The Effect of Board of Directors’ Independence on Classification Shifting: Evidence from U.S.
Firms
Name: Sjoerd Veldhuizen Student number: 12917036 Thesis supervisor: Dennis Jullens Date: June 17, 2021
Word count: 12202
MSc Accountancy & Control Specialization: Accountancy Amsterdam Business School
Faculty of Economics and Business, University of Amsterdam
2
Statement of Originality
This document is written by student Sjoerd Veldhuizen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This master thesis examines the use of classification shifting among U.S. firms and the effect of board independence on this form of earnings management. McVay (2006) concludes that managers opportunistically shift core expenses to special items to increase core earnings, which results in a positive relationship between unexpected core earnings and special items.
However, less is known about mitigating this form of earnings management. Therefore, this study investigates whether board independence affects classification shifting practices. The sample of this study consists of U.S. firms within the S&P 1500 in the period of 2008-2018. I find evidence that U.S. firms see classification shifting as a viable earnings management method to inflate core earnings. The results show that, in presence of classification shifting, firms classify core expenses as special items to increase core earnings in the current year. I also find evidence that this abnormally high core earnings in the current year are not
maintained, because a part of the shifted core expenses reverses in the following year. Further results indicate that board independence can affect classification shifting practices. This study confirms that CEO duality (CEO of the company is also the chairman of the board) is
positively associated with classification shifting. This indicates that firms with a “dual” CEO are more likely to engage in classification shifting practices. However, the results show that the proportion of independent board members is not associated with classification shifting.
4
Contents
1 Introduction ... 6
2 Literature Review and Hypothesis Development ... 9
2.1 Agency Theory ... 9
2.2 Earnings Management ... 10
2.2.1 Accrual-Based Earnings Management ... 10
2.2.2 Real Earnings Management ... 11
2.2.3 Classification Shifting ... 11
2.2.4 Motivations for Earnings Management ... 13
2.3 Board Characteristics and Earnings Management ... 15
2.3.1 Board Independence and Earnings Management ... 15
2.3.2 CEO Duality... 16
2.3.3 Independent Directors ... 16
2.4 Hypotheses Development ... 17
2.4.1 CEO Duality and Classification Shifting ... 17
2.4.2 Independent Board Members and Classification Shifting ... 17
3 Data and Research Method ... 19
3.1 Data and Sample Selection ... 19
3.2 Research Method ... 20
3.3 Testing Classification Shifting ... 23
4 Results ... 25
4.1 Descriptive Statistics and Correlation Matrix ... 25
4.2 Expectation Models ... 28
4.2.1 Model of Expected Core Earnings ... 28
4.2.2 Model of Expected Change in Core Earnings... 29
4.3 Measuring Classification Shifting ... 30
5
4.3.1 Misclassification of Core Expenses as Special Items ... 30
4.3.2 Reversal of Special Items... 31
4.3.3 Effect Board Independence on Classification Shifting ... 32
4.4 Robustness Analysis McVay model ... 35
5 Conclusion and Discussion ... 37
References ... 39
Appendix ... 43
6
1 Introduction
This study examines the effect of board independence on the level of classification shifting.
Opportunistic behaviour by managers in the field of financial reporting has been a concern for many years (Callao & Jarne, 2010). According to Jensen and Meckling (1976), the separation of ownership and control creates an agency problem. This agency problem produces difficulties if the agent and the principal may not have the same interests. In the context of earnings management firms may have incentives to create an opportunistic picture of their financial information at the expense of the users of this financial information.
Prior research shows that earnings management can be divided into real earnings management and accrual-based earnings management (Ho et al., 2015). McVay (2006) examines a third potential earnings management tool, called classification shifting.
Classification shifting is a lesser-known form of earnings management and in comparison to accrual-based earnings management and real earnings management, also less researched.
According to McVay (2006) classification shifting is the deliberate misclassification of core expenses as noncore special items within the income statement to increase core earnings.
Classification shifting differs from accrual-based and real earnings management, because it does not change bottom line earnings but overstates “core earnings” (McVay, 2006). She argues that classification shifting is caused by managers wishing to maximize reported performance to present a picture that is not consistent with the economic reality.
Klein (2002) shows that there is a relation between corporate governance and the likelihood firms engage in earnings management. The board of directors has been regarded as a key corporate governance mechanism as it decreases agency costs, rewarding top executives and leads to maximization of shareholders wealth (Kanojia, Sharma & Jain, 2020). They are responsible for setting up objectives, controlling and monitoring actions of the firm which lead to alignment between the interests of managers and shareholders (Fama & Jensen, 1983). An important aspect of increasing the effectiveness of boards is board independence (Beiner et al., 2004). Klein (2002) concludes that the independence of the board is negatively associated with the level of accrual-based earnings management. This finding is supported by Xie, Davidson
& DaDalt (2003) who also found that earnings management is less likely to occur in companies whose boards include more independent outside directors. In addition, Osma (2008) examined that firms with more independent board members are also less likely to engage in real earnings management. This paper investigates if this relation between earnings management and board independence is also present in the case of classification shifting.
7
This study focuses on a specific part of corporate governance, namely the independence of the board of directors. According to the findings above, more research is needed to find out the association of board independence on classification shifting. In this study, board independence is classified by two measures: (1) CEO duality and (2) proportion of independent directors on the board. To examine the effect of board independence on classification shifting the following research question is stated:
“What is the relation between board independence and classification shifting?”
This study contributes to existing literature in several ways. First, it builds on existing earnings management literature that examined the role of board characteristics on accrual- based earnings management and real earnings management (Klein, 2002; Xie et al., 2003;
Osma, 2008). This study extends this literature by examining classification shifting instead of accrual-based earnings management or real earnings management. Second, it expands the existing literature of classification shifting. Despite the prevalence of classification shifting (McVay, 2006) in the United States, less is known about mitigating this form of earnings management by internal corporate governance mechanisms. Zalata and Roberts (2016) conclude that boards and audit committees of higher quality can mitigate classification shifting practices. But their study is limited to an IFRS setting (sample of UK firms) and less is known about this mitigation effect among U.S. firms. Finally, this study provides insights whether board independence improves the financial reporting quality. This can provide insights for investors, because classification shifting results in an increase of core earnings, so investors should probably also take in consideration to critically look at the characteristics of the board when making investment decisions. In addition, this study also provides insights for standard- setters because it contains information about the effect of board independence on financial reporting quality. This information can be used in decision making regarding future corporate governance or financial reporting related standards.
