Does executive career horizon influence classification shifting?

(1)

Does executive career horizon influence

classification shifting?

Master thesis, Msc Accountancy and Controlling, specialization Controlling University of Groningen, Faculty of Economics and Business

June 24, 2019

Els van Minnen Studentnumber: S2351706

Professor Rankestraat 31a 9713 GD Groningen Tel.: +31 611412411

E-mail: e.c.j.minnen@student.rug.nl

Supervisor Prof. C. K. (Stan) Hoi

Acknowledgement: I thank prof. C. K. (Stan) Hoi for his guidance, support and valuable comments on earlier drafts of this thesis. I would also like to thank V. A. Porumb for his useful instructions on how

(2)

Does executive career horizon influence

classification shifting?

Abstract

Classification shifting is an earnings management method that involves deliberate misclassification of items within the income statement to manipulate either core earnings or operating cash flows. This study is novel by not treating CEOs as a homogeneous group. Instead, this study examines the impact of CEO career horizons on classification shifting. The career horizon used is an industry-adjusted compound measure that includes both CEO tenure and age. I find statistically significant evidence that there is a relationship between CEO career horizon and classification shifting using longitudinal data from 1,246 different U.S. firms between 2002 and 2017. I find that classification shifting is less prevalent among firms with CEOs with a relative short career horizon compared to firms with CEOs with a relatively longer career horizon. The results indicate that while CEOs with a relative short career horizon are more risk-averse and focused on maintaining their status quo, they do not consider classification shifting as a viable earnings management tool. This suggests that the behavioral attributes of these CEOs might have outweighed the relatively low-risk character of classification shifting. These findings remain unchanged in a range of sensitivity tests and robustness checks.

(3)

1. Introduction

Earnings management include various accounting techniques that executives can use to manipulate earnings. By so doing, executives are masking the extent to which earnings reflect true economic performance of the company, potentially reducing financial statement informativeness of the company (McVay, 2006). The extant accounting research has focused on earnings management tools such as accruals management and real earnings management (Cohen, Dey and Lys, 2008; Zang, 2012). McVay (2006) provides evidence of classification shifting in U.S. companies. Classification shifting is an earnings management method that involves deliberate misclassification of items within the income statement to manipulate either core earnings (McVay, 2006; Fan, Barua, Cready and Thomas, 2010) or operating cash flows (Lee, 2012). To date, classification shifting remains a relatively unexplored area of research in the earnings management literature. This notwithstanding, the Securities Exchange Commission (SEC) has underpinned the relevance of classification shifting: “The appropriate classification of amounts within the income statement is as important as the appropriate measurement or recognition of such amounts” (SEC, 2000). In particular, the SEC has accused companies such as SafeNet Inc., Symbol Technologies, and Dell Inc. of misclassifying operating expenses as non-recurring expenses (SEC, 2009; SEC, 2010a; SEC, 2010b).

Prior research shows that classification shifting is not restricted to U.S. firms but occurs in a wide international setting (Behn, Gotti, Herrmann and Kang, 2013; Haw, Ho and Li, 2011; Shirato and Nagata 2012; Athanasakou, Strong and Walker, 2009). Much of this literature treats Chief Executive Officers (CEOs) as a homogenous group (Zalata, Ntim, Aboud and Gyapong, 2018). This, despite that Francis, Huang, Rajgopal and Zang (2008), Ali and Zhang (2015), and Huang, Rose-Green and Lee (2012) find that CEO reputation, CEO tenure and CEO age affect accrual and real earnings management. This study therefore examines the impact of the individually determinant CEO career horizon on classification shifting. The purpose of this study is to examine the following: does the career horizon of CEOs influence engagement in classification shifting? The career horizon used is an industry-adjusted compound measure that involves both CEO tenure and CEO age, where a long career horizon relative to industry peers reflects CEOs with shorter tenure and younger age. The opposite applies for a short career horizon. Both CEO tenure and CEO age could affect the CEO risk preference, which in turn influences corporate decisions and choices. Because classification shifting is a relatively low-cost and flexible instrument to manage earnings, it is less risky than other tools. CEOs with a short career horizon are more risk-averse and focused on maintaining their status quo than their counterparts with a long career horizon. Consequently, I argue that CEOs with a relative short career horizon are more likely to adopt the less risky classification shifting tool.

I find statistically significant evidence that there is a relationship between CEO career horizon and classification shifting using 1,246 unique U.S. firms and over 15,000 firm-year observations over the period 2002 to 2017. Furthermore, I find that CEO career horizons are negatively and significantly associated with classification shifting, indicating that classification shifting is less prevalent in firms with CEOs with a relative short career horizon compared to firms with CEOs with a relatively long career horizon. The findings indicate that while CEOs with a relative short career horizon are more risk-averse and focused on maintaining their status quo, they do not regard classification shifting as a viable earnings management tool, suggesting that the behavioral attributes of these CEOs might have outweighed the relatively low-risk character of classification shifting as a tool to manage earnings.

This study adds to the accounting research in several respects and offers important insights for multiple financial statements and governance stakeholders. Earnings management is a much-studied subject in the financial accounting field (Rath and Sun, 2008), it directly impacts the credibility of financial reporting and therefore it can undermine the usefulness of accounting

(5)

information in well-functioning financial markets (Rath and Sun, 2008; Haw et al., 2011). First, this study expands the classification shifting literature directly, which in turn contributes to the literature on earnings management as most classification shifting research focuses on neoclassical firm-level determinants (Zalata and Roberts, 2016) or country-level determinants (Behn et al. 2013; Haw et al. 2011). My study is the first to explore the effects of CEO career horizon on classification shifting, hence one of the few that does not treat CEOs as a homogeneous group. In particular, my study investigates the effect on the use of classification shifting of CEOs with a short career horizon relative to their industry peers. Second, this study extends the prior research of Zalata et al. (2018). They show that gender affects classification shifting, where female executives engage considerably less in classification shifting than male CEOs because they are more risk-averse executives compared to male CEOs. Consistent with their findings I find that CEOs with a relative short career horizon are more risk-averse executives and are therefore less likely to adopt classification shifting. Third, my study substantiates earlier research of Ali and Zhang (2015), who find less pronounced evidence of accrual and real earnings management in the later years of CEOs’ service compared to their early years of service as they are more risk-averse and more concerned about their reputation. I find that this also applies to classification shifting focusing on the duration of the CEO career horizon. Finally, my study also contributes to the literature on the relationship between CEO career horizon and earnings management. Few studies have taken into account the use CEO age in analytical models, particularly as opposed to the variable CEO tenure (Miller, 1991; Shen and Cannella, 2002). CEO age is normally used as a control variable (Krause and Semadeni, 2014). This study contributes to the aforementioned shortcoming by providing additional evidence that CEOs with a relative short career horizon engage less in classification shifting than CEOs with a relative long career horizon, even with distinct measures of career horizon. The remainder of this thesis is structured as follows. In the following chapter I examine the earnings management literature and, in particular classification shifting literature. Furthermore, I examine the literature on CEO career horizons and build on those literatures to develop my hypothesis. Next, I test my prediction by using U.S. companies between 2002 and 2017 for which the requisite data on classification shifting and CEO career horizon are available. Finally, I discuss the results and limitations of the study.

(6)

2. Prior literature and hypothesis development

This chapter provides a review of the earnings management literature, with particular focus on classification shifting. Moreover, I develop my hypothesis that expounds the effect of CEO career horizon on classification shifting.

