Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Corporate governance, earnings manipulation, and information quality in capital markets Gao, Yanmin 2006

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2006-199787.pdf [ 5.42MB ]
Metadata
JSON: 831-1.0092927.json
JSON-LD: 831-1.0092927-ld.json
RDF/XML (Pretty): 831-1.0092927-rdf.xml
RDF/JSON: 831-1.0092927-rdf.json
Turtle: 831-1.0092927-turtle.txt
N-Triples: 831-1.0092927-rdf-ntriples.txt
Original Record: 831-1.0092927-source.json
Full Text
831-1.0092927-fulltext.txt
Citation
831-1.0092927.ris

Full Text

Corporate Governance, Earnings Manipulation, and Information Quality in Capital Markets by Yanmin Gao B.A., Northeastern University of Finance and Economics, 1992 M.A. , Northeastern University of Finance and Economics, 1995 M.A., University of Windsor, 2000 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF T H E REQUIREMENTS FOR T H E DEGREE OF DOCTOR OF PHILOSOPHY in T H E F A C U L T Y OF G R A D U A T E STUDIES (Business Administration) THE UNIVERSITY OF BRITISH COLUMBIA September 2006 © Yanmin Gao, 2006 ABSTRACT This thesis was inspired by the accounting scandals around the world at the beginning of the 21st century and the reforms in corporate governance law that followed. We use both theoretical and empirical analysis to focus on several aspects of corporate governance mechanisms. The first study focuses on the board of directors. We develop a dynamic agency model with renegotiation to analyze the interaction between the board's monitoring and the manager's earnings manipulation in a firm. The second study broadens the perspective to consider the information in the capital markets with a focus on the role of financial disclosure regulation. It examines how financial disclosure regulation affects the information quality in capital markets, using a strategic investor model in an insider trading setting. The third study investigates these ideas empirically. We construct measures of public, private, and total information quality, and study how they change following the Sarbanes-Oxley Act of 2002 (SOX) in the United States. The key findings of the thesis are as follows. The first study shows that the dynamic incentive contract motivates the manager to work hard on the one hand and to manipulate earnings on the other hand. Even though the board anticipates the earnings manipulation, the board does not unambiguously choose to monitor the manager. The second study demonstrates that the equilibrium market price reflects both the public and a fraction of the private information. While tightened disclosure regulation might improve the quality of public information, the changes in the quality of private information and of total information are ambiguous due to the subtle interplay between public reporting and private information acquisition. The third study provides empirical evidence showing a temporary improvement in the quality of public information following the SOX. The quality of private information decreases and there is no significant change in total information quality. The thesis contributes to the literatures of earnings management and of corporate governance, adding to the growing debate on the costs and benefits of governance reforms. By identifying important indirect effects, we stress the need to consider "substance over form". ii TABLE OF CONTENTS Abstract ii Table of Contents iii List of Tables v List of Figures vi Acknowledgement vii Chapter 1 Introduction 1 Chapter 2 Board Monitoring and Earnings Manipulation 8 2.1 Introduction 8 2.2 Related Literature Review 12 2.3 Dynamic Agency Model with Renegotiation 14 2.4 Characterization of Equilibrium 22 2.5 The Interaction between Corporate Governance and Earnings Manipulation 29 2.6 Window Dressing 33 2.7 Real Earnings Manipulation 37 2.8 The Impact of Corporate Governance Reforms 42 2.9 Concluding Remarks 44 Chapter 3 Corporate Governance and Information Quality 49 3.1 Introduction 49 3.2 The Strategic Investor Model 53 3.3 Characterizing Market Equilibrium and Information Quality 57 3.4 Endogenous Private Signals Quality 62 3.5 The Effect of Corporate Governance on Information Quality 64 3.6 Concluding Remarks : 71 Chapter 4 Empirical Evidence 72 4.1 Introduction 72 4.2 Institutional History and Literature Review 74 4.3 Research Design and Variable Measurement 79 4.4 Empirical Results 89 iii 4.5 Sensitivity Checks 96 4.6 Concluding Remarks 99 Bibliography .....118 Appendix 1: Proofs for Chapter 2 125 Appendix 2: Proofs for Chapter 3 131 iv LIST OF TABLES Table 3.1: Comparison of equilibrium level of information quality 70 Table 4.1: Sample selection and descriptive statistics 104 Table 4.2: Descriptive statistics for variables used in the analyses 105 Table 4.3: Univariate comparisons of information quality variables surrounding SOX 106 Table 4.4: Univariate comparisons of board characteristics variables before and after SOX 107 Table 4.5: Pearson (Spearman) correlations above (below) the diagonal 108 Table 4.6: Ordinary least squares regressions examining the impact of SOX on the information quality in capital markets 109 Table 4.7: The association between board characteristics and information quality 110 Table 4.8: Correlations between time dummies and board interaction variables . . . 112 Table 4.9: Descriptive statistics for information quality variables - Gu measures.. 113 Table 4.10: Univariate comparisons of information quality variables surrounding SOX - Gu measures 114 Table 4.11: Ordinary least squares regressions examining the impact of SOX on the information quality - Gu measures 115 Table 4.12: The association between board characteristics and information quality - Gu measures 116 v LIST OF FIGURES Figure 2.1: Timeline of the agency model 47 Figure 3.1: Timeline of the strategic investor model 55 Figure 3.2: Effect of corporate governance on total information quality 70 Figure 4.1: Timeline of major accounting scandals in 2001 - 2002 102 Figure 4.2: Timeline of major government events 103 Figure 4.3: The event periods 81 vi ACKNOWLEDGEMENTS I am greatly indebted to my co-supervisors Gerald Feltham and Joy Begley for all of their help, insightful and challenging discussions, and valuable advice at many stages during the development of this thesis. I also want to acknowledge my debt to my committee members, Tom Ross and Qiang Cheng, who significantly contributed in many ways. I appreciate the comments from Amin Amershi, Paul Anglin, Sudipta Basu, Sandra Chamberlain, Xia Chen, Christian Hofmann, Chia-Chun Hsieh, John Hughes, Kin Lo, Alfred Wagenhofer, and Shuo Wu. The thesis benefited from discussions with seminar audiences at the University of Alberta, University of British Columbia, University of California - Los Angeles, University of Illinois - Urbana Champaign, University of Toronto, the 2005 C A A A Annual Conference, and the 2006 A A A FARS Midyear Conference. I would also like to thank the C M A B C for their contribution towards the IRRC data. Finally, I am grateful to my parents Peng Gao and Fengshu Tian for being the first to teach me and their loving encouragement ever since. vii CHAPTER 1 Introduction The corporate governance failures at the beginning of the twenty-first century often involved significant accounting scandals. For example, Enron used opaque financial disclosures related to special purpose entities (SPEs) and off-balance-sheet activities; WorldCom, Tyco, Swissair, Vivendi, and Daewoo Group manipulated their financial statements before going bankrupt. The collapse of these prominent companies caused investors to become concerned that similar failures might occur at other companies. Such concerns are bringing greater awareness and changes around the world. In Europe, investor activism and corporate control activities are becoming more common. In Asia, certain Japanese companies have implemented structural changes after the Commercial Code of 2002, including a reduction in board size, more outside directors, and an improvement in disclosure. The Korean government has also embarked on a series of governance reforms on increased financial transparency and disclosure, and board independence. In the United States, the Sarbanes-Oxley Act (SOX) of 2002 created strict rules on financial disclosure and 1 increased monitoring by boards of directors to facilitate the detection of disclosure failures.1 Academics have debated which approach to regulating a corporation is most appropriate and recent governance reforms have led to an increase in research on this important issue, e.g., Holmstrom and Kaplan (2003), Mitchell (2003), Ribstein (2002, 2005) and Romano (2004). Some argue that the scandals demonstrated the need for reform. Others argue that the costs of the changes mandated by SOX are much larger than the benefits, if those benefits actually exist. The same concerns apply to governance reforms elsewhere in the world. Motivated by these accounting scandals and the current debate on governance reforms, this thesis uses both theoretical and empirical analysis to study the effects of corporate governance. In broad terms, corporate governance is based on both internal and external mechanisms. Internal mechanisms are the ways by which individual firms implement control and accountability, including the board of directors, internal control, and ethical codes/cultures. External mechanisms include regulation, legal and bankruptcy systems, external audits, and investor activism. The SOX is comprehensive in terms of reforming both internal and external governance mechanisms. It tightened financial disclosure regulation with the aim of creating an environment with accurate and timely disclosure of more financial information. For example, Section 203 states "The CEO and CFO of each issuer shall prepare a statement to accompany the audit report to certify the 'appropriateness of the financial statements and disclosures . . . and that those financial statements and 1 For a summary of Sarbanes-Oxley Act of 2002, see http://www.aicpa.org/info/sarbanes-oxley-summary.htm. For more background on changes internationally, see Banks (2004). 2 disclosures fairly, present in all material respects, the operations and financial condition of the issuer.' " The maximum penalties for willful and knowing violations of this section are a fine of up to $5 million and/or imprisonment of up to 20 years. In order to enhance the internal monitoring function by the board of directors, Section 301 states "Each member of the (board) audit committee shall be . . . independent." The New York Stock Exchange (NYSE) and NASDAQ have rules ensuring the "independence of directors . . . and to strengthen corporate governance practices". This thesis adds to the corporate governance literature by providing three studies focusing on several different aspects of governance mechanisms. The first study focuses on the board of directors because the board is an important element of internal governance. A board's duties include defining corporate strategy and philosophy, compensating executive management, as well as implementing internal controls. Our model focuses on the last two tasks. We analyze the interaction between the board and the manager in a firm by using a dynamic agency setting. It offers a comprehensive analysis by finding the Renegotiation-Proof Nash Equilibrium. By jointly explaining the compensation contract, board monitoring and earnings manipulation, this study provides insight into how well corporate governance mechanisms limit earnings management and whether increased board monitoring benefits shareholders. We relate the parameters of the model to various kinds of changes in governance measures mandated by governments in many countries and use its predictions to comment on the likely impacts of those reforms. While the first study focuses on the quality of accounting information within a firm, the second study broadens the perspective to consider the information in the 3 capital market with a focus on the role of financial disclosure regulation. Many recent corporate governance reforms around the world seek to increase public disclosure by firm management which in the hope that disclosure change will make the market prices more informative about the future values of firms. The total information impounded in the market price includes both the public information disclosed by firm management and the private information acquired by investors. This study considers the direct and indirect effects of tighter disclosure regulation on information quality through an analysis of the interplay between public reporting and private information acquisition. The study develops a strategic investor model in an insider trading setting based on Kyle (1985). The model includes four types of traders: a firm's manager (i.e., an insider), multiple privately informed rational investors, one privately uninformed rational investor (called the "market maker"), as well as liquidity traders. As an insider, the manager has knowledge about her firm that others do not. The manager can disclose some information to the public and withhold the remainder for the purpose of personal trading. Private investors can incur costs to acquire private information. Corporate governance reforms change the amount of information that a manager can withhold and, thus, change the investor's decision to acquire information. We characterize these behaviors in rational expectations equilibrium. The reforms surrounding SOX provide a natural experiment to investigate the relationship between corporate governance and information quality. The third study investigates these ideas empirically using corporate governance data from IRRC, financial analyst data from I/B/E/S, and accounting data from Compustat. We use 4 earnings forecasts of financial analysts to capture the underlying information sets available to informed investors in capital markets. The qualities of public, private and total information are estimated by employing the measures proposed in Barron, Kim, Lim, and Stevens (1998) and Gu (2005). Both measures derive the properties of unobservable information quality by using observable data on the forecast dispersion, forecast error and the number of forecasts. The key findings of the thesis are as follows. The first study shows that the dynamic incentive contract motivates the manager to work hard on the one hand and to manipulate earnings on the other hand. Earnings manipulation consisting of earnings shifting arises in equilibrium. Interestingly, even though the board anticipates the earnings manipulation, the board does not unambiguously choose to monitor the manager. This paper explores four different settings including accounting and real earnings manipulation, costly window dressing, and costless manipulation. The board prefers to not monitor at all if the manipulation is costless to the manager and has no effect on the underlying economic outcome. However, the board prefers to monitor at least a little if the manipulation is costly to the manager and/or affects the economic outcome. The mixed results imply that, while board monitoring limits manipulation, it may not always be beneficial to shareholders. The second study demonstrates that the equilibrium market price reflects both the public and a fraction of the private information. In our stylized single-period model setting, there is prior uncertainty about the terminal firm value and it can be reduced either by releasing public information or by obtaining private information. The quality of that information is defined to be the extent to which that information 5 reduces prior uncertainty about the terminal value of the firm. In particular, a comprehensive measure, referred to as total information quality, is defined to be the extent to which the market price reveals the terminal firm value. The analysis shows that, while tightened regulation might improve the quality of public information, the changes in the quality of private information and of total information are ambiguous due to the subtle interplay between public reporting and private information acquisition. We show cases where tighter regulation could produce changes which conflict with the aims of SOX. The third study provides empirical evidence showing that the quality of public information improved following the SOX. However, the increase in public information quality is offset by a decrease in private information quality. There is no significant change in total information quality in the year following the passage of the SOX. Further, the improvement of public information does not persist. We fail to find a significant association between board characteristics and information qualities during the sample period. These results are contrary to the aims and intended effects of the reforms. The thesis makes the following contributions. First, it contributes to the earnings management literature. We explore earnings manipulation in a variety of settings, including accounting and real manipulation, costly window dressing and costless manipulation. The analysis captures earnings manipulation consisting of earnings shifting over time, and consumption smoothing without smoothing firm's accounting earnings. 6 It contributes to the governance literature and to the current debate on the net benefits of governance reforms. We show that the extent to which the board engages in monitoring does not depend on earnings manipulation per se, but rather on its real economic effect. Due to the subtle interplay between different sources of information, we show the need to identify and consider indirect adjustments when governance mechanisms change. Many of the reforms introduced as a result of SOX are intended to increase the quality of public information. For this reason, it is noteworthy that, according to our empirical analysis, the improvement in public information quality does not persist. The change may have been temporary because media coverage of the scandals and the failure of Arthur Andersen changed management's awareness of the issue rather than the actual passage of SOX. Writing at the time, Cunningham (2002, p. 19) argues "The changes are more likely to have psychological rather than substantive effects." The results of our studies suggest a need to consider "substance over form." When implementing a new law, as seen in the new wave of worldwide governance reforms at the beginning of this century, proper practices should be strong yet flexible in terms of time, location, industry and country. 7 CHAPTER 2 Board Monitoring and Earnings Manipulation 2.1 Introduction This chapter offers a theoretical perspective to study the interaction between a manager's earnings management behavior and corporate governance mechanisms, with a focus on the activities of a board of directors. We address the following questions. How well do corporate governance mechanisms limit earnings management? Does increased board monitoring benefit shareholders? To study these questions, we develop a dynamic agency model with renegotiation between the board and the manager. The principal is the board representing the firm's shareholders, and the agent is the manager hired to perform productive effort. The main duties of a board include defining corporate strategy and philosophy, compensating executive management, as well as implementing internal controls.2 The model focuses on two of these tasks: compensating and monitoring management. For the task of compensating management, the board writes a performance-based compensation contract. Since the manager's productive effort is not observable and the real outcome is not contractible, we assume that accounting earnings are used as performance measures. 2 See Banks (2004), chapter 2, p.34. 8 The incentive compensation contract is designed to motivate the manager (i.e., agent) to work hard but might also induce the agent to manipulate earnings. If the earnings manipulation causes a transfer of wealth from shareholders to the manager, then board monitoring becomes necessary. In practice, the financial monitoring by the board is often implemented by the board's audit committee. Generally, the audit committee creates internal audit mechanisms, ensures the appropriateness of reporting policies, processes, and infrastructure for producing financial statements, and reviews financial data, such as quarterly and annual financial statements, etc. In addition, an agent who is thinking about manipulating earnings needs to be aware of external governance mechanisms such as the regulatory environment, financial disclosure regulations and civil and criminal liabilities for fraud. A penalty function is used to capture both the internal and external governance mechanisms. Unlike a standard agency model, the principal in our model setting exerts monitoring effort in addition to offering compensation contracts, and the monitoring effort is costly to the principal. Next, both the principal and the agent commit to stay for two periods but, after observing the public report at the end of the first period, the principal might renegotiate the compensation contract with the agent. Allowing for renegotiation is important, since we try to create a setting where the revelation principle does not hold, and hence earnings management can be captured.3 The other distinguishing feature is that, instead of the time-multiplicative preferences frequently The revelation principle (Myerson 1979) states that any equilibrium outcome of any mechanism can be replicated by a truth-telling equilibrium in another mechanism. However, the revelation principle is not a general result since it assumes: (1) communication is not blocked, (2) the form of the contract is not restricted, (3) the principal can commit to use the reports submitted by the agents in any pre-specified manner (Arya et al. 1998, p.10). In the literature, the existing papers study earnings management under settings where the revelation principle does not hold. 9 used in the earlier literature, we use time-additive preferences. The time-multiplicative form implies that the timing of consumption is immaterial, provided it is interest-rate adjusted. A time-additive form is more interesting in a dynamic setting since it captures the importance of the timing and the incentive to smooth consumption. With access to borrowing and lending opportunities in financial markets, the agent does not need to rely on the compensation contract to smooth income. This fact allows us to focus on other reasons for earnings manipulation. The chapter provides the following results. Because of the principal's inability to commit to not renegotiate the contract, along with economic differences of periods, the optimal incentive rate changes dynamically except under knife-edge conditions. This incentive scheme benefits shareholders by inducing the agent to implement productive effort but it also leads the agent to manipulate earnings (i.e., performance measures). In equilibrium, earnings manipulation consists of earnings being shifted to the period with a higher incentive rate in order to increase the agent's compensation. This earnings manipulation does not include earnings smoothing. Consumption smoothing still occurs without smoothing firm's accounting earnings because of borrowing and lending in financial markets. Interestingly, even though the principal anticipates the earnings manipulation, the principal does not unambiguously choose to monitor the agent. The extent to which the board engages in monitoring depends on the real economic effect of earnings manipulation, i.e., whether there are welfare losses for shareholders. This chapter explores four different settings, including accounting earnings manipulation, real earnings manipulation, costly window dressing, and costless manipulation. The 10 board prefers to not monitor at all under the setting where the manipulation is costless to the agent and has no effect on the underlying economic outcome. Not monitoring is preferred in this setting because, in equilibrium, the principal anticipates and prices the amount of manipulation directly through the fixed compensation payment. Using the optimal compensation scheme is better than exerting costly monitoring since monitoring would just cause a deadweight loss to the firm's shareholders. In the settings where the manipulation is costly to the agent (referred to as "window dressing" in the literature4) or affects the underlying economic outcome (i.e., real earnings manipulation) or both, the principal prefers to monitor at least a little because the manipulation is costly to the firm's shareholders, either directly or indirectly. The mixed results imply that, while board monitoring restricts earnings manipulation, it does not always benefit shareholders. Recent worldwide corporate governance reforms have the effect of changing certain parameters in the model. Hence, the model offers several predictions concerning the impact of various kinds of changes in governance measures mandated by governments in many countries. The remainder of the chapter proceeds as follows. The next section reviews the related literature. Section 2.3 develops a two-period agency model with renegotiation between the board and the manager, and outlines the event structure. Section 2.4 characterizes the equilibrium, focusing on the optimal compensation contract, on induced earnings manipulation, and on optimal board monitoring. Section 2.5 provides a complete analysis of the interaction between corporate governance and earnings manipulation. Section 2.6 and Section 2.7 extend the basic model to 4 Feltham and Xie (1994) define window dressing as: "performance measures, including accounting earnings, are often subject to manipulation in the sense that the agent takes actions that improve his performance measure but contribute little or nothing to the principal's gross payoff'. 11 window dressing and real earnings manipulation settings. The discussion of the model's implications for recent governance reforms is offered in Section 2.8. The last section contains concluding remarks. 