THE EFFECT OF INCOME-INCREASING EARNINGS MANAGEMENT ON ANALYSTS RESPONSES. Jomo Sankara. A Dissertation Submitted to the Faculty of

Transcription

1 THE EFFECT OF INCOME-INCREASING EARNINGS MANAGEMENT ON ANALYSTS RESPONSES by Jomo Sankara A Dissertation Submitted to the Faculty of the College of Business in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy Florida Atlantic University Boca Raton, Florida August 2012

2 Copyright by Jomo Sankara 2012 ii

3

4 ACKNOWLEDGEMENTS I wish to thank my program advisor, and dissertation committee chair, Dr. Mark Kohlbeck for all the support and guidance he provided throughout this process and over the last five years. I am grateful to have had the opportunity to learn from him first hand. I also wish to thank my other dissertation committee members, Dr. Kimberly Dunn, Dr. Jian Cao, and Dr. Ky-Hyang Yuhn for contributing their time and expertise to this study. I want to give a special thanks to Dr. Terrance Skantz, Dr. Julia Higgs and Dr. Robin Radtke for investing in my successful completion of this program. Without their guidance I will not have achieved this goal. I also wish to extend special thanks to Errol for his support throughout this long and rewarding journey. I am fortunate he traveled this path with me. I am thankful to all of the other fellow Ph.D. students, especially Yezen, Ivy and Joe, who always provided encouragement and never hesitated to provide assistance when needed. Finally, I wish to thank my family for their love, encouragement and support. iv

5 ABSTRACT Author: Title: Institution: Dissertation Advisor: Degree: Jomo Sankara The Effect of Income-Increasing Earnings Management on Analysts Responses Florida Atlantic University Dr. Mark Kohlbeck Doctor of Philosophy Year: 2012 As a consequence of financial analysts joint role as information intermediaries and firm monitors, I investigate analysts responses to opportunistic corporate earnings management as firm mispricing increases. While firms management have capital markets and executive equity incentives to manage earnings, financial analysts have trading volume, investment banking, and management information incentives which result in analysts optimism bias. However, prior research also finds that analysts have reputational incentives, which motivate them to provide accurate and profitable outlooks. Using a generalized linear model (GLM), I estimate analysts stock recommendation (price targets) responses for earnings management firms. I use the residual income model to compute fundamental value and I add proxies for earnings v

6 management to my analyst-responses models. I find strong evidence of a negative association between the favorableness of analysts stock recommendation (price target) responses and corporate earnings management as mispricing increases. I also find a negative association between the favorableness of analysts responses and corporate earnings management in those firms analysts believe to be overvalued. These negative associations are increased as the length of consecutive periods of earnings management increases. However, I only find partial support for my hypotheses that the negative associations increase as the cost of the earnings management mechanism used to manage earnings increases. The main implications of my findings are that analysts use corporate earnings management and firm fundamental value in their stock recommendations (price targets) responses. In addition, my results provide evidence that, after controlling for earnings quality, analysts stock recommendations (price targets) are consistent with strategies based on residual income models. These findings will be of interest to shareholders, regulators, and researchers as well as to finance and accounting practitioners. vi

7 DEDICATION To Zalika and Shomari

8 THE EFFECT OF INCOME-INCREASING EARNINGS MANAGEMENT ON ANALYSTS RESPONSES LIST OF FIGURES... x LIST OF TABLES... xi Chapter 1: INTRODUCTION... 1 Chapter 2: LITERATURE REVIEW Analyst Reports Dual Role of Financial Analysts Analysts Incentives Earnings Management Capital Markets Explanation for Managing Earnings Executive Compensation Explanation for Managing Earnings Meeting or Beating Analysts Forecasts by Expectations Management Earnings Management and Analyst Response Overvalued Equity Agency Costs of Overvalued Equity Earnings Management Overvaluation and Undervaluation Overvalued Equity and Analysts Responses Summary Chapter 3: HYPOTHESES DEVELOPMENT Analysts Responses in Firms that Manage Earnings vii

9 3.2 Analysts Responses in Overvalued Firms that Manage Earnings as the Length of Overvaluation Increases Summary of Hypotheses Chapter 4: RESEARCH DESIGN Earnings Management Proxies Analysts Responses to Earnings Management Main Mispricing Proxy Alternate Mispricing Proxy Dependent Variables Regression Models Consecutive Periods of Overvaluation Regression Models Post-Regulation FD Analyses...79 Chapter 5: DATA Sample Selection Descriptive Statistics Annual Trends Correlation Matrices Chapter 6: EMPIRICAL RESULTS GLM Regression Models for Mispriced Firms Multiple Regression Models for H1 and H Multiple Regression Models for H Multiple Regression Models for H viii

10 6.2 Alternate Tests of Hypotheses 1, 2 and Multiple Regression Models using Overvalued Firms to Test H1 and H Multiple Regression Models using Overvalued Firms to Test H Overvaluation Duration Multiple Regression Models for H Summary Chapter 7: SENSITIVITY ANALYSES P/V Ratio Computation Firm Overvaluation Continuous Variable to Represent Firm Overvaluation Firm Overvaluation Indicator Variable Long-Term Growth Model Controlling for Interaction Effects of Income-Decreasing Earnings Management Changes to Income-Increasing Earnings Management Growth and Economic Downturn Time Periods Petersen Cluster Regressions and Newey-West Standard Error Correction Chapter 8: CONCLUDING REMARKS APPENDIX A: FIGURES APPENDIX B: TABLES REFERENCES ix

11 LIST OF FIGURES Figure 1: Overvalued Equity, Earnings Management and Analysts Stock Recommendation Figure 2: Overvalued Equity, Earnings Management and Analysts Price Targets Figure 3 Panel A: Market Value of Equity Panel B: Return on Equity Panel C: Analysts Consensus Stock Recommendations Panel D: Analysts Consensus Price Targets Panel E: Income-increasing Discretionary Accruals Panel F: Income-increasing Real Transaction Management x

12 LIST OF TABLES Table 1: Summary of Hypotheses Table 2: Sample Selection Summary Table 3: Descriptive Statistics Panel A: Variable Definitions Panel B: Pooled Analysis-Descriptive Statistics Table 4: Correlation Matrices Panel A - Recommendations Correlation Matrix Panel B- Price Targets Correlation Matrix Table 5: Main Model: Regression (Hypotheses 1 and 2) Table 6: Main Model: GLM Regression (H3) Table 7: Main model: GLM Regression (H4) Table 8: Overvaluation Model: GLM Regression (Hypothesis 2) Table 9: Overvalued Model: GLM Regression (H3) Table 10: Overvaluation Model: GLM Regression (H5) Table 11: Sensitivity Analysis: Terminal Value Based on Fade Rate (H1 and H2) Table 12: Sensitivity Analysis: Terminal Value Based on Fade Rate (H3) Table 13: Sensitivity Analysis: Terminal Value Based on Fade Rate (H4 and H5) Table 14: Sensitivity Analysis: Continuous Overvaluation Variable (H2) Table 15: Sensitivity Analysis: Continuous Overvaluation Variable (H3) xi

13 Table 16: Sensitivity Analysis: Continuous Overvaluation Variable (H5) Table 17: Sensitivity Analysis: Long Term Growth Forecast (H1 and H2) Table 18: Sensitivity Analysis: Long Term Growth Forecast (H3) Table 19: Sensitivity Analysis: Long Term Growth Forecast (H4 and H5) Table 20: Sensitivity Analysis: Income-decreasing Interactions (H1 and H2) Table 21: Sensitivity Analysis: Income-decreasing Interactions (H3) Table 22: Sensitivity Analysis: Income-decreasing Interactions (H4 and H5) Table 23: Sensitivity Analysis: Change Models (H3) Table 24: Sensitivity Analysis: Growth and Economic Downturn (H1 and H2) Table 25: Sensitivity Analysis: Growth and Economic Downturn Time Periods (H3) Table 26: Sensitivity Analysis: Growth and Economic Downturn Time Periods (H4) Table 27: Sensitivity Analysis: Growth and Economic Downturn Time Periods (H5) xii

14 CHAPTER 1 INTRODUCTION There are many studies on earnings management for capital markets purposes, and prior research has found a growing trend in managing earnings to meet earnings thresholds. These studies have concluded that opportunistic earnings management is potentially costly to a firm. Specifically, earnings management choices such as real transactions management and non-gaap earnings management (hereafter referred to as financial irregularities) may result in the destruction of a firm s value (Jensen, 2005; Badertscher, 2011). This value destruction reduces the fundamental value of firms as computed by residual income models and may result in firm mispricing. Prior research finds that stock recommendation strategies based on residual income models result in positive abnormal returns but prior research has not investigated how earnings quality affects the relationships between analysts stock recommendations (price targets) responses and firm mispricing. This paper fills that gap and also provides evidence on how analysts respond to earnings management. Bradshaw (2004) finds that analysts stock recommendations are inconsistent with stock-picking strategies based on residual income models. In a related perspective, Schipper (1991) calls for more research on the inputs into stock recommendations. If earnings management is considered by analysts when making stock recommendations or setting price targets, then analysts responses may be consistent with strategies based on residual income models after 1

15 controlling for earnings quality. This is important to the accounting literature as it provides further evidence on how financial analysts (hereafter referred to as analysts and sell-side analysts) use the earnings information generated and reported through the accounting function. Furthermore, this paper provides further evidence on the profitability of analysts outlook, analysts biases and analysts monitoring role. I focus on sell-side analysts responses to earnings management because of analysts multi-faceted roles in the capital markets. Sell-side analysts enjoy a joint role as information intermediaries and as firm monitors. Jensen and Meckling (1976) argue that security analysts are one of the parties who may have a comparative advantage in monitoring companies, and this monitoring helps to reduce agency costs by limiting the non-profit maximizing behavior of management. Prior research suggests that analysts monitoring role includes both the detection and the prevention of earnings management (Yu, 2008; Healy and Palepu, 2001). It follows that sell-side analysts could help to decrease agency costs through their monitoring role. Conversely, researchers also argue that analysts pressure management to meet targets, which may result in an increase of earnings management. Analysts also bring information to capital markets, which contributes to making capital markets informationally efficient and may also contribute to investment strategies and price discovery (Guan, 2006; Gleason and Lee, 2004). Prior research has found significant market reaction to analysts price targets, earnings forecasts, and stock recommendations (Womack, 1996; Asquith et al., 2005). Given analysts dual role within capital markets, analysts may be expected to alarm investors of the costs associated with earnings management. 2

16 Analysts incentives to curry favor with management, gain investment banking business for their firms and increase trading activity (Ke and Yu, 2006; Michaely and Womack, 1999; Cowen et al., 2006; Irvine, 2004; Jackson, 2005; Agrawal and Chen, 2008) can all result in analyst optimism which may bias analysts reports. This bias may prevent or reduce analysts responses to earnings management. However, analysts also have incentives to maximize their reputation, which should motivate analysts to release accurate reports that reflect the impact of earnings management. Prior research finds that analysts reduce their earnings forecasts when faced with earnings management (Guan, 2006; Louis, 2004), which suggests that analysts may reduce their stock recommendations or price targets of firms where earnings are managed. However, even if analysts incentives result in accurate earnings forecasts, these forecasts may not lead to reliable price targets or profitable stock recommendations. For instance, prior empirical evidence that analysts stock recommendations are inconsistent with investment strategies based on present value models (Barniv et al., 2009; Bradshaw, 2004) generates uncertainty about the analysts responses. This line of research finds that residual income models are positively associated with future abnormal returns, but still negatively associated with stock recommendations despite recent changes in capital markets regulations (Barniv et al., 2009). Bradshaw (2004) argues that the discrepancy between stock recommendations and stock picks based on residual income models may be due to analysts reliance on heuristics. Does this mean that analysts make more optimistic stock recommendations and price targets when faced with earnings management? Loh and Mian (2006) provide evidence that the most accurate analysts make the more profitable recommendations, a 3

17 finding which is consistent with residual income valuation models. Similarly, Ertimur et al. (2007) find that the value relevance of earnings helps to explain the profitability of stock recommendations. Based on this stream of research and assuming analysts use residual income models based on fundamental values, I posit that earnings quality partially explains analysts stock recommendations and price targets given analysts monitoring role. Specifically, as a result of analysts monitoring role, analysts incentives, and empirical studies revealing a positive association between earnings forecasts and recommendation profitability, I hypothesize that financial analysts actually reduce the level of their stock recommendations when faced with earnings management as firm mispricing increases. In addition, I also expect that price targets will be lower for earnings management firms as mispricing increases. Firms may use different earnings management mechanisms as the length of consecutive periods of earnings management increases. The use of a specific earnings management method may therefore reveal the length of consecutive periods of earnings management. Based on the pecking order of earnings management (Badertscher, 2011; Cohen et al., 2008; Ettredge et al., 2010) and the differential cost of different types of earnings management (Zang, 2012; Cohen and Zarowin, 2010), I hypothesize that financial analysts responses to earnings management is dependent on the type of earnings management mechanism and the duration of the income-increasing earnings management. Specifically, financial analysts reduce their optimism in their stock recommendations (price targets) responses more when faced with more costly methods of earnings management. Similarly, financial analysts reduce their optimism in their stock 4

18 recommendation (price targets) responses more when faced with a greater length of consecutive annual earnings management. I separately investigate analysts recommendations and price targets responses in overvalued firms that manage earnings. The fundamental value of the firm can be derived from the Ohlson model by using analysts own earnings forecasts (Lo and Lys, 2000; Frankel and Lee, 1998). Therefore, firms that analysts believe to be overvalued can be inferred by comparing the stock price with the computed fundamental value. Hereinafter, any reference to overvalued firms or overvalued equity refers to firms that analysts believe to be overvalued, and this categorization of firms in no way contradicts the efficient market hypothesis. Overvalued firms are a special subset of mispriced firms because management in overvalued firms has strong incentives to maintain high stock prices (Jensen, 2005; Efendi et al., 2007; Chi and Gupta, 2009) 1. Thus, analysts monitoring role is especially important in these firms. Prior research finds that overvalued firms are more likely to manage earnings. Jensen (2005) argues that management, the agents of overvalued firms, may be forced to cook the books to justify the stock price when the firm is overvalued. These earnings management actions taken to artificially maintain the high stock price are therefore a result of the agency costs of overvaluation of equity, and these actions lead to the destruction of firm value. Gupta and Chi (2009) find that overvalued firms are positively associated with income-increasing discretionary accruals. Similarly, Badertscher (2011) argues that there is a pecking order of earnings management, and higher levels of overvalued equity are positively associated with more costly types of earnings 1 These incentives include equity-based compensation, bonuses, job security, and reputation. 5

19 management and the duration of the earnings management. Thus, analysts monitoring role may be more important for overvalued firms as investors downside risks in their investments is greatest in overvalued firms. As a result, I expect the negative association between analysts responses and earnings management in overvalued firms to be greater as the cost of the earnings management mechanism increases and as the length of overvaluation increases. My sample, which represents earnings management firms followed by sell-side analysts, consists of all the firms on the Institutional Brokers' Estimate System (I/B/E/S) database from 1992 to Firms management are prohibited from disclosing selective information to analysts after the enactment of the Regulation Fair Disclosure Act 2000 (hereafter referred to as Regulation FD). Therefore, I also use a post-regulation FD sample because analysts have fewer incentives to make optimistic responses after the passage of this regulation. I use different proxies for analysts responses to earnings management: the favorableness of analysts consensus stock recommendations responses and the favorableness of analysts consensus price targets responses. I extend Bradshaw s (2004) stock recommendation model by adding earnings management variables to test my hypotheses. I expect similar results whether stock recommendations or price targets are used as the dependent variable. Using generalized linear model (GLM) regressions, I estimate analysts stock recommendations and price targets for earnings management firms. I find evidence that the relationship between the favorableness of analysts stock recommendations (price targets) responses and income-increasing earnings management is more negative as firm mispricing increases. I also find evidence that the negative association between the 6

20 favorableness of analysts responses and income-increasing earnings management in mispriced firms is greater as the length of the consecutive periods of income-increasing earnings management increases. In my overvaluation analysis, I find that earnings management in overvalued firms is negatively associated with the favorableness of analysts responses. I only find evidence in partial supporting that the negative association between the favorableness of the analysts responses and income-increasing earnings management as mispricing increases is greater as the cost of the method used to manage earnings increases. Similarly, I only find evidence in partial supporting that the favorableness of the analysts responses is negatively associated with income-increasing earnings management in a change specification. Finally, I do not find any evidence that the negative association between the favorableness of the analysts responses and earnings management in overvalued firms is greater as the cost of the method used to manage earnings increases or the length of the overvaluation increases. Taken together, these results provide evidence that analysts consider earnings management, fundamental value, and the length of consecutive periods of opportunistic earnings management when making stock recommendations and price targets. Furthermore, they consider the costs of the different GAAP-based earnings management mechanisms used to manage earnings. The impact of earnings management on stock recommendation and analysts price targets is important for several reasons. First, although Schipper (1991) called for more research into the information analysts use to make stock recommendations, this appeal has remained largely unanswered. I provide evidence of the extent to which analysts use 7

21 earnings management information in making investment recommendations and price targets, which will be of interest to researchers and investors. Second, I extend Bradshaw (2004), Barniv et al., (2009) and Chen and Chen (2009) who find that stock recommendations are inconsistent with investment strategies based on residual income models. Using levels data, this paper provides evidence that stock recommendations and price targets are consistent with investment strategies based on residual income models after controlling for earnings quality. Investors will be interested in how earnings management in firms affects the profitability of stock recommendations and accuracy of price targets. Finally, this paper extends Jensen (2005) and Badertscher (2011). Jensen (2005) argues that there are agency costs of the overvaluation of equity, and analysts are named as contributors to firm overvaluation. I provide evidence on analysts monitoring role in reducing their outlook for firms they believe to be overvalued and firms that manage earnings. Therefore, both investors and regulators will be interested in this study as I provide evidence on analysts monitoring role and the extent analysts contribute to the alleviation of the agency cost of overvalued equity. Furthermore, this paper extends Badertscher (2011) by providing evidence on analysts differential response to earnings management as management utilizes the options on their earnings management pecking order. The rest of the paper is organized as follows. In chapter 2, I review earnings management, overvaluation of equity, and financial analyst literature. I develop my hypotheses in chapter 3. In chapter 4, I describe the research design used in this study. In chapter 5, I provide details of the sample. The tests of my hypotheses and related 8

22 sensitivity analyses are provided in chapters 6 and 7. My concluding remarks follow in chapter 8. 9

23 CHAPTER 2 LITERATURE REVIEW My analysis of analysts responses to earning management builds upon three strands of research. This chapter summarizes the studies most relevant to my research questions and includes the most recent advances in each strand. Section 2.1 reviews the relevant financial analysts literature. Section 2.2 summarizes the most relevant and recent earnings management literature. This section also reviews the relationship between earnings management and analysts responses. Section 2.3 summarizes research on the overvaluation of equity. This section also discusses the relationships between overvaluation and earnings management, and overvaluation and analysts responses. I summarize this chapter in section Analyst Reports Both buy-side analysts and sell-side analysts evaluate firms to identify investment opportunities for their clients or employer. They both communicate their analysis and recommendations, which include providing overviews of firms businesses, forecasts of earnings, cash flows and/or revenues, target stock prices, stock recommendations, and justifications for the recommendations. I focus on sell-side analysts reports in this paper because only sell-side analysts reports are publicly available. Analysts stock recommendations (price targets) response to earnings management may be determined by analysts characteristics, their job roles, and their 10

24 incentives. Therefore, in this section, I discuss the dual role of the analyst and analysts economic incentives. Sell-side analysts (hereafter financial analysts, analysts or sell-side analysts) are employed by brokerage houses, investment-banking firms, or independent research institutes. Researchers argue that it is the differences in employment that create differential incentives for analysts, which ultimately result in differential analyst behavior. However, analysts dual role also impacts analysts behavior and makes them important players in the capital markets Dual Role of Financial Analysts Analysts responses to financial reporting are important because of the key roles financial analysts perform in capital markets. Analysts are both information intermediaries and firm monitors and, therefore, contribute to making both individual firms and the capital markets more efficient. First, I review information intermediaries literature. Second, I review the analysts monitoring literature. Fama (1970) argues that, in an efficient market, security prices fully reflect all available information. Securities analysts contribute to the efficiency of markets because they provide information to the marketplace. Specifically, analysts are information intermediaries who collect, process, and distribute information to individuals and institutional investors. Analysts firm-specific and industry-specific findings are typically conveyed to the public in research reports, which may include revenue, cash flow and/or earnings forecasts, price targets, stock recommendations, overviews of firms, and justifications for recommendations. These reports incorporate both publicly available information and analysts private information. Bhushan (1989) argues that analysts information is important because it is demanded by capital market participants. It is 11

25 analysts expertise in information processing and private information which makes their reports so useful to capital markets. In fact, analysts have diverse information-processing skills, which, together with prior beliefs, are used to generate private (idiosyncratic) information (Kim and Verrecchia, 1997; Barron et al., 2005). Factors such as experience, access to resources, skill, and forecasting method determine the precision of this private information. Thus, analysts are more likely to follow firms where the potential for accumulating private information is greater. Barth et al. (2001) find evidence that analysts have a greater potential for profitable private information acquisition in firms with more information asymmetry. Similarly, Kirk (2008) finds that firms with less visibility and more information uncertainty are more likely to hire a research firm. These findings suggest that the demand and supply of analysts is greatest where private information is most needed, and a strong market reaction should be expected from such information. Consequently, evidence of analysts role as information providers can also be inferred from investors reaction to analysts reports. Prior research finds significant market reaction to analyst forecast revisions (Elton et al., 1981; Brown et al., 1985; Givoly and Lakonishok, 1979; Gonedes et al., 1976), analyst recommendations (Stickel, 1995; Womack, 1996), and price targets (Brav and Lehavy, 2003). Givoly and Lakonishok (1984) argue that, since many clients are willing to pay for forecasting services, analysts research must be relevant. For example, Givoly and Lakonishok (1979) find that positive (negative) abnormal returns are associated with upward (downward) revisions in the months surrounding the forecast revision. The abnormal return over the revision month and the two following months is 2.2 percent. Furthermore, 12

26 Givoly and Lakonishok (1980) find that an investment strategy utilizing information on forecast revisions can earn abnormal returns ranging from 1.22 to percent depending on the strategy chosen and the level of transaction costs. Gleason and Lee (2003) investigate the post-earnings forecast revision announcement drift reported by Givoly and Lakonishok (1980) and others. Using a sample of analyst forecast data from October 1993 to December 1998, they find that the post-revision price drift is greater for firms with larger good-news earnings revisions. Furthermore, they find that the post-revision price drift is greater for high-innovation revisions, revisions made by superior analysts and firms with lower analyst coverage. However, revisions made by celebrity analysts are associated with lower post-revision price drifts. Gleason and Lee (2003) suggest that individual financial analysts bring relevant information to capital markets, and multiple analysts enhance the speed of the price discovery process. Similarly, prior research finds price targets to be both informative and incrementally informative. Using a large database of price targets from 1997 to 1999, Brav and Lehavy (2003) find that price target revisions are associated with significant abnormal returns, and these returns are positively associated with the favorableness of the price target revision. Specifically, the average abnormal return for portfolios based on ratio of price target revisions to stock price ranges from percent for the lowest price target ratio to 3.21 percent for the most favorable revisions. Furthermore, the researchers provide evidence that price targets are incrementally informative when issued contemporaneously with stock recommendation and earnings forecast revisions. While 13

