CLASSIFICATION SHIFTING IN AN
INTERNATIONAL CONTEXT: THE EFFECT OF
NATIONAL CULTURE
Master thesis, MSc Accountancy & Controlling Track: Accountancy & Controlling
University of Groningen, Faculty of Economics and Business
June 22, 2020 MILOU PAUS Student number: 2911345 Paterswoldseweg 18c 9726 BD Groningen Tel: +31 (0)631299848 E-mail: m.e.g.m.paus@student.rug.nl Supervisor Prof. Dr. C. K. Hoi Word count: 11,199
Acknowledgement: I thank Prof. Dr. C. K. Hoi for his guidance in the process of writing this thesis, and for his helpful and valuable comments on earlier drafts of this thesis.
1
CLASSIFICATION SHIFTING IN AN
INTERNATIONAL CONTEXT: THE EFFECT OF
NATIONAL CULTURE
ABSTRACT: Prior studies demonstrate that managers in the U.S. and other countries
opportunistically shift core expenses to inflate core earnings. Building on this research, I investigate the effects of national culture on classification shifting practices. In particular, I study the influence of individualism and uncertainty avoidance on classification shifting using a sample of 20,205 firms with 101,792 firm-years across 73 countries. I find that individualism is positively associated with classification shifting. This relation is robust to controlling for formal institutions, the level of economic freedom, other aspects of culture, clustering standard errors by country and controlling for country overrepresentation. However, I find no significant association between uncertainty avoidance and classification shifting. Overall, these findings indicate that certain national cultural dimensions, such as individualism, affect firm-level classification shifting.
Key words: Earnings management; Classification shifting; Informal institutions; Culture; Individualism; Uncertainty avoidance.
2
Table of Contents
1. Introduction ... 3
2. Prior Literature and Hypothesis Development ... 5
2.1. Earnings Management Tools ... 5
2.2. The Low Costs of Classification Shifting ... 6
2.3. Agency Theory versus Institutional Theory ... 7
2.4. Individualism ... 9
2.5. Uncertainty Avoidance ... 12
3. Methodology ... 15
3.1. Data Collection and Sample ... 15
3.2. Measurement of Classification Shifting ... 15
3.2. Measurement of Individualism and Uncertainty Avoidance ... 16
3.3. Control Variables ... 17
3.4. Regression Model ... 18
4. Results ... 19
4.1. Descriptive Statistics ... 19
4.2. Unexpected Core Earnings ... 23
4.3. Primary Tests ... 24
4.4. Sensitivity Tests ... 26
5. Discussion and Conclusion ... 32
6. References ... 35
7. Appendices ... 49
Appendix A: Sample Selection... 49
Appendix B: Description of Variables ... 49
Appendix C: Descriptives by Country ... 52
3 1. Introduction
Earnings management, the “purposeful intervention [by management] in the external financial reporting process with the intent of obtaining private gain” (Schipper, 1989, p.92), is an important research construct in the accounting literature. To date, earnings management is studied in the form of accrual-based earnings management, real earnings management and the focus of this paper: classification shifting (Abernathy, Beyer & Rapley, 2014). First introduced by McVay (2006), classification shifting pertains to the opportunistic shift of core expenses and revenues in a company’s income statement, in order to present a more favorable position of operating profit. More specifically, management engages in classification shifting because the public bases its decisions primarily on core profitability in lieu of (unaltered) bottom-line GAAP profit (Fan, Thomas & Yu, 2019). The existence of classification shifting has been widely validated in a single-country setting (e.g. Fan, Barua, Cready & Thomas, 2010; Alfonso, Cheng & Pan, 2015; Zalata & Roberts, 2016, 2017). However, cross-country research on classification shifting is extremely scarce (Behn, Gotti, Herrmann & Kang, 2013). This, whereas Kury (2007) and Gray, Kang, Lin and Tang (2015) signify that country-level institutions essentially establish the opportunity and incentives for managerial earnings manipulation. More specifically, firms are embedded within a national institutional environment (Ioannou & Serafeim, 2012), that comprehends norms, values, policies and regulations (Kostova, 1997). This institutional environment serves as a foundation for organizational activities (Holm, 1995), including earnings management practices (Kury, 2007). As such, it is important to study classification shifting in light of broader country-level institutions (Behn et al., 2013).
Only three existing studies consider classification shifting in an international context (Haw, Ho & Li, 2011; Behn et al., 2013; Baik, Cho, Choi & Lee, 2016) and all investigate the effect of formal institutions, namely the rule of law, investor protection and IFRS adoption. However, these studies have largely ignored the influence of informal institutions in general, and culture in particular. Nevertheless, culture is a critical country-level determinant of firm-level earnings management (Doupnik, 2008). As such, this paper provides novelty in that it studies the influence of national culture on classification shifting behavior. In particular, I investigate two specific aspects of culture that are more likely to affect decision-making on accounting practices and reporting (Zarzeski, 1996; Hope, 2003): individualism and uncertainty avoidance. On the one hand, individualism and uncertainty avoidance lead to risk-averse and law-abiding behavior, and thereby potentially mitigate earnings management (Getz and Volkema, 2001;
4
Han, Kang, Salter & Yoo, 2010). On the other hand, individualism facilitates and motivates the pursuit of opportunistic self-interest, and uncertainty avoidance could materialize in a preference for consistent and smooth earnings, which ultimately results in higher levels of earnings management (Nabar & Boonlert-U-Thai, 2007; Lewellyn & Bao, 2017). In light of these conflicting arguments, how country-level individualism and uncertainty avoidance influence firm-level classification shifting is a question of empirical interest.
Based on a sample of 101,792 firm-year observations originating from 73 countries, I find robust and statistically significant evidence that classification shifting is more pervasive among firms in individualist nations. In essence, individualism induces risk-averse behavior and lends credence to, as well as facilitates the pursuit of self-interest (Sharp & Salter, 1997; Wang & Fischbeck, 2008). Given that classification shifting is a relatively low risk earnings management technique (Abernathy et al., 2014), it is a viable tool to pursue self-interest. Moreover, my findings show that uncertainty avoidance exhibits a positive, but not significant relationship with classification shifting. Potentially because the ability of uncertainty avoidance to determine behavior has decreased over time (Smith, Dugan & Trompenaars, 1996). These findings add to the accounting literature in various respects. First, I directly expand the literature on classification shifting as I provide novel evidence on individualism as a classification shifting determinant. More broadly, I contribute to the debate on the relationship between culture and business developments as initiated by Kwok and Tadesse (2006), and I endorse Gray’s (1988) propositions by showing that individualism determines accounting classification-related decision-making. Last, I substantiate Sharp and Salter’s (1997) conclusion that individualism motivates self-interested behavior. That is, I find that this heightened self-interest materializes in higher levels of classification shifting. Moreover, I complement more recent findings by Lewellyn and Bao (2017), who show that individualism and the related tendency to pursue self-interest result in more accrual-based earnings management. I find that this also applies to classification shifting.
The remainder of this paper is organized as follows. In the next section, I provide a comprehensive analysis of the literature earnings management, institutions and culture. Ultimately, I relate the literature on individualism and uncertainty avoidance to classification shifting to establish testable hypotheses. Next, I expound on the sampling procedure, measurement methods and regression models established to test my hypotheses. Thereafter, I report the results of the hypotheses tests and submit the findings to several sensitivity checks. Lastly, I conclude my study and provide a discussion of the results and research limitations.
5
2. Prior Literature and Hypothesis Development
This chapter outlines a review of the literature on earnings management in general, and classification shifting in particular. I elaborate on the relative costs of classification shifting, and explain the salience of institutions in determining incentives for classification shifting. Finally, I develop hypotheses for the relationship between classification shifting and two institutional variables: individualism and uncertainty avoidance.
2.1. Earnings Management Tools
Prior research reveals three forms of earnings management: accrual-based earnings management, real earnings management and classification shifting (McVay, 2006; Abernathy et al., 2014). First, management manipulates earnings through applying opportunistic judgement and estimation techniques in establishing the value of accruals (Enomoto, Kimura & Yamaguchi, 2015). Examples of such accrual-based earnings management include altering depreciation or amortization write-offs (Roychowdhury, 2006). An alternative tool is real earnings management, which involves the manipulation of actual firm-level operations and activities (Enomoto et al., 2015). Other than accrual-based earnings management, real earnings management directly affects cash flows. Examples include disinvestment in R&D, manipulation of shipment and production schedules, and provision of discounts to accelerate sales (Fudenberg & Tirole, 1995; Healy & Wahlen, 1999; Dechow & Skinner, 2000). Lastly, classification shifting refers to the practice of moving core income statement items to special items in order to present more favorable core earnings (McVay, 2006). More specifically, core earnings refer to the sales generated by core business activities, minus core expenses, i.e. costs of goods sold and selling general and administrative expenses. Special items differ from core earnings in that they are uncommon and non-recurring in nature (McVay, 2006). Thus, with classification shifting, management shifts core expenses that generally persist into future periods, such as marketing expenses, to non-persisting special items1, such as restructuring
costs (Fairfield, Sweeney & Yohn 1996; McVay, 2006).