2.2 Related Literature Review In the accounting literature, there is no consensus on the definition of earnings management. One representative definition comes from Healy and Wahlen (1999, p.368): "Earnings management occurs when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on the reported accounting numbers".5 Extensive empirical evidence of earnings management is well documented, e.g., reviews of Healy and Wahlen (1999), Dechow and Skinner (2000), Fields et al. (2001), and Schipper and Vincent (2003). In contrast to ample empirical research, there is little theoretical modeling on earnings management. A troublesome issue for earnings management modeling is the revelation principle (RP). The existing models study earnings management by violating RP. The internal and external sources of earnings management are studied in several papers: Dye (1988), Evans and Sridhar (1996), Demski (1998), Arya et al. (1998), and Kirschenheiter and Melumad (2002). The "internal" source is related to optimal contracting and minimizing the expected cost of getting a manager to adopt the shareholders' preferred action. The "external" source is related to the capital market and is based on current shareholders' desire to 5 The other representative definition is by Schipper (1989, p.92): "...a purposeful intervention in the external financial reporting process, with the intent of obtaining some private gain (as opposed to, say, merely facilitating the neutral operation of the process)...." 12 influence perceptions of the firm's value. A popular form of earnings management, i.e., income smoothing, is well studied. Several papers show different motives to smooth income or earnings, including consumption smoothing incentive in Lambert (1984) and Suh (1990), career concerns in Fudenberg and Tirole (1995), leverage concerns in Trueman and Titman (1988), and stock price concerns in Goel and Thakor (2003). Earnings management can be undesirable or desirable to shareholders, depending on whether earnings management is opportunistic or informative of the firm's economic value. Focusing on opportunistic earnings management, we extend the earlier literature to establish a link between earnings management and corporate governance.6 In broad terms, corporate governance is based on both internal and external mechanisms. Internal mechanisms are the ways by which individual firms implement control and accountability, including the board of directors, internal control, and ethical codes/cultures. External mechanisms include regulation, legal and bankruptcy systems, external audits, and investor activism. Our work is related to several recent papers, e.g., Ewert and Wagenhofer (2005), Christensen, Frimor and Sabac (2004), Liang (2004), and Goldman and Slezak (2003). While their work focuses on the effect of audit control, accounting standards or regulation on earnings management, we particularly focus on the roles of board of directors. The board of directors is an important element of internal governance, attracting considerable 6 According to the corporate governance literature, there are several definitions of corporate governance. In broad terms, Banks (2004, p.3) defines corporate governance as "the structure and function of a corporation in relation to its stake-holders generally, and its shareholders specifically"'. Focusing more narrowly on investor value, Shleifer and Vishny (1997, p.737) define corporate governance as "the way in which suppliers of finance to corporations assure themselves of getting a return on their investment"'. 13 attention from practitioners and academics. The other distinguishing feature is that, instead of time-multiplicative preferences that are frequently used in the literature, time-additive preferences are used in our model. Therefore this chapter captures earnings shifting, similar to Christensen, Frimor and Sabac (2004) and Liang (2004), and also captures consumption smoothing over time. Furthermore, the model endogenously determines earnings manipulation, board monitoring, and compensation contracts. This facilitates a comprehensive analysis of their interplay and the development of insight from that interplay. Finally, we explore a variety of settings of earnings manipulation including accounting and real earnings manipulation, costly window dressing and costless manipulation. 2.3 Dynamic Agency Model with Renegotiation We consider a two-period agency model where the board of directors (later referred to as the "principal"), acting on behalf of the firm's shareholders, hires the manager (later referred to as the "agent") to provide productive effort in each of the two periods. Both the principal and agent commit to stay for two periods. The principal offers the agent a compensation contract which is subject to renegotiation at the start of the second period. Unlike standard agency models, the principal also exerts effort to monitor the agent, denoted as e (e > 0). The principal commits to the specified level of monitoring ex ante, and e is observable by the agent. That is, this is a game in which the principal moves first. 7 See Banks (2004), Hermalin and Weisbach (2003), Fields and Keys (2003). 14 The agent exerts product ive effort at in per iod r. The expected outcome f r om a, is E[xt \at] = pt, t = 1,2, where y (y > 0 ) measures the agent's product iv i ty . T h e pr inc ipa l can not observe the agent's product ive effort, and the real outcome is not contract ible. In pract ice, account ing earnings are often used as per formance measures in management compensat ion contracts. The agent has abi l i ty to manipulate earnings. The ex is t ing literature dist inguishes between account ing and real earnings management. In general , "account ing earnings management" changes reported earnings wi thout changing the under ly ing cash f l ows . " R e a l earnings management" changes operat ing decis ions so that there are changes in both the under ly ing cash f lows and the reported earnings (Z i v , 1998). In the bas ic mode l , we consider account ing earnings manipu la t ion that is costless to the agent (i.e., the agent does not exert personal ly cost ly manipu la t ive effort). The manipu la t ion is achieved s imp ly by changing account ing po l i c ies , such as depreciat ion methods, bad debt a l lowance rates, or t im ing sales discount. The per formance measures are assumed to be represented by yx = x{ + D and y2 =x2-RD, where D is the amount o f earnings manipu la t ion, R = l + i is the r iskless one-per iod return and I is the r iskless one-per iod interest rate. R is used to account for the t ime value o f money where R = 1 i f shi f t ing pure account ing numbers (e.g., through accruals manipulat ion) , and R > 1 i f shi f t ing real dol lars (e.g., through earl ier sales discount) . A pos i t ive (or negative) D indicates upward (or downward) manipu la t ion in the first per iod. T h e speci f icat ion of the performance measures impl ies that the amount o f earnings manipu la t ion in the first per iod is reversed in the second per iod. 15 T h e agent's incentive to manipulate earnings is l i m i t e d by various corporate governance mechanisms. B o a r d m o n i t o r i n g , often implemented b y the audit committee, plays a signif icant role i n discouraging and detecting earnings manipulat ion . In addit ion, the overal l external regulatory environment also helps deter an agent f r o m manipulat ing earnings, e.g., through f inancia l disclosure regulations, as w e l l as c i v i l and c r i m i n a l l iabi l i t ies for fraud. T o represent the effect o f these governance mechanisms, w e assume manipulat ion results i n a penalty o f P-/2(p + e)D2, where p (p>0) represents the tightness o f external regulatory environment. T h e amount o f the penalty is increasing i n the absolute value o f earnings manipulat ion , the tightness o f regulat ion, and the intensity o f board m o n i t o r i n g . In practice, the penalty that the agent faces depends o n the legal system. Shareholders, or their representative, the board, may sue the agent for her manipulat ing earnings, and the penalty to be p a i d by the agent to shareholders is decided b y a court. A court is l i k e l y to impose a larger penalty i f the leve l o f earnings m a n i p u l a t i o n is larger. S i m i l a r l y , tighter external regulat ion tends to result i n more severe punishment. T h e size o f the penalty m a y also depend o n the intensity o f internal board m o n i t o r i n g because more intense m o n i t o r i n g is more l i k e l y to produce Q the k i n d o f c o n v i n c i n g evidence that leads a court to impose a larger penalty. F o r mathematical tractabil i ty, a d y n a m i c L E N m o d e l , i.e., l inear contract, exponential ut i l i ty , and n o r m a l l y distributed performance measures, is adopted. 9 T h e 8 While the penalty will only be imposed if manipulation is detected and prosecution is successful, we simplify the analysis by ignoring the uncertainty of the penalty. 9 It is well-known that LEN models provide simplification of mathematical representations, and are widely used in the literature. Recently, several papers employ the LEN framework in dynamic settings, such as, Christensen et al. (2005), Feltham et al. (2004), Dutta and Reichelstein (2003). 16 L E N model exogenously specifies the compensation contract as a linear function of the performance measures. Further, based on the concept of renegotiation-proofness of Fudenberg and Tirole (1990), we solve for a linear renegotiation-proof contract. The following describes the timeline of the model which highlights the sequence of events within each period, and provides important assumptions with respect to the performance measures, the agent's preferences, and the principal's preferences. Timeline of the Model (see Figure 2.1): Stage 1: At date zero, the principal designs an initial compensation contract c' (f ,v[,vl2) and offers it to the agent, where / ' is the initial fixed payment paid at the end of the first period, and v,', v'2 are the incentive rates for the first and second period respectively. At the same time, the principal commits to the level of monitoring e. Stage 2: At the start of the first period, the agent chooses her productive effort a{, which affects the first-period performance measure and outcome to the principal. Stage 3: At the end of the first period, after privately observing the outcome the agent decides whether to manipulate earnings. If so, she chooses the amount of earnings manipulation D, and then issues a public report y , . Stage 4: Since the principal can't observe the agent's productive effort and the real outcome, the principal uses the public report y, to pay the agent her first-period compensation cox = f + v[yl. 17 Stage 5: After the agent has received her compensation, and paid her first-period personal productive effort cost k(a{) as well as the penalty if she manipulates earnings, the agent chooses her first-period consumption c, . The agent is allowed to borrow and lend in financial markets. Stage 6: At the start of the second period, based on the information released at the end of the first period, the principal may offer a renegotiated compensation contract cr (Af ,v2), where Af r represents a change in the fixed payment, and vr2 is the revised incentive rate for the second period. The agent can accept or refuse the new contract. If she refuses, then the initial contract stays in effect. Stage 7: The agent chooses her second-period productive effort a2, which affects the second-period performance measure and outcome to the principal. Stage 8: At the end of the second period, the outcome x2 is realized and privately observed by the agent. The amount of earnings manipulation D in the first period is reversed, and the agent issues the second-period public report y2. Stage 9: Based on y2 and the possible renegotiated contract, the principal pays the agent second-period compensation a>2 = Af' + vr2y2. Stage 10: After receiving her second-period compensation and paying her second-period personal productive effort cost k(a2), the agent consumes her remaining bank balance. 18 Performance Measures The contractible performance measures yt, t = 1,2, reported at the end of each period, are assumed to take the following forms: yl=xl+D, x{ = yal + ex, (2.1) y2-x2-RD, x2=ya2+e2, (2.2) where er, t = 1,2, are normally distributed noise terms with zero mean and covariance matrix erf paxa2 paxo2 o\ with pe (-!,+!) The performance measures have the following features. The effect of the productive effort a, on the expected performance measure yt is linear. The noise term et is additive and, hence, the covariance matrix of performance measures is independent of the productive effort. The two noise terms are potentially correlated with the correlation coefficient p. Further, the conditional expected performance measures given the productive effort and earnings manipulation are as follows. E[yl\al,D] = yal+D, Var[y, | ax,D] = a], (2.3) E[y 2 |a 1 ,a 2 ,D,y , ] = ya2 -RD + p(<J2l<7x)(yx -yax -D) , (2.4) Var[y2 \ al,a2,D,yx] = (\ -p2)cr22. (2.5) Note that the posterior mean for the second-period performance measure y2 depends on the first-period performance measure y,, the productive effort ax, a2, as well as the earnings manipulation D. The posterior variance of y2 is independent of the 19 specific values of y,, a{, a2 and D, and lower than its prior variance unless the inter-period correlation equals zero, i.e., p = 0 . Agent's Preferences The agent is assumed to be risk and effort averse. Her preferences for the contract period are represented by the following time-additive exponential utility function: M " exp(-rc,), r=l where r>0 is the agent's risk aversion parameter, and ct represents consumption net of productive effort cost and the penalty. For simplicity, the market discount factor P = \IR is used as the agent's discount factor. Let k(at) be the agent's productive effort cost, where k(at) = /2af is a quadratic function as commonly assumed in the agency literature. The agent can borrow and lend in financial markets. Let \ be the amount of borrowing or lending. After the agent has received her compensation and paid her personal effort cost, as well as the penalty at the end of the first period, she has a cumulative-consumption bank balance:10 Bx=a>x- k(a{ )-P = c ] + \ . Similarly, at the end of the second period, the cumulative-consumption bank balance is B2 = R\ +co2 -k(a2). The agent is assumed to choose her consumption before accepting the contract at date zero. 2 0 The agent is assumed to consume her bank balance at the end of the contract, i.e., c2 = B2, and, without loss of generality, zero reservation consumption is assumed, i.e., c° = 0 . One feature of the agent's preferences is the time-additive form of the utility function. Mathematical complications explain why time-additive preferences have not been used as frequently as traditional time-multiplicative preferences even if the time-additive utility function is more realistic. It induces the phenomenon of consumption smoothing in dynamic settings and introduces borrowing and lending opportunities in financial markets which allow us to capture earnings management for reasons other than income smoothing. Principal's Preferences The board of directors, i.e., principal, represents the firm's well-diversified shareholders and is assumed to be risk neutral. The market discount rate /? is applied to the principal as it is to the agent.11 A key feature is that the principal takes costly monitoring, and the monitoring cost is represented by M(e) = /2mee2, where me > 0 is the parameter of marginal monitoring cost, stated in date zero dollars. Let X = x, + fk2 be the total gross payoff to the principal, and W = (Ox + fi(Q2 be the total compensation payment to the agent, both are stated in period one dollars. Thus, at the contract date, the principal's preferences are represented by: EU0P =/]E0[X]-/3E0[W-P]-M(e). " A different discount rate can be used, but it is not essential for the subsequent results. 21 That is, the principal's expected utility is the expected gross payoff minus expected compensation payment (adjusted by the penalty) and monitoring cost as well. Notice that the model assumes that the board, representing the firm's shareholders, may sue the agent, and that any penalty paid by the agent goes to the shareholders. In some cases, penalties may be paid to a court or other third party. 2.4 Characterization of Equilibrium As is commonly done in dynamic models, the equilibrium is characterized by backward induction. We discuss the concept of renegotiation-proof contracts first and then characterize the equilibrium. According to Fudenberg and Tirole (1990), a contract is renegotiation-proof if the principal will not choose to alter it at the renegotiation stage. Given the L E N (e.g., linear contract, exponential utility, and normally distributed performance measures) assumption in the model setting, we show there is no loss of generality in restricting attention to linear renegotiation-proof contracts. Renegotiation-proof Contract The initiablinear compensation contract is c'(f,v\,v2). At the start of t = 2, i.e., the renegotiation stage, / 'and v( are irrevocable and the principal offers a renegotiated linear compensation contract cr(Afr,vr2) with a fixed payment Af and a revised incentive rate for the second period v2. If the agent accepts the new contract, then her second-period compensation will be Q)2(cr) = Af + vr2y2. The 22 L E N assumption implies that, conditional on the information at the end of t = 1, her expected net compensation (i.e., the compensation net of productive effort cost) is £,[o)2 -k(a 2)] = Af + vr2[pi2-RD + p«J21a{)(y, -ja{-D)]-/2a2. (2.6) and that the risk premium for the net compensation based on the posterior variance of y2 is RPy=y2r[v2f(\-p2)(T22. (2.7) Therefore, the agent's certainty equivalent with respect to the net compensation, conditional on the information at the end of the first period, is CEx(cr) = pE,[a)2-k(a2)]- f3RPx. (2.8) The agent is assumed to consume her bank balance at the end of t = 2, i.e., her second-period consumption c2=B2, and B2= R(Bl-cl) +co2-k(a2). Thus, the agent's utility maximization decision with respect to her second-period productive effort a2 is arg max Ex [-/? exp(-rc2)] (2.9) = arg max{5, - c, + C£ , } " 2 The first-order condition for the problem yields a2 - yv2. This indicates that the agent optimally chooses her second-period productive effort based on the revised incentive rate v2. Given the initial contract, the public report y,, and the principal's conjecture of the agent's first-period productive effort a, and earnings manipulation D, the principal chooses A / r and v2 so as to maximize his second-period expected utility, subject to acceptance by the agent. Formally, the principal solves 23 maxEUf (cr) = /?{p2 - Ex\co2(cr)]}, &fy2 (2.10a) subject to CEx(cr)>CEx(cl), (2.10b) (2.10c) Solving the principal's decision problem yields (see the Appendix), r2 + r(l-p2)(722  ' and AT = yiiia,]2 -[ai2]2} + /2r(l-p2)a2{[vr2]2 -[v2]2} - {vr2[yar2 -RD + p(a21 cr, )(y, - jfi, - D)] - v2\ya2 -RD + p(a21 cr, )(y, - ft - D)]}. The solution indicates that A / r is a linear function of yl,dl,D. Furthermore, if the initial contract specifies v'2 = v2, then Af r=0. Thus, the linear renegotiation-proof contract is crp (f, v,, v2). It is worth emphasizing that this result is not true in general. The paper by Feltham, Indjejikian and Nanda (2004) shows that, even though the initial contract is linear, the renegotiation-proof contract is a quadratic function of the first-period performance measure. The difference occurs because the second-period payoff parameter is random and is correlated with the first-period performance measure in their model setting. Equilibrium Now focusing on the renegotiation-proof contract crp(f,vx,vr2), backward induction works in the following way. Since the agent is assumed to consume all the remaining bank balance at the end of the second period (c 2 = B2), the second-period 24 consumption c 2 is not a decision variable. We start by characterizing the agent's second-period productive effort choice a2, followed by the principal's renegotiated incentive rate vr2 at t = 2. Then we move back to t = 1 to characterize the agent's first-period consumption decision c,, earnings manipulation D, and first-period productive effort a,. Finally, we end with the principal's compensation contract terms ( / , v,) and his monitoring choice e at t - 0. The discussion of renegotiation-proof contracts in the previous subsection shows that the agent's optimal second-period productive effort is a2 = yv2, and the second-l y 2 period incentive rate is v2 = — — - . Increasing the incentive v2 would f + r(\-p2)(j22 induce the agent to exert more productive effort a2, but would also impose higher risk as well as more effort cost on the agent. So the principal determines an optimal v2 by trading off between the benefit of the agent's second-period productive effort and the cost of inducing it (implied by the second-period risk premium and effort cost). At t = 1, the agent chooses consumption c, so as to maximize her expected utility over the two-period consumption, conditional on the information available at this time, and anticipating the equilibrium solutions in the second period. Borrowing and lending opportunities are available to the agent. Given c2 = B2 = RXX +co2- k(a2) and Xl =Bx-c{, the agent's decision problem with respect to c, is max{ - exp(-rc,) - pEy [exp(-rc2)]} = {-exp(-rc,) - fiexp[-r(/?(B, - c,) + RCEX)]}. (2.11) 25 The first-order condition for this problem is characterized by c, = R(B{ - cx + CEX). Rearranging it gives c, = A, (5, + CEi) with the annuity amortization factor A, = (1 + /J)"1. Thus, it is optimal for the agent to consume an amount equal to an annuity based on the sum of first-period bank balance and second-period certainty equivalent. This suggests that the agent smoothes consumption over time by borrowing and lending. Substituting the optimal c, into Equation (2.11) produces {- exp(-rc,) - pE{ [exp(-rc2)]} = (! + /?){-expt-M, (Bx + CEX)]}, (2.12) where rA, is the effective risk aversion parameter, which is the original consumption risk aversion coefficient adjusted by the annuity factor. The agent chooses the amount of earnings manipulation D after the outcome x{ has been realized. At this point of time, x{ is observed only by the agent, and the agent anticipates her first-period compensation Oil(D) = f + v\(xi + D) > m e penalty P = y2(p + e)D2, as well as the second-period certainty equivalent CE[(D) = pvr2\ya\ -RD + p(a21a{ ) (JC, - ) « , ) ] - plc(a2) - fiR^. (2.13) Based on Equation (2.12), the agent's decision problem with respect to D is reduced to the following expression: max{a);(D)-k(a1)-P(D) + CE^(D)}. (2.14) Then the first-order condition for the maximization problem is given by v, -v2 -(p + e)D = 0. (2.15) It yields D = —-—(v, - v[). In equilibrium, the agent trades off the benefit obtained p + e through earnings manipulation (resulting in more compensation) and the expected 26 cost incurred due to the penalty. The optimal amount of manipulation is driven by the difference between the incentive rates of the two periods, and is restricted by external regulation and internal board monitoring. For a given compensation contract, the agent chooses her first-period productive effort fl[ at (=1 to maximize her expected utility over the two periods. This is equivalent to maximizing her total certainty equivalent CE0 under the L E N assumption, maxCE0 = 0Eo[W - K(ax,ar2)-P]-0RPo, (2.16) "i where K(ax,a2) = k(al) + /3k{a[) represents the total productive effort cost, and RP0 = y2{rAl[vl + p(<72 /ax)p\2]2a2 + rfi[v2]2(l- p2)a2} is the total risk premium for the compensation, which takes account of the effect of inter-temporal correlation captured by p . The first-order condition for the problem implies ax = yvx, i.e., the agent chooses an optimal first-period productive effort based on the incentive rate of this period. Finally, we consider the compensation terms (f,vx) and monitoring choice e by the principal at t = 0. Anticipating the optimal solutions in the next two periods, the principal chooses (f,vx,e) to maximize his expected utility at date zero, i.e., the expected gross payoff minus expected compensation payment (adjusted by the penalty) and monitoring cost, subject to the agent's participation and incentive constraints: max EU0P = /3EQ[X]- 0Eo[W - P]- M(e), (2.17a) subject to CE0 =/3E0[W - K(ax,a2)-P]- /3RP0 > c°, (2.17b) 27 ax=yvx, (2.17c) a2=y>r2. (2.17d) The principal chooses the fixed wage / to be just sufficient to induce the agent to accept the compensation contract, i.e., the participation constraint is binding. This implies / = K(ax,a2) + RP0 -E0[OJX + j3co2 -P], where a>x-vxyl is the variable first-period wage. Using the binding participation constraint, we re-express the principal's decision problem as the following unconstrained optimization problem: maxJT£ = fiEo[X]-0[K(ax,ar2) + RPo]-M(e). (2.18) v,,e Therefore, ex ante, the principal chooses vx and e so as to maximize his expected gross payoff net of the agent's total productive effort cost and total consumption risk premium, minus his monitoring cost as well. Then, the first derivatives of rig with respect to v, and e are: anp _ dE0[X] dal ' dK(ax,ar2) dax | dRP0 ^ dvx dax 3v, da{ dvt oV, .= P\f -ir\ + rAx CJ2(vx+p{cJ2 /cr,)^)]}, (2.19a) anp -2 -=-m e e<0. (2.19b) de 1 Setting the first derivative with respect to v, equal to zero yields v, = — y - A,[j3v2 ], where Q = y2 + rAxo~2, X = -^rAlpalO'2. The optimal first-period incentive rate v, is chosen by balancing the benefit provided by the first-period productive effort and the cost of inducing this effort, as determined by the first-period effort cost and risk 28 premium. A key feature of the solution for v, is that it depends on the variance of the first-period performance measure y, and is adjusted for the inter-period correlation p. Furthermore, the first derivative with respect to e indicates that the level of monitoring takes a corner solution, e = 0. That is, the principal optimally chooses to not monitor the agent. This result is justified by Equation (2.18) which shows that ex ante, any resource spent on monitoring creates only a deadweight loss to the principal. 