THE EFFECT OF MANAGERIAL OWNERSHIP
ON CLASSIFICATION SHIFTING: AN
EMPIRICAL STUDY
Master thesis, MSc Accountancy
University of Groningen, Faculty of Economics and Business June 22, 2020
Ferdi van der Spoel * Studentnumber: S3792633 Lissabonstraat 36 9718 AZ Groningen Tel: + 31 (0)620216736 E-mail: f.van.der.spoel.1@student.rug.nl Supervisor Prof. Dr. C.K. Hoi Word count: 10,766
* The author would like to thank Prof. Dr. C. K. Hoi for his guidance in writing this thesis, and helpful comments on previous versions.
THE EFFECT OF MANAGERIAL OWNERSHIP
ON CLASSIFICATION SHIFTING: AN
EMPIRICAL STUDY
ABSTRACT
Previous literature has focused on three techniques of earnings management: accrual-based earnings management, real earnings management, and classification shifting. Classification shifting involves the deliberate misclassification of items within the income statement to manipulate core earnings. This study examines the impact of managerial ownership on classification shifting. We use a sample of 12,669 firm-year observations in 1,630 firms in the U.S. for the period 2003 to 2018. We find evidence of a non-linear relationship between managerial ownership and classification shifting. For lower levels of ownership, managerial ownership is significantly negatively related to classification shifting. On the contrary, for higher levels of ownership, managerial ownership is significantly positively related to classification shifting. Specifically, we found that the turning point is at an ownership level of 10.73%, after which an additional increase in managerial ownership is associated with higher levels of classification shifting. These findings suggest that at lower levels of managerial ownership, the effect of increasing managerial ownership could lead to agreement of interests between the managers and stockholders. Entrenchment exceeds incentive alignment at the found turning point. Overall, these findings suggest that managerial ownership gives rise to both the incentive alignment and entrenchment effects on classification shifting.
JEL Classification M40, M41
Keywords Earnings Management, Classification Shifting, Managerial Ownership
TABLE OF CONTENTS
I. INTRODUCTION ... 4
II. PRIOR LITERATURE AND HYPOTHESIS DEVELOPMENT ... 7
Earnings Management Tools ... 7
Managerial Ownership ... 9
Hypothesis Development ... 13
III. METHODOLOGY ... 15
Sample & Data ... 15
Empirical Method ... 16
Measuring Managerial Ownership ... 18
Test Design ... 19
IV. RESULTS ... 20
Descriptive Statistics ... 20
Regression Results ... 24
Sensitivity Analysis ... 26
V. CONCLUSION AND DISCUSSION ... 30
REFERENCES ... 33
APPENDIX A: VARIABLE DEFINITIONS ... 36
I.
INTRODUCTION
Executives, the board of directors, investors, and analysts are convinced that earnings are the most fundamental performance indicator in the financial statements since earnings provide critical information for investment decisions (DeGeorge, Patel, and Zeckerhauser, 1999). Graham, Harvey, and Rajgopal (2005) reported that the majority of the executives they surveyed prefer smooth earnings and admit to sacrificing long-term economic value to smooth earnings. Earnings smoothing is one form of earnings management. More generally, earnings management is defined as “a condition when managers use judgment in their reporting and the organizing of transactions to adjust financial statements to misinform stakeholders about the underlying economic performance of the firm or to influence contractual results that be contingent on accounting records.” (Healy and Wahen, 1999).
Traditionally, earnings management can be classified into two main categories; accrual-based earnings management and real earnings management. However, McVay (2006) examined an alternative earnings management tool named classification shifting, which refers to the deliberate misclassification of core expenses as special items within the income statement with the purpose of inflating core earnings. As defined by Blitzer, Friedman, and Silverblatt (2002), core earnings refer to the earnings generated from a firm’s main business. Core earnings are valued as one of the leading indicators of a company’s financial performance, as Fairfield, Sweeney, and Yohn (1996) show that the predictive content of earnings increases from the classification scheme. Existing literature shows that core earnings influence investors, particularly less sophisticated investors, judgment and decisions through an unintentional cognitive effect and the emphasis managers place on core earnings (Bradshaw and Sloan, 2002; Elliott, 2006). McVay (2006) found evidence that this earnings management tool is indeed used by managers, and therefore concluded that there are many additional settings of classification shifting that future research could investigate. Specifically, prior literature showed that classification shifting seems to be used to achieve predetermined earnings targets, particularly for managers who are not able to use accrual earnings management (Fan, Barua, and Cready, 2010; McVay, 2006). In conclusion, classification shifting can be used to increase core earnings and provide managers with a relatively low-cost tool to meet analyst forecasts (McVay, 2006). The role of corporate governance instruments in financial reporting is to compliance with the financial accounting regulations and to assure the reliability of financial statements (Bushman and Smith, 2003). Hence, as a result of corporate governance quality, deviations could appear in earnings management across firms. Properly structured corporate governance
mechanisms are expected to reduce the use of earnings management since they come up with management control. Empirical evidence is consistent with this argument. For example, Davidson, Goodwin-Stewart, and Kent (2005) found that most non-executive directors on the board and the audit committee are found to be significantly related with a lower probability of earnings management, as measured by discretionary accruals. These results are in line with Klein (2002). Xie, Davidson, and Dadalt (2003) found that board and audit committee members with corporate or financial backgrounds are associated with firms that have less discretionary accruals. These examples of studies have examined the effect of corporate governance mechanisms on earnings management, using accrual-based or real earnings management. However, studies that examined the effect of corporate governance mechanisms on classification shifting are limited. In this regard, there is evidence that legal institutions and independent boards of directors and independent audit committees constrain classification shifting (Haw, Ho, and Li, 2011; Zalata and Roberts, 2016).
Managerial ownership is also observed as an effective corporate governance mechanism. For example, Morck, Shleifer, and Vishny (1988) found evidence that managerial ownership has a non-linear effect on corporate performance. Managerial ownership could effectively limit managers’ opportunistic behavior through incentive alignment. However, there is theoretically a contradiction in both theories and empirical evidence of the impact of managerial ownership on firm performance (e.g., Morck et al., 1988; McConnell and Servaes, 1990; Short and Keasey, 1999). Increasing levels of managerial ownership could encourage managers to improve the value of the firm since managers carry a part of ownership as a stockholder. On the contrary, the entrenchment theory proposes that higher stock ownership levels of management would provide deeper entrenchment, which results in an expanding span for opportunistic behavior.
To date, there has been no research that specifically studied the competing effects of incentive alignment and entrenchment effects on classification shifting arising from managerial ownership. Accordingly, the objective of this study is to empirically examine whether and how managerial ownership affects the extent of classification shifting.
