The Ethics of Tax Avoidance The moderating effect of internationalization on the relationship between CSR and tax avoidance

(1)

avoidance

By Willem Sträter

s2192647

MSc. International Financial Management Faculty of Economics and Business

University of Groningen 13 January 2016

ABSTRACT

This paper examines how economic, environmental and social CSR activities are related to tax avoidance. Subsequently, this study examines whether internationalization moderates the relationship between the different CSR activities and tax avoidance. A matched sample of 266 firm-year observations was formed; equally split between tax avoidant firms as well as tax compliant firms by employing a novel approach to identify corporate tax avoidance based on tax disputes. The logit regression results shows that the more firms engage in social CSR activities the more likely they are to avoid taxes. Moreover, the main regression also supports the positive moderation effect of internationalisation on the relationship between social CSR and tax avoidance

Keywords:

Tax avoidance • Business ethics • Corporate social responsibility • Internationalization

(2)

1

List of Tables

Table 1: Matched sample mean, median and standard deviation statistics……… 14

Table 2: Descriptive statistics……… 21

Table 3: Matched sample descriptive statistics……….. 23

Table 4: Correlation matrix and multicollinearity statistics………...24

Table 5: Binominal Logit Regression ………28

Table 6: OLS Regression – Total book-tax difference………..32

Table 7: OLS Regression – Residual book-tax-difference……….33

Table 8: Paired samples t-test for Matched sample descriptive statistics……….. 42

Table 9: Paired samples t-test for Matched sample mean, median and standard deviation statistics……….….42

Table 10: Sign-test of Matched sample descriptive statistics ………43

Table 11: Sign-test of Matched sample mean, median and standard deviation statistics….. 43

Table 12: Box-Tidwell Linearity test……….44

(4)

3

List of Abbreviations

BTD Book-tax-difference

CAPINT Capital Intensity

CFO Cash Flow from Operations

CSR Corporate Social Responsibility EBIT Earnings before Interest and Taxes ECNSCORE Economic CSR score

ENVSCORE Environmental CSR score

ETR Effective tax rate

INSIDST Managerial stockownership INTERZ Internationalization

INVINT Inventory Intensity

LEV Leverage

MTB Market-to-book

OECD Organisation for Economic Co-operation and Development

ROA Return-on-assets

SIZE Firm size

SOCSCORE Social CSR score

TA Total accruals

TAXD Tax disputes

UN United Nations

(5)

4

1. Introduction

Over the last fifty years the distribution of the tax burden for US citizens has shifted dramatically. Today, individual tax payers account for a substantially bigger share of total tax income than they did in the past (Christensen and Murphy, 2004). Not only did the corporate tax rates drop drastically over the last decades, corporations also greatly increased their levels of tax avoidance (Sikka, 2010). Especially tax havens have enabled organizations to decrease their tax bill, as it is estimated that 42 percent of net income earned by U.S. organisations is earned in tax havens (Dharmapala and Dhammika, 2014). In today’s world tax havens are more easily accessible as previous studies have shown that globalization, communication’s innovation and capital mobility have played their part in enable organizations to avoid more taxes (Rego, 2003; Christensen and Murphy, 2004; Wilson, 2009). Especially the improvements in capital mobility have enabled organisations to cherry pick particular jurisdictions with favourable tax laws.

The damaging effects of tax avoidance have been discussed thoroughly in previous studies (McGee, 2006; Sikka, 2010; Stephenson and Vracheva, 2015; Hanlon and Heitzman, 2010). The link between corporate social responsibility (CSR) and tax avoidance has only been explored recently by Lanis and Richardson (2015). In this context, CSR is the belief that companies go beyond that what is required of them by law, by taking responsibility of social wellbeing. Therefore, in light of the damaging effect of tax avoidance, one would expect to see that the increase in socially responsible corporations leads to less corporate tax avoidance.

