MSc Thesis
Tax avoidance by CEOs with after-tax
compensation metrics
J.T. van Krevel
Faculty
: Faculty of Economics and Business
Program
: MSc Accountancy & Control, specialization Accountancy
Supervisor
: prof. dr. L.R.T van der Goot
Second Reader
: dr. W.H.P. Janssen
Date
: 31
stof July, 2015
Version
: 4
thDraft
Word count
: 18.749
Statement of originality
This document is written by student Jivan van Krevel 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.
Abstract
This study examines the effects of after-tax metrics in CEO bonus plans on tax avoidance behaviour. For a starting sample of S&P500 firm, financial statement data is collected from COMPUSTAT for fiscal year 2013. Information about the pre- and after-tax metrics in CEO compensation plans is collected from annual DEF14A filings. Effective tax rate models and the DTAX model are used to estimate tax avoidance. This study finds no evidence that firms with after-tax metrics in their CEO bonus plans are associated with lower GAAP ETRs. For the Cash ETR, Current ETR and DTAX, the study also fails to find a significant relationship. Overall, the findings indicate that firms with after-tax metrics in CEO bonus contracts are not associated with tax avoidance.
Contents
1. Introduction ...6
2. Literature review...9
2.1 Explanation of SFAS 109 ‘Accounting for Income Taxes’……...…9
2.2 Tax avoidance ………...….10
2.3 Types of executive compensation………...12
2.4 Economic theory...15
3. Methodology and hypothesis development ...17
3.1 Effective tax rate models...17
3.2 Book-tax differences………...…….…20
4. Data and descriptives... 23
5. Results... 27
5.1 Simple correlations...27
5.2 Regression analysis ...29
6. Additional analysis...34
6.1 Current ETR...34
6.2 Alternative DTAX model...35
6.3 Robustness test with more observations...37
6.4 Alternative approach for missing values...40
7. Conclusion...42
References...45
Appendix I - Variable Computation and Definition...48
Appendix II - COMPUSTAT missing values...49
List of Tables and Figures
Table 1 – U.S. Corporate tax rates...9
Figure 1 – CEO pay mix ...13
Table 2 – Sample selection ...24
Table 3 – Final sample by industry ...24
Table 4 –Descriptive statistics ...25
Table 5 – Descriptive statistics of the PERMDIFF model...26
Table 6 – Correlation matrix ...28
Table 7 – OLS regression of ETR models...31
Table 8 – First-order OLS regression of the PERMDIFF model...32
Table 9 – OLS regression of the DTAX model ...33
Table 10 – OLS regression of Current ETR...35
Table 11 – Alternative first-order OLS regression of the PERMDIFF model ...36
Table 12 – OLS regression of the alternative DTAX model ...36
Table 13 – OLS regression results of alternative analysis...38
1. Introduction
‘’Finding ways of reducing tax payments has become a fair game to erode the state’s capacity to provide social stability conducive to smoother accumulation of economic surpluses’ (Sikka 2010, p.3).
From 1998 to 2013, the effective tax rates by US-owned firms have been reduced by a third,
decreasing from 30 to 20 percent. Had these firms paid their taxes using the normal rate of 30 percent, ceteris paribus, it would have resulted in $200 billion in additional taxes for just the year of 2013 (Zucman 2014). If, in some hypothetical context, this money would have been allocated to the US military budget, it would’ve resulted in an increase of 30 percent. This example illustrates the velocity of the damages of tax avoidance. The increased use of tax-based stimulus packages in the U.S.
motivates firms to reduce tax payments. In addition, the steadily-developing globalization of the world economy provides increasing opportunities for multinational firms to apply foreign tax strategies as most tax rules are still determined on a national level (Hanlon, Heitzman 2010, Newman 1989). As Cressey’s Fraud Triangle suggested, besides having opportunities, one of the important factors that explains behaviour are the incentives. In search of incentives for tax avoidance, researchers often investigate CEO bonus plans. These bonus plans usually contain several different elements with which the compensation committee tries to motivate CEOs to act in the best interest of the firm. A vast amount of evidence is provided about CEOs managing earnings with the use of accruals for bonus plan purposes. A well-known example is the study by (Healy 1985) which provides evidence that setting caps and bogeys in bonus plans appear to be an effective means of influencing executives in their accounting policy decision-making. Aside from setting caps and bogeys, the compensation committee also has to decide on what type of metrics is used as the determinants of the total bonus. In particular, if financial metrics are used, the committee has to choose between pre- and after-tax metrics. Economic theory states that executives make choices that are consistent with maximizing the value of their compensation plans (Wallace 1997). This suggests that CEOs who are compensated with after-tax metrics, will have an incentive to implement tax strategies that lower the tax liability.
Prior literature encourages accounting researchers to focus on relevant tax and incentive policy debates as taxes potentially affect many real corporate decisions (Hanlon, Heitzman 2010).
Specifically, Shackelford and Shevlin (2001) call for more research on why some firms compensate on pre-tax metrics and other on after-tax metric. Thus far, there exist a limited amount of evidence that links elements of CEO compensation plans to tax avoidance (Armstrong, Blouin & Larcker 2012). Desai and Dharmapala (2006) have found a negative relation between the use of equity-based
compensation and aggressive tax planning. They attribute their finding to the theory that managerial diversion (i.e. the deviation from shareholder interests) will decrease with compensation plans that include shares because the interests of executives are more aligned with those of the shareholders. Yet, they do argue that there is also a reverse effect because the equity options also increase tax sheltering behaviour. Rego and Wilson (2012) find that stock option convexity (i.e. option price volatility) leads to more tax aggressiveness by CFO’s and CEO’s. This contrast with the Desai and Dharmapala (2006) study could perhaps be explained by the fact that executives are only faced with the upside potential of options while not bearing the downside risk as with shares. Robinson, Sikes and Weaver (2010) and Armstrong, Blouin and Larcker (2012) look at a relation between tax avoidance and compensation of a specific executive which they believe has an significant influence on setting tax policy; the tax
director. They find that tax executives’ compensation is negatively related to GAAP effective tax rates. Similarly, Gaertner (2014) finds evidence of CEOs with after-tax incentives being negatively associated with effective tax rates suggesting that annual bonus plans are an efficient way for firms to minimize tax exposure. However, other studies do not find a link between compensation plans and tax avoidance. Frank, Lynch and Rego (2009) also do not find evidence that executives are provided with incentives for tax avoidance. Phillips (2003) was the first study to investigate if compensating CEOs and business managers on a pre- or after-tax basis affects the firm’s effective tax rate. With the use of a survey, he finds that for business managers after-tax compensation leads to lower effective tax rates. In contrast with the study by Gaertner (2014), he did not have similar findings with respect to CEOs. Overall, the evidence that links compensation elements to tax avoidance behaviour of executives is limited and debated.
The objective of this study is to provide additional evidence of the effects of after-tax metrics in CEO bonus plans on tax avoidance behaviour. In doing so, this study aims at contributing to the debate by creating a better understanding. The research question is stated as follows; are after-tax
metrics in CEO bonus plans related to tax avoidance behaviour for US S&P500 firms in 2013?. The
question is answered with the use of four empirical models; three types of effective tax rates and a model that estimates the discretionary permanent book-tax differences (DTAX). For a starting sample of 500 firms, financial statement data is acquired from COMPUSTAT database and information about CEO compensation is hand-collected from annual DEF14A proxy statements. This study finds no evidence that firms that use after-tax metrics in their CEO bonus plans have lower GAAP ETR. For the Cash ETR and Current ETR, this study also fails to find a significant relationship. Overall, these findings imply that having after-tax metrics in bonus plans does not lead to tax avoidance behavior. Furthermore, this study finds no evidence that firms that use after-tax metrics in their CEO bonus plans are associated with higher discretionary permanent book-tax differences. This implies that these firms are not linked to tax avoidance behaviour belonging to the more aggressive part of the spectrum (Frank, Lynch & Rego 2009). Overall, the findings of this study suggest that firms that use after-tax
metrics are not associated with tax avoidance. The findings are consistent of the results of the study by Phillips (2003). However, the results are in contrast with the findings of Gaertner (2014) which found an negative association between CEO after-tax compensation and GAAP ETR and Cash ETR.
