shows the importance of controlling for this variable. Probit Results Ex ante overauditing Table 5 9 presents the probit results of equation ( 5 1). As stated earlier the investigation is exploratory therefore different variations of equation ( 5 1) are presented in the table. The interpretation of the results will be based on the complete model which is model (8) shown in the last column of T able 5 9 First note that the model is reasonably well -specified with a pseudo R2 of 29 percent. The results of the probit regression show that competition (FIRSTYEAR and EFFIC) negatively affects the likelihood of expost overauditing. This is consistent with the theoretical prediction that competition mitigates overauditing ( DeAngelo 1981a, Wolinsky 1993, Hubbard 1998). In terms of the measures that proxy for the cost of an audit, the results show that the risk of material misstatement (ROMM) and the number of si gnificant deficiencies ( SD) are positive and significant suggesting that client risk (and hence the cost of auditing) are positively associated with the likelihood of ex ante overauditing. In addition, the results also show that being a subsidiary of anot her organization is negatively associated with the likelihood of exante overauditing. Altogether, these findings are consistent with the theoretical prediction that experts are more likely to exaggerate the needed service when they face high -cost consumer s (Fong 2005) For example, subsidiaries may be cheaper to service because they may be smaller and less complex than parent organizations and therefore less likely to be targets of overauditing. Clients

PAGE 139

139 profitability (ROA) and the extent of reliance on in ternal controls (HRELY) do not seem to be associated with ex ante overauditing. In regards to the auditors private information about the client, the results indicate that the percent of time spent understanding the clients business (BUS) is positively an d significantly associated with the likelihood of ex ante overauditing. This result is a bit unexpected because it suggests that overauditing is more likely to occur when auditors lack client -specific information which is contrary to what theory predicts. However, percent of time spent understanding a clients business could also proxy for risk. Prior research in auditing finds that auditors spend a lot of time understanding their clients business during the early years of auditor -client tenure (i.e., shor t tenures). Since short tenure engagements are associated with a higher risk of litigation it then follows that engagements with a high BUS are risky. In this perspective, the positive and significant coefficient on BUS conforms to the theoretical prediction that clients that are risky will experience more overauditing. Interestingly, time spent talking to non accounting personnel (OUTSD) negatively affects the likelihood of exante overauditing. This finding is consistent with the theoretical prediction th at less client -specific information reduces the likelihood of overauditing. Another interesting result is that overauditing is less likely to occur in clients that are under pressure to release earnings as measured by DAYSE. As argued earlier both large f irms and firms with high litigation risk are under pressure to release their earnings on a timely fashion. The result on DAYSE is more consistent with the large firms reporting early because they are the ones least likely to experience overauditing. It is interesting to also note that the number of reports issued for the client (NREPTS) and interim work (INTRM) are not associated with the likelihood of ex ante overauditing. The coefficient on client preparedness (LPREP) is

PAGE 140

140 positive and significant indicatin g that ex ante overauditing is more likely to occur when the client is well -prepared for the external audit Ex post overauditing Table s 5 10, 5 11, and 5 12 present the probit regression results of equation (5 2) when the dependent variable is equal to P_ EXCD, A_EXCD, and U_EXCD, respectively. Because equation (5 2) includes many proxies that measure the same underlying constructs I also run a stepwise probit regression. The results of the stepwise probit are shown in the last column of T able 5 10, 5 11, and 5 12. Recall that the dependent variable of interest, P_ EXCD, A_EXCD, and U_EXCD equals 1 if actual hours exceed planned hours and if profits are high. Five different variations of equation (5 2) are estimated including the stepwise probit regression model. Interpretation of the results will be based primarily on t he stepwise regression outcome. Results in Table 5 10 show the probit regression results for P_EXCD, the case where ex post overaudits are identified based on the engagement profitability (PR OF). The model is reasonably well -specified with a pseudo R2 of 11 percent. The results indicate that ex-post overauditing is less likely to occur in highly competitive settings (based on models 1 through 4), as measured by FIRSTYEAR, although this variable is not significant in the stepwise probit. In terms of var iables that proxy for the audit costs results indicate that only PUBLIC and HRELY are significant. More specifically, public companies seem to experience more overauditing as suggested by the positive and significant coefficient on PUBLIC. Also, there see ms to be less overauditing when the auditor places a high reliance on a clients internal controls, as measured by HRELY. Both these results are consistent with the theoretical prediction in Fong (2005) since PUBLIC indicate high -cost audits whereas HLREY could indicate low -cost audits. Various client characteristics that are observable to researchers like profitability (ROA), complexity

PAGE 141

141 (COMPLX), and leverage (LEV) are not significantly associated with the likelihood of ex -post overauditing. In regards to the auditors private information, the results indicate that when auditors perform interim work on the client (INTRM) there is a higher chance of ex -post overauditing. This is consistent with the prediction that client -specific information learned through performing interim work increases the likelihood of ex-post overauditing. On the other hand, information as measured by the percent of total time spent by partners (PTIME) is negatively associated with the likelihood of ex-post overauditing. This result s uggests that information learned through tasks performed by partners makes overauditing less likely relative to information learned through tasks performed by the lower ranks like staff. Moreover, performing tax services (TAX) makes ex -post overauditing le ss likely although this result is not significant in the stepwise regression. Other measures of the auditors information like the number of reports (NREPTS), percent of time spent on understanding the clients business (BUS), percent of time spent with no n accounting personnel (OUTSD), the clients pressure to release earnings (DAYSE) and performance of nonaudit services (NAS) do not seem to be associated with ex -post overauditing. This is somewhat surprising because the provision of nonaudit services ( including tax) may enrich the auditors information about the client and can therefore potentially facilitate strategic behavior. However, this paper only focuses on the strategy of overstating audit hours (i.e., overauditing), so it could very well be tha t the provision of nonaudit service s affects other potential strategic choices. Nevertheless, the results (or lack of) associated with the non audit services could also be explained on the grounds that since nonaudit services may be a lucrative business for the auditor, he may have less incentives to overstate audit hours in audit client s that also purchase non audit services.

PAGE 142

142 Interestingly, the preparation level of the client (LPREP) is negatively associated with the ex -post overstatement of audit hours. This may be the case because client preparation could be indicative of how seriously the client considers the ext ernal audit and also about her own knowledge as to whether additional audit hours are actually necessary. To a certain extent, client s prepar edness can proxy for the clients ability to monitor the external auditor therefore limiting the auditors ability to act strategically. A clients bargaining power does not seem to affect the likelihood of ex-post overauditing as indicated by the insignif icant coefficient on LASSETS. Finally, SCOPE and OPIN are not significantly associated with the p robability of overauditing. This finding could indicate that P_EXCD is able to accurately identify strategic over audits. The results for A_EXCD presented in Ta ble 5 11 are similar to those in Table 5 10. The only difference in this case is that the number of reports (NREPTS) seems to increase the likelihood of ex-post overauditing which is more consistent with the expectation because as previously stated the num ber of reports can be used to justify an additional increase in audit hours. The results of U_EXCD are presented in Table 5 12. First, note that consistent with the prediction, overaudits are less likely to occur when competition is high as measured by the negative and significant coefficient on EFFIC. In terms of the variables that measure the cost of performing an audit, ROA is negative and significant. This result is similar to P_EXCD in Table 5 10 and is consistent with the expectation that overauditing is less likely to occur in less risky clients which in our case is proxied by high ROA. When it comes to the effect of the auditors private information it is shown that interim work (INTRM) and performance of nonaudit services (NAS) positively and significantly affect the likelihood of ex -post overauditing. The

PAGE 143

143 result of NAS is interesting because it is consistent with the theoretical prediction that clientspecific information (in this case learned through NAS provision) is associated with more overauditing (strategic actions). Finally, SCOPE is positive and significant suggesting that the likelihood of overauditing is higher if the auditor de clared a scope expansion. Overall, the results in this section show that different client characteristics drive ex ante and ex -post overauditing. It is especially important to note that client characteristics observable to researchers like client size, complexity, and leverage are not significantly associated with overauditing. This is a very meaningful result provi ding empirical support for the prediction that an experts strategic behavior depends on what the expert learns privately through his own interaction with the client. Robustness Tests In the main analysis, ex ante overaudits are identified as engagements w here planned hours exceed the benchmark, whereas ex post overaudits are identified based on the median levels of engagement profitability (PROF), hourly audit fee (AFH), and the unexpected fee (UFEE). In order to check the robustness of the results, all te sts are replicated using different cut offs. More specifically, EXAGG equals 1 if the difference, HPLAN -PRED, is greater than the median, and then again greater than the 75th percentile and zero otherwise. The results are consistent with the results presen ted in T able 5 9 The only difference is that INTRM, DAYSE, OUTSD are significant only at the 10 percent level, and LPREP is not significant. In addition, P_EXCD, A_EXCD, and U_EXCD are defined to equal 1 in cases where actual hours exceed planned hours and for which PROF, AFH, and UFEE is greater than the 75th percentile, respectively. The results of stepwise probit based on P_EXCD show that consistent with the main analysis, PUBLIC and INTRM are positively associated whereas LPREP is negatively associated with the likelihood of ex -post overauditing. In addition, HRELY and