2.1 Earnings management tools

Earnings management refers to corporate practices that misrepresent or mask a firm’s true economic performance. According to McVay (2006) there are three earnings management tools to distinguish: accrual earnings management, real earnings management and classification

shifting. Healy (1985) and Roychowdhury (2006) define accrual-based earnings management

as a tool that involves manipulation of accruals like declaration of expenses or acceleration of revenues to boost current earnings with no direct cash flow effects. Real earnings management is defined by Schipper (1989) as the manipulation of real activities with cash flow effects by altering reported earnings or some set of it through temporal arrangement of investment or financing decisions. For example, boosting sales by providing discounts or cutting into R&D expenses to manage earnings (Dechow and Sloan, 1991; Baber, Fairfield and Haggard, 1991). McVay (2006) shows that classification shifting is another tool through which managers manipulate earnings. She defines classification shifting as “the deliberate misclassification of items within the income statement” (McVay 2006, pp. 501). McVay (2006) uses special items to detect classification shifting and provides evidence that managers shift core expenses (i.e. cost of goods sold or selling, general and administrative expenses) to special items in order to increase core earnings (see Appendix A Figure A1). This practice means that cost items are excluded from the routine (recurring) core expenses base and reported as non-routine (non-recurring) income-decreasing special items instead1. Special items are by definition either unusual in nature or infrequent in occurrence – not both at the same time (McVay 2006). By moving core expenses to special items, the core earnings (sales minus core expenses) will be reported more favorable as a result, however the bottom-line net income is not impacted (see Appendix A Figure A1). Classification shifting differs from accrual and real earnings management. First, both present and future earnings are affected by these instruments, whereas classification shifting does not change future or previous earnings. Second, classification shifting does not modify the GAAP net-income (McVay, 2006). Finally, Nelson, Elliott and Tarpley (2002) show that classification shifting is less likely to be detected by auditors because it has no impact on future earnings as GAAP net-income does not change and has a limiting effect on regulators and (external) auditors’ scrutiny.

Classification shifting

Prior research supported the viability of classification shifting (Barnea, Ronen and Sadan, 1976; Givoly, Hayn and D’Souza, 1999; Davis, 2002). McVay (2006) finds the first evidence of this tool’s application amongst U.S. firms. With subsequent research, Haw et al. (2011), Shirato and Nagata (2012), Behn et al. (2013) demonstrate that classification shifting is not only limited to U.S. firms, but exists in a broad international setting. Haw et al. (2011) and Behn et al. (2013) also find that classification shifting can be mitigated by strong legal institutions and higher financial analyst following primarily in weak investor protection countries, respectively. Athanasakou et al. (2009), Barua, Davidson and Rama (2010a) and Fan et al. (2010) find that classification shifting is used to meet analysts’ expectations. Zalata and Roberts (2016) extend the classification literature on firm-level by examining the role of internal corporate governance on classification shifting and find that strong internal governance (e.g. quality of board and

1_{This thesis focuses only on classification shifting using income-decreasing special items, income-increasing}

(7)

audit committees) constrains classification shifting. Barua and Cready (2008) and Fan et al. (2010) focus on improving the estimation methodology of McVay’s (2006), including mitigating issues about potential model prejudice.

The emerging research on classification shifting treat classification shifting as a tool that motivates managers to misclassify recurring expenses as special items to meet their earnings benchmarks, particularly firms that are constrained to use accrual-based earnings management (for example McVay, 2006; Anthanasakou et al., 2009; Barua et al., 2010a; Fan et al., 2010; Haw et al., 2011). These studies cover the influence of neoclassical firm-level determinants (Zalata and Roberts 2016) or country (Haw et al., 2011; Shirato and Nagata, 2012; Behn et al., 2013) whereby all CEOs are treated as homogenous (Zalata et al., 2018). This, despite the fact that academic research has already established that CEOs motivation and ability to get involved in accrual and real earnings management can depend on more individually related determinants. Francis et al. (2008), Ali and Zhang (2015), and Huang et al. (2012) find that CEO reputation, CEO tenure, and CEO age influence accrual and real earnings management. Francis et al. (2008) show that more reputed CEOs are associated with poorer earnings quality, which is among others measured by discretionary accruals. Ali and Zhang (2015) find less pronounced evidence of accrual and real earnings management in the later years of CEOs’ service compared to their early years of service. Huang et al. (2012) find a positive association between CEO age and financial reporting quality because older CEOs engage less in aggressive earnings management leading to higher-quality financial reporting. To date, only one classification shifting study does not treat CEOs as a homogeneous group. Zalata et al. (2018) examined the effect of CEO gender on classification shifting and find much engagement in classification shifting by female CEOs in the pre-Sarbanes-Oxley act (SOX) period. However, after SOX was implemented, female CEOs engaged considerably less in classification shifting than male CEOs as female CEOs are more risk-averse than male CEOs.

2.2 Behavioral attributes and risk preference of CEOs

Jensen and Meckling (1979) and Hambrick and Mason (1984) suggest that the priorities and incentives of CEOs will change as they become longer tenured and older. CEO tenure is defined as the number of years the CEO has fulfilled the function. This study captures both CEO tenure and age in the CEO career horizon, a compound measure adjusted to the industry (see section 3.3). The interaction between CEO tenure, CEO age and CEO career horizon is illustrated in Figure A2 (Appendix A). This figure shows that if the CEO is young or away from retirement, has a short tenure or is in the early years of service than the CEO career horizon will be long. The reverse applies as well.

Hambrick and Mason (1984), Finkelstein and Hambrick (1988) and Finkelstein, Hambrick and Cannella (2009) provide evidence that differences in CEO attributes such as beliefs, values, and experiences affect their motivation and decisions. In particular, Hambrick and Fukutomi (1991) argue that CEOs with a long tenure are more conservative towards change than short-tenured CEOs because of (1) commitment to their organizational paradigm, (2) neglect of data that disconfirms their paradigm, (3) lower interest in their job, and (4) enhanced authority to evade calls for change. The ‘fixed paradigm issue’ is linked to the phenomenon that longer-tenured CEOs are less open to change (Miller, 1991) and more risk-averse (Simsek, 2007). This issue includes bounded rationality (Cyert and March, 1963), meaning that CEOs experience cognitive and time-limited decision-making e.g. in their working environment and when mapping available strategies (Hambrick and Fukutomi, 1991). The riskiness of CEO choices can change over the tenure of a CEO based on this paradigm. CEOs with a short tenure are open to environmental and organizational changes leading to a relatively flexible paradigm, especially as flexibility is reduced over time with CEO tenure. This is consistent with the life cycle of a

(8)