This study has potential implications for compensation committees as well as shareholders. Evidence of the negative relationship between the use of after-tax metrics and certain tax planning strategies can be useful when designing and voting on compensation plans. For instance, despite the gains in after-tax firm value, shareholders might not want executives to engage in tax avoidance because the opportunities that arise can cause managerial diversion of rents. Also, it can lead to potential risks due to additional (unforeseen) taxes, interest and penalties representing a decrease in cash flows and investor wealth (Hanlon, Heitzman 2010). Investors can also benefit from the evidence that links these firms with increased tax avoidance. Hanlon and Slemrod (2009) find that the market responded with a negative stock price reaction of 1,04% when news about a firm’s involvement in a tax shelter was announced. Opposed to this, firms that appear less likely to avoid tax have less negative event returns. Finally, policy makers at governmental organisations whose aim is to uphold (or perhaps increase) tax revenues might also want to be aware of the effects of using after-tax metrics on tax avoidance behaviour.
The next chapter reviews relevant prior literature. Chapter 3 discusses the methodology and the hypothesis development. In Chapter 4, the data and descriptives are presented. Chapter 5 discusses the main results. Chapter 6 presents the additional analysis. Chapter 7 concludes.
2. Literature review
2.1 Explanation of SFAS 109 ‘Accounting for Income Taxes’
In the United States, companies are required to compute income for taxes under a different set of rules (SFAS 109) than the reported income in their annual statements (which is computed in accordance with U.S. GAAP). The U.S. GAAP requires firms to record an income tax expense; an accrual-based accounting estimate that allocates the right amount of income tax to the period to which the income statement belongs. The amount of the tax expense is calculated by multiplying the taxable income with the applicable tax rate provided by the IRS. The tax table for 2013 are shown in Table 1. With
exception of the last category of incomes, the U.S. have a progressive tax system. This represents a system where firms with higher income pay relatively more taxes compared to firms with lower incomes. It is important to note that income tax expense is not the actual cash expenditure or obligation to the tax authorities at that moment. Due to differences in tax treatment of items that are included in (or excluded from) the financial statements, the tax obligation belonging to the current period (i.e. current tax expense) differs from the calculated total tax expense. The financial statements contain a note where details about the current and accrued part of the income tax expense are
disclosed. Moreover, the ‘current tax’ amount contains some accrued parts as well so technically it is not the same as the actual cash taxes paid to the tax authorities (Hanlon, Heitzman 2010).
Table 1.
U.S. Corporate tax rates.
Over- But not
over-
Tax is: Of the amount
over- 0 50,000 15% 0 50,000 75,000 7,500 + 25% 50,000 75,000 100,000 13,750 + 34% 75,000 100,000 335,000 22,250 + 39% 100,000 335,000 10,000,000 113,900 + 34% 335,000 10,000,000 15,000,000 3,400,000 + 35% 10,000,000 15,000,000 18,333,333 5,150,000 + 38% 15,000,000 18,333,333 - - - 35% 0
Note: U.S. tax rate schedule for corporate income in 2013. Values shown are in U.S. dollars. Table is included in the Instructions of Form 1120 with can be found on www.irs.gov
There exist several differences between the computation of taxable income and book income. Most differences have a temporary nature. These reflect a difference in the timing of recognition; the deduction in tax accounting in the current period results in an equal increase in future periods
(Stolowy, Lebas 2006). Common examples that cause differences are depreciation methods, deferred revenue, and warranty and bad debt expense. The differences between the two income measures is accounted for by recording a deferred tax liability or asset (Hanlon, Heitzman 2010). While deferred tax appears as an expense in the income statement, it is not accompanied by a cash outflow. There are also permanent differences between the two income measures that do not reverse. Examples include municipal bond interest or meals and entertainment disallowances. Additional differences between taxable and book income also derive from items that are related to taxes (and are thereby excluded from tax computations). An example is the valuation allowance; this account reduces a deferred tax asset to the extent to which the firm estimates it might not be recoverable in the future (Stolowy, Lebas 2006). Furthermore, differences might also arise due to different consolidation rules for book and tax purposes. Book-tax differences create a gap between the corporate income for accounting purposes and tax purposes. Consequently, the effective tax rate is different from the nominal tax rate. For instance, if income for tax purposes is lower compared to book income, the corporate tax rate set by the IRS is applied to a lower income figure resulting in a lower tax obligation. At the same time, financial statements still disclose higher book income. Hence, by looking at the calculated tax obligation and the higher book income, the effective tax rate is lower than the corporate tax rate (see Appendix I for the computation and definition of different types of variables).
Another complexity for U.S. multinational firms is the rules for foreign subsidiary earnings. In principal, U.S firms have to pay taxes on foreign earnings in foreign countries as well as in the U.S. (but usually they will receive a tax credit for foreign paid taxes). In addition, there is the so-called ’deferral’ rule that states that the taxation of the foreign earnings is deferred until cash is actually repatriated to the U.S. parent (Hanlon, Heitzman 2010). When foreign earnings are included in the consolidated financial statements, a deferred tax liability has to be recorded. However, APB 23 provides an exception to this rule which allows firms to not recognize this liability if they decide that the foreign earnings will not be repatriated and, thereby, be permanently reinvested in the foreign country. This results in a lower tax expense which, in turn, results in an effective tax rate which is lower than the domestic U.S. corporate tax rate.
2.2 Tax avoidance
In prior literature there has been some inconsistency about what falls under tax avoidance. Also, in some cases, seemingly interchangeable terms are used such as ‘tax evasion’, ‘aggressiveness’, ‘noncompliance’ and ‘sheltering’. Dyreng, Hanlon and Maydew (2008) were the first to adopt a relatively broad definition of ‘tax avoidance’ which formed the basis for later definitions. Their definition was ‘’the ability to pay a low amount of tax per dollar of reported pre-tax financial
accounting income’’ (p.3). Chen et al. (2010) later adapted this to the behaviour that ‘encompasses tax planning activities that are legal, or that may fall into the gray area, as well as activities that are illegal’
(p. 42). It is important to note that this definition includes the illegal part along the tax behaviour continuum. This differs from Weisbach’s (2005; in Salihu, Obid & Annuar 2013) interpretation because it limits tax avoidance to the legal activities while labelling tax evasion as illegal. Elaborating on Chen et al.'s (2010) definition, Hanlon and Heitzman (2010) define tax avoidance as ‘the reduction of explicit taxes’ (p. 137). This definition is also supported by Salihu, Obid and Annuar (2013). Hence, this thesis will define tax avoidance as ‘the behavior aimed at reducing the explicit tax liabilities’.