PAGE 144

144 PTIME are not sig nificant. The results based on A _EXCD show that in addition to the variables shown in T able 5 9 as significantly affecting the likelihood of ex-post overauditing, other variables are also significant More specifically, the nu mber of significant deficiencies (SD) and size (LASSETS) are positively associated with the probability of ex -post overauditing, whereas percent of time spent with non accounting personnel (OUTSD) and the return on assets (ROA) are negatively related to ex -post overauditing. The results based on U_EXCD show that subsidiaries (SUB), and time spent understanding a clients business (BUS) are negatively, whereas the clients complexity (COMPLX) is positively associated with ex -post overauditing. These results are consistent with the theoretical predictions. Conclusion and D iscussion Regulators have expressed concern that audits are becoming increasingly inefficient and that auditors frequently prescribe more audit service than a client might want or need (i.e., overauditing) This paper investigates engagement -specific characteristics associated with the likelihood of overauditing where two strategies of overauditing are identified. It is argued that the auditor can engage in either, ex ante overauditing, recomm ending more audit than necessary prior to the start of the audit, or ex -post overauditing, increasing audit hours after the start of the audit. The analysis is exploratory in that theory does not provide a clear guidance as to which factors might exacerbat e the incentive to overaudit. Results show that ex ante overaudits are positively related to the risk of material misstatement, significant number of deficiencies, and time spent understanding a clients business, and negatively related to competition, cli ent preparedness, time spent with nonaccounting personnel, the number of days till the client releases earnings and whether the client is a subsidi ary of another organization. On the other hand, ex -post overauditing is more likely to occur in clients tha t are publicly traded, and in clients for whom multiple reports are issued, but

PAGE 145

145 is less likely to occur w hen tax services are performed and when the client is highly prepared for the external audit. Moreover, some results show that ex post overaudits have a disproportionate amount of effort allocated to cheaper ranks of labor. The results in this paper have important implications for policies directed towards the auditing function. The theory of credence goods argues that auditors are likely to strategicall y utilize the superior information they have about the audit and any information they learn about the client. One such strategic action has to do with supplying and charging for a larger number of audit hours than is necessary resulting in inefficient audi ts. The goal of this paper is to raise awareness about this feature of audits and to emphasize that increased regulation might inadvertently exacerbate this feature leading to additional inefficiencies in the marketplace Even though overauditing increases welfare in the form of higher assurance about the reliability of financial statements, this could come at increased costs to buyers of audits having to pay for more service that they want. Another contribution of this study is that it provides initial evidence on the issue of overauditing. Importantly, overauditing is not associated with client characteristics that are observable to outsiders, but it seems to be a function of the auditors private information about the client.

PAGE 146

146 Table 5 1. Descriptive statistics of continuous variables Variables Mean Median Standard deviation P10 P90 EFFIC 0.91 0.92 0.09 0.76 1 BUS 35.66 30 .00 85.41 15 .00 50 .00 SD 7.96 4 .00 9.73 0 .00 21.00 NREPORTS 5.3 2 .0 10. 0 1 .0 10.0 DAYSE 47.37 37 .00 81 .00 0 .00 98 .00 COMPLX 2.61 3 .00 0.88 2 .00 4 .00 PTIME 0.06 0.06 0.03 0.03 0.1 1 MTIME 0.17 0.16 0.06 0.09 0.24 ITIME 0.33 0.32 0.14 0.17 0.51 STIME 0.37 0.39 0.16 0.05 0.54 HACT 2,252 1,346 3,035 453 4,612 HPLAN 2,027 1,100 3,347 400 3,400 PRED 1,920 1,226 1,795 629 4,231 HPLAN -PRED 106.2 191.8 2,408.3 1,123.7 961.9 HACT HPLAN 225. 5 194 .0 2,154.5 120 .0 1 100 .0 All variables are defined in Appendix C.

PAGE 147

147 Table 5 2. Frequency distribution of engagement characteristics Variables N Percent FIRSTYEAR 87 51 ROMM 88 51 INTRM 28 16 LPREP 75 4 4 SUB 31 18 OUTSD 66 39 HRELY 18 1 1 OPIN 149 87 SCOPE 89 52 PUBLIC 112 65 NAS 79 46 TAX 120 70 ICE 58 3 4 CIB 113 66 All variables are defined in Appendix C.

PAGE 148

148 Table 5 3. Engagement characteristics by EXAGG Variables EXAGG = 0 EXAGG = 1 Difference HPLAN PRED 613.05 1 544.78 2 157.83*** PROF 0.63 0.61 0.02 AFH 154.7 196 .0 41.3*** UFEE 0.23 0.47 0.7 0 *** EFFIC 0.93 0.85 0.08*** BUS 29.87 47.21 17.34 SD 6.7 0 10.47 3.77*** NREPTS 5.15 5.61 0.46 DAYSE 52.93 36.23 16.7 ROA 0.02 0.04 0.02 N 114 57 T tests are two tailed. All variables are defined in Appendix C.

PAGE 149

149 Table 5 4. Engagement characteristics by P_EXCD Variables P_EXCD =0 P_EXCD=1 Difference HACT HPLAN 21.05 658.94 679.99** PROF 0.52 0.8 0 0.28*** AFH 163 177 14 UFEE 0.08 0.15 0.23*** ASSETS 1 915.76 1 304.43 611.32 COMPLX 2.6 0 2.63 0.03 LEV 0.27 0.29 0.02 ROA 0.003 0.080 0.083 NREPTS 4.73 6.3 0 1.57 PTIME 0.066 0.063 0.003 MTIME 0.168 0.165 0.003 ITIME 0.35 0.31 0.04* STIME 0.36 0.39 0.03 EFFIC 0.91 0.91 0 .00 SD 6.65 10.26 3.61** DAYSE 46.03 49.73 3.7 0 BUS 39.03 29.73 9.3 0 N 109 62 T tests are two tailed. All variables are defined in Appendix C.

PAGE 150

150 Table 5 5. Engagement characteristics by A_EXCD A_EXCD =0 A_EXCD=1 Difference HACT HPLAN 83.01 570.3 0 487.29 PROF 0.55 0.79 0.24*** AFH 155.8 0 198.8 0 43 .00 *** UFEE 0.07 0.18 0.25*** ASSETS 1 798.4 9 1 441.51 356.97 COMPLX 2.59 2.64 0.05 LEV 0.28 0.28 0 .00 ROA 0.02 0.06 0.04 NREPTS 4.03 8.38 4.35*** PTIME 0.065 0.064 0.001 MTIME 0.17 0.16 0.01 ITIME 0.34 0.31 0.03 STIME 0.36 0.38 0.02 EFFIC 0.90 0.91 0.01 SD 7.4 9.3 1.9 DAYSE 48.2 45.3 2.9 BUS 38.4 28.9 9.5 N 121 50 T tests are two tailed. All variables are defined in Appendix C.

PAGE 151

151 Table 5 6. Engagement characteristics by U_EXCD Variables U_EXCD =0 U_EXCD=1 Difference HACT HPLAN 77.53 758.24 835.77*** PROF 0.58 0.7 0 0.12*** AFH 171.32 163.37 7.95 UFEE 0.21 0.37 0.58*** ASSETS 916.8 0 3 ,060.67 2 143.8 7 COMPLX 2.64 2.55 0.09 LEV 0.3 0 0.24 0.06* ROA 0.01 0.09 0.1 0 NREPTS 4.77 6.24 1.47 PTIME 0.066 0.063 0.003 MTIME 0.16 0.17 0.01 ITIME 0.35 0.3 0 0.05*** STIME 0.35 0.4 0 0.05* EFFIC 0.93 0.86 0.07*** SD 7.7 8.4 0.7 DAYSE 51.3 40.5 10.8 BUS 39.1 29.7 9.4 N 109 62 T tests are two tailed. All variables are defined in Appendix C.