2.5 The Interaction between Corporate Governance and Earnings Manipulation This section uses the equilibrium derived in the previous section to provide a complete analysis about the interplay of the compensation contract, earnings manipulation, and board monitoring. Proposition 2.1 The principal's compensation contract choices are characterized as follows: (2.1a) The first-period incentive rate v, = -^y2 - A[p\2], where Q = y2 + rAxax, ^ = ^ r A x p o x o 2 , A, = (1 + / j ) - 1 . r2 (2.1b) The second-period incentive rate v2 = — —- . y2 + r(\-p2)cy2 (2.1c) The fixed payment / = K(ax,a2) + RP0 - E0[cox + j5a>2 -P]. 2 9 Due to renegotiation and inter-temporal correlation, the principal changes the incentive rate over time. Proposition (2.1b) demonstrates that the second-period incentive rate v2 is based only on the posterior variance of the second-period performance measure because, at the time of renegotiation, only the second-period risk premium matters. However, the principal chooses the first-period incentive rate vx based on the total compensation risk, and hence the total risk premium paid to the agent. Specifically, Proposition (2.1a) shows that v, reflects not only the prior variance, but also the adjustment for the inter-period correlation p captured by the term Xp\\. This adjustment can be positive (or negative), depending on the sign of P-Proposition 2.2 The agent's choices with respect to her productive effort, consumption and earnings manipulation, given incentive rate v,, v2, are characterized by (2.2a) The first-period productive effort a, =}Vl. (2.2b) The second-period productive effort a2 = yv2. (2.2c) The first-period consumption c, = A, (5, + CEX). (2.2d) The earnings manipulation D - — (v, -v2). P Consistent with a common result in the agency literature, the agent's productive effort depends on the incentive provided by the principal shown in Proposition (2.2a) and (2.2b). The incentive contract motivates the agent to work hard, and the higher incentive rates, the more productive effort. However, this incentive contract also induces the agent to manipulate earnings. This effect is highlighted by Proposition 3 0 (2.2d). The agent shifts earnings between the two periods, and an upward (or a downward) shift occurs when v, > v2 (or v, < v2). That is, the agent always shifts earnings to the period with a higher incentive rate so as to maximize her total compensation over the two periods. Note that the agent has a time-additive utility function, and Proposition (2.2c) implies that she smoothes consumption over time. Interestingly, the consumption smoothing is not done by smoothing accounting earnings since earnings shifting arises as an equilibrium phenomenon. Rather, the agent achieves her consumption smoothing by borrowing and lending in financial markets. Consider a special case where (5 -1, i.e., there is a zero interest rate, and the noise is identical in the two periods cr, = <r2 = a, thus Corollary 2.1: i. If -\<p< Y2, then D > 0 . ii. \fy^<p<\, then D < 0. iii. If P~Yi o r P ~^ 0 0 ' t n e n D = 0. Proof: See the Appendix. This corollary establishes that, if the inter-period correlation is negative or positive but low, -1 < p < ' t n e n a relatively strong first-period incentive is necessary in order to induce efficient productive effort. As a result, v, > v2 and the higher first-period incentive leads the agent to shift earnings to the first period, i.e., D > 0 . If the correlation is positive and sufficiently high, p > , then the second-period incentive 31 becomes relatively strong v, < vr2, consequently, earnings are shifted to the second period and hence D < 0. This corollary also identifies the knife-edge conditions under which no earnings manipulation exists in equilibrium i.e., D = 0. When P~Yl' ^ e P r m c ^P a ^ s e t s e c l u a l incentives for both periods vl=v2r, therefore, the agent has no desire to shift earnings. Under an extreme condition p —> °°, the threat of infinitely tight regulation deters the agent from manipulating earnings. Note that this earnings manipulation, i.e., earning shifting, is mainly driven by contract renegotiation. From an ex post perspective, renegotiation is beneficial to the principal. However, if the principal can fully commit to a contract with equal incentive rates for both periods, where no earnings shifting is induced, then the principal is generally better off ex ante. On the other hand, according to Christensen and Feltham (2005, Chapter 28, p.513), "while renegotiation is generally ex ante inefficient, it may be difficult (or impossible) for the principal and the agent to commit themselves not to engage in ex post mutually beneficial renegotiation.'" Proposition 2.3 The principal's monitoring choice is e = 0 . Interestingly, the principal optimally allows the agent to manipulate earnings without any monitoring. The reason for no monitoring is that, in equilibrium, the principal anticipates the amount of earnings manipulation and prices it directly through the fixed compensation payment. This is justified by Proposition (2.1c). Re-expressing this proposition yields 32 f = K(ai,ar2) + RP0-y2{vf+/3[vr2)2}-y2pD2. (2.20) Using the optimal compensation scheme is better than exerting costly monitoring since monitoring would just cause a deadweight loss to the firm's shareholders. 2.6 Window Dressing In the basic model, the accounting earnings manipulation is costless to the agent, i.e., the manipulation is achieved without personally costly manipulative effort. However, in practice, some types of accounting earnings manipulation are costly to the agent. For example, in order to shift next period's earnings to the current period, the agent may have to take significant time and effort to talk to some friendly customers at the end of the current period in order to arrange for shipping goods. This is referred to as costly window dressing. In the window dressing setting, in addition to productive effort, the agent exerts manipulative effort. The manipulative effort is costly, and the cost (later referred to as "window dressing cost") is assumed to have a quadratic functional form, i.e., k(Dw) = y2kw[Dw]2 with Jfew>0, stated in period t = l dollars. The window dressing cost reduces the agent's wealth available for consumption. Formally, the agent's utility function becomes uaw exp( -<) , . (2.21) where c,w is the consumption net of both productive effort cost and window dressing cost. Consequently, when making an earnings manipulation decision, the agent maximizes 33 max{co; (Dw) - k « ) - k{Dw) - P(Dw) + CEXX(Dw)}. (2.22) D"' Comparing the agent's decision problem under costless and costly manipulation setting, i.e., (2.14) versus (2.22), shows the impact of window dressing. Now the first-order condition with respect to Dw is daj?(Dw) dk(Dw) dP(Dw) {dCE;(Dw)=Q dDw dDw dDw dDw This equation indicates that, in equilibrium, the agent has to trade off the benefit of window dressing and the associated cost, as determined by both the penalty and the manipulative effort cost. The first-order condition leads to the following proposition. Proposition 2.4 In the window dressing setting, the agent's earnings manipulation choice, given incentive rates v,"', v[ and monitoring ew, is characterized by DW =—-—-(vr-vo. kw + p + e Proof: See the Appendix. Observe that the optimal earnings manipulation Dw depends on the manipulative effort cost. Holding the other parameters constant, a higher window dressing cost parameter kw results in a lower level of manipulation. Furthermore, the agent's total certainty equivalent is adjusted downward by &(Dw),ihatis, CE; = /3E0[WW -K(axw,ar2)-k(Dw)-Pw]~/3RP0W. (2.24) The adjusted total certainty equivalent alters the participation constraint (PCW) in the principal's decision problem at t = 0: 34 CE? = J3E0[Ww -K(a?,a2)-k(Dw)-PW]-J3RP0W >C°, (PCW) and, as a result, the principal's unconstrained optimization problem takes the following form (see the Appendix for the derivation): m a x n f = p\E0[Xw]-fi[K(a?,a r2) + k(Dw) + RP0W]-M(ew). (2.25) Comparing equation (2.18) in the costless manipulation setting, with equation (2.25) demonstrates that, in addition to compensating the agent for her productive effort cost and consumption risk premium, the principal must compensate the agent for her window dressing cost. Differentiating II^ with respect to v?,ew establishes: dll£_ = dE0[X"]da? _ dk(a?)da? + dk{Dw)dDw + dRP^ dv?  P da? dv? da? dv? dDw dv? dv? anr = a ^ ^ _ 3 M ( 0 ( 2.26b) dew dDw dew dew H\kw + p + e w f ' 2 • e , , dE0[Xw]da? dk(a?)da? dRPQ\, Observe that the term {— - — — - [ — — - + —1} represents the net da? dv? da? dv? dv? marginal benefit provided by the agent's productive effort induced by the incentive c)k (D  w}  w rate v?, and the term { } is the marginal cost caused by the agent's 1 dDw dv?  5 y & manipulative effort induced by v?. In equilibrium, the principal determines an optimal v? by balancing the dual effects of the incentive rate, i.e., inducing the agent to provide beneficial productive effort on the one hand, and costly window dressing on the other hand. Letting (2.26a) equal zero yields v ? = ^ f - ( X w - ^ ) W 2 ] , W i i h Q w ^ Q + G, (2.27) 35 k 1 where Q = V2 + rA a2 G= , / T = — rA pa, a,. Notice that, the (K + p + ew)2 Qw first-period incentive rate vtw is.adjusted by the window dressing cost parameter kw, and also an implicit function of ew. If we define MB:=-/3aK 0 =j3 r ( v ; - v 2 r ) 2 , (2-28) dDw dew (kw + p + ew) 1 2 then the first derivative with respect to ew leads to the following proposition. Proposition 2.5 In the window dressing setting, the principal chooses to monitor, and the optimal level of monitoring ew is characterized by the first-order condition12 MB?-meew=0. The first-order condition expresses ew as an implicit function of incentive rates v{w, v[. The first term MB^ represents the marginal benefit of monitoring implied by the lower induced payment to the agent to cover the window dressing cost. The second term represents the marginal cost of monitoring. In equilibrium, the principal chooses a non-zero optimal monitoring level such that the marginal benefit and the marginal cost are equal. Unlike the costless earnings manipulation setting, the principal prefers to monitor at least a little. The reason for the non-zero monitoring is that, ex ante, window dressing induces more payment to the agent, and hence lowers the shareholders' value. 1 2 Note that an interior solution e" is implicitly assumed. This requires the second derivative to be negative, i.e., 32nj" _ 9 M B ; < Q d[e"]2 ~ de"' '"' 3 6 A related interesting question is: what happens if the principal refuses to pay for the agent's window dressing effort? Mathematically, the fixed compensation payment becomes / = cost of productive effort + risk premium - E0 [Wv \ axw, a2 ], Then the total compensation payment is E0[W \ a™,a2] = f + E0[WV \ a™,a2] = cost of productive effort + risk premium. However, window dressing occurs anyway even though the principal ignores it, and the total payment considering window dressing is E0[W\a;,ar2,Dw] = f +E0[W" \ aiw,a^ + (Vlw-vr2)Dw = cost of productive effort + risk premium + (v{w -v2)Dw . Since (v,w - v2)Dw = (*„ + p + ew)[Dwf > k(Dw), the principal ends up paying too much compensation if he tries not to pay for the window dressing effort. Therefore it is optimal for the principal to reduce the pay to the level which just covers the window dressing cost k(Dw). 2.7 Real Earnings Manipulation This section extends accounting earnings manipulation to real earnings manipulation. Real earnings manipulation implies that the agent deviates from the optimal operating plan to change the underlying economic outcome, and hence manage the reported earnings. For example, by giving an early special discount, the agent can shift next-period sales to the current period. The manipulation results in a real cost to shareholders as long as the discount rate is greater than j/^ . Of course, the manipulation can go the other way, i.e., current sales are shifted to the next period 37 by delaying the special discount offer. This type of real manipulation affects the underlying real outcome, but is costless to the agent in terms of manipulative effort. Formally, we model this real earnings manipulation as follows. Let DR be the amount of real earnings manipulation, and let H(DR) = /2hR[DR]2, with hR > 0, be the detrimental effect of real manipulation on the underlying economic outcome. Then = *, + DR, (2.29) xR =x2-RDR -RH(DR). (2.30) Now the total gross payoff to the principal is XR = xR + /3x2 =x{+ J3x2 -H(DR), which is directly reduced by H(DR). Further, let the performance measures be yR = x R , and y2 = x2 . Then the expected performance measures given the productive effort and real manipulation are E[yR \al,DR] = yal+DR, (2.31) E[yR \ay,a2,DR,yR] = ya2-RDR -RH(DR) + p(cj21CJx){yR -yax-DR). (2.32) At the time when the agent chooses the amount of real manipulation DR (i.e., after xy is realized), the agent anticipates her first-period compensation, the penalty, and the second-period certainty equivalent as follows, 0){(DR) = f + v?(xl+DR), P(DR) = /2(p + eR)[DR]2, (2.33) CEf(DR) = pv2[ya[ -RDR -RH(DR) + p(cr2 /cr,)(*, - yal)]- pk(ar2)- . (2.34) Her decision problem with respect to DR is to maximize max{co; (DR)-k(aR )-P(DR) + CE?(DR)}. (2.35) 38 The first-order condition yields DR = - ( v * -vr2). In equilibrium, the vr2hR + p + e agent takes account of the negative effect of the real manipulation on performance measures, and the amount of real manipulation is decreasing in hR . At date zero, the principal chooses vR and eR to maximize his expected gross payoff net of the agent's total productive effort cost and total compensation risk premium, as well as his monitoring cost. That is, m a x n f = /3E0[XR]-/3[K(aR,ar2) + RPR]-M(eR). (2.36) v, ,e The key feature is that the principal's gross payoff XR = xl + fk2 -H(DR), and the equilibrium DR is a function of vR and eR . Hence, the first-order conditions with respect to vR, and eR are as follows:13 911* _ aidE0[XR]daR dH(DR)dDR rdk(aR)daR , dRPR„_n dv* -P[^aT~dvJ ~ ^ b ^ ^ ~ [ ~ ^ d v J + ~dvT']}~0- ( 2 > 3 7 a ) deR ~ P dDR deR deR = P( r u K ^ 3 ( v f - v O 2 - m . e * = 0 . (2.37b) (v[hR +p + eR)3 The following proposition summarizes the main results. Proposition 2.6 In the real earnings manipulation setting that is costless to the agent, (2.6a) The principal chooses to monitor, and the optimal level of monitoring eR is characterized as: MBR -meeR = 0 , 1 3 Interior solutions are implicitly assumed. This requires the second derivatives to be negative. 39 l t r n R n dH(D ) dZ) _ hR / R r x 2 where MB* =-0 K R = yg— R- ^ ( v -v )2. d£> de ( v ^ + p + e y (2.6b) Given incentive rates vR ,v 2 , and monitoring efi , the equilibrium real earnings manipulation is characterized by DR = —(vR - v 2 ) . v[hR + p + e The real earnings manipulation is costly to the principal directly through the term H(DR). In equilibrium, the principal engages in monitoring so as to reduce this cost. The optimal level of monitoring occurs when the marginal benefit (implied by the lowered real cost caused by the manipulation) equals the marginal cost of monitoring. Ex post, the degree of real manipulation is limited by the principal's monitoring and external regulation. In practice, some real earnings manipulation could be costly to the agent directly, i.e., it requires the agent's manipulative effort. Following the analysis in the window dressing section, let k(DRw) = /2kw[DRwf be the manipulative effort cost. Then the agent's decision problem with respect to DRw is m a x { < ( D R w ) - k ( a R w ) - k ( D R w ) - P ( D R w ) + CEXX (DRw)}. (2.38) Solving this maximization problem produces DRw = — (v,R,v - v 2) K + v i h R + p + e w (see the Appendix for the derivation). More importantly, when the principal chooses his ex ante monitoring eRw, he must take account of both the direct cost H(DRw) and the induced cost k(DRw). Formally, the principal maximizes the following decision problem, max n£*w = /3E0[XRw]- P[K(aRw,a[) + k(DRw) + RPRw] -M(eRw)., (2.39) 40 where XRw = *, + px2 -H{DRw). Differentiating Ylfw with respect to vRw,eRw establishes the following first-order conditions:14 dvRw ~ P daRw dvRw dDRw dvRw dk{afn daR" dk(DR") dPRw dRPRw _ anfw = dH{PRw)dPRw gdk(DRw) dPRw dM(eRw) deRw dPRw deRw dPRw deRw deRw =p(t + * ; t \ R ^ ( v r - v > ) 2 - m < e R W = 0 • ( 2 - 4 0 b ) (kw +v2hR +p + e ) We summarize the characteristics of the equilibrium as follow. Proposition 2.7 In the real earnings manipulation setting that is costly to the agent, (2.7a) The equilibrium monitoring eRw is characterized by MBRw - meeRw = 0 , where H dDRw deRw H dDRw deRw P (kw + vr2hR + p + eRw)3 1 2 (2.7b) The real earnings manipulation is characterized by PRw = ^ ( v * w - v 2 r ) . K + vr2hR+p + eRwK 1 2 J In this real earnings manipulation setting that is also costly to the agent, the principal bears double losses implied by the detrimental effect of manipulation on the underlying real outcome and the agent's manipulative effort cost. Compared with the other earnings manipulation settings, the role of the principal's monitoring here 1 4 As elsewhere, interior solutions are implicitly assumed, and the second derivatives are required to be negative. 41 becomes more important. The marginal benefit of monitoring comes from a reduced real cost H(DRw), and a reduced induced cost k(DRw). Therefore intense monitoring benefits shareholders both directly and indirectly. 2.8 The Impact of Corporate Governance Reforms This section briefly discusses the model's implications for recent corporate governance reforms. First, comparative static analysis leads to the following proposition. Proposition 2.8: (2.8a) In the setting where earnings manipulation is costless to the agent and has no effect on the underlying economic outcome, i. & / =0. /dme ii- 3 1 D / d p < 0 if [(1 - p2 )o\ - A (a,2 + f5poxo1)] * 0 . (2.8b) In the costly window dressing and real earnings manipulation settings, i. d e / ^ m < 0, where e* = ew, eR , or eRw. ii. Ex post, given the incentive rates which are not equal, ^ I ^ }/dp<^ anc* ^ D < 0 , where (D*,e*) = (Dw,ew), (DR,eR),or (DRw,eRw). Proof: See the Appendix. When the earnings manipulation is costless to the agent and has no effect on the underlying economic outcome, the optimal board monitoring is zero. But when the earnings manipulation is costly to the agent (i.e., window dressing) and/or affects the 42 underlying economic outcome, the board chooses to monitor management, and Proposition (2.8b - i) shows that a decrease in the monitoring cost parameter (me) induces the board to monitor more intensely. Further, Proposition (2.8b - ii) shows that, ex post, either more internal board monitoring (e* ) or tighter external regulation ( p) induces a lower level of earnings manipulation. Recent worldwide corporate governance reforms mandate various kinds of changes in governance measures. For example, Sarbanes-Oxley Act of 2002 in U.S.A. makes significant changes in corporate responsibility, financial disclosure, the legal status of auditors, analysts and especially board of directors. In Europe, investor activism and corporate control activity are becoming more common. In Asia, certain Japanese companies have implemented structural changes after the Commercial Code of 2002, including board size reduction, more outside directors and committees, and disclosure improvements. The Korean government has also embarked on a series of governance reforms on increased financial transparency and disclosure, board independence. These reforms have the effect of changing certain parameters in this model. For example, the monitoring cost parameter me might be reduced by the proposal that the audit committee structure includes members with financial background since financial experts work more efficiently than non-financial experts when dealing with issues related to financial and audit control. The suggestions and requirements of greater board independence might also lower board's overall monitoring cost. If so, the changes may facilitate more intense board monitoring under the situations in which boards choose to monitor. In addition, the requirement that Chief Executive Officers (CEOs) and Chief Financial Officers (CFOs) certify the financial statements 43 of their companies with severe penalties in the event of disclosure failures results in a higher value of regulation parameter p. Consequently, these changes will help deter management from manipulating earnings. 2.9 Concluding Remarks This chapter uses a dynamic agency model with renegotiation to study the interaction between corporate governance and earnings management. The model captures both internal and external governance mechanisms, specifically, board of directors and overall regulatory environment. Our analysis makes the following contributions. First, it contributes to the earnings management literature. We explore a variety of types of earnings manipulation, including accounting and real manipulation, costly window dressing and costless manipulation. The equilibrium captures earnings manipulation consisting of earnings shifting over time, and consumption smoothing without smoothing firm's accounting earnings. Another contribution is that the model shows that the extent to which the board engages in monitoring does not depend on earnings manipulation per se, but rather on its real economic effect. For the four different earnings manipulation settings explored in the chapter, the board prefers to monitor if the manipulation is costly to the agent and/or affects the underlying economic outcome, but it prefers to not monitor at all if the manipulation is costless to the agent and does not affect the economic outcome. In the real world, some accounting accruals manipulations fall into the latter category. Through optimal contracting, this type of manipulation has 44 zero economic effect on shareholders' value and any resources used in monitoring create a deadweight loss to the firm. The study contributes to the current debates about corporate governance. The chapter stresses the importance of "substance" over "form" because a regulatory change would have direct and induced effects that need to be considered. When implementing a new law, as seen in the new wave of worldwide governance reforms at the beginning of this century, proper practices should be strong yet flexible in terms of time, location, industry and country. For example, the SOX Act may be appropriate for U.S. companies, but some of its rules might not be suitable for European or Asian companies due to differences in economy, politics and culture. One limitation of the chapter is that the board's incentive is exogenous. The board is assumed to maximize shareholders' value, and controls all of the bargaining power. In practice, this might not be true. For example, directors often desire to be re-appointed to the board, and top executives often play a considerable role in the nominating process. Subject to managerial power, the board members' own interests can be different from that of shareholders. This might be especially true for companies using the single board system which combine the roles of chairperson and C E O . 1 5 While this chapter does not explicitly model managerial power, certain parameters in the model capture this feature indirectly. For example, managerial power has the potential to increase board members' opportunity cost of monitoring due to the career concerns, that changes the monitoring cost parameter me. An increased value of mc induces the board to exert less intense monitoring in 1 5 Bebchuk and Fried (2003) review paper notes how managerial power affects an executive's own pay and the board of directors' monitoring of management. 45 equilibrium. This suggests that managerial power can distort the board's incentive to act in the shareholders' interest. 4 6 Figure 2.1 Timeline of the agency model Date zero: - The principal designs and offers an initial compensation contract to the agent c / ( / / , v ; , v 2 ) . - The principal commits to the level of monitoring e. Period one: - The agent chooses productive effort ax. - Outcome xx is realized and privately observed by the agent. - The agent decides on the amount of earnings manipulation D. - Public report y, is issued. - The agent receives her first-period compensation cox = / ' + v,' y,. - Penalty P. - Bank balance available for consumption Bx=cox- k(ax)-P.]6 - The agent chooses consumption c,. Period two: - Renegotiated compensation contract cr ( A / r , v 2 ) . - The agent chooses second-period productive effort a2. - Outcome x2 is realized and privately observed by the agent. 1 6 The agent is assumed to choose her date zero consumption before accepting the contract. The agent reverses the amount of earnings manipulation D in the previous period. Public report y2 is issued. The agent receives her second-period compensation a>2 = Af + v2y2. Bank balance available for consumption B2 - R\{ +co2- k(a2). The agent consumes her remaining bank balance c2 = B2. 