As mentioned earlier, there is a difference between non-conforming and conforming tax avoidance. Non-conforming tax avoidance deals with tax avoidance transactions accounted for differently for book and tax purposes while conforming tax avoidance deals with both tax and book incomes being reduced when the tax strategy is applied (Hanlon, Heitzman 2010, Salihu, Obid & Annuar 2013). So, conforming tax avoidance involves the reduction of taxable income even at the expense of reducing book income while non-conforming creates a book-tax difference (Armstrong, Blouin & Larcker 2012). Together, both these subtypes make up total tax avoidance. With an
increased amount of aggressiveness, there are several examples do be mentioned along the continuum of non-conforming tax avoidance. An example is a tax strategy that involves the use of municipal bond investments. The interest received on these types of investments (e.g. government bonds) is generally tax-favoured, meaning that the extra income will be permanently omitted from taxable income resulting in a lower tax expense (Hanlon, Heitzman 2010). This often serves as a paragon for a perfectly legal tax strategy. Furthermore, Newman (1989) provides some examples of tax strategies available to the multinational firm manager. He mentions the use of transfer pricing policies. If, for instance, the tax rate in a foreign country in which the firm owns a subsidiary is lower than the domestic tax rate, the manager can set transfer prices relatively high so that most of the profits will be taxed with the lower tax rate. A second possibility is that a manager can implement foreign tax rates in the tax strategy and allocate resources among different subsidiaries so that the marginal after-tax returns are equalized and after-tax profits are maximized (Newman 1989). This way, the benefits of a possible progressive tax system can be exploited. Third, Newman (1989) also mentions that if dividends are regularly paid, managers may make use of holdings in a third country (with favorable tax legislation) to lower foreign taxes. These strategies can involve business decisions that on itself are not value-creating activities while reducing the tax liabilities. For instance, setting up a holding company is initially a value-destroying activity were it not for benefits in the form of a reduced tax liability. In addition, when opportunity costs are taken into account, such decisions may be deemed as even more value-destroying. Even though some of the above mentioned strategies are perfectly legal examples of non-conforming tax avoidance, it might be considered a grey area because they are not in line with the spirit of tax legislation. Another example which can be considered very aggressive non-conforming tax avoidance is the use of abusive tax shelters. The IRS estimates that for 1999 the
revenue loss from abusive tax shelters was between $14.5 to $18.4 billion (Crocker, Slemrod 2005). While tax sheltering isn’t illegal, abusive tax shelters are described as ‘very complicated transactions promoted to corporations and wealthy individuals to exploit tax loopholes and provide large,
unintended tax benefits.’ (US General accounting office 2003, p. 13 in; Crocker, Slemrod 2005). The IRS actively searches for the use of abusive tax sheltering.
When looking at total tax avoidance there also exist other types of tax strategies that be addressed as conforming tax avoidance. These strategies reduce the tax liability by simultaneously reducing accounting income. For instance, firms might choose to lower accounting earnings to benefit from the lower tax liability because the benefits of the reduced tax liability outweigh the costs
involved with less relevant book income figures (McGuire, Wang & Wilson 2014). Private or family firms might not rely on the public market to provide them with capital. These type of firms usually acquire capital from banks with which they share private information. As a consequence, they do not have to inform the market with statements, making accounting earnings less relevant. Hence, if earnings are lower, the effect on their going concern is less severe.
2.3 Types of executive compensation
The compensation executives receive can take on various forms including base salary, short and long term cash bonuses, stock options, restricted stock, performance shares and several other fringe benefits and perquisites (Murphy, 2012). Since the 1980s, most forms of compensation are linked to the performance of the firm to some extent. The massive switch to performance-based compensation was due to an imposed tax law; Internal Revenue Code 162 (m). This law states that publicly held firms cannot deduct non-performance-based employee remuneration as a business expense when such remuneration exceeds the amount of $1,000,000. Congress passed this new legislation to target a small group of individuals who had abusively high salaries. Ironically, the law actually caused overall CEO pay to increase as firms were encouraged to grant more stock options and, in many cases, increase CEO pay to exactly $1 million (Murphy, 2012). Figure 1 shows the realized average and median CEO pay mix of 463 S&P500 firms in 2011. Total average (median) compensation amounted to $12.3 million ($7.8 million). There are six basic components: (1) base salaries; (2) discretionary bonuses; (3) non-equity incentives; (4) stock options; (5) equity grants; and (6) other pay.
Base salaries are the wages that are paid out on a non-formulaic basis and are therefore not performance-based. Together with bonuses it was the only type of compensation used by public firms to pay their CEOs until the 1940s. In 2011, base salary only represents a mere 14% of total pay as other types of compensation where introduced after WOII. Like in most countries, wages are subject to income tax in the U.S. Depending on your filing status (e.g. single or married) a progressive tax
system applies with minimum and maximum rates of respectively 25% and 39.6% in 2013 (IRS, 2013a). As mentioned above, most firms choose to limit base salaries to $1 million.
Fig. 1
CEO pay mix.
Note: Figure 1 shows the ‘realized’ CEO pay mix of 463 S&P500 firms in 2011 as reported by Murphy (2012). Percentages relate to CEO total average (median) compensation of 12.3 (7.8) million U.S. dollars. Note that median base salaries are more or less equal to the imposed limit on deductible business expenses in the Internal Revenue Code 162 (m). Also note that, as of 2011, most firms do not make use of stock options in compensation contracts
Another common form of compensation are bonuses. Similar to base salaries, discretionary bonuses distinguish themselves from other types of bonuses because they are not based on a formula but rather on a board’s subjective assessment of value creation (Murphy, 2012). Hence, firms ought to be aware of the effects of granting such a bonus on the deductibility of their expenses. Short-term and long-term bonus are other types of bonuses and are performance-based because these bonus plans contain predetermined performance goals for the CEO to achieve. As suggested, long-term bonus plans usually contain performance goals that relate to several years after the fiscal year. For instance, the proxy statement of Nike for fiscal year 2013 states that CEO Mark Parker is entitled to a bonus of $3.5 million when certain targets for revenue growth and earnings-per-share (EPS) are achieved in the next three years. Murphy (2012) characterizes bonus plans on de basis of three dimensions; performance measures (i.e. metrics), performance thresholds and structure of the pay-performance relation. While most firms use several financial and non-financial measures in bonus plans, it is common for firms to use an accounting income based metric (e.g. net income or EBITDA) (Murphy, 2012). Also, in most cases, bonuses are determined based on a performance in relation to certain benchmark (Murphy, 2000). For instance, instead of looking at absolute EPS, bonus is determined on EPS growth compared to last year or compared to peer-group performance. See Appendix III for common bonus plan metrics. Thresholds in bonus contracts, also called caps and bogeys by Healy (1985), are set during the firm’s
14% 3%
25% 19%
34% 5%
Average pay mix
13%
20% 31%
2%
Median pay mix
Base salary Discretionary bonus Non-equity incentives Stock options Equity grants Other pay
budgeting process. When chosen performance measures are beyond the bogey, the executive will receive a bonus. The bonus reaches its limit at the cap of the bonus plan. The pay-performance relation is described as the function of the total bonus with all performance outcomes as inputs. For instance, the pay-performance relation may have a linear trend if the bonus increases evenly when performance outcomes increase. Pay-performance relations can also be convex or concave. In the U.S, bonuses are taxed as supplemental wages in line with Publication 15 (IRS, 2013b). Two taxation methods are possible; (1) a method where the supplemental wage is taxed with a flat rate of 25%; and (2) a method where the supplemental wage is added to regular pay which likely results in a higher tax rate because of the progressive tax system discussed above. Hence, a flat tax rate of 25% is relatively low
considering the first bracket of regular income tax has the same rate. A special rule applies when the bonus exceeds the amount of $1 million during the calendar year. All supplemental wages above this threshold will be taxed with at 39.6% (or the highest applicable rate of income tax for the year).