However, as mentioned before, tax avoidance is on the rise. This counterintuitive link between CSR and tax avoidance can be viewed from a shareholder and a stakeholder perspective. Under the shareholder perspective CSR is viewed as a voluntary activity with moral undertones (Timonen, 2008) where shareholders’ money is spent on CSR activities (Friedman, 1970). On the contrary, stakeholder theorists argue that organisations have a moral obligation to their stakeholders to engage in CSR activities (Sikka, 2010). Avi-Yonah (2008) argues that regardless which perspective a firm holds on taxation, CSR is linked to taxation because tax income is used to fund the development of society and its stakeholders, but also numerous services demanded by organisations such as legal systems and oversight (Avi-Yonah, 2008).

This study builds on previous studies in the area of CSR and tax avoidance by Hanlon and Heitzman (2010) and Lanis and Richardson (2012; 2015). These studies have called for a more thorough analysis of which CSR activities are associated with the likelihood of

(6)

5 corporate tax avoidance. A study by Huseynov and Klamm (2012) found that there is a link between tax avoidance and CSR by testing the effects of KLD database constructs corporate governance, community and diversity on tax avoidance. A study from the same year by Lanis and Richardson (2012) found that social investment commitment was negatively related to tax avoiding corporate behaviour. This was supported by Hoi, Wu, and Zhang (2013) who showed that firms with excessive irresponsible CSR activities are more likely to avoid taxes.

They scored firms on their amount of negative CSR ratings from the KLD database and did not study the effect of responsible CSR activities. Contrary to these findings a study by Davis et al. (2013) shows that CSR has a negative relation with five year effective tax rates and has a positive relation to tax lobbying. The study suggests that firms do not consider taxes a socially responsible construct even though they engage in other CSR activities. Recently two studies came yet again to the conclusion that CSR is negatively related to CSR. The first by Muller and Kolk (2015) found that multinationals in India had a higher effective tax rate when they had a higher CSR reputation. Lanis and Richardson (2015) studied the effect of CSR on tax avoidance and took the CSR concept as a whole. They used the same KLD dataset as used by Huseynov and Klamm (2012), Hoi, Wu, and Zhang (2013), and Davis et al. (2013) and find that CSR is associated with less tax avoidant behaviour. They improve on previous studies by using a direct measure of tax avoidance based around firm tax disputes instead of the previous employed indirect measures.

This study will answer the call by Hanlon and Heitzman (2010) and Lanis and Richardson (2015) for a more thorough analysis of CSR activities and its effect on tax avoidance. In this study the concept CSR will be dissected into three parts, economic, environmental and social. It tries to extend on previous studies by using a different metric of CSR and a direct measure of tax avoidance in favour of the indirect measures such as effective tax rates and book-tax differences. This study also aims to find if internationalization moderates the relationship between CSR and tax avoidance. Previous studies have shown that more internationalized firms engage in more CSR, but academics also found that globalization and tax avoidance are positively related. The moderation effect of internationalization will therefore be tested on the relationship between CSR and tax avoidance.

The paper is structured as follows. The next section provides a review of the existing literature on the ethics of tax avoidance, the different corporate social responsibility components and the moderating effect of internationalization on the relationship between tax

(7)

6 avoidance and CSR. Section three discusses the research design and explains the regression model. While section four describes the empirical results from the main regression model as well as the robustness tests. To conclude, section five discusses the results and puts them in perspective, while also explaining the limitations inherent to this study and suggestions for future research.

(8)

7

2. Literature Review

2.1 The ethics of tax avoidance

A very narrow division exists between tax avoidance and tax evasion. However, it is generally agreed upon that tax avoidance is a legitimate way to minimize taxes, whereas tax evasion entails practices that contravene the law (Sikka, 2010). However, even though tax avoiding strategies do not contravene the letter of the law, they do break the spirit of the law.

Thus, making them unjustifiable for many scholars and legislators as they argue that these tax avoiding practices are unethical (McGee, 2006).

The ethics of tax avoidance have been discussed for a considerable amount of time.

