The effect of firm-level political
uncertainty on accrual-based earnings
management and income smoothing
AbstractThis research focuses on effects of firm-level political uncertainty on earnings management, and specifically income smoothing. This study is motivated by renewed concerns about the effect of political uncertainty on firm behavior, because of recent political events such as the longest government shutdown in the US. I use agency theory to predict that higher firm-level political uncertainty increases information asymmetry, which is a prerequisite for earnings management. Besides, higher firm-level political uncertainty increases the volatility in future earnings, creating incentives for managers to smooth their earnings. The conditioning role of CEO power is examined on firm-level political uncertainty and earnings management. Since powerful CEOs particularly have the discretion to make decisions, I expect that more powerful CEOs are associated with even more pronounced degrees of earnings management relative to their less powerful colleagues. The sample consists of 8309 firm-year observations of the period 2002-2016. This research uses the new measure of firm-level political uncertainty of Hassan et al. (2017), based on conference calls of listed US firms. Discretionary accruals are based on the modified Jones model, income smoothing on the correlation between discretionary accruals and pre-managed income, and CEO power on a constructed index. After controlling for firm size, growth, ROA, financial leverage, negative income, audit committees, board size, auditor type, female directors, age of the CEO and gender of the CEO, insignificant results are found. These results remain unchanged for using different regressions, different models for measuring discretionary accruals and extra analysis on the main variables. These results may be due to the sample period used or the construction of the sample which predominantly includes large listed companies. Moreover, firm-level political uncertainty may only have an effect on earnings management and income smoothing if it is included in a measure of overall external uncertainty a firm faces.
Rachel Stevelink S2977737
Master Thesis Accountancy & Controlling Supervisor: R. Hooghiemstra
24-06-2019
1
Table of Contents
Introduction ... 2 Theoretical framework ... 5 Research design ... 14 Sample ... 14 Dependent variables ... 15 Independent variables ... 16 Control variables ... 19 Theoretical model ... 20 Results ... 21 Univariate results ... 21 Multivariate results ... 26 Additional analysis ... 29Discussion and conclusion ... 33
Summary and conclusion ... 33
Limitations ... 35
Future research ... 36
2
Introduction
At the beginning of 2019, the longest governmental shutdown in the US history took place (Wagner, Rocha, & Wills, 2019). During the shutdown United States president Donald Trump claimed he was willing to keep the government closed for months or even years (Mills, 2019). The past years, similar events like the vote from the UK to leave the European Union and the Arab spring have renewed concerns about the effects of political uncertainty on investment, employment, and firm behavior. Economists, business leaders, and politicians are involved in intense debates about the question which aspects of political decision-making might be most disruptive to businesses and what is the size of such effects (Hassan, Hollander, van Lent, & Tahoun, 2017).
Political decisions on regulation, taxation, expenditures, and the enforcement of rules have major impact on the business environment. Even in well-functioning democracies, the outcomes of these decisions are often hard to predict, generating risk (Hassan, Hollander, van Lent, & Tahoun, 2017). Uncertainties associated with possible changes in government policy or national leadership have implications for the behavior of firms. Especially, during times of crisis and recession uncertainty about government policies increases (Julio & Yook, 2012). Firm-level political uncertainty is likely to affect the magnitude and the variability of firm performance. A way to potentially reduce this variability is the use of earnings management, especially income smoothing.
As discussed later, according to agency theory, firm-level political uncertainty increases the information asymmetry between shareholders and managers. Given the self-interest of the manager, the agent may not behave as agreed with the principal (Eisenhardt, 1989). Managers have several incentives to engage in earnings management. One of these incentives is reducing the variability of earnings. Generally Accepted Accounting Principles (GAAP) allow a degree of flexibility and discretion to manage earnings in response to firm-level political uncertainty. Consequently, more discretionary accruals may be evident during higher firm-level political uncertainty. A sample comprising 8309 firm-year observations for the period 2002-2016 is used to assess whether higher firm-level political uncertainty increases earnings management and income smoothing.
3 Finance literature has documented the effects of aggregate political uncertainty on financial decisions and firm value (Boutchkova, Doshi, Durnev, & Molchanov, 2012; Pastor & Veronesi, 2012; Pastor & Veronesi, 2013; Brogaard & Detzel, 2015). Studies have found that aggregate political uncertainty leads to lower corporate investments (Julio & Yook, 2012; An, Chen, Luo, & Zhang, 2016; Jens, 2017), lower cash holdings (Xu, Chen, Xu, & Chan, 2016), and a higher cost of equity (Li, Luo, & Chan, 2018). Ghosh & Olsen (2009) examined the relationship between overall external uncertainty, consisting of the unpredictability of actions of customers, suppliers, competitors and regulatory groups, and earnings management.
Absent in the literature is evidence on how changes in firm-level political uncertainty affect the use of earnings management, thereby possibly misleading users of accounting information or signal private information to shareholders. Investors rely on financial statements, including earnings, to assess the real value of corporations (Armstrong, Guay, & Weber, 2010) which is essential for making optimal investment decisions (Chen, Chen, Wang, & Zheng, 2018). Therefore, it is important to investigate the effects of firm-level political uncertainty on the use of accrual-based earnings management and income smoothing.
The following research question is examined:
“What is the effect of firm-level political uncertainty on accrual-based earnings management in general, and income smoothing in particular?”
Due to the information asymmetry between managers and shareholders, the intentions of managers on decisions about discretion over accruals are unobservable. Their intent cannot be assessed or verified using ex post accounting information. Therefore, Healy & Whalen (1999) state that future research is needed to identify conditions in which discretion in accrual accounting is more likely to occur. My research will add to the ongoing stream of earnings management literature by examining the effect of firm-level political uncertainty on earnings management. Particularly, the contribution focuses on the relationship between firm-level political uncertainty and income smoothing, since income smoothing is rarely investigated.
4 Additionally, this research will contribute to the research on political uncertainty by using a new measure focusing on individual firm-level political uncertainty. Quantifying the effects of political risk associated with specific political decisions has often proven difficulty, partially, due to a lack of measurement. Existing empirical evidence on political uncertainty focuses on broad metrics that capture the effect of a variety of types of uncertainty (Baker, Bloom, & Davis, 2016; Gulen & Ion, 2016). This research will focus on firm-level political uncertainty that individual US firms face based on the new measure constructed by Hassan et al. (2017). They used tools from computational linguistics to calculate the share of quarterly earnings conference calls that are devoted to political risks.
Lastly, this research also takes into account the effect of CEO power. The relationship between CEO power and financial reporting quality has been extensively studied. For example, research on firm characteristics associated with earnings management found that powerful CEOs are more likely to manipulate earnings through share repurchases (Farrell, Yu, & Zhang, 2013). Other research suggests that CEOs with greater power generally, and greater power over board directors, are associated with lower quality of financial reporting (Carcello, Neal, Palmrose, & Scholz, 2011; Bruynseels & Cardinaels, 2014; Lisic, Neal, Zhang, & Zhang, 2016).
However, the moderating effect of CEO power on the effect of earnings management has never been studied. Since, particularly powerful CEOs are able to successfully influence decisions (Shin, 2016), they have the ability to exert their will and thereby influence financial reporting to a greater extent than less powerful CEOs (Adams, Almeida, & Ferreira, 2005). Additionally, due to information asymmetries, a powerful CEO has the ability to meet personal preferences (Abernethy, Kuang, & Qin, 2015), for example in managing earnings. Accordingly, higher firm-level political uncertainty increases information asymmetry, giving powerful CEOs more incentive to use earnings management.