In the beginning of the 1950s, a new type of compensation entered CEO bonus plans; stock options. With this type of compensation, a CEO acquires the right (and not the obligation) to purchase shares at an agreed-upon price (i.e. the strike price). This way, if the firm’s stock performs well, it will rise above the strike price resulting in an increase in the value of the option contract. A notable
difference between options and regular shares is that the holder of an option is not subjected to the downside risk while benefitting from the upside potential. As mentioned above, stock options gained popularity due to the imposed Internal Revenue Code 162 (m) in the 1980s. Stock options are considered to be performance-based because the value of the option contract depends on the firm’s stock market performance. This popularity reached its peak in 2001 when base salaries made up only 18% of total pay while options accounted for more than 50% of pay (Murphy, 2012). However, in the period that followed, stock options lost their popularity quickly due to backdating scandals. In these wrongdoings, individuals falsified option contract grant-date’s by cherry picking dates on which the stock price was relatively low. In other words, they manipulated the contracts so they could set the strike price equal to the lowest stock price they could find. As presented in Figure 1, as of 2011, most firms do not make use of stock options in their CEO bonus plans. Many firms switched to equity incentives. One form is restricted stock; shares with a restriction to sell for a determined amount of time. As of 2011, it is the biggest part of CEO compensation (Murphy, 2012). With restricted shares, not only are holders subjected to the upside potential and the downside risk, holders are also forced to hold the shares. Especially in unstable and uncertain times (e.g. mergers or increased stock price volatility), it can be beneficial that (new) large shareholder are not allowed to sell their shares (Murphy, 2012). Another common form of equity compensation are performance shares. These are defined as shares that are granted upon achieving a certain performance goal. Shares as well as options are presented at their fair value on grant date in annual proxy statements. For options it is common to use the Black-Scholes model to value the contract (or a similar methodology). While shares and
options are given a certain value when included in total CEO compensation information at grant-date, the IRS decided that they are not taxed unless the holder sells the shares. In case the holder decides to sell, any income with be classified as a short- or long-term capital gain. Depending on the balance between short- and long-term gains, tax rates can vary between 0% and 20% for 2013 (IRS, 2015). When compared with regular income tax rates, these rates are relatively low. Given the favorable tax rates of both equity incentives and bonuses (i.e. supplemental wages) it is perhaps not surprising that these types of compensation account for the largest part of median and average compensation.
Other forms of compensation (denoted as ‘other pay’ in Figure 1) includes perks, termination and signing bonuses and pension benefits. Some types of this extraordinary pay are classified as supplemental wage with rates of 25% to 39.6%. However, there exist an extensive amount of
legislation on the various types of fringe benefits which is beyond the scope of this research. Refer to IRS Publication 15b for more detailed information.
2.4 Economic theory
There exists a well-established idea that executives make choices that maximize the value of their contracts by increasing the value of their compensation (Gaertner 2014). The bonus plan hypothesis, for instance, states that managers are more likely to shift income from future periods into the current period to maximize their remuneration (Watts, Zimmerman 1978). Also, Wallace (1997) suggests that managers make choices that are consistent with maximizing the value of their compensation.
Furthermore, contract maximizing can also be explained from an agency theory perspective (Jensen, Meckling 1979). Due to the separation of ownership (i.e. shareholders) and control (i.e. executives), shareholders want to make sure that management acts in the best interest of the firm as a whole. They do this with the use of compensation contracts that align the interests of the executives with those of the firm (Fama, Jensen 1983). These contracts can also include elements that motivate executives to reduce the tax liability as long as the incremental profit exceeds the costs of doing so. By having effective incentives, agency costs such as moral hazard problems and adverse selection are reduced. In particular, moral hazard problems are reduced because part of the risk is transferred to the manager. Consequently, an executive’s compensation will be lower when he decides to shirk. For example, if shareholder do not want an executive to pursue aggressive tax strategies but the executive neglects this preference, an effective incentive program will limit an executive’s bonus. In addition, having
effective incentives reduces adverse selection costs because only executives who believe they can profit from the contract will sign it. For non-capable executives, the risks of such a contract are too high. Consequently, only capable executives who believe that they can maximize the contract will self-select (Demski, Feltham 1978). This suggests that in the case of after-tax incentives, only
managers who believe they are able to lower the tax liability will accept the contract. Hence, based on the above mentioned theories it is predicted that managers with after-tax incentives try to maximize
this contract by reducing explicit tax liabilities. With regard to annual bonus plans, Murphy (2012) states that annual bonus plans based on accounting measures are likely to be an important factor in influencing behavior of CEOs. First of all, executives direct their efforts towards elements of a bonus contract that they understand. He argues that performance metrics in bonus plans are often easier to understand than, for instance, stock prices. This can be interpreted such that bonus plans may provide stronger incentives. Second, the immediacy and tangibility of a cash awards can also provide
additional incentives over other forms of compensation.
Phillips (2003) is the first study to investigate this relation. He investigates if after-tax incentives have an effect in increasing tax avoidance, measured with GAAP ETR, by business unit managers and CEO’s. For S&P 500 firms in 1996, he finds that compensating business unit managers on an after-tax basis leads to a lower ETR while compensating CEOs in the same manner does not; with the latter running contrary to predictions. The author suggests that CEOs might have other incentives that motivate them to focus on after-tax results. Also, he does not rule out the possibility that CEO’s still have an indirect effect on ETR through they subordinate business managers, whose after-tax incentives are significant. The recent study of Gaertner (2014) also examines the relationship between after-tax incentives of CEOs and tax avoidance, again measured by GAAP ETR. His sample size includes 354 firm observations from the year 2010. By hand-collection, 215 firms with after-tax incentives were identified. The author finds a significantly negative relationship between GAAP ETR and the use of after-tax incentives for CEOs which he claims is the first evidence that is in accordance with economic theory. By 2011, most firms did not use stock options in their CEO bonus plans (see Fig. 1 for the median pay mix).
3. Methodology and hypothesis development
There exist several ways to measure tax avoidance. Salihu, Obid and Annuar (2013) and Hanlon and Heitzman (2010) provide a clear overview of measures. They argues that researchers must carefully select measures that are most appropriate to their research question because not all measures capture the same aspects of tax avoidance.
3.1 Effective tax rate models
ETRs are commonly used for describing tax avoidance (Armstrong, Blouin & Larcker 2012, Gaertner 2014, Dyreng, Hanlon & Maydew 2008, Chen et al. 2010, Salihu, Obid & Annuar 2013, Mills 1998). The GAAP ETR is calculated by multiplying the corporate tax rate with taxable income and dividing this term pre-tax (accounting) profits. This measure captures the average rate of tax per dollar of income (Hanlon, Heitzman 2010). It is important to note that tax strategies that involve the deferral of income are not captured by the GAAP ETR. This is because the tax expense estimation includes the accrual part of estimated tax liability. Yet, changes that are not tax strategies such as changes in tax contingency reserves or valuation allowances will affect GAAP ETR. This is because changes in these accounts are transferred to the ‘tax expense account’. Both Phillips (2003) and Gaertner (2014) make use of GAAP ETR to measure the effect of after-tax compensation in compensation plans. As
mentioned earlier, Phillips did not find significant evidence for S&P 500 firms in 1996. Gaertner (2014) mentions a possible reason for the insignificant results. He argues that Phillips (2003) uses a proprietary dataset (i.e. a survey) with a relatively small sample size of 209 observations in total. The survey shows that 39 percent of the managers in the sample (i.e. N=81) has no after-tax incentives. When a statistical power test is applied to the sample to examine the power of the performed t-test, the results is 0.64 (Cohen 2013). This result implies that the t-test has only a 64% chance of finding an existing relationship (Gaertner 2014). In other words, it is likely that Phillips (2003) did not find a significant result because his sample size was too small. Also, the study could not control for current economic trends because data was only collected from one year (1997). Gaertner (2014) also
investigates the effects of having after-tax metrics in bonus plans and he finds a negatively significant result on GAAP ETR. Unlike the Phillips study, his information is hand-collected from annual proxy statements and makes use of a larger S&P500 sample. However, his study also is limited to data from the year 2010 with makes it harder to make interferences about the effect of after-tax metrics in general. Furthermore, he includes firms in the sample that are subject to different tax legislation. Hence, the relationship of after-tax metrics on tax avoidance is still unclear.
This study also investigates the relationship between after-tax metrics and tax avoidance with use of GAAP ETR. As predicted by economic theory and in accordance with the results of Gaertner (2014), it is predicted that tax avoidance will increase when after-tax metrics are used. This leads to the following hypothesis;
H1: The use of after-tax metrics in CEO bonus plans is negatively related to a firm’s GAAP ETR
To test this hypothesis, the model as shown below is used. The hand-collected variable of interest
CEOATAX is discussed in Chapter 4. The computations and definitions for the other variables can be
found in Appendix I.
