The Effect of Shareholder-Initiated Corporate Governance. On Accrual-Based and Real Earnings Management *

Transcription

1 The Effect of Shareholder-Initiated Corporate Governance On Accrual-Based and Real Earnings Management * Jeffrey Ng School of Accounting and Finance, The Hong Kong Polytechnic University tee-yong-jeffrey.ng@polyu.edu.hk Hong Wu School of Accounting and Finance, The Hong Kong Polytechnic University hong.87.wu@polyu.edu.hk Weihuan Zhai School of Accounting and Finance, The Hong Kong Polytechnic University zhai.zhai@polyu.edu.hk Jing Zhao School of Accounting and Finance, The Hong Kong Polytechnic University jingzhao@polyu.edu.hk October 2017 Shareholder activism is an important source of corporate governance. Using a dynamic regression discontinuity design on shareholder proposals that pass or fail by a small margin of votes in shareholder meetings, we analyze the effect of shareholder-initiated corporate governance on earnings management. We find that both accrual-based and real earnings management diminish after shareholder proposals are passed. Consistent with improved financial reporting quality, we show that the audit fees are higher following passage of shareholder proposals. We obtain similar reduction in earnings management when we examine the subsample of proposals that focus on changing governance structures (e.g., board independence). Interestingly, when we focus on proposals to improve financial reporting (auditor independence and clawback adoptions), we find that accrual-based earnings management decreases while real earnings management increases, suggesting a substitution effect. This substitution effect is stronger when there are entrenched auditors, consistent with firms shifting away from accrualsbased earnings management when shareholders act to improve auditor independence. Overall, our paper demonstrates that shareholder-initiated corporate governance can affect earnings management and that the effect can vary depending on the nature of the activism. JEL: M41 Keywords: shareholder activism, shareholder proposals, accrual-based earnings management, real activities manipulation * We appreciate the helpful comments and suggestions from Samir Trabelsi, Feng Tian, Shivaram Rajgopal, and seminar participants at The Hong Kong Polytechnic University.

2 1. Introduction The agency problem the conflict created by the misalignment of incentives between owners and managers lies at the heart of the theory of the firm (Berle and Means, 1932; Jensen and Meckling, 1976). The design of a firm s corporate governance structure should therefore seek to minimize the costs that conflict creates. Shareholder activism has been growing rapidly in recent years, as shareholders seek to improve corporate governance and encourage superior performance (Pound, 1992; Black, 1992). Of the many forms of shareholder activism, the submission of shareholder proposals is the least costly and most common way for shareholders to put forward initiatives to improve corporate governance (Gillian and Starks, 2000; IIiev et al. 2015). Cuñat, Gine, and Guadalupe (2012) report that 3,984 shareholder proposals were filed between 1997 and 2007 among S&P 1500 firms, with the trend increasing over the period. Shareholder proposals are submitted under SEC Rule 14a-8, which allows the initiating shareholders to include, at the company s expense, a 500-word supporting argument in the firm s proxy materials. Shareholder votes on these proposals are not binding. Nonetheless, passing a shareholder proposal may impose external pressure for managers to improve corporate governance. Recent research shows that shareholder proposals are associated with small positive valuation effects and better long-run performance (Thomas and Cotter, 2007; Renneboog and Szilagyi, 2011; Guercio, Seery, and Woidtke, 2008; Cuñat et al., 2012, 2016). In particular, Cuñat et al. (2012) find that passing a corporate governance proposal generates a 1.3% abnormal return on the day of the vote, with an implied increase in market value of 2.8% per proposal. Bach and Metzger (2017) explore the shareholder defiance mechanism through non-binding governance proposals create value and find empirical evidence that majority support for a 1

3 shareholder proposal essentially forces management to follow shareholders will by implementing the proposal. 1 To the extent that passing a shareholder-initiated governance proposal effectively improves corporate governance by strengthening the alignment between managerial incentives and shareholder interests, we should expect such proposals to reduce earnings management, which is a prevalent management behavior driven by agency conflicts (e.g., Healy, 1985; Schipper, 1989; Graham, Harvey, and Rajgopal, 2005). For example, after passing a shareholder proposal to improve board independence, the power of CEO may be constrained and the corporate governance of a firm can be improved. Thus, managers may have less discretionary power to manipulate earnings numbers, and firms will be less likely to engage in earnings management. However, the effect of the passage of shareholder proposals on earnings management is not without tension. First, the non-binding nature of shareholder proposals might mean that their passage might lack the necessary teeth to affect earnings management. Second, prior studies have provided evidence that efforts to improve corporate governance can have the opposite effect on accrual-based and real earnings management (Cohen and Zarowin, 2010; Zang, 2012; Chan et al., 2015). Zang (2012) suggests that there is an equilibrium between accrual-based and real earnings management and finds that managers trade off the two earnings management methods based on their relative costs. To the extent that the pressure to meet earnings targets has not changed, the passage of shareholder proposals that specifically target accrual-based earnings management shifts managerial preferences toward real earnings management (Cohen and Zarowin, 2010; Chan et al., 2015). If so, one might expect the passage of such proposals to 1 When management refuses to implement a majority-supported proposal, the management is intensely publicized by shareholder organizations, undermining the confidence that investors had previously put into management and trigger activist campaigns. 2