27 price targets and earnings forecasts provide important information to capital markets, stock recommendations are most likely to be used in a trading strategy. Stock recommendations are the main output from the analysts reports (Schipper, 1991) and should therefore be the most useful. Stickel (1995) provides evidence of stock recommendations usefulness as a result of market reaction to stock recommendations from brokerage houses. Using buy and sell recommendations from 1988 to 1991, Stickel (1995) finds that the short-term price reaction is determined by the size of the brokerage house, firm size, the reputation of the analyst, the strength of the recommendation, the magnitude of the change in recommendation, and contemporaneous earnings forecast revisions. In addition, Stickel finds that firm size, contemporaneous earnings forecast revisions and the strength of the recommendation provide permanent price shocks, whereas, the magnitude of the change in recommendation, broker size and analyst reputation provide temporary price shocks. Womack (1996) finds market under-reaction to new analysts buy and sell recommendations. Using 1989 to 1991 analyst recommendations from the top 16 brokerage research departments, Womack (1996) finds initial size-adjusted price increases of 3 percent for buy recommendations and initial size-adjusted decreases of 4.7 percent for sell recommendations. He also finds significant trading volume reactions to these stock recommendations. However, there is also a market reaction drift as prices increase another 2.4 percent in the month following buy recommendations while prices fall an additional 9 percent in the six months following sell recommendations. However, using consensus stock recommendations as a trading strategy may not be profitable after transaction costs. Using recommendations from 1985 to 1996, Barber 14

28 et al. (2001) find that investors earn average abnormal returns of more than 4 percent by buying the most highly recommended stocks and selling short the least favorably recommended stocks. In fact, using analysts consensus recommendations to pick stocks provide returns four times better than choosing stocks based on book to market or firm size. However, high trading levels, such as daily rebalancing, are required to obtain the excess return, resulting in substantial transactions costs such as brokerage commission and the bid-ask spread. After transaction costs, this trading strategy does not beat the market index (i.e., there are no abnormal returns). Mikhail et al. (2004) also find that positive abnormal returns from trading on stock recommendations from the analysts with the best prior performance is insufficient to cover transaction costs. These researchers use a database of individual analysts recommendations from 1985 to They provide evidence that analysts issuing more (less) profitable recommendation revisions in the past continue to do so. Further, the market reacts more positively to recommendation revisions from high-performing analysts and there is a post-recommendation revision drift for these analysts. However, a trading strategy based on these high performing analysts will be unprofitable as a consequence of transaction costs. Overall, information in stock recommendations is useful to capital markets, even though a trading strategy based on this information may not be profitable as a result of transaction costs. While elements of the analysts report are individually informative, elements of the report are also incrementally informative. Using analyst data from 1997 to 1999, Asquith et al. (2005) find that changes in summary price targets, earnings forecasts, and stock recommendations all provide independent information to the capital markets. 15

29 Further, they find that price targets, earnings forecasts, and stock recommendations are all incremental to market reaction, and the stronger the justification provided in the report, the stronger the market reaction to the report. The researchers provide evidence that supporting data are most important for downgrades, and market reaction is significantly greater for smaller firms and firms with smaller analyst followings. In summary, all aspects of analysts reports are informative to capital markets. Analysts information can also be used to value firms. Using a residual income model such as the Ohlson model, a stock s fundamental value can be expressed as the reported book value plus the present value of expected abnormal returns (Ohlson, 1995). Using analysts consensus earnings forecasts in their abnormal earnings calculations, Frankel and Lee (1998) calculate firms fundamental values (V) and find that V is highly correlated with current stock price. Using a sample period ranging from 1979 to 1991, the researchers find that V explains more than 70% of the cross-sectional variation in stock price. Furthermore, the researchers find that the fundamental value to price (V/P) ratio is a good predictor of long-term cross-sectional returns. Using a V/P strategy of a long position in the top quintile firms and a short position in the bottom quintile firms yields cumulative buy-and-hold returns of 35% over a three-year period. I now review analysts monitoring roles within capital markets. Jensen and Meckling (1976) argue that security analysts are one of the parties who may have a comparative advantage in monitoring companies, and this monitoring helps to reduce agency costs by limiting the non-profit maximizing behavior of management. Specifically, Healy and Palepu (2001) suggest that information 16

30 intermediaries such as analysts help to detect managers misbehavior by engaging in private information production. Yu (2008) argues there are three reasons analysts may be effective monitors against earnings management. First, analysts are expected to provide information in the interests of prospective shareholders as well as of other participants in the market, and not only current shareholders like internal governance bodies. Second, analysts usually have financial training in accounting and finance, and often have a background in the industries they cover. They generally have the resources and skills to analyze tedious financial statements and complicated footnotes. Third, analysts produce regular updates; therefore, they continuously scrutinize management behavior and explore financial reporting irregularities. Recent studies have focused on analysts monitoring role in detecting earnings manipulation. Dyck et al. (2010) study all reported fraud cases in large American companies from 1996 to They find that equity holders agents (auditors and analysts) collectively account for 24% of the cases in their sample. Analysts success in detecting fraud is somewhat due to their need to gather a lot of relevant information as a byproduct of their normal work. Reputational incentives, improved career prospects, and access to inside information are other factors in detecting fraud. Overall, Dyck et al. (2010) find that the most efficient external whistleblowers for corporate fraud are analysts. The detection of earnings manipulation is likely to lead to accounting restatements. Zeng et al. (2010) evaluates restatements announced between January 2003 and December They find that in eleven percent of the sample, analysts discuss 17

31 accounting problems in their reports before these problems are publicly disclosed and the financial statements are restated. There is, on average, a negative market reaction of 4 percent to these analyst reports, indicating that investors take note of these warnings in analysts reports. Several factors determine the likelihood of analysts identifying accounting problems, such as the analysts reputation, experience with the firm, access to firm information, firm visibility, and size of the accounting manipulation. In addition to detecting accounting irregularities, analysts are also associated with lower levels of GAAP-based earnings management. Using a sample from 1988 to 2002, Yu (2008) asks if analysts serve as external monitors or if they put excessive pressure on managers to meet performance benchmarks. He finds that firms followed by more analysts, more experienced analysts, and analysts from top brokers are associated with lower levels of earnings management. Guan (2006) uses a sample of individual earnings forecasts and finds that experienced analysts adjust more for earnings management while analysts following more firms adjust less for earnings management. On the other hand, analysts adjust less for earnings management when the firm has a history of managing earnings. Guan also investigated how Regulation FD and analyst affiliation affect the relationship between analyst following and earnings management. Prior to Regulation FD, management s practice of selectively disclosing private information to a select group of analysts was believed to bias analysts reports. Regulation FD was issued in October 2000 by the Securities and Exchange Commission (SEC) to prohibit management from selectively disclosing information to analysts. While as expected, analysts adjusted more for earnings management following Regulation FD, surprisingly, Guan (2006) did not find 18

32 that affiliated analysts adjusted any less for earnings management than unaffiliated analysts. While analysts responses were not affected by investment banking incentives in Guan s study, analysts incentives are likely to have a significant effect on analysts firm monitoring Analysts Incentives Analysts responses to earnings management in overvalued firms are likely to be affected by their incentives. Financial analysts have incentives to maximize their reputation, optimize the receivers reaction to their reports, and maximize the value of their forecasts to investors (Beyer et al., 2010). Accordingly, analysts actions to optimize reaction to their reports may depend on the type of their employer, which may give rise to conflicts of interest. As sell-side analysts are employed by brokerage houses, independent research institutes, or investment banking firms, analysts incentives to optimize the receivers reaction may include objectives such as securing investment banking business and increasing trading volume. These incentives may lead to analysts optimistic recommendations, price targets, and/or forecast revisions. Optimistic recommendations increase trading volume because risk-averse investors and short-selling constraints create a demand for favorable stocks which can be purchased, and optimistic recommendations encourage investors to buy stocks. First, I review analysts incentives to optimize reaction to their reports. Second, I review analysts incentive to optimize their reputation. Analysts recommendation optimism is evidenced from the skewed distribution of analysts stock ratings. Barber et al. (2006) use a timeframe from 1996 to June They find that the percentage of buy recommendations initially increased but steadily 19

33 decreased from mid-2000, partly as a result of NASD Rule 2711 and NYSE Rule 472 which required public dissemination of ratings distribution. The researchers find buy recommendations increased from 60 percent to a high of 74 percent by June 2000, before reducing to 42 percent by June Conversely, sell recommendations decreased from 4 percent to 2 percent before increasing to a high of 17 percent in June Hold recommendations followed a similar pattern to sell recommendations, changing from 36 percent at the start of the sample period to 41 percent by June Barber et al. (2006) find that brokers who previously made the smallest percentage of buy recommendations significantly outperform brokers with the greatest percentage of buy recommendations. Clearly, analysts stock recommendations are optimistic even though the level of optimism is decreasing. Conflicts of interest arising from investment banking fees have historically been singled out as the main impetus for analyst optimism. Analysts working at investment banks had conflicts of interest as a result of the way research was funded. Prior to the Global Research Analyst Settlement, research was commonly funded by investment banking fees because research was used to attract new banking clients and to market new offers to investors (Cowen et al., 2006). Prior research has found the use of underwriting fees to fund research to be a conflict of interest that creates incentives for analysts to be optimistic. Such optimism resulted in analysts making optimistic recommendations and long-term earnings growth forecasts (Cowen, 2006). For example, using a sample of 391 initial public offerings (IPOs) in 1990 and 1991, Michaely and Womack (1999) find that investment banking underwriters buy recommendations perform more poorly than buy recommendations from unaffiliated brokers. 20

34 However, recent studies now suggest brokerage firms may be more responsible for analyst optimism than investment banking. Besides, Hong and Kubik (2003) find that, in their sample, underwriting relationships only account for 3 percent of an analysts portfolio. Therefore, conflicts of interest arising from investment banking may no longer be a major factor in analysts optimism. For example, Cowen et al. (2006) argue that IPOs require more from their investment bank than simply optimistic forecasts and recommendations. Using long-term earnings growth forecasts and stock recommendations from January 1996 to December 2002, Cowen et al. (2006) argue that analyst optimism is at least partially driven by trading incentives. They find that firms that serve both retail and institutional investors are more optimistic than other employers of analysts, while they find analysts working at full-service banks to be the least optimistic. Also, they argue that brokerage firms are more likely to drop coverage of stocks they consider to be less attractive than investment banks as a result of their incentive to cover firms for which they are optimistic. Thus, they argue that self-selection may be partially responsible for analyst optimism. Similarly, Irvine (2004) finds that analysts can generate higher trading fees by issuing more optimistic earnings forecasts than the consensus earnings forecast, and by issuing more optimistic stock recommendations. Using data from the Toronto stock exchange, Irvine (2004) finds that buy recommendations generate more trading through the brokerage firm. However, forecast errors (the difference between earnings' forecasts and actual EPS) do not generate more commissions, suggesting the conflict of interest does not motivate inaccurate earnings forecasts. 21

35 Jackson (2005) finds that forecast optimism can exist even if investment banking affiliation is removed. The researcher provides evidence that optimistic analysts generate higher trading volumes for their firms, which provides a conflict of interest and motivation for biased earnings forecasts and stock recommendations. However, the analysts reputation acts as a constraint on optimism as greater analyst reputation generates greater trading volume for the firm. Clarke et al. (2007) studied all-star analysts who switch investment banks and do not find that analysts change their optimism when joining a new bank. Analyst reputation is positively associated with gaining investment banking business for equity offerings. Jacob et al. (2008) find investment bank analysts earnings forecasts are less optimistic than those working for non-investment banks. Furthermore, the researchers find that investment bank analysts earnings forecasts are more accurate than those made by other analysts and affiliated analysts issue the most accurate forecasts. Similarly, Agrawal and Chen (2005) find that optimism in long-term forecasts is positively associated with the importance of the brokerage business to the analysts employer. In a later study, using a manually generated dataset of 110,000 stock recommendations issued from 1994 to 2003, Agrawal and Chen (2008) find that analysts stock recommendations are optimistic as a result of investment banking and brokerage trading conflicts of interest. Furthermore, the researchers find that investors adjust for the biases, and, one year after the recommendation revisions, the investment performance is not associated with the magnitude of the conflicts of interest. Thus, both investment banking and brokerage trading services contribute to analyst optimism. As previously alluded to, 22

36 researchers also argue that self-selection may be a factor in the observed analyst optimism. McNichols and O Brien (1997) argue that analyst optimism may be a result of self-selection rather than analyst opportunism. The researchers argue that analysts are more likely to provide forecasts and recommendations for stocks for which they have favorable expectations. This may be because they wish to build a favorable relationship with management or because they wish to increase trading volume. Therefore, initiation (ending) of coverage is associated with relative optimism (pessimism) of the stocks. The researchers use a proprietary database containing stock recommendations for more than 4,000 analysts from July 1987 to December They find that the ratings of stocks just added are more likely to be strong buy, and the analysts tend to drop stocks with lower rating than those whose coverage continues. Therefore, analysts portfolios of shares are likely to be more favorable than the market index, and changes in analysts forecasts and recommendations are likely to be more optimistic than the market as a whole. McNichols and O Brien (1997) find subsequent return on equity for the stocks analysts added to be better than the performance of the stocks analysts covered in the past or the stocks analysts dropped, providing evidence that analysts did add better-performing stocks and were not unjustifiably optimistic about the firms they added. Consistent with McNichols and O Brien (1997), Das et al. (2006) find that IPOs with higher residual coverage have greater returns and operating performance than those with a lower residual coverage within 3 years of the public issue. This provides further evidence that sell-side analysts cover firms expected to provide superior performances. 23

37 Analyst optimism may also be determined by analysts incentives to have a positive relationship with management. Using a sample which straddles Regulation FD, Chen and Matsumoto (2006) find that analysts receive less information following a stock recommendation downgrade compared to the information received after a stock recommendation upgrade. The researchers use forecast accuracy as a proxy for private information from management, and find that analysts providing more favorable recommendations or more optimistic recommendations compared to the mean recommendation have more accurate forecasts. Further, they argue that bold forecasts, forecasts made on days when no other analyst makes a forecast for the same firm, are evidence of private information from management. Consequently, Chen and Matsumoto (2006) find that analysts providing less favorable recommendations receive less managerial information as they make less bold forecasts. However, the researchers do not find any significant change in this practice after Regulation FD. Using a pre-regulation FD sample (January 1983 to June 2000), Ke and Yu (2006) find that analysts use biased earnings forecasts to gain better access to private information from management in order to improve their forecast accuracy and job security. Analysts initially issue optimistic earnings forecasts to curry favor from management, followed by pessimistic earnings forecasts before the earnings announcement. This strategy entails both optimism and pessimism, which enables analysts to produce more accurate earnings forecasts and makes them less likely to be fired. On the other hand, firms receive both initial optimistic forecasts and are then later able to meet or beat a more pessimistic revision. Ke and Yu s (2006) results hold 24

38 regardless of investment banking affiliation. Next, I turn to analysts incentives to maximize the value of their forecasts to investors. Irvine et al. (2007) find abnormally high institutional purchases and trading volume from 5 days before the announcement of stock recommendations. The researchers use a dataset of stock recommendations for new stocks from March 1996 to December 1997 and from March 2000 to December They argue that analysts have economic incentives to tip their preferred clients on the contents of upcoming initiations as a result of the trading commissions earned. Furthermore, Irvine et al. (2007) find the increase in institutional buying before the recommendation release is positively related to the optimism of the stock recommendation, the quality of the brokerage firm, and the rank of the analyst. This finding suggests that investors have foreknowledge of contents of the analysts report, and those institutions that buy in advance of initiation do earn abnormal returns. Finally, I review analysts incentive to optimize their reputation. Jackson (2005) finds that analysts with better reputations create higher future trading volume for their employer. Thus, a strong reputation provides long-term gains, providing evidence that analysts have economic incentives to develop their reputations. Beyer et al. (2010) argue analysts reputation is tied to the precision of their private information, the timeliness of information production, and forecast accuracy. Thus, forecast accuracy is a key component of analysts reputations. Using a database of quarterly earnings forecasts from 1980 to 1995, Mikhail et al. (1999) argue that forecast accuracy is important to analysts. The researchers find that the likelihood of analyst turnover is greater if the analyst s forecast accuracy is lower than his/her peers. However, there is no association between the likelihood of analysts 25

39 turnover and absolute forecast accuracy, absolute profitability of stock recommendation or relative profitability of stock recommendation. Similarly, Hong and Kubik (2003) find that relatively accurate forecasters are more likely to experience favorable career outcomes. In fact, analysts who are extremely accurate compared to other analysts are 52 percent more likely to move up the brokerage house hierarchy. However, using a dataset of earnings forecasts from 1983 to 2000, they also find that optimistic analysts (relative to the consensus forecast) are more likely to experience favorable turnover, and optimistic analysts who promote stocks are rewarded by brokerage houses. Thus, accuracy is not a key factor for brokerage houses analysts evaluation, and accuracy mattered less from 1996 to 2000 than from 1986 to Being named as an All-Star analyst or part of an All-American research team is also a proxy for analyst reputation (Stickel, 1992). Stickel (1992) argues that an analyst s position on the Institutional Investor (II) All-American Research Team can be viewed as a proxy for relative compensation and reputation. Furthermore, the research finds that these analysts supply more accurate and more frequent earnings forecasts than other analysts. In addition, market reaction to II All-American analysts large upward forecast revisions is greater than to other analysts. Thus, Stickel (1992) argues that as II All- American status is a proxy for reputation, greater analyst reputation is positively associated with performance. Researchers also investigate why some analysts report more accurate earnings forecasts than others. Using a sample from the IBES database from 1983 to 1994, Clement (1999) finds that analysts earnings forecast accuracy is positively related to analysts general and firm-specific forecasting experience as well as employer size. The 26

40 researcher argues that these variables represent analysts ability and resources. However, analysts forecast accuracy is negatively correlated with task complexity, as represented by the number of firms and industries covered by the analyst. Jacobs et al. (1999) also finds that analysts earnings forecast accuracy is associated with brokerage size. Furthermore, they find brokerage industry specialization impacts analysts forecasting accuracy. However, contrary to Clement (1999), they do not find that experience significantly improves forecast accuracy. These analyst attributes not only produce superior earnings forecasts, prior research also finds that they may result in superior stock recommendations, and reporting profitable stock recommendations also builds the analyst s reputation. Prior research finds that earnings forecast accuracy is positively correlated with stock recommendation profitability (Loh and Mian, 2006; Ertimur et al., 2007). Loh and Mian (2006) find that the analysts with the most accurate earnings forecasts also issue the most profitable stock recommendations. Using data from the IBES and CRSP databases from 1994 to 2000, this study finds that analysts in the top accuracy quintile make recommendations that earn average monthly returns of 0.737%. In contrast, the lowest accuracy quintile s stock recommendations earn negative average monthly returns. Using a dataset of stock recommendations from 1993 to 2004, Ertimur et al. (2007) build on Loh and Mian s (2006) findings and find that only more accurate analysts in firms with value relevant earnings make more profitable recommendations after controlling for analysts expertise. Ertimur et al. (2007) also categorize their sample into analysts working for brokerage firms with differing reputation for investment banking business, whereby analysts working for brokerage firms with a high reputation for 27

41 investment banking are deemed to be conflicted. The researchers find that the positive correlation between earnings forecast accuracy and buy recommendations profitability only apply to non-conflicted analysts. On the other hand, greater earnings forecast accuracy is associated with greater profitability from hold recommendations for conflicted analysts, where a hold recommendation is treated as a sell. In the post- Regulation FD period, the study shows that stock recommendation profitability is positively associated with earnings forecast accuracy for both conflicted and nonconflicted analysts, providing evidence that the recent financial regulations have improved the relationship between analysts accuracy and the profitability of their recommendations. Prior research also suggests that analysts who provide superior stock recommendations may do so persistently. In fact, researchers argue that analysts past performance is a predictor of current performance (Mikhail et al., 2004; Li, 2005). Using a database of individual analysts recommendations from 1985 to 1999, Mikhail et al. (2004) find that the relative analysts stock recommendations performances are persistent as analysts issuing more (less) profitable recommendation changes in the past continue to do so. Furthermore, the market reacts more positively to changes in recommendations from high-performing analysts, suggesting it is aware of the performance differences. However, Mikhail et al. (2004) find that a trading strategy based on analysts prior performance is unprofitable as a result of transaction costs. Alternatively, Li (2005) uses recommendation levels from a 1993 to 1999 sample to investigate the persistence of relative stock recommendation performance. The researcher finds that analysts with high risk-adjusted abnormal returns in the estimation period persistently outperform low performing analysts in the 28

42 subsequent holdout period. The performance persistence is greater for superior analysts buy recommendations than for their sell recommendations. Overall, analysts have incentives to provide superior forecast accuracy in order to build and maintain their reputation, and those with high forecast accuracy may also produce profitable stock recommendations on a consistent basis. Analyst experience, industry specialization, and brokerage size are some of the attributes that may improve analyst forecasting accuracy. 2.2 Earnings Management Earnings management is extremely important to the accounting profession because of its impact on financial reporting and the usefulness of the net income figure. Prior research has used contracting theory and positive accounting theory to explain earnings management practice. However, there has recently been an increase in corporate earnings management, whereby management manages earnings for capital market purposes Capital Markets Explanation for Managing Earnings Managers have incentives to manage earnings because firms are rewarded for meeting or beating earnings threshold. Using financial reporting data from 1982 to 1992, Barth et al. (1999) find that firms with annual increases in earnings per share (EPS) have higher price-earnings multiples than other firms. The researchers also found that the price earnings multiples decline significantly when earnings decrease after an earnings string. Similarly, Myers et al. (2006) find that firms enjoy on average over 20 percent annual abnormal returns during the first five years of a consecutive seasonal quarterly earnings string. Using data from 1963 to 2004, Myers et al. (2006) argue that five years of 29

43 consecutive quarterly earnings increases could only be achieved through earnings management. Furthermore, they find that their proxy for earnings strings firms is highly correlated with the use of earnings management strategies such as the use of income smoothing, special items and share repurchases, all of which boost EPS. Myers et al. (2006) find that firms achieve greater rewards for quarterly earnings strings than annual strings, and they argue that the market premium for earnings strings together with the penalty for earnings string breaks provide managers with incentives to maintain and extend earnings strings. Skinner and Sloan (2002) focus on the penalty for growth stocks missing the earnings threshold. They find that the negative price response resulting from a negative earnings surprise is larger than the positive price response connected to a positive earnings surprise. They also find that growth stocks have a relatively low occurrence of negative earnings surprises in boom periods, and consequently outperform other stocks in these periods. These studies provide compelling evidence that firms have strong incentives to meet earnings thresholds Executive Compensation Explanation for Managing Earnings Prior research also finds that executive compensation provides an incentive for corporate earnings management. Ke (2001) find that managers equity incentives are positively correlated to the length of a string of consecutive quarterly earnings increases. Similarly, Cheng and Warfield (2005) use stock-based compensation data from 1993 to 2000 to find that managers with high equity incentives are more likely to report earnings that at least meet analysts earnings forecast after controlling for firm performance. To provide evidence that managers benefit from this earnings management; the researchers 30