𝐺𝐴𝐴𝑃 𝐸𝑇𝑅𝑖,𝑡 = 𝛽0+ 𝛽1 𝐶𝐸𝑂𝐴𝑇𝐴𝑋𝑖,𝑡+ 𝛽2 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+ 𝛽3 𝐿𝑛𝑀𝑉𝑖,𝑡+ 𝛽4 𝑅𝑂𝐴𝑖,𝑡+ 𝛽5 𝑂𝐶𝐹𝑖,𝑡 + 𝛽6 𝐿𝐸𝑉𝑖,𝑡+ 𝛽7 𝐶𝐴𝑃𝑖,𝑡+ 𝛽8 𝑅𝐷𝑖,𝑡+ 𝛽9 𝐼𝑁𝑇𝐴𝑁𝑖,𝑡+ 𝛽10 𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝛽11 𝑁𝑂𝐿𝑖,𝑡
+ 𝜀𝑖,𝑡
In order to control for specific effects that might drive the change in effective tax rates, a number of control variables used in prior literature is included in the regression model. To control for size of a firm 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠 and 𝐿𝑛𝑀𝑉 are included. Larger firms have the ability to profit from economies of scale which could enhance their tax planning opportunities as well as their lobbying power (Gupta, Newberry 1997). At the same time, prior literature suggests that larger firms are associated with higher political costs because of higher tax payments (Omer, Molloy & Ziebart 1993, Zimmerman 1983). Hence, the risk of negative publicity reduces tax avoidance behaviour. 𝑅𝑂𝐴 is included to control for underlying economic activity due to a difference in tax planning opportunities for more profitable firms (Armstrong, Blouin & Larcker 2012). 𝑂𝐶𝐹 is included to control for the operating cash position (Armstrong, Blouin & Larcker 2012, Gaertner 2014). 𝐿𝐸𝑉 is included to control for leverage and the extent of a firm’s tax shield. In particular, when the tax shield is bigger, the incentive for tax planning is less (Armstrong, Blouin & Larcker 2012, Mackie‐Mason 1990). The 𝐶𝐴𝑃 variable is included to control for capital intensity. Firms with higher property, plant and equipment have greater tax-planning opportunities (Gupta, Newberry 1997). 𝑅𝐷 is included to control for the amount of research and development investments a firm made because of the tax-favoured legislation regarding
accompanied expenses (Gaertner 2014, Armstrong, Blouin & Larcker 2012, Berger 1993). 𝐼𝑁𝑇𝐴𝑁 controls for the extra opportunities of intangible intensive firms for income-shifting (Desai, Hines Jr 2002). FOREIGN is included to control for the international activities of firms by measuring their foreign income. Internationally active firms not only have greater opportunities for tax planning but also have lower costs to implement these strategies. Furthermore, the risk of tax rate changes for these firms is less because they are not bound to a single jurisdiction (Newman 1989). Finally, 𝑁𝑂𝐿 is included to control for net operating losses by looking at the change tax loss carry forwards.
An alternative effective tax rate measure is the Cash ETR. This measure is calculated by dividing the cash taxes paid by pre-tax profits. A disadvantage is that paid cash outflows might not belong to the same period as the pre-tax accounting earnings, creating a mismatch. However, this measure has the advantage that accrual elements are not included in the numerator and that it is affected by deferral tax strategies (Hanlon, Heitzman 2010). Hence, the Cash ETR is able to measure other types of tax avoidance. This leads to the following hypothesis;
H2: The use of after-tax metrics in CEO bonus plans is negatively related to a firm’s Cash ETR
To test the hypothesis, the following empirical models are used. The variables included in the model resemble those of the GAAP ETR model. The computations can be found in Appendix I.
𝐶𝑎𝑠ℎ 𝐸𝑇𝑅𝑖,𝑡= 𝛽0+ 𝛽1 𝐶𝐸𝑂𝐴𝑇𝐴𝑋𝑖,𝑡+ 𝛽2 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+ 𝛽3 𝐿𝑛𝑀𝑉𝑖,𝑡+ 𝛽4 𝑅𝑂𝐴𝑖,𝑡+ 𝛽5 𝑂𝐶𝐹𝑖,𝑡 + 𝛽6 𝐿𝐸𝑉𝑖,𝑡+ 𝛽7 𝐶𝐴𝑃𝑖,𝑡+ 𝛽8 𝑅𝐷𝑖,𝑡+ 𝛽9 𝐼𝑁𝑇𝐴𝑁𝑖,𝑡+ 𝛽10 𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝛽11 𝑁𝑂𝐿𝑖,𝑡 + 𝜀𝑖,𝑡
In general, there consist some drawbacks when measuring tax avoidance with ETRs. First of all,
GAAP ETR does not capture tax deferral strategies in order to lower current tax expenses (i.e. GAAP ETR includes the accrual element) (Hanlon, Heitzman 2010). Hence, interferences about total tax
avoidance would not be appropriate. Similarly, ETR’s in general may not be suited to measure tax avoidance in all contexts because their denominator only captures non-conforming tax avoidance. Second, ETR’s not only include the tax aspect but also earnings management. For instance, managers can lower the ETR by shifting profits to tax havens (i.e. countries with a low corporate tax rate) but they can also lower ETR by increasing pre-tax accounting income while maintaining the value of taxable income. Since an accounting measure is located in the denominator of the ETRs formulas, the measure is also affected. Hence, measuring tax avoidance with just ETRs might not be appropriate as it can include noise that originates from other sources than tax avoidance. Therefore, this study will make use of another type of measure that is able to compensate for the drawbacks of ETRs.
