The Effect of Chief Operating Officers on Real Earnings Management

Transcription

1 The Effect of Chief Operating Officers on Real Earnings Management Abstract: Because Chief Operating Officers (COOs) are responsible for internal operations and because the use of real earning management (REM) can have negative consequences on longterm operating performance, we posit that firms with COOs will be less likely to use REM to inflate near-term earnings. Consistent with this, we find that the level of REM is lower for firms employing COOs than for other firms. In subsequent tests, we investigate whether the use of REM varies with the likelihood that COOs are heir apparent to the Chief Executive Officer (CEO) position and find that heir apparent COOs engage in more REM than do non-heir apparent COOs. Overall, our results suggest that the presence of a COO restricts the use of REM on average, and that this effect is driven by COOs who are less likely to ascend to the CEO role in the near term.

2 I. Introduction Most publicly traded firms employ a Chief Executive Officer (CEO) and a Chief Financial Officer (CFO). In addition, many publicly traded firms (approximately 36 percent in our sample) also employ a Chief Operating Officer (COO). While a large body of research investigates the roles of the CEO and CFO, relatively few studies address the role of the COO. According to EY (2014, 17), the COO controls the allocation and prioritization of corporate resources and assets in order to achieve the strategic goals. Once a decision has been made, he or she must ensure that the right resources are in the right places. 1 Consistent with this, Hambrick and Cannella (2004) find that COOs are generally responsible for internal operations, allowing CEOs to focus on overall firm strategy. Because COOs are heavily involved in internal operations, we investigate whether COO firms differ from non-coo firms in terms of their propensity to manage earnings through real activities manipulation or real earnings management (REM). 2 There are several reasons why the presence of a COO should reduce the use of REM. First, agency theory suggests that managers consider their reputations when deciding whether to engage in myopic, rent-seeking behaviors (Fama 1980; Francis, Huang, Rajgopal, and Zang 2008). Assuming that COO reputations are related to operating performance, engaging in practices that reduce future operating performance (such as REM) is likely to have a high reputational cost for COOs. Second, COOs should be in a better position than non-coos to identify and quantify the real costs of REM. This is important because the actual costs of REM 1 See The DNA of the COO: Time to Claim the Spotlight, available at 2 Roychowdhury (2006, 337) defines REM as departures from normal operational practices, motivated by managers desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations. 1

3 are not known ex ante (i.e., when decisions about REM are made). 3 Moreover, if COOs identify more potential costs of REM, their estimates of the total costs of REM should be higher than estimates developed by non-coo managers. Finally, the presence of a COO has the potential to increase task-oriented conflict when top management decides whether to engage in REM. 4 This is especially true because, as discussed above, COOs have both the incentive and the ability to raise concerns about the implications of REM during such discussions. Although task-oriented conflict arises from management disagreement, it also increases information exchange and improves decision quality (Amason and Sapienza 1997; Janssen et al. 1999). Because REM decreases long-term firm value (Cohen and Zarowin 2010; Francis, Hasan, and Li 2011; Kim and Park 2014), heightened task-oriented conflict could result in management being less likely to engage in REM. To conduct our analyses, we use a sample of 17,658 firm-year observations from 1992 through 2014, 36 percent of which have a COO (6,372 firm-year observations). We find that COO firms engage in lower levels of REM through abnormal production, abnormal discretionary expenses (i.e., research and development (R&D), selling, general and administrative (SG&A), and advertising expenses), and through abnormal cash flows from operations. In all specifications, we follow Roychowdhury (2006), Zang (2012), Chan, Chen, Chen, and Yu 3 For example, although it is clear ex ante that REM through the overproduction of inventory leads to increased carrying costs (Roychowdhury 2006; Brüggen, Krishnan, and Sedatole 2011), it is difficult to accurately forecast the magnitude of the increase in warehouse space, human capital, obsolescence, and financial resources required in future periods without a strong understanding of firm-wide operations. 4 Task-oriented conflict is a disagreement that arises among managers because of a divergence in opinion about firm-level issues such as the allocation of resources or the development of firm strategy (Janssen, Van De Vliert, and Veenstra 1999). 2

4 (2014), and Cheng, Lee, and Shevlin (2016) and control for factors previously shown to influence REM. In all specifications, we also include industry and year fixed effects. 5, 6 In subsequent tests, we investigate whether the extent to which COO firms engage in REM varies with the career prospects of the current COO. Specifically, we posit that the preference of COOs to avoid REM could be weaker if the COO expects to ascend to the CEO role in the near term. This is due to the relatively short period over which the operating performance of heir apparent COOs is likely to be evaluated. To proxy for COOs that expect to ascend to the CEO role in the near term, we use the following two empirical proxies: 1) an ex ante measure, computed as COO compensation relative to CEO compensation (Bebchuk, Cremers, and Peyer (2011) and Feng, Ge, Luo, and Shevlin (2011)); 7 and 2) an ex post measure that indicates whether the COO is promoted to CEO in the following three years. The results are consistent with our expectations. Specifically, among firms that employ a COO, REM increases with the ratio of COO compensation to CEO compensation. Similarly, REM is higher among firms with COOs that ascend to CEO within three years. Thus, we conclude that career prospects have a significant effect on the incentives and behavior of COOs. This paper contributes to the literature by providing evidence on the effects of COOs on firms REM behavior. This is important because REM has important long-term implications for firm performance and because the COO position is previously unexplored. Thus, our findings 5 Our inferences do not change when we restrict our sample to suspect firms, defined as those where earnings surprise (i.e., actual earnings minus the consensus analyst forecast prior to the earnings announcement) is between zero and one percent of stock price (Cheng et al. 2016). Additionally, our inferences do not change when we use performance-adjusted measures of REM following Cohen, Pandit, Wasley, and Zach (2015) or when we use performance-adjusted measures and restrict the sample to suspect firms. 6 Ewert and Wagenhofer (2005) and Zang (2012) argue that when the cost of one form of earnings management increases, firms will substitute that form with another form of earnings management. Thus, we also test whether COO firms substitute accruals-based earnings management (AM) for REM. However, we find no evidence that COO firms engage in higher levels of AM than do non-coo firms. 7 Thus, we expect that more powerful COOs (i.e., those that are paid well relative to the CEO) are more likely to ascend to the CEO role. 3

