Description
Securities fraud, also known as stock fraud and investment fraud, is a deceptive practice in the stock or commodities markets that induces investors to make purchase or sale decisions on the basis of false information, frequently resulting in losses, in violation of securities laws.
ABSTRACT
Title of dissertation:
Securities Fraud: An Economic Analysis Yue Wang, Doctor of Philosophy, 2005
Dissertation directed by:
Professor Lemma Senbet, Professor Nagpurnanand Prabhala Department of Finance
This thesis develops an economic analysis of securities fraud. The thesis consists of a theory essay and an empirical essay. In the theory essay, I analyze a ?rm’s propensity to commit securities fraud and the real consequences of fraud. I show that fraud can lead to investment distortions. I characterize the nature of the distortions, and show that it results from fraud-induced market mispricing and management’s ability to in?uence the ?rm’s litigation risk through investment. The theory also characterizes the equilibrium supply of fraud. I demonstrate the linkages between a ?rm’s fraud propensity and the structure of its assets in place and growth options, and analyze the e?ect of corporate governance on fraud. The theory provides testable implications on crosssectional determinants of ?rms’ fraud propensities and the relation between fraud and investment. In the empirical essay, I test my main model predictions, using a new hand-compiled fraud data set. I use econometric methods to account for the unobservability of undetected frauds, and disentangle the e?ects of cross-sectional variables into their e?ect on the probability of committing fraud and the e?ect on the probability of detecting fraud. I ?nd that the level, type, and ?nancing of investment all matter in determining the probability of fraud and the likelihood of detection. I also examine the monitoring roles of large shareholders, institutional owners, independent auditors, and corporate boards. I ?nd that large block or institutional holdings tend to discourage fraud by increasing the detection likelihood. The roles of independent auditors and corporate board are weaker. Finally, insider equity incentives, growth potential, external ?nancing needs and pro?tability all in?uence a ?rm’s propensity to commit fraud. The paper also demonstrates the importance of separating fraud commitment and fraud detection, because cross-sectional variables
can have opposing e?ects on these two components, and therefore can be masked in their overall e?ect on the incidence of detected fraud.
Securities Fraud : An Economic Analysis by Yue Wang
Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial ful?llment of the requirements for the degree of Doctor of Philosophy 2005
Advisory Committee: Professor Lemma Senbet, Co-Chair/Co-Advisor Nagpurnanand Prabhala, Co-Chair/Co-Advisor Professor Vojislav Maksimovic Professor Je?rey Smith Professor Nengjiu Ju
c Copyright by Yue Wang 2005
This dissertation is dedicated to my grandmother and grandfather.
ACKNOWLEDGMENTS
I owe my gratitude to all the people who have made this thesis possible and because of whom my graduate experience has been one that I will cherish forever. First and foremost I’d like to thank my advisors, Professor Lemma Senbet and Professor Nagpurnanand Prabhala, for all the stimulating advices and consistently strong support in the past ?ve years. It has been great pleasure of mine to work with and learn from these extraordinary individuals. I would like to thank Professor Vojislav Maksimovic for all the inspiring discussions we had during my time in the PhD program. I am also extremely grateful to Professor Je?rey Smith for his kind help and insightful comments on my empirical essay. I thank Professor Nengjiu Ju for agreeing to serve on my thesis committee and sparing time to review my manuscript. I also owe my gratitude to all the other faculty members in the Finance Department. I would not have been able to reach this far without their kind encouragement and helpful advice. I wish to convey my special thanks to Professor Alexander Triantis, who has been a great mentor and a wonderful friend during the past two years. I owe my deepest thanks to my family - my mother and father who have always stood by me and guided me through my career, and have pulled me through against impossible odds at times. I’d also like to express my gratitude to my husband Yingkai Liu for walking through all the ups and downs with me in the past four years, and for believing in me even when I do not believe in myself. It is impossible to remember all, and I apologize to those I’ve inadvertently left out. Lastly, thank you all and thank God!
iii
TABLE OF CONTENTS List of Figures 1 Introduction 1.1 Theory of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Empirical Investigation of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi 1 2 4 7 8 8 9 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16 16 17 19 20 24 25 25 28 30 31 33 34 41 41 43 44 44 45 45 47 47 48 49 54 57 58 58 61 61 62 63 65 65 66 67 67
2 Securities Fraud & Securities Litigation 2.1 Securities Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Common Types of Alleged Securities Fraud . . . . . . . . . . . . . . . . . . . . . . 3 Related Literature 4 A Model of Securities Fraud 4.1 Model Framework . . . . . . . . . . . . . . 4.1.1 The Firm . . . . . . . . . . . . . . . 4.1.2 Time Line and Assumptions . . . . . 4.2 Cost and Bene?t of Fraud . . . . . . . . . . 4.2.1 Litigation Cost of Fraud . . . . . . . 4.2.2 Fraud Incentives . . . . . . . . . . . 4.3 Securities Fraud and Investment Incentives 4.3.1 Investment Distortions . . . . . . . . 4.3.2 A Numerical Illustration . . . . . . . 4.4 Disclosure Strategy . . . . . . . . . . . . . . 4.4.1 Equilibrium Misreporting . . . . . . 4.4.2 Fraud Propensity . . . . . . . . . . . 4.5 Model Implications and Discussion . . . . .
5 An Empirical Investigation of Securities Fraud 5.1 Fraud Sample . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Time Trends and Firm Characteristics . . . . 5.1.2 Industry Distribution . . . . . . . . . . . . . 5.1.3 The Nature of Fraud . . . . . . . . . . . . . . 5.2 Empirical Methodology . . . . . . . . . . . . . . . . 5.2.1 A Model with Partial Observability of Fraud 5.2.2 Model Identi?cation and Estimation . . . . . 5.2.3 Comparison with Straight Probit Model . . . 5.3 Hypothesis Development and Model Speci?cation . . 5.3.1 Probability of Fraud Detection . . . . . . . . 5.3.2 Propensity to Commit Fraud . . . . . . . . . 5.3.3 Control Variables . . . . . . . . . . . . . . . . 5.3.4 Summary of Model Speci?cation . . . . . . . 5.4 Descriptive Information and Univariate Analysis . . 5.5 Multivariate Analysis . . . . . . . . . . . . . . . . . . 5.5.1 Firm Characteristics and Fraud . . . . . . . . 5.5.2 Investment and Fraud . . . . . . . . . . . . . 5.5.3 Equity Ownership and Fraud . . . . . . . . . 5.5.4 Auditor, Board and Fraud . . . . . . . . . . . 5.5.5 Summary of Results . . . . . . . . . . . . . . 5.5.6 Comparison with Simple Probit Models . . . 5.6 Robustness Checks . . . . . . . . . . . . . . . . . . . 5.6.1 Frivolous Lawsuits . . . . . . . . . . . . . . .
iv
5.6.2 5.6.3 5.6.4 6 Conclusion
Timing of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Industry and Business Cycle e?ects . . . . . . . . . . . . . . . . . . . . . . . Di?erent Model Speci?cations . . . . . . . . . . . . . . . . . . . . . . . . . .
68 69 70 87 89 96
7 Appendix: Proofs of Propositions Bibliography
v
LIST OF FIGURES Figure 1: Model Time Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figure 2: Probability of Fraud Detection . . . . . . . . . . . . . . . . . . . . . . . 40 40 85 85 86
Figure 3: Determination of the Beginning Fiscal Year of Fraud . . . . . . . . . . . Figure 4: Identi?cation Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figure 5: Timing of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
Chapter 1 Introduction
In recent years, a string of high pro?le corporate scandals like those of Global Crossing, Enron, Tyco, and Worldcom has brought securities fraud and corporate governance to the forefront of public attention and policy debate. The magnitude of the alleged securities fraud is stunning. According to Stanford Securities Class Action Clearinghouse and Cornerstone Research, 224 securities lawsuits in 2002 in the United States were associated with a total $206 billion loss of market capitalization in the defendant ?rms.1 The governance crisis was followed by rapid and substantial legislative and regulatory changes that aimed to restor investor con?dence in the capital markets. The movement was so fast that 9 months after the Enron debacle, President Bush signed the Sarbanes-Oxley bill into law. Securities fraud is a very serious issue. It undermines a core value in capital markets, the integrity of public companies, which is essential to investor con?dence in those markets and the e?cient allocation of capital. Furthermore, we also observe ine?cient investments and serious value destructions in many fraudulent ?rms (e.g., Enron, Nortel, eToys), which implies that there could be large real economic cost associated with fraud. The governance crisis and the on-going governance reform call for careful economic re?ections on what have happened, because the exact nature, signi?cance, and consequences of securities fraud and the economics underlying the legislative and regulatory changes are still incompletely understood. This thesis develops an economic framework to characterize the determinants and consequences of securities fraud. I de?ne securities fraud as deliberate and material misrepresentation of corporate performance, and thus use fraud and misreporting interchangeably. The thesis consists of a theoretical model of fraud and empirical analysis.
1 Cornerstone
Research, “Securities Class Action Case Filings 2002: A Year in Review.”
1
1.1 Theory of Fraud
The theory part of the thesis builds on Gary Becker’s (1968) economic analysis of crime. Following Becker’s approach, one can view fraudulent behavior as an economic activity, whose equilibrium supply depends on a rational calculation of the expected bene?ts and costs from engaging in it. Di?erent ?rms have di?erent propensities to commit fraud because they face di?erent cost-bene?t tradeo?s. In this paper, the bene?t from fraud is that ?nancial misreporting can create (or sustain) short-term market overvaluation of the ?rm. The cost of fraud is litigation risk. With some positive probability, fraudulent activities will be uncovered, resulting in a fraud penalty (which includes both explicit monetary ?nes and other implicit costs, such as loss of reputation). Within this framework, the ?rm’s propensity for fraud, the magnitude of fraud, and the ?rm’s investment incentives are analyzed. The theory demonstrates an interesting link between a ?rm’s ?nancial disclosure incentive and its real investment decision. First, ?nancial misreporting can a?ect the short-term market valuation of the ?rm and allow the ?rm to invest using cheap outside capital. Second, after committing fraud, the ?rm has incentive to cover things up. Such incentive can motivate the ?rm to strategically use investment to mask fraud and reduce its litigation risk. The basic intuition is that stochastic cash ?ows from a new investment can decrease the precision of the ?rm’s total cash ?ow and create inference problems for the market. In sum, investment can a?ect both the ?rm’s ex ante bene?t from committing fraud and its ex post probability of being detected. The model predicts that fraudulent ?rms tend to overinvest in the sense that they would undertake some negative NPV projects that destroy shareholder value. In particular, fraud can induce a preference for risky (in terms of high return volatility) or uncorrelated projects (uncorrelated with the cash ?ow from existing assets), because these types of investment can better disguise fraud than others. The investment distortion can lead to serious value destructions in the ?rm, which is the real economic cost of fraud. Furthermore, the cost of ine?ciency is borne by not only shareholders of fraudulent ?rms but also those of honest ?rms, because ex ante the market cannot perfectly distinguish between the two types of ?rms.
2
The theory further characterizes the ?rm’s equilibrium disclosure strategy. The model shows that the ?rm will honestly reveal performance if its performance is very good or if it is desperately bad. The former case is associated with low bene?t from fraud, and the latter is associated with high litigation risk. The ?rm’s propensity to commit fraud and the magnitude of fraud depend on the nature of the ?rm’s assets and growth opportunities. The model predicts that fraudulent ?rms tend to have high growth potential but experience negative pro?tability shocks. Growth potential can positively in?uence the ?rm’s payo? from fraud and negatively in?uence its litigation risk. In addition, litigation events tend to cluster in certain industries during some speci?c time period, because ?rms’ cost-bene?t tradeo?s of fraud are correlated within an industry. The theory also generates implications about the role of corporate governance in the context of corporate fraud. The model shows that good corporate governance can increase the likelihood of fraud detection and thus deter fraud ex ante. However, corporate governance may also fail to prevent fraud if it is just about aligning the interest of the management with that of incumbent shareholders. This is because even when such alignment is perfect, fraud can still emerge in equilibrium. Finally, the theory demonstrates the e?ect of the endogenous detection risk on the crosssectional variations in ?rms’ fraud propensities. While the penalty for fraud (at least the explicit liability provisions) is largely determined by securities laws and thus is exogenous to the ?rm, the probability of detection can be in?uenced by the ?rm’s endogenous actions (e.g., investment, disclosure) as well as ?rm-speci?c attributes. This endogeneity implies that the detection risk is more important in determining cross-sectional variations in ?rms’ fraud propensities than are penalty provisions. Therefore, without increasing the probability of detection, enhanced liability standards alone may achieve only limited deterrence, because ?rms can undo some e?ects of tightened penalty by adjusting their probabilities of getting caught. More important, fraudulent ?rms’ incentive to decrease their likelihood of being detected can be a potential source of value destruction. Therefore, there is a real danger associated with over-regulation.
3
1.2 Empirical Investigation of Fraud
I then empirically test some of my key model predictions. Speci?cally, I examine the e?ects of real investment, corporate monitoring, and ?rm characteristics on a ?rm’s cost-bene?t tradeo? of committing fraud. The analysis is based on a new hand-compiled fraud data set, which consists of private securities class action lawsuits ?led between 1996 and 2003 against US public companies with allegations of accounting irregularities. The next contribution of the paper is methodological. In assessing a ?rm’s propensity to commit fraud, we face an identi?cation problem because we only observe detected fraud. A nonlitigated ?rm can be either an honest ?rm or an undetected fraudulent ?rm. This implies that the probability of a ?rm committing fraud and the probability of observing the ?rm as fraudulent can be very di?erent. This study utilizes statistical methods to control for this problem. In essence, I model the probability of detected fraud (what we observe) as the product of two latent probabilities: the probability of committing fraud and the probability of detecting fraud conditional on fraud occurrence. Then I use econometric methods to back out these two latent probabilities. Disentangling fraud commitment and fraud detection provides two advantages. First, it allows me to control for the unobservability of frauds committed but not detected. Second and more important, it allows me to examine the economics of each probability as well as their interactions.2 Using the above methodology, I examine the link between real investment and the incidence of fraud. There has been surprisingly little exploration on the relation between corporate fraud and investment. This is, however, an important issue. We have observed ine?cient investments and serious value destructions in many fraudulent ?rms (e.g., Enron, Nortel, eToy). Hence, there can be large real economic cost associated with fraud. Wang (2004) theorizes that fraud can induce overinvestment incentives for two reasons. First, fraud can create (or sustain) market overvaluation and decrease the external ?nancing cost of investment. Second, after committing fraud, the ?rm has incentive to strategically use investment to mask fraud and decrease its litigation
2 In
a concurrent paper, Li (2004) uses a simultaneous model with partial observability to analyze the role of the
SEC in detecting fraud.
4
risk. I ?nd evidence that supports this theory. First, I ?nd that the alleged fraudulent ?rms on average have larger investment expenditures than a random sample of non-convicted ?rms and a sample of size and age matched comparison ?rms. Second, di?erent types of investment appear to have di?erential e?ects on the probability of fraud detection. Risky investment (e.g., investment in R&D) and uncorrelated investment (e.g., diversifying acquisition) tend to decrease the probability of detection, while straightforward investment (e.g., capital expenditures) and correlated investment (e.g., focused acquisition) do not. Lastly, di?erent investments also in?uence a ?rm’s propensity to commit fraud di?erently. This is either because of the way the investments are ?nanced (e.g., stock-based vs. cash-based acquisitions) or because of the di?erential litigation risk they induce. Overall, the empirical results imply that investment is associated with both a ?rm’s ex-ante bene?t from committing fraud and its ex-post litigation risk, and thus is an important determinant of the ?rm’s fraud incentives. Corporate securities fraud also provides a new and interesting angle to examine the roles of di?erent corporate monitors in determining ?rms’ incentives and behavior. E?ective monitoring should increase the probability of uncovering fraudulent corporate activities and discourage fraud ex ante. I investigate four types of corporate monitors: large shareholders, institutional owners, independent auditors, and board of directors. I ?nd that the presence of block equity holders and large institutional ownership are associated with high probability of fraud detection and low probability of fraud. For example, increasing block ownership by 10% on average tends to increase the probability of detection by 1% and decrease the probability of fraud by 4%. This supports the view of enhancing shareholder monitoring in combatting corporate fraud. The roles of independent auditors and corporate boards seem to be much weaker. I ?nd no evidence that auditors’ opinions increase the likelihood of fraud detection. There is some weak evidence that reputable independent auditors and large corporate boards are related to higher likelihood of fraud detection. The role of insider equity incentives has received a great amount of public attention following the recent wave of corporate scandals. Several studies have documented that large executive pay for performance sensitivity is associated with high probability of corporate fraudulent reporting
5
(see, e.g., Johnson, Ryan and Tian (2003), Peng and R¨ oell (2004), and Burns and Kedia (2004)). In this paper, since I separate the probability of fraud from the probability of detected fraud, I am able to more directly examine the e?ect of insider equity incentive on a ?rm’s propensity to commit fraud. Interestingly, I ?nd a concave relation between the two. When insider equity incentive is small, the probability of fraud increases as the equity incentive increases. When insider equity incentive becomes large, the positive relation weakens and eventually reverses. Overall, this result seems to support the predictions of the agency theory. However, it also implies that equity incentive can be a double-edged sword when it is used to align managerial and shareholder interests in dispersedly-owned ?rms. Finally, I examine how ?rm characteristics in?uence a ?rm’s cost-bene?t tradeo? of engaging in fraud. I ?nd that high growth potential and large external ?nancing need are two important motivational factors for fraud. Alleged fraudulent ?rms on average grow much faster than comparison ?rms and have larger portion of the growth supported by external capital. There is also indirect evidence that fraudulent ?rms generally experience negative pro?tability shocks in the year when fraud begin. Most existing studies on corporate fraud have focused on the bene?t side of the tradeo?. The literature on earnings management and corporate fraud has provided evidence that managers misreport corporate performance in order to facilitate external ?nancing activities, to avoid violations of debt covenants, or to increase performance-related compensation (see, Healy and Wahlen (1999) for a review). On the cost side of the tradeo?, a few papers have examined the consequences following the revelation of fraud. For example, Dechow, Sloan, and Sweeney (1996) show that the revelation of fraud leads to persistent increase in the ?rm’s cost of capital. Baucus and Baucus (1997) ?nd that ?rms convicted for illegal corporate behavior su?er from prolonged poor operating performance. Gande and Lewis (2005) document signi?cant negative abnormal returns upon the ?ling of securities lawsuits. The only paper I know that studies the probability of fraud detection is Li (2004). Li emphasizes the strategic role of the SEC in detecting corporate fraud, and documents that a larger SEC budget increases the probability of fraud detection and deters fraud. My paper
6
further demonstrates the importance of understanding ?rm-level economic determinants of fraud detection and how the detection risk in?uences its ex-ante propensity to commit fraud. I show that the probability of detection depends on ?rms’ investment decisions, the strength of corporate monitoring, and ?rm-speci?c attributes. The cross-sectional variations in the detection risk help to explain the variations in ?rms’ fraud propensities. I also demonstrate the importance of disentangling the probability of committing fraud from the probability of detecting fraud. Cross-sectional variables can have opposing e?ects on the two latent probabilities, and thus can be masked in their overall e?ect on the incidence of detected fraud. For example, this paper shows that large institutional ownership is associated with high probability of fraud detection and low probability of fraud. The e?ect on detection tends to dominate and thus we observe a positive relation between institutional ownership and the compound probability (incidence of detected fraud). This may lead us to draw incorrect inferences. Distinguishing the probability of fraud from the probability of detected fraud is not only important for understanding the economics of fraud, but also relevant from a regulatory point of view in setting policies that deal with fraud.
1.3 Thesis Structure
The thesis is structured as follows. Chapter 2 introduces the basic institutional knowledge about securities fraud and securities class action litigation. Chapter 3 reviews the related literature. Chapter 4 develops an analytical model to characterize the economic determinants of corporate fraud propensity and the real consequences of fraud. Chapter 5 empirically investigate ?rms’ fraud incentives and fraud detection. Finally, Chapter 6 concludes.
7
Chapter 2 Securities Fraud & Securities Litigation
The purpose of this chapter is to provide some general knowledge about securities fraud, securities laws and regulations, and securities litigation. The structure of this chapter is as follows. Section 2.1 presents a de?nition of securities fraud, and describes the major anti-fraud laws and regulations that govern the securities industry. Section 2.2 describes the common types of fraud allegations in the private securities class action litigation between 1996 and 2002.
2.1 Securities Fraud
A thorough understanding of the nature, signi?cance and consequences of securities fraud requires a proper de?nition of securities fraud. Fraud, in general, as de?ned in Webster’s Universal College Dictionary, is deceit or trickery perpetrated for pro?t or to gain some unfair or dishonest advantage. I de?ne securities fraud as follows based on the description of the SEC and the Securities Exchange Act of 1934. Securities fraud refers to the use of any manipulative and deceptive devices, in connection with the purchase or sale of any security, that are in contravention of such rules and regulations as the Commission may prescribe as necessary or appropriate in the public interest or for the protection of investors. The term “security” means any note, stock, treasury stock, bond, debenture, derivative securities, certi?cate of interest, or in general, any instrument commonly known as a security. Section 10(b) of the Securities Exchange Act of 1934 and the rules promulgated thereunder (especially Rule 10(b)-5) build the major substance of the broad anti-fraud provisions that make it unlawful for anyone to engage in fraud or misrepresentation in connection with the purchase or sale of a security. Violations of these provisions include employment of any devices, schemes or arti?ce to defraud, misrepresentation and/or omission of material information, or engaging in any act, practice or course of business which operates or would operate as a fraud or deceit upon
8
any person, in connection with the purchase or sale of any security. The essence of the above regulations is to prohibit deliberate and material information misrepresentation in any form of public communications between the ?rm and its investors, and between the ?rm and its regulators. There are two major types of securities litigation: the SEC’s enforcement actions and the private class action litigation. According to Securities Class Action Clearinghouse (SCAC) established by the Stanford Law School, a securities class action is a case brought pursuant to Federal Rule of Civil Procedure 23 on behalf of a group of persons who purchased the securities of a particular company during a speci?ed period of time (the class period). The complaint generally contains allegations that the company and/or certain of its o?cers and directors violated one or more of the federal or state securities laws. A suit is ?led as a class action because the members of the class are so numerous that joinder of all members is impracticable.
2.2 Common Types of Alleged Securities Fraud
Table 1 lists the types of commonly alleged securities fraud in class action lawsuits between 1996 and 2002 and the frequency distribution. The litigation information is retrieved from SCAC. I identify the speci?c nature of fraud allegations based on information extracted from the case complaints and/or press releases. In each year there was a small number of cases that did not provide enough information for us to determine the nature of the allegations. Therefore, the information provided in this section is based on the identi?able ?lings. 1. Financial statement fraud, which refers to the deliberate and material misstatement of ?nancial statements issued by publicly traded companies to mislead the ?nancial statements users (Rezaee [2002]). 2. Misrepresentation or concealment of material facts (excluding misreporting in the ?nancial statements). Material facts are the ones that, if made available, would cause the information receivers to change their judgment or decision. This category of securities fraud includes a public ?rm issuing false information and/or omit important information in security registration statements/prospectus (section 11, 12(a) of the Securities Act of 1933), in proxy 9
statements (section 14 of the Exchange Act of 1934) and other important public documents, as well as false and misleading oral communications at press releases and conference calls. Many allegations in this category also frequently involves a?rmative fraud, i.e., the release of false forward-looking statements to the investing public. An example of a?rmative fraud is that a public ?rm issue glowing but misleading projections about the ?rm’s future business prospects and competitive position. 3. Illegal insider trading. According to the SEC’s de?nition, illegal insider trading refers generally to buying or selling a security, in breach of a ?duciary duty or other relationship of trust and con?dence, while in possession of material, non-public information about the security. Insider trading violations may also include “tipping” such information, securities trading by the person “tipped”, and securities trading by those who misappropriate such information. “Insiders” generally include o?cers, directors, and individuals who hold more than 10 percent of the company’s stock (regardless of whether they work for the company). 4. Investment bank fraud. This category of fraud refers to the unfair dealings in investment banking activities. Most commonly alleged investment bank frauds include unfair IPO allocations and misleading analyst reports. 5. Breach of ?duciary duty. This category generally involves violations of section 14 of the Exchange Act. Section 14 prohibits any information misrepresentation in the proxy statements, particularly information related to tender o?ers, management buyouts, and other merger/acquisition activities. Most of the cases in this category alleged that the management or controlling shareholders expropriated minority shareholders in merger/acquisition activities, and misled minority shareholders to tender or exchange their shares at unfairly low prices. 6. Stock price manipulation, which refers to deliberate buying or selling of a security, or deliberate intervention of other people’s buying or selling of a security, in order to control the price of the security.
10
Table 2.1: Commonly Alleged Securities Fraud This table presents the types of commonly alleged securities fraud in 1268 private securities class action lawsuits between 1996 and 2002. Pure investment bank fraud cases (i.e., cases that allege unfair IPO allocations by securities underwriters and untrue securities analyst reports) are excluded. “Other information misrepresentation/omission” means material non-accounting related information misreporting or omission. Nature of Fraud Number of observations Accounting irregularity Other information misrepresentation/omission Illegal insider trading Breach of ?duciary duty Stock price manipulation # of Filings 1268 596 486 337 49 9 % of Total 47.08 38.33 26.58 3.90 0.71
11
Chapter 3 Related Literature
This thesis is related to several strands of research: (1) the accounting literature on earnings management and ?nancial disclosure; (2) the literature on agency theory; and (3) recent research on corporate fraud. The economics of corporate misreporting is examined in the accounting disclosure literature. Dye (1988) analyzes two conditions under which earnings management may exist in equilibrium. First, the cost-minimizing contract that induces preferred action from the manager may not prevent earnings management, which leads to the internal demand for earnings management. Second, incumbent shareholders may attempt to alter the perceptions of prospective investors through managed earnings, which creates the external demand for earnings management. In line with Dye’s notion of internal demand for earnings management, Lacker and Weinberg (1989) show that the optimal risk sharing contract may not prevent the agent from falsifying the outcome. Goldman and Slezak (2003) show that the optimal equity compensation contract that induces the desired managerial e?ort may not prevent (and may even encourage) the agent from misreporting. Several other papers together with my paper are consistent with Dye’s notion of external demand for earnings management. Stein (1989) argues that capital market pressure can induce the management to in?ate current pro?tability at the expense of forgoing future cash ?ows. Bebchuk and Bar-Gill (2002) present a model in which ?rms’ needs for external ?nancing and insiders’ bene?t from informed trading can motivate management to misreport corporate performance. Jensen (2004) argues that corporate fraud can result from a dramatic form of capital market pressure. When the market substantially overvalues a ?rm’s equity, the ?rm may feel forced to defraud investors in order to defend such overvaluation, and this can lead to serious value destructions in the ?rm. I show that overvaluation can result from the ?rm’s endogenous choice, and an important
12
source of value destruction is the fraud-induced investment distortions. There has been a large body of empirical research on earnings management. Earnings management does not necessarily imply the existence of securities fraud. Earnings management re?ects discretionary managerial judgment (or managerial ?exibility) in corporate ?nancial reporting.1 However, both securities fraud and earnings management involve some information misrepresentation. A number of studies have examined di?erent incentives for earnings management, including capital market needs, contracts written on accounting numbers and government regulations (see Healy and Wahlen (1999) for a review of empirical work on earnings management). The literature provides evidence that managers have incentives to manipulate earnings in an attempt to in?uence short-term stock price performance before major external ?nancing activities or externally-?nanced investment (see, e.g., Teoh, Welch and Wong (1998a,b) on public equity o?ers; Erickson and Wang (1998) on stock-?nanced acquisitions). Efendi, Srivastava and Swanson (2004) ?nd that the likelihood of an earnings restatement is signi?cantly higher for ?rms that make one or more sizable acquisitions. Several studies have examined the relation between the structure of managerial compensation contracts and the likelihood of earnings management, and ?nd that the pay-performance sensitivity induced by stock options seems to increase earnings management (see, e.g., Gao and Shrieves (2003)). Research directly on corporate fraud has been sparse, but has started to attract academic interest after the explosion of corporate scandals and the recent legislation movement. Most of the recent studies focus on the e?ects of insider equity incentives on ?rms’ incentives to misreport. Johnson, Ryan, and Tian (2003), Peng and R¨ oell (2004), Burns and Kedia (2004), and Efendi, Srivastava and Swanson (2004) all ?nd that large executive pay-for-performance sensitivity is positively associated with fraudulent reporting. These results seem to support the over-incentivization argument that insider equity incentive is a double-edged sword. It may induce managerial misreporting incentive rather than managerial e?ort in creating shareholder wealth. Alexander and Cohen (1999), however, documents a negative relation between insider ownership and the likelihood
1 Schipper
[1989] de?nes earnings management as “purposeful intervention in the external reporting process, with
the intent of obtaining some private gain to managers or shareholders”.
13
of fraud and provide some support for the classic agency theory. Several studies have examined the relation between the characteristics of the board and the probability of corporate fraudulent reporting. Beasley (1996) studies a sample of ?rms subject to SEC’s AAERs and ?nds that board independence (proxied by the percentage of outside directors in the board) is signi?cantly negatively related to the likelihood of ?nancial statement fraud. Klein (2002) ?nds an inverse relation between board independence and abnormal accruals. Dechow, Sloan and Sweeney (1996) ?nd that ?rms committing ?nancial statement fraud are likely to have a board dominated by insiders and have a CEO who is also the chairman of the board or the founder of the company. Agrawal and Chadha (2004) examine the incidence of accounting restatements, and ?nd that board independence is irrelevant, but the presence of independent directors with ?nancial or accounting expertise on the audit committee is associated with signi?cantly lower probability of accounting restatements. The major contributions of my thesis to the literature are threefold. First, my thesis is the ?rst paper that seriously analyzes the role of real investment in the context of corporate fraud. My theory model shows that investment in?uences both a ?rm’s ex-post probability of fraud detection and its ex-ante propensity to commit fraud. Fraud can induce distorted investment incentives, which is the real economic cost of fraud. Second, I empirically examine the model predictions on the relation between fraud and corporate investment incentives, and ?nd strong support for my theory. For example, I ?nd that risky investment and uncorrelated investment have a strong negative e?ect on the probability of fraud detection, while straightforward investment and correlated investment do not. This implies that the type of investment matters in determining the ?rm’s detection risk. I also ?nd that acquisition expenditures in?uence the probability of fraud only if it is (at least partially) ?nanced by stock, indicating that the ?nancing of investment matters. The third contribution of my thesis is methodological. I introduce a new empirical methodology to analyze fraud. The existing literature has ignored the fact that we only observe detected fraud. That is, the outcome we observe depends on the outcome of two latent processes: fraud commitment and fraud detection. If we ignore this structure, we could draw incorrect inferences, because the same
14
variable can have opposing e?ects on the two latent processes and thus get masked in its overall e?ect on the outcome we observe. This study utilizes statistical methods to disentangling fraud commitment and fraud detection. This allows me to control for the unobservability of undetected frauds. More important, this allows me to examine the economics of each component as well as their interactions. My thesis provides new insights about corporate fraud incentives that cannot be obtained using the models in the existing literature.
15
Chapter 4 A Model of Securities Fraud
This chapter develops an economic framework of securities fraud. I analyzes the interaction between the ?rm’s ?nancial disclosure and its real investment decision. I also characterize the ?rm’s equilibrium disclosure strategy, probability of committing fraud and the magnitude of fraud. The chapter is structured as follows. Section 4.1 introduces the model framework and assumptions. Section 4.2 characterizes the ?rm’s cost-bene?t tradeo? of committing fraud. Section 4.3 examines the ?rm’s investment incentives in the presence of fraud. Section 4.4 derives the ?rm’s equilibrium disclosure strategy. Section 4.5 discusses model implications and possible extensions.
4.1 Model Framework 4.1.1 The Firm
Consider a typical public ?rm whose market value consists of both its assets in place and
2 1 growth opportunities. The asset value is normally distributed, A ? N (A, ?A ). The growth
opportunity takes the form of a possible new investment project in the future whose value is also
2 normally distributed, G ? N (G, ?G ). The market knows the distributions of A and G, but does
not observe the realizations of each component. The market value of the ?rm is the expected discounted value of future cash ?ows. For simplicity, I assume that investors are risk neutral, and the discount rate is zero. Therefore, the ?rm value is simply E (V ) = A + G. The ?rm is operated by a manager who owns a fraction 0 < ? < 1 of the ?rm. I assume that the manager holds restricted stock and thus is not allowed to trade any of his own equity shares. This simplifying assumption allows abstraction from the incentive and signalling e?ects of insider trading, and it also implies that the manager maximizes the wealth of long-term shareholders.2
1I 2I
can always choose reasonable values for A and ?A such that negative asset values have almost zero probabilities. assume that there is no opportunity for perquisite consumption. This type of agency problem is not the focus
16
The accounting and auditing literature has provided evidence that both capital market activities (see the citations in the introduction) and pro?ts from informed trading (e.g., Summers and Sweeney (1998)) can motivate fraudulent reporting. According to my study of private securities class action litigation against US public companies between 1996 and 2002, about 68% of the securities lawsuits involved misreporting surrounding major capital market activities (external ?nancing or externally-?nanced investment), and about 29% of the cases involved allegations of illegal insider trading and insider personal gains. This paper focuses on fraud and ?rm investment, and thus analyzes the former scenario. The model shows that even when the manager’s interest is perfectly aligned with that of existing shareholders, fraud can still exist in equilibrium. Adding managerial agency problem could, of course, further exacerbate the manager’s fraud incentives.
4.1.2 Time Line and Assumptions
There are four periods in this model, t = 0, 1, 2, 3. The sequence of events is described below (also see Figure 1 at the end of the chapter for an outline). Time 0: Institutional Arrangements At time 0, the institutional arrangement of the ?rm
is established. The strength of the ?rm’s internal corporate governance is indicated by p ? [0, 1]. Higher p represents better governance and also higher likelihood of internal detection of fraud.3 Time 1: Disclosure of Earnings At time 1, the manager privately observes the realization
of the intermediate earnings generated by the ?rm’s assets.4 The earnings realization is drawn from the following process. e = q A + u. (4.1)
q indicates the average productivity of the ?rm’s assets in place, of which the market is aware. u
2 is a white noise term, u ? N (0, ?u ). Equation (4.1) shows that the realized intermediate earnings
(e) contain useful information about the value of the ?rm’s assets. Let the signal-to-noise ratio be
of this paper.
3 Section 4 The
4.5 will discuss the possibility of endogenizing this parameter.
intermediate information does not have to be earnings. It can be any valuable piece of accounting in-
formation, or even more general information about the ?rm’s overall ?nancial condition, operational condition, or business prospects.
17
? ?
2 q?A 2 +? 2 . q 2 ?A u
Then the expected value of the assets conditional on the earnings realization e is
˜|e) = A + ? (e ? e). E (A After observing the intermediate earnings, the manager makes a disclosure decision, y (e) = e + ?. (4.2)
? represents the amount of distortion in the reported earnings. ? = 0 means that the manager chooses to truthfully reveal the earnings realization. ? > 0 implies that the manager in?ates earnings. ? is assumed to be nonnegative. That is, this paper focuses on overreporting of earnings. It is possible that managers may intentionally understate earnings (e.g., the case of Freddie Mac). Empirical studies on earnings management as well as SEC accounting and auditing enforcement actions, however, indicate that accounting overstatement is much more frequently observed than understatement (see, e.g., Feroz, Park, and Pastena (1991); Rezaee (2002)), and thus it is a more interesting subject for research. Once the earnings disclosure is made, the market prices the ?rm’s equity based on the reported earnings y (e), but the market does not have to take the earnings announcement at face value. Investors are generally aware of the possibility of misreporting. The market’s prior belief about the ?rm’s likelihood of misreporting is ?0 ? [0, 1], and the expected amount of misreporting ˜|y (e)], where the expectation is ? . Then the time 1 market value of the ?rm’s assets is V1 = E?0 [A incorporates the market’s prior belief about fraud. Time 2: Investment Decision In this period, a new investment opportunity arrives with
2 probability ?, requires an initial outlay of $I , and will generate a gross return R, R ? N (R, ?R ).
For simplicity, I set R = 1, which allows me to parameterize the pro?tability of the new investment in a straightforward way. Once the new investment opportunity arrives, the manager observes the gross return as r, the realization of R. The market does not observe this but knows the return distribution (i.e., the mean and variance of R). The manager makes an investment decision: whether to take the new project or not. If he decides to take it, the ?rm needs to raise $I as the initial capital. I assume that new equity shares are issued. I will discuss the robustness of the model results with respect to this assumption in 18
section 4.2.2. Time 3: Liquidation time 2, V = A + IR = If the ?rm does not invest, V =A= 1 1 e ? u. q q (4.4) 1 1 e + I R ? u. q q (4.3) At time 3, the ?rm has a liquidating cash ?ow V . If the ?rm invests at
The market is able to observe this ?nal cash ?ow and can use this information to update its belief about the probability of fraud at time 1. How the market interprets a particular ?nal cash ?ow realization depends on the market’s expectation about V . The following table lists four distributions of V : the perceived distribution (conditional on y (e)), the true distribution (conditional on e), the distribution given that the ?rm invests (I), and the one given not (N). Investment (I) True E (V |I, e) = E (A + I R|I, e) V ar(V |I, e) = V ar(V |I ) Perceived E (V |I, y ) = E (A + I R|I, y ) V ar(V |I, y ) = V ar(V |I ) No Investment (N) E (V |N, e) = E (A|e) V ar(V |N, e) = V ar(V |N ) E (V |N, y ) = E (A|y ) V ar(V |N, y ) = V ar(V |N )
2 2 2 V ar(V |I ) = ?e /q + (I?R )2 + 2?I?R ?e /q + ?u /q 2 , where ? is the correlation between e and R. 2 2 2 V ar(V |N ) = ?e /q + ?u /q 2 . We can see that misreporting only distorts the expected value of the
?rm’s ?nal cash ?ow, not the variance of it.
4.2 Cost and Bene?t of Fraud
This section characterizes the cost-bene?t tradeo? of committing fraud. The litigation cost of fraud is derived in section 4.2.1. The bene?t from fraud and the manager’s optimization problem are presented in section 4.2.2.
19
4.2.1 Litigation Cost of Fraud
At time 3, after the realization of the ?nal cash ?ow, the market may unearth the manager’s misreporting at time 1 with some probability. If fraud is detected, the ?rm will be subject to a fraud penalty. The expected litigation cost, which is the product of the detection likelihood and the penalty after detection, is the cost of committing fraud.
Probability of Fraud Detection
This model considers two fraud detection mechanisms: detection through cash ?ow and detection through internal corporate governance.5 At time 3, after observing the ?rm’s ?nal cash ?ow, the uninformed outsiders rationally choose an investigation strategy that maximizes their payo? from litigation.6 More speci?cally, the market chooses a threshold v such that it will investigate the manager’s time 1 disclosure whenever the ?nal cash ?ow realization V falls below this threshold. I assume that any misreporting, if it exists, will be discovered upon investigation (i.e., the conditional probability of fraud detection upon investigation is 1). Thus, I will use the probability of fraud investigation and the probability of fraud detection interchangeably. I call the region {V : V ? v } the cash ?ow detection region. If this region is not reached (i.e., V > v ), no external investigation will be triggered, but detection of fraud is still possible. In this situation, the probability of fraud detection solely depends on the ?rm’s quality of corporate governance (p). That is, p indicates the likelihood of an internal investigation of fraud when the cash ?ow realization does not automatically reveal fraud. In sum, the likelihood of detection conditional on V ? v is 1, and the likelihood conditional on V > v is p. Then the e?ective probability of fraud detection is P = P rob.(V > v ) × p + P rob.(V ? v ) × 1. Probability of Cash Flow Detection
5 Of
(4.5)
At time 3, if the market investigates the ?rm, the
course, there are other detection forces, such as regulators (SEC) and independent auditors. This paper
focuses on the role of capital markets and internal corporate governance in discovering fraud.
6 Here
the outsiders can be the ?rm’s outside (and uninformed) investors or the regulators such as the SEC.
Therefore, the role of outsiders represents general capital market monitoring.
20
expected payo? from the e?ort is f E (? |V ) ? C , where C > 0 is the investigation cost. Therefore, the market will examine the ?rm’s disclosure practice if and only if f E (? |V ) ? C ? 0, or E (? |V ) = y ? E (e|V ) ? De?ne ?V = have E (e|V ) = e + ?V [V ? E (V |y )]. (4.7)
cov (e,V ) V ar (V ) .
C . f
(4.6)
Then under the perceived cash ?ow distribution (the one based on y (e)), we
When we substitute this expression into equation (4.6), we can see that an external investigation will be triggered if and only if V ? v = E (V |y ) ? e + C/f ? y . ?V (4.8)
This condition implies that when the ?nal cash ?ow realization is su?ciently below the market expectation, outside investors will rationally think they have been misled and will start an investigation. De?ne vc = v ? E ( V |y ) V ar(V ) ,
and let ? denote the standard normal cumulative distribution function. Then the ?rm’s probability of facing an outside investigation under the perceived distribution is7 P rob.[V ? v |y ] = ?(vc ). (4.9)
Yet, the ?rm’s true probability of having an external investigation is not simply ?(vc ). Let ? = ?
1 V ar (V )
be the precision of the ?rm’s ?nal cash ?ow. Then, under the true cash ?ow
distribution (the one based on e), we have P rob.[V ? v |e] = ?(vc + K ), (4.10)
where K = [E (V |y ) ? E (V |e)]? . We can see that when K is positive, the ?rm’s actual probability of cash ?ow detection is strictly greater than ?(vc ). In other words, the more the manager can
7 Since
?(vc ) is not necessarily zero, even an honest ?rm may face an outside investigation. However, if the ?rm
has not misreported, the investigation will not lead to discovery of fraud. Thus the honest ?rm will not be punished even if it may face an outside investigation.
21
raise the market’s expectation about V by false disclosure (E (V |y ) > E (V |e)), the more likely is an outside investigation of fraud (see Figure 2 at the end of the chapter for an illustration). This implies that the bene?t and cost of fraud are endogenously related to each other, and there exists an optimal size of fraud. In sum, the essential point underlying the cash ?ow detection mechanism is that the ?nal cash ?ow realization V is a function of the true earnings realization e, not the reported earnings y (e) (see equations (4.3) and (4.4)). Therefore, investors can update their belief about the probability of misreporting after observing V , whose realization the fraudulent manager cannot fully control. This implies that fraud can be partially self-revealing, which is supported by securities litigation in the United States. Table 2 at the end of the paper lists the corporate events or entities that precipitated the federal private securities class action lawsuits ?led in 1996 and 1997 in the United States. Among the 187 lawsuits, at least 132 cases (or 70.6% of the total) were ?led after some unexpectedly disappointing earnings realizations. Expected Probability of Fraud Detection At time 1, when the manager makes the disclosure
decision y (e), what matters is his expected fraud detection likelihood P . Essentially, P tells the manager how risky it is to commit fraud. Let ?I (?N ) be the probability of cash ?ow detection if the ?rm invests (does not invest). ?I ?N vc,I vc,N KI KN where ?I = ?
1 V ar (V |I )
= = = = = =
?(vc,I + KI ), ?(vc,N + KN ), ? e + C/f ? y ?I , ?V,I e + C/f ? y ? ?N , ?V,N
(4.11) (4.12) (4.13) (4.14) (4.15) (4.16)
[E (V |y ) ? E (V |e)]?I , [E (V |y ) ? E (V |e)]?N ,
and ?N = ?
1 . V ar (V |N )
Now let PI (PN ) denote the e?ective probability of
fraud detection, given that the ?rm invests (does not invest) at time 2. Then according to equation
22
(4.5). We have PI PN = (1 ? ?I )p + ?I , = (1 ? ?N )p + ?N . (4.17) (4.18)
These two equations imply that the probability of fraud detection within the ?rm depends on ?rm-speci?c attributes, such as the quality of corporate governance and the nature of cash ?ows. More important, ?I and ?N depend on the manager’s disclosure strategy (y (e)) and the market’s response (E [V |y (e)]). This implies that the likelihood of detection and thus the litigation cost are endogenous to the manager’s decision making. At time 1, the manager’s expected probability of fraud detection (P ) is simply a weighted average of PI and PN . Let x be the probability that the ?rm will undertake a newly arrived investment project (x will be endogenously determined in section 4.3). Then ?x is the probability that the ?rm will exercise a growth option at time 2. Then we have P = ?xPI + (1 ? ?x)PN . (4.19)
Fraud Penalty
Once fraud is discovered, the ?rm is subject to a legal ?ne of f ? . That is, the fraud penalty is assumed to be proportional to the amount of distortion in the earnings announcement. The ?ne is paid out of the company’s ?nal cash ?ow V . Monetary settlement is a prevailing means of fraud punishment. Of course, there are other negative consequences of fraud such as the negative price response to securities litigation (Gri?n, Grundfest and Perino (2003)), loss of the ?rm’s reputation, persistent increase in the cost of capital (Dechow, Sloan, and Sweeney (1996)), and long-run poor ?rm performance (Baucus and Baucus (1997)). I incorporate all the explicit and implicit fraud consequences in the marginal fraud penalty parameter f and measure them in terms of money. In order to understand the nature of securities fraud and the role of securities litigation (or fraud detection), it is important to know who bears the litigation cost of fraud (i.e., who pays the ?ne and who receives the compensation). Let us consider a typical private securities class action
23
litigation. The plainti? (or class members) is a group of the ?rm’s outside security holders (e.g., equity holders, debt holders) who purchase the ?rm’s public securities during some speci?c time period (class period). Once the lawsuit is settled, the defendant ?rm (or its existing shareholders) pays the settlement to the plainti? investors. In this model, the class period would start at time 1 if the manager makes false disclosure and end at time 3 if the fraud is uncovered. The class members would be the new (and uninformed) shareholders who ?nance the ?rm’s new project at time 2.
4.2.2 Fraud Incentives
If a new investment opportunity arrives at time 2 and the ?rm takes it, the market value of the ?rm based on its earnings disclosure and investment decision is E (V |I, y ), while the true value of the ?rm is E (V |I, e). The di?erence between E (V |I, y ) and E (V |I, e) results from the misreporting of earnings at time 1. In order to undertake the new investment, the ?rm needs to raise $I by issuing a fraction ? (y ) = I E (V |I, y )
of new equity. ? is the percentage ownership of the new shareholders. The expected value to existing shareholders at time 3 is thus (1 ? ? )E (V |I, e). The value of ? indicates the cost of external ?nancing. A high ? means that the incumbent shareholders need to sacri?ce a large fraction of the ?nal cash ?ows in order to raise $I , or a high cost of external capital. We can see that ? is a function of the reported earnings y (e). If E (V |I, y ) increases in y (e), then ? decreases in y (e). This implies that a potential bene?t of committing fraud is that ?nancial misreporting may create (or sustain) short-term market overvaluation of the ?rm’s equity and thus reduce the ?rm’s cost of external ?nancing.8 Of course, there may exist other motives for fraud, such as incentive
8I
assume that the ?rm has to ?nance the new project by raising new equity. Since the bene?t of fraud derives
from the e?ect of ?nancial misreporting on the short-term market valuation of the ?rm’s outside security, the insight of the model will not change if the ?rm can use debt ?nancing. In the debt context, there is still an external ?nancing cost, which is the interest rate the ?rm pays.