CEO: in the early years a CEO gains a lot of knowledge and experience and takes initiatives to positively influence corporate outcomes. However, over time, the risk preference of the same CEO changes as the CEO becomes more risk-averse, hesitant towards change and wants to keep the status quo (Dechow and Sloan 1991; Barker and Mueller 2002; Musteen, Barker and Baeten 2006; Matta and Beamish, 2008). Several studies support Hambrick and Mason’s (1984) proposition that older CEOs are less likely to pursue risky strategies or investments involving risk than younger CEOs. Dechow and Sloan (1991) and Barker and Mueller (2002) find that older CEOs spend less on the riskier advertising, capital and research and development (R&D) expenditure, respectively. Matta and Beamish (2008) examine the effect of career horizon, defining the latter as the time to retirement on international acquisitions and find that younger CEOs are associated with more international acquisitions hence more risky behavior (Carpenter, Pollock and Leary, 2003). Musteen et al. (2006) exploring the association of CEO tenure with the CEO attitude towards change find that CEOs become more conservative as their tenure increases. McClelland, Barker and Oh (2012) find that CEOs with longer tenure adopt less risky strategies which adversely impact future firm performance (e.g. future return on assets and market-based performance). Furthermore, Hambrick and Fukutomi (1991), Simsek (2007) and McClelland, Liang and Barker (2010) find that older and longer tenured CEOs are more concerned about maintaining their status quo, leading to a more fixed paradigm. Moreover, previous studies show that commitment to the status quo is related to long-tenured CEOs (Hambrick, Geletkanycz, Fredrickson, 1993), cultural values (Geletkanycz, 1997), longer industry tenure and not having distinct functional experience (Geletkanycz and Black, 2001). CEOs with a longer tenure and older age are therefore likely to favor their status quo as compared to the unknown results of the change execution (McClelland, Barker and Oh, 2012). Thus, as outlined above, multiple studies have established a relationship between the behavioral attributes of CEOs and the CEO’s tenure or age, which attributes i.e. risk-aversion, change resistance and focus on status quo affect their risk preference. In particular, these studies find that a longer tenure or older age is leading to more conservative and risk-averse CEO behavior and with a focus on keeping the status quo, leading to a preference for lower risk. Furthermore, the attitude of a CEO towards change may have a significant impact on an organization’s strategy. A general view of the literature indicates that CEOs display different kinds of conduct towards pursuing strategies at distinct phases of their careers, which can also impact the engagement in earnings management. Ali and Zhang (2015) examined the CEOs tenure and CEOs reputational incentives for earnings management, but only in conjunction with accrual and real earnings management and without classification shifting. Ali and Zhang (2015) build on the argument that the market uses the performance of the company as a measure to evaluate the newly designated CEOs’ capacity. CEOs in their early years of service want to prevent a reputation of a low capacity and are thus more inclined to overstate earnings to report good performance (Holmstrom, 1982). However, to safeguard their established reputation, CEOs with a longer tenure are less inclined in risky, opportunistic behavior. Hence, Ali and Zhang (2015) find that earnings overstatement in terms of accruals and real earnings management is higher in the early years than in the later years of the service of a CEO.

2.3 Classification shifting: a relatively low-risk earnings management tool

Prior research find that the selection of the earnings management instrument is influenced by the constraints, costs and timing of the earnings management instrument (for example Ewert and Wagenhofer, 2005; Cohen et al., 2008; Badertscher, 2011; Zang, 2012; Abernathy, Beyer and Rapley, 2014). Compared with accruals management and real earnings management, classification shifting has several advantages. Cohen et al. (2008) find that the use of accrual-based earnings management has decreased since the implementation of SOX owing to greater

(9)

regulatory scrutiny. Nelson et al. (2002) show that classification shifting is less likely to be detected by auditors because it has no consequences for future earnings as the GAAP net-income does not change. In addition, the interpretation of expense classification standards can be subjective. Hence, auditors have limited ability to detect classification shifting or less incentive to require adjustments (Cohen et al., 2008). Therefore, accrual earnings management has higher detection costs as compared to classification shifting (Alfonso, Cheng and Pan, 2015). According to Alfonso et al. (2015) accrual earnings management has potentially higher litigation and reputation costs due to increased attention by both investors and regulators following the large number of financial scandals since 2000 (Bruns and Merchant, 1990; Merchant and Rockness, 1994; Barua, Lin and Sbaraglia, 2010b;). Huang, Roychowdhury and Sletten (2018) find that this also applies to real earnings management, but to a lower degree than accrual management. Moreover, classification shifting does not involve real business transactions compared to real earnings management and is therefore only an accounting manipulation technique with lower expenses for executives to attain their earnings management objectives (McVay, 2006; Alfonso et al., 2015). Therefore, classification shifting is less expensive than accrual and real earnings management, at least in terms of detection, reputation and litigation risk (Behn et al., 2013; McVay, 2006; Alfonso et al., 2015). Furthermore, Zang (2012) has underpinned the importance of the timing of accrual and real earnings management. Accrual earnings management must be carried out at the end of the period, but within the boundaries of the accounting system (i.e. before the books close) and as a consequence has low accounting system flexibility, whereas real earnings management must be carried out during the fiscal year. Unlike these instruments, classification shifting is done outside of the accounting system as it occurs before earnings are announced but at the end of the period (Abernathy et al., 2014).

Classification shifting is therefore a relatively low-cost and a more flexible earnings management tool, hence a less risky instrument to manage earnings than accrual and real earnings management tools.

2.4 Hypothesis development

This study intends to explore the effect of CEO career horizon, including both CEO tenure and age on classification shifting i.e. their ability and motivation to intentionally misclassify special items. According to several studies both CEO tenure and CEO age may affect the CEO risk preference, which in turn influences corporate decisions and choices, including those related to earnings management strategies. CEOs who have a short career horizon are longer tenured or near-retirement become more risk-averse, reluctant towards change and focused on maintaining their status quo than their younger counterparts who have a short tenure and are far from retirement (long career horizon). These behavioral attributes make older and longer tenured CEOs less inclined to engage in accrual or real earnings management (Ali and Zhang, 2015), but they are still under pressure to deliver good economic performance. Since classification shifting is a relatively low-cost and more flexible instrument it is a less risky tool for earnings management than other tools. The relatively low-risk character of classification shifting may influence the readiness of risk-averse, older and longer tenured CEOs to adopt classification shifting to manage earnings. Building on these arguments regarding the CEO career horizon and the relatively low-risk character of classification shifting, I propose the following hypothesis:

Hypothesis 1. CEOs with a relative short career horizon are more engaged in classification shifting than CEOs with a relative long career horizon.

(10)

3. Research design and summary statistics

This chapter explains the sampling procedure, defines the variables, explains the empirical model I use to explore the effect of CEO career horizon on classification shifting, and provides the summary statistics.

3.1 Sampling procedure

The sample used in this study includes financial data collected from North America’s annual Compustat file for the years 2002 to 20172. Data on CEO tenure and age are taken from the ExecuComp and BoardEx databases to construct the CEO career horizon measure. Each observation is required to have sufficient data to calculate the models’ variables. Like previous research (McVay, 2006; Haw et al., 2011), I use the Fama and French (1997) industry classification. Following Zalata et al. (2018) I exclude financial firms because of their distinct financial reporting environment (Zalata and Roberts, 2016). Following McVay (2006) I remove observations from the sample for the following reasons: 1) annual sales of less than US$ 1 million to mitigate the risk of outliers (since sales is the denominator of most variables), 2) fiscal year-end changes during the year to ensure comparable years, or 3) less than fifteen-firm-year observations per industry per fiscal fifteen-firm-year to ensure that there is sufficient data to estimate expected core earnings.

The final data set includes 15,275 firm-year observations for 1,246 unique firms and 2,698 unique CEOs between 2003 and 2017 for which all regression data are available. The sample selection procedure is summarized in Appendix B Table B1, and the variables used in my analyses are defined in Table B2.

3.2 Classification shifting measure: core earnings expectation model

To provide empirical evidence in support of classification shifting, I will use the core earnings expectation model as developed by McVay (2006). The model first estimates the expected core earnings (CEt) using Eq (1). Then, the “unexpected core earnings” are calculated as the difference between the reported core earnings and expected core earnings. Furthermore, the model expects the core earnings to be overstated (i.e. increase unexpected core earnings) in these years when cost items are labeled as income-decreasing special items, hence an expected positive relationship of the unexpected core earnings with special items. Following McVay (2006), I use the following equation to estimate unexpected core earnings.