To observe the relationship between managerial ownership and classification shifting, data is acquired from the Annual Compustat Database and ExecuComp Database. This results in a final sample of 12.669 firm-year observations of U.S. listed firms containing the period 2003 to 2018. We use the approach established by McVay (2006) for measuring classification shifting, and managerial ownership is measured as the stocks owned by the top five executives. We provide evidence that for lower levels of ownership, managerial ownership is significantly
negatively related to classification shifting. On the contrary, for higher levels of ownership, managerial ownership is significantly positively related to classification shifting. Theoretically, this regression suggests that at lower levels of ownership, the incentive alignment effect is dominant, and at higher levels of ownership, the entrenchment effect is dominant. These findings are in line with the studies that examine managerial ownership and corporate performance (e.g., Morck et al., 1988; McConnell and Servaes, 1990; Short and Keasey, 1999). Specifically, we found that the positive relationship exceeds the negative relationship at 10.73%, indicating that at this level, the entrenchment effect exceeds the incentive alignment effect. These findings are in line with Short and Keasey (1999), who found a 15.58% cutoff where a negative relationship exceeds a positive relationship between managerial ownership and corporate performance, and Teshima and Shuto (2008), who found a 13.6% cutoff where a positive relationship exceeds a negative relationship between managerial ownership and earnings management.
This study could be a valuable contribution to the limited existing literature on classification shifting and under which conditions classification shifting occurs. Primary, this paper is the first to research the association between managerial ownership and classification shifting, including both two competing effects of managerial ownership. We indicate that managerial ownership influences classification shifting in non-linear form, as the incentive alignment theory is dominant for lower levels, and the entrenchment theory for higher levels. Consequently, managerial ownership must be precisely taken into account when observing the relationship with classification shifting. Additionally, financial reporting is of significant value to the financial statement users in decision-making. Therefore, investors and other users that are concerned about earnings management could also benefit from the findings for the reason that they supply awareness of managerial ownership’s influence on the quality of the earnings. Besides that, the findings are applicable for countries with an institutional environment identical to that of U.S. listed firms.
The next section provides a review of the prior literature on managerial ownership and classification shifting and develops the hypothesis. The third section elaborates on the sample, data, and methodology applied to test the hypothesis. The fourth section provides the results, including descriptive statistics and sensitivity analysis. Finally, the fifth section provides a conclusion and discussion of the study.
II.
PRIOR LITERATURE AND HYPOTHESIS DEVELOPMENT
This chapter explains the different earnings management tools, in particular, classification shifting. Furthermore, it examined the underlying theories of managerial ownership; incentive alignment theory and entrenchment theory. As last, it develops the hypothesis.
Earnings Management Tools
Earnings management is described as “the purposeful intervention in the external financial reporting process with the intent of obtaining private gains” (Schipper, 1989). Healy and Wahlen (1999) define earnings management as a condition where managers use judgment in their reporting and organization of transactions to adjust financial statements to misinform stakeholders about the firm’s underlying economic performance or to influence contractual results that be contingent on accounting records, such as balance sheets and income statements. As such, earnings management occurs when information asymmetry is present, and managers have the discretion to use judgments to alter financial reporting while stockholders are unable to detect and monitor such activities at a low cost effectively. In a survey of 421 financial executives, Graham et al. (2005) find that eighty percent of the interviewed executives indicate that they would use earnings management tools to meet analyst expectations. Fifty-five percent would even go as far as delaying a new plan if this meant the firm would meet analyst expectations. Conclusively, it is reasonable to state that earnings management is common in the financial world.
Three different practices of earnings management have appeared in prior literature: accrual-based earnings management, real earnings management, and classification shifting. As defined by Braam et al. (2015), accrual-based earnings management purposes to conceal true economic performance by altering accounting approaches or estimations within the generally accepted accounting principles. Zang (2012) gives the following definition of real earnings management: “the purposeful action to alter reported earnings in a particular direction, which is achieved by changing the timing or structuring of an operation, investment, or financial transaction, and which has suboptimal business consequences.” Chi, Lisic, and Pevzner (2011) explain that the main aspect of real earnings management is that it is accomplished by business decisions, whereas accrual-based earnings management is accomplished by accounting decisions.
Classification Shifting
Classification shifting is quite a new phenomenon in literature, as it was introduced by McVay (2006). Before this paper, research was solely performed into accrual-based earnings management and real earnings management. McVay (2006) defines classification shifting and delivers evidence that classification shifting is a widespread earnings management tool amongst U.S. firms. She defines classification shifting as shifting income-decreasing core expenditures such as cost of goods sold and selling, general, and administrative expenditures to special items with the purpose of inflating core earnings, consequently creating a positive association between unexpected core earnings and income-decreasing special items. McVay (2006) was the first to create a regression-based methodology to measure classification shifting and demonstrated that classification shifting occurs inside companies as she discovered that unexpected core earnings are increasing in the year of the special item. This unexpectedly high performance reverses in the subsequent year, and therefore found U.S. evidence of manager’s participation with this measurement approach. From then on, literature has been developed on classification shifting.
To date, researchers have examined both the antecedents and the consequences of classification shifting. Fan et al. (2010) have discovered a positive association between classification shifting and companies that just meet or beat analysts’ predictions. Investors consequently reward firms for meeting expectations, as Alfonso, Cheng, an Pan (2015) found evidence that U.S. investors considerably overvalue firms’ core earnings as when current core earnings are higher, they expect increasing future earnings. While these aforementioned studies focus on U.S. firms, several recent studies have extended the analysis to include firms in East Asia (Haw et al., 2011), in the U.K. (Zalata and Roberts, 2017), and in a global setting that includes firms in up to 40 countries (Behn, Gotti, Hermann, and Kang, 2013)1. Overall, these studies associate by demonstrating that classification shifting is a deliberate choice of a firm’s managers.
Among these studies, a central theme is the relationship between corporate governance mechanisms and classification shifting. Haw et al. (2011) found that corporate governance mechanisms, namely well-functioning legal institutions, mitigate the use of classification shifting by managers in East Asian firms. Zalata and Roberts (2016) found that the overall
1 In addition, Fan et al. (2010) discloses that classification shifting is more dominant in the fourth quarter than in
interim quarters. Abernathy, Beyer, and Rapley (2014) were the first to examine the trade-off decision between all three earnings management techniques. They provided evidence that a firm’s probability of using classification shifting increases with the relative costliness of real- and accrual earnings management, indicating that classification shifting is used as a substitute for both.
quality of board and audit committees reduces classification shifting. Besides that, Behn et al. (2013) found higher financial analyst monitoring diminishes classification shifting, but such association is mainly observed in countries with weak investor protection environments. These findings collectively suggest that external and internal corporate governance mechanisms could constrain the extent of classification shifting. Of these studies, Zalata and Roberts (2016) provided the most proximate evidence that bears on the association between managerial ownership and classification shifting. Nevertheless, they focused on stock ownership of non-executive directors. Thus, we can state that the effect of managerial ownership on classification shifting is still largely unexplored.
Managerial Ownership
The discussion regarding the effect of corporate governance instruments on earnings management should be positioned in the framework of the agency problem emerging from the ownership and control separation, forming interests misalignment among management and stockholders. Separation of ownership and control within firms could ensure that management’s performance is shaped by self-interest that could diminish their objectives of maximizing firm wealth and, therefore, the interests of stockholders or other stakeholders (Fama and Jensen, 1983). Managers can participate in earnings management caused by information asymmetry and separation between ownership and control. In this vein, prior literature on earnings management has mainly established an agency perspective (e.g., Healy, 1985; Dechow and Sloan, 1991). Healy (1985) provided evidence that managers use earnings management to maximize their compensation. Similarly, Dechow and Sloan (1991) found that executives incline to reduce costs on R&D activities in their final years, possibly to increase reported earnings.