To test whether U.S. firms engage in classification shifting I use a sample of U.S. firms within the S&P 1500 for the period of 2008-2018. To test if these firms engage in classification shifting practices, I use the models of McVay (2006). Before classification shifting practices can be investigated, two models are used to determine the expected core earnings and the expected change in core earnings. Secondly, a model is used to identify whether firms classify core expenses as special items. An additional model is used to determine whether the presence of misclassification of core expenses is indeed caused by classification shifting instead of other real economic events. If the misclassification of core expenses as special items is caused by
8
classification shifting, it is expected that this misclassification will reverse in the following year. Finally, I investigate whether board independence is associated with the classification of core expenses as special items.
Consistent with McVay (2006) I find evidence that classification shifting is a viable earnings management method in the U.S. The results show a positive relationship between unexpected core earnings and special items in the current year, which indicates that managers shift some of the core expenses to special items. In addition, I find that these misclassifications of core expenses as special items are partially reversed in the following year. I also find evidence that firms with a CEO who also serves as the chairman of the board are more likely to engage in classification shifting practices. Finally, the results show that the proportion of independent board members is not associated with classification shifting.
The following parts of this study consists of four more sections. Section 2 describes the related literature and the formulated hypotheses. Section 3 provides information about the data and research method. Section 4 consists of the descriptive statistics, correlation matrix, results and a robustness analysis. Finally, section 5 provides the conclusion and discussion.
9
2 Literature Review and Hypothesis Development
In this section, based on literature, the theories and concepts surrounding this study are discussed. First, the underlying theory of this research will be discussed. Secondly, it describes the meaning of earnings management and the different ways earnings management can be applied. Then, classification shifting is described in more detail and how classification shifting practices differ from other commonly used earnings management methods. At last, motives for earnings management and the relation between board independence and earnings management will be discussed. After all the necessary theorical concepts have been discussed, the hypotheses of this research will be developed and shown.
2.1 Agency Theory
The key theory for a better understanding of the concepts discussed in this study is the agency theory. The agency theory is first introduced by Jensen and Meckling (1976) and explains the relationship between the principal and the agent and the problems arising from misalignment of their interests. Jensen and Meckling (1976, p. 308) formulated the following definition
regarding the agency theory:
“We define an agency relationship as a contract under which one or more persons (the principal(s)) engage another person (the agent) to perform some service on their behalf which involves delegating some decision-making authority to the agent. If both parties to the relationship are utility maximizers there is good reason to believe that the agent will not always act in the best interests of the principal.”
The agency theory is focused on resolving two problems that might occur in agency relationships (Eisenhard, 1989). The first problem arises when the goals and desires of the agent and the principal conflict and it is expensive or difficult for the principal to verify what the agent is doing. The first problem is caused by the fact that the principal is unable to verify that the agent has behaved correctly. The second is the problem of risk sharing. This problem arises when the principal and agent have different risk attitudes. This can lead to agency problems because the agent and the principal may prefer different actions towards risk because of the difference in their attitudes towards risk.
According to Jensen and Meckling (1976) the principal can mitigate conflicts from his interest by providing incentives for the agent and/or by monitoring the agent that he will not take actions that would harm the principal. In addition, in some situations the principal will pay
10
the agent to expend resources (bonding costs) to guarantee that the agent will not take actions that are not aligned with the interests of the principal (Jensen & Meckling, 1976).
2.2 Earnings Management
Earnings management is in general a much-discussed phenomenon in the existing literature.
Man and Wong (2013) define earnings management as the choice of a manager of accounting policies or other actions to intentionally affect earnings. According to Healy and Wahlen (1999) earnings management occurs when managers use judgement in financial reporting practices, structuring transactions to modify financial reporting, mislead stakeholders about the underlying economic performance of the firm and/or to influence contractual numbers that depend on the reported accounting information of the firm. Healy and Wahlen (1999) suggest earnings management is caused by managers’ discretion in accounting related issues and this discretion has both benefits and costs. Benefits of earnings management include potential improvements in credible communication of private information to their external stakeholders.
Costs of earnings management are related to the potential of misallocation of resources.
2.2.1 Accrual-Based Earnings Management
Prior research shows that earnings management can be divided into real and accrual-based earnings management (Ho et al., 2015). Accrual-based earnings management refers to the intentionally changing of accruals in a particular direction that is accomplished when managers adjust expense or revenue accruals to modify financial reporting (Badertscher, 2011).
Examples of accrual earnings management are expense manipulation, revenue manipulation and margin manipulation (Dechow, Sloan & Sweeney, 1995). Expense manipulation refers to the delayed recognition of expenses. This approach is implemented by adding the supposed amount of expense manipulation to total accruals in the current year (when earnings management occurs) and subtracting this amount of expense manipulation in the following year (Dechow et al., 1995). Revenue manipulation is the premature recognition of revenue, assuming all costs are fixed. This form of accrual management increases revenue and receivables in the current year and in a later moment in time revenue and receivables will be decreased with the same amount (Dechow et al., 1995). Finally, margin manipulation refers to the untimely recognition of revenue, assuming that all costs are variable. Margin manipulation allows for shifting of profit margins. For example, increasing accruals by 1% of lagged assets and revenue and accounts receivable by 10% of lagged assets for a firm with a net income ratio
11
of ten percent. The same amount of these addition in the current year is subtracted in the following year (Dechow et al., 1995).
2.2.2 Real Earnings Management
Real earnings management refers to the deliberate adjustment of reported earnings in a particular direction by changing the timing or structure of a business, financing or investment decision (Badertscher, 2011). Ge and Kim (2013) consider three forms of real earnings management: overproduction, sales manipulation and the abnormal reduction of discretionary expenditures.
Overproduction means that more goods are produced than necessary to increase earnings. The costs of products sold appears in the income statement as the costs of goods sold (COGS) and the costs of the products that are unsold appears as inventory in the balance sheet.
In case of overproduction, the overhead costs spread over more products, which results in lower costs per product. This gives managers the opportunity to report a lower COGS and increase earnings while leaving a remarkable proportion of the production costs in the inventory in the balance sheet. Boosting of earnings by overproduction is less sustainable and the overproduced inventory may be obsolete later, which could result in future losses (Ge and Kim (2013).