The remainder of the paper is structured as followed: section 2 reviews the literature and develops the hypotheses. Section 3 introduces the research design. Section 4 describes the data, presents the main empirical results and gives details on additional analyses. Finally, section 5 gives a conclusion and discussion.
5
Theoretical framework
The role of agency theory on earnings management
According to agency theory, a principal-agent problem arises due to the separation of ownership and control of the firm (Jensen & Meckling, 1976; Fama & Jensen, 1983). Agency theory assumes that both the principal (shareholder) and the agent (manager) are motivated by self-interest. An agency problem exists when the agent has incentives or goals that conflict with those of the principal (Boyd, 1994). Given the self-interest of the agent, the agent may not behave as agreed with the principal (Eisenhardt, 1989).
A main cause of agency problems is the existence of information asymmetry (Eisenhardt, 1989), indicating that shareholders have less information than managers (Dai, Kong, & Wang, 2013). The existence of information asymmetry hampers shareholders’ ability to verify if the agent has behaved appropriately (Jensen & Meckling, 1976; Zajac & Westphal, 1994) and leads to moral hazard problems. First, shareholders cannot directly observe CEO behavior and the resulting outcomes, giving CEOs relative impunity to act in their own interest (Ndofor, Wesley, & Priem, 2015). Second, CEOs have access to information that is not directly available to shareholders. Although management must report financial information to shareholders, that information is highly refined, incomplete, and subject to manipulation (Desai, Hogan, & Wilkins, 2006). This offers the opportunity for managers to engage in opportunistic behavior without fear of discovery (Ndofor, Wesley, & Priem, 2015).
Thus, various studies suggest that information asymmetry is a prerequisite for opportunistic behavior (such as earnings management) to materialize as under such circumstances principals do not have access to relevant information to monitor manager’s actions, while at the same time the majority of the shareholders lack sufficient resources and incentives to actually monitor the managers (Dye, 1988; Trueman & Titman, 1988; Richardson, 2000). Thereby, providing self-interested managers with the opportunity to behave opportunistically and increase their personal wealth at the expense of shareholders (Eisenhardt, 1989).
6
Earnings management
A manifestation of an agency problem between managers and shareholders is earnings management as it leads to lower earnings quality (Hooghiemstra, Hermes, Oxelheim, & Randoy, 2019). Earnings management occurs when managers try to control or adjust reported accounting earnings to maximize their own interests (Dai, Kong, & Wang, 2013). Healy & Wahlen (1999) define earnings management as “Managers using judgement in financial reporting and in structuring transactions to alter financial reports to either mislead stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers” (p. 368). In the same vein, Scott (2015) describes earnings management as “The choice by a manager of accounting policies, or real actions, affecting earnings so as to achieve some specific reported earnings objective” (p. 445).
An often studied form relating to accounting policy choices is the use of discretionary accruals, such as provisions for credit losses, warranty costs, inventory values and provisions for restructuring (Scott, 2015). This is called accrual-based earnings management and aims to obscure true economic performance by changing accounting methods or estimates within the generally accepted accounting principles (GAAP) (Dechow & Skinner, 2000). GAAP provides a degree of flexibility that allows opportunities for managers to use discretion in reporting earnings. Only, extreme forms of earnings management can be seen as fraud, where shareholders are purposely misled (Erickson, Hanlon, & Maydew, 2006). An example of accrual-based earnings management is executives using a less conservative estimation method to reduce the amount of bad debt expenses, so that earnings appear higher (Chen, Luo, Tang, & Tong, 2015). Accrual-based earnings management does not substantively modify the firm’s strategy nor does it change the basic parameters of operation1.
1 In contrast, real earnings management alters the execution of real business transactions by adapting the timing or
structure of these transactions. The operating activities of the firm are changed to meet or beat short-term earnings targets, which has direct cash flow consequences and potential long-term consequences for their economic value (Braam, Nandy, Weitzel, & Lodh, 2015). Examples include managing expenditures on advertising, R&D, maintenance, timing of purchases and disposal of capital assets, and stuffing the value channels (Scott, 2015). Real earnings management strategies are considered to be relatively costly compared to accrual-based earnings management, but are more difficult to detect (Graham, Harvey, & Rajgopal, 2005).
7
Incentives for earnings management
Prior research has found different motives for managers to conduct earnings management. Managers may want to obtain their own personal benefits from using earnings management. Healy & Wahlen (1999) find that managers manage earnings to increase their bonus rewards. Other research finds that managers manage earnings towards firm’s financial goals to obtain the target payout form their bonus plan (Gaver, Gaver, & Austin, 1995). Executives may also engage in earnings management to extract rents from shareholders in the form of increased equity compensation when this is based on the financial performance of the firm (Healy, 1985; Watts & Zimmerman, 1990). A large body of literature presents empirical evidence that CEO equity incentives influence accruals management (Glaum, Lichtblau, & Lindemann, 2004; Bergstresser & Phillippon, 2006). For example, Bergstresser & Phillippon (2006) find that CEOs with compensation plans that depend on the share price of the company will engage in earnings management to attain their desired compensation. Managers also use earnings management to reduce the likelihood of dismissal when firm performance is low (Weisbach, 1988).
Another motive for earnings management is meeting or beating investors’ earnings expectations. Firms that report earnings greater than expected generally enjoy a share price increase, as investors upwardly adjust their probabilities of good future performance. In contrast, firms with earnings lesser than expected suffer a share price decrease (Bartov, Givoly, & Hayn, 2002; Skinner & Sloan, 2002). Besides, managers may feel pressure to adjust earnings to meet certain targets to prevent being fired by the board of directors (Peasnell, Pope, & Young, 2005; Osma, 2008). Therefore, managers have strong incentives to ensure that earnings expectations are met.
Managers can also choose to use earnings management for managing share prices. When a firm plans to issue new or additional shares to the public, management faces a temptation to manage earnings upward to maximize the amount received from the share issue (Scott, 2015). Besides, earnings management may arise as a device to reduce the probability of debt covenant violation (Scott, 2015). Firms that are approaching violations of accounting-based restrictions are more likely to make income-increasing discretionary accounting changes (Sweeney, 1994). Even if default cannot be avoided, discretionary accruals are used to try improve the bargaining position in the event of renegotiation (DeFond & Jiambalvo, 1994).
8
The role of political uncertainty
Uncertainty places significant constraints on the firm, affecting its strategy and decision-making (Child, 1972). Uncertainty refers to the degree of change or variability from external factors relevant to an organization’s operations. This could stem from the unpredictability of the actions of customers, suppliers, competitors, and regulatory groups to which probabilities cannot be attached due to their constant change (Child, 1972; Dess & Beard, 1984). Agency theory emphasizes that firms operate under conditions of uncertainty, which lead to potential information asymmetries between the executives who manage the firm and the firm’s shareholders (Walker, 2013).
A recent literature has emerged that focuses on the effects of firm-level political uncertainty. By levying taxes, enforcing laws, providing subsidies, regulating competition, and defining environmental policies, governments shape the environment in which businesses operate (Pastor & Veronesi, 2012). By doing so, governments have the potential to influence various factors that affect firm behavior. For example, governments frequently modify tax laws with the intent of stimulating investment (Hassett & Metcalf, 1999). Furthermore, government influence may affect the cost structure of firms through federal contracts, entry and exit barriers, antitrust legislation, and regulation pertaining to employment and health care (Jurado, Ludvigson, & Ng, 2015).