5 should be of interest to a number of potential stakeholders including regulators, investors, auditors, and researchers. In addition, our findings should be of interest to boards of directors who are charged with making decisions about CEO succession plans and top management composition. Given the considerable influence of COOs on firm decisions and outcomes, we hope that our study will motivate other researchers to look beyond the CEO and CFO in subsequent C-suite research. The remainder of the paper is organized as follows: Section II discusses prior literature and develops our hypotheses. Section III presents our research design. Section IV describes our sample selection and data. Section V discusses our results. Section VI provides a summary and conclusion. II. Prior Literature and Development of Hypotheses Prior Research on the COO Position Although the COO position has received little attention in the accounting literature, management research examines the COO role and its effect on firm performance. Hambrick and Cannella (2004) formally define the responsibilities of the COO and distinguish between the roles of the COO and the CEO. They explain that the COO is responsible for internal operations (e.g., communicating and implementing strategy, allocating resources, etc.), while the CEO focuses on external relationships and decisions. Evidence on the effect of COOs on firm performance is mixed. Hambrick and Cannella (2004) find that COO firms experience poorer performance (i.e., they generate a lower return on assets (ROA) and have lower market-to-book ratios) than non-coo firms. In contrast, Marcel (2009) finds that COO firms experience higher ROA and market-to-book ratios when there is 4

6 less functional diversity between the CEO and COO. Taken together, these findings indicate that the impact of having a COO on firm decisions and on firm outcomes can vary. Real Earnings Management Extant research finds that firms engage in earnings management in order to meet or just beat earnings benchmarks. Earnings management often takes one of two forms. Accruals-based earnings management (AM) occurs when management uses discretion in their choice of accounting methods or valuations; this form typically takes place at the end of the reporting period (Jones 1991; Kothari, Leone, and Wasley 2005). Real earnings management (REM) occurs when a firm diverges from its normal business practices during the year in order to artificially bolster current period earnings (Roychowdhury 2006; Zang 2012). Roychowdhury (2006) identifies several methods by which firms engage in REM, including overproducing inventory in order to reduce the cost of goods sold, reducing discretionary expenses, and offering discounts or lenient credit terms in order to increase current period revenues. Prior research suggests that REM is associated with adverse outcomes. For example, in the analytical model in Ewert and Wagenhofer (2005, 1112) firm value strictly decreases as REM increases. In addition, Cohen and Zarowin (2010) find that firms that engage in REM before a seasoned equity offering (SEO) experience lower levels of post-seo operating performance than do SEO firms not engaging in REM. This latter result is particularly important since COO reputations should be affected by operating performance. The Effect of COOs on REM We expect that COOs prefer to avoid engaging in REM for several reasons. First, prior research suggests that managers consider potential reputational costs when making decisions (Fama 1980; Francis et al. 2008). Because COOs are typically evaluated on their ability to 5

7 manage operations (Zhang 2006), COO reputations should be influenced by operating performance. This, coupled with evidence that REM is detrimental to future operating performance (as discussed above), suggests that COOs may prefer to avoid REM in order to protect and build their reputations as effective managers. Second, due to the nature of their responsibilities, COOs should be in a better position than other managers to identify and quantify the long-term costs of REM, which may dissuade COO firms from engaging in REM. Specifically, if COOs identify and quantify more potential costs of REM, then the hurdle rate for (i.e., required benefit of) potential REM activities will increase, and fewer of these activities will be implemented. Consistent with cost considerations playing a significant role in REM decision making, Graham, Harvey, and Rajgopal (2005) report that 55.3 percent of CFO survey respondents stated that they would use REM to meet an earnings target, but only if the resulting sacrifice in firm value is small. Finally, because COOs can quantify the potential implications of REM and face incentives to avoid REM, we expect them to be more likely to voice concerns when such activities are proposed by other managers. Thus, the presence of a COO should increase the level of task-oriented conflict among top management (Janssen et al. 1999), leading to improved decision making (Amason 1996). This improvement occurs because task-oriented conflict allows senior management to voice concerns and provide alternative perspectives, which leads to more information exchange before decisions are made (Amason 1996; Amason and Sapienza 1997; Marcel 2009). Since REM is detrimental to future operating performance and firm value, heightened task-oriented conflict should result in less REM. Collectively, the preceding discussion forms the basis for our first hypothesis (stated in the alternative form): 6