24
compensation and insider trading pro?t. This paper, however, focuses on ?rms’ ?nancing and investment incentives in the presence of fraud. Misreporting also comes with a cost: the expected litigation liability. Both the fraud penalty and the probability of detection are functions of ? = y (e) ? e. The cost-bene?t tradeo? leads to the following maximization problem for the manager at time 1. max ? = ?x[1 ? ? (y )]E (V |I, e) + (1 ? ?x)E (V |N, e) ? P (? )f ?,
? ?0
(4.20)
where P (? ) = ?xPI + (1 ? ?x)PN . The expected value to long-term shareholders is their expected ?nal cash ?ow net of the litigation cost. The solution to this problem, ? ? , is the optimal amount of misreporting.
4.3 Securities Fraud and Investment Incentives
In order to solve the manager’s optimization problem in equation (4.20), I need to derive the manager’s investment incentive x in the presence of fraud. Recall that x is the probability that the manager will undertake a newly arrived investment project at time 2. Section 4.3.1 derives the ?rm’s investment incentive at time 2, given its disclosure strategy at time 1. Section 4.3.2 presents a numerical illustration.
4.3.1 Investment Distortions
Suppose that a new investment opportunity does arrive at time 2. The manager privately observes the gross return to the new project as r. If the ?rm issues new equity and invests, the market value of the ?rm’s equity will be E (V |I, y ) = E (A|I, y ) + IE (R|I ). (4.21)
The true value of the ?rm is, however, E (A|e) + Ir. In order to invest, the ?rm needs to issue a fraction ? = I/E (V |I, y ) of new equity. The ?rm also faces the potential litigation liability PI f ? , if ? = 0. Then the expected ?nal payo? to the existing shareholders is (1 ? ? )[E (A|e) + Ir] ? PI f ? if the ?rm invests, or E (A|e) ? PN f ? if the ?rm does not issue and invest. Therefore, for the ?rm 25
to issue and invest, we need (1 ? ? )[E (A|e) + Ir] ? PI f ? > E (A|e) ? PN f ?. (4.22)
A cuto? investment pro?tability rc can be derived such that the above condition is satis?ed when r > rc . In other words, the ?rm will invest if and only if the return to the new investment exceeds some threshold level rc . rc = 1 means that the ?rm will strictly follow the positive NPV rule when making new investment. rc > 1 implies that the ?rm tends to underinvest in the sense that it will pass up some positive NPV projects. rc < 1 implies that the ?rm tends to overinvest in the sense that it will undertake some negative NPV projects. Therefore, the manager’s investment incentive is re?ected in his choice of the cuto? pro?tability to new investments. The model results about the manager’s investment decision are presented in the following propositions. Detailed proofs are provided in the appendix. Proposition 1 Financial misreporting can a?ect the ?rm’s investment incentives. Speci?cally,
? ? the ?rm’s cuto? pro?tability to new investments (rc ) depends on its magnitude of fraud (? ). rc (? )
is the solution to the following equation. rc = where zc = (rc ? R)/?R , and m(zc ) = ?(zc )/[1 ? ?(zc )]. Given the manager’s misreporting strategy ? at time 1, the probability that the ?rm will undertake a newly arrived investment opportunity at time 2 is
? ? x = P rob.[r > rc (? )] = 1 ? ?[zc (? )].
E (A|e) E (A|I, y ) + I?R m(zc )
?
(PN ? PI )f ? , (1 ? ? )I
(4.23)
(4.24)
The lower the cuto? investment pro?tability, the more likely is the ?rm to exercise its growth option at time 2. 26
Proposition 2 Making a new investment can decrease the ?rm’s probability of being investigated at time 3 if the ?rm can boost its market value by overstating its earnings, and either the correlation between the cash ?ow from the new investment and that from the existing assets is in a neighborhood around zero or the cash ?ow from the new investment is volatile enough and the correlation is in some certain range. Speci?cally, PI < PN if E [V |y (e)] > E [V |e] when ? > 0 and one of the following conditions is satis?ed: (1) ? ? [? , ], where is an arbitrary small positive number;
?e (2) max(?1, ? qI? ) < ? < ? ? 1, and I?R > I?R . R
Proposition 3 If the ?rm can boost its market value by overstating its earnings, then the ?rm has
? an incentive to overinvest. That is, if E [V |y (e)] > E (V |e) when ? > 0, then rc < 1. The larger
the magnitude of fraud, the lower is the fraudulent ?rm’s threshold return to new investments,
? ?rc < 0. ??
(4.25)
The essential message in these propositions is that ?nancial misreporting can in?uence the ?rm’s investment incentives in two ways. First, misreporting can in?uence the short-term ?rm value and thus the ?rm’s short-term external ?nancing cost. This e?ect is re?ected in the ?rst term on the right-hand side of equation (4.23). If a low-earnings ?rm overstates its earnings (y (e) > e) to pool with a high-earnings ?rm, and if the market cannot fully see through this, then we have E (A|y ) > E (A|e) for the low-earnings and dishonest ?rm. This implies that the market on average will overvalue the equity of the fraudulent ?rm. This overvaluation lowers the ?rm’s external ?nancing cost and thus gives the ?rm a larger incentive to raise money and invest, resulting in overinvestment. The high-earnings ?rm, however, will su?er from some market undervaluation due to the cross-subsidization between the good ?rm and the fraudulent ?rm. The good ?rm cannot ?nance the new investment on reasonable terms and therefore has less incentive to issue and invest. This is consistent with the underinvestment argument in Myers and Majluf (1984). In sum, the fraud-induced market mispricing implies that a fraudulent ?rm tends to overinvest, and a good and honest ?rm tends to underinvest.
27
Second, ?nancial misreporting can also a?ect the ?rm’s investment decision through the in?uence of investment on the ?rm’s litigation risk. The second term on the right-hand side of equation (4.23) represents the change in the expected litigation cost per investment dollar if the ?rm invests rather than not. If this change is negative, then the reduction in litigation risk will
? push the fraudulent ?rm’s pro?tability threshold rc further down below 1. This means that the
potential negative e?ect of making a new investment on the ?rm’s litigation risk will exacerbate the investment distortion. Given any ? > 0, Proposition 2 states that PI < PN if the investment is uncorrelated with the ?rm’s existing assets or if the investment is risky enough. The basic intuition is as follows. The market observes the combined cash ?ow from the ?rm’s assets in place and from the new investment, and draw inference about the magnitude of misreporting on the asset value based on the total cash ?ow. On one hand, given the level of cash ?ow volatility of the new investment, the inference problem will be most di?cult for the market when the cash ?ow correlation between the new investment and the existing assets is low around zero. On the other hand, given the level of correlation, high cash ?ow volatility from the new investment will decrease the valuation precision of the ?rm’s total cash ?ow and make it harder for the outsiders to see through fraud. Therefore, the incentive to disguise fraud will induce the fraudulent manager to overinvest in risky (high cash ?ow volatility) and uncorrelated projects. In the following analysis, I will focus on the case in which PI < PN . In sum, the key insight in Propositions 1 to 3 is that securities fraud can lead to real value losses. The distorted investment incentive can arise both from the fraud-induced market misvaluation of the ?rm’s assets (E [A|y (e)] = E [A|e]) and from the e?ect of investment on the ?rm’s litigation risk (PI = PN ). Securities fraud can lead to underinvestment by good and honest ?rms and overinvestment by fraudulent ?rms.
4.3.2 A Numerical Illustration
This section presents a numerical example to illustrate the relationship between fraud and the ?rm’s investment incentives. Two levels of earnings realization are considered: eL and eH ,
28
eL < eH . Based on the ?rm’s true earnings realization (e) and its reported earnings (y ), I label the ?rm as one of the following three types. LH ?rm: low earnings (e = eL ) are honestly revealed (y = eL ); HH ?rm: high earnings (e = eH ) are honestly revealed (y = eH ); LD ?rm: low earnings (e = eL ) are reported as high earnings (y = eH ).
? Table 3 presents each type of ?rm’s cuto? return to new investments rc and probability of
making a new investment x (in parentheses). The numerical example reveals the following patterns with respect to the ?rm’s investment incentives in the presence of securities fraud.
? ? (1) The HH ?rm tends to underinvest (rc > 1), and the LD ?rm tends to overinvest (rc < 1).
Put di?erently, the LD ?rm is more likely to exercise its growth option than the HH ?rm and the LH ?rm. These distortions emerge in all three panels. (2) Holding other parameters constant, an increase in the magnitude of misreporting (? ) worsens both the underinvestment problem of the HH ?rm and the overinvestment problem of the LD ?rm (as shown in panel A). This clearly demonstrates the investment distortion spillover between fraudulent and honest ?rms. (3) Holding other parameters constant, an increase in the volatility of the investment return (I?R ) helps to mitigate the underinvestment problem of the HH ?rm but exacerbates the overinvestment problem of the LD ?rm (as shown in panel B). This is because higher investment volatility is associated with higher value of the ?rm’s growth option, which to some extent lessens the market undervaluation of the HH ?rm but worsens the overvaluation of the LD ?rm.9 Furthermore, according to Proposition 2, large I?R also strengthens the negative e?ect of investment on the ?rm’s litigation risk, which motivates the fraudulent ?rm to overinvest. (4) Holding other parameters constant, larger asset volatility (?A ) exacerbates both the underinvestment problem of the HH ?rm and the overinvestment problem of the LD ?rm (as shown
9 The
market’s expected NPV of the new project is I [E (R|I ) ? 1] = I?R m(zc ). Since m(zc ) > 0, a large I?R
scales up the market value of the new project.
29
in panel C). The intuition is that high volatility of the asset value implies less valuation precision of the ?rm’s cash ?ows, which can not only worsen the misvaluation of the ?rm’s assets in place but also decrease the litigation risk of the fraudulent ?rm. (5) Even the LH ?rm tends to overinvest slightly, but this distortion has nothing to do with securities fraud. It arises solely from the e?ect of asymmetric information about the investment return, as shown in Myers and Majluf (1984).10 What is important is the di?erence
? of the LH ?rm and of the LD ?rm, because this di?erence measures the e?ect between the rc
of misreporting on the investment incentive of a low-earnings ?rm. In sum, the numerical illustrations demonstrate that ?nancial misreporting can distort investment decisions in both fraudulent and honest ?rms. The degree of distortion depends on the magnitude of fraud as well as the characteristics of the ?rm’s assets and growth options.
4.4 Disclosure Strategy
? Section 4.3 shows that the manager’s investment incentive (rc or x) can be in?uenced by
?nancial misreporting (? ). Now I move back to time 1 and examine the manager’s disclosure strategy y (e), taking into consideration her investment incentives at time 2. At time 1, the manager privately observes the earnings (e) generated from the ?rm’s assets and makes an earnings announcement y (e) = e + ? (e). That is, given any earnings realization e, the manager optimally chooses the amount of misstatement ? such that the expected value to long-term shareholders at time 3 is maximized. The manager’s objective function is speci?ed in equation (4.20) in section 4.2.2. Now I substitute equation (4.24) into (4.20) and rewrite the manager’s maximization problem as follows.
? ? max ? = ?[1 ? ?(zc )][1 ? ? (y )]E (V |I, e) + {1 ? ?[1 ? ?(zc )]}E (V |N, e) ? P (? )f ?, ? ?0
10 For
(4.26)
? = 1 if and only if E (R|I ) = 1, that is, the market believes new investments are on average the LH ?rm, rc
zero NPV projects. If the market has bullish expectations and believes new projects on average have strictly positive NPV (E (R|I ) > 1), then the LH ?rm will have an incentive to overinvest.
30
? ? where zc ? [rc (? ) ? R]/?R . In sum, misreporting a?ects the manager’s objective function in three
ways. First, it can directly a?ect the short-term market valuation of the ?rm V2 (I, y ) and thus its external ?nancing cost ? (y ). Second, it can indirectly in?uence the long-term performance of the
? ?rm V through the endogenous investment decision rc (? ). Third, misreporting brings a potential
litigation liability P (? )f ? . The optimal strategy balances the bene?t of misreporting and the cost of it. Section 4.4.1 describes a perfect Bayesian equilibrium disclosure strategy y ? (e) = e + ? ? (e). Section 5.3.2 analyzes some important properties of the ?rm’s fraud propensity and the fraud magnitude.
4.4.1 Equilibrium Misreporting
I adopt the perfect Bayesian equilibrium concept to study the manager’s equilibrium misreporting strategy. A perfect Bayesian equilibrium has two requirements. First, the market forms expectations on the ?rm value [E (V |y )] using Bayes’s rule whenever possible. Second, given the market’s beliefs, the manager’s disclosure strategy y (e) maximizes her objective function in (4.27). Proposition 4 An equilibrium disclosure strategy involves partitioning the earnings space into fraud region(s) and nonfraud region(s). Speci?cally, there are three cuto? earnings realizations ?? < el < ec < eh < +?, and the manager’s earnings disclosure strategy is as follows. y ? (e) = e, if e ? ec ,
? y ? (e) = e + ?1 (e) > ec , if el ? e < ec ,
y ? (e) = e, if e < el . Let e denote the earnings value the market infers from y (e). Then the market value of the ?rm’s assets in place after the earnings announcement is ˜|e , e = y ), if y > eh , V1 (y ) = E (A ˜|e , e = y ) + ?1 E [A ˜|e , e = y ?1 (e)], if ec ? y ? eh , V1 (y ) = (1 ? ?1 )E (A 1 ˜|e , e = y ?1 (e)], if el ? y < ec , V1 (y ) = E [A 2 ˜|e , e = y ), if y < el , V1 (y ) = E (A 31
?1 ?1 where ?1 ? P rob.(misreporting |ec ? y ? eh ), y1 (e) = y (e) ? ? 1 (e), and y2 (e) = y (e) ? ? 2 (e). ? 1
and ? 2 are the market’s expected amount of misreporting when ec ? y ? eh and when el ? y < ec , respectively. Detailed proof of this proposition is provided in the appendix. Here I discuss the implications. Proposition 4 implies that the manager will honestly reveal intermediate earnings when the true earnings realization is very good or desperately bad. The manager has an incentive to overstate earnings when the earnings realization is mediocre or fairly disappointing. The intuition is as follows. When the ?rm is in good shape (e > ec ), the manager does not need to overreport earnings at the cost of incurring future litigation liability. When the ?rm is in a shaky condition (el ? e < ec ) but faces some possible future growth opportunities, the manager will rationally want to take the chance and dress up short-term ?rm appearance so that the future growth options can be exercised on favorable terms (i.e., a lower external ?nancing cost). When intermediate earnings happen to be stunningly bad (e < el ), however, moderate overreporting of earnings will not change the picture much. In this case, in order to mimic a high-earnings ?rm, the low-earnings ?rm has to engage in substantial overstatement of earnings, which implies a large potential litigation cost. When the expected cost of fraud exceeds the bene?t, the manager and the shareholders are better o? by honestly revealing the earnings. Proposition 4 shows that outside investors will rationally discount the ?rm’s earnings announcement if el ? y ? eh . When ec ? y ? eh , the fraudulent ?rm pools with high-earnings ?rms. The market value of the ?rm’s assets re?ects a weighted average of the two types. When el ? y < ec , the market fully discounts the reported earnings because the ?rm has an incentive to overreport when its true earnings realization is in this region. Proposition 4 implies, however, that ˜|e , e = y ?1 (e)] if el ? y < ec is el ? y < ec will not be observed in equilibrium. So V1 (y ) = E [A 2 an o?-equilibrium speci?cation.
32
4.4.2 Fraud Propensity
Given any cuto? value ec and el , the ?rm’s probability of misreporting is simply P rob.(f raud) = P rob.(el ? e < ec ). The combination of a high ec and a low el implies a high fraud propensity. Di?erent ?rms can have di?erent cuto? values and thus di?erent likelihoods of misreporting. The fraud region as well as the magnitude of misreporting depend on the structural parameters in the model. The following proposition presents some comparative-static results for ? ? and P rob.(f raud) with respect to some important bene?t and cost parameters. Proof is provided in the appendix. Proposition 5 The ?rm’s fraud propensity and the magnitude of fraud are related to its pro?tability, growth potential, and quality of corporate governance. Speci?cally,
? /?e < 0; (1) ??1
? (2) If PI < PN , then ??1 /?? > 0, and ?P rob.(f raud)/?? > 0;
? (3) ??1 /?p < 0,
?P rob.(f raud)/?p < 0;
The ?rst result states that in the fraud region, the magnitude of misreporting increases as earnings realization decreases. This is because a low-earnings ?rm has high marginal bene?t from misreporting,
?2? ???e
< 0.
The second result shows that if exercising a growth opportunity can decrease the probability of fraud detection, then both the ?rm’s fraud propensity and the amount of misreporting increase in its growth potential (?). In this model, growth can a?ect both the bene?t and cost of engaging in fraud. First, for a rapidly growing but cash-poor ?rm, misreporting business prospects and conditions can create a short-term bene?t by enabling the ?rm to raise external capital on favorable terms to support its growth. Second, growth can decrease the ?rm’s litigation risk, if it can decrease the valuation precision of the ?rm’s cash ?ows,
?P ??
= ?x(PN ? PI ) < 0.
The last result relates the ?rm’s fraud propensity to the quality of corporate governance. Good corporate governance implies more e?ective monitoring of management and thus a better 33
chance that any fraudulent activities within the ?rm will be discovered,
?P ?p
> 0.11
4.5 Model Implications and Discussion
The cost-bene?t analysis of securities fraud provides testable implications for (1) the relation between fraud and investment incentives and (2) the economic determinants of the cross-sectional di?erences in ?rms’ fraud propensities. Fraud and Ine?cient Investment This theory predicts that fraudulent ?rms tend to over-
invest. Yet, the investment can be ine?cient and can lead to serious value destructions. The telecommunications industry is a good illustration. Sidak (2003) o?ers evidence that the prevailing ?nancial misrepresentations in this industry during the past 7 years (particularly by WorldCom) have led to excessive investment and overbuilding. The Eastern Management Group estimates that a signi?cant percentage of the $90 billion invested in that industry was misallocated because of fraudulent growth projections.12 Moeller, Schlingemann, and Stulz (2004) document that in the recent merger wave (1998-2001), acquiring ?rms lost a total of $240 billion surrounding the announcement of acquisitions, and the acquisitions resulted in a net synergy loss of $134 billion (compared to a net synergy gain of $11.5 billion in the 1980s). This implies that the market did not see those investments as value-increasing. Interestingly, Wang (2004b) shows that this period appeared to be fraud-prevailing. Jensen (2004) also provides some good examples of bad investments and value destruction in fraudulent ?rms such as Nortel Networks and eToy. The theory argues that part of the overinvestment incentives arise from the negative e?ect of investment on the ?rm’s detection risk. The model predicts that the type of investment that produces the most valuation imprecision will have the strongest e?ect on detection likelihood. The theory also implies there is investment distortion spillover between fraudulent and honest ?rms. Overinvestment by fraudulent ?rms can crowd out investment by good and honest
11 This
study mainly focuses on the monitoring role of corporate governance, and does not incorporate the broader
functions of governance such as designing executive compensation structures.
12 Eastern
Management Group, supra note 42, at 2 (quoting Joelle Tessler, “WorldCom Spine UUNET is Critical
Part of Internet,” San Jose Mercury News, September 1, 2002).
34
?rms. This implies that fraud-induced real value losses are borne not only by shareholders of fraudulent ?rms but also by those of ?rms that have no intention to misreport. Fraud Propensity and Firm Attributes The theory shows that ?rm characteristics can
in?uence the ?rm’s likelihood of engaging in fraud. Speci?cally, fraudulent ?rms tend to be those who have good growth prospects and large external ?nancing needs, but experience negative profitability shocks. Growth itself is not a bad thing, but this model shows that it can have a signi?cant e?ect on the manager’s fraud incentives (both on the bene?t and cost of fraud). The model predictions are consistent with many ?ndings in the accounting literature on earnings management and corporate fraud. Loebbecke, Eining, and Willingham (1989) study a small sample of managerial frauds and conclude that the most signi?cant “red ?ags” for fraud are rapid company growth and poor accounting performance. The National Commission on Fraudulent Financial Reporting (1987) states that young public companies have a proportionately greater risk of ?nancial statement fraud. Young ?rms generally have higher growth potential than mature ?rms. Litigation Across Industries The model predicts an industry e?ect in the cross-sectional
distribution of securities fraud. That is, there will be “litigation clustering” in certain industries during a speci?c time period. This is because both ?rms’ bene?t from fraud (such as asset profitability and growth potential) and litigation risk are correlated within an industry, which implies that ?rms’ fraud propensities will be in?uenced by industry factors. E?ect of Increasing Disclosure The model shows that increasing the informativeness of
the earnings has an ambiguous e?ect on the ?rm’s likelihood of committing fraud. This implies that imposing heavy disclosure requirements on public ?rms may not produce the expected e?ects. The reason is that increased disclosure could give the market an illusion of increased transparency, which could actually decrease market vigilance. Fraud Detection Likelihood This theory shows that while the fraud penalty (f ) is largely
determined by securities laws and regulations, fraud detection likelihood (P ) is substantially in?uenced by the ?rm’s endogenous actions as well as ?rm-speci?c attributes. This implies that the probability of detection is more important than the penalty in determining cross-sectional
35
di?erences in ?rms’ fraud propensities. The policy implication is that raising litigation liability standards alone will achieve only limited deterrence, because ?rms may adjust P to o?set some e?ect of increased f on their expected litigation cost. More important, the theory shows that ?rms may even destroy value in order to decrease their detection risk, which can be an unintended consequence of imposing heavy penalty. Ine?cient investment is one example. Leuz, Triantis and Wang (2004) provide possibly another. They document that since the passage of Sarbanes-Oxley Act there has been a dramatic surge in the number of public ?rms that voluntarily deregistered their common stock and ceased to ?le regular reports with the SEC (they call this “going dark” transactions). They also document substantial negative abnormal returns and loss of liquidity associated with deregistration and continued drop in the ?rms’ market capitalization after deregistration. Their ?ndings imply that insiders of those companies may have sacri?ced shareholders’ interest in order to hide from market scrutiny. Internal Corporate Governance and Extensions This paper shows that even when the
manager’s interest is perfectly aligned with that of shareholders, fraudulent behavior can still emerge, because incumbent shareholders may ?nd it advantageous to defraud prospective investors. Good corporate governance will not completely prevent fraud if it is under the control of existing shareholders. In fact, Table 2 shows that the likelihood of fraud detection is much lower from within the ?rm than from outside. Therefore, enhancing other detection forces such as capital market vigilance, responsibility of “gatekeepers” (e.g., auditors and lawyers) and securities regulation is necessary in combating corporate fraud. In the present model, the quality of internal corporate governance p is exogenously determined, and I focus on detection by capital markets. A more general model can allow shareholders of the company to choose the level of p, and allow the market to incorporate this information into its belief about the likelihood of fraud (?0 = g (p), g (p) < 0). Therefore, a higher p corresponds to a higher ex ante bene?t from fraud because it leads to a lower ?0 and thus a smaller discount of the ?rm’s earnings report (the signalling e?ect). As illustrated by Figure 2, however, a larger di?erence between E (V |y ) and E (V |e) also implies a higher likelihood of cash ?ow detection. This
36
means that a higher p will increase the likelihood of both internal and external fraud detection (the litigation e?ect). The optimal quality of internal corporate governance p? balances the signalling e?ect with the litigation e?ect. Since in this paper the manager represents the interests of incumbent long-term shareholders, the extension is equivalent to having a model in which the manager chooses ? and p at the same time (i.e., time 0 and time 1 are combined). The manager’s optimization problem can be as follows. max
? ? = E (V |N, e) + ?[1 ? ?(zc )][?0 ? ? (y, p)]E (V |I, e) ? P (?, p)f ? ? h(p),
? ?0,0?p?1
(4.27)
where h(p) is the cost of building the quality of internal corporate governance. p? depends on the functional form of g (p) and h(p). For example, if the market is not sensitive to corporate governance (at least for some range of p realizations), then the ?rm will choose a p as low as possible, regardless of its fraud propensity. If the market values good corporate governance but it is very costly to build up the quality, then the ?rm may still lean towards a low p. If the market values good governance and the cost of establishing good governance is reasonable, then the choice of p will depend on the ?rm’s ex ante fraud incentives.
37
Table 4.1: Fraud Discovery (1996-1997) This table lists the various corporate events or entities that precipitated the 187 federal securities class action lawsuits during 1996 and 1997. The litigation information is retrieved from Stanford Securities Class Action Clearinghouse. Information about the triggering events of each lawsuit is extracted from the relevant case documents (i.e., the case complaints, the press releases, and the court decisions). The ?rst column of the table lists the event or entity that precipitated or initiated the securities lawsuits. The triggering events can overlap in some lawsuits. Precipitator Number of observation Devastating news announcement Regulators (mostly SEC) Independent auditors Business journal articles Board/internal investigation Securities analysts Shareholder/Investor Stock Exchanges/credit rating services Management turnover 1996 93 63 6 10 7 7 1 3 0 2 1997 94 69 6 7 5 4 3 4 1 1 Total 187 132 12 17 12 11 4 7 1 3 % of Total 70.59 6.42 9.09 6.42 5.88 2.14 3.74 0.53 1.60
38
Table 4.2: A Numerical Illustration of Investment Incentives I assume the following parameter values. The value of the ?rm’s assets in place is normally distributed with expectation A = 100 and volatility ?A = 30. The average return on assets is q = 0.16. The earnings noise u is normally distributed with zero mean and volatility ?u = 4. 2 + ? 2 = 6.25. The size of The expected earnings is e = qA = 16, and volatility is ?e = q 2 ?A u the new investment is I = 25. The volatility of investment return is I?R = 25 ? 0.3 = 7.5. The correlation coe?cient between R and e is ? = 0.3. The market’s prior belief about the probability of misreporting is ?0 = 0.5. The marginal fraud penalty is f = 1.5. The institutional e?ciency is p = 0.3. The cost of investigation is C = E (f ? ) = f ? . In panel A, I set eL = e ? ?e = 9.75. I consider two levels of eH . First, eH = e = 16, which means that ? = 6.25. Second, eH = e + ?e = 22.25, which means that ? = 12.5. In panels B-C, ? = 12.5. Panel A: Fraud Magnitude and Investment Bias ? = 6.25 0.83 (71%) 0.94 (58%) 1.05 (43%) ? = 12.5 0.73 (81%) 0.94 (58 %) 1.14 (33%)
LD LH HH
Panel B: Investment Volatility and Investment Bias I?R LD LH HH 2.5 0.76 (79%) 0.98 (53%) 1.19 (26%) 7.5 0.73 (81%) 0.94 (58%) 1.14 (33%) 12.5 0.71 (83%) 0.91 (62%) 1.09 (38%) 17.5 0.69 (85%) 0.89 (65%) 1.06 (42%)
Panel C: Asset Volatility and Investment Bias ?A 1.14 0.74 0.94 = 30 (31%) (81%) (58%) ?A 1.17 0.69 0.94 = 40 (28%) (85%) (58%)
HH LD LH
39
Figure 1: Model Time Line
0 Quality of governance 0? p? 1 is set, which later determines the prob. of internal fraud detection.
1 The manager privately observes the intermediate earnings e from existing assets A. The manager makes a disclosure decision y(e) =e + ?.
2 A new investment opportunity comes with probability ?, requires an initial cost of $I, and generates gross return R. The manager observes the realization of R as r, and makes the investment and financing decisions.
3 The firm generates a liquidating cash flow V. Misreporting is detected with prob. P. A penalty f? is imposed upon detection.
Figure 2: Probability of Fraud Detection
True
Report
vc
E (V | e)
E (V | y )
Note: In this figure, the shaded area represents the probability of cash flow detection.
40
Chapter 5 An Empirical Investigation of Securities Fraud
This chapter empirically investigates the economic determinants of ?rms’ fraud propensity and the fraud detection likelihood. More speci?cally, I address the following research questions: 1. How does investment in?uence the ?rm’s fraud incentives and their detection risk? 2. What are the roles of di?erent corporate monitors in the context of fraud? What type of corporate monitor has been e?ective in discovering corporate fraudulent activities? 3. What is the role of insider equity incentives in determining the ?rm’s propensity to commit fraud? 4. How are ?rm characteristics related to the ?rm’s likelihood of engaging fraud and the ?rms’ likelihood of getting caught? The structure of this chapter is as follows. Section 5.1 describes the accounting fraud sample and presents some stylized facts about accounting-related class action lawsuits from 1996 to 2003. Section 5.2 presents the empirical model of fraud. Section 5.3 discusses the related literature and develops the empirical hypotheses. Section 5.4 reports the results from univariate comparisons between the fraud sample and the comparison sample. Section 5.5 reports the multivariate analysis on the determinants of ?rms’ propensity to commit fraud and the likelihood of fraud detection. Section 5.6 presents robust checks on the model results regarding the possibility of false detection, the timing of fraud, and di?erent model speci?cations.
5.1 Fraud Sample
The fraud sample in this study is based on Securities Class Action Clearinghouse (SCAC) established by Stanford Law School. This clearinghouse provides a comprehensive database of
41
federal private securities class action lawsuits ?led since 1996 in the United States. A private securities class action is a case brought pursuant to Federal Rule of Civil Procedure 23 on behalf of a group of persons (class members) who purchased the securities of a particular company during a speci?ed time (class period). A suit is ?led as a class action because the members of the class are so numerous that joinder of all members is impracticable. I went through the details of all the available case documents associated with each lawsuit (e.g., case complaints, press releases, court decisions, etc.) to identify the nature of fraud allegations. As a result, I singled out 684 lawsuits ?led against 660 US public companies during 1996 to 2003 involving allegations of accounting irregularities. For ?rms that had multiple securities lawsuits, I only use the earliest one in the analysis. Existing studies mostly rely on the SEC’s Accounting and Auditing Enforcement Releases (AAERs) to identify accounting frauds. Several recent studies use accounting restatements to proxy for fraudulent ?nancial reporting (Agrawal and Chadha (2004), Burn and Kedia (2004), Efendi, Srivastava and Swanson (2004)). This paper is the ?rst to study class action litigation involving accounting-related allegations. Private class action litigation has long been an important concomitant to the enforcement of securities laws (Cox and Thomas (2003)). The volume of class action lawsuits is also comparable to that of the SEC’s enforcement actions. More important, class action litigation can provide new insights for understanding market forces in securities litigation, because class action suits generally involve the interests of thousands of investors, and key plainti? investors play an important role in the litigation. My class action sample does overlap with the SEC’s AAER sample and the accounting restatement sample that have been used in the existing studies. Among the 660 fraudulent ?rms in my sample, 207 ?rms were subject to parallel SEC’s AAERs, and 334 ?rms had accounting restatements according to the General Accounting O?ce’s October 2003 report.1 In Section 5.6, I will use these two subsamples to check the robustness of my results across di?erent proxies of
1 The
General Accounting O?ce’s October 2003 report lists all the accounting restatements between January
1997 and June 2003. My sample period is from January 1996 to December 2003. Therefore, there can potentially be more than 334 restatements in my sample.
42
securities fraud. The following sections provides detailed descriptive information about the fraud sample.
5.1.1 Time Trends and Firm Characteristics
Table 1 describes the evolution of class action litigation over time. Panel A shows that accounting-related frauds have on average accounted for about 47% of the total litigation activities (excluding litigation against investment banks) over the past 8 years. The number of accounting frauds peaked in 2002, where it represented 56.13% of all the securities class action ?lings. Interestingly, accounting-related litigation substantially decreased in 2003 (only 40% of all lawsuits), which may have resulted from tightened securities regulation and increased market vigilance. Panel B shows the distribution of the class periods associated with the 660 lawsuits. Every class action lawsuit speci?es a class period. The beginning of a class period shows the earliest time a fraud a?ects the market, based on the judgment of securities attorneys. A class period generally ends at the time of some major events that precipitate the litigation. The length of the class period provides some information about the duration of fraud. The average length of the class period is a little more than one year, but there is substantial variation. Some frauds a?ected the market for more than ?ve years, while some less than a quarter. Panel B also shows that ?rms in the fraud sample were largely young public companies. The median age was only 3.45 years, and more than 60% of the sample ?rms were less than 5 years old. About 64% of the alleged fraudulent ?rms were listed on NASDAQ when fraud began. Panel C shows the distribution of the ?scal year in which fraud began. I label the beginning ?scal year of fraud as year 0. I determine year 0 based on the beginning of the class periods and the ?rms’ ?scal year end. The beginning of a class period indicates the earliest time fraud a?ected the market, but does not necessarily indicate the beginning of fraud. In general, an accounting fraud starts to a?ect the market when a fraudulent ?nancial report is released to the public. Given that there is about one month’s lag for quarterly reports and a two-to-three-month lag for annual reports, year 0 can be the same ?scal year in which the class period starts, or the previous ?scal
43
year. Figure 1 illustrates the two scenarios. If a ?rm was subject to both private litigation and the SEC’s enforcement action, I cross check the beginning year with that speci?ed by the SEC. The information about the SEC’s enforcement actions is retrieved from the SEC’s litigation archive.
5.1.2 Industry Distribution
Table 2 presents the industry distribution of fraud. I classify the alleged fraudulent ?rms into 24 industry categories. The primary classi?cation is based on two-digit SIC codes, but in some instances, I use three-digit SIC codes, as this is more informative about the types of companies that engaged in fraud. Table 2 shows evidence of signi?cant industry patterns in securities fraud litigation. First, technology ?rms are disproportionately more involved in accounting-related securities litigation. In particular, ?rms in software and programming alone accounted for 17.42% of all accounting fraud cases in the past 8 years. Electronic parts, computer manufacturing, and telecommunications companies represent another 19% of the litigation activities. Second, the service sector and particularly the ?nancial services and the business services industries also show a high litigation concentration. In total, the technology (including bio-technology ?rms) and service sectors account for 67% of all securities lawsuits studied in this paper.
5.1.3 The Nature of Fraud
Table 3 lists some speci?c accounting items that are often manipulated, based on the relevant case documents in 563 class action lawsuits.2 Allegations of improper revenue recognition are most common, accounting for 67.44% of all the accounting fraud allegations. Operational expenses are also likely to be manipulated by managers to reach desired earnings targets (17.26% of the 563 cases). As for the balance sheet items, misstatements of assets are more frequently observed than misstatements of liabilities and equity. Among the di?erent types of assets, accounts receivable and inventory seem to be frequently misstated. This observation is consistent with the ?ndings in Chan et al. (2005) that changes in inventory and accounts receivables are closely related to
2I
am only able to clearly identify the speci?c accounting items in 563 out of 660 cases.
44
the earnings quality and thus can help to predict future stock returns. Finally, understatement of reserves and allowances is also fairly often, accounting for about 9% of the 563 lawsuits.
5.2 Empirical Methodology 5.2.1 A Model with Partial Observability of Fraud
In implementing comparisons between the fraud sample and any sample of non-convicted ?rms, we face an identi?cation problem because we only observe detected fraud. That is, we only observe frauds that have been committed and subsequently detected. Firms that have not been sued in securities litigation are either innocent ?rms or undetected fraudulent ?rms (see Figure 2 for an illustration). This implies that the probability of detected fraud (what we observe) is di?erent from the probability of fraud (what we are interested to estimate but cannot observe), unless detection is perfect. To address this identi?cation problem, I use a bivariate probit model with partial observability as discussed in Poirier (1980) and Feinstein (1990). In essence, this technique models the observed outcome (detected fraud) as a function of the joint realizations of two latent processes.
? Let Fi? denote ?rm i’s potential to commit fraud, and Di denote the ?rm’s potential of getting
caught conditional on fraud being committed. Then consider the following reduced form model: Fi?
? Di
= =
xF,i ?F + ui ; xD,i ?D + vi ,
(5.1) (5.2)
where xF,i contains variables that help explain ?rm i’s potential to commit fraud, and xD,i contains variables that help explain the ?rm’s detection risk. ui and vi are zero-mean disturbance terms, and follows a bivariate normal distribution. Their variances have been normalized to equal unity. The correlation between ui and vi is ?. Now I de?ne the following binary variables. Fraud occurrence: Fi = 1 if Fi? > 0, and Fi = 0 if otherwise;
? Fraud detection : Di = 1 if Di > 0, and Di = 0 if otherwise.
45
We, however, do not directly observe the realizations of Fi and Di . What we observe is Zi = Fi Di Zi = 1 if ?rm i has committed fraud and has been detected, and Zi = 0 if ?rm i has not committed fraud or has committed fraud but has not been detected. Let ? denote the bivariate standard normal cumulative distribution function. The empirical model for Zi is P (Zi = 1) = P (Fi Di = 1) (5.3)
= P (Fi = 1, Di = 1)
? = P (Fi? > 0, Di > 0)
= ?(xF,i ?F , xD,i ?D , ?); P (Zi = 0) = P (Fi Di = 0) (5.4)
= P (Fi = 0, Di = 0) + P (Fi = 1, Di = 0) = 1 ? ?(xF,i ?F , xD,i ?D , ?). An implicit assumption in the model is that false detection of fraud is not allowed for
? (P (Fi = 0, Di = 1) = 0), because the process of Di is only de?ned conditional on Fi = 1.
Extension of the above model to statistically control for false detection is possible, but it tends to complicate the estimation.3 I will come back to the issue of false detection in Section 5.6.
? Although I de?ne Di conditional on Fi = 1, the correlation between the two disturbance
terms ? may not necessarily be zero. As discussed in Feinstein (1990), a non-zero correlation may arise for a number of reasons, particularly when the potential fraud-doer and the detection force possess information about one another.
3 Let
Di = 1 indicate false detection. Then P (Zi = 1) = P (Fi = 1, Di = 1) + P (Fi = 0, Di = 1); P (Zi = 0) = P (Fi = 0, Di = 0) + P (Fi = 1, Di = 0) ? P (Fi = 0, Di = 1).
In a well-functioning legal environment, P (Di = 1) should be very small, much smallers than P (Di = 1). Then assuming P (Di = 1) = 0 will not substantially bias the model estimation.
46
5.2.2 Model Identi?cation and Estimation
The partial observability of fraud raises a model identi?cation issue. This is because we only observe the joint outcome of two latent processes, and the decomposition between the two latent components may not be unique. According to Poirier (1980), the conditions for full identi?cation of the model parameters are (1) xF,i and xD,i do not contain exactly the same variables; and (2) the explanatory variables exhibit substantial variations in the sample. The above model can be estimated using the maximum-likelihood method. The log-likelihood function for the model is L(?F , ?D , ?) =
zi =1
log[P (zi = 1)] +
zi =0
log[P (zi = 0)]
(5.5)
=
i=1,...,n
{zi ln[?(xF,i ?F , xD,i ?D , ?) + (1 ? zi )ln[1 ? ?(xF,i ?F , xD,i ?D , ?)]}.
I use the ?ling of class action lawsuits to proxy for detected fraud (i.e., Z = 1). The partial observability model implies that the appropriate comparison sample (Z = 0) should be a random sample of non-litigated ?rms. I therefore use all the ?rms in the COMPUSTAT database that have not been subject to any private securities litigation (accounting-related or not) or the SEC’s AAERs between 1996 and 2003.
5.2.3 Comparison with Straight Probit Model
A straight probit model, which has been used in many existing studies on fraud, is as follows. For ?rms i=1,...,n, P (Di = 1) = 1; P (Zi = 1) = P (Fi = 1) = ?(xF,i ?F ). The log likelihood function associated with this model is L(?F ) =
i=1,...,n
{zi ln[?(xF,i ?F ) + (1 ? zi )ln[1 ? ?(xF,i ?F )]}.
(5.6)
We can see that as long as detection is not perfect (i.e., P (Di = 1) ? 1), the straight probit model will systematically understate the true probability of fraud. 47
An inference problem could also arise when (5.6) is estimated instead of (5.5). For example, we want to examine the marginal e?ect of an explanatory variable xi on the probability of fraud P (Fi = 1). Let us take partial derivative of xi on both sides of equation (5.3). ?P (Zi = 1) ?P (Fi = 1) ?P (Di = 1|Fi = 1) = P (Di = 1|Fi = 1) + P (Fi = 1). ?xi ?xi ?xi If this variable has opposite e?ects on P (Fi = 1) and P (Di = 1|Fi = 1), then
?P (Zi =1) ?xi ?P (Fi =1) ?xi
(5.7) and
can even have di?erent signs, not to mention that the magnitude will be di?erent. This
may lead us to draw incorrect inference about the role of xi . Section 5.5.6 provides concrete examples of the discussions here.
5.3 Hypothesis Development and Model Speci?cation
Following the framework of fraud in Wang (2004), a ?rm’s propensity to commit fraud depends on its expected bene?t and cost from engaging in fraud. The expected cost of fraud is the litigation risk: with some positive probability, fraudulent activities will be uncovered, resulting in a penalty. Wang (2004) argues that while the penalty (at least the explicit liability provision) is largely determined by securities laws and thus exogenous to the ?rm, the probability of detection depends on the ?rm’s endogenous actions (e.g., investment, disclosure) as well as ?rm-speci?c attributes. This implies that the detection risk is a more important determinant of the crosssectional variations in ?rms’ fraud propensities than are penalty provisions. Therefore, I focus on the likelihood of detection for the cost side of the tradeo?. A factor will positively in?uence a ?rm’s fraud propensity if it can increase the ?rm’s bene?t from committing fraud, or if it can decrease the ?rm’s expected probability of getting caught, or both. The structure of this section is as follows. Sections 5.3.1 and 5.3.2 discuss factors that can potentially a?ect a ?rm’s detection risk and its bene?t from fraud, respectively. Section ?? discusses the control variables. Section 5.3.4 summaries the model speci?cation.
48
5.3.1 Probability of Fraud Detection
The probability of fraud detection essentially determines how risky it is for a ?rm to engage in fraud. If a factor can signi?cantly in?uence such probability and if its e?ect can be anticipated at the time the ?rm makes the fraud decision, then this factor will in?uence the ?rm’s ex-ante propensity to commit fraud (in the opposite direction). Therefore, I start with the determinants of the fraud detection likelihood (i.e., xD ), and then move to the determinants of fraud propensity (i.e., xF ) in Section 5.3.2.
Investment
Wang (2004) argues that fraudulent ?rms tend to overinvest. The overinvestment incentive is twofold. First, fraud can create short-term market overvaluation of the ?rm and thus decrease the external ?nancing cost of investment. Second, after committing fraud, the fraudulent ?rm has incentive to cover things up. This incentive can motivate the management to strategically use investment to disguise fraud. Wang shows that investment with high uncertainty and/or low correlation with current activities can mask fraud better than others, because these types of investment can decrease the precision of the ?rm’s cash ?ows and create inference problems for the market. Wang’s argument has the following three testable implications: (1). Fraudulent ?rms have larger investment expenditures than comparable honest ?rms; (2). Di?erent types of investment have di?erential e?ects on a ?rm’s probability of being detected and the probability of fraud. Risky investments and uncorrelated investments have stronger negative e?ects on the detection likelihood than other types of investments; (3). Financing of the investment in?uences a ?rm’s probability of committing fraud. Externally-?nanced investment will motivate fraud better than internally-?nanced investment. To test the above implications, I investigate three types of investment: investment in research & development (R&D), capital expenditures, and mergers/acquisitions. These investments can substantially di?er in their e?ects on a ?rm’s valuation precision. Investment outcome of R&D
49
projects is generally highly uncertain. It is di?cult for the market to fully understand and correctly value its impact on the ?rm value. Capital expenditures tend to be more straightforward. COMPUSTAT de?nes capital expenditures as the funds used for additions to the company’s property, plant and equipment. Mergers and acquisitions, in theory, should fall in the middle, because the investment is to acquire an existing asset rather than to create something new. However, the true value of the acquired assets and the synergy between the acquired and the existing assets may not be correctly understood by the market or even the acquirer. I further distinguish between cash-based acquisitions and stock-based acquisitions. The earnings management literature has provide evidence that stock-based acquisitions are associated with higher incentive of earnings management (e.g., Erickson and Wang (1998)). In this study, I examine the e?ect of stock-based acquisitions on both the probability of fraud and the probability of fraud detection. I also distinguish between focused acquisitions and diversifying acquisitions. I de?ne focused acquisitions as acquisitions within the same two-digit SIC codes. According to Wang (2004), focused acquisitions should be associated with higher probability of detection than diversifying ones are.
Corporate Monitoring
E?ective monitoring over the management should increase the likelihood of fraud detection and deter fraud ex ante. In this study, I examine the roles of four types of corporate monitors in the context of corporate securities fraud: large shareholders, institutional owners, independent auditors, and board of directors.
Monitoring by Shareholders: A ?rm’s ownership structure is important in determining both the ?rm’s bene?t from committing fraud and its detection risk. This is because the ownership structure is crucially related to the incentive structure within the ?rm, including the incentive of the management to defraud outside investors and the incentive of shareholders to monitor the management and detect fraud. The monitoring role of large shareholders has received a great amount of attention in the 50
?nance and economics literature. Shleifer and Vishny (1997) argue that concentrated ownership is a key element of a good corporate governance system because large shareholders have high incentive and power to impose e?ective monitoring over the management. There has been quite some empirical evidence on the role of large shareholders in corporate governance (see a recent survey by Holderness (2003)). For example, Bethel, Liebeskind and Opler (1998) ?nd that company performance improves after an activist investor purchases a block of shares. Bertrand and Mullainathan (2001) ?nd that the presence of a large shareholder on the board is associated with tighter control over executive compensation. In the context of corporate fraud, it is also intuitive that large shareholders should go against fraudulent reporting, because they cannot cash out in a short period of time to catch the windfall from fraud, and they will likely su?er a lot from the severe consequences of fraud. Therefore, I expect a positive relation between block ownership holding and the likelihood of fraud detection. Large shareholders are often institutional investors. Monitoring by institutional shareholders has attracted growing public and academic interest, as institutional ownership skyrocketed over the past two decades in the United States. The Private Securities Litigation Reform Act (PSLRA), which was passed in December 1995, explicitly encourages more active participation of institutional investors in securities litigation by requiring each class action lawsuit to specify a lead plainti?. William Lerach, a partner in Milberg Weiss Bershad Hynes & Lerach LLP and a leader in representing investors in securities class action suits, points out that some large pension funds have actively participated in securities litigation and have successfully established corporate governance enhancements in class action settlements.4 Therefore, I expect institutional equity holdings to have a positive e?ect on fraud detection.
Monitoring by Independent Auditors: Independent auditors are probably the most important corporate “gatekeepers”. They pledge their reputational capital and provide protections to dispersed investors by verifying and assessing the quality of ?rms’ disclosures. In the late 1990s, however, such protections seemingly failed. The most notorious example is Arthur Andersen’s role in the
4 Keynote
address by William S. Lerach in council of institutional investors spring 2001 meeting.
51
Enron scandal and its subsequent criminal indictment. The increasing importance of non-audit services in auditing ?rms’ total revenue has also led to widespread market concern about auditor independence. Frankel, Johnson, and Nelson (2002) ?nd that auditor independence is negatively associated with the probability of earnings management. Anup and Chadha (2004) ?nd a negative but insigni?cant relation between auditor independence and the probability of accounting restatements. Bajaj, Gunny and Sarin (2003) examine a sample of class action lawsuits that involve allegations of accounting irregularities, and ?nd no signi?cant di?erence in auditors’ compensation (audit vs. non-audit fees) between the fraud sample and the comparison sample. However, for ?rms with large market reaction to the alleged fraud, their auditors have signi?cantly higher non-audit income. In this study, I directly examine whether higher auditor reputational capital leads to higher likelihood of fraud detection. First, I examine whether ?rms whose independent auditor is one of the ?ve largest accounting ?rms (Arthur Andersen, PricewaterhouseCoopers, Deloitte & Touche, Ernst & Young, KPMG) have a higher probability of fraud detection. Second, I examine the role of auditor opinion in fraud detection. If the independent auditors exert due diligence in certifying disclosures, then I expect that adverse auditor opinions to increase fraud detection.