44 show that managers are more likely to sell some shares in subsequent periods, after the earnings management. Burns and Kedia (2006) use a sample of restatements from 1995 to 2002 and find that the sensitivity of a CEO s option portfolio to stock price is significantly and positively related to the likelihood of misreporting. The researchers also find a positive association between option sensitivity and the magnitude of the restatement. Similarly, using a sample of restatement announcements and executive stock option-based compensation in 2001 and 2002, Efendi et al. (2007) find that in-the-money stock options help to explain misstatements in overvalued firms. The researchers argue that management takes action to support the firm s short-term stock price in overvalued firms. Just as there are capital market penalties for missing earnings targets, there are also executive compensation penalties for missing earnings targets. Matsunaga and Park (2001) find that CEO annual bonuses are reduced if quarterly consensus earnings forecasts are missed for at least two quarters Meeting or Beating Analysts Forecasts by Expectations Management. Matsumoto (2002) finds that firms use both earnings management and expectation management to meet or beat analysts forecasts. However, using a sample of reported earnings mainly from 1993 to 1997, the researcher finds that high-growth firms tend to manage earnings upwards but not to guide analysts forecasts down. This study focuses on firms that have abnormally high earnings management; therefore, expectations management is not a major concern. Furthermore, decreases in analysts earnings forecasts are likely to be costly and reduce firm value as prior research provides evidence that investors react to analysts reports (Asquith et al., 2005). However, since the passage 31

45 of the Sarbanes-Oxley Act (SOX) 2002, trends in earnings management and expectations management have changed. Koh et al. (2008) find that earnings management has decreased post-sox whereas expectation management has increased. Using a sample from 1987 to 2006, the researchers find that the capital market premium for meeting or narrowly beating market expectations no longer exists, and fewer firms are now narrowly beating analysts earnings forecasts. Somewhat consistent with these findings, Cohen et al. (2008) find that decreases in accruals management post-sox has been replaced by increases in real transactions management, and firms continue to have incentives to meet or beat earnings thresholds. Overall, there are mixed findings on whether firms continue to have incentives to meet or beat market expectations. However, researchers continue to find evidence of corporate earnings management, and for firms that continue to meet or beat earnings thresholds, expectation management is unlikely to impact my results Earnings Management and Analyst Response There has surprisingly been little research on the association between analysts reports and earnings management. Prior research has found that analysts behavior can motivate managers to manage earnings. Using analysts stock recommendations from 1985 to 1998, Abarbanell and Lehavy (2003) find that sell recommendations motivate firms to report extreme income-decreasing earnings management. This suggests that sell recommendations provide incentives for firms to take a big bath. Similarly, buy recommendations motivate firms to meet or beat the consensus on earnings forecasts. Thus, there is a positive relationship between analyst recommendations and earnings management. In addition, Guan (2006) investigates the association between analysts 32

46 earnings forecast and earnings management. Using individual analysts earnings forecasts from 1993 to 2004, Guan (2006) finds that discretionary accruals are negatively associated with analysts earnings forecast revisions. 2.3 Overvalued Equity In terms of overvalued firms, the relationship between overvaluation and earnings management and analysts responses to overvaluation. Overvalued equity produces organizational forces reliant on the firm s overvaluation. These forces increase agency conflicts and reduce firm value as management attempts to maintain the firm s overvaluation. The agency costs of the overvaluation of equity are important because hundreds of billions of dollars in losses were suffered by investors as a result of these agency costs (Jensen, 2005; Efendi, 2007). First, I review the literature on the agency costs of overvalued equity. Second, I review the relationship between overvaluation and earnings management. Third, I review the literature on overvalued equity and analysts responses Agency Costs of Overvalued Equity Contrary to general agency costs, both security markets and incentive-based management compensation intensify the conflicts between owners and management as a result of the agency costs of overvalued equity. Jensen (2005) argues that securities markets increase the conflicts of interest between firm owners and managers when the firm is overvalued and agency costs increase. These conflicts arise because of the rewards and penalties for meeting or missing market expectations respectively. Overvalued firms are under intense pressure to meet market expectations in order to maintain their stock 33

47 price. Management is also under pressure to meet market expectations because their welfare is tied to the firms performance (Weisbach, 1988; Efendi, 2007). Specifically, incentive-based compensation, including money, stock options, bonuses, and job security, motivate management to meet market expectations and attempt to maintain their firms overvaluation (Jensen, 2005). However, by definition, these firms are unable to meet the expectations implicit in their stock price without earnings manipulation because, if they could, they would not be overvalued. In an efficient market, the firm s stock price would be the unbiased estimate of the true underlying value of the firm. Badertscher (2011) argues that it, therefore, follows that 50 percent of firms will be overvalued at any time. The problem is the market is unsure which firms are overvalued and which are undervalued. Using analysts own forecasts, the Ohlson model can be used to identify firms that analysts have forecast as mispriced. Prior research has found that a trading strategy based on these firms which analysts believe to be mispriced earns positive returns. For example, Frankel and Lee (1998) find that analysts forecasts can be used to estimate the fundamental value of securities, and the fundamental value-to-price ratio is a good predictor of long-term cross-sectional returns. This classification of firms as firms perceived to be overvalued does not challenge efficient market assumptions. Management and analysts may believe a firm is overvalued, and management may act to maintain or take advantage of a stock price it believes to be unsustainable; however, the capital markets in general may have no reason to believe that the firm is overvalued or was previously overvalued. An alternative explanation for management s behavior is that management may simply be inefficient or shirking, and therefore be 34

48 forced to use earnings management to maintain the firm value. Similarly, analysts may simply be wrong or conservative in forecasting earnings. These overvalued firms are expected to have negative abnormal returns and lower operating performance in the future. Using a sample of domestic non-financial companies from 1975 to 1993, Frankel and Lee (1998) provide evidence that firms in the lowest quintile of fundamental valueto-stock price (V/P) ratio earn lower abnormal returns in the subsequent 12 months, 24 months, and 36 months. Similarly, using a shorter sample period from January 1994 to June 1998, Bradshaw (2004) finds that the V/P ratio is positively associated with oneyear ahead excess returns. As firms with lower V/P ratios are likely to be overvalued, this finding suggests that overvalued firms are associated with lower future excess returns. The finance literature argues that overvalued firms are more likely to issue stock and use stock in merger and acquisition transactions. Myers and Majluf (1984) argue that a firm issues stock only when it is overvalued. Loughran and Ritter (1995) also suggest that firms take advantage of overvaluation to issue equity. These researchers find that both IPOs and SEOs have poor long-run stock price performances relative to non-issuing firms. SEOs also have poor long-run operating performance after the first six months of the issue. Likewise, using a sample of post-acquisition returns from 1970 to1989, Loughran and Vijh (1997) find that firms that complete stock mergers earn on average negative excess returns of -25 percent whereas firms completing cash tender offers earn on average 61.7 percent positive excess returns over 5 years. The stock merger returns are not significantly different than returns from stock issues. These studies provide evidence 35

49 that overvalued firms are associated with poor stock market and operating performance in the future Earnings Management Overvaluation and Undervaluation Management may use real transaction management, accrual management or financial irregularities to manage earnings. Chi and Gupta (2009) find that overvaluation is related to subsequent income-increasing earnings management, and this incomeincreasing earnings management is negatively related to future abnormal returns and operating performance. Using a sample spanning 1964 to 2003, the researchers find that overvalued firms with high discretionary accruals underperform the low discretionary accruals of overvalued firms by approximately 12 percent. Furthermore, the difference in performance between overvalued firms with higher discretionary accruals increases as the overvaluation length increases. In addition, Badertscher (2011) finds that the longer a firm is overvalued, the greater the total amount of earnings management and the greater the lack of operating flexibility. Using a sample of domestic nonfinancial firms from fiscal year 1994 to 2008, Badertscher (2011) finds there is a pecking order to managing earnings. Managers initially use accruals management, followed by real transaction management and they finally resort to non-gaap earnings management or financial irregularities. Therefore, overvaluation is a major determinant of using financial irregularity, and a major method used to maintain overvaluation is financial irregularities. In fact, over 60 percent of restatement firms were overvalued before engaging in financial irregularities, and over 60 percent of restatement firms remained overvalued after the non-gaap earnings management. 36

50 Financial irregularities are deemed the most costly earnings management method because of the associated capital market penalties and its impact on managerial reputation. Efendi et al. (2007) find that the firms in their sample of restated firms from 2001 and 2002, where CEOs had substantial in the money stock options, were substantially overvalued before they misstated their financial reports. The researchers suggest that, as these firms had significant positive abnormal market adjusted returns in the year before first misstating their financial statement, and as they had significantly higher market to book ratios than a matched sample at the end of the year before first misstating their financial reports, these firms were substantially overvalued. Furthermore, the researchers find significant negative market-adjusted abnormal returns in the five days surrounding the restatement announcements for the restating firms, which provide evidence of the cost of managing earnings for overvalued firms. Specifically, they argue that market capitalization losses of approximately $100 billion were incurred as a result of restatement announcements. Kothari et al. (2006) argue that, although overvalued firms are over-represented in high accrual portfolios, the opposite is not true as managers of undervalued firms do not have incentives to report low accruals. Managers of overvalued firms have incentives to maintain their stock price, and to use earnings management in order to achieve their goals whereas undervalued firms do not have similar incentives to maintain their valuation status, and do not use earnings management in a symmetrical manner to maintain undervaluation. In fact, some undervalued firms may report high accruals in order to correct the mispricing and avoid the penalties of a low valuation, whereas other undervalued firms simply do not have incentives to manage earnings or may decide to 37

51 take a big bath. Therefore, there is no prima facie reason to expect undervalued firms to have future positive abnormal returns as a result of changes in accruals. As a result of the differential incentives and the asymmetric relationship between firm mispricing and earnings management, I do not investigate analysts responses to earnings management in undervalued firms Overvalued Equity and Analysts Responses Researchers have recently found that, using stock recommendations as the proxy for analysts responses, analysts respond inconsistently or insignificantly to a strategy based on present value models of firm value. As a result, overvalued firms generally have more optimistic stock recommendations, and undervalued firms have more pessimistic stock recommendation, which suggests a positive relationship between overvaluation and stock recommendations. Specifically, Bradshaw (2004) uses analysts earnings forecasts in residual income models to calculate the fundamental values of a firm s stock. He finds that, while price-earnings-to-growth models and analysts projections of long-term earnings growth models are positively associated with analysts recommendation, the analysts recommendations are not explained by residual income models. In further analysis, Bradshaw (2004) finds that the heuristic valuation models provide the least profitable strategy for investors, while long-term investors would gain higher future returns by using the residual income models for their investment decisions. Using a wider time horizon, Barniv et al. (2009) find that the negative relation between analysts stock recommendations and the V/P ratio based on residual income valuations is decreasing following financial regulations. The researchers also find stock recommendations continue to be associated with negative future returns. Similarly, Chen 38

52 and Chen (2009) find a stronger relationship between analysts stock recommendation and fundamental value and a weaker relationship between analysts stock recommendation and conflicts of interest after financial regulations. Overall, these studies suggest a positive analysts stock recommendation response to perceived overvaluation. Conversely, it is uncertain the extent to which analysts contribute to overvaluation. As previously suggested, the overvaluation of equity is inconsistent with the efficient market hypothesis and there may be alternative explanations for price anomalies. However, prior research has documented analyst optimism especially in longterm forecasts and stock recommendation (Barber et al., 2006; Jackson, 2005; Cowen et al., 2006, Agrawal and Chen, 2008). To the extent that analyst optimism increases firm value, analysts behavior may contribute positively to firm overvaluation. Given documented market reaction to analysts reports (e.g., Asquith et al., 2005), analyst optimism may indeed contribute at least temporarily to firm overvaluation. Prior research has documented instances of analysts inefficiency, which may contribute to optimistic forecasts. For example, Bradshaw et al. (2001) find that analysts do not fully adjust for transitory working capital accrual components of earnings, which may result in optimistic earnings forecasts. Similarly, Easterwood and Nutt (1999) argue that, as a result of conflicts of interest, analysts appear to overreact to good news and under-react to bad news. While Elgers et al. (2003) find analysts are more effective than investors in recognizing the difference in the persistence of the accruals and cash flow components of earnings, Kang and Yoo (2007) contradict their finding and argue that analysts interpretation of accruals are more biased than stock prices. Consistent with Kang and Yoo (2007), Ali et al. (2000) find the negative correlation between accruals and 39

53 future abnormal returns is greater for firms with greater analyst following and institutional ownership. This finding may suggest that greater analyst coverage produces greater pressure on management to meet targets, which may result in overvaluation; therefore, firms with greater analyst following are more likely to be overvalued. However Ali et al. (2000) do not use discretionary accruals in their study, so their study does not directly relate to earnings management. 2.4 Summary Prior research has investigated analysts earnings forecast response to earnings management (Guan, 2006; Louis, 2004). The literature finds that analysts respond negatively to earnings management and have a delayed negative response to earnings management in overvalued firms. Figure 1 graphically presents the relationships between earnings management, overvaluation of equity, and analysts consensus stock recommendation response. The chart indicates that overvaluation is positively associated with both earnings management and a more optimistic consensus stock recommendation (Chi and Gupta, 2009; Badertscher, 2010; Bradshaw, 2004; Barniv et al., 2009). It is also possible that earnings management increases overvaluation as a result of the mispricing of accruals, and analysts stock recommendations may increase overvaluation as a result of analysts optimism. Finally, Abarbanell and Lehavy (2003) find that buy (sell) recommendations are associated with income-increasing (decreasing) accruals providing evidence that analysts stock recommendations increase earnings management. However, to the best of my knowledge, the association between earnings management and the analysts stock 40

54 recommendation response has never been investigated. Similarly, the association between earnings management in overvalued firms and analysts consensus stock recommendations response has not previously been tested. The literature suggests that analysts can play a role in reducing the value of firms believed to be overvalued. Analysts are effective information intermediaries and firm monitors who should constrain both earnings management and stock mispricing. On the other hand, analysts incentives may restrict their ability to provide useful information or to effectively monitor firms. While prior research documents analysts optimistic bias to obtain investment banking business, increase trading volume, and optimize their relationship with management to obtain private information, analysts also have incentives to maximize their reputation which provide greater job security and better career prospects. In fact, analysts earnings, forecast accuracy, and stock recommendations profitability are important facets of analysts reputation, which suggests analysts will reflect the implications of earnings management in firms they believe to be overvalued in their stock recommendations. It follows that analysts reputation will also impact their stock recommendation response to earnings management in overvalued firms. Evaluating analysts stock recommendation response to earnings management in overvalued firms will determine the extent to which analysts use earnings management in their stock recommendations, and whether stock recommendation optimism or analysts reputation is more important when faced with earnings manipulation. Figure 2 graphically presents the relationships between earnings management, overvaluation of equity and analysts consensus price target responses. There has been much less research on analysts price targets compared to research on analysts earnings 41

55 forecast and analysts stock recommendations. Thus, the association between overvalued equity and analysts price target response is unknown. Similarly, researchers have not investigated the impact of analysts price targets on earnings management. To the best of my knowledge, there have also not been any studies on analysts price targets responses to earnings management or to overvaluation. However, analysts roles as information intermediaries and firm monitors, as well as their optimism and reputation, will also impact analysts price targets responses to earnings management. Furthermore, Bradshaw (2002) uses a random sample of analysts reports mainly from 1998 and 1999 and finds that price target percentages (price target scaled by price) are positively correlated with more optimistic stock recommendations. Thus, analysts price targets responses may be correlated with analysts stock recommendations responses. 42

56 CHAPTER 3 HYPOTHESES DEVELOPMENT In this section, I develop my predictions for analysts responses to earnings management in all mispriced firms, and I separately develop predictions for analysts responses to earnings management in overvalued firms. In subsection 3.1, I develop my predictions for analysts responses to earnings management in all mispriced firms. First, I create my hypothesis for the relationship between earnings management and analysts responses. I then predict analysts responses to earnings management as firm mispricing increases. Finally, I predict the impact the earnings management method and duration has on analysts responses. In subsection 3.2, I develop my predictions for analysts responses to earnings management in overvalued firms as the length of consecutive periods of overvaluation increases. 3.1 Analysts Responses in Firms that Manage Earnings The separation of ownership and control of the firm creates information asymmetry between management and owners because management has detailed information about the firm that the owners do not have access to (Jensen and Meckling, 1976). Agency theory posits that management is able to use information asymmetry to manage earnings. However, earnings management can be both harmful and useful to firms. The more recent research has focused on the harmful effects of earnings management. For example, prior research finds that firms manage earnings before issuing 43

57 shares (mergers), and post-issue (post-merger) firms tend to perform poorly after the event (Teoh and Wong, 2002; Louis, 2004). Healy (1985) also argues that earnings management can arise from opportunistic management aiming to maximize their bonuses. On the other hand, earnings management is useful when used as an efficient contracting vehicle. It can also be used to unblock communication between management and owners (Demski and Sappington, 1990). Jensen and Meckling (1976) argue that security analysts may have a comparative advantage in monitoring companies. This external monitoring contributes to detecting management misbehavior such as corporate fraud (Dyck et al., 2006; Zeng et al., 2010) and GAAP-based earnings management (Yu, 2008; Guan, 2006). However, researchers have also argued that analyst coverage creates excessive pressure on managers to manage earnings. Brown and Caylor (2005) find that meeting quarterly analysts earnings forecasts is more important to managers than avoiding either quarterly losses or quarterly earnings decreasing. Prior research provides evidence that investors reward (penalize) firms for meeting (missing) analysts quarterly earnings forecasts more than they did for meeting (missing) the other two thresholds (Brown and Caylor, 2005; Matsumoto, 2002). Analysts are also under pressure to provide optimistic forecasts as a result of their need to pursue investment banking business, their need to build and maintain positive relationships with managers, and their need to maximize trading fees (Irvine, 2004; Cowen et al., 2006; Chen and Matsumoto, 2006; Ke and Yu, 2006; Michaely and Womack, 1999). Regulation FD, which prohibits non-public disclosures to financial analysts, could reduce the pressure to curry favor with management. Recent research also suggests a reduction in the premium for, and the likelihood of, meeting analysts earnings 44

58 forecasts post the Sarbanes-Oxley Act of 2002 (Bartov and Cohen, 2007; Koh et al., 2008). However, the use of real transactions management remains prevalent (Cohen et al., 2008; Bartov and Cohen, 2007) and management continues to face pressure to meet analysts optimistic forecasts. Prior research finds a negative association between earnings management and analysts subsequent earnings forecasts (Guan, 2006; Louis, 2004). However, prior research suggests that analyst optimism may reduce any negative analyst response to earnings management. Both Bradshaw et al. s (2001) finding that analysts do not fully adjust for transitory working capital accrual components of earnings and Kang and Yoo s (2007) finding that analysts have a more biased interpretation of accruals than investors suggest that analysts responses to earnings management may not be very accurate. Given analysts monitoring role and empirical evidence that earnings management is negatively associated with analysts earnings forecasts, I predict a negative relationship between income-increasing earnings management and analysts stock recommendations (price targets). As prior research has used GAAP-based earnings management to test the association between earnings management and analysts forecasts, I use income-increasing GAAP-based earnings management to test this relationship. I expect analysts optimism to result in stock recommendations that encourage the purchase of stocks and discourage the sale of stocks. Similarly, analysts optimism should result in higher consensus price targets. I refer to these higher price targets and stock recommendations that encourage (discourage) investors to buy stock (from selling stock) as favorable analyst responses. I also refer to the favorableness of the analysts responses as the degree to which the analyst has a positive outlook on the firm 45

59 and therefore the degree to which analysts set high price targets and recommend (do not recommend) buying (selling) firm stock. My first hypothesis stated in the alternative is as follows: H 1 : The favorableness of analysts responses is negatively associated with incomeincreasing GAAP-based earnings management. I now discuss how the combination of earnings management and fundamental value impacts stock recommendations (price targets). Prior research finds that incorporating analysts earnings forecasts into residual income models results in profitable stock-picking strategies; however, it is unclear whether analysts use their own earnings forecasts in stock recommendations. For example, Bradshaw (2004) finds that analysts stock recommendations are inconsistent with a stock-picking strategy based on a residual income model. Instead, analysts rely on heuristics such as the price-earnings to growth (PEG) or long-term growth forecasts when making stock recommendations (Bradshaw, 2004). These heuristics in decision making suggest analysts may have a positive stock recommendation (and price target) response to earnings management as firm mispricing increases. Specifically, analysts may provide more pessimistic (optimistic) recommendations (price targets) for firms believed to be undervalued (overvalued). In a related perspective, prior research provides evidence that optimistic investor sentiment is associated with more favorable and less profitable stock recommendations (Bagnoli et al., 2010). However, researchers have not previously segregated elements of the financial report on which the analysts stock recommendations or price target are based. Specifically, it is unknown how earnings quality will impact the profitability of analysts stock recommendations and price targets responses. 46

60 Analysts may incorporate earnings quality into their stock recommendations and price targets because analysts have incentives to maximize stock recommendation profitability. Given that analysts have economic incentives to maximize their reputation and that forecast accuracy is a key component of an analysts reputation (Beyer et al., 2010; Jackson, 2005), analysts have incentives to incorporate the negative implications of earnings management into their earnings forecasts, stock recommendations and price targets. Furthermore, Loh and Mian (2006) and Ertimur et al. (2007) find that earnings forecast accuracy is positively associated with stock recommendation profitability, especially in firms with value-relevant earnings. Income-increasing earnings management should adversely affect all firms, especially mispriced firms as managers of these firms are more likely to opportunistically manage earnings for managerial compensation and capital market purposes. Thus, I expect the effect of analysts monitoring role and their incentives to maximize their reputation outweigh the effect of analysts use of heuristics. Consistent with hypothesis 1, I use GAAP-based earnings management to test the relationships. I therefore predict a greater negative relationship between opportunistic GAAP-based earnings management and analysts responses (that is, stock recommendations and prices targets) as mispricing increases. H 2 : The negative association between the favorableness of the analysts responses and income-increasing GAAP-based earnings management is greater as firm mispricing increases. Prior literature finds that various earnings management mechanisms are used by management at the same time and these mechanisms may impact the firm in different ways. The main earnings management choices are accruals management, real transactions 47