Sales manipulation refers to the managers’ attempts to increase sales during the year by offering, for example, more lenient credit terms or price discounts. The increased sales, due to these attempts, most likely disappear once the firm reverts to their original prices. Offering more lenient credit terms, longer payments periods for example, increases the risk of higher uncollectible accounts (Ge and Kim (2013).
Discretionary expenses include employee training, maintenance, advertising, and other expenses. Firms often pay these discretionary expenses by cash. Reducing these kind of expenditures lowers the outflow of cash and has a positive effect on abnormal cash flows, probably at the risk of lower cash flows in the future. For example, an abnormal reduction of advertising expenditures may result in lower future sales and therefore also lower cash flows (Ge and Kim (2013).
2.2.3 Classification Shifting
According to McVay (2006) the bulk of earnings management literature has focused on two earnings management tools: accrual management and the manipulation of real economic activities. In this study a third form of earnings management, called classification shifting, is examined. Classification shifting was first introduced by McVay (2006). She defined
12
classification shifting as the deliberate misclassification of items within the income statement.
This misclassification is caused by managers’ opportunistically shifting of expenses from core expenses (cost of goods sold and selling, general and administrative expenses) to special items.
According to McVay (2006) classification shifting does not change the bottom-line earnings number but overstates “core earnings”. For example, she states that managers use this form of earnings management to meet the analyst forecast earnings benchmark. Abernathy, Beyer and Rapley (2004) mention that classification shifting misrepresents the persistence of items within the income statement, which could mislead investors about the performance of the firm. According to Abernathy et al. (2004) this misrepresentation of persistence is caused by the placement of items within the income statement in a way that more persistent items are placed higher in the income statement. This is consistent with the view of McVay (2006) who states that classification shifting means shifting recurring expenses downwards in the income statement to special items. Earnings management through classification shifting could potentially undermine the credibility of financial statements, and these statements are an essential part of useful accounting information in capital markets (Haw et al., 2011). According to Zalata and Roberts (2016) non-recurring (non-persistent) expenses are, unlike recurring (persistent) expenses, transitory or infrequent. Users of financial statement, especially less sophisticated investors, seem not to understand the difference in nature and weight of individual categories within the income statement. As a result, this could motivate managers to misclassify a part of their recurring expenses as non-recurring expenses, which inflates the core earnings of the firm (Zalata & Roberts, 2016).
McVay (2006) focuses on the shifting of expenses between core expenses (cost of goods sold and selling, general and administrative expenses). Special items are events that are a result of a firm’s ongoing and continuous activities, but these activities are infrequent or unusual in nature and must be disclosed within the income statement as a separate line item as a part of continuing operations or in the footnotes (McVay, 2006). Examples of these infrequent or unusual events that are recognized as special items are gains or losses from the sale of investments or equipment, write-offs or write-downs of receivables, intangibles, inventories or equipment and special one-time charges arising from corporate restructurings (McVay, 2006).
McVay (2006) argues that classification shifting is different from manipulation of real activities and accrual management in three ways. First, classification shifting does not change GAAP earnings and classification shifting would be pointless if users of financial statements only focus on the earnings number. Therefore, classification shifting focuses on the shifting of individual components within the income statement that are meant to be informative to users
13
of the financial statements. Secondly, all three forms of earnings management raise expectations of future performance, but accrual management and manipulation of real activities also decrease earnings in future periods. There is no “settling up” using classification shifting, so next period’s earnings are the same as actual earnings instead of actual earnings less the cost of earnings management in the previous period. Finally, classification shifting does not affect GAAP net income. According to Nelson, Elliot, and Tarpley (2002) the scrutiny of auditors and regulators is limited when GAAP net income does not change. As a result of this, classification shifting may be an attractive earnings management tool and can be preferred by managers over accrual management or the manipulation of real economic activities. In addition, Abernathy et al. (2014) mention that the use of classification shifting as an earnings management strategy may be increased because of the attention that has been given to accrual- based earnings management by regulators and auditors. Also, McVay (2006) mentions that classification shifting may receive less attention by auditors because the appropriate categorization of expenses may not be noticeably clear to auditors. The procedures followed by the auditors might identify unrecorded expenses, but in the setting of classification shifting it is not the recognition that is in question but the classification. Documentation of expenditures may be too general for the auditor to investigate the appropriate classification.
2.2.4 Motivations for Earnings Management
An important reason managers engage in earnings management is the existence of compensation schemes (Healy, 1985). Healy (1985) was one of the first who provides evidence of contractual motivation to engage in earnings management. When managers have inside information, they could manage net income to increase their own bonuses. Therefore, it is expected that managers will manage earnings in the current period. In contrast, new managers are less like to engage in income-increasing earnings management in the current period and more likely to manage earnings downwards in the current period to take a “big bath”, because this may increase their bonuses in the future periods (Man and Wong, 2013).
Graham, Harvey and Rajgopal (2005) conclude that an important reason for managers to engage in earnings management is to meet or beat earnings benchmarks. Based on the study of Graham et al. (2005) managers want to meet or beat earnings benchmarks for several reasons. They state managers want to maintain or increase the stock price, build creditability with the capital market, convey future growth prospects and improve the (external) reputation of the management team. Moreover, Graham et al. (2005) mention that failure to hit the earnings benchmarks could create uncertainty about a firm’s prospects and may be an indicator
14
of hidden, deeper problems in the firm. In addition, managers are concerned about spending more time after the earnings announcement explaining why they missed the earnings benchmark rather than presenting their vision of the future of the firm (Graham et al, 2005).
Another motivation to engage in earnings management is to reflect a good picture about the future income flows of a firm to their investors by earnings smoothing (Aljifri, 2007). He mentions that managers may use income smoothing to reduce fluctuations in earnings rather than to increase reported income. Subramanyam (1996) argues that, because of decreased fluctuations, smoothing of earnings improves the quality of earnings. Defond and Park (1997) provide evidence that managers appear to shift earnings from good years to bad years. For example, if current earnings are “good” and expected future earnings are expected to be “poor”, managers have an incentive to shift some of the earnings from the good year to the poor year (Aljifri, 2007).