Martin Crowe wrote an article in 1944 on different ethical perspectives on tax avoidance, which is still relevant today. He argued that the ethics of tax avoidance can be examined from three of perspectives. First, by the relationship between citizens and their religion or belief system. Second, the relationship between the citizens and the state. Third, the relationship between the citizens and taxpayers in society (Crowe, 1944). Most studies have focused on the second relationship, as to why citizens do not have a moral duty to pay taxes to the state (Ballas and Tsoukas, 1998). The majority of these have focused on government corruption, the illegitimacy of the state or the inability of the state to properly collect taxes (McGee, 1999). However, in democracies the belief is that tax avoidance is almost always unethical (McGee, 2006). Most strong democracies have low government corruption (Kolstad and Wiig, 2015), the state is legitimate and there is a strong belief that individuals should conform to majority rule. Nevertheless, even if there is no moral duty to pay taxes to the government there is a moral duty to other taxpayers (Lehmkuhl, 1902). This argument is in accordance with the third perspective on the ethics of tax avoidance, the relationship between the citizens and taxpayers in society. Lehmkuhl (1902) argues that there is a moral obligation to pay taxes based on the tax burden shift that occurs when taxes are avoided in society. This is supported by Davis (1938) who argues that taxpayers have to make up for the tax avoidance of others. Tax avoidance in these societies allows companies to become economic free-riders. They enjoy the benefits of corporate citizenship in society, but do not pay their fair share. Which causes a tax income void, left behind by these organizations, to be transferred to individual tax payers (Christensen and Murphy, 2004). Thus, the social tax burden increases for regular citizens which cannot benefit from the same tax advantages as their corporate counterparts. However, these are just the direct effects of tax avoidance on tax payers and citizens as a whole. Just as important are the indirect effects such as the societal

(9)

8 implications of tax avoidance as taxes are a fundamental way to finance the provision of public goods in society (Freedman, 2003; Friese, Link, and Mayer, 2008). Christensen and Murphy (2004) even argue that paying taxis is the most fundamental way of engaging with broader society for private and corporate citizens. This argument is supported by Hutton (2002), who argues that tax revenues are vital to the development and maintenance of physical infrastructure, institutions and justice that underpin the market economy. Thus, based on the ethics of tax avoidance this study will choose the third approach. This approach is the most suitable research perspective as it examines tax avoidance from the perspective of citizens and taxpayers in society. Therefore, providing an ideal starting point to gain a more detailed understanding of how CSR and tax avoidance are associated.

To conclude, not paying taxes has an impact on communities especially because organisations benefit from what society provides, such as a well-educated labour force, good infrastructure, strong legal systems and institutions paid for by government tax income. This has resulted in many tax avoiding strategies to be deemed tax evasion after these strategies were scrutinized in court. Especially tax avoiding schemes with hardly any economic substance have been considered unacceptable based on ethical grounds (Aid, 2009).

2.2 Tax avoidance as a CSR component

The first notable mentions of the concept corporate social responsibility (CSR) emerged around the 1950s. Bowen and Johnson (1953) laid the foundation for CSR research with their book ‘The social responsibilities of the businessman’. In his book he questions “What responsibilities to society may a businessman reasonably be expected to assume?” and by doing so he marks the beginning of the modern period of CSR literature (Carroll, 1999).

During the 1960s and 70s the definition of CSR was further explored until Thomas J. Jones (1980) formulated an interesting perspective on CSR. His view on CSR was that

“corporations have an obligation to constituent groups in society other than stockholders and beyond that prescribed by law and union contract” and most importantly that CSR should be viewed as a process, not as a set of outcomes as was previously assumed. After the 1990s not many new insights on the concept CSR have been added to the existing literature and the focus shifted towards other themes s1uch as business ethics and stakeholder theory which are often complementary (Carroll, 1999). In the broadest of terms CSR firms are defined as companies that should strive to “make a profit, obey the law, be ethical and be a good corporate citizen” (Carroll, 2006). No more specific and universally accepted definition exists

(10)

9 on the subject of CSR. However, it is universally agreed upon that CSR incorporates issues such as “sustainability, sustainable development, environmental management, business ethics, philanthropy and community investment, worker rights and welfare, human rights, corruption, corporate governance, legal compliance and animal rights” (Dillard and Murray, 2013). Following Moser and Martin (2012), this study will adopt a broad perspective on CSR activities in which these corporate activities affect all of the firm’s stakeholders.