61 management, and financial irregularities. First, I discuss the choice between accruals management and real transaction management. Second, I discuss the choice between financial irregularities and the other GAAP-based earnings management mechanisms. Zang (2012) investigates the managerial use of real transactions management (RTM) and accruals management from 1987 to The researcher argues that any or both of these two earnings management mechanisms can be employed to enable firms to just meet or beat an earnings benchmark. Therefore, these earnings management choices are substitutes and should be considered jointly when evaluating earnings management. Zang (2012) finds that the earnings management choice is determined by the relative cost of each earnings management mechanism. The costs associated with real transaction management include market share, financial health, institutional ownership and the firms marginal tax rates. Firms with greater market share incur lower RTM costs because they are generally market leaders with competitive advantages over other firms. However, firms with poor financial health incur higher RTM costs because these firms need to optimize operational efficiency. Similarly, firms with greater institutional ownership incur greater RTM costs because of institutional owners monitoring capacity and superior understanding of the long term implications of the firms operating decisions. Finally, firms with higher marginal tax rates incur greater RTM costs because RTM manipulation will result in greater unwarranted tax payments. Alternatively, the costs associated with accruals management (AM) are BigN auditors, auditor tenure, the Sarbanes Oxley legislation, beginning net operating assets, and the length of the operating cycle. Zang (2012) argues a BigN auditor is an AM cost 48

62 because it may be harder to convince a high-quality auditor to accept aggressive estimates. Similarly, as audit quality increases with tenure, auditor tenure is another relative AM cost. Post SOX observations are also categorized as an AM cost as SOX heightened scrutiny of firms accounting practices resulting in decreasing AM (Cohen et al., 2008). Prior period AM is an accruals management cost because prior period AM constrains AM in the current period (Barton and Simko, 2002). Finally, a shorter operating cycle is a relative AM cost because the firm has smaller accrual accounts and the accruals reverse more quickly. Zang (2012) does not rank the costs but finds evidence that the idiosyncratic cost of each mechanism impacts the firm s choice of the amount of each earning management mechanism to utilize. An analysis of the deterrents suggests that real transaction management is more costly to a firm as the real transaction management cost variables are more likely to be associated with the destruction of firm value. Badertscher (2011) also argues that each earnings management mechanism has costs and benefits. He argues that accruals management may be the least costly because it has no first-order effect on cash flows and therefore has a smaller impact on firm value. Accruals management is also within the boundaries of GAAP and can be performed at period-end to ensure targets are met. However, the reversing nature of accruals creates a constraint on the ability of management to indefinitely use accruals management. RTM is an alternative earnings management choice which alters reported earnings by changing the timing or structuring of operating, investing or financing decisions. Badertscher (2011) argues that RTM is more costly than accruals management because of its adverse 49

63 impact on optimal business operations. RTM therefore impacts both cash flows and earnings and may therefore potentially destroy long term firm value. In addition, Cohen et al. (2008) provide evidence that accruals management decreased and real transaction management increased after the passage of the Sarbanes Oxley (2002) Act. Cohen et al. (2008) argue that firms switched from accruals management to real transaction management (RTM) after SOX even though real transaction management is more costly to shareholders because RTM is also harder to detect. Finally, Cohen and Zarowin (2010) provide evidence that real transaction management is more costly than accrual management at seasoned equity offerings. Using U.S. seasoned equity offering (SEO) from 1987 to 2006, the researchers find that firms use real transaction management as well as accruals management around SEOs. Furthermore, they provide evidence that real transaction management has a greater economic impact than accruals management on the decline in the firms operating performance after the SEO. These studies provide compelling evidence that real transaction management is more costly to firms than accruals management. Managers can also utilize non-gaap earnings management, which can be categorized as financial irregularities and may result in accounting restatements 2. Non- GAAP earnings management may be less likely to be detected than RTM, and it enables management to manage earnings by large amounts. However, non-gaap earnings management is the most costly form of earnings management because of capital market 2 Some restatements may have contributed to earnings management in prior periods and will be included in the GAAP earnings management measure. This noise in the GAAP earnings management measure is a limitation of this study. 50

64 costs of detection, and management may suffer reputational costs in the managerial labor market. Ettredge et al. (2010) provide evidence that restatements follow years of incomeincreasing accruals management and real transactions management. The researchers argue that managers prefer to utilize GAAP-based earnings management before non- GAAP earnings management because non-gaap earnings management results in severe penalties when they are detected and reported. These penalties include increased regulation, legal and debt costs as well as negative market reactions and possible job termination for executives. Using a sample of restatement announcements from 1995 to 2003, Ettredge et al. (2010) find greater balance sheet bloat preceding fraud restatements than non-fraud restatements, greater balance sheet bloat for restatement firms than a control group, and greater past real transaction management for restatement firms than for a control group. Therefore, prior research provides evidence that financial irregularities (as proxied by restatements) are more costly than accruals management and real transaction management. The costs of earnings management are important to analysts as these costs may impact the profitability of stock recommendations and the accuracy of price targets. Specifically, if an earnings management mechanism (EM 1) destroys more firm value than an alternative mechanism (EM 2) ceteris paribus, then the analyst needs to provide a lower stock recommendation for firms using EM 1 in order to optimize the profitability of his/her stock recommendation. Similarly, firms using EM 1 should be priced lower than EM 2 firms; therefore, the analyst s price target should be lower for EM 1 firms than for EM 2 firms. As a result of the differential impact on firm value of the different 51

65 earnings management mechanisms, together with analysts monitoring role and analysts incentives to maximize their reputation, I hypothesize that financial analysts responses to earnings management is dependent on the type of earnings management. Specifically, financial analysts reduce their optimism in their stock recommendation more when faced with more costly methods of earnings management. The most costly earnings management mechanisms are financial irregularities followed by RTM. My third hypothesis stated in the alternative is as follows: H 3 : The negative association between the favorableness of the analysts responses and income-increasing earnings management as mispricing increases is greater as the cost of the method used to manage earnings increases. Prior research also suggests that the longer a firm manages earnings, the more costly the earnings management choice. Badertscher (2011) argues management use more costly methods of earnings management as the earnings management string lengthens. Furthermore, Badertscher (2011) argues that firms use more costly earnings management mechanisms as a result of the constraints caused by high net operating assets. Barton and Simko (2002) argue that a firm s net operating assets are evidence of prior accruals management. As accruals reverse, net operating assets inflated by accruals management act as a constraint on future accruals management. Zang (2012) argues that a firms net operating asset is therefore a cost of accruals management, which may deter future accruals management and instigate greater real transaction management. As real transaction management is more costly than accruals management (Cohen and Zarowin, 52

66 2010), it follows that longer periods of earnings management result in greater destruction of firm value. As a result of this evidence that the length of the consecutive periods of earnings management is positively associated with the destruction of firm value, agency theory, and analysts incentives to make more profitable stock recommendations in order to maximize their reputation, my next hypothesis stated in the alternative is as follows: H 4 : The negative association between the favorableness of the analysts stock responses and income-increasing earnings management as mispricing increases is greater as the length of the consecutive periods of income-increasing earnings management increases. 3.2 Analysts Responses in Overvalued Firms that Manage Earnings as the Length of Overvaluation Increases. In addition, I investigate the relationship between earnings management in overvalued firms and analysts responses as the length of consecutive years of overvaluation increases. Mispriced firms are either overvalued or undervalued and analysts are likely to respond more strongly to earnings management in overvalued firms than to earnings management in other firms. As stated above, prior research finds an insignificant or positive relationship between firms price to fundamental value (P/V) ratios and analysts stock recommendations. A positive relationship suggests that analysts have more optimistic (pessimistic) stock recommendations for overvalued (undervalued) firms. I separately evaluate the association between earnings management as the length of 53

67 overvaluation increases and analysts responses because of the significance of earnings management in overvalued firms. The agency cost of overvalued equity theory (Jensen, 2005) suggests that overvaluation increases the conflicts of interest between the companies owners (the principal) and managers (the agent). This agency cost of overvalued equity is incremental to regular agency costs because securities markets and equity based compensation intensifies, rather than reduces, these conflicts. To motivate management to meet these targets, owners write contracts that tie the meeting of market expectations to management s welfare. Consequently, management s equity incentives, bonuses and even job security are often dependent on the firm s performance (Efendi et al., 2007; Weisbach, 1988). As a result, these contracts motivate management to meet market expectations. For example, managers equity incentives are positively correlated with meeting analysts earnings forecasts and achieving earnings strings (Cheng and Warfield; 2005; Ke, 2001). Therefore, both security markets pressure and equity-based managerial compensation contribute to the agency costs of overvalued equity. Overvalued firms are not able to maintain their stock price via normal operations because if they could, they would not be overvalued. Similarly, if the analysts earnings forecasts cannot explain the stock price, then it is unlikely the firm will be able to report the performance necessary to maintain the stock price. As a result, management has strong incentives to manage earnings in order to meet expectations. The agency costs of overvalued equity ultimately results in the destruction of firm value not only because management is unduly compensated for managed earnings, but mainly because of the costly methods used to manage earnings such as the use of real transaction management 54

68 and the exercise of value destroying acquisitions (Jensen, 2005; Badertscher, 2011). Furthermore, income increasing earnings management in firms with consecutive periods of overvaluation lead to even greater destruction of firm value. Chi and Gupta (2009) find that the longer a firm is overvalued, the greater the effect discretionary accruals have on future operating profit. Using a sample of firms in the Compustat database from 1963 to 2003, Chi and Gupta (2009) find high abnormal accrual overvalued firms underperform low abnormal accrual overvalued firms by about 13 points using industry-adjusted unmanaged EBITDA to assets as the performance measure. Similarly, the researchers find that the length of the overvaluation is negatively associated with future abnormal returns for high abnormal accruals firms. Chi and Gupta (2009) find that the abnormal stock return for high abnormal accrual overvalued firms is about 12% lower during the following year than for low abnormal accrual overvalued firms. In addition, Badertscher (2011) finds that the total amount of earnings management is positively associated with the duration of the overvaluation. The researcher uses seven annual categories to measure the effect of the duration of overvaluation on the amount of earnings management and generally finds that longer periods of overvaluation have a greater effect on accruals management, RTM, and total earnings management. Further, Badertscher (2011) finds that the longer a firm is overvalued, the more likely the firm will be engaged in non-gaap earnings management. In response to the agency costs of overvalued equity, Louis (2004) finds a negative association between income-increasing accruals in overvalued firms undergoing a merger and analysts subsequent earnings forecasts. This finding suggests that analysts 55

69 recognize the negative impact of earnings management on future operating performance when firms are overvalued, and they value forecast accuracy. Therefore, as a result of this evidence that the length of the overvaluation is positively associated with the amount of earnings management, agency theory which posits that analysts have incentives to monitor the firms they cover, and analysts incentives to optimize their reputation, my fifth hypothesis stated in the alternative is as follows: H 5 : The negative association between the favorableness of the analysts responses and income-increasing earnings management in overvalued firms is greater as the length of the overvaluation in firms increases. 3.3 Summary of Hypotheses Table 1 provides a summary of my hypotheses. Hypothesis 1 is a prediction for the overall relationship income-increasing earnings management and analysts responses are negatively related. Hypotheses 2 to 4 provide predictions for analysts responses to income-increasing earnings management as mispricing increases. Hypothesis 2 builds on hypothesis 1 by predicting that, given the negative association between income-increasing earnings management and analysts responses, analysts responses will be incrementally lower as the mispricing increases in earnings management firms. Similarly, hypotheses 3 and 4 build on hypothesis 2. These hypotheses predict a greater negative association between income-increasing earnings management in mispriced firms and analysts responses as the cost of the earnings management mechanisms (the length of earnings management) increases. Finally, hypothesis 5 predicts a greater negative association between income-increasing earnings 56

70 management in mispriced firms and analysts responses as the length of overvaluation increases. I use overvalued firms to represent a subsection of mispriced firms in hypothesis 5. My predictions for all the hypotheses are negative which signifies the reductions in the favorableness of analysts responses and therefore the analysts role in monitoring corporate earnings management. Finding support of these hypotheses suggests the use of earnings management and fundamental value computations in analysts responses. Support of these hypotheses also suggests effective analysts monitoring of opportunistic corporate earnings management. 57

71 CHAPTER 4 RESEARCH DESIGN In this section of the paper, I define my earnings management proxies. I then develop the empirical models used to test the hypotheses, and discuss the dependent, explanatory and control variables. 4.1 Earnings Management Proxies I use three main variables to proxy for earnings management. These measures represent different earnings management mechanisms. The different earnings management mechanisms also signify different costs of earnings management and may reveal the extent of the earnings management as some earnings management methods are more likely to be used after the firm has been managing earnings (or overvalued) for several years. As the focus of this dissertation is on firm mispricing and earnings management, and how they relate to analysts responses, I focus on income-increasing earnings management as mispriced firms have strong incentives to manage earnings upwards in order to maintain high security valuations (Kothari et al., 2006; Badertscher, 2011). I do not expect income-decreasing earnings management to have a significant association with my dependent variables. However, I include income-decreasing earnings management in my models in order to capture the total effects of earnings management on analysts responses. 58

72 My first measure of earnings management (EM) is accruals management, proxied by discretionary accruals. I adapt the modified Jones model to estimate discretionary accruals as follows: 3 ACCRUALS it = α + β 1 ( REV it - AR it ) + β 2 PPE it + β 3 ROA it + є it (1) where: ACCRUALS it is the total accruals of firm i in year t (see equation 2 below) REV it is the change in sales revenue of firm i in year t AR it is the change in accounts receivable of firm i in year t PPE it is gross property, plant and equipment for firm i at the end of year t ROA it is return on assets for firm i in year t All the variables, including the constant term, are scaled by total assets for firm i at the beginning of year t (i.e., A it-1 ). The model is estimated annually for each three-digit industry. 4 The residual represents accruals management as captured by abnormal accruals (ABNACCRUALS). My first abnormal accruals variable captures positive abnormal accruals (PABNACCRUALS) whereby all negative values of abnormal accruals are set to 0 to reflect only income-increasing earnings management. For negative abnormal accruals (NABNACCRUALS), all positive values of abnormal accruals are set to 0 to reflect only income-decreasing earnings management. Total accruals are calculated as follows: 3 The model is adapted for performance since it is important to control for financial performance when earnings management may be correlated with financial performance (Dechow et al., 1995; Kothari et al, 2005) 4 For abnormal accruals and RTM proxies, I require a minimum of ten observations in each three-digit industry-year in order to estimate abnormal accruals and RTM. If ten observations are not available, then I use a two-digit industry-year specification. 59

73 ACCRUALS it = INCOME it - CFO it (2) where: INCOME is the income before extraordinary items for firm i in year t. CFO is the cash flow from operations for firm i in year t. Following Roychowbury (2006), I use three empirical proxies for real transaction management (RTM): abnormal cash flow, abnormal discretionary expenditures, and abnormal inventory. Sales manipulation can be identified by computing abnormal operating cash flows. This manipulation could include offering more discounts, providing lenient credit terms as well as channel stuffing. I expect sales manipulation to reduce operating cash flow after controlling for the level of and change in sales. Normal cash flow is estimated by regressing sales and change in sales on the annual operating cash flow as follows in Equation (3): CFO it = α + β 1 REV it + β 2 REV it + є it (3) where: REV is the net sales revenue of firm i in year t All other variables are previously defined. All the variables, including the constant term, are scaled by total assets for firm i at the beginning of year t (i.e., A it-1 ). The model is estimated annually within each three-digit industry. The residual represents my proxy for RTM sales manipulation as captured by abnormal cash flow (ABNCFO). I expect firms increasing earnings by manipulating sales to have a negative abnormal cash flow. My second proxy for RTM is abnormal discretionary expenditures. Discretionary expenditures are defined as R&D expense, advertising expense and selling, general and administrative (SG&A) expenses (Roychowbury, 2006; Badertscher, 2011). Firms can increase reported earnings and meet earnings benchmarks by reducing discretionary 60

74 expenditures. Normal discretionary expenditures are estimated in EQUATION (4) as follows: DISEXP it = α + β 1 REV it + β 2 REV it-1 + є it (4) where: DISEXP is the sum of R&D expense, advertising expense and SG&A for firm i in year t. REV it-1 is the change in net sales revenue of firm i in year t-1 All other variables are previously defined. Once again, all the variables are scaled by total assets for firm i at the beginning of year t and the model is estimated annually within each three-digit industry. The residual represents my proxy for RTM discretionary expenses manipulation (ABNDISPEXP). I expect firms manipulating discretionary expenditure to manage earnings to have a negative abnormal discretionary expenditure. My third proxy for RTM is abnormal inventory production. Management can increase profits by increasing inventory which has the effect of capitalizing a greater proportion of the fixed costs. Using Roychowbury (2006), I estimate the normal level of production costs in equation (5) as follows: PROD it = α + β 1 REV it + β 2 REV it + β 3 REV it-1 + є it (5) where: PROD is the cost of goods sold plus change in inventory for firm i in year t. All other variables are previously defined. All the variables are scaled by total assets for firm i at the beginning of year t and the model is estimated annually within each three-digit industry. The residual represents my proxy for RTM inventory manipulation (ABNPROD). I expect firms manipulating inventory to have a positive abnormal inventory. 61

75 Consistent with prior research, I construct a composite abnormal RTM variable (ABNRTM). First, I multiply ABNCFO and ABNDISPEXP by negative one so that negative abnormal cash flows and discretionary expenditures represent higher levels of income-increasing RTM. Next, I sum ABNCFO, ABNDISPEXP and ABNPROD. Greater levels of ABNRTM represent greater use of RTM to manage earnings. PABNRTM captures income-increasing earnings real transaction management by setting all negative values of abnormal RTM to 0, whereas NABNRTM captures incomedecreasing earnings real transaction management by setting all positive values of abnormal RTM to 0. To compute total GAAP earnings management, I create composite GAAP earnings management variables: GAAPEM, PGAAPEM and NGAAPEM. GAAPEM is the sum of accruals management (ABNACCRUALS) and total abnormal RTM (ABNRTM). PGAAPEM captures income-increasing GAAP-based earnings management by setting all negative values of GAAPEM to 0. This variable represents greater use of GAAP-based earnings management to manage earnings upwards. NGAAPEM, which captures income-decreasing GAAP-based earnings management by setting all positive values of GAAPEM to 0, represents greater use of GAAP-based earnings management to manage earnings downwards. Finally, I use restatements as my proxy for the most costly form of earnings management. Restatements represent financial irregularities or non-gaap earnings management, which can be very costly to a firm as evidenced by large negative abnormal returns on restatement announcements. I use restatements that correspond to the effective 62

76 year of the financial irregularity in my models; however, these restatements may be unobservable in the financial period. I expect financial analysts to adjust for restatements because of their role as firm monitors. First, analysts may be able to identify accounting irregularities as they are incurred and before they are reported. For example, Zeng et al. (2010) find that analysts report accounting problems in 11 percent of their sample of restatements before the problems were publicly disclosed. Also, analysts may surmise that the managed or fraudulent results are unusual or unsustainable based on their expectations and knowledge of the firm and industry. I use an indicator variable (RESTATE) to identify restatements by the client related to the fiscal year. I also separate restatements into income-increasing restatements 5 (PRESTATE) and income-decreasing restatements (NRESTATE), which are also indicator variables. Consistent with Badertscher (2011) and Ettredge et al. (2010), I include only restatements that are described as an accounting irregularity and not those described as an accounting error. Thus, I focus on only those restatements that question the quality of financial reporting. Hennes et al. (2008) argue that distinguishing between irregularities and errors is important when drawing inferences and significantly enhances the power of tests related to restatements. 5 Income-increasing restatements are adverse restatements. These restatements are likely to be a result of managerial behavior to increase earnings in the past. 63

77 4.2. Analysts Responses to Earnings Management In this section, I outline the proxies used to capture mispricing and overvaluation in order to test the hypotheses. I then present the dependent variables and regression models Main Mispricing Proxy My main mispricing proxy is the stock price to fundamental value (P/V) ratio 6. The P/V ratio is calculated by dividing the current price of the security by the fundamental value of the security. Following Bradshaw (2004), I estimate a firms fundamental value as the present value of expected residual income for the next five years plus a terminal value (TV). P/V ratios are based on the terminal value assumption that abnormal returns persists in perpetuity 7 (Ohlson, 1995; Frankel and Lee, 1998). The residual income model is specified in equation (6) as follows: V t = BVPS t + E t[ri t+n ] 5 n=1 (1+r) n + E t[ri t+5 ] r(1+r) 5 (6) where: V t is the estimate of the intrinsic value (fundamental value) of the stock at time t BVPS is book value per share at time t E t [ ] is the expectation based on available information at time t. 6 The alternative proxy used for mispricing is firm overvaluation (see subsection 4.2.2). 7 This is variable V R11 in Bradshaw (2004). The terminal value uses the assumption that abnormal returns persists in perpetuity. In Bradshaw (2004), the correlation between V R12 and stock recommendation is not significant. 64

78 r is equity cost of capital n is the time period used in estimating the fundamental value. I use the Fama and French (1997) industry classification to compute firms equity cost of capital and CRSP monthly indices file to provide the 30-day Treasury bill rates used in the equity cost of capital calculations. RI is residual income calculated as specified in equation (7) below: RI t+n = EPS t+n r*bvps t+n-1 (7) where: EPS is financial analysts computed earnings per share at time t using analysts earnings forecasts as expected future earnings. All other variables are previously defined. The model requires earnings forecasts for up to five years. If long horizon earnings forecasts are unavailable, then these forecasts are estimated by using the longterm growth forecast (LTG). If LTG is unavailable, then earnings forecasts are computed assuming the return on equity remains constant over the forecast horizon. Future book values are estimated from beginning period book values, forecasted earnings and expected dividend payouts consistent with the clean surplus assumption. Consistent with prior research (Frankel and Lee, 1998; Bradshaw, 2004; Barniv et al., 2009), the dividend payout ratio is assumed to remain constant (based on the most recent fiscal year). This residual income valuation model assumes that the residual income in the terminal year persists in perpetuity (Bradshaw, 2004; Badertscher, 2011). The assumption under which the terminal value persists in perpetuity is a less conservative assumption than the fade rate assumption. Therefore, this terminal value assumption reduces the amount of overvaluation of equity compared to the fade rate terminal value assumption. 65

79 Under the assumption that the terminal value persists in perpetuity, Bradshaw (2004) finds analysts stock recommendations to not be significantly associated with stockpicking strategies based on a residual income model. As the forecast horizon in this study is 5 years, the terminal value is based on residual income in year t+5. The equity cost of capital is based on Fama and French s full factor 3-year industry cost of equity (Fama and French, 1997). Fama and French (1997) use 5-year rolling regressions of excess returns 8 to calculate their industry risk premium. Equity cost of capital is the industry risk premium plus the risk-free rate, proxied by the 30-day Treasury bill rate at the beginning of the recommendation month. Fama and French (1997) argue that, while the cost of equity for industries are imprecise, mainly due to uncertainty about the true factor risk premiums, firms cost of equity would unquestionably be less precise as their factor risk premium would be greater Alternate Mispricing Proxy Mispriced firms that have a greater market value than fundamental value are, by definition, overvalued. P/V ratios can therefore be used to identify both firm mispricing and firm overvaluation. My alternate mispricing proxy is a firm analysts believe to be overvalued. Following Badertscher (2011), I define equity believed by analysts to be overvalued as the firms in the highest quintile of stock price to fundamental value (P/V) ratios. These firms are expected to have P/V ratios much greater than one and are the most highly valued firms in the sample. 8 Excess returns are regressed on market returns, and two mimicking portfolios that proxy for size and book-to-market. 66