Classification shifting differs from accrual-based and real earnings management in three ways (McVay, 2006; Behn et al., 2013). First, whereas accrual-based earnings management and real earnings management involve the modification of bottom-line earnings, classification shifting solely affects individual segments of income, and does not impact ultimate bottom-line earnings (Bradshaw & Sloan, 2002; McVay, 2006). Furthermore, even though all three types
1 In line with McVay’s (2006) methodology, I focus on income-decreasing special items in my analysis on
6
of earnings management generally influence expectations of the organization’s future earnings, accrual-based and real earnings management reverse in future periods, whereas classification shifting within the current period’s income statement does not influence future earnings (McVay, 2006; Behn et al., 2013). Lastly, since classification shifting does not affect net income, Nelson, Elliot and Tarpiey (2002) propose that it is less subject to regulatory scrutiny or audit-related investigations, and likewise has a lower risk of litigation than the other two forms of earnings management.
2.2. The Low Costs of Classification Shifting
Following McVay’s (2006) methodology, a multitude of researchers studied both antecedents and implications of classification shifting in within-country and cross-country settings. First, Athanasakou, Strong and Walker (2009), evidence that U.K. firms engage in core earnings manipulation to fulfil earnings expectations, and that investors subsequently reward shifters for fulfilling those expectations. Similarly, Alfonso et al. (2015), demonstrate that U.S. investors significantly overprice shifters’ core earnings because they expect future earnings to be higher when current core earnings are higher. Fan et al. (2010) modified McVay’s (2006) methodology and show that management performs classification shifting to satisfy analyst expectations. Siu and Faff (2013) also employ this modified model and find subsequent evidence that management maximizes equity value by shifting expense classifications before issuing equity. Most importantly, these studies support McVay’s (2006) belief that classification shifting is a deliberate decision of management, and performed because investors base decisions on core, rather than bottom-line, earnings.
However, this decision is affected by the relative costs of each technique, and classification shifting has several advantages compared to accrual-based and real earnings management (Athanasakou et al., 2009; Alfonso et al., 2015). First, accrual-based and real earnings management are rigid tools and reverse in the future, whereas classification shifting is more flexible and does not reverse in the future (Zang, 2012; Behn et al., 2013). Furthermore, classification shifting is subject to less regulatory and investigatory scrutiny than real and accrual-based earnings management (Abernathy et al., 2014; Francis, Hasan & Li, 2016). Nelson et al. (2002) and Zang (2012) show that auditors pay more attention to real and accrual-based earnings management, with which bottom-line earnings are affected. On the contrary, classification shifting is less thoroughly examined, as it solely impacts core, and not bottom-line, earnings (Fan et al., 2010). These findings were supported by Cohen, Dey and Lys (2008), who argue that classifying costs involves judgement and subjectivity. Likewise, from an auditor
7
perspective, it is harder to identify shifting of expense classification. Additionally, accrual-based earnings management faces enhanced regulatory scrutiny, stemming from directives such as the Sarbanes-Oxley Act (Cohen et al., 2008; Alfonso et al., 2015), and international equivalents. Compared to accrual-based earnings management, real earnings management is easier to conceal as it can be disguised as an ordinary business activity (Kothari, Mizik & Roychowdhury, 2016). Nevertheless, real earnings management faces more regulatory scrutiny than classification shifting (Nelson et al., 2002; Huang, Roychowdhury & Sletten, 2019). Moreover, real earnings management often materializes in diminished future-oriented investments, thereby destroying rather than creating value (Chen, Rees & Sivaramakrishnan, 2010). These relatively low costs of classification shifting make it a viable tool to manipulate earnings. However, rather than being produced in a vacuum, the core of all managerial practices, including performing classification shifting, lies in the institutional environment (Lindberg, Campbell & Hollingsworth, 1991; Kostova, 1997; Berthod, 2016). Accordingly, I approach the agency-based incentives for classification shifting in the light of broader institutions.
2.3. Agency Theory versus Institutional Theory
Research on earnings management in general (e.g. Healy & Palepu, 1993; Holthausen, Larker & Sloan, 1995; Teoh, Welch & Wong, 1998), and classification shifting in particular (e.g. Lail, Thomas & Winterbotham., 2014; Joo & Chamberlain, 2017; Fan et al., 2019) has primarily taken an agency orientation. For instance, Lail et al. (2014) show that management opportunistically shifts core segment expenses to other costs to present higher segment profits, and Fan et al. (2019) demonstrate that management opportunistically manipulates expense classification to prevent debt covenant violation. From an agency perspective, management has the ability to realize this due to (a) a distinction between ownership and day-to-day organizational control and (b) information asymmetry (Beatty & Harris, 1999; Lail et al., 2014). That is, management (the agent) is more informed than the firm’s shareholders (the principal) and has the ability to exercise control in the pursuance of self-interested objectives (Jensen & Meckling, 1976; Eisenhardt, 1989). Whereas empirical research supports the notion that executives significantly impact their organizations and activities (Wiersema & Bantel, 1992), their freedom in decision-making is limited because of constraints faced within, and opportunities granted by, external institutions (Crossland & Hambrick, 2011). Thus, the ability to pursue self-interest moves beyond the sole consideration of information asymmetry.
8
Strictly speaking, management faces pressure from its institutional environment to maintain legitimacy (Kury, 2007), understood as “a generalized perception or assumption that the actions of an entity are desirable, proper or appropriate within some socially constructed system of norms, values, beliefs and definitions” (Suchman, 1995: p.574). Institutions thus shape what is perceived as legitimate behavior and thereby set the stage for management to behave in a self-interested manner (Kury, 2007). Institutions can be either informal or formal (North, 1990). Informal institutions are cognitive, and refer to “socially shared rules, usually unwritten, that are created, communicated, and enforced outside officially sanctioned channels” (Helmke & Levitsky, 2004, p.727). On a more concrete level, formal institutions derive their structure from the state (Mantzavinos, 2001), and encompass “rules and procedures that are created, communicated and enforced through channels widely accepted as official” (Helmke & Levitsky, 2004, p.727). Research has addressed formal institutions with regards to classification shifting. Haw et al. (2011) for instance investigate the impact of the rule of law on misclassification of expenses. Behn et al. (2013) demonstrate that classification shifting differs at different levels of investor protection. Additionally, Baik et al. (2016) study how the adoption of IFRS affects classification shifting. Nevertheless, informal institutions, such as Hofstede’s cultural dimensions, have solely been considered with respect to accrual-based and real earnings management (e.g. Doupnik, 2008; Callen, Morel & Richardson, 2011; Paredes & Wheatley, 2017). This, whereas researchers (Gray, 1988; Hope, 2003; Doupnik, 2008) claim that culture is a critical determinant in accounting-related decision-making. In fact, Doupnik (2008) demonstrates that culture is able to explain 49 percent of cross-country differences in earnings smoothing and earnings discretion practices in general. As such, it is important to consider culture as an opportunity-granting and incentivizing device for classification shifting in particular.