Politicians and regulatory institutions frequently make decisions that alter the environment. Businesses often face a significant amount of uncertainty regarding the timing, content, and potential impact of policy decisions (Gulen & Ion, 2016). Uncertainties about which policies will be adopted also arise because of the differences in political preferences and objectives of individual policymaker’s (Pastor & Veronesi, 2012). Firm-level political uncertainty is defined as the risks associated with politics, including for example concerns about regulation, ballot initiatives, and government funding individual firms face (Hassan, Hollander, van Lent, & Tahoun, 2017).
Accounting and finance literature provide evidence that nationwide political uncertainty affects firm behavior. Firms that face nationwide political uncertainty lower their investment level (Kang, Lee, & Ratti, 2014; Wang, Chen, & Huang, 2014; Gulen & Ion, 2016). Julio and Yook (2012) find that firms reduce investment expenditures when nationwide political uncertainty is rising. Firms defer investments until this uncertainty is resolved. An et al. (2016) studied the corporate investment decisions surrounding turnover of local politicians in China. They find that firms
9 significantly reduce investments as these turnovers created political uncertainty. Another study finds that nationwide political uncertainty is strongly negatively associated with merger and acquisition activity at the firm level (Bonaime, Gulen, & Ion, 2018). This is consistent with theory that predicts that high levels of uncertainty will increase the option value of waiting, thus delaying investments, mergers, and acquisitions (Bloom, 2009). Chen et al. (2018) find that during periods of political turnovers, indicating increased nation-wide political uncertainty, firms have higher absolute levels of discretionary accruals and higher income decreasing accruals. Therefore, exposure to firm-level political risk can create both incentives and opportunities for firms to engage in earnings management (Gross, Königsgruber, Pantzalis, & Perotti, 2016). In the remainder of this sub section, I will provide arguments why firm-level political uncertainty creates incentives and opportunities.
Managers have opportunities to respond strategically to firm-level political uncertainty, earnings management being one of them. First, firm-level political uncertainty increases information asymmetry (Akerlof, 1970; Cheung & Krinsky, 1994; Umanath, Ray, & Campbell, 1996). Empirical evidence indicates that the greater information asymmetry between managers and shareholders, the greater the likelihood of the firm to manage earnings (Richardson, 2000). Second, due to higher firm-level political uncertainty, a lack of stability in accounting figures arises. Therefore, it is more difficult for shareholders to assess accounting numbers and, hence, to detect earnings management (Ghosh & Olsen, 2009).
Firm-level political uncertainty creates incentives for managers to manage earnings upward to avoid reporting losses and earnings declines (Park & Shin, 2004). As mentioned before, managers use earnings management to reduce the likelihood of being fired when firm performance is low (Weisbach, 1988; Peasnell, Pope, & Young, 2005; Osma, 2008). Furthermore, managers may use earnings management to still reach their financial targets and thus obtain their equity compensation goal.
In conclusion, managers have opportunities and incentives due to higher firm-level political uncertainty to engage in earnings management. Accordingly, the following hypothesis is formulated:
H1: Higher firm-level political uncertainty is positively related to accrual-based earnings management.
10
Income smoothing
Firm-level political uncertainty has the potential to affect the magnitude and the variability of firm performance (Cheng & Kesner, 1997). Higher firm-level political uncertainty implies an increase in the expected volatility of future reported earnings. This would lead to less smooth earnings figures being reported to investors and motivates managers to manage earnings (Schipper, 1989; Healy & Wahlen, 1999; Dechow & Skinner, 2000; Graham, Harvey, & Rajgopal, 2005), where earnings management is used as a buffer against the potential earning effects of uncertainty (McAnally, Srivastava, & Weaver, 2008). Income smoothing involves making use of the discretion provided by GAAP such that current (high) earnings are dampened and recorded in subsequent periods if the operating environment is expected to become tougher (Healy, 1985; Trueman & Titman, 1988).
Prior research shows that managers use income smoothing via discretionary accruals to reduce the variability in reported earnings imposed by uncertainty (Ghosh & Olsen, 2009). They find that external uncertainty, including the unpredictability of the actions of regulatory groups, induce greater variability in reported earnings. Subsequently, managers reduce this variability by using discretionary accruals to smooth earnings.
Graham et al. (2005) provide evidence that executives have strong preferences for reducing the variability in earnings. They find that more than three-fourths of managers are willing to sacrifice some economic value to achieve smooth earnings paths. Managers could be tempted to smooth earnings to stay between the lower and upper bonus compensation boundaries to maximize their awards over time (Dye, 1988). Evidence also suggests that managers smooth income for their job security (Fudenberg & Tirole, 1995; DeFond & Park, 1997; Riahi-Belkaoui, 2003). Furthermore, income smoothing can portray a less risky image of the firm (Wang & Williams, 1994; Gul, Cher, & Tsui, 2003) by reducing the perceived bankruptcy probability of the firm and, hence, the firm’s borrowing cost (Trueman & Titman, 1988). Besides, managers believe that investors pay more for a firm with a smoother income stream (Trueman & Titman, 1988), since smoother earnings improve the predictability of future earnings, which increase the stock price (Graham, Harvey, & Rajgopal, 2005).
11 Additionally, income smoothing can also benefit shareholders. Income smoothing can serve as a way to unblock communication to outsiders. Managers frequently obtain specialized information as part of their expertise. This information is often costly and difficult to communicate to the shareholders, therefore the communication is blocked. Discretionary accrual management is a way to credibly reveal management’s inside information about earnings expectations (Scott, 2015). Current and prospective investors will benefit if it enhances the information value of reported earnings (Wang & Williams, 1994). Uninformed investors will benefit since they could incur greater trading losses, when they need to trade for liquidity reasons, resulting from the unnecessary variability if earnings are not smoothed (Goel & Thakor, 2003).
In conclusion, due to flexible accounting standards, managers may use their discretion to reduce the additional variability that firm-level political uncertainty may impose on a firm (Bannister & Newman, 1996; Ghosh & Olsen, 2009). Therefore, the following hypothesis is formulated.
H2: Higher firm-level political uncertainty is positively related to income smoothing.
The moderating effect of CEO power
Board of directors play an important role in corporate governance. From an agency perspective, boards are regarded as monitoring devices that have to make sure that managers serve shareholders’ interests (Fama & Jensen, 1983). Effective board monitoring can help to reduce the self-serving behavior of managers. According to the literature, the board of directors is also important in constraining earnings management activities. Prior research shows a negative association between the proportion of independent directors on the board and earnings management. Independent directors provide more effective monitoring mainly because they have fewer conflicts of interests than inside or affiliated directors (Weisbach, 1988; Rosenstein & Wyatt, 1990). Besides, outside directors have incentives to be effective monitors because they want to maintain their reputational capital (Fama, 1980; Fama & Jensen, 1983).
The effectiveness of board monitoring might also depend on the structuring of the board. Boards often delegate work on important tasks to standing committees, for example the audit committee is responsible for overseeing the financial reporting (Klein, 1998). Higher audit committee quality, measured by independence and meeting frequency, are negatively associated with earnings
12 management (Beasley, 1996; Klein, 2002; Farber, 2005; Krishnan, 2005). Thus, the board of directors is a crucial corporate governance mechanism to limit earnings management.
While in theory, a more independently functioning board helps to mitigate earnings management, recent research shows that the accumulated amount of power of the CEO carries important implications for the effectiveness of boards. The CEO is generally regarded as the most powerful member in the organization (Daily & Johnsen, 1997). Power is the capacity of individual actors to exert their will (Finkelstein, 1992). CEO power is the ability to make and carry out strategic decisions regardless of the board’s position or influence (Finkelstein & Hambrink, 1996; Feng, Ghosh, & Sirmans, 2005).