4 reduce accrual-based earnings management but to increase real earnings management. In fact, this presumably unintended substitution effect is an example of why well-intended shareholder proposals, even if they achieve their goals, might not always increase firm value. It is challenging to estimate the causal effect of shareholder proposals on earnings management. This is because the outcomes of proposals are arguably endogenous and correlated with other firm characteristics. Thus, comparing the earnings management measures of firms with different vote outcomes is likely to capture the effect of those unobserved characteristics, rather than the effect of the shareholder proposals. To overcome this limitation, we need a setting in which the proposal outcomes are exogenous or random. We follow Cellini, Ferreira, and Rothstein (2010) and Cuñat et al. (2012) in adopting dynamic Regression Discontinuity Design (RDD) model. 2 This empirical strategy essentially compares the earnings management of firms with proposals that pass by a small margin to those with proposals that fail by a small margin. Near the passing threshold, a small increase in vote shares (say, from 40% to 60%) would lead to a discontinuity of change in the probability of implementing a proposal (Ertimur, Ferri, and Stubben, 2010 and Bach and Metzger, 2017). 3 Thus, passing a shareholder proposal is "locally" exogenous and uncorrelated with firms characteristics, allowing us to draw causal inferences about its passage. We employ all shareholder-sponsored governance proposals of firms included in the S&P 1500, in addition to another 500 widely held firms between 1997 and Our results show that both accrual-based and real earnings management decrease after shareholder proposals are 2 This design alleviates the significant problem of endogeneity between actions to improve corporate governance and corporate outcomes. For example, given the number of concurrent events that occur just before and during shareholder meetings, it might be difficult to attribute corporate outcomes to shareholder proposals per se. Further, these outcomes could arise from other (uncontrolled) factors that could be related to shareholder proposals and corporate outcomes. 3 In section 4.2, we employ a RDD approach and confirm that passage of shareholder proposals indeed increases the likelihood of implementing those proposals. 3

5 passed. The reduction in accrual-based earnings management is significant for up to two years after the annual meeting, while the reduction in real activities manipulation remains significant for three to four years after the annual meeting. The result is robust to various measures of accrual-based and real earnings management. As firms put more effort to improve financial reporting quality, the demand for higher audit quality increases. Previous papers have shown that audit quality and auditor s effort are positively related with audit fees. For example, Srinidhi, He, and Firth (2014) show that strongly governed family firms demand for higher audit effort, pay more audit fees, and have better earnings quality compared to the weakly governed family firms. Consistent with their study, we document that audit fees increase after passing shareholder proposals. We further explore the effect of shareholder proposals and find an interesting heterogeneity among different types of proposals. Proposals that aim to change the firm s governance structures reduce both accrual-based and real earnings management. In contrast, those that focus mostly on improving the firm s financial reporting have opposite effects on accruals-based and real earnings management: Accrual-based earnings management declines but real activities manipulation increases. The financial reporting proposals include proposals that aim to improve auditor independence and those that intend to implement clawback provisions policies that authorize the board to recoup the compensation paid to executives based on misstated financial reports. As is shown in with Cohen, Dey, and Lys (2008) and Zang (2012), it is harder for managers to beef up the accounting numbers via discretionary accruals with strengthened monitoring from the auditors. Similarly, with the clawback provisions in place, managers would like to avoid high accounting accruals, which may attract more scrutiny from the SEC and lead to subsequent accounting restatements and claw backs (Chan et al., 2015). In other words, these 4

6 financial reporting proposals are likely to put a severe constraint on managers abilities to engage in accruals management. However, as managers are still under pressure to boost earnings numbers, these proposals may drive them to switch to real earnings management, which is harder to detect but potentially more costly to shareholders. The substitution effect between accrual-based and real earnings management would represent an unintended consequence of particular types of shareholder proposals. To further substantiate this substitution effect, we conduct a difference-in-difference analysis by examining the role of entrenched auditors. Auditors play an important role in ensuring the integrity of financial reporting process. When a firm has an auditor that has been with the firm for a long period and there are nevertheless shareholder proposals to address weaknesses in the accounting process, one implication is that the entrenched auditor is not sufficiently monitoring managers opportunistic accounting (Carey and Simnett, 2006). Consistent with the corporate governance role of shareholder activism in terms of reducing such managerial opportunism, we find that for firms with entrenched auditors, accrual-based earnings management is reduced even more after proposals to improve auditors independence and implement clawback provisions. In contrast, real earnings management shows an even greater increase after such proposals, indicating that shifts to real earnings management are induced by shareholders attempts to improve the financial reporting process. We make several contributions to the previous literature. First, we add to the literature on shareholder-initiated corporate governance. Cuñat et al. (2012) show that passing a governance proposal enhances corporate governance and firm value. However, little is known about the specific mechanisms through which corporate governance is impacted by governance proposals. Our paper shows that mitigating an important agency issue, i.e. the earnings management, can be 5