According to Aljifri (2007) there are two major underlying motives of earnings smoothing. The first underlying motive of earnings smoothing is to increase investor confidence in the future firms’ financial situation. Earnings smoothing leads to a more stable flow of earnings to support a high dividend payout, which will positively affect the firm’s stock price. Because dividends are adopted from current income, many investors base their selling and buying decisions on the dividend yield. Dividends in general are not only paid to meet the expectations of shareholders but also to attract potential investors who are interested in returns on their capital. As a result of this, managers may smooth earnings to pay dividends to increase stock prices at the end (Aljifri, 2007). The second major underlying motive of earnings smoothing is to enhance the ability of investors to predict future cash flows. This motive is consistent with one of the objectives of financial statements which states that financial statements must provide useful information to creditors and investors in comparing, predicting, and evaluating future cash flows. Therefore, financial statements should help investors in making their predictions about future cash flows. This would help managers to provide positive information about future cash flows to investors, which increases the ability of investors to predict future cash flows (Aljifri, 2007). Finally, Noronha, Zeng and Vinten (2008) suggest that external contracts, for instance dividend covenants, debt contracts and supplying contracts can incentivize managers to manipulate financial data to meet the requirements of the contract.
15
2.3 Board Characteristics and Earnings Management
The board of directors has been regarded as a key corporate governance mechanism as it decreases agency costs, rewarding top executives and leads to maximization of shareholders wealth (Kanojia, Sharma & Jain, 2020). They are responsible for setting up objectives, controlling and monitoring actions of the firm which lead to alignment between the interests of managers and shareholders (Fama & Jensen, 1983). Shareholders elect board members on their behalf and in return the board delegates power to top management, while they still monitoring the performance of the management and ratifying decisions that shows lack of attention to shareholders (Man and Wong, 2013). If board members fail to monitor the managers’ behaviour, the shareholders can vote out and replace board members. According to Kao and Chen (2004), the board of directors might be able to monitor the management and prevent them from engaging in earnings management. Therefore, the board of directors could also play an importing role, through their monitoring authority, to reduce earnings management by the management of the firm. Prior studies examine how specific board characteristics are effective in mitigating earnings management (Jensen and Fama, 1983; Peasnell and Young, 2005; Alves, 2014).
2.3.1 Board Independence and Earnings Management
An important aspect of increasing the effectiveness of boards is board independence (Beiner et al., 2004). The board of directors is a collective body that should act in the interests of their shareholders (Fuzi, Halim & Julizaerma, 2016). According to Fuzi et al. (2016) the board of directors requires to be composited of executive and non-executive directors to pursue the interests of the shareholders. The non-executive board members will not be able to effectively exercise their duties unless they are independent from the management and provide unbiased opinions. Independent directors are the members of the board to represent the shareholders and reduce agency problems (Fuzi et al., 2016). In addition, also the Code of Corporate Governance and regulators recommend a balanced composition of the board which also consists of independent directors (Fuzi et al., 2016). According to Man and Wong (2013) this mitigation effect exists because independent directors do not pursue self-interests such as misappropriation of assets, executive compensation, pressure from shareholders to meet or beat expectations of firm performance and the need to maintain their personal reputation to the public. In contrast, when the CEO comes from the founding family, the firm lacks
16
independence and the likelihood of restatements of accounting information increases (Agrawal and Chadha, 2005).
2.3.2 CEO Duality
Two important measures of board independence are the presence of CEO duality in the board and the proportion of independent directors on the board. CEO duality refers to the situation when the chairman of the board of directors is also the CEO of the firm (Johnson, 2019). In the case of CEO duality, the board of directors is considered to be less independent of firm management, because it is led by a manager of the firm instead of an external individual who is more independent of the firm (Rechner & Dalton, 1991). Aktas et al. (2019) use two competing theories to describe the relation between CEO duality and firm performance, namely the agency theory and stewardship theory. Agency theory states that agents commit to opportunistic behaviour and are focused on their own excessive benefits, at the expense of the interests of the shareholders. Based on the agency theory, CEO duality is therefore undesirable because it assigns excess power to a single executive, weakening board oversight and promotes management entrenchment (Aktas et al., 2019). In contrast, stewardship theory predicts that CEO duality can be beneficial for firms, because it enhances cohesive leadership, signals firm stability, and creates confidence in the company management (Aktas et al., 2019). They also mention that faster decision making, knowledge and expertise can result from CEO duality.
2.3.3 Independent Directors
Another important aspect of board independence is the presence of independent directors in the board. Independent board members can contribute by helping to ensure that managers act in the interests of stakeholders (Fama & Jensen, 1983) and therefore reduce existing agency problems. Martin (2011) states that most boards consist of independent, outside directors (who have no connection to the company other than their board seat) and a few inside directors. He concludes that inside board members are not well positioned to protect the interests of shareholders, because in many cases they act in their own interest and receive meaningful stock-based compensation. The effectiveness of the monitoring role of the board of directors is determined by inside as well as outside board members. However, Klein (2002) finds a positive relationship between independent boards and their monitoring activities. According to Klein (2002) the reduction of independence of boards is the largest when the board is comprised of a minority of outside directors. In addition, Peasnell, Pope and Young (2005) conclude that
17
outside directors also appear to play an important role in monitoring the integrity and credibility of financial statements.
2.4 Hypotheses Development
In this section, hypotheses related to the influence of CEO duality and the presence of independent board members on classification shifting will be formulated based on the literature and theories discussed in the previous section.
2.4.1 CEO Duality and Classification Shifting
An important board characteristic is CEO duality. As mentioned before, CEO duality refers to the situation when the chairman of the board is also the CEO of the firm (Johnson, 2019).
Agency theory suggests that separation of duties may lead to better monitoring of the board process (Fama & Jensen, 1983; Jensen, 1993). In the absence of a separation between the CEO and the chairman of the board the monitoring function of the board over earnings management may be threatened because the CEO has more discretion to manage financial reporting (Finkelstein & D’Aveni, 1994). In case of CEO duality, the CEO has more power over the firm and board without being supervised by a chairman.