A powerful CEO has the ability to even engage in potentially value-destroying activities (Finkelstein, 1992; Finkelstein & Hambrink, 1996). Agency theory (Jensen & Meckling, 1976) argues that an increase in CEO power exacerbates agency problems. Managers could be motivated to take advantage of information asymmetries and use their influence on the board to maximize their personal wealth at the expense of shareholders’ utility (Morse, Nanda, & Seru, 2011). The managerial power theory (Bebchuk, Fried, & Walker, 2002) suggests that CEO power affects the design of compensation contracts, and powerful CEOs are more likely to receive an increase in bonus and equity compensation (Henderson, Masli, Richardson, & Sanchez, 2010). Research shows that firms with powerful CEOs exhibit lower firm value, lower profitability, more negative market reactions to acquisition announcements, poorer credit ratings, and higher costs of debt (Bebchuk, Fried, & Walker, 2002; Bebchuk, Cohen, & Ferrell, 2009; Liu & Jiraporn, 2010). Accordingly, I also expect that more powerful CEOs prop up the firm’s earnings more than their less powerful colleagues.
Finkelstein et al. (2009) argue that CEO power is a multidimensional construct consisting of structural, ownership, expert, and prestige power. Structural power is based on the formal position of the CEO in the firm (D'Aveni & Kesner, 1993; Daily & Johnsen, 1997). Generally, the CEO has legitimate authority over others due to the nature of the position (Astley & Sachdeva, 1984; Finkelstein, 1992). CEO ownership power reflects that the CEO may be part of the firm’s founding family or owns a substantial proportion of the firm’s shares. On the one hand, a CEO who is also the founder or related to the founder may gain power through their often long-term interaction with the board (Finkelstein, 1992). On the other hand, prior research suggests that equity ownership by
13 a CEO may partially shield her from internal monitoring mechanisms, to the extent that it may be difficult for the board to fire the CEO (Denis, Denis, & Sarin, 1997). CEO expert power entails having comprehensive knowledge and understanding of firm operations and its environment (Firstenberg & Malkiel, 1994). Increased experience as CEO causes greater familiarity with firm capabilities and operations (Alderfer, 1986; Singh & Harianto, 1989). CEO prestige power refers to the reputation or status of the CEO. Directors may monitor prestigious CEOs less, because they will equate their image with successful leadership and performance (Hengartner, 2006). Together, these four dimensions define CEO power.
Prior research shows that boards are less effective monitors in case the firm is led by a powerful CEO. Specifically, an increasing body of empirical research demonstrates that powerful CEOs are associated with lower quality of earnings, and are more likely to restate financial statements (Adams, Almeida, & Ferreira, 2005; Efendi, Srivastava, & Swanson, 2007; Feng, Ge, Luo, & Shevlin, 2011). Besides, CEOs with power are more likely to manipulate earnings to meet or beat analysts’ forecasts (Mande & Son, 2012). A potential reason for this positive association between CEO power and the degree of earnings management is that CEOs tend to appoint directors who share their style and preferences, allowing them to exert more pressure on reporting judgements and decisions (Carcello, Neal, Palmrose, & Scholz, 2011; Bishop, DeZoort, & Hermanson, 2017).
Research finds that particularly powerful CEOs are able to successfully influence decisions (Shin, 2016). Individuals cannot take actions based on motivation alone if they lack sufficient and appropriate ability to control resources and influence others (Reinholt, Pedersen, & Foss, 2011). Thus, powerful CEOs have the ability to exert their will and thereby influence financial reporting to a greater extent than less powerful CEOs (Adams, Almeida, & Ferreira, 2005). Additionally, due to information asymmetry between shareholders and CEOs, a powerful CEO has the ability to meet personal preferences (Abernethy, Kuang, & Qin, 2015), for example, in managing earnings. Since, higher firm-level political uncertainty increases information asymmetry, powerful CEOs have more incentive to use earnings management.
Therefore the following hypothesis is formulated:
H3: CEO power strengthens the positive association between firm-level political uncertainty and accrual-based earnings management.
14
Research design
Sample
The initial sample consists of the dataset of Hassan et al. (2017) which contains listed firms in the United States during the period 2002 until 2016. All other data is merged with this dataset. The financial data needed for the measure of earnings management and firm-level control variables is collected from Compustat. The data on corporate governance variables is collected from GMI, data about CEOs is collected from Execucomp and GMI.
First, I delete observations because of missing data in determining the extent of earnings management and income smoothing. The estimations of earnings management and income smoothing are based on all data available in Compustat, because this gives the most accurate representation of these variables. Due to the unique nature of accounting of financial firms it is hard to estimate accruals and discretionary accruals for them. Therefore, the data on earnings management excludes banks (SIC-codes 6000-6199) and insurance companies (SIC-codes 6300-6411). In total 20232 observations are deleted due to missing earnings management data. Second, 3 non-US firms are deleted from the sample, since this research focuses on the USA. Third, 16157 observations are deleted due to insufficient data on the variables needed for the CEO power index. Fourth, missing data on corporate governance, CEO and firm-level control variables are deleted, these are 1, 81 and 6 observations respectively. After eliminating data that does not suit the requirements, the final sample consists of 8309 observations. Table 1 summarizes how the final sample is constructed.
Table 1 Sample construction
Observations Initial sample for 2002-2016
Missing earnings management data & deletion of SIC codes 6000-6199 & 6300-6411 Non-US firms
Missing data on CEO power index
Missing corporate governance control variables data Missing CEO control variables data
Missing firm-level control variables data
44789 (20232) (3) (16157) (1) (81) (6) Final sample 8309
15
Dependent variables
Earnings management
Since only management knows which part of reported accruals is economically justified and which part has been manipulated, it is hard in practice to measure the proportion of discretionary accruals. The modified Jones model by Dechow et al. (1995) is one of the most commonly used models to measure discretionary accruals. The Jones (1991) model takes into consideration the effect of changes in a firm’s economic circumstances on non-discretionary accruals by incorporating total assets, property, plant and equipment, and changes in revenues as variables. The modified Jones model incorporates the change in receivables since management can exercise discretion over revenues (Dechow, Sloan, & Sweeney, 1995).
Focusing on total accruals to identify accruals is a simple and easy way in many circumstances (Dechow & Schrand, 2004). This is supported by Dechow et al. (2003), who show that the correlation between estimated discretionary accruals and total accruals from the modified Jones model is more than 80%. Therefore, total accruals are used as a measure for earnings manipulation.
Earnings management will be measured for all firms k in industry j for year t based on the expected accrual-based modified Jones model (Dechow, Sloan, & Sweeney, 1995).
TACjk,t = αj,t[1/TAjk,t-1] + βj,t[ΔREVjk,t - ΔRECjk,t] + γj,t[PPEjk,t] + εjk,t (1)
The total accruals (TACjk,t), change in revenues (ΔREVjk,t), change in net sales (ΔRECjk,t) and gross property, plant and equipment (PPEjk,t) are each deflated by lagged total assets (TAjk,t-1). To check for expected, for example economic-based, components in total accruals, the changes in revenues and receivables, and PPE are used.
Thereafter, the discretionary accrual is calculated for each firm-year ij,t of the sample with the following equations:
EARNMANij,t = TACij,t -{αj,t[1/ TAij,t-1] + βj,t[ΔREVij,t - ΔRECij,t ] + γj,t[PPEij,t]} (2)
in this equation αj,t, βj,t and γj,t are the estimated coefficients (per 2-digit SIC industry) from the first equation.