In light of the level of attention given to corporate social responsibility by organizations (Matten and Moon, 2008) it should be expected that firms are mindful of their level of tax avoidance. The foundation of CSR is built around sustainable organizational behaviour and as argued before not paying taxes has an impact on communities and is widely regarded as unethical (Velasquez, 2012). However, activities solely designed to reduce the corporate tax bill are becoming more widespread across the corporate landscape (Lanis &

Richardson, 2015). Thus, on one hand CSR activities are more prevalent among organizations while on the other hand tax avoidance is on the rise. This is contradictory with recent findings (Lanis and Richardson, 2012; 2015; Watson, 2011; Hoi, Wu, and Zhang, 2013; Muller and Kolk, 2015) which show there is a negative relationship between CSR and tax avoidance. Thus, one would expect that an increased attention to CSR would result in less corporate tax avoidance. However, CSR is a broad concept which contains many different socially responsible activities, so it would not be surprising if each of these activities has a different effect on the level of corporate tax avoidance.

2.3 Economic, environmental and social CSR activities

Corporate social responsibility can be divided among three main categories, namely economic, environmental and social CSR activities. Firm have different motivations why to engage in either of these three CSR activities, which will be discussed in turn.

Firms have an obligation to make a profit, not only to their shareholders, but also to their other stakeholders. Economic CSR activities ensure that an organisation is able to generate sustainable growth and a high return on its investments by allocating its resources efficiently. It ensures that the firm maintains in strong financial health and is able to generate long term shareholder value by using best management practices (Thomas Reuters, 2016).

The largest benefits of tax avoidance are that it is a relatively inexpensive source of financing (Armstrong, Blouin and Larcker, 2012) and legal way to boost profits (Scholes et al., 1992).

However, tax avoidance can also have a negative effect on the sustainable growth and long

(11)

10 term shareholder value of a firm. Firms that have been identified by tax authorities as tax avoiding may be forced to pay additional taxes, interest and penalties (Crocker and Slemrod, 2005). Which results in decreased cash flows and lower shareholder value and thus tax avoidance can be detrimental to the economic sustainability of the firm. In addition, Hanlon and Slemrod (2009) found that news about the tax avoidance practices of a firm have a small but negative reaction on the stock price of said firm of -1.04 percent. They also found in support of Hardeck and Hertl (2014) that this negative reaction is more prominent for firms in more consumer oriented fields. Furthermore, the activities which fall under the umbrella of economic CSR are more than just financial sustainability. These activities foster innovation, job creation, capital investments and limit anti-competitive behaviour and price fixing. Thus, tax avoidance is most likely incompatible with economic CSR activities and therefore this study hypothesizes that:

H1: Firms engaging in more economic CSR activities are less likely to engage in tax avoidance

Corporate reputation is a vital attribute to the success of a company (Vonwil and Wreschniok, 2009) and a very important reason for companies to engage in environmental and social CSR activities (Babiak and Trendafilova, 2011). Both the environmental and the social CSR activities have an effect on how a firm interacts directly with its surroundings. As opposed to economic CSR activities they are under more scrutiny due to the increased visibility and impact of these activities. For example, social CSR activities ensure that an organization gains and retains the trust and loyalty of its workforce, customers and society by focussing on human rights, employment quality, training, but also trying to limit misconduct and corruption. Environmental CSR activities span from the reduction of waste and greenhouse gases to spills and water usage (Thomas Reuters, 2016).

As mentioned before the motivation to engage in environmental and social CSR activities stems mainly from the concept of reputational costs as the reputation of a company can be severely damaged if customers come to know about the tax aggressive behaviour by a firm.

This was supported by Hardeck and Hertl (2014) who found that customers punish tax avoiding companies by purchase intent and reputational harm. Thus, tax avoidance would affect customer loyalty and therefore decrease the level of sales of products or services, which opposes the intention of social CSR. However, even though the backlash of tax

(12)

11 avoidance seems severe, companies motivated by cost-benefit analyses might still avoid taxes if the benefits outweigh the costs incurred by tax avoidance.