7 one important channel. We also contribute to the earnings management literature by showing how corporate governance affects earnings management. A large literature has examined the relation between corporate governance and earnings management. 4 While prior studies have examined many measures of corporate governance, their results are frequently contradictory. The reasons for these contradictions include the measurement errors inherent in different corporate governance constructs and correlated omitted variable problems (Larcker, Richardson, and Tuna, 2007). Establishing causal estimates can be difficult in this setting as a firm s corporate governance structure is usually correlated with its fundamental characteristics. To the best of our knowledge, we are the first paper to employ the dynamic RDD approach to identify the effect of shareholder-initiated corporate governance on earnings management. This approach has been used by Cuñat et al. (2012, 2016) and Flammer (2015) to investigate the stock market reaction to shareholder proposals and by Armstrong, Gow, and Larcker (2013) to examine the effect of sayon-pay proposals on the level or composition of future CEO incentive compensation. We complement these studies and show that governance proposals have causal effects on earnings management. Further, we add to the recent discussion on how improved corporate governance can have different effects on accrual-based and real earnings management. In general, prior studies have found evidence of the substitution between accrual-based and real earnings management following attempts to improve firm s financial reporting (e.g., Cohen et al., 2008; Chan et al., 2015). We complement and present a contrast to this literature by studying the effect of shareholder-initiated corporate governance via shareholder proposals and relying on the voting outcomes of these proposals to identify their effects. 4 Some of these studies include Morck, Shleifer, and Vishny (1988), Byrd and Hickman (1992), Brickley, Coles, and Terry (1994), Yermack (1996), Core, Holthausen, and Larcker (1999), Klein (2002), and Gompers, Ishii, and Metrick (2003). 6

8 Our findings also highlight the importance of understanding the heterogeneity in shareholder proposals and, more broadly, in different corporate governance mechanisms. Specifically, when we drill down into specific proposal types, we document results that contrast with findings that treat all proposals as a single group. Such findings might have important policy implications for shareholders and policymakers: Managers can and will respond to shareholder activism in ways that might be different from the initial intensions of shareholders. The rest of this paper is organized as follows. Section 2 provides institutional backgrounds of the shareholder proposals and reviews the prior literature. Section 3 discusses the empirical design. We present our results in Section 4 and conclude in Section Institutional Background and Literature Review 2.1 Institutional Background of Shareholder Proposals Our empirical analysis regarding the impact of shareholder votes on earnings management relies heavily on features of shareholder voting in the US. Therefore, in this subsection, we offer a brief description of the shareholder voting process. More details on shareholder proposals can be found in IIiev et al (2015), Cuñat, Gine, and Guadalupe (2012), and Bach and Metzger (2017), etc. On the occasion of general assemblies, shareholders can be asked to vote on many different matters. They may elect directors or vote on specific proposals, sponsored either by the management or by a shareholder. Our study is centered on shareholder-sponsored governance proposals because, unlike management sponsored ones, shareholder-sponsored proposals cannot be removed strategically by the firm s management, and their vote distribution is not affected by selective withdrawal around the majority threshold (Listokin, 2008). By SEC rule 14a-8, any 7

9 shareholder with a holding in the company worth at least $2,000 or 1% of outstanding shares can submit such proposals. Those proposals are different in several dimensions from managementsponsored proposals that have been studied in related papers (e.g., Popadak, 2013). The main difference is that shareholder-sponsored proposals are not usually binding the management. This means that even if the approval threshold set by the corporate charter has been passed, managers and the board of directors have discretion over whether or not to implement the proposal. Whether the non-binding votes carry real effects is an interesting empirical question that motivates our focus on this type of governance proposal. Even though an approved proposal is not binding, it may still have monitoring power over the firm. While passage of a shareholder proposal according to the charter never binds the board s final decision, a refusal to implement it may still be a valid proof of breach of fiduciary. The Council of Institutional Investors (CII), which by 2004 represented over 140 pension funds, including many of the biggest ones, would push for implementation based on a bar for approval that is uniform across firms (e.g., voting support for the proposal reaching more than 50% of votes cast for and against ). CII provides an annual list of the proposals that have reached this level of support to its member institutions, and management need to provide solid justifications if they decide not to implement a proposal. Management s refusal to implement passed shareholder proposals might bear significant consequences for the management and board of directors. In companies in which a shareholder proposal reaches 50% of votes cast for and against, CII considers that a board is in breach of CII policies, a list of best governance practices, if the proposal is then not implemented by the board. Members of CII use the organization as a forum to discuss their voting decisions for each firm in which they hold a stake. Because CII policies serve as a guideline for those discussions, 8

10 breaching the policies is very likely to trigger sanctions in the form of a high number of votes withheld by funds which belong to CII. In order to detect breaches to its policies, CII keeps track of the implementation of majority-supported proposals using both public (SEC filings and news releases) and private sources of information. It then issues a list of complying and noncomplying firms and makes it available to its members. The latter can then individually select companies from this list and launch vote-no campaigns against them, or the list may serve as an input in the discussions that CII organizes between its members ahead of each proxy season (Anand and Givant Star, 1994). Importantly, the organization treats the chief executive officer of companies where a majority vote took place as the main individual accountable for the decision to implement shareholder proposals. For instance, when there is no direct evidence that a company has implemented a passing proposal, the CII staff sends a letter directly to the CEO asking for a justification; the resulting correspondence is made public to all members of the shareholder organization. Apart from the CEO, these activists may in addition target directors who form the governance committee since decisions to reform the corporate charter are taken there. This makes it likely that decisions to withhold a vote made by CII members may particularly target the incumbent CEO and governance committee members. Overall, the critical features of the shareholder voting system imply that passed shareholder proposals do bear real consequences as management is faced with significant pressure from third parties and shareholders and hence is more likely to implement the proposals. In Section 4.3, we analyze the implementation probability of passed proposals and provide evidence for the monitoring power of these proposals. 9