According to Dechow, Sloan and Sweeney (1996) the likelihood of earnings manipulation is systematically related to weaknesses in the oversight of management. For instance, their study concludes that firms more often engage in earnings manipulation if the CEO also serves as the chairman of the board. Also, Sharma (2004) finds evidence that firms with greater board independence are less likely to engage in fraudulent reporting than those with CEO duality. However, prior research does not always show a negative impact arising from CEO duality (Xie et al., 2003; Davidson et al., 2005; García-Meca & Sánchez-Ballesta, 2009). Based on agency theory and previous research, CEO duality leads more often to an increase in earnings manipulation. Therefore, I expect the existence of a comparable effect between CEO duality and classification shifting. The following hypothesis is stated:
H1: Classification shifting is positively associated with CEO duality
2.4.2 Independent Board Members and Classification Shifting
Another important mechanism to evaluate board independence is the proportion of independent directors on the board. According to the agency theory, independent board members are necessary because there is a reason to believe that the agent will not always act in the best
18
interest of the principal (Jensen & Meckling, 1976). Results of Klein (2002) suggest that boards structured to be more independent from the CEO are more effective in monitoring the financial accounting process. She also argues that firms with a higher proportion of independent directors are less likely to engage in accrual-based earnings management. In addition, she concludes that this effect is the most pronounced when the board consist of a majority of independent directors.
Results of Xie et al. (2003) also show that firms are less likely to engage in accrual- based earnings management if their boards include more independent outside directors.
According to Peasnell et al. (2005) the likelihood a manager engages in accrual-based earnings management to avoid loss reporting and earnings reductions is negatively associated to the proportion of independent board members. Moreover, Osma (2008) finds evidence that firms with a higher proportion of outside board members are also less likely to engage in real earnings management. Based on this prior research regarding accrual-based and real earnings management, I expect a comparable effect of the proportion of independent board members on classification shifting. Therefore, the following hypothesis is formulated:
H2: Classification shifting is negatively associated with the proportion of independent board members
19
3 Data and Research Method
The purpose of this study is to examine whether U.S. companies use classification shifting as an earnings management method and whether the degree of board independence affects classification shifting practices used to inflate core earnings. This section provides the data and methodology used in this study to answer the research question. First, information will be provided about the data and sample selection used to test the hypotheses. The remaining parts describes how classification shifting and board independence are empirically measured.
3.1 Data and Sample Selection
This study focuses on data of U.S. firms within the S&P 1500. The S&P 1500 includes all stocks in the S&P 500, S&P 400 and S&P 600 and are therefore representative for the U.S.
economy. The data, used to test the hypotheses, is collected from Institutional Shareholder Services (ISS) and Compustat. Compustat is used to retrieve financial data to determine the presence of classification shifting in the U.S. ISS – Directors is used to collect data about CEO duality and the proportion of independent board members. The data is obtained from 2007 up to and including 2019, because data about board independence is available from 2007 and 2019 is the most recent year of which the data is available.
Following prior research (McVay, 2006; Haw et al., 2011) financial firms and firms with sales less than $1 million are eliminated. Observations with sales less than $1 million are removed to avoid outliers, because sales are used as a deflator for many of the variables. Also consistent with these prior studies, firms with fiscal year-end changes are removed to ensure the comparability of different fiscal years. Because this study focuses on U.S. firms, firms from other countries are removed. Also firms with another currency than U.S. dollar are removed from the sample to ensure comparability between firms. I require a minimum of 15 observations per industry per fiscal year to ensure a data pool that is necessarily large to estimate core earnings from the expectation models. Each firm- year observation is required to have enough data to test hypothesis 1 and 2. Finally, the regression models require one year of lagged data and one-year of ahead data. Therefore, the actual years examined are 2008-2018.
The final sample consists of 9,374 firm- year observations. Table 1 provides the sample selection procedure of the final sample.
20
Table 1
Sample Selection Procedure
Sample Selection Number of observations
Original sample Compustat 2007-2019 173.622
Financial firms (72.296)
Sales less than $1 million (19.680)
Observations of countries other than U.S. (23.815)
Observations with a currency other than U.S. Dollar (138)
Observations with fiscal year-end changes (2.591)
Observations with insufficient data to test hypothesis 1 and 2 (26.651) Observations within industries with less than 15 observations
per fiscal year
(1.997)
Merging with ISS - Directors (17.080)
Final Sample 9.374
3.2 Research Method
To test whether CEO duality and the proportion of independent board members affect classification shifting, I will focus on the misclassification of core expenses and the reversal of this misclassification in the next year. The association between unexpected core earnings and special items will be tested and in case of classification shifting, I expect that firm’s core earnings will be overstated when special items are recognized in the current year and reversed in the following year. To test whether S&P 1500 firms engage in classification shifting, the two-step procedure of McVay (2006) is used. This model of McVay (2006) tests the relation between special items and unexpected core earnings/unexpected change in core earnings. To determine unexpected core earnings, expected core earnings must be determined first. Expected core earnings (to examine year t) can be derived from model 1 (variables are defined in Table 2):
𝐶𝐸𝑡 = β0 + β1𝐶𝐸t–1 + β2𝐴𝑇𝑂t + β3𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 + β4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + β5∆𝑆𝐴𝐿𝐸𝑆𝑡 +
β6𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + εt (1)
21
According to McVay (2006) the association between unexpected core earnings in year t and special items in year t can also be alternatively explained. This association can also be caused by unexpectedly high core earnings due to immediate benefits of the restructuring charge or other real economic events. To distinguish opportunistic behavior from managers and real economic changes, it is expected that the increase associated with special items in year t reverses in year t+1. Therefore, firms that classification shift have (1) higher core earnings than the expected level of core earnings in year t and (2) a lower change in core earnings than expected in year t +1. To test whether the improvement of special items in year t reverses in year t+1 I will further follow the model of McVay (2006). To test this part, the change in core earnings is modelled. It is expected that the unexpected change in core earnings from year t to t+1 is declining in special items in year t. Expected change of core earnings (to examine year t+1) can be derived from model 2 (variables are defined in Table 2):
∆𝐶𝐸𝑡 = φ0 + φ1𝐶𝐸t–1 + φ2∆𝐶𝐸t–1 + φ3∆𝐴𝑇𝑂t + φ4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 + φ5𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + φ6∆𝑆𝐴𝐿𝐸𝑆𝑡
+ φ7𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + εt (2)
Table 2
Variable Definitions with Corresponding Compustat Data Item Numbers
Variable Definition
𝐶𝐸 Core earnings, defined as Sales – COGS – Selling, General and Administrative Expenses (#13) / Sales (#12)
∆𝐶𝐸𝑡+1 Change in Core Earnings (𝐶𝐸t+1 – CEt)
UE_CEt Unexpected Core Earnings, defined as the difference between reported and predicted Core Earnings
UE_∆CEt Unexpected Change in Core Earnings is the difference between reported and predicted Change in Core Earnings
%SIt Income-Decreasing Special Items as a Percentage of Sales (Special Items (#17) × –1 / Sales (#12)) if Special Items are income decreasing and 0 otherwise.