48 CHAPTER 3 Corporate Governance and Information Quality 3.1 Introduction Many recent corporate governance reforms around the world aim to increase public disclosure by firm management. Making more information publicly available to the capital market is generally expected to make current market prices more informative about the future value of the firm. However, the market price is based on total information: both the public information disclosed by firm management and the private information acquired by investors. For this reason, it is relevant to ask: does forcing management to make more information available publicly change the amount of information acquired by investors? Once this question has been answered, we can ask: do the changes in public and private information result in a net increase or a net decrease in total information? McNichols and Trueman (1994) note that both academics and regulators are interested in the answers to these questions. Based on Kyle (1985), we develop a strategic investor model in an insider trading setting. In this single-period model, four sets of traders are included: a firm's manager (i.e., an insider), multiple privately informed rational investors, one privately uninformed rational investor called the "market maker", as well as liquidity traders. The manager has knowledge about the terminal firm value. We assume that she 49 discloses at least some of that information and we consider the possibility that she is able to withhold some of the information and use it for personal trading, without incurring penalties for insider trading. The privately informed investors acquire costly private information. The manager and informed investors use their private information (in addition to the public report) when determining their market trades. While the privately uninformed investor, i.e. the market maker, can not observe their private information, he can imperfectly infer that information from the aggregated trading quantities. Hence, the market clearing price depends on both the public report and the private information. In our stylized single-period model setting, there is prior uncertainty about the terminal firm value and it can be reduced either by releasing public information or by obtaining private information. The quality of that information is defined to be the extent to which that information reduces prior uncertainty about the terminal value of the firm. In particular, a comprehensive measure, referred to as total information quality, is defined to be the extent to which the market price reveals the terminal firm value. It is derived based on the expected square of the difference between the equilibrium price and the terminal value. An increase in total information quality means that the market price is more informative about the terminal firm value. The equilibrium market price in our model reflects both the public and private information, but in a different way. We show that, while the public report is fully revealed, the private signals are partially revealed since the manager and informed investors partially hide their private information through strategic trading. 50 Consequently, the total information impounded in the market price is the aggregate of the public information and a fraction of the private information. The manager knows more about the future events than is publicly disclosed. As a corporate insider, she may have many motives to trade, such as, realizing stock-based compensation, shedding firm-specific risk, or meeting liquidity needs. In this model, we only consider insider trading, i.e., a manager who trades with the intent to profit from her possession of nonpublic information. Such trades are illegal and we consider the implications of corporate governance mechanisms used to enforce the ban on insider trading. While tighter governance forces management to disclose inside information, and hence improve the quality of public information, the analyses show that its effect on the equilibrium total information quality is more subtle. At the same time as corporate governance changes the public information directly, the interaction between the public reporting and private information acquisition changes the amount of private information acquired in equilibrium. More specifically, tighter governance could lead to higher qualities in both public reports and private signals if the investors obtain private signals by processing public reports. Or, tighter governance could increase the quality of public information but lower the quality of private information if both sets of information are from the same sources. In other case, more informative public reports make it more costly for the investors to obtain additional information. Of course, tighter governance could improve the quality of the public reports and have no effect on the private signals if the different types of information come from different sources. In all cases, the total information impounded in the market price is the aggregate of public and private 51 information. We show that the effect of tighter governance on total information quality in capital markets is case-specific. The link between governance and total information quality depends on the interaction between public disclosure and private information acquisition. In particular, the quality of private information is an endogenous variable which is a function of the quality of public information. In the accounting literature, Demski and Feltham (1994), Kim and Verrecchia (1994), and McNichols and Trueman (1994) study similar questions in various settings. Demski and Feltham (1994) use a Grossman and Stiglitz (1980) type model to note that changes in the information content of the public report affect an investor's decision to privately acquire this costly signal. A more informative disclosure has an ambiguous effect on price change variances around the date of the announcement. Both Kim and Verrecchia (1994), and McNichols and Trueman (1994) use Kyle (1985) type models. Kim and Verrecchia focus on the issue of information asymmetry, created by investors who may process the information in financial reports to obtain extra information, and study its link to market liquidity and the bid-ask spread. They conclude that more public information reduces information asymmetries and makes the market more liquid. McNichols and Trueman study the importance of short term trading behavior prior to an announcement and show how this behavior affects price reactions to a disclosure. They conclude that an informed trader prefers to receive information about the future disclosure than about the firm's value. We use a Kyle type model and consider the role played by some investors acquire private information when insiders are endowed with information. The role of insiders is important since we 52 compare the equilibrium in a model which allows insiders to trade and withhold information with the equilibrium in a model in which insiders cannot trade and must disclose all inside information. The remainder of the chapter proceeds as follows. The next section develops a model of strategic investors and outlines the information structure. Section 3 characterizes the equilibrium market behaviors. Section 4 derives the endogenous informed investor's private signal precision. Section 5 considers the effects of corporate governance on various information qualities in capital markets, i.e., public, private and total information. The last section provides the concluding remarks. 3.2 The Strategic Investor Model In the model, there are four sets of risk neutral traders: a firm's manager (i.e., an insider), a single privately uninformed rational investor, multiple privately informed rational investors (/), and liquidity traders. The manager and the privately informed investors trade to maximize profits based on their own private information. The liquidity traders (referred to as noise traders) trade for reasons that are independent of the market price. The shares held by the liquidity traders is exogenously specified as 2 17 Zt ~ N ( 0 , o~z). The privately uninformed investor does not observe the individual quantities traded by the manager and the privately informed investors as well as the liquidity traders, but observes the aggregate quantity. The uninformed investor imperfectly infers the manager's and the informed investors' private information based on the aggregate quantity, and sets the market price so that he expects to break 1 7 As is common in Kyle type model, the preferences of liquidity traders are not modeled, so it is impossible to analyze social welfare issues. 53 even. The uninformed investor is called the market maker. The introduction of noisy trades is important since it ensures that market price is not fully revealing of private information. The timeline and the sequence of events are described as follows (see Figure 3.1). At date zero, the manager has endowed information about the terminal firm value, the information set is (ya,ym). At date one, the manager disclose ya to the public (all investors observe the public report) and withholds ym for the purpose of personal trading. Then the multiple investors (7) acquire costly private signal y(.. Based on their own private information, the manager and informed investors place orders simultaneously along with liquidity traders. The uninformed investor, i.e., the market maker, does not know ym, y(., the individual quantities traded by the manager zm, the privately informed investors zt, and the liquidity traders (Z,), but observes the aggregate quantity (z u ) . He absorbs the remaining shares to clear the market and sets the market price. At date two, the terminal firm value is realized. Several features of the model are as follows. First, since the manager and informed investors use their private information (in addition to the public report) to determine the trading quantities, the market clearing price depends on that information. As is common in Kyle type models, both the manager and informed investors anticipate that their choices of demand functions influence the market price, and act strategically. Therefore, instead of acting as a price taker, they have incentives to restrain their trades so as to partially hide their private information and maximize trading profits. 54 Figure 3.1 Timeline of the strategic investor model t=0 t=l t=2 - manager endowed with info. ( ya , ym) -manager issues public report yt (withholds ym) a — terminal value (v = m0 + e) -investors acquire private signals ys -trade zm, Zt, Zl -market maker absorbs the remaining shares and sets market price px by observing ya and z, Information Structure The terminal value of the firm is represented by v = m0 + e, where £ is a zero mean, normally distributed random variable, i.e., e ~ N ( 0 , ( T 0 2 ) . The random trade of Zt is not correlated with the firm value v, Cov[v,Zl] = 0. Further, the prior beliefs with respect to the public report ya issued and the information withheld ym by the manager are ya ~ N ( 0 , ( 7 a 2 ) , ym ~ N ( 0 , C T 2 ) , and Cov[ya,ym] = 0 . For simplicity, the multiple investors (7) are assumed to receive the same private signal y(., and y. ~ N ( 0 ,o~f). The investors' private signal is independent of the manager's information, i.e., Cov[ya,yi] = 0 , Cov[ym,yi] = 0. This implies that the manager 55 and the identically informed investors are informed about different components of the noise in the firm value. Given the public report, the posterior belief about the noise in the firm value is £ \ya ~ N( ya, CT 2 , ). The private signals held by the manager and informed investors are informative about the noise in the public report. Specifically, e\ya,ym ~ N( ya + ym. <?li) , and £ \ya, ym, yt ~ N( yu + ym + yt, cr,2). Similar to Demski and 18 Feltham (1994), our specified information structure is chosen for its algebraic simplicity: °l = <*l + = <r2a + °l + <*lx = <*\ + °l + of + of. (3.1) The total uncertainty about the terminal firm value, cr 2 , is assumed to be a constant throughout the model. The total uncertainty resolved by the public report ya is CT2 . The incremental uncertainty resolved by the information withheld by the manager ym is O" 2 . And the further incremental uncertainty resolved by the informed investor's private signal y(. is o~2. Since cr2 is constant, we interpret either <72 or hal = \lo~]{ as a measure of the quality (or informativeness) of the public report with respect to e. An increase in <r2 represents an increase in quality implying a greater reduction in uncertainty by the public report. Similarly, cr2 and af measure the qualities (or The set up of the model is stylized. The main feature of our set up is that the signals are set equal to the posterior expectation about the noise in the terminal value. Other papers, such as McNichols and Trueman (1994), use an alternative set up in which the signals provide information about the terminal value but they are not equal to the posterior expectation. The two set ups are qualitatively similar. See Demski and Feltham (1994) and Christensen and Feltham (2003) for a discussion of the fact that the scale of a signal is arbitrary and it is always possible to represent a signal in terms of the posterior mean it produces. 56 informativeness) of the information withheld by the manager and the private signal acquired by the informed investor, respectively. 3.3 Characterizing Market Equilibrium and Information Quality Initially, we treat the informativeness a? as exogenous, and consider the informed investor's private signal acquisition decision in the next section. zm and zi are the orders placed by the manager and the informed investors, respectively, and the shares sold by the liquidity trade is Z , . Hence, the aggregate net quantity traded by the three sets of traders is The uninformed investor does not observe the individual quantities traded by them, but observes the aggregate quantity zu . He acquires those shares to clear the market, and sets a market price, denoted as px. Then, ignoring information acquisition costs, the trading profits for the manager, as well as for the informed and uninformed investors are The uninformed investor sets the market price so that he expects to breakeven, that is, price px equals the expected terminal value of the firm based on the information set the uniformed investor observes at this time, which includes the public report ya, and the aggregate quantity zu. That is, (3.2) 1=1 nm =(v~Pl )Z„ , 7Ci = 0 - p, )Z,, Ku =(v-pl )ZU . (3.3) (3.4) 57 where juul(zu) = E[eav \zu], and eav =e-ya represents the prior belief about the noise in the public report. The term juul(zu) is the uninformed investor's posterior mean with respect to eav conditional on zu. Moreover, the manager and the informed investors conjecture the posterior mean is proportional to the aggregate quantity: Hul{zu) = E[em\zH] = -bzu. (3.5) At the same time, the manager and the informed investors conjecture what the other will demand as a result of their private information. The conjectured demands are zm=oym, z,=0y,- (3-6) Given these conjectures, the manager and the informed investors choose trading quantities to maximize their profits by anticipating the influence of their trades on the market price. Specifically, the manager's expected price resulting from an order of zm units, given her information set (ya , ym) is E\-P\ I ya. ym. z„ ] = mo + ya + *>Zm + blp\E\\yi \ ya, ym] = mQ+ya+bzm. (3.7) The key here is that since ya , ym and yi are all independent, E[yi \ yu, ym ] = 0 . As a result, the manager's expectation with respect to the orders placed by the informed investor is zero. Further, the manager's trading profit is calculated as Km = ( v -Pi ) z m . Then based on her information set (ya,ym), the manager's expected trading payoff is U m \ . y,n . ) = ^ V I «^ • 1 " EiPl I . . Zm ] }Zm = (ym-bzm)zm. (3.8) 58 See the Appendix for the derivations. The manager maximizes her payoff by choosing her order quantity zm, then the first-order condition yields zm(ym) = — ym . 2b It indicates that the manager's demand is a function of her private signal. Similarly, informed investor f s expected price resulting from an order of zt units, given his information set ( ya, yi) is E[px\ya,yi,zi]=m,+ya+bzi+b{I-l)Pyi. (3.9) Note that since all investors receive the same private signal by assumption, investor f s expectation with respect to the orders placed by the other informed investors is E[zJ | y,] = py,•. Therefore, given the informed investor fs trading profit is calculated as 7ti = (v - px )z., his expectation of the trading payoff conditional on his information set (ya , yt) is un (ya. y, >Zi••) = i y, - W - Dffy, - bz, }z,. (3. io) Maximizing the expected profit produces z,(y,) = —[1-^(7 -l)y^]y,. As with the 2b manager, the informed investor fs demand is a function of his private signal. Finally, based on the conjectured trading quantities by the manager and the informed investors, the uninformed investor's posterior mean with respect to eav is See the Appendix for the derivation. Rational expectations require that, the conjectures must be consistent with the optimal choices. Therefore, the following proposition characterizes the market equilibrium. 59 Proposition 3.1 In the strategic investor model with insider trading, the equilibrium demands and market price are characterized as: (3.1a) the manager's demand: zm - ocym. (3.1b) the informed investor i's demand: z,. = (jy,, for / = 1,2, . . . , /. (3.1c) the market price: pl-m0 + ya -bzu. Proof: see the Appendix. Properties of the Equilibrium: Observe that, the demand functions for the manager and the informed investor differ because they have different private information. Simple comparative static analyses suggest that the orders placed by the manager and informed investors are increasing in the liquidity noise o\, but decreasing in the precisions of their private signals <72m and <jf. In this model setting, the market maker imperfectly infers the private signals. More inferred private signals lead to lower trading profits to the manager and informed investors. Therefore, they strategically restrict their trades so as to hide their private information. On the other hand, the noise created by liquidity traders helps them to hide their private signals and hence induce more trading quantities. Furthermore, substituting parameters a , (5 , and b into the equilibrium price expression yields where a = —, B = 2b 1 , and b2 (I + l)b 60 p ^ ^ + ya+^ym+j-^yi-bZt. (3.12) Grossman and Stiglitz (1980) define price-informativeness as the squared correlation between the informed posterior mean and the price. McNichols and Trueman (1994) study price-informativeness by calculating the expected change in price before and after an announcement. Similarly, we calculate the expected squared difference between px and v is 1 9 Based on the information structure, CT 2 = <r2 - (CT 2 + <T2 + < T 2 ) . In equilibrium, b2 = \—<J2 +—-—r-cT 2 ]—!— . Thus, the above expression becomes 4 m (/ + 1)2 ' CT 2 E[Pi-v]2 =<72 - ( C T 2 +^cr 2 +y~-l<7?)- (3-14) It implies that, the total prior uncertainty with respect to the terminal value CT 2 is partially resolved by the public report yu and private information ym, y(., and the amount of uncertainty resolved is O = a2 +—<72 + ^ a2. As O increases, the " 2 m I+l ' market price moves closer to the terminal firm value. O represents the aggregate of the informativeness of all types of information, i.e., public and private information. Therefore, we refer it to as "total information quality". The following proposition summarizes the result. From the information structure, v = mn+e = mn + y +v +v.+£... 61 Proposition 3.2 In the strategic investor model with insider trading, the equilibrium total information quality impounded in the market price is O s CT; +-cr„ + at. The total information quality measures the extent to which the market price reflects the terminal firm value. The public information is fully impounded in the equilibrium price, but the private signals are only partially impounded. As the number of informed investors / goes to infinity, their private information becomes closer to being fully revealed. 3.4 Endogenous Private Signals Quality The previous sections treat the informativeness of the informed investors' private signals as exogenous. This section endogenizes erf to consider information acquisition costs. Suppose the number of the informed investors is fixed, but each individual chooses the precision of his signal. For simplicity, the cost of the signal for each informed investor is assumed to be the same, and represented by the following functional form: /r,.(cT;,cr,2) = ( l+^7 a 2 )cT, 2 . (3.15) where y is the private information cost parameter, which can be positive, negative, or zero, and (1 + yu]) > 0. The specification indicates that the private acquisition cost increases in its own informativeness o~f. On the other hand, the marginal information cost is affected by the informativeness of the public disclosure cr2 when 62 y*0, and increased public information quality decreases (or increases) the information cost when y < 0 (or y > 0). Considering the information cost, informed investor fs net trading profit is n?' =(v-P{)Zi-Kr (3.16) Based on equilibrium derived in the previous section, the informed investor f s ex ante expected equilibrium net payoff, given that I investors are informed, is u r ( c r < } = „ * 2 h <JzG\ - * , o -17> (V + 1) I 1 2 . ' _2 See the Appendix for the derivation. Maximizing the ex ante net payoff with respect to erf yields the following proposition. Proposition 3.3 In the strategic investor model with insider trading, given a fixed /, informed investor f s optimal choice of private signal quality is characterized as / 2 i / I 2* <'+•>' - < ' + 1 > ' 0 + r f . ) . L / 4 < J m (Z+I) 2 ' 7 ' - ] Comparative statics show that the optimal level of private information quality erf* is (3.3a) increasing (or decreasing) in erf if y<0 (or 7 > 0 ) ; (3.3b) unaffected by af if 7 = 0 . Proof: see the Appendix. 2 0 Note that when the informed investor makes decision on a2 , he anticipates the equilibrium in the trading stage. It is a sequential game. 63 The proposition demonstrates that an informed investor acquires a less informative signal if public disclosure increases his information cost. This could happen when the public report and the private information are from the same source, and hence increased informativeness of the public report makes it more difficult for the investor to acquire additional information. In contrast, if the investor obtains private information by analyzing the public report, then a more informative public report could lead to a better quality private signal. Finally, private signal acquisition is not affected by the public report if the two information sources are independent. Notice that this intuition does not necessarily contradict the previous assumption that the public and private information are independent. For example, an expert who studies an annual report carefully may understand the implications of certain details, such as unrealized gains or special accounting practices, that affect firm value. These details could be in addition to and independent of what other investors understand from the public report. We use the simpler independence assumption for tractability considerations and because it does not affect the results qualitatively. 3.5 The Effect of Corporate Governance on Information Quality Accounting scandals and self dealing by corporate insiders have shaken investor confidence in the integrity of securities markets. For this reason, there have been worldwide corporate governance reforms that aim at improving the quality of public disclosure through tightened regulations. Certainly, some of the new rules will help, such as, the requirement of Chief Executive Officers (CEOs) and Chief Financial 2 1 See Kim and Verrecchia (1994, esp. p. 42). 64 Officers (CFOs) to certify the financial statements of their companies, and the imposition of severe penalties in the event of disclosure failures. As well, revised rules on insider trading will help to prohibit unlawful trades by insiders. Based on the above model, we consider the setting in which the firm's manager, i.e., the insider, is not allowed to trade, and discloses all the endowed information publicly. The main differences from the previous setting with insider trading are as follows. First, the public information set is increased from one signal ya to two signals (y a, ym), so both signals are observed by the investors. Second, traders are reduced to three sets: / privately informed investors, the market maker, and the liquidity traders. Thus, the aggregate trading quantity absorbed by the market maker is zu = Z[ - ^ z,- • The market maker sets the market price based on the two public 1=1 signals and the aggregate trading quantity, i.e., Pi = Euiiv I y„.ym^u] = m0 + ya+ ym + //„,(z„). (3.18) Finally, the total informativeness of the public signals is (a 2 + a2m) and, hence, the private information cost function becomes kt (a2, a2m, af) = [1 + y(a2a + a2m )]a2'. (3.19) Following the same characterizing process as in the previous sections, the following proposition describes the market behavior. Proposition 3.4 In the strategic investor model without insider trading, the market equilibrium is (3.4a) the informed investor f s demand: z,(y,) = < T z /— y, , for i = 1, 2, . . . , /. 65 yfl a (3.4b) the market price: px = m0 + ya + ym -b zu where b = . I + 1 CTZ Proof: see the Appendix. Proposition 3.4 implies that the informed investors trade more aggressively if there is more liquidity noise, and hence have a larger impact on the market price. The informed trading is decreasing in the number of informed investors in the market due to competition. Further, substituting parameters b into the equilibrium price expression yields P\=m0 + ya+ ym + y-^ y, -b'Z,. (3.20) As in the insider trading setting, we calculate the expected squared difference between p\ and v: E[p\ - v]2 = Jf^fVi + b'2(7z + < • (3-21> Given the information structure, o]x = <T02 - (CT2 + r j 2 + a]), and the equilibrium condition, b - ^ , the above expression becomes 7 + 1 crz E[P\ - v ] 2 =cr02 - ( C T 2 +cr2 +y^,2')- (3-22) Proposition 3.5 In the strategic investor model without insider trading, the equilibrium total information quality impounded in the market price is 2 2 I 2' 6 6 Compared to insider trading setting, the inside information ym is now fully captured in the market price. In order to compare the total information quality in the two settings, we need to solve for the equilibrium level of private signal quality without insider trading. Given that / investors are informed, informed investor i's ex ante expected equilibrium net payoff is: U~' (CT<} = 7T\TT7 ^ - ^-') • (3-23) (/ + 1)V/ See the Appendix for the derivation. Maximizing the ex ante net payoff produces the following proposition. Proposition 3.6 In the strategic investor model without insider trading, given a fixed I, the optimal level of informed investor f s private signal quality is , 1 a ' "2[l + ^(cTa2+cT,;)](/ + l )V7 £ J Z ' Simple comparative statics demonstrate that private signal precision o~* is increasing (or decreasing) in the informativeness of public information (a] + cr 2) if y < 0 (or y > 0) and unaffected if y = 0. This result is same as in the insider trading setting. In addition, investors acquire a less precise signal as more investors become informed, because more competition leads to lower profit. 