2.3.1. An informal institutional perspective: culture
As an informal institution, culture can shape individual behavior, including managerial behavior (Hambrick & Finkelstein, 1987; North, 1990;Crossland & Hambrick, 2011). That is, culture provides incentives and justification for behavior congruent with the values, beliefs and principles existing within a country (Licht, Goldschmidt & Schwartz, 2005, 2007; Deephouse, Newburry & Solemani, 2016). Hofstede (1980) identified six cultural dimensions that govern individual behavior: uncertainty avoidance, individualism, power distance, masculinity,
long-9
term orientation and indulgence2. Prior research, however, finds that individualism and
uncertainty avoidance are particularly relevant to accounting-related decision-making. For instance, Hope (2003) shows that individualism causes those involved with accounting to engage more intensively in exercising independent professional judgement and self-control (professionalism), whereas higher levels of uncertainty avoidance are associated with a close adherence to rules and regulations rather than independent judgement (statutory control). Askary, Yazdifar and Askarany (2008) and Perera and Matthews (1990) demonstrate that individualists deal more flexibly with accounting practices (flexibility), and that uncertainty avoidant nations have a preference for more uniform and universal application of accounting practices (uniformity). Moreover, Gray and Vint (1995), Zarzeski (1996) and Hope (2003) find that individualists disclose accounting information more openly and extensively (transparency), whereas higher levels of uncertainty avoidance are associated with more confidentiality in reporting on accounting practices and results (secrecy). Furthermore, Perera and Matthews (1990) and Askary et al. (2008) demonstrate that individualists present accounting numbers more optimistically (optimism), and that uncertainty avoidant nations employ more rigid and prudent accounting practices (conservatism). These studies validate Gray’s (1988) premise that societal cultural values permeate into accounting values, a notion backed-up by several researchers (Gray & Vint, 1995; Guan, Pourjalali, Sengupta & Teruya, 2005; Doupnik, 2008; Han et al., 2010; Gray et al., 2015). Building on and extending these studies, I focus on the effects of individualism and uncertainty avoidance on classification shifting practices in a cross-country setting.
2.4. Individualism
2.4.1. Individualism and agency-based self-interest
Individualism can be understood as “the extent to which individuals' self-interests are prioritized over the concerns of a group” (McCoy, Galetta & King, 2005, p.213). In highly individualistic cultures, people act primarily in the best interest of themselves and their direct family members (Hofstede, 2001). Furthermore, people behave more autonomously, and value the pursuit of personal ends over the pursuit of group goals (Hofstede, 1980; Triandis, 1994, 2001; Ramamoorthy & Carroll, 1998). As argued by Waterman (1981), the pursuance of self-interest under individualism involves manipulating others to engender private benefits, also referred to as moral hazard. Accordingly, individualism could give rise to agency problems.
2 Long-term versus short-term orientation was added later in Hofstede’s (2001) work. In 2010, a sixth dimension,
10
Sharp and Salter (1997) for instance demonstrate that individualism yields a culture wherein the pursuit of self-interest is the norm and perceived as legitimate behavior. On the contrary, collectivist nations disdain interested decision-making, and contemplate this self-centeredness as illegitimate behavior. Moreover, Sharp and Salter (2001) show that a positive association between individualism and the pursuit of personal ends in the context of cross-border investments even holds across small differences in individualism (U.S. versus Canada). Additionally, individualism proves to demand individually-oriented recruitment, reward, performance appraisal and promotion strategies (Ramamoorthy & Carroll, 1998); demand pay-for-performance based contracts (Schuler & Rogovsky, 1998); require personal involvement in decision-making (Chow, Kato & Merchant, 1996) and result in managerial misbehavior (Han et al., 2010). Thus, individualism lends legitimacy to the pursuit of personal ends, which reflects into firm structure and managerial decision-making.
2.4.2. Individualism and earnings management
From an agency perspective, self-interested management engages in earnings management to manipulate firm profitability in the pursuit of personal benefits (Nabar & Boonlert-U-Thai, 2007; Jiraporn, Miller, Yoon & Kim, 2008; Han et al., 2010). Burgstahler & Dichev (1997), Das and Zhang (2003) and Graham, Harvey and Rajgopal (2005) find evidence of earnings management to avoid negative earnings surprises and to smoothen results. McVay (2006), Athanasakou et al. (2009) and Fan et al. (2010), show that earnings manipulation is performed to meet analysts’ forecasts and benchmarks. Moreover, Cornett, McNutt and Tehranian (2009) demonstrate that performance benchmarks provoke earnings management. Inferring from these findings, management is triggered to engage in earnings management primarily by the desire to benefit oneself. This incentive is likely heightened in individualist countries, where self-interested decision-making is the standard, appropriate and legitimate state of affairs. As such, individualist nations provide room and motivation to engage in self-interested behavior, including earnings management (Lewellyn & Bao, 2017).
By building upon this argument, analysis on international (Doupnik, 2008; Han et al., 2010; Lewellyn & Bao, 2017), European (Gray et al., 2015) and Asia-Pacific (Guan et al., 2005) samples shows that individualism exacerbates accrual-based earnings management, also in a single-GAAP (IFRS) environment (Gray et al., 2015). Albeit evidence on the relationship between individualism and classification shifting does not exist, firms in individualist nations might be more likely to engage in classification shifting, since classification is performed in the pursuit of personal benefits (e.g. McVay, 2006; Fan et al., 2010). Furthermore, I previously
11
established that individualism is associated with more professional judgement, flexibility and optimism in accounting practices (Gray, 1988; Perera & Matthews, 1990; Hope, 2003; Askary et al., 2008). Likewise, one can expect that individualists have more room to exercise opportunistic discretion in classifying expenses, and thus to engage in classification shifting. Yet, studies investigating the role of individualism in terms of accrual-based and real earnings management also present evidence in the other direction. More specifically, Callen et al. (2011), Desender, Castro and De León (2011), and Zhang, Liang and Sun (2013) find that firms embedded in collectivist cultures perform more accrual-based earnings management than individualist-oriented firms. Furthermore, Lyu, Yuen and Zhang (2015) provide international evidence that individualism significantly mitigates the degree of earnings management practices induced by concentrated ownership.
2.4.3. Individualism and risk-avoidance
The diverging results can potentially be explained by the risk-avoiding effect of individualism (Lee & Guven, 2013). Albeit earnings management falls within the guidelines established by generally accepted accounting principles, it is a corruptive and risky activity (Liu, Wei & Xie, 2016). As demonstrated by Lee and Guven (2013) risk preference lies on the root of corruptive activities. More specifically, risk aversion inhibits the level of corruptive and risky activities (Djawadi & Fahr, 2013). With respect to individualism, Wang and Fischbeck (2008) demonstrate that collectivist nations (China) exhibit more risk seeking behavior than individualist nations (US). Similarly, Statman (2008) and Rieger, Wang and Hens (2015) show that individualist nations are more risk averse than collectivist nations. Weber, Hsee and Sokolowska (1998) share this finding, and evidence that collectivists are risk-seeking especially with respect to the financial domain. These findings are grounded in the cushion hypothesis, as proposed and validated by Hsee and Weber (1999), Fan & Xiao (2006), Grable, Joo & Park (2010), Marshall, Huan, Xu & Nam (2011) and Iliashenko (2019). That is, the emphasis on and strength of the social networks in collectivist nations function as a “cushion” against possible financial catastrophes and thus prompt risk-taking activities. As established previously, those engaging in real and accrual-based earnings management face significant risks because of high auditor and regulatory scrutiny (Roychowdhury, 2006; Huang et al., 2019). Thus, if individualism inhibits engagement in risky practices, it could reduce accrual-based and real earnings management. Indeed, the evidence from Callen et al. (2011), Desender et al. (2011) and Zhang et al. (2013) shows a negative linkage between individualism and accrual-based earnings management, which is consistent with this argument. However, classification shifting
12
is a relatively low-risk earnings management technique, compared to accrual-based and real earnings management. As such, individualism could cause managers to use classification shifting as a preferred, low-risk earnings management tool.
Taken together, all of the aforementioned arguments imply a positive association between individualism and classification shifting. First, individualism facilitates, promotes and lends credence to the pursuit of self-interest. If managers perform classification shifting to obtain personal benefits (e.g. McVay, 2006; Fan et al., 2010), individualism could cause management to perform more classification shifting. Second, on condition that classification shifting is a relatively low-risk earnings management technique, individualism, which is associated with risk aversion, might lead managers to shift toward classification shifting as a preferred earnings management tool. Thus, I propose the following testable hypothesis:
Hypothesis 1: In highly individualist nations, firms perform more classification shifting.