𝐴𝑇𝑂 Asset Turnover, defined as Sales (#12) / Average Net Operating Assets. Net Operating Assets is the difference between Operating Assets – Operating Liabilities. Operating Assets is calculated as Total Assets (#6) minus Cash and Short–Term Investments (#1).
Operating Liabilities is calculated as Total Debt (#9 + #34) minus Book Value of Common and Preferred Equity (#60 + #130) minus Minority Interest (#38). Average Net Operating Assets is required to be positive.
∆𝐴𝑇𝑂t Change in Asset Turnover, calculated as 𝐴𝑇𝑂t – 𝐴𝑇𝑂t–1
22
Consistent with the approach of McVay (2006) the explanatory variables of model 1 consists of six continuous variables. Lagged core earnings (𝐶𝐸t–1) is included because core earnings tend to be very consistent. The second included variable is asset turnover ratio (𝐴𝑇𝑂t), which has been proven to be inversely related to profit margin (McVay, 2006) and the definition of core earnings in this study is almost the same as profit margin. Model 1 also include prior-year accruals (𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1) because future performance is related to past accruals. Current year accruals (𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡) are included to control for extreme performance by accrual management. This is relevant because it is possible that extreme accruals could be caused by accrual management and therefore controlling for accruals results in a stronger prediction of core earnings associated with classification shifting. Sales growth (∆𝑆𝐴𝐿𝐸𝑆) is included to control for the impact of sales growth on fixed costs (as sales grow, fixed costs decrease per sales dollar). To allow the slope to differ between sales increases and decreases, negative sales (𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆) is included in the model.
To model the change in core earnings, lagged core earnings (𝐶𝐸t–1) and the change in 𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆 Operating Accruals, defined as ((Net Income before
Extraordinary Items #123 – Cash from Operations (#308 –
#124)) / Sales (#12)
∆𝑆𝐴𝐿𝐸𝑆 Percent Change in Sales = (Sales – Salest–1) / Salest–1
𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆 If ∆𝑆𝐴𝐿𝐸𝑆 is less than 0, and 0 otherwise
SIZE Firm Size, defined as the natural log of Total Assets (#6) LEV Leverage, defined as Total Debt (#9 + #34) / Total Assets (#6) CFO Cash Flow from Operations scaled by lagged Total Assets (#6) ROA Return on Assets, defined as Net Income (#172) / Average Total
Assets (Total Assetst – Total Assetst–1) / 2
BMV Book to Market Value, defined as Market Capitalization at year–
end ((Market Value of Equity (#24 x #25) / Book Value of Equity (#60))
CEOD CEO Duality when the CEO of the company is also the chairman of the board. This variable takes the value 1 when CEO Duality is present and 0 otherwise.
IND Percentage Independent Board Members, defined as Independent Members / Total Board Members. This variable takes the value 1 if the percentage of independent board members in relation to the total members on the board is higher than 70% and 0 otherwise.
All variables are winsorized at 1 percent and 99 percent. I assigned a zero to special items (#17), preferred stock at carrying value (#130) and minority interest (#38) if that data item is missing. Definitions of the Compustat data item numbers are included in Appendix I.
23
core earnings from year t–2 to t–1 (∆𝐶𝐸t–1) are included in model 2 to allow the model to vary the degree of main reversion based on the core earnings of the prior year. Asset turnover (𝐴𝑇𝑂t) is replaced for the change in asset turnover (∆𝐴𝑇𝑂t) in model 2 and the other variables are retained.
3.3 Testing Classification Shifting
To test whether S&P 1500 firms misclassify core expenses as special items, I will follow the models of McVay (2006). Consistent with Zalata and Roberts (2016) I added control variables to control for firm performance. I add the following variables to the models: firm size (SIZE), leverage (LEV), operating cash flow (CFO), return on assets (ROA) and book to market value (BMV):
UE_CEt = α0+ α1%SIt + α2SIZEt + α3LEVt + α4CFOt+ α5ROAt + α6BMVt + εt (3a)
UE_∆CEt+1 = n0+ n1%SIt + n2SIZEt+1 + n3LEVt+1 + n4CFOt+1 + n5ROAt+1 + n6BMVt+1 + εt
(3b)
Where UE_CEt refers to the unexpected core earnings in year t and UE_∆CEt+1 is the unexpected change in core earnings in year t+1. Unexpected core earnings and unexpected change in core earnings are calculated as the difference between the reported and predicted core earnings and change in core earnings, based on the models 1 and 2 above. %SIt is defined as income decreasing special items and is scaled by sales, both in year t. A positive special item refers to an income-decreasing special item, income-increasing special items are set to zero.
When firms engage in classification shifting, the unexpected core earnings increase with special items and therefore α1 is expected to be positive. If this result is indeed caused by managers misclassifying expenses, it is expected that n1 is negative because of the reversal in year t+1 of special items in year t. To test if board independence affects the misclassification of core expenses as special items, I include the interactions of special items with board independence proxies in model 4:
UE_CEt = α0+ α1%SIt + α2CEOD + α3IND + α4%SI × CEOD + α5%SI × IND + α6SIZE +
α7LEV + α8CFO + α9ROA + α10BMV + εt (4) Where CEOD refers to CEO duality. I categorize a firm as having a “dual CEO” when a CEO
also occupies board chair positions. This variable takes the value 1 when CEO duality is present
24
and 0 otherwise. IND is the proportion of independent board members, determined as a percentage of independent board members in relation to the total members on the board. This variable takes the value 1 if the percentage of independent board members in relation to the total members on the board is higher than 70% and 0 otherwise.