11 2.2 Previous Literature on the Effect of Shareholder Proposals In recent years, shareholder activism has been a rapidly growing trend, a response to statelevel and firm-level adoption of various provisions that insulate managers from the monitoring and control of shareholders. Among many forms of shareholder activism, the submission of shareholder proposals is the least costly and most common way for shareholders to put forward initiatives to improve corporate governance (Gillian and Starks, 2000; IIiev et al. 2015). In general, studies on the effect of corporate governance on managerial behavior and firm performance have produced mixed evidence (Morck et al., 1988; Byrd and Hickman, 1992; Brickley et al., 1994; Yermack, 1996; Core et al., 1999; Klein, 2002; Gompers et al., 2003). The reasons for the varied outcomes include the measurement errors inherent in different corporate governance constructs and correlated omitted variable problems (Larcker et al., 2007). Establishing causal estimates can be difficult in this setting as corporate governance structure is usually correlated with firms fundamental characteristics. Recent research shows that shareholder proposals are associated with small positive valuation effects and better long-run performance (Thomas and Cotter, 2007; Renneboog and Szilagyi, 2011; Guercio et al. 2008; Cuñat et al., 2012). In particular, Cuñat et al. (2012) find that passing corporate governance proposals generates a 1.3% abnormal return on the day of the vote with an implied increase in market value of 2.8% per proposal. They also document that successful proposals lead to less inefficient investments and value-destroying acquisitions. Sun et al. (2013) find that discretionary accruals increase for firms that receive pay-for-performance shareholder proposals as these proposals introduce incentives to manage earnings for short-term gains. 10

12 Given the mixed evidence in the prior literature on the effectiveness of the passage of shareholder proposals and the potentially different effects of corporate governance on accrualbased and real earnings management, we state our hypothesis in the null form: H0: The passage of shareholder-initiated corporate governance proposals has no effect on earnings management. 3. Research Design A common challenge for the existing literature on shareholder activism is to establish a causal relationship between the activism and performance outcomes. Shareholder intervention is arguably endogenous and correlated with other firm characteristics. For instance, activists have been shown to target firms with high liquidity, poor stock performance, and low leverage and market valuation (Norli, Ostergaard, and Schindele, 2015; Fos, 2017). Bourveau and Schoenfeld (2017) argue that activists target mispriced firms and that managers can preempt activism at their firm by providing more voluntary disclosure, which brings about positive price correction. Therefore, comparing firms earnings management measures with different levels of shareholder activism is likely to capture the effect of those unobserved characteristics rather than the effect of the activism. To overcome this limitation, we employ the dynamic RDD model in Cellini et al. (2010). This empirical strategy essentially compares firms earnings management with proposals that pass by a small margin to those with proposals that fail by a small margin. In the spirit of RDD, within the close margin of the passing threshold, whether a proposal passed or failed can be treated as random. For example, whether a proposal will pass with 60% votes or fail with 40% cannot be predicted ex-ante. In other words, firms with proposals within this close margin have 11

13 similar characteristics. Hence, our estimate captures a causal effect of shareholder proposals and does not incorporate any endogenous confounding factors as long as their effect is continuous around the threshold. A similar RDD has been used by Cuñat et al. (2012, 2016), Flammer (2015), and Ge et al. (2017) to investigate the effectiveness of shareholder proposals Regression Discontinuity in Shareholder Votes We closely follow Cuñat et al. (2012, 2016) to implement a multi-period version of the dynamic RDD. The benefit of this model is that it allows us to estimate the effect of shareholder proposals on earnings management activities for each year after the proposal s passage: t+1, t+2, etc. Another benefit is that it aggregates the votes for a given firm and meeting date, which takes into consideration the fact that shareholders may vote on more than one proposal during an annual meeting. Hence, the multi-period dynamic RDD model can be expressed as: y i,t+τ = β τ PASS τ n i,t + [P l ( v k k n k=1 i,t, γ l,τ )] + [P r ( v k k k=1 i,t, γ r,τ )] + α i,t + α τ + α c + α φ + ε i,t+τ. (1) y i,t+τ is the level of earnings management for firm i at time t + τ. PASS τ i,t is a dummy variable for whether a shareholder proposal is passed. If a proposal s vote shares (v k i,t ) are equal to or greater than the majority threshold (50%), PASS τ i,t is defined as 1, and 0 otherwise. To expand to the full sample from proposals in the close margin of the threshold, we follow the standard approach in Cuñat et al. (2012, 2016) which uses the polynomials of vote shares to approximate the continuous underlying relationship between y i,t+τ and v k i,t, allowing for a discontinuous jump at the majority threshold v. P l ( n k=1 v i,t shares for observations on the left-hand side of the threshold and P r ( k k, γ l,τ ) is the polynomial in vote n k=1 v i,t k k, γ r,τ ) is the 5 RDD was first introduced by Thistlethwaite and Campbell (1960) to analyze the impact of merit awards on future academic outcomes. The gist of RDD was that individuals with scores just below the cutoff (who did not receive the award) were good comparisons to those just above the cutoff (who did receive the award). 12