Monitoring by Board of Directors:
The monitoring role of the board of directors is an impor-
tant component of corporate governance. The board is presumed to monitor the management on behalf of shareholders, because di?use ownership makes direct shareholder control di?cult. The economics and ?nance literature on the board starts with the assumption that the board’s monitoring e?ectiveness is a function of the board’s independence from the management. Two characteristics of the board, size and composition, are related to board independence. Empirical research in this area ?nds that board size and composition a?ect the observable board actions such as the board’s decision on CEO turnover, executive compensation, and merger/acquisitions (see surveys by John and Senbet (1998) and Hermalin and Weisbach (2003)). The recent wave of high-pro?le corporate scandals has brought the e?ectiveness of board monitoring to the center of securities legislation and governance reform. The newly-passed Sarbanes52
Oxley Act (SOX) and the NYSE and NASDAQ’s new corporate governance guidance mandate a number of changes that are aimed to improve board monitoring. For example, SOX requires that the audit committee consist entirely of independent directors and the audit committee hire the outside auditor. Both SEC and the national stock exchanges strongly recommend overall board independence. Several studies have examined the relation between the characteristics of the board and the probability of corporate fraudulent reporting. Beasley (1996) studies a sample of ?rms subject to SEC’s AAERs and ?nds that board independence (proxied by the percentage of outside directors in the board) is signi?cantly negatively related to the likelihood of ?nancial statement fraud. Klein (2002) ?nds an inverse relation between board independence and abnormal accruals. Dechow, Sloan and Sweeney (1996) ?nd that ?rms committing ?nancial statement fraud are likely to have a board dominated by insiders and have a CEO who is also the chairman of the board or the founder of the company. Agrawal and Chadha (2004) examine the incidence of accounting restatements, and ?nd that board independence is irrelevant, but the presence of independent directors with ?nancial or accounting expertise on the audit committee is associated with signi?cantly lower probability of accounting restatements. In this study, I examine the e?ect of board independence on the likelihood of fraud detection. Following the literature, I use board size and the percentage of outside directors to proxy for board independence. “Grey” directors who are not employees of a ?rm but have some business relation with the ?rm are not counted as outside directors.
Unexpected Performance Shock
Wang (2004) argues that fraud can be partially self-revealing. If the manager in?ates the earnings and misleads the market to have a high expectation on the ?rm’s future cash ?ows, then if later the cash ?ow realization turns out to be comparably bad (which the manager cannot fully control), outside investors will rationally think that they probably have been fooled and will start an investigation. Therefore, unexpected bad performance (unexpected by the market) after the commencement of fraud will increase the probability of fraud detection.
53
To proxy for such unexpected performance shock, I use the regression residual term from the following simple prediction model. ROAi,1 = ?0 + ?1 ROAi,0 + ?2 ROAi,?1 + i . (5.8)
ROAi,t is ?rm i’s return on asset in year t, which is de?ned as the ratio of operating income after depreciation over the average total assets from year t ? 1 to year t. ROAi,1 is used as the dependent variable because the average length of the class period is about one year.
i
(the residual ROA)
will be low if ?rm i’s performance in year 1 is bad compared with the (reported) performance in the previous two years. The realizations of this variable cannot be fully expected in year 0 when the management makes the fraud decision. Therefore, although this variable may signi?cantly in?uence the ?rm’s detection risk, its e?ect is ex-post and thus should not a?ect the ?rm’s ex-ante fraud decision.5
5.3.2 Propensity to Commit Fraud
The equilibrium supply of fraud depends on the expected bene?t and cost of engaging in fraud. Therefore, xF should include factors that can a?ect either the bene?t from fraud, or the litigation risk, or both. The previous section discusses some potential determinants of the detection risk. Now I turn to factors that can potentially in?uence a ?rm’s bene?t from committing fraud.
5 There
are two caveatees associated with this variable. First, this variable is not completely exogenous. The
direction of causality, however, is not ambiguous. It is intuitive that bad operating performance eventually reveals fraud. Detection of fraud may result in immediate plunge in stock returns and may a?ect the long-run performance of the fraudulent ?rm, but it is hard to believe that the revelation of fraud leads to immediate bad operating performance. Second, the management could have better information about future abnormal bad performances than the market does. Therefore, expectation of the residual ROA may impose some ex ante deterrence. However, a reasonable counter argument is that the managers commit fraud because they believe that the current bad performance is only temporary and things should go back to normal later. I will discuss the robustness of the results regarding this variable in Section 5.6.
54
Pro?tability and Growth Potential
Wang (2004) and Bebchuk and Bar-Gill (2002) predict that ?rms that have high growth potential but experience negative pro?tability shocks have high propensity to commit fraud. The intuition is that for such ?rms misreporting short-term ?rm performance can allow them to raise external capital and exercise their growth options on sweet terms. A problem emerges when we test the above prediction. We cannot directly observe the negative pro?tability shocks because they are covered by fraud. A possible solution is to use the ex-post restated ?nancial data rather than the originally reported data. However, to my knowledge, the restatement data in COMPUSTAT is not as comprehensive and complete as the original data. Therefore, I try to infer the existence of pro?tability shocks by comparing the pro?tability before the commencement of fraud and that at the revelation of fraud. The di?erence between the two pro?tability levels can imply hidden performance changes when fraud is alive. I use return on asset ROA as the pro?tability measure. I use two proxies for growth potential, the annual asset growth rate and the book-to-market ratio.
External Financing Needs
The combination of low asset pro?tability and high growth implies large reliance of a ?rm on the external capital markets. Stein (1989) argues that the lack of ?nancial slack can expose the manager to capital market pressure and can motivate the manager to in?ate short-term performance at the cost of forfeiting long-term values. The earnings management literature has provided evidence that managers tend to overreport earnings prior to major external ?nancing activities such as public equity o?erings (see, e.g., Teoh, Welch and Wong (1998a,b)). I construct two variables to proxy for a ?rm’s external ?nancing needs. The ?rst variable, externally ?nanced growth rate, is constructed based on Demirg¨ uc ¸-Kunt and Maksimovic (1998) to proxy for a ?rms’ projected need for outside capital. Speci?cally, the externally-?nanced growth rate is a ?rm’s asset growth rate in excess of the maximum growth rate that can be supported by the ?rm’s internally available
55
capital (ROA/(1-ROA)).6 The second variable, EF , is constructed following Richardson and Sloan (2003) to measure a ?rm’s actual net external ?nancing cash ?ows. Speci?cally, EFt = ?CEt + ?P Et + ?DEBTt , ASSET St
where ?CEt , ?P Et , and ?DEBTt are the changes in the book value of common equity, preferred equity, and total debt in year t, respectively. ASSET St is the book value of assets.7 This variable can be viewed as a measure of a ?rm’s realized external ?nancing need. Since the second variable is an outcome-based measure, I focus on the ?rst variable in order to reduce endogeneity, and use the second measure only as a robustness check.
Financial Distress
Another factor that is closely related to ?nancial slack and external ?nancing need is the degree of ?nancial distress. Maksimovic and Titman (1991) theorize that ?nancial di?culties can a?ect a ?rm’s incentive to honor its implicit contracts and in other ways maintain a favorable reputation. In their model, both ?nancial shortfalls and overall debt overhang can induce the distressed ?rm to increase current cash ?ow at the cost of losing reputation and long-term profitability. Several accounting studies ?nd some evidence that avoidance of penalties associated with the violations of debt covenants is a motivation to manage earnings (Sweeney (1994), DeFond and Jiambalvo (1994), and Dechow et al. (1996)). These studies imply that ?nancial distress can increase ?rms’ incentives to misreport. I use the ratios of long-term debt and short-term debt to total assets to proxy for the degree of ?nancial distress.
Insider Equity Incentives
The relation between insiders’ equity incentives and the incidence of corporate fraud has been at the center of the current debate and reform on corporate governance. There are two forces associated with insiders’ equity stake. On one hand, the classic agency theory implies that
6 See
Demirg¨ uc ¸-Kunt and Maksimovic (1998) for assumptions and justi?cations for this measure. According to
the discussion in that paper, ROA here is the ratio of income before extraordinary items over assets.
7 See
Richardson and Sloan (2003) for a discussion of some possible limitations of this measure.
56
higher percentage insider ownership can better align insiders’ incentives to that of the shareholders. Since fraud is outright contravention of shareholders’ interest, high insider ownership should be associated with low fraud propensity. The agency view is supported by the work of Alexander and Cohen (1999). They examine public ?rms convicted of federal crimes in 1984-1990, and ?nd that crime occurs less frequently among ?rms in which management has a larger ownership stake. On the other hand, large equity incentives can be a double-edged sword, because the positive relation between ?rm performance and insiders’ compensation (or wealth) can induce distorted managerial reporting incentives (see, e.g., Goldman and Slezak (2003)). The second force seems to be supported by the ?ndings in some recent empirical work such as Johnson, Ryan and Tian (2003), Peng and R¨ oell (2004), and Burns and Kedia (2004). These papers ?nd that high pay-forperformance ratio (as a result of large equity-based compensation) is related to high probability of fraud or earnings manipulation, indicating over-incentivization of the management. In this study, I examine the role of insider percentage stock ownership and executive equity compensation in the context of accounting fraud. Executive equity compensation is measured as the value of restricted stock and stock options (using the Black-Scholes model) over an executive’s total compensation. I then compute the sum and average of the ratios across the ?ve top executive o?cers in the ?rm.
5.3.3 Control Variables
Some previous studies on ?nancial statement fraud ?nd that ?rms tend to commit fraud at a very early stage of their business cycle. Beasley, Carcello and Hermanson (1999) document that ?rms that have engaged in ?nancial statement fraud are generally small. The National Commission on Fraudulent Financial Reporting (AICPA 1987, 29) states that young public ?rms may face greater pressure to dress up ?rm appearance and thus have higher likelihood of engaging in fraud. There are also clear industry patterns in securities litigation (see Table 2). Technology ?rms (software & programming, computer and electronic parts, biotech), service ?rms (?nancial services, business services, utility, and telecommunication services) and the trade industries (whole sales
57
and retails) appear to have disproportionately high fraud concentration. This implies that these industries tend to have either large bene?t from fraud, or high detection risk, or both. Furthermore, ?rm size, age and industry segments are likely to be correlated with ?rms’ pro?tability, growth potential, external ?nancing need and ownership structure. Therefore, I control for ?rm size (log of total assets), age (as a public company), and ?rms’ membership in the technology, service and trade sectors.
5.3.4 Summary of Model Speci?cation
Factors Growth Potential External Financing Need Financial Distress Pro?tability Pro?tability (ex post) Insider Equity Incentive Investment Shareholder Monitoring Board Monitoring Independent Auditor Control Variables Variables Asset Growth, Book-to-Market Ext. Fin. Growth, Ext. Fin. C.F. Leverage, ST Debt ROA Residual ROA Insider Own, Equity Compensation R&D, Capital Exp., Acquisition Block Own, Institution Own Board Size, Outside Director Big Five, Auditor Opinion Firm Size, Age, Industry +/+ + + + ?F + + + ?D
5.4 Descriptive Information and Univariate Analysis
This section presents univariate comparisons between the fraud sample and the comparison sample. The explanatory variables are grouped into ?ve categories: (1) ?rm size and age; (2) profitability and growth; (3) external ?nancing needs; (4) investment; and (5) corporate monitoring. Table 4 Panel A reports the median and mean of each variable for both samples and the nonparametric Wilcoxon z-statistics for testing di?erences between the two samples. All the ?nancial 58
information is retrieved from COMPUSTAT database. Information on stock-based acquisition and acquisition volume is from SDC Platinum database. Ownership information is from CDA Spectrum. Information on executive equity compensation is from ExecuComp database. Information on board of directors is from EdgarPro database. To facilitate my analysis, I use the following ?scal year counting. For the fraud sample, I label the ?scal year in which fraud begins as year 0. The determination of year 0 is discussed in Section 5.1.1. Then the ?scal year prior to year 0 is year -1, and the one after is year 1. Since the comparison sample consists of all the non-litigated ?rms, all the comparison ?rms enter each relevant ?scal year. For example, ?scal year -1 spans from 1991 to 2002 for the fraud sample. Then all the observations of the comparison ?rms in year 1991 to year 2002 are labelled as information from year -1 and are used in the analysis. In this study, all the variables on pro?tability, growth, and external ?nancing needs are measured at the average level from year -2 to year -1. Using pre-fraud information helps to mitigate the e?ect of fraud on those measures. Information on investment is from year 0. The reason is that those investments were made around the time when fraud was committed, and therefore could have been used strategically by the management to disguise fraud. Corporate monitoring variables are measured at the average level from year -1 to year 0. Year 0 information is incorporated to strengthen the deterrence e?ect of monitoring on ?rms’ decision to commit fraud. The fraud sample on average appears to be larger but younger than the comparison sample. Studies on SEC’s AAERs generally ?nd that alleged ?rms are small (see, e.g., Beasley, Carcello and Hermanson (1999)). Firms that are subject to private class action litigation can be larger because class action lawsuits tend to target ?rms with “deeper pockets” (Cox and Thomas (2003)). The fraud sample seems to have outperformed the comparison sample in the two years before the commencement of fraud, and underperformed the comparison sample in year 1. This is consistent with the argument in Section 5.3.2. Fraudulent ?rms experienced some negative performance shock in year 0 but chose to cover up the problems by false ?nancial disclosure. Then fraud got uncovered in year 1, and the concealed bad performance was revealed.
59
Table 4 Panel A also shows that fraudulent ?rms tend to have signi?cantly higher growth rate and lower book-to-market ratio than the comparison ?rms. The median asset growth rate is 46% for the fraud sample, and only 9% for the comparison sample. High growth and low internal pro?tability naturally leads to large need for outside capital. According to the argument in Demirg¨ uc ¸-Kunt and Maksimovic (1998), on average only 13% of the growth in the fraudulent ?rms could be supported by internal funds, resulting in a high projected need for outside capital. The fraudulent ?rms also raised more external capital even before the commencement of fraud. The median ratio of net external ?nancing cash ?ow to total assets is 19% for the fraud sample, and only 4% for the comparison sample. The fraud sample, however, does not seem to be more burdened by debt than the comparison sample. The di?erence in growth opportunities across the two samples is further re?ected in investment expenditures. The fraud sample on average invested more than the comparison sample did in the year when fraud occurred. For instance, the median ratio of net investing cash out?ow to total assets is 11% for the fraud sample, and 6% for the comparison sample. However, the univariate comparisons do not control for factors that may in?uence the size of investment. We know that di?erent industries have di?erent investment patterns, and young ?rms tend to invest more than mature ?rms do. Firm size is also a potential determinant of investment size. Since all the investment variables have been normalized by the book value of assets, the size e?ect is already taken into account. Therefore, in order to have a more direct test of the overinvestment prediction in Wang (2004), I construct a control sample that is matched with the fraud sample in terms of industry distribution (two-digit SIC codes) and ?rm age at the end of year -1. Table 4 Panel B shows that the fraud sample on average had a much higher investment intensity than the control sample did both before and after the commencement of fraud. Except for capital expenditures, the di?erences across the two samples are statistically signi?cant, and particularly so for merger/acquisition-related expenditures. Finally, Table 4 Panel A shows that the fraud sample on average has more concentrated ownership, more institutional holdings, more large insider equity incentives. The fraud sample also
60
tends to have a smaller board and lower percentage of independent directors.
5.5 Multivariate Analysis
This section presents evidence from multivariate tests to simultaneously assess the e?ects of ?rm characteristics, investment, and corporate monitoring on a ?rm’s propensity to commit fraud and the probability of fraud detection.
5.5.1 Firm Characteristics and Fraud
Table 5 reports the e?ects of pro?tability, growth and external ?nancing need on a ?rm’s fraud incentives. We can see that ROA is positively associated with the likelihood of fraud. This result may seem counterintuitive at ?rst glance. However, it is actually intuitive because it is di?cult for a (known) troubled ?rm to sell a good earnings report. A ?rm will have incentive to fool the market and may easily succeed when the market believes that the ?rm is pro?table based on previous years’ performance, while deterioration in pro?tability has already started. The concealed performance deterioration, if it continues, will eventually lead to the revelation of fraud. The average marginal e?ect of residual ROA on P (D|F ) across all the models is -0.26, which means that a 10% unexpected decrease in ROA in year 1 is associated with an average 2.6% increase in the probability of detection. This result supports the argument in Wang (2004) that fraud is, to some extent, self-revealing. The more the manager is able to raise the market’s expectation by fraudulent reporting, the more likely the market will later see inconsistency between ?rm performance and what it has been guided to expect. The inconsistency leads to the discovery of fraud. Table 5 also shows that a ?rm’s growth potential and external ?nancing need are important motivational factors for fraud. Models 1 and 2 indicate that higher asset growth rate and lower book-to-market ratio are related to higher probability of fraud. Model 3 implies that fraudulent ?rms are likely to have a growth rate higher than what can be supported by their internal funds. The average marginal e?ect of externally ?nanced growth on P (F ) across models is 0.64, which means that increasing the externally ?nanced growth by 10% tends to increase a ?rm’s probability
61
of misreporting by 6.4%. Models 4-6 further show that fraudulent ?rms on average raise more external capital, but they do not appear to be more burdened by debt. This implies that fraudulent ?rms may pursue more equity ?nancing than debt ?nancing. Overall, results in Table 5 imply that rapidly growing ?rms with insu?cient internal capital are likely to misreport their ?nancial performance, because fraud enables them to exercise their growth options on favorable terms.
5.5.2 Investment and Fraud
Table 6 reports the relation between ?rms’ investment expenditures and their fraud incentives. Several interesting results emerge. First, I ?nd that di?erent types of investment have di?erential e?ects on the likelihood of fraud detection. Investment in R&D has the strongest negative e?ect on the probability of fraud detection. The relation is statistically and economically signi?cant. The average marginal e?ect of R&D expenditures on P (D|F ) across models is -0.17, which means that a 10% higher R&D expenditures is on average associated with a 1.7% lower probability of detection. Note that a ?rm’s total litigation cost is the probability of detection times the penalty upon detection. Suppose that the penalty can be completely measured in terms of money, then a 1.7% decrease in the detection likelihood can correspond to a substantial reduction in the dollar value of litigation cost. The e?ect of net investing cash ?ow on the probability of detection is also signi?cantly negative but much weaker than that of R&D expenditures. The average marginal e?ect of net investing cash ?ow on P (D|F ) is -0.05. Straightforward investment like capital expenditures does not seem to in?uence the likelihood of detection. Neither do acquisition expenditures. I further examine some di?erent measures of merger/acquisition intensity. Model 10 shows that larger the number of acquisitions in year 0, higher the probability of detection. A possible explanation for this result is that the regulators and the market may pay more attention to ?rms that are active in mergers/acquisitions. Furthermore, Model 11 shows that more focused acquisitions (acquisitions within the same two-digit SIC codes), higher the probability of detection. It is consistent with
62
the argument in Wang (2004) that investments that have high correlation with a ?rm’s current business can increase the ?rm’s litigation risk. Second, di?erent types of investment also have di?erential e?ects on ?rms’ propensity to commit fraud. The di?erences can stem from two sources: either di?erent investments a?ect ?rms’ bene?t from fraud di?erently, or they a?ect ?rms’ risk of being detected di?erently. Let us compare cash-based acquisitions and stock-based acquisitions. Table 6 shows that these two types of acquisitions have di?erent e?ects on P (F ), but not on P (D|F ). Stock-based acquisitions have a signi?cant positive relation with P (F ), while cash-based acquisitions do not. This implies that the ?nancing of the investment in?uences ?rms’ bene?t from committing fraud, but not the detection risk. Then let us compare R&D expenditures and capital expenditures. In all models, R&D expenditures are signi?cantly positively associated with P (F ), while capital expenditures do not in?uence P (F ). If these two types of investment are not generally ?nanced di?erently, then their di?erential e?ects on P (F ) should largely arise from their di?erential e?ects on P (D|F ). Holding other factors constant, ?rms that invest more in R&D tend to have lower litigation risk. Low litigation risk can encourage fraud.
5.5.3 Equity Ownership and Fraud
Table 7 presents the roles of insider equity incentives and shareholder monitoring in determining a ?rm’s fraud incentives. First, insider percentage stock ownership has a signi?cant concave relation with the probability of fraud. That is, when insider ownership is small, the probability of fraud increases as insider ownership increases. When insider ownership is large, however, the probability of fraud decreases as insider ownership increases. Given the dramatic increase in the use of stock options in managers’ compensation, the percentage stock ownership will not capture the full impact of managers’ equity incentives. Therefore, I construct an executive equity compensation variable, which is the total value of an executive’s restricted stock and stock options over her total compensation and sum over all the key executives in the company.8 Model 14 shows that
8 For
insider equity incentives, I have also examined various speci?cations of executive equity compensation other
than the one reported in Model 14. For example, I compute the average ratio rather than the sum across executives
63
executive equity compensation exhibits a similar but slightly weaker concave shape. The concavity implies that insider equity ownership (or equity compensation) can be a double-edged sword when it is used to align the interest of the managers to that of the outside shareholders. When insiders hold small stakes in the ?rm, the agency problem due to separation of ownership and control can be severe. However, steep equity incentive scheme may not solve the problem, because it can induce insiders to misreport rather than to work harder for the interest of outside shareholders. Interestingly, equity incentive seems to work well only when insiders already have substantial equity stakes in the ?rm.9 Second, I ?nd that the presence of large shareholders and institutional shareholders increases fraud detection and discourages fraud. The marginal e?ects of institutional ownership on P (D|F ) and P (F ) are 0.14 and -0.27, respectively. This means that a 10% increase in institutional share holdings is associated with an average 1.4% increase in the probability of fraud detection, and an average 2.7% decrease the probability of fraud.10 Block ownership has a similar e?ect. In general, however, block ownership has a slightly stronger e?ect on P (F ), while institutional ownership has a slightly stronger e?ect on P (D|F ). These results imply that the strength of shareholder monitoring in?uences ?rms’ propensity to commit fraud through their impact on the likelihood of fraud detection, and provide support for enhancing shareholder monitoring in the on-going corporate governance reform.
in a company. I use the value of exercised stock options rather than the Black-Scholes value of all stock option holdings. I also examine equity ownership and equity compensation of CEO. Overall, the results are qualitatively consistent across di?erent speci?cations.
9I
separately examine the subsample of ?rms that have 20% or higher insider ownership. I ?nd a signi?cant
negative relation between insider ownership and the probability of fraud. This result is not reported in the tables.
10 There
is a caveatee regarding the interpretation of the result. The Private Securities Litigation Reform Act
(PSLRA) that was passed in December 1995 requires that every class action lawsuit appoint a lead plainti?. PSLRA encourages large institutional investors to be lead plainti?s. Therefore, class action suits could be more likely to go through for ?rms that have large institutional investors. This may lead to the positive relation between institutional ownership and P (D|F ).
64
5.5.4 Auditor, Board and Fraud
Table 8 presents the e?ects of independent auditors and corporate boards on corporate fraud incentives. The auditor being one of the ?ve largest accounting ?rms appears to be related to higher likelihood of fraud detection.11 The deterrence e?ect, however, is not statistically signi?cant. Auditor opinions seem to have no in?uence on detection. The reason is that auditor opinions do not exhibit much variation at all. For the fraud sample, 79% of the auditor opinions in the year when fraud occurred were unquali?ed opinions, and the rest 21% were unquali?ed opinions with some explanations. The uniformly unquali?ed auditor opinions themselves show the problem: Why do independent auditors seldom disagree with their clients regarding the quality of disclosure? Are they truly independent? On monitoring by the board of directors, models 16-17 show that board size and the percentage of outside directors are positively associated with the probability of detection and negatively associated with the probability of fraud. However, the relations are not statistically signi?cantly. This could be due to the power issue. Unfortunately, I do not have board data for a large number of ?rms in my sample (see Table 4).
5.5.5 Summary of Results
In sum, Tables 5-8 present the multivariate analysis on the e?ects of ?rm characteristics, investment, and corporate monitoring on a ?rm’s probability of committing fraud and the probability of fraud detection. I ?nd that fraudulent ?rms are likely to be high-growth ?rms that have large needs for external capital but experience negative shocks in pro?tability. Performance deterioration, although temporarily concealed by fraud, tends to reveal itself and increase fraud detection. Second, I ?nd that investment can in?uence both ?rms’ ex-ante bene?t from committing fraud (e.g., through the ?nancing of the investment) and their ex-post detection risk. Therefore,
11 I
take out Arthur Andersen and ?nd similar result on the big four accounting ?rms. I also examine Arthur
Andersen separately, and ?nd no signi?cant result. Since these results are similar to what is reported in Table 8 Model 15, they are not reported.
65
it is an important determinant of ?rms’ incentives to defraud investors. Investments with high degree of uncertainty and/or low correlation with existing assets tend to negatively in?uence the likelihood of fraud detection. These results, together with the evidence of overinvestment in Panel B of Table 4, imply that fraud can be associated with investment distortions and thus real economic costs. Finally, di?erent types of corporate monitors also appear to have di?erent e?ects on fraud propensity and fraud detection. The presence of block equity holders and high institutional holdings is associated with high probability of fraud detection and low probability of fraud. There is weak evidence that reputable independent auditors and large corporate boards increase the likelihood of fraud detection.
5.5.6 Comparison with Simple Probit Models
Existing studies on fraud have used straight probit models to assess the e?ect of a factor on a ?rm’s probability of committing fraud. As discussed in Section 5.2.3, the straight probit model equates the probability of detected fraud to the probability of fraud. Therefore, it can not only underestimate the probability of fraud, but also lead to incorrect inferences. Table 9 compares the results from the straight probit model and the bivariate probit model, and demonstrates the problems associated with the straight probit model. Using the full sample of comparison ?rms as in the previous models (i.e., using multiple years’ data for every comparison ?rm) leads to very low marginal e?ects of all the variables in the straight probit model. Therefore, in order to better illustrate the di?erences between the straight probit and the bivarate probit models, I randomly choose one year for every comparison ?rm. That is, each comparison ?rm only enters the regression once in Table 9. First, let us look at the results on R&D expenditures. The straight probit model shows no signi?cant e?ect of R&D expenditures on P (F ), while the bivariate probit model shows a strong positive e?ect. The reason is that investment in R&D has opposing e?ects on P (F ) and P (D|F ). The two forces roughly o?set each other, resulting in no e?ect on the probability of detected fraud
66
P (Z ). Second, the marginal e?ects of net investing cash ?ows on P (F ) have consistent signs across the two models, but substantially di?er in magnitude (0.10 in probit and 0.43 in bivariate probit). The straight probit model underestimates the marginal e?ect of this variable. Third, the two models show opposing e?ects of institutional ownership on P (F ). The straight probit model reports a positive e?ect, while the bivariate probit model reports a negative one. Again, the reason is that institutional ownership has opposite e?ects on P (F ) and P (D|F ), and for this variable the positive e?ect on detection dominates. The comparisons in Table 9 clearly show that disentangling the e?ect of a factor on the probability of detecting fraud and its e?ect on the probability of committing fraud is important for us to draw sensible conclusions.
5.6 Robustness Checks 5.6.1 Frivolous Lawsuits
In this study, I use the ?ling of securities class action lawsuits to proxy for detected fraud. However, the ?ling of a lawsuit does not necessarily indicate that the alleged ?rm is fraudulent, because allegations could be frivolous or mistaken. Therefore, the fraud sample could be subject to biases due to possible false detections. Many studies in the legal literature have argued that The Private Securities Litigation Reform Act (PSLRA), which was passed in December 1995, makes it more di?cult for shareholders to sue a public company (see., e.g., Choi (2004)). My sample consists of litigation suits since 1996 (post PSLRA). Therefore, the probability of frivolous lawsuits in my sample should be lower than it was before PSLRA. In order to further mitigate the bias of false detection in the estimation, I separately examine the following three subsamples. The ?rst subsample has 334 ?rms that announced accounting restatements surrounding the securities lawsuits. The accounting restatement information is from General Accounting O?ce (GAO)’s October 2003 report. Since I study accounting-related fraud, the fact that the alleged ?rms restated their ?nancial reports provides support to the allegations. 67
The second subsample contains 207 ?rms that were subject to parallel SEC’s AAERs. Information on AAERs is retrieved from SEC’s web site. If frivolous lawsuits could result from the pro?t-orientation of private securities lawyers, then having parallel SEC’s litigation increases the credibility of the lawsuits, because SEC is not pro?t-oriented. Several papers in the legal literature (see, e.g., Johson, Nelson and Pritchard (2002), Choi (2004)) have viewed suits that result in dismissal or a low value settlement ($2 million or less) as “nuisance”. Therefore, in the last subsample, I exclude 27 cases that were either later dismissed by the court or had a settlement less than $2 million. The dismissal and settlement information is retrieved from the Securities Class Action Clearinghouse (SCAC). Table 10 shows that the main model results hold qualitatively across all three subsamples. This implies that the possible existence of false detection does not drive the results.
5.6.2 Timing of Fraud
The beginning of a ?rm’s fraudulent scheme is generally a little fuzzy due to the di?culty of tracing evidence far back in time. For accounting-related frauds, identifying the timing of fraud can be even more di?cult because the border line between aggressive accounting and securities fraud is not always a clear cut. In this study, I determine the beginning ?scal year of fraud (year 0) based on the speci?cation of class periods and ?rms’ ?scal year ending months. For ?rms that are subject to both private class action litigation and SEC’s AAERS, I also cross check the timing of fraud using the information in SEC’s litigation ?lings. To further examine the validity of the year 0 speci?cations, I compare ?rms’ ROA based on the originally reported accounting data with ROA based on the restated data from COMPUSTAT. Figure 5 plots the median historic and restated ROA for both the fraud and comparison samples from year -2 to year 2. We can see that for the fraud sample, the historic ROA and the restated ROA are consistent with each other in years -2 and -1, start to diverge in year 0, and then re-converge in year 2. This implies that the determination of year 0 is on average valid.
68
5.6.3 Industry and Business Cycle e?ects
So far, this paper has focused on ?rm-level economic determinants of fraud. Several papers have argued that industry and market environment also in?uences ?rms’ fraud incentives. Wang (2004) predicts that fraudulent events tend to cluster in certain industries during certain time period, because both the bene?t from fraud and the litigation risk are correlated among ?rms in the same industry. Gande and Lewis (2005) empirically document the industry spillover e?ect in securities litigation. That is, the ?ling of lawsuits on one ?rm signi?cantly negatively a?ects the stock performances of other ?rms in the same industry. I have controlled for industry distribution in the analysis. Here I further control for the industry securities litigation environment. I use the logarithm of the total market value of fraudulent ?rms in an industry in year -1 to proxy for industry litigation intensity. A high total market value can result from either a large number of frauds or the existence of some mega cases. Poval, Singh and Winton (2004) argue that ?rms’ fraud incentives are in?uenced by businesscycle factors. Their model shows that corporate fraud incentives are low when the economic condition is very good (investors are highly optimistic) or when it is very bad (investors are highly skeptical). The fraud incentives are high when the economic condition is switching from good to bad. In order to understand the e?ect of market-wide determinants on the probability of fraud, I construct a business cycle variable that equals -1 if the year in which fraud begins is between 1992 and 1994 or between 2001 and 2002 (bust), equals 0 if year 0 is between 1995 and 1997 or in 2003, and equals 1 if year 0 is between 1998 and 2000 (boom). Table 11 shows that the main model results remain unchanged after incorporating the industry and business cycle e?ects. In addition, both industry litigation intensity and business cycle variables tend to be positively related to P (D|F ) and negatively related to P (F ). The likelihood of fraud detection is high in industries with high litigation intensity. Good economic conditions are also related to higher ex post detection risk. This is actually intuitive. First, if a fraud begins in a very good year, this implies that the fraudulent ?rm has some negative idiosyncratic shocks. Second, very good economic conditions may not continue. The problems
69
concealed by fraud are likely to be revealed as the overall condition weakens, which leads to the discovery of fraud.12
5.6.4 Di?erent Model Speci?cations
The model speci?cation described in Section 5.3.4 is mainly from the ?rm’s viewpoint. Companies rationally compare the expected bene?t and litigation risk of engaging in fraud. The explanatory variables in the P (F ) equation consist of variables that either in?uence ?rms’ bene?t from committing fraud or in?uence their litigation risk. We can extend the model into a strategic two-party game: The ?rm calculates its risk of being detected when it makes the fraud decision. The detection forces also anticipate the ?rm’s likelihood of committing fraud when allocating their resources. For example, the market may be more vigilant with ?rms with high externally ?nanced growth if those ?rms tend to have high propensity to commit fraud. This implies that the externally ?nanced growth can be positively associated with the probability of fraud detection. Therefore, in Table 11 Speci?cation 3, if a factor a?ects a ?rm’s bene?t from committing fraud and its e?ect can be anticipated ex ante by the detection forces, then this factor is in both the fraud commitment and fraud detection equations. The results show that although high externally ?nanced growth is an important motivational factor for fraud, it does not appear to signi?cantly in?uence ?rms’ probability of being detected. A possible explanation for this is that growth itself is not necessarily a bad thing, and therefore does not necessarily trigger investor vigilance. As discussed in Section 5.3.1, the residual ROA variable is not completely exogenous. This variable, however, appears in all the models. In Table 11 Speci?cation 4, I take out this variable and examine whether the results on other variables still hold. The main results are qualitatively unchanged. The statistical signi?cance of variables is consistent with previous models. However, the marginal e?ects of variables in the P (F ) equation are lower.
12 I
also use the return to a market portfolio to proxy for overall business conditions. The results are consistent with
those reported in Table 11. High market return in the beginning year of fraud is associated with high probability of fraud detection.
70
Table 5.1: Allegations of Accounting Fraud : 1996 – 2003 Panel A: Litigation Filings by Calender Year The fraud sample consists of 684 class action lawsuits against 660 US public companies. The total number of lawsuits each year does not include cases ?led against private companies or cases against investment companies for pure fraudulent investment banking activities (such as unfair allocation of IPO shares and misleading analyst reports).
Year Accounting fraud Total # of lawsuits % of total
1996 45 100 45.00
1997 70 163 42.94
1998 103 232 44.40
1999 80 195 36.92
2000 107 206 51.94
2001 88 168 52.38
2002 119 212 56.13
2003 71 177 40.11
1996-2003 684 1454 47.04
Panel B: Class Periods, Age, and Stock Exchange The information on class periods is retrieved from the class action lawsuits. Age is de?ned as the number of years between a ?rm’s IPO date and the beginning of its class period. A ?rm’s stock exchange is identi?ed as of the beginning of the class period. Class Period # of obs. mean median maximum minimum (days) 660 471 354 2040 13 Age (years) # of obs. 652 mean 8.16 median 3.45 age10 years 22.66% Stock Exchange # of obs. 660 NYSE 30.3% AMEX 3.7% NASDAQ 64.0% Other 2.0%
Panel C: Accounting Fraud by the Beginning Fiscal Year The beginning ?scal year of a fraud is identi?ed based on the speci?cation of the class period and the ?rm’s ?scal year ending month. Fiscal year # of cases Fiscal year # of cases 1992 1 1998 97 1993 8 1999 117 1994 15 2000 108 1995 47 2001 57 1996 86 2002 16 1997 107 2003 1
71
Table 5.2: Industry Distribution of Accounting Fraud This table reports the distribution of accounting-fraud ?rms across industry segments. I classify ?rms into 24 industry segments based on 2-digit or 3-digit SIC codes, as detailed in the table. Percentage of total is computed based on the total number of public ?rms in each industry in the COMPUSTAT database.
Industry Agriculture (100-900) Mining (1000-1400) Construction (1520-1731) Food & Tobacco (2000-2111) Fabrics & Textile Products (2200-2390) Wood & Furniture (2400-2590) Paper & Printing (2600-2790) Chemicals (2800-2821, 2840-2990) Pharmaceutical (2833-2836) Materials & Related Products (3011-3490) Industry Manuf. (3510-3569, 3578-3590, 3711-3873) Computer-related Hardware (3570-3577) Electronics (3600-3695) Miscellaneous Manuf. (3910-3990) Transportation (4011-4731) Telecommunications (4812-4899) Utilities (4900-4991) Wholesales (5000-5190) Retails (5200-5990) Financial Services (6021-6799) Services (7000-7361, 7380-7997, 8111-8744) Software & Programming (7370-7377) Healthcare Services (8000-8093) Others (8880-9995) Total
Fraud Events 1 10 1 11 12 2 3 4 22 18 49 33 64 2 11 31 29 31 36 73 66 115 36 0 660
% of Sample 0.15 1.52 0.15 1.67 1.82 0.30 0.45 0.61 3.33 2.73 7.42 5.00 9.70 0.30 1.67 4.70 4.39 4.70 5.45 11.06 10.00 17.42 5.45 0.00 100
% of Total 1.18 0.74 0.78 2.59 3.79 1.06 0.72 0.82 2.81 1.89 2.44 6.82 5.13 0.91 2.24 3.65 4.37 3.64 2.70 1.51 3.72 6.13 9.57 0.00 2.96
72
Table 5.3: Table 3: Nature of Accounting Fraud This table presents the nature of the alleged ?nancial misrepresentations in 563 securities lawsuits studied in this paper. I am only able to identify the exact nature of the misrepresentation in 563 cases based on the information in relevant case documents (e.g., case complaints, press releases and court decisions). I categorize these 563 cases into 11 groups based on the accounting items that have been manipulated. I report the number of ?lings and the frequency of each category. Allegations # of identi?ed cases Improper revenue recognition Understatement of expenses Non-recurring items Overstatement of account receivables Overstatement of inventory Overstatement of intangibles Overstatement of investment Overstatement of other assets Understatement of reserves/allowances Understatement of liability Other # of Filings 563 380 97 4 53 38 13 9 72 49 20 24 % of Sample 67.50 17.26 0.71 9.43 6.76 2.31 1.60 12.81 8.72 3.56 4.27
73
Table 5.4: Univariate Comparisons of Firm Characteristics For each variable, the median, the mean (in the brackets) and the z -statistics for Wilcoxon tests are reported. ** and * indicate signi?cance at 1 and 5% levels, respectively. “ROA”=(operating income after depreciation)/assets. “Res. ROA” is the residual from regression: ROA1 =?0 + ?1 ROA0 + ?2 ROA?1 + . “B-M”=(assets)/( assets-equity+market value). “EF. Growth”=“Asset Growth” ROA2 - 1? ROA2 , where ROA2=(income before extraordinary items)/assets. “EF. C.F.”=(?common stock+?preferred stock+?debt)/assets. “Leverage”=(LT debt)/assets. “ST Debt”= (debt in current liabilities)/debt. “Bank/Debt”=(bank loan)/debt. “Invest. C.F.”= -(net investing cash ?ow)/assets. “Focused Acquis” is the percentage of acquisitions in which the target ?rm is within the same two-digit SIC codes as the acquirer. “Insider” is the percentage ownership of o?cers and directors. “Block” is the total percentage ownership of the shareholders who own at least 5% of the ?rm’s equity. “Institution” is the percentage ownership of ?nancial institutions. “Big Five”=1 if the independent auditor is one of the biggest ?ve accounting ?rms, and 0 if otherwise. “Opinion” goes from 1 (best) to 5 (worst). “B-Independ.” is the fraction of independent directors. “Equity Comp.” is executives’ value of stock and stock option over total compensation. Fraud Sample 192 (5515) 496 (5102) 157 (2057) 2.89 (7.64) 0.08 (-0.00) -0.01 (-0.07) 0.46 (1.09) 0.50 (0.53) 0.41 (1.08) 0.19 (0.23) 0.10 (0.17) 0.27 (0.36) 0.00 (0.05) 0.11 (0.14) 0.04 (0.06) 0.00 (0.04) 0.01 (0.11) 0.00 (0.79) 0.00 (0.22) 0.15 (0.21) 0.35 (0.38) 0.36 (0.39) 1.00 (0.81) 1.00 (1.21) 2.67 (3.59) 0.25 (0.29) 2.31 (2.31) # of obs. 631 535 627 630 616 545 563 521 562 551 626 561 632 611 614 579 587 631 631 599 602 572 631 533 273 273 223 Nonfraud Sample 136 (3538) 93 (1764) 90 (1520) 6.25 (8.73) 0.05 (-0.06) 0.02 (-0.00) 0.09 (0.36) 0.77 (0.74) 0.07 (0.39) 0.04 (0.03) 0.11 (0.18) 0.24 (0.35) 0.00 (0.05) 0.06 (0.08) 0.04 (0.06) 0.00 (0.02) 0.00 (0.03) 0.00 (0.14) 0.00 (0.05) 0.09 (0.18) 0.26 (0.32) 0.19 (0.26) 1.00 (0.71) 1.00 (1.29) 4.33 (4.75) 0.29 (0.34) 1.42 (1.52) # of obs. 68202 56493 65696 63338 66634 49764 61470 53270 59819 59893 66873 59676 64389 59165 56480 54566 55785 65047 65047 37526 37569 37506 68202 56210 2312 2617 12178 Wilcoxon z 5.44** 15.64** 6.44** -9.00** 8.03** -10.22** 20.03** -14.90** 17.94** 19.05** 0.45 1.12 5.04** 9.92** 2.06* 15.91** 17.37** 24.07** 21.72** 5.58** 6.88** 11.38** 5.06** -3.91** -10.05** -6.30** 9.78**
Assets($106 ) Market Value($106 ) Sales($106 ) Age ROA Res. ROA [1] Asset Growth B-M EF. Growth EF. CF. Leverage ST Debt R&D Invest. C.F. Capital Exp. Acquis.(cf.) Acquis.(cf.+stock) # of Acquis. Focused Acquis. Insider Block Institution Big Five Opinion B-Size B-Independ. Equity Comp.
74
Table 5.5: Investment: Fraud Sample vs. Industry-Age Matched Sample The control sample is matched with the fraud sample based on two-digit SIC codes and ?rm ages (the number of years since IPO date).