8 H1: Firms that employ a COO engage in less REM than do firms that do not employ a COO. Heir Apparent COOs and REM Some organizations use the COO position as a training ground for internal successors to the CEO position (Hambrick and Cannella 2004; Zhang 2006). Consistent with this, a recent EY survey reports that only 14 percent of COOs C-Suite peers expect them to hold the same role for at least five years, and that only 30 percent of COOs aspire to hold their positions for at least five years, with a majority aspiring to more powerful positions. 8 Because internal promotions are more likely following favorable performance (Zhang and Rajagopalan 2004), heir apparent COOs may be more willing to engage in REM in order to improve short-term firm performance. It may also be the case that COOs who are nearing promotion to CEO are more likely to allow the CEO to engage in REM in order to avoid friction with the CEO, since the current CEO may have input into his eventual successor. It is important to note, however, that not all COOs aspire to CEO positions. 9 While we expect that COO reputations (both internal and external) are influenced by operating performance, and that the resulting incentives should encourage COOs to avoid REM, we also expect that the incentives faced by COOs change as their career prospects within the firm change. Specifically, we expect that incentives related to future operating performance decrease and that incentives related to short-term performance increase as the likelihood of COO 8 See The DNA of the COO: Time to Claim the Spotlight, available at 9 For example, when asked, Does the COO inevitably want the CEO role? John Thompson, Vice Chairman of Heidrick and Struggles, replied, Potentially. There are some people that I think are natural second bananas, to use a slang term, and they don t have any real interest in being number one. (See page 101 of Riding Shotgun: The Role of the COO, available at 7

9 promotion to CEO increases. Thus, our second hypothesis (stated in the alternative form) is as follows: H2: Firms with heir apparent COOs engage in more REM than do firms without heir apparent COOs. III. Research Design Measuring Real Earnings Management (REM) Following Roychowdhury (2006), we measure REM using abnormal production costs (AbnProd), abnormal discretionary expenses (AbnDisExp), and abnormal cash flow from operating activities (AbnCFO). Abnormal production costs The REM technique that results in abnormal production costs occurs when a firm increases the production of inventory so that the fixed overhead costs allocated to the cost of goods sold decrease in the current period (because the fixed overhead cost related to unsold units remains in inventory), increasing reported earnings. This REM technique is problematic because inventory overproduction leads to increased inventory carrying costs in subsequent periods (Brüggen et al. 2011). Following Roychowdhury (2006), we estimate normal production as a function of sales, change in current year sales, and change in prior year sales for each two-digit SIC code industry-year with at least 15 observations as follows: Prod it / Assets it-1 = δ 1 (1/Assets it-1 ) + δ 2 (Sales it / Assets it-1 ) + δ 3 (ΔSales it / Assets it-1 ) + δ 4 (ΔSales it-1 / Assets it-1 ) + ε it (1) where: Prod = production costs calculated as the sum of cost of goods sold and the change in inventory from the prior year (Compustat: COGS, INVT); Assets = total assets (Compustat: AT); Sales = total net sales (Compustat: SALE); 8

10 ΔSales = the change in total net sales relative to the prior year; and i and t indicate firm and fiscal year, respectively. Abnormal production (AbnProd) is measured as the residual from Equation (1). A larger residual indicates more overproduction, which increases reported earnings in the current period. Abnormal discretionary expenses The REM technique that results in abnormal discretionary expenses involves reducing the amount spent on R&D, SG&A, and advertising in order to increase current period income. Following Roychowdhury (2006), we estimate normal discretionary expenses as a function of lagged sales for each industry-year with at least 15 observations as follows: DisExp it / Assets it-1 = ζ 1 (1/Assets it-1 ) + ζ 2 (Sales it-1 / Assets it-1 ) + ε it (2) where: DisExp = the sum of R&D, advertising, and selling, general, and SG&A expenses (Compustat: XSGA, XAD, XRD); 10 and all other variables are as previously defined. Abnormal discretionary expenses (AbnDisExp) is measured as the residual from Equation (2). A more negative residual indicates lower abnormal discretionary expenses, which increases reported earnings in the current period. Abnormal cash flow from operating activities The REM technique that results in abnormal cash flow from operating activities involves accelerating the timing of sales by offering lenient credit terms and price discounts in order to inflate current year earnings. Following Roychowdhury (2006), we estimate cash flow from operating activities as a function of sales and the change in sales for each industry-year with at least 15 observations as follows: 10 We set advertising and R&D expenses to zero if they are missing, following Roychowdhury (2006). 9

11 CFO it / Assets it-1 = γ 1 (1/Assets it-1 ) + γ 2 (Sales it / Assets it-1 ) + γ 3 (ΔSales it / Assets it-1 ) + ε it (3) where: CFO t = cash flow from operating activities (Compustat: OANCF); and all other variables are as previously defined. Abnormal cash flow from operating activities (AbnCFO) is measured as the residual from Equation (3). A more negative residual indicates more REM via lenient credit terms and price discounts, which increases reported earnings in the current period. We multiply AbnDisExp and AbnCFO by -1 so that larger values indicate more incomeincreasing REM. Following prior literature (e.g., Cohen and Zarowin 2010; Chan et al. 2014; Cheng et al. 2016), we also create two composite measures of REM defined as the sum of AbnProd and AbnDisExp (REM1) and the sum of AbnDisExp and AbnCFO (REM2). 11 The Impact of COOs on REM To examine whether the presence of a COO is associated with REM (H1), we follow the model in Chan et al. (2014) but include CEOChair to control for CEO power as follows: REM it = κ 0 + κ 1 COO it + κ 2 MktShare it-1 + κ 3 CEOChair it + κ 4 Inst it-1 + κ 5 ETR it + κ 6 BigN it + κ 7 AuditTenure it + κ 8 NOA it-1 + κ 9 Cycle it-1 + κ 10 ROA it + κ 11 Size it + κ 12 BtM it + κ 13 Earnings it-1 + κ 14SalesGrowth it + κ 15Restructure it + κ 16Writedown it + κ 17 Suspect it + κ j IndustryFE + κ k YearFE + ε it (4) where REM = one of the five measures of REM: AbnProd, AbnDisExp, AbnCFO, REM1, or REM2; COO = an indicator variable set equal to one if the firm employs a COO, and zero otherwise; MktShare = net sales divided by the sum of net sales in the three-digit SIC code industry, following Zang (2012); 11 We do not include AbnProd and AbnCFO in the same composite measure because Roychowdhury (2006) notes that activities which increase the level of AbnProd can also increase the level of AbnCFO, leading to double counting when these variables are combined. 10