3.2 Book-tax differences
Commonly used alternatives to ETRs when measuring tax avoidance are models that include book-tax differences (Chen et al. 2010, Desai, Dharmapala 2009, Kim, Li & Zhang 2011). The most simplistic measure of a book-tax difference (BTD) is, in essence, very similar to the ETR measures. With BTD, two types of income, taxable income and accounting income, are subtracted while with ETR’s two figures originating from those types of income are dividend. As mentioned before, the differences can be classified as either permanent or temporary. Two common types of BTD are total BTD and temporary BTD (measured with deferred tax expense). Desai and Dharmapala (2006) also use an alternative measure by regressing total book-tax differences on total accruals which controls for earnings management. The residual is then used to make interferences about tax avoidance. Frank, Lynch and Rego (2009) introduce a model which is similar to the Jones Model (1991) of discretionary accruals. The Jones Model tries to estimate the discretionary part of total accruals. In order to do so, first the nondiscretionary accruals are estimated with the use of a regression model which contains variables that are highly correlated. The discretionary part can be found by looking at the difference with the total accruals. Frank, Lynch and Rego's (2009) DTAX model tries to apply the same mechanics of the Jones model onto the tax avoidance context. They try to estimate the discretionary portion (DTAX) of the PERMDIFF, the total permanent book-tax differences. In doing so, the authors attempt to remove underlying determinants that are not driven by intentional tax avoidance (Hanlon, Heitzman 2010). The PERMDIFF measure is similar to GAAP ETR when it comes to measuring (non-conforming) tax avoidance but, in addition, it measures only tax avoidance that helps increase after-tax book income by reducing GAAP ETR (Hanlon, Heitzman 2010). Frank, Lynch and Rego's (2009) argue that more aggressive tax avoidance creates permanent differences. Armstrong, Blouin and Larcker (2012) apply a slightly modified version of this model to investigate compensation of tax directors. They introduce another control variable into the model that controls for the amount of foreign operations a firm is actively involved in. This study also uses a modified DTAX model by including a variable into the model that controls for foreign operations. The model is as follows;
𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹𝑖,𝑡= ∝0+ ∝1𝐼𝑁𝑇𝐴𝑁𝐺𝑖,𝑡+∝2𝑈𝑁𝐶𝑂𝑁𝑖,𝑡+ ∝3𝑀𝐼𝑖,𝑡+ ∝4𝐶𝑆𝑇𝐸𝑖,𝑡+ ∝5∆𝑁𝑂𝐿2𝑖,𝑡 + ∝6𝐿𝐴𝐺𝑃𝐸𝑅𝑀𝑖,𝑡 + ∝7𝐿𝐴𝐺𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝜀𝑖,𝑡
The benefit of including a model that measures permanent differences is that the effect of deferral tax planning strategies on overall tax avoidance is excluded. Consequently, when this model is used in addition to ETR models, an enhanced understanding can be acquired about the type and nature of tax avoidance behavior. The dependent variable in this model is 𝑃𝐸𝑅𝑀𝐷𝐼𝐹𝐹 which is defined as the total book-tax difference minus temporary book-tax differences. The independent variables are all highly correlated as they serve to estimate the non-discretionary portion of the total permanent book-tax differences (Frank, Lynch and Rego, 2009). 𝐼𝑁𝑇𝐴𝑁𝐺 includes goodwill and other intangibles because
differences between accounting and tax regulation in this area is associated with permanent differences that are unrelated to aggressive tax avoidance. 𝑈𝑁𝐶𝑂𝑁 and 𝑀𝐼 are included to control for income reported under the equity method (equity in earnings) and to control for minority interests because tax regulation for partially-owned firms can give rise to permanent differences unrelated to aggressive tax avoidance. They control for 𝐶𝑆𝑇𝐸 because state tax expense causes nondiscretionary permanent differences. ∆𝑁𝑂𝐿2 captures the real change in tax loss carry forwards. Prior literature suggests that a difference tends to affect the valuation allowance account which, in turn, affects the PERMDIFFs. However, these changes are not related to aggressive tax avoidance (Frank, Rego 2006, Miller, Skinner 1998, Schrand, Wong 2003). Note that ∆𝑁𝑂𝐿2 differs from the control variable included in the ETR models. 𝐿𝐴𝐺𝑃𝐸𝑅𝑀 (i.e. lagged PERMDIFF) controls for permanent differences that exist throughout time but are not likely the result of aggressive tax planning strategies. Finally,
𝐿𝐴𝐺𝐹𝑂𝑅𝐸𝐼𝐺𝑁 includes lagged foreign income into the model to control for a firms international activities. Firms which are more internationally orientated have more tax planning opportunities. This control variable is similar to the foreign assets variable used by Armstrong, Blouin and Larcker (2012). By including this control as an estimator for the nondiscretionary accruals, Armstrong, Blouin and Larcker (2012) suggest that having international activities and benefitting from international tax legislation is not labelled as aggressive. The remaining 𝜀𝑖,𝑡 is the DTAX. As with the Jones Model, the coefficients from the first-order OLS regression are used to calculate the nondiscretionary portion of the permanent differences. When these are subtracted from the total PERMDIFF’s, the discretionary portion (DTAX) is what remains. Firms with a higher DTAX will have bigger permanent differences resulting from aggressive tax avoidance (Frank, Lynch & Rego 2009). When this model is applied to the context of this study, this leads to the following hypothesis;
H3: The use of after-tax metrics in CEO bonus plans is positively related to a firm’s DTAX
To test the hypothesis the model as depicted below is used. Similar to the Armstrong, Blouin and Larcker (2012) study, the control variable which controls for foreign operations is also included in the final regression model. The controls are discussed above. Computations can be found in Appendix I. 𝐷𝑇𝐴𝑋𝑖,𝑡 = 𝛽0+ 𝛽1 𝐶𝐸𝑂𝐴𝑇𝐴𝑋𝑖,𝑡+ 𝛽2 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+ 𝛽3 𝐿𝑛𝑀𝑉𝑖,𝑡+ 𝛽4 𝑅𝑂𝐴𝑖,𝑡+ 𝛽5 𝑂𝐶𝐹𝑖,𝑡+ 𝛽6 𝐿𝐸𝑉𝑖,𝑡
+ 𝛽7 𝐶𝐴𝑃𝑖,𝑡+ 𝛽8 𝑅𝐷𝑖,𝑡+ 𝛽9 𝐼𝑁𝑇𝐴𝑁𝑖,𝑡+ 𝛽10 𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝛽11 𝑁𝑂𝐿𝑖,𝑡+ 𝜀𝑖,𝑡
As mentioned earlier, the DTAX measure is similar to the GAAP ETR measure and, therefore, it does not capture deferral tax strategies. However, an advantage of the DTAX measure is that it only captures tax avoidance that results in a lower GAAP ETR by increasing book income (Hanlon, Heitzman 2010). Hence, it is a more narrow measure than the GAAP ETR as it tends to measure the more aggressive part of tax avoidance. Furthermore, book-tax differences, by definition, do not capture conforming tax avoidance since this behaviour is characterized by both reducing taxable income and accounting
income (i.e. not affecting the width of the book-tax gap). Book-tax measures are therefore not suited to measure tax avoidance of firms who place various levels of importance on accounting earnings. However, given the economic theory it is highly unlikely that executives of public firms who are being compensated on accounting earnings would lower earnings for the tax benefit. This is mainly because the incremental earnings due to the lower tax liability would not offset the costs of reducing book income when shareholders place much emphasis on earnings figures in financial statements. Consequently, there exists no incentive for executives to apply such a strategy because it would not maximize compensation a way that is suggested by previously mentioned economic theory.
4. Data and descriptives
The data for this study is collected from COMPUSTAT. In order to acquire the most relevant evidence, data from the most recent year available is used, which is fiscal year 2013. The starting sample contains the S&P 500 firms. This is due to a number of reasons. First of all, large and multinational companies have the most resources to apply effective tax strategies (Newman 1989, Dyreng, Hanlon & Maydew 2008). Second, smaller firms are not suited because they place various levels of importance on accounting earnings which complicates the analysis due to the possibility of conforming tax avoidance (Salihu, Obid & Annuar 2013). Third, US firms have the most data available in COMPUSTAT. Fourth, public firms disclose compensation information in proxy
statements which are needed for this study. The Electric, Gas and Sanitary Services firms (SIC-codes 49) and the Finance, Insurance and Real Estate firms (SIC-codes 60-69) are excluded from the sample because these firms are subjected to different tax regulation (Frank, Lynch & Rego 2009). For
instance, banks have specific bank tax regulations and energy firms are exempted of taxation. Data from COMPUSTAT also had several missing values. In accordance with prior literature, missing values are excluded from the analysis on a list wise basis (Philips, 2003; Gaertner 2014; McGuire, Wang & Wilson, 2014). However, for variables of the DTAX model, different choices are made in accordance with Frank, Lynch and Rego (2009). Appendix II shows the frequencies of the missing values per input. Also, more specific information about actions undertaken to deal with these missing values is included in the description.
The dummy variable CEOATAX is the main variable of interest and is hand-collected from publicly available proxy statements (DEF 14A). These filings contain a chapter ‘Compensation, Discussion and Analysis’ which presents information about the pay mix, the amount of cash
compensation, the metrics used to determine the bonus and, in most cases, the weight of these metrics. The content and the mechanics of the annual cash incentive program for the CEO is examined. If the cash compensation is determined based on after-tax metrics (e.g. earnings-per-share or return-on-equity) the dummy gets the value of 1. If pre-tax metrics are used (e.g. pre-tax income or operating income) the dummy variable gets the value of 0. In case the bonus plan contains several pre- and – after-tax financial metrics the total weight of both categories is taken into account; CEOATAX is valued 1 (0) if the after-tax (pre-tax) weight of the financial metrics is at least 60%. Appendix III shows the categorization used for the most common metrics. In a number of cases an observation is removed from the sample. The most common reason is that proxy statements were missing. Other reasons are; the CEO did not receive any cash compensation; the CEO retired during fiscal year 2013; cash compensation was not based on accounting metrics (e.g. strategic or safety metrics); or the proxy
statement provided insufficient information. Another category of observations was temporary labelled as 2. In these cases, the annual bonus program contained both pre-tax and after-tax metrics in an equal manner but the overall effect was not clear enough to set CEOATAX as either 0 or 1. In the main analysis of this study these observations are excluded from the sample. However, these observations are reintroduced in the additional analyses of Chapter 6 where they are transformed in either 0, 1 or the sample mean.The adjustment of the starting sample into the final sample is shown in Table 2.