80 In order to identify the top quintile of P/V ratios, I use P/V rankings based on annual portfolios for fiscal year t+1. This enables the fundamental value calculation to include the effect of the fiscal year financial report, which includes the disclosure of any earnings management. Although the firms have different year-end dates, the P/V ratios are comparable because any variations in risk-free interest rate will be incorporated in both the price and the fundamental value calculation Dependent Variables I use two proxies for analysts responses to earnings management. The consensus stock recommendations favorableness ratio is my main dependent variable. 9 First, consistent with prior research (e.g., Bradshaw, 2004; Chen and Chen, 2009), I invert the coding of the stock recommendations in I/B/E/S so that Strong Sell is represented by the number 1, Sell is 2, Hold is 3, Buy is 4 and Strong Buy is represented by the number 5. This coding is more intuitive, as a larger number represents a more optimistic recommendation. Next, I reduce each number by 1 so that the stock recommendations range from 0 to 4 (where 0 is Strong Sell and 4 is Strong Buy). This variable is referred to as REC. Finally, I divide REC by 4 to provide a favorableness ratio (FAVREC). For example, a favorableness ratio of 1 represents a consensus stock recommendation of Strong Buy whereby a favorableness ratio of 0.25 represents an average stock recommendation of Sell. My second proxy for analysts responses is the favorableness of analysts consensus price targets (FAVPTG). FAVPTG is a ratio calculated by taking the first 9 I use the first consensus stock recommendation in the following year as the dependent variable. This ensures analysts first response to reported earnings management is captured. 67

81 consensus price target of the following year scaled by the stock price at the date of the analysts report. I expect analysts to forecast a lower price target ratio in the presence of value-destroying earnings management. This proxy is a continuous variable Regression Models Prior research finds that analysts own forecasts can be used to calculate fundamental firm value using the Ohlson model (Ohlson, 1995; Frankel and Lee, 1998). Therefore, I am able to test analysts responses to earnings management in firms that analysts have forecast as mispriced. This research design does not challenge any efficient market assumptions. I start with the base model developed in Bradshaw (2004). The base model is as follows: RESPONSE it = α + β 1 V/P it + YEARS+ ε (8) where: V/P represents the fundamental value determined in equation 6 scaled by price. The Ohlson (1995) model is used to calculate the fundamental value, using analysts consensus earnings forecasts as expected earnings. YEARS represent a vector of indicator variables representing each year. This captures a different intercept for each year. Earlier studies included variables such as earnings growth, earnings-to-price ratio, and market-to-book value as control variables for the stock recommendation level in their regression models. However, Dong et al. (2006) argue that the P/V ratio filters out the growth expectations and it is a pure measure of overvaluation. Further, prior research has found that analysts fail to take advantage of market anomalies found in variables such as book-to-market, price momentum and earnings momentum (Finger and Landsman, 1998). Thus, it is unlikely that these variables explain stock recommendations or price 68

82 targets, and they are unlikely to add to my empirical models. Besides, if an investor s objective is to maximize wealth, the stock s fundamental value, as indicated by the V/P ratio, is the only information required to make the best investment decision. Therefore, to make profitable stock recommendations, the analysts simply need to provide stock recommendations based on the V/P ratio (Frankel and Lee, 1998; Bradshaw, 2004). For example, the analysts forecast errors help to explain current analysts earnings forecasts. However, this variable is not needed to explain analysts stock recommendations because current earnings are incorporated in the stock price as well as the fundamental value of the stock, and the previous analyst earnings forecast is irrelevant to any stock trading decision. In other words, the relevant effects of forecast error are already included in the V/P ratio. As a result, this model is parsimonious. Barniv et al. (2009) and Chen and Chen (2009) use similar parsimonious models in their investigations of the impact of changes in regulations on analysts stock recommendations. The researchers add explanatory variables such as Regulation FD and analysts affiliation to their models but do not add any additional control variables. I modify equation 8 to include my earnings management variables of interest to test my hypotheses. First, I inverse the V/P ratio and use the price-to-fundamental value ratio (P/V ratio) so that higher values of the measure suggest that analysts should provide more negative analysts responses. I use the composite measure based on abnormal RTM and discretionary accruals discussed above that is considered GAAP-based earnings management. The analyst response levels model is as follows: RESPONSE it+1 = α + β 1 PGAAPEM it + β 2 NGAAPEM it + β 3 MISPRICING it+1 + β 4 MISPRICING it+1 *PGAAPEM it + ε it (9) 69

83 where: RESPONSE represents proxies for analysts responses using a levels model. This dependent variable is either level of favorableness in stock recommendation (FAVREC) or level of favorableness in price target (FAVPTG). Analyst response is estimated by firm i at year t + 1, where t+1 represents the first consensus analysts stock recommendation (price target) for the following fiscal year. MISPRICING is the P/V ratio computed as the market price at the date of the analyst s response from the CRSP database and value is based on the fundamental value of the stock at the date of the analyst s response derived from Ohlson (1995) as discussed in the preceding section. PGAAPEM represents income-increasing GAAP-based earnings management and is the composite measure summing income-increasing abnormal RTM and income-increasing discretionary accruals. This variable is positive or equal to 0 (see subsection 4.1). NGAAPEM represents income-decreasing GAAP-based earnings management and is the composite measure summing income-decreasing abnormal RTM and income-decreasing discretionary accruals. This variable is negative or equal to 0 (see subsection 4.1). Bradshaw (2004) hypothesized a positive association between analysts stock recommendations and fundamental value based on the V/P ratio. This hypothesis was consistent with Frankel and Lee (1998), who found that the V/P ratio predicted future cross-sectional returns. However, Bradshaw (2004) found a negative or insignificant relationship between stock recommendations and the V/P ratio, although subsequent 70

84 research suggests this negative relationship may be dwindling post-financial regulations. As I invert the V/P ratio to find the firm s P/V ratio, a positive or insignificant association between the P/V ratio and stock recommendations (price targets) is expected. I use the PGAAPEM as my proxy for income-increasing earnings management. This earnings management composite is used to test the relationship between analysts responses (either stock recommendations or price targets) and earnings management. A negative coefficient is expected for PGAAPEM consistent with the first hypothesis. The interaction of the income-increasing composite for GAAP-based earnings management with the ratio of price-to-fundamental value is included to test the second hypothesis regarding whether there is a differential analysts responses to income-increasing earnings management as the mispricing in firms increases. I expect a negative association consistent with the second hypothesis. Income-decreasing earnings management (NGAAPEM) may reflect the fact that accruals reverse. Therefore, income-decreasing earnings management may indicate constraints on earnings management (Barton and Simko, 2002) which would suggest an association with lower stock recommendations (price targets). On the other hand, income-decreasing accruals and excess spending could reflect big bath behavior or reactions to poor operational performance. Big bath behavior is undertaken to enable firms to meet their targets in the future, which should be associated with higher stock recommendations (price targets). However, write-offs, excess spending or lower production levels as a result of poor operational performance may be associated with lower stock recommendations (price targets). Therefore, I am unable to predict the sign of the relationship between income-decreasing earnings management and stock 71

85 recommendations (price targets). As previously discussed, I do not test analysts responses to income-decreasing earnings management as mispricing increases (that is, I do not interact NGAAPEM with the P/V ratio) because the relationship between earnings management and firm mispricing is asymmetric. I modify equation 9 to address the differential cost of the different mechanisms used to manage earnings (hypothesis 3). I expect accruals management to be the least costly earnings management mechanism, followed by real transactions management. Financial irregularities are predicted to be the most costly earnings management mechanism. Similar to equation 9, I split my earnings management measures into income-increasing and income-decreasing elements. As a consequence of analysts monitoring roles and their incentive to maximize their reputation, I expect the negative relationship between earnings management and analysts responses to be greater for more costly income-increasing earnings management mechanisms 10. Equation 10 uses variables representing accruals management, real transactions management, and financial irregularities to model analysts responses to earnings management as follows: RESPONSE it+1 = α + β 1 MISPRICING it+1 + β 2 PABNACCRUALS it + β 3 PABNRTM it + β 4 NABNACCRUALS it + β 5 NABNRTM it + β 6 PRESTATE it + β 7 NRESTATE it + β 8 MISPRICING it+1 *PABNACCRUALS it + β 9 MISPRICING it+1 *PABNRTM it + β 10 MISPRICING it+1 *PRESTATE it + ε it (10) 10 As a consequence of the idiosyncratic impact of earnings management on firms and firms differential P/V ratios, it is impossible to exactly assess the costs of different types of earnings management. My comparisons of the earnings management mechanisms are based on mean P/V ratios. 72

86 where: PABNACCRUALS is income-increasing discretionary accruals (scaled by total assets at the beginning of the fiscal year) for firm i in year t based on the modified Jones model and adjusted for performance. This variable is positive or set to 0. : NABNACCRUALS is income-decreasing discretionary accruals (scaled by total assets at the beginning of the fiscal year) for firm i in year t based on the modified Jones model and adjusted for performance. This variable is negative or set to 0. PABNRTM is the composite measure summing abnormal cash flows (proxy for sales manipulation), abnormal discretionary expenses and abnormal inventory (all scaled by total assets at the beginning of the fiscal year) when the composite value is positive. These variables are estimated using RTM models from Roychowbury (2006). This variable is positive or set to 0. NABNRTM is the composite measure summing abnormal cash flows (proxy for sales manipulation), abnormal discretionary expenses, and abnormal inventory (all scaled by total assets at the beginning of the fiscal year) when the composite value is negative. This variable is negative or set to 0. PRESTATE is an indicator variable taking the value of 1 if the firm reported an adverse restatement which relates to the period before the current years earnings announcement and after the previous years earnings announcement, and 0 otherwise. 73

87 NRESTATE is an indicator variable taking the value of 1 if the firm reported a favorable restatement which relates to the period before the current earnings announcement and after the previous years earnings announcement, and 0 otherwise. All other variables are previously defined. I use the variables PABNACCRUALS, PABNRTM, NABNACCRUALS, NABNRTM, PRESTATE and NRESTATE as my proxies for earnings management in equation As a result of analysts monitoring roles and evidence in prior research that earnings management is negatively associated with analysts earnings forecasts, I predict that β 2, β 3 and β 6 are all negative. Income-decreasing earnings management may indicate constraints on earnings management (Barton and Simko, 2002), big bath behavior, or reactions to poor operational performance. Thus, as the possible reasons for reporting income-decreasing earnings management ranges from signaling actions to enable steady, improving earnings in the future to reporting poor operational performance, I am unable to predict the sign of the relationship between the variables NABNACCRUALS and NABNRTM, and stock recommendations (price targets) responses. In a similar fashion, I am unable to predict the sign of the relationship between NRESTATE and analysts responses. 11 NRESTATE represents a restatement which increases earnings in the current period. However, this restatement is a result of a financial irregularity which decreased earnings in the current or prior period. Therefore, NRESTATE is considered income decreasing for the purpose of this study. Similarly, PRESTATE is considered income increasing. 74

88 The interaction of the separate income-increasing earnings management methods (PABNACCRUALS, PABNRTM and PRESTATE) with the P/V ratio is primarily used to test hypothesis 3. Consistent with hypothesis 2, I expect the coefficient on these interactions to be negative as the combination of income-increasing or opportunistic earnings management in mispriced firms may signal the destruction of firm value and a negative association is consistent with residual income model predictions. I expect negative analysts responses to the destruction of firm value. Specifically, analysts are firm monitors, and they have incentives to optimize their reputation by producing profitable recommendations and accurate price targets (Loh and Mian, 2006; Ertimur et al., 2007). Furthermore, I expect β 10 to have greatest effect on the dependent variables. Next, I expect β 9 to have a greater effect on the dependent variables than β 8, consistent with analysts adjusting their recommendations and price targets more for the most costly earnings management choices (Cohen and Zarowin, 2010; Cohen et al., 2008, Ettredge et al., 2010). These predictions are consistent with hypothesis 3. My next model is used to test hypothesis 4 by investigating the relationship between the length of earnings management as firms mispricing increases and analysts responses. This model builds on equation 9 as it seeks to test whether the negative association between analysts responses and GAAP-based earnings management in mispriced firms is greater as the length of consecutive periods of income-increasing earnings management (PGAAPEMN) increases. I expect the coefficient on the interaction between MISPRICING, PGAAPEM and PGAAPEMN to be negative. Equation 11 models analysts responses and uses an earnings-management length variable to proxy for the length of the earnings management as follows: 75

89 RESPONSE it+1 = α + β 1 MISPRICING it+1 + β 2 PGAAPEMN it + β 3 NGAAPEMN it + β 4 PGAAPEM it + β 5 NGAAPEM it + β 6 MISPRICING it+1 *PGAAPEM it + β 7 MISPRICING it+1 *PGAAPEMN it *PGAAPEM it + ε it (11) where: PGAAPEMN it represents the number of years of consecutively reporting income-increasing GAAP-based earnings management. NGAAPEMN it represents the number of years of consecutively reporting income-decreasing GAAP-based earnings management. All other variables are previously defined. PGAAPEMN (NGAAPEMN) captures the length of time (in number of years) that a firm has been reporting income-increasing (decreasing) GAAP-based earnings management. As income-increasing earnings management is expected to be negatively associated with the RESPONSE, I predict a negative relationship between the number of years of income-increasing GAAP-based earnings management and analysts stock recommendations (price targets) responses. Consistent with hypothesis 2, β 6 tests the association between analysts responses and earnings management as mispricing increases. β 7 is used to test hypothesis 4. I predict that the negative association between analysts responses and income-increasing GAAP-based earnings management as mispricing increases to be greater as the length of income-increasing earnings management increases consistent with hypothesis 4. However, I am unable to predict the sign of the relationship between NGAAPEMN, and stock recommendations (price targets) as firms have several competing reasons for reporting income-decreasing earnings management. 76

90 4.3. Consecutive Periods of Overvaluation. In this section, I present the regression models used to test analysts stock recommendations (price targets) responses to earnings management in firms believed to be overvalued as the consecutive periods of overvaluation increases Regression Models In these regression models I use the alternate mispricing proxy, firm overvaluation (OVER), to represent firm mispricing. Overvalued firms are mispriced firms in the top quintile of P/V ratios. Therefore, I replace MISPRICING with OVER in these models. Consistent with the other regression models, I also build on equation 9 and include my other variables of interest to test hypothesis 5. This hypothesis investigates the relationship between the length of overvaluation (OVERN) in firms that manage earnings and analysts responses. I expect the coefficient on the interaction between OVERN and income-increasing GAAP-based earnings management (PGAAPEM) to be negative as the number of years of overvaluation increases. Equation 12 models analysts responses and uses an overvaluation of equity variable to proxy for the number of years of overvaluation as follows: RESPONSE it+1 = α + β 1 OVER it +1 + β 2 OVERN it+1 + β 3 PGAAPEM it + β 4 NGAAPEM it +β 5 OVER it+1 *PGAAPEM it +β 6 OVERN it *PGAAPEM it + ε it (12) where: OVER is an indicator variable equal to 1 if the firm was in the top quintile of P/V ranking (based on annual portfolios formed for each fiscal year t+1) and 0 otherwise as discussed. OVERN is a discrete variable which represents the number of years of consecutive overvaluation. A firm is overvalued if it was in the top 77

91 quintile of P/V ranking (based on annual portfolios formed for each fiscal year t+1). All other variables are previously defined. I use the indicator variable OVER, which represents overvalued firms, as my alternative proxy for mispricing. Prior research has found stock recommendations to be negatively or insignificantly associated with residual income model predictions based on V/P ratios (Bradshaw, 2004; Barniv et al., 2009; Chen and Chen, 2009). Thus, one may expect OVER to be insignificantly or positively associated to stock recommendations based on P/V ratios. However, prior research has not investigated whether it is overvalued firms or undervalued firms that are driving these results. Thus, I make no prediction about the sign of the coefficient on this dichotomous variable in any of the models. A negative coefficient is expected for PGAAPEM consistent with hypothesis 1. The interaction of the income-increasing composite for GAAP-based earnings management with overvaluation is included to provide additional evidence on hypothesis 2 regarding whether there is a differential analysts response to income-increasing earnings management for mispriced firms. I expect a negative association consistent with the second hypothesis. OVERN captures the number of years a firm has been overvalued. Although Badertscher (2011) finds that the longer a firm is overvalued, the more severe its choice of earnings management, there is no evidence that an overvalued firm is managing earnings. It is conceivable that a firm has maintained a high P/V ratio without managing 78

92 earnings, and therefore no destruction of value can be predicted. Thus, I am unable to predict the sign of the coefficient for this variable. However, the interaction between OVERN and the PGAAPEM variable is expected to lead to greater destruction of firm value as OVERN increases. Thus, I predict a negative coefficient on the interaction between OVERN and PGAAPEM consistent with hypothesis Post-Regulation FD Analyses Regulation Fair Disclosure 2000 prohibited management from selectively disclosing private information to a select group of analysts. The intent of Regulation FD is therefore to reduce analysts reliance on management for private information and ultimately to reduce analysts bias. Using all the models outlined in this section, I separately test for the favorableness of analysts stock recommendations (price targets) responses to earnings management as mispricing increases and in overvalued firms after Regulation FD. Prior research has provided evidence that stock recommendations have become less optimistic and more consistent with residual income models after the financial regulations. Similarly, Guan (2006) finds analysts adjusting their earnings forecasts more for earnings management following Regulation FD. However, it is unknown how Regulation FD will affect analysts stock recommendations and price targets responses to earnings management. Therefore, although I expect Regulation FD to make the association between analysts responses and earnings management in mispriced firms (overvalued firms) more negative, I make no formal prediction for the impact of Regulation FD on analysts stock recommendations (price targets) responses. 79

93 CHAPTER 5 DATA In this section of the paper, I describe the sample selection procedure used to generate my samples. I generate separate samples to analyze analysts level of (changes in) recommendations and analysts level of (changes in) price targets. I then analyze the descriptive statistics and discuss any trends in the data. Finally, I discuss the correlations between the variables used in the regression models. 5.1 Sample Selection I obtain analyst consensus earnings forecast data, recommendations data, and price target data from Institutional Brokers' Estimate System (I/B/E/S) summary files, annual financial data from Compustat North America, and firm stock price data from the Center for Research in Security Prices (CRSP) databases from fiscal years 1992 to For financial irregularities, I use Audit Analytics restatement data from 2000 to 2010 and exclude all restatements resulting from accounting errors (Ettredge et al., 2010; Badertscher, 2011). Table 2 describes my sample selection procedure. The initial sample begins with 113,260 consensus earnings forecast firm observations from the I/B/E/S database for the fiscal years 1992 to I lose 23,160 observations as a result of merging the I/B/E/S database and Compustat North America. I also lose 4,367 observations as a result of merging the data with the CRSP database. In addition, 15,262 observations from financial 80

94 services and utilities industry companies (SIC Codes and , respectively) are excluded as a result of regulations and industry-specific financial reporting issues. To employ the residual income model, I require that analyst forecasts for the firm are included in the I/B/E/S database and that firms have both one-year-ahead and twoyear-ahead earnings-per-share forecasts reported in I/B/E/S. Also, in order to be included in this study, each observation must have sufficient financial information to compute earnings management in Compustat from 1994 to Furthermore, I remove firms with negative book values as the return on equity (ROE) for these firms cannot be meaningfully interpreted (Frankel and Lee, 1999; Cao, 2007; Badertscher, 2011). I also remove firms with dividend payout ratios or ROEs greater than 100% as these extremely high values are most likely the result of extremely low book values or earnings. Finally, I remove all firms that have a stock price or a fundamental value per share of less than $1. Excluded as well are 2,820 observations as a result of missing I/B/E/S analysts forecast data; 21,530 are excluded due to missing financial information, and I exclude 5,389 observations as a result of extreme observations. This procedure results in a sample of 40,732 observations from 9,965 firms. Next, I merge Compustat and analyst forecast data with analyst recommendations data (analyst price target data) on I/B/E/S. My recommendations sample excludes 4,380 observations for missing recommendation data, leaving a final recommendations levels sample of 36,352. In addition, 18,054 observations are excluded from my price target sample as a result of missing price target data, leaving a final price target levels sample of 22,

95 As I seek to investigate the association between earnings management as mispricing increases (in overvalued firms) and the favorableness of analysts responses after controlling for Regulation FD. I create post-regulation FD samples and run separate post-regulation FD regressions. Table 2 indicates that there are 21,890 post-regulation FD stock recommendations observations and 19,572 post-regulation FD price target observations. As Audit Analytics only began reporting restatement data in 2000, restatements are only included in post-regulation FD regression models. 5.2 Descriptive Statistics Table 3 shows descriptive statistics for the variables used in this study. Panel A defines the variables. All variables are winsorized at the 1% and 99% levels to reduce the impact of outliers while retaining observations (Kennedy et al., 1992). In Panel B, I compare firm observations in the full sample with the post- Regulation FD sample. Firms had a larger mean market value of equity but a lower mean ROE post-regulation FD compared to the whole sample period. This may be the result of the lower mean cost of equity post Regulation FD. Firms also reported larger income before extraordinary items, larger book values per share, and a higher fundamental value post-regulation FD. The increases in these performance indicators may be partially due to inflation. However, post-regulation FD, firms had lower growth potential as reflected in the lower long-term growth forecasts and higher book-to-market ratios. Mainly because the difference between mean prices reported for the full sample and those reported post-regulation FD is insignificant, the P/V ratio and the price to earnings ratio are both lower after Regulation FD. This finding suggests that mispricing 82

96 and especially overvaluation is lower post-regulation FD. However, the average length of consecutive periods of overvaluation increased post-regulation FD which suggests a trend of maintaining overvaluation once it is achieved. Consistent with a decrease in analysts optimism bias post-regulation FD, the post-regulation FD sample shows lower recommendations, price targets, and buy percentages. Significantly lower income-increasing earnings management (especially discretionary accruals) is also reported in the post-regulation FD sample, consistent with trends reported in recent research (e.g., Cohen et al., 2008). Income-increasing restatement are over six times as frequent as increase-decreasing restatement, which suggests that management are more likely to use non-gaap earnings management opportunistically. Interestingly, the descriptive statistics also show significant increases in income-increasing and income-decreasing earnings management strings post- Regulation FD. This trend could suggest more consistent reporting or more opportunistic managerial behavior. 5.3 Annual Trends As I seek to investigate analysts responses to earnings management in mispriced firms, I also use trend data to report the differences between firms with the greatest levels of earnings management and those with the lowest earnings management levels over time. This analysis may shed light on the likely analysts responses and the changes in analysts responses over time. Prior research has found that corporate regulations such as Regulation FD and the Sarbanes-Oxley Act (2002) impact variables such the favorableness of analysts responses and the levels of earnings management (Barniv et 83