CEt = β0 + β1CEt-1 + β2ATOt + β3ACCRUALSt-1 + β4ACCRUALSt + β5ΔSALESt + β6NEG_ΔSALESt + εt (1)

Where CEt is core earnings, calculated as sales minus both cost of goods sold and selling, general and administrative expenses scaled by sales. ATOt is asset turnover ratio, defined as sales divided by the average of net operating assets and expenses. ACCRUALSt is operating accruals, calculated as net income before extraordinary items minus cash from operations scaled by sales. ΔSALESt is the percentage change in sales. To allow for different slope coefficients for sales increases and decreases, NEG_ΔSALESt represents the percentage change in sales when the sales change is negative. Although some research (e.g. Fan et al., 2010) question whether McVay’s method of including accruals in calculating CEt, creates bias in estimating classification shifting, I will retain McVay’s Eq. (1) specification for comparability purposes and will address the concerns through additional robustness checks3_.

2_{As some variables require one year of lagged data, the actual period examined is 2003 – 2017.}

3_{I performed several robustness checks to test the validity of McVay’s (2006) model and the results obtained are}

(11)

Following McVay (2006), I estimate Eq. (1) for a firm using Ordinary Least Square (OLS) by fiscal year and industry, excluding the respective firm in each regression. The residual (Ɛ) of each regression provides an estimation of the unexpected core earnings (UE_CEt) for a firm in a year, where UE_CEt = CEt – E(CEt).

3.3 CEO career horizon measure

Antia, Pantzalis and Park (2010) and Lee, Park and Folta (2018) examined the impact of CEO career horizon on agency costs as well as firm performance and real option investments, respectively. They argue that CEO career horizon should be approximated by the CEOs expected tenure with the firm, as CEOs may expect their tenure to be longer if they are younger and newer in the position compared to peers in similar firms (Jensen and Meckling, 1979). Their measure of CEO career horizon is therefore based on a combination of CEO tenure and age relative to the median tenure and age of CEOs in the industry. A general view from the literature is that both CEO tenure and CEO age affect earnings management (Francis et al., 2008; Ali and Zhang, 2015; Huang et al., 2012) and that career horizons for CEOs across industries may differ (Hambrick et al., 1993). My hypothesis assumes a relationship between classification shifting and CEO career horizon. This implies the necessity of a measure to capture the CEO’s career horizon incorporating the effect of both industry-adjusted tenure and age. Hence, following Antia et al. (2010) and Lee et al. (2018) I use the following equation to construct the career horizon of a CEO (CEOHORIZON):

CEOHORIZON = [CEOTENUREind,t – CEOTENUREi,t] + [CEOAGEind,t – CEOAGEi,t] (2) Where CEOTENURE is defined as the number of years the CEO has fulfilled the function and CEOAGE is the age of the CEO working for firm i in year t (Antia et al. 2010). CEO TENUREind,t and CEOAGEind,t is based on the industry median. The formula may result in positive or negative values. A positive value indicates a longer CEO career horizon relative to the industry median because the CEO is younger and/or is less experienced in the function. Likewise, a negative value indicates a short CEO career horizon because the CEO is older and/or more experienced relative to the median of CEOs in competitor firms.

Eq. (2) is more complete than normal horizon measures as suggested by previous studies using CEO tenure, CEO age or the number of years before reaching the assumed retirement age of 704 (Matta and Beamish, 2008; Lee et al., 2018). The CEO career horizon (illustrated in Figure A2 Appendix A) may be longer if the CEOs are younger and/or less experienced in the position than to their peers in similar firms. Industries can be very different when it comes to employing young or older CEOs (Hambrick et al., 1993), for example internet-related industries have younger CEOs compared to mature industries like General Motors and GE. Academics and professionals acknowledge the fundamentally distinct career horizons of CEOs across industries (Hambrick et al., 1993; Lucier, Spiegel and Schuyt, 2002). Therefore, Eq. (2) is an industry-adjusted measure and not a surrogate for the age of the industry (Antia et al., 2010) as it controls for the impact of the industry on CEO tenure and age. The four-digit Standard Industrial Classification (SIC) code is used to assign firms to the Fama and French (1997) industry hierarchy and enables comparison of CEOs in the same industry.

4_{I performed a range of sensitivity tests with alternate definitions of CEO career horizon (using tenure only, age}

only, industry average instead of median values, retirement age and a standardized measure by dividing the difference in tenure and age by the industry median before the ratios are summed). The sensitivity analysis confirms the robustness of the baseline regression results.

(12)

3.4 Regression model and control variables

Following McVay (2006), Barua et al. (2010b) and Zalata and Roberts (2017), I use the following specification to examine the association between CEO career horizon and classification shifting5_:

UE_CEt = α0 + α1%SIt + α2CEOHORIZONt + α3%SIt x CEOHORIZONt + α4Size + α5Leverage + α6MBV + α7OCF + α8ROA + (firm and year dummies) + εt (3)

For ease of exposition, from this point forward I drop the t subscript, which represents a given year during the sample period between 2003 and 2017. The study centers around shifting core expenses to special items. Following McVay (2006) I predict an association of unexpected core earnings (UE_CE) with special items (%SI) when firms engage in classification shifting. Where %SI is defined as income-decreasing special items scaled by sales, both in the current year, calculated as (Special Items * -1) / Sales when special items are income-decreasing and zero otherwise (McVay, 2006; Fan et al., 2010; Zalata et al., 2018). The connection between UE_CE and %SIin Eq. (3) is the element which indicates classification shifting (Zalata and Roberts, 2017). Hence, I expect α1 to be positive and significant. To test my hypothesis that CEOs with a relative short career horizon have a relationship with the amount of classification shifting, I expect the interaction coefficient between special items (%SI) and CEO career horizon (CEOHORIZON) α3 to be positive and significant.

Following McVay (2006), Barua et al. (2010b) and Zalata et al. (2018), I include a range of firm-level control variables (see Eq. (3)), and no prediction is made for the sign. Where Size is calculated as the natural log of total assets. Leverage is the ratio of long-term debt to stockholder’s equity. MBV is the market-to-book-value ratio calculated as market value scaled by book value of common equity. OCF represents operating cash flow defined as cash flow from operating activities scaled by lagged total assets and ROA is return on assets calculated as net income divided by average total assets. In addition, I include dummy variables to control for unobserved firm and year fixed effects6. All variables in Eq. (1) and (2) are winsorized at 1 percent and 99 percent to mitigate the influence of outliers. Appendix B Table B2 contains detailed definitions and calculations of all variables.

3.5 Summary statistics

The descriptive statistics for the main variables used in the regression analysis are provided in Table 1. Mean (median) sample CE values are 0.156 (0.148), indicating that the core earnings represent approximately 16% of the sample’s net sales. This is substantially higher than the 7% reported by McVay (2006), although in line with the growing trend reported in latest research (for example Behn et al., 2013; Abernathy et al., 2014). The mean (median) UE_CE values and %SI values are respectively 0.003 (-0.001) and 0.020 (0.002). The reported firm characteristics are used to control for performance. The descriptive statistics are within the range of those reported in other studies (McVay, 2006; Fan et al., 2010; Zalata et al., 2018). CEOHORIZON has a negative mean and median value (-2.302 and -0.537) showing that the sample includes more CEOs with a relative short career horizon than CEOs with a long career horizon7.

5_{Following prior research (Fan et al. 2010; Behn et al. 2013), I will examine unexpected core earnings using the}

levels model and not investigate changes in core earnings.

6_{I conducted a Hausman test to compare the estimates of the random-effects and the fixed-effects model in the}

panel structure, the test indicated that the fixed-effects model is preferable.