According to Agrawal and Mandelker (1987), the agency problem can be resolved once both parties have similar interests. Prior studies have shown that well-functioning corporate governance mechanisms can align interests, and therefore help to control earnings management (e.g., Zalata and Roberts, 2016; Davidson et al., 2005; Xie et al., 2003; Behn et al., 2013; Haw et al., 2011). As described by Bushman and Smith (2003), corporate governance mechanisms’ role in financial reporting is to assure compliance with the financial accounting regulatory and to assure the reliability of financial statements. Hence, effective and properly structured corporate governance mechanisms are expected to reduce the use of earnings management since they provide functional control of management. Managerial ownership is also observed as a fundamental corporate governance mechanism and therefore can play a part in reducing
earnings management. For example, Morck et al. (1988) found that managerial ownership has a positive effect on corporate performance for lower ownership levels. Overall, managerial ownership could be an effective mechanism in limiting managers’ opportunistic behavior.
Incentive Alignment and Entrenchment Theory
Two main explanations of managerial ownership have emerged from prior literature. The first explanation is the incentive alignment theory. The agency theory proposes that there is a divergence between the interests of managers and stockholders. Ownership encourages managers to improve the value of the firm since managers carry a part of the wealth outcomes as stockholder (Jensen and Meckling, 1976). Hence, managerial ownership partially reduces the agency concerns, whereas concerns arise through a separation of ownership and control. The effect of managerial ownership could lead to agreement of interests between the managers and stockholders. Managers who possess firm equity could behave to maximize the firm’s value and stockholders’ value because their personal interests are aligned with stockholders’ interests. Therefore, corporate governance instruments such as managerial ownership could improve the alignment of managers and stockholders’ interests and diminish opportunistic conduct following from interests’ conflict. According to the incentive alignment theory, the effect has more effect as managerial ownership increases, proposing that as managerial ownership grows, the effectivity of incentive alignment increases, and opportunistic managerial behavior reduces consistently. Therefore, according to this alignment of interests, if managerial ownership increases, earnings management decreases. Williamson (1964) hypothesized already in an early study that managers without stock ownership would have different objectives than maximizing the firm’s value, which could lead to firms without managerial ownership to be less effective. Warfield, Wild, and Wild (1995) suggested that managerial ownership levels influence earnings’ informativeness and the extent of accrual earnings management. Their study provides evidence that managerial ownership is positively related to earnings’ explanatory power for returns and negatively related to the extent of accrual-based earnings management. They expound their outcomes as being consistent with managerial ownership behaving as a disciplining instrument.
In contradiction to incentive alignment, there is the entrenchment theory. If the interests of both management and stockholders are not completely aligned, expanding ownership of the management could result in more power to use their discretion in their own interests and objectives without the fear of disciplinary actions. The more shares a manager holds, the less power the other owners of the company have upon a manager’s decisions. The entrenchment
theory bears the aforementioned thought. Morck et al. (1988) claim that when managerial voting power grows, managers could turn out to be entrenched in a firm as the market for corporate control’s disciplinary power declines and that higher managerial ownership would provide further entrenchment, which results in an expanding span for earnings management. For example, Healy (1985) suggested that managers use earnings management to maximize their compensation. He found evidence that managers employ earnings management to improve their total compensation, as a positive association between managerial ownership and earnings manipulation was found, similar to Cheng and Warfield (2005). Therefore, it potentially gives rise to an entrenchment effect. In addition, it is also more expected that managers with higher ownership are more difficult to remove even if corporate performance is low, as Conyon and Florou (2002) found that ownership is negatively related to threat of dismissal. This could represent entrenchment, and thus a greater incentive to manage earnings. In conclusion to the entrenchment theory, managerial entrenchment could lead to more earnings management.
Empirical Findings on Managerial Ownership and Firm Performance
According to the arguments, the impact of managerial ownership could be divergent. Regardless of the valuable perceptions that prior theoretical research gives, the empirical literature contains contradicting findings on the effects of managerial ownership. Several studies have examined the effects of managerial ownership in linear regressions. Morck et al. (1988) re-investigated the relationship between managerial ownership and corporate performance. Similar to Demsetz and Lehn (1985), they found no significant relationship with linear regressions as this regression is incapable of representing both the conflicting incentive alignment and entrenchment effects. Consequently, they estimated a piecewise linear regression of managerial ownership on corporate performance and provided evidence of a non-linear relation with corporate performance2. Morck et al. (1988) suggest that high levels of managerial ownership could result in entrenchment, as external stockholders find it hard to control the activities of managers with high levels of managerial ownership. Therefore, at certain levels of managerial ownership, management uses privileges which reduce the company’s value and additionally have sufficient control to fulfill their own goals without the threat of disciplinary actions. However, management who already is fully entrenched does not obtain any further control from additional increases in ownership. On the other side, management with insignificant ownership does not obtain any significant increase in control from a minor
2 In all mentioned studies, corporate performance is measured as Tobin’s Q, which is the ratio of the market
increase in managerial ownership. These effects propose that increases in managerial ownership attend to increase managerial entrenchment in an intermediate level of ownership only. Overall, the combination of the incentive alignment and entrenchment effects ensures a non-linear relation. Morck et al. (1988) examined this combination with piecewise linear regression practices (allowing the coefficients of the managerial ownership variable to change at certain percentages) to estimate the relation between these variables. They used the board of directors as measurement for managerial ownership and found evidence that in the 0% to 5% range of ownership, the incentive alignment theory is leading. Corporate performance is decreasing with growing managerial ownership up to 20%, indicating that the entrenchment theory is leading at intermediate levels of ownership. After that, corporate performance rises slightly at higher ownership levels, indicating a return of the incentive alignment theory.
Following Morck et al. (1988), other large studies have examined non-linear relationships between managerial ownership and corporate performance as well (McConnell and Servaes, 1990; Short and Keasey, 1999; Hermalin and Weisbach, 1988). McConnell and Servaes (1990) came up with evidence that also assisted a non-linear relationship between managerial ownership and corporate performance. However, although their empirical results are consistent with the theoretical argument that the relation is non-linear, their empirical results are different. Short and Keasey (1999) extended their research to the U.K., where McConnell and Servaes (1990) and Morck et al. (1988) were based on U.S. listed firms. Their results provided similarity, as it showed a non-linear relationship between managerial ownership and corporate performance. However, they also found different turning points than the study of Morck et al. (1988). Hermalin and Weisbach (1987) performed similar regressions for 134 NYSE firms. Their research reveals a similar non-linear relation between managerial ownership and corporate performance by all current and former CEO’s still on the board of directors. They found that the relationship between CEO ownership and corporate performance is positive between 0% and l%, negative between 1% and 5%, positive between 5% and 20%, and negative after 20%. Table 1 presents the different turning points of large studies that found a non-linear relationship.