67 Comparison of Information Quality in Two Settings Based on the above analyses, we compare the equilibrium information qualities in the settings with and without insider trading. First, tighter regulation prohibits unlawful insider trading and induces higher quality public information. Specifically, the informativeness of public disclosure is increased from <r2 in the insider trading setting to (<r2 + CT 2 ) in the setting without insider trading. Further, regulation changes the informed investors' trading profits and information acquisition costs. The change in the trading profits is due to the differential revealing of the signal ym in the market price. And, the change in the information costs is due to the interactive effect of public information on private information acquisition. The interaction is ambiguous since the change in marginal information cost yu2m could be positive, negative, or zero, depending the sign of parameter y. That is, increased informativeness of public reporting may lead to a higher / lower information cost, or have no effect. As a result, the optimal levels of private signal precisions differ substantially. Mathematically, they are characterized by / y4(7l+/2(I + l ) 2 ( 7 r (I + l ) 2 n . , 1 -a + y&2) vs. CT. = — — , [y4cr2m + ^ 7 ,<r f ] * ° z " ' ' 2 [ l-r r ( c T f l 2 + c T „ 2 , ) ] ( / + l)v7 z ' Then, the change of total information quality is A . S O - * = i c r 2 + -J-JUjf* - <J2* ]. (3.24) The total information quality improves if the private information quality increases. However, if the increased level of public information quality is offset by (or lower 68 than) the reduced level of private information quality, then the total information quality is unchanged (or decreases). If decreasing, it indicates that the market price is less informative about the terminal firm value. In sum, the net effect of tightened governance on information efficiency and market price is case-specific. Table 3.1 illustrates this ambiguity with numerical examples. Both private signal qualities af and af are decreasing in the information cost parameter y. That is, the more costly of the private information, the less precise of the signal. In the table, when y increases from 0.002 to 0.004, the precision af decreases from 302.1 to 255.9 and af decreases from 276.8 to 225.8. The effect of increased public disclosure on the endogenous quality of private signals is ambiguous. Comparing the settings with insider trading and without insider trading, the public information quality improves by 10 but the quality of private information does not consistently increase. The effect is ambiguous because it varies with the sign and the value of y. As shown in the table, private signal quality increases by 42.2 when ^=0.004, but decreases by 30 when y= -0.004. Consequently, the total information quality could be better, worse or unchanged, as shown in Figure 3.2. 69 Table 3.1 Comparison of equilibrium level of information quality With Insider Trading (1) Without Insider Trading (2) Change (2-1) 7 <*'a of of of A,- A * -0.006 50 725.4 659.5 60 847.7 766.4 10 122.4 107.0 -0.004 50 558.9 520.8 60 601.2 561.0 10 42.2 40.2 -0.002 50 444.7 425.6 60 448.4 433.7 10 3.7 8.1 0.000 50 362.8 357.4 60 347.2 349.4 10 -15.6 -8.0 0.002 50 302.1 306.8 60 276.8 290.7 10 -25.3 -16.1 0.004 50 255.9 268.2 60 225.8 248.2 10 -30.0 -20.0 0.006 50 219.8 238.1 60 187.7 216.4 10 -32.0 -21.7 * The table shows the equilibrium values for varying information cost parameter y. The other parameters: o~z =500, / =5. Here, A, = erf - o f , A,j, = <t> - (I>. Figure 3.2 Effect of corporate governance on total information quality Change in total information 1201 -40 J Information cost parameter * The figure shows the equilibrium values of change in total information quality for varying information cost parameter y. The other parameters are oz =500, / =5. The horizontal axis is y*1000-70 3.6 Concluding Remarks In this chapter, we analyze a simple insider trading model to examine how corporate governance affects information quality in capital markets. One feature of the model is to study the effect of governance through the analysis of the interaction between public reporting and private investor information acquisition. The model helps us to understand the direct and indirect roles of governance on the total information quality impounded in the market prices. Directly, better corporate governance helps to improve the quality of public disclosure. Indirectly, governance affects private information acquisition through its effect on public disclosure. The interplay of public disclosure and private information acquisition is subtle. Increased informativeness of public reports can lead to more or less informative private signals. Both public and private information are impounded in market prices; hence, the effects of governance on the total information quality and market prices are ambiguous. We show cases where more public disclosure can actually make the total information quality decrease. The model is simple, and the set up is stylized, since its purpose is to focus attention on the conceptual issues surrounding corporate governance, public reporting, and private information acquisition. For this reason, and since there is a large literature which investigates related aspects, we do not consider the social costs or benefits of the proposed policy change. As in Grossman and Stiglitz (1980), there is a general presumption that it is better to have a more informationally efficient market but this remains an open question. 71 CHAPTER 4 Empirical Evidence 4.1 Introduction The key prediction from the theory in the previous chapter is that, while tightened corporate governance might improve the quality of public disclosure, its effect on the total information quality is ambiguous. The reason is that increased public disclosure may lead to more or less private information acquisition. The corporate governance reforms of Sarbanes-Oxley Act (SOX) of 2002 provide a natural experiment to investigate the relationship between corporate governance and information quality in capital markets. This chapter examines two research questions. First, we investigate the information quality surrounding the SOX reforms. In particular, we examine whether public disclosure improved after the passage of the SOX. What is the effect of SOX on private information quality? Does the total information quality increase after the SOX? The second research question considers whether boards of directors contribute to the changes in the information quality during this period. We focus on boards since, although the reforms around SOX are comprehensive, the board of directors is one of the most important internal governance mechanisms. The SOX and related reforms emphasized independent directors as a way to monitor financial information 72 processing and detect disclosure failures. The previous chapter proposed a stylized model of events in a single period. In it, we defined the quality of public (private) information as the extent to which that public (private) information reduces prior uncertainty about the terminal firm value. The market price reflects both public and private information. A comprehensive measure, referred to as total information quality, is defined as the extent to which the market price reveals the terminal firm value. It is calculated as the expected square of the difference between the market price and the terminal value. An increase in total information quality means that the market price is more informative about terminal firm value. This chapter implements this idea by using earnings forecasts of financial analysts to capture the underlying public and private information set available to informed investors in capital markets. The literature supports the use of this kind of proxy (e.g. Botosan, Plumlee and Xie, 2004; Barron, Harris and Stanford, 2003). We compute the different precisions of public, private and total information levels by employing two different information quality measures, one proposed by Barron, Kim, Lim and Stevens (1998) (hereafter BKLS), and the other proposed by Gu (2005). Both measures derive unobservable analysts' information properties by using observable analysts forecast data, i.e., forecast dispersion, forecast error, and the number of forecasts. The empirical results document that, the quality of public disclosure improves following the SOX reforms. However, the quality of private information decreases. Further, since total information is the aggregate of the public and private information, 73 the increase in public information quality is offset by the decrease in private information quality, so that there is no significant change in total information quality in the year following the passage of the SOX. Moreover, the public disclosure improvement only lasts one year. Following the prior literature on corporate governance, we measure board characteristics in terms of board independence, and board meeting frequency. We fail to find a significant positive association between board characteristics and public information quality during the sample period. The result suggests that reforms to board independence and board meeting times may not have contributed to the improvement in public disclosure after the passage of the SOX. If true, then other enhanced governance mechanisms have played a dominant role in increasing the quality of public information immediately after the passage of the SOX. The remainder of the chapter proceeds as follows. The next section presents a history of the events surrounding the Sarbanes-Oxley Law and some related literature. Section 4.3 develops the research design and describes how the variables are measured. In particular, it illustrates the calculation of the BKLS (1998) information quality variables. Section 4.4 presents and discusses empirical results including data and sample description. Sensitivity checks are offered in Section 4.5. The last section provides the concluding remarks. 4.2 Institutional History and Literature Review The SOX and related reforms have attracted a lot of attention from the academics, raising many debates regarding the mandated governance changes (Ribstein 2002, 74 Mitchell 2003, etc.). The SOX was approved at the end of July of 2002 and, in the words of Li , Pincus and Rego (2006), it is "the most important legislation affecting financial corporate reporting passed in the United States since the 1930s." Figure 4.1 lists a number of high profile accounting scandals at large companies that occurred during the period leading up to the passage of SOX. The Enron scandal of October 2001 may be the best known of these scandals. Figure 4.2 lists related critical events in government that occurred at the same time. Starting in December 2001, Congress held hearings on the collapse of Enron. These hearings led to over 30 bills being proposed but it was unclear whether the problems were widespread or just in a few "rotten apples" (Jain and Razaee, 2005). A bill on the issue was passed by the House of Representatives in April 2002, but was not immediately considered by the Senate. The scandal involving WorldCom occurred in May 2002, less than two months before SOX was passed. In the middle of June 2002, a jury convicted Arthur Andersen of obstructing the SEC's investigation of Enron, just before the Senate Banking Committee met. Romana (2004) notes that the debates in the Senate and in the House of Representatives on the Sarbanes-Oxley Act itself took less than a week. The reason for the surprising speed is because events, such as those involving WorldCom and Arthur Andersen plus Global Crossing and Tyco, "changed the political landscape" (Li, Pincus and Rego, 2006, p. 12). Evidence of the high pressure to pass a law quickly can be seen in the comment of former SEC Chief Accountant Lynn Turner that "much of the content . . . had been discussed previously but lacked support in Congress" (Li, Pincus and Rego, 2006, p. 12). Romano argues that the quick consideration indicates that the law has flaws. These flaws can be seen 75 in current attempts to revise the law , especially for smaller firms, and in estimates showing that the cost of compliance is high (cited in Bryan and Lilien, 2006). Ribstein (2002, p. 67) notes that some firms may be willing to pay such costs to demonstrate integrity while some commentators worry about a "race to the bottom" among securities regulators seeking to lower compliance costs. The SOX aims to create an environment with accurate and timely disclosure of more financial information. For example, Section 203 states "The CEO and CFO of each issuer shall prepare a statement to accompany the audit report to certify the 'appropriateness of the financial statements and disclosures . . . and that those financial statements and disclosures fairly present, in all material respects, the operations and financial condition of the issuer." The maximum penalties for willful and knowing violations of this section are a fine of up to $5 million and/or imprisonment for up to 20 years. This Section came into force in August 2002. In addition to existing extensive quarterly and annual filings, Section 401 requires more reports on off-balance-sheet transactions and on pro-forma earnings. Section 409 requires real time disclosure on material changes in financial conditions.23 Some of the SOX reforms dealt with the board of directors. In order to enhance the monitoring function by the board of directors, Section 301 states "Each member of the (board) audit committee shall be ... independent. 'Independent' is defined as not receiving, other than for service on the board, any consulting, advisory, or other compensatory fee from the issuer, and as not being an affiliated person of the issuer, or any subsidiary thereof." In August 2002, the New York Stock Exchange (NYSE) 2 2 See thomas.loc.gov/cgi-bin/query/D?cl09:2:./temp/~cl09RRDpVp. 2 3 www.aicpa.org/info/sarbanes_oxley_summary.htm. 76 filed a proposed rule change with the SEC aimed at ensuring the "independence of directors . . . and to strengthen corporate governance practices". In October 2002, the NASDAQ offered a similar proposal (Carter, Lynch and Zeldman, 2005). These rules became effective in January 2003.24 The results of recent event studies concerning the market effects of SOX on firm value have been mixed, according to Ribstein (2005) and Romano (2004). Li , Pincus and Rego (2006) find that passing the Act was associated with significantly positive abnormal stock return and that investors anticipated that SOX would constrain earnings management. Rezaee and Jain (2003) find that abnormal returns of firms with weak governance were negative and suggested that the costs of implementing the SOX rules for such firms would exceed the benefits. Bhattacharya, Groznik and Haslem (2002) find that executive certification of financial statements were not significantly priced by investors. Cohen, Dey and Lys (2005) investigate the trends in earnings management in the pre and post SOX periods. Distinguishing between accounting and real earnings management, they suggest that the SOX caused firms to switch from accounting to real earnings management methods. An extensive governance literature has studied how board characteristics are related to corporate performance, and how board characteristics affect the observable actions of boards, such as CEO replacement, executive compensation and 25 acquisitions. The literature has shown that there is no overall positive relation For more details on the time line, see www.businessweek.com/adsections/2003/pdf/0331so.pdf, www.goodwinprocter.com/pdfs/S_0_tirneline.pdf, www.forbes.com, http://www.ffhsj.com/cmemos/020802_sarb_timeline.pdf, and Romano (2005). 2 5 Hermalin and Weisbach (2003) survey the economics literature and offer several key findings. For example, board composition is not related to corporate performance, but board size is negatively 77 between the degree of board independence and various measures of firm performance, such as earnings, Tobin's q, and stock price (Romano, 2004). In the accounting literature, some papers study how board characteristics are related to earnings management and the quality of financial disclosures, instead of corporate performance. Klein (2002) finds a negative relation between audit committee independence and abnormal accruals. Agrawal and Chadha (2003) finds that having at least one member of the audit committee with accounting or financial expertise is more important than the independence of the committee. They also find that firms having the CFO on the audit committee were more likely to restate earnings. Romano (2004) discusses these results and worries that the findings may be biased because of self-selection, i.e., that better managed firms may choose directors with more expertise. Few studies consider financial disclosure practices, although some papers have begun to focus on management earnings forecasts: Ajinkya, Bhojraj and Sengupta (2005) and Karamanou and Vafeas (2005). Ajinkya, Bhojraj and Sengupta (2005) finds that firms with more outside directors tend to issue management earnings forecasts more frequently, and they tend to be more accurate and less optimistically biased. Karamanou and Vafeas (2005) considers additional characteristics of the board and the audit committee, such as size, percentage of outside directors and the number of meetings. Overall, they find that effective governance is associated with more accurate forecasts. However, Felo, Krishnamurthy and Solieri (2003) find no association between board independence and the quality of financial reporting, related to corporate performance. As well, both board size and board composition appear to be related to the quality of the board's decision on CEO replacement, executive compensation, and acquisitions. 78 measured b y A I M R scores. O u r w o r k considers the effect o f the governance reforms around S O X on the i n f o r m a t i o n qual i ty i n capital markets. Instead o f f o c u s i n g o n an i n d i v i d u a l source o f informat ion, such as management earnings forecasts, w e consider more comprehensive measures that reflect the overa l l qual i ty o f p u b l i c disclosure. In addi t ion to testing the l i n k between governance reforms and p u b l i c in format ion qual i ty , w e also test the l i n k between governance reforms and private in format ion quality. T h i s second test is based o n our theory w h i c h stresses the interaction between p u b l i c and private informat ion. W e complete the analysis b y l o o k i n g at the total in format ion qual i ty , w h i c h is arguably more important because it captures the overal l in format ion ef f ic iency i n capital markets. M o r e o v e r , since the S O X and related governance reforms proposed strict rules o n board structure a i m i n g to enhance the board's m o n i t o r i n g funct ion. W e therefore investigate whether board characteristics change f o l l o w i n g S O X , and whether those board characteristics contributed to the changes i n p u b l i c , private and total i n f o r m a t i o n quality. O u r study adds to the g r o w i n g debate o n the costs and benefits o f S O X and related governance reforms. 4.3 Research Design and Variable Measurement T h e reforms surrounding the S O X provide a natural setting to e m p i r i c a l l y examine the association between corporate governance and informat ion quality i n capital markets. A s noted above, both internal and external governance mechanisms were enhanced under the n e w rules. 79 There was a lot of pressure on Congress to pass a law, so the wording of the Sarbanes-Oxley Act that was passed on July 30, 2002 was chosen relatively quickly.26 The requirement that the CEO and CFO certify the quarterly and annual reports was implemented starting from August 2002 and was expected to immediately affect the quality of public disclosure. For this reason, we use August 2002 as the critical time of the "event" that we study. Our empirical analysis examines a three-year window, from August 2001 to July 2004, surrounding the passage of the Act. Year -1 corresponds to the fiscal year ended between August 2001 and July 2002, year 0 corresponds to the fiscal year ended between August 2002 and July 2003, and year +1 corresponds to the fiscal year ended between August 2003 and July 2004 (see Figure 4.3). The post-SOX periods are year 0 and year +1.27 Note that we also tried other months as the cutoff to define the event years, such as October 2002. This yields regression results that are similar to those reported in the chapter. 2 6 Ribstein (2005, p. 14) comments that "it may be difficult to decide on a relevant date for measuring the effect of SOX given continuous media reports bearing on the likelihood of adoption". 2 7 We do not go back further because of the consideration that the effect of SOX will be confounded with the effect of Regulation Fair Disclosure (Reg. FD). Reg. FD was approved on August 10, 2000 and went into effect on October 23, 2000. It requires that all material information be communicated to all investors at the same time to eliminate selective disclosure of information to certain preferred analysts and institutional shareholders. Mohanram and Sunder (2003) and Gu (2005) document a decrease in the precision of common information but an increase in the precision of analysts' private information after the Reg. FD. 80 Figure 4.3 The Event Periods Year-l YearO Year +1 08/2001 08/2002 08/2003 08/2004 This chapter studies two research questions. First, we investigate the information qualities surrounding the S O X reforms. In particular, we examine whether public disclosure improved after the passage of the S O X . Our first hypothesis is that the S O X and related reforms increase the quality of public information. Based on the theory in the previous chapter, which discussed the interaction between public reporting and private information acquisition, we also examine whether the S O X and related reforms increase or decrease the quality of private information. To complete the analysis, since there is a general presumption and hope that S O X and related reforms have increased the quality of total information available in capital markets, we test this hypothesis also. The following regression model is used to investigate these questions. IQt = a0i + auD_ yrO + a2iD_ yrl + a3iControls + ei. (4 .1) where IQi represents the information quality measures (i = public, private, or total). D _ yrO and D _ yrl are time dummies for year 0 and year +1. They are included to capture the impact of S O X in the year immediately after the Act was passed and in 81 the following year. Including D _ yrl allows us to examine whether the impact was permanent. The following control variables are included in the model: firm size, growth, D_loss (equals one if the firm reports a loss), D_miss (equals one if the firm's actual earnings is below the mean analysts forecasts), and D_litigate (equals one if the firm 98 is in an industry with high litigation risk). The prior literature suggests that larger firms and high-growth firms attract more analysts following, and more information is 9Q available on these firms. Performance dummies are included because uncertainty about earnings is higher when earnings performance is poor. The literature typically uses mean analysts forecasts and zero profit as benchmarks to capture earnings performance. As well, existing evidence in the literature shows that firms operating in industries that are subject to higher litigation risks are likely to provide higher quality public information. The second research question explored in this chapter investigates whether characteristics of the board are related to information quality during the governance reform period. The reforms around SOX are comprehensive, affecting both external and internal governance mechanisms. The board of directors is presumed to be an important internal governance mechanism monitoring management and overseeing disclosure policy. The strict rules proposed by SOX and related governance reforms on board structure are intended to enhance the board's monitoring function. The " The control variables are defined more fully later in the chapter. 2 9 We did not control for the number of analysts following a company since we include firm size and growth as control variables, and large firms and high-growth firms are associated with more analysts following. 82 following regression model is used to investigate the board's impact on information quality: IQ, = J30i + fiuD_ yrO + P2iD_ yrl + py Board + PMBoardxD_yrO + P5iBoardxD_yrl + P6iControls + 77,.. (4.2) The regression model aims to capture the effect of the board structure on information quality after controlling for other governance mechanisms, such as financial disclosure regulation implied by the time dummies. The variables which represent the interaction between board characteristics and the time dummies are used to capture the effect of any change in the relationship between board and information quality after the passage of the SOX. Measurement of Information Quality In the previous chapter, we offered a single period model and defined the quality of public (private) information as the extent to which public (private) information reduces prior uncertainty about the terminal value of the firm. In general, a financial analyst's information set about earnings consists of common (public) and idiosyncratic (private) information. These analysts are employed by investment firms and their recommendations are used by investors when making decisions. This chapter uses earnings forecasts of financial analysts to capture the underlying public and private information set available to informed investors in capital markets. 