(13)

TABLE 1 Descriptive statistics

Variable Mean Median Standard

Deviation 25% 75% CE 0.156 0.148 0.195 0.083 0.239 UE_CE 0.003 -0.001 0.091 -0.020 0.020 %SI 0.020 0.002 0.055 0.000 0.015 CEOHORIZON -2.302 -0.537 11768 -8084 5638 Size 7699 7.612 1688 6485 8858 Leverage 0.620 0.358 2069 0.009 0.831 OCF 0.112 0.104 0.095 0.063 0.161 ROA 0.046 0.053 0.107 0.019 0.095 MBV 3.103 2.248 4680 1407 3736

See Appendix B Table B2 for variable definitions and calculations. The full sample consists of 15,275 firm-year observations. All variables are winsorized at 1 percent and 99 percent.

The Spearman/Pearson correlation matrix for the main variables is reported in Table 2. It shows that the %SI and UE_CE for Spearman and Pearson is positively and significantly correlated (0.016, p < 0.05 and 0.036, p < 0.01). This indicates that some firms may have misclassified core expenses as special items. It also shows a negative and significant correlation between %SI and CEOHORIZON (-0.071, p < 0.01 and -0.050, p < 0.01 respectively for Spearman and Pearson) indicating that my hypothesis is not supported.

In general, the correlation matrix does not indicate severe potential issues with multicollinearity8. An assessment of the variance inflation factor (VIF) was conducted. The VIF values (1.01 to 2.24 with mean of 1.45) confirm that it is appropriate to include all independent variables in one regression model in order to estimate the dependent variables9. The VIF and the related tolerance rates are reported in Appendix C Table C1.

8_{Dohoo, Ducrot, Fourichon, Donald and Hurnik (1997) argued that multicollinearity is certain at the 0.90 level of} a correlation coefficient or higher.

(14)

TABLE 2

Spearman/Pearson Correlation Matrix

Variable CE ACCRUALS ATO ΔSALES NEG_

ΔSALES

UE_CE %SI CEOHORIZON Size Leverage OCF ROA MBV

CE 1.000** 0.145*** -0.166*** 0.134*** 0.283*** 0.212*** -0.091*** 0.009 0.319*** 0.049* 0.482*** 0.539*** 0.047* ACCRUALS -0.301*** 1.000** 0.192*** 0.102*** 0.314*** -0.072** -0.468*** 0.030*** -0.022*** -0.037*** 0.013* 0.570*** 0.037*** ATO -0.467*** 0.440*** 1.000** 0.028*** 0.071*** 0.012 -0.115*** -0.030*** -0.179*** -0.079*** 0.110*** 0.111*** 0.103*** ΔSALES 0.158*** 0.139*** 0.064*** 1.000** 0.587*** 0.056*** -0.099*** 0.012 -0.083*** -0.023*** 0.189*** 0.196*** 0.105*** NEG_ΔSALES 0.168*** 0.174*** 0.098*** 0.783*** 1.000** 0.020** -0.207*** 0.023*** 0.046*** -0.004 0.234*** 0.350*** 0.083*** UE_CE 0.255*** -0.074*** -0.049*** -0.010* -0.010* 1.000** 0.036*** -0.025*** 0.042*** -0.006 0.120*** 0.088*** 0.013 %SI 0.023*** -0.278*** -0.178*** -0.154*** -0.166*** 0.016** 1.000** -0.050*** 0.009 0.035*** -0.191*** -0.494*** -0.041* CEOHORIZON 0.012 0.015* 0.000 0.030*** 0.039*** -0.024*** -0.071*** 1.000** 0.002 -0.022*** 0.020** 0.033*** 0.010 Size 0.325*** -0.109*** -0.260*** -0.082*** -0.006* 0.069*** 0.106*** -0.006 1.000** 0.145*** 0.025*** 0.099*** 0.008 Leverage 0.143*** -0.114*** -0.290*** -0.073*** -0.031*** 0.013* 0.148*** -0.008 0.451*** 1.000** -0.059*** -0.048*** 0.536*** OCF 0.428*** -0.162*** 0.189*** 0.250*** 0.229*** 0.149*** -0.179*** 0.032*** -0.026*** -0.187*** 1.000** 0.631*** 0.215*** ROA 0.383*** 0.429*** 0.286*** 0.290*** 0.317*** 0.142*** -0.360*** 0.036*** -0.016** -0.232*** 0.652*** 1.000** 0.172*** MBV 0.225*** 0.076*** 0.188*** 0.252*** 0.235*** 0.084*** -0.088*** 0.030*** -0.001 0.110*** 0.436*** 0.448*** 1.000**

See Appendix B Table B2 for variable definitions and calculations. The full sample consists of 15,275 firm-year observations. Spearman (Pearson) correlations are below (above) the diagonal. All variables are winsorized at 1 percent and 99 percent. Significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

(15)

4. Results

This chapter begins with the results of the hypothesis tests, followed with the results of a number of sensitivity tests. Finally, the results of several robustness analyses are described.

4.1 Multiple regression analysis

The results of the baseline regression model Eq. (3) using OLS regressions with firm-level clustered robust standard errors for the full sample and an alternative sample are presented in Table 3 under models (1) and (2).

TABLE 3

Baseline regressions CEO career horizon and classification shifting

Dependent variable UE_CE All Firms Firms with

income-decreasing special items only Model (1) (2) %SI 0.180*** 0.121*** (0.040) (0.032) CEOHORIZON -0.006*** -0.004** (0.002) (0.002) %SI x CEOHORIZON -0.008*** -0.014*** (0.001) (0.001) Size -0.010*** -0.013*** (0.003) (0.003) Leverage -0.001 -0.000 (0.001) (0.001) MBV 0.000 -0.000 (0.000) (0.000) OCF 0.082*** 0.080*** (0.026) (0.029) ROA 0.093*** 0.062** (0.031) (0.029) _cons 0.062*** 0.087*** (0.021) (0.024)

Firm-fixed effects Yes Yes

Year-fixed effects Yes Yes

Number of Observations 15,275 9,479

R-squared 0.022 0.025

I estimate the parameters using the following regression model:

UE_CEt = α0 + α1%SIt + α2CEOHORIZONt + α3%SIt x CEOHORIZONt + α4Size + α5Leverage + α6MBV +

α7OCF + α8ROA + (firm and year dummies) + εt

See Appendix B Table B2 for variable definitions and calculations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on ordinary least squares regression with robust standard errors adjusted for

heteroskedasticity and within-firm clustering. Standard errors are in parenthesis. Significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

The coefficient of %SI is positive (0.180) and significant at the 1% level, showing misclassification of core expenses as special items in the income statement i.e. classification shifting. The interaction between special items and CEO career horizon (%SI x CEOHORIZON), which is the variable of interest, is negative (-0.008) and significant at the

(16)

1% level. This demonstrates that there is a distinction between CEOs with a short career horizon relative to the industry and a relatively long career horizon in terms of their preference for classification shifting as a low-risk earnings management technique. Interestingly, the inference is that classification shifting is less prevalent among firms with CEOs with a relatively short career horizon. Therefore, although there is a relationship with CEO career horizon and classification shifting, my hypothesis is not confirmed as I predicted a positive relation for %SI x CEOHORIZON (see α3 in Eq. (3)).

McVay (2006) and Zalata and Roberts (2017) find that some firms have more opportunities to use classification shifting than other firms, particularly firms with special items. Following their approach, I restrict the sample only to firms with income-decreasing special items, results of which are presented in Table 3 model (2). Indeed, with the coefficient of %SI still being positive and significant (0.121, p < 0.01) and the %SI x CEOHORIZON being negative and significant (-0.014, p < 0.01) I find similar results as to the use of the full sample.

4.2 Sensitivity analysis

I find my main findings for the full sample in a range of sensitivity tests to be robust to alternate definitions of CEO career horizon and a low-leverage firm test using a subsample.