97 al., 2009; Chen and Chen, 2009; Bartov and Cohen, 2007; Cohen et al., 2008). Economic conditions such as economic booms and regressions are also likely to impact these key variables. I focus my attention on trends for my dependent variables, the explanatory variables and key firm characteristics to shed further light on possible reasons for changes over time. Figure 3 reports trend analysis for the fifth quintile of income-increasing earnings management firms (hereafter 5QEM firms) compared to the first quintile of earnings management firms (hereafter 1QEM firms). The fifth quintile of income-increasing earnings management represents the highest levels of income-increasing earnings management. Panels A and B show trend data for the mean market value of equity and ROE respectively. High-earnings management firms are consistently smaller than the low-earnings management firms. The market value of equity trend chart shows many smaller firms became low-earnings management firms during 2008, most likely as a result of the recession. High-earnings management firms also consistently report a lower return on equity than 1QEM firms. Once again, the recession of 2008 resulted in a large reduction in ROE for larger firms that report lower earnings management. This is likely the result of big bath behavior; however, smaller firms that report high levels of earnings management did not appear to also take a big bath. Panels C and D show trend data for analysts consensus recommendations and analysts consensus price targets respectively. High-earnings management firms consistently receive higher stock recommendations than low-earnings management firms. The high stock recommendations are most likely due to the high-growth potential of these firms, which may also partially explain the high level of accruals. This chart 84

98 provides evidence of lower recommendations during economic downturns and recessions, and higher recommendations during economic upturns. Conversely, mean price targets are higher for low-earnings management firms mainly because these firms are generally larger than high-earnings management firms. I now turn to the earnings management trends found in my sample. Panels E and F report discretionary accruals trends and real transactions management trends for 5QEM firms compared to 1QEM firms respectively. Highearnings management s income-increasing discretionary accruals range from approximately 16% of lagged assets to just over 5% of lagged assets. By comparison, high-earnings management s income-increasing real transaction management ranges from about over 50% of lagged assets to over 90% of lagged assets. Both incomeincreasing discretionary accruals and real transaction management appears to have decreased after Regulation FD. Next, I discuss the correlation between the variables in the regression models. 5.4 Correlation Matrices Table 4 shows the correlation matrices for the recommendations (price targets) regression models. Panel A reports the correlation for the recommendations regression model. The favorableness of the recommendation (FAVREC) is positively correlated with the P/V ratio consistent with prior research (Bradshaw, 2004; Badertscher, 2011; Barniv et al., 2009). FAVREC is positively (negatively) correlated with incomeincreasing (income-decreasing) earnings management contrary to hypothesis 1. In addition, contrary to expectations, FAVREC is negatively correlated with consecutive 85

99 periods of income-increasing earnings management and consecutive periods of overvaluation. FAVREC s relationship with earnings management, while contrary to my hypothesis, is not totally unexpected as a result of analysts incentives to optimize reaction to their reports. As expected, the P/V ratio is positively correlated with the overvaluation variable and the consecutive periods of overvaluation variable. Panel B reports the correlation for the price target regression model using levels variables. Contrary to expectations, the favorableness of price targets (FAVPTG) is negatively correlated with the P/V ratio, suggesting that analysts price targets are consistent with stock-picking strategies based on residual-value models. While this is surprising, it is not totally unexpected as prior research has not investigated the relationship between price targets and the P/V ratio. This finding does not impact my predictions within a multiple regression setting, as analysts may still be expected to reduce their price targets further in the presence of income-increasing earnings management as mispricing increases. Also, inconsistent with expectations, the favorableness in price targets is positively (negatively) correlated with income-increasing (income-decreasing) earnings management. This is also not completely surprising as I predicted that analysts price targets responses will be influenced by analysts optimism. This result is unlikely to affect the multiple regression analyses as I focus on how the interaction of mispricing and income-increasing earnings management reduces the relationship between earnings management and analysts responses. Similar to the recommendations levels correlation matrix, FAVPTG is negatively correlated with any restatement, consecutive periods of income-increasing earnings management and consecutive periods of overvaluation. The 86

100 remaining variables in the price targets levels model correlation matrix are the same as variables in the recommendations levels model correlation matrix. In the next section, I discuss my empirical results. 87

101 CHAPTER 6 EMPIRICAL RESULTS This chapter presents empirical results and discusses their implications. First, I assess analysts consensus stock recommendation (price target) response to incomeincreasing GAAP-based earnings management as mispricing increases. I also assess analysts responses to different methods used to manage earnings upwards as well as their responses to consecutive periods of income-increasing earnings management to provide additional evidence on analysts responses to earnings management as mispricing increases. Furthermore, I separately evaluate the results using post-regulation FD samples, and I provide interpretation of the results. Second, I use an alternative proxy for mispricing to test my hypotheses. Specifically, I investigate analysts consensus stock recommendation (price target) response to income-increasing earnings management in firms that analysts perceived to be overvalued. Again, I also assess analysts responses to different methods used to manage earnings upwards as well as their responses to consecutive periods of overvaluation to provide further evidence on analysts responses to earnings management in overvalued firms. I also use separate post-regulation FD samples to test my hypotheses. 88

102 As my sample is panel data, which combine all cross-sectional and time series observations, my sample provides more external validity than pure cross-sectional or time series-pooled data alone. However, it is unlikely that consecutive observations are independent because the sample contains the same firms represented in multiple years. Panel data may result in within-firm (year-specific) serial correlation within the error term (fixed factors) or unobserved factors which are part of the error term but uncorrelated with the explanatory variables (random factors). Therefore, panel data are expected to suffer from fixed effects or random effects. I use the Hausman (1981) test to test for random or fixed effects in the sample data. The Hausman (1981) test rejects the null hypothesis that there are no fixed effects in the sample. Therefore, I use a generalized linear model (GLM) to adjust for fixed effects in the samples 12. The GLM procedure uses the method of least squares to fit general linear models. It is based on ordinary least squares but interprets all effects as fixed and is therefore specifically designed for the fixed-effects model. 6.1 GLM Regression Models for Mispriced Firms Multiple Regression Models for H1 and H2 Table 5 reports the results of the GLM regressions of equation 9 which test hypothesis 1 (the association between the favorableness of analysts responses and income-increasing GAAP-based earnings management) and hypothesis 2 (the association between the favorableness of analysts responses and income-increasing GAAP-based 12 An alternative approach for addressing cross-sectional and time-series dependence standard errors is to use year dummies and cluster standard errors by firm. 89

103 earnings management as mispricing increases). The variance inflation factors (VIFs) in all the regression models are less than 10, which suggest that multicollinearity is not a problem in any of the regression models. The F-test statistics for all the models drawn from equation 9 are significant at a 1% level of significance, which suggests that all the overall models are significant. The adjusted R 2 for the full sample when using the favorableness of analysts stock recommendation responses (hereafter referred to as FAVREC) as the dependent variable is 7.7%. Contrary to prior research, MISPRICING is negatively associated with the FAVREC. This is important as it provides evidence that analysts stock recommendations may be consistent with stock-picking strategies based on residual income models when earnings quality is considered in the model. Consistent with univariate results, I find that income-increasing GAAP-based earnings management (hereafter referred to as PGAAPEM) is positively associated with the FAVREC. This result suggests that analysts optimism bias and analysts heuristics may be driving their responses to income-increasing GAAP-based earnings management. These results are contrary to hypothesis 1. I also find that the interaction of MISPRICING and PGAAPEM is not significantly associated with the FAVREC. Therefore, hypothesis 2 is not supported. Interestingly, income-decreasing GAAP-based earnings management (NGAAPEM) is negatively associated with FAVREC, which reinforces the argument that analysts optimism bias and heuristics may be driving their responses to earnings management. This result could also suggest that analysts treat NGAAPEM as evidence of poor performance. 90

104 I now analyze regression results which test the association between the favorableness of analysts price target responses (hereafter referred to as FAVPTG), MISPRICING and PGAAPEM using the full sample. The R 2 for this estimation model is 10.98%. Consistent with the results from the FAVREC model using the full sample, I also find that MISPRICING and NGAAPEM are negatively associated with FAVPTG. Consistent with univariate results and the FAVREC model, I find that incomeincreasing earnings management is positively associated with the FAVPTG. The first hypothesis is again not supported. However, I find that the interaction of PGAAPEM and MISPRICING is negatively associated with the favorableness of the analysts price targets responses, which provides support for hypothesis 2 using the full sample time period. Next, I analyze the regression results which test the association between FAVREC, MISPRICING and PGAAPEM using the post-regulation FD sample. The R 2 for this estimation model is 5.46%. Regulation FD prohibited the transfer of private information from management to financial analysts and is therefore expected to reduce analysts optimism as analysts have fewer incentives to curry favor from management. For example, Chen and Matsumoto (2006) find that analysts who issued more favorable recommendations received more information from management before, but not after, Regulation FD. Consistent with the results from the FAVREC model using the full sample, I find that NGAAPEM is negatively associated with FAVREC. I also find 91

105 stronger evidence that MISPRICING is negatively associated with FAVREC using the post-regulation FD sample. 13 Consistent with univariate results and contrary to hypothesis 1, I find that PGAAPEM is positively associated with the favorableness of stock recommendations. However, I find that FAVREC is negatively associated with the interaction of PGAAPEM and MISPRICING, which provides support for hypothesis 2 using the post- Regulation FD sample period. This suggests that after Regulation FD, analysts provide a greater monitoring role and are more likely to reduce their stock recommendations as a result of PGAAPEM in mispriced firms. Finally, I analyze the regression results which test the association between FAVPTG, MISPRICING and PGAAPEM using the post-regulation FD sample. The R 2 for this estimation model is 11.56%. Consistent with the results from FAVPTG model using the full sample, I find that NGAAPEM is negatively associated with the favorableness of price targets and MISPRICING is negatively associated with FAVPTG. Consistent with univariate results and contrary to hypothesis 1, I find that PGAAPEM is positively associated with the FAVPTG. Also, consistent with the full sample results, I find stronger evidence 14 that the interaction of income-increasing earnings management and MISPRICING is negatively associated with FAVPTG, which 13 The P/V ratio s coefficient (β3) is greater in the post-regulation FD model than in the full sample model. This larger coefficient provides evidence of a greater analyst response to rising P/V ratios. 14 The coefficient on the interaction variable is greater in the post-regulation FD model than in the full sample model. This larger coefficient provides evidence of greater analysts responses to earnings management as mispricing increases. 92

106 provides support for hypothesis 2 using the post-regulation FD sample time period. This result provides further evidence that Regulation FD had the effect of improving analysts monitoring role. Taken together, these results suggest that PGAAPEM is not negatively associated with analysts stock recommendations or price targets responses, contrary to hypothesis 1. However, my results suggest that, after Regulation FD, the favorableness of the analysts responses is negatively associated with the interaction of PGAAPEM and MISPRICING using the levels sample. Thus, hypothesis 2 is supported after Regulation FD in the levels sample, which provides further evidence that Regulation FD has reduced analysts optimism and improved the accuracy of analysts reports. I also find that the association between MISPRICING and stock recommendations (price targets) is generally negative and significant in the levels sample data when I include earnings management in the model. This negative relationship is consistent with stock-picking strategies based on residual income models. This is a very interesting finding. Bradshaw (2004) finds that the association between the V/P ratio and stock recommendations was either insignificant or contrary to stock-picking strategies based on residual income models. More recently, Barniv et al. (2009) and Chen and Chen (2009) find the inconsistency between stock recommendations and stock-picking strategies based on residual income models remain even after recent regulatory changes. To the best of my knowledge, this is the first study to provide evidence of consistency between analysts stock recommendations and a stock-picking strategy based on fundamental value. This finding suggests that analysts may incorporate fundamental value 93

107 considerations into their stock recommendations and price targets. I now analyze tests of hypothesis Multiple Regression Models for H3 Table 6 reports the results of the GLM regressions which tests hypothesis 3 (the association between the favorableness of analysts responses and earnings management in mispriced firms as the cost of the method used to manage earnings increases). These models are based on equation 10. The F-test statistics for all the models drawn from equation 10 are significant at a 1% level of significance which suggests all the overall models are significant. As Audit Analytics only began reporting restatement data in 2000, restatement data are not included in the full sample regressions. β 6, β 7, and β 10 are, therefore, all set to zero in the full sample regressions. The adjusted R 2 for the full sample when using FAVREC as the dependent variable is 8.43%. Income-decreasing earnings management is negatively associated with FAVREC, which is consistent with the results reported in Table 5. Also, contrary to prior research and consistent with results in Table 5, MISPRICING is negatively associated with FAVREC. Consistent with univariate results, I find that the different income-increasing earnings management mechanisms (PABNACCRUALS and PABNRTM) are positively associated with FAVREC. These results are contrary to hypothesis 1. For my tests of hypothesis 3, I find that the interaction of MISPRICING and income-increasing accruals management (PABNACCRUALS) is negatively and significantly associated with the FAVREC. However, I find that the interaction of MISPRICING and income-increasing real transaction management (PABNRTM) is not significantly associated with the 94

108 FAVREC. As PABNRTM is more costly than PABNACCRUALS, I conclude that the most costly methods used to manage earnings do not increase the negative association between the FAVREC and income-increasing earnings management more than other methods. Thus, hypothesis 3 is not supported for stock recommendations when using the full sample. I now analyze the regression results which test the association between FAVPTG, MISPRICING and different income-increasing earnings management mechanisms using equation 10 for the full sample. The R 2 for this estimation model is 12.86%. Consistent with the results from the FAVREC model using the full sample, I find that incomedecreasing earnings management is negatively associated with FAVPTG. Also consistent with the results from the FAVREC model using the full sample, I find that MISPRICING is negatively associated with FAVPTG. Consistent with univariate results and contrary to hypothesis 1, I find that the different income-increasing earnings management mechanisms (PABNACCRUALS and PABNRTM) are positively associated with FAVPTG. For my tests of hypothesis 3, I find that the interaction of PABNACCRUALS (PABNRTM) and MISPRICING is negatively associated with FAVPTG. I find that PABNACCRUALS s regression slope is given by the formula ( *MISPRICING) * PABNACCRUALS, whereas the formula for PABNRTM s regression slope is ( *MISPRICING) * PABNRTM. Using the mean P/V ratio of (see descriptive statistics), a one-unit increase in the PABNACCRUALS interaction reduces FAVPTG by 19% on average whereas a one-unit increase in the PABNRTM interaction only reduces FAVPTG by 4.6% on average. Thus, hypothesis 3 is not supported when using the full price target sample. These results 95

109 suggest that, in the overall sample, analysts may not consider the costs of the methods used to manage earnings when making price targets. Next, I analyze the regression results which test the association between FAVREC, MISPRICING, and income-increasing earnings management mechanisms using the post Regulation FD sample. As previously indicated, Regulation FD is expected to reduce analysts optimism as analysts have fewer incentives to curry favor from management. The R 2 for this estimation model is 6.31%. Consistent with the results from the FAVREC model using the full sample, I find that income-decreasing earnings management is negatively associated with the favorableness of stock recommendation or insignificant after Regulation FD. Also consistent with the results from the FAVREC model using the full sample, I find additional evidence that MISPRICING is negatively associated with FAVREC using the post-regulation FD sample. PABNACCRUALS, PABNRTM and PRESTATE are each positively associated with FAVREC consistent with the full sample findings. For my tests of hypothesis 3, I find that the interaction of PABNACCRUALS and MISPRICING is not significantly associated with the favorableness of the analysts stock recommendation responses. However, I find that both the interaction of PABNRTM and PRESTATE with MISPRICING are negatively associated with FAVREC. As the interaction of PABNRTM (PABNACCRUALS) and MISPRICING is significant (insignificant), hypothesis 3 is partially supported when using the post-regulation FD stock recommendation sample because the more costly GAAP-based income-increasing earnings management creates a more negative analysts responses. PABNRTM s regression slope is given by the formula ( *MISPRICING) * PABNRTM, 96

110 whereas the formula for PRESTATE s regression slope is ( *MISPRICING) * PRESTATE. Using the mean P/V ratio of in the post-regulation FD sample (see descriptive statistics), a one-unit increase in the PABNRTM interaction reduces FAVREC by 1.9%, whereas an income-increasing restatement interaction only reduces FAVREC by 1.5%. Therefore, hypothesis 3 is not fully supported in the post-regulation FD sample. These results suggest that Regulation FD may have had the effect of making analysts more responsive to more costly earnings management methods. Analysts may have become more responsive to real transactions management than accruals management because of the documented decrease (increase) in accruals management (real transaction management). Analysts may not be as responsive to adverse restatements because restatements may not be reported in the same accounting period that the income-increasing financial irregularity occurs; therefore, analysts may not be aware of many financial irregularities. Finally, I analyze the regression results which test the association between FAVPTG, MISPRICING and different income-increasing earnings management mechanisms using the post-regulation FD sample. The R 2 for this estimation model is 13.81%. Consistent with the results from the FAVREC model using the full sample, I also find that income-decreasing earnings management is negatively associated with FAVPTG or not significant and MISPRICING is negatively associated with the favorableness of price targets. I find that PABNACCRUALS, PABNRTM and PRESTATE are all positively associated with FAVPTG. For my tests of hypothesis 3, I find that the interaction of 97

111 PABNACCRUALS and PABNRTM with MISPRICING is negatively associated with FAVPTG. Similarly, I find that the interaction of PRESTATE and MISPRICING is negatively associated with FAVPTG. Using the mean P/V ratio of 1.345, I find evidence that a one-unit increase in the PABNACCRUALS interaction (PABNRTM interaction) reduces FAVPTG by 8% (11.2%). Similarly, the adverse restatements interaction reduces FAVPTG by 4%. As the PABNRTM interaction reduces FAVPTG more than the PABNACCRUALS interaction on average, I find that hypothesis 3 is partially supported when using the full price target sample. However, as both the PABNRTM and PABNACCRUALS interactions reduce FAVPTG more than the PRESTATE interaction, hypothesis 3 is not fully supported. These results suggest that in the post-regulation FD sample, analysts may consider the costs of the GAAP-based methods used to manage earnings when making price targets as hypothesized. However, analysts may not be as responsive to adverse restatements because financial irregularities may be less observable than GAAP-based earning management. In summary, the negative association between FAVREC (FAVPTG) and incomeincreasing earnings management is greater for real transaction management than for accruals management post-regulation FD. As prior research has found real transaction management to be more costly than accruals management, hypothesis 3 is partially supported. However, the negative association between the favorableness of the analysts stock recommendation (price target) responses and income-increasing earnings management is not greater for financial irregularities than for GAAP-based earnings management. This may be because financial irregularities may not be observable in the period of the indiscretion, which may limit analysts ability to respond accordingly. 98

112 6.1.3 Multiple Regression Models for H4 Table 7 reports the results of the GLM regressions which test hypothesis 4 (the association between the favorableness of analysts responses and earnings management in mispriced firms as the length of the consecutive periods of income-increasing GAAPbased earnings management increases). These models are based on equation 11. The F-test statistics for all the models drawn from equation 11 are significant at a 1% level of significance, which suggests all the overall models are significant. The R 2 for the full sample when using the FAVREC as the dependent variable is 8.08%. Consistent with univariate results, a string of income-decreasing GAAP-based earnings management is negatively associated with FAVREC, which suggests analysts may perceive such a string as evidence of poor performance. In addition, consistent with Table 5 results, I find that NGAAPEM is negatively associated with FAVREC and MISPRICING is negatively associated with FAVREC. Also consistent with results tabulated in Table 5 but contrary to hypothesis 1, I find that PGAAPEM is positively associated with FAVREC. However, consistent with univariate results, I find that length of the consecutive periods of income-increasing GAAP-based earnings management (PGAAPEMN) is negatively associated with FAVREC. This result is consistent with expectations and suggests that PGAAPEMN is considered negatively by analysts. This result could be due to analysts concern that PGAAPEMN may be associated with value destruction. Contrary to hypothesis 2, the interaction of PGAAPEM and MISPRICING is positively associated with FAVREC at a one tail test. However, I find that the interaction of MISPRICING, PGAAPEM, and PGAAPEMN is negatively and significantly associated with FAVREC. Therefore, 99

113 hypothesis 4 is supported for stock recommendations using the full sample. This finding suggests that it is the length of the string of income-increasing earnings management that drives the negative association between income-increasing earnings management and the favorableness of analysts stock recommendations in mispriced firms. I now analyze the regression results which test the association between FAVPTG, PGAAPEM, MISPRICING, and the PGAAPEMN using the full sample. The R 2 for this estimation model is 11.9%. Consistent with the results from the FAVREC model using the full sample, I also find that a string of income-decreasing earnings management, NGAAPEM, and MISPRICING are all negatively associated with the FAVPTG. Contrary to hypothesis 1 but consistent with results from equation 9 (see Table 5), I find that PGAAPEM is positively associated with FAVPTG. Consistent with univariate results, I find that PGAAPEMN is negatively associated with FAVPTG. When PGAAPEMN is added to the model, the interaction between PGAAPEM and MISPRICING is not statistically significant, contrary to hypothesis 2. However, I find that the interaction of PGAAPEM, PGAAPEMN and MISPRICING is negatively associated with FAVPTG, which provides further support for hypothesis 4 using the full sample time period. Next, I analyze the regression results which test the association between FAVREC, PGAAPEM, MISPRICING, and PGAAPEMN using the post-regulation FD sample. The R 2 for this estimation model is 5.81%. These results are statistically similar to the corresponding results using the full sample. Specifically, a string of incomeincreasing (income-decreasing) GAAP-based earnings management is negatively associated with FAVREC and the association between FAVREC and income-increasing 100

114 earnings management is more negative as the length of the consecutive periods of income-increasing earnings management increases. Therefore, this result is consistent with hypothesis 4. Finally, I analyze the regression results which test the association between FAVPTG, PGAAPEM, MISPRICING, and a string of PGAAPEM using the post- Regulation FD sample. The R 2 for this estimation model is 12.53%. These results are statistically similar to the corresponding results using the full sample. Specifically, a string of income-increasing (income-decreasing) GAAP-based earnings management is negatively associated with FAVPTG, the association between the FAVPTG and PGAAPEM is more negative as the length of the consecutive periods of incomeincreasing earnings management increases and MISPRICING is negatively associated with FAVPTG using the post-regulation FD sample. Once again, hypothesis 4 is supported in this model. Interestingly, the interaction of MISPRICING and PGAAPEM is negative at a one tail test in this model which provides evidence that the association between income-increasing earnings management as mispricing increases and analysts responses is more negative post-regulation FD. Taken together, these results provide evidence that the negative association between the favorableness of analysts responses and PGAAPEM in mispriced firms is more negative as the length of the consecutive periods of income-increasing earnings management increases. Thus, hypothesis 4 is supported. In fact, it appears that it is the length of the consecutive periods of earnings management that is driving the negative relationship. Furthermore, these results provide further evidence that MISPRICING is negatively associated with the favorableness of analysts responses after including 101