In recent years, the environment has been the most important concern for stakeholders in company’s CSR efforts (Kassinis and Vafeas, 2006). Because stakeholders, such as customers, value organisations which undertake environmental CSR activities it is not surprising that the focus of organisations on environmental CSR activities has increased (Babiak and Trendafilova, 2011). However, corporate motives for engaging in environmental CSR do not necessarily stem from the moral values an organisations holds. An emerging perspective on CSR is that of risk management (Hoi, Wu, and Zhang, 2013). As discussed before, the reputational damages of being associated with tax avoidance as a company can be severe. This perspective views CSR as a method to increase a firm’s reputation as a good corporate citizen and therefore is able to mitigate negative corporate publicity and events.

Godfrey (2005) supports this idea and argues that stakeholders might temper the negative backlash towards the organization due to goodwill. Thus, CSR can be used as a pre-emptive method to mitigate the negative reaction of stakeholders by either decreasing the amount of irresponsible activities or increasing the responsible CSR initiatives (Minor and Morgan, 2011; Godfrey, 2005). In the context of this paper, these environmental CSR activities can be used to obscure socially irresponsible activities such as tax avoidance. Consistent with the risk management approach is the cost-benefit analysis of CSR activities in regard to environmental CSR. As these activities are costly, a firm could support these activities by engaging in more tax avoidance. In this instance, CSR is merely used as a method to enhance its reputation as a corporate citizen. If so, a firm pretends to act in the interest of other stakeholders, whilst mostly caring about its shareholders. Which supports the idea of Sikka (2010) when he conjured the term ‘organized hypocrisy’.

Thus, in the face of risk-management theory and the larger visibility of environmental and social activities, this study hypothesizes that:

H2: Firms engaging in more environmental CSR activities are more likely to engage in tax avoidance

H3: Firms engaging in more social CSR activities are more likely to engage in tax avoidance

(13)

12 2.4 Internationalization, CSR and tax avoidance

As mentioned before the studies have shown that globalization, communication’s innovation and capital mobility have played their part in enable organizations to avoid more taxes (Rego, 2003; Christensen and Murphy, 2004; Wilson, 2009). Especially capital mobility was mentioned as an enabler of tax avoidance. Most multinational firms have the advantage over purely national companies that they have assets and sales abroad, providing more opportunities to funnel profits between different tax jurisdictions. Thus, one would expect internationalized firm to be more tax avoidant. However, this is most likely not the case as higher internationalization also leads to more governmental and social scrutiny. First, governmental scrutiny results in more disclosure obligations and subsequently transparency of a firms activities. Second, due to internationalization firms are more likely to engage in CSR activities both actively and passively. Actively as internationalized firms need to satisfy a more and diverse group of stakeholders. Internationalized firms choose to engage in CSR due to an increase in public visibility, making it easier for the public to monitor and be informed about a company’s activities (Liang et al., 2014). Furthermore, not engaging in CSR activities, under these conditions, will result in negative reputational costs (Hardeck and Hertl, 2014). On top of that, firms engage in CSR activities passively due to increased external pressure from NGO’s and governments, due to their increased visibility and global activities. For example, more internationalized firms need to also adhere to guidelines, principles and declarations issued by the UN and OECD, which dictate more socially responsible behaviour (Kercher, 2007). Thus, the opportunities of globalization to avoid taxes are offset by greater visibility and governmental scrutiny due to negative reputational costs and increase transparency.

Therefore, it is expected that the association between tax avoidance and different CSR activates is contingent on the extent to which companies are internationalized.

Thus, this study hypothesizes as follows:

H4a: The relationship between tax avoidance and economic CSR activities is moderated by a firm’s internationalization; Firms that undertake economic CSR activities with higher levels of internationalization are less likely to engage in tax avoidant behaviour.

(14)

13 H4b: The relationship between tax avoidance and environmental CSR activities is moderated by a firm’s internationalization; highly internationalized companies which undertake environmental CSR activities are less likely to engage in tax avoidant behaviour.

H4c: The relationship between tax avoidance and social CSR activities is moderated by a firm’s internationalization; highly internationalized companies which undertake social CSR activities are less likely to engage in tax avoidant behaviour.