PAGE 152

152 Table 5 7. Correlations related to variables in equation (5 1) EXAGG FIRSTYEAR EFFIC ROMM BUS SD NREPTS INTRM DAYSE LPREP ROA SUB OUTSD HRELY EXAGG 1.00 FIRSTYEAR 0.07 1.00 EFFIC 0.39 0.12 1.00 ROMM 0.09 0.09 0.16 1.00 BUS 0.10 0.09 0.01 0.07 1.00 SD 0.18 0.11 0.02 0.19 0.08 1.00 NREPTS 0.02 0.02 0.08 0.06 0.02 0.02 1.00 INTRM 0.04 0.08 0.04 0.05 0.04 0.04 0.09 1.00 DAYSE 0.10 0.06 0.01 0.05 0.04 0.08 0.03 0.01 1.00 LPREP 0.05 0.08 0.04 0.25 0.07 0.33 0.01 0.07 0.11 1.00 ROA 0.04 0.08 0.02 0.07 0.00 0.02 0.05 0.10 0.00 0.05 1.00 SUB 0.04 0.13 0.06 0.03 0.15 0.02 0.08 0.16 0.13 0.01 0.00 1.00 OUTSD 0.05 0.58 0.12 0.17 0.07 0.10 0.08 0.07 0.11 0.17 0.04 0.03 1.00 HRELY 0.04 0.07 0.04 0.27 0.03 0.12 0.08 0.00 0.02 0.23 0.07 0.01 0.08 1.00 All variables are defined in Appendix C.

PAGE 153

153 Table 5 8. Correlations related to variables in equation (5 2) P_EXCD A_EXCD U_EXCD LASSETS COMPLX LEV PUBLIC ROA NAS TAX PTIME MTIME ITIME EFFIC P_EXCD 1.00 A_EXCD 0.67 1.00 U_EXCD 0.39 0.37 1.00 LASSETS 0.07 0.10 0.04 1.00 COMPLX 0.02 0.02 0.05 0.50 1.00 LEV 0.03 0.00 0.13 0.25 0.13 1.00 PUBLIC 0.16 0.17 0.03 0.16 0.08 0.17 1.00 ROA 0.12 0.05 0.14 0.16 0.10 0.04 0.04 1.00 NAS 0.00 0.06 0.08 0.31 0.26 0.01 0.10 0.03 1.00 TAX 0.04 0.05 0.07 0.05 0.09 0.05 0.23 0.06 0.07 1.00 PTIME 0.04 0.01 0.03 0.01 0.09 0.10 0.29 0.08 0.01 0.02 1.00 MTIME 0.02 0.10 0.04 0.05 0.07 0.04 0.12 0.13 0.04 0.16 0.45 1.00 ITIME 0.14 0.04 0.19 0.22 0.15 0.03 0.26 0.09 0.14 0.14 0.04 0.02 1.00 EFFIC 0.00 0.04 0.36 0.06 0.15 0.07 0.22 0.02 0.02 0.06 0.04 0.02 0.17 1.00 FIRSTYEAR 0.07 0.16 0.00 0.02 0.07 0.04 0.06 0.08 0.03 0.08 0.02 0.09 0.12 0.12 NREPTS 0.07 0.19 0.07 0.24 0.17 0.13 0.05 0.05 0.08 0.11 0.03 0 0 7 0.07 0.08 LPREP 0.24 0.18 0.13 0.04 0.17 0.00 0.05 0.05 0.10 0.02 0.02 0.08 0.03 0.04 INTRM 0.19 0.13 0.19 0.02 0.00 0.05 0.25 0.10 0.04 0.15 0.22 0.08 0.00 0.04 OPIN 0.07 0.06 0.18 0.12 0.11 0.12 0.12 0.10 0.02 0.13 0.05 0.01 0.09 0.00 SCOPE 0.11 0.05 0.03 0.17 0.14 0.10 0.26 0.02 0.19 0.19 0.21 0.15 0.07 0.17 SD 0.18 0.09 0.04 0.12 0.26 0.12 0.02 0.02 0.22 0.05 0.09 0.12 0.14 0.02 ROMM 0.12 0.19 0.07 0.02 0.10 0.08 0.06 0.07 0.03 0.12 0.09 0.05 0.05 0.16 SUB 0.01 0.03 0.07 0.03 0.01 0.14 0.48 0.00 0.28 0.12 0.12 0.08 0.18 0.07 HRELY 0.18 0.14 0.02 0.12 0.08 0.02 0.07 0.07 0.03 0.06 0.17 0.09 0.05 0.04 BUS 0.05 0.05 0.05 0.03 0.03 0.13 0.12 0.00 0.06 0.13 0.06 0.01 0.24 0.01 OUTSD 0.05 0.02 0.05 0.03 0.07 0.01 0.02 0.04 0.16 0.15 0.07 0.01 0.07 0.12 DAYSE 0.02 0.02 0.06 0.01 0.05 0.11 0.03 0.00 0.11 0.09 0.07 0.01 0.05 0.01

PAGE 154

154 Table 5 8 Continued FIRSTYEAR NREPTS LPREP INTRM OPIN SCOPE SD ROMM SUB HRELY BUS OUTSD DAYSE FIRSTYEAR 1.00 NREPTS 0.02 1.00 LPREP 0.08 0.01 1.00 INTRM 0.08 0.09 0.07 1.00 OPIN 0.01 0.08 0.09 0.07 1.00 SCOPE 0.22 0.04 0.17 0.20 0.08 1.00 SD 0.11 0.02 0.33 0.04 0.03 0.19 1.00 ROMM 0.09 0.06 0.25 0.05 0.15 0.12 0.19 1.00 SUB 0.12 0.09 0.01 0.16 0.22 0.13 0.02 0.03 1.00 HRELY 0.07 0.08 0.23 0.00 0.06 0.01 0.12 0.27 0.01 1.00 BUS 0.09 0.02 0.07 0.04 0.03 0.08 0.08 0.07 0.15 0.03 1.00 OUTSD 0.57 0.08 0.17 0.07 0.05 0.13 0.10 0.17 0.03 0.08 0.07 1.00 DAYSE 0.06 0.03 0.11 0.00 0.00 0.02 0.08 0.05 0.13 0.02 0.05 0.12 1.00 All variables are defined in Appendix C.

PAGE 155

155 Table 5 9. Probit results of equation (5 1) Variables Model (1) Model (2) Model (3) Model (4) Model (5) Model (6) Model (7) Model (8) FIRSTYEAR 0.35* 0.34 0.38* 0.42* 0.44** 0.46** 0.82*** 0.92*** EFFIC 6.18*** 6.74*** 6.23*** 6.5*** 7.02*** 7.11*** 8.42*** 8.5*** ROMM --0.47** ----0.41* 0.41* 0.73*** 0.7*** SD ------0.03*** 0.02** 0.03** 0.03*** 0.03*** ROA --------------0.42 SUB --------------0.85** HRELY --------------0.57 NREPTS ----------0.01 0.01 0.01 BUS ----0.002*** --0.002*** 0.002** 0.002** 0.002*** INTRM ----------0.28 0.3 0.41 OUTSD ------------0.64** 0.59** DAYSE ----------0.005 0.004 0.01** LPREP ------------0.47* 0.56** Intercept 5.3*** 5.55*** 5.29*** 5.37*** 5.61*** 5.88*** 6.78*** 7.23*** N 171 171 171 171 171 171 171 171 Pseudo R 2 0.13 0.16 0.14 0.17 0.19 0.21 0.26 0.29 This table presents the probit regression results of equation ( 5 1 ) where the dependent variable is EXAGG. Columns (1) through (8 ) present different variations of equation (5 1). All tests are two -tailed. ***, **, and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. Variables are defined in Appendix C