2.5. Uncertainty Avoidance
2.5.1. Uncertainty avoidance and uniformity
Hofstede (1984) approaches uncertainty avoidance as the degree to which a society attempts to strengthen security and bypass risk by governing and managing the future. Uncertainties are part of our daily lives. However, countries differ in the extent to which they develop rules and practices that ought to protect them from these uncertainties (Hofstede, 1984). Strongly uncertainty avoidant nations are characterized by an extensive regulatory and legislative framework (Guan & Pourjalali, 2010). Contrarily, in weak uncertainty avoidant nations greater risks are taken, rule sets are scarce, and there is more leeway towards deviation in beliefs and behavior (Johnson & Droege, 2004). Armstrong (1996) and Getz and Volkema (2001) argue that the search for stability that stems from uncertainty avoidance results in a set of detailed laws, norms, policies and rules. With respect to accounting, Hussein (1996) proposes that strongly uncertainty avoidant countries have more detailed rule sets surrounding financial disclosures. That is, accounting is characterized by more uniformity, abundant law and regulation enforcement, as well as more extensive and comprehensive sets of rules (Gray, 1998; Guan & Pourjalali, 2010). Besides, Perera (1989) and Chand, Cummings and Patel (2012) find that uncertainty avoidance results in more conservative and secretive accounting practices. In addition to the existence of ample laws and regulations, the enforcement of rules is stronger in uncertainty avoidant nations (Vitell, Nwachukwu & Barnes, 1993). Likewise, Vitell et al. (1993) and Armstrong (1996) propose that firms in strongly uncertainty avoidant nations are
13
more likely to operate in compliance with existing formal principles than their counterparts in weak uncertainty avoidant nations. Thus, research demonstrates that uncertainty avoidance results in an extensive accounting framework, close adherence to accounting rules and regulation, rigidity and prudence in accounting practices, as well as a uniform application of accounting standards (Gray, 1988; Gray & Vint, 1995).
2.5.2. Uncertainty avoidance and earnings management
Accrual-based and real earnings management are corruptive, risky activities (Liu et al., 2016), and, relative to classification shifting, ample rules and standards exist to avoid real and accrual-based earnings management. Thus, given that uncertainty avoidance results in risk-averse and rule-binding managerial decision-making, it might mitigate accrual-based and real earnings management. This notion is supported in the literature. More specifically, Han et al. (2010) find international evidence that uncertainty avoidance mitigates accrual-based earnings management, since uncertainty avoidance materializes in higher uniformity, regulatory control and high conservatism in accounting and reporting (Gray, 1988). Based on this same premise, Guan et al. (2005) provide Asian-based evidence that firms in strongly uncertainty avoidant nations perform less discretionary accruals manipulation. Kanagaretnam, Lim and Lobo (2011) find that these results hold within the banking industry. Furthermore, Paredes and Wheatley (2017) demonstrate that uncertainty avoidance and the related avoidance of future ambiguity mitigates real earnings management, since real earnings management damages future firm performance.
Though, evidence on the contrary also exists. Nabar and Boonlert-U-Thai (2007), Doupnik (2008), Callen et al. (2011) and Ugrin, Mason and Emley (2017) evidence that strong uncertainty avoidance results in opportunistic earnings smoothing, because individuals from strongly uncertainty avoidant nations prefer consistency in rewards (Schuler, Jackson, Jackofsky & Slocum Jr., 1996). For instance, Dou, Truong and Veeraraghavan (2016) find that stock-owners in strongly uncertainty avoidant nations respond more negatively when confronted with earnings surprises and inconsistency, as opposed to weakly uncertainty avoidant nations. Schmelling (2009) demonstrates that investors in strongly uncertainty avoidant nations typically exhibit herd behavior and overreact when earnings are unexpected. Zouaoui, Nouyirgat and Beer (2010) even go as far as considering uncertainty avoidance as a proxy for investor overreaction. Moreover, Friedman (2007) signifies the importance of job certainty, financial stability and the attainment of expected returns in highly uncertainty avoidant cultures. Thus, uncertainty avoidant management attaches value to stable
14
remuneration and job security. Since a portion of compensation is generally based on stock-value (Baum, Ford & Zhao, 2012), this creates impetus to maintain stable stock prices. Thus, uncertainty avoidance offers considerable incentives to smooth earnings through earnings management.
2.5.3. Uncertainty avoidance and classification shifting
However, no evidence exists surrounding the impact of uncertainty avoidance on classification shifting practices. Albeit research shows that the law-binding, risk-avoiding effect of uncertainty avoidance mostly inhibits real and accrual-based earnings management, this might not necessarily be the case for classification shifting. That is, classification shifting is less regulated than accrual-based and real earnings management, and thus faces less statutory control (Cohen et al., 2008; Zalata & Roberts, 2017). Accordingly, classification shifting is less subject to regulatory and investigatory scrutiny than accrual-based and real earnings management (Abernathy et al., 2014; Francis et al., 2016). Hence, the scarcity of rules and investigations surrounding classification shifting leads me to believe that the rule-binding effect of uncertainty avoidance is less applicable and is not necessarily as relevant for classification shifting. Moreover, as argued by Geiger et al. (2006), earnings management is only viable and beneficial in uncertainty avoidant nations as long as it does not mitigate future options and opportunities. Whereas real and accrual-based earnings management reverse in future periods, classification shifting does not (McVay, 2006). Furthermore, real earnings management has negative effects on future generation of wealth as it generally involves the divestment of forward-looking capital (Chen et al., 2010). Hence, classification shifting is less detrimental to future wealth creation than the other two forms of earnings management. Lastly, classification shifting is a relatively low-risk earnings management tool, and does not face as much regulatory scrutiny as accrual-based and real earnings management. Uncertainty avoidant nations are more conservative and risk-averse (Gray, 1988; Kanagaretnam et al., 2011). Consequently, the lower costs and lower perceived riskiness of classification shifting compared to the other earnings management tools lead me to believe that classification shifting is potentially the preferred tool to perform earnings management with the aim of mitigating uncertainty and smoothing core earnings in high uncertainty avoidant nations. Hence, I hypothesize:
Hypothesis 2: In highly uncertainty avoidant nations, firms perform more classification shifting.
15
3. Methodology
This chapter delineates the data collection procedure, the methods with which I measure the variables, as well as the baseline regression model employed to assess the influence of individualism and uncertainty avoidance on classification shifting.
3.1. Data Collection and Sample
The ensuing empirical analyses are based on a sample of 101,792 firm-years in 20,205 unique firms from 73 countries for the 2010-2018 period. Data is gathered from the Compustat Global database, the Security Daily database, and Hofstede’s 2010 country-level data scores. Following Behn et al. (2013) Fan et al. (2010), McVay (2006), and others, I gather data based on industries as defined by four-digit SIC codes (Fama & French, 1997). In line with Han et al. (2010), Athanasakou et al. (2011), Abernathy et al. (2014) and Gray et al. (2015), I exclude financial organizations (SIC 6000-6999) and utility organizations (SIC 4000-4999), since their reporting and regulatory environments are very distinct (Zalata & Roberts, 2017; Abernathy et al., 2014; Athanasakou et al., 2009). Following McVay (2006), I exclude firm-years with total sales below 1 million US$ to avoid bias from outliers, I exclude consecutive firm-years for which a change in fiscal year-end occurs between the years to ensure comparability, and I require a minimum of 15 observations per industry-fiscal year to assure expected core earnings can be adequately determined. Appendix A, Table A1 summarizes this sample selection procedure.
3.2. Measurement of Classification Shifting
To measure classification shifting empirically, I employ McVay’s (2006) expected core earnings model. McVay (2006) assumes that firms employing classification shifting overstate core earnings, i.e. the sales generated by core business activities minus costs of goods sold and selling, general and administrative expenses, in the year that an income-decreasing special item is recognized. More explicitly, I first model core earnings (𝐶𝐶𝐶𝐶𝑡𝑡), and seek to explain it as a function of firm performance. An explicit oversight of the variables that reflect firm performance and their underlying reasoning is presented in Table 1. Following McVay (2006), I run an ordinary least squares regression on each industry-year based on the underlying model (Eq. 1) to estimate the coefficients with which I ultimately determine expected core earnings:
𝐶𝐶𝐶𝐶𝑡𝑡 = 𝛼𝛼0+ 𝛼𝛼1𝐶𝐶𝐶𝐶𝑡𝑡−1+ 𝛼𝛼2𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡+ 𝛼𝛼3𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡−1+ 𝛼𝛼4𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡+ 𝛼𝛼5∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑡𝑡
+ 𝛼𝛼6𝑁𝑁𝐶𝐶𝑁𝑁∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑡𝑡+ 𝜀𝜀𝑡𝑡
16
Appendix B, Table B1 provides a detailed explanation of the variables under analysis and their measurement methods.
TABLE 1
Variables for measuring expected core earnings (𝑬𝑬(𝑪𝑪𝑬𝑬)𝒕𝒕)
Variable name Purpose References
Lagged core earnings (𝐶𝐶𝐶𝐶𝑡𝑡−1)
Capture earnings persistence over time. McVay (2006) Asset turnover ratio
(𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡)
Capture negative relation between profit margin and core earnings.