Morck et al. (1988) found turning points of 5% and 20% as they applied a selection of piecewise points. These turning points provided the most effective regression model due to its smallest sum of squared errors. However, using piecewise linear models of low, intermediate, and high ownership levels could be a potential problem because it allows the coefficients of the managerial ownership variables to change only at these levels of ownership. Besides that, Morck et al. (1988) note that there is no theoretical foundation for the selection of turning points
Table 1. Managerial ownership turning points
Study Incentive Alignment Entrenchment
Morck et al. (1988) 0% - 5% / 20% - 100% 10% - 20%
McConnell and Servaes (1990) 1976 sample 0% – 100% -
McConnell and Servaes (1990) 1986 sample 0% - 25% 25% - 100%
McConnell and Servaes (1990) 1976 samplea
0% - 49.4% 49.4% - 100% McConnell and Servaes (1990) 1986 samplea
0% - 37.6% 37.6% - 100%
Hermalin and Weisbach (1988) 0% - 1% / 5% - 20% 1% - 5% / 20% - 100%
Short and Keasey (1999) 0% - 15.58% / 41.84% - 100% 15.58% - 41.84%
Himmelberg, Hubbard, and Palia (1999) 0% - 29.1% 29.1% - 100%
Teshima and Shuto (2008) 0% - 13.6% / 38.8% - 100% 13.6% - 38.8%
a McConnell and Servaes (1990) performed additional curvilinear regressions instead of piecewise regressions, to
find the exact turning points of managerial ownership.
on the piecewise regression, which means that the relationship’s form is mainly an empirical matter. Therefore, several studies (e.g., McConnell and Servaes, 1990; Short and Keasey, 1999; Teshima and Shuto, 2008) solved this problem using curvilinear functional forms instead of piecewise forms. As shown in Table 1, there are differences between the turning points of each study. Kole (1995) examined the differences in data sources used in several studies and concluded the size of sample firms could account for the established differences in these studies. In brief, prior literature shows both entrenchment and incentive alignment effects on firm performance, but no consensus has emerged on the appropriate level of ownership cutoff when one effect dominates the other effect.
Hypothesis Development
In conclusion, the incentive alignment theory proposes that managerial ownership constrains management’s opportunistic behavior, suggesting a negative relationship between managerial ownership and earnings management. The entrenchment theory suggests that high levels of managerial ownership could become ineffective in aligning managers to make decisions that maximize the value of firms, suggesting a positive relationship with earnings management. In accordance with the prior literature review, Figure 1 presents the predicted relationship between managerial ownership and earnings management theoretically, indicating a combination of both incentive alignment and entrenchment effects. In the low area, there is a negative incentive alignment effect, hence earnings management is reducing as managerial ownership grows. In the intermediate area, there is a positive entrenchment effect, which
Figure 1. The expected relationship between managerial ownership and earnings management
indicates that earnings management is increasing as managerial ownership increases. In the high area, there is a negative incentive alignment effect, and there is no entrenchment effect since managers are already fully entrenched. As a result, earnings management decreases as managerial ownership increases.
According to the discussion, managerial ownership’s effect on earnings management could be divergent. Prior literature mainly focused on linear regressions between ownership and managing earnings. As aforementioned, Warfield et al. (1995) found a negative relation between managerial ownership and the magnitude of discretionary accruals, suggesting the dominance of the incentive alignment effect. Furthermore, Larcker et al. (2007) found that insider power, primarily measured by managerial ownership, is positively associated with discretionary accruals and restatements, which suggests entrenchment effects. In addition, Huang, Wang, and Zhou (2013) studied the role of managerial ownership in the association between stockholder rights and discretionary accruals. They concluded that managers with high levels of ownership diminish the negative relation between stockholder rights and discretionary accruals, which indicates that managers are expected to use ownership against stockholders’
E ar ni ng s m an ag em en t Managerial ownership Entrenchment effect Incentive alignment effect
disciplinary actions and obtain more freedom in managing earnings. These studies examined the effect of managerial ownership with linear regressions.
Nevertheless, Teshima and Shuto (2008) were the first to consider both incentive alignment and entrenchment effects in examining a non-linear relationship between managerial ownership and earnings management. They found a negative association among managerial ownership and discretionary accruals for low and high levels of stock ownership in Japanese companies, suggesting the existence of the incentive alignment theory at these levels. For intermediate levels of ownership, they found a positive relation between managerial ownership and accrual-based earnings management. These results are in line with theoretical implications. More specifically, Teshima and Shuto (2008) found a turning point of 13.6%, where entrenchment effects exceed incentive alignment effects, as shown in Table 1.
Overall, prior literature found evidence on the negative (positive) relationship between managerial ownership and earnings management (corporate performance) for low and high levels of managerial ownership, suggesting the dominance of the incentive alignment theory. Contrary, there is a positive (negative) relationship between managerial ownership and earnings management (corporate performance) at intermediate levels of ownership, consistent with the entrenchment theory being strong for these levels of ownership. Yet, each aforementioned study has found different turning points of managerial ownership effects. Given these arguments, it eventually is an empirical question whether and how managerial ownership influences classification shifting. Our study only focuses on the specific level of managerial ownership where the entrenchment effect exceeds the incentive alignment effect. Therefore, the following hypothesis is stated:
Hypothesis 1. At lower levels of ownership, managerial ownership is negatively related to
classification shifting, and at higher levels of ownership, managerial ownership is positively related to classification shifting.
III. METHODOLOGY
In this chapter, the methodology is explained. First, the sample and data are elaborated. After that, the regression-based method for classification shifting of McVay (2006) is further explained. As last, the test design with the managerial ownership variables is developed.
Sample & Data
This subsection describes the sample and data that is utilized to test the hypothesis. The sample used contains financial data gathered from North America’s Annual Compustat
database on U.S. listed firms from 2002 to 20183. Table 2 summarizes the sample selection
procedure. All firm-year observations need to have adequate data, and if not, the observation is excluded from the sample. According to Zalata and Roberts (2016) and Anthanasakou et al. (2011), financial firms have a different financial reporting setting, and utility firms are more regulated. Hence, financial firms and utility firms are dropped. Similar to McVay (2006), firm-years are excluded from the sample for the following explanations; 1) sales is less than $1 million to diminish the outliers, 2) fiscal year-end changes to make comparable years, and 3) less than 15 observations per industry and fiscal year to provide enough data for expected core earnings’ estimation. We provide the Fama and French (1997) industry categorization, which is consistent with prior classification shifting studies (McVay 2006; Fan et al., 2010). In addition, the year 2019 is not included in the sample procedure for the reason that there is a reduction in observations since Compustat does not hold the financial data for all firms so far4. The ExecuComp data set contains information on executive compensation, including the proportion of total stocks owned by the top five executives to measure managerial ownership. Starting with an initial sample of 92.260 firm-year observations, all exclusions result in a sample of 12.669 firm-year observations, of which 1.630 firms.