Age [-1] R&D [-2,-1] R&D [0] R&D [1] Capital Exp. [-2,-1] Capital Exp. [0] Capital Exp. [1] Acquis(cf.) [-2,-1] Acquis(cf.) [0] Acquis(cf.) [1] Acquis.(cf.+stock) [-2,-1] Acquis.(cf.+stock) [0] Acquis.(cf.+stock) [1] Invest. C.F. [-2,-1] Invest. C.F. [0] Invest. C.F. [1]
Fraud Sample 2.89 (7.64) 0.00 (0.07) 0.00 (0.05) 0.00 (0.06) 0.05 (0.06) 0.04 (0.06) 0.04 (0.06) 0.00 (0.03) 0.00 (0.04) 0.00 (0.03) 0.01 (0.09) 0.01 (0.11) 0.00 (0.07) 0.11 (0.14) 0.11 (0.14) 0.07 (0.09)
# of obs. 630 627 630 577 608 614 560 589 579 553 595 587 544 604 611 553
Control Sample 2.90 (7.00) 0.00 (0.06) 0.00 (0.05) 0.00 (0.06) 0.05 (0.06) 0.04 (0.06) 0.04 (0.06) 0.00 (0.02) 0.00 (0.02) 0.00 (0.02) 0.00 (0.04) 0.00 (0.05) 0.00 (0.04) 0.09 (0.10) 0.07 (0.10) 0.07 (0.07)
# of obs. 630 573 571 565 552 548 545 544 536 523 550 542 509 547 537 511
Wilcoxon z 0.13 2.14* 2.11* 2.77** 0.87 1.03 1.57 3.78** 6.47** 4.65** 6.53** 9.00** 5.10** 3.81** 4.84** 1.67
75
Table 5.6: Pro?tability, Growth, & Fraud This table reports the relation between ?rms’ pro?tability, growth potential, external ?nancing need and their propensity to commit accounting fraud. Probit coe?cient estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **,* indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA Asset Growth B-M EF. Growth Res. ROA Log(Assets) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 1 P (F ) P ( D |F ) 1.09/0.40 (3.61)** 1.81/0.67 (5.27)**
Model 2 P (F ) P (D|F ) 1.11/0.10 (3.85)**
Model 3 P (F ) P (D|F ) 2.16/0.71 (4.81)**
-2.03/-0.19 (-10.24)** 1.88/0.62 (4.87)** -0.03/-0.01 (-0.41) 0.01/0.00 (2.49)* 0.03/0.01 (0.08) -0.41/-0.14 (-1.12) -0.63/-0.23 (-1.59) 0.25 (0.32)
-0.03/-0.01 (-0.44) 0.01/0.01 (2.57)* 0.19/0.07 (0.56) -0.36/-0.13 (-0.97) -0.57/-0.22 (-1.40) 0.02 (0.02)
-1.16/-0.10 (-6.07)** 0.07/0.01 (1.80) -0.00/-0.00 (-1.12) 0.25/0.02 (1.76) 0.24/0.02 (1.46) 0.51/0.06 (2.48)* -2.26 (-18.97)** -0.49 (0.06) -2431.56 112.43 (13) 50137
0.27/0.02 (7.97)** -0.01/-0.00 (-3.35)** -0.28/-0.02 (-2.12)* -0.16/-0.01 (-1.02) 0.37/0.04 (2.02)* -1.43 (-6.63)**
-5.72/-0.72 (-8.04)** -0.14/-0.02 (-3.87)** 0.02/0.00 (3.38)** 0.58/0.09 (5.05)** 0.53/0.08 (4.34)** 0.21/0.03 (1.68) -1.25 (-4.15)** -0.33 (0.05) -2201.25 425.41 (13) 45354
-1.13/-0.09 (-5.16)** 0.06/0.01 (1.77) -0.00/-0.00 (-0.98) 0.30/0.03 (2.23)* 0.26/0.02 (1.62) 0.51/0.06 (2.63)** -2.27 (-19.48)** -0.56 (0.06) -2415.48 107.44 (13) 49019
76
Table 5.7: External Financing & Fraud
ROA Asset Growth EF. C.F. Leverage ST Debt Res. ROA Log(Assets) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 4 P (F ) P ( D |F ) 0.39/0.07 (2.14)* 0.06/0.01 (1.99)* 2.63/0.46 (6.83)**
Model 5 P (F ) P (D|F ) 1.08/0.39 (3.35)** 1.80/0.66 (5.20)**
Model 6 P (F ) P (D|F ) 1.18/0.45 (3.94)** 1.80/0.69 (5.45)**
0.08/0.02 (0.35) 0.07/0.03 (0.21) -0.03/-0.01 (-0.33) 0.01/0.01 (2.24)* 0.21/0.08 (0.58) -0.36/-0.14 (-0.87) -0.57/-0.22 (-1.28) -0.12 (-0.17)
0.30/0.05 (9.15)** -0.02/-0.00 (-4.03)** -0.00/-0.00 (-0.02) -0.56/-0.09 (-2.64)** -0.49/-0.07 (-1.79) -2.51 (-9.67)**
-3.60/-0.71 (-8.75)** -0.18/-0.04 (-5.58)** 0.02/0.00 (4.45)** 0.43/0.10 (3.12)** 0.60/0.13 (3.90)** 0.72/0.19 (3.12)** -0.78 (-2.39)* -0.19 (0.32) -2335.23 352.63 (14) 49146
-0.03/-0.01 (-0.47) 0.01/0.01 (2.59)** 0.18/0.07 (0.55) -0.34/-0.13 (-0.93) -0.56/-0.22 (-1.39) 0.03 (0.04)
-1.17/-0.10 (-5.70)** 0.07/0.01 (1.82) -0.00/-0.00 (-1.11) 0.24/0.02 (1.75) 0.24/0.02 (1.44) 0.50/0.06 (2.45)* -2.25 (-18.75)** -0.50 (0.08) -2424.07 105.96 (14) 49585
-1.29/-0.11 (-6.53)** 0.07/0.01 (1.56) -0.00/-0.00 (-0.95) 0.24/0.02 (1.62) 0.25/0.02 (1.34) 0.51/0.06 (2.33)* -2.26 (-17.81)** -0.45 (0.07) -2417.36 124.91 (14) 49677
77
Table 5.8: Investment, Fraud Propensity & Detection This table reports the regression results on the relation between investment and fraud. Probit coe?cient estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **,* indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA EF. Growth R&D Invest. CF. Capital Exp. Acquis.(cf.) Acquis.(cf.+stock) Res. ROA Log(Asset) Age Tech Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 7 P (F ) P (D|F ) 1.74/0.29 (4.86)** 1.61/0.27 (3.43)** 3.42/0.58 -1.50/-0.13 (4.70)** (-5.06)** 0.99/0.17 -0.32/-0.03 (2.27)* (-2.10)*
Model 8 P (F ) P (D|F ) 1.84/0.35 (4.97)** 1.64/0.31 (3.48)** 3.71/0.71 -1.56/-0.14 (4.69)** (-4.91)**
Model 9 P (F ) P ( D |F ) 12.71/0.80 (4.48)** 2.27/0.67 (4.38)** 2.61/0.63 -1.57/-0.16 (3.24)** (-3.48)**
-0.57/-0.11 (-0.48) 1.19/0.23 (0.49)
0.15/0.01 (0.26) 0.38/0.03 (0.52)
-0.97/-0.29 (-0.70)
0.26/0.02 (0.46)
0.00/0.00 (0.02) 0.01/0.00 (1.90) -0.32/-0.06 (-1.18) -0.25/-0.04 (-0.86) -0.56/-0.12 (-1.68) 0.65 (1.18)
-1.03/-0.09 (-7.71)** 0.04/0.00 (1.18) -0.00/-0.00 (-0.75) 0.40/0.04 (3.10)** 0.24/0.02 (1.67) 0.46/0.05 (2.58)** -2.06 (-15.07)** -0.80 (0.00) -2300.92 134.03 (17) 43920
-0.01/-0.00 (-0.20) 0.01/0.00 (2.10)* -0.24/-0.05 (-0.82) -0.23/-0.05 (-0.82) -0.49/-0.12 (-1.44) 0.69 (1.25)
-1.13/-0.10 (-7.42)** 0.06/0.00 (1.64) -0.00/-0.00 (-0.81) 0.37/0.04 (2.55)* 0.24/0.02 (1.44) 0.44/0.05 (2.31)* -2.14 (-14.94)** -0.79 (0.00) -2187.11 129.77 (19) 41482
2.43/1.06 (2.45)* 0.05/0.01 (0.58) 0.01/0.00 (1.23) -0.19/-0.06 (-0.53) -0.36/-0.11 (-1.08) -0.28/-0.09 (-0.71) -0.19 (-0.17)
0.43/0.03 (1.61) -1.44/-0.10 (-4.93)** 0.03/0.00 (1.02) 0.00/0.00 (0.24) 0.37/0.03 (3.12)** 0.30/0.02 (2.29)* 0.32/0.03 (1.89) -2.23 (-15.21)** -0.54 (0.19) -2099.20 114.44 (19) 41930
78
Table 5.9: Investment, Fraud Propensity & Detection (Continued) Model 10 P (F ) P (D|F ) 1.75/0.25 (3.45)** 1.80/0.25 (3.00)** 3.11/0.44 -1.37/-0.10 (2.69)** (-3.14)** 0.80/0.11 -0.31/-0.02 (1.66) (-1.96)* 0.36/0.05 0.09/0.01 (1.63) (4.60)** Model 11 P (F ) P ( D |F ) 1.77/0.22 (3.87)** 1.83/0.23 (2.90)** 3.34/0.42 -1.42/-0.09 (3.35)** (-3.88)** 1.09/0.14 -0.41/-0.03 (2.29)* (-2.46)* 0.34/0.04 0.07/0.01 (1.58) (3.47)** -0.73/-0.09 0.37/0.02 (-2.67)** (3.80)** -1.02/-0.07 (-6.22)** 0.17/0.02 -0.03/-0.00 (1.82) (-1.28) -0.00/-0.00 0.00/0.00 (-0.29) (0.70) -0.59/-0.09 0.49/0.04 (-2.20)* (4.83)** -0.61/-0.09 0.35/0.03 (-1.84) (3.16)** -0.98/-0.21 0.59/0.06 (-2.56)* (3.44)** 0.23 -1.96 (0.28) (-16.00)** -0.75 (0.02) -2252.92 209.68 (21) 43920
ROA EF. Growth R&D Invest. CF. # of Acquis Focused Acquis Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
0.15/0.02 (1.62) -0.00/-0.00 (-0.45) -0.64/-0.11 (-2.57)* -0.57/-0.09 (-1.79) -0.76/-0.16 (-2.11)* 0.21 (0.20)
-1.04/-0.07 (-4.87)** -0.03/-0.00 (-1.04) 0.00/0.00 (0.80) 0.51/0.05 (5.27)** 0.33/0.03 (2.99)** 0.50/0.05 (3.10)** -1.94 (-14.84)** -0.75 (0.07) -2262.09 164.86 (19) 43920
79
Table 5.10: Insider Equity Incentive, Corporate Monitoring & Fraud
ROA EF. Growth R&D Invest. CF. Insider (Insider)2 Equity Comp. (Equity Comp.)2 Block Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 12 P (F ) P ( D |F ) 2.06/0.82 (5.05)** 1.63/0.65 (4.97)** 3.32/1.32 -2.11/-0.28 (3.52)** (-4.53)** 1.70/0.67 -0.54/-0.07 (2.90)** (-2.26)* 1.72/0.68 (2.90)** -2.45/-0.97 (-3.00)**
Model 13 P (F ) P (D|F ) 2.11/0.81 (5.76)** 1.75/0.68 (6.90)** 2.83/1.09 -1.89/-0.25 (2.83)** (-3.96)** 1.50/0.58 -0.69/-0.09 (2.50)* (-2.82)** 1.81/0.70 (2.96)** -2.28/-0.88 (-2.68)**
Model 14 P (F ) P (D|F ) 2.87/1.14 (3.80)** 2.09/0.84 (2.53)* 1.77/0.71 -2.40/-0.20 (1.50) (-2.46)* 1.63/0.65 -1.08/-0.18 (1.98)* (-2.04)*
2.35/0.94 (2.45)* -2.17/-0.86 (-1.88) -1.02/-0.40 (-2.40)* 0.84/0.11 (3.71)** -0.69/-0.27 (-2.03)* 0.23/0.09 (4.07)** 0.00/0.00 (0.04) 0.03/0.01 (0.10) -0.32/-0.12 (-0.99) -0.76/-0.26 (-2.23)* -2.01 (-3.32)** 1.08/0.14 (6.39)** -2.06/-0.28 (-5.93)** -0.08/-0.01 (-2.86)** 0.00/0.00 (0.71) 0.40/0.06 (2.66)** 0.50/0.08 (2.98)** 0.73/0.14 (3.44)** -1.54 (-8.94)** -0.36 (0.11) -1923.87 236.42 (21) 29330 -0.68/-0.27 (-1.25) 0.11/0.05 (1.31) 0.01/0.00 (1.49) 0.21/0.08 (0.43) -0.21/-0.08 (-0.40) -1.05/-0.38 (-1.79) -1.66 (-1.80) 0.35/0.06 (1.08) -3.42/-0.60 (-5.83)** -0.05/-0.01 (-1.02) 0.01/0.00 (2.86)** 0.43/0.09 (1.57) 0.79/0.18 (2.75)** 1.16/0.31 (2.48)* -1.53 (-3.40)** -0.51 (0.02) -765.64 115.10 (21) 8747
0.26/0.10 (3.81)** -0.01/-0.00 (-1.17) -0.29/-0.11 (-1.07) -0.59/-0.23 (-1.98)* -0.74/-0.27 (-2.33)* -1.59 (-2.66)**
-1.98/-0.27 (-7.20)** -0.06/-0.01 (-1.71) 0.01/0.00 (1.66) 0.60/0.10 (4.66)** 0.59/0.10 (4.06)** 0.65/0.12 (3.59)** -1.75 (-8.73)** -0.39 (0.05) -1999.62 256.28 (21) 29439
80
Table 5.11: Independent Auditor, Corporate Board & Fraud
ROA EF. Growth R&D Invest. CF. Insider (Insider)2 Block Big Five Opinion B-Size B-Independ. Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 15 P (F ) P ( D |F ) 2.08/0.83 (5.00)** 1.63/0.65 (5.38)** 3.33/1.33 -2.04/-0.24 (3.27)** (-4.39)** 1.67/0.66 -0.53/-0.06 (2.84)** (-2.25)* 1.61/0.64 (2.53)* -2.38/-0.95 (-2.69)** -1.04/-0.41 0.79/0.09 (-2.26)* (3.44)** -0.04/-0.01 0.30/0.03 (-0.16) (1.94) -0.01/-0.00 (-0.12)
Model 16 P (F ) P (D|F ) 1.72/0.63 (2.46)* 3.64/1.35 (4.89)** 5.28/1.95 -3.14/-1.24 (3.05)** (-2.58)** 1.42/0.52 -0.81/-0.32 (1.71) (-1.46) 1.21/0.45 (1.03) 1.45/0.54 (0.76) -0.26/-0.09 0.26/0.10 (-0.61) (0.66)
Model 17 P (F ) P (D|F ) 1.79/0.69 (2.56)* 3.60/1.38 (4.56)** 5.21/2.00 -3.01/-1.18 (3.12)** (-2.58)** 1.45/0.56 -0.78/-0.30 (1.71) (-1.43) 1.58/0.61 (1.33) 0.96/0.37 (0.51) -0.24/-0.09 0.19/0.08 (-0.58) (0.50)
-0.02/-0.01 (-0.35)
0.10/0.04 (1.91) -0.78/-0.30 (-0.98) 0.08/0.03 (0.70) 0.01/0.00 (1.70) -0.02/-0.01 (-0.05) -0.50/-0.18 (-0.96) -1.06/-033 (-1.78) -2.08 (-3.55)** 0.36/0.14 (0.48) -2.16/-0.85 (-3.96)** -0.30/-0.12 (-5.82)** 0.00/0.00 (0.62) 0.51/0.20 (1.91) 1.52/0.54 (4.52)** 1.86/0.59 (3.06)** 1.21 (2.68)** -0.38 (0.35) -482.16 105.99 (23) 2186
0.26/0.10 (3.79)** -0.01/-0.00 (-1.66) -0.22/-0.09 (-0.76) -0.43/-0.17 (-1.33) -0.69/-0.25 (-1.99)* -1.52 (-2.32)*
-1.94/-0.23 (-7.01)** -0.07/-0.01 (-2.19)* 0.01/0.00 (1.78) 0.53/0.08 (3.82)** 0.47/0.07 (3.05)** 0.60/0.10 (3.07)** -1.97 (-8.18)** -0.38 (0.42) -1833.71 229.88 (24) 29225
0.09/0.03 (0.65) 0.01/0.00 (1.78) 0.09/0.03 (0.21) -0.32/-0.11 (-0.52) -0.88/-0.27 (–1.28) -2.01 (-3.52)**
-2.28/-0.90 (-3.90)** -0.36/-0.14 (-5.61)** 0.00/0.00 (0.69) 0.49/0.19 (1.92) 1.43/0.51 (4.51)** 1.83/0.58 (3.02)** 1.34 (2.70)** -0.30 (0.56) -481.81 109.18 (23) 2186
81
Table 5.12: Bivariate Probit Model vs. Straight Probit Model This table compares the results from the following two statistical models: Bivariate probit model: P (Zi = 1) = P (Fi = 1)P (Di = 1); Straight probit model: P (Zi = 1) = P (Fi = 1). Both models are estimated using a random comparison sample without repetition. That is, every comparison ?rm only enters the estimation once. The probit coe?cent estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **, * indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2). Probit P (F ) 1.04/0.18 (4.16)** 0.24/0.04 (6.97)** -0.49/-0.08 (-1.17) 0.59/0.10 (2.59)** 1.56/0.27 (3.74)** -1.74/-0.30 (-2.88)** 0.98/0.17 (7.58)** -2.12/-0.36 (-8.09)** 0.04/0.01 (2.18)* 0.00/0.00 (1.13) 0.53/0.11 (6.17)** 0.43/0.08 (5.34)** 0.27/0.05 (2.73)** -2.39 (-17.78)** -1100.36 275.66 (13) 3336 Bivariate Probit P (F ) P (D|F ) 2.34/0.65 (5.45)** 2.65/0.74 (5.41)** 3.51/1.29 -2.71/-0.97 (3.42)** (-4.72)** 1.46/0.43 -0.80/-0.25 (2.59)** (-2.54)** 2.33/0.71 (2.44)* -2.74/-0.84 (-2.70)** -0.61/-0.20 1.30/0.48 (-1.96)* (5.82)** -2.94/-0.85 (-6.31)** 0.28/0.07 -0.11/-0.03 (4.12)** (-2.86)** 0.00/0.00 0.01/0.00 (0.26) (2.03)* 0.13/0.04 0.41/0.17 (0.44) (2.13)* -0.17/-0.07 0.50/0.17 (-0.55) (2.48)** -0.53/-0.30 0.66/0.37 (-1.61) (2.58)** -2.26 -0.44 (-4.32)** (-1.88) -0.49 (0.03) -1007.66 219.26 (21) 3336
ROA EF. Growth R&D Invest. CF. Insider Own (Insider)2 Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
82
Table 5.13: Frivolous Lawsuits This table presents robustness checks of the main results over three subsamples: (1) 334 out of 660 ?rms that announced accounting restatements before or after the lawsuits; (2) 207 out of 660 ?rms that were subject to both private class action litigation and the SEC’s Accounting and Auditing Enforcement; (3)exclusion of 27 nuisance cases. A case is considered as a nuisance case if it is later dismiss by the court or if it leads to a less than two million dollar settlement. The probit coe?cent estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **, * indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA EF. Growth R&D Invest. C.F. Insider (Insider)2 Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Restatements P (F ) P (D|F ) 2.55/0.66 (4.39)** 1.89/0.49 (5.48)** 3.31/0.86 -2.01/-0.14 (3.33)** (-3.83)** 1.97/0.51 -0.53/-0.04 (2.78)** (-1.96)* 1.91/0.50 (2.09)* –2.27/-0.59 (-1.82) -0.63/-0.16 1.11/0.08 (-1.50) (5.49)** -2.47/-0.17 (-7.07)** 0.18/0.05 -0.05/-0.00 (1.83) (-0.99) 0.01/0.00 0.00/0.00 (0.97) (0.20) 0.14/0.04 0.35/0.03 (0.37) (1.75) -0.17/-0.4 0.26/0.02 (-0.38) (1.15) -0.74/-0.14 0.73/0.08 (-1.65) (2.82)** -2.62 -2.06 (-4.17)** (-9.04)** 0.02 (0.93) -1141.10 196.18 (21) 29117
SEC Enforcement P (F ) P ( D |F ) 2.40/0.66 (3.62)** 1.45/0.40 (3.55)** 3.18/0.87 -2.25/-0.14 (1.98)* (-2.11)* 2.00/0.55 -0.96/-0.06 (1.84) (-2.00)* 1.83/0.50 (1.60) -2.58/-0.71 (-1.40) -0.88/-0.24 1.21/0.08 (-0.99) (2.93)** -2.31/-0.15 (-4.49)** 0.09/0.03 -0.01/-0.00 (0.66) (-0.09) 0.01/0.00 -0.00/-0.00 (1.38) (-0.65) 0.45/0.13 -0.01/-0.00 (1.01) (-0.02) 0.17/0.05 -0.15/-0.01 (0.17) (-0.23) -0.21/-0.05 0.26/0.02 (-0.26) (0.48) -2.20 -1.98 (-1.83) (-6.82)** -0.15 (0.72) -740.56 113.22 (21) 29021 83
Non-Nuisance Suits P (F ) P (D|F ) 1.99/0.77 (5.60)** 1.67/0.64 (6.84)** 2.81/1.08 -1.90/-0.25 (2.49)** (-3.49)** 1.31/0.51 -0.68/-0.09 (2.02)* (-2.51)* 1.78/0.69 (2.96)** -2.22/-0.85 (-2.62)** -0.73/-0.28 1.12/0.15 (-2.02)* (6.38)** -2.05/-0.27 (-5.19)** 0.23/0.09 -0.08/-0.01 (4.06)** (-2.79)** 0.00/0.00 0.00/0.00 (0.14) (0.74) 0.05/0.02 0.39/0.06 (0.16) (2.46)* -0.31/-0.12 0.50/0.08 (-0.95) (2.72)** -0.76/-0.26 0.74/0.15 (-2.20)* (3.28)** -1.96 -1.55 (-3.15)** (-8.78)** -0.37 (0.11) -1859.72 239.00 (21) 29312
Table 5.14: Di?erent Model Speci?cations This table presents robustness checks of the main results across di?erent model speci?cations. “Ind. Lit” is the logarithm of the total market value of fraudulent ?rms in an industry in year -1. “Cycle”=-1 for years between 1992 and 1994 and between 2001 and 2002 (bust), =0 for years between 1995 and 1997, and =1 for years between 1998 and 2000 (boom).
ROA EF. Growth R&D Invest. C.F. Insider (Insider)2 Block Ind. Lit. (*103 ) Cycle Res. ROA Log(Asset) Age Tech Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Speci?cation 2 P (F ) P (D|F ) 1.90/0.75 (4.93)** 1.56/0.62 (5.42)** 3.07/1.22 -1.93/-0.23 (2.98)** (-3.79)** 1.68/0.67 -0.60/-0.07 (2.92)** (-2.49)* 1.71/0.68 (2.88)** -2.41/-0.96 (-2.95)** -1.00/-0.40 0.78/0.09 (-2.21)* (3.24)** -0.77/-0.31 0.70/0.08 (-0.99) (1.75) -0.21/-0.08 0.31/0.04 (-1.73) (5.70)** -1.89/-0.23 (-6.43)** 0.26/0.10 -0.05/-0.01 (3.46)** (-1.48) -0.01/-0.00 0.01/0.00 (-1.04) (1.50) -0.24/-0.10 0.52/0.08 (-0.93) (3.96)** -0.54/-0.21 0.52/0.08 (-1.81) (3.51)** -0.82/-0.31 0.67/0.12 (-2.51)* (3.66)** -1.21 -1.92 (-1.79) (-9.05)** -0.45 (0.04) -1959.12 291.98 (25) 29439
Speci?cation 3 P (F ) P ( D |F ) 1.61/0.64 (2.66)** 1.34/0.53 -0.03/-0.00 (2.52)* (-0.67) 2.78/1.10 -1.97/-0.29 (2.78)** (-3.10)** 1.58/0.63 -0.66/-0.10 (2.53)** (-1.80) 2.45/0.97 -0.59/-0.09 (3.26)** (-1.85) -2.31/-0.91 (-2.91)** -1.31/-0.52 1.04/0.15 (-2.65)** (2.76)** -1.00/-0.40 0.90/0.13 (-1.20) (1.64) -0.19/-0.07 0.30/0.04 (-1.19) (4.34)** -1.86/-0.28 (-5.88)** 0.28/0.11 -0.08/-0.01 (2.60)** (-1.00) -0.01/-0.00 0.01/0.00 (-0.70) (0.91) -0.24/-0.10 0.52/0.09 (-1.01) (3.71)** -0.54/-0.21 0.53/0.10 (-1.90) (2.91)** -0.84/-0.32 0.73/0.15 (-2.73)** (3.45)** -1.21 -1.63 (-1.85) (-3.26)** -0.55 (0.02) -1955.92 311.90 (27) 29439 84
Speci?cation 4 P (F ) P (D|F ) 0.39/0.04 (1.97)* 0.55/0.06 (2.27)* 2.34/0.26 -1.56/-0.20 (3.69)** (-4.20)** 1.02/0.11 -0.65/-0.08 (3.39)** (-3.37)** 0.44/0.05 (1.98)* -0.65/-0.07 (-1.93) -0.75/-0.08 0.71/0.09 (-1.98)* (2.26)* -0.97/-0.11 0.89/0.12 (-2.02)* (2.54)* -0.38/-0.04 0.38/0.05 (-6.05)** (7.00)** 0.04/0.00 (0.83) 0.00/0.00 (0.16) -0.43/-0.06 (-2.36)* -0.50/-0.07 (-2.61)** -0.81/-0.14 (-3.49)** 1.71 (5.18)** 0.00/0.00 (0.04) 0.00/0.00 (0.06) 0.46/0.07 (3.28)** 0.48/0.08 (3.24)** 0.72/0.14 (3.77)** -2.05 (-11.09)** -0.99 (0.00) -2183.13 336.68 (24) 31696
Figure 3: Determination of the Beginning Fiscal Year of Fraud
Fiscal year ending month Class Period
12
6 Year 0
12
Fiscal year ending month
Class Period
4 Year 0
4
6
4
Figure 4: Identification Problem
No fraud (F=0) Z=0 The Firm Not detected (D=0|F=1) Fraud (F=1) Detected (D=1|F=1) Z=1
85
Figure 5: Timing of Fraud
Historic vs. Restated ROA 4 3 Median ROA 2 1 0 -1 -2 -3 -4 Fiscal Year -2 -1 0 1 2
Fraud-Historic Fraud-Restated Control-Historic Control-Restated
Note: ROA is the ratio of net income over total assets. I use net income because the restated information on this variable is more complete than the one on other accounting measures such as income before extraordinary items and operating income. The purpose here is to compare the originally reported data with the restated data.
86
Chapter 6 Conclusion
This thesis analyzes corporate securities fraud and its consequences. The theory model shows that fraud can lead to investment distortions in both fraudulent ?rms and honest ?rms, which is the real economic cost of fraud. The investment distortion is twofold. On one hand, fraud can in?ate short-term ?rm value and allow the ?rm to invest using cheap outside capital. On the other hand, once committed fraud, the ?rm has incentive to strategically use investment to mask fraud. The incentive to disguise fraud can not only induce the ?rm to overinvest, but also gives the ?rm a preference for risk and suboptimal diversi?cation. The theory model also characterizes the endogenous cost-bene?t tradeo? of committing fraud and derives the ?rm’s equilibrium disclosure strategy. The model shows that the cost and bene?t of fraud are endogenously related, which results in the optimal size of fraud and the ?rm’s equilibrium fraud propensity. In particular, the theory demonstrates the important role of the endogenous detection risk in determining the cross-sectional variations in ?rms’ fraud incentives. The model generates testable implications about the economic determinants of cross-sectional di?erences in fraud propensities and the relationship between fraud and corporate investment incentives. The theory predicts that fraudulent ?rms tend to have good growth prospects, but experience negative pro?tability shocks. Litigation events tend to cluster in certain industries during some speci?c time period. The theory also predicts that fraudulent ?rms tend to overinvest. Investment can negatively in?uence the ?rm’s litigation risk. The type of investment that introduces the most valuation imprecision has the strongest e?ect on the likelihood of fraud detection. The investment, however, can be ine?cient and can result in long-term underperformance of fraudulent ?rms. I also empirically investigate the economic determinants of ?rms’ propensity to commit accounting fraud and the probability of fraud detection, using a sample of public companies that
87
were subject to federal private securities class action litigation between 1996 and 2003. I use econometric methods to control for the unobservability of undetected frauds, and disentangle the e?ect of a factor on a ?rm’s probability of committing fraud and its e?ect on the ?rm’s probability of being detected. The separation of fraud commitment and fraud detection allows me to examine the economics of each probability as well as their interactions. The results of this study show that investment, strength of corporate monitoring, insider equity incentives, and some ?rm characteristics signi?cantly in?uence a ?rm’s cost-bene?t tradeo? of engaging in fraud. First, the level, type, and ?nancing of investment types of investment all matter in dertermining a ?rm’s ex-post probability of fraud detection and ex-ante propensity to commit fraud. Second, di?erent types of corporate monitors have di?erent e?ects on ?rms’ fraud incentives. The presence of block equity holders and large institutional ownership tends to increase the likelihood of fraud detection and discourage fraud. The roles of independent auditors and board of directors appear to be weaker. Third, there is a concave relation between insider equity incentive and the probability of fraud. When insider equity incentive is small, increasing equity incentive can have the unintended e?ect of increasing the probability of fraud. When insider equity incentives is already large, such e?ect disappears. This implies that insider equity incentive can be a double-edged sword when it is used to align managerial and shareholder interests in dispersely-owned ?rms. Finally, high growth potential, large external ?nancing need, and (hidden) negative pro?tability shocks seem to be important motivational factors for fraud. This study also demonstrates the importance of disentangling the probability of committing fraud and the probability of detecting fraud, because cross-sectional variables can have opposing e?ects on the two latent probabilities, and therefore can be masked in their overall e?ect on the incidence of detected fraud. Ignoring this structure can lead us to draw incorrect inferences about the determinants of corporate fraud.
88
Chapter 7 Appendix: Proofs of Propositions
Proof of Proposition 1 The market value of the ?rm at time 2 after the investment announcement is V2 (I, y ) E (R|I ) zc = E (V3 |I, y ) = E (A|I, y ) + IE (R|I ), = E (R|R > rc ) = R + I?R m(zc ), = (rc ? R)/?R , ?(zc ) . 1 ? ?(zc )
m(zc ) = The investment condition is
(1 ? ? )[E (A|e) + Ir] ? PI f ? > E (A|e) ? PN f ?, where ? = I/V2 (I, y ). Solving for r, we get r> E (A|e) E (A|I, y ) + I?R m(zc ) ? (PN ? PI )f ? . (1 ? ? )I
(7.1)
(7.2)
This leads to equation (4.23). The left-hand side of equation (4.23) monotonically increases in rc , while the right-hand side monotonically decreases in rc . Therefore, there exists a unique solution
? to equation (4.23), rc .
Proof of Proposition 2 PI < PN if and only if ?I < ?N , or equivalently vc,I +KI < vc,N +KN . vc,I +KI is a function of I?R , while vc,N + KN is not. Let M = e + C/f ? y , and ?I = cov (e, V3 |I )/(?e Take the derivative of vc,I + KI with respect to I?R . ? (vc,I + KI ) ? (I?R ) = = ?vc,I ?KI + ? (I?R ) ? (I?R ) M ??I ??I + [E (V3 |y ) ? E (V3 |e)] . 2 ?e ?I ? (I?R ) ? (I?R ) 89 (7.3) (7.4) V ar(V3 |I )).
? 4 2 ?u +(2q?e I?R )2 ??u ??I ?e I If max(?1, ? qI? ) < ? < < 1, then ? (?? 2q?e I?R I?R ) < 0 and ? (I?R ) < 0, and therefore R ? 4 2 ?u +(2q?e I?R )2 ??u ? (vc,I +KI ) ??I I < 0. If ? ? ? 1, then ? (?? ? (I?R ) 2q?e I?R I?R ) > 0 but ? (I?R ) < 0. Therefore, there ? 4 2 ? +(2q?e I?R )2 ??u ? (v +KI ) ?e exists ? ? [ u 2q?e I?R , 1] such that when max(?1, ? qI? ) < ? < ?, ?c,I < 0. Since (I?R ) R vc,N + KN does not depend on I?R and vc,I + KI decreases with I?R , there exists a cuto? value I?R , such that when I?R ? I?R , vc,I + KI ? vc,N + KN . Proof of Proposition 3 Note that E (A|I, y ) + I?R m(zc ) = V2 (I, y ) ? I . Let us take derivative with respect to ? on both sides of equation (4.23). ?rc ?? = ? ? E (A|e) ?V2 (I, y ) (PN ? PI )f ? + (PN ? PI )f ? [V2 (I, y ) ? I ]2 ?? (1 ? ? )I (PN ? PI )f ? (???/?? ) . (1 ? ? )2 I ?PN ?PI ?V2 (I, y ) ? = (1 ? p)(?N ?N ? ?I ?I ) , ?? ?? ??
(7.5)
PN ? PI = where ? = ? ?(s)/?s. ? =
I V2 (I,y )
is the fractional ownership of the new shareholders. V2 (I, y ) does not directly
depend on ? , since the market does not observe ? . From the manager’s point of view, however, what is important is how much V2 (I, y ) will be di?erent from V2 (I, e) if the manager reports one more unit of earnings above the true realization e. So let us de?ne ?V2 (I, y ) V2 (I, y ) ? V2 (I, e) = lim . ?? y?e (y ?e)?0 Then ?? I ?V2 (I, y ) = (? ). 2 ?? V2 (I, y ) ?? (7.7) (7.6)
If y (e) = e does not generate any e?ect on the market valuation (i.e., V2 (I, y ) = V2 (I, e)), then ??/?? = 0. As long as misreporting can increase the market value of the ?rm’s assets, i.e., then ??/?? < 0. Substitute these relations into (7.5), and we have
?rc ?? ?V2 (I,y ) ??
? 0,
< 0.
Proof of Proposition 4
90
The ?rst-order condition for the maximization problem (4.20) is ?? +g ?? g? = = 0; 0, (7.8) (7.9)
where g is the Lagrange multiplier for the nonnegativity constraint on ? . ?? ?? ?z ?rc )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?rc + ?[1 ? ?(zc )][(? )E (V3 |I, e) + (1 ? ? )Im (zc ) ] ?? ?? ?z ?rc ? {? (PN ? PI ) + ?[1 ? ?(zc )]PI + (1 ? ?[1 ? ?(zc )])PN }f ? ?R ?? = ?( ? ? P f. (7.10)
The following steps present the derivations of the equilibrium strategy speci?ed in Proposition 4. Step 1: A Conjecture. Suppose that there exists a cuto? earnings realization ec such that
the manager will honestly reveal the earnings if the true earnings realization is above ec , and the manager would like to overreport earnings if the true realization is below ec . Mathematically, y (e) = e or ? (e) = 0, if e ? ec ; y (e) > e or ? (e) > 0, if e < ec . Given the above conjecture about the manager’s fraud incentives, the market’s reaction to an earnings announcement can be as follows. When investors observe the announced earnings y (e), they rationally infer e = y (e) ? ? , using their prior belief about the probability of misreporting ?0 . ? is the market’s expected amount of misreporting. The time 1 conditional probability of fraud is ?1 = P rob.(misreporting |y ? ec ). Therefore, whenever y ? ec , investors believe that e = y > ec , with probability (1 ? ?1 ); e = y ? ? 1 < ec , with probability ?1 . When investors observe y < ec , they rationally discount the earnings announcement, and e = y ? ? 2 . Then the market value of the ?rm’s assets in place after the earnings announcement is V1 (y ? ec ) V1 (y < ec ) = (1 ? ?1 )E (A|e = y ) + ?1 E (A|e = y ? ? 1 ); (7.11) (7.12)
= E (A|e = y ? ? 2 ). 91
? 1 and ? 2 are the market’s expected amount of misreporting given y ? ec and y < ec , respectively. In equilibrium, they should be equal to the manager’s optimal choice of misreporting in the two earnings announcement scenarios. ? ? 0 and the structure of litigation cost of fraud naturally leads to a conjecture that ? (e) is monotonic in e in each di?erent region speci?ed above. This does not imply, however, that y (e) is always monotonic in e (due to the pooling of the two types of ?rm). Then in each of the two scenarios (fraud or honest) there is a one-for-one mapping between e and y (e). This implies that under each scenario, e = y (e) ? ? is still normally distributed. Therefore, given the true realization of earnings e, when y ? ec , ?V2 (I, y ) = ? (1 ? ?1 ) > 0. ?? When y < ec , ?V2 (I, y ) = 0. ?? Step 2: Deriving ec . (7.14) (7.13)
Let us plug equation (7.13) and (7.14) into (7.10) and di?erentiate with
respect to ? on both sides. Then use the following relationships: ?rc ?? PN PN PN PI we can ?nd that ?2? < 0. ?? 2 This means that the objective function is globally concave. There exists a unique maximizer
? ? ? = ?1 . The concavity and the nonnegative ? constraint imply that
< = = =
0; (1 ? p)? (1 ? ?1 )?N ?N > 0; (1 ? p)? (1 ? ?1 )?N |vc,N + KN |?N > 0; (1 ? p)? (1 ? ?1 )?I ?I > 0;
= (1 ? p)? (1 ? ?1 )?I |vc,I + KI |?I > 0,
?? ? |?=0 > 0 ? ?1 > 0, ?? ?? ? |?=0 ? 0 ? ?1 = 0. ?? 92
I de?ne the following notations. ?0 = ? (y = e) = I/V2 (I, e), rc,0 = rc (? = 0), zc,0 = (rc,0 ? R)/?R , ?0 = ?(zc,0 ), ?0 = ?(zc,0 ), and m0 = m(zc,0 ). Then plug ? = 0 into equation (7.10), and we have ?? ?rc |?=0 = ?(1 ? ?0 ){Im0 (? |?=0 )[?0 m0 + (1 ? ?0 )zc,0 ] + ?0 ? (1 ? ?1 )} ? pf, ?? ?? where ?rc E (A|e)? (1 ? ?1 ) |?=0 = ? . ?? [V2 (I, e) ? I ]2 The ?rst term on the right-hand side of equation (7.15) decreases as e increases, while the second term does not depend on e. Therefore, we can ?nd a cuto? ec , such that
?? ?? |? =0 ?? ?? |? =0
(7.15)
> 0 if e < ec , and
? 0 if e ? ec . ec is the solution to ?? |?=0 = 0. ??
Step 3: Deriving eh .
To facilitate the analysis below, it is convenient to decompose
?? ??
into
a marginal bene?t of fraud term and a marginal cost of fraud term. Let MB ?z ?rc )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?m(zc ) ?rc + ?[1 ? ?(zc )][(? )E (V3 |I, e) + (1 ? ? )I ]; ?? ?zc ?? ?z ?rc ?PI ?PN = {? (PN ? PI ) + ?[1 ? ?(zc )] + (1 ? ?[1 ? ?(zc )]) }f ? ?R ?? ?? ?? = ?( ? + P f. (7.16)
MC
(7.17)
Then let us take the ?rst and the second derivatives of both MB and MC with respect to e. We can ?nd that
?M B ?e ?M C ?e
< 0, > 0,
?2M B ?e2 ?2M C ?e2
< 0; > 0.
The relations about the ?rst derivatives mean that when the true earnings realization is low, the marginal bene?t of fraud is relatively high, while the marginal cost of fraud is relatively low. This implies that
? ??1 < 0. ?e
The relations about the second derivatives imply that
? ? 2 ?1 > 0. 2 ?e
93
? Given that ?1 (e) is a decreasing and concave function of e, there exists
? (e)]. eh ? max [e + ?1 e eh , the market rationally believes that y = e. Step 4: Deriving el .
? Similarly, given that ?1 (e) is a decreasing and concave function of
? e, there also exists a lower bound el such that when e < el , y (e) = e + ?1 (e) < ec , and when ? el ? e < ec , y (e) = e + ?1 (e) ? ec . el is the solution to the following equation:
M B [?1 (el )] = M C [?1 (el )], where ?1 (el ) = ec ? el . When the ?rm announces y < ec , however, the market reaction changes, because now the low-earnings ?rm is not pooled with the high-earnings ?rm. In other words, the bene?t-cost tradeo? is di?erent, which implies that the optimal amount of misreporting in this region should
? . be di?erent from ?1
Let ?2 (e) be the manager’s misreporting strategy when e < el . If ?2 (e) is a monotonic function of e, then y (e) = e + ?2 (e) < ec is also a monotonic function of e. In other words, y (e) is a su?cient statistic of e, and thus E (A|y ) = E (A|e). Substitute E (A|y ) = E (A|e) into equations (4.17), (4.18), (4.23), (7.16) and (7.17), and we get
?rc ??
= 0, M B = 0, and M C = pf > 0. Since the
marginal bene?t is less than the marginal cost regardless of ? , the optimal amount of misreporting
? = 0. Put di?erently, if the manager chooses a monotonic disclosure strategy when e < el , is ?2
then the optimal monotonic strategy is y (e) = e. Step 5: Possibility of ? (e) as a nonmonotonic function of e. Let us also consider whether
there exists an equilibrium in which ?2 (e) is a nonmonotonic function of e. Since ? ? 0 (which means y (e) ? e), and the litigation cost is an increasing and monotonic function of ? , I can make the following conjecture about ?2 (e). I can partition the earnings space {e : e < el } into many intervals, [e1 , el ), [e2 , e1 ), [e3 , e2 ).... In each earnings interval, y (e) equals the upper bound of that interval. The lower bound of each interval is determined, such that the earnings realization at the 94
lower bound plus the optimal amount of misreporting equals the upper bound earnings value. Take the ?rst interval [e1 , el ) for an example. If the true earnings realization is in this interval, then the manager announces y (e) = el . The market rationally infers that e = E (e|e1 ? e < el ) and uses e to price the ?rm’s assets in place. It is easy to see that ?rms with e < e < el get worse o? by reporting y (e) = el than reporting y (e) = e, because the ?rm’s asset value is underpriced by the market, and the ?rm faces potential litigation cost. Then these ?rms would rather honestly reveal their earnings, and the conjectured equilibrium collapses. This happens to any nonmonotonic ?2 (e). Proof of Proposition 5 1. e 2. ? ?M B ?? = ?z ?rc (? )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?rc + (1 ? ?(zc ))[(? )E (V3 |I, e) + (1 ? ? )Im (zc ) ] ?? ?? (7.18) (7.19) Proof is shown in the proof of Proposition 4 (step 3).
> 0; ?M C ?? = ??[1 ? ?(zc )](PN ? PI )f ? < 0.
? (e) Since the marginal bene?t of fraud increases in ? and the marginal cost decreases in ?, ?1
increases in ? for any given e. This also implies that ?M B |?=0 ?? = (1 ? ?0 ){Im0 (? > 0; ?M C |?=0 ?? This implies that = 0.
?ec ??
?el ??
< 0.
?rc |?=0 )[?0 (m0 + (1 ? ?0 )zc,0 ] + ?0 ? (1 ? ?1 )} ?? (7.20) (7.21)
> 0.
95
BIBLIOGRAPHY [1] Agrawal, Anup, and Sahiba Chadha, 2004, “Corporate Governance and Accounting Scandals,” Journal of Law and Economics Forthcoming. [2] Agrawal, Anup, Je?rey F. Ja?e, and Jonathan M. Karpo?, 1999, “Management turnover and corporate governance changes following the revelation of fraud,” Journal of Law and Economics April, 309-342. [3] Alexander, Cindy R., and Mark A. Cohen, 1999, “Why do corporations become criminals? Ownership, hidden actions, crime as an agency cost,” Journal of Corporate Finance 5, 1-34. [4] American Institute of Certi?ed Public Accountants, National Commission on Fraudulent Financial Reporting (AICPA), 1987, Reporting of the National Commission on Fraudulent Financial Reporting, New York, NY. [5] Bajaj, Mukesh, Katherine Gunny, and Atulya Sarin, 2003, “Auditor compensation and auditor failure: An empirical analysis,” working paper, University of California-Berkeley. [6] Baucus, Melissa S., and David A. Baucus, 1997, “Paying the piper: An empirical examination of long-term ?nancial consequences of illegal corporate behavior,” Academy of Management Journal 40 (February), 129-151. [7] Beasley, Mark S., 1996, “An empirical analysis of the relation between the board of director composition and ?nancial statement fraud,” The Accounting Review 71, No. 4, 443-465. [8] Beasley, Mark S., Joseph V. Carcello, and Dana R. Hermanson, 1999, “Fraudulent ?nancial reporting: 1989-1997,” Research commissioned by the Treadway Commission. [9] Bebchuk, Lucian A., and Oren Bar-Gill, 2002, “Misreporting corporate performance,” working paper, Harvard University. [10] Becker, Gary S., 1968, “Crime and punishment: An economic approach,” Journal of Political Economy , 169-217.
96
[11] Bertrand, M., and S. Mullainathan, 2001, “Are CEOs rewarded for luck? The ones without principals are,” Quarterly Journal of Economics 116, 901-932. [12] Bethel, Jennifer, J. Liebeskind, and T. Opler, 1998, “Block share purchases and corporate performance,” Journal of Finance 53, 605-635. [13] Burns, Natasha, and Simi Kedia, 2004, “Do executive stock options generate incentives for earnings management? Evidence from accounting restatements,” working paper, Harvard University. [14] Chan, Konan, Louis K. C. Chan, Narasimhan Jegadeesh, and Josef Lakonishok, 2005, “Earnings quality and stock returns,” Journal of Business Forthcoming. [15] Choi, Stephen J., 2004, “Do the merits matter less after the Private Securities Litigation Reform Act?” working paper, University of California-Berkeley. [16] Co?ee, John C. Jr., 1991, “Liquidity versus control: The institutional investor as corporate monitor,” Columbia Law Review 91, No. 6, 1277-1367. [17] Cox, James D, and Randall S. Thomas, 2003, “SEC enforcement heuristics: An empirical inquiry,” Duke Law Journal 53, 737-780. [18] Dechow, Patricia M., Richard G. Sloan, and Awy P. Sweeney, 1996, “Causes and consequences of earnings manipulation An analysis of ?rms subject to enforcement actions by the SEC,” Contemporary Accounting Research 13, No. 1, 1-36. [19] DeFond, M. and J. Jiambalva, 1994,“Debt Covenant E?ects and the Manipulation of Accruals,” Journal of Accounting and Economics 17, 145-176. [20] Demirg¨ uc ¸-Kunt, Asli and Vojislav Maksimovic, 1998, “Law, ?nance, and ?rm growth,” Journal of Finance 53, 2107-2137. [21] Dye, Ronald A., 1988, “Earnings management in an overlapping generations model,” Journal of Accounting Research 26, No. 2, 195-235.
97
[22] Efendi, Jap, Anup Srivastava and Edward P. Swanson, 2004, “Why do corporate managers misstate ?nancial statements? The role of option compensation, corporate governance, and other factors,” working paper, Texas A&M University. [23] Erickson, Merle., and Shiing-wu Wang, 1999, “Earnings management by acquiring ?rms in stock for stock mergers,” Journal of Accounting and Economics 18, 3-28. [24] Feinstein, Jonathan S., 1990, “Detection Controlled Estimation,” Journal of Law and Economics 33, No. 1, 233-276. [25] Feroz, Ehsan, Kyungjoo Park, and Victor Pastena, 1991, “The ?nancial and market e?ects of the SEC’s accounting and auditing enforcement releases,” Journal of Accounting Research 29, 107-142. [26] Frankel, Richard M., Marilyn F. Johnson, and Karen K. Nelson, 2002, “The relation between auditors’ fee for non-audit services and earnings management,” Accounting Review 77 Supplement, 71-105. [27] Gande, Amar, and Craig M. Lewis, 2005, “Shareholder initiated class action lawsuits: Shareholder wealth e?ect and industry spillovers,” working paper, Vanderbilt University. [28] Gao, Pengjie, and Ronald E. Shrieves, 2003, “Earnings management and executive compensation: A case of overdose of option and underdose of salary?” working paper, University of Tennessee. [29] oldman, Eitan, and Steve L. Slezak, 2003, “The economics of fraudulent misreporting,” working paper, University of North Carolina at Chapel Hill. [30] Healy, Paul M., and James M. Wahlen, 1999, “A review of the earnings management literature and its implications for standard setting,” Accounting Horizons 13, No. 4, 365-383. [31] Hermalin, Benjamin E., and Michael Weisbach, 2003, “Boards of directors as an endogenously determined institution: A survey of the economic literature,” Economic Policy Review April, 7-26. 98
[32] Holderness, Cli?ord G., 2003, “A survey of blockholders and corporate control,” Economic Policy Review April, 51-64. [33] Jensen, Michael C., 2004,“Agency costs of overvalued equity,” working paper, Harvard University. [34] John, Kose, and Lemma W. Senbet, 1998, “Corporate governance and board e?ectiveness,” Journal of Banking & Finance 22, 371-403. [35] Johnson, Shane A., Harley E. Ryan, and Yisong S. Tian, 2003, “Executive compensation and corporate fraud,” working paper, Louisiana State University. [36] Johnson, Marilyn F., Karen K. Nelson, and A.C. Pritchard, 2002, “Do the merits matter more? Class action under the Private Securities Litigation Reform Act,” working paper, Michigan State University. [37] Lacker, Je?rey M., and John A. Weinberg, 1989, “Optimal contracts under costly state falsi?cation,” The Journal of Political Economy 97, No. 6, 1345-1363. [38] Li, Si, 2004, “Corporate fraud and costly monitoring: An empirical analysis of a simultaneous system with partial observability,” working paper, Duke University. [39] Klein, April, 2002, “Audit committee, board of director characteristics, and earnings management,” Journal of Accounting & Ecnomics 33, No. 3, 375-400. [40] Leuz, Christian, Alexander Triantis, and Tracy Yue Wang, 2004, “Why do ?rms go dark? Causes and economic consequences of voluntary SEC deregistrations,” working paper, University of Maryland. [41] Loebbecke, James K., Martha M. Eining, and John J. Willingham, 1989, “Auditors’ experience with material irregularities: Frequency, nature, and detectability,” Auditing: A Journal of Practice & Theory 9, No. 1, 1-28. [42] Maksimovic, Vojislav, and Sheridan Titman, 1991, “Financial policy and reputation for product quality,” The Review of Financial Studies 4, No. 1, 175-200. 99
[43] Moeller, Sara, Frederik P. Schlingemann and Rene M. Stulz, 2004, “Wealth destruction on a massive scale? A study of acquiring-?rm returns in the recent merger wave,” Journal of Finance forthcoming. [44] Myers, Stewart C., and Nicholas S. Majluf, 1984, “Corporate ?nancing and investment decisions when ?rms have information that investors do not have,” Journal of Financial Economics 13, 187-221. [45] Peng, Lin, and Ailsa R¨ oell, 2004, “Executive pay, earnings manipulation and shareholder litigation,” working paper, Princeton University. [46] Poirier, Dale J., 1980, “Partial observability in bivariate probit models,” Journal of Econometrics 12, 209-217. [47] Povel, Paul, Rajdeep Singh, and Andrew Winton, 2004, “Booms, busts, and fraud,” working paper, University of Minnesota. [48] Richardson, Scott A., and Richard G. Sloan, 2003, “External ?nancing and future stock returns,” working paper, University of Pennsylvania. [49] Shleifer, Andrei, and Robert W. Vishny, 1997, “A survey of corporate governance,” Journal of Finance 52, No. 2, 737-783. [50] Sidak, Gregory J., 2003, “The failure of good intentions: The WorldCom fraud and the collapse of American telecommunications after deregulation,” Yale Journal on Regulation 20, 207-267. [51] Stein, Jeremy, 1989, “E?cient capital markets, ine?cient ?rms: A model of myopic corporate behavior,” Quarterly Journal of Economics 104, 655-669. [52] Sweeney, A.P., 1994,“Debt Covenant Violations and Manager’s Accounting Responses,” Journal of Accounting and Economics 17, 281-308.
100
[53] Teoh, Siew Hong, Ivo Welch, and T. J. Wong, 1998a, “Earnings management and the postissue performance of seasoned equity o?erings,” Journal of Financial Economics 50, No. 1 63-99. [54] Teoh, Siew Hong, Ivo Welch, and T. J. Wong, 1998b, “Earnings management and the longterm market performance of initial public o?erings,” Journal of Finance 53, No. 6, 1935-1974.
101
doc_817204068.pdf
Securities fraud, also known as stock fraud and investment fraud, is a deceptive practice in the stock or commodities markets that induces investors to make purchase or sale decisions on the basis of false information, frequently resulting in losses, in violation of securities laws.