12 CEOChair = an indicator variable set equal to one if the CEO also serves as the chairman of the board, and zero otherwise; Inst = the percentage of outstanding shares owned by institutional owners (from the Thomson Reuters Institutional Holdings database); ETR = total income taxes divided by pre-tax income and constrained between zero and one, following Abernathy, Beyer, and Rapley (2014) (Compustat: TXT, PI); BigN = an indicator variable set equal to one if the firm is audited by a Big N auditor, and zero otherwise; AuditTenure = an indicator variable set equal to one if the length of the auditor-client relationship is above the sample median, and zero otherwise; NOA = an indicator variable set equal to one if net operating assets (NOA), calculated as shareholder s equity minus cash and marketable securities plus total debt divided by lagged total assets, are above the industry-year median, and zero otherwise (Compustat: CEQ, CHE, DLTT, AT); Cycle = the days receivable plus the days sales in inventory, following Dechow (1994) (Compustat: RECT, INVT, SALE, COGS); ROA = income before extraordinary items divided by lagged total assets (Compustat: IB, AT); Size = the natural log of total assets (Compustat: AT); BtM = the book value of equity divided by the market value of equity (Compustat: CEQ, PRCC_F, CSHO); Earnings = earnings before extraordinary items less total accruals and production costs, plus discretionary expenses, following Zang (2012) (Compustat: IB); SalesGrowth = current year net sales less prior year net sales, scaled by prior year net sales (Compustat: SALE); Restructure = an indicator variable set equal to one if a firm takes a restructuring charge, and zero otherwise (Compustat: RCP); Writedown = an indicator variable set equal to one if a firm reports a writedown, and zero otherwise (Compustat: WDP); Suspect = an indicator variable set equal to one if earnings per share minus the most recent mean consensus analyst forecast of earnings per share (relative to the earnings 11

13 announcement date) is between zero and one percent of share price, and zero otherwise, following Cheng et al. (2016); IndustryFE = industry fixed effects, where industries are defined using two-digit SIC codes; and YearFE = year fixed effects. The coefficient of interest in Equation (4) is κ 1 (the coefficient on COO). Consistent with H1, we expect a negative and significant coefficient on COO. Because more powerful CEOs may be more likely to exert influence on earnings management strategy, we control for CEO power (CEOChair). Following Zang (2012), we control for costs associated with REM. These costs should be lower for firms with a higher market share (MktShare) and higher for firms with higher institutional holdings (Inst) and effective tax rates (ETR). Following Zang (2012), we also control for the costs of AM because as AM becomes costlier, firms are more likely to use REM. Here, we control for auditor type (BigN), long auditor tenure (AuditTenure), net operating assets (NOA), and the length of the operating cycle (Cycle). Because firms engaging Big N auditors and firms with long-tenured auditors are less likely to engage in AM (Stice 1991; Francis, Maydew, and Sparks 1999), we expect a positive relation between these controls and REM. We expect a positive relation between beginning-of-the-year NOA and REM because high NOA constrains management s ability to manage earnings using current-period accruals. Because firms with longer operating cycles have more ability to manage earnings using AM (Dechow 1994), we expect a positive relation between Cycle and REM. In addition to these variables, we include return on assets (ROA), total assets (Size), and book-to-market (BtM) to control for differences in firm performance, size, and growth opportunities, respectively. Following Zang (2012), we control for pre-managed earnings (Earnings) because prior literature suggests that the level of REM is related to the magnitude of 12

14 earnings management needed to meet benchmarks (Beatty, Chamberlain, and Magliolo 1995; Hunt, Moyer, and Shevlin 1996). Following Chan et al. (2014), we control for sales growth (SalesGrowth), restructuring charges (Restructure), and asset write-downs (Writedown). Finally, following Cheng et al. (2016), we control for the overall level of earnings management by including Suspect. 12 The Effect of Heir Apparent COOs To test whether firms with heir apparent COOs engage in more REM than do firms without heir apparent COOs (H2), we estimate the following models using the sample of firmyear observations with a COO: REM it = λ 0 + λ 1 Relative it + λ 2 MktShare it-1 + λ 3 CEOChair it + λ 4 Inst it-1 + λ 5 ETR it + λ 6 BigN it + λ 7 AuditTenure it + λ 8 NOA it-1 + λ 9 Cycle it-1 + λ 10 ROA it + λ 11 Size it + λ 12 BtM it + λ 13 Earnings it-1 + λ 14 SalesGrowth it + λ 15 Restructure it + λ 16 Writedown it + λ 17 Suspect it + λ j IndustryFE + λ k YearFE + ε it (5) REM it = β 0 + β 1 Heir it + β 2 MktShare it-1 + β 3 CEOChair it + β 4 Inst it-1 + β 5 ETR it + β 6 BigN it + β 7 AuditTenure it + β 8 NOA it-1 + β 9 Cycle it-1 + β 10 ROA it + β 11 Size it + β 12 BtM it + β 13 Earnings it-1 + β 14 SalesGrowth it + β 15 Restructure it + β 16 Writedown it + β 17 Suspect it + β j IndustryFE + β k YearFE + ε it (6) where: Relative = the COO s total compensation divided by the CEO s total compensation (Execucomp: TDC1); Heir = an indicator variable set equal to one if the current COO becomes CEO within three years, and zero otherwise; and all other variables are as previously defined. The coefficient of interest in Equation (5) is λ 1 (the coefficient on Relative), and the coefficient of interest in Equation (6) is β 1 (the coefficient on Heir). Positive coefficients would indicate that firms with heir apparent COOs engage in more REM than other COO firms, consistent with our 12 In additional analyses, we limit our sample to suspect firms and all inferences are unchanged. 13