Fan et al. (2010), Nissim & Penman (2001)
Lagged operating accruals
(𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡−1)
Capture the low persistence of operating accruals into future core earnings.
Sloan (1996)
Current-year accruals (𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡)
Account for unusual performance that affects core earnings, as reflected in accruals.
DeAngelo, DeAngelo & Skinner (1994)
Change in sales (∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑡𝑡)
Capture sales as a core earnings determinant.
McVay (2006) Slope of changes
(𝑁𝑁𝐶𝐶𝑁𝑁∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑡𝑡)
Capture varying proportional changes in fixed expenses as activities increase and decrease.
Anderson, Banker & Janakiraman (2003)
For a specification on the measurement techniques of the variables delineated in Table 1, I refer to Appendix B, Table B1.
After obtaining the coefficients, I determine firm-level expected core earnings (𝐶𝐶(𝐶𝐶𝐶𝐶)𝑡𝑡). Ultimately, I expect the level of unexpected core earnings (reported (𝐶𝐶𝐶𝐶𝑡𝑡) minus predicted (𝐶𝐶(𝐶𝐶𝐶𝐶)𝑡𝑡) core earnings) in year 𝑡𝑡 to increase with the level of income-decreasing special items in year 𝑡𝑡 under classification shifting. As such, my dependent variable, the unexplained portion of core earnings (𝑈𝑈𝐶𝐶𝐶𝐶𝑡𝑡), is captured by:
𝑈𝑈𝐶𝐶𝐶𝐶𝑡𝑡= 𝐶𝐶𝐶𝐶𝑡𝑡− 𝐶𝐶(𝐶𝐶𝐶𝐶)𝑡𝑡 (2)
3.2. Measurement of Individualism and Uncertainty Avoidance
To assess whether individualism and uncertainty avoidance impact classification shifting, I employ Hofstede’s country-level scores effectual since 2010 (see e.g. Gray et al., 2015). Hofstede’s dimensions have been criticized for representativeness, as they were derived from studying IBM employees, rather than nations. However, Hofstede (1980, 2001) refutes this criticism, and argues that the homogeneity of the subjects’ non-national characteristics allowed him to detect cultural differences with compelling precision and accuracy. Albeit other attempts to operationalize culture have been made (e.g. Schwartz, 1994; House, Hanges,
17
Javidan, Dorfman & Gupta, 2004), to date no study has been recognized as a successor of Hofstede’s scores (Doupnik, 2008). More importantly, Smith and Bond (1999) and Kirkman, Lowe and Gibson (2006) conclude that all other attempts to operationalize culture have supported Hofstede’s (1980; 2001) work, rather than refuted it. The vast majority of the research on culture and earnings management has employed Hofstede’s (1980; 2001) work to operationalize culture (e.g. Doupnik, 2008; Han et al., 2010; Desender et al., 2011; Zhang et al., 2013; Gray et al., 2015). Moreover, employing Hofstede’s dimensions for this research has the added benefit that it has been theoretically (Gray, 1988) and empirically (Doupnik & Tsakumis, 2004; Gray & Vint, 2005; Tsakumis, 2007) linked to accounting practices. Building on these arguments, I employ Hofstede’s uncertainty avoidance and individualism dimensions in shedding light on classification shifting. Since culture is a rather time-invariant concept, the country-level scores are presumed to remain constant over time (see e.g. Gray & Vint, 1995; Jaggi & Low, 2000). The indices range from 0-1003, where 0 is highly collectivist (weakly uncertainty avoidant) and 100 highly individualist (strongly uncertainty avoidant). I base the country-level scores on the country wherein the firm is headquartered (see e.g. Kollmann, Kuckertz & Breugst, 2009).
3.3. Control Variables
I control for country, firm-level and industry effects associated with discretion in earnings, as proposed by McVay (2006), Desender et al. (2011); Barua, Lin and Sbaraglia (2010) and Han et al. (2010) without predicting a direction. First, I control for gross national product (GNP), since it determines ownership, financing, and payout structures of firms and thereby affects the propensity to perform earnings management (Leuz, Nanda & Wysocki., 2003; Desender et al., 2011). Second, I control for firm size as large firms face more external monitoring, which affects its opportunity to engage in earnings management in the first place (Han et al., 2010). Third, I control for leverage, as levels of debt affect a firm’s tendency and potential to engage in earnings management (De Angelo et al., 1994; Saleh & Ahmed, 2005). Fourth, I control for book-to-market value, a proxy for firm-level growth (Skinner & Sloan, 2002), which affects management’s incentive to manipulate earnings. Furthermore, I control for return on assets and operating cash flows, as Barua et al. (2010) show that these variables significantly impact classification shifting practices. Lastly, I will include industry, countryand
3 A 2014 re-measurement of Hofstede’s cultural indices led to an uncertainty avoidance of 112 in Greece, 104 in
Portugal, 101 in Guatemala and 100 in Uruguay. Untabulated results do not change when I run the analysis with these values.
18
year fixed effects4 to control for imperceptible industry, year and country effects (Han et al.,
2010; Behn et al., 2013). A detailed explanation of all control variables and their measurement methods can be found in Appendix B, Table B1.
3.4. Regression Model
Following Behn et al. (2013) and Fan et al. (2010), I employ McVay’s (2006) core earnings-level model. I first analyze the direct effect of special items on classification shifting without controls, because there are already many variables included in establishing unexpected core earnings (McVay, 2006). I do include industry, year and country fixed effects for validity purposes5. This results in the following regression model:
𝑈𝑈𝐶𝐶𝐶𝐶𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1%𝑆𝑆𝑆𝑆𝑡𝑡+ (𝐴𝐴𝑐𝑐𝐴𝐴𝑐𝑐𝑡𝑡𝐴𝐴𝑐𝑐, 𝑖𝑖𝑐𝑐𝑖𝑖𝐴𝐴𝐴𝐴𝑡𝑡𝐴𝐴𝑐𝑐 𝐴𝐴𝑐𝑐𝑖𝑖 𝑐𝑐𝑆𝑆𝐴𝐴𝐴𝐴 𝑓𝑓𝑖𝑖𝑓𝑓𝑆𝑆𝑖𝑖 𝑆𝑆𝑓𝑓𝑓𝑓𝑆𝑆𝐴𝐴𝑡𝑡𝐴𝐴) + 𝜀𝜀𝑡𝑡 (3)
First, special items scaled by sales6 (%𝑆𝑆𝑆𝑆𝑡𝑡) comprehend income-decreasing special items. More specifically, %𝑆𝑆𝑆𝑆𝑡𝑡 is calculated by multiplying special items with -1, and dividing it by sales when special items are decreasing. When special-items are income-increasing, special items equal zero (McVay, 2006; Behn et al., 2013). For a clear specification on the exact measurement method, I refer to Appendix B, Table B1. The relationship between %𝑆𝑆𝑆𝑆𝑡𝑡 and 𝑈𝑈𝐶𝐶𝐶𝐶𝑡𝑡 is the central point of interest, and represents presence of classification shifting.
When 𝛽𝛽1 is positive and significant, I conclude that management performs classification shifting to present higher core earnings (McVay, 2006; Behn et al., 2013). More specifically, a positive 𝛽𝛽1 implies that firms shifted expenses to special items, leading to an increase in the level of unexpected core earnings (Eq. 2). Thus, I expect a positive and significant sign for 𝛽𝛽1. When 𝛽𝛽1 is negative and significant, special items decrease unexpected core earnings. As argued by Behn et al. (2013) bad performance is likely the reason for a negative significant 𝛽𝛽1. That is, organizations with large negative special-items are generally also organizations that perform poorly, i.e. perform below expected core earnings.
To investigate the impact of individualism (𝑆𝑆𝐼𝐼𝐼𝐼) and uncertainty avoidance (𝑈𝑈𝐴𝐴𝑆𝑆) on classification shifting, I run my regression model with the controls established earlier (GNP (𝑁𝑁𝑁𝑁𝐺𝐺𝑡𝑡), firm size (𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶𝑡𝑡), leverage (𝐿𝐿𝐶𝐶𝐼𝐼𝑡𝑡), return on assets (𝑅𝑅𝐴𝐴𝐴𝐴𝑡𝑡), cash from
4 My main analysis does not include firm-level fixed effects because cultural dimensions are firm fixed effects,
causing them to be dropped from analysis. However, special items, and the interactions with individualism and uncertainty avoidance remain, and the coefficients and significance resemble those in the main regression model.