ABSTRACT
Title of dissertation:
Securities Fraud: An Economic Analysis Yue Wang, Doctor of Philosophy, 2005
Dissertation directed by:
Professor Lemma Senbet, Professor Nagpurnanand Prabhala Department of Finance
This thesis develops an economic analysis of securities fraud. The thesis consists of a theory essay and an empirical essay. In the theory essay, I analyze a ?rm’s propensity to commit securities fraud and the real consequences of fraud. I show that fraud can lead to investment distortions. I characterize the nature of the distortions, and show that it results from fraud-induced market mispricing and management’s ability to in?uence the ?rm’s litigation risk through investment. The theory also characterizes the equilibrium supply of fraud. I demonstrate the linkages between a ?rm’s fraud propensity and the structure of its assets in place and growth options, and analyze the e?ect of corporate governance on fraud. The theory provides testable implications on crosssectional determinants of ?rms’ fraud propensities and the relation between fraud and investment. In the empirical essay, I test my main model predictions, using a new hand-compiled fraud data set. I use econometric methods to account for the unobservability of undetected frauds, and disentangle the e?ects of cross-sectional variables into their e?ect on the probability of committing fraud and the e?ect on the probability of detecting fraud. I ?nd that the level, type, and ?nancing of investment all matter in determining the probability of fraud and the likelihood of detection. I also examine the monitoring roles of large shareholders, institutional owners, independent auditors, and corporate boards. I ?nd that large block or institutional holdings tend to discourage fraud by increasing the detection likelihood. The roles of independent auditors and corporate board are weaker. Finally, insider equity incentives, growth potential, external ?nancing needs and pro?tability all in?uence a ?rm’s propensity to commit fraud. The paper also demonstrates the importance of separating fraud commitment and fraud detection, because cross-sectional variables
can have opposing e?ects on these two components, and therefore can be masked in their overall e?ect on the incidence of detected fraud.
Securities Fraud : An Economic Analysis by Yue Wang
Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park in partial ful?llment of the requirements for the degree of Doctor of Philosophy 2005
Advisory Committee: Professor Lemma Senbet, Co-Chair/Co-Advisor Nagpurnanand Prabhala, Co-Chair/Co-Advisor Professor Vojislav Maksimovic Professor Je?rey Smith Professor Nengjiu Ju
c Copyright by Yue Wang 2005
This dissertation is dedicated to my grandmother and grandfather.
ACKNOWLEDGMENTS
I owe my gratitude to all the people who have made this thesis possible and because of whom my graduate experience has been one that I will cherish forever. First and foremost I’d like to thank my advisors, Professor Lemma Senbet and Professor Nagpurnanand Prabhala, for all the stimulating advices and consistently strong support in the past ?ve years. It has been great pleasure of mine to work with and learn from these extraordinary individuals. I would like to thank Professor Vojislav Maksimovic for all the inspiring discussions we had during my time in the PhD program. I am also extremely grateful to Professor Je?rey Smith for his kind help and insightful comments on my empirical essay. I thank Professor Nengjiu Ju for agreeing to serve on my thesis committee and sparing time to review my manuscript. I also owe my gratitude to all the other faculty members in the Finance Department. I would not have been able to reach this far without their kind encouragement and helpful advice. I wish to convey my special thanks to Professor Alexander Triantis, who has been a great mentor and a wonderful friend during the past two years. I owe my deepest thanks to my family - my mother and father who have always stood by me and guided me through my career, and have pulled me through against impossible odds at times. I’d also like to express my gratitude to my husband Yingkai Liu for walking through all the ups and downs with me in the past four years, and for believing in me even when I do not believe in myself. It is impossible to remember all, and I apologize to those I’ve inadvertently left out. Lastly, thank you all and thank God!
iii
TABLE OF CONTENTS List of Figures 1 Introduction 1.1 Theory of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Empirical Investigation of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi 1 2 4 7 8 8 9 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16 16 17 19 20 24 25 25 28 30 31 33 34 41 41 43 44 44 45 45 47 47 48 49 54 57 58 58 61 61 62 63 65 65 66 67 67
2 Securities Fraud & Securities Litigation 2.1 Securities Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Common Types of Alleged Securities Fraud . . . . . . . . . . . . . . . . . . . . . . 3 Related Literature 4 A Model of Securities Fraud 4.1 Model Framework . . . . . . . . . . . . . . 4.1.1 The Firm . . . . . . . . . . . . . . . 4.1.2 Time Line and Assumptions . . . . . 4.2 Cost and Bene?t of Fraud . . . . . . . . . . 4.2.1 Litigation Cost of Fraud . . . . . . . 4.2.2 Fraud Incentives . . . . . . . . . . . 4.3 Securities Fraud and Investment Incentives 4.3.1 Investment Distortions . . . . . . . . 4.3.2 A Numerical Illustration . . . . . . . 4.4 Disclosure Strategy . . . . . . . . . . . . . . 4.4.1 Equilibrium Misreporting . . . . . . 4.4.2 Fraud Propensity . . . . . . . . . . . 4.5 Model Implications and Discussion . . . . .
5 An Empirical Investigation of Securities Fraud 5.1 Fraud Sample . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Time Trends and Firm Characteristics . . . . 5.1.2 Industry Distribution . . . . . . . . . . . . . 5.1.3 The Nature of Fraud . . . . . . . . . . . . . . 5.2 Empirical Methodology . . . . . . . . . . . . . . . . 5.2.1 A Model with Partial Observability of Fraud 5.2.2 Model Identi?cation and Estimation . . . . . 5.2.3 Comparison with Straight Probit Model . . . 5.3 Hypothesis Development and Model Speci?cation . . 5.3.1 Probability of Fraud Detection . . . . . . . . 5.3.2 Propensity to Commit Fraud . . . . . . . . . 5.3.3 Control Variables . . . . . . . . . . . . . . . . 5.3.4 Summary of Model Speci?cation . . . . . . . 5.4 Descriptive Information and Univariate Analysis . . 5.5 Multivariate Analysis . . . . . . . . . . . . . . . . . . 5.5.1 Firm Characteristics and Fraud . . . . . . . . 5.5.2 Investment and Fraud . . . . . . . . . . . . . 5.5.3 Equity Ownership and Fraud . . . . . . . . . 5.5.4 Auditor, Board and Fraud . . . . . . . . . . . 5.5.5 Summary of Results . . . . . . . . . . . . . . 5.5.6 Comparison with Simple Probit Models . . . 5.6 Robustness Checks . . . . . . . . . . . . . . . . . . . 5.6.1 Frivolous Lawsuits . . . . . . . . . . . . . . .
iv
5.6.2 5.6.3 5.6.4 6 Conclusion
Timing of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Industry and Business Cycle e?ects . . . . . . . . . . . . . . . . . . . . . . . Di?erent Model Speci?cations . . . . . . . . . . . . . . . . . . . . . . . . . .
68 69 70 87 89 96
7 Appendix: Proofs of Propositions Bibliography
v
LIST OF FIGURES Figure 1: Model Time Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figure 2: Probability of Fraud Detection . . . . . . . . . . . . . . . . . . . . . . . 40 40 85 85 86
Figure 3: Determination of the Beginning Fiscal Year of Fraud . . . . . . . . . . . Figure 4: Identi?cation Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . Figure 5: Timing of Fraud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
Chapter 1 Introduction
In recent years, a string of high pro?le corporate scandals like those of Global Crossing, Enron, Tyco, and Worldcom has brought securities fraud and corporate governance to the forefront of public attention and policy debate. The magnitude of the alleged securities fraud is stunning. According to Stanford Securities Class Action Clearinghouse and Cornerstone Research, 224 securities lawsuits in 2002 in the United States were associated with a total $206 billion loss of market capitalization in the defendant ?rms.1 The governance crisis was followed by rapid and substantial legislative and regulatory changes that aimed to restor investor con?dence in the capital markets. The movement was so fast that 9 months after the Enron debacle, President Bush signed the Sarbanes-Oxley bill into law. Securities fraud is a very serious issue. It undermines a core value in capital markets, the integrity of public companies, which is essential to investor con?dence in those markets and the e?cient allocation of capital. Furthermore, we also observe ine?cient investments and serious value destructions in many fraudulent ?rms (e.g., Enron, Nortel, eToys), which implies that there could be large real economic cost associated with fraud. The governance crisis and the on-going governance reform call for careful economic re?ections on what have happened, because the exact nature, signi?cance, and consequences of securities fraud and the economics underlying the legislative and regulatory changes are still incompletely understood. This thesis develops an economic framework to characterize the determinants and consequences of securities fraud. I de?ne securities fraud as deliberate and material misrepresentation of corporate performance, and thus use fraud and misreporting interchangeably. The thesis consists of a theoretical model of fraud and empirical analysis.
1 Cornerstone
Research, “Securities Class Action Case Filings 2002: A Year in Review.”
1
1.1 Theory of Fraud
The theory part of the thesis builds on Gary Becker’s (1968) economic analysis of crime. Following Becker’s approach, one can view fraudulent behavior as an economic activity, whose equilibrium supply depends on a rational calculation of the expected bene?ts and costs from engaging in it. Di?erent ?rms have di?erent propensities to commit fraud because they face di?erent cost-bene?t tradeo?s. In this paper, the bene?t from fraud is that ?nancial misreporting can create (or sustain) short-term market overvaluation of the ?rm. The cost of fraud is litigation risk. With some positive probability, fraudulent activities will be uncovered, resulting in a fraud penalty (which includes both explicit monetary ?nes and other implicit costs, such as loss of reputation). Within this framework, the ?rm’s propensity for fraud, the magnitude of fraud, and the ?rm’s investment incentives are analyzed. The theory demonstrates an interesting link between a ?rm’s ?nancial disclosure incentive and its real investment decision. First, ?nancial misreporting can a?ect the short-term market valuation of the ?rm and allow the ?rm to invest using cheap outside capital. Second, after committing fraud, the ?rm has incentive to cover things up. Such incentive can motivate the ?rm to strategically use investment to mask fraud and reduce its litigation risk. The basic intuition is that stochastic cash ?ows from a new investment can decrease the precision of the ?rm’s total cash ?ow and create inference problems for the market. In sum, investment can a?ect both the ?rm’s ex ante bene?t from committing fraud and its ex post probability of being detected. The model predicts that fraudulent ?rms tend to overinvest in the sense that they would undertake some negative NPV projects that destroy shareholder value. In particular, fraud can induce a preference for risky (in terms of high return volatility) or uncorrelated projects (uncorrelated with the cash ?ow from existing assets), because these types of investment can better disguise fraud than others. The investment distortion can lead to serious value destructions in the ?rm, which is the real economic cost of fraud. Furthermore, the cost of ine?ciency is borne by not only shareholders of fraudulent ?rms but also those of honest ?rms, because ex ante the market cannot perfectly distinguish between the two types of ?rms.
2
The theory further characterizes the ?rm’s equilibrium disclosure strategy. The model shows that the ?rm will honestly reveal performance if its performance is very good or if it is desperately bad. The former case is associated with low bene?t from fraud, and the latter is associated with high litigation risk. The ?rm’s propensity to commit fraud and the magnitude of fraud depend on the nature of the ?rm’s assets and growth opportunities. The model predicts that fraudulent ?rms tend to have high growth potential but experience negative pro?tability shocks. Growth potential can positively in?uence the ?rm’s payo? from fraud and negatively in?uence its litigation risk. In addition, litigation events tend to cluster in certain industries during some speci?c time period, because ?rms’ cost-bene?t tradeo?s of fraud are correlated within an industry. The theory also generates implications about the role of corporate governance in the context of corporate fraud. The model shows that good corporate governance can increase the likelihood of fraud detection and thus deter fraud ex ante. However, corporate governance may also fail to prevent fraud if it is just about aligning the interest of the management with that of incumbent shareholders. This is because even when such alignment is perfect, fraud can still emerge in equilibrium. Finally, the theory demonstrates the e?ect of the endogenous detection risk on the crosssectional variations in ?rms’ fraud propensities. While the penalty for fraud (at least the explicit liability provisions) is largely determined by securities laws and thus is exogenous to the ?rm, the probability of detection can be in?uenced by the ?rm’s endogenous actions (e.g., investment, disclosure) as well as ?rm-speci?c attributes. This endogeneity implies that the detection risk is more important in determining cross-sectional variations in ?rms’ fraud propensities than are penalty provisions. Therefore, without increasing the probability of detection, enhanced liability standards alone may achieve only limited deterrence, because ?rms can undo some e?ects of tightened penalty by adjusting their probabilities of getting caught. More important, fraudulent ?rms’ incentive to decrease their likelihood of being detected can be a potential source of value destruction. Therefore, there is a real danger associated with over-regulation.
3
1.2 Empirical Investigation of Fraud
I then empirically test some of my key model predictions. Speci?cally, I examine the e?ects of real investment, corporate monitoring, and ?rm characteristics on a ?rm’s cost-bene?t tradeo? of committing fraud. The analysis is based on a new hand-compiled fraud data set, which consists of private securities class action lawsuits ?led between 1996 and 2003 against US public companies with allegations of accounting irregularities. The next contribution of the paper is methodological. In assessing a ?rm’s propensity to commit fraud, we face an identi?cation problem because we only observe detected fraud. A nonlitigated ?rm can be either an honest ?rm or an undetected fraudulent ?rm. This implies that the probability of a ?rm committing fraud and the probability of observing the ?rm as fraudulent can be very di?erent. This study utilizes statistical methods to control for this problem. In essence, I model the probability of detected fraud (what we observe) as the product of two latent probabilities: the probability of committing fraud and the probability of detecting fraud conditional on fraud occurrence. Then I use econometric methods to back out these two latent probabilities. Disentangling fraud commitment and fraud detection provides two advantages. First, it allows me to control for the unobservability of frauds committed but not detected. Second and more important, it allows me to examine the economics of each probability as well as their interactions.2 Using the above methodology, I examine the link between real investment and the incidence of fraud. There has been surprisingly little exploration on the relation between corporate fraud and investment. This is, however, an important issue. We have observed ine?cient investments and serious value destructions in many fraudulent ?rms (e.g., Enron, Nortel, eToy). Hence, there can be large real economic cost associated with fraud. Wang (2004) theorizes that fraud can induce overinvestment incentives for two reasons. First, fraud can create (or sustain) market overvaluation and decrease the external ?nancing cost of investment. Second, after committing fraud, the ?rm has incentive to strategically use investment to mask fraud and decrease its litigation
2 In
a concurrent paper, Li (2004) uses a simultaneous model with partial observability to analyze the role of the
SEC in detecting fraud.
4
risk. I ?nd evidence that supports this theory. First, I ?nd that the alleged fraudulent ?rms on average have larger investment expenditures than a random sample of non-convicted ?rms and a sample of size and age matched comparison ?rms. Second, di?erent types of investment appear to have di?erential e?ects on the probability of fraud detection. Risky investment (e.g., investment in R&D) and uncorrelated investment (e.g., diversifying acquisition) tend to decrease the probability of detection, while straightforward investment (e.g., capital expenditures) and correlated investment (e.g., focused acquisition) do not. Lastly, di?erent investments also in?uence a ?rm’s propensity to commit fraud di?erently. This is either because of the way the investments are ?nanced (e.g., stock-based vs. cash-based acquisitions) or because of the di?erential litigation risk they induce. Overall, the empirical results imply that investment is associated with both a ?rm’s ex-ante bene?t from committing fraud and its ex-post litigation risk, and thus is an important determinant of the ?rm’s fraud incentives. Corporate securities fraud also provides a new and interesting angle to examine the roles of di?erent corporate monitors in determining ?rms’ incentives and behavior. E?ective monitoring should increase the probability of uncovering fraudulent corporate activities and discourage fraud ex ante. I investigate four types of corporate monitors: large shareholders, institutional owners, independent auditors, and board of directors. I ?nd that the presence of block equity holders and large institutional ownership are associated with high probability of fraud detection and low probability of fraud. For example, increasing block ownership by 10% on average tends to increase the probability of detection by 1% and decrease the probability of fraud by 4%. This supports the view of enhancing shareholder monitoring in combatting corporate fraud. The roles of independent auditors and corporate boards seem to be much weaker. I ?nd no evidence that auditors’ opinions increase the likelihood of fraud detection. There is some weak evidence that reputable independent auditors and large corporate boards are related to higher likelihood of fraud detection. The role of insider equity incentives has received a great amount of public attention following the recent wave of corporate scandals. Several studies have documented that large executive pay for performance sensitivity is associated with high probability of corporate fraudulent reporting
5
(see, e.g., Johnson, Ryan and Tian (2003), Peng and R¨ oell (2004), and Burns and Kedia (2004)). In this paper, since I separate the probability of fraud from the probability of detected fraud, I am able to more directly examine the e?ect of insider equity incentive on a ?rm’s propensity to commit fraud. Interestingly, I ?nd a concave relation between the two. When insider equity incentive is small, the probability of fraud increases as the equity incentive increases. When insider equity incentive becomes large, the positive relation weakens and eventually reverses. Overall, this result seems to support the predictions of the agency theory. However, it also implies that equity incentive can be a double-edged sword when it is used to align managerial and shareholder interests in dispersedly-owned ?rms. Finally, I examine how ?rm characteristics in?uence a ?rm’s cost-bene?t tradeo? of engaging in fraud. I ?nd that high growth potential and large external ?nancing need are two important motivational factors for fraud. Alleged fraudulent ?rms on average grow much faster than comparison ?rms and have larger portion of the growth supported by external capital. There is also indirect evidence that fraudulent ?rms generally experience negative pro?tability shocks in the year when fraud begin. Most existing studies on corporate fraud have focused on the bene?t side of the tradeo?. The literature on earnings management and corporate fraud has provided evidence that managers misreport corporate performance in order to facilitate external ?nancing activities, to avoid violations of debt covenants, or to increase performance-related compensation (see, Healy and Wahlen (1999) for a review). On the cost side of the tradeo?, a few papers have examined the consequences following the revelation of fraud. For example, Dechow, Sloan, and Sweeney (1996) show that the revelation of fraud leads to persistent increase in the ?rm’s cost of capital. Baucus and Baucus (1997) ?nd that ?rms convicted for illegal corporate behavior su?er from prolonged poor operating performance. Gande and Lewis (2005) document signi?cant negative abnormal returns upon the ?ling of securities lawsuits. The only paper I know that studies the probability of fraud detection is Li (2004). Li emphasizes the strategic role of the SEC in detecting corporate fraud, and documents that a larger SEC budget increases the probability of fraud detection and deters fraud. My paper
6
further demonstrates the importance of understanding ?rm-level economic determinants of fraud detection and how the detection risk in?uences its ex-ante propensity to commit fraud. I show that the probability of detection depends on ?rms’ investment decisions, the strength of corporate monitoring, and ?rm-speci?c attributes. The cross-sectional variations in the detection risk help to explain the variations in ?rms’ fraud propensities. I also demonstrate the importance of disentangling the probability of committing fraud from the probability of detecting fraud. Cross-sectional variables can have opposing e?ects on the two latent probabilities, and thus can be masked in their overall e?ect on the incidence of detected fraud. For example, this paper shows that large institutional ownership is associated with high probability of fraud detection and low probability of fraud. The e?ect on detection tends to dominate and thus we observe a positive relation between institutional ownership and the compound probability (incidence of detected fraud). This may lead us to draw incorrect inferences. Distinguishing the probability of fraud from the probability of detected fraud is not only important for understanding the economics of fraud, but also relevant from a regulatory point of view in setting policies that deal with fraud.
1.3 Thesis Structure
The thesis is structured as follows. Chapter 2 introduces the basic institutional knowledge about securities fraud and securities class action litigation. Chapter 3 reviews the related literature. Chapter 4 develops an analytical model to characterize the economic determinants of corporate fraud propensity and the real consequences of fraud. Chapter 5 empirically investigate ?rms’ fraud incentives and fraud detection. Finally, Chapter 6 concludes.
7
Chapter 2 Securities Fraud & Securities Litigation
The purpose of this chapter is to provide some general knowledge about securities fraud, securities laws and regulations, and securities litigation. The structure of this chapter is as follows. Section 2.1 presents a de?nition of securities fraud, and describes the major anti-fraud laws and regulations that govern the securities industry. Section 2.2 describes the common types of fraud allegations in the private securities class action litigation between 1996 and 2002.
2.1 Securities Fraud
A thorough understanding of the nature, signi?cance and consequences of securities fraud requires a proper de?nition of securities fraud. Fraud, in general, as de?ned in Webster’s Universal College Dictionary, is deceit or trickery perpetrated for pro?t or to gain some unfair or dishonest advantage. I de?ne securities fraud as follows based on the description of the SEC and the Securities Exchange Act of 1934. Securities fraud refers to the use of any manipulative and deceptive devices, in connection with the purchase or sale of any security, that are in contravention of such rules and regulations as the Commission may prescribe as necessary or appropriate in the public interest or for the protection of investors. The term “security” means any note, stock, treasury stock, bond, debenture, derivative securities, certi?cate of interest, or in general, any instrument commonly known as a security. Section 10(b) of the Securities Exchange Act of 1934 and the rules promulgated thereunder (especially Rule 10(b)-5) build the major substance of the broad anti-fraud provisions that make it unlawful for anyone to engage in fraud or misrepresentation in connection with the purchase or sale of a security. Violations of these provisions include employment of any devices, schemes or arti?ce to defraud, misrepresentation and/or omission of material information, or engaging in any act, practice or course of business which operates or would operate as a fraud or deceit upon
8
any person, in connection with the purchase or sale of any security. The essence of the above regulations is to prohibit deliberate and material information misrepresentation in any form of public communications between the ?rm and its investors, and between the ?rm and its regulators. There are two major types of securities litigation: the SEC’s enforcement actions and the private class action litigation. According to Securities Class Action Clearinghouse (SCAC) established by the Stanford Law School, a securities class action is a case brought pursuant to Federal Rule of Civil Procedure 23 on behalf of a group of persons who purchased the securities of a particular company during a speci?ed period of time (the class period). The complaint generally contains allegations that the company and/or certain of its o?cers and directors violated one or more of the federal or state securities laws. A suit is ?led as a class action because the members of the class are so numerous that joinder of all members is impracticable.
2.2 Common Types of Alleged Securities Fraud
Table 1 lists the types of commonly alleged securities fraud in class action lawsuits between 1996 and 2002 and the frequency distribution. The litigation information is retrieved from SCAC. I identify the speci?c nature of fraud allegations based on information extracted from the case complaints and/or press releases. In each year there was a small number of cases that did not provide enough information for us to determine the nature of the allegations. Therefore, the information provided in this section is based on the identi?able ?lings. 1. Financial statement fraud, which refers to the deliberate and material misstatement of ?nancial statements issued by publicly traded companies to mislead the ?nancial statements users (Rezaee [2002]). 2. Misrepresentation or concealment of material facts (excluding misreporting in the ?nancial statements). Material facts are the ones that, if made available, would cause the information receivers to change their judgment or decision. This category of securities fraud includes a public ?rm issuing false information and/or omit important information in security registration statements/prospectus (section 11, 12(a) of the Securities Act of 1933), in proxy 9
statements (section 14 of the Exchange Act of 1934) and other important public documents, as well as false and misleading oral communications at press releases and conference calls. Many allegations in this category also frequently involves a?rmative fraud, i.e., the release of false forward-looking statements to the investing public. An example of a?rmative fraud is that a public ?rm issue glowing but misleading projections about the ?rm’s future business prospects and competitive position. 3. Illegal insider trading. According to the SEC’s de?nition, illegal insider trading refers generally to buying or selling a security, in breach of a ?duciary duty or other relationship of trust and con?dence, while in possession of material, non-public information about the security. Insider trading violations may also include “tipping” such information, securities trading by the person “tipped”, and securities trading by those who misappropriate such information. “Insiders” generally include o?cers, directors, and individuals who hold more than 10 percent of the company’s stock (regardless of whether they work for the company). 4. Investment bank fraud. This category of fraud refers to the unfair dealings in investment banking activities. Most commonly alleged investment bank frauds include unfair IPO allocations and misleading analyst reports. 5. Breach of ?duciary duty. This category generally involves violations of section 14 of the Exchange Act. Section 14 prohibits any information misrepresentation in the proxy statements, particularly information related to tender o?ers, management buyouts, and other merger/acquisition activities. Most of the cases in this category alleged that the management or controlling shareholders expropriated minority shareholders in merger/acquisition activities, and misled minority shareholders to tender or exchange their shares at unfairly low prices. 6. Stock price manipulation, which refers to deliberate buying or selling of a security, or deliberate intervention of other people’s buying or selling of a security, in order to control the price of the security.
10
Table 2.1: Commonly Alleged Securities Fraud This table presents the types of commonly alleged securities fraud in 1268 private securities class action lawsuits between 1996 and 2002. Pure investment bank fraud cases (i.e., cases that allege unfair IPO allocations by securities underwriters and untrue securities analyst reports) are excluded. “Other information misrepresentation/omission” means material non-accounting related information misreporting or omission. Nature of Fraud Number of observations Accounting irregularity Other information misrepresentation/omission Illegal insider trading Breach of ?duciary duty Stock price manipulation # of Filings 1268 596 486 337 49 9 % of Total 47.08 38.33 26.58 3.90 0.71
11
Chapter 3 Related Literature
This thesis is related to several strands of research: (1) the accounting literature on earnings management and ?nancial disclosure; (2) the literature on agency theory; and (3) recent research on corporate fraud. The economics of corporate misreporting is examined in the accounting disclosure literature. Dye (1988) analyzes two conditions under which earnings management may exist in equilibrium. First, the cost-minimizing contract that induces preferred action from the manager may not prevent earnings management, which leads to the internal demand for earnings management. Second, incumbent shareholders may attempt to alter the perceptions of prospective investors through managed earnings, which creates the external demand for earnings management. In line with Dye’s notion of internal demand for earnings management, Lacker and Weinberg (1989) show that the optimal risk sharing contract may not prevent the agent from falsifying the outcome. Goldman and Slezak (2003) show that the optimal equity compensation contract that induces the desired managerial e?ort may not prevent (and may even encourage) the agent from misreporting. Several other papers together with my paper are consistent with Dye’s notion of external demand for earnings management. Stein (1989) argues that capital market pressure can induce the management to in?ate current pro?tability at the expense of forgoing future cash ?ows. Bebchuk and Bar-Gill (2002) present a model in which ?rms’ needs for external ?nancing and insiders’ bene?t from informed trading can motivate management to misreport corporate performance. Jensen (2004) argues that corporate fraud can result from a dramatic form of capital market pressure. When the market substantially overvalues a ?rm’s equity, the ?rm may feel forced to defraud investors in order to defend such overvaluation, and this can lead to serious value destructions in the ?rm. I show that overvaluation can result from the ?rm’s endogenous choice, and an important
12
source of value destruction is the fraud-induced investment distortions. There has been a large body of empirical research on earnings management. Earnings management does not necessarily imply the existence of securities fraud. Earnings management re?ects discretionary managerial judgment (or managerial ?exibility) in corporate ?nancial reporting.1 However, both securities fraud and earnings management involve some information misrepresentation. A number of studies have examined di?erent incentives for earnings management, including capital market needs, contracts written on accounting numbers and government regulations (see Healy and Wahlen (1999) for a review of empirical work on earnings management). The literature provides evidence that managers have incentives to manipulate earnings in an attempt to in?uence short-term stock price performance before major external ?nancing activities or externally-?nanced investment (see, e.g., Teoh, Welch and Wong (1998a,b) on public equity o?ers; Erickson and Wang (1998) on stock-?nanced acquisitions). Efendi, Srivastava and Swanson (2004) ?nd that the likelihood of an earnings restatement is signi?cantly higher for ?rms that make one or more sizable acquisitions. Several studies have examined the relation between the structure of managerial compensation contracts and the likelihood of earnings management, and ?nd that the pay-performance sensitivity induced by stock options seems to increase earnings management (see, e.g., Gao and Shrieves (2003)). Research directly on corporate fraud has been sparse, but has started to attract academic interest after the explosion of corporate scandals and the recent legislation movement. Most of the recent studies focus on the e?ects of insider equity incentives on ?rms’ incentives to misreport. Johnson, Ryan, and Tian (2003), Peng and R¨ oell (2004), Burns and Kedia (2004), and Efendi, Srivastava and Swanson (2004) all ?nd that large executive pay-for-performance sensitivity is positively associated with fraudulent reporting. These results seem to support the over-incentivization argument that insider equity incentive is a double-edged sword. It may induce managerial misreporting incentive rather than managerial e?ort in creating shareholder wealth. Alexander and Cohen (1999), however, documents a negative relation between insider ownership and the likelihood
1 Schipper
[1989] de?nes earnings management as “purposeful intervention in the external reporting process, with
the intent of obtaining some private gain to managers or shareholders”.
13
of fraud and provide some support for the classic agency theory. Several studies have examined the relation between the characteristics of the board and the probability of corporate fraudulent reporting. Beasley (1996) studies a sample of ?rms subject to SEC’s AAERs and ?nds that board independence (proxied by the percentage of outside directors in the board) is signi?cantly negatively related to the likelihood of ?nancial statement fraud. Klein (2002) ?nds an inverse relation between board independence and abnormal accruals. Dechow, Sloan and Sweeney (1996) ?nd that ?rms committing ?nancial statement fraud are likely to have a board dominated by insiders and have a CEO who is also the chairman of the board or the founder of the company. Agrawal and Chadha (2004) examine the incidence of accounting restatements, and ?nd that board independence is irrelevant, but the presence of independent directors with ?nancial or accounting expertise on the audit committee is associated with signi?cantly lower probability of accounting restatements. The major contributions of my thesis to the literature are threefold. First, my thesis is the ?rst paper that seriously analyzes the role of real investment in the context of corporate fraud. My theory model shows that investment in?uences both a ?rm’s ex-post probability of fraud detection and its ex-ante propensity to commit fraud. Fraud can induce distorted investment incentives, which is the real economic cost of fraud. Second, I empirically examine the model predictions on the relation between fraud and corporate investment incentives, and ?nd strong support for my theory. For example, I ?nd that risky investment and uncorrelated investment have a strong negative e?ect on the probability of fraud detection, while straightforward investment and correlated investment do not. This implies that the type of investment matters in determining the ?rm’s detection risk. I also ?nd that acquisition expenditures in?uence the probability of fraud only if it is (at least partially) ?nanced by stock, indicating that the ?nancing of investment matters. The third contribution of my thesis is methodological. I introduce a new empirical methodology to analyze fraud. The existing literature has ignored the fact that we only observe detected fraud. That is, the outcome we observe depends on the outcome of two latent processes: fraud commitment and fraud detection. If we ignore this structure, we could draw incorrect inferences, because the same
14
variable can have opposing e?ects on the two latent processes and thus get masked in its overall e?ect on the outcome we observe. This study utilizes statistical methods to disentangling fraud commitment and fraud detection. This allows me to control for the unobservability of undetected frauds. More important, this allows me to examine the economics of each component as well as their interactions. My thesis provides new insights about corporate fraud incentives that cannot be obtained using the models in the existing literature.
15
Chapter 4 A Model of Securities Fraud
This chapter develops an economic framework of securities fraud. I analyzes the interaction between the ?rm’s ?nancial disclosure and its real investment decision. I also characterize the ?rm’s equilibrium disclosure strategy, probability of committing fraud and the magnitude of fraud. The chapter is structured as follows. Section 4.1 introduces the model framework and assumptions. Section 4.2 characterizes the ?rm’s cost-bene?t tradeo? of committing fraud. Section 4.3 examines the ?rm’s investment incentives in the presence of fraud. Section 4.4 derives the ?rm’s equilibrium disclosure strategy. Section 4.5 discusses model implications and possible extensions.
4.1 Model Framework 4.1.1 The Firm
Consider a typical public ?rm whose market value consists of both its assets in place and
2 1 growth opportunities. The asset value is normally distributed, A ? N (A, ?A ). The growth
opportunity takes the form of a possible new investment project in the future whose value is also
2 normally distributed, G ? N (G, ?G ). The market knows the distributions of A and G, but does
not observe the realizations of each component. The market value of the ?rm is the expected discounted value of future cash ?ows. For simplicity, I assume that investors are risk neutral, and the discount rate is zero. Therefore, the ?rm value is simply E (V ) = A + G. The ?rm is operated by a manager who owns a fraction 0 < ? < 1 of the ?rm. I assume that the manager holds restricted stock and thus is not allowed to trade any of his own equity shares. This simplifying assumption allows abstraction from the incentive and signalling e?ects of insider trading, and it also implies that the manager maximizes the wealth of long-term shareholders.2
1I 2I
can always choose reasonable values for A and ?A such that negative asset values have almost zero probabilities. assume that there is no opportunity for perquisite consumption. This type of agency problem is not the focus
16
The accounting and auditing literature has provided evidence that both capital market activities (see the citations in the introduction) and pro?ts from informed trading (e.g., Summers and Sweeney (1998)) can motivate fraudulent reporting. According to my study of private securities class action litigation against US public companies between 1996 and 2002, about 68% of the securities lawsuits involved misreporting surrounding major capital market activities (external ?nancing or externally-?nanced investment), and about 29% of the cases involved allegations of illegal insider trading and insider personal gains. This paper focuses on fraud and ?rm investment, and thus analyzes the former scenario. The model shows that even when the manager’s interest is perfectly aligned with that of existing shareholders, fraud can still exist in equilibrium. Adding managerial agency problem could, of course, further exacerbate the manager’s fraud incentives.
4.1.2 Time Line and Assumptions
There are four periods in this model, t = 0, 1, 2, 3. The sequence of events is described below (also see Figure 1 at the end of the chapter for an outline). Time 0: Institutional Arrangements At time 0, the institutional arrangement of the ?rm
is established. The strength of the ?rm’s internal corporate governance is indicated by p ? [0, 1]. Higher p represents better governance and also higher likelihood of internal detection of fraud.3 Time 1: Disclosure of Earnings At time 1, the manager privately observes the realization
of the intermediate earnings generated by the ?rm’s assets.4 The earnings realization is drawn from the following process. e = q A + u. (4.1)
q indicates the average productivity of the ?rm’s assets in place, of which the market is aware. u
2 is a white noise term, u ? N (0, ?u ). Equation (4.1) shows that the realized intermediate earnings
(e) contain useful information about the value of the ?rm’s assets. Let the signal-to-noise ratio be
of this paper.
3 Section 4 The
4.5 will discuss the possibility of endogenizing this parameter.
intermediate information does not have to be earnings. It can be any valuable piece of accounting in-
formation, or even more general information about the ?rm’s overall ?nancial condition, operational condition, or business prospects.
17
? ?
2 q?A 2 +? 2 . q 2 ?A u
Then the expected value of the assets conditional on the earnings realization e is
˜|e) = A + ? (e ? e). E (A After observing the intermediate earnings, the manager makes a disclosure decision, y (e) = e + ?. (4.2)
? represents the amount of distortion in the reported earnings. ? = 0 means that the manager chooses to truthfully reveal the earnings realization. ? > 0 implies that the manager in?ates earnings. ? is assumed to be nonnegative. That is, this paper focuses on overreporting of earnings. It is possible that managers may intentionally understate earnings (e.g., the case of Freddie Mac). Empirical studies on earnings management as well as SEC accounting and auditing enforcement actions, however, indicate that accounting overstatement is much more frequently observed than understatement (see, e.g., Feroz, Park, and Pastena (1991); Rezaee (2002)), and thus it is a more interesting subject for research. Once the earnings disclosure is made, the market prices the ?rm’s equity based on the reported earnings y (e), but the market does not have to take the earnings announcement at face value. Investors are generally aware of the possibility of misreporting. The market’s prior belief about the ?rm’s likelihood of misreporting is ?0 ? [0, 1], and the expected amount of misreporting ˜|y (e)], where the expectation is ? . Then the time 1 market value of the ?rm’s assets is V1 = E?0 [A incorporates the market’s prior belief about fraud. Time 2: Investment Decision In this period, a new investment opportunity arrives with
2 probability ?, requires an initial outlay of $I , and will generate a gross return R, R ? N (R, ?R ).
For simplicity, I set R = 1, which allows me to parameterize the pro?tability of the new investment in a straightforward way. Once the new investment opportunity arrives, the manager observes the gross return as r, the realization of R. The market does not observe this but knows the return distribution (i.e., the mean and variance of R). The manager makes an investment decision: whether to take the new project or not. If he decides to take it, the ?rm needs to raise $I as the initial capital. I assume that new equity shares are issued. I will discuss the robustness of the model results with respect to this assumption in 18
section 4.2.2. Time 3: Liquidation time 2, V = A + IR = If the ?rm does not invest, V =A= 1 1 e ? u. q q (4.4) 1 1 e + I R ? u. q q (4.3) At time 3, the ?rm has a liquidating cash ?ow V . If the ?rm invests at
The market is able to observe this ?nal cash ?ow and can use this information to update its belief about the probability of fraud at time 1. How the market interprets a particular ?nal cash ?ow realization depends on the market’s expectation about V . The following table lists four distributions of V : the perceived distribution (conditional on y (e)), the true distribution (conditional on e), the distribution given that the ?rm invests (I), and the one given not (N). Investment (I) True E (V |I, e) = E (A + I R|I, e) V ar(V |I, e) = V ar(V |I ) Perceived E (V |I, y ) = E (A + I R|I, y ) V ar(V |I, y ) = V ar(V |I ) No Investment (N) E (V |N, e) = E (A|e) V ar(V |N, e) = V ar(V |N ) E (V |N, y ) = E (A|y ) V ar(V |N, y ) = V ar(V |N )
2 2 2 V ar(V |I ) = ?e /q + (I?R )2 + 2?I?R ?e /q + ?u /q 2 , where ? is the correlation between e and R. 2 2 2 V ar(V |N ) = ?e /q + ?u /q 2 . We can see that misreporting only distorts the expected value of the
?rm’s ?nal cash ?ow, not the variance of it.
4.2 Cost and Bene?t of Fraud
This section characterizes the cost-bene?t tradeo? of committing fraud. The litigation cost of fraud is derived in section 4.2.1. The bene?t from fraud and the manager’s optimization problem are presented in section 4.2.2.
19
4.2.1 Litigation Cost of Fraud
At time 3, after the realization of the ?nal cash ?ow, the market may unearth the manager’s misreporting at time 1 with some probability. If fraud is detected, the ?rm will be subject to a fraud penalty. The expected litigation cost, which is the product of the detection likelihood and the penalty after detection, is the cost of committing fraud.
Probability of Fraud Detection
This model considers two fraud detection mechanisms: detection through cash ?ow and detection through internal corporate governance.5 At time 3, after observing the ?rm’s ?nal cash ?ow, the uninformed outsiders rationally choose an investigation strategy that maximizes their payo? from litigation.6 More speci?cally, the market chooses a threshold v such that it will investigate the manager’s time 1 disclosure whenever the ?nal cash ?ow realization V falls below this threshold. I assume that any misreporting, if it exists, will be discovered upon investigation (i.e., the conditional probability of fraud detection upon investigation is 1). Thus, I will use the probability of fraud investigation and the probability of fraud detection interchangeably. I call the region {V : V ? v } the cash ?ow detection region. If this region is not reached (i.e., V > v ), no external investigation will be triggered, but detection of fraud is still possible. In this situation, the probability of fraud detection solely depends on the ?rm’s quality of corporate governance (p). That is, p indicates the likelihood of an internal investigation of fraud when the cash ?ow realization does not automatically reveal fraud. In sum, the likelihood of detection conditional on V ? v is 1, and the likelihood conditional on V > v is p. Then the e?ective probability of fraud detection is P = P rob.(V > v ) × p + P rob.(V ? v ) × 1. Probability of Cash Flow Detection
5 Of
(4.5)
At time 3, if the market investigates the ?rm, the
course, there are other detection forces, such as regulators (SEC) and independent auditors. This paper
focuses on the role of capital markets and internal corporate governance in discovering fraud.
6 Here
the outsiders can be the ?rm’s outside (and uninformed) investors or the regulators such as the SEC.
Therefore, the role of outsiders represents general capital market monitoring.
20
expected payo? from the e?ort is f E (? |V ) ? C , where C > 0 is the investigation cost. Therefore, the market will examine the ?rm’s disclosure practice if and only if f E (? |V ) ? C ? 0, or E (? |V ) = y ? E (e|V ) ? De?ne ?V = have E (e|V ) = e + ?V [V ? E (V |y )]. (4.7)
cov (e,V ) V ar (V ) .
C . f
(4.6)
Then under the perceived cash ?ow distribution (the one based on y (e)), we
When we substitute this expression into equation (4.6), we can see that an external investigation will be triggered if and only if V ? v = E (V |y ) ? e + C/f ? y . ?V (4.8)
This condition implies that when the ?nal cash ?ow realization is su?ciently below the market expectation, outside investors will rationally think they have been misled and will start an investigation. De?ne vc = v ? E ( V |y ) V ar(V ) ,
and let ? denote the standard normal cumulative distribution function. Then the ?rm’s probability of facing an outside investigation under the perceived distribution is7 P rob.[V ? v |y ] = ?(vc ). (4.9)
Yet, the ?rm’s true probability of having an external investigation is not simply ?(vc ). Let ? = ?
1 V ar (V )
be the precision of the ?rm’s ?nal cash ?ow. Then, under the true cash ?ow
distribution (the one based on e), we have P rob.[V ? v |e] = ?(vc + K ), (4.10)
where K = [E (V |y ) ? E (V |e)]? . We can see that when K is positive, the ?rm’s actual probability of cash ?ow detection is strictly greater than ?(vc ). In other words, the more the manager can
7 Since
?(vc ) is not necessarily zero, even an honest ?rm may face an outside investigation. However, if the ?rm
has not misreported, the investigation will not lead to discovery of fraud. Thus the honest ?rm will not be punished even if it may face an outside investigation.
21
raise the market’s expectation about V by false disclosure (E (V |y ) > E (V |e)), the more likely is an outside investigation of fraud (see Figure 2 at the end of the chapter for an illustration). This implies that the bene?t and cost of fraud are endogenously related to each other, and there exists an optimal size of fraud. In sum, the essential point underlying the cash ?ow detection mechanism is that the ?nal cash ?ow realization V is a function of the true earnings realization e, not the reported earnings y (e) (see equations (4.3) and (4.4)). Therefore, investors can update their belief about the probability of misreporting after observing V , whose realization the fraudulent manager cannot fully control. This implies that fraud can be partially self-revealing, which is supported by securities litigation in the United States. Table 2 at the end of the paper lists the corporate events or entities that precipitated the federal private securities class action lawsuits ?led in 1996 and 1997 in the United States. Among the 187 lawsuits, at least 132 cases (or 70.6% of the total) were ?led after some unexpectedly disappointing earnings realizations. Expected Probability of Fraud Detection At time 1, when the manager makes the disclosure
decision y (e), what matters is his expected fraud detection likelihood P . Essentially, P tells the manager how risky it is to commit fraud. Let ?I (?N ) be the probability of cash ?ow detection if the ?rm invests (does not invest). ?I ?N vc,I vc,N KI KN where ?I = ?
1 V ar (V |I )
= = = = = =
?(vc,I + KI ), ?(vc,N + KN ), ? e + C/f ? y ?I , ?V,I e + C/f ? y ? ?N , ?V,N
(4.11) (4.12) (4.13) (4.14) (4.15) (4.16)
[E (V |y ) ? E (V |e)]?I , [E (V |y ) ? E (V |e)]?N ,
and ?N = ?
1 . V ar (V |N )
Now let PI (PN ) denote the e?ective probability of
fraud detection, given that the ?rm invests (does not invest) at time 2. Then according to equation
22
(4.5). We have PI PN = (1 ? ?I )p + ?I , = (1 ? ?N )p + ?N . (4.17) (4.18)
These two equations imply that the probability of fraud detection within the ?rm depends on ?rm-speci?c attributes, such as the quality of corporate governance and the nature of cash ?ows. More important, ?I and ?N depend on the manager’s disclosure strategy (y (e)) and the market’s response (E [V |y (e)]). This implies that the likelihood of detection and thus the litigation cost are endogenous to the manager’s decision making. At time 1, the manager’s expected probability of fraud detection (P ) is simply a weighted average of PI and PN . Let x be the probability that the ?rm will undertake a newly arrived investment project (x will be endogenously determined in section 4.3). Then ?x is the probability that the ?rm will exercise a growth option at time 2. Then we have P = ?xPI + (1 ? ?x)PN . (4.19)
Fraud Penalty
Once fraud is discovered, the ?rm is subject to a legal ?ne of f ? . That is, the fraud penalty is assumed to be proportional to the amount of distortion in the earnings announcement. The ?ne is paid out of the company’s ?nal cash ?ow V . Monetary settlement is a prevailing means of fraud punishment. Of course, there are other negative consequences of fraud such as the negative price response to securities litigation (Gri?n, Grundfest and Perino (2003)), loss of the ?rm’s reputation, persistent increase in the cost of capital (Dechow, Sloan, and Sweeney (1996)), and long-run poor ?rm performance (Baucus and Baucus (1997)). I incorporate all the explicit and implicit fraud consequences in the marginal fraud penalty parameter f and measure them in terms of money. In order to understand the nature of securities fraud and the role of securities litigation (or fraud detection), it is important to know who bears the litigation cost of fraud (i.e., who pays the ?ne and who receives the compensation). Let us consider a typical private securities class action
23
litigation. The plainti? (or class members) is a group of the ?rm’s outside security holders (e.g., equity holders, debt holders) who purchase the ?rm’s public securities during some speci?c time period (class period). Once the lawsuit is settled, the defendant ?rm (or its existing shareholders) pays the settlement to the plainti? investors. In this model, the class period would start at time 1 if the manager makes false disclosure and end at time 3 if the fraud is uncovered. The class members would be the new (and uninformed) shareholders who ?nance the ?rm’s new project at time 2.
4.2.2 Fraud Incentives
If a new investment opportunity arrives at time 2 and the ?rm takes it, the market value of the ?rm based on its earnings disclosure and investment decision is E (V |I, y ), while the true value of the ?rm is E (V |I, e). The di?erence between E (V |I, y ) and E (V |I, e) results from the misreporting of earnings at time 1. In order to undertake the new investment, the ?rm needs to raise $I by issuing a fraction ? (y ) = I E (V |I, y )
of new equity. ? is the percentage ownership of the new shareholders. The expected value to existing shareholders at time 3 is thus (1 ? ? )E (V |I, e). The value of ? indicates the cost of external ?nancing. A high ? means that the incumbent shareholders need to sacri?ce a large fraction of the ?nal cash ?ows in order to raise $I , or a high cost of external capital. We can see that ? is a function of the reported earnings y (e). If E (V |I, y ) increases in y (e), then ? decreases in y (e). This implies that a potential bene?t of committing fraud is that ?nancial misreporting may create (or sustain) short-term market overvaluation of the ?rm’s equity and thus reduce the ?rm’s cost of external ?nancing.8 Of course, there may exist other motives for fraud, such as incentive
8I
assume that the ?rm has to ?nance the new project by raising new equity. Since the bene?t of fraud derives
from the e?ect of ?nancial misreporting on the short-term market valuation of the ?rm’s outside security, the insight of the model will not change if the ?rm can use debt ?nancing. In the debt context, there is still an external ?nancing cost, which is the interest rate the ?rm pays.
24
compensation and insider trading pro?t. This paper, however, focuses on ?rms’ ?nancing and investment incentives in the presence of fraud. Misreporting also comes with a cost: the expected litigation liability. Both the fraud penalty and the probability of detection are functions of ? = y (e) ? e. The cost-bene?t tradeo? leads to the following maximization problem for the manager at time 1. max ? = ?x[1 ? ? (y )]E (V |I, e) + (1 ? ?x)E (V |N, e) ? P (? )f ?,
? ?0
(4.20)
where P (? ) = ?xPI + (1 ? ?x)PN . The expected value to long-term shareholders is their expected ?nal cash ?ow net of the litigation cost. The solution to this problem, ? ? , is the optimal amount of misreporting.