3 0 As Botosan, Plumlee and Xie (2004) argue, the characteristics of analysts' information sets may differ from the information sets used by other informed investors and this difference may introduce measurement error. However, they think that this error would not introduce bias. Barron, Harris and 83 In the literature, Barron, Kim, Lim and Stevens (1998) (hereafter BKLS) derives a measure of unobservable analysts' information properties by using observable analysts forecast data, i.e., forecast dispersion, forecast error, and the number of forecasts. In their theoretical model, N analysts forecast future earnings that is normally distributed. The analysts have a common prior and each receives a private signal which equals the earnings plus noise. The noise terms can be correlated. Based on the key assumption that the precision of the analysts' information is identical, BKLS provide the following formulas for the precision of analysts' public and private information, SE-DIN h = , (4.3) [(1-1/N)D + SE]2 s = T , (4.4) [(1-1/N)D + SE]2 where h and s are the precision of public and private information, respectively. SE is the expected realized squared error in the mean forecast, and D is the expected forecast dispersion. More specifically, SE = E[(Ajt-Fjt)2], (4.5) D = E[-^-it(FijI-Fjl)2], (4.6) where Ajt is the actual earnings for firm j in year t, Fjjt is analyst i's forecast of earnings for firm j in year t, and Fjt is the mean of the forecasts for firm j in year t. In their model, the precision of total information is the inverse of the analysts' Stanford (2003) study this question more carefully and show that the trading behavior of investors is consistent with the changes in the information environment captured by the BKLS measure. 84 total uncertainty, V. The total uncertainty is the sum of two parts: (i) the idiosyncratic uncertainty among analysts (i.e., private), represented by the expected forecast dispersion D, and (ii) the uncertainty common to all analysts (i.e., public), represented by the expected squared error in the mean forecast SE adjusted by the fraction of idiosyncratic uncertainty that cannot be diversified away (i.e., DIN). The total information precision is decomposed into public and private information precision based on the public-to-total precision ratio/?, that is, h = p(l/V) and s = (1 - p)(l/V). Under the BKLS assumptions, the public-to-total precision ratio p , i U 1 . , . . . SE-DIN is equal to the public-to-total uncertainty ratio, i.e., p = . (l-l/N)D + SE When estimating SE and D using the empirical data, it is important to recognize that a data set consists of ex post realizations rather than the ex ante expectations. Substituting the realized values for expected values introduces measurement error into our measures of information quality. Barron, Byard and Kim (2002, p. 828) point out that "such measurement error is ameliorated in large samples." Like many previous studies employing the BKLS measure, such as Mohanram and Sunder (2005), Botosan and Stanford (2005), Christensen, Gaver and Stuerke (2005), Botosan and Plumlee (2004), Barron, Harris and Stanford (2003), Byard and Shaw (2003), and Barron, Byard and Kim (2002), we do not expect it to introduce bias. To implement the BKLS measures, an important feature in the analysis of analyst forecasts literature is the choice of the scaling variable. Generally, the stock price is used. We scale forecasts and actual earnings by the stock price at t-1, and then multiply by 100 (i.e., expressed in percentages). Next, to meaningfully calculate the information measures, we make the following restrictions. An analyst can issue 85 earnings forecasts several times during a single firm-year, we use the most recent forecast before the actual earnings announcement date. This procedure mitigates data interdependence and selects the most up to date forecast. We also require that firms have at least two different analysts produce forecasts for a given year since the denominator for the calculation of the forecast dispersion D would be zero if N = 1. Further, in expectation, all information precisions should be positive. But the estimated measures can sometimes take on negative values. For example, the precision of public information will be negative (h<0) if (SE-D/N)<0, i.e., when the mean forecast error is small relative to the forecast dispersion. It implies that there is little common (public) information. So a zero value is assigned to h for those cases. In addition, since the estimated values of h and s are highly skewed, we take square roots of them as in Gu (2005). Thus, the information quality variables are measured as Pub_IQ = Jh, ¥xi_IQ = Js~, Total_IQ = Jh +s, (4.7) where Pub _IQ, Pr i _ IQ and Total _ IQ represent the quality of public, private and total information under BKLS measurement. Finally, we winsorize all of the information quality variables at the 1% and 99% level to reduce the impact of outliers. It is worth mentioning that the measures rely on the following assumptions: (i) all analysts announce their forecasts simultaneously; (ii) analysts issue unbiased forecasts; (iii) analysts' private information is of equal precision. Ivkovic and Jegadeesh (2004, p. 10) point out, when analysts make forecasts sequentially instead simultaneously, it is not clear whether the BKLS measures are applicable. If 86 analysts' have an optimistic bias, then one of the BKLS assumptions is not maintained, and the measures may not be valid. In addition, Gu (2005) criticizes the assumption of identical precision of private information across all analysts. In section 4.5, we use Gu's proposal of an alternate measure to check the sensitivity of our results. Measurement of Board Characteristics In the corporate governance literature, board characteristics are measured along several dimensions, such as board size, board composition (e.g., insider-outsider ratio), board meetings, and board members with financial backgrounds. We use two board characteristics measures drawn from the literature to proxy for how boards monitor management: board independence measured by the percentage of outside directors, and board meeting frequency.31 A common view is that outside directors are independent and are effective monitors of managerial actions. A body of empirical work finds evidence to support this view, such as, Dechow, Hutton, and Sloan (1995), Beasley (1996), Klein (2002) and Ajinkya, Bhojraj and Sengupta (2005). However, an alternative view is that outside directors may be ineffective if they are appointed by firms' managers, or if the board culture discourages conflict. The evidence from Yermack (1996) and Bhagat and Black (1997) is consistent with this argument. Therefore, the correlation between board independence and the quality of public disclosure is unclear. Board meeting frequency is a proxy for the time that directors devote to monitor management. In expectation, boards that meet 3 1 We also used audit committee independence and meetings. It provides similar results as reported in the chapter. 87 more frequently are more likely to exercise more effective control over firms' disclosure practices. So, a positive relation between number of board meetings and the quality of public disclosure would be expected. On the other hand, boards may meet more often if they need to fix past disclosure flaws or deal with other financial emergencies. For example, Vafeas (1999) finds that boards meet more often after crises. Therefore, it is not clear that more board meetings necessarily lead to better quality public disclosure. Measurement of Control Variables Based on prior research, the following variables are used to control for other possible determinants of the properties of information in capital markets. Ln(ta) = Log of the total assets of a firm at the beginning of the fiscal year. Market-to-book ratio = market value / book value, where, book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt. D_loss = One if a firm reports losses in a given year, and zero otherwise. D_miss = One if a firm's actual earnings per share is below the mean forecast for a given year, and zero otherwise. D_litigate = One if the firm is in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577, 7370-7374), electronics (3600- 3674), and retail (5200-5961) industries, and zero otherwise. " Note that, in general, mean / median forecast is the normal target that researchers use to see if managers try to meet or beat consensus of analysts' forecasts. The measure used here is based on the mean of the forecasts that are used to calculate the information quality variables. 88 Prior literature suggests that there is a positive association between firm size and the information quantity or quality available to investors. In line with the literature, we use the log of total assets to reduce the skewness in firm size. The market-to-book ratio at the end of the fiscal year is used to control for firms' growth opportunities. Existing evidence suggests that there is more information demand for high-growth firms (Barron, Byard, Kile and Reidl, 2002). We do not control for the number of analysts following the firms since large firms and high-growth firms are associated with more analysts following. Earnings have been shown to be more difficult to predict when earnings performance is poor (Mohanram and Sunder 2003, Gu 2005). The literature typically uses dummy variables D_loss and D_miss to capture this feature. Prior work also documents information differences among industries. Firms in the industries that are subject to higher litigation risk, such as, biotechnology, computer, electronics industries, tend to provide higher quality of public information. Therefore, we use D_litigate to control for the litigation effect. 4.4 Empi r ica l Results Data and Sample Description The sample period corresponds to the fiscal year ends from August 2001 to July 2004. The data is obtained from IRRC, Compustat, and I/B/E/S databases. We start with the IRRC corporate governance database, which covers the companies that are included in the S&P 500, Midcap, or Smallcap indexes. The companies in the IRRC database are matched first with the Compustat database and then with financial 3 3 For example, Atiase (1985) shows that larger firms have smaller price reactions to earnings announcements. 89 analyst data retrieved from the I/B/E/S database. Specifically, we use the detailed analyst forecast history and actual earnings per share from I/B/E/S. Table 4.1 Panel A shows that the initial sample of available IRRC and Compustat matched data is 4,086 firm-year observations. To calculate meaningful information quality measures, we select data from the I/B/E/S database which meets the restrictions noted in the previous section. The final sample includes 2,931 firm-years. Panel B of Table 4.1 shows that the sample size increases over time, from 919 firm-years in year -1 to 1,032 firm-years in year +1. Panel C in Table 4.1 presents the summary statistics for the sample. Due to high skewness, large differences exist between the mean and median values with regard to total assets, sales, book value of equity, and market value of equity. For example, while the median total assets (sales) is 2,461 million (1,729 million), the mean value is 15,370 million (6,225 million). But, there is less skewness in the number of analysts following, with a mean of 12.25, and a median of 10. Overall, the sample seems to include larger firms due to the above selection restrictions. Tests and Results Table 4.2 describes the summary statistics of all variables used in the analysis over the whole sample period, including information quality variables, board characteristics, and control variables. Panel A shows the estimated quality of public information has a mean of 6.29 and a median of 2.91, which is lower than the quality of private information which has a mean of 10.07 and median of 4.44. The total information quality aggregates the public and private qualities. The mean total 90 information quality is 13.29, and the median is 7.13. Panel B indicates that small differences exist between the mean and median values of board characteristics variables. The mean value of board independence (and the number of board meetings) is 0.69 (7.26) and the median is 0.71 (7). This suggests that, on average, about 70% of board members are outsiders and they meet seven times per fiscal year. Finally, Panel C shows that 11% of sample firms report losses, 33% of firms report earnings below the mean forecasts and 28% of firms belong to the industries with high litigation risks. The sample firms have an average market-to book ratio of 1.88 and a median of 1.47. We are interested in the trend of information quality surrounding the passage of the SOX. Table 4.3 shows that the public information quality improves from a mean of 6.72 (median of 2.65) in year -1 to 6.95 (3.48) in year 0. The improvement is significant at 1% level using a Wilcoxon test (the Z statistic is 3.19) but it is not significant using a mean test (the t statistic is 0.49). Surprisingly, this improvement is not sustained. In year +1, the mean (median) drops to 5.29 (2.55). This decline is significant at the 1% level for both the mean and Wilcoxon test. While the quality of public information increases in the year immediately after the passage of the SOX, the quality of private information decreases. On average, the quality of private information decreases from 12.25 in year -1 to 10.50 in year 0. The change is significant based on the mean test but is not significant based on the Wilcoxon test. Furthermore, in year +1, the quality of private information continues to decline to a mean of 7.71. 91 The total information aggregates public and private information. From year -1 to year 0, improved public disclosure is offset by the decline in the private information quality, Table 4.3-2 shows that there is no significant change in the median total information quality immediately after the passage of the SOX. Specifically, the median total information quality is 8.00 in year -1 vs. 8.30 in year 0. The Z statistic of Wilcoxon test is 0.69. Consistent with the declines in public and private information quality in year +1, the quality of total information significantly decreases, to a median of 5.60 in year +1. The mean tests show weak evidence (t = -1.73) of a decline in total information in year 0 and significant evidence (t = -5.32) of further decline in year +1. In summary, the univariate analysis indicates that the governance reforms around SOX induced a temporary increase in the quality of public disclosure. The increase in public information is associated with less private information acquisition and hence lower quality of private signals. This could happen if both sets of information are from the same sources. If so, more public information is likely to make it more costly for investors to acquire additional private information. Furthermore, the opposite changes in the public and private information quality results in no significant change in the total information quality immediately following SOX. Another observation is that the improvement in public disclosure is not permanent, lasting only for one year. All three types of information quality drop in year +1. Table 4.4 reports board characteristics by year. Board characteristics increase significantly after the SOX. Specifically, the portion of independent board members increases from an average 0.67 in year -1 to 0.69 in year 0, and then 0.72 in year +1. 92 The t test of the difference in the means is significant at 1% level. Similarly, the median jumps from 0.67 up to 0.71, and then 0.73 which is significant at the 1% level using a Wilcoxon rank test. The results suggest that firms changed their board composition to meet the requirements of SOX and the new NYSE and NASDAQ regulations. The number of board meetings increases slightly across the three periods. The median (mean) board meetings increase from 6 (7.10) in year -1 to 7 (7.26) in year 0. In year +1, the median stays unchanged although the mean increases to 7.41. The changes in the mean and median board meetings from year -1 to year 0 are not significant. However, the change from year -1 to year +1 is significant at the 5% (1%) level using a t-test (Wilcoxon test). In general, firms changed their governance structure under the heightened regulation climate. Table 4.5 displays Pearson and Spearman correlations for variables used in the regression analysis. Results between Pearson and Spearman correlations are similar. Board variables, i.e., board independence and board meetings are positively correlated, but the Pearson correlation coefficient is only 0.10. This suggests that each measure captures a distinct dimension of a board's monitoring process. The public information quality is positively (negatively) correlated with time dummy D_yr0 (D_yrl). Overall, the control variables, i.e., firm size, growth, D_loss D_miss and D_litigate, are strongly related to all three information quality variables as expected. The regression results in Table 4.6 are consistent with the preliminary univariate analyses in Table 4.3. The year 0 dummy variable (D_yr0) has a positive and significant coefficient on the public information quality, and the year +1 dummy 93 variables (D _ yr\) has a negative and significant coefficient. This suggests that the increased scrutiny associated with the introduction of SOX might have a significantly positive impact on firms' public disclosure quality presumably because of the enhanced external and internal governance mechanisms. However, the positive effect is not maintained. The year 0 coefficients in the private and total information regression are insignificant (the t statistics are -0.19 and 0.70 respectively), and the year +1 coefficients are significantly negative. This indicates that the SOX did not improve total information quality in capital markets. One of the goals of the SOX is to motivate information efficiency in capital markets through increased public disclosure. However, the empirical results indicate that even though the quality of public disclosure improved, the total information quality has not improved. As expected, control variables are mostly significant at the 1% level. The changes in information qualities surrounding SOX could be attributed to several simultaneous changes in governance mechanisms. Boards of directors are presumed to play an important role in monitoring firm management. The preliminary analysis in Table 4.4 demonstrates a significant change in board characteristics around the introduction of SOX. Another question raised in this chapter is whether boards of directors contributed to the changes in the information environment around the introduction of SOX. Table 4.7 re-estimates the regressions in Table 4.6 adding board independence and board meetings, both on their own and interacted with time dummy variables. In the public information regression, the coefficient on year 0 continues to be positive and significant but the coefficient on year +1 is no longer negative. Private information quality continues to decline in year +1, however the 94 decline in total information quality in year +1 is no longer significant. The board independence and board meeting variables that are not interacted with the time dummies capture the relation between board characteristics and information quality in the base year, i.e., year -1. The interacted variables indicate how the relationship between information quality and board characteristics differs between year 0 or year +1 and the base year. Most of the board variables are individually insignificant, except for the effects of board independence on private and total information quality in year -1. The effects of board independence on private and total information quality are significantly negative. This suggests that, prior to the introduction of SOX, informed investors acquired less information about firms with more independent boards. The insignificant interacted variables imply that the introduction of SOX did not change this relation. The non-association of board characteristics and public information quality is not surprising since a change in board characteristics does not necessarily lead to a change in actual board monitoring behavior. Effective board monitoring depends on several factors such as the skills of board members, their time and incentives, and the information available to the board. The literature has shown that there is no overall positive relation between the degree of board independence and various measures of firm performance, such as earnings, Tobin's q, and stock price. Moreover, the result suggests that board of directors have not contributed to the improvement of public disclosure after the passage of the SOX, implying that other enhanced governance mechanisms might have played a more dominant role in temporarily improving the quality of public information. 95 We should be cautious in our interpretation of individual coefficients because of omitted variables or the potential of multi-collinearity. Table 4.8 shows that some of the interaction variables are highly correlated. For this reason, we tested whether the year dummies and the board characteristics, including interactions, are jointly insignificant using an F-test (Greene, 2000, p. 283). For both the public (F = 5.27) and total (F = 10.34) information regressions, we strongly reject the hypothesis at the 1% level (critical value F(7, 2917) = 2.66). 4.5 Sensitivity Checks This section uses the information measures introduced in a recent working paper by Gu (2005) to check the robustness of the previous results. While the BKLS information measures are popular, they rely on some rather restrictive assumptions. In particular, the assumption of identical precision of private information across all analysts has been criticized in the recent literature. Gu (2005, p.l) argues that "... analysts' private information can come from their private communication with managers or from individual analytical skills and efforts. Neither needs to be the same across individuals...." By relaxing this assumption, Gu (2005) generalizes the BKLS (1998) measures with the following formulas. »•= W-WE-DW ( 4 8 ) N2[(l-l/N)D + SEf - J>,.2 /=i h = y - h \ fori=l,...,/V, (4.9) where Vi = E[(Ajt - Fijt)2} is analyst i 's expected forecast error for firm j in year t. The formulas show that an individual analyst's forecast error reflects both the public 96 information error and the analysts own private information error. The main difference between the G u and B K L S measures is that G u utilizes individual forecast error Vj to extract the differential properties of private information, and B K L S focus on aggregate forecast properties to extract aggregate properties of the information environment. The B K L S measures are a special case of the G u measures. Similar to B K L S , we replace the expected values with the realized values to re-calculate the information quality measures, based on the G u ( 2 0 0 5 ) formulas. When the estimated measures take on negative values, a zero value is assigned to those measures. If the actual earnings exactly meet a forecast, i.e., (Ajt - Fjjt)2 = 0 , the estimated precision of individual private information goes to infinity, i.e., s, —» °° . For this case, si is assigned a value that is twice as large as the maximum precision of the remaining analysts with finite precision measures.3 4 Consistent with the treatment used for the B K L S measures, we take the square roots of h and s,. The aggregate properties of information are measured as where Pub_Gu, Pr / _Gu and Total _ Gu represent the quality of public, private and total information using the G u measures. A l l information measures are winsorized at the 1% and 9 9 % level to reduce the impact of outliers. Table 4 . 9 reports the summary statistics of information quality measures. Compared with the B K L S measures in Table 4 . 2 , the magnitudes of G u measures are 3 4 Removing such observations yields regression results that are qualitatively similar to those reported in the chapter. ( 4 . 1 0 ) 9 7 higher for all three information quality measures. Specifically, the mean (median) of the public information quality is 6.98 (3.35) vs. 6.29 (2.91) in Table 4.2. Similarly, the private information quality has a mean (median) of 16.44 (9.88) that is higher than under BKLS measurement at 10.07 (4.44). Consequently, there is a big difference in total information quality, 20.31 vs. 13.29 with regard to the mean, and 13.42 vs. 7.13 with regard to the median. While the magnitudes are different, the Gu measures reflect similar time trends to the BKLS measures. Table 4.10-2 shows the median of public information quality increases from 3.24 to 4.13 after the passage of the SOX, but decrease to 2.87 in year +1. The changes are significant at the 1% level. However, as in Table 4.3, the increase in the mean is not significant. Like the BKLS measures, the Gu measures reflect a significant decline of the private information quality from year -1 to year 0 and with a higher level of significance (the t statistic for the difference of mean test is -4.18, and the Wilcoxon Z statistic is -2.52). All information quality levels are decreasing in year +1. Consistent with the univariate analyses, the regression results in Table 4.11 demonstrate that the governance reforms around SOX significantly changed both public and private information quality, but in opposite directions. Public information quality improves (the coefficient of D_yr0 is 1.21 with a t statistic of 2.84), but private information quality significantly declines (the coefficient of D _ yrO is -2.18 with a t statistic of -2.62). As a result, the change in total information quality is insignificant. When board characteristics are included in the information quality regressions in Table 4.12, the results are qualitatively the same as for the BKLS 98 measures. That is, we still fail to find a significant positive relation between board independence (board meet frequency) and the public information quality. However, the significant negative relations between board independence and both private and total information reported in Table 4.7 are no longer significant in Table 4.12. Overall, both information measures provide consistent results. 