14 polynomial in vote shares for observations on the equation s right-hand side. Throughout this paper, we use second-order polynomials to test the effect. 6 We employ a panel data set to estimate this regression. For each firm-meeting ( i, t ), observations at time t + τ are pooled for multiple τ, including τ < 0. Specifically, we use data from the years relative to the meeting -2 through +4. For τ < 0, the coefficient of the dummy variable PASS k i,t and the parameters of polynomials γ k l,τ k and γ r,τ are constrained to zero; they are allowed to vary for τ > 0. α i,t is the firm-meeting fixed effects to capture any unobservable firm characteristics. Following Cellini et al. (2010), we further include fixed effects for the time period relative to the meeting α τ ( distance-to-the-election fixed effects), industry fixed effects α c, firm-meeting fixed effects and fiscal year fixed effects α φ. Standard errors are clustered at the firm-year level and second-order polynomials are used throughout. Assume the outcome variable y i,t+τ is the earnings management of firm i in year t + τ, a positive coefficient, β τ, implies that passing a proposal leads to an increase in earnings management in year t + τ. Following Cuñat et al. (2012, 2016), we also examine the average effect over four years on earnings management after a proposal passes. To do so, we set τ = 0 and omit distance-toelection fixed effects α τ, as follows: n y i,t+τ = β 0 PASS i,t + [P l ( v k k n k=1 i,t, γ l,τ )] + [P r ( v k k k=1 i,t, γ r,τ )] + α i,t + α c + α φ + ε i,t+τ. (2) A positive β 0 thus implies that passing a shareholder proposal leads to an average increase in earnings management in the four years after the annual meeting. 6 The results are qualitatively similar if we use first-, third-, or fourth-order polynomials. 13

15 3.2 Measures of Accrual-Based Earnings Management Following Kothari, Leone, and Wasley (2005), we use discretionary accruals to capture accrual-based earnings management. Discretionary accruals are the difference between the normal level of accruals and firms actual accruals level. We estimate the latter using the modified Jones (1991) model and the performance adjusted modified Jones model. Performance is adjusted by ROA in the last or current year. We estimate the following model for each industry-year, where the industry is defined as the first two digits of the SIC code, with at least 15 observations: ACCR i,t = α 0 + ACCR i,t = α 0 + ACCR i,t = α 0 + β 0 ASSETS i,t 1 + β 1 ( SALES i,t AR i,t ) + β 2 PPE i,t + ε i,t. (3) β 0 ASSETS i,t 1 + β 1 ( SALES i,t AR i,t ) + β 2 PPE i,t + β 3 ROA i,t 1 + ε i,t. (4) β 0 ASSETS i,t 1 + β 1 ( SALES i,t AR i,t ) + β 2 PPE i,t + β 3 ROA i,t + ε i,t. (5) ACCR i,t is total accruals, measured as the change in non-cash current assets minus the change in current non-interest-bearing liabilities, minus depreciation and amortization expenses for firm i in year t, scaled by lagged total assets; SALES i,t is the annual change in sales scaled by lagged total assets; AR i,t is the annual change in accounts receivable scaled by lagged total assets; PPE i,t is property, plant, and equipment for firm i in year t, scaled by lagged total assets; ROA i,t is the return on assets for firm i in year t; and ROA i,t 1 is the return on assets for firm i in year t 1, The estimated residuals are discretionary accruals, which proxy for accrual-based earnings management. 3.3 Measures of Real Earnings Management We construct real earnings management measures following the prior literature (Roychowdhury 2006; Cohen et al., 2008; Cohen and Zarowin, 2010). We follow 14

16 Roychowdhury (2006) in calculating three individual real earnings management measures (AB_CFO, AB_EXP, and AB_PROD). AB_CFO is the abnormal discretionary operating cash flow, multiplied by -1. AB_EXP is abnormal discretionary expenses, multiplied by -1. AB_PROD is the abnormal discretionary production cost. We then follow Cohen and Zarowin (2010) in constructing two aggregate real earnings management measures, REM1 and REM2. REM1 is the sum of AB_CFO and AB_EXP, and REM2 is the sum of AB_EXP and AB_PROD. Finally, following Cohen et al. (2008), we compute REM_PROXY as the sum of AB_CFO, AB_EXP, and AB_PROD. For each measure, a higher value indicates a greater likelihood of engaging in real earnings management. The estimation of AB_CFO follows the equation below: CFO i,t = α 0 + β 0 ASSETS i,t 1 + β 1 SALES i,t + β 2 SALES i,t + ε i,t, (6) where CFO i,t is the cash flow from operations scaled by lagged total assets; SALES i,t is annual sales scaled by lagged total assets for firm i in year t, and SALES i,t is the annual change in sales scaled by lagged total assets for firm i in year t,. For each firm-year, AB_CFO is the residual of regression (7), multiplied by -1. The estimation of AB_EXP follows the equation below: EXP i,t = α 0 + β 0 ASSETS i,t 1 + β 1 SALES i,t 1 + ε i,t, (7) where EXP i,t is the sum of advertising expenses, research, and development expenses and selling, and general and administrative expenses scaled by lagged total assets for firm i in year t; SALES i,t 1 is sales in the last year scaled by lagged total assets for firm i in year t. AB_EXP is the residual of regression (8), multiplied by -1. The estimation of AB_PROD follows the equation below: 15