PAGE 156

156 Table 5 10. Probit results of equation (52) when the dependent variable is P_EXCD Variables Model (1) Model (2) Model (3) Model (4) Step wise probit FIRSTYEAR 0.573** 0.584** 0.601** 0.429* --EFFIC 1.112 -------ROMM 0.168 0.123 ------SD 0.014 0.014 0.016 0.016 --ROA 0.300 0.317 0.378 0.352 --SUB 0.301 0.306 ------HRELY 1.057** 0.978** ----0.795* COMPLX 0.287* 0.247 0.194 0.186 --LEV 0.047 0.135 0.112 0.054 --PUBLIC 0.441 0.534* 0.441* 0.444* 0.431** NREPTS 0.013 0.014 0.0136 0.122 --BUS 0.001 0.001 0.001 ----INTRM 0.501 0.498 0.435 0.448 0.621** OUTSD 0.274 0.280 0.325 ----DAYSE 0.000 0.000 0.000 ----NAS 0.266 0.235 0.183 0.157 --TAX 0.508* 0.480* 0.503* 0.529** --PTIME 8.315* 8.029* 5.962 6.72 6.807* MTIME 1.455 1.456 1.328 1.395 --ITIME 1.337 1.156 1.072 1.149 --LPREP 0.585** 0.554** 0.626*** 0.663*** 0.532*** LASSETS 0.079 0.061 0.044 0.046 --OPIN 0.093 0.093 0.063 0.076 --SCOPE 0.283 0.230 0.193 0.204 --Intercept 0.757 0.186 0.235 0.103 0.026 N 171 171 171 171 171 Pseudo R 2 0.19 0.19 0.17 0.16 0.11 This table presents the probit regression results of equation ( 5 2 ) where the dependent variable is P_EXCD. Columns (1) through (4 ) present different variations of equation (5 2). All tests are two tailed. ***, **, and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. Variables are defined in Appendix C

PAGE 157

157 Table 5 11. Probit results of equation (52) when the dependent variable is A_EXCD Variables Model (1) Model (2) Model (3) Model (4) Step wise probit FIRSTYEAR 0.763*** 0.767*** 0.756*** 0.640*** 0.554*** EFFIC 0.205 --------ROMM 0.265 0.271 ------SD 0.006 0.006 0.007 0.007 --ROA 0.08 0.085 0.121 0.115 --SUB 0.036 0.036 ------HRELY 0.884** 0.873** 0.998*** 1.026*** 0.848** COMPLX 0.162 0.154 0.156 0.156 --LEV 0.174 0.157 0.116 0.169 --PUBLIC 0.621* 0.639** 0.666** 0.641** 0.658*** NREPTS 0.0372*** 0.037*** 0.037*** 0.036*** 0.035*** BUS 0.000 0.000 0.000 ----INTRM 0.281 0.279 0.260 0.279 --OUTSD 0.325 0.326 0.242 ----DAYSE 0.001 0.001 0.001 ----NAS 0.127 0.133 0.111 0.141 --TAX 0.002 0.002 0.047 0.079 --PTIME 4.725 4.599 4.404 4.811 --MTIME 0.630 0.607 0.447 0.448 --ITIME 1.102 1.078 1.017 1.063 --LPREP 0.333 0.329 0.382 0.407 0.422* LASSETS 0.076 0.073 0.074 0.080 --OPIN 0.154 0.154 0.178* 0.179* 0.193* SCOPE 0.025 0.014 0.059 0.063 --Intercept 0.945 0.767 0.705 0.849 0.943*** N 171 171 171 171 171 Pseudo R 2 0.18 0.18 0.17 0.17 0.14 This table presents the probit regression results of equation ( 5 2 ) where the dependent variable is A_EXCD. Columns (1) through (4 ) present different variations of equation (52). All tests are two tailed. ***, **, and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. Variables are defined in Appendix C

PAGE 158

158 Table 5 12. Probit results of equation (52) when the dependent variable is U_EXCD Variables Model (1) Model (2) Model (3) Model (4) Step wise probit FIRSTYEAR 0.401 --------EFFIC 5.508*** 5.341*** 5.346*** 5.267*** 5.863*** ROMM 0.204 0.216 ------SD 0.003 0.004 0.007 0.007 --ROA 0.661* 0.616* 0.551 0.548 0.819** SUB 0.085 0.045 ------HRELY 0.236 0.208 ------COMPLX 0.004 0.015 0.026 0.032 --LEV 0.730 0.756 0.778 0.889* --PUBLIC 0.062 0.089 0.918 0.145 --NREPTS 0.019* 0.017 0.0167 0.016 --BUS 0.001 0.002 0.002 ----INTRM 0.431 0.465 0.484 0.500 0.531* OUTSD 0.173 0.042 0.015 ----DAYSE 0.003 0.003 0.003 ----NAS 0.604** 0.620** 0.616** 0.574** 0.443** TAX 0.090 0.058 0.021 0.045 --PTIME 4.352 3.949 4.810 4.891 --MTIME 0.260 0.188 0.065 0.107 --ITIME 1.911** 1.825* 1.893** 1.907** --LPREP 0.471* 0.483** 0.409* 0.378 --LASSETS 0.128 0.118 0.114 0.099 --OPIN 0.019 0.026 0.012 0.002 --SCOPE 0.482* 0.403 0.391 0.415 --Intercept 7.301*** 6.907*** 6.855*** 6.531*** 4.583*** N 171 171 171 171 171 Pseudo R sq 0.26 0.25 0.25 0.24 0.18 This table presents the probit regression results of equation ( 5 2 ) where the dependent variable is U_EXCD. Columns (1) through (4 ) present different variations of equation (52). All tests are two tailed. ***, **, and indicate coefficient significance at p -values less than 0.01, 0.05, and 0.10, respectively. Variables are defined in Appendix C

PAGE 159

159 CHAPTER 6 CONLUDING REMARKS This study seeks to expand our understanding of auditing by proposing a different theoretical framework under which to view the audit production process. Tradition al audit research makes two assumptions regarding the audit production process: first, it ass umes that firms can determine their own demand for a ssurance with certainty (Simunic 1980) and that auditor reputation and size ensure high quality audits (DeAngelo 1981). The first assumption invalidates the role of an auditor as an expert in the marketpl ace, while the second predicts that auditors who care about their reputation always provide effective and efficient audits. However, in addition to supplying audits, auditors advise their client firms of the appropriate amount of audit needed to achieve th e desired level of assurance. Moreover, recent audit production research and anecdotal evidence suggest s that audits are generally not conducted on an efficient basis so these assumptions may be unrepresentative of the actual audit environment that buyers and sellers face (Dopuch et al. 2003, Kaplan et al. 2007, Knechel et al. 2009). In this perspective, audits exhibit credence attributes. It is therefore important to analyze how our understanding of audits changes when these assumptions are relaxed. The an alysis in this study is grounded on the theory of credence goods which argues that an expert -seller such as an auditor, possesses an information advantage with respect to the service he provides and has incentives to act strategically by advising c lients to purchase more of a service than needed (overauditing) and/or by providing less service than initially recommended (overcharging) or even less than the client needs (underauditing) (Dulleck and Kerschbamer 2006). In the first set of hypotheses this study argues that the auditor gains private information about a clients audit needs through repeated interaction with the client, and may be able to use this information to maximize his own profit (Beck and Wu 2006, Fong 2005). Therefore, as

PAGE 160

160 private informati o n increases with tenure, so do the engagement profits. However, a clients own learning and switching costs may mitigate strategic acts over time (Lewis and Yildirim 2002, Fong 2005, Dulleck and Kerschbamer 2006). As a result of the interaction between th e auditors private information and a clients switching costs over time, the engagement profits are predicted to increase with tenure, but eventually decrease. The theory of credence goods also argues that sellers are less able to extract rents from consu mers who are well informed about the products they demand. Following a similar rationale, it is predicted in this paper that a clients own knowledge about the audit process is associated with lower profits. Using proprietary audit engagement data from one of the international auditing firms, I find that engagement profits are highest for mid tenure clients relative to the short and the long tenured ones. Further, I also find that engagement profits are lower when clients have internal audit departments and are well prepared for the external audit, where the presence of internal audit departments and the clients level of preparedness proxy for the clients knowledge about the audit process. These results are consistent with the hypotheses generated from the credence goods theory. Further, assessment of the effort and fee variables reveals that audit production is most consistent with overcharging. More specifically, higher audit profitability is achieved by exerting less effort than expected and by charging a higher fee than expected, where the expected effort and fees are calculated as the predicted values of the regressions that model audit effort and audit fee s as a function of client characteristics. Overcharging may be the natural outgrowth of fixed fee contracting where the auditor uses his superior information to negotiate an arrangement that protects him from the possibility of unforeseen cost overruns that would be difficult to recover from the client. In the second set of hypotheses I explore engage ment characteristics associated with inefficient audits (i.e., overauditing ) It is shown that competition and a clients own knowledge