25
4 Results
This section contains the results of the study. First, the descriptive statistics and the correlation matrix of the main variables of interest are presented. The second part of this section contains the test results of the models to predict core earnings and the change in core earnings. Thirdly, this section provides the results that clarify whether classification shifting is a viable earnings management method in the U.S. and the influence of board independence on mitigation this form of earnings management. The last paragraph consists of the results of the robustness test regarding the model of expected core earnings/change in core earnings that is used to measure the degree of classification shifting in this study.
4.1 Descriptive Statistics and Correlation Matrix
Table 3 provides descriptive statistics for the main variables, which are winsorized at 1 percent and 99 percent to eliminate outliers. The mean (median) core earnings (𝐶𝐸t), as a percentage of sales, is 17.3% (15.7%). The mean of unexpected core earnings (UE_CEt) is 1.3% and is higher in relation to prior studies, which find a mean of or close to 0% (McVay, 2006;
Athanasakou et al., 2009; Zalata & Roberts, 2015). Mean income-decreasing special items (%SIt), as a percentage of sales is 2.10%, which is consistent with income-decreasing special items of 2.70% that McVay (2006) found. In line with McVay (2006) income-increasing special items are not included in the analysis and are set to zero. The mean of CEO Duality (CEOD) is 0.504, which means that in 50,4% of the firm years examined a CEO also served as the chairman of the board. The mean of firm years with more than 70% independent directors on the board (IND) is 0.820. This means that for 82% of the firm years examined, the board of directors consists of at least 70% independent board members.
Table 4 shows the correlation matrix of the main variables. Current year core (𝐶𝐸t) earnings are positively correlated (0.613) by last year core earnings (𝐶𝐸t–1), showing that core earnings are persistent. Consistent with the findings of McVay (2006), special items as a percentage of sales (%SIt) are negatively associated with core earnings (𝐶𝐸t). Special items as a percentage of sales (%SIt) are positively correlated (0.142) with unexpected core earnings (UE_CEt), which already can be an indication that firms engage in classification shifting to increase their core earnings.
26
Table 3
Descriptive Statistics
Mean Median Standard
deviation 25% 75%
SALESt (in millions) 8,740.065 2,179.391 26,350.907 841.045 6,160.388
∆𝑆𝐴𝐿𝐸𝑆t 6.80% 5.50% 0.203 –1.50% 13.30%
𝐶𝐸t 0.173 0.157 0.017 0.099 0.238
𝐶𝐸t–1 0.173 0.157 0.171 0.100 0.236
∆𝐶𝐸𝑡 0.000 0.002 0.107 –0.011 0.014
∆𝐶𝐸t–1 0.001 0.002 0.105 –0.012 0.014
UE_CEt 0.013 0.005 0.079 –0.014 0.032
UE_∆CEt+1 0.003 0.001 0.069 –0.019 0.022
%SIt 2.10% –0.40% 0.060 –0.00% –1.70%
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 –0.087 –0.055 0.154 –0.110 –0.021
𝐴𝑇𝑂t 2.436 –1.647 3.066 –1.003 –2.774
CEODt 0.504 1.000 0.500 0.000 1.000
INDt 0.820 1.000 0.384 1.000 1.000
All variables are winsorized at 1 percent and 99 percent. Variables are defined in Table 2.
Table 4 Correlation Matrix
SALESt ∆𝑆𝐴𝐿𝐸𝑆t 𝐶𝐸t 𝐶𝐸t–1 ∆𝐶𝐸𝑡 ∆𝐶𝐸t–1 UE_CEt UE_∆CEt+1 %SIt 𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 𝐴𝑇𝑂t
SALESt 1.000
∆𝑆𝐴𝐿𝐸𝑆t –0.010 1.000
𝐶𝐸t 0.007 0.218 1.000
𝐶𝐸t–1 0.004 –0.030 0.613 1.000
∆𝐶𝐸𝑡 0.002 –0.138 –0.311 –0.122 1.000
∆𝐶𝐸t–1 0.003 0.316 0.374 –0.433 –0.120 1.000
UE_CEt 0.018 –0.189 0.349 0.099 –0.114 0.266 1.000
UE_∆CEt+1 0.018 –0.068 –0.024 0.028 0.410 –0.037 0.007 1.000
%SIt –0.044 –0.103 –0.080 0.001 0.011 –0.100 0.142 –0.039 1.000
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 0.045 0.158 0.105 –0.106 –0.194 0.214 –0.208 –0.037 –0.487 1.000
𝐴𝑇𝑂t 0.142 0.059 –0.177 –0.206 –0.009 0.033 –0.059 –0.040 –0.109 0.168 1.000
All variables are winsorized at 1 percent and 99 percent. Variables are defined in Table 2.
4.2 Expectation Models
This paragraph provides results of the models of expected core earnings/expected change in core earnings, presented in Table 5 and 6, respectively. The first subparagraph consists of the results of the model of expected core earnings and the second subparagraph provides the results of the model of expected change in core earnings. Expected core earnings/expected change in core earnings have to be derived first before classification shifting practices can be measured.
4.2.1 Model of Expected Core Earnings
To calculate unexpected core earnings, expected core earnings must be derived first. Expected core earnings are derived from model 1:
𝐶𝐸𝑡 = β0 + β1𝐶𝐸t–1 + β2𝐴𝑇𝑂t + β3𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 + β4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + β5∆𝑆𝐴𝐿𝐸𝑆𝑡 + β6𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + εt (1)
Table 5 summarizes the regression results of the estimation model of expected core earnings.
Table 5
Model of Expected Core Earnings–Level Dependent variable: 𝐶𝐸t
Independent
variables Predicted sign Mean
Coefficient t-statistic p-value
𝐶𝐸t–1 + 0.786*** 362.87 <0.001
𝐴𝑇𝑂t – –0.006*** –16.10 <0.001
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 – –0.127*** –49.88 <0.001
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + ?0.221*** 11164.01 <0.001
∆𝑆𝐴𝐿𝐸𝑆𝑡 + ?0.125*** 165.06 <0.001
𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + ?0.380*** 157.75 <0.001
Adjusted R2 84.2%
There are 9,374 firm-year observations and 401 industry-year regressions for the 2008-2018 period.
Regressions are estimated cross-sectionally by industry and fiscal year, following model 1. For each of the independent variables, t-statistics and p-values are based on one-tailed t-tests due to the separate industry year regressions performed. Variables are defined in Table 2. *, **, *** are significant at the 10%, 5%, and 1%
level, respectively.