17 PROD i,t = α 0 + β 0 ASSETS i,t 1 + β 1 SALES i,t + β 2 SALES i,t + β 3 SALES i,t 1 + ε i,t, (8) where PROD i,t is the sum of the cost of goods and the change in inventory for firm i in year t, scaled by lagged total assets; SALES i,t is annual sales scaled by lagged total assets for firm i in year t; SALES i,t is the annual change in sales scaled by lagged total assets for firm i in year t; and SALES i,t 1 is the change in sales in the last year scaled by the lagged total assets for firm i in year t. AB_PROD is the residual of regression (9). We estimate the above three models for each industry (defined using two-digit SIC codes) and year with at least 15 observations. 3.4 Sample Selection We obtained shareholder proposals from ISS (formerly Riskmetrics). The data set includes all S&P 1500 companies and 500 widely held firms. We start from a sample of 6,038 shareholder proposals from 1997 to After requiring firm-level financial variables to be available in Compustat, we have 5,837 proposals. Following Roychowdhury (2006), we exclude firms in regulated industries (SIC codes between ) and financial institutions (SIC codes between ). Following some prior literature (e.g., Fresard, 2010; and Almeida, Kim, and Kim, 2015), we also delete firms with negative book equity values because these are unusual values (possibly indicating firms in unusual conditions). In addition, we control for firm growth and investment opportunities using the book-to-market of equity. Following Kothari et al. (2005), we exclude observations if the absolute value of total accruals scaled by total assets exceeds one. This helps to mitigate the effects of outliers. We also impose a requirement of at least 15 observations for each industry-year grouping as a reasonable number of observations are needed to estimate our regression-based measures of earnings management. Finally, we are left with 2,258 shareholder proposals, or 1,420 firm-year observations from 1997 to

18 There are 85 types of shareholder proposals in the RiskMetrics classification. Because of the limited number of observations in each classification, we reclassify these proposals into 3 groups: governance structure (N=1,447), financial reporting proposals (N=46), and other proposals that cannot be classified into either of the above two categories (N=765). Governance structure proposals include two types: board structure proposals (N=323), which aim to separate the chairman and the CEO and improve board independence; G-index proposals (N=1,097), which seek to promote shareholder rights through changing governance provisions, as illustrated in Gompers, Ishii and Metrick (2003). Financial reporting proposals include two types of proposals: audit proposals that aim to limit non-audit fees or services and clawback proposals that intend to implement clawback provisions. 4. Empirical Results 4.1 Summary Statistics As mentioned in the research design, we employ a panel data set including observations in the [t-2, t+4] window around the meeting year. Table 2 reports the summary statistics. The magnitudes of our accrual-based and real earnings management measures are largely similar to those in the previous literature (Cohen et al., 2008; Zang, 2012). 4.2 RDD Validity The identification assumption of RDD is that there is no discontinuity in vote shares around the threshold (McCrary, 2008) and vote shares near the threshold cannot be predicted (Lee and Lemieux, 2010). In this section, we examine the validity of RDD by plotting the vote share distribution. If the density around the threshold is discontinuous, passing or failing a proposal would be endogenous and violate the RDD s fundamental assumption. 17

19 Figure 2 plots the discontinuity of the accrual earnings management measure (the modified Jones model with lagged ROA in Panel A) and the real earnings management measure (REM_PROXY in Panel B) at the 50% vote share threshold, one year after an annual meeting. We observe that the distribution of vote shares is smooth around the threshold, while the average level of accrual-based and real earnings management for proposals that marginally pass is substantially higher than those marginally fail. Another assumption of RDD is that before the annual meeting, there are no systematic differences in the characteristics of firms that marginally pass proposals and those with proposals that marginally fail. Table 3 reports the probit model regressions of the passing dummy and our measures of accrual-based and real earnings management. Each row corresponds to a different dependent variable and each entry comes from a separate regression. Each entry in the table reports the coefficient on whether a proposal passed. Models (1) and (2) ((3) and (4)) report the estimated effect of passing a vote on the outcome variable levels (changes) the year before the annual meeting, T -1 (between T -2 and T -1). Models (1) and (3) report estimates without controlling for a polynomial in the vote share and therefore estimate the average effect of passing relative to not passing. Models (2) and (4) include the polynomial in the vote share of order four on each side of the threshold such that it effectively estimates the effect at the discontinuity. All models control for year fixed effects; standard errors are clustered at the firm level. We observe that there is no significant difference in all the accrual-based and real earnings management measures between the passed and unpassed groups one or two years before the meeting. These results support the validity of using RDD in our setting. 18