16
Income smoothing
Following Leuz et al. (2003), Tucker & Zarowin (2006), and Myers et al. (2007), income smoothing will be measured by the (negative) correlation between the change in discretionary-accruals (ΔDAP) and the change in pre-discretionary income (ΔPDI). The assumption of this measure is that there is an underlying pre-managed income series and that managers use discretionary accruals to make the reported series smooth.
To estimate discretionary accruals the cross-sectional modified Jones model is used. The discretionary accruals (DAP) are calculated following equation 2. The pre-discretionary income (PDI) is calculated as net income minus discretionary accruals (PDI = NI – DAP).
Income smoothing (INCOMESMOOTH) is the correlation between the change in discretionary accruals and the change in pre-discretionary income: Corr(ΔDAP, ΔPDI). Following Tucker & Zarowin (2006), the current year and past four years’ observations are used to calculate the change. Therefore data from 1998 until 2016 is used for the measure. The more negative the correlation, the higher the degree of income smoothing in a certain year.
The parameters in both models are estimated by industry and each firm-year has to have at least eight observations with the same two-digit SIC code. All the variables used in the regression are winsorized at the 1st and 99th percentiles, to mitigate the effects of outliers. Finally, the earnings management and income smoothing proxies are merged with the dataset of Hassan et al. (2017).
Independent variables
Firm-level political uncertainty
The measure of firm-level political uncertainty is obtained from the database of Hassan et al. (2017). They use textual analysis of transcripts of quarterly earnings conference-calls between participants and firm management to construct a firm-level measure of political risk faced by individual listed firms in the US. Prior research finds that the discussions in conference calls typically centers around uncertainties the firm is facing (Bowen, Davis, & Matsumoto, 2002; Hollander, Pronk, & Roelofsen, 2010).
17 A pattern-based sequence-classification method developed in computational linguistics is adapted to distinguish between language associated with political versus non-political topics. A training library of political text (an undergraduate political science textbook, text from the political section of newspapers) and a training library of political text (an accounting textbook, text from non-political sections of newspapers, and transcripts of speeches on non-non-political texts) are used to identify two-word combinations (bigrams) that are frequently used in political texts. The number of instances in which conference-call participants use these bigrams in conjunction with synonyms for risk or uncertainty is counted and divided by the total length of the conference call to obtain a measure of the share of the conversation that is concerned with political risks, PRISK (Hassan, Hollander, van Lent, & Tahoun, 2017).
Since this database contains quarterly measures of PRISK, the quarter containing the ending month of the fiscal year equals the yearly measure of PRISK. The end fiscal year month data is gathered from Compustat. PRISK is the proxy for firm-level political uncertainty where a higher value indicates more firm-level political uncertainty.
CEO power
There is no one generally agreed upon definition of CEO power. In general, CEOs are perceived powerful if they can influence strategic decisions despite potential opposition from board members or other top executives. Finkelstein (1992) identified four sources of power: structural power, ownership power, expert power, and prestige power. Tang et al. (2011) modified Finkelstein’s (1992) measure by omitting the prestige source, arguing that it’s not a proximal measure of executive power when compared to the other measures.
The most commonly used proxies for structural power are CEO duality and board independence. CEO duality refers to boards where the CEO is the chairman. CEO duality impairs board independence and the monitoring role of the board, which results in an increase in CEO power (Jensen, 1993; Boyd, 1994; Efendi, Srivastava, & Swanson, 2007). A dummy variable for CEO duality is created coding 0 for separated CEO and board chair roles, and 1 for combined CEO/chair role. Board independence is used since research shows that boards with a higher ratio of independent directors are better able to keep managers in check and less likely to support decisions
18 that could harm shareholders in the long run (Rosenstein & Wyatt, 1990; Larcker & Tayan, 2011). In contrast, boards with a higher ratio of inside directors are less effective in monitoring (Beatty & Zajac, 1994). Thus, the higher the proportion of dependent directors on the board, the more likely the CEO can exert influence over them (Ryan & Wiggings, 2004; Morse, Nanda, & Seru, 2011). The dummy variable Board independence is created equaling 1 if the proportion of dependent directors of the firm is above 50% and 0 otherwise.
Ownership power is measured by the stock ownership of the CEO. Greater CEO stock ownership reduces the influence of the board and enables the CEO to exercise more discretion in making decisions, thus increasing CEO power (Finkelstein, 1992). CEO ownership is a dummy variable that will equal 1 if the total amount of shares held by the CEO is greater than 5 percent of the outstanding shares. Research suggests that entrenched CEOs become more powerful once they own above 5% of the firm (Hermalin & Weisbach, 1988).
Expert power will be measured by CEO tenure because it is associated with comprehensive knowledge and understanding of the company’s operating environment (macroeconomic, industry, supply chain, product market competition) and long term relationships with key stakeholders (Firstenberg & Malkiel, 1994). CEO tenure increases the CEO’s influence over the board and therefore increases CEO power (Ryan & Wiggings, 2004). Besides, longer tenure provides CEOs with the opportunity to accumulate social capital and knowledge which serves as an important source of informal power (Greve & Mitsuhasi, 2007). The tenure of the CEO is based on the number of years the CEO is active in the function of CEO in the firm. The number of years is measured by taking the observation year and subtracting the year the CEO was appointed. CEO tenure is a dummy variable equaling 1 if CEO tenure is above the sample median tenure and 0 otherwise.
CEOPOWER is the index variable of CEO power and therefore the sum of each of the variables listed. The variable ranges from 0 to 4. A higher index value indicates greater CEO power.
19
Control variables
There are other variables that potentially could influence the level of earnings management. Therefore, a number of firm-specific, corporate governance and CEO-specific variables are included in this research.
Firm-specific control variables
The firm-specific control variables include: firm size (FIRMSIZE) measured by the logarithm of the total assets, firm growth (GROWTH) measured by the sales growth in the current year compared to the previous year, return on assets (ROA) measured by EBIT divided by total assets and financial leverage (FINLEV) measured by total debt divided by total assets. Next to that, the dummy variable LOSS_D is created equaling 1 if the firm experienced a loss in the observation year and 0 otherwise. Prior studies show that these characteristics are associated with earnings management (Xie, Davidson III, & DaDalt, 2003; Peasnell, Pope, & Young, 2005; Larcker, Richardson, & Tuna, 2007; Dhaliwal, Naiker, & Navissi, 2010; Srinidhi, Gul, & Tsui, 2011; Chen, Folsom, Paek, & Sami, 2013; Chiu, Teoh, & Tian, 2013).
Corporate governance related control variables
Several corporate governance control variables are included. First, AC is included as a dummy variable assuming the value of 1 if the company has an audit committee and 0 otherwise. An audit committee composed of members of the board of directors supervises the internal control process and carries responsibilities related to external auditors and the accounting process. Prior research demonstrates that this is an important mechanism to reduce earnings management (Beasley, 1996; Klein, 2002; Bédard, Chtourou, & Courteau, 2004). Second, I control for board size as this serves as a measure of board effectiveness (Peasnell, Pope, & Young, 2005; Chiu, Teoh, & Tian, 2013). BOARDSIZE is measured as the logarithm of the number of directors on the firm’s board. Third, the presence of female directors on the board is included. Previous research proved that having one or more female directors is associated with tougher monitoring and reduced earnings management (Srinidhi, Gul, & Tsui, 2011; Adams & Ferreira, 2009). The variable FEMDIR is the number of female directors on the board. Fourth, firms that are audited by Big 4 auditors (EY, KPMG, PWC & Deloitte) are considered to have less discretionary accruals (Watkins, Hillison, & Morecroft, 2004). The variable BIG4 is a dummy variable equaling 1 if the firm is audited by one of the Big 4 firms and 0 otherwise.