29
The adjusted R2 of the model of expected core earnings is 84.2%, which is quite high in comparison to prior studies where the adjusted R2 is 75.5% at the highest (McVay, 2006; Haw et al., 2011; Zalata & Roberts, 2015). All coefficients of the independent variables have the predicted signs and are significant at the 1% level. As predicted, prior year core earnings (𝐶𝐸t–
1) are a strong predictor of core earnings (CEt) with a coefficient of 0.786. The asset turnover ratio (𝐴𝑇𝑂t) is negatively associated with core earnings (CEt) with a coefficient of –0.006.
Prior-year accruals (𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1) are negatively associated with core earnings (CEt), which is consistent with higher level of accruals having lower earnings persistence (McVay, 2006).
Also, the positive coefficient of 0.221 on current-year accruals (𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡) is as predicted.
Consistent with Anderson et al. (2003) and McVay (2006) the coefficient on firms that face a sales decline (𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡) is 0.380 and therefore significantly larger than the coefficient of 0.125 for firms that experienced sales growth (∆𝑆𝐴𝐿𝐸𝑆𝑡).
4.2.2 Model of Expected Change in Core Earnings
To derive the unexpected change in core earnings, expected change in core earnings must be calculated first. The expected change in core earnings is derived from model 2:
∆𝐶𝐸𝑡 = φ0 + φ1𝐶𝐸t–1 + φ2∆𝐶𝐸t–1 φ3∆𝐴𝑇𝑂t + φ4𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 + φ5𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + φ6∆𝑆𝐴𝐿𝐸𝑆𝑡
+ φ7𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + εt (2)
Table 6 provides the regression results of the model of the expected change in core earnings. The adjusted R2 of the model of expected change of core earnings is 55.5%, which is also quite high in comparison to prior studies where the adjusted R2 is 51.7% at the highest (McVay, 2006; Haw et al., 2011; Zalata & Roberts, 2015). All coefficients of the independent variables have the predicted signs and are statistically significant at the 1% level. Consistent with mean reversion, the level of current core earnings is negatively associated with core earnings from the previous year (𝐶𝐸t–1) with a coefficient of –0.114. Also, the change in core earnings (∆𝐶𝐸t–1) is negatively associated with the change in core earnings in the prior year (coefficient of –0.027). As predicted, the change in asset turnover (∆𝐴𝑇𝑂t) ratio is positively related with the change in core earnings. Overall, the results in Table 5 and 6 are consistent with the results of McVay (2006) and indicate that the expectation models 1 and 2 work well for the sample of U.S. firms. Also, the expectation models have a higher level of explanatory power (adjusted R2) in comparison to prior research.
30
4.3 Measuring Classification Shifting
This paragraph consists of the results of the presence of classification shifting practices in the U.S. The first subparagraph provides the results of the misclassification of core expenses as special items and the reversal of special items in the next year (year t+1). The remainder of this section provides results of the impact of board independence on classification shifting practices.
4.3.1 Misclassification of Core Expenses as Special Items
Before investigating whether board independence mitigates classification shifting, I first investigate whether U.S. firms see classification shifting as a viable earnings management method. To test whether S&P 1500 firms misclassify core expenses as special items, model 3a is used:
UE_CEt = α0+ α1%SIt + α2SIZEt + α3LEVt + α4CFOt + α5ROAt + α6BMVt + εt (3a) Table 6
Model of Expected Core Earnings–Change Dependent variable: ∆𝐶𝐸𝑡
Independent
variables Predicted sign Mean
Coefficient t-statistic p-value
𝐶𝐸t–1 – –0.114*** –66.25 <0.001
∆𝐶𝐸t–1 – –0.027*** –7.65 <0.001
∆𝐴𝑇𝑂t + 0.001*** 6.65 <0.001
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡–1 – –0.114*** –53.62 <0.001
𝐴𝐶𝐶𝑅𝑈𝐴𝐿𝑆𝑡 + 0.180*** 76.82 <0.001
∆𝑆𝐴𝐿𝐸𝑆𝑡 + 0.155*** 21.59 <0.001
𝑁𝐸𝐺_∆𝑆𝐴𝐿𝐸𝑆𝑡 + 0.305*** 36.12 <0.001
Adjusted R2 55.3%
There are 9,374 firm-year observations and 401 industry-year regressions for the 2008-2018 period. Regressions are estimated cross-sectionally by industry and fiscal year, following model 1. For each of the independent variables, t-statistics and p-values are based on one-tailed t-tests due to the separate industry year regressions performed. Variables are defined in Table 2. *, **, *** are significant at the 10%, 5%, and 1% level, respectively.
31
Table 7 summarizes the regression results that test whether there is a relationship between special items (SIt) and unexpected core earnings (UE_CEt). When firms engage in classification shifting, the unexpected core earnings increase with special items and therefore α1 is expected to be positive. As expected, there is a positive significant relationship between special items and unexpected core earnings (a coefficient of 0.246; p-value <0.001), which suggests that some companies might have shifted some of their core expenses to special items to inflate core earnings.
4.3.2 Reversal of Special Items
When firms engage in classification shifting, the unexpected core earnings increase with special items and therefore, as mentioned before, α1 is expected to be positive. To distinguish opportunistic behavior from managers and real economic changes, it is expected that the increase associated with special items in year t reverses in year t +1. To test if there is a reversal in year t +1 of special items in year t, the following model is used:
UE_∆CEt+1 = n0+ n1%SIt + n2SIZEt+1 + n3LEV + n4CFOt+1 + n5ROAt+1 + n6BMV t+1 + εt (3b) Table 7
Regression of Unexpected Core Earnings on Special Items Dependent variable: UECEt
Independent variables Coefficient t-statistic p-value
Intercept –0.067*** –14.93 <0.001
%SIt –0.246*** –15.59 <0.001
SIZEt –0.007*** –12.18 <0.001
LEVt –0.004*** –0.86 0.392
CFOt –0.163*** –12.49 <0.001
ROAt –0.036*** –2.43 0.015
BMVt –0.000*** 0.71 0.480
Adjusted R2 6.80%
All variables are winsorized at the 1% and 99% level. Variables are defined in Table 2. *, **, *** are significant at the 10%, 5%, and 1% level, respectively.