Empirical Method
McVay (2006) established an approach to measure classification shifting. While some studies (e.g., Fan et al., 2010) reflect that McVay’s methodology of including accruals in the measurement of expected core earnings, creates bias for the estimation of classification shifting, we maintain McVay’s Eq. (1) methodology for comparability considerations and deal with other concerns through further sensitivity analysis. In McVay’s (2006) methodology, current core earnings are expected to be overstated in the year income-decreasing special items occur. Therefore, McVay (2006) intends that when firms use classification shifting, unexpected core earnings increase with income-decreasing special items in the same year. The model of McVay (2006) described that the amount of expected core earnings is made by estimating Eq. (1), where after the predicted value is applied as the expected level of core earnings. Eq. (1) is estimated per industry and year to control for industry and macroeconomic concerns:
CEt = α0 + α1CEt-1 + α2ATOt + α3ACCRUALSt-1 + α4ACCRUALSt
+ α5∆SALESt + α6NEG_∆SALESt + εt (1)
3 Since several variables need one year of lagged data, the real period tested is 2003 to 2018. 4 Results are robust when 2019 is included in the sample.
Table 2. Sample selection procedure Firm-year observations % Total firms
Initial sample from Compustat North America -
Annual Updates for Jan 1st, 2002 - Dec 31st, 2018 92.260 100 12.797
Financial and utilities firms 85.726 Sales under $1 million 68.989 Fiscal year-end change
Missing ownership data
Less than 15 observations per industry
68.493 14.140 12.669
Final Sample (Jan 1st, 20035 - Dec 31st, 2018) 12.669 13.7 1.630
where CEt is the real amount of core earnings at year t, ATOt is the asset turnover at year t,
ACCRUALSt is the amount of accruals at year t, ∆SALESt the percent change in sales at year t,
and NEG_∆SALESt is the percent change in sales when this change is less than zero and zero
otherwise at year t. Appendix A Table A1 shows a detailed explanation of the variable measurements. This model is estimated cross-sectional by industry and fiscal year, excluding firm i, using Ordinary Least Squares (OLS). Unexpected core earnings are the difference between actual core earnings and expected core earnings. In vein with McVay (2006), classification shifting takes places if there is a statistically positive relationship between unexpected core earnings and income-decreasing special items as a percentage of sales, which is tested by Eq. (2)
UE_CE= β0 + β1%SI+ Control variables + v (2)
where UE_CEt is the difference between actual core earnings and predicted core earnings as
estimated by Eq. (1). %SIt is defined as income-decreasing special items as a percentage of
sales if special items are income-decreasing, and zero otherwise. In accordance with McVay (2006), β1 is expected to be significant and positive when classification shifting is used. Similar
to Zalata and Roberts (2016), we have added control variables to the model. These control variables are elaborated in the test design.
5 The year 2002 was used as a lagged year in variable calculations. It was automatically dropped from the sample
Measuring Managerial Ownership
Prior literature has provided a large variation in managerial ownership measurements. Chung and Pruitt (1996) define managerial ownership generally as the stocks owned by the management of a firm. More specifically, Tushman and Rosenkopf (1996) suggest that managers are the individuals at the highest level of management; top executives of the firm. However, this definition of managerial ownership differs from that of Morck et al. (1988) who define managerial ownership as ownership by members of the board of directors, and McConnell and Servaes (1990) who define managerial ownership as ownership by both executive officers and the board of directors’ members. Concerning managerial ownership, we distinguish executive managers and the board of directors. For instance, the effect of managerial ownership by non-executive directors on classification shifting has been studied by Zalata and Roberts (2016). The literature supporting the effect of ownership of the board of directors is different from that of ownership by executive managers, as these two groups could have different interests and incentives (Demsetz and Villalonga, 2001). As given as an example by Demsetz and Villalonga (2001), a board member could have a board position as he represents someone who has large stock ownership or possesses stocks himself, and therefore board members do not have interests identical to those of executive managers. Furthermore, Kole (1995) made a comparison, and distinguished managerial ownership in directors, insiders, and executive managers, and found that executive management generally has higher proportions of ownership than the board of directors.
Consequently, we only consider executive managers to measure managerial ownership. The firm’s CEO is a key agent of the stockholder and responsible for the organizations' strategies and policies. CEOs are likely to have the most impact on decision-making (Hall and Liebman, 1998). However, CEOs often work as part of a team involving key line managers in shaping the firms' operational strategies (Lafond and Roychowdhury, 2008). Therefore, we provide stocks owned by the top five executives as measurement for managerial ownership. Additionally, we use the stocks owned, excluding stock options, which is similar to Warfield et al. (1995), Lafond and Roychowdhury (2008), and Jelinek and Stuerke (2009). This definition could be linked to the ExecuComp database, which includes compensation items of the top five executive officers per firm-year. Overall, managerial ownership refers to the proportion of stocks owned (excluding stock options) by the top five executives.
Test Design
To test the hypothesis, managerial ownership and the interaction of income-decreasing special items with managerial ownership are included in our regression model, as shown in Eq. (3). Following Morck et al. (1988), we assume a non-linear relationship for the effects of managerial ownership. As aforesaid, using piecewise linear models of low, intermediate, and high ownership levels could be a potential problem, because it ensures that the coefficients of the managerial ownership variables only change at predetermined ownership levels. Therefore, we avoid this problem by using curvilinear functional forms rather than piecewise forms. Hence, we add the square of managerial ownership to our estimation model to test the non-linear relationship and find the exact turning point. Furthermore, comparable to Zalata and Roberts (2016) and Barua, Lin, and Sbaraglia (2010), control variables are attached to control for performance. The following firm-level control variables are attached to the model: firm size (SIZ), leverage (LEV), operating cash flow (OCF), return on assets (ROA), and book to market value (BMV). We perform OLS regressions in combination with robust standard errors for heteroskedasticity, as suggested by White (1980). In addition, year-fixed effects are added to control for unobserved year-fixed effects6. The regression model has the following equation:
UE_CE= 𝛾0 + 𝛾1%SI+ 𝛾2MAN + 𝛾3MAN2 + 𝛾4%SI x MAN + 𝛾5%SI x MAN2
+ 𝛾6SIZ + 𝛾7LEV + 𝛾8OCF + 𝛾9ROA + 𝛾10BMV + v (3)
where
UE_CE = unexpected core earnings, measured as the difference between actual core earnings and expected core earnings at year t as estimated by Eq. (1),
%SI = income-decreasing special items as percentage of sales, measured as special items divided by sales in year t, when special items are income-decreasing, and zero otherwise.
MAN = managerial ownership, measured as the proportion of stocks excluding stock options owned by the top five executives,
MAN2 = the square of MAN,
SIZ = size of the firm, calculated as the natural log of total assets,
LEV = leverage, measured as total debts scaled by total assets,
6 A Hausman test is performed for the comparison of the estimates of the random-effects model and the
OCF = operating cash flow, measured as cash flows from operations scaled by lagged total assets,
ROA = return on assets, measured as net income scaled by total assets on average,
BMV = book to market value, measured as total assets scaled by market capitalization as at year-end.