CEO career horizon sensitivities

The significance of tenure and age in the CEO career horizon was underpinned by Matta and Beamish (2008), Antia et al. (2010) and Krause and Semadeni (2014). Tenure and age are evaluated individually to determine their effect on the baseline results. In addition, in the CEO career horizon variable, I tested the impact of using the industry average rather than the industry median values for tenure and age. I then tested an alternative CEO career horizon measure using the number of years remaining before the CEO reached the assumed retirement age of 65 (Murphy, 1999; Murphy and Zimmerman, 1993; Matta and Beamish, 2008; McClelland et al., 2012). I also tested the impact of using a standardized measure, in which the difference in tenure and age is divided by the median of tenure and age of the industry before the ratios are summed to minimize the potential effect of tenure and age correlation.

The test results of the CEO career horizon sensitivity analysis are presented in Table 4 models (1) to (4) which show that the coefficient of %SI is positive and significant (range from 0.180 to 0.182, p < 0.01) and the appropriate interaction coefficients are negative (range from -0.008 to -0.010) and significant at 1%. This confirms that the primary results for alternative definitions of CEO career horizon are robust throughout this assessment.

Low-leverage firms’ sensitivity

DeAngelo, DeAngelo and Skinner (1994), Peltier-Rivest (1999), Jaggi and Lee (2002) and Saleh and Ahmed (2005) find that high-debt and low-debt firms vary in their engagement in earnings management. High-debt firms are more scrutinized than low-debt firms with less ability to inflate core earnings. I expect to find more pronounced results in low-debt firms following this rationale. I classify a firm as low-leverage when their long-term debt-to-equity ratio is lower than the sample median. To test this conjecture, I use a distinct subsample of leverage firms to repeat my primary assessment. The test results of the sensitivity of low-leverage firms are provided in Table 4 model (5), showing that the coefficient of %SI is positive and significant (0.183, p < 0.05), and that the appropriate interaction coefficient is negative (-0.009) and significant at the 1% level. This confirms the robustness of the main findings for the sensitivity of low-leverage firms.

(17)

TABLE 4

CEO career horizon and Low-leverage firms sensitivity tests Dependent variable

UE_CE

Tenure only

Age only Industry

average Retire- ment age Standard- ized measure Low-leverage firms Model (1) (2) (3) (4) (5) (6) %SI 0.181*** 0.182*** 0.180*** 0.182*** 0.181*** 0.183** (0.039) (0.039) (0.039) (0.039) (0.039) (0.092) CEOHORIZON -0.005*** -0.003* -0.003** -0.003* -0.005*** -0.004 (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) %SI x CEOHORIZON -0.010*** -0.009*** -0.008*** -0.010*** -0.010*** -0.009*** (0.002) (0.001) (0.001) (0.002) (0.002) (0.002) Size -0.010*** -0.011*** -0.010*** -0.011*** -0.010*** -0.014** (0.003) (0.003) (0.003) (0.003) (0.003) (0.006) Leverage -0.001 -0.001 -0.001 -0.001 -0.001 -0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) MBV 0.000 0.000 0.000 0.000 0.000 0.001 (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) OCF 0.081*** 0.081*** 0.082*** 0.081*** 0.082*** 0.123** (0.026) (0.026) (0.026) (0.026) (0.026) (0.050) ROA 0.094*** 0.094*** 0.094*** 0.094*** 0.094*** 0.101* (0.031) (0.031) (0.031) (0.031) (0.031) (0.058) _cons 0.062*** 0.063*** 0.062*** 0.063*** 0.062*** 0.069* (0.021) (0.021) (0.021) (0.021) (0.021) (0.037)

Firm-fixed effects Yes Yes Yes Yes Yes Yes

Year-fixed effects Yes Yes Yes Yes Yes Yes

Number of Observations 15,275 15,275 15,275 15,275 15,275 7,294

R-squared 0.022 0.022 0.022 0.022 0.023 0.026

UE_CEt = α0 + α1%SIt + α2CEOHORIZONt + α3%SIt x CEOHORIZONt + α4Size + α5Leverage + α6MBV + α7OCF +

α8ROA + (firm and year dummies) + εt

See Appendix B Table B2 for variable definitions and calculations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on ordinary least squares regression with robust standard errors adjusted for heteroskedasticity and within-firm clustering. Standard errors are in parenthesis. Significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

Industry sensitivity tests

The baseline regression model may not have included unknown industry characteristics that affect classification shifting and the CEO career horizon. I conducted additional sensitivity tests to explore whether the main findings are sensitive to such unknown industry-level factors. Appendix C Table C2 provides a summary, ranked by mean value, of classification shifting per industry using the Fama and French (1997) industry classification. First, I excluded the top three industries with the highest unexpected core earnings from this overview and retested the baseline regression model. Second, I run the model using the full sample based on the four-digit SIC code with robust standardized errors clustered at industry level. Next, the OLS regression with standard errors clustered at industry level was used in an industry-year panel data set. Fourth, I used a firm-level random-effects model with industry fixed effects. Finally, I used industry-level random-effects with standard errors clustered at industry level.

The results of this industry sensitivity analysis are presented in Table 5 models (1) to (5). The positive and significant %SI values (range 0.139 to 0.180, p < 0.01) and negative and significant

(18)

intercept coefficients (range -0.007 to -0.008, p < 0.01) confirm that my findings are robust for all five industry sensitivity tests.

TABLE 5

Industry sensitivity tests Dependent variable UE_CE Excluding top 3 classification shifting industries Firm fixed Industry clustering Industry fixed Industry clustering Industry fixed Firm random Industry random Industry clustering Model (1) (2) (3) (4) (5) %SI 0.139*** 0.180*** 0.160*** 0.180*** 0.155*** (0.041) (0.054) (0.040) (0.057) (0.037) CEOHORIZON -0.004 -0.006*** -0.004** -0.006*** -0.004** (0.001) (0.002) (0.002) (0.002) (0.002) %SI x CEOHORIZON -0.008*** -0.008*** -0.007*** -0.008*** -0.007*** (0.001) (0.001) (0.001) (0.001) (0.001) Size -0.012*** -0.010*** 0.001** -0.010*** 0.002*** (0.002) (0.003) (0.000) (0.003) (0.000) Leverage -0.001 -0.001 0.000 -0.001 0.000 (0.001) (0.001) (0.001) (0.001) (0.001) MBV 0.000 0.000 -0.000 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) OCF 0.068*** 0.082** 0.082** 0.082** 0.088** (0.026) (0.035) (0.037) (0.037) (0.042) ROA 0.119*** 0.093*** 0.077*** 0.093*** 0.065** (0.033) (0.020) (0.025) (0.021) (0.032) _cons 0.072*** 0.062*** -0.018*** 0.066*** -0.022*** (0.019) (0.016) (0.006) (0.017) (0.006)

Firm-fixed effects Yes Yes No No No

Firm-random effects No No No Yes No

Industry-fixed effects No No Yes Yes No

Industry-random effects No No No No Yes

Year-fixed effects Yes Yes Yes Yes Yes

Number of Observations 13,934 15,275 15,275 15,275 15,275

R-squared 0.026 0.022 0.026 0.139 0.026

UE_CEt = α0 + α1%SIt + α2CEOHORIZONt + α3%SIt x CEOHORIZONt + α4Size + α5Leverage + α6MBV + α7OCF

+ α8ROA + (firm and year dummies) + εt

See Appendix B Table B2 for variable definitions and calculations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on ordinary least squares regression with robust standard errors adjusted for heteroskedasticity. Standard errors are in parenthesis. Significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

4.3 Robustness checks

I find my main findings to be robust when comparing alternative model variables and regression technique in a range of robustness checks for which the results are included in Table 6.