4.6 Concluding Remarks The purpose of this chapter is to examine how corporate governance reforms surrounding the SOX affect the quality of information in capital markets. We investigate information quality at all three levels: public, private, and total information. The theory suggests that tightened corporate governance has an ambiguous effect on the total information, because while public disclosure is expected to increase, private information acquisition could either increase or decrease as a result. To carry out the empirical tests, we calculate information quality measures using the BKLS (1998) and Gu (2005) methods. We find that the events associated with the introduction of SOX temporarily improved public disclosure quality, but the improvement is not maintained after the first year. While the precision of public information increases with the introduction of SOX, the precision of private information declines. As a result, there is no significant change of total information quality in the period immediately after SOX. Since one of the goals of SOX is to promote information efficiency through enhanced public disclosure, the results in the chapter show no evidence that SOX achieves this goal in the longer term. 99 Furthermore, the analyses of governance structure reports a significant change in board characteristics after the SOX. Overall, boards became more independent and met more frequently. However, the change in board independence has no significant effect on firms' public disclosure quality. Many of the reforms introduced as a result of SOX are intended to increase the quality of public information. For this reason, our finding that the change in public information quality does not persist is noteworthy. The increase in public information quality in year 0, but decrease in year +1, may be because media coverage of the scandals and the failure of Arthur Andersen during 2002 changed management's awareness of the issue, rather than the actual passage of SOX. Writing at the time, Cunningham (2002, p. 19) argues "The changes are more likely to have psychological rather than substantive effects." Our results suggest that the introduction of the SOX had a short term impact on public information quality, but the regulated changes in board structure may not have resulted in any lasting improvement in disclosure. Like previous studies, the measures that we use are less than ideal. For example, our board meeting measure does not distinguish between whether an increase in meetings is the result of increased scrutiny or is needed to fix past disclosure flaws, or to deal with other financial emergencies. As Vafeas (1999) notes, boards meet more often after crises. The measures are also crude in the sense that it may be difficult to distinguish between the timing of information about the Act and firm-specific effects (Jain and Razaee, 2004). The early 2000s were a time of great change for many firms. Thus, our regressions reveal correlations but not necessarily causation. 100 The fact that our results in year 0 do not repeat in year +1 implies that, as more data becomes available, an important question for future research would be to consider which of the new procedures initiated by SOX have lasting effects. We do not study the costs of complying with SOX. These costs may explain why some firms "went dark" (Ribstein, 2005, p. 17- 18) in the second year (i.e., year +1) after the passage of SOX and this change may explain the decrease in information quality between year 0 and year +1. A firm can choose to go dark, i.e., be a public company without having to satisfy the Securities and Exchange Act of 1934, if the number of nominal shareholders falls below 300. Ribstein (2005) notes that 200 firms went dark in 2003 and that 44 of the 114 firms that went private cited the costs of complying with SOX as a reason. Since our data does not allow us to identify which firms went dark, we cannot test for the significance of this decrease in disclosure. 101 Figure 4.1: Timeline of major accounting scandals in 2001- 2002 Month Company involved in the accounting scandal October 2001 Enron November 2001 Arthur Anderson 2002 January 2002 Kmart; Homestore.com February 2002 Global Crossing; Qwest March 2002 Xerox April 2002 A O L Time Warner; Bristol-Myers Squibb; Duke Energy; Merck; Mirant; Nicor Energy L L C May 2002 WorldCom June 2002 Adelphia July 2002 CMS Energy; Dynergy; El Paso; Halliburton; Peregrine Systems; Reliant Energy; Tyco Sources: Cohen, Dey and Lys (2005); www.forbes.com. 102 Figure 4.2: Timeline of major government events Month Event December 2001 US Congress holds hearings on corporate scandals 2002 March 2002 President Bush offers "Ten-Point Plan" to improve corporate responsibility April 2002 House of Representatives passes a bill June 2002 Lots of discussions concerning various drafts of a bill July 2002 President Bush signs Sarbanes-Oxley Act into law August 2002 CEOs and CFOs must certify annual and quarterly reports, NYSE changes rules to strengthen independence of directors and corporate governance practices October 2002 Appointment of members of the Public Company Accounting Oversight Board (PCAOB) NASDAQ offers rules to strengthen independence of directors and corporate governance practices 2003 January 2003 Responsibilities for auditors increased, NYSE and NASDAQ require regular meetings of non-management directors without management present, NYSE requires the nominating committee and the compensation committee of Board of Directors to be composed entirely of independent directors, NASDAQ requires majority of board members be independent April 2003 SEC adopts rules increasing independence of Boards of Directors May 2003 Mandatory rotation of lead audit partner July 2003 Audit committee required to have person with financial and accounting expertise October 2003 Enhanced conflict of interest rules Sources: www.businessweek.com/adsections/2003/pdf/0331so.pdf accessed on Aug. 9, 2006; www.forbes.com accessed on Aug. 9, 2006; http://www.ffhsj.com/cmemos/020802_sarb_timeline.pdf accessed on Aug. 9, 2006; www.goodwinprocter.com/pdfs/S_0_tirneline.pdf accessed on Aug. 9, 2006; Romano (2005). 103 Table 4.1 Sample selection and descriptive statistics The sample includes firm-years in the period of August 2001- July 2004 that have sufficient IRRC, I/B/E/S, and Compustat data to meaningfully estimate the variables used in the regressions. Panel A: Sample selection Criterion Number of firm years IRRC data and Compustat data available 4,086 I/B/E/S data that meet the requirements to meaningfully estimate the information quality 2,931 variables* Final sample 2,931 * Each analyst may issue earnings forecasts several times for a firm-year, we restrict to the most recent forecast before the actual earnings announcement. We also require that the forecasts are made within 180 days prior to the announcement of actual earnings. To meaningfully estimate the information quality variables, firms are kept if they have at least two different analysts produce forecasts for a given year. Panel B: Sample distribution over time Event year -1 0 +1 Total Number of firms 919 980 1,032 2,931 Panel C: Descriptive statistics for sample firms (n = 2,931) Mean Std. Min. Q l Med. Q3 Max. Total assets 15,370 64,992 28 825 2,461 9,397 1,264,032 Sales (net) 6,225 16,158 19 662 1,729 5,253 257,157 Book value of equity 2,772 6,476 -4,734 389 893 2,371 96,889 Market value of equity 8,734 25,577 43 861 2,154 6,429 397,831 Analysts following 12.25 7.92 2 6 10 17 51 * All variables except analyst following are measured in $ millions. Analysts following are the number of unique analysts issuing annual earnings forecasts in a firm-year. 104 Table 4.2 Descriptive statistics for variables used in the analyses This table reports the descriptive statistics of board characteristics, information quality and control variables for the sample of 2,931 firm-years used in the analyses. Mean Std. Min. Q l Med. Q3 Max. Panel A: Information quality variables Public information quality 6.29 9.24 0 0.57 2.91 7.85 65.89 Private information quality 10.07 15.54 0 1.48 4.44 12.03 125.0! Total information quality 13.29 17.71 0.14 3.01 7.13 16.22 131.6 Panel B: Board characteristics variables Board independence 0.69 0.15 0.13 0.60 0.71 0.82 1 Board meetings 7.26 3.06 0 5 7 9 36 Panel C: Control variables Firm size 7.90 1.66 0.30 6.65 7.74 9.02 13.91 Growth 1.88 1.14 0.40 1.14 1.47 2.14 11.10 D_loss 0.11 0.31 0 0 0 0 1 D_miss 0.33 0.47 0 0 0 1 1 D_litigate 0.28 0.45. 0 0 0 1 1 Variable measurement (all variables are measured for a firm-year): Board independence - Percent of independent directors on the board; Number of regular and special board meetings held through a year; Log of the total assets of a firm at the beginning of a given year; The market-to-book ratio, where book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt; One if the firm report losses in a given year, and zero otherwise; One if a firm's actual earnings is below the mean forecast for a given year, and zero otherwise; One if the firm in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577, 7370-7374), electronics (3600-3674), and retail (5200-5961) industries, and zero otherwise; are measured based on BKLS (1998) model. See the main text for the detailed variable calculations. Board meetings = Firm size = Growth = DJoss = D_miss= DJitigate = Information quality variables: 105 Table 4.3 Univariate comparisons of information quality variables surrounding SOX This table compares the mean and median values for information quality variables surrounding SOX. T statistics are from pooled difference of mean test, and Z statistics are for Wilcoxon rank test. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively (two-tailed). Table 4.3 -1 Univariate comparision of the mean value of information quality Year-1 YearO Year +1 T stat. T stat. T stat. mean mean mean for diff. for diff. for diff. 1 2 3 2&1 3&2 3&1 Sample size 919 980 1032 Public information quality 6.72 6.95 5.29 0.49 -4.47*** -3.44*** Private information quality 12.25 10.50 7.71 -2.21** -4.81*** -6.29*** Total information quality 15.62 14.07 10.47 -1.73* -5.32*** -6.37*** Table 4.3 -2 Univariate comparison of the median value of information quality Year-1 YearO Year+1 Wilcoxon Wilcoxon Wilcoxon median median median Z stat. Z stat. Z stat. 1 2 3 2&1 3&2 3&1 Sample size 919 980 1032 Public information quality 2.65 3.48 2.55 3 [9 * * * -3.84*** -0.14 Private information quality 5.43 5.34 3.16 -0.44 -6.34*** -6 4 9 * * * Total information quality 8.00 8.30 5.60 0.69 -6.54*** -5 42*** Variable measurement (all variables are measured for a firm-year): Information quality variables: are measured based on BKLS (1998) model. See the main text for the detailed variable calculations. 1 0 6 Table 4.4 Univariate comparisons of board characteristics variables before and after SOX This table compares the mean and median value for board characteristics variables before and after SOX. T statistics are from pooled difference of mean test, and Z statistics are for Wilcoxon two-sample rank test. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively (two-tailed). Table 4.4 -1 Univariate comparision of the mean values of board variables Year-1 mean 1 YearO mean 2 Year +1 mean 3 T stat. for diff. 2&1 T stat. for diff. 3&2 T stat. for diff. 3&1 Sample size 919 980 1032 Board independence 0.67 0.69 0.72 3.68*** 3.30*** 7 03*** Board meetings 7.10 7.26 7.41 1.17 1.10 2.27** Table 4.4 -2 Univariate comparison of the median values of board variables Year-1 YearO Year+1 Wilcoxon Wilcoxon Wilcoxon median median median Z stat. Z stat. Z stat. 1 2 3 2&1 3&2 3&1 Sample size 919 980 1032 Board independence 0.67 0.71 0.73 3.35*** 2.64*** 5.98*** Board meetings 6 7 7 1.48 1.02 2.48** Variable measurement (the variables are measured for a firm-year): Board independence = Percent of independent directors on the board; Board meetings = Number of regular and special board meetings held through a year. 107 Table 4.5 Pearson (Spearman) correlations above (below) the diagonal This table displays Pearson (spearman) correlations for the sample of 2,931 firm-years in the period August 2001-July 2004. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively (two-tailed). Public Private Total Board Board Firm D_ D_ D_ info. info. info. D^yrO D_yrl inde. meet Size Growth loss miss litigate Public info. 0.42 0.65 0.05 -0.08 -0.07 -0.11 -0.01 0.35 -0.17 -0.16 0.07 *** *** *** *** *** *** *** *** *** *** Private info. 0.29 0.94 0.02 -0.11 -0.07 -0.07 0.08 0.31 -0.16 -0.02 0.05 *** *** *** *** *** *** *** *** ** Total info. 0.52 0.91 0.03 -0.12 -0.08 -0.10 0.06 0.34 -0.19 -0.06 0.05 *** *** * *** *** *** *** *** *** *** *** D_yr0 0.08 0.07 0.08 -0.52 0.00 -0.00 -0.01 -0.13 0.03 0.01 0.00 *** *** *** *** *** * £>_yr7 -0.04 -0.14 -0.13 -0.52 0.11 0.04 0.01 0.05 -0.05 -0.05 -0.01 ** *** *** *** *** * *** *** ** Board inde. -0.09 -0.07 -0.09 0.01 0.09 0.10 0.16 -0.08 0.02 0.00 0.10 *** *** *** *** *** *** *** *** Board meet -0.13 -0.08 -0,13 0.00 0.04 0.11 0.22 -0.14 0.08 0.05 -0.01 *** *** *** ** *** *** *** *** *** Firm size -0.01 0.09 0.05 -0.01 0.01 0.17 0.23 -0.24 -0.16 -0.01 -0.22 *** *** *#* *** *** *** *** Growth 0.31 0.35 0.42 -0.16 0.11 -0.06 -0.18 -0.32 -0.09 -0.08 0.29 *** *** *** *** *** *** *** *** *** *** *** D_loss -0.21 -0.28 -0.33 0.03 -0.05 0.01 0.08 -0.16 -0.13 0.15 0.11 *** *** *** * *** *** *** *** *** *** D_miss -0.23 -0.06 -0.13 0.01 -0.05 -0.00 0.04 -0.01 -0.11 0.15 -0.04 *** *** *** ** ** *** *** ** DJtitigate 0.09 0.05 0.06 0.00 -0.01 -0.01 -0.01 -0.22 0.31 0.11 -0.04 *** *** *** *** *** *** *** ** * D_yr0 and D_yrl are dummy variables for year 0 and +1. See Table 4.2 for the detailed definitions of other variables. 108 Table 4.6 Ordinary least squares regressions examining the impact of SOX on the information quality in capital markets This table reports results from the following regression: IQt - a0i + auD _ yrO + a2iD _ yr\ + a}iSize + aMGrowth + a5iD _ loss + CX6iD _ miss + OCliD _ litigate + ei Public info. Private info. Total info, quality quality quality Intercept -0.21 -7.31*** -5.97*** (-0.21) (-4.44) (-3.26) Time dummy variables D_yr6» 1.15*** -0.13 0.51 (2.94) (-0.19) (0.70) Djyrl -1.56*** -4.51*** -5.22*** (-4.08) (-6.94) (-7.23) Control variables Firm size 0.30*** I 34*** 1 34*** (3.06) (7.95) (7.18) Growth 2 93*** 6.21*** (19.73) . (18.83) (22.20) D_loss -3.40*** -5.89*** -7.69*** (-6.56) (-6.68) (-7.86) D_miss -2.26*** 0.85 -0.65 (-6.72) (1.48) (-1.02) D_litigate -0.43 -0.42 -1.11 (-1.14) (-0.66) (-1.58) Sample size 2931 2931 2931 F-statistic 86.87 76.19 102.28 Adjusted R 2 17.02% 15.23% 19.48% * Figures in brackets are -^statistics for OLS regressions. One-sided (two-sided) critical f-values for significance levels of 10%, 5%, and 1% are 1.28, 1.65, 2.33 (1.65, 1.96, 2.58), respectively. Variable measurement (all variables are measured for a firm-year): D_yr0--Djyrl •• Firm size • Growth •• D_loss -D_miss= D_litigate = Information quality IQt One if year 0, and zero otherwise; One if year +1, and zero otherwise; Log of the total assets of a firm at the beginning of a given year; The market-to-book ratio, where book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt; One if the firm report losses in a given year, and zero otherwise; One if a firm's actual earnings is below the mean forecast for a given year, and zero otherwise; One if the firm in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577,7370-7374), electronics (3600-3674), and retail (5200-5961) industries, and zero otherwise; are measured based on BKLS (1998). See the main text for the detailed variable calculations. 109 Table 4.7 The association between board characteristics and information quality This table reports results from the following regression: /_>. = p0i + PUD _ yrO + p2iD _ yr\ + pvInde. + pMMeet. + P5iInde. X D _ yrO + P6iMeet. X D _ yrO + PltInde. xD_yrl + PSiMeet. x D _ yrl + fi9iSize + pmGrowth + /?, UD _ loss + Pl2iD _ miss + Pl3iD _ litigate + 77, Public info. Private info. Total info. Quality Quality Quality Intercept -0.15 -2.63 -2.14 (-0.10) (-1.02) (-0.75) Time dummy variables D_yrO 4.76** -1.90 1.46 (2.56) (-0.60) (0.42) D_yrl 1.27 -6.29* -4.57 (0.65) (-1.90) (-1.24) Board characteristics variables Board independence -0.87 _7 79*** -6.83** (-0.52) (-2.71) (-2.14) Board meetings -0.02 -0.08 -0.09 (-0.25) (-0.50) (-0.52) Board independences D_yrO -3.31 5.96 2.83 (-1.36) (1.44) (0.62) Board meetings x D_yrO -0.18 -0.30 -0.38 (-1.39) (-1.36) (-1.55) Board independencex D_yrl -1.96 4.14 1.77 (-0.77) (0.96) (0.37) Board meetings x D_yrl -0.19 -0.10 -0.21 (-1.47) (-0.48) (-0.87) Control variables Firm size 0.40*** I 49*** I 54*** (3.91) (8.59) (7.97) Growth 2.86*** 4.68*** 6.13*** (19.42) (18.54) (21.86) D_loss -3.21*** -5.52*** -7.25*** (-6.17) (-6.24) (-7.38) Djniss -2.24*** 0.87 -0.61 (-6.65) (1.53) (-0.95) DJitigate -0.43 -0.48 -1.16* (-1.15) (-0.77) (-1.65) Sample size 2931 2931 2931 F-statistic 48.56 42.51 67.19 Adjusted R 2 17.42% 15.55% 19.90% * Figures in brackets are /-statistics for OLS regressions. One-sided (two-sided) critical /-values for significance levels of 10%, 5%, and 1% are 1.28, 1.65, 2.33 (1.65, 1.96, 2.58), respectively. 110 Variable measurement in Table 4.7 (all variables are measured for a firm-year): D_jrO-D^rl: Board independence • Board meetings • Firm size • Growth • D_loss = D miss= DJitigate -Information quality lQi: One if year 0, and zero otherwise; One if year +1, and zero otherwise; Percent of independent directors on the board; Number of regular and special board meetings held through a year; Log of the total assets of a firm at the beginning of a given year; The market-to-book ratio, where book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt; One if the firm report losses in a given year, and zero otherwise; One if a firm's actual earnings is below the mean forecast for a given year, and zero otherwise; One if the firm in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577, 7370-7374), electronics (3600-3674), and retail (5200-5961) industries, and zero otherwise; are measured based on BKLS (1998). See the main text for the detailed variable calculations. I l l Table 4.8 Correlations between time dummies and board interaction variables This table reports Pearson (Spearman) correlations above (below) the diagonal for the time dummies and board interaction variables. All correlations are significant at < 0.0001 level. Board Board Board Board D_yr0 D_yrl inde. meet. inde. meet. xD_yrO xD_yrO xD_yrl xD_yrl D^rO -0.52 0.97 0.89 -0.51 -0.46 D_yri -0.52 -0.50 -0.46 0.97 0.88 Board inde. xD_yr0 0.97 -0.51 0.87 -0.49 -0.45 Board meet. xD_yr0 0.97 -0.51 0.95 -0.45 -0.41 Board inde. xD_yrl -0.51 0.97 -0.49 -0.49 0.87 Board meet. xD_yri -0.51 0.97 -0.49 -0.49 0.94 Variable measurement (all variables are measured for a firm-year): D_yr0 - One if year 0, and zero otherwise; D_yrl = One if year +1, and zero otherwise; Board independence - Percent of independent directors on the board; Board meetings = Number of regular and special board meetings held through a year; 112 Table 4 . 9 Descriptive statistics for information quality variables - Gu measures This table reports the descriptive statistics of information quality variables based on Gu measures for the sample of 2,931 firm-years. Mean Std. Min. Q i Med. Q3 Max. Public information quality 6.98 10.06 0 0.72 3.35 8.87 70.96 Private information quality 16.44 19.74 0 2.82 9.88 22.75 143.91 Total information quality 20.31 21.57 0.22 5.60 13.42 27.46 143.91 * Information quality variables are measured for a firm-year. They are calculated by using Gu (2005) model. See the main text for the detailed variable calculations. 113 Table 4.10 Univariate comparisons of information quality variables surrounding SOX - Gu measures This table compares the mean and median value for information quality variables based on Gu measures surrounding SOX. T statistics are from pooled difference of mean test, and Z statistics are for Wilcoxon rank test. • *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively (two-tailed). Table 4.10-1 Univariate comparision of the mean value of information quality Year -1 mean • 1 YearO mean 2 Year +1 mean 3 T stat. for diff. 2&1 T stat. for diff. 3&2 T stat. for diff. 3&1 Sample size 919 980 1032 Public information quality 7.48 7.72 5.82 0.46 -4.68*** -3.67*** Private information quality 20.82 16.68 12.32 -4.18*** -5 97*** -9 31*** Total information quality 24.83 20.96 15.68 -3.60*** -6.48*** -9 23*** Table 4.10 -2 Univariate comparison of the median value of information quality Year-1 YearO Year+1 Wilcoxon Wilcoxon Wilcoxon median median median Z stat. Z stat. Z stat. 1 2 3 2&1 3&2 3&1 Sample size 919 980 1032 Public information quality 3.24 4.13 2.87 3.02*** -4.23*** -0.68 Private information quality 12.83 11.06 6.71 -2.52** -6.11*** -8.17*** Total information quality 16.11 15.02 10.21 -1.67* .7.17*** -8.24*** * Information quality variables are measured for a firm-year. They are calculated by using Gu (2005) model. See the main text for the detailed variable calculations. 114 Table 4.11 Ordinary least squares regressions examining the impact of SOX on the information quality - Gu measures This table reports results from the following regression: IQj _Gu-a0j +aljD_ yrO + a2jD_ yrl + a3jSize + a4JGrowth + a5jD _loss + a6JD _miss + CtljD _litigate + £j Public info. Private info. Total info. quality Quality Quality Intercept 0.55 -10.33*** -7.54*** (0.52) (-5.01) (-3.43) Time dummy variables D^rO 1 2 i * * * -2.18*** -1.38 (2.84) (-2.62) (-1.56) D^rl -1.81*** -8.48*** -9 23*** (-4.32) (-10.42) (-10.64) Control variables Firm size 027*** 2.60*** 244*** (2.49) (12.31) (10.87) Growth 3 11*** 5.60*** 7 30*** (19.19) (17.74) (21.70) DJtoss -3.84*** -8.54*** -11.01*** (-6.78) (-7.74) (-9.37) D_miss -2.35*** 1 95*** 0.53 (-6.38) (2.72) (0.69) DJtitigate -0.45 -0.82 -1.40* (-1-10) (-1.04) (-1.67) Sample size 2931 2931 2931 F-statistic 83.67 91.34 117.50 Adjusted R 2 16.49% 17.75% 21.77% * Figures in brackets are /-statistics for OLS regressions. One-sided (two-sided) critical r-values for significance levels of 10%, 5%, and 1% are 1.28, 1.65, 2.33 (1.65, 1.96, 2.58), respectively. Variable measurement (all variables are measured for a firm-year): D_yr0 = One if year 0, and zero otherwise; D_yrl = One if year +1, and zero otherwise; Firm size = Log of the total assets of a firm at the beginning of a given year; Growth = The market-to-book ratio, where book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt; DJoss = One if the firm report losses in a given year, and zero otherwise; D_miss= One if a firm's actual earnings is below the mean forecast for a given year, and zero otherwise; DJitigate = One if the firm in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577, 7370-7374), electronics (3600-3674), and retail (5200-5961) industries, and zero otherwise; Information quality IQ _Gu: a r e measured based on Gu (2005). See the main text for the detailed ' variable calculations. 115 Table 4.12 The association between board characteristics and information quality - Gu measures This table reports results from the following regression: IQj-Gu = 60J + px.D_yrG+ P2jD_yr\ + P3jInde. + P4jMeet. + P5j Inde. xD_ yrO + P6jMeet. xD_ yrO + P7j Inde. xD_yr\ + P%Meet. xD _yr\ + Pqj Size + Pi0JGrowth + PnjD _ loss + PX2jD _ miss + PnjD _ litigate + 7]. Public info. Private info. Total info. quality quality Quality Intercept 0.53 -8.86*** -6.25* (0.32) (-2.74) (-1.81) Time dummy variables D^yrO 5.03** -1.71 0.84 (2.47) (-0.43) (0.20) D_yrl 1.26 -5.84** -4.86 (0.59) (-1.40) (-1.10) Board characteristics variables Board independence -0.75 -3.43 -3.39 (-0.41) (-0.95) (-0.88) Board meetings -0.02 0.01 -0.02 (-0.23) (0.03) (-0.10) Board independencex D_yrO -3.77 3.44 1.38 (-1.42) (0.66) (0.25) Board meetings x D_yrO -0.16 -0.38 -0.43 (-1.16) (-1.39) (-1.45) Board independencex D_yrl -2.33 -1.84 -3.03 (-0.84) (-0.34) (-0.52) Board meetings x D_yrl -0.18 -0.15 -0.27 (-1.32) (-0.58) (-0.95) Control variables Firm size 0.36*** 2 70*** 2.60*** (3.25) (12.42) (11.20) Growth 3 07*** 5.55*** 7 23*** (18.91) (17.52) (21.42) D_loss -3.65*** -8.26*** -10.65*** (-6.43) (-7.45) (-9.02) D_miss -2 32*** 1.96*** 0.55 (-6.31) (2.73) (0.73) DJitigate -0.45 -0.84 -1.41* (-1.11) (-1.06) (-1.67) Sample size 2931 2931 2931 F-statistic 46.54 49.75 64.19 Adjusted R 2 16.81% 17.78% 21.90% * Figures in brackets are f-statistics for OLS regressions. One-sided (two-sided) critical r-values for significance levels of 10%, 5%, and 1% are 1.28, 1.65, 2.33 (1.65, 1.96, 2.58), respectively. 116 Variable measurement in Table 4.12 (all variables are measured for a firm-year): D_yrO = D_yrl = Board independence = Board meetings = Postsox= Firm size -Growth = DJtoss -D miss= DJtitigate = Information quality lQj _ Gu : One if year 0, and zero otherwise; One if year +1, and zero otherwise; Percent of independent directors on the board; Number of regular and special board meetings held through a year; One if year 0 or year +1, and zero otherwise; Log of the total assets of a firm at the beginning of a given year; The market-to-book ratio, where book value is the book value of total assets in a given year, and market value is the market value of equity plus book value of debt; One if the firm report losses in a given year, and zero otherwise; One if a firm's actual earnings is below the mean forecast for a given year, and zero otherwise; One if the firm in the biotechnology (four-digit SICs are 2833-2836, 8731-8734), computer (3570-3577, 7370-7374), electronics (3600-3674), and retail (5200-5961) industries, and zero otherwise; are measured based on Gu (2005). See the main text for the detailed .variable calculations. 