(15)

14

3. Research Design

3.1 Research data

The sample consist of firms on the S&P 500 index during the 2004–2009 period. This period was chosen due to data accessibility restrictions on the variable measuring tax disputes. The data was retrieved from the Thomas Reuters Wordscope Database on 13 October. The data regarding the tax disputes was retrieved from the WRDS MSCI KLD database on the 30^th of October. The retrieved data covers a period from 2003 to 2009 of which the data from 2003 is only used to calculate book-tax-differences in the robustness tests based on lagged total assets. The total retrieved raw data sample comprised of 2973 firm-year observations of which 193 had been involved in a tax dispute with a governmental body.

3.2 Matched sample

Following the research by Lanis and Richardson (2015) a matched sample was created. The goal of matching observations is too improve the validity by eliminating possible effects of other variables not under investigation (OECD, 2004)¹. This is especially important when using the tax disputes as a measure of tax avoidance as there is a possible selection bias present in the sample. For instance only firms that have been formally investigated or charged by governmental bodies over their tax obligations end up in the sample. This leads to tax avoidant firms that have not yet been in a tax dispute being identified as tax compliant. Thus, by using a matched sample the validity of the sample is drastically improved as it ensures that no tax avoidant firms are misidentified as non-tax avoidant. This is done by excluding all observations of single firm if it has been identified as tax avoidant in one year. This ensures that tax avoidant firm year observations are matched with firms that have never been in a tax dispute during the sample period.

The matched sample was set up in the following manner; every tax avoidant firm was matched with a non-tax avoidant firm from the same year and industry. The final selection was made by comparing the market value of equity between the tax avoidant firm and the possible matches. The non-tax avoidant firm with the smallest difference in market value of equity was matched with the tax avoidant firm. The largest deviation in market value of equity between two matched firms is 32.5%, which is well below the cut-off point used in previous studies (Kaplan and Reishus, 1990; Lanis and Richardson, 2015).

1 OECD , 2004). Retrieved 28 December 2016, from https://stats.oecd.org/glossary/detail.asp?ID=3709

(16)

15 This results in a matched sample consisting of 272 firm year observations of which three firm year observations with their corresponding match have been removed from the sample due to their studentized residuals being above 2,5 standard deviations (Cohen et Al., 2003). Thus, making the final sample consist of 266 firm year observations of which 140 are different firms over the 2004-2009 period.

Table 1

Matched sample mean, median and standard deviation statistics ($1=$1000)

Variable TAXD=1 TAXD=0

Total Assets 52.397 52.804

(25.488) (21.418)

[105.637] [128.311]

Net Sales 38.007 33.222

(14.628) (12.802)

[70.235] [63.153]

Market value of equity 54.108 52.278

(20.183) (19.273)

[80.091] [73.212]

GAAP ETR² 33% 27%

(34%) (30%)

[22%] [30%]

ROA³ 13% 11%

(11%) (10%)

[15%] [8%]

Note: Medians in parentheses, standard deviations in square brackets

A paired samples t-test was used to determine whether the mean difference between paired observations is statistically significantly different from zero. The results show that the means from the two groups do not significantly differ from each other. A sign test was also carried out to determine whether the median difference between the matched observations is significantly different from zero. The test shows that the effective tax rate (GAAP ETR) medians are statistically different from one another (ρ < 0.05). This is a surprising result as one would expect the group which has been identified as tax avoidant to have a lower effective tax rate.

2 GAAP ETR = income tax expense divided by pre-tax accounting income

3 ROA = pre-tax accounting income divided by total assets

(17)

16 3.3 Dependent variable

The dependent variable tax avoidance is difficult to measure as tax fillings are secret. Most studies measure the level of tax avoidance using proxy measures obtained from financial statements. The most common methods are book-tax difference measures (Desai and Dharmapala, 2006; Frank, Lynch, and Rego, 2009) and effective tax rates (ETR) (McGuire, Omar, and Whang, 2006; Chen et al., 2014). The first method measures the difference between the reported GAAP financial income and the income tax expense worldwide plus the statutory tax rate (Manzon and Plesko, 2001). The second method, ETR, equals the total income tax expense divided by the worldwide pre-tax accounting income; where the difference between the GAAP and the statutory tax rate is the amount of tax avoidance