PAGE 161

161 about the audit process is negatively related, whereas client risk and the auditors private information are positively related to the likelihood of overauditing. These results are consistent with the theoretical predictions. There is also a more subtle contribution of this paper on our understanding of audits. The theory of credence goods argues that an auditors strategic behavior is a function of the auditors knowledge about each specific client, therefore it predicts that strategic behavior and therefore audit production quality and efficiency are client -specific. This reasoning provides a basis for quality differentiation at the engagement -specific level. In contrast to this perspective, traditional audit research argues that audit firm size is an effective mechanism at preserving quality (DeAngelo 1981), an argument that provides a basis for quality differentiation acr oss audit firms, but not within an audit firm The results in the study show that indeed an auditors strategic behavior is primarily driven by variables that proxy for his client -specific information, consistent with theorys prediction. There are limitat ions associated with this study. The data used to test the hypotheses are from one audit firm. If the audit processes are unique to this firm then the results might not generalize to other audit firms. Moreover, the data pertain to audits surrounding a per iod of heightened attention and sweeping regulatory changes on the audit profession and therefore the results may not generalize to other time periods. Additional research may be helpful to resolving these issues. The purpose of this study is to understand how audits are conducted by using the theory of credence goods as the basis for explaining auditors incentives. At its core the theory states that auditors are economic agents that provide a valuable product (service) in the marketplace in exchange for p rofit, therefore it is only rational that they would act like any other economic

PAGE 162

162 agent by maximizing their profits. In the audit markets profit maximization is achieved through adjustments in the audit production process which are possible due to the unobs ervable (i.e., credence) nature of audits. Such adjustments however have important implications on the (actual) level of assurance supplied to investors regarding the quality and the reliability of the financial disclosures and thus o n the efficient alloca tion of resources in the market. One of the goal s of this study is to inform on the credence feature of audits because this feature may be important for policy decisions. Recent regulation like the PCAOB inspection of audit firms can play a powerful role i n mitigating the credence nature of auditing. For example, by disclosing the actual quality of audits the inspection reports may reduce the incentive to underaudit. However, regulation can be a double -edged sword in that too much of it increases the incent ive to overaudit leading to inefficient audits as evidenced by recent controversy surrounding SOX section 404.

PAGE 163

163 APPENDIX A AUDITORS RATIONAL B EHAVIOR Case 1: H -Type Client Define: S1 = (H, FH, CL), S2 = (H, FL, CL), S3 = (H, FH, CH) Net profit under S1: NP1 = FHCL E( m ) E( d ) where E( m d d Net profit under S2: NP2 = FL-CL E( m ) Net profit under S3: NP3 = FHCH Gross margin under S11= FHCL Gross margin under S22 = FL-CL Gross margin under S33 = FH-CH We know that FH-CL>FL-CL because FH>FL 12; Also FH-CL>FHCH because CH>CL, thus 13. Auditors Decision: Choose First between S1 and S2 If NP1>NP2 then choose S1; otherwise choose S2 => [FHCL E( m ) E( d )] > [FL-CL E( m )] => [FHCL E( m ) E( d )] [FL-CL E( m )] simplify terms => [FHCL-E( d )] [FL-CL] Since FHCL 1> FL-CL 2 the final outcome depends on the magnitude of E( d ) such that: Case 1a: If E( d 12 then S2 dominates Case 1b: If E( d 12 then S1 dominates If the client can observe the audit process th en overcharging is not possible. In that case, E( d ) is too large and S2 dominates. If, on the other hand, it is assumed that underauditing is impossible then E( d ) is not large enough to wipe out the gross margin advantage of overcharging and thus S1 dominates. Case 1a: auditor chooses between S2 and S3 Auditor compares NP2 and NP3 [FL-CL-E( m )] [FHCH] 2 3 1 23 then S3 dominates 2 2 3 then outcome depends on the magnitude of E( m ) such that a If E( m 23 then S3 dominates b If E(m 23 then S2 dominates Case 1b: auditor chooses between S1 and S3 Auditor compares NP1 and NP3 [FHCL-E( m ) -E( d )] [FHCH] 13 then FH-CL >FHCH and the outcome depends on the magnitude of E( m ) + E( d ) such that: 1b 1. If [E( m ) + E( d 13 then S3 dominates

PAGE 164

164 1b 2. If [E( m ) + E( d 13 then S1 dominates Case 2: L -Type Client Define: S4 = (L, FH, CH), S5 = (L, FH, CL), S6 = (L, FL, CL) Net profit under S4: NP4 = FHCH E( d ) Net profit under S5: NP5 = FHCL E( d ) Net profit under S6: NP6= FL-CL Gross margin under S44= FHCH Gross margin under S55 = FH-CL Gross margin under S66 = FL-CL We know that FH-CL>FHCH because CH>CL 5> 4; Also FHCL>FL-CL because FH>FL, thus 56. Auditors Decision: Choose First between S4 and S5 Case 2a: assume first that overcharging is not possible In this case S5 is not possible and S4 dominates. Then the auditor chooses between S4 and S6 and compares NP4 and NP6 [FHCH-E( d )] [FL-CL] 4 6 46 then S6 dominates 46 then outcome depends on the magnitude of E( d ) such that a If E(d 46 then S6 dominates b If E(d 46 then S4 dominates Case 2b: assume that overcharging is possible In this case, it is obvious that S5 dominates because NP5=FH-CL E( d ) > FH-CH E( d ) = NP4 Then, the auditor compares NP5 and NP6 [FHCL-E( d )] [FL-CL] 56 then FHCL >FL-CL and the outcome depends on the magnitude of E( d ) such that: 2b 1. If E( d 56 then S6 dominates 2b 2. If E( d 56 then S5 dominates

PAGE 165

165 APPENDIX B THE BENCHMARK AUDIT Equation (B 1) is used to calculate the benchmark audit LHRSi = c0 + c1(LASSETSi) + c2(FRGNi) + c3(COMPLXi) + c4(IAUDi) + c5(NREPTSi) + c6(LEVi) + c7(PUBLICi) + c8(FIRSTYEARi) + c9(ABRi) + c10(ROMMi) + c11(MRELYi) + c12(HRELYi) + c13(MASi) + c14(R TAXi) + c15(COVBOUN Dii (B 1) Where LHRS = t he natural log of total hours LASSETS = the natural log of the total assets FRGN = percent of client total assets located outside the US COMPLX = clients operational complexity as assessed by the auditor. This variable ranges from 1=very simple to 5=very complex. IAUD = equals 1 for firms with an internal audit department ; 0 otherwise NREPTS = the total number of audit reports prepared for the c lient LEV = the ratio of longterm debt to total assets PUBLIC = equals 1 for engagements of public companies; 0 otherwise FIRSTYEAR = equals 1 for first year engagements ; 0 otherwise ABR = equals 1 if the auditor business risk as assessed by the auditor is high; 0 otherwise. ROMM = the risk of material misstatement as assessed by the auditor MRELY = equals 1 if the auditor places a moderate level of reliance on the clients internal control ; 0 otherwise HRELY = equals 1 if the auditor places a high reliance on the clients internal controls; 0 otherwise MAS = the ratio of fees related to management advisory services to audit fees RTAX = the ratio of tax fees over total audit fees COVBOUND = e quals 1 of the client is bound by significant restr ictive debt covenants ; 0 otherwise

PAGE 166

166 Equation (B 1) regresses audit hours on various client characteristics including the natural log of total assets (LASSETS), the percent of total assets located outside the US (FRGN) the clients operational complexity (C OMPLX), whether the client has an internal audit department (IAUD), the number of reports issued by the auditor (NREPTS), the ratio of long -term debt to total assets (LEV), whether the client is a public company (PUBLIC), whether the client is a first year client (FIRSTYEAR), auditor business risk (ABR), the risk of material misstatement (ROMM), whether auditor places moderate (MRELY) or high reliance (HRELY) on the clients internal controls, the ratio of management consulting fees to total fees (MAS), the ratio of tax services fees to total fees ( RTAX), and whether the client is bound by debt covenants (COVBOUND) (see Simunic 1980, OKeefe et al. 1994, Bell et al. 2001, Whisenant et al. 2003, Bedard and Johnstone 2004, Hay et al. 2006, and Bell et al. 2008 ). Table B -1 shows the outcome of this regression. The table shows that the model is well -specified with coefficients and overall model fit consistent with prior literature (see Bell et al. 2001 and Bell et al. 2008). Using the coefficients from T able B 1 I calculate the predicted value of audit hours for each engagement as a function of that engagements characteristics. The predicted value is then used as a proxy for the expected audit effort given client characteristics (PRED).