15 prediction that COO incentives to avoid REM are weaker when COOs expect to ascend to the CEO role in the near term. 13 IV. Sample Selection and Data We begin with all public firms available from Compustat for years 1992 through We exclude financial institutions (SIC codes ) and firms in regulated industries (SIC codes ) from the sample. To identify COO observations, we use the Execucomp database and include only firm-years with at least five officers with non-missing titles to reduce the likelihood of type II error in our identification of COOs. 14 We code officers as COOs if they hold the title of Chief Operating Officer, President of Operations, Executive Vice President of Operations, or Senior Vice President of Operations. To ensure that officers coded as COOs are actually functioning as dedicated COOs, we do not include officers with additional titles that are unrelated to operations (e.g., President of Marketing). In addition, we do not include officers serving as COOs of operating units or of geographic regions. 15 Because our analyses require data on COO and CEO compensation, we omit 32 observations where the COO previously served as CEO (because COO compensation is likely to be impacted by this atypical career path), and we follow Bebchuk et al. (2011) and Feng et al. (2011) and exclude 5,428 CEO turnover years to remove the effect of partial-year 13 Because Equation (6) uses a forward-looking measure of heir apparent COOs (Heir), the sample period for this test ends in Thus, our estimation of Equation (6) uses a smaller sample. 14 In untabulated analyses, we examine the COO s pay-rank in the top 5 compensated officers for our sample of 6,372 COO observations. The distribution of COO pay rank is as follows: highest paid = 7.12%, second highest paid = 57.85%, third highest paid = 20.21%, fourth highest paid = 9.45%, and fifth highest paid = 5.37%. This suggests that, although some observations classified as not having a COO may in fact have a COO, the sample of incorrectly classified observations is likely to be small. 15 Because we identify COOs using an index search of officer titles provided in Execucomp (Execucomp: TITLEANN), we hand-collected a subsample of 200 random observations and assessed the validity of our classification. We ensured that each observation was classified correctly by comparing the title listed in Execucomp with the title and biography obtained from the corresponding firm-year proxy filing (DEF 14A) obtained from the Security Exchange Commission s EDGAR database and we found no incidences of either type I or type II errors. 14

16 compensation. 16 We exclude 1,227 observations without a unique COO for at least 360 days in the fiscal year to ensure proper identification of our variable of interest. 17 Finally, we exclude 105 observations with missing compensation data necessary to calculate Relative and 12 observations where the COO s total compensation is more than 10 times that of the CEO (to remove observations that are unlikely to reflect relative COO-CEO power). 18 The resulting sample consists of 17,658 observations, 6,372 with a COO and 11,286 without a COO. In order to test H2, we use the sample of 6,372 observations where a COO is present. We exclude 524 additional observations when the variable of interest is Heir because this variable requires forward-looking information, resulting in a sample of 5,848 observations for this analysis. We winsorize all continuous variables at ± 1% to mitigate the effect of extreme observations. Table 1, Panel A presents the percent of sample firms with a COO by year. We find that COOs are more common in the earlier portion of our sample period, followed by a decline. This is consistent with recent reports documenting a decline in the use of COOs. 19 Panel B compares 16 Although compensation is not included in the model used to test H1 (Equation (4)), we exclude these firms to maintain a consistent sample across our tests. Our inferences are unchanged, however, when we do not impose this sample constraint. 17 Execucomp reports officers as of the annual meeting date, so it is possible that the named officers did not serve in that capacity for the entire year. When a COO is not in place for the entire year, s/he may be unable to affect REM in that year. To ensure that all COO firms are appropriately identified, we hand collected the dates on which COOs started and left the position. If a unique COO was not present for at least 360 days, we eliminated the observation. Our inferences are unchanged, however, when we retain observations with a unique COO present for at least 180 days and when we retain observations where non-unique individuals serve in the COO position for 360 or for 180 days. 18 In some firm-years, CEO total compensation is abnormally low. For example, the CEO of Apple Inc. received $1 in total compensation from 2006 through 2009 and the CEO of Fossil Group Inc. received $4,266 in total compensation in For other firm-years, COO total compensation is abnormally high because of a one-time payment. For example, the COO of Biogen Inc. received a one-time stock option grant with a Black-Scholes value of $13.38 million in 1994 but the CEO did not receive any stock options in the year. This results in large values of Relative. Because these arrangements do not provide a good indication of the CEO s relative power, we omit these observations from our analyses. 19 For example, a 2014 article by PwC s advisory practice, Strategy+Business, discusses the decline in COOs among Fortune 500 firms in the U.S., from 48 percent in 2000 to 36 percent in 2014 (see The Decline of the COO, available at Moreover, an article in The Washington Post partially attributes the decline in COOs to the growing importance of CFOs (see The Case of the Disappearing Chief Operating Officer, available at 15