5 I include industry, year and country fixed effects as failure to control for these influences in a wide cross-country
sample with panel data could potentially result in erroneous conclusions (MacKay & Philips, 2005; Fischer, 2010).
19
operations (𝐶𝐶𝐶𝐶𝐴𝐴𝑡𝑡) and book-to-market value (𝐵𝐵𝐵𝐵𝐼𝐼𝑡𝑡)), and country7, industry and year fixed
effects. This allows me to control for firm-level and country-level performance and development. The regression model looks as follows:
𝑈𝑈𝐶𝐶𝐶𝐶𝑡𝑡 = 𝛾𝛾0+ 𝛾𝛾1%𝑆𝑆𝑆𝑆𝑡𝑡+ 𝛾𝛾2𝑆𝑆𝐼𝐼𝐼𝐼 + 𝛾𝛾3𝑈𝑈𝐴𝐴𝑆𝑆+ 𝛾𝛾4𝑆𝑆𝐼𝐼𝐼𝐼 × %𝑆𝑆𝑆𝑆𝑡𝑡+ 𝛾𝛾5𝑈𝑈𝐴𝐴𝑆𝑆 × %𝑆𝑆𝑆𝑆𝑡𝑡+
𝛾𝛾6𝑁𝑁𝑁𝑁𝐺𝐺𝑡𝑡+ 𝛾𝛾7𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶𝑡𝑡+ 𝛾𝛾8𝐿𝐿𝐶𝐶𝐼𝐼𝑡𝑡+ 𝛾𝛾9𝑅𝑅𝐴𝐴𝐴𝐴𝑡𝑡+ 𝛾𝛾10𝐶𝐶𝐶𝐶𝐴𝐴𝑡𝑡+ 𝛾𝛾11𝐵𝐵𝐵𝐵𝐼𝐼𝑡𝑡+
(𝐴𝐴𝑐𝑐𝐴𝐴𝑐𝑐𝑡𝑡𝐴𝐴𝑐𝑐, 𝑖𝑖𝑐𝑐𝑖𝑖𝐴𝐴𝐴𝐴𝑡𝑡𝐴𝐴𝑐𝑐 𝐴𝐴𝑐𝑐𝑖𝑖 𝑐𝑐𝑆𝑆𝐴𝐴𝐴𝐴 𝑓𝑓𝑖𝑖𝑓𝑓𝑆𝑆𝑖𝑖 𝑆𝑆𝑓𝑓𝑓𝑓𝑆𝑆𝐴𝐴𝑡𝑡𝐴𝐴) + 𝜀𝜀𝑡𝑡
(4)
As before, a positive estimate on 𝛾𝛾1 indicates the presence of classification shifting. Moreover, a positive estimate on 𝛾𝛾4 (𝛾𝛾5), implies classification shifting is more pervasive in highly individualist (strongly uncertainty avoidant) nations. For the remainder of this research, I will omit the time-indicating subscript 𝑡𝑡, which refers to the year to which the data is applicable between 2010 and 2018, for ease of readability.
4. Results
This section discusses the results of the regression analyses and submits these results to several sensitivity analyses.
4.1. Descriptive Statistics
Table 2 presents the summary statistics of the main variables under analysis. The statistics are in line with existing cross-country research on classification shifting (e.g. Haw et al., 2011; Behn et al., 2013). Core earnings on average comprise 9.61 percent of sales (with a median of 10.60 percent). The mean and median of unexpected core earnings (𝑈𝑈𝐶𝐶𝐶𝐶) are close to zero (they equal 0.0008 and -0.0022 respectively) and special items (%𝑆𝑆𝑆𝑆) represent on average 1.70 percent of sales. The mean (median) individualism (𝑆𝑆𝐼𝐼𝐼𝐼) and uncertainty avoidance (𝑈𝑈𝐴𝐴𝑆𝑆) scores equal approximately 52 (46) and 54 (46) respectively, and resemble statistics in existing research on culture and earnings management (Nabar & Boonlert-U-Thai, 2007; Doupnik, 2008).
7 In addition to controlling for country-level GNP, I control for country-level fixed effects in my study on informal
cultural institutions. I do so as Fischer (2010) states that it is essential to control for imperceptible country heterogeneity when studying institutional influences in a cross-country fashion. Failure to control for it impairs the legitimacy of any conclusions drawn.
20 TABLE 2 Descriptive Statistics
Variable Name Mean Median Standard
Deviation 25% 75% 𝐶𝐶𝐶𝐶 0.0961 0.1060 0.2800 0.0452 0.1945 𝑈𝑈𝐶𝐶𝐶𝐶 0.0008 -0.0022 0.1115 -0.0325 0.0351 %𝑆𝑆𝑆𝑆 0.0170 0.0000 0.0660 0.0000 0.0040 𝑆𝑆𝐼𝐼𝐼𝐼 52.2480 46 29.1564 20 89 𝑈𝑈𝐴𝐴𝑆𝑆 54.0133 46 23.5288 36 81 𝑁𝑁𝑁𝑁𝐺𝐺 10.2900 10.5503 0.7185 9.7273 10.8359 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶 6.1111 6.0549 1.9488 4.7877 7.4099 𝐿𝐿𝐶𝐶𝐼𝐼 0.2648 0.2422 0.1961 0.1175 0.3769 𝐵𝐵𝐵𝐵𝐼𝐼 0.5584 0.2355 1.0310 0.0794 0.6026 𝑅𝑅𝐴𝐴𝐴𝐴 0.0049 0.0258 0.1281 -0.0051 0.0595 𝐶𝐶𝐶𝐶𝐴𝐴 0.0548 0.0608 0.1125 0.0114 0.1105
All variables and their measurement techniques are delineated in Appendix B, table B1. The full sample comprises 101,792 observations of 20,205 firms from 73 countries. All variables are winsorized at the 1 and 99 percent level.
Appendix C, Table C1 reports country-level summary statistics for the 73 countries under analysis. Mean core earnings (𝐶𝐶𝐶𝐶) are highest in Tanzania (0.3948) and Norway (0.2701), whereas Ghana and Tanzania exhibit the highest mean unexpected core earnings (𝑈𝑈𝐶𝐶𝐶𝐶) (0.0950 and 0.0659 respectively). Furthermore, firms in Latvia (0.1015) and Hong Kong (0.0530) on average display the highest level of special items (%𝑆𝑆𝑆𝑆). Individualism (𝑆𝑆𝐼𝐼𝐼𝐼) ranges from low (collectivist) in Panama (11) and Venezuela (12), to high (individualist) in the United States (91) and Australia (90). Moreover, both highly uncertainty avoidant (𝑈𝑈𝐴𝐴𝑆𝑆) nations, like Greece (100) and Portugal (99) and low uncertainty avoidant nations, like Singapore (8) and Jamaica (13) are represented in this study. This is in line with the ranges presented in previous research on culture and earnings management (e.g. Nabar & Boonlert-U-Thai, 2007; Doupnik, 2008).
Table 3 presents a Pearson correlation matrix for the main variables. Since all correlation coefficients are below the multicollinearity threshold of 0.90 (Dohoo, Ducrot, Donald and Hurnik, 1997), there are no indications of multicollinearity. An inspection of the variance inflation factors (VIF), depicted in Appendix C, Table C1, confirms this finding8. In line with
8 The VIF values should be below 10 and have a tolerance level above 0.10 to avoid multicollinearity (Hair Jr.,
21
McVay (2006), core earnings (𝐶𝐶𝐶𝐶) relates negatively and significantly and unexpected core earnings (𝑈𝑈𝐶𝐶𝐶𝐶) relates positively and significantly to special items (%𝑆𝑆𝑆𝑆) (see also Behn et al., 2013). Furthermore, individualism (𝑆𝑆𝐼𝐼𝐼𝐼) correlates positively and significantly and uncertainty avoidance (𝑈𝑈𝐴𝐴𝑆𝑆) correlates negatively and significantly with percentage of special items (%𝑆𝑆𝑆𝑆), indicating a potential relationship between country-level culture and the level of special items reported by firms.