Validity McVay’s 2006 model: variables

To address the concerns of Barua and Cready (2008) and Fan et al. (2010) I perform three kinds of robustness checks. These authors question whether the McVay’s method of estimating CE, namely the specification of Eq. (1), might create bias in estimating classification shifting.

(19)

Therefore, following Haw et al. (2011) and Alfonso et al. (2015) I repeat my main analysis using three tests of the McVay model to determine if my main results hold.

I extract ACCRUALS (lagged and current) from Eq. (1)10 for the first two tests. This to address Barua and Cready (2008) and Fan et al.’s (2010) concern that inclusion of special items accruals creates model bias. Then for the first test, I re-estimate classification shifting using a subsample of firms with positive core earnings as McVay (2008) shows empirical evidence of classification shifting under these conditions. Second, I use Behn et al. (2013) and Zalata et al. (2018) classification shifting studies that use the full sample to re-estimate the model. Third, I re-estimate McVay’s Eq. (1) replacing total (lagged and current) ACCRUALS with Working Capital Accruals (WCA)11, 12. This addresses Athanasakou et al. (2009) concern that ACCRUALS as used include depreciation expenses and special item accruals, which may introduce bias. The results of the robustness check are shown in Table 6 (1), (2) and (3) models. The positive and significant %SI values (range 0.077 to 0.195, p < 0.01) and negative and significant intercept coefficients (-0.010, p < 0.01) are qualitatively comparable to the primary analysis outcomes. Consequently, the results after the elimination of accruals verify that model bias does not drive my findings on classification shifting. In addition, the results of the robustness check with working capital accruals suggest that my main findings are not sensitive to McVay’s (2006) model’s accrual definition.

Validity McVay’s 2006 model: regression method

I find that CEOs with a relatively short career horizon are less likely to engage in classification shifting than CEOs with a long career horizon. This result is based on the McVay method of excluding firm i when the Eq. (1) coefficients per industry, per firm and fiscal year are calculated. Some research on classification shifting (for example Zalata et al. 2018) did not exclude firm i when determining the Eq. (1) coefficients, resulting in industry-level rather than firm-level coefficients. This may therefore have an effect on UE_CE. Therefore, I repeat my main analysis by not excluding firm i in the Eq. (1) coefficient. Table 6 model (4) presents the outcome. The positive and significant %SI value of 0.131 at the 1% level and the negative and significant intercept coefficient of -0.007 at the 1% level are qualitatively similar to the main analysis. Thus, suggesting that the technique of regression to exclude firm i is not per se essential.

10_{The revised Eq (1) is as follows: CE}

t = β0 + β1CEt-1 + β2ATOt + β5ΔSALESt + β6NEG_ΔSALESt + εt

11_{Definition WCA: increase in accounts receivable + increase in inventory – decrease in accounts payable –}

decrease in income taxes payable + increase in other current net assets (Cheng and Thomas, 2006).

12_{The revised Eq (1) is as follows: CE}

t = β0 + β1CEt-1 + β2ATOt + β3WCAt-1 + β4WCAt + β5ΔSALESt +

(20)

TABLE 6

Robustness checks on validity McVay’s 2006 model: accruals and regression method Dependent variable UE_CE Excluding

accruals with positive earnings firms only Excluding accruals with full sample Accruals replaced by working capital accruals full sample Regression Eq. (1) by not excluding firm i Model (1) (2) (3) (4) %SI 0.077*** 0.190*** 0.195*** 0.131*** (0.022) (0.036) (0.037) (0.028) CEOHORIZON -0.004*** -0.005*** -0.005*** -0.004*** (0.001) (0.002) (0.002) (0.001) %SI x CEOHORIZON -0.010*** -0.010*** -0.010*** -0.007*** (0.021) (0.001) (0.001) (0.001) Size -0.009*** -0.012*** -0.012*** -0.008*** (0.002) (0.002) (0.002) (0.002) Leverage 0.000 0.000 0.000 -0.000 (0.000) (0.001) (0.001) (0.001) MBV 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) OCF -0.053*** -0.037 -0.037 0.066*** (0.014) (0.024) (0.025) (0.018) ROA 0.163*** 0.276*** 0.281*** 0.037* (0.016) (0.032) (0.033) (0.022) _cons 0.067*** 0.084*** 0.082*** 0.050*** (0.012) (0.018) (0.019) (0.015)

Firm-fixed effects Yes Yes Yes Yes

Year-fixed effects Yes Yes Yes Yes

Number of Observations 14,428 15,275 15,275 15,275

R-squared 0.037 0.051 0.051 0.021

I estimate the parameters using the following regression equation for models (1), (2), (3):

UE_CEt = α0 + α1%SIt + α2CEOHORIZONt + α3%SIt x CEOHORIZONt + α4Size + α5Leverage + α6MBV +

α7OCF + α8ROA + (firm and year dummies) + εt

and for model (4) I use the equation:

UE_CEi,t = α0 + α1%SIi,t + α2CEOHORIZONt + α3%SIi,t x CEOHORIZONt + α4Size + α5Leverage + α6MBV

+ α7OCF + α8ROA + (firm and year dummies) + εt

See Appendix B Table B2 for variable definitions and calculations. All variables are winsorized at 1 percent and 99 percent. Estimates are based on ordinary least squares regression with robust standard errors adjusted for

heteroskedasticity and within-firm clustering. Standard errors are in parenthesis. Significance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively.

(21)

5. Discussion and conclusion

A large body of literature has documented that managers tend to manipulate earnings through accruals management and even real activities (Cohen et al., 2008; Zang, 2012). A nascent stream of this earnings management literature has provided some evidence on classification shifting (McVay, 2006; Fan et al., 2010; Haw et al., 2011; Behn et al., 2013). However, much of this literature treats CEOs as a homogenous group (Zalata et al. 2018). This, despite that Francis et al. (2008), Ali and Zhang (2015), and Huang et al. (2012) find that CEO reputation, CEO tenure, and CEO age affect earnings management via accruals management and manipulation of real activities. In this context, this study is novel in that I do not treat CEOs as a homogeneous group. Instead, I examine the effect of CEO career horizons on classification shifting. Classification shifting is a relatively low-cost and more flexible earnings management tool than accrual and real earnings management (McVay, 2006; Behn et al., 2013; Alfonso et al., 2015; Zalata, 2018). Thus, conceivable less risky than other tools, that lead risk-averse executives to adopt classification shifting. Prior research demonstrated that CEOs with long tenure and CEOs with age near retirement are more risk-averse and focused on maintaining their status quo (Hambrick and Fukutomi, 1991; McClelland et al., 2010; Simsek, 2007) than their counterparts who are young, have shorter tenure, and are far from retirement age. Consequently, I argue that both CEO tenure and CEO age could affect the CEOs risk preference, which in turn affects corporate decisions in using classification shifting. This study captures both CEO tenure and age in the CEO career horizon, a compound measure adjusted to the industry. I hypothesize that CEOs with a relative short career horizon (i.e. long tenure and age near retirement relative to their industry peers) are more engaged in classification shifting than CEOs with a relative long career horizon.