17 firm characteristics across COO and non-coo observations. On average, COO firms have higher market share, have shorter operating cycles, higher institutional ownership, and higher effective tax rates, are more likely to engage a Big N auditor and to have a CEO serve as the chairman of the board of directors, are less likely to incur a restructuring charge or asset writedown, and exhibit lower levels of pre-managed earnings. In addition, we find univariate evidence that COO firms exhibit lower levels of REM through abnormal discretionary expenses (AbnDisExp). V. Empirical Results The Impact of COOs on REM Table 2 presents the results from estimating equation (4) to determine whether the presence of a COO is associated with REM. 20 In Column (1), where the dependent variable is AbnProd, the coefficient on COO is negative and significant (-0.009, p = 0.008), indicating that COO firms engage in lower levels of REM through overproduction. In addition, in Column (2), where the dependent variable is AbnDisExp, the coefficient on COO is negative and significant (-0.010, p = 0.034), indicating that COO firms engage in lower levels of REM through abnormal discretionary expenses. Finally, in Column (3), where the dependent variable is AbnCFO, the coefficient on COO is negative and significant (-0.004, p = 0.066), indicating that COO firms engage in a lower level of REM through price discounts and lenient credit terms. The coefficients on COO are also negative and significant (-0.019, p = 0.011, and , p = 0.007, respectively) when the dependent variables are REM through AbnProd and AbnDisExp (REM1) and REM through AbnCFO and AbnDisExp (REM2). Taken together, our results support H1 and reveal that COO firms engage in less REM through abnormal production, abnormal discretionary expenses, and abnormal cash flows from operations. 20 Recall that we multiply AbnDisExp and AbnCFO by -1 so that for all three measures of REM (i.e., AbnProd, AbnDisExp, and AbnCFO), a positive coefficient represents income-increasing REM. 16

18 Heir Apparent COOs and REM Tables 3 and 4 report the results from the estimations of Equations (5) and (6), respectively, which we use to test H2. In Table 3, the coefficients on Relative are positive and significant in Columns (1) and (2), where AbnProd and AbnDisExp are the dependent variables, respectively (0.009, p = 0.018, and 0.009, p = 0.081). This indicates that firms with heir apparent COOs engage in higher levels of REM, both through abnormal production and through a reduction in discretionary expenses. However, the results in Column (3) reveal no relation between whether the COO is an heir apparent and REM through abnormal cash flow from operations. In Column (4), where the dependent variable is REM1, the coefficient on Relative is positive and significant (0.019, p = 0.029), again suggesting that heir apparent COOs engage in more REM. Finally, in Column (5), we find a positive and significant (0.011, p=0.040) coefficient on Relative when the dependent variable is REM2. In Table 4, we find a positive and significant coefficient on Heir when AbnProd (0.022, p = 0.007) or AbnDisExp (0.029, p = 0.005) is the dependent variable, indicating that firms with heir apparent COOs engage in higher levels of REM relative to firms where the COO does not ascend to the CEO role. Similar to Table 3, we find no relation between Heir and AbnCFO. Finally, in Columns (4) and (5), we find positive and significant associations between Heir and REM1 (0.052, p = 0.002) and between Heir and REM2 (0.017, p = 0.098). Overall, our results indicate that, although firms with COOs engage in lower levels of REM, firms with COOs who are more likely to ascend to the CEO position engage in more REM, presumably to improve current earnings and influence the board of directors to promote a successor from within the firm. The Tradeoff Between REM and AM 17

19 The model in Ewert and Wagenhofer (2005) suggests that when one form of earnings management becomes costly or otherwise unusable, firms will engage in another form of earnings management to meet or just beat earnings benchmarks. Thus, we expect that COO firms will increase AM to compensate for decreased REM. Following prior literature, we use discretionary accruals (the difference between the firm s actual accruals and an estimate of normal accruals) to proxy for AM. We estimate normal accruals using the performance-adjusted modified Jones (1991) model from Badertscher (2011). Specifically, we estimate the following cross-sectional model for industry-years with at least 15 observations: Accruals it / Assets it-1 = η 0 + η 1 (1/Assets it-1 ) + η 2 (ΔSales it / Assets it-1 ) + η 3 (PPE it / Assets it-1 ) + ε it (7) where Accruals = total accruals calculated as earnings before extraordinary items less cash flow from operating activities before extraordinary items and discontinued operations (Compustat: IBC, OANCF, XIDOC); PPE = property, plant, and equipment (Compustat: PPENT); and all other variables are as previously defined. We measure performance-adjusted discretionary accruals as the difference between the estimated residual from Equation (7) for firm i and the median residual of the observations in the same industry-year-roa-quintile as firm i, where firm i is omitted from the calculation of the median following Badertscher (2011). Because prior literature shows that firms will compensate for lower REM by increasing accruals, we estimate the following regression: AM it = ν 0 + ν 1 COO it + ν 2 MktShare it-1 + ν 3 CEOChair it + ν 4 Inst it-1 + ν 5 ETR it + ν 6 BigN it + ν 7 AuditTenure it + ν 8 NOA it-1 + ν 9 Cycle it-1 + ν 10 ROA it + ν 11 Size it + ν 12 BtM it + ν 13 Earnings it-1 + ν 14 SalesGrowth it + ν 15 Restructure it + ν 16 Writedown it + ν 17 Suspect it 18