If classification shifting takes place within firms, 𝛾1 has to be significant and positive,
in accordance with McVay (2006). Therefore, 𝛾1 is expected to be positive and significant.
Furthermore, to test the hypothesis that managerial ownership has an association with the extent of classification shifting, we regress between the unexpected core earnings (UE_CE) and the interaction of income-decreasing special items (%SI) and managerial ownership (MAN). We expect that at low levels of managerial ownership, unexpected core earnings (UE_CE) is negatively related to the interrelated variable (SI x MAN), which means that 𝛾4 is expected to
be significant and negative, suggesting the dominance of incentive alignment. We also assume that at a specific point, entrenchment dominates incentive alignment. Therefore, we expect that at higher levels of managerial ownership, unexpected core earnings (UE_CE) is positively related to the interrelated variable (SI x MAN2), which means that 𝛾5 is expected to be significant
and positive.
IV.
RESULTS
This chapter begins with the descriptive statistics, providing summary statistics, a Pearson correlation matrix, and the model of expected core earnings. After that, the results of the main regression are shown. Finally, sensitivity analyses are performed to check robustness.
Descriptive Statistics
The descriptive statistics for the variables provided in the regression analysis are shown in Table 3. The mean (median) of sample CE values is 0.158 (0.142), which shows that the core earnings represent nearly 16 percent of sales. This is noticeably larger than the 7 percent stated by McVay (2006), but on the other hand similar to the increasing development in recent literature (e.g., Abernathy et al., 2014; Behn et al., 2013). The mean (median) of sample UE_CE values is 0.000 (0.000), whereas the mean (median) of sample %SI values is 0.021 (0.004). The presented firm characteristics are provided to control for performance. These classification shifting descriptive statistics are similar to prior large studies of classification shifting (McVay, 2006; Fan et al., 2010; Zalata and Roberts, 2016). Furthermore, the mean (median) of sample
MAN values is 0.034 (0.009). This indicates that the top five executives own approximately 3.4
percent of the total firm’s stocks. For example, Morck et al. (1988) and Short and Keasey (1999) show larger managerial ownership values, respectively 0.133 (0.056) and 0.106 (0.034). However, Short and Keasey (1999) have already shown that managerial ownership values are decreasing slightly over the period. Besides that, these studies have used other proxies for managerial ownership, which ensures that managerial ownership values are less useful for comparison. In a more recent study, Lafond and Roychowdhury (2008) showed that the mean of managerial ownership, with also the top five executives of U.S. listed firms as a proxy (period 1994 to 2004), is 0.045, which is similar to our descriptive statistics.
The Pearson correlation matrix for the main variables is shown in Table 4. It reports that the percentage of income-decreasing special items and unexpected core earnings for Pearson is positively and significantly correlated (0.021), which shows that companies might have misclassified core expenses as special items. Furthermore, the matrix shows that the correlations between ATO and the UE_CE are not significant, which is similar to McVay (2006). Additional calculations of the variance inflation factor (VIF) and related tolerance were done to indicate that there are no potential issues with multicollinearity, which is shown in Appendix B Table B1. The VIF values (1.28 to 6.63, mean 3.42) approve that each independent variable in the regression model could be included to estimate the dependent variable without multicollinearity concerns7.
Eq. (1) is estimated for all industry and year combinations to predict the expected core earnings. The regression outcomes are not all shown. In its place, the mean and median of each coefficient and the corresponding p-value are presented in Table 5. Similar estimates are established as in McVay (2006) for all explanatory variables. Core earnings are persistent as core earnings’ lagged value is statistically significant and positive for approximately 99% of the 320 regressions, which is comparable to McVay (2006). Moreover, ATO does not seem to be an effective predictor of core earnings, also comparable to that of McVay (2006). Lastly, a high adjusted R2 shows that Eq. (1) can predict core earnings effectively.
Table 3. Descriptive statistics
Variable Mean Median Standard
Deviation 25% 75% CE 0.158 0.142 0.153 0.081 0.222 UE_CE 0.000 0.000 0.044 -0.015 0.015 %SI 0.021 0.004 0.060 0.000 0.017 MAN 0.034 0.009 0.067 0.003 0.028 MAN2 0.006 0.000 0.023 0.000 0.001 SIZ 7.478 7.363 1.619 6.317 8.528 LEV 0.217 0.194 0.197 0.027 0.332 OCF 0.116 0.108 0.089 0.066 0.162 ROA 0.048 0.055 0.099 0.017 0.095 BMV 1.309 0.865 1.723 0.524 1.470
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations. All variables are winsorized at 1 percent and 99 percent.
Table 4. Pearson correlation matrix
Variable SALES ∆SALESt CEt CEt-1 UE_CEt %SIt ACCRSt ATOt
SALESt 1.000 ∆SALESt -0.040 1.000 (0.000) CEt 0.017 0.229 1.000 (0.052) (0.000) CEt-1 0.018 0.040 0.621 1.000 (0.048) (0.000) (0.000) UE_CEt 0.001 0.010 0.351 0.032 1.000 (0.915) (0.258) (0.000) (0.000) %SIt -0.046 -0.094 -0.061 -0.001 0.013 1.000 (0.000) (0.000) (0.000) (0.927) (0.145)
Table 4 (Continued)
Variable SALES ∆SALESt CEt CEt-1 UE_CEt %SIt ACCt ATOt
ACCRSt 0.062 0.132 0.142 -0.097 -0.005 -0.449 1.000
(0.000) (0.000) (0.000) (0.000) (0.560) (0.000)
ATOt 0.117 0.0300 -0.237 -0.246 -0.003 -0.120 0.191 1.000
(0.000) (0.001) (0.000) (0.000) (0.718) (0.000) (0.000)
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations. Significance is provided in the parentheses. All variables are winsorized at 1 percent and 99 percent.
ACC is ACCRUALS. The p-values are shown in parentheses.
Table 5. Model of expected core earnings
Independent Variables
Sign Mean Median Percent
Significant Predicted Direction Intercept 0.031 0.015 (0.152) (0.100) CEt-1 + 0.865 0.888 99.0% 99.3% (0.002) (0.000) ATOt - -0.003 -0.000 30.7% 54.2% (0.223) (0.220) ACCRUALSt-1 - -0.089 -0.066 48.3% 72.9% (0.150) (0.114) ACCRUALSt + 0.079 0.076 52.2% 68.6% (0.147) (0.093) ∆SALESt + 0.040 0.038 50.7% 73.1% (0.154) (0.096) NEG_∆SALESt + 0.137 0.090 47.7% 63.2% (0.152) (0.107) Adjusted R2 85.3% 88.2%
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations and 320 industry-year regressions. All variables are winsorized at 1 percent and 99 percent. The p-values are provided in parentheses and are formed on one-tailed tests for the independent variables. As the sample sizes per industry and per year differ, the p-values rather than standard errors are shown.