115 PGAAPEMN in the model, and therefore consistent with recommendations based on residual income models. I now use my alternate proxy for additional evidence on hypotheses 1 to 3 and I test hypothesis 5 in order to determine if analysts respond differently to firms they believe to be overvalued. 6.2 Alternate Tests of Hypotheses 1, 2 and 3. In this section I use firms analysts believe to be overvalued (OVER) as my proxy for mispricing. Overvalued firms are the mispriced firms that are most likely to manage earnings upwards and most likely to destroy firm value (Kothari et al., 2006; Jensen, 2005; Efendi et al., 2007). Therefore, analysts responses to income-increasing earnings management are expected to be greater in overvalued firms than other firms. As firm overvaluation is my alternative proxy for mispricing, I simple replace MISPRICING with OVER in these models Multiple Regression Models using Overvalued Firms to Test H1 and H2 Table 8 reports the results of the GLM regressions which tests hypothesis 1 (the association between the favorableness of analysts responses and earnings management) and hypothesis 2 (the association between.the favorableness of analysts responses and earnings management in mispriced firms). Although these models are based on equation 9, I adapt equation 9 by replacing MISPRICING with OVER. The F-test statistics for all the models drawn from equation 9 are significant at a 1% level of significance, which suggests all the overall models are significant. The adjusted R 2 for the full sample when using the favorableness of analysts stock recommendation as the dependent variable is 7.83%. I find that overvalued firms are 102

116 negatively associated with the favorableness of analysts stock recommendations which suggests that analysts may use residual income models in making recommendations. Consistent with the main model results in Table 5, I find that income-increasing (income-decreasing) earnings management is positively (negatively) associated with FAVREC. These results are contrary to hypothesis 1 as previously discussed. Also, similar to the main model results in Table 5, I find that the interaction of the overvaluation dummy variable and PGAAPEM is not significantly associated with FAVREC. Therefore, hypothesis 2 is not supported for stock recommendations when using the full sample. This result may suggest that analysts do not perceive PGAAPEM in overvalued firms as any more value-destroying than overvalued firms in general. I now analyze the regression results which test the association between FAVPTG, firm overvaluation and PGAAPEM using the full sample. The R 2 for this estimation model is 10.94%. Similar to the findings for stock recommendations, I find that overvalued firms are negatively associated with the FAVPTG, which suggests that analysts may deem that these firms are relatively overpriced. This finding reinforces prior evidence that price targets may be consistent with strategies based on residual income models when income-increasing earnings management and an overvalued firms indicator variable are included in the model. Consistent with the main model results and univariate results but contrary to hypothesis 1, I find that income-increasing (income-decreasing) earnings management is positively (negatively) associated with FAVPTG. However, similar to the main model results, I find that the interaction of PGAAPEM and the overvaluation indicator variable is negatively associated with FAVPTG, which provides additional support for hypothesis 103

117 2 using the full sample time period. Next, I analyze the regression results which test the association between FAVREC, firm overvaluation, and PGAAPEM using the post- Regulation FD sample. The R 2 for this estimation model is 5.66%. Consistent with the full sample results, I find that overvalued firms are negatively associated with the FAVREC. Consistent with results pertaining to the full sample and contrary to hypothesis 1, I find that income-increasing (income-decreasing) earnings management is positively (negatively) associated with FAVREC. However, I find that the interaction of PGAAPEM and the firm overvaluation indicator variable is negatively associated with the favorableness of analysts stock recommendations responses, which provides support for hypothesis 2 using the post-regulation FD sample period. This suggests that, after Regulation FD, analysts provide a greater monitoring role and are more likely to reduce their stock recommendations as result of PGAAPEM in overvalued firms. Finally, I assess the regression results which test the association between FAVPTG, firm overvaluation, and PGAAPEM using the post-regulation FD sample. The R 2 for this estimation model is 11.40%. These results are similar to the full sample results, which provides additional support for hypothesis 2 because FAVPTG is negatively associated with PGAAPEM in overvalued firms. Taken together, these results suggest that PGAAPEM is not negatively associated with analysts stock recommendation responses contrary to hypothesis 1. However, my results suggest that after Regulation FD, the favorableness of the analysts stock recommendations (price targets) responses is negatively associated with PGAAPEM in overvalued firms. Thus, hypothesis 2 is supported after Regulation FD, which provides 104

118 further evidence that Regulation FD has reduced analysts optimism and improved the accuracy of analysts reports. I now analyze additional tests of hypothesis 3 by using overvalued firms as my proxy for mispricing Multiple Regression Models using Overvalued Firms to Test H3 Table 9 reports the results of the GLM regressions which tests hypothesis 3 (the association between the favorableness of analysts responses and earnings management in mispriced firms as the cost of the method used to manage earnings increases). I adapt equation 9 by replacing MISPRICING with OVER in these models. Restatements are excluded from the full samples as a result of insufficient restatement observations before Regulation FD. The F-test statistics for all the models drawn from equation 9 are significant at a 1% level of significance. The R 2 for the full sample when using FAVREC as the dependent variable is 8.56%. Consistent with results in Table 8, the overvaluation indicator variable is negatively associated with the FAVREC. This result suggests that overvalued firms may be driving analysts stock recommendation positive relationship with fundamental value after controlling for earnings management. Consistent with the correlation matrices univariate results, I find that the different income-increasing earnings management mechanisms (PABNACCRUALS and PABNRTM) are positively associated with FAVREC contrary to hypothesis 1. Similarly, income-decreasing earnings management mechanisms are negatively associated with FAVREC, consistent with univariate results. For my additional tests of hypothesis 3, I find that the interaction of firm overvaluation and PABNACCRUALS is negatively and significantly associated with FAVREC. However, I find that the interaction of firm overvaluation and PABNRTM is not significantly associated with FAVREC. As 105

119 PABNRTM is more costly than PABNACCRUALS, I conclude that the most costly methods used to manage earnings do not increase the negative association between FAVREC and income-increasing earnings management more than alternative methods. Thus, hypothesis 3 is not supported for stock recommendations when using the full sample. I now analyze the regression results which test the association between FAVPTG, firm overvaluation, and different income-increasing earnings management mechanisms using the full sample. The R 2 for this estimation model is 12.66%. These results are similar to the corresponding main model results in Table 6. Specifically, I find that the different income-increasing (income-decreasing) earnings management mechanisms are positively (negatively) associated with FAVPTG. For my tests of hypothesis 3, I find that the interactions of both income-increasing GAAP-based earnings management variables and the overvaluation indicator variable are negatively associated with FAVPTG. Additionally, PABNACCRUALS s regression slope is given by the formula ( *OVER) * PABNACCRUALS, whereas the formula for PABNRTM s regression slope is ( *OVER) * PABNRTM. Therefore, in overvalued firms, a one-unit increase in PABNACCRUALS increases FAVPTG by 81.9% whereas a one-unit increase in PABNRTM decreases FAVPTG by 5.4%. However, hypothesis 3 tests the effects of the interaction terms. For overvalued firms, a one-unit increase in the PABNACCRUALS interaction reduces FAVPTG by 50.2% whereas a one-unit increase in PABNRTM decreases FAVPTG by 10.2%. Thus, hypothesis 3 is not supported when using the full price target sample because PABNACCRUALS decreases FAVPTG more in overvalued firms than PABNRTM. These results suggest that, in the full sample, 106

120 analysts may not consider the costs of the methods used to manage earnings in overvalued firms when making price targets. Consistent with the corresponding results in Table 8, I find that firm overvaluation is negatively associated with FAVPTG, which suggests analysts reduce price targets further as a result of perceived overvaluation. Next, I analyze the regression results which test the association between FAVREC, firm overvaluation, and income-increasing earnings management mechanisms using the post-regulation FD sample. The R 2 for this estimation model is 6.51%. These results are similar to the results for the full sample. Specifically, I find that the different income-increasing (income-decreasing) earnings management mechanisms are positively (negatively or not significantly) associated with FAVREC. Furthermore, I find that the interaction of income-increasing accruals (real transaction) management and the overvaluation indicator variable is negatively (not significantly) associated with FAVREC. As PABNRTM is more costly than income-increasing accruals management, I conclude that the most costly methods used to manage earnings in overvalued firms do not increase the negative association between the favorableness of the analysts stock recommendation responses and income-increasing earnings management more than alternative methods. Thus, using overvalued firms, hypothesis 3 is not supported for stock recommendations in the post-regulation FD sample. In addition, I find that the interaction of PRESTATE and the overvaluation indicator variable is negatively associated with FAVREC using a one-tail test. However, in overvalued firms, the PRESTATE interaction decreases FAVREC by 1.5% whereas the PABNACCRUALS interaction decreases FAVREC by 11.2%. Therefore, income-increasing discretionary accruals have a greater impact on the negative association between FAVREC and 107

121 earnings management than adverse restatements, contrary to hypothesis 3. Also consistent with the full sample results, I find further evidence that the overvaluation indicator variable is negatively associated with FAVREC using the post-regulation FD sample. Finally, I analyze the regression results which test the association between FAVPTG, firm overvaluation, and different income-increasing earnings management mechanisms using the post-regulation FD sample. The R 2 for this estimation model is 13.64%. These results are consistent with the full sample results which are contrary to hypotheses 1 and 3. In this test, I find that the interaction of PRESTATE and the overvaluation indicator variable is negatively associated with FAVPTG; however, this interaction has a smaller impact on the dependent variable than the PABNACCRUALS and PABNRTM interaction variables. All other results are consistent with the full sample results. In summary, the negative association between the favorableness of the analysts stock recommendation (price target) responses and income-increasing earnings management is greater for accruals management than for real transaction management. As prior research has found real transaction management to be more costly than accruals management, hypothesis 3 is not supported using the alternate mispricing proxy. After Regulation FD, the association between the favorableness of the analysts stock recommendation (price target) responses and PRESTATE is at least weakly significant ; however, the negative association between the favorableness of the analysts stock recommendation (price target) responses and income-increasing earnings management is not greater for financial irregularities than for GAAP-based accruals management. This 108

122 may be because financial irregularities are not observable in the period of the indiscretion, which may limit analysts ability to respond accordingly. 6.3 Overvaluation Duration Multiple Regression Models for H5 Table 10 reports the results of the GLM regressions which tests hypothesis 5 (the association between the favorableness of analysts responses and earnings management in overvalued firms as the length of the consecutive periods of overvaluation increases) using levels data. These models are based on equation 12. The F-test statistics for all the models drawn from equation 12 are significant at a 1% level of significance which suggests all the overall models are significant. The R 2 for the full sample when using the FAVREC as the dependent variable is 7.87%. Consistent with univariate results, I find that the length of the consecutive periods of overvaluation (OVERN) is negatively associated with the FAVREC. This result suggests that after including the GAAP-based earnings management variable in the model and controlling for OVER, a string of consecutive years of overvaluation is considered negatively by analysts. Also, consistent with prior research, mispricing as represented by OVER is not associated with the FAVREC in this regression analysis. Contrary to expectations but consistent with prior results (see Table 8), I find that income-increasing (income-decreasing) GAAP-based earnings management is positively (negatively) and significantly associated with FAVREC. Consistent with hypothesis 2, FAVREC is negatively associated with PGAAPEM in overvalued firms. However, contrary to hypothesis 5, I find that the interaction of PGAAPEM and OVERN is not 109

123 significantly associated with FAVREC using the full sample. This finding suggests that, before Regulation FD, analysts did not respond more negatively to income- increasing earnings management as the length of overvaluation increases. I now analyze the regression results which test the association between FAVPTG, OVER, PGAAPEM, and OVERN using the full sample. The R 2 for this estimation model is 10.94%. Consistent with prior results in Table 8, OVER is negatively associated with the FAVPTG, which suggests analysts may consider fundamental value when setting price targets after controlling for the length of overvaluation and including earnings management in the model. Contrary to the stock recommendation sample results, I find that OVERN is not associated with FAVPTG. However, consistent with the stock recommendation sample results, I find that income-increasing (income-decreasing) GAAP-based earnings management is positively (negatively) and significantly associated with FAVPTG. This result may suggest that, after modeling GAAP-based earnings management and controlling for OVER, a string of consecutive years of overvaluation is not considered by analysts when setting price targets. Consistent with hypothesis 2, FAVREC is negatively associated with PGAAPEM in overvalued firms. However, contrary to hypothesis 5, I find that the interaction of PGAAPEM and OVERN is not associated with FAVPTG using the full sample. Next, I analyze the regression results which test the association between FAVREC, OVER, PGAAPEM, and OVERN using the post-regulation FD sample. As previously discussed, Regulation FD is expected to reduce analysts optimism as analysts have fewer incentives to curry favor from management. The R 2 for this estimation model 110

124 is 5.73%. Contrary to the full sample results and consistent with the results from FAVPTG model, I find evidence that OVER is negatively associated with FAVREC using the post-regulation FD sample. Similar to the results in the full sample, income-increasing (income-decreasing) GAAP-based earnings management is positively (negatively associated) with FAVREC and OVERN is negatively associated with the favorableness of the analysts stock recommendations responses. In addition, I find that the interaction of PGAAPEM and OVER (OVERN) is negatively (not associated) with FAVREC using the post-regulation FD sample. This finding provides further evidence that, even after Regulation FD, analysts stock recommendations is not more negatively associated to income-increasing earnings management as the length of overvaluation increased. Finally, I analyze the regression results which test the association between FAVPTG, OVER, PGAAPEM, and OVERN using the post-regulation FD sample. The R 2 for this estimation model is 11.41%. These results are statistically similar to the corresponding results using the full sample. Specifically, income-increasing (incomedecreasing) GAAP-based earnings management is positively (negatively) associated with FAVPTG and OVER (OVERN) is negatively (not significantly) associated with FAVPTG using the post-regulation FD sample. Furthermore, the negative association between the favorableness of the analysts price targets responses and income-increasing earnings management is greater for overvalued firms (which supports hypothesis 2) but not greater as the length of firm overvaluation increases (contrary to hypothesis 5). Taken together, these results provide evidence that the association between the favorableness of the analysts stock recommendations (price targets) responses and 111

125 income-increasing earnings management is not more negative as the length of firm overvaluation increases even after Regulation FD. Thus, hypothesis 5 is not supported. However, post-regulation FD, firm overvaluation is negatively associated to analysts stock recommendations (price targets) responses which is consistent with analysts use of fundamental value in making recommendations (price targets). 6.4 Summary Overall, I find support 15 for hypothesis 2 using the stock recommendations sample post-regulation FD and using both price targets samples. I find partial support for hypothesis 3 in the main model using both stock recommendations and price targets samples only after Regulation FD and only in relation to the costs of GAAP-based earnings management mechanisms. I also find evidence supporting hypothesis 4 using both stock recommendations and price targets samples and in both sample periods. However, I find that income-increasing earnings management is positively associated with the favorableness in analysts responses, contrary to hypothesis 1. These results provide compelling evidence that the relationship between analysts responses and income increasing earnings management is more negative as firm mispricing increases. Furthermore, this negative association is greater as costs of the GAAP-based earnings mechanisms increases and the length of consecutive periods of income-increasing GAAP-based earnings management increases. 15 In many cases, although I find statistical significance in support of my hypothesis, the economic significance of the finding is small. 112

126 Using firm overvaluation as my proxy for mispricing, I find further support for hypothesis 2 post-regulation FD in the stock recommendations sample and for both price targets samples. However, I do not find support for hypothesis 5 suggesting that the negative association between analysts responses and income-increasing earnings management in overvalued firms is not greater as the length of overvaluation increases. Similarly, I do not find any support for hypotheses 1 and 3 using overvalued firms. Therefore, I conclude that the relationship between analysts responses and incomeincreasing earnings management is more negative in overvalued firms. However, this negative association is neither dependent on the length of consecutive periods of overvaluation or the cost of the earnings management mechanism used to manage earnings. 113

127 CHAPTER 7 SENSITIVITY ANALYSES I make a number of assumptions in the development of the analysts responses models and the computation of fundamental value. In this chapter, I report the results of sensitivity analyses regarding these assumptions. These results reinforce many of my findings on analysts responses to income-increasing earnings management after controlling for the firm mispricing and overvaluation. Chapter 6 reports results using data from the full sample period and separately from the post-regulation FD sample. I find more results in support of my hypotheses and more significant results using the post-regulation FD sample. I therefore limit my sensitivity analyses to post-regulation FD samples. 7.1 P/V Ratio Computation In Chapter 4, I use a residual income model to calculate the fundamental value of a firm (e.g., Ohlson, 1995; Frankel and Lee, 1998). The firm value is based on the firm s current book value, the expected value of residual income for the next 5 years and a terminal value assumption. In my main analyses, I use the terminal value assumption that abnormal returns persists in perpetuity (Bradshaw, 2004; Badertscher, 2011). However, in his seminal paper, Bradshaw (2004) also computed fundamental value using the terminal value assumption that abnormal returns revert to zero, which is hereafter also referred to as the fade rate assumption. The fade rate assumption is consistent with the 114

128 argument that competition drives abnormal returns to zero in the long run. An autoregressive parameter (ω) of 0.68 reverts the residual income to zero (where the return on equity is equal to the cost of capital). Thus, the residual income valuation model based on the fade rate assumes that the residual income in the terminal year reverts to zero over ten years (Bradshaw, 2004; Barniv et al., 2009). Under the fade rate assumption, the firm s fundamental value is lower than under the perpetuity terminal value assumption, which means the firm will appear to be more overvalued. The residual income model using the fade rate assumption is specified in equation 13 as follows: V t = BVPS t + E t[ri t+n ] + 5 n=1 (1+r) n ωe t [RI t+5 ] (1+r ω)(1+r) 5 (13) where: ω is the autoregressive parameter (fade-rate) based on the assumption that abnormal returns disappear within ten years. All other variables are previously defined (see equation 6 in 4.2.1). Tables 11 to 13 report the results of the sensitivity analyses based on the fade rate assumption. The F-test statistics for all the models drawn from Tables 11 to 13 are significant at a 1% level of significance, which suggests all the overall models are significant. I split the tables into main models and overvaluation models with separate analyses for stock recommendation and price target samples. Table 11 reports the results of the GLM regressions which test hypotheses 1 and 2 using the fade rate assumption. The main model uses equation 9 but MISPRICING is redefined with the terminal value calculation based on the fade rate assumption. The overvaluation model uses equation 9 but MISPRICING is replaced with OVER to test the effect of overvalued firms on 115

129 analysts response. In this model, the computation of OVER is based on the fade rate assumption. Under the fade rate assumption, Bradshaw (2004) finds analysts stock recommendations to be negatively associated with stock-picking strategies based on a residual income model. Therefore, I expect that MISPRICING based on the fade rate assumption will be positively associated with the favorableness of stock recommendations (price targets). Consistent with expectations, MISPRICING is positively associated with the favorableness of analysts stock recommendations responses. Consistent with the main results (see Table 5, post-regulation FD) but contrary to hypothesis 1, income-increasing (decreasing) GAAP-based earnings management is positively (negatively) associated with the favorableness of analysts stock recommendations responses. However, contrary to the main results and hypothesis 2, under the fade rate assumption the interaction of income-increasing earnings management and MISPRICING is positively associated with FAVREC. Similarly, using the overvaluation model, OVER is positively associated with the favorableness of analysts stock recommendations responses and the interaction of income-increasing earnings management and firm overvaluation is not associated with FAVREC. Overall, contrary to the main results, hypothesis 2 is not supported in the stock recommendations sample under the fade rate assumption. However, the regression results under the fade rate assumption are similar to the main results using the price target sample for both the main and overvaluation model. Specifically, OVER is negatively associated with FAVPTG and the interaction of income-increasing earnings management and firm overvaluation is negatively with FAVPTG in a one tail test. Therefore, hypothesis 2 is supported for price targets using the fade rate assumption. 116

130 Table 12 reports the results of the GLM regressions which test hypothesis 3. Consistent with prior research and Table 11, MISPRICING is positively associated with FAVREC. Contrary to the main results tabulated in Tables 6 and 9, the interaction of MISPRICING (firm overvaluation) and income-increasing accruals management is negatively associated with FAVREC in both the main and overvaluation models. Also, the interaction of MISPRICING (firm overvaluation) and income-increasing real transaction management is positively associated with FAVREC, contrary to hypothesis 3. On the other hand, the regression results under the fade rate assumption continue to be similar to the main results using the price target sample for both the main and overvaluation model. Therefore, these findings provide partial support for hypothesis 3 using the P/V ratio as the proxy for mispricing in the price targets sample under the fade rate assumption. However, hypothesis 3 is not supported using overvalued firms in the price targets sample. Finally, Table 13 reports the results of the GLM regressions which test hypotheses 4 and 5, using the fade rate assumption in the terminal value computation. As in Tables 11 and 12, the sensitivity results based on the stock recommendations sample are generally contrary to the main results and inconsistent with hypotheses 4 and 5. However, the sensitivity results based on the price targets sample are generally consistent with the main results and support hypothesis 4 but not hypothesis 5. In summary, the favorableness of stock recommendations results appear to be sensitive to the terminal value assumptions used to calculate fundamental value. However, the favorableness of price targets regression results appear to be robust. 117

131 7.2 Firm Overvaluation Continuous Variable to Represent Firm Overvaluation In Chapter 6, I use a dummy variable to represent the change in the favorableness of analysts responses for an overvalued firm. This model assumes that analysts have a monolithic response to all overvalued firms. However, analysts may have a differential response to overvalued firms depending on the extent of the overvaluation. Specifically, analysts may amend their stock recommendations (price targets) more for the most overvalued firms. To test whether the extent of the overvaluation matters, I use a continuous variable in my analysts response models to represent firm overvaluation. Table 14 reports the results of the GLM regressions which provide additional evidence on hypothesis 2 using the continuous overvaluation variable. The F-test statistics for all the models drawn from Table 14 are significant at a 1% level of significance. Consistent with Table 8, I find that firm overvaluation using the continuous variable is negatively associated with the favorableness of analysts responses. Consistent with hypothesis 2, this sensitivity analysis reinforces the findings that the association between income-increasing earnings management and the favorableness of financial analysts responses is more negative for overvalued firms. Table 15 reports the results of the GLM regressions which tests hypothesis 3 using the continuous overvaluation variable. The F-test statistics for all the models drawn from Table 15 are significant at a 1% level of significance which suggests all the overall models are significant. The firm overvaluation variable (OVERC) is negatively association with FAVREC, consistent with main results in tables 6 and 9. The interactions of income-increasing accruals, real transactions management and 118

132 restatements with overvaluation are negatively associated with the favorableness of stock recommendation responses consistent with hypothesis 2. A one-unit increase in the PRESTATE interaction reduces FAVREC by 1.0% whereas a one-unit increase in PABNRTM decreases FAVREC by 0.7% for overvalued firms. This result provides partial support for hypothesis 3 using the continuous overvaluation variable as incomeincreasing restatements are more costly than GAAP-based earnings management. However, a one-unit increase in the PABNACCRUALS interaction reduces FAVREC by 0.9% whereas a one-unit increase in PABNRTM decreases FAVREC by 0.7% for overvalued firms. Thus, hypothesis 3 is not fully supported because PABNACCRUALS decreases FAVPTG more in overvalued firms than PABNRTM. Similarly, firm overvaluation is negatively associated with the favorableness of price targets when using the continuous overvaluation variable. Similarly to the results using the stock recommendation sample, the interactions of income-increasing earnings management and firm overvaluation are negative but the most costly earnings management mechanisms do not have the greatest effect on FAVPTG on average. Thus, consistent with the main overvaluation model results (see Table 9), this sensitivity analysis reinforces the findings that analysts do not reduce their price targets more for overvalued firms reporting the most costly earnings management, inconsistent with hypothesis 3. Table 16 reports the results of the GLM regressions which tests hypothesis 5 using the continuous overvaluation variable. The F-test statistics for all the models drawn from Table 16 are significant at a 1% level of significance which suggests all the overall models are significant. In these models the firm overvaluation variable (OVERC) is not 119