20
CEO-specific control variables
The first CEO-specific control variable is the age of the CEO. Research suggests that the older CEOs get, the more ethical and more conservative they become. Therefore, older CEOs are less likely to engage in earnings management (Huang, Rose-Green, & Lee, 2012). CEOAGE is the logarithm of the number of years the CEO has been alive. The second control variable is if the CEO is female. Prior research shows that female presence in top management is negatively associated with discretionary accruals (Na & Hong, 2017). CEOFEM is the dummy variable equaling 1 if the CEO is a woman and 0 otherwise.
Lastly, I include year and industry dummies.
Theoretical model
Based on the literature review provided above, the following equations can be used to answer the research question:
EARNMAN i,t = α + β1*Firm-level political uncertainty i,t + β2*CEO power i,t + β3* β4* (Firm-level
political uncertainty i,t * CEO power i,t) + β5* FIRMSIZE i,t + β6* GROWTH i,t + β7* ROA i,t + β8*
FINLEV i,t+ β9* LOSS_D i,t+ β10* AC i,t+ β11* BOARDSIZE i,t+ β12* FEMDIR i,t+ β13* BIG4 i,t+
β14* CEOAGE i,t+ β15* CEOFEM i,t+ β16* INDUSTRY i,t + β17* YEAR i,t+ ε i,t
INCOMESMOOTH i,t= α + β1*Firm-level political uncertainty i,t+ β2* FIRMSIZE i,t+ β3* GROWTH i,t
+ β4* ROA i,t+ β5* FINLEV i,t+ β6* LOSS_D i,t+ β7* AC i,t+ β8* BOARDSIZE i,t + β9* FEMDIR i,t
+ β10* BIG4 i,t+ β11* CEOAGE i,t+ β12* CEOFEM i,t+ β13* INDUSTRY i,t β14* YEAR i,t+ ε i,t
In which i is a subscript of firms, t is a subscript of time, β is the estimated coefficient and ε the error term.
The ordinary least squares regression with standard errors clustered by firm is used to test the hypothesis.
21
Results
Univariate results
The univariate results present the descriptive statistics and the correlation of the variables used in this paper. Table 2 presents the mean and standard deviation (SD) values for all variables. Outliers and their potential effects on the results are accounted for by winsorizing the data at 1% at both tails. Only EARNMAN and PRISK are winsorized at 2.5% at both tails. The number of observations (N) is 83092.
Table 2 shows that the average discretionary accruals is 0,017 varying between -1,676 and 1,873. The average of absolute discretionary accruals is 0,346. The average of income smoothing is -0,933 with a minimum of -1 and a maximum of 0,225. Other studies with income smoothing have an average of -0,709 (Tucker & Zarowin, 2006). The average of PRISK is 93,571 with a minimum of 0 and a maximum of 412,831. The average of the CEO Power index is 1,799 with a minimum of 0 and maximum of 4. This is in line with earlier research on CEO power with an average of 1,885 and a SD of 0,848 (Tien, Chen, & Chuang, 2013).
With regard to the control variables, the average firm size is 6531,46 with a minimum of 89,665 and a maximum of 92385. The average sales growth is 9,1% compared to the previous year. The average ROA is 9,4% and the financial leverage 0,181. This indicates that the firms in the sample, on average, efficiently use their assets and have relatively low debt compared to their assets.
87,8% of the firm-years have an audit committee present. 91,7% of the firm-years is audited by a BIG 4 auditor. The average age of the CEO is 57,760, and in 2,5% of the firm-years the firm is led by a female CEO. The average board consists of 8,954 people with a minimum of 5 and a maximum of 14. On average, 1,817 women are members of the board with a minimum of 0 and a maximum of 8.
2 Winsorizing PRISK at 2.5% at both tails is in accordance with Hassan et al. (2017). EARNMAN is winsorized at
22 Table 2
Descriptive statistics
Variable N Mean SD Min Max
EARNMAN 8309 0,017 0,688 -1,676 1,873 ABS(EARNMAN) 8309 0,346 0,573 0,000 1,873 INCOMESMOOTH 8309 -0,933 0,202 -1 0,260 TA 8309 -0,062 0,068 -0,315 0,146 PRISK 8309 93,571 110,993 0 412,831 CEOPOWER 8309 1,799 0,927 0 4 FIRMSIZE 8309 6531,46 14247,38 89,665 92385 GROWTH 8309 0,091 0,206 -0,444 0,984 ROA 8309 0,094 0,089 -0,238 0,367 FINLEV 8309 0,181 0,176 0 0,757 LOSS_D 8309 0,150 0,357 0 1 AC 8309 0,878 0,327 0 1 BOARDSIZE 8309 8,954 2,059 5 14 BIG4 8309 0,917 0,276 0 1 FEMDER 8309 1,817 1,745 0 8 CEOAGE 8309 56,760 7,334 41 78 CEOFEM 8309 0,025 0,157 0 1
EARNMAN are the discretionary accruals from the accruals prediction error from the Modified Jones model, e.g. the difference between total accruals and estimated expected accruals. See Eq. (2)
ABS are the absolute values.
ÍNCOMESMOOTH is the Pearson correlation between the change in discretionary accruals and the change in pre-discretionary income over a window of 5 years.
TA are the total accruals calculated as the difference between net income minus cash flow from operations, deflated by lagged total assets.
PRISK is the amount of political bigrams in conjunction with synonyms for risk or uncertainty divided by the total length of the conference call.
CEOPOWER is the index variable for CEO power consisting of the sum of the dummy’s for CEO duality, board independence, CEO stock ownership and CEO tenure.
FIRMSIZE are the total assets.
GROWTH is the change in sales divided by lagged sales
ROA is the return on assets calculated as EBIT divided by total assets.
FNLEV is the financial leverage of the firm calculated as total debt divided by total assets. LOSS_D is a dummy variable equaling one if a loss occurred in the fiscal year.
AC is a dummy variable equaling one if an audit committee is present. BOARDSIZE is the number of directors on the board.
BIG4 is a dummy equaling one if the firm is audited by KPMG, Deloitte, PWC or EY. FEMDER is the number of female directors on the board.
CEOAGE is the number of years the CEO is alive.
23 Table 3 provides details on year basis for the main variables used in this research. When looking at the development of absolute discretionary accruals, it’s striking to see that during the period 2002 to 2007, the absolute earnings seem to be stable with an average of 0,261. This is lower than the average of the entire sample (0,346). In the period 2008-2016, the average discretionary accrual is 0,402. Especially during the period 2010-2014, high discretionary accruals are found with an average of 0,433. Contrary, in the period 2002-2005 low discretionary accruals are reported with an average of 0,235.
Looking at the year averages of the measure of income smoothing, the values seem to be stable. Income smoothing starts with an average value during the period 2002-2006 of -0,948. Following lower values during 2007-2010 with an average of -0,913 and thereafter increasing values with an average of -0,934 during 2011-2016. However, these values are all close to the mean of the entire sample of -0,933.
The average values of firm-level political uncertainty show that in 2002 the firm-level political uncertainty starts with an average of 95,087, which is close to the sample average. During 2003-2007 the average value is very low, with an average of 78,144. However, in the period 2008-2012, an increase of 38% is reported with an average of 107,821. Following the period 2013-2016, where the values decrease with a period average of 94,663. Again, similar to the period average.