Regression Results
The outcomes of Eq. (2) and (3) using OLS regressions are presented in Table 6. Both regressions provide the full sample, as shown in Table 2. Model 1 (Eq. 2) provides the findings when only income-decreasing special items are included as an independent variable. Model 2 (Eq. 3) uses OLS regression with robust standard errors for the full sample. The coefficient of
%SI is positive and significant in model 1 (0.073) and model 2 (0.078), indicating
misclassification of core expenses as special items. These outcomes are consistent with the provided evidence of McVay (2006). The regression between UE_CE and interrelated variable
%SI x MAN is negative and significant (-0.361). Additionally, the regression between UE_CE
and interrelated variable %SI x MAN2 is positive and significant (1.682). The results of our main regression suggest that the association between managerial ownership and classification shifting is non-linear. Therefore, regarding our hypothesis, we could state that at lower levels of ownership, managerial ownership is negatively related to classification shifting, and at higher levels of ownership, managerial ownership is positively related to classification shifting. Theoretically, this regression suggests that at lower levels of ownership, the incentive alignment effect is dominant, and at higher levels of ownership, the entrenchment effect is dominant. These findings are in line with the aforementioned studies that examined managerial ownership and corporate performance (e.g., Morck et al., 1988; McConnell and Servaes, 1990; Short and Keasey, 1999). More specifically, classification shifting is negatively related to managerial ownership up to 10.73% ownership8. After this percentage, classification shifting is positively
related to managerial ownership. Hence, we can state that at approximately 11%, the entrenchment effect exceeds the incentive alignment effect. This turning point is consistent with Short and Keasey (1999), who found a 15.58% cutoff, and Teshima and Shuto (2008), who found a 13.6% cutoff. Regarding the firm-characteristic control variables that are included for each regression models, we found that all control variables are significant, except for BMV. Furthermore, the adjusted R2 is significantly higher than McVay (2006). However, this could be assigned to additional control variables in our model.
As proposed by McVay (2006) and Zalata and Roberts (2017), some companies experience less opportunities to make use of classification shifting than others, especially companies without special items. Therefore, an alternative sample is added under model 3. This
8 The turning point is calculated by re-writing the equation into a completed square form. The turning points are
found by differentiating y (UE_CE) with respect to x, letting Coefficient y
Coefficient x = 0, and solving for x. In this case: UE_CE
Table 6. Regression of unexpected core earnings, special items, and managerial ownership Dependent Variable = UE_CE
Independent Variables Full Sample (1) Full Sample (2) Income-decreasing Special Items Only
(3) Intercept -0.019*** -0.017*** -0.020*** (0.002) (0.002) (0.003) %SI 0.073*** 0.078*** 0.078*** (0.013) (0.017) (0.014) MAN -0.018 0.004 (0.011) (0.015) MAN2 0.023 -0.021 (0.029) (0.040) %SI x MAN -0.361*** -0.387*** (0.098) (0.118) %SI x MAN2 1.682*** 1.740*** (0.306) (0.315) SIZ 0.001*** 0.001** 0.001** (0.000) (0.000) (0.000) LEV 0.008*** 0.008*** 0.006* (0.002) (0.002) (0.003) OCF 0.053*** 0.053*** 0.056*** (0.008) (0.009) (0.011) ROA 0.067*** 0.068*** 0.066*** (0.012) (0.009) (0.009) BMV 0.000 0.000 0.000* (0.000) (0.000) (0.001)
Year-fixed effects No Yes Yes
Number of observations 12.669 12.669 8.553
Adjusted R2 4.62% 4.70% 4.73%
Model 1: UE_CE= β0 + β1%SI+ β2SIZ + β3LEV + β4OCF + β5ROA + β6BMV + v
Model 2/3: UE_CE= y0 + y1%SI+ y2MAN + y3MAN2 + y4%SI x MAN + y5%SI x MAN2+ y6SIZ + y7LEV +
y8OCF + y9ROA + y10BMV + v
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on OLS regression with robust standard errors adjusted for heteroskedasticity, and year-fixed effects. Robust standard errors are provided in parentheses. Significance at the 10%, 5%, and 1% level is shown by *, **, and ***, respectively.
regression has the same model as Eq. (3), however only includes a sample of firms with decreasing special items. Hence, on the contrary to the other regressions where income-increasing special items are set to zero, firms with income-income-increasing special items are excluded. The outcomes are similar to the regression of the full sample. The coefficient of %SI is positive and significant (0.078) and significant as well. Furthermore, the regression between
UE_CE and interrelated variable %SI x MAN is negative and significant (-0.387). The
regression between UE_CE and interrelated variable %SI x MAN2 is positive and significant (1.740). In this model, the turning point of managerial ownership is 11.12%, which is similar to our main regression. Hence, we could propose that our main results are similar to the sample of firms with income-decreasing special items only.
Sensitivity Analysis
In this section, we completed a sensitivity analysis to observe if the findings reported in the main analysis are sensitive to the model conditions of McVay (2006), and other possible measurements of managerial ownership.
Validity of McVay’s (2006) Classification Shifting Model
The main regression’s results employ on the unexpected core earnings’ measurement using the expectations model of McVay (2006). According to McVay (2006), accruals include both normal accruals and special item accruals. Both of these accruals are predicted to be related to performance. Nevertheless, the relation with performance might not be similar for both types of accruals, though McVay’s (2006) model is limited to handling each type of accrual equally. Hence, if special items are more or less related to performance than normal accruals, the residuals from the models could differ systematically with accrual special items. Furthermore, Fan et al. (2010) and Barua and Cready (2008) also claim that the presence of special item accruals will result in a positive mechanical relationship between special items and unexpected core earnings. Therefore, McVay’s (2006) findings are more illustrative of model bias than of classification shifting.
Consequently, similar to McVay (2006), we re-estimate core earnings under three conditions, which are presented in Table 7. The first alternative condition (Excluding Accruals) excludes accruals from the model. The second alternative condition (Accruals = CE - CFO) includes accruals before special items, where accruals are the difference between core earnings (CE) and cash from operations (CFO). This ensures that all special items are treated as accruals. The third alternative condition (Accruals = NI - CFO - SI_ACC) tries to withdraw special items that are expected to be accruals, rather than all special items. In this method, accruals are similar
Table 7. Sensitivity analysis for validity McVay’s (2006) expectation model Dependent Variable = UE_CE
Independent Variables Excluding Accruals (1) Accruals = CE - CFO (2) Accruals = NI – CFO – SI_ACC (3) Excluding Current Accruals (4) Intercept -0.019*** -0.009*** -0.019*** -0.020*** (0.002) (0.002) (0.002) (0.002) %SI 0.032*** 0.069*** 0.047*** 0.029** (0.007) (0.010) (0.008) (0.010) MAN -0.020** -0.015* -0.013 -0.020** (0.008) (0.007) (0.008) (0.008) MAN2 0.032 0.025 0.015 0.034 (0.023) (0.025) (0.023) (0.023) %SI x MAN -0.527* -0.614** -0.314 -0.418 (0.254) (0.223) (0.303) (0.279) %SI x MAN2 2.007** 2.061** 1.244 1.455 (0.876) (0.892) (1.431) (1.056) SIZ 0.001*** 0.000** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) LEV 0.012*** 0.015*** 0.009*** 0.009*** (0.002) (0.001) (0.002) (0.002) OCF 0.037*** -0.057*** 0.048*** 0.028*** (0.010) (0.005) (0.006) (0.008) ROA 0.080*** 0.123*** 0.051*** 0.090*** (0.008) (0.008) (0.006) (0.006) BMV 0.000 0.001 0.000 0.000 (0.000) (0.000) (0.000) (0.000)
Year-fixed effects Yes Yes Yes Yes
Number of observations 12.006 11.998 10.817 11.998
Adjusted R2 5.25% 6.66% 4.54% 6.33%
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on OLS regression with robust standard errors adjusted for heteroskedasticity, and year-fixed effects. Robust standard errors are provided in parenthesis. Significance at the 10%, 5%, and 1% level is shown by *, **, and ***, respectively.