133 significantly association with analysts responses, contrary with the main results in table 10. However, consistent with the main results, hypothesis 5 is not supported providing further evidence that the association between analysts responses and income-increasing earnings management is not more negative as the length of overvaluation increases Firm Overvaluation Indicator Variable I define firm overvaluation as the top quintile of P/V rankings in this study. However, firms in the top 30% percent of P/V rankings have P/V rankings greater than 1 and may be perceived as overvalued by some analysts. Alternatively, other analysts may believe that only firms with extremely high P/V rankings are overvalued. To test the sensitivity of the definition of firm overvaluation to my results, I first redefine firm overvaluation as the top decile of P/V rankings (OVER10). Next, I redefine firm overvaluation as the top 30% of P/V rankings (OVER30) to test my hypotheses. Using OVER10 as my overvaluation proxy in the stock recommendations sample, I find that only hypothesis 3 is partially supported. Both hypotheses 2 and 5 are not supported. However, using OVER10 as my overvaluation proxy in the price targets sample, I find that none of the hypotheses using the overvaluation model are supported. These results suggest that my main results are not robust to using a more extreme definition of firm overvaluation although hypothesis 3 results may be robust using the stock recommendations sample. These results are not tabulated. Using OVER30 as my overvaluation proxy in the stock recommendations sample, I find that hypothesis 2 is moderately supported. However, consistent with the main results, hypotheses 3 and 5 are not supported. I find similar but stronger results using the price targets sample. Specifically, hypothesis 2 is supported but hypotheses 3 and 5 are 120

134 not supported. These results suggest that my main results are robust to using a less extreme definition of firm overvaluation in both the stock recommendations and price targets samples. These results are not tabulated. 7.3 Long-Term Growth Model Analysts reports, which consist of many forecasts including earnings forecasts, price targets, cash flow forecasts, and long-term growth forecasts, culminate in the stock recommendation. However, Bradshaw (2004) finds that it is variables such as long-term growth forecasts, and not fundamental value that explain analysts stock recommendations. I, therefore, test whether the combination of fundamental value and income-increasing earnings management explain analysts long-term growth forecasts. Specifically, I use the analysts consensus long-term growth forecasts as my proxy for analysts responses. Tables report the results of the GLM regressions of analysts long-term growth responses to earnings management after controlling for mispricing. The F-test statistics for all the models drawn from these tables are significant at a 1% level of significance which suggests all the overall models are significant. Furthermore, the R 2 s for these regression models range from 10.08% to 13.78% which provides evidence that these models explain a significant amount of the variation in long-term growth forecasts. As before, I split the tables into the main models and the overvaluation models. Overall, the results are generally consistent with the main results (see Tables 5 to 10 using the post-regulation FD samples) and hypotheses 2 and 4 are generally supported (using a one-tail test). 121

135 7.4 Controlling for Interaction Effects of Income-Decreasing Earnings Management Kothari et al. (2006) find that analyst forecast optimism is concentrated among the high, but not the low, accrual portfolio. This finding is based on the argument that undervalued firms do not have incentives to report low accruals. Therefore, in this study, I do not investigate the association between income-decreasing earnings management as mispricing changes and analysts responses. However, it is possible that, by excluding the interaction of income-decreasing earnings management and MISPRICING in my models, I am omitting a control factor from my analysts responses models. Tables report the results of the GLM regressions using income-decreasing interactions in the analysts responses models. The F-test statistics for all the models drawn from these tables are significant at a 1% level of significance which suggests all the overall models are significant. Table 20 tests hypotheses one and two. The R 2 s for these regression models range from 5.48% to 11.71%. As before, I split the tables into the main models and the overvaluation models. The results are generally consistent with the main results (see Tables 5 and 8 using the post-regulation FD samples). Specifically, hypothesis 2 is supported and MISPRICING s (OVER s) coefficient is negative. The income-decreasing earnings management interaction variables are not significant in the overvaluation models. Table 20 tests hypothesis 3. The R 2 s for these regression models range from 6.36% to 13.94%. My main model findings using the stock recommendations sample are similar to my findings in the main results. Specifically, the negative association between income-increasing earnings management and the favorableness of analysts stock recommendation responses as mispricing increases is greater for the more costly GAAP- 122

136 based methods used to manage earnings. This finding provides partial support for hypothesis 3. However, using the favorableness of price targets as the dependent variable, I do not find support for hypothesis 3 even though the association between analysts price targets responses and income-increasing earnings management is more negative as mispricing increases for each of the earnings management mechanisms. These findings are contrary to my main results using the price targets sample. Using the overvaluation models, my findings from the sensitivity analyses are consistent with the main results (see Table 9 using the post-regulation FD samples). Specifically hypothesis 3 is not supported in both the stock recommendations and price targets samples. In the models reported in this table, some of the income-decreasing earnings management interactions variables are significantly associated with analysts responses. Table 22 test hypotheses 4 and 5 using the stock recommendations and price targets samples respectively. The R 2 s for these regression models range from 5.73% to 12.81%. My findings from the main model results for both the stock recommendations and price targets samples are consistent with my findings in the main results. Specifically, the negative association between income-increasing earnings management and the favorableness of analysts responses is greater as the duration of consecutive periods of income-increasing earnings management increases. This finding is consistent with hypothesis 4. In addition, my findings from the overvaluation model results are also consistent with my main results. Thus, I find that the association between incomeincreasing earnings management and the favorableness of analysts responses is not more negative as the duration of consecutive periods of overvaluation increases. This finding is 123

137 inconsistent with hypothesis 5. The income-decreasing interaction variables are generally not significant in the overvaluation models. In summary, interacting income-decreasing earnings management with MISPRICING (OVER) does not impact most of my findings. However, using this model, hypothesis 3 is not supported using the price targets sample. 7.5 Changes to Income-Increasing Earnings Management It is possible that analysts may respond to a change from income-decreasing to income-increasing earning management, especially when management has incentives to change to a strategy of opportunistic earnings management. In my analyses, I assume that greater mispricing provides greater incentives to use opportunistic earnings management. Thus, I investigate whether analysts responses are caused by the combination of changes in income-increasing earnings management and changes in mispricing. As it is not possible to test the cause of a phenomenon using a levels model, I also consider a changes specification to provide evidence that would be consistent with causal effects. I modify equation (10) to focus on the changes in analysts responses and the independent variables between years. For my test variable, I focus on whether the firm s GAAP-based earnings management mechanisms were non-income-increasing in the prior year and income-increasing in the current year ( PSABNACCRUALS, PSABNRTM and PRESTATE). That is, I am capturing the change to income-increasing earnings management only. Consistent with hypothesis 1, I expect a negative association between a change to income-increasing earnings management and analysts responses. 124

138 Furthermore, similar to the second hypothesis, I expect that the interaction term representing the interaction between the change to income-increasing earnings management and the change in the ratio of price to fundamental value to be negative, indicating that the association between the change in the favorableness of the analysts responses and the change in MISPRICING is more negative if the firm has a change in income-increasing earnings management. Finally, I expect changes to accruals management to be the least costly, followed by changes to real transactions management, and changes to financial irregularities are predicted to be the most costly earnings management mechanism. It follows that, consistent with hypothesis 3, I expect the relationship between changes in earnings management and changes in analysts responses to be more negative as a result of changes in the more costly earnings management mechanism. The changes specifications are as follows: RESPONSE it+1 = α + β 1 MISPRICING it+1 + β 2 PSABNACCRUALS it + β 3 PSABNRTM it + β 4 PRESTATE it + β 5 MISPRICING it+1 * PSABNACCRUALS it + β 6 MISPRICING it+1 * PSABNRTM it + β 7 MISPRICING it+1 * PRESTATE it + ε it (25) where: RESPONSE represents proxies for analysts responses using a change model. This dependent variable is either change in stock recommendation ( REC) or change in price target ( PTG). See Table 3 for definitions. ΔMISPRICING is the change in P/V ratio by taking the P/V ratio in year t away from the P/V ratio in year t+1, all scaled by the P/V ratio in year t. PSABNACCRUALS is an indicator variable taking the value of 1 if firm i s total abnormal accruals in year t-1 (the preceding fiscal year) is not 125

139 income-increasing and the total abnormal accruals in year t (the current fiscal year) is income-increasing, and 0 otherwise. PSABNRTM is an indicator variable taking the value of 1 if firm i s total abnormal real transaction management in year t-1 (the preceding fiscal year) is not income-increasing and the total abnormal real transaction management in year t (the current fiscal year) is income-increasing, and 0 otherwise. PRESTATE is an indicator variable taking the value of 1 if firm i s restatement in year t-1 (the preceding fiscal year) is not adverse (or if there was no restatement in year t-1) and firm i s restatement in year t is adverse, and 0 otherwise. All other variables are previously defined. I add change in MISPRICING ( MISPRICING) as a control variable in the change models. Forward-looking analysts may identify trends and make changes to their recommendations based on current trends including changes in P/V ratios. As a result of analysts heuristics in decision making, I do not predict the sign of the relationship between change in MISPRICING and the change in the favorableness of the stock recommendation (price target). I use change to adverse restatement ( PRESTATE), change to income-increasing abnormal RTM ( PSABNRTM) and change to income-increasing discretionary accruals ( PSABNACCRUALS), as my changes in earnings management explanatory variables. Consistent with hypothesis 1, I expect coefficients β 2, β 3 and β 4 to be negative in equation 25. In addition, consistent with hypothesis 3, I expect the negative association between 126

140 changes in earnings management as firm mispricing increases and changes in the favorableness of analysts responses to be greater for the most costly earnings management mechanisms. Specifically, I expect β 7 to have greatest effect on the dependent variables. Next, I expect β 6 to have a greater effect on the dependent variables than β 5, consistent with analysts changing their recommendations and price targets more for changes in the most costly earnings management choices as mispricing increases. For the overvaluation model, I expect a similar relationship between β 7 to β 5, consistent with hypothesis 3. Table 23 reports the results of the GLM regressions using my changes samples. The F-test statistics (untabulated) for all the models drawn from equation 25 are significant at a 1% level of significance, which suggests all the overall models are significant. Furthermore, the R 2 s for these regression models range from 3.72% to 21.45%. In the overvaluation model, I replace MISPRICING with OVER as both variables are proxies for mispricing. Consistent with hypothesis 1, I find a change to income-increasing abnormal RTM ( PSABNRTM) is negatively associated with a change in the favorableness of stock recommendations ( REC). This finding suggests that changes to RTM may cause lower analysts stock recommendations after Regulation FD. However, contrary to hypothesis 1, a change to income-increasing accruals management ( PSABNACCRUALS) and a change to an adverse restatement ( PRESTATE) are not negatively associated with REC. For my tests of hypothesis 3 using the main model, I find that the interaction of MISPRICING and PSABNRTM is negatively and significantly associated with REC. However, the interaction of MISPRICING and PSABNACCRUALS ( PRESTATE) is not significantly 127

141 associated with REC. As income-increasing real transaction management is more costly than income-increasing accruals management, I conclude that the most costly methods used to manage earnings do increase the negative association between a change in the favorableness of the analysts stock recommendation responses and a change in incomeincreasing earnings management. Thus, hypothesis 3 is partially supported for stock recommendations. However, as the coefficient for PRESTATE is not significant, I find that the most costly method used to manage earnings does not increase the negative association between REC and a change in income-increasing earnings management; therefore, hypothesis 3 is not fully supported. Taken together, these results suggest that analysts appear to be mainly concerned with real transaction management and do therefore take the cost of the earnings management method used into consideration. Analysts may not adjust as much for adverse restatements because the restatements may be made in later periods, and, as the actions are outside GAAP, analysts may not be aware of the earnings management behavior. Using my price target sample, I find that changes to the different incomeincreasing earnings management mechanisms ( PSABNACCRUALS, PSABNRTM and PRESTATE) are negatively associated with a change in the favorableness of price targets ( PTG) using a one-tail test. This finding is consistent with hypothesis 1. For my tests of hypothesis 3 using the main model, I find that the interaction of PSABNACCRUALS ( PSABNRTM) and MISPRICING is not significantly (negatively) associated with PTG. As income-increasing real transaction management is more costly than income-increasing accruals management, I conclude that hypothesis 3 is partially supported when using the full price target sample. However, the interaction of 128

142 PRESTATE and MISPRICING is positively associated with PTG. Therefore, the most costly method of managing earnings does not increase the negative association between a change in the favorableness of the analysts stock recommendation responses and a change in income-increasing earnings management and hypothesis 3 is not fully supported. I now turn to my overvaluation models. I find that firm overvaluation is negatively (positively) associated with REC ( PTG). The only other significant relationship in the regressions is the negative association between a change to incomeincreasing real transaction management ( PSABNRTM) and changes in analysts responses. All the other regression relationships are insignificant or contrary to expectations, including all the interaction relationships. Thus, hypothesis 3 is not supported in the overvaluation models using a change sample. This result means I did not find any evidence that the cost of the method used to manage earnings in overvalued firms causes analysts to reduce stock recommendations or price targets. In summary, this sensitivity analysis provides evidence on hypotheses 1 and 3. I find partial support for hypotheses 1 and 3 using these change models but no support for hypothesis 3 in the overvaluation models. I also find evidence which may provide indirect support for hypothesis 2 in these regressions. Overall, the results suggests that changes to real transaction management may cause changes in analysts responses post- Regulation FD and the negative association between income-increasing GAAP-based earnings management in mispriced firms may be greater as the cost of the method used to manage earnings increases. However, the negative association between REC ( PTG) and a change to income-increasing earnings management is not greater for changes to 129

143 financial irregularities than for changes to GAAP-based earnings management. This may be because financial irregularities are not observable in the period of the indiscretion, which may limit analysts ability to respond accordingly. 7.6 Growth and Economic Downturn Time Periods I also investigate whether my hypotheses hold during different time periods. It is possible that analysts are more optimistic during economic booms or periods of growth, and corresponding less optimistic during economic downturns. Therefore, I split the post- Regulation FD period into economic growth and economic downturn time periods. I define economic growth as a positive change in GDP and economic downturn as a negative change in GDP. I ignore 2000 to 2001 because that period consists of frequent changes between economic growth and economic downturn, which results in relatively small sample sizes. The time period spanning 2002 to 2007 represents my economic growth period, with annual changes in GDP ranging from 1.8% to 3.5% in The sample sizes in this economic growth time period is 13,564 stock recommendation firm observations and 12,226 price target firm observations. The time period 2008 to 2009 represents my economic downturn period. The sample sizes in this economic downturn time period is 4,144 stock recommendations firm observations and 4,006 price target firm observations. Tables report the results of the GLM regressions separated into the different economic time periods. The F-test statistics for all the models drawn from these tables are significant at a 1% level of significance which suggests all the overall models are significant. Table 24 tests hypotheses one and two. The R 2 s for these regression models 130

144 range from 2.53% to 12.07%. I split the table into the economic growth time period and the economic downturn time period. The stock recommendation results in both time periods are generally consistent with the main results (see Table 5 using the post- Regulation FD samples). Specifically, hypothesis 2 is supported. However, using the price target sample, hypothesis 2 is not supported during the economic growth years but is supported during the economic downturn period. In untabulated results, using firm overvaluation as the proxy for mispricing, hypothesis 2 is supported in both stock recommendation and price target samples and for both time periods. Table 25 tests hypothesis 3. The R 2 s for these regression models range from 3.42% to 14.19%. My findings using the stock recommendations sample during both time periods are similar to my findings in the main results. Specifically, the negative association between income-increasing earnings management and the favorableness of analysts stock recommendation responses as mispricing increases is greater for the more costly GAAP-based methods used to manage earnings. This finding provides partial support for hypothesis 3. Similarly, hypothesis 3 is partially supported using the price targets sample during the economic downturn time period. However, using the favorableness of price targets as the dependent variable during the economic growth period, I do not find support for hypothesis 3. In untabulated results, using firm overvaluation as the proxy for mispricing, hypothesis 3 is not supported in the stock recommendation samples for either time periods. However, hypothesis 3 is partially supported in the price target samples for both time periods. Table 26 tests hypothesis 4. The R 2 s for these regression models range from 2.97% to 13.37%. My findings for the economic growth time period results for both the 131

145 stock recommendations and price targets samples are consistent with my findings in the main results. Specifically, the negative association between income-increasing earnings management and the favorableness of analysts responses is greater as the duration of consecutive periods of income-increasing earnings management increases. This finding is consistent with hypothesis 4. However, during the economic downturn (2008 to 2009), I find that for both stock recommendations and price targets, hypothesis 2 is supported but hypothesis 4 is not supported. This finding may be result of a reduction of incomeincreasing earnings management during economic downturns as a result of big bath behavior. This big bath behavior is likely to be more prevalent in firms that have managed earnings in the past. Finally, table 27 tests hypothesis 5. The R 2 s for these regression models range from 2.29% to 11.14%. My findings for all the regressions are consistent with my findings in the main results. Specifically, the association between income-increasing earnings management and the favorableness of analysts responses is not more negative as the duration of consecutive periods of overvaluation increases. This finding is at least moderately consistent with hypothesis 2 but is inconsistent with hypothesis 5. In summary, my stock recommendation results are generally robust to the economic climate, however, the price target results mainly hold during economic downturns or in the overvaluation model. Interestingly, economic downturns may reduce strings of income-increasing earnings management, which impacts both hypotheses 2 and

146 7.7 Petersen Cluster Regressions and Newey-West Standard Error Correction An alternative approach for addressing the cross-sectional and time-series dependence standard errors as a result of panel data is to cluster standard errors (Petersen, 2009). In this sensitivity analysis, I correct for correlations among firms and within years by using clustered standard errors in two dimensions to test my hypotheses. Using the clustering approach with stock recommendations post-regulation FD sample, I find results similar to my main results. Specifically, I find that hypothesis 3 is partially supported, hypothesis 4 is supported but hypotheses 1 and 5 are not supported. However, hypothesis 2 is only supported in the overvaluation model and is not significant at a 10% level of significance in the main model. Alternatively, using the clustering approach with the price targets post-regulation FD sample, I find results consistent with all the main results. Specifically, I find that hypothesis 2 is supported in both the main and overvaluation models, hypothesis 3 is partially supported in the main model only, hypothesis 4 is supported but hypotheses 1 and 5 are not supported. Finally, when using time series data, the error terms may be correlated over time which may result in autocorrelation and heteroskedasticity. Therefore, I test my hypotheses by using Newey-West standard errors to correct for the effects of correlation in the error terms. Using Newey-West standard errors with both stock recommendations and price targets post-regulation FD samples; I find results consistent with my main results. Specifically, hypothesis 2 is supported using both the main mispricing proxy and the overvaluation model, hypothesis 3 is at least partially supported using both the main and overvaluation models and hypothesis 4 is supported. Consistent with the main results, 133

147 hypotheses 1 and 5 are not supported. Interestingly, using Newey-West standard errors, hypothesis 3 is fully supported in the main model using the stock recommendation sample and hypothesis 3 is partially supported in the overvaluation models. 134

148 CHAPTER 8 CONCLUDING REMARKS Prior research has found opportunistic earnings management to be potentially costly to firms. In addition, the method used to manage earnings and the number of consecutive periods of earnings management has been found to increase the costs of earnings management. I examine whether analysts consider earnings quality in their outlooks and whether analysts responses are consistent with strategies based on residual income models after controlling for earnings quality. As a result, I examine the monitoring role of financial analysts and hypothesize that analysts have incentives to reduce the favorableness of their stock recommendations (price targets) as a result of income-increasing earnings management as mispricing increases. Furthermore, I hypothesize that analysts stock recommendations (price targets) are further impacted by the costs of the methods used to manage earnings and the length of consecutive periods of earnings management. I separately investigate analysts stock recommendation (price target) response to earnings management in overvalued firms. Prior research provides evidence that earnings management in overvalued firms is potentially costly and I hypothesize financial analysts will reduce their stock recommendations (price targets) more as a result of earnings management as the length of consecutive periods of overvaluation increases. I use the base model designed by Bradshaw (2004) to test my hypotheses. I add 135

149 earnings management variables to the model and separately test analysts responses after Regulation FD. Furthermore, I use levels data and change data (in sensitivity analyses) to test my hypotheses to provide evidence on both association and causation effects. To address the fixed effects in my panel data, I use a generalized linear model. Contrary to hypothesis 1, income-increasing earnings management is not correlated with a reduction in stock recommendations (price targets) using levels data. However, I find partial support for hypothesis 1 because analysts reduce their change in stock recommendations (price targets) when firms report a change to income-increasing RTM. Thus, hypothesis 1 is partially supported by the use of change models. Consistent with hypothesis 2, I find that the interaction between earnings management and mispricing is negatively associated with the favorableness of stock recommendations (price targets). I find this relationship throughout the sample period, extending to after Regulation FD. This finding provides additional evidence that analysts are effective firm monitors and may be able to identify the destruction of firm value. I also find evidence that analysts reduce stock recommendations (price targets) more as a result of income-increasing real transactions earnings management than as a result of income-increasing discretionary accruals, which provides partial support for hypothesis 3. However, I do not find that adverse restatements reduce the favorableness of stock recommendations more than real transactions management, contrary to hypothesis 3. Finally, I find strong evidence that the interaction between the length of the consecutive periods of income-increasing earnings management and mispricing is negatively associated with the favorableness of stock recommendations (price targets), consistent with hypothesis 4. I find that some results only hold in samples taken after the 136

150 enactment of Regulation FD or are stronger after Regulation FD. Thus, this study provides evidence that Regulation FD improved analysts firm monitoring by removing some of the incentives to produce biased reports. In separate analysis, I find further evidence in support of hypothesis 2 as earnings management in overvalued firms is found to be negatively associated with the favorableness of stock recommendations (price targets). However, I do not find that the favorableness of analysts stock recommendation (price target) responses reduces as the cost of the method used to manage earnings in overvalued firms increases, contrary to hypothesis 3. I also do not find any support for hypothesis 5 (that the association between income-increasing earnings management and the favorableness of stock recommendations [price targets] is more negative as length of the consecutive periods of overvaluation firms increases). An important goal of this study is to shed further light on the association between stock recommendations (price targets) and fundamental value. This study sought to desegregate fundamental value and investigate whether analysts stock recommendations (price targets) are consistent with stock-picking strategies based on residual income after controlling for the quality of earnings (as proxied through earnings management). Prior research found that stock recommendations are inconsistent with stock-picking strategies based on fundamental value. However, this study provides evidence that, after including earnings management in the regression model, stock recommendations (price targets) are indeed consistent with stock picking strategies based on residual income computations. 137

151 APPENDIX A: FIGURES 138

152 Figure 1: Relationships between Overvalued Equity, Earnings Management and Analysts Stock Recommendation Overvaluation 139 Earnings management + Abarbanell and Lehavy (2003) Analyst Stock Recommendation Response Key = Not Previously Tested

153 Figure 2: Relationships between Overvalued Equity, Earnings Management and Analysts Price Targets Overvaluation 140 Earnings management? Analyst Price Target Response Key = Not Previously Tested