117 Bibliography Ajinka, B., S. Bhojraj and P. Sengupta. 2005. The association between outside directors, institutional investors and the properties of management earnings forecasts. Journal of Accounting Research 43 (3): 343- 376. Arya, A., J. Glover and S. Sunder. 1998. Earnings management and the revelation principle. Review of Accounting Studies 3: 7-34. Bainbridge, S. M . , 2000. Insider trading: An overview. Working paper, University of California- Los Angeles. Banks, E. 2004. Corporate Governance, Financial Responsibility, Controls and Ethics. Basingstoke, New York: Palgrave Macmillan. Barron, O., D. Byard and O. Kim. 2002. Changes in analysts' information around earnings announcements. The Accounting Review 11 (4): 821- 846. Barron, O., D. Byard, C. Kile and E. Riedl. 2002. High-technology intangibles and analysts' forecasts. Journal of Accounting Research 40 (2): 289- 312. Barron, O., O. Kim, S. Lim and D. Steven. 1998. Using analysts' forecasts to measure properties of analysts' information environment. The Accounting Review 73 (4): 421-433. Baums, T. and K. E. Scott. 2003. Taking shareholder protection seriously? Corporate governance in the United States and Germany. Working paper, European Corporate Governance Institute; Stanford Law School. Bebchuk, L. A., and O. B. Gill. 2003. Misreporting corporate performance. Working paper, Harvard Law School and NBER. Bebchuk, L. A. and J. M . Fried. 2003. Executive compensation as an agency problem. Journal of Economic Perspectives 17 (3): 71-92. Becker, G. S. 1993. Nobel lecture: The economic way of looking at behavior. The Journal of Political Economy 101(3): 385-409. Beneish M . D. 2001. Earnings management: A perspective. Managerial Finance 21 (12): 3-17. Benston, G., M . Bromwich, R.E. Litan and A. Wagenhofer. 2003. Following the Money: The Enron Failure and the State of Corporate Disclosure. AEI-Brookings Joint Center for Regulatory Studies, Washington, D.C.. 118 Bhagat, S., and B. S. Black, 1997. Do independent directors matter? Working paper, Columbia University. Botosan, C. A., M . A. Plumlee and Y. Xie. 2004. The role of information precision in determining the cost of equity capital. Review of Accounting Studies 9: 233- 259. Bryan, S., and Lilien. 2006. Characteristics of firms with material weaknesses in internal control: An assessment of Section 404 of Sarbanes Oxley. Working paper, Baruch College. Bushman, R. M . , and A. Smith, 2001. Financial accounting information and corporate governance Journal of Accounting and Economics 32: 237- 333. Carter, M . E. , L. Lynch and S. Zechman. 2005. The relationship between executive compensation and earnings management: Changes in the post- Sarbanes- Oxley era. Working paper, Wharton School University of Pennsylvania. Christensen, P. O., and G. A. Feltham. 2000. Market performance measures and disclosure of private management information in capital markets. Review of Accounting Studies 5: 301- 329. Christensen, P. O., and G. A. Feltham. 2003. Economics of Accounting: Volume I -Information in Markets. Boston: Kluwer Academic Publishers. , and . 2005. Economics of Accounting: Volume II - Performance Evaluation. New York: Springer. Christensen, P. O., G. A. Feltham, C; Hofmann and F. Sabac. 2003. Timeliness, accuracy, and relevance in dynamic incentive contracts. Working paper, University of British Columbia. Christensen, P. O., G. A. Feltham and F. Sabac. 2005. A contracting perspective on earnings quality. Journal of Accounting and Economics 39 (2): 265-294. Christensen, P. O., H. Frimor and F. Sabac. 2004. Earnings management, revenue recognition, and depreciation in dynamic agency with renegotiation. Working paper, University of Alberta. Cohen, D., A. Dey and T. Lys. 2005. Trends in earnings management in the pre- and post- Sarbanes-Oxley Periods. Working paper, Northwestern Unversity. Dechow, P. M . , and D. J. Skinner. 2000. Earnings management: Reconciling the views of accounting academics, practitioners, and regulators. Accounting Horizons 14 (2): 235-250. 119 DeFond, M . , R. Hann, and X. Hu. 2004. Does the market value financial expertise on audit committees of Board of Directors. Working paper, University of Southern California. Demski, J. S. 1998. Performance measure manipulation. Contemporary Accounting Research 15 (3): 261-285. Demski, J. S., and G. A. Feltham. 1994. Market response to financial reports. Journal of Accounting and Economics 17: 3-40. Dutta, S., and S. Reichelstein. 2003. Leading indicator variables, performance measurement and long-term versus short-term contracts. Journal of Accounting Research 41 (3): 837-866. Dye, R. A. 1988. Earnings management in an overlapping generation model. Journal of Accounting Research 26 (2): 195-235. Ewert R., and A. Wagenhofer. 2005. Economic effects of tightening accounting standards to restrict earnings management. The Accounting Review 80 (4): 1101-1124. Evans, J. H., and S. S. Sridhar. 1996. Multiple control systems, accrual accounting, and earnings management. Journal of Accounting Research 34 (1): 45-65. Feltham, G. A., R. Indjejikian and D. Nanda. 2004. Dual purpose measures. Working paper, University of British Columbia. Feltham, G. A., and J. Pae. 2000. Analysis of the impact of accounting accruals on earnings uncertainty and response coefficients. Journal of Accounting, Auditing & Finance 15 (3): 199-220. Feltham, G. A., and M . G. H. Wu. 2000. Public reports, information acquisition by investors, and management incentives. Review of Accounting Studies 5: 155-190. Feltham, G. A., and M . G. H. Wu. 2001. Incentive efficiency of stock versus options. Review of Accounting Studies 6: 7- 28. Feltham, G. A., and J. Xie. 1994. Performance measure congruity and diversity in multi-task principal/agent relations. The Accounting Review 69 (3): 429-453. Fields, M . A., and P. Y. Keys. 2003. The emergence of corporate governance from Wall St. to Main St.: Outside directors, board diversity, earnings management, and managerial incentives to bear risk. The Financial Review 38: 1-24. Fields T. D., T. Z. Lys and L. Vincent. 2001. Empirical research on accounting choice. Journal of Accounting and Economics 31: 255-307. 120 Fudenberg, D., and J. Tirole. 1990. Moral hazard and renegotiation in agency contracts. Econometrica 58 (6): 1279-1319. , and . 1995. A theory of income and dividend smoothing based on incumbency rents. The Journal of Political Economy 103 (1): 75-93. Greene, W. H. 2000. Econometric Analysis. Upper Saddle River, New Jersey: Prentice Hall. Goel, A. M . , and A. V. Thakor. 2003. Why do firms smooth earnings? Journal of Business 76(1): 151-192. Goldman E. , and S. L. Slezak. 2003. The economics of fraudulent misreporting. Working paper, University of North Carolina, Chapel Hill. Grossman, S. J., and J. E. Stiglitz. 1980. On the impossibility of informationally efficient markets. American Economic Review 70: 393-408. Gu, Z. 2005. Measuring the precision of analysts private and common information: Generalization and an application. Working paper, Carnegie Mellon University. Healy, P. M . 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics 7: 85-107. Healy, P. M . , and J. M . Wahlen. 1999. A review of the earnings management literature and its implications for standard setting. Accounting Horizons 13 (4): 365-383. Hermalin, B. E. , and M . S. Weisbach. 1998. Endogenously chosen boards of directors and their monitoring of the CEO. The American Economic Review 88(1): 96-118. , and . 2003. Boards of directors as an endogenously determined institution: A survey of the economic literature. Economic Policy Review 9 (1): 7-26. Holmstrom, B., and S. N. Kaplan. 2003. The state of U.S. corporate governance: What's right and what's wrong? Working paper, the University of Chicago. Huddart, S., J. S. Hughes and M . Williams. 2004. Pre-announcement of insiders' trades. Working paper, Pennsylvania State University. Hughes, J., J. Liu and J. Liu. 2005. Information, diversification and asset pricing. Working paper, University of California- Los Angeles. Insider trading, http://www.sec.gov/. 121 Issaevitch, T. A. 2006. Insider trading, earnings management and informational efficiency. Working paper, Columbia University. Ivkovich, Z., and N. Jegadeesh, 2004. The Timing and Value of Forecast and Recommendation Revisions: Do Analysts Receive Early Peek at Good News? Working Paper, University of Illinois- Urbana Champaign. Jain, P., and Z. Rezaee, 2005. The Sarbanes-Oxley Act of 2002 and security market behavior: Early evidence. Working paper, University of Memphis. Jensen, M . , and W. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3: 305-360. Karamanou, I., and N. Vafeas. 2005. The association between corporate boards, audit committees, and management earnings forecasts: An empirical analysis. Journal of Accounting Research 43 (3): 453- 486. Kim, O., and R. E. Verrecchia. 1994. Market liquidity and volume around earnings announcements. Journal of Accounting and Economics 17: 41-67. Kirschenheiter, M . , and N. D. Melumad. 2002. Can "big bath" and earnings smoothing co-exist as equilibrium financial reporting strategies? Journal of Accounting Research 40 (3): 761-796. Kyle, A. S. 1985. Continuous auctions and insider trading. Econometrica 53: 1315-1335. Kyle, A. S. 1989. Informed speculation with imperfect competition. Review of Economic Studies 56: 317-356. Kwon, Y. K. 2001. Book value, residual earnings, and equilibrium firm value with asymmetric information. Review of Accounting Studies 6: 387-395. Lambert, R. A. 1984. Income smoothing as rational equilibrium behavior. The Accounting Review 59 (4): 604-618. . 2001. Contracting theory and accounting. Journal of Accounting and Economics 32: 3- 87. Leuz, C , and R. E. Verrecchia. 2004. Firms' capital allocation choices, information quality and the cost of capital. Working paper, Wharton School. Li , H., M . Pincus and S. O. Rego. 2006. Market reaction to events surrounding the Sarbanes-Oxley Act of 2002: Overall and as a function of earnings management. Working paper, University of Iowa. 122 Li , Y., and J. Weintrop. 2005. The effectiveness of NYSE guidelines to regulate CEO compensation through independent compensation committees. Working paper, Barach College. Liang, P. J. 2004. Equilibrium earnings management, incentive contracts, and accounting standards. Contemporary Accounting Research 21 (3): 685-717. Lobo, G. J., and J. Zhou. Did conservatism in financial reporting increase after the Sarbanes-Oxley Act? Initial evidence. Working paper, University of Houston. Hughes, J., J. Liu and J. Liu. 2005. Information, diversification and asset pricing. Working paper, University of California- Los Angeles. McNichols, M . , and B. Trueman. 1994. Public disclosure, private information collection, and short-term trading. Journal of Accounting and Economics 17: 69-94. Mitchell, L. E. 2003. The Sarbanes-Oxley Act and the reinvention of corporate governance? Villanova Law Review 48 (4): 1189-1215. Mohanram, P., and S. Sunder. 2003. How has Regulation Fair Disclosure affected the functioning of financial analysts?. Working paper, Columbia Business School. Morris, S., and H. S. Shin. 2002. Social Value of Public Information. American Economic Review 92 (5): 1521- 1534. Myerson, R. 1979. Incentive compatibility and the bargaining problem. Econometrica 47 (1): 61-74. Piotroski, J., and D. Roulstone. 2004. The influence of analysts, institutional investors, and insiders on the incorporation of market, industry and firm specific information into stock prices. Working paper, University of Chicago. Povel, P, R. Singh and A. Winton. 2004. Boom, busts and fraud. Working paper, University of Minnesota. Pritchard, A. 2003. Self-regulation and securities markets. Regulation Spring 32- 39. Ribstein, L. E. 2002. Market vs. regulatory responses to corporate fraud: A critique of the Sarbanes-Oxley Act of 2002. Working paper, University of Illinois, College of Law. Ribstein, L., 2005. Sarbanes-Oxley after three years. Working paper, University of Illinois. 123 Romano, R., 2004. The Sarbanes-Oxley Act and the making of quack corporate governance. Working paper, Yale Law School. Ronen, T., and V. Yaari. 2002. On the tension between full revelation and earnings management: A reconsideration of the revelation principle. Journal of Accounting, Auditing & Finance 17 (4): 273-294. Schipper, K. 1989. Commentary on earnings management. Accounting Horizons (December), 91-102. Schipper, K., and L. Vincent. 2003. Earnings quality. Accounting Horizons 17 (Suppl.): 97-110. Shleifer, A., and R. W. Vishny. 1997. A survey of corporate governance. The Journal of Finance 52 (2): 737-783. Stein, J. C. 1989. Efficient capital markets, inefficient firms: A model of myopic corporate behavior. The Quarterly Journal of Economics 104 (4): 655-669. Suh, Y. S. 1990. Communication and income smoothing through accounting method choice. Management Science 36 (6): 704-723. Summary of Sarbanes-Oxley Act of 2002. http://www.aicpa.org/info/sarbanes-oxley-summary.htm. Tirole, J. 2001. Corporate Governance. Econometrica 69 (1): 1-35. Trueman, B., and S. Titman. 1988. An explanation for accounting income smoothing. Journal of Accounting Research 26: 127-139. Vafeas, N. 1999. Board meeting frequency and firm performance. Journal of Financial Economics 53: 113- 142. Yermack, D. 1996. Higher valuation of companies with a small board of directors. Journal of Financial Economics 40: 185- 211. Zhang, G. 2001. Private information production, public disclosure, and the cost of capital: Theory and implications. Contemporary Accounting Research 18: 363-384. Ziv, A. 1998. Discussion of "earnings management and the revelation principle". Review of Accounting Studies 3: 35-40. 124 APPENDIX 1 Proofs for Chapter 2 Solving the Principal's Decision Problem at the Renegotiation Stage At the start of t = 2, i.e., the renegotiation stage, the principal chooses A / r and v2 so as to maximize his second-period expected utility given the initial contract, the public report y,, as well as his conjecture of the agent's first-period productive effort a{ and earnings manipulation D, subject to acceptance by the agent, max EUf (cr) = /?{yar2 - £,[OJ 2(c r)]}, A T . " ! subject to CEl (cr) > CE{ (c'), a[ = yvr2. Let the participation constraint be binding. Thus A / r = ^ { [ a 2 ] 2 - K ] 2 } + / 2Kl-p 2)o- 2 2{[v 2 r] 2-[v 2] 2} -{vr2[yar2 -RD + p{o2J(JX)(y, -ya x - D ) ] - v 2 [ j o 2 -RD + p(a2lox)(y, -yd{ -£>)]}. ( A l . l ) Substituting (Al. l ) and the incentive constraint into the objective function, the principal's decision problem becomes an unconstrained optimization problem, maxnf = (3{y2vr2-y2{/2[vr2]2 - K ] 2 } - ^ r ( l - / ? 2 ) c r 2 { [ v 2 r ] 2 -[v 2] 2} -vi2[ya2-Rb + p(a2lcrx){yx-ya{-b)}}. (A1.2) The first-order condition with respect to v2 is f-y2vr2-r(l-p2)a22vr2 =0. (A1.3) 125 Y1 Therefore, the revised incentive rate is v2 = 2 y2+r(\-p2)cj22 Q .E.D. Proof of Corollary 2.1 The agent's equilibrium earnings manipulation isD = — (v, - v 2 ) . First, it is P obvious that if p —» °° , i.e., infinitely tight regulation, then D = 0 . Next, since parameter p > 0, Sign (D) = Sign (v,-v 2 r ) . (A1.4) Given equilibrium solutions v, = ^y2 - A[p\2] with Q = y2 + rAx(?2, 1 r A, = —rA,pcrx(77, A = (1 + B) ', and v2 = —5 ; — - , then G y2+r(l-p2)a22 ,2 „ z T 2, 2M-p2)v22-Mcrx2+PpcTXCJ2)]. (A1.5) v. — v2 = 9 , Since y > 0,r > 0,/? > 0, pe (-1,1), Sign (v,-v 2 r) =S ign{ ( l -p 2 ) cT 2 -A 1 (cT 2 +/J /?cT 1 _- 2 ) } . (A1.6) Let A s f l - y O 2 ) ^ 2 " A C 0 " 2 + Pp^x^i)^ if /? = 1 and = CT2 = <T, then A = <7 20_->_p-p 2). (A1.7) Hence Sign (D) = Sign (Y2 - - p2), we obtain the following relationship between inter-temporal correlation /? and earnings' manipulation D under /? = 1 and 126 i. I f - \ < p < Y 2 , then D>0. ii. If Y2<p<\, then D<0. iii. If p = Yo o r P ~~* 0 0 ' t n e n D = 0 Q.E.D. Proof of Proposition 2.4 In window dressing setting, the agent's decision problem with respect to earnings manipulation is max{OJXX (D w ) - k(a?) - k(Dw) - P(Dw) + C £ , A ( D * ) ) . where = / + v,w(x, + D w ) , fc(Dw) = /2kw[Dwf, P(ZT) = + ew)[DwY, and C£, v (D w ) = [pr2 - RDW + p(cr21al)(*, - jo,)] - A(a 2 r ) - / ^ i • Differentiating it with respect to D" yields: dtfiD") dk(Dw) dP(Dw) dCEi\Dw) dDw dDw dDw + dDw = v? ~kwDw-(p + ew)Dw-vr2 = 0. (A1.8) Thus, the optimal earnings manipulation in window dressing setting is given by D W = U 1 >r -v 2 r ) -K + P + e Q.E.D. Derivation of n , T in Window Dressing Setting At t = 0, the principal's decision problem with respect to fw,v?,ew in window dressing setting is 127 max EU0pw =/3E0[XW]-/3E0[WW -Pw]-M(ew), (A1.9a) /">i"',e"' subject to CE™ = /3E0[WW - K(a?,a[)-k(Dw)-Pw]-0RPow > c°, (A1.9b) < = . (AI.9C) a2=yv2. (A1.9d) Given zero reservation consumption assumption, i.e., c° = 0, binding participation constraint (A 1.9b) implies fw = K(alw,a2) + k(Dw) + RP0w-E0[a™+/3co2-Pw]. (ALIO) Substituting the binding participation constraint (ALIO) and incentive constraints (A1.9c) and (A1.9d) into the objective function (A1.9a), the principal's decision problem becomes the following unconstrained optimization problem, max Ylpw =/3E0[Xw]- p\K(a{w ,ar2) + k(Dw) + RP™ ] - M (ew). That is, the principal chooses vxw,ew to maximize his expected gross payoff net of compensation for the agent's productive effort cost, window dressing cost and consumption risk premium, minus his monitoring cost. Q.E.D. Derivation of D w in Real Earnings Manipulation Setting The agent's decision problem with respect to DRw in the real earnings manipulation setting that is costly to the agent is 128 max{of (£) ) — ) - k{D ) — P(D ) + CEXX(DRw)}. D"" where cox{{DRw) = f + vRw(xx + DRw), k(DRw) = y2kw[DRw]2, CE,X (DRW ) =• p\[ [pr2 — RDRW — RH(DRW) + p{cr21 CT, )(•*, -ft)]- J3k(ar2) - pRPx, H(DRW) = y2hR[DRwf, P(DRw) = /2(p + eRW)[DRW]2. The first-order condition for the maximization problem is dco;(DRw) dk(DRw) dP(DRw) dCE;(DRw) dDRw dDRw dDRw + dDRw = vRw-kwDRw-(p + eRw)DRw-vr2-vr2hRDRw =0. (A l . l l ) Therefore the optimal earnings manipulation in the real manipulation setting is characterized by DRw = — (vRw -vr2). kw+vr2hR + P + eRwK 1 Q.E.D. Proof of Proposition 2.8 (2.8a) In the setting where earnings manipulation is costless to the agent and has no effect on the underlying economic outcome, the equilibrium earnings manipulation is D = — (v, - v2). Based on the proof of Corollary 2.1, P 2 v, - v2 '- —— [(1 - p2 )a\ - A, (of + Ppcrx<J2)]. Q[f +r(l-p 2)cr 2]  1 2 and Sign (D) = Sign (v, - v2). Further, given y>0, r>0, /?e(- l , l ) , /J>0 and Q = y 2 + rA]crf >0 , therefore d I < 0 as long as 129 [(1 - p2 )a\ - A, (cr,2 + /Jpcr,a2)] * 0. (Al . 12) (2.8b) In the costly window dressing setting, the equilibrium ew is characterized by the following equation, MB? -meew = 0, where MB? =. 3 T(v? -vr2)2. (kw + p + ewY Differentiating the equation with respect to me yields ™!L*?l-le~+mtK.] = o. ( A i . i 3 ) dew dme dme The existence of the interior solution of ew requires d2np0w dMB? N , A , ^ °— = e--m<0. (A1.14) d[ew]2 dew This leads to ^ y ^ m <0. Similarly, we prove ^ey^m <0; ^ y ^ m < 0 in the real earnings manipulation settings. Q.E.D. 130 APPENDIX 2 Proofs for Chapter 3 Derivation of Equation 3.7 & 3.8 According to equation (3.4), the market price px is Px = Euiiv\ ya^u] = m0 + ya +fiul(z„). where the aggregate trading quantity equals z„ =Z,-(zm +zZz^-The conjectures are Hu\(zu) = -bzu, and z,=/iy,. The manager anticipates the market price based on her own information set ( ya, ym). Then given the assumptions Cov[ya, y(.] = 0, Cov[ym, y, ] = 0, as well as E[Z, ] = 0, the manager's expected price resulting from an order of zm units, given her information set ( ya, ym), EiPx I yu.ym>zJ = "h + y a + *>zm + bIpE[yi \ ya,ym] = m0+ya+bzm. (A2.1) The key here is that since ya, ym and y(. are all independent, E[yt \ ya, ym ] = 0. As a result, the manager's expectation with respect to the orders placed by the informed investor is zero. Further, the manager's trading profit is calculated as n , n =(v~Pi)zm• I n addition, v = m0 + e and E[e\ya,ym] = ya+ym. Based on her information set (y a, ym), the manager's expected trading payoff is Um\ (ya • ym. zm) = {E[v\ya, ym] - E[px \ ya, ym, zm ] }zm 131 = (ym -bzjz. (A2.2) Q.E.D. Derivation of Equation 3.9 & 3.10 The derivation is similar as for Equation (3.7) and (3.8). By assumption, all informed investors receive the same private signal. Then the investor f s expectation with respect to the orders placed by the other informed investors are E[zj\ya,yt] = 0yr (A2.3) Thus, the informed invetsor fs expected price resulting from an order of z, units, given his information set (y a , yi), is E[Pl | ya, y,-,zi] = m0 + ya-b{E[Zt ] - ( z. + Z j | y,] + E[zm | y,.]} = mo+ya+bzi+b(l-l)0yl. (A2.4) Further, given the informed investor f s trading profit is calculated as ni = (v - p{ )zt, and E[e|yfl,y,] = ya + y, , his expectation of the trading payoff conditional on his information set (y a, y,.) is un(ya > yi>Zi) = {E[v | ya, yt] - E\px \ ya, y,,zt] = { y , - M / - l ) / 3 v , - f e , k - (A2.5) Q.E.D. 132 Derivation of Equation 3.11 The uninformed investor's posterior mean with respect to £ a v conditional on zu is calculated as MAz.) = E[£av \zj = E[EM] + {zu-E[z.J}. (A2.6) i Here zu = Z,-(zm+ ^izi) with Z , ~ N ( 0 , < T 2 ) and the conjectures zm=aym, ;=i Zi=pyr Further, £av=£-ya~ N(0 ,cr 2 , ) , and the random trade of Z , is not correlated with the firm value. Therefore, we have Cov(eav,(-aym-YJpyi)) / U O = T izu-E[zu]}. (A2.7) Var[Zl-(aym+Ydj3yl)] 1=1 Given the assumptions £ a v | y „ , , y , - ~ N( y„, + y, , c r 2 ) , y,. ~ N ( 0 , r j , 2 ) , Cov[y m , y, ] = 0 , and identical y , , we obtain M,Azu) = ~ 2 2 ^ , 2 , ^ 2 V (A2.8) Q.E.D. Proof of Proposition 3.1 In rational expectations equilibrium, the demand functions and the uninformed investor's posterior belief must be consistent with conjectures. This implies zm =aym=^7 ym. Zi = pyt = ^ - [1 - b(I - l)/?]y,, 2b 2b 133 a2crl+/32I2crf+c72z Mu\(zu)- bzu- 2 2 "' 2 2 2' , _2 z « - (A2.9) Solving the equations provides the following solutions: 1 a 1 _ 2 r 1 2 / 2 n 1 tt = ^ ' ' - < / — * • " = t 4 < T " , + (77i)7<T Q.E.D. Derivation of Equation 3.17 The informed investor f s expected trading profit at date one is (see equation (3.10)) Un(ya,yi,zl) = {yl-b(I-l)0yl-bzi}zi. Given equilibrium z,- = /3y (, and yi ~ N( 0 , <72), taking expectation yields EWn (ya, y,-, z,)] = (1 - W/Jcr 2 . (A2.10) Further, substituting equilibrium parameters /? = ^ and &2 - [—cr2m + — — — C T 2 ] — - , the above expression becomes z 2 _ r l _2 , ______ _2^ 1 (/ + D2 ' W / i ( y . , . z , ) ] = x „ 2 n • (A2.11) 1 CT 7 CT, 2 ( / + 1 ) 2 / I 2 , / _2 Considering the information acquisition cost Kt, the informed investor f s ex ante expected equilibrium net payoff, given that / investors are informed, is u r ( < T ' 2 } = TTZ^ h G z ° \ ~ r< ( < T» * } ' Q E D " \ ' ' *•) /A 2 ' 2 134 Proof of Proposition 3.3 The informed investor investor f s decision with respect to the informativeness of private signal af is 1 C T 7 t T , 2 max ur (erf) = ' - (1 + ya\ )*f (/ + 1)Z 1 , I 2 \ 4 C T ' " + ( 7 7 T ) y C 7 ' Thus, the first order condition is •a + ri). (A2.12) (7 + 1)2 ' _ ( 7 + l ) 2 , t , \y(j2 i 7 f T 2 * l x ° z Then the partial derivative of cr2* with respect to <T2 is 2 I 2* dcrf* 4 ( / + i ) 4 u ^ ° ' m + ( 7 7 r 7 0 ' ' ' - ^ w h e r e Q = K—-^ > 0 . (A2.13) 3<T" fKcTfn+—^C7rf ^ / r . 1 \ I < J 3cT 2* That is, the sign of ' depends on the sign of information cost parameter y, and OCT,, erf is increasing (or decreasing) in erf if y < 0 (or y > 0) , and unaffected by cr 2 i f 7 = 0 . Q .E .D. Proof of Proposition 3.4 In the setting without insider trading, the manager, i.e., the insider, discloses all the endowed information to the public (ya, ym). Since the manager does not engage in trade, there are only three sets of traders in the market: / privately informed investors, 135 the market maker, and the liquidity traders. Then the aggregate trading quantity absorbed by the market maker is zu=Z,-^Zi. (A2.14) Since he observes the two public signals (y f l ,y ,„) and the aggregate trading quantity zu, he uses this information to set the market price, i.e., Pi = Eui [v I y„ > ym. z'„ ] = m0 + ya+yn+ //„, (zu). (A2.15) As in the setting with insider trading, the conjecture of //„, (zu) is Mui(0 = E[£av\zu] = -bzu. (A2.16) The informed investor i's expected price resulting from an order of zt units, given his information set ( ya, ym, y,) is E[pl\ya,ym,yi,z]]=m0+ya+ym+b'z,+b^zj. (A2.17) Therefore, his expectation of the trading payoff is un (y«. ym. * . *i ) = { y, >zi • <A2-18> ;=1 1 • Maximizing the expected profit produces z'i(yi) = -^r[yi -b"^z)]. The 7 informed investors are identical, zt =zr Hence, z, (y.) = — r y , . Finally, the uninformed (/ + \)b investor's posterior mean with respect to eav is 136 R a t i o n a l expectation e q u i l i b r i u m requires that, the u n i n f o r m e d investor 's posterior mean should be consistent w i t h the conjecture, w h i c h impl ies / b'=- ( I + 1 ) h ( ) — r - r C T +CTZ 7 + 1 [bf • b ' ( A 2 . 2 0 ) 7 + 1CTZ T o summarize , the market e q u i l i b r i u m is characterized as: z] (y,) = —^7= y,, p\=m0 + ya+ ym - b z„, where b = v7 CT] C T , V / / + 1 ( J 2 Q.E.D. Derivation of Equation 3.23 In the strategic investor m o d e l without insider trading, the in formed investor /'s expectation o f the gross trading payof f at date one is (see equation (A2.18)) / - i un(ya>yn,>yi>Zi) = {yi -b'z, hi-J=I Substituting the e q u i l i b r i u m solutions z,(y ;) = °"z y ; , b - G>' , z\ =z'-, and taking expectation gives E[Un (ya, ym, y,., z])] = * crzcr,:. (A2.21) (/ + l ) v / 137 Taking the information acquisition cost Ki into account, the informed investor V s ex ante expected equilibrium net payoff, given that / investors are informed, is U'C"' ((7>} = „ \l7 a**> ~ K> ((T«' a>»' ) • ( A 2 ' 2 2 ) (/ + 1)V/ Q.E.D. 138 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0092927/manifest

Comment

Related Items