20 + ν j IndustryFE + ν k YearFE + ε it (8) where AM = accruals management as estimated using Model (7); we estimate Model (8) separately for signed, positive (income increasing), and negative (income decreasing) values of accruals management separately; all other variables are as previously defined. Table 5 presents the results from estimating Equation (8). In Column (1), we find that when the dependent variable is signed abnormal accruals (SignedAbnAcc), the coefficient on COO is not significant. Columns (2) and (3) present results using income-increasing (PosAbnAcc) and income-decreasing (NegAbnAcc) accruals, respectively. Again, we do not find a significant relation between COO and signed accruals. Together, these results indicate that COO firms do not differ from non-coo firms in their use of AM to manage earnings. Analyses Using Suspect Firms We conduct our main analyses on a sample of all firms with available data. However, not all firms have incentives to engage in income-increasing earnings management. For example, if managers expect to beat benchmarks by a large margin, they may avoid earnings management activities entirely or may engage in income-decreasing earnings management to smooth earnings (DeFond and Park 1997; Roychowdhury 2006). Thus, we follow Cheng et al. (2016) and reestimate our REM tests after restricting our sample to suspect firms. Specifically, we limit our sample to firms with positive consensus analyst forecast errors between 0 and 1 percent of stock price, such that these firms meet or just beat earnings expectations. We present the results of these analyses in Tables 6 through 8. Table 6 presents the results from our primary analysis, where we re-estimate Equation (4) using the Suspect sample. Consistent with the results in Table 2, we find that COO firms engage 19

21 in lower levels of REM through abnormal production, abnormal discretionary expenses, abnormal cash flows, and the composite measures (REM1 and REM2). Tables 7 and 8 present the results from our analyses related to heir apparent COOs using the Suspect sample. Here, we find results that are generally consistent with those in Tables 3 and Performance-Adjusted Measures of REM Cohen et al. (2015) argue that traditional measures of REM may result in misspecified tests. In order to address this concern, we create performance-adjusted measures of REM using an approach analogous to that used in Badertscher (2011) to performance-adjust AM. Specifically, we estimate Equations (1) through (3) for each industry-year with at least 15 observations and measure performance-adjusted REM as each firm s residual from this estimation less the corresponding industry-year-roa-quintile median residual. We then estimate Equations (4) through (6) using performance-adjusted REM metrics as the dependent variables. We draw similar inferences from the results, which are presented in Tables 9 through 11. VI. Conclusion In this study, we examine whether the presence of a COO affects the level of real earnings management. Using a sample of 17,658 observations from 1992 through 2014, we find that, on average, COO firms engage in lower levels of REM than do non-coo firms. However, using the sample of COO firms, we also find that heir apparent COOs, measured using both an ex ante measure (i.e., COOs whose compensation is high relative to that of the CEO) and an ex post measure (i.e., COOs who actually ascend to the CEO position within three years), engage in more REM. These findings suggest that COOs are willing to manage earnings through REM when faced with incentives to engage in myopic behavior but on average, COOs could be 21 One exception is that the coefficient on Heir in Table 8 is no longer significant at conventional levels when the dependent variable is REM2 (p = 0.155). 20

22 beneficial because they appear to reduce the extent of real earnings management on average. These findings should be of interest to investors, standard setters, and most notably, boards of directors as they make decisions related to executive succession planning and composition arrangements. They should also be of interest to CEOs as they structure their top management teams. 21

23 References Abernathy, J. L., B. Beyer, and E. T. Rapley Earnings management constraints and classification shifting. Journal of Business Finance & Accounting 41 (5-6): Amason, A. C., Distinguishing the effects of functional and dysfunctional conflict on strategic decision making: Resolving a paradox for top management teams. The Academy of Management Journal 39 (1): , and H. J. Sapienza The effects of top management team size and interaction norms on cognitive and affective conflict. Journal of Management 23 (4): Badertscher, B. A Overvaluation and the choice of alternative earnings management mechanisms. The Accounting Review 86 (5): Beatty, A., S. L. Chamberlain, and J. Magliolo Managing financial reports of commercial banks: The influence of taxes, regulatory capital, and earnings. Journal of Accounting Research 33 (2): Bebchuk, L. A., K. J. M. Cremers, and U. C. Peyer The CEO pay slice. Journal of Financial Economics 102 (1): Brüggen, A., R. Krishnan, and K. L. Sedatole Drivers and consequences of short-term production decisions: Evidence from the auto industry. Contemporary Accounting Research 28 (1): Chan, L. H., K. C. W. Chen, T-Y. Chen, and Y. Yu Substitution between real and accruals-based earnings management after voluntary adoption of compensation clawback provisions. The Accounting Review 90 (1): Cheng, Q., J. Lee, and T. Shevlin Internal governance and real earnings management. The Accounting Review, forthcoming. Cohen, D. A., and P. Zarowin Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics 50 (1): , S. Pandit, C. E. Wasley, and T. Zach Measuring real activity management. Working paper, University of Texas at Dallas, University of Illinois at Chicago, University of Rochester, and The Ohio State University. Dechow, P. M Accounting earnings and cash flows as measures of firm performance: The role of accounting accruals. Journal of Accounting and Economics 18 (1): DeFond, M. L., and C. W. Park Smoothing income in anticipation of future earnings. Journal of Accounting and Economics 23 (2):