4.3 Securities Fraud and Investment Incentives
In order to solve the manager’s optimization problem in equation (4.20), I need to derive the manager’s investment incentive x in the presence of fraud. Recall that x is the probability that the manager will undertake a newly arrived investment project at time 2. Section 4.3.1 derives the ?rm’s investment incentive at time 2, given its disclosure strategy at time 1. Section 4.3.2 presents a numerical illustration.
4.3.1 Investment Distortions
Suppose that a new investment opportunity does arrive at time 2. The manager privately observes the gross return to the new project as r. If the ?rm issues new equity and invests, the market value of the ?rm’s equity will be E (V |I, y ) = E (A|I, y ) + IE (R|I ). (4.21)
The true value of the ?rm is, however, E (A|e) + Ir. In order to invest, the ?rm needs to issue a fraction ? = I/E (V |I, y ) of new equity. The ?rm also faces the potential litigation liability PI f ? , if ? = 0. Then the expected ?nal payo? to the existing shareholders is (1 ? ? )[E (A|e) + Ir] ? PI f ? if the ?rm invests, or E (A|e) ? PN f ? if the ?rm does not issue and invest. Therefore, for the ?rm 25
to issue and invest, we need (1 ? ? )[E (A|e) + Ir] ? PI f ? > E (A|e) ? PN f ?. (4.22)
A cuto? investment pro?tability rc can be derived such that the above condition is satis?ed when r > rc . In other words, the ?rm will invest if and only if the return to the new investment exceeds some threshold level rc . rc = 1 means that the ?rm will strictly follow the positive NPV rule when making new investment. rc > 1 implies that the ?rm tends to underinvest in the sense that it will pass up some positive NPV projects. rc < 1 implies that the ?rm tends to overinvest in the sense that it will undertake some negative NPV projects. Therefore, the manager’s investment incentive is re?ected in his choice of the cuto? pro?tability to new investments. The model results about the manager’s investment decision are presented in the following propositions. Detailed proofs are provided in the appendix. Proposition 1 Financial misreporting can a?ect the ?rm’s investment incentives. Speci?cally,
? ? the ?rm’s cuto? pro?tability to new investments (rc ) depends on its magnitude of fraud (? ). rc (? )
is the solution to the following equation. rc = where zc = (rc ? R)/?R , and m(zc ) = ?(zc )/[1 ? ?(zc )]. Given the manager’s misreporting strategy ? at time 1, the probability that the ?rm will undertake a newly arrived investment opportunity at time 2 is
? ? x = P rob.[r > rc (? )] = 1 ? ?[zc (? )].
E (A|e) E (A|I, y ) + I?R m(zc )
?
(PN ? PI )f ? , (1 ? ? )I
(4.23)
(4.24)
The lower the cuto? investment pro?tability, the more likely is the ?rm to exercise its growth option at time 2. 26
Proposition 2 Making a new investment can decrease the ?rm’s probability of being investigated at time 3 if the ?rm can boost its market value by overstating its earnings, and either the correlation between the cash ?ow from the new investment and that from the existing assets is in a neighborhood around zero or the cash ?ow from the new investment is volatile enough and the correlation is in some certain range. Speci?cally, PI < PN if E [V |y (e)] > E [V |e] when ? > 0 and one of the following conditions is satis?ed: (1) ? ? [? , ], where is an arbitrary small positive number;
?e (2) max(?1, ? qI? ) < ? < ? ? 1, and I?R > I?R . R
Proposition 3 If the ?rm can boost its market value by overstating its earnings, then the ?rm has
? an incentive to overinvest. That is, if E [V |y (e)] > E (V |e) when ? > 0, then rc < 1. The larger
the magnitude of fraud, the lower is the fraudulent ?rm’s threshold return to new investments,
? ?rc < 0. ??
(4.25)
The essential message in these propositions is that ?nancial misreporting can in?uence the ?rm’s investment incentives in two ways. First, misreporting can in?uence the short-term ?rm value and thus the ?rm’s short-term external ?nancing cost. This e?ect is re?ected in the ?rst term on the right-hand side of equation (4.23). If a low-earnings ?rm overstates its earnings (y (e) > e) to pool with a high-earnings ?rm, and if the market cannot fully see through this, then we have E (A|y ) > E (A|e) for the low-earnings and dishonest ?rm. This implies that the market on average will overvalue the equity of the fraudulent ?rm. This overvaluation lowers the ?rm’s external ?nancing cost and thus gives the ?rm a larger incentive to raise money and invest, resulting in overinvestment. The high-earnings ?rm, however, will su?er from some market undervaluation due to the cross-subsidization between the good ?rm and the fraudulent ?rm. The good ?rm cannot ?nance the new investment on reasonable terms and therefore has less incentive to issue and invest. This is consistent with the underinvestment argument in Myers and Majluf (1984). In sum, the fraud-induced market mispricing implies that a fraudulent ?rm tends to overinvest, and a good and honest ?rm tends to underinvest.
27
Second, ?nancial misreporting can also a?ect the ?rm’s investment decision through the in?uence of investment on the ?rm’s litigation risk. The second term on the right-hand side of equation (4.23) represents the change in the expected litigation cost per investment dollar if the ?rm invests rather than not. If this change is negative, then the reduction in litigation risk will
? push the fraudulent ?rm’s pro?tability threshold rc further down below 1. This means that the
potential negative e?ect of making a new investment on the ?rm’s litigation risk will exacerbate the investment distortion. Given any ? > 0, Proposition 2 states that PI < PN if the investment is uncorrelated with the ?rm’s existing assets or if the investment is risky enough. The basic intuition is as follows. The market observes the combined cash ?ow from the ?rm’s assets in place and from the new investment, and draw inference about the magnitude of misreporting on the asset value based on the total cash ?ow. On one hand, given the level of cash ?ow volatility of the new investment, the inference problem will be most di?cult for the market when the cash ?ow correlation between the new investment and the existing assets is low around zero. On the other hand, given the level of correlation, high cash ?ow volatility from the new investment will decrease the valuation precision of the ?rm’s total cash ?ow and make it harder for the outsiders to see through fraud. Therefore, the incentive to disguise fraud will induce the fraudulent manager to overinvest in risky (high cash ?ow volatility) and uncorrelated projects. In the following analysis, I will focus on the case in which PI < PN . In sum, the key insight in Propositions 1 to 3 is that securities fraud can lead to real value losses. The distorted investment incentive can arise both from the fraud-induced market misvaluation of the ?rm’s assets (E [A|y (e)] = E [A|e]) and from the e?ect of investment on the ?rm’s litigation risk (PI = PN ). Securities fraud can lead to underinvestment by good and honest ?rms and overinvestment by fraudulent ?rms.
4.3.2 A Numerical Illustration
This section presents a numerical example to illustrate the relationship between fraud and the ?rm’s investment incentives. Two levels of earnings realization are considered: eL and eH ,
28
eL < eH . Based on the ?rm’s true earnings realization (e) and its reported earnings (y ), I label the ?rm as one of the following three types. LH ?rm: low earnings (e = eL ) are honestly revealed (y = eL ); HH ?rm: high earnings (e = eH ) are honestly revealed (y = eH ); LD ?rm: low earnings (e = eL ) are reported as high earnings (y = eH ).
? Table 3 presents each type of ?rm’s cuto? return to new investments rc and probability of
making a new investment x (in parentheses). The numerical example reveals the following patterns with respect to the ?rm’s investment incentives in the presence of securities fraud.
? ? (1) The HH ?rm tends to underinvest (rc > 1), and the LD ?rm tends to overinvest (rc < 1).
Put di?erently, the LD ?rm is more likely to exercise its growth option than the HH ?rm and the LH ?rm. These distortions emerge in all three panels. (2) Holding other parameters constant, an increase in the magnitude of misreporting (? ) worsens both the underinvestment problem of the HH ?rm and the overinvestment problem of the LD ?rm (as shown in panel A). This clearly demonstrates the investment distortion spillover between fraudulent and honest ?rms. (3) Holding other parameters constant, an increase in the volatility of the investment return (I?R ) helps to mitigate the underinvestment problem of the HH ?rm but exacerbates the overinvestment problem of the LD ?rm (as shown in panel B). This is because higher investment volatility is associated with higher value of the ?rm’s growth option, which to some extent lessens the market undervaluation of the HH ?rm but worsens the overvaluation of the LD ?rm.9 Furthermore, according to Proposition 2, large I?R also strengthens the negative e?ect of investment on the ?rm’s litigation risk, which motivates the fraudulent ?rm to overinvest. (4) Holding other parameters constant, larger asset volatility (?A ) exacerbates both the underinvestment problem of the HH ?rm and the overinvestment problem of the LD ?rm (as shown
9 The
market’s expected NPV of the new project is I [E (R|I ) ? 1] = I?R m(zc ). Since m(zc ) > 0, a large I?R
scales up the market value of the new project.
29
in panel C). The intuition is that high volatility of the asset value implies less valuation precision of the ?rm’s cash ?ows, which can not only worsen the misvaluation of the ?rm’s assets in place but also decrease the litigation risk of the fraudulent ?rm. (5) Even the LH ?rm tends to overinvest slightly, but this distortion has nothing to do with securities fraud. It arises solely from the e?ect of asymmetric information about the investment return, as shown in Myers and Majluf (1984).10 What is important is the di?erence
? of the LH ?rm and of the LD ?rm, because this di?erence measures the e?ect between the rc
of misreporting on the investment incentive of a low-earnings ?rm. In sum, the numerical illustrations demonstrate that ?nancial misreporting can distort investment decisions in both fraudulent and honest ?rms. The degree of distortion depends on the magnitude of fraud as well as the characteristics of the ?rm’s assets and growth options.
4.4 Disclosure Strategy
? Section 4.3 shows that the manager’s investment incentive (rc or x) can be in?uenced by
?nancial misreporting (? ). Now I move back to time 1 and examine the manager’s disclosure strategy y (e), taking into consideration her investment incentives at time 2. At time 1, the manager privately observes the earnings (e) generated from the ?rm’s assets and makes an earnings announcement y (e) = e + ? (e). That is, given any earnings realization e, the manager optimally chooses the amount of misstatement ? such that the expected value to long-term shareholders at time 3 is maximized. The manager’s objective function is speci?ed in equation (4.20) in section 4.2.2. Now I substitute equation (4.24) into (4.20) and rewrite the manager’s maximization problem as follows.
? ? max ? = ?[1 ? ?(zc )][1 ? ? (y )]E (V |I, e) + {1 ? ?[1 ? ?(zc )]}E (V |N, e) ? P (? )f ?, ? ?0
10 For
(4.26)
? = 1 if and only if E (R|I ) = 1, that is, the market believes new investments are on average the LH ?rm, rc
zero NPV projects. If the market has bullish expectations and believes new projects on average have strictly positive NPV (E (R|I ) > 1), then the LH ?rm will have an incentive to overinvest.
30
? ? where zc ? [rc (? ) ? R]/?R . In sum, misreporting a?ects the manager’s objective function in three
ways. First, it can directly a?ect the short-term market valuation of the ?rm V2 (I, y ) and thus its external ?nancing cost ? (y ). Second, it can indirectly in?uence the long-term performance of the
? ?rm V through the endogenous investment decision rc (? ). Third, misreporting brings a potential
litigation liability P (? )f ? . The optimal strategy balances the bene?t of misreporting and the cost of it. Section 4.4.1 describes a perfect Bayesian equilibrium disclosure strategy y ? (e) = e + ? ? (e). Section 5.3.2 analyzes some important properties of the ?rm’s fraud propensity and the fraud magnitude.
4.4.1 Equilibrium Misreporting
I adopt the perfect Bayesian equilibrium concept to study the manager’s equilibrium misreporting strategy. A perfect Bayesian equilibrium has two requirements. First, the market forms expectations on the ?rm value [E (V |y )] using Bayes’s rule whenever possible. Second, given the market’s beliefs, the manager’s disclosure strategy y (e) maximizes her objective function in (4.27). Proposition 4 An equilibrium disclosure strategy involves partitioning the earnings space into fraud region(s) and nonfraud region(s). Speci?cally, there are three cuto? earnings realizations ?? < el < ec < eh < +?, and the manager’s earnings disclosure strategy is as follows. y ? (e) = e, if e ? ec ,
? y ? (e) = e + ?1 (e) > ec , if el ? e < ec ,
y ? (e) = e, if e < el . Let e denote the earnings value the market infers from y (e). Then the market value of the ?rm’s assets in place after the earnings announcement is ˜|e , e = y ), if y > eh , V1 (y ) = E (A ˜|e , e = y ) + ?1 E [A ˜|e , e = y ?1 (e)], if ec ? y ? eh , V1 (y ) = (1 ? ?1 )E (A 1 ˜|e , e = y ?1 (e)], if el ? y < ec , V1 (y ) = E [A 2 ˜|e , e = y ), if y < el , V1 (y ) = E (A 31
?1 ?1 where ?1 ? P rob.(misreporting |ec ? y ? eh ), y1 (e) = y (e) ? ? 1 (e), and y2 (e) = y (e) ? ? 2 (e). ? 1
and ? 2 are the market’s expected amount of misreporting when ec ? y ? eh and when el ? y < ec , respectively. Detailed proof of this proposition is provided in the appendix. Here I discuss the implications. Proposition 4 implies that the manager will honestly reveal intermediate earnings when the true earnings realization is very good or desperately bad. The manager has an incentive to overstate earnings when the earnings realization is mediocre or fairly disappointing. The intuition is as follows. When the ?rm is in good shape (e > ec ), the manager does not need to overreport earnings at the cost of incurring future litigation liability. When the ?rm is in a shaky condition (el ? e < ec ) but faces some possible future growth opportunities, the manager will rationally want to take the chance and dress up short-term ?rm appearance so that the future growth options can be exercised on favorable terms (i.e., a lower external ?nancing cost). When intermediate earnings happen to be stunningly bad (e < el ), however, moderate overreporting of earnings will not change the picture much. In this case, in order to mimic a high-earnings ?rm, the low-earnings ?rm has to engage in substantial overstatement of earnings, which implies a large potential litigation cost. When the expected cost of fraud exceeds the bene?t, the manager and the shareholders are better o? by honestly revealing the earnings. Proposition 4 shows that outside investors will rationally discount the ?rm’s earnings announcement if el ? y ? eh . When ec ? y ? eh , the fraudulent ?rm pools with high-earnings ?rms. The market value of the ?rm’s assets re?ects a weighted average of the two types. When el ? y < ec , the market fully discounts the reported earnings because the ?rm has an incentive to overreport when its true earnings realization is in this region. Proposition 4 implies, however, that ˜|e , e = y ?1 (e)] if el ? y < ec is el ? y < ec will not be observed in equilibrium. So V1 (y ) = E [A 2 an o?-equilibrium speci?cation.
32
4.4.2 Fraud Propensity
Given any cuto? value ec and el , the ?rm’s probability of misreporting is simply P rob.(f raud) = P rob.(el ? e < ec ). The combination of a high ec and a low el implies a high fraud propensity. Di?erent ?rms can have di?erent cuto? values and thus di?erent likelihoods of misreporting. The fraud region as well as the magnitude of misreporting depend on the structural parameters in the model. The following proposition presents some comparative-static results for ? ? and P rob.(f raud) with respect to some important bene?t and cost parameters. Proof is provided in the appendix. Proposition 5 The ?rm’s fraud propensity and the magnitude of fraud are related to its pro?tability, growth potential, and quality of corporate governance. Speci?cally,
? /?e < 0; (1) ??1
? (2) If PI < PN , then ??1 /?? > 0, and ?P rob.(f raud)/?? > 0;
? (3) ??1 /?p < 0,
?P rob.(f raud)/?p < 0;
The ?rst result states that in the fraud region, the magnitude of misreporting increases as earnings realization decreases. This is because a low-earnings ?rm has high marginal bene?t from misreporting,
?2? ???e
< 0.
The second result shows that if exercising a growth opportunity can decrease the probability of fraud detection, then both the ?rm’s fraud propensity and the amount of misreporting increase in its growth potential (?). In this model, growth can a?ect both the bene?t and cost of engaging in fraud. First, for a rapidly growing but cash-poor ?rm, misreporting business prospects and conditions can create a short-term bene?t by enabling the ?rm to raise external capital on favorable terms to support its growth. Second, growth can decrease the ?rm’s litigation risk, if it can decrease the valuation precision of the ?rm’s cash ?ows,
?P ??
= ?x(PN ? PI ) < 0.
The last result relates the ?rm’s fraud propensity to the quality of corporate governance. Good corporate governance implies more e?ective monitoring of management and thus a better 33
chance that any fraudulent activities within the ?rm will be discovered,
?P ?p
> 0.11
4.5 Model Implications and Discussion
The cost-bene?t analysis of securities fraud provides testable implications for (1) the relation between fraud and investment incentives and (2) the economic determinants of the cross-sectional di?erences in ?rms’ fraud propensities. Fraud and Ine?cient Investment This theory predicts that fraudulent ?rms tend to over-
invest. Yet, the investment can be ine?cient and can lead to serious value destructions. The telecommunications industry is a good illustration. Sidak (2003) o?ers evidence that the prevailing ?nancial misrepresentations in this industry during the past 7 years (particularly by WorldCom) have led to excessive investment and overbuilding. The Eastern Management Group estimates that a signi?cant percentage of the $90 billion invested in that industry was misallocated because of fraudulent growth projections.12 Moeller, Schlingemann, and Stulz (2004) document that in the recent merger wave (1998-2001), acquiring ?rms lost a total of $240 billion surrounding the announcement of acquisitions, and the acquisitions resulted in a net synergy loss of $134 billion (compared to a net synergy gain of $11.5 billion in the 1980s). This implies that the market did not see those investments as value-increasing. Interestingly, Wang (2004b) shows that this period appeared to be fraud-prevailing. Jensen (2004) also provides some good examples of bad investments and value destruction in fraudulent ?rms such as Nortel Networks and eToy. The theory argues that part of the overinvestment incentives arise from the negative e?ect of investment on the ?rm’s detection risk. The model predicts that the type of investment that produces the most valuation imprecision will have the strongest e?ect on detection likelihood. The theory also implies there is investment distortion spillover between fraudulent and honest ?rms. Overinvestment by fraudulent ?rms can crowd out investment by good and honest
11 This
study mainly focuses on the monitoring role of corporate governance, and does not incorporate the broader
functions of governance such as designing executive compensation structures.
12 Eastern
Management Group, supra note 42, at 2 (quoting Joelle Tessler, “WorldCom Spine UUNET is Critical
Part of Internet,” San Jose Mercury News, September 1, 2002).
34
?rms. This implies that fraud-induced real value losses are borne not only by shareholders of fraudulent ?rms but also by those of ?rms that have no intention to misreport. Fraud Propensity and Firm Attributes The theory shows that ?rm characteristics can
in?uence the ?rm’s likelihood of engaging in fraud. Speci?cally, fraudulent ?rms tend to be those who have good growth prospects and large external ?nancing needs, but experience negative profitability shocks. Growth itself is not a bad thing, but this model shows that it can have a signi?cant e?ect on the manager’s fraud incentives (both on the bene?t and cost of fraud). The model predictions are consistent with many ?ndings in the accounting literature on earnings management and corporate fraud. Loebbecke, Eining, and Willingham (1989) study a small sample of managerial frauds and conclude that the most signi?cant “red ?ags” for fraud are rapid company growth and poor accounting performance. The National Commission on Fraudulent Financial Reporting (1987) states that young public companies have a proportionately greater risk of ?nancial statement fraud. Young ?rms generally have higher growth potential than mature ?rms. Litigation Across Industries The model predicts an industry e?ect in the cross-sectional
distribution of securities fraud. That is, there will be “litigation clustering” in certain industries during a speci?c time period. This is because both ?rms’ bene?t from fraud (such as asset profitability and growth potential) and litigation risk are correlated within an industry, which implies that ?rms’ fraud propensities will be in?uenced by industry factors. E?ect of Increasing Disclosure The model shows that increasing the informativeness of
the earnings has an ambiguous e?ect on the ?rm’s likelihood of committing fraud. This implies that imposing heavy disclosure requirements on public ?rms may not produce the expected e?ects. The reason is that increased disclosure could give the market an illusion of increased transparency, which could actually decrease market vigilance. Fraud Detection Likelihood This theory shows that while the fraud penalty (f ) is largely
determined by securities laws and regulations, fraud detection likelihood (P ) is substantially in?uenced by the ?rm’s endogenous actions as well as ?rm-speci?c attributes. This implies that the probability of detection is more important than the penalty in determining cross-sectional
35
di?erences in ?rms’ fraud propensities. The policy implication is that raising litigation liability standards alone will achieve only limited deterrence, because ?rms may adjust P to o?set some e?ect of increased f on their expected litigation cost. More important, the theory shows that ?rms may even destroy value in order to decrease their detection risk, which can be an unintended consequence of imposing heavy penalty. Ine?cient investment is one example. Leuz, Triantis and Wang (2004) provide possibly another. They document that since the passage of Sarbanes-Oxley Act there has been a dramatic surge in the number of public ?rms that voluntarily deregistered their common stock and ceased to ?le regular reports with the SEC (they call this “going dark” transactions). They also document substantial negative abnormal returns and loss of liquidity associated with deregistration and continued drop in the ?rms’ market capitalization after deregistration. Their ?ndings imply that insiders of those companies may have sacri?ced shareholders’ interest in order to hide from market scrutiny. Internal Corporate Governance and Extensions This paper shows that even when the
manager’s interest is perfectly aligned with that of shareholders, fraudulent behavior can still emerge, because incumbent shareholders may ?nd it advantageous to defraud prospective investors. Good corporate governance will not completely prevent fraud if it is under the control of existing shareholders. In fact, Table 2 shows that the likelihood of fraud detection is much lower from within the ?rm than from outside. Therefore, enhancing other detection forces such as capital market vigilance, responsibility of “gatekeepers” (e.g., auditors and lawyers) and securities regulation is necessary in combating corporate fraud. In the present model, the quality of internal corporate governance p is exogenously determined, and I focus on detection by capital markets. A more general model can allow shareholders of the company to choose the level of p, and allow the market to incorporate this information into its belief about the likelihood of fraud (?0 = g (p), g (p) < 0). Therefore, a higher p corresponds to a higher ex ante bene?t from fraud because it leads to a lower ?0 and thus a smaller discount of the ?rm’s earnings report (the signalling e?ect). As illustrated by Figure 2, however, a larger di?erence between E (V |y ) and E (V |e) also implies a higher likelihood of cash ?ow detection. This
36
means that a higher p will increase the likelihood of both internal and external fraud detection (the litigation e?ect). The optimal quality of internal corporate governance p? balances the signalling e?ect with the litigation e?ect. Since in this paper the manager represents the interests of incumbent long-term shareholders, the extension is equivalent to having a model in which the manager chooses ? and p at the same time (i.e., time 0 and time 1 are combined). The manager’s optimization problem can be as follows. max
? ? = E (V |N, e) + ?[1 ? ?(zc )][?0 ? ? (y, p)]E (V |I, e) ? P (?, p)f ? ? h(p),
? ?0,0?p?1
(4.27)
where h(p) is the cost of building the quality of internal corporate governance. p? depends on the functional form of g (p) and h(p). For example, if the market is not sensitive to corporate governance (at least for some range of p realizations), then the ?rm will choose a p as low as possible, regardless of its fraud propensity. If the market values good corporate governance but it is very costly to build up the quality, then the ?rm may still lean towards a low p. If the market values good governance and the cost of establishing good governance is reasonable, then the choice of p will depend on the ?rm’s ex ante fraud incentives.
37
Table 4.1: Fraud Discovery (1996-1997) This table lists the various corporate events or entities that precipitated the 187 federal securities class action lawsuits during 1996 and 1997. The litigation information is retrieved from Stanford Securities Class Action Clearinghouse. Information about the triggering events of each lawsuit is extracted from the relevant case documents (i.e., the case complaints, the press releases, and the court decisions). The ?rst column of the table lists the event or entity that precipitated or initiated the securities lawsuits. The triggering events can overlap in some lawsuits. Precipitator Number of observation Devastating news announcement Regulators (mostly SEC) Independent auditors Business journal articles Board/internal investigation Securities analysts Shareholder/Investor Stock Exchanges/credit rating services Management turnover 1996 93 63 6 10 7 7 1 3 0 2 1997 94 69 6 7 5 4 3 4 1 1 Total 187 132 12 17 12 11 4 7 1 3 % of Total 70.59 6.42 9.09 6.42 5.88 2.14 3.74 0.53 1.60
38
Table 4.2: A Numerical Illustration of Investment Incentives I assume the following parameter values. The value of the ?rm’s assets in place is normally distributed with expectation A = 100 and volatility ?A = 30. The average return on assets is q = 0.16. The earnings noise u is normally distributed with zero mean and volatility ?u = 4. 2 + ? 2 = 6.25. The size of The expected earnings is e = qA = 16, and volatility is ?e = q 2 ?A u the new investment is I = 25. The volatility of investment return is I?R = 25 ? 0.3 = 7.5. The correlation coe?cient between R and e is ? = 0.3. The market’s prior belief about the probability of misreporting is ?0 = 0.5. The marginal fraud penalty is f = 1.5. The institutional e?ciency is p = 0.3. The cost of investigation is C = E (f ? ) = f ? . In panel A, I set eL = e ? ?e = 9.75. I consider two levels of eH . First, eH = e = 16, which means that ? = 6.25. Second, eH = e + ?e = 22.25, which means that ? = 12.5. In panels B-C, ? = 12.5. Panel A: Fraud Magnitude and Investment Bias ? = 6.25 0.83 (71%) 0.94 (58%) 1.05 (43%) ? = 12.5 0.73 (81%) 0.94 (58 %) 1.14 (33%)
LD LH HH
Panel B: Investment Volatility and Investment Bias I?R LD LH HH 2.5 0.76 (79%) 0.98 (53%) 1.19 (26%) 7.5 0.73 (81%) 0.94 (58%) 1.14 (33%) 12.5 0.71 (83%) 0.91 (62%) 1.09 (38%) 17.5 0.69 (85%) 0.89 (65%) 1.06 (42%)
Panel C: Asset Volatility and Investment Bias ?A 1.14 0.74 0.94 = 30 (31%) (81%) (58%) ?A 1.17 0.69 0.94 = 40 (28%) (85%) (58%)
HH LD LH
39
Figure 1: Model Time Line
0 Quality of governance 0? p? 1 is set, which later determines the prob. of internal fraud detection.
1 The manager privately observes the intermediate earnings e from existing assets A. The manager makes a disclosure decision y(e) =e + ?.
2 A new investment opportunity comes with probability ?, requires an initial cost of $I, and generates gross return R. The manager observes the realization of R as r, and makes the investment and financing decisions.
3 The firm generates a liquidating cash flow V. Misreporting is detected with prob. P. A penalty f? is imposed upon detection.
Figure 2: Probability of Fraud Detection
True
Report
vc
E (V | e)
E (V | y )
Note: In this figure, the shaded area represents the probability of cash flow detection.
40
Chapter 5 An Empirical Investigation of Securities Fraud
This chapter empirically investigates the economic determinants of ?rms’ fraud propensity and the fraud detection likelihood. More speci?cally, I address the following research questions: 1. How does investment in?uence the ?rm’s fraud incentives and their detection risk? 2. What are the roles of di?erent corporate monitors in the context of fraud? What type of corporate monitor has been e?ective in discovering corporate fraudulent activities? 3. What is the role of insider equity incentives in determining the ?rm’s propensity to commit fraud? 4. How are ?rm characteristics related to the ?rm’s likelihood of engaging fraud and the ?rms’ likelihood of getting caught? The structure of this chapter is as follows. Section 5.1 describes the accounting fraud sample and presents some stylized facts about accounting-related class action lawsuits from 1996 to 2003. Section 5.2 presents the empirical model of fraud. Section 5.3 discusses the related literature and develops the empirical hypotheses. Section 5.4 reports the results from univariate comparisons between the fraud sample and the comparison sample. Section 5.5 reports the multivariate analysis on the determinants of ?rms’ propensity to commit fraud and the likelihood of fraud detection. Section 5.6 presents robust checks on the model results regarding the possibility of false detection, the timing of fraud, and di?erent model speci?cations.
5.1 Fraud Sample
The fraud sample in this study is based on Securities Class Action Clearinghouse (SCAC) established by Stanford Law School. This clearinghouse provides a comprehensive database of
41
federal private securities class action lawsuits ?led since 1996 in the United States. A private securities class action is a case brought pursuant to Federal Rule of Civil Procedure 23 on behalf of a group of persons (class members) who purchased the securities of a particular company during a speci?ed time (class period). A suit is ?led as a class action because the members of the class are so numerous that joinder of all members is impracticable. I went through the details of all the available case documents associated with each lawsuit (e.g., case complaints, press releases, court decisions, etc.) to identify the nature of fraud allegations. As a result, I singled out 684 lawsuits ?led against 660 US public companies during 1996 to 2003 involving allegations of accounting irregularities. For ?rms that had multiple securities lawsuits, I only use the earliest one in the analysis. Existing studies mostly rely on the SEC’s Accounting and Auditing Enforcement Releases (AAERs) to identify accounting frauds. Several recent studies use accounting restatements to proxy for fraudulent ?nancial reporting (Agrawal and Chadha (2004), Burn and Kedia (2004), Efendi, Srivastava and Swanson (2004)). This paper is the ?rst to study class action litigation involving accounting-related allegations. Private class action litigation has long been an important concomitant to the enforcement of securities laws (Cox and Thomas (2003)). The volume of class action lawsuits is also comparable to that of the SEC’s enforcement actions. More important, class action litigation can provide new insights for understanding market forces in securities litigation, because class action suits generally involve the interests of thousands of investors, and key plainti? investors play an important role in the litigation. My class action sample does overlap with the SEC’s AAER sample and the accounting restatement sample that have been used in the existing studies. Among the 660 fraudulent ?rms in my sample, 207 ?rms were subject to parallel SEC’s AAERs, and 334 ?rms had accounting restatements according to the General Accounting O?ce’s October 2003 report.1 In Section 5.6, I will use these two subsamples to check the robustness of my results across di?erent proxies of
1 The
General Accounting O?ce’s October 2003 report lists all the accounting restatements between January
1997 and June 2003. My sample period is from January 1996 to December 2003. Therefore, there can potentially be more than 334 restatements in my sample.
42
securities fraud. The following sections provides detailed descriptive information about the fraud sample.
5.1.1 Time Trends and Firm Characteristics
Table 1 describes the evolution of class action litigation over time. Panel A shows that accounting-related frauds have on average accounted for about 47% of the total litigation activities (excluding litigation against investment banks) over the past 8 years. The number of accounting frauds peaked in 2002, where it represented 56.13% of all the securities class action ?lings. Interestingly, accounting-related litigation substantially decreased in 2003 (only 40% of all lawsuits), which may have resulted from tightened securities regulation and increased market vigilance. Panel B shows the distribution of the class periods associated with the 660 lawsuits. Every class action lawsuit speci?es a class period. The beginning of a class period shows the earliest time a fraud a?ects the market, based on the judgment of securities attorneys. A class period generally ends at the time of some major events that precipitate the litigation. The length of the class period provides some information about the duration of fraud. The average length of the class period is a little more than one year, but there is substantial variation. Some frauds a?ected the market for more than ?ve years, while some less than a quarter. Panel B also shows that ?rms in the fraud sample were largely young public companies. The median age was only 3.45 years, and more than 60% of the sample ?rms were less than 5 years old. About 64% of the alleged fraudulent ?rms were listed on NASDAQ when fraud began. Panel C shows the distribution of the ?scal year in which fraud began. I label the beginning ?scal year of fraud as year 0. I determine year 0 based on the beginning of the class periods and the ?rms’ ?scal year end. The beginning of a class period indicates the earliest time fraud a?ected the market, but does not necessarily indicate the beginning of fraud. In general, an accounting fraud starts to a?ect the market when a fraudulent ?nancial report is released to the public. Given that there is about one month’s lag for quarterly reports and a two-to-three-month lag for annual reports, year 0 can be the same ?scal year in which the class period starts, or the previous ?scal
43
year. Figure 1 illustrates the two scenarios. If a ?rm was subject to both private litigation and the SEC’s enforcement action, I cross check the beginning year with that speci?ed by the SEC. The information about the SEC’s enforcement actions is retrieved from the SEC’s litigation archive.
5.1.2 Industry Distribution
Table 2 presents the industry distribution of fraud. I classify the alleged fraudulent ?rms into 24 industry categories. The primary classi?cation is based on two-digit SIC codes, but in some instances, I use three-digit SIC codes, as this is more informative about the types of companies that engaged in fraud. Table 2 shows evidence of signi?cant industry patterns in securities fraud litigation. First, technology ?rms are disproportionately more involved in accounting-related securities litigation. In particular, ?rms in software and programming alone accounted for 17.42% of all accounting fraud cases in the past 8 years. Electronic parts, computer manufacturing, and telecommunications companies represent another 19% of the litigation activities. Second, the service sector and particularly the ?nancial services and the business services industries also show a high litigation concentration. In total, the technology (including bio-technology ?rms) and service sectors account for 67% of all securities lawsuits studied in this paper.
5.1.3 The Nature of Fraud
Table 3 lists some speci?c accounting items that are often manipulated, based on the relevant case documents in 563 class action lawsuits.2 Allegations of improper revenue recognition are most common, accounting for 67.44% of all the accounting fraud allegations. Operational expenses are also likely to be manipulated by managers to reach desired earnings targets (17.26% of the 563 cases). As for the balance sheet items, misstatements of assets are more frequently observed than misstatements of liabilities and equity. Among the di?erent types of assets, accounts receivable and inventory seem to be frequently misstated. This observation is consistent with the ?ndings in Chan et al. (2005) that changes in inventory and accounts receivables are closely related to
2I
am only able to clearly identify the speci?c accounting items in 563 out of 660 cases.
44
the earnings quality and thus can help to predict future stock returns. Finally, understatement of reserves and allowances is also fairly often, accounting for about 9% of the 563 lawsuits.
5.2 Empirical Methodology 5.2.1 A Model with Partial Observability of Fraud
In implementing comparisons between the fraud sample and any sample of non-convicted ?rms, we face an identi?cation problem because we only observe detected fraud. That is, we only observe frauds that have been committed and subsequently detected. Firms that have not been sued in securities litigation are either innocent ?rms or undetected fraudulent ?rms (see Figure 2 for an illustration). This implies that the probability of detected fraud (what we observe) is di?erent from the probability of fraud (what we are interested to estimate but cannot observe), unless detection is perfect. To address this identi?cation problem, I use a bivariate probit model with partial observability as discussed in Poirier (1980) and Feinstein (1990). In essence, this technique models the observed outcome (detected fraud) as a function of the joint realizations of two latent processes.
? Let Fi? denote ?rm i’s potential to commit fraud, and Di denote the ?rm’s potential of getting
caught conditional on fraud being committed. Then consider the following reduced form model: Fi?
? Di
= =
xF,i ?F + ui ; xD,i ?D + vi ,
(5.1) (5.2)
where xF,i contains variables that help explain ?rm i’s potential to commit fraud, and xD,i contains variables that help explain the ?rm’s detection risk. ui and vi are zero-mean disturbance terms, and follows a bivariate normal distribution. Their variances have been normalized to equal unity. The correlation between ui and vi is ?. Now I de?ne the following binary variables. Fraud occurrence: Fi = 1 if Fi? > 0, and Fi = 0 if otherwise;
? Fraud detection : Di = 1 if Di > 0, and Di = 0 if otherwise.
45
We, however, do not directly observe the realizations of Fi and Di . What we observe is Zi = Fi Di Zi = 1 if ?rm i has committed fraud and has been detected, and Zi = 0 if ?rm i has not committed fraud or has committed fraud but has not been detected. Let ? denote the bivariate standard normal cumulative distribution function. The empirical model for Zi is P (Zi = 1) = P (Fi Di = 1) (5.3)
= P (Fi = 1, Di = 1)
? = P (Fi? > 0, Di > 0)
= ?(xF,i ?F , xD,i ?D , ?); P (Zi = 0) = P (Fi Di = 0) (5.4)
= P (Fi = 0, Di = 0) + P (Fi = 1, Di = 0) = 1 ? ?(xF,i ?F , xD,i ?D , ?). An implicit assumption in the model is that false detection of fraud is not allowed for
? (P (Fi = 0, Di = 1) = 0), because the process of Di is only de?ned conditional on Fi = 1.
Extension of the above model to statistically control for false detection is possible, but it tends to complicate the estimation.3 I will come back to the issue of false detection in Section 5.6.
? Although I de?ne Di conditional on Fi = 1, the correlation between the two disturbance
terms ? may not necessarily be zero. As discussed in Feinstein (1990), a non-zero correlation may arise for a number of reasons, particularly when the potential fraud-doer and the detection force possess information about one another.
3 Let
Di = 1 indicate false detection. Then P (Zi = 1) = P (Fi = 1, Di = 1) + P (Fi = 0, Di = 1); P (Zi = 0) = P (Fi = 0, Di = 0) + P (Fi = 1, Di = 0) ? P (Fi = 0, Di = 1).
In a well-functioning legal environment, P (Di = 1) should be very small, much smallers than P (Di = 1). Then assuming P (Di = 1) = 0 will not substantially bias the model estimation.
46
5.2.2 Model Identi?cation and Estimation
The partial observability of fraud raises a model identi?cation issue. This is because we only observe the joint outcome of two latent processes, and the decomposition between the two latent components may not be unique. According to Poirier (1980), the conditions for full identi?cation of the model parameters are (1) xF,i and xD,i do not contain exactly the same variables; and (2) the explanatory variables exhibit substantial variations in the sample. The above model can be estimated using the maximum-likelihood method. The log-likelihood function for the model is L(?F , ?D , ?) =
zi =1
log[P (zi = 1)] +
zi =0
log[P (zi = 0)]
(5.5)
=
i=1,...,n
{zi ln[?(xF,i ?F , xD,i ?D , ?) + (1 ? zi )ln[1 ? ?(xF,i ?F , xD,i ?D , ?)]}.
I use the ?ling of class action lawsuits to proxy for detected fraud (i.e., Z = 1). The partial observability model implies that the appropriate comparison sample (Z = 0) should be a random sample of non-litigated ?rms. I therefore use all the ?rms in the COMPUSTAT database that have not been subject to any private securities litigation (accounting-related or not) or the SEC’s AAERs between 1996 and 2003.
5.2.3 Comparison with Straight Probit Model
A straight probit model, which has been used in many existing studies on fraud, is as follows. For ?rms i=1,...,n, P (Di = 1) = 1; P (Zi = 1) = P (Fi = 1) = ?(xF,i ?F ). The log likelihood function associated with this model is L(?F ) =
i=1,...,n
{zi ln[?(xF,i ?F ) + (1 ? zi )ln[1 ? ?(xF,i ?F )]}.
(5.6)
We can see that as long as detection is not perfect (i.e., P (Di = 1) ? 1), the straight probit model will systematically understate the true probability of fraud. 47
An inference problem could also arise when (5.6) is estimated instead of (5.5). For example, we want to examine the marginal e?ect of an explanatory variable xi on the probability of fraud P (Fi = 1). Let us take partial derivative of xi on both sides of equation (5.3). ?P (Zi = 1) ?P (Fi = 1) ?P (Di = 1|Fi = 1) = P (Di = 1|Fi = 1) + P (Fi = 1). ?xi ?xi ?xi If this variable has opposite e?ects on P (Fi = 1) and P (Di = 1|Fi = 1), then
?P (Zi =1) ?xi ?P (Fi =1) ?xi
(5.7) and
can even have di?erent signs, not to mention that the magnitude will be di?erent. This
may lead us to draw incorrect inference about the role of xi . Section 5.5.6 provides concrete examples of the discussions here.
5.3 Hypothesis Development and Model Speci?cation
Following the framework of fraud in Wang (2004), a ?rm’s propensity to commit fraud depends on its expected bene?t and cost from engaging in fraud. The expected cost of fraud is the litigation risk: with some positive probability, fraudulent activities will be uncovered, resulting in a penalty. Wang (2004) argues that while the penalty (at least the explicit liability provision) is largely determined by securities laws and thus exogenous to the ?rm, the probability of detection depends on the ?rm’s endogenous actions (e.g., investment, disclosure) as well as ?rm-speci?c attributes. This implies that the detection risk is a more important determinant of the crosssectional variations in ?rms’ fraud propensities than are penalty provisions. Therefore, I focus on the likelihood of detection for the cost side of the tradeo?. A factor will positively in?uence a ?rm’s fraud propensity if it can increase the ?rm’s bene?t from committing fraud, or if it can decrease the ?rm’s expected probability of getting caught, or both. The structure of this section is as follows. Sections 5.3.1 and 5.3.2 discuss factors that can potentially a?ect a ?rm’s detection risk and its bene?t from fraud, respectively. Section ?? discusses the control variables. Section 5.3.4 summaries the model speci?cation.
48
5.3.1 Probability of Fraud Detection
The probability of fraud detection essentially determines how risky it is for a ?rm to engage in fraud. If a factor can signi?cantly in?uence such probability and if its e?ect can be anticipated at the time the ?rm makes the fraud decision, then this factor will in?uence the ?rm’s ex-ante propensity to commit fraud (in the opposite direction). Therefore, I start with the determinants of the fraud detection likelihood (i.e., xD ), and then move to the determinants of fraud propensity (i.e., xF ) in Section 5.3.2.
Investment
Wang (2004) argues that fraudulent ?rms tend to overinvest. The overinvestment incentive is twofold. First, fraud can create short-term market overvaluation of the ?rm and thus decrease the external ?nancing cost of investment. Second, after committing fraud, the fraudulent ?rm has incentive to cover things up. This incentive can motivate the management to strategically use investment to disguise fraud. Wang shows that investment with high uncertainty and/or low correlation with current activities can mask fraud better than others, because these types of investment can decrease the precision of the ?rm’s cash ?ows and create inference problems for the market. Wang’s argument has the following three testable implications: (1). Fraudulent ?rms have larger investment expenditures than comparable honest ?rms; (2). Di?erent types of investment have di?erential e?ects on a ?rm’s probability of being detected and the probability of fraud. Risky investments and uncorrelated investments have stronger negative e?ects on the detection likelihood than other types of investments; (3). Financing of the investment in?uences a ?rm’s probability of committing fraud. Externally-?nanced investment will motivate fraud better than internally-?nanced investment. To test the above implications, I investigate three types of investment: investment in research & development (R&D), capital expenditures, and mergers/acquisitions. These investments can substantially di?er in their e?ects on a ?rm’s valuation precision. Investment outcome of R&D
49
projects is generally highly uncertain. It is di?cult for the market to fully understand and correctly value its impact on the ?rm value. Capital expenditures tend to be more straightforward. COMPUSTAT de?nes capital expenditures as the funds used for additions to the company’s property, plant and equipment. Mergers and acquisitions, in theory, should fall in the middle, because the investment is to acquire an existing asset rather than to create something new. However, the true value of the acquired assets and the synergy between the acquired and the existing assets may not be correctly understood by the market or even the acquirer. I further distinguish between cash-based acquisitions and stock-based acquisitions. The earnings management literature has provide evidence that stock-based acquisitions are associated with higher incentive of earnings management (e.g., Erickson and Wang (1998)). In this study, I examine the e?ect of stock-based acquisitions on both the probability of fraud and the probability of fraud detection. I also distinguish between focused acquisitions and diversifying acquisitions. I de?ne focused acquisitions as acquisitions within the same two-digit SIC codes. According to Wang (2004), focused acquisitions should be associated with higher probability of detection than diversifying ones are.
Corporate Monitoring
E?ective monitoring over the management should increase the likelihood of fraud detection and deter fraud ex ante. In this study, I examine the roles of four types of corporate monitors in the context of corporate securities fraud: large shareholders, institutional owners, independent auditors, and board of directors.
Monitoring by Shareholders: A ?rm’s ownership structure is important in determining both the ?rm’s bene?t from committing fraud and its detection risk. This is because the ownership structure is crucially related to the incentive structure within the ?rm, including the incentive of the management to defraud outside investors and the incentive of shareholders to monitor the management and detect fraud. The monitoring role of large shareholders has received a great amount of attention in the 50
?nance and economics literature. Shleifer and Vishny (1997) argue that concentrated ownership is a key element of a good corporate governance system because large shareholders have high incentive and power to impose e?ective monitoring over the management. There has been quite some empirical evidence on the role of large shareholders in corporate governance (see a recent survey by Holderness (2003)). For example, Bethel, Liebeskind and Opler (1998) ?nd that company performance improves after an activist investor purchases a block of shares. Bertrand and Mullainathan (2001) ?nd that the presence of a large shareholder on the board is associated with tighter control over executive compensation. In the context of corporate fraud, it is also intuitive that large shareholders should go against fraudulent reporting, because they cannot cash out in a short period of time to catch the windfall from fraud, and they will likely su?er a lot from the severe consequences of fraud. Therefore, I expect a positive relation between block ownership holding and the likelihood of fraud detection. Large shareholders are often institutional investors. Monitoring by institutional shareholders has attracted growing public and academic interest, as institutional ownership skyrocketed over the past two decades in the United States. The Private Securities Litigation Reform Act (PSLRA), which was passed in December 1995, explicitly encourages more active participation of institutional investors in securities litigation by requiring each class action lawsuit to specify a lead plainti?. William Lerach, a partner in Milberg Weiss Bershad Hynes & Lerach LLP and a leader in representing investors in securities class action suits, points out that some large pension funds have actively participated in securities litigation and have successfully established corporate governance enhancements in class action settlements.4 Therefore, I expect institutional equity holdings to have a positive e?ect on fraud detection.
Monitoring by Independent Auditors: Independent auditors are probably the most important corporate “gatekeepers”. They pledge their reputational capital and provide protections to dispersed investors by verifying and assessing the quality of ?rms’ disclosures. In the late 1990s, however, such protections seemingly failed. The most notorious example is Arthur Andersen’s role in the
4 Keynote
address by William S. Lerach in council of institutional investors spring 2001 meeting.
51
Enron scandal and its subsequent criminal indictment. The increasing importance of non-audit services in auditing ?rms’ total revenue has also led to widespread market concern about auditor independence. Frankel, Johnson, and Nelson (2002) ?nd that auditor independence is negatively associated with the probability of earnings management. Anup and Chadha (2004) ?nd a negative but insigni?cant relation between auditor independence and the probability of accounting restatements. Bajaj, Gunny and Sarin (2003) examine a sample of class action lawsuits that involve allegations of accounting irregularities, and ?nd no signi?cant di?erence in auditors’ compensation (audit vs. non-audit fees) between the fraud sample and the comparison sample. However, for ?rms with large market reaction to the alleged fraud, their auditors have signi?cantly higher non-audit income. In this study, I directly examine whether higher auditor reputational capital leads to higher likelihood of fraud detection. First, I examine whether ?rms whose independent auditor is one of the ?ve largest accounting ?rms (Arthur Andersen, PricewaterhouseCoopers, Deloitte & Touche, Ernst & Young, KPMG) have a higher probability of fraud detection. Second, I examine the role of auditor opinion in fraud detection. If the independent auditors exert due diligence in certifying disclosures, then I expect that adverse auditor opinions to increase fraud detection.