20 4.3 Implementation of Shareholder Proposals To support our conjecture that shareholder-initiated governance proposals general real consequences and affect firms earnings management policies, we analyze the implementation probability of passed proposals versus failed proposals. Table 4 reports the results. In Column (2), we employ the RDD approach and find that the passage of the shareholder-initiated governance proposal is positively associated with the likelihood of implementing the proposal. The passed proposals have significantly higher likelihood of being implemented. 4.4 Shareholder Proposals and Accrual-Based Earnings Management Table 5 reports the results on the effect of passing a proposal on accrual-based earnings management. The dependent variables in Panels A, B, and C are DA, DA_LROA, and DA_ROA, respectively. DA is the discretionary accrual estimated using the modified Jones model; DA_LROA is the discretionary accrual estimated using the modified Jones model that controls for lagged ROA; DA_ROA is the discretionary accrual estimated using the modified Jones model that controls for ROA. Following the earnings management literature (e.g., Healy and Wahlen, 1999; Fields, Lys, and Vincent, 2001; Cohen et al., 2008; Zang, 2012; Chan et al., 2015; among others), we include the following control variables: BTM (Book-to-market ratio), SIZE (natural log of total assets), LEV (leverage ratio), and ROA (return on asset). For each panel, we employ two models. In Model (1), we use a single PASS dummy that measures the average post-meeting effect of passing a proposal in the four years after the meeting. In Model (2), we include five dummy variables: PASS_T, Pass_T+1, PASS_ T+2, Pass_ T+3, and PASS_ T+4. Each variable equals 1 if a firm passes a proposal in the meeting year and the observation occurs during T, T+1, T+2, T+3, or T+4 years after the meeting, 19

21 respectively, zero otherwise. We follow Cuñat et al. (2012) and include second-order polynomials in the vote shares on each side of the threshold. Model (1) controls for industry fixed effects, fiscal year fixed effects, and firm-meeting fixed effects. Standard errors are clustered at the firm-year level. In addition, Model (2) of each panel controls for distance-to-theelection fixed effects. In Model (1), we find that the coefficient on PASS is negative (-0.008) and significant (p=0.034). This suggests that the passing of a proposal is associated with a decrease in discretionary accruals in the four years after the annual meeting. In Model (2), we find that the coefficients on PASS_T, PASS_T+1, and PASS_T+2 are negative and significant, suggesting the decrease in discretionary accruals after the meeting mainly occurs during the meeting year or in the first and second years after it. Moreover, the decrease in discretionary accruals is most pronounced for the meeting year: the passing of a proposal is associated with the largest decrease in discretionary accruals in the first year. Similarly, in columns (3) to (6), our main results hold for the two alternative measures of discretionary accruals. Passing a shareholder proposal is associated with decreased accrual-based earnings management. Across all models, the coefficients of ROA are positive and significant, which is consistent with the results in Kothari et al. (2005) and Ali and Zhang (2015). To summarize, we find that passing a shareholder proposal leads to a significant decrease in discretionary accruals in the two years after the annual meeting and that the decrease is most pronouced for the year of the annual meeting. 4.5 Shareholder Proposals and Real Earnings Management Next, we examine the relation between the passage of shareholder-initiated governance proposals and real earnings management. The results are reported in Table 6. For columns (1) to 20

22 (6), the dependent variables are AB_CFO, AB_EXP, AB_PROD, REM1, REM2, and REM_PROXY, respectively. Similar to Table 5, we employ two models for each panel. In Model (1), we use a single PASS dummy that measures the average post-meeting effect of passing a proposal in the four years after the meeting. In Model (2), we include five dummy variables: PASS_T, PASS_T+1, PASS_ T+2, PASS_T+3, and PASS_T+4. Each variable equals 1 if the proposal passed in the meeting year and the observation is in T, T+1, T+2, T+3, or T+4 years after the meeting, respectively, zero otherwise. These models follow Cuñat et al. (2012), which includes second-order polynomials in the vote shares on each side of the threshold. Model (1) controls for industry fixed effects, fiscal year fixed effects, and firm-meeting fixed effects. Standard errors are clustered at the firm-year level. In addition, Model (2) in each panel controls for distance-to-the-election fixed effects. In Table 6, we show that the coefficients on PASS are negative but not very significant in Models (1) and (2), suggesting that a proposal s passing is not significantly related to the increase in abnormal cash flow. The coefficients on PASS and PASS_T through PASS_T+4, in Models (3) and (4) respectively, are negative and significant at the 5-10% level, which suggests that passing a shareholder proposal is associated with an increase in abnormal expenses. In particular, the increase in abnormal expenses is present for each year from the year of the meeting through the fourth year after it. Further, we find that the coefficients on PASS and PASS_T through PASS_T+4, in Models (5) and (6) respectively, are in general negative and significant, suggesting that passing a proposal is associated with a decrease in abnormal production costs. In columns (7) to (12), where the dependent variables are the sum of the individual real earnings management measures, we find similar but stronger results: real earnings management 21