Using U.S. firms for the sample period 2003–2017, 15,275 firm-year observations, I find that there is a statistically significant relationship with the relative duration of CEO career horizons and classification shifting. My hypothesis is not confirmed as I find that classification shifting is less prevalent among firms with CEOs with a relatively short career horizon compared to firms with CEOs with a longer career horizon. This finding complements the study of Ali and Zhang (2015), who find less pronounced evidence of accrual and real earnings management in the later years of CEOs’ service compared to their early years of service. My finding is also consistent with Zalata et al. (2018) who find that more risk-averse executives engage significantly less in classification shifting. The results indicate that while CEOs with a relative short career horizon are more risk-averse and focused on maintaining their status quo, they do not consider classification shifting as a viable earnings management tool, suggesting that the behavioral attributes of these CEOs might have outweighed the relatively low-risk character of classification shifting as a tool to manage earnings.

This research has implications for investors, auditors and board of directors. For example, users of financial statements (i.e. analysts and investors) as they use disclosed pro-forma earnings in their investment decisions (Chen, Krishnan and Peyzner, 2012) and for auditors who assess and approve these statements. They need to consider the relationship of CEO career horizon and classification shifting. In particular, that classification shifting is less prevalent among firms with CEOs with a relatively short career horizon as opposed to their counterparts with a relative longer career horizon. Furthermore, the governance role of the board of directors as well as their decisions on CEO appointments may be affected whilst recognizing the CEO’s career horizon association with classification shifting.

The study outcome may be subject to limitations and provide opportunities for future research. First, prior research find that the use of earnings management techniques is influenced by the governance environment (Cheng, Lee, and Shevlin, 2015; Zalata et al., 2018). The sample

(22)

period used in this study covers the post-SOX13 period. The subsequent strong legal environment may have affected the behavioral attitude of CEOs with a short career horizon towards the low-risk nature of classification shifting. Therefore, future research can enhance my findings by replicating my study using the pre-SOX period. If the findings show that classification shifting is more prevalent in firms with CEOs with a relative short career horizon than this demonstrates that the strong legal environment affects the CEOs behavior towards classification shifting. Second, this study is limited to CEOs rather than other executives. Jiang, Petroni and Wang (2010) suggest that motivations of CEOs and CFOs to engage in earnings management are inclined to be different. Future research may replicate my study by focusing on the career horizon of other executives. Finally, as we know little about classification shifting whereby CEOs are not treated as a homogeneous group, there are many directions for future research to develop the classification shifting literature for example the impact of cultural factors or political preference of CEOs et cetera.

(23)

References

Abernathy, J. L., Beyer, B., & Rapley, E. T. (2014). Earnings management constraints and classification shifting. Journal of Business Finance and Accounting, 41(5–6): 600–626. Alfonso, E., Cheng, C. A., & Pan, S. (2015). Income classification shifting and mispricing of core earnings. Journal of Accounting, Auditing and Finance. Retrieved from https://doi.org/10.1177/0148558X15571738.

Ali, A., & Zhang, W. (2015). CEO tenure and earnings management. Journal of Accounting and Economics, 59(1): 60–79.

Antia, M., Pantzalis, C., & Park, J. C. (2010). CEO decision horizon and firm performance: An empirical investigation. Journal of Corporate Finance, 16(3): 288–301.

Athanasakou, V., Strong, N. C., & Walker, M. (2009). Earnings management or forecast guidance to meet analyst expectations? Accounting and Business Research, 39: 3–35.

Baber, W. R., Fairfield, P. M., & Haggard, J. A. (1991). The effect of concern about reported income on discretionary spending decisions: the case of research and development. The Accounting Review, 66: 818–829.

Badertscher, B. (2011). Overvaluation and the choice of alternative earnings management mechanisms. The Accounting Review, 86(1): 491–518.

Barker, V. L., & Mueller, G. C. (2002). CEO characteristics and ﬁrm R&D spending. Management Science, 48(6): 782–801.

Barnea, A., Ronen J., & Sadan, S. (1976). Classificatory smoothing of income with extraordinary items. The Accounting Review, 51(1): 110–122.

Barua, A., & Cready, W. (2008). Classification shifting and special items. Evidence of earnings management or research design consequence. SSRN Electronic Journal. Retrieved from http://dx.doi.org/10.2139/ssrn.1139231.

Barua, A., Davidson, L. F., Rama, D. V., & Thiruvadi, S. (2010a). CFO gender and accruals quality. Accounting Horizons, 24(1): 25–39.

Barua, A., Lin, S., & Sbaraglia, A. S. (2010b). Earnings management using discontinued operations. Accounting Review, 85(5): 1485–1509.

Behn, B. K., Gotti, G., Herrmann, D., & Kang, T. (2013). Classification shifting in an international setting: investor protection and financial analysts monitoring. Journal of international accounting research, 12(2): 27–50.

Bruns, W. J., & Merchant, K. A. (1990). The dangerous morality of managing earnings. Management Accounting, 72(2): 22–25.

Carpenter, M. A., Pollock, T. G., & Leary, M. M. (2003). Testing a model of reasoned risk-taking: governance, the experience of principals and agents, and global strategy in high-technology IPO ﬁrms. Strategic Management Journal, 24(9): 803–820.

(24)

Chen, L., Krishnan, G., & Pevzner, M. (2012). Pro-forma disclosures, audit fees, and auditor resignations. Journal of Accounting and Public Policy, 31: 237–257.

Cheng, C. S., & Thomas, W. (2006). Evidence of the abnormal accrual anomaly incremental to operating cash flows. The Accounting Review, 81: 1151–1167.

Cheng, Q., Lee, J., & Shevlin, T. (2015). Internal governance and real earnings management. The Accounting Review, 91(4), 1051–1085.

Cohen, D., Dey, A., & Lys, T. (2008). Real and accrual-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review, 83: 757–878.

Cyert, R. M., & March, J. G. (1963). A behavioral theory of the firm. Englewood Cliffs, NJ: Prentice Hall.

Davis, A. (2002). The value relevance of revenue for Internet firms: Does reporting grossed-up or barter revenue make a difference? Journal of Accounting Research, 40(2): 445–477.

Dechow, P. M., & Sloan, R. G. (1991). Executive incentives and the horizon problem—an empirical-investigation. Journal of Accounting and Economics, 14(1): 51–89.

DeAngelo, H., DeAngelo, L., & Skinner, D. J. (1994). Accounting choice in troubled companies. Journal of Accounting and Economics, 17(1–2): 113–143.

Dohoo, I., Ducrot, C., Fourichon, C., Donald, A., & Hurnik, D. (1997). An overview of techniques for dealing with large numbers of independent variables in epidemiologic studies. Preventive Veterinary Medicine, 29(3): 221–239.

Ewert, R., & Wagenhofer, A. (2005). Economic effects of tightening accounting standards to restrict earnings management. The Accounting Review, 80(1): 101–124.

Fama, E. G., & French, K. (1997). Industry costs of equity. Journal of Financial Economics, 43: 153–193.

Fan, Y., Barua, A., Cready, W. M., & Thomas, W. (2010). Managing earnings using classification shifting: Evidence from quarterly special items. The Accounting Review, 85: 1303–1323.

Finkelstein, S., & Hambrick, D. C. (1988). Chief executive compensation—A synthesis and reconciliation. Strategic Management Journal, 9: 543–558.

Finkelstein, S., Hambrick, D. C., & Cannella, B. (2009). Strategic leadership: Theory and research on executives, top management teams, and boards. New York: Oxford University Press.

Francis, J., Huang, A. H., Rajgopal, S., & Zang, A. Y. (2008). CEO reputation and earnings quality. Contemporary Accounting Research, 25(1): 109–147.

Does executive career horizon influence classification shifting?

Does executive career horizon influence

classification shifting?

Does executive career horizon influence

classification shifting?

Table of contents

1.

Introduction

2.

Prior literature and hypothesis development

3.

Research design and summary statistics

4.

Results

5.

Discussion and conclusion

References