24 Ewert, R., and A. Wagenhofer Economic effects of tightening accounting standards to restrict earnings management. The Accounting Review 80 (4): Fama, E. F Agency problems and the theory of the firm. Journal of Political Economy 88 (2): Feng, M., W. Ge, S. Luo, and T. Shevlin Why do CFOs become involved in material accounting manipulations? Journal of Accounting and Economics 51 (1): Francis, B. B., I. Hasan, and L. Li Firms real earnings management and subsequent stock price crash risk. Working paper, Rensselaer Polytechnic Institute. Francis, J., A. H. Huang, S. Rajgopal, and A. Y. Zang CEO reputation and earnings quality. Contemporary Accounting Research 25 (1): Francis, J. R., E. L. Maydew, and H. C. Sparks The role of Big 6 auditors in the credible reporting of accruals. Auditing: A Journal of Practice & Theory 18 (2): Graham, J. R., C. R. Harvey, and S. Rajgopal The economic implications of corporate financial reporting. Journal of Accounting and Economics 40 (1): Hambrick, D. C., and A. A. Cannella CEOs who have COOs: Contingency analysis of an unexplored structural form. Strategic Management Journal 25 (10): Hunt, A., S. E. Moyer, and T. Shevlin Managing interacting accounting measures to meet multiple objectives: A study of LIFO firms. Journal of Accounting and Economics 21 (3): Janssen, O., E. Van De Vliert, and C. Veenstra How task and person conflict shape the role positive interdependence in management teams. Journal of Management 25 (2): Jones, J. J Earnings management during import relief investigations. Journal of Accounting Research 29 (2): Kim, Y., and M. S. Park Real activities manipulation and auditors client-retention decisions. The Accounting Review 89 (1): Kothari, S. P., A. J. Leone, and C. E. Wasley Performance-matched discretionary accrual measures. Journal of Accounting and Economics 39 (1): Marcel, J. J Why top management team characteristics matter when employing a chief operating officer: A strategic contingency perspective. Strategic Management Journal 30 (6):

25 Roychowdhury, S Earnings management through real earnings management. Journal of Accounting and Economics 42 (3): Stice, J. D Using financial and market information to identify pre-engagement factors associated with lawsuits against auditors. The Accounting Review 66 (3): Zang, A. Y Evidence on the trade-off between real activities manipulation and accrualbased earnings management. The Accounting Review 87 (2): Zhang, Y., and N. Rajagopalan When the known devil is better than an unknown god: An empirical study of the antecedents and consequences of relay CEO successions. The Academy of Management Journal 47 (4): The presence of a separate COO/president and its impact on strategic change and CEO dismissal. Strategic Management Journal 27 (3):

26 Table 1 Panel A: Percent of firms with a COO by year Year N COO = 1 COO = 0 % of Obs % % % % % % % % % % % % % % % % % % % % % % % Total 17,658 6,372 11,286 36% 25

27 Panel B: Summary Statistics Total COO = 1 COO = 0 Mean Mean Mean Difference p-value REM REM AbnProd AbnDisExp * AbnCFO MktShare *** CEOChair *** Inst * ETR *** BigN *** AuditTenure NOA *** Cycle *** ROA Size BtM Earnings *** SaleGrowth * Restructure *** Writedown Suspect Number of Observations 17,658 6,372 11,286 Note: p-values are one-tailed for variable of interest, and two-tailed otherwise 26

28 Table 2 The Impact of COOs on REM (1) (2) (3) (4) (5) AbnProd AbnDisExp AbnCFO REM1 REM2 COO *** ** * ** *** (0.008) (0.034) (0.066) (0.011) (0.007) MktShare *** * (0.923) (0.729) (0.000) (0.870) (0.087) CEOChair ** (0.745) (0.889) (0.021) (0.927) (0.436) Inst *** * (0.736) (0.624) (0.000) (0.938) (0.077) ETR *** ** (0.001) (0.951) (0.991) (0.220) (0.023) BigN * (0.564) (0.063) (0.406) (0.120) (0.955) AuditTenure (0.287) (0.292) (0.246) (0.233) (0.192) NOA 0.018*** 0.049*** *** 0.067*** 0.010* (0.000) (0.000) (0.001) (0.000) (0.053) Cycle ** (0.108) (0.025) (0.583) (0.389) (0.182) ROA *** 0.125*** *** *** *** (0.000) (0.000) (0.000) (0.005) (0.000) Size 0.018*** 0.033*** *** 0.051*** 0.011*** (0.000) (0.000) (0.000) (0.000) (0.000) BtM 0.038*** 0.032*** 0.011*** 0.070*** 0.049*** (0.000) (0.000) (0.000) (0.000) (0.000) Earnings *** *** *** *** *** (0.000) (0.000) (0.000) (0.000) (0.000) SalesGrowth ** * (0.530) (0.018) (0.233) (0.051) (0.294) Restructure *** *** (0.764) (0.759) (0.000) (0.980) (0.004) Writedown *** 0.008* *** *** (0.002) (0.087) (0.000) (0.551) (0.000) Suspect *** *** *** *** *** (0.000) (0.000) (0.000) (0.000) (0.000) Constant *** *** *** *** (0.000) (0.000) (0.235) (0.000) (0.000) Year Indicators Yes Yes Yes Yes Yes Industry Indicators Yes Yes Yes Yes Yes Observations 17,658 17,658 17,658 17,658 17,658 Adjusted R-squared Note: Robust p-values are in parentheses below coefficient estimates are one-tailed for variable of interest, and two-tailed otherwise *** p<0.01, ** p<0.05, * p<0.1 t-tests are computed using robust standard errors correcting for firm clusters 27