PAGE 167

167 T able B 1. R egression of hours on client characteristics Variable Coefficient t -stat p value LASSETS 0.251 7.69 0.00 FRGN 0.148 0.60 0.55 COMPLX 0.153 2.88 0.00 IAUD 0.322 2.59 0.01 NREPTS 0.005 1.10 0.27 LEV 0.287 2.15 0.03 PUBLIC 0.477 4.66 0.00 FIRSTYEAR 0.224 2.39 0.02 ABR 0.129 1.68 0.09 ROMM 0.086 0.87 0.38 MRELY 0.135 1.17 0.24 HRELY 0.070 0.40 0.69 MAS 0.089 0.63 0.53 R TAX 0.122 1.61 0.10 COVBOUND 0.081 0.88 0.38 Intercept 2.757 7.29 0.00 N 171 Adjusted R 2 0.69

PAGE 168

168 APPENDIX C DEFINITIONS OF VARIA BLES Table C 1. Definition of variables used in the empirical analyses Variables Definitions MEDTEN Equals 1 for engagements with tenure 3 7 (or 3 8 years); 0 otherwise PROF Ratio of actual audit fee s to standard fee s AFH Ratio of actual audit fees to audit hours IAUD Equals 1 for clients that have an internal audit department; 0 otherwise LPREP Equals 1 if the client is well-prepared for the external audit as assessed by the auditor; 0 otherwise LASSETS The natural log of total assets NREPTS The total number of reports prepared for the client FRGN The percent of a client s total assets located outside the US COMPLX Client's operational complexity as assesse d by the auditor. This variable ranges from 1=very simple to 5=very complex LEV Ratio of long term debt to total assets PUBLIC Equals 1 for public companies; 0 otherwise HRELY Equals 1 if the auditor places a high reliance on a client's internal controls; 0 otherwise MAS Ratio of fees related to ma n agement advisory services to audit fees ABR Equals 1 if the auditor business risk as assessed by the auditor is high; 0 otherwise COVIOL Equals 1 for clients that violated debt covenants during the fiscal year subject to this audit; 0 otherwise PTIME Ratio of hours expended by the partner to total audit hours MTIME Ratio of hours expended by managers to total audit hours ITIME Ratio of hours expended by incharge to total audit hours STIME Ratio of hours expended by staff to total audit hours SHRTEN Equals 1 for engagements with tenure 2 years or less; 0 otherwise LNGTEN Equals 1 for engagements with tenure greater than 7 (or 8) years; 0 otherwise LHRS The natural log of total hours FIRSTYEAR Equals 1 for first year engagements; 0 otherwise ROMM Equals 1 if the risk of material misstatement as assessed by the auditor is high; 0 otherwise MRELY Equals 1 if the auditor places a moderate level of reliance on the client's internal controls ; 0 otherwise RTAX Ratio of tax fees to total audit fees COVBOUND Equals 1 if the client is bound by significant restrictive debt covenants; 0 otherwise

PAGE 169

169 Table C 1. Continued Variables Definitions HPARTNER The natural log of partner hours HMANAGER The natural log of manager hours HINCHARGE The natural log of incharge hours HSTAFF The natural log of staff hours LAF The natural log of audit fee ICE Equals 1 for engagements in the information, communications and entertainment industry; 0 otherwise CIB Equals 1 for engagements in the consumer and industrial business industry ; 0 otherwise UEFF Residuals from regression (4 3) RUFEE Residual s from regression (4 4) divided by total assets UFEE Residuals from regression (4 4) TUFEE Residual s from regression (4 5) HPLAN Planned audit hours HACT Actual audit hours PRED P redicted audit hours (see Appendix B) EXAGG Equals 1 if planned audit hours are greater than predicted; 0 otherwise EXCD Equals to P_EXCD, A_EXCD, and U_EXCD one at a time; 0 otherwise P_EXCD Equals 1 if actual audit hours are greater than planned and engagement profitability (PROF) is greater than the median; 0 otherwise A_EXCD Equals 1 if actual audit hours are greater than planned and hourly audit fees (AFH) are greater than the median; 0 otherwise U_EXCD Equals 1 if actual audit hours are greater than planned and audit fee residuals (UFEE) are positive; 0 otherwise (see Appendix D) EFFIC Efficiency score from the Data Envelopment Analysis ( see Appendix E) SD The number of significant deficiencies discovered by the auditor ROA Ratio of net income to total asset s SUB Equals 1 if the client is a subsidiary of another organization BUS Percent of time spent on understanding the client's business INTRM Equals 1 if the auditor performed i nterim work; 0 otherwise OUTSD Equals 1 if percent of time spent with nonaccounting personnel is high; 0 otherwise DAYSE The number of calendar days lapsed after the client's fiscal year end up to the date of the client's issuance of the earnings press release NAS Equals 1 if the auditor provides other non audit services; 0 otherwise TAX Equals 1 if the auditor provides tax services; 0 otherwise

PAGE 170

170 Table C 1. Continued Variables Definitions RNAS Ratio of non audit fees to audit fees O PIN Equals 1 if the client receives an unqualified opinion; 0 otherwise SCOPE Equals 1 if there was a scope expansion; 0 otherwise

PAGE 171

171 APPENDIX D THE CALCULATION OF U NEXPECTED FEES (UFEE ) Equation (D 1) is used to calculate the unexpected audit fees (UFEE) : LAFi = d0 + d1(LASSETSi) + d2(FRGNi) + d3(COMPLXi) + d4(I A UDi) + d5(NREPTSi) + d6(LEVi) + d7(PUBLICi) + d8(FIRSTYEARi) + d9(ABRi) + d10(ROMMi) + d11(MRELYi) + d12(HRELYi) + d13(MASi) + d14(R TAXi) + d15(COVBOUNDi) + wi (D 1 ) Where LAF = the natural log of total audit fees LASSETS = the natural log of the total assets FRGN = the percent of client total assets located outside the US COMPLX = clients operational complexity as assessed by the auditor. This variable ranges from 1=very simple to 5=very complex. IAUD = e quals 1 for firms with an internal audit department ; 0 otherwise NREPTS = the total number of audit reports prepared for the client LEV = the ratio of longterm debt to total assets PUBLIC = equals 1 for engagements of public companies; 0 otherwise FIRSTYEAR = equals 1 for first year engagements ; 0 otherwi se ABR = equals 1 if the auditor business risk as assessed by the auditor is high; 0 otherwise. ROMM = the risk of material misstatement as assessed by the auditor MRELY = equals 1 if the auditor places a moderate level of reliance on the clients inter nal control ; 0 otherwise HRELY = e quals 1 if the auditor places a high reliance on the clients internal controls; 0 otherwise MAS = the ratio of fees related to management advisory services to total audit fees RTAX = the ratio of tax fees to total audi t fees COVBOUND = equals 1 i f the client is bound by significant restrictive debt covenants ; 0 otherwise

PAGE 172

172 Equation ( D 1 ) regresses audit fees on various client characteristics including the natural log of total assets (LASSETS), the percent of total assets located outside the US, the clients operational complexity (COMPLX), whether the client has an internal audit department (IAUD), the number of reports issued by the auditor (NREPTS), the ratio of longterm debt to total assets (LEV), whether the client is a public company (PUBLIC), whether the client is a firstyear client (FIRSTYEAR), auditor business risk (ABR), the risk of material misstate ment (ROMM), whether auditor places moderate (MRELY) or high reliance (HRELY) on the clients internal controls, the ratio of management consulting fees to total fees (MAS), the ratio of tax services fees to total fees ( RTAX), and whether the client is bound by covenants (COVBOUND) (see Simunic 1980, OKeefe et al. 1994, Bell et al. 2001, Whisenant et al. 2003, Bedard and Johnstone 2004, Hay et al. 2006, and Bell et al. 2008). Table D 1 shows the outcome of this regression. The table shows that the model is well -specified with coefficients and overall model fit consistent with prior literature (see Bell et al. 2001 and Bell et al. 2008). The residuals of this regression are used as a proxy for unexpected audit fees (UFEE).