to net income before extraordinary items (NI) less cash from operations (CFO), and accrual special items (SI_ACC)9. Furthermore, to follow the explanation that the model is weak as a
result of ineffectively controlling for performance, we only use firms with positive core earnings, as these firms are expected to be less prone to the effects of performance. McVay (2008), in response to Barua and Cready (2008), presents empirical evidence against their statements. She showed use of classification shifting under these conditions in a model of expected core earnings for a sample of firms with positive core earnings in which the performance effect is mitigated. As last, Behn et al. (2013) also excluded current accruals from the main model for core earnings to address Fan et al.’s (2010) concerns with McVay’s (2006) model. However, they included lagged accruals. Hence, we use this model of core earnings for the fourth alternative condition.
The results are shown in Table 7. The coefficient of %SI of each model is similarly positive (ranging from 0.032 to 0.069) and significant, indicating the use of classification shifting for each alternative. The interaction between UE_CE and the interrelated variable %SI
x MAN is negative for each measurement (ranging from -0.314 to -0.614), which is in line with
the results of our main regression, but only statistically significant for model 2. Besides that, regressions between UE_CE and the interrelated variable %SI x MAN2 provide positive coefficients (ranging from 1.244 to 2.061), whereas only model 1 and 2 are statistically significant. In conclusion, the findings of model 1 and model 2 show similar regression values as our main regression, and therefore are not sensitive to the accrual definition of McVay’s (2006) model. However, the findings of model 3 and model 4 are not statistically significant, and therefore not robust to the accrual definition.
Alternative Measurements
Further sensitivity analysis was conducted using other measurements of the independent variable. Each regression is shown in Table 8. In our main regression, managerial ownership is measured as the stocks owned by the top five executives. Nevertheless, Kim and Lu (2011) claim that changes in the number of executive managers or the board of directors lead to differences in the measures applied by Morck et al. (1988) and Himmelberg et al. (1999), however, these changes do not need to result in shifts in incentives. Hence, managerial ownership measures depending on ownership of the board of directors or ownership of the top
9 Similar to McVay (2006), we estimate accrual special items (SI_ACC) to be the sum of Loss (Gain) on sale of
assets and Funds from Operations, Other. This value is winsorized at special items and set to 0 if special items are income-increasing.
Table 8. Sensitivity analysis for alternative measurements
Dependent Variable = UE_CE
Independent Variables CEO Only (1) Including Options (2) Firm-fixed Effects (3) Industry-fixed Effects (4) Intercept -0.017*** -0.017*** 0.020 -0.019*** (0.002) (0.002) (0.010) (0.003) %SI 0.072*** 0.082*** 0.118*** 0.079** (0.014) (0.020) (0.020) (0.030) MAN -0.042* -0.018 0.003 -0.016 (0.019) (0.011) (0.028) (0.010) MAN2 0.118 0.020 -0.024 0.020 (0.075) (0.029) (0.059) (0.032) %SI x MAN -0.098 -0.418* -0.326 -0.356* (0.444) (0.208) (0.364) (0.204) %SI x MAN2 1.211 1.632*** 1.794** 1.673*** (2.160) (0.391) (0.822) (0.386) SIZ 0.001** 0.001** -0.005*** 0.001 (0.000) (0.000) (0.001) (0.001) LEV 0.008*** 0.008*** 0.001* 0.010*** (0.002) (0.002) (0.006) (0.003) OCF 0.053*** 0.052*** 0.080*** 0.057*** (0.009) (0.009) (0.011) (0.014) ROA 0.067*** 0.068*** 0.109*** 0.068** (0.009) (0.009) (0.015) (0.025) BMV 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.001)
Year-fixed effects Yes Yes Yes Yes
Number of observations 12.669 12.669 12.669 12.669
Adjusted R2 4.65% 4.70% 6.23% 4.89%
See Appendix A Table A1 for variable definitions and measurements. The full sample contains 12.669 firm-year observations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on OLS regression with robust standard errors adjusted for heteroskedasticity, and year-fixed effects are added. Robust standard errors are provided in parenthesis. Significance at the 10%, 5%, and 1% level is shown by *, **, and ***, respectively.
five executive managers could not be a solid and consistent measure, according to Kim and Lu (2011). Hence, we apply the ownership of CEOs as an alternative measure for managerial ownership, which is in line with Zhou (2001), Kim and Lu (2011), and Benson and Davidson (2009). This variable is not affected by changing numbers of executive managers or the board of directors. Besides that, CEOs are likely to have the most impact on decision-making (Hall and Liebman, 1998), suggesting that they most thoroughly affect corporate strategy and culture, and control and monitor the activities of other key executive managers. Therefore, we use stocks owned by the CEO as an alternative measurement for managerial ownership (model 1). The findings present dissimilar regressions values and are not statistically significant, and therefore not robust.
Another managerial ownership measurement is the stocks owned, including stock options. To increase managerial ownership levels, managers are awarded stocks or stock options (Core and Guay, 1999). According to Zhou (2001), overlooking ownership of options could result in incorrect inferences. Therefore, they claim that literature must integrate the incentive effect of options when examining the relationship between managerial ownership and performance. Hence, stocks owned, including options is used as an alternative measurement of managerial ownership (model 2). The results of this sensitivity analysis provide similar regression values as our main regression, and therefore it is not sensitive to the measurement of managerial ownership.
Furthermore, the main regression might not have incorporated an unknown firm and industry effect that influence managerial ownership and classification shifting. Therefore, we provide extra sensitivity analysis to examine if the findings are sensitive to these unknown firm-level and firm-level factors. We used additional firm-fixed effects (model 3) and industry-fixed effects (model 4). Although our regressions present similar values, the findings are not similarly statistically significant. This could be due to that firm- and industry-fixed effects are overly restrictive.
V.
CONCLUSION AND DISCUSSION
This study provides evidence of the association between managerial ownership and classification shifting. There has been no research that studied the relationship between managerial ownership and classification shifting. Prior literature on the relationship between managerial ownership and corporate performance and discretionary accruals has found a non-linear relation. These studies (e.g., Morck et al. 1988; McConnell and Servaes, 1990; Teshima and Shuto, 2008) found that the incentive alignment theory is dominant at lower levels.