Table 3
Details on main variables
Year Mean ABS(EARNMAN) Mean INCOMESMOOTH Mean PRISK 2002 0,162 -0,934 95,087 2003 0,260 -0,944 87,058 2004 0,262 -0,959 81,879 2005 0,254 -0,960 72,500 2006 0,377 -0,942 73,227 2007 0,252 -0,916 76,058 2008 0,398 -0,907 110,574 2009 0,344 -0,917 107,484 2010 0,438 -0,914 104,411 2011 0,342 -0,930 101,509 2012 0,449 -0,939 115,126 2013 0,508 -0,934 98,848 2014 0,428 -0,940 90,858 2015 0,321 -0,921 89,845 2016 0,392 -0,938 99,099 Average 0,346 -0,933 93,571
24 The correlation matrix is presented in table 4. I find high correlations between FIRMSIZE and BOARDSIZE (0,595) and between ROA and LOSS_D (0,537), which potential suggest multicollinearity concerns. However, all VIFs remain well below the value of 10, which has been recommended as the maximum level (O'Brien, 2007). This indicates that multicollinearity should not pose a severe threat to the validity of the results.
25 Table 4 Correlation matrix (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) ABS(EARNMAN) (1) 1 INCOMESMOOTH (2) -0,195 ** 1 PRISK (3) 0,049 ** -0,003 1 CEOPOWER (4) 0,004 -0,001 0,008 1 FIRMSIZE (5) -0,031 ** -0,037 ** 0,008 0,024 * 1 GROWTH (6) 0,049 ** -0,055 ** -0,028 * -0,016 -0,067 ** 1 ROA (7) 0,011 -0,018 -0,035 ** 0,028 * 0,080 ** 0,185 ** 1 FINLEV (8) -0,014 0,069 ** -0,044 ** -0,040 ** 0,271 ** 0,003 -0,025 * 1 LOSS_D (9) 0,011 0,082 ** 0,023 -0,027 * -0,187 ** -0,179 ** -0,537 ** 0,098 ** 1 AC (10) -0,081 ** 0,008 0,014 -0,027 * -0,072 ** 0,037 ** 0,002 -0,026 * 0,021 1 BOARDSIZE (11) -0,013 0,020 0,026 * -0,043 ** 0,595 ** -0,104 ** 0,069 ** 0,157 ** -0,114 ** -0,035 ** 1 FEMDIR (12) 0,023 * 0,019 -0,011 -0,021 0,425 ** -0,132 ** 0,066 ** 0,087 ** -0,088 ** -0,097 ** 0,474 ** 1 BIG4 (13) 0,001 -0,033 ** 0,002 -0,062 ** 0,267 ** -0,025 * 0,010 0,119 ** -0,045 ** 0,016 0,202 ** 0,163 ** 1 CEOAGE (14) -0,030 ** 0,030 ** 0,005 0,294 ** 0,160 ** -0,051 ** 0,031 ** 0,057 ** -0,056 ** -0,031 ** 0,114 ** 0,034 ** 0,013 1 CEOFEM (15) -0,004 0,003 -0,062 ** -0,047 ** -0,006 -0,006 0,010 -0,020 -0,014 -0,013 -0,001 0,273 ** 0,007 -0,076 ** 1 ** and * correlation coefficient is significant with respectively 1 and 5 percent.
ABS (EARNMAN) are the absolute discretionary accruals from the accruals prediction error from the Modified Jones model, e.g. the difference between total accruals and estimated expected accruals. See Eq. (2)
INCOMESMOOTH is the Pearson correlation between the change in discretionary accruals and the change in pre-discretionary income over a window of 5 years. PRISK is the LOG amount of political bigrams in conjunction with synonyms for risk or uncertainty divided by the total length of the conference call.
CEOPOWER is the index variable for CEO power consisting of the sum of the dummy’s for CEO duality, board independence, CEO stock ownership and CEO tenure. FIRMSIZE is the LOG of total assets.
GROWTH is the change in sales divided by lagged sales
ROA is the return on assets calculated as EBIT divided by total assets.
FNLEV is the financial leverage of the firm calculated as total debt divided by total assets. LOSS_D is a dummy variable equaling one if a loss occurred in the fiscal year.
AC is a dummy variable equaling one if an audit committee is present. BOARDSIZE is the number of directors on the board.
BIG4 is a dummy equaling one if the firm is audited by KPMG, Deloitte, PWC or EY. FEMDER is the number of female directors on the board.
CEOAGE is the LOG number of years the CEO is alive.
26
Multivariate results
Table 5 presents the main results of the first linear regression model with absolute discretionary accruals based on the Modified Jones model as the dependent variable. The OLS regression with standard errors clustered by firm is used to test the empirical predictions. Additionally, I control for other factors which are likely to affect the absolute discretionary accruals, since research shows that falsely rejecting the null hypothesis for no discretionary accruals can be prevented by controlling for confounding factors (Bartov, Gul, & Tsui, 2000).
Table 5 model 2 includes the independent variable firm-level political uncertainty and adds 0,003 explanatory power compared to model 1 (with only the control variables). This brings the R-squared value to 0,323. The results show that an insignificant positive coefficient is found between PRISK and absolute discretionary accruals (β = 0,001; not significant). Since the relationship is insignificant, hypothesis 1 is rejected.
Table 5 model 3 consists of the control variables and the moderating CEO power variable. In comparison with model 1, this model adds 0,003, bringing the R-squared to 0,323. For the interaction of CEO power on the relationship between PRISK and absolute discretionary accruals a positive insignificant coefficient is found (β = 0,006; not significant). Due to the insignificant relationship, hypothesis 3 is also rejected. 3
The coefficients regarding the control variables FIRMSIZE and ROA are in line with prior research and statistically significant. The coefficients for CEOFEM, GROWTH and LOSS_D correspond with prior research, but are not significant. Table 5 provides no support for AC and BIG4. The coefficient of AC and BIG4 have the wrong sign and are statistically insignificant. Since only 12,2% of firms did not have an audit committee and only 8,3% of firms did not have a BIG4 auditor, these result may be due to a lack of statistical power. Furthermore, I predicted that a higher leverage is associated with higher discretionary accruals. However, the coefficient is negative and not significant. The variables BOARDSIZE, CEOAGE, and FEMDER have positive coefficients, however negative ones were predicted. A smaller board may be less obstructed with bureaucratic problems and more functional. Therefore, smaller boards may provide better financial reporting oversight (Xie, Davidson III, & DaDalt, 2003).
3 For the interaction of CEO power on the relationship between PRISK and INCOMESMOOTH a positive
27
Table 5
Earnings management
Model 1 Model 2 Model 3
Intercept 0,161 0.106 0,122 [0,215] [0,218] [0,227] PRISK (H1) - 0,001 -0,010 [0,005] [0,011] CEOPOWER - - -0,030 [0,025] Interaction (H3) - - 0,006 [0,006] Control variables FIRMSIZE -0,015 *** -0,013 ** -0,013 ** [0,005] [0,006] [0,006] GROWTH 0,048 0,024 0,025 [0,031] [0,033] [0,033] ROA 0,156 ** 0,196 ** 0,196 ** [0,080] [0,084] [0,084] FINLEV -0,017 -0,020 -0,021 [0,043] [0,044] [0,044] LOSS_D 0,002 0,003 0,003 [0,019] [0,020] [0,020] AC 0,027 0,037 0,038 [0,066] [0,069] [0,069] BOARDSIZE 0,007 ** 0,005 0,005 [0,004] [0,004] [0,004] BIG4 0,034 0,026 0,026 [0,023] [0,024] [0,024] FEMDER 0,003 0,001 0,001 [0,004] [0,005] [0,005] CEOAGE 0,036 0,051 0,061 [0,046] [0,048] [0,051] CEOFEM -0,022 -0,049 -0,051 [0,036] [0,044] [0,045]
INDUSTRY F.E. YES YES YES
YEAR F.E. YES YES YES
R2 0,320 0,323 0,323
Adjusted R2 0,314 0,315 0,315
N 8309 8309 8309
***, ** and * coefficient is significant at respectively 1, 5, and 10 percent. In brackets are the standard errors adjusted for firm clustering. Industry and year fixed effects are suppressed for brevity. Definitions of variables are similar to the definitions used in table 4.