PAGE 173

173 Table D 1 Regression of audit fees on client characteristics Variable Coefficient t -stat p value LASSETS 0.334 10.09 0.00 FRGN 0.002 0.01 0.99 COMPLX 0.122 1.76 0.08 INTAUD 0.132 0.96 0.33 NREPTS 0.007 1.82 0.07 LEV 0.260 1.82 0.07 PUBLIC 0.505 4.67 0.00 FIRSTYEAR 0.078 0.76 0.44 ABR 0.116 1.40 0.16 ROMM 0.048 0.42 0.67 MRELY 0.001 0.01 0.99 HRELY 0.248 1.13 0.26 MAS 0.008 0.05 0.95 R TAX 0.188 3.26 0.00 COVBOUND 0.049 0.52 0.52 Intercept 7.064 17.48 0.00 N 171 Adjusted R 2 0.69

PAGE 174

174 APPENDIX E AUDIT EFFICIENCY USING DATA ENVELOPMENT ANALYSIS I use Data Envelopment Analysis (DEA) to calculate the relative audit production efficiency. Data Envelopment Analysis (DEA) is a non -parametric technique that is used to estimate the maximum amount of output, or the efficient frontier, using any combinati on of inputs (see Charnes et al. 1978). This method assumes that inputs and outputs are linearly related, however no specific assumptions are made as to the functional form of this relationship. DEA estimates an efficient benchmark for every engagement (us ually referred to as a decision making unit), by comparing the input -output relations of this particular audit engagement with all other audits in the sample. The comparison yields an efficiency score for each engagement which can range from 0 (highly inef ficient) to 1 (very efficient). The specification of the DEA model used in this study closely follows Dopuch et al. (2003). More specifically, seven different client characteristics are defined to be outputs of the production process including the natural log of total assets (LASSETS), the percent of foreign assets (FRGN), client complexity (COMPLX), the number of audit reports issued by the auditor (NREPTS), leverage (LEV), whether a firm is publicly traded (PUBLIC), and the risk of material misstatement ( ROMM). Four different inputs are used in the model including hours worked by partners (HPAR T NER), managers (HMANAGER), in -charge (HINCHARGE) and staff (HSTAFF). The DEA model is applied to 171 observations using an input oriented approach and assuming cons tant returns to scale. An input oriented approach refers to the fact that the linear programming underlying the DEA model minimizes the inputs for a given level of output, whereas the constant returns to scale refers to the assumption that a change in inpu ts leads to a proportional change in outputs. DEA yields client -specific parameters that measure the inefficiency of the audit production for that client. The results show that the mean (median) efficiency score is 0.91 (0.92) suggesting that on

PAGE 175

175 average, t he audits in the sample are very efficient. However, the high efficiency scores could also result from the small sample size. Moreover, the minimum efficiency score is 0.61 suggesting that some audits are highly inefficient.

PAGE 176

176 LIST OF REFERENCES Alger, I. and F. Salanie. 2006. A theory of fraud and overtreatment in experts markets. Journal of Economics and Management Strategy 15 (Winter) : 853881. American Institute of Certified Public Accountants (AICPA). 1978. The commission on auditors r esponsibilities: report, conclusions, and recommendations New York, NY: AICPA. Antle, R., E. Gordon, G. Narayanamoorthy, and L. Zhou. 2006. The joint determination o f audit fees, non audit fees, and abnormal accruals. Review of Quantitative Finance and Account ing 27: 235266. Arens, A. A., R. J. Elder, and M.S. Beasley. 2001. Auditing and A ssurance S ervices: A n Integrated A pproach, 9th Edition Upper Saddle River, NJ : Prentice Hall. Arrunada, B. and C. Paz -Ares. 1997. Mandatory rotation of company auditors: A critical examination. International Review of Law and Economics 17 (March): 31 61. A s thana, S., S. Balsam, and S. Kim. 2004. The Effect of Enron, Andersen, and Sarbanes Oxley on the market for audit services. Working paper, University of Texas San Antonio. Bartels, R., D. G. Fiebig, and A. Van Soest. 2006. Consumers and experts: An econometric analysis of the demand for water heaters. Empirical Economics 31 (May): 369391. Barton, J. 2005. Who cares about auditor reputation? Contemporary Accounting Research 22 (Fall): 549 586. Beck, P. J. and M. Wu. 2006. Learning by doing and audit quality. Contemporary Accounting Research 23 (Spring): 1 30. Bedard, J. C. and K.M. Johnstone. 2004. Earnings manipulation risk, corporate governance risk, and auditors planning and pricing decisions. The Accounting Review 79 (April): 277304. Bell, T.B., W. R. Landsman, and D. A. Shackelford. 2001. Auditors perceived business risk and audit fees: Analysis an d evidence. Journal of Accounting Research 39 (Spring): 35 43. Bell, T. B., R. Doogar, and I. Solomon. 2008. Audit labor usage and fees under business risk auditing. Journal of Accounting Research 46 (September): 729760. Belsky, G. 1996. Watch out: car repair crooks have some new tricks up their grimy sleeves. Money Magazine 25 (June). Bluethgen, R., S. Meyer, and A. Hackethal. 2008. High quality financial advice needed. Working paper, European Business School. Boone, J. P., I. K. Khurana, and K. K. Raman 2008. Audit firm tenure and the equity risk premium. Journal of Accounting, Auditing and Finance 23 (Winter): 115140.

PAGE 177

177 Brownstone, D. and R. Valletta. 2001. The bootstrap and multiple imputations: Harnessing increased computing power for improved statis tical tests. Journal of Economic Perspectives 15 (Autumn): 129 141. Brozovsky, J. A. and F. M Richardson. 1998. The effects of information availability on the benefits accrued from enhancing audit -firm reputation. Accounting, Organizations and Society 23 (November): 767779. Butterworth, S. and K. A. Houghton. 1995. Auditor switching: The pricing of audit services. Journal of Business Finance and Accounting 22 (April): 323344. Cameron, A. C. and P. K. Trivedi. 2005. Microeconometrics methods and applications. New York, NY: Cambridge University Press. Carcello, J. V. and Z. V. Palmrose. 1994. Auditor litigation and modified reporting on bankrupt clients. Journal of Accounting Research 32 (Supplement): 1 30. Carey, P. and R. Simnett. 2006. Audit part ner tenure and audit quality. The Accounting Review 81 (May): 653676. Castarella, J. R., J. Francis, B. Lewis, and P. Walker. 2004. Auditor industry specialization, client bargaining power, and audit pricing. Auditing: A Journal of Practice and Theory 23 (Spring): 123140. Charnes, A., W.W. Cooper, and E. Rhodes. 1978. Measuring the efficiency of decision making units. European Journal of Operational Research 2 (November) : 429444. Darby, M. R. and E. Karni. 1973. Free competition and the optimal amount of fraud. Journal of Law and Economics 16 (April): 67 88. Davis, L. R., B. S. Soo, and G.M. Trompeter. 2007. Auditor tenure and the ability to meet or beat earnings forecasts. Working paper, Michigan Technological University. DeAngelo, L. 1981a. Auditor independence, low balling, and disclosure regulation. Journal of Accounting and Economics 3 (August): 113127. DeAngelo, L. 1981b. Auditor size and audit quality. Journal of Accounting and Ec onomics 3 (December): 183 199. Dechow, P. M., R. G. Sloan, and A. P. Sweeney. 1996. Causes and consequences of earnings manipulation: an analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13 (Spring): 1 36. Defond, M. L. and J. R. Francis. 2005. Audit research after Sarbanes Oxley. Auditing: A Journal of Practice and Theory 24 (Supplement): 5 30. Dopuch, N., M. Gupta, D.A. Simunic, and M.T.Stein. 2003. Production efficiency and the pricing of audit services. Conte mporary Accounting Research 20 (Spring): 47 77.