Monitoring by Board of Directors:
The monitoring role of the board of directors is an impor-
tant component of corporate governance. The board is presumed to monitor the management on behalf of shareholders, because di?use ownership makes direct shareholder control di?cult. The economics and ?nance literature on the board starts with the assumption that the board’s monitoring e?ectiveness is a function of the board’s independence from the management. Two characteristics of the board, size and composition, are related to board independence. Empirical research in this area ?nds that board size and composition a?ect the observable board actions such as the board’s decision on CEO turnover, executive compensation, and merger/acquisitions (see surveys by John and Senbet (1998) and Hermalin and Weisbach (2003)). The recent wave of high-pro?le corporate scandals has brought the e?ectiveness of board monitoring to the center of securities legislation and governance reform. The newly-passed Sarbanes52
Oxley Act (SOX) and the NYSE and NASDAQ’s new corporate governance guidance mandate a number of changes that are aimed to improve board monitoring. For example, SOX requires that the audit committee consist entirely of independent directors and the audit committee hire the outside auditor. Both SEC and the national stock exchanges strongly recommend overall board independence. Several studies have examined the relation between the characteristics of the board and the probability of corporate fraudulent reporting. Beasley (1996) studies a sample of ?rms subject to SEC’s AAERs and ?nds that board independence (proxied by the percentage of outside directors in the board) is signi?cantly negatively related to the likelihood of ?nancial statement fraud. Klein (2002) ?nds an inverse relation between board independence and abnormal accruals. Dechow, Sloan and Sweeney (1996) ?nd that ?rms committing ?nancial statement fraud are likely to have a board dominated by insiders and have a CEO who is also the chairman of the board or the founder of the company. Agrawal and Chadha (2004) examine the incidence of accounting restatements, and ?nd that board independence is irrelevant, but the presence of independent directors with ?nancial or accounting expertise on the audit committee is associated with signi?cantly lower probability of accounting restatements. In this study, I examine the e?ect of board independence on the likelihood of fraud detection. Following the literature, I use board size and the percentage of outside directors to proxy for board independence. “Grey” directors who are not employees of a ?rm but have some business relation with the ?rm are not counted as outside directors.
Unexpected Performance Shock
Wang (2004) argues that fraud can be partially self-revealing. If the manager in?ates the earnings and misleads the market to have a high expectation on the ?rm’s future cash ?ows, then if later the cash ?ow realization turns out to be comparably bad (which the manager cannot fully control), outside investors will rationally think that they probably have been fooled and will start an investigation. Therefore, unexpected bad performance (unexpected by the market) after the commencement of fraud will increase the probability of fraud detection.
53
To proxy for such unexpected performance shock, I use the regression residual term from the following simple prediction model. ROAi,1 = ?0 + ?1 ROAi,0 + ?2 ROAi,?1 + i . (5.8)
ROAi,t is ?rm i’s return on asset in year t, which is de?ned as the ratio of operating income after depreciation over the average total assets from year t ? 1 to year t. ROAi,1 is used as the dependent variable because the average length of the class period is about one year.
i
(the residual ROA)
will be low if ?rm i’s performance in year 1 is bad compared with the (reported) performance in the previous two years. The realizations of this variable cannot be fully expected in year 0 when the management makes the fraud decision. Therefore, although this variable may signi?cantly in?uence the ?rm’s detection risk, its e?ect is ex-post and thus should not a?ect the ?rm’s ex-ante fraud decision.5
5.3.2 Propensity to Commit Fraud
The equilibrium supply of fraud depends on the expected bene?t and cost of engaging in fraud. Therefore, xF should include factors that can a?ect either the bene?t from fraud, or the litigation risk, or both. The previous section discusses some potential determinants of the detection risk. Now I turn to factors that can potentially in?uence a ?rm’s bene?t from committing fraud.
5 There
are two caveatees associated with this variable. First, this variable is not completely exogenous. The
direction of causality, however, is not ambiguous. It is intuitive that bad operating performance eventually reveals fraud. Detection of fraud may result in immediate plunge in stock returns and may a?ect the long-run performance of the fraudulent ?rm, but it is hard to believe that the revelation of fraud leads to immediate bad operating performance. Second, the management could have better information about future abnormal bad performances than the market does. Therefore, expectation of the residual ROA may impose some ex ante deterrence. However, a reasonable counter argument is that the managers commit fraud because they believe that the current bad performance is only temporary and things should go back to normal later. I will discuss the robustness of the results regarding this variable in Section 5.6.
54
Pro?tability and Growth Potential
Wang (2004) and Bebchuk and Bar-Gill (2002) predict that ?rms that have high growth potential but experience negative pro?tability shocks have high propensity to commit fraud. The intuition is that for such ?rms misreporting short-term ?rm performance can allow them to raise external capital and exercise their growth options on sweet terms. A problem emerges when we test the above prediction. We cannot directly observe the negative pro?tability shocks because they are covered by fraud. A possible solution is to use the ex-post restated ?nancial data rather than the originally reported data. However, to my knowledge, the restatement data in COMPUSTAT is not as comprehensive and complete as the original data. Therefore, I try to infer the existence of pro?tability shocks by comparing the pro?tability before the commencement of fraud and that at the revelation of fraud. The di?erence between the two pro?tability levels can imply hidden performance changes when fraud is alive. I use return on asset ROA as the pro?tability measure. I use two proxies for growth potential, the annual asset growth rate and the book-to-market ratio.
External Financing Needs
The combination of low asset pro?tability and high growth implies large reliance of a ?rm on the external capital markets. Stein (1989) argues that the lack of ?nancial slack can expose the manager to capital market pressure and can motivate the manager to in?ate short-term performance at the cost of forfeiting long-term values. The earnings management literature has provided evidence that managers tend to overreport earnings prior to major external ?nancing activities such as public equity o?erings (see, e.g., Teoh, Welch and Wong (1998a,b)). I construct two variables to proxy for a ?rm’s external ?nancing needs. The ?rst variable, externally ?nanced growth rate, is constructed based on Demirg¨ uc ¸-Kunt and Maksimovic (1998) to proxy for a ?rms’ projected need for outside capital. Speci?cally, the externally-?nanced growth rate is a ?rm’s asset growth rate in excess of the maximum growth rate that can be supported by the ?rm’s internally available
55
capital (ROA/(1-ROA)).6 The second variable, EF , is constructed following Richardson and Sloan (2003) to measure a ?rm’s actual net external ?nancing cash ?ows. Speci?cally, EFt = ?CEt + ?P Et + ?DEBTt , ASSET St
where ?CEt , ?P Et , and ?DEBTt are the changes in the book value of common equity, preferred equity, and total debt in year t, respectively. ASSET St is the book value of assets.7 This variable can be viewed as a measure of a ?rm’s realized external ?nancing need. Since the second variable is an outcome-based measure, I focus on the ?rst variable in order to reduce endogeneity, and use the second measure only as a robustness check.
Financial Distress
Another factor that is closely related to ?nancial slack and external ?nancing need is the degree of ?nancial distress. Maksimovic and Titman (1991) theorize that ?nancial di?culties can a?ect a ?rm’s incentive to honor its implicit contracts and in other ways maintain a favorable reputation. In their model, both ?nancial shortfalls and overall debt overhang can induce the distressed ?rm to increase current cash ?ow at the cost of losing reputation and long-term profitability. Several accounting studies ?nd some evidence that avoidance of penalties associated with the violations of debt covenants is a motivation to manage earnings (Sweeney (1994), DeFond and Jiambalvo (1994), and Dechow et al. (1996)). These studies imply that ?nancial distress can increase ?rms’ incentives to misreport. I use the ratios of long-term debt and short-term debt to total assets to proxy for the degree of ?nancial distress.
Insider Equity Incentives
The relation between insiders’ equity incentives and the incidence of corporate fraud has been at the center of the current debate and reform on corporate governance. There are two forces associated with insiders’ equity stake. On one hand, the classic agency theory implies that
6 See
Demirg¨ uc ¸-Kunt and Maksimovic (1998) for assumptions and justi?cations for this measure. According to
the discussion in that paper, ROA here is the ratio of income before extraordinary items over assets.
7 See
Richardson and Sloan (2003) for a discussion of some possible limitations of this measure.
56
higher percentage insider ownership can better align insiders’ incentives to that of the shareholders. Since fraud is outright contravention of shareholders’ interest, high insider ownership should be associated with low fraud propensity. The agency view is supported by the work of Alexander and Cohen (1999). They examine public ?rms convicted of federal crimes in 1984-1990, and ?nd that crime occurs less frequently among ?rms in which management has a larger ownership stake. On the other hand, large equity incentives can be a double-edged sword, because the positive relation between ?rm performance and insiders’ compensation (or wealth) can induce distorted managerial reporting incentives (see, e.g., Goldman and Slezak (2003)). The second force seems to be supported by the ?ndings in some recent empirical work such as Johnson, Ryan and Tian (2003), Peng and R¨ oell (2004), and Burns and Kedia (2004). These papers ?nd that high pay-forperformance ratio (as a result of large equity-based compensation) is related to high probability of fraud or earnings manipulation, indicating over-incentivization of the management. In this study, I examine the role of insider percentage stock ownership and executive equity compensation in the context of accounting fraud. Executive equity compensation is measured as the value of restricted stock and stock options (using the Black-Scholes model) over an executive’s total compensation. I then compute the sum and average of the ratios across the ?ve top executive o?cers in the ?rm.
5.3.3 Control Variables
Some previous studies on ?nancial statement fraud ?nd that ?rms tend to commit fraud at a very early stage of their business cycle. Beasley, Carcello and Hermanson (1999) document that ?rms that have engaged in ?nancial statement fraud are generally small. The National Commission on Fraudulent Financial Reporting (AICPA 1987, 29) states that young public ?rms may face greater pressure to dress up ?rm appearance and thus have higher likelihood of engaging in fraud. There are also clear industry patterns in securities litigation (see Table 2). Technology ?rms (software & programming, computer and electronic parts, biotech), service ?rms (?nancial services, business services, utility, and telecommunication services) and the trade industries (whole sales
57
and retails) appear to have disproportionately high fraud concentration. This implies that these industries tend to have either large bene?t from fraud, or high detection risk, or both. Furthermore, ?rm size, age and industry segments are likely to be correlated with ?rms’ pro?tability, growth potential, external ?nancing need and ownership structure. Therefore, I control for ?rm size (log of total assets), age (as a public company), and ?rms’ membership in the technology, service and trade sectors.
5.3.4 Summary of Model Speci?cation
Factors Growth Potential External Financing Need Financial Distress Pro?tability Pro?tability (ex post) Insider Equity Incentive Investment Shareholder Monitoring Board Monitoring Independent Auditor Control Variables Variables Asset Growth, Book-to-Market Ext. Fin. Growth, Ext. Fin. C.F. Leverage, ST Debt ROA Residual ROA Insider Own, Equity Compensation R&D, Capital Exp., Acquisition Block Own, Institution Own Board Size, Outside Director Big Five, Auditor Opinion Firm Size, Age, Industry +/+ + + + ?F + + + ?D
5.4 Descriptive Information and Univariate Analysis
This section presents univariate comparisons between the fraud sample and the comparison sample. The explanatory variables are grouped into ?ve categories: (1) ?rm size and age; (2) profitability and growth; (3) external ?nancing needs; (4) investment; and (5) corporate monitoring. Table 4 Panel A reports the median and mean of each variable for both samples and the nonparametric Wilcoxon z-statistics for testing di?erences between the two samples. All the ?nancial 58
information is retrieved from COMPUSTAT database. Information on stock-based acquisition and acquisition volume is from SDC Platinum database. Ownership information is from CDA Spectrum. Information on executive equity compensation is from ExecuComp database. Information on board of directors is from EdgarPro database. To facilitate my analysis, I use the following ?scal year counting. For the fraud sample, I label the ?scal year in which fraud begins as year 0. The determination of year 0 is discussed in Section 5.1.1. Then the ?scal year prior to year 0 is year -1, and the one after is year 1. Since the comparison sample consists of all the non-litigated ?rms, all the comparison ?rms enter each relevant ?scal year. For example, ?scal year -1 spans from 1991 to 2002 for the fraud sample. Then all the observations of the comparison ?rms in year 1991 to year 2002 are labelled as information from year -1 and are used in the analysis. In this study, all the variables on pro?tability, growth, and external ?nancing needs are measured at the average level from year -2 to year -1. Using pre-fraud information helps to mitigate the e?ect of fraud on those measures. Information on investment is from year 0. The reason is that those investments were made around the time when fraud was committed, and therefore could have been used strategically by the management to disguise fraud. Corporate monitoring variables are measured at the average level from year -1 to year 0. Year 0 information is incorporated to strengthen the deterrence e?ect of monitoring on ?rms’ decision to commit fraud. The fraud sample on average appears to be larger but younger than the comparison sample. Studies on SEC’s AAERs generally ?nd that alleged ?rms are small (see, e.g., Beasley, Carcello and Hermanson (1999)). Firms that are subject to private class action litigation can be larger because class action lawsuits tend to target ?rms with “deeper pockets” (Cox and Thomas (2003)). The fraud sample seems to have outperformed the comparison sample in the two years before the commencement of fraud, and underperformed the comparison sample in year 1. This is consistent with the argument in Section 5.3.2. Fraudulent ?rms experienced some negative performance shock in year 0 but chose to cover up the problems by false ?nancial disclosure. Then fraud got uncovered in year 1, and the concealed bad performance was revealed.
59
Table 4 Panel A also shows that fraudulent ?rms tend to have signi?cantly higher growth rate and lower book-to-market ratio than the comparison ?rms. The median asset growth rate is 46% for the fraud sample, and only 9% for the comparison sample. High growth and low internal pro?tability naturally leads to large need for outside capital. According to the argument in Demirg¨ uc ¸-Kunt and Maksimovic (1998), on average only 13% of the growth in the fraudulent ?rms could be supported by internal funds, resulting in a high projected need for outside capital. The fraudulent ?rms also raised more external capital even before the commencement of fraud. The median ratio of net external ?nancing cash ?ow to total assets is 19% for the fraud sample, and only 4% for the comparison sample. The fraud sample, however, does not seem to be more burdened by debt than the comparison sample. The di?erence in growth opportunities across the two samples is further re?ected in investment expenditures. The fraud sample on average invested more than the comparison sample did in the year when fraud occurred. For instance, the median ratio of net investing cash out?ow to total assets is 11% for the fraud sample, and 6% for the comparison sample. However, the univariate comparisons do not control for factors that may in?uence the size of investment. We know that di?erent industries have di?erent investment patterns, and young ?rms tend to invest more than mature ?rms do. Firm size is also a potential determinant of investment size. Since all the investment variables have been normalized by the book value of assets, the size e?ect is already taken into account. Therefore, in order to have a more direct test of the overinvestment prediction in Wang (2004), I construct a control sample that is matched with the fraud sample in terms of industry distribution (two-digit SIC codes) and ?rm age at the end of year -1. Table 4 Panel B shows that the fraud sample on average had a much higher investment intensity than the control sample did both before and after the commencement of fraud. Except for capital expenditures, the di?erences across the two samples are statistically signi?cant, and particularly so for merger/acquisition-related expenditures. Finally, Table 4 Panel A shows that the fraud sample on average has more concentrated ownership, more institutional holdings, more large insider equity incentives. The fraud sample also
60
tends to have a smaller board and lower percentage of independent directors.
5.5 Multivariate Analysis
This section presents evidence from multivariate tests to simultaneously assess the e?ects of ?rm characteristics, investment, and corporate monitoring on a ?rm’s propensity to commit fraud and the probability of fraud detection.
5.5.1 Firm Characteristics and Fraud
Table 5 reports the e?ects of pro?tability, growth and external ?nancing need on a ?rm’s fraud incentives. We can see that ROA is positively associated with the likelihood of fraud. This result may seem counterintuitive at ?rst glance. However, it is actually intuitive because it is di?cult for a (known) troubled ?rm to sell a good earnings report. A ?rm will have incentive to fool the market and may easily succeed when the market believes that the ?rm is pro?table based on previous years’ performance, while deterioration in pro?tability has already started. The concealed performance deterioration, if it continues, will eventually lead to the revelation of fraud. The average marginal e?ect of residual ROA on P (D|F ) across all the models is -0.26, which means that a 10% unexpected decrease in ROA in year 1 is associated with an average 2.6% increase in the probability of detection. This result supports the argument in Wang (2004) that fraud is, to some extent, self-revealing. The more the manager is able to raise the market’s expectation by fraudulent reporting, the more likely the market will later see inconsistency between ?rm performance and what it has been guided to expect. The inconsistency leads to the discovery of fraud. Table 5 also shows that a ?rm’s growth potential and external ?nancing need are important motivational factors for fraud. Models 1 and 2 indicate that higher asset growth rate and lower book-to-market ratio are related to higher probability of fraud. Model 3 implies that fraudulent ?rms are likely to have a growth rate higher than what can be supported by their internal funds. The average marginal e?ect of externally ?nanced growth on P (F ) across models is 0.64, which means that increasing the externally ?nanced growth by 10% tends to increase a ?rm’s probability
61
of misreporting by 6.4%. Models 4-6 further show that fraudulent ?rms on average raise more external capital, but they do not appear to be more burdened by debt. This implies that fraudulent ?rms may pursue more equity ?nancing than debt ?nancing. Overall, results in Table 5 imply that rapidly growing ?rms with insu?cient internal capital are likely to misreport their ?nancial performance, because fraud enables them to exercise their growth options on favorable terms.
5.5.2 Investment and Fraud
Table 6 reports the relation between ?rms’ investment expenditures and their fraud incentives. Several interesting results emerge. First, I ?nd that di?erent types of investment have di?erential e?ects on the likelihood of fraud detection. Investment in R&D has the strongest negative e?ect on the probability of fraud detection. The relation is statistically and economically signi?cant. The average marginal e?ect of R&D expenditures on P (D|F ) across models is -0.17, which means that a 10% higher R&D expenditures is on average associated with a 1.7% lower probability of detection. Note that a ?rm’s total litigation cost is the probability of detection times the penalty upon detection. Suppose that the penalty can be completely measured in terms of money, then a 1.7% decrease in the detection likelihood can correspond to a substantial reduction in the dollar value of litigation cost. The e?ect of net investing cash ?ow on the probability of detection is also signi?cantly negative but much weaker than that of R&D expenditures. The average marginal e?ect of net investing cash ?ow on P (D|F ) is -0.05. Straightforward investment like capital expenditures does not seem to in?uence the likelihood of detection. Neither do acquisition expenditures. I further examine some di?erent measures of merger/acquisition intensity. Model 10 shows that larger the number of acquisitions in year 0, higher the probability of detection. A possible explanation for this result is that the regulators and the market may pay more attention to ?rms that are active in mergers/acquisitions. Furthermore, Model 11 shows that more focused acquisitions (acquisitions within the same two-digit SIC codes), higher the probability of detection. It is consistent with
62
the argument in Wang (2004) that investments that have high correlation with a ?rm’s current business can increase the ?rm’s litigation risk. Second, di?erent types of investment also have di?erential e?ects on ?rms’ propensity to commit fraud. The di?erences can stem from two sources: either di?erent investments a?ect ?rms’ bene?t from fraud di?erently, or they a?ect ?rms’ risk of being detected di?erently. Let us compare cash-based acquisitions and stock-based acquisitions. Table 6 shows that these two types of acquisitions have di?erent e?ects on P (F ), but not on P (D|F ). Stock-based acquisitions have a signi?cant positive relation with P (F ), while cash-based acquisitions do not. This implies that the ?nancing of the investment in?uences ?rms’ bene?t from committing fraud, but not the detection risk. Then let us compare R&D expenditures and capital expenditures. In all models, R&D expenditures are signi?cantly positively associated with P (F ), while capital expenditures do not in?uence P (F ). If these two types of investment are not generally ?nanced di?erently, then their di?erential e?ects on P (F ) should largely arise from their di?erential e?ects on P (D|F ). Holding other factors constant, ?rms that invest more in R&D tend to have lower litigation risk. Low litigation risk can encourage fraud.
5.5.3 Equity Ownership and Fraud
Table 7 presents the roles of insider equity incentives and shareholder monitoring in determining a ?rm’s fraud incentives. First, insider percentage stock ownership has a signi?cant concave relation with the probability of fraud. That is, when insider ownership is small, the probability of fraud increases as insider ownership increases. When insider ownership is large, however, the probability of fraud decreases as insider ownership increases. Given the dramatic increase in the use of stock options in managers’ compensation, the percentage stock ownership will not capture the full impact of managers’ equity incentives. Therefore, I construct an executive equity compensation variable, which is the total value of an executive’s restricted stock and stock options over her total compensation and sum over all the key executives in the company.8 Model 14 shows that
8 For
insider equity incentives, I have also examined various speci?cations of executive equity compensation other
than the one reported in Model 14. For example, I compute the average ratio rather than the sum across executives
63
executive equity compensation exhibits a similar but slightly weaker concave shape. The concavity implies that insider equity ownership (or equity compensation) can be a double-edged sword when it is used to align the interest of the managers to that of the outside shareholders. When insiders hold small stakes in the ?rm, the agency problem due to separation of ownership and control can be severe. However, steep equity incentive scheme may not solve the problem, because it can induce insiders to misreport rather than to work harder for the interest of outside shareholders. Interestingly, equity incentive seems to work well only when insiders already have substantial equity stakes in the ?rm.9 Second, I ?nd that the presence of large shareholders and institutional shareholders increases fraud detection and discourages fraud. The marginal e?ects of institutional ownership on P (D|F ) and P (F ) are 0.14 and -0.27, respectively. This means that a 10% increase in institutional share holdings is associated with an average 1.4% increase in the probability of fraud detection, and an average 2.7% decrease the probability of fraud.10 Block ownership has a similar e?ect. In general, however, block ownership has a slightly stronger e?ect on P (F ), while institutional ownership has a slightly stronger e?ect on P (D|F ). These results imply that the strength of shareholder monitoring in?uences ?rms’ propensity to commit fraud through their impact on the likelihood of fraud detection, and provide support for enhancing shareholder monitoring in the on-going corporate governance reform.
in a company. I use the value of exercised stock options rather than the Black-Scholes value of all stock option holdings. I also examine equity ownership and equity compensation of CEO. Overall, the results are qualitatively consistent across di?erent speci?cations.
9I
separately examine the subsample of ?rms that have 20% or higher insider ownership. I ?nd a signi?cant
negative relation between insider ownership and the probability of fraud. This result is not reported in the tables.
10 There
is a caveatee regarding the interpretation of the result. The Private Securities Litigation Reform Act
(PSLRA) that was passed in December 1995 requires that every class action lawsuit appoint a lead plainti?. PSLRA encourages large institutional investors to be lead plainti?s. Therefore, class action suits could be more likely to go through for ?rms that have large institutional investors. This may lead to the positive relation between institutional ownership and P (D|F ).
64
5.5.4 Auditor, Board and Fraud
Table 8 presents the e?ects of independent auditors and corporate boards on corporate fraud incentives. The auditor being one of the ?ve largest accounting ?rms appears to be related to higher likelihood of fraud detection.11 The deterrence e?ect, however, is not statistically signi?cant. Auditor opinions seem to have no in?uence on detection. The reason is that auditor opinions do not exhibit much variation at all. For the fraud sample, 79% of the auditor opinions in the year when fraud occurred were unquali?ed opinions, and the rest 21% were unquali?ed opinions with some explanations. The uniformly unquali?ed auditor opinions themselves show the problem: Why do independent auditors seldom disagree with their clients regarding the quality of disclosure? Are they truly independent? On monitoring by the board of directors, models 16-17 show that board size and the percentage of outside directors are positively associated with the probability of detection and negatively associated with the probability of fraud. However, the relations are not statistically signi?cantly. This could be due to the power issue. Unfortunately, I do not have board data for a large number of ?rms in my sample (see Table 4).
5.5.5 Summary of Results
In sum, Tables 5-8 present the multivariate analysis on the e?ects of ?rm characteristics, investment, and corporate monitoring on a ?rm’s probability of committing fraud and the probability of fraud detection. I ?nd that fraudulent ?rms are likely to be high-growth ?rms that have large needs for external capital but experience negative shocks in pro?tability. Performance deterioration, although temporarily concealed by fraud, tends to reveal itself and increase fraud detection. Second, I ?nd that investment can in?uence both ?rms’ ex-ante bene?t from committing fraud (e.g., through the ?nancing of the investment) and their ex-post detection risk. Therefore,
11 I
take out Arthur Andersen and ?nd similar result on the big four accounting ?rms. I also examine Arthur
Andersen separately, and ?nd no signi?cant result. Since these results are similar to what is reported in Table 8 Model 15, they are not reported.
65
it is an important determinant of ?rms’ incentives to defraud investors. Investments with high degree of uncertainty and/or low correlation with existing assets tend to negatively in?uence the likelihood of fraud detection. These results, together with the evidence of overinvestment in Panel B of Table 4, imply that fraud can be associated with investment distortions and thus real economic costs. Finally, di?erent types of corporate monitors also appear to have di?erent e?ects on fraud propensity and fraud detection. The presence of block equity holders and high institutional holdings is associated with high probability of fraud detection and low probability of fraud. There is weak evidence that reputable independent auditors and large corporate boards increase the likelihood of fraud detection.
5.5.6 Comparison with Simple Probit Models
Existing studies on fraud have used straight probit models to assess the e?ect of a factor on a ?rm’s probability of committing fraud. As discussed in Section 5.2.3, the straight probit model equates the probability of detected fraud to the probability of fraud. Therefore, it can not only underestimate the probability of fraud, but also lead to incorrect inferences. Table 9 compares the results from the straight probit model and the bivariate probit model, and demonstrates the problems associated with the straight probit model. Using the full sample of comparison ?rms as in the previous models (i.e., using multiple years’ data for every comparison ?rm) leads to very low marginal e?ects of all the variables in the straight probit model. Therefore, in order to better illustrate the di?erences between the straight probit and the bivarate probit models, I randomly choose one year for every comparison ?rm. That is, each comparison ?rm only enters the regression once in Table 9. First, let us look at the results on R&D expenditures. The straight probit model shows no signi?cant e?ect of R&D expenditures on P (F ), while the bivariate probit model shows a strong positive e?ect. The reason is that investment in R&D has opposing e?ects on P (F ) and P (D|F ). The two forces roughly o?set each other, resulting in no e?ect on the probability of detected fraud
66
P (Z ). Second, the marginal e?ects of net investing cash ?ows on P (F ) have consistent signs across the two models, but substantially di?er in magnitude (0.10 in probit and 0.43 in bivariate probit). The straight probit model underestimates the marginal e?ect of this variable. Third, the two models show opposing e?ects of institutional ownership on P (F ). The straight probit model reports a positive e?ect, while the bivariate probit model reports a negative one. Again, the reason is that institutional ownership has opposite e?ects on P (F ) and P (D|F ), and for this variable the positive e?ect on detection dominates. The comparisons in Table 9 clearly show that disentangling the e?ect of a factor on the probability of detecting fraud and its e?ect on the probability of committing fraud is important for us to draw sensible conclusions.
5.6 Robustness Checks 5.6.1 Frivolous Lawsuits
In this study, I use the ?ling of securities class action lawsuits to proxy for detected fraud. However, the ?ling of a lawsuit does not necessarily indicate that the alleged ?rm is fraudulent, because allegations could be frivolous or mistaken. Therefore, the fraud sample could be subject to biases due to possible false detections. Many studies in the legal literature have argued that The Private Securities Litigation Reform Act (PSLRA), which was passed in December 1995, makes it more di?cult for shareholders to sue a public company (see., e.g., Choi (2004)). My sample consists of litigation suits since 1996 (post PSLRA). Therefore, the probability of frivolous lawsuits in my sample should be lower than it was before PSLRA. In order to further mitigate the bias of false detection in the estimation, I separately examine the following three subsamples. The ?rst subsample has 334 ?rms that announced accounting restatements surrounding the securities lawsuits. The accounting restatement information is from General Accounting O?ce (GAO)’s October 2003 report. Since I study accounting-related fraud, the fact that the alleged ?rms restated their ?nancial reports provides support to the allegations. 67
The second subsample contains 207 ?rms that were subject to parallel SEC’s AAERs. Information on AAERs is retrieved from SEC’s web site. If frivolous lawsuits could result from the pro?t-orientation of private securities lawyers, then having parallel SEC’s litigation increases the credibility of the lawsuits, because SEC is not pro?t-oriented. Several papers in the legal literature (see, e.g., Johson, Nelson and Pritchard (2002), Choi (2004)) have viewed suits that result in dismissal or a low value settlement ($2 million or less) as “nuisance”. Therefore, in the last subsample, I exclude 27 cases that were either later dismissed by the court or had a settlement less than $2 million. The dismissal and settlement information is retrieved from the Securities Class Action Clearinghouse (SCAC). Table 10 shows that the main model results hold qualitatively across all three subsamples. This implies that the possible existence of false detection does not drive the results.
5.6.2 Timing of Fraud
The beginning of a ?rm’s fraudulent scheme is generally a little fuzzy due to the di?culty of tracing evidence far back in time. For accounting-related frauds, identifying the timing of fraud can be even more di?cult because the border line between aggressive accounting and securities fraud is not always a clear cut. In this study, I determine the beginning ?scal year of fraud (year 0) based on the speci?cation of class periods and ?rms’ ?scal year ending months. For ?rms that are subject to both private class action litigation and SEC’s AAERS, I also cross check the timing of fraud using the information in SEC’s litigation ?lings. To further examine the validity of the year 0 speci?cations, I compare ?rms’ ROA based on the originally reported accounting data with ROA based on the restated data from COMPUSTAT. Figure 5 plots the median historic and restated ROA for both the fraud and comparison samples from year -2 to year 2. We can see that for the fraud sample, the historic ROA and the restated ROA are consistent with each other in years -2 and -1, start to diverge in year 0, and then re-converge in year 2. This implies that the determination of year 0 is on average valid.
68
5.6.3 Industry and Business Cycle e?ects
So far, this paper has focused on ?rm-level economic determinants of fraud. Several papers have argued that industry and market environment also in?uences ?rms’ fraud incentives. Wang (2004) predicts that fraudulent events tend to cluster in certain industries during certain time period, because both the bene?t from fraud and the litigation risk are correlated among ?rms in the same industry. Gande and Lewis (2005) empirically document the industry spillover e?ect in securities litigation. That is, the ?ling of lawsuits on one ?rm signi?cantly negatively a?ects the stock performances of other ?rms in the same industry. I have controlled for industry distribution in the analysis. Here I further control for the industry securities litigation environment. I use the logarithm of the total market value of fraudulent ?rms in an industry in year -1 to proxy for industry litigation intensity. A high total market value can result from either a large number of frauds or the existence of some mega cases. Poval, Singh and Winton (2004) argue that ?rms’ fraud incentives are in?uenced by businesscycle factors. Their model shows that corporate fraud incentives are low when the economic condition is very good (investors are highly optimistic) or when it is very bad (investors are highly skeptical). The fraud incentives are high when the economic condition is switching from good to bad. In order to understand the e?ect of market-wide determinants on the probability of fraud, I construct a business cycle variable that equals -1 if the year in which fraud begins is between 1992 and 1994 or between 2001 and 2002 (bust), equals 0 if year 0 is between 1995 and 1997 or in 2003, and equals 1 if year 0 is between 1998 and 2000 (boom). Table 11 shows that the main model results remain unchanged after incorporating the industry and business cycle e?ects. In addition, both industry litigation intensity and business cycle variables tend to be positively related to P (D|F ) and negatively related to P (F ). The likelihood of fraud detection is high in industries with high litigation intensity. Good economic conditions are also related to higher ex post detection risk. This is actually intuitive. First, if a fraud begins in a very good year, this implies that the fraudulent ?rm has some negative idiosyncratic shocks. Second, very good economic conditions may not continue. The problems
69
concealed by fraud are likely to be revealed as the overall condition weakens, which leads to the discovery of fraud.12
5.6.4 Di?erent Model Speci?cations
The model speci?cation described in Section 5.3.4 is mainly from the ?rm’s viewpoint. Companies rationally compare the expected bene?t and litigation risk of engaging in fraud. The explanatory variables in the P (F ) equation consist of variables that either in?uence ?rms’ bene?t from committing fraud or in?uence their litigation risk. We can extend the model into a strategic two-party game: The ?rm calculates its risk of being detected when it makes the fraud decision. The detection forces also anticipate the ?rm’s likelihood of committing fraud when allocating their resources. For example, the market may be more vigilant with ?rms with high externally ?nanced growth if those ?rms tend to have high propensity to commit fraud. This implies that the externally ?nanced growth can be positively associated with the probability of fraud detection. Therefore, in Table 11 Speci?cation 3, if a factor a?ects a ?rm’s bene?t from committing fraud and its e?ect can be anticipated ex ante by the detection forces, then this factor is in both the fraud commitment and fraud detection equations. The results show that although high externally ?nanced growth is an important motivational factor for fraud, it does not appear to signi?cantly in?uence ?rms’ probability of being detected. A possible explanation for this is that growth itself is not necessarily a bad thing, and therefore does not necessarily trigger investor vigilance. As discussed in Section 5.3.1, the residual ROA variable is not completely exogenous. This variable, however, appears in all the models. In Table 11 Speci?cation 4, I take out this variable and examine whether the results on other variables still hold. The main results are qualitatively unchanged. The statistical signi?cance of variables is consistent with previous models. However, the marginal e?ects of variables in the P (F ) equation are lower.
12 I
also use the return to a market portfolio to proxy for overall business conditions. The results are consistent with
those reported in Table 11. High market return in the beginning year of fraud is associated with high probability of fraud detection.
70
Table 5.1: Allegations of Accounting Fraud : 1996 – 2003 Panel A: Litigation Filings by Calender Year The fraud sample consists of 684 class action lawsuits against 660 US public companies. The total number of lawsuits each year does not include cases ?led against private companies or cases against investment companies for pure fraudulent investment banking activities (such as unfair allocation of IPO shares and misleading analyst reports).
Year Accounting fraud Total # of lawsuits % of total
1996 45 100 45.00
1997 70 163 42.94
1998 103 232 44.40
1999 80 195 36.92
2000 107 206 51.94
2001 88 168 52.38
2002 119 212 56.13
2003 71 177 40.11
1996-2003 684 1454 47.04
Panel B: Class Periods, Age, and Stock Exchange The information on class periods is retrieved from the class action lawsuits. Age is de?ned as the number of years between a ?rm’s IPO date and the beginning of its class period. A ?rm’s stock exchange is identi?ed as of the beginning of the class period. Class Period # of obs. mean median maximum minimum (days) 660 471 354 2040 13 Age (years) # of obs. 652 mean 8.16 median 3.45 age10 years 22.66% Stock Exchange # of obs. 660 NYSE 30.3% AMEX 3.7% NASDAQ 64.0% Other 2.0%
Panel C: Accounting Fraud by the Beginning Fiscal Year The beginning ?scal year of a fraud is identi?ed based on the speci?cation of the class period and the ?rm’s ?scal year ending month. Fiscal year # of cases Fiscal year # of cases 1992 1 1998 97 1993 8 1999 117 1994 15 2000 108 1995 47 2001 57 1996 86 2002 16 1997 107 2003 1
71
Table 5.2: Industry Distribution of Accounting Fraud This table reports the distribution of accounting-fraud ?rms across industry segments. I classify ?rms into 24 industry segments based on 2-digit or 3-digit SIC codes, as detailed in the table. Percentage of total is computed based on the total number of public ?rms in each industry in the COMPUSTAT database.
Industry Agriculture (100-900) Mining (1000-1400) Construction (1520-1731) Food & Tobacco (2000-2111) Fabrics & Textile Products (2200-2390) Wood & Furniture (2400-2590) Paper & Printing (2600-2790) Chemicals (2800-2821, 2840-2990) Pharmaceutical (2833-2836) Materials & Related Products (3011-3490) Industry Manuf. (3510-3569, 3578-3590, 3711-3873) Computer-related Hardware (3570-3577) Electronics (3600-3695) Miscellaneous Manuf. (3910-3990) Transportation (4011-4731) Telecommunications (4812-4899) Utilities (4900-4991) Wholesales (5000-5190) Retails (5200-5990) Financial Services (6021-6799) Services (7000-7361, 7380-7997, 8111-8744) Software & Programming (7370-7377) Healthcare Services (8000-8093) Others (8880-9995) Total
Fraud Events 1 10 1 11 12 2 3 4 22 18 49 33 64 2 11 31 29 31 36 73 66 115 36 0 660
% of Sample 0.15 1.52 0.15 1.67 1.82 0.30 0.45 0.61 3.33 2.73 7.42 5.00 9.70 0.30 1.67 4.70 4.39 4.70 5.45 11.06 10.00 17.42 5.45 0.00 100
% of Total 1.18 0.74 0.78 2.59 3.79 1.06 0.72 0.82 2.81 1.89 2.44 6.82 5.13 0.91 2.24 3.65 4.37 3.64 2.70 1.51 3.72 6.13 9.57 0.00 2.96
72
Table 5.3: Table 3: Nature of Accounting Fraud This table presents the nature of the alleged ?nancial misrepresentations in 563 securities lawsuits studied in this paper. I am only able to identify the exact nature of the misrepresentation in 563 cases based on the information in relevant case documents (e.g., case complaints, press releases and court decisions). I categorize these 563 cases into 11 groups based on the accounting items that have been manipulated. I report the number of ?lings and the frequency of each category. Allegations # of identi?ed cases Improper revenue recognition Understatement of expenses Non-recurring items Overstatement of account receivables Overstatement of inventory Overstatement of intangibles Overstatement of investment Overstatement of other assets Understatement of reserves/allowances Understatement of liability Other # of Filings 563 380 97 4 53 38 13 9 72 49 20 24 % of Sample 67.50 17.26 0.71 9.43 6.76 2.31 1.60 12.81 8.72 3.56 4.27
73
Table 5.4: Univariate Comparisons of Firm Characteristics For each variable, the median, the mean (in the brackets) and the z -statistics for Wilcoxon tests are reported. ** and * indicate signi?cance at 1 and 5% levels, respectively. “ROA”=(operating income after depreciation)/assets. “Res. ROA” is the residual from regression: ROA1 =?0 + ?1 ROA0 + ?2 ROA?1 + . “B-M”=(assets)/( assets-equity+market value). “EF. Growth”=“Asset Growth” ROA2 - 1? ROA2 , where ROA2=(income before extraordinary items)/assets. “EF. C.F.”=(?common stock+?preferred stock+?debt)/assets. “Leverage”=(LT debt)/assets. “ST Debt”= (debt in current liabilities)/debt. “Bank/Debt”=(bank loan)/debt. “Invest. C.F.”= -(net investing cash ?ow)/assets. “Focused Acquis” is the percentage of acquisitions in which the target ?rm is within the same two-digit SIC codes as the acquirer. “Insider” is the percentage ownership of o?cers and directors. “Block” is the total percentage ownership of the shareholders who own at least 5% of the ?rm’s equity. “Institution” is the percentage ownership of ?nancial institutions. “Big Five”=1 if the independent auditor is one of the biggest ?ve accounting ?rms, and 0 if otherwise. “Opinion” goes from 1 (best) to 5 (worst). “B-Independ.” is the fraction of independent directors. “Equity Comp.” is executives’ value of stock and stock option over total compensation. Fraud Sample 192 (5515) 496 (5102) 157 (2057) 2.89 (7.64) 0.08 (-0.00) -0.01 (-0.07) 0.46 (1.09) 0.50 (0.53) 0.41 (1.08) 0.19 (0.23) 0.10 (0.17) 0.27 (0.36) 0.00 (0.05) 0.11 (0.14) 0.04 (0.06) 0.00 (0.04) 0.01 (0.11) 0.00 (0.79) 0.00 (0.22) 0.15 (0.21) 0.35 (0.38) 0.36 (0.39) 1.00 (0.81) 1.00 (1.21) 2.67 (3.59) 0.25 (0.29) 2.31 (2.31) # of obs. 631 535 627 630 616 545 563 521 562 551 626 561 632 611 614 579 587 631 631 599 602 572 631 533 273 273 223 Nonfraud Sample 136 (3538) 93 (1764) 90 (1520) 6.25 (8.73) 0.05 (-0.06) 0.02 (-0.00) 0.09 (0.36) 0.77 (0.74) 0.07 (0.39) 0.04 (0.03) 0.11 (0.18) 0.24 (0.35) 0.00 (0.05) 0.06 (0.08) 0.04 (0.06) 0.00 (0.02) 0.00 (0.03) 0.00 (0.14) 0.00 (0.05) 0.09 (0.18) 0.26 (0.32) 0.19 (0.26) 1.00 (0.71) 1.00 (1.29) 4.33 (4.75) 0.29 (0.34) 1.42 (1.52) # of obs. 68202 56493 65696 63338 66634 49764 61470 53270 59819 59893 66873 59676 64389 59165 56480 54566 55785 65047 65047 37526 37569 37506 68202 56210 2312 2617 12178 Wilcoxon z 5.44** 15.64** 6.44** -9.00** 8.03** -10.22** 20.03** -14.90** 17.94** 19.05** 0.45 1.12 5.04** 9.92** 2.06* 15.91** 17.37** 24.07** 21.72** 5.58** 6.88** 11.38** 5.06** -3.91** -10.05** -6.30** 9.78**
Assets($106 ) Market Value($106 ) Sales($106 ) Age ROA Res. ROA [1] Asset Growth B-M EF. Growth EF. CF. Leverage ST Debt R&D Invest. C.F. Capital Exp. Acquis.(cf.) Acquis.(cf.+stock) # of Acquis. Focused Acquis. Insider Block Institution Big Five Opinion B-Size B-Independ. Equity Comp.
74
Table 5.5: Investment: Fraud Sample vs. Industry-Age Matched Sample The control sample is matched with the fraud sample based on two-digit SIC codes and ?rm ages (the number of years since IPO date).