TABLE 3
Pearson Correlation Matrix
𝐶𝐶𝐶𝐶 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝐴𝐴𝐴𝐴𝐴𝐴 ∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 𝑁𝑁𝑆𝑆𝑁𝑁 ∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 𝑈𝑈𝐶𝐶𝐶𝐶 %𝑆𝑆𝑆𝑆 𝑆𝑆𝐼𝐼𝐼𝐼 𝑈𝑈𝐴𝐴𝑆𝑆 𝑁𝑁𝑁𝑁𝐺𝐺 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶 𝐿𝐿𝐶𝐶𝐼𝐼 𝐶𝐶𝐶𝐶𝐴𝐴 𝑅𝑅𝐴𝐴𝐴𝐴 BMV 𝐶𝐶𝐶𝐶 1.000 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 0.367** 1.000 𝐴𝐴𝐴𝐴𝐴𝐴 -0.005 0.094** 1.000 ∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 0.016** 0.055** 0.010** 1.000 𝑁𝑁𝑆𝑆𝑁𝑁 ∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 0.271** 0.271** 0.078** 0.449** 1.000 𝑈𝑈𝐶𝐶𝐶𝐶 0.496** -0.022** -0.004 -0.051** -0.042** 1.000 %𝑆𝑆𝑆𝑆 -0.214** -0.478** -0.069** -0.024** -0.191** 0.022** 1.000 𝑆𝑆𝐼𝐼𝐼𝐼 -0.048** -0.098** -0.048** 0.009** 0.056** 0.010** 0.097** 1.000 𝑈𝑈𝐴𝐴𝑆𝑆 0.010** 0.030** 0.090** -0.085** 0.047** -0.004 -0.083** 0.005† 1.000 𝑁𝑁𝑁𝑁𝐺𝐺 -0.102** -0.081** -0.011** -0.027** 0.016** -0.019** 0.089** 0.579** 0.137** 1.000 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶 0.290** 0.072** -0.027** -0.031** 0.145** 0.086** -0.053** 0.045** 0.064** 0.084** 1.000 𝐿𝐿𝐶𝐶𝐼𝐼 0.016** -0.148** -0.222** -0.040** -0.056** 0.026** 0.088** -0.042** 0.077** -0.067** 0.169** 1.000 𝐶𝐶𝐶𝐶𝐴𝐴 0.506** -0.049** 0.044** -0.018** 0.181** 0.251** -0.140** 0.083** 0.043** 0.012** 0.259** -0.099** 0.506 𝑅𝑅𝐴𝐴𝐴𝐴 0.601** 0.546** 0.082** 0.036** 0.285** 0.174** -0.432** -0.094** 0.015** -0.118** 0.287** -0.192** 0.569** 1.000 𝐵𝐵𝐵𝐵𝐼𝐼 -0.114** -0.150** -0.133** -0.061** -0.171** -0.016** 0.116** 0.032** -0.037** 0.063** -0.179** 0.005** -0.105** -0.172** 1.000
A specification of how the variables in table 3 are measured can be found in Appendix B, Table B1. The sample consists of 101,792 firm-year observations of 20,205 firms from 73 countries. All variables are winsorized at the 1 and 99 percent level. Significance is indicated at the 1 percent (**), 5 percent (*) and 10 percent (†) level.
23 4.2. Unexpected Core Earnings
To estimate Model 1 and compose unexpected core earnings, I run 414 industry-year regressions. The mean and median regression output is presented in Table 4. With an adjusted 𝑅𝑅2 of 73.34 percent, Model 1 is a strong predictor of core earnings. All estimates resemble those
found by McVay (2006). Lagged core earnings (𝐶𝐶𝐶𝐶𝑡𝑡−1) is a powerful and persistent predictor of core earnings as the median coefficient is highly significant (0.7312, 𝑝𝑝 < 0.01) and positive for all observations in the 414 industry-year regressions. On the contrary, the asset turnover ratio (𝐴𝐴𝐴𝐴𝐴𝐴) is not strongly related to core earnings, probably because the industry drives the association between the profit margin and asset turnover (Soliman, 2004).
TABLE 4
Model (1): Unexpected Core Earnings
Dependent Variable: 𝑬𝑬(𝑪𝑪𝑬𝑬𝒕𝒕) Independent Variables Sign Mean Coefficient (one-tailed p-value) Median Coefficient (one-tailed p-value) Percent significant (p-value ≤ 0.10, one-tailed test) Percent with sign in the predicted direction Intercept 0.0447 (0.0363) 0.0359 (0.0001) 𝐶𝐶𝐶𝐶𝑡𝑡−1 + (0.0001) 0.7205 (0.0000) 0.7312 99.69% 100% 𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 - (0.2380) -0.0004 (0.2285) -0.0003 23.35% 72.24% 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡−1 - (0.0659) -0.0958 (0.0004) -0.0931 77.95% 81.33% 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑡𝑡 + (0.0508) 0.1362 (0.0000) 0.1251 84.81% 85.97% ∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 + 0.0691 (0.0767) 0.0576 (0.0009) 75.22% 89.75% 𝑁𝑁𝐶𝐶𝑁𝑁∆𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴 + 0.2851 (0.0560) 0.2377 (0.0003) 83.63% 90.07% Adjusted R2 71.62% 73.34%
The measurement methods for the variables in Table 4 are presented in Appendix B, Table B1. This model was run on 101,792 observations of 20,205 firms from 73 countries which formed 414 industry-year regressions. The predicted sign is determined on the grounds of theory, and depicted in the second column. The third and fourth column represent
respectively the mean and median coefficients for all variables in Model 1. The 𝑝𝑝-values are estimated by means of
one-tailed tests and presented in parentheses. The fifth column shows the percentage of the 414 industry-year
regressions of which the coefficient is significant at the 10 percent level (𝑝𝑝 < 0.10). The last column shows the
percentage of the 414 industry-year regressions of which the value of the variable is in line with the predicted direction. All variables are winsorized at the 1 and 99 percent level.
24 4.3. Primary Tests
I estimate the results for my baseline regressions, as described in Eq. (3) and Eq. (4), by means of ordinary least squares regression with robust standard errors clustered on firm-level. Results are presented in Table 5. I first study whether firms in my complete sample exhibit classification shifting behavior. Model 2 reports the regression results with special items (%𝑆𝑆𝑆𝑆) as the sole independent variable. As discussed previously, I expect unexpected core earnings to become higher with special items when management misclassifies core expenses. Similar to McVay (1006) and Behn et al. (2013), I find that the coefficient of special items (%𝑆𝑆𝑆𝑆) is positive and significant at the 1 percent level (0.0359), indicating that firms classification shift. The proportion of explained variance equals 0.84 percent, which is substantially higher than McVay’s (2006) 0.03 percent, potentially because I study a cross-country sample, and include year, industry and country fixed effects.
Model 3 assesses the variables of interest, i.e. individualism (𝑆𝑆𝐼𝐼𝐼𝐼), uncertainty avoidance (𝑈𝑈𝐴𝐴𝑆𝑆) and their interactions with special items (%𝑆𝑆𝑆𝑆 × 𝑆𝑆𝐼𝐼𝐼𝐼 and %𝑆𝑆𝑆𝑆 × 𝑈𝑈𝐴𝐴𝑆𝑆 respectively) and includes the controls (𝑁𝑁𝑁𝑁𝐺𝐺, 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶, 𝐿𝐿𝐶𝐶𝐼𝐼, 𝐶𝐶𝐶𝐶𝐴𝐴, 𝑅𝑅𝐴𝐴𝐴𝐴, 𝐵𝐵𝐵𝐵𝐼𝐼). As illustrated in Table 5, special items (%𝑆𝑆𝑆𝑆) is positive and significant at the 10 percent level (0.0620), indicating that firms misclassify core expenses as special items. In support of hypothesis 1, I find that the coefficient of interaction between special items and individualism (%𝑆𝑆𝑆𝑆 × 𝑆𝑆𝐼𝐼𝐼𝐼) is positive and significant at the 1 percent level (0.0012). This indicates that country-level individualism significantly increases the level of classification shifting performed by management. The coefficient of the interaction between classification shifting and uncertainty avoidance (%𝑆𝑆𝑆𝑆 × 𝑈𝑈𝐴𝐴𝑆𝑆) is not significant. Thus, hypothesis 2, which predicts that classification shifting is more pervasive in countries scoring high on uncertainty avoidance, is not supported. Notably, the adjusted 𝑅𝑅2 is substantially higher for my research than for Behn et al. (2013), who report an average adjusted 𝑅𝑅2 of 2.18 percent, while controlling for firm-level performance in addition to assessing the
influence of investor protection on classification shifting. This indicates that culture is an important determinant to consider in studying classification shifting internationally.