Table 2
Sample selection.
S&P 500 firms 500
Less:
Missing financial statement for fiscal year 2013 Removed SIC-codes (49 and 60-67)
CEOATAX unable to determine CEOATAX doubtful observations Firms in final sample
(1) (125) (47) (26) 301
Note: The actual number of observations included in the following analyses may differ from the number shown of ‘firms in final sample’ in this table. This is mainly because observations that contain missing values are excluded from the analysis. Missing values are presented in greater detail in Appendix II.
Table 3
Final sample by industry.
Industry N Percentage
Agriculture, Forestry, Fishing Mining
Construction Manufacturing
Transportation & Public Utilities Wholesale Trade Retail Trade Services Total 1 28 6 166 23 8 29 40 301 0.33 % 9.30 % 1.99 % 55.15 % 7.64 % 2.66 % 9.63 % 13.29 % 100 %
Table 3 represents the composition of the sample by industry. As with other studies, the most significantly represented industry is manufacturing (Armstrong, Blouin & Larcker 2012, Gaertner 2014, Phillips 2003). However, an important difference with most prior literature is that the industries with different tax regulation were not excluded as Frank, Lynch and Rego (2009) recommended. Given that the total share of these deviant firms within the sample can be substantial, this results in a potentially distorted view of tax avoidance. For instance, if these firms were not excluded, the total number of observations in the final sample would increase with nearly 42% (125 observations; see Table 1). Consequently, when the final sample is compared with the sample from the Gaertner (2014)
study, this study has fewer observations. However, the difference is limited to only 53 observations because more CEO bonus plans were based on accounting metrics.
The descriptive statistics are shown in Table 4. Compared to recent studies, the number of firms with after-tax metrics in bonus plans in the sample is surprisingly low. While Gaertner (2014) and Phillips (2003) both recorded percentages around 60%, this study only finds 92 firms which results in a percentage of nearly 31%. However, older studies like Healy (1985), Newman (1989) and Carnes and Guffey (2000) report similar percentages respectively 47.3%, 33.9%, and 30.1%. Ceteris paribus, this implies a shift in the type of bonus metrics used in bonus plans from after-tax to more pre-tax oriented plans. However, differences in methodologies among studies might also be the cause of differences in reported percentages. For instance, a survey-based study like Phillips (2003) might identify after-tax bonus plans in a different way than studies like Carnes, Guffey (2000), Gaertner (2014) and Newman (1989) which all investigate proxy statements. Furthermore, the manner in which researcher deal with cases in that are ambiguous might also differ. And as mentioned earlier, the proposition of the sample can have an effect on the reported outcomes. For instance, by choosing not to exclude financial service and utility firms one can significantly increase the sample size.
Table 4
Descriptive statistics.
Variable N Mean Std. dev Q1 Median Q3
CEOATAX 301 0.306 0.461 0.000 0.000 1.000 GAAP ETR 301 0.286 0.161 0.205 0.296 0.378 Cash ETR 301 0.253 0.161 0.158 0.250 0.334 DTAX 154 0.000 0.032 -0.010 0.000 0.012 LnAssets 301 9.451 1.033 8.721 9.350 10.138 LnMV 301 9,801 0.947 9.084 9.670 10.415 ROA 301 0.107 0.089 0.059 0.097 0.150 OCF 301 0.122 0.063 0.081 0.114 0.157 LEV 300 0.238 0.166 0.135 0.217 0.312 CAP 301 0.264 0.233 0.086 0.181 0.363 RD 199 0.063 0.079 0.004 0.024 0.091 INTAN 301 0.245 0.211 0.055 0.216 0.402 FOREIGN 259 0.042 0.493 0.011 0.034 0.063 NOL 202 0.876 0.330 1.000 1.000 1.000
Note: Q1 and Q3 represent the first and the third quartile, respectively. Q1 implies that 25% of the firms have a value below the shown value in the table. Similarly, Q3 implies that 25% of the firms have a value above the shown value for Q3 in the table. Computations and definitions of all variables can be found in Appendix I.
Examining the GAAP ETR, the mean appears to be 1 percent point lower compared to the recent studies of Armstrong, Blouin and Larcker (2012) and Gaertner (2014). Also, a lower Cash ETR is found compared to the Armstrong, Blouin & Larcker (2012) study with a difference of more than 2 percent points. Overall, this suggests that firms in the sample have succeeded more in avoiding tax,
unintentionally or intentionally. Surprisingly, the DTAX measure shows a rounded mean of 0% which implies that, on average, firms in the sample did not have any discretionary permanent differences. Consequently, this suggests that the average firm is not associated with aggressive tax avoidance. The third quartile (Q3) of the DTAX shows that only 25% of the sample, around 75 firms, have more than 1,2% of discretionary permanent differences which is surprisingly low considering Frank, Lynch and Rego (2009) report a DTAX Q3 of 4% and Armstrong, Blouin and Larcker (2012) show a Q3 of 7,5%. Furthermore, due to missing values, DTAX, LEV, RD, FOREIGN and NOL have fewer observations than the initial final sample. Especially for the DTAX model, the reduced amount of observations can have consequences for the following analysis.
Table 5
Descriptive statistics of the PERMDIFF model
Variable N Mean Std. dev Q1 Median Q3
PERMDIFF 301 0.032 0.068 0.008 0.023 0.045 INTANG 301 0.270 0.238 0.062 0.223 0.428 UNCON 301 0.002 0.007 0.000 0.000 0.001 MI 301 0.001 0.007 0.000 0.000 0.006 CSTE 301 0.003 0.005 0.000 0.001 0.004 NOL2 170 0.014 0.082 -0.003 0.002 0.010 LAGPERM 300 0.029 0.061 0.006 0.023 0.041 LAGFOREIGN 259 0,046 0.054 0.012 0.036 0.068
Note: Q1 and Q3 represent the first and the third quartile, respectively. Q1 implies that 25% of the firms have a value below the shown value in the table. Similarly, Q3 implies that 25% of the firms have a value above the shown value for Q3 in the table. Computations and definitions of all variables can be found in Appendix I.
Table 5 shows the descriptives of the components of the first-order OLS regression of the DTAX model. All variables are scaled by lagged assets. For instance, the INTANG with a value of 0.270 means that, on average, intangibles made up 27% of beginning-of-the-year assets. The positive value for the mean of NOL2 implies that, on average, the firms in the sample had an increase in the tax loss carry forwards which indicates a net operating loss. The mean for the PERMDIFF is only 0.032 which indicates that the total permanent differences are 3.2 % of assets at the beginning of the year. This small proportion is consistent with the small mean found for the discretionary differences in Table 4 (i.e. DTAX with a mean of 0.0 %).