Age [-1] R&D [-2,-1] R&D [0] R&D [1] Capital Exp. [-2,-1] Capital Exp. [0] Capital Exp. [1] Acquis(cf.) [-2,-1] Acquis(cf.) [0] Acquis(cf.) [1] Acquis.(cf.+stock) [-2,-1] Acquis.(cf.+stock) [0] Acquis.(cf.+stock) [1] Invest. C.F. [-2,-1] Invest. C.F. [0] Invest. C.F. [1]
Fraud Sample 2.89 (7.64) 0.00 (0.07) 0.00 (0.05) 0.00 (0.06) 0.05 (0.06) 0.04 (0.06) 0.04 (0.06) 0.00 (0.03) 0.00 (0.04) 0.00 (0.03) 0.01 (0.09) 0.01 (0.11) 0.00 (0.07) 0.11 (0.14) 0.11 (0.14) 0.07 (0.09)
# of obs. 630 627 630 577 608 614 560 589 579 553 595 587 544 604 611 553
Control Sample 2.90 (7.00) 0.00 (0.06) 0.00 (0.05) 0.00 (0.06) 0.05 (0.06) 0.04 (0.06) 0.04 (0.06) 0.00 (0.02) 0.00 (0.02) 0.00 (0.02) 0.00 (0.04) 0.00 (0.05) 0.00 (0.04) 0.09 (0.10) 0.07 (0.10) 0.07 (0.07)
# of obs. 630 573 571 565 552 548 545 544 536 523 550 542 509 547 537 511
Wilcoxon z 0.13 2.14* 2.11* 2.77** 0.87 1.03 1.57 3.78** 6.47** 4.65** 6.53** 9.00** 5.10** 3.81** 4.84** 1.67
75
Table 5.6: Pro?tability, Growth, & Fraud This table reports the relation between ?rms’ pro?tability, growth potential, external ?nancing need and their propensity to commit accounting fraud. Probit coe?cient estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **,* indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA Asset Growth B-M EF. Growth Res. ROA Log(Assets) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 1 P (F ) P ( D |F ) 1.09/0.40 (3.61)** 1.81/0.67 (5.27)**
Model 2 P (F ) P (D|F ) 1.11/0.10 (3.85)**
Model 3 P (F ) P (D|F ) 2.16/0.71 (4.81)**
-2.03/-0.19 (-10.24)** 1.88/0.62 (4.87)** -0.03/-0.01 (-0.41) 0.01/0.00 (2.49)* 0.03/0.01 (0.08) -0.41/-0.14 (-1.12) -0.63/-0.23 (-1.59) 0.25 (0.32)
-0.03/-0.01 (-0.44) 0.01/0.01 (2.57)* 0.19/0.07 (0.56) -0.36/-0.13 (-0.97) -0.57/-0.22 (-1.40) 0.02 (0.02)
-1.16/-0.10 (-6.07)** 0.07/0.01 (1.80) -0.00/-0.00 (-1.12) 0.25/0.02 (1.76) 0.24/0.02 (1.46) 0.51/0.06 (2.48)* -2.26 (-18.97)** -0.49 (0.06) -2431.56 112.43 (13) 50137
0.27/0.02 (7.97)** -0.01/-0.00 (-3.35)** -0.28/-0.02 (-2.12)* -0.16/-0.01 (-1.02) 0.37/0.04 (2.02)* -1.43 (-6.63)**
-5.72/-0.72 (-8.04)** -0.14/-0.02 (-3.87)** 0.02/0.00 (3.38)** 0.58/0.09 (5.05)** 0.53/0.08 (4.34)** 0.21/0.03 (1.68) -1.25 (-4.15)** -0.33 (0.05) -2201.25 425.41 (13) 45354
-1.13/-0.09 (-5.16)** 0.06/0.01 (1.77) -0.00/-0.00 (-0.98) 0.30/0.03 (2.23)* 0.26/0.02 (1.62) 0.51/0.06 (2.63)** -2.27 (-19.48)** -0.56 (0.06) -2415.48 107.44 (13) 49019
76
Table 5.7: External Financing & Fraud
ROA Asset Growth EF. C.F. Leverage ST Debt Res. ROA Log(Assets) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 4 P (F ) P ( D |F ) 0.39/0.07 (2.14)* 0.06/0.01 (1.99)* 2.63/0.46 (6.83)**
Model 5 P (F ) P (D|F ) 1.08/0.39 (3.35)** 1.80/0.66 (5.20)**
Model 6 P (F ) P (D|F ) 1.18/0.45 (3.94)** 1.80/0.69 (5.45)**
0.08/0.02 (0.35) 0.07/0.03 (0.21) -0.03/-0.01 (-0.33) 0.01/0.01 (2.24)* 0.21/0.08 (0.58) -0.36/-0.14 (-0.87) -0.57/-0.22 (-1.28) -0.12 (-0.17)
0.30/0.05 (9.15)** -0.02/-0.00 (-4.03)** -0.00/-0.00 (-0.02) -0.56/-0.09 (-2.64)** -0.49/-0.07 (-1.79) -2.51 (-9.67)**
-3.60/-0.71 (-8.75)** -0.18/-0.04 (-5.58)** 0.02/0.00 (4.45)** 0.43/0.10 (3.12)** 0.60/0.13 (3.90)** 0.72/0.19 (3.12)** -0.78 (-2.39)* -0.19 (0.32) -2335.23 352.63 (14) 49146
-0.03/-0.01 (-0.47) 0.01/0.01 (2.59)** 0.18/0.07 (0.55) -0.34/-0.13 (-0.93) -0.56/-0.22 (-1.39) 0.03 (0.04)
-1.17/-0.10 (-5.70)** 0.07/0.01 (1.82) -0.00/-0.00 (-1.11) 0.24/0.02 (1.75) 0.24/0.02 (1.44) 0.50/0.06 (2.45)* -2.25 (-18.75)** -0.50 (0.08) -2424.07 105.96 (14) 49585
-1.29/-0.11 (-6.53)** 0.07/0.01 (1.56) -0.00/-0.00 (-0.95) 0.24/0.02 (1.62) 0.25/0.02 (1.34) 0.51/0.06 (2.33)* -2.26 (-17.81)** -0.45 (0.07) -2417.36 124.91 (14) 49677
77
Table 5.8: Investment, Fraud Propensity & Detection This table reports the regression results on the relation between investment and fraud. Probit coe?cient estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **,* indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA EF. Growth R&D Invest. CF. Capital Exp. Acquis.(cf.) Acquis.(cf.+stock) Res. ROA Log(Asset) Age Tech Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 7 P (F ) P (D|F ) 1.74/0.29 (4.86)** 1.61/0.27 (3.43)** 3.42/0.58 -1.50/-0.13 (4.70)** (-5.06)** 0.99/0.17 -0.32/-0.03 (2.27)* (-2.10)*
Model 8 P (F ) P (D|F ) 1.84/0.35 (4.97)** 1.64/0.31 (3.48)** 3.71/0.71 -1.56/-0.14 (4.69)** (-4.91)**
Model 9 P (F ) P ( D |F ) 12.71/0.80 (4.48)** 2.27/0.67 (4.38)** 2.61/0.63 -1.57/-0.16 (3.24)** (-3.48)**
-0.57/-0.11 (-0.48) 1.19/0.23 (0.49)
0.15/0.01 (0.26) 0.38/0.03 (0.52)
-0.97/-0.29 (-0.70)
0.26/0.02 (0.46)
0.00/0.00 (0.02) 0.01/0.00 (1.90) -0.32/-0.06 (-1.18) -0.25/-0.04 (-0.86) -0.56/-0.12 (-1.68) 0.65 (1.18)
-1.03/-0.09 (-7.71)** 0.04/0.00 (1.18) -0.00/-0.00 (-0.75) 0.40/0.04 (3.10)** 0.24/0.02 (1.67) 0.46/0.05 (2.58)** -2.06 (-15.07)** -0.80 (0.00) -2300.92 134.03 (17) 43920
-0.01/-0.00 (-0.20) 0.01/0.00 (2.10)* -0.24/-0.05 (-0.82) -0.23/-0.05 (-0.82) -0.49/-0.12 (-1.44) 0.69 (1.25)
-1.13/-0.10 (-7.42)** 0.06/0.00 (1.64) -0.00/-0.00 (-0.81) 0.37/0.04 (2.55)* 0.24/0.02 (1.44) 0.44/0.05 (2.31)* -2.14 (-14.94)** -0.79 (0.00) -2187.11 129.77 (19) 41482
2.43/1.06 (2.45)* 0.05/0.01 (0.58) 0.01/0.00 (1.23) -0.19/-0.06 (-0.53) -0.36/-0.11 (-1.08) -0.28/-0.09 (-0.71) -0.19 (-0.17)
0.43/0.03 (1.61) -1.44/-0.10 (-4.93)** 0.03/0.00 (1.02) 0.00/0.00 (0.24) 0.37/0.03 (3.12)** 0.30/0.02 (2.29)* 0.32/0.03 (1.89) -2.23 (-15.21)** -0.54 (0.19) -2099.20 114.44 (19) 41930
78
Table 5.9: Investment, Fraud Propensity & Detection (Continued) Model 10 P (F ) P (D|F ) 1.75/0.25 (3.45)** 1.80/0.25 (3.00)** 3.11/0.44 -1.37/-0.10 (2.69)** (-3.14)** 0.80/0.11 -0.31/-0.02 (1.66) (-1.96)* 0.36/0.05 0.09/0.01 (1.63) (4.60)** Model 11 P (F ) P ( D |F ) 1.77/0.22 (3.87)** 1.83/0.23 (2.90)** 3.34/0.42 -1.42/-0.09 (3.35)** (-3.88)** 1.09/0.14 -0.41/-0.03 (2.29)* (-2.46)* 0.34/0.04 0.07/0.01 (1.58) (3.47)** -0.73/-0.09 0.37/0.02 (-2.67)** (3.80)** -1.02/-0.07 (-6.22)** 0.17/0.02 -0.03/-0.00 (1.82) (-1.28) -0.00/-0.00 0.00/0.00 (-0.29) (0.70) -0.59/-0.09 0.49/0.04 (-2.20)* (4.83)** -0.61/-0.09 0.35/0.03 (-1.84) (3.16)** -0.98/-0.21 0.59/0.06 (-2.56)* (3.44)** 0.23 -1.96 (0.28) (-16.00)** -0.75 (0.02) -2252.92 209.68 (21) 43920
ROA EF. Growth R&D Invest. CF. # of Acquis Focused Acquis Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
0.15/0.02 (1.62) -0.00/-0.00 (-0.45) -0.64/-0.11 (-2.57)* -0.57/-0.09 (-1.79) -0.76/-0.16 (-2.11)* 0.21 (0.20)
-1.04/-0.07 (-4.87)** -0.03/-0.00 (-1.04) 0.00/0.00 (0.80) 0.51/0.05 (5.27)** 0.33/0.03 (2.99)** 0.50/0.05 (3.10)** -1.94 (-14.84)** -0.75 (0.07) -2262.09 164.86 (19) 43920
79
Table 5.10: Insider Equity Incentive, Corporate Monitoring & Fraud
ROA EF. Growth R&D Invest. CF. Insider (Insider)2 Equity Comp. (Equity Comp.)2 Block Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 12 P (F ) P ( D |F ) 2.06/0.82 (5.05)** 1.63/0.65 (4.97)** 3.32/1.32 -2.11/-0.28 (3.52)** (-4.53)** 1.70/0.67 -0.54/-0.07 (2.90)** (-2.26)* 1.72/0.68 (2.90)** -2.45/-0.97 (-3.00)**
Model 13 P (F ) P (D|F ) 2.11/0.81 (5.76)** 1.75/0.68 (6.90)** 2.83/1.09 -1.89/-0.25 (2.83)** (-3.96)** 1.50/0.58 -0.69/-0.09 (2.50)* (-2.82)** 1.81/0.70 (2.96)** -2.28/-0.88 (-2.68)**
Model 14 P (F ) P (D|F ) 2.87/1.14 (3.80)** 2.09/0.84 (2.53)* 1.77/0.71 -2.40/-0.20 (1.50) (-2.46)* 1.63/0.65 -1.08/-0.18 (1.98)* (-2.04)*
2.35/0.94 (2.45)* -2.17/-0.86 (-1.88) -1.02/-0.40 (-2.40)* 0.84/0.11 (3.71)** -0.69/-0.27 (-2.03)* 0.23/0.09 (4.07)** 0.00/0.00 (0.04) 0.03/0.01 (0.10) -0.32/-0.12 (-0.99) -0.76/-0.26 (-2.23)* -2.01 (-3.32)** 1.08/0.14 (6.39)** -2.06/-0.28 (-5.93)** -0.08/-0.01 (-2.86)** 0.00/0.00 (0.71) 0.40/0.06 (2.66)** 0.50/0.08 (2.98)** 0.73/0.14 (3.44)** -1.54 (-8.94)** -0.36 (0.11) -1923.87 236.42 (21) 29330 -0.68/-0.27 (-1.25) 0.11/0.05 (1.31) 0.01/0.00 (1.49) 0.21/0.08 (0.43) -0.21/-0.08 (-0.40) -1.05/-0.38 (-1.79) -1.66 (-1.80) 0.35/0.06 (1.08) -3.42/-0.60 (-5.83)** -0.05/-0.01 (-1.02) 0.01/0.00 (2.86)** 0.43/0.09 (1.57) 0.79/0.18 (2.75)** 1.16/0.31 (2.48)* -1.53 (-3.40)** -0.51 (0.02) -765.64 115.10 (21) 8747
0.26/0.10 (3.81)** -0.01/-0.00 (-1.17) -0.29/-0.11 (-1.07) -0.59/-0.23 (-1.98)* -0.74/-0.27 (-2.33)* -1.59 (-2.66)**
-1.98/-0.27 (-7.20)** -0.06/-0.01 (-1.71) 0.01/0.00 (1.66) 0.60/0.10 (4.66)** 0.59/0.10 (4.06)** 0.65/0.12 (3.59)** -1.75 (-8.73)** -0.39 (0.05) -1999.62 256.28 (21) 29439
80
Table 5.11: Independent Auditor, Corporate Board & Fraud
ROA EF. Growth R&D Invest. CF. Insider (Insider)2 Block Big Five Opinion B-Size B-Independ. Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Model 15 P (F ) P ( D |F ) 2.08/0.83 (5.00)** 1.63/0.65 (5.38)** 3.33/1.33 -2.04/-0.24 (3.27)** (-4.39)** 1.67/0.66 -0.53/-0.06 (2.84)** (-2.25)* 1.61/0.64 (2.53)* -2.38/-0.95 (-2.69)** -1.04/-0.41 0.79/0.09 (-2.26)* (3.44)** -0.04/-0.01 0.30/0.03 (-0.16) (1.94) -0.01/-0.00 (-0.12)
Model 16 P (F ) P (D|F ) 1.72/0.63 (2.46)* 3.64/1.35 (4.89)** 5.28/1.95 -3.14/-1.24 (3.05)** (-2.58)** 1.42/0.52 -0.81/-0.32 (1.71) (-1.46) 1.21/0.45 (1.03) 1.45/0.54 (0.76) -0.26/-0.09 0.26/0.10 (-0.61) (0.66)
Model 17 P (F ) P (D|F ) 1.79/0.69 (2.56)* 3.60/1.38 (4.56)** 5.21/2.00 -3.01/-1.18 (3.12)** (-2.58)** 1.45/0.56 -0.78/-0.30 (1.71) (-1.43) 1.58/0.61 (1.33) 0.96/0.37 (0.51) -0.24/-0.09 0.19/0.08 (-0.58) (0.50)
-0.02/-0.01 (-0.35)
0.10/0.04 (1.91) -0.78/-0.30 (-0.98) 0.08/0.03 (0.70) 0.01/0.00 (1.70) -0.02/-0.01 (-0.05) -0.50/-0.18 (-0.96) -1.06/-033 (-1.78) -2.08 (-3.55)** 0.36/0.14 (0.48) -2.16/-0.85 (-3.96)** -0.30/-0.12 (-5.82)** 0.00/0.00 (0.62) 0.51/0.20 (1.91) 1.52/0.54 (4.52)** 1.86/0.59 (3.06)** 1.21 (2.68)** -0.38 (0.35) -482.16 105.99 (23) 2186
0.26/0.10 (3.79)** -0.01/-0.00 (-1.66) -0.22/-0.09 (-0.76) -0.43/-0.17 (-1.33) -0.69/-0.25 (-1.99)* -1.52 (-2.32)*
-1.94/-0.23 (-7.01)** -0.07/-0.01 (-2.19)* 0.01/0.00 (1.78) 0.53/0.08 (3.82)** 0.47/0.07 (3.05)** 0.60/0.10 (3.07)** -1.97 (-8.18)** -0.38 (0.42) -1833.71 229.88 (24) 29225
0.09/0.03 (0.65) 0.01/0.00 (1.78) 0.09/0.03 (0.21) -0.32/-0.11 (-0.52) -0.88/-0.27 (–1.28) -2.01 (-3.52)**
-2.28/-0.90 (-3.90)** -0.36/-0.14 (-5.61)** 0.00/0.00 (0.69) 0.49/0.19 (1.92) 1.43/0.51 (4.51)** 1.83/0.58 (3.02)** 1.34 (2.70)** -0.30 (0.56) -481.81 109.18 (23) 2186
81
Table 5.12: Bivariate Probit Model vs. Straight Probit Model This table compares the results from the following two statistical models: Bivariate probit model: P (Zi = 1) = P (Fi = 1)P (Di = 1); Straight probit model: P (Zi = 1) = P (Fi = 1). Both models are estimated using a random comparison sample without repetition. That is, every comparison ?rm only enters the estimation once. The probit coe?cent estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **, * indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2). Probit P (F ) 1.04/0.18 (4.16)** 0.24/0.04 (6.97)** -0.49/-0.08 (-1.17) 0.59/0.10 (2.59)** 1.56/0.27 (3.74)** -1.74/-0.30 (-2.88)** 0.98/0.17 (7.58)** -2.12/-0.36 (-8.09)** 0.04/0.01 (2.18)* 0.00/0.00 (1.13) 0.53/0.11 (6.17)** 0.43/0.08 (5.34)** 0.27/0.05 (2.73)** -2.39 (-17.78)** -1100.36 275.66 (13) 3336 Bivariate Probit P (F ) P (D|F ) 2.34/0.65 (5.45)** 2.65/0.74 (5.41)** 3.51/1.29 -2.71/-0.97 (3.42)** (-4.72)** 1.46/0.43 -0.80/-0.25 (2.59)** (-2.54)** 2.33/0.71 (2.44)* -2.74/-0.84 (-2.70)** -0.61/-0.20 1.30/0.48 (-1.96)* (5.82)** -2.94/-0.85 (-6.31)** 0.28/0.07 -0.11/-0.03 (4.12)** (-2.86)** 0.00/0.00 0.01/0.00 (0.26) (2.03)* 0.13/0.04 0.41/0.17 (0.44) (2.13)* -0.17/-0.07 0.50/0.17 (-0.55) (2.48)** -0.53/-0.30 0.66/0.37 (-1.61) (2.58)** -2.26 -0.44 (-4.32)** (-1.88) -0.49 (0.03) -1007.66 219.26 (21) 3336
ROA EF. Growth R&D Invest. CF. Insider Own (Insider)2 Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
82
Table 5.13: Frivolous Lawsuits This table presents robustness checks of the main results over three subsamples: (1) 334 out of 660 ?rms that announced accounting restatements before or after the lawsuits; (2) 207 out of 660 ?rms that were subject to both private class action litigation and the SEC’s Accounting and Auditing Enforcement; (3)exclusion of 27 nuisance cases. A case is considered as a nuisance case if it is later dismiss by the court or if it leads to a less than two million dollar settlement. The probit coe?cent estimates/marginal e?ects and their t-statistics (in parentheses), the Wald Chi-squared statistics and the degree of freedom (in parentheses) are reported. **, * indicate signi?cance at 1 and 5% levels, respectively. ? is correlation between u and v in equations (1) and (2).
ROA EF. Growth R&D Invest. C.F. Insider (Insider)2 Institution Res. ROA Log(Asset) Age Tech. Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Restatements P (F ) P (D|F ) 2.55/0.66 (4.39)** 1.89/0.49 (5.48)** 3.31/0.86 -2.01/-0.14 (3.33)** (-3.83)** 1.97/0.51 -0.53/-0.04 (2.78)** (-1.96)* 1.91/0.50 (2.09)* –2.27/-0.59 (-1.82) -0.63/-0.16 1.11/0.08 (-1.50) (5.49)** -2.47/-0.17 (-7.07)** 0.18/0.05 -0.05/-0.00 (1.83) (-0.99) 0.01/0.00 0.00/0.00 (0.97) (0.20) 0.14/0.04 0.35/0.03 (0.37) (1.75) -0.17/-0.4 0.26/0.02 (-0.38) (1.15) -0.74/-0.14 0.73/0.08 (-1.65) (2.82)** -2.62 -2.06 (-4.17)** (-9.04)** 0.02 (0.93) -1141.10 196.18 (21) 29117
SEC Enforcement P (F ) P ( D |F ) 2.40/0.66 (3.62)** 1.45/0.40 (3.55)** 3.18/0.87 -2.25/-0.14 (1.98)* (-2.11)* 2.00/0.55 -0.96/-0.06 (1.84) (-2.00)* 1.83/0.50 (1.60) -2.58/-0.71 (-1.40) -0.88/-0.24 1.21/0.08 (-0.99) (2.93)** -2.31/-0.15 (-4.49)** 0.09/0.03 -0.01/-0.00 (0.66) (-0.09) 0.01/0.00 -0.00/-0.00 (1.38) (-0.65) 0.45/0.13 -0.01/-0.00 (1.01) (-0.02) 0.17/0.05 -0.15/-0.01 (0.17) (-0.23) -0.21/-0.05 0.26/0.02 (-0.26) (0.48) -2.20 -1.98 (-1.83) (-6.82)** -0.15 (0.72) -740.56 113.22 (21) 29021 83
Non-Nuisance Suits P (F ) P (D|F ) 1.99/0.77 (5.60)** 1.67/0.64 (6.84)** 2.81/1.08 -1.90/-0.25 (2.49)** (-3.49)** 1.31/0.51 -0.68/-0.09 (2.02)* (-2.51)* 1.78/0.69 (2.96)** -2.22/-0.85 (-2.62)** -0.73/-0.28 1.12/0.15 (-2.02)* (6.38)** -2.05/-0.27 (-5.19)** 0.23/0.09 -0.08/-0.01 (4.06)** (-2.79)** 0.00/0.00 0.00/0.00 (0.14) (0.74) 0.05/0.02 0.39/0.06 (0.16) (2.46)* -0.31/-0.12 0.50/0.08 (-0.95) (2.72)** -0.76/-0.26 0.74/0.15 (-2.20)* (3.28)** -1.96 -1.55 (-3.15)** (-8.78)** -0.37 (0.11) -1859.72 239.00 (21) 29312
Table 5.14: Di?erent Model Speci?cations This table presents robustness checks of the main results across di?erent model speci?cations. “Ind. Lit” is the logarithm of the total market value of fraudulent ?rms in an industry in year -1. “Cycle”=-1 for years between 1992 and 1994 and between 2001 and 2002 (bust), =0 for years between 1995 and 1997, and =1 for years between 1998 and 2000 (boom).
ROA EF. Growth R&D Invest. C.F. Insider (Insider)2 Block Ind. Lit. (*103 ) Cycle Res. ROA Log(Asset) Age Tech Service Trade Constant ? (p-value) Log Likelihood ?2 (d.f.) # of obs.
Speci?cation 2 P (F ) P (D|F ) 1.90/0.75 (4.93)** 1.56/0.62 (5.42)** 3.07/1.22 -1.93/-0.23 (2.98)** (-3.79)** 1.68/0.67 -0.60/-0.07 (2.92)** (-2.49)* 1.71/0.68 (2.88)** -2.41/-0.96 (-2.95)** -1.00/-0.40 0.78/0.09 (-2.21)* (3.24)** -0.77/-0.31 0.70/0.08 (-0.99) (1.75) -0.21/-0.08 0.31/0.04 (-1.73) (5.70)** -1.89/-0.23 (-6.43)** 0.26/0.10 -0.05/-0.01 (3.46)** (-1.48) -0.01/-0.00 0.01/0.00 (-1.04) (1.50) -0.24/-0.10 0.52/0.08 (-0.93) (3.96)** -0.54/-0.21 0.52/0.08 (-1.81) (3.51)** -0.82/-0.31 0.67/0.12 (-2.51)* (3.66)** -1.21 -1.92 (-1.79) (-9.05)** -0.45 (0.04) -1959.12 291.98 (25) 29439
Speci?cation 3 P (F ) P ( D |F ) 1.61/0.64 (2.66)** 1.34/0.53 -0.03/-0.00 (2.52)* (-0.67) 2.78/1.10 -1.97/-0.29 (2.78)** (-3.10)** 1.58/0.63 -0.66/-0.10 (2.53)** (-1.80) 2.45/0.97 -0.59/-0.09 (3.26)** (-1.85) -2.31/-0.91 (-2.91)** -1.31/-0.52 1.04/0.15 (-2.65)** (2.76)** -1.00/-0.40 0.90/0.13 (-1.20) (1.64) -0.19/-0.07 0.30/0.04 (-1.19) (4.34)** -1.86/-0.28 (-5.88)** 0.28/0.11 -0.08/-0.01 (2.60)** (-1.00) -0.01/-0.00 0.01/0.00 (-0.70) (0.91) -0.24/-0.10 0.52/0.09 (-1.01) (3.71)** -0.54/-0.21 0.53/0.10 (-1.90) (2.91)** -0.84/-0.32 0.73/0.15 (-2.73)** (3.45)** -1.21 -1.63 (-1.85) (-3.26)** -0.55 (0.02) -1955.92 311.90 (27) 29439 84
Speci?cation 4 P (F ) P (D|F ) 0.39/0.04 (1.97)* 0.55/0.06 (2.27)* 2.34/0.26 -1.56/-0.20 (3.69)** (-4.20)** 1.02/0.11 -0.65/-0.08 (3.39)** (-3.37)** 0.44/0.05 (1.98)* -0.65/-0.07 (-1.93) -0.75/-0.08 0.71/0.09 (-1.98)* (2.26)* -0.97/-0.11 0.89/0.12 (-2.02)* (2.54)* -0.38/-0.04 0.38/0.05 (-6.05)** (7.00)** 0.04/0.00 (0.83) 0.00/0.00 (0.16) -0.43/-0.06 (-2.36)* -0.50/-0.07 (-2.61)** -0.81/-0.14 (-3.49)** 1.71 (5.18)** 0.00/0.00 (0.04) 0.00/0.00 (0.06) 0.46/0.07 (3.28)** 0.48/0.08 (3.24)** 0.72/0.14 (3.77)** -2.05 (-11.09)** -0.99 (0.00) -2183.13 336.68 (24) 31696
Figure 3: Determination of the Beginning Fiscal Year of Fraud
Fiscal year ending month Class Period
12
6 Year 0
12
Fiscal year ending month
Class Period
4 Year 0
4
6
4
Figure 4: Identification Problem
No fraud (F=0) Z=0 The Firm Not detected (D=0|F=1) Fraud (F=1) Detected (D=1|F=1) Z=1
85
Figure 5: Timing of Fraud
Historic vs. Restated ROA 4 3 Median ROA 2 1 0 -1 -2 -3 -4 Fiscal Year -2 -1 0 1 2
Fraud-Historic Fraud-Restated Control-Historic Control-Restated
Note: ROA is the ratio of net income over total assets. I use net income because the restated information on this variable is more complete than the one on other accounting measures such as income before extraordinary items and operating income. The purpose here is to compare the originally reported data with the restated data.
86
Chapter 6 Conclusion
This thesis analyzes corporate securities fraud and its consequences. The theory model shows that fraud can lead to investment distortions in both fraudulent ?rms and honest ?rms, which is the real economic cost of fraud. The investment distortion is twofold. On one hand, fraud can in?ate short-term ?rm value and allow the ?rm to invest using cheap outside capital. On the other hand, once committed fraud, the ?rm has incentive to strategically use investment to mask fraud. The incentive to disguise fraud can not only induce the ?rm to overinvest, but also gives the ?rm a preference for risk and suboptimal diversi?cation. The theory model also characterizes the endogenous cost-bene?t tradeo? of committing fraud and derives the ?rm’s equilibrium disclosure strategy. The model shows that the cost and bene?t of fraud are endogenously related, which results in the optimal size of fraud and the ?rm’s equilibrium fraud propensity. In particular, the theory demonstrates the important role of the endogenous detection risk in determining the cross-sectional variations in ?rms’ fraud incentives. The model generates testable implications about the economic determinants of cross-sectional di?erences in fraud propensities and the relationship between fraud and corporate investment incentives. The theory predicts that fraudulent ?rms tend to have good growth prospects, but experience negative pro?tability shocks. Litigation events tend to cluster in certain industries during some speci?c time period. The theory also predicts that fraudulent ?rms tend to overinvest. Investment can negatively in?uence the ?rm’s litigation risk. The type of investment that introduces the most valuation imprecision has the strongest e?ect on the likelihood of fraud detection. The investment, however, can be ine?cient and can result in long-term underperformance of fraudulent ?rms. I also empirically investigate the economic determinants of ?rms’ propensity to commit accounting fraud and the probability of fraud detection, using a sample of public companies that
87
were subject to federal private securities class action litigation between 1996 and 2003. I use econometric methods to control for the unobservability of undetected frauds, and disentangle the e?ect of a factor on a ?rm’s probability of committing fraud and its e?ect on the ?rm’s probability of being detected. The separation of fraud commitment and fraud detection allows me to examine the economics of each probability as well as their interactions. The results of this study show that investment, strength of corporate monitoring, insider equity incentives, and some ?rm characteristics signi?cantly in?uence a ?rm’s cost-bene?t tradeo? of engaging in fraud. First, the level, type, and ?nancing of investment types of investment all matter in dertermining a ?rm’s ex-post probability of fraud detection and ex-ante propensity to commit fraud. Second, di?erent types of corporate monitors have di?erent e?ects on ?rms’ fraud incentives. The presence of block equity holders and large institutional ownership tends to increase the likelihood of fraud detection and discourage fraud. The roles of independent auditors and board of directors appear to be weaker. Third, there is a concave relation between insider equity incentive and the probability of fraud. When insider equity incentive is small, increasing equity incentive can have the unintended e?ect of increasing the probability of fraud. When insider equity incentives is already large, such e?ect disappears. This implies that insider equity incentive can be a double-edged sword when it is used to align managerial and shareholder interests in dispersely-owned ?rms. Finally, high growth potential, large external ?nancing need, and (hidden) negative pro?tability shocks seem to be important motivational factors for fraud. This study also demonstrates the importance of disentangling the probability of committing fraud and the probability of detecting fraud, because cross-sectional variables can have opposing e?ects on the two latent probabilities, and therefore can be masked in their overall e?ect on the incidence of detected fraud. Ignoring this structure can lead us to draw incorrect inferences about the determinants of corporate fraud.
88
Chapter 7 Appendix: Proofs of Propositions
Proof of Proposition 1 The market value of the ?rm at time 2 after the investment announcement is V2 (I, y ) E (R|I ) zc = E (V3 |I, y ) = E (A|I, y ) + IE (R|I ), = E (R|R > rc ) = R + I?R m(zc ), = (rc ? R)/?R , ?(zc ) . 1 ? ?(zc )
m(zc ) = The investment condition is
(1 ? ? )[E (A|e) + Ir] ? PI f ? > E (A|e) ? PN f ?, where ? = I/V2 (I, y ). Solving for r, we get r> E (A|e) E (A|I, y ) + I?R m(zc ) ? (PN ? PI )f ? . (1 ? ? )I
(7.1)
(7.2)
This leads to equation (4.23). The left-hand side of equation (4.23) monotonically increases in rc , while the right-hand side monotonically decreases in rc . Therefore, there exists a unique solution
? to equation (4.23), rc .
Proof of Proposition 2 PI < PN if and only if ?I < ?N , or equivalently vc,I +KI < vc,N +KN . vc,I +KI is a function of I?R , while vc,N + KN is not. Let M = e + C/f ? y , and ?I = cov (e, V3 |I )/(?e Take the derivative of vc,I + KI with respect to I?R . ? (vc,I + KI ) ? (I?R ) = = ?vc,I ?KI + ? (I?R ) ? (I?R ) M ??I ??I + [E (V3 |y ) ? E (V3 |e)] . 2 ?e ?I ? (I?R ) ? (I?R ) 89 (7.3) (7.4) V ar(V3 |I )).
? 4 2 ?u +(2q?e I?R )2 ??u ??I ?e I If max(?1, ? qI? ) < ? < < 1, then ? (?? 2q?e I?R I?R ) < 0 and ? (I?R ) < 0, and therefore R ? 4 2 ?u +(2q?e I?R )2 ??u ? (vc,I +KI ) ??I I < 0. If ? ? ? 1, then ? (?? ? (I?R ) 2q?e I?R I?R ) > 0 but ? (I?R ) < 0. Therefore, there ? 4 2 ? +(2q?e I?R )2 ??u ? (v +KI ) ?e exists ? ? [ u 2q?e I?R , 1] such that when max(?1, ? qI? ) < ? < ?, ?c,I < 0. Since (I?R ) R vc,N + KN does not depend on I?R and vc,I + KI decreases with I?R , there exists a cuto? value I?R , such that when I?R ? I?R , vc,I + KI ? vc,N + KN . Proof of Proposition 3 Note that E (A|I, y ) + I?R m(zc ) = V2 (I, y ) ? I . Let us take derivative with respect to ? on both sides of equation (4.23). ?rc ?? = ? ? E (A|e) ?V2 (I, y ) (PN ? PI )f ? + (PN ? PI )f ? [V2 (I, y ) ? I ]2 ?? (1 ? ? )I (PN ? PI )f ? (???/?? ) . (1 ? ? )2 I ?PN ?PI ?V2 (I, y ) ? = (1 ? p)(?N ?N ? ?I ?I ) , ?? ?? ??
(7.5)
PN ? PI = where ? = ? ?(s)/?s. ? =
I V2 (I,y )
is the fractional ownership of the new shareholders. V2 (I, y ) does not directly
depend on ? , since the market does not observe ? . From the manager’s point of view, however, what is important is how much V2 (I, y ) will be di?erent from V2 (I, e) if the manager reports one more unit of earnings above the true realization e. So let us de?ne ?V2 (I, y ) V2 (I, y ) ? V2 (I, e) = lim . ?? y?e (y ?e)?0 Then ?? I ?V2 (I, y ) = (? ). 2 ?? V2 (I, y ) ?? (7.7) (7.6)
If y (e) = e does not generate any e?ect on the market valuation (i.e., V2 (I, y ) = V2 (I, e)), then ??/?? = 0. As long as misreporting can increase the market value of the ?rm’s assets, i.e., then ??/?? < 0. Substitute these relations into (7.5), and we have
?rc ?? ?V2 (I,y ) ??
? 0,
< 0.
Proof of Proposition 4
90
The ?rst-order condition for the maximization problem (4.20) is ?? +g ?? g? = = 0; 0, (7.8) (7.9)
where g is the Lagrange multiplier for the nonnegativity constraint on ? . ?? ?? ?z ?rc )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?rc + ?[1 ? ?(zc )][(? )E (V3 |I, e) + (1 ? ? )Im (zc ) ] ?? ?? ?z ?rc ? {? (PN ? PI ) + ?[1 ? ?(zc )]PI + (1 ? ?[1 ? ?(zc )])PN }f ? ?R ?? = ?( ? ? P f. (7.10)
The following steps present the derivations of the equilibrium strategy speci?ed in Proposition 4. Step 1: A Conjecture. Suppose that there exists a cuto? earnings realization ec such that
the manager will honestly reveal the earnings if the true earnings realization is above ec , and the manager would like to overreport earnings if the true realization is below ec . Mathematically, y (e) = e or ? (e) = 0, if e ? ec ; y (e) > e or ? (e) > 0, if e < ec . Given the above conjecture about the manager’s fraud incentives, the market’s reaction to an earnings announcement can be as follows. When investors observe the announced earnings y (e), they rationally infer e = y (e) ? ? , using their prior belief about the probability of misreporting ?0 . ? is the market’s expected amount of misreporting. The time 1 conditional probability of fraud is ?1 = P rob.(misreporting |y ? ec ). Therefore, whenever y ? ec , investors believe that e = y > ec , with probability (1 ? ?1 ); e = y ? ? 1 < ec , with probability ?1 . When investors observe y < ec , they rationally discount the earnings announcement, and e = y ? ? 2 . Then the market value of the ?rm’s assets in place after the earnings announcement is V1 (y ? ec ) V1 (y < ec ) = (1 ? ?1 )E (A|e = y ) + ?1 E (A|e = y ? ? 1 ); (7.11) (7.12)
= E (A|e = y ? ? 2 ). 91
? 1 and ? 2 are the market’s expected amount of misreporting given y ? ec and y < ec , respectively. In equilibrium, they should be equal to the manager’s optimal choice of misreporting in the two earnings announcement scenarios. ? ? 0 and the structure of litigation cost of fraud naturally leads to a conjecture that ? (e) is monotonic in e in each di?erent region speci?ed above. This does not imply, however, that y (e) is always monotonic in e (due to the pooling of the two types of ?rm). Then in each of the two scenarios (fraud or honest) there is a one-for-one mapping between e and y (e). This implies that under each scenario, e = y (e) ? ? is still normally distributed. Therefore, given the true realization of earnings e, when y ? ec , ?V2 (I, y ) = ? (1 ? ?1 ) > 0. ?? When y < ec , ?V2 (I, y ) = 0. ?? Step 2: Deriving ec . (7.14) (7.13)
Let us plug equation (7.13) and (7.14) into (7.10) and di?erentiate with
respect to ? on both sides. Then use the following relationships: ?rc ?? PN PN PN PI we can ?nd that ?2? < 0. ?? 2 This means that the objective function is globally concave. There exists a unique maximizer
? ? ? = ?1 . The concavity and the nonnegative ? constraint imply that
< = = =
0; (1 ? p)? (1 ? ?1 )?N ?N > 0; (1 ? p)? (1 ? ?1 )?N |vc,N + KN |?N > 0; (1 ? p)? (1 ? ?1 )?I ?I > 0;
= (1 ? p)? (1 ? ?1 )?I |vc,I + KI |?I > 0,
?? ? |?=0 > 0 ? ?1 > 0, ?? ?? ? |?=0 ? 0 ? ?1 = 0. ?? 92
I de?ne the following notations. ?0 = ? (y = e) = I/V2 (I, e), rc,0 = rc (? = 0), zc,0 = (rc,0 ? R)/?R , ?0 = ?(zc,0 ), ?0 = ?(zc,0 ), and m0 = m(zc,0 ). Then plug ? = 0 into equation (7.10), and we have ?? ?rc |?=0 = ?(1 ? ?0 ){Im0 (? |?=0 )[?0 m0 + (1 ? ?0 )zc,0 ] + ?0 ? (1 ? ?1 )} ? pf, ?? ?? where ?rc E (A|e)? (1 ? ?1 ) |?=0 = ? . ?? [V2 (I, e) ? I ]2 The ?rst term on the right-hand side of equation (7.15) decreases as e increases, while the second term does not depend on e. Therefore, we can ?nd a cuto? ec , such that
?? ?? |? =0 ?? ?? |? =0
(7.15)
> 0 if e < ec , and
? 0 if e ? ec . ec is the solution to ?? |?=0 = 0. ??
Step 3: Deriving eh .
To facilitate the analysis below, it is convenient to decompose
?? ??
into
a marginal bene?t of fraud term and a marginal cost of fraud term. Let MB ?z ?rc )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?m(zc ) ?rc + ?[1 ? ?(zc )][(? )E (V3 |I, e) + (1 ? ? )I ]; ?? ?zc ?? ?z ?rc ?PI ?PN = {? (PN ? PI ) + ?[1 ? ?(zc )] + (1 ? ?[1 ? ?(zc )]) }f ? ?R ?? ?? ?? = ?( ? + P f. (7.16)
MC
(7.17)
Then let us take the ?rst and the second derivatives of both MB and MC with respect to e. We can ?nd that
?M B ?e ?M C ?e
< 0, > 0,
?2M B ?e2 ?2M C ?e2
< 0; > 0.
The relations about the ?rst derivatives mean that when the true earnings realization is low, the marginal bene?t of fraud is relatively high, while the marginal cost of fraud is relatively low. This implies that
? ??1 < 0. ?e
The relations about the second derivatives imply that
? ? 2 ?1 > 0. 2 ?e
93
? Given that ?1 (e) is a decreasing and concave function of e, there exists
? (e)]. eh ? max [e + ?1 e eh , the market rationally believes that y = e. Step 4: Deriving el .
? Similarly, given that ?1 (e) is a decreasing and concave function of
? e, there also exists a lower bound el such that when e < el , y (e) = e + ?1 (e) < ec , and when ? el ? e < ec , y (e) = e + ?1 (e) ? ec . el is the solution to the following equation:
M B [?1 (el )] = M C [?1 (el )], where ?1 (el ) = ec ? el . When the ?rm announces y < ec , however, the market reaction changes, because now the low-earnings ?rm is not pooled with the high-earnings ?rm. In other words, the bene?t-cost tradeo? is di?erent, which implies that the optimal amount of misreporting in this region should
? . be di?erent from ?1
Let ?2 (e) be the manager’s misreporting strategy when e < el . If ?2 (e) is a monotonic function of e, then y (e) = e + ?2 (e) < ec is also a monotonic function of e. In other words, y (e) is a su?cient statistic of e, and thus E (A|y ) = E (A|e). Substitute E (A|y ) = E (A|e) into equations (4.17), (4.18), (4.23), (7.16) and (7.17), and we get
?rc ??
= 0, M B = 0, and M C = pf > 0. Since the
marginal bene?t is less than the marginal cost regardless of ? , the optimal amount of misreporting
? = 0. Put di?erently, if the manager chooses a monotonic disclosure strategy when e < el , is ?2
then the optimal monotonic strategy is y (e) = e. Step 5: Possibility of ? (e) as a nonmonotonic function of e. Let us also consider whether
there exists an equilibrium in which ?2 (e) is a nonmonotonic function of e. Since ? ? 0 (which means y (e) ? e), and the litigation cost is an increasing and monotonic function of ? , I can make the following conjecture about ?2 (e). I can partition the earnings space {e : e < el } into many intervals, [e1 , el ), [e2 , e1 ), [e3 , e2 ).... In each earnings interval, y (e) equals the upper bound of that interval. The lower bound of each interval is determined, such that the earnings realization at the 94
lower bound plus the optimal amount of misreporting equals the upper bound earnings value. Take the ?rst interval [e1 , el ) for an example. If the true earnings realization is in this interval, then the manager announces y (e) = el . The market rationally infers that e = E (e|e1 ? e < el ) and uses e to price the ?rm’s assets in place. It is easy to see that ?rms with e < e < el get worse o? by reporting y (e) = el than reporting y (e) = e, because the ?rm’s asset value is underpriced by the market, and the ?rm faces potential litigation cost. Then these ?rms would rather honestly reveal their earnings, and the conjectured equilibrium collapses. This happens to any nonmonotonic ?2 (e). Proof of Proposition 5 1. e 2. ? ?M B ?? = ?z ?rc (? )[(1 ? ? )E (V3 |I, e) ? E (V3 |N, e)] ?R ?? ?? ?rc + (1 ? ?(zc ))[(? )E (V3 |I, e) + (1 ? ? )Im (zc ) ] ?? ?? (7.18) (7.19) Proof is shown in the proof of Proposition 4 (step 3).
> 0; ?M C ?? = ??[1 ? ?(zc )](PN ? PI )f ? < 0.
? (e) Since the marginal bene?t of fraud increases in ? and the marginal cost decreases in ?, ?1
increases in ? for any given e. This also implies that ?M B |?=0 ?? = (1 ? ?0 ){Im0 (? > 0; ?M C |?=0 ?? This implies that = 0.
?ec ??
?el ??
< 0.
?rc |?=0 )[?0 (m0 + (1 ? ?0 )zc,0 ] + ?0 ? (1 ? ?1 )} ?? (7.20) (7.21)
> 0.
95
BIBLIOGRAPHY [1] Agrawal, Anup, and Sahiba Chadha, 2004, “Corporate Governance and Accounting Scandals,” Journal of Law and Economics Forthcoming. [2] Agrawal, Anup, Je?rey F. Ja?e, and Jonathan M. Karpo?, 1999, “Management turnover and corporate governance changes following the revelation of fraud,” Journal of Law and Economics April, 309-342. [3] Alexander, Cindy R., and Mark A. Cohen, 1999, “Why do corporations become criminals? Ownership, hidden actions, crime as an agency cost,” Journal of Corporate Finance 5, 1-34. [4] American Institute of Certi?ed Public Accountants, National Commission on Fraudulent Financial Reporting (AICPA), 1987, Reporting of the National Commission on Fraudulent Financial Reporting, New York, NY. [5] Bajaj, Mukesh, Katherine Gunny, and Atulya Sarin, 2003, “Auditor compensation and auditor failure: An empirical analysis,” working paper, University of California-Berkeley. [6] Baucus, Melissa S., and David A. Baucus, 1997, “Paying the piper: An empirical examination of long-term ?nancial consequences of illegal corporate behavior,” Academy of Management Journal 40 (February), 129-151. [7] Beasley, Mark S., 1996, “An empirical analysis of the relation between the board of director composition and ?nancial statement fraud,” The Accounting Review 71, No. 4, 443-465. [8] Beasley, Mark S., Joseph V. Carcello, and Dana R. Hermanson, 1999, “Fraudulent ?nancial reporting: 1989-1997,” Research commissioned by the Treadway Commission. [9] Bebchuk, Lucian A., and Oren Bar-Gill, 2002, “Misreporting corporate performance,” working paper, Harvard University. [10] Becker, Gary S., 1968, “Crime and punishment: An economic approach,” Journal of Political Economy , 169-217.
96
[11] Bertrand, M., and S. Mullainathan, 2001, “Are CEOs rewarded for luck? The ones without principals are,” Quarterly Journal of Economics 116, 901-932. [12] Bethel, Jennifer, J. Liebeskind, and T. Opler, 1998, “Block share purchases and corporate performance,” Journal of Finance 53, 605-635. [13] Burns, Natasha, and Simi Kedia, 2004, “Do executive stock options generate incentives for earnings management? Evidence from accounting restatements,” working paper, Harvard University. [14] Chan, Konan, Louis K. C. Chan, Narasimhan Jegadeesh, and Josef Lakonishok, 2005, “Earnings quality and stock returns,” Journal of Business Forthcoming. [15] Choi, Stephen J., 2004, “Do the merits matter less after the Private Securities Litigation Reform Act?” working paper, University of California-Berkeley. [16] Co?ee, John C. Jr., 1991, “Liquidity versus control: The institutional investor as corporate monitor,” Columbia Law Review 91, No. 6, 1277-1367. [17] Cox, James D, and Randall S. Thomas, 2003, “SEC enforcement heuristics: An empirical inquiry,” Duke Law Journal 53, 737-780. [18] Dechow, Patricia M., Richard G. Sloan, and Awy P. Sweeney, 1996, “Causes and consequences of earnings manipulation An analysis of ?rms subject to enforcement actions by the SEC,” Contemporary Accounting Research 13, No. 1, 1-36. [19] DeFond, M. and J. Jiambalva, 1994,“Debt Covenant E?ects and the Manipulation of Accruals,” Journal of Accounting and Economics 17, 145-176. [20] Demirg¨ uc ¸-Kunt, Asli and Vojislav Maksimovic, 1998, “Law, ?nance, and ?rm growth,” Journal of Finance 53, 2107-2137. [21] Dye, Ronald A., 1988, “Earnings management in an overlapping generations model,” Journal of Accounting Research 26, No. 2, 195-235.
97
[22] Efendi, Jap, Anup Srivastava and Edward P. Swanson, 2004, “Why do corporate managers misstate ?nancial statements? The role of option compensation, corporate governance, and other factors,” working paper, Texas A&M University. [23] Erickson, Merle., and Shiing-wu Wang, 1999, “Earnings management by acquiring ?rms in stock for stock mergers,” Journal of Accounting and Economics 18, 3-28. [24] Feinstein, Jonathan S., 1990, “Detection Controlled Estimation,” Journal of Law and Economics 33, No. 1, 233-276. [25] Feroz, Ehsan, Kyungjoo Park, and Victor Pastena, 1991, “The ?nancial and market e?ects of the SEC’s accounting and auditing enforcement releases,” Journal of Accounting Research 29, 107-142. [26] Frankel, Richard M., Marilyn F. Johnson, and Karen K. Nelson, 2002, “The relation between auditors’ fee for non-audit services and earnings management,” Accounting Review 77 Supplement, 71-105. [27] Gande, Amar, and Craig M. Lewis, 2005, “Shareholder initiated class action lawsuits: Shareholder wealth e?ect and industry spillovers,” working paper, Vanderbilt University. [28] Gao, Pengjie, and Ronald E. Shrieves, 2003, “Earnings management and executive compensation: A case of overdose of option and underdose of salary?” working paper, University of Tennessee. [29] oldman, Eitan, and Steve L. Slezak, 2003, “The economics of fraudulent misreporting,” working paper, University of North Carolina at Chapel Hill. [30] Healy, Paul M., and James M. Wahlen, 1999, “A review of the earnings management literature and its implications for standard setting,” Accounting Horizons 13, No. 4, 365-383. [31] Hermalin, Benjamin E., and Michael Weisbach, 2003, “Boards of directors as an endogenously determined institution: A survey of the economic literature,” Economic Policy Review April, 7-26. 98
[32] Holderness, Cli?ord G., 2003, “A survey of blockholders and corporate control,” Economic Policy Review April, 51-64. [33] Jensen, Michael C., 2004,“Agency costs of overvalued equity,” working paper, Harvard University. [34] John, Kose, and Lemma W. Senbet, 1998, “Corporate governance and board e?ectiveness,” Journal of Banking & Finance 22, 371-403. [35] Johnson, Shane A., Harley E. Ryan, and Yisong S. Tian, 2003, “Executive compensation and corporate fraud,” working paper, Louisiana State University. [36] Johnson, Marilyn F., Karen K. Nelson, and A.C. Pritchard, 2002, “Do the merits matter more? Class action under the Private Securities Litigation Reform Act,” working paper, Michigan State University. [37] Lacker, Je?rey M., and John A. Weinberg, 1989, “Optimal contracts under costly state falsi?cation,” The Journal of Political Economy 97, No. 6, 1345-1363. [38] Li, Si, 2004, “Corporate fraud and costly monitoring: An empirical analysis of a simultaneous system with partial observability,” working paper, Duke University. [39] Klein, April, 2002, “Audit committee, board of director characteristics, and earnings management,” Journal of Accounting & Ecnomics 33, No. 3, 375-400. [40] Leuz, Christian, Alexander Triantis, and Tracy Yue Wang, 2004, “Why do ?rms go dark? Causes and economic consequences of voluntary SEC deregistrations,” working paper, University of Maryland. [41] Loebbecke, James K., Martha M. Eining, and John J. Willingham, 1989, “Auditors’ experience with material irregularities: Frequency, nature, and detectability,” Auditing: A Journal of Practice & Theory 9, No. 1, 1-28. [42] Maksimovic, Vojislav, and Sheridan Titman, 1991, “Financial policy and reputation for product quality,” The Review of Financial Studies 4, No. 1, 175-200. 99
[43] Moeller, Sara, Frederik P. Schlingemann and Rene M. Stulz, 2004, “Wealth destruction on a massive scale? A study of acquiring-?rm returns in the recent merger wave,” Journal of Finance forthcoming. [44] Myers, Stewart C., and Nicholas S. Majluf, 1984, “Corporate ?nancing and investment decisions when ?rms have information that investors do not have,” Journal of Financial Economics 13, 187-221. [45] Peng, Lin, and Ailsa R¨ oell, 2004, “Executive pay, earnings manipulation and shareholder litigation,” working paper, Princeton University. [46] Poirier, Dale J., 1980, “Partial observability in bivariate probit models,” Journal of Econometrics 12, 209-217. [47] Povel, Paul, Rajdeep Singh, and Andrew Winton, 2004, “Booms, busts, and fraud,” working paper, University of Minnesota. [48] Richardson, Scott A., and Richard G. Sloan, 2003, “External ?nancing and future stock returns,” working paper, University of Pennsylvania. [49] Shleifer, Andrei, and Robert W. Vishny, 1997, “A survey of corporate governance,” Journal of Finance 52, No. 2, 737-783. [50] Sidak, Gregory J., 2003, “The failure of good intentions: The WorldCom fraud and the collapse of American telecommunications after deregulation,” Yale Journal on Regulation 20, 207-267. [51] Stein, Jeremy, 1989, “E?cient capital markets, ine?cient ?rms: A model of myopic corporate behavior,” Quarterly Journal of Economics 104, 655-669. [52] Sweeney, A.P., 1994,“Debt Covenant Violations and Manager’s Accounting Responses,” Journal of Accounting and Economics 17, 281-308.
100
[53] Teoh, Siew Hong, Ivo Welch, and T. J. Wong, 1998a, “Earnings management and the postissue performance of seasoned equity o?erings,” Journal of Financial Economics 50, No. 1 63-99. [54] Teoh, Siew Hong, Ivo Welch, and T. J. Wong, 1998b, “Earnings management and the longterm market performance of initial public o?erings,” Journal of Finance 53, No. 6, 1935-1974.
101
doc_817204068.pdf