McVay (2006) and Zalata and Roberts (2016) argue that firms reporting income-decreasing special items have more opportunity to perform classification shifting to begin with. Hence, I re-estimate the coefficients for solely those firms reporting special items that decrease income. The results are presented in Model 4. The coefficient of special items (%𝑆𝑆𝑆𝑆) (0.0283) is positive but not significant. Thus, there is no direct evidence that firms reporting special items are also
25
more likely to classification shift. The interaction effect between individualism and special items (%𝑆𝑆𝑆𝑆 × 𝑆𝑆𝐼𝐼𝐼𝐼) remains positive and significant at the 1 percent level (0.0009).
TABLE 5
Baseline regression models
Dependent variable: Unexpected Core Earnings
Full Sample Full Sample Non-Zero Income-decreasing special items Model (2) (3) (4) %𝑆𝑆𝑆𝑆 0.0359** (0.0123) (0.0356) 0.0620† (0.0231) 0.0283 𝑆𝑆𝐼𝐼𝐼𝐼 -0.0019** (0.0005) -0.0026** (0.0004) 𝑈𝑈𝐴𝐴𝑆𝑆 -0.0020** (0.0003) -0.0027** (0.0004) %𝑆𝑆𝑆𝑆 × 𝑆𝑆𝐼𝐼𝐼𝐼 0.0012** (0.0004) 0.0009** (0.0003) %𝑆𝑆𝑆𝑆 × 𝑈𝑈𝐴𝐴𝑆𝑆 (0.0007) 0.0004 (0.0005) -0.0002 𝑁𝑁𝑁𝑁𝐺𝐺 0.0529** (0.0058) 0.0802** (0.0114) 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶 (0.0002) 0.0006* 0.0010** (0.0004) 𝐿𝐿𝐶𝐶𝐼𝐼 0.0354** (0.0026) 0.0323** (0.0040) 𝐶𝐶𝐶𝐶𝐴𝐴 0.2190** (0.0066) 0.2046** (0.0110) 𝑅𝑅𝐴𝐴𝐴𝐴 0.0892** (0.0082) 0.0887** (0.0127) 𝐵𝐵𝐵𝐵𝐼𝐼 0.0018** (0.0005) 0.0033** (0.0009) Intercept -0.0138† (0.0074) -0.2973** (0.0534) -0.4921** (0.0712)
Year fixed effects Yes Yes Yes
Industry fixed effects Yes Yes Yes
Country fixed effects Yes Yes Yes
Adjusted 𝑅𝑅2 0.84% 8.40% 7.93% Observations (Countries) 101,792 (73) 101,792 (73) 39,238 (72)9
The measurement methods for the variables in Table 5 are presented in Appendix B, table B1. The sample consists of 101,792 firm-year observations of 20,205 firms from 73 countries. All variables are winsorized at the 1 and 99 percent level. The coefficients are estimated by means of an ordinary least squares regression. Standards errors (presented in parentheses) are robust and clustered on firm-level to address potential heteroscedasticity. Significance is indicated at the 1 percent (**), 5 percent (*) and 10 percent (†) level.
9 Ghana is dropped from the sample when I solely analyze non-zero income-decreasing special items.
26 4.4. Sensitivity Tests
4.4.1. Additional controls for formal institutions
Whereas this study controls for country fixed effects, I do not address time-variant institutional influences. However, research shows that national formal institutions that vary over time could also significantly influence firm-level engagement in earnings management. Dyreng, Hanlon & Maydew (2009) for instance show that regulatory scrutiny and penalties on earnings management are lower in countries with a poor rule of law. Guenther (1994) and Desai and Dharmapala (2009) demonstrate that the tax burden placed on firms can induce earnings management. Furthermore, Anokhin and Schulze (2008) show that corruption among those concerned with property rights, contract enforcement and courts, as well as lack of government policy implementation and effectiveness results in higher levels of information asymmetry and opportunistic behavior. Moreover, Manzetti and Wilson (2007) found that corruption and poor rule of law aggravate agency problems. As evidenced by Lourenço, Rathke, Santana & Branco (2018), this materializes in earnings management. González and García-Meca (2014) illustrate that no corruption, high government effectiveness and a strong rule of law mitigate earnings management. By contrast, Shen and Chih (2007) argue that a strong rule of law can also motivate earnings management, as penalties for negative earnings are higher.
These formal institutions are integrated in the Heritage Foundation’s annual index of economic freedom10. Employing this aggregated measure, Riahi-Belkaoui (2004) evidences that economic freedom mitigates earnings management, whereas Baatour and Othman (2016) demonstrate that earnings management is more pervasive in economically free nations. Following these findings, I expect that economic freedom (𝐶𝐶𝐶𝐶𝐶𝐶) (as reflected in the rule of law (𝑅𝑅𝐴𝐴𝐿𝐿), government size (𝑁𝑁𝑆𝑆𝑆𝑆), regulatory efficiency (𝑅𝑅𝐶𝐶𝐶𝐶) and openness of markets (𝐴𝐴𝐺𝐺𝐵𝐵)) significantly influences the pervasiveness of classification shifting. To test this conjecture, I re-run Model 3 including these four formal institutions, as well as with the overall score on economic freedom. The results are provided in Table 6. The coefficient for special items as a (%𝑆𝑆𝑆𝑆) is positive and significant at the 10 percent level for including the score on economic freedom (0.0624) and for including the individual institutions (0.0611). The interaction coefficient of special items and individualism is positive and significant at the 1 percent level for including economic freedom (0.0012) and the individual institutions (0.0012). Thus, my results are robust for including controls on formal institutions.
10 Which is calculated as the average of country-level indices on rule of law, government size, regulatory efficiency
27 TABLE 6
Sensitivity: Formal Institutions
Dependent variable: Unexpected Core Earnings Economic Freedom Separate Variables
Model (5) (6) %𝑆𝑆𝑆𝑆 (0.0356) 0.0624† (0.0356) 0.0611† 𝑆𝑆𝐼𝐼𝐼𝐼 -0.0022** (0.0005) -0.0023** (0.0005) 𝑈𝑈𝐴𝐴𝑆𝑆 -0.0025** (0.0004) -0.0029** (0.0004) %𝑆𝑆𝑆𝑆 × 𝑆𝑆𝐼𝐼𝐼𝐼 0.0012** (0.0004) 0.0012** (0.0004) %𝑆𝑆𝑆𝑆 × 𝑈𝑈𝐴𝐴𝑆𝑆 (0.0007) 0.0004 (0.0007) 0.0003 𝐶𝐶𝐶𝐶𝐶𝐶 -0.0007** (0.0002) 𝑅𝑅𝐴𝐴𝐿𝐿 (0.0001) 0.0003* 𝐴𝐴𝐺𝐺𝐵𝐵 -0.0003† (0.0002) 𝑅𝑅𝐶𝐶𝐶𝐶 -0.0021** (0.0002) 𝑁𝑁𝑆𝑆𝑆𝑆 0.0003** (0.0001) 𝑁𝑁𝑁𝑁𝐺𝐺 0.0646** (0.0069) 0.0750** (0.0073) 𝑆𝑆𝑆𝑆𝑆𝑆𝐶𝐶 (0.0002) 0.0006* (0.0002) 0.0005* 𝐿𝐿𝐶𝐶𝐼𝐼 0.0353** (0.0026) 0.0350** (0.0026) 𝐶𝐶𝐶𝐶𝐴𝐴 0.2190** (0.0066) 0.2188** (0.0066) 𝑅𝑅𝐴𝐴𝐴𝐴 0.0892** (0.0082) 0.0897** (0.0082) 𝐵𝐵𝐵𝐵𝐼𝐼 0.0018** (0.0005) 0.0018** (0.0005) Intercept -0.3196** (0.0540) -0.3081** (0.0590)
Year fixed effects Yes Yes
Industry fixed effects Yes Yes
Country fixed effects Yes Yes
Adjusted 𝑅𝑅2 8.41% 8.50% Observations (Countries) 101,792 (73) 101,792 (73)
The measurement methods for the variables in Table 6 are presented in Appendix B, Table B1. For a clear specification on the methodology of the Heritage Foundation’s score on economic freedom and individual formal institutions, I refer to the methodology presented on the website of the Heritage Foundation (https://www.heritage.org/index/). The sample consists of 101,792 firm-year observations of 20,205 firms from 73 countries. All variables are winsorized at the 1 and 99 percent level. The coefficients are estimated by means of an ordinary least squares regression. Standards errors (presented in parentheses) are robust and clustered on firm-level to address potential heteroscedasticity. Significance is indicated at the 1 percent (**), 5 percent (*) and 10 percent (†) level