5. Results
5.1 Correlation matrix
Table 6 shows the correlation matrix for all measures of tax avoidance including all independent variables. For each combination, the Pearson correlation and its p-value are shown. Significance levels of 10%, 5% and 1% are indicated. The matrix shows that the main variable of interest, CEOATAX, is negatively correlated with a firm’s GAAP ETR with a significance level up to 5%. This is in line with H1 which states that firms that have after-tax metrics in their CEO bonus plans, have lower GAAP effective tax rates. However, both the Cash ETR and the DTAX measure show no significant results. This implies that coefficients cannot be considered different from zero and, therefore, this finding is not in line with H2 and H3. While these correlations can give a clear overview of the overall relationships, they do not control for other effects that might influence the relationship as with the OLS regression analyses. Furthermore, LnAssets is significantly correlated with CEOATAX showing a positive coefficient of 0.095. Overall, this indicates that the bigger firms, measured by total assets, tend to have after-tax metrics in bonus plans. Similarly, LnMV is positively significant indicating that the bigger firms in the sample, measured by market value, tend to have after-tax metrics in bonus plans. Also, both INTAN and FOREIGN are positively correlated with a significance level of 5% and 10%, respectively. This implies that higher intangible assets and higher foreign income are associated with having after-tax metrics in CEO bonus plans. Furthermore, the three models used in this study are all significantly correlated with each other. This is to be expected because these models all measure a part of tax avoidance to some extent. Evidently, the strongest relationship is found between the GAAP
ETR and the DTAX model as these models measure tax avoidance in the same way. Correlations with
the DTAX are negative because higher discretionary permanent differences show tax avoidance whereas with ETRs, lower effective tax rates indicate tax avoidance. Surprisingly, the correlation of the DTAX model and the NOL control is missing. The reason for this result is that the tax loss carry forward has too many missing values. See Appendix II for a detailed summary of missing values.
Furthermore, Table 6 yellow-flags several correlations with a value of 0.3 of higher. Three correlations with a value of 0.5 or higher are red-flagged. First, LnAssets and LnMV have a high correlation of 0.771, which implies that both variables explain the variance of the dependent variable in the same manner. Consequently, considering multicollinearity, this means that one of the two has to be excluded from the analysis. LnMV is dropped because of the higher correlation with FOREIGN. Second, the correlation between ROA and OCF also has a high value which results in the exclusion of one of the two variables. ROA is dropped due to the high correlations (i.e. yellow-flagged) with
Notes: P-values listed below correlations coefficients.*,**,*** denote significance at the 10% level, 5% level and 1% level, respectively. Coefficients between 0.3 and 0.5 are flagged yellow. Coefficients above 0.5 are flagged red.Computations of all variables can be found in Appendix I.
(1) CEOATAX 1 (2) CASH ETR 0.073 1 0.205 (3) GAAP ETR -0.140** -0.249*** 1 0.015 0.000 (4) DTAX -0.013 -0.132** -0.364*** 1 0.872 0.103 0.000 (5) LnAssets 0.095* -0.008 -0.019 0.013 1 0.100 0.895 0.747 0.878 (6) LnMV 0.139** -0.017 -0.012 0.003 0.771*** 1 0.016 0.767 0.830 0.971 0.000 (7) ROA -0.006 0,134** 0.040 0.003 -0.226*** 0.197*** 1 0.922 0.020 0.486 0.966 0.000 0.001 (8) OCF -0.057 0.038 0.084 -0.049 -0.211*** 0.182*** 0.704*** 1 0.323 0.515 0.146 0.547 0.000 0.002 0.000 (9) LEV -0.057 -0.020 0.043 0.047 0.022 -0.162*** -0.104* -0.074 1 0.328 0.735 0.455 0.559 0.706 0.005 0.073 0.202 (10) CAP -0.076 -0.046 0.141** -0.004 0,215*** 0.005 -0.169*** 0.089 0.161*** 1 0.187 0.426 0.015 0.960 0.000 0.935 0.003 0.125 0.005 (11) RD -0.062 -0.006*** -0.256*** 0.044 -0.077 0,151*** -0.320*** -0.004 -0,191*** -0.371*** 1 0.573 0.938 0.000 0.654 0.281 0.009 0.000 0.953 0.007 0.000 (12) INTAN 0,144** -0.019 -0.027 -0.046 0.041 0.027 -0.095 -0.235*** 0,132** -0.546*** 0.044 1 0.012 0.749 0.637 0.570 0.479 0.637 0.101 0.000 0.022 .000 0.538 (13) FOREIGN 0.106* 0.078 -0.172*** 0.000 0.020 0.234*** 0.466*** 0.399*** -0.113* -0.077 -0.144* -0.121* 1 0.089 0.211 0.006 0.998 0.744 0.000 0.000 0.000 0.071 0.218 0.053 0.052 (14) NOL 0.051 -0.084 -0.012 .c -0.045 -0.078** -0.157** -0.035 0.058 -0.097 0.178* 0.096 0.088 1 0.473 0.234 0.861 0.000 0.521 0.269 0.26 0.619 0.416 0.171 0.043 0.175 0.244
OCF and FOREIGN (i.e. 0.399). While both operating cash flow and foreign income capture
profitability to some extent, both are included in the analysis because the FOREIGN variable controls specifically for the international profitability of firms. Hence, in the context of measuring tax
avoidance, it is essential that international activity is taken into account. The models of Chapter 3 are modified as follows;
𝐸𝑇𝑅 = 𝛽0+ 𝛽1 𝐶𝐸𝑂𝐴𝑇𝐴𝑋𝑖,𝑡+ 𝛽2 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+ 𝛽3 𝑂𝐶𝐹𝑖,𝑡+ 𝛽4 𝐿𝐸𝑉𝑖,𝑡+ 𝛽5 𝑅𝐷𝑖,𝑡+ 𝛽6 𝐼𝑁𝑇𝐴𝑁𝑖,𝑡
+ 𝛽7 𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝛽8 𝑁𝑂𝐿𝑖,𝑡+ 𝜀𝑖,𝑡
𝐷𝑇𝐴𝑋 = 𝛽0+ 𝛽1 𝐶𝐸𝑂𝐴𝑇𝐴𝑋𝑖,𝑡+ 𝛽2 𝐿𝑛𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+ 𝛽3 𝑂𝐶𝐹𝑖,𝑡+ 𝛽4 𝐿𝐸𝑉𝑖,𝑡+ 𝛽5 𝑅𝐷𝑖,𝑡+ 𝛽6 𝐼𝑁𝑇𝐴𝑁𝑖,𝑡
+ 𝛽7 𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑖,𝑡+ 𝜀𝑖,𝑡
The ETR as the dependent variable can either be GAAP ETR or CASH ETR in accordance with the models described in Chapter 3. Note that the NOL control is excluded from the DTAX model due to a missing correlation (see correlation matrix of Table 6). Computations and definitions of all variables are presented in Appendix I.
5.2 Regression results
Table 7 shows the main results for OLS regressions which included the two ETR models. For the
GAAP ETR model, the F-statistic is significant up to 1% which indicates that the explained variance is
relatively large compared to the unexplained variance. In contrast, the value of the F-statistic is 1.535 and not significant for Cash ETR. This implies that the unexplained variance in the model is almost equal to the explained variance. The adjusted R² indicates a models’ ability to estimate the variance of the observations. The GAAP ETR model has a higher value (15.1%) compared to the Cash ETR (3.5%). Other studies have reported similar adjusted R² for GAAP ETR (Gaertner 2014, Armstrong, Blouin & Larcker 2012, Hope, Ma & Thomas 2013, Phillips 2003). However, for Cash ETR the R² is relatively low. Another troublesome result is the number of observations in both models. While the initial final sample contained 301 observations, the OLS regression of the ETR models only includes up to 117. The reason for the exclusion of 61.1% of the sample is the number of missing values. Specifically, the tax loss carry forward, R&D expense and foreign income have the most missing values in the COMPUSTAT database. The additional analyses of Chapter 6 try to solve this issue by introducing new observations. See Appendix II for specific information about missing values.
In contrast to the result in the correlation matrix, the dummy variable CEOATAX is not significant in the GAAP ETR model. Similarly, the coefficient of CEOATAX is not significant in the Cash ETR model. These findings suggest that there is no evidence that firms with after-tax metrics in their CEO bonus plans have lower ETRs. This result is inconsistent with H1 and H2 and suggests that