23 decreases significantly after a passing proposal. In particular, Model (10) shows that passing a proposal is significantly associated with a decrease in real earnings management passing from the meeting year up to four years after the meeting. In terms of the control variables, BTM is generally positively associated with real earnings management. BTM controls for firms growth opportunity. Growth firms are less likely to take real economic actions to manipulate earnings because these real actions can damage firms competitive ability and growth potential in the long term, consistent with Cheng, Lee, and Shevlin (2016). SIZE is positively associated with real earnings management: larger firms are more capable of manipulating cash flow, expenses, and production costs. LEV is positively associated with real earnings management: firms with higher leverage have more of a need to manipulate earnings through real activities. The signs on these coefficients are all consistent with Cheng et al. (2016). To summarize, we find that passing a shareholder proposal is associated with a significant decrease in real earnings management in the four years after the meeting. 4.6 Alternative Regression Specifications In this subsection, we examine our main result using alternative specifications of regression models. We focus on the subsample, where vote percentage is within 10% around the majority passing threshold. Table 7 reports the results. In Panel A, we use a simple difference-indifference approach to estimate the effect of passing a proposal on accrual-based and real earnings management. In Panel B, we use the regression discontinuity method again, but within 10% around the passing threshold. Overall, we find that our main result is robust to alternative specifications: the passing of governance proposals is associated with lower accrual-based and real earnings management. 22

24 4.7 Shareholder Proposals and Audit Fees To complement our main result on earnings management, we further examine the effect of passing a shareholder-initiated governance proposal on. To the extent that the passage of governance proposals improves corporate governance and helps reduce opportunistic accounting behavior, we should expect firms spend more in engaging auditors to monitor the financial statements. Table 8 reports the results on audit fees. We find that, consistent with our argument, audit fees significantly increase after the passage of governance proposals. 4.8 Heterogeneous Effects of Governance Proposals on Accrual-based and Real Earnings Management In this subsection, we examine the effects of different types of proposals on earnings management. Shareholder proposals can be categorized into three main types: governance structure proposals, compensation proposals, and financial reporting proposals. Governance structure proposals are likely to improve the general corporate governance of the firm (e.g. improve board independence and diversity, separate CEO and Chairman, etc). To the extent that our main result illustrates the improvement of corporate governance reduces earnings management, we should expect our main result is prevalent in the subsample of governance structure proposals, where shareholders try to implement changes that align the incentives of managers and shareholders. Financial reporting proposals include proposals that intend to improve auditor independence and implement clawback provisions. These proposals are likely to impose severe constraint on managers discretion over accruals management. In particular, audit proposals are about limiting non-audit fees or services. The economic bond created between auditors and clients through the 23

25 provision of non-audit services can impair auditor independence, thus compromising audit quality (Tepalagul and Lin, 2015). Hence, limiting non-audit fees or services would constrain managers from manipulating earnings via the accrual-based management method. With enhanced monitoring by auditors, managers may switch to real earnings management to achieve their earnings targets. Clawback proposals are intended to recoup or claw back bonuses if a misstatement of financial reports occurs, thus making it costly for managers to manipulate accruals. Recent literature (Cohen et al., 2008; Cohen and Zarowin, 2010; Gunny, 2010; Kothari, Mizik, and Roychowdhury, 2016; Zang, 2012; Ge and Kim, 2014; Chan et al., 2015) argues that managers may engage in real earnings management when accrual-based earnings manipulation is restricted. Hence, we should expect such proposals passing to result in a substantial substitution from accrual-based earnings management to real earnings management. In other words, for both audit and clawback proposals, accrual-based earnings management should still decrease but real earnings management is likely to increase. Table 9 reports the effect of passing a proposal on accrual-based and real earnings management for the subsample of governance structure and financial reporting proposals. In all models, the dependent variables are DA, DA_LROA, DA_ROA, REM1, REM2 and REM_PROXY, respectively. Panel A reports the results on governance structure proposals. We find that the coefficients on PASS are consistently negative across all models, suggesting that our main result is prevalent among governance-structure proposals. Panel B reports the results for financial reporting proposals, i.e. audit and clawback proposals. We find that coefficients on PASS are negative for accrual-based earnings management yet the coefficients on PASS are positive for real earnings management. These results suggest that when shareholders target the 24

26 management s financial reporting incentives, similar to those in Zang (2012), there appears to be a potential transfer between accrual-based and real earnings management. 5. Conclusion Shareholder activism is an important source of corporate governance and earnings management is a serious agency problem. In this paper, we analyze the effect of shareholder proposals on both accrual-based and real earnings management. The results show that after passing proposals, both accrual-based and real earnings management decrease. As a result, we also find that that investors trust firms reported earnings more, reflected in an increase in ERC. Firms also spend more in audit fees, suggesting the strengthening of efforts to ensure the financial reporting quality. Finally, we also find some heterogeneous effects of different types of governance proposals. We find that governance structure proposals reduce both accrual-based and real earnings management. Proposals that focus on improving firms governance structure reduce both accrual-based and real earnings management. In contrast, proposals that focus mostly on improving firms financial reporting can lead to unintended consequences. Our results show that after passing a proposal that seeks to improve auditor independence or approve a clawback provision, firms tend to switch from accrual-based earnings management to real earnings management. Further analyses indicate that this substitution effect varies with the pre-proposal levels of auditor entrenchment. We contribute to the literature by estimating a causal effect of passing proposals on both accrual-based and real earnings management. The use of dynamic RDD arguably rules out any unobserved firm characteristics that jointly determine the shareholder proposal outcome and the 25