PAGE 178

178 Doyle, J., W. Ge, and S. McVay. 2007. Accruals quality and internal control over financial reporting. The Accounting Review 82 (October): 11411170. Dulleck, U. and R. Kerschbamer. 2005. Price discrimination via the choice of distribution channels. Working paper, University of Vienna. Dulleck, U. and R. Kerschbamer. 2006. On Doctors, mechanics, and computer specialists: The economics of credence goods. Journal of Economic Literature 44 (March): 5 42. Dullec k, U., and R. Kerschbamer. 2009. Experts vs. discounters: Consumer free riding and experts withholding advice in markets for credence goods. International Journal of Industrial Organization 27 (January): 12 23. Dulleck, U., R. Kerschbamer, and M. Sutter. 2009. The economics of credence goods: On the role of liability, verifiability, reputation and competition. Working paper, Queensland University of Technology. Dye, R. A. 1993. Auditing standards, legal liability, and auditor wealth. Journal of Political E conomy 101 (October): 887914. Emons, W. 1997. Credence goods and fraudulent experts. The RAND Journal of Economics 28 (Spring): 107119. Fong, Y. 2005. When do experts cheat and whom do they target? The RAND Journal of Economics 36 (Spring): 113130. Francis, J. R. 1984. The effect of audit firm size on audit prices: A study of the Australian market. Journal of Accounting and Economics 6 (August): 133151. Francis, J. R. 2004. What do we know about audit quality? The British Accounting Review 36 (Decem ber): 345368. Ghosh, A. and D. Moon. 2005. Auditor tenure and perceptions of audit quality. The Accounting Review 80 (April): 585612. Greene, W. H. 2003. Econometric Analysis, 5th Ed ition New York, NY: Pearson Education Inc. Hackenbrack, K. E. and C .E. Hogan. 2002. Market response to earnings surprises conditional on reasons for an auditor change. Contemporary Accounting Research 19 (Summer): 195223. Hackenbrack, K. E. and C.E. Hogan. 2005. Client retention and engagement -level pricing. Auditing: A Journal of Practice and Theory 24 (May): 7 20. Hackenbrack, K. E. and W.R. Knechel. 1997. An empirical examination of time usage in the audit process. Contemporary Accounting Research 14 (Fall): 481 499. Hay, D. A. and G. S. Liu. 1997. The efficiency of firms: W hat difference d oes competition make? The Economic Journal 107 (May): 597 617.

PAGE 179

179 Hay, D., W. R. Knechel and N. Wong. 2006. Audit fees: A meta analysis of the effect of s upply and demand attributes. Contemporary Accounting Research 23 (1):141191. Hoyle, J. 1978. Mandatory auditor rotation: T he arguments and an alternative. Journal of Accountancy 145 (May): 6978. Hubbard, T.N. 1998. An empirical examination of moral hazard in the vehicle inspection market. The RAND Journal of Economics 29 (Summer): 406426. Johnson, V. E., I. K. Khurana, and J. K. Reynolds. 2002. Audit -firm tenure and the quality of financial reports. Contemporary Accounting Research 19 (Winter): 637660. Khurana, I. and K. Raman. 2004. Litigation risk and the financial reporting credibility of Big 4 vs. N onBig 4 audits: Evidence from Anglo American countries. The Accounting Review 79 (April): 473495. Kaplan, S. E., P. B. Roush, and L. Thorne. 2007. Andersen and the market for lemons in audit reports. Journal of Business Ethics 70 (): 363373. Kim, J., B. Y. Song, and J. L. Tsui. 2008. Auditor quality, tenure, and bank loan pricing. Working paper, Hong Kong Polytechnic University. Klein, B. and K. B. Leffler. 1981. The role of market forces in assuring contractual performance. Journal of Political Economy 89 (August): 615 641. Knechel, W. R., S. E. Salterio, and B. Ballou. 2007. Auditing: A ssurance and Risk, 3rd Edition Toronto, CA : Thompson SouthWestern. Knechel, W. R., P. Rouse, and C. Schelleman. 2009. A modified audit production framework: Evaluating the relative efficiency of audit engagements. The Accounting Review, Forthcoming. Lewis, T. R. and H. Yildirim. 2002. Managing dynamic competition. American Economic Review 92 (September): 779797. Mayhew, B. W. 2001. Auditor reputation building. Journal of Accounting Research 39 (December): 599 617. Mayhew, B. W., J. W. Schatzberg, and G. R. Sevcik. 2001. The effect of accounting uncertainty and auditor reputation on auditor objectivity. Auditing: A Journal of Practice and Theory 20 (Supplement): 4970. McLaren, N. L. 1958. Rotation of auditors. Journal of Accountancy 106 (July): 41 44. Nelson, P. 1970. Information and consumer behavior. Journal of Political Econom y 78 (2): 311329.

PAGE 180

180 Nazareth, A. L. 2007. Remarks before the Council of Institutional Investors, Washington D.C. March 20. Found at http://www.sec.gov/news/speech/2007/spch032007aln.htm Net er, J., W. Wasserman and M. H. Kutner. 1985. Applied linear statistical models. Homewood, IL: Richard D. Irwin. Ogneva, M., K. R. Subramanyam, and K. Raghunandan. 2007. Internal control weakness and cost of equity: E vidence from SOX section 404 disclosure s. The Accounting Review 82 (October): 12551297. OKeefe, T. B., D. A. Simunic, and M. T. Stein. 1994. The production of audit services: evidence from a major public accounting firm. Journal of Accounting Research 32 (Autumn): 241261. Pitchik, C. and A. Schotter. 1987. Honesty in a model of strategic information transmission. American Economic Review 77 (5): 10321036. Pitchik, C. and A. Schotter. 1993. Information transmission in regulated markets. Canadian Journal of Economics 26(4): 815829. Richardson, H. 1999. The credence good problem and the organization of health care markets. Working Paper, Texas A&M University. Rubin, P. H. 2004. Regulation of information and advertising. Working paper, Emory University. Schelleman, C. and W. R. Knechel 2007. Earnings management and the pricing and production of audit services. Working paper, University of Florida. Sengupta, P. 2004. Disclosure timing: determinants of quarterly earnings release dates. Journal of Accounting and Public Policy 23 (November December): 457482. Shapiro, C. 1983. Premiums for high quality products as returns to reputations. The Quarterly Journal of Economics 98 (November): 659680. Shockley, R. A. 1981. Perceptions of auditors independence: An empirical analysis. The Accounti ng Review 56 (October): 785 800. Simunic, D. A. 1980. The pricing of audit services: T heory and evidence. Journal of Accounting Research 18 (Spring): 161190. Simunic, D. A. 1984. Auditing, c onsulting, and a uditor i ndependence. Journal of Accounting Resear ch 22 (2): 679702. Simunic, D. A. and M. T. Stein. 1996. The impact of litigation risk on audit pricing: A review of the economics and the evidence. Auditing: A Journal of Practice and Theory 15 (Supplement): 119134.

PAGE 181

181 Skinner, D. 1994. Why firms voluntar ily disclose bad news? Journal of Accounting Research 32 (Spring): 38 60. Skinner, D. 1997. Earnings disclosure and stockholder lawsuits. Journal of Accounting and Economics 5 (November): 249283. Smieliauskas, W. 1996. Discussion of The impact of litigation risk on audit pricing: A review of the economics and evidence. Auditing: A Journal of Practice and Theory 15 (Supplement): 139144. Stice, J. D. 1991. Using financial and market information to identify pre -engagement factors associated with lawsuits against auditors. The Accounting Review 66 (July): 516533. St. Pierre, K. and J. A. Anderson. 1984. An analysis of the factors associated with lawsuits against public accountants. The Accounting R eview 59 (April): 242263. U.S. Chamber of Commerce. 2006. Auditing: a profession at risk. Washington, D.C : U.S. Chambers of Commerce. U.S. General Accounting Office (GAO). 2003. Public Accounting Firms: Required S tudy on the P otential E ffects of M andatory F irm Rotation. Washington, D.C.: Government Printing Office. Wade, B. 1999. Practical Traveler. New York Times August 29. Watts, R. L. and J. L. Zimmerman. 1983. Agency problems, auditing, and the theory of the firm: S ome evidence. Journal of Law and Economics 26 (October): 613634. Watts, R. L. and J. L. Zimmerman. 1986. Positive Accounting Theory. Englewood Cliffs, NJ: Prentice Hall. Whisenant, S., S. Sankaraguruswamy, and K. Raghunandan. 2003. Evidence on the joint determination of audit and non audit fees. Journal of Accounting Research 41 (September): 721 744. Willenborg, M. 1999. Empirical analysis of the economic demand for auditing in the initial public offerings market. Journal of Accounting Research 37 (Spring): 225238. Wolinsky, A. 1993. Competition in a market for informed experts services. The RAND Journal of Economics 24 (Autumn): 380398. Wolinsky, A. 1995. Competition in markets for credence goods. Journal of Institutional and Theoretical Economics 151 (1): 117 131.

PAGE 182

182 BIOGRAPHICAL SKETCH Monika Causholli received her Bachelor of Science in Finance from Oregon State University in 2000 and began working as an analyst at Enron Corp. in Portland, Oregon and then in Houston, Texas. Monika received her Masters of Science in Accounting from the University of Texas at Arlington in 2004. She then joined the PhD program with a concentration in Accounting at the University of Florida in Gainesville Monika complete d her PhD in August of 2009 and began her academic care er as an A ssistant P rofessor of Accounting at the University of Kentucky in Lexington.