28 Table 6 presents the main results of the second linear regression model with income smoothing as the dependent variable. The OLS regression with standard errors clustered by firm is used to test the empirical predictions. I control for other factors which are likely to affect discretionary accruals used in the measure for income smoothing.
Table 6 Income Smoothing Model 1 Model 2 Intercept -0,898 *** -0,918 *** [0,099] [0,101] PRISK (H2) - 0,001 [0,002] Control variables FIRMSIZE -0,004 -0,004 [0,003] [0,003] GROWTH -0,005 -0,008 [0,009] [0,009] ROA -0,014 -0,006 [0,036] [0,035] FINLEV 0,012 0,018 [0,022] [0,023] LOSS_D 0,045 *** 0,047 *** [0,009] [0,010] AC 0,003 0,004 [0,013] [0,013] BOARDSIZE 0,000 0,001 [0,002] [0,002] BIG4 -0,024 * -0,027 * [0,014] [0,015] FEMDER -0,000 0,001 [0,002] [0,002] CEOAGE -0,020 -0,018 [0,025] [0,025] CEOFEM -0,015 -0,008 [0,018] [0,020]
INDUSTRY F.E. YES YES
YEAR F.E. YES YES
R2 0,315 0,324
Adjusted R2 0,308 0,316
N 8309 8309
***, ** and * coefficient is significant at respectively 1, 5, and 10 percent. In brackets are the standard errors adjusted for firm clustering. Industry and year fixed effects are suppressed for brevity. Variable definitions are similar to definitions used in table 4.
29 Table 6 model 2 includes the independent variable firm-level political uncertainty and adds 0,009 explanatory power compared to model 1 (with only the control variables). This brings the R-squared value to 0,324. The results show that an insignificant positive coefficient is found between PRISK and INCOMESMOOTH (β = 0,001; not significant). Since the relationship is insignificant, hypothesis 2 is rejected.
The control variables LOSS_D and BIG4 are significant and in line with the expectations. The variables FIRMSIZE, FINLEV, CEOAGE, and CEOFEM are consistent with prior research, however not significant. GROWTH, AC, BOARDSIZE, and FEMDIR have the wrong sign, as said before this result could be due to a lack of statistical power.
Additional analysis
I considered some additional tests to check the robustness of the previous findings. In this paper the modified Jones model to calculate the discretionary accruals is used. Since the results of this study depend on this measure, it is important to ensure that this metric actually measures discretionary accruals and not other firm characteristics included in the model. A different regression and new measures for discretionary accruals are used to test if this influences the results. Additionally, extra analysis on the main variables are performed to try to explain the insignificant results.
Jones model to measure earnings management
Another common approach to calculate discretionary accruals is using the Jones (1991) model. This model is the basis for the modified Jones model used in this study, only in the equation the change in receivables is not subtracted from the change in revenues. The formula for measuring discretionary accruals becomes:
EARNMAN ij,t = ACCRij,t/ TAij,t-1-{αj,t[1/ TAij,t-1] + βj,t[ΔREVij,t / TAij,t-1] + γj,t[PPEij,t/ TAij,t-1]} (3)
Table 7 model 1 presents the main results of the first robustness check. This OLS regression with absolute discretionary accruals based on the Jones (1991) model as the dependent variable is clustered with standard errors by firm. The multivariate results are almost identical to those presented in table 4 model 2. Kothari (2001) also find statistically insignificant differences between the discretionary accruals from the Jones and modified Jones model.
30
Kothari model to measure earnings management
Kothari also modified the Jones (1991) model adding ROA as an extra control variable. Previous research found that the modified Jones model is miss specified for well-performing or poorly performing firms (Kothari, Leone, & Wasley, 2005).
Earnings management will be measured for all firms k in industry j for year t based on the expected accrual-based Kothari model.
Accrualsjk,t = aj,t[1/TAjk,t-1] + bj,tΔREVjk,t + cj,tPPEjk,t + dj,tROAjk,t+ µjk,t (4)
The total accruals (Accrualsjk,t), change in revenues (ΔREVjk,t) and gross property, plant and equipment (PPEjk,t) are each deflated by lagged total assets (TAjk,t-1).
The discretionary accruals (DAP) are calculated for each firm-year ij,t of the sample by the following equation:
DAPij,t = Accrualsij,t -{ aj,t[1/TAij,t-1] + bj,tΔREVij,t + cj,tPPEij,t + dj,tROAij,t} (5)
in this equation aj,t, bj,t, cj,t and dj,t are the estimated coefficients (per 2-digit SIC industry) from equation 4.
Cross-sectional regression is used on all firms in the same industry (two-digit SIC) for each year. Table 7 model 2 presents the main results. This is an OLS regression with absolute discretionary accruals based on the Kothari model as the dependent variable. The regression is clustered with standard errors by firm. As with the other proxies for earnings management, no significant relationship is found between firm-level political uncertainty and earnings management.
Different regression for earnings management
The results of the third robustness check for earnings management can be found in table 7 model 3. The absolute discretionary accruals of the modified Jones model are used as the dependent variable. Only this time a panel regression with year and industry fixed effects is used. Still, the relationship between firm-level political uncertainty and earnings management is insignificant. However, the constant coefficient changes from a positive value to a negative value. Besides, other control variables have significant results compared to the main results.
31
Different regression for income smoothing
Table 7 model 4 presents the main results of the fourth robustness check. This time a panel regression with year and industry random effects is used to test hypothesis 3 about income smoothing. This has little impact on the results. Still, no significant relationship is found between firm-level political uncertainty and income smoothing. Only, the control variables FIRMSIZE, FINLEV are now significant at the 1% level.
Kothari model for measure income smoothing
The proxy for income smoothing used the modified Jones model to calculate discretionary accruals. For the fifth robustness check the Kothari model is used to calculate these. The Kothari model takes into account the effect of firms with better and poorer performance. Therefore, I include this measure as an additional analysis. Equation 5 is used to calculate the discretionary accruals. Table 7 model 5 contains the results of the last robustness check. The dependent variable is income smoothing, which is now based on the correlation between discretionary accruals and pre-discretionary income calculated with the Kothari model. The OLS regression is clustered with standard errors by firm. Model 5 shows that there is still no significant relationship between firm-level political uncertainty and income smoothing.
Table 7 Robustness checks
Model 1 Model 2 Model 3 Model 4 Model 5
Intercept 0,079 0.707 -0,118 -0,894 *** -0,876 *** [0,589] [0,419] [0,378] [0,093] [0,100] PRISK -0,000 0,002 -0,000 0,002 0,002 [0,015] [0,012] [0,006] [0,002] [0,002] CONTROL VARIABLES
YES YES YES YES YES
R2 0,250 0,284 0,026 0,319 0,332
Adjusted R2 0,242 0,276 - - 0,324
N 8309 8309 8309 8309 8309
***, ** and * coefficient is significant at respectively 1, 5, and 10 percent. In brackets are the standard errors adjusted for firm clustering. Control variables are suppressed for brevity. Variable definitions are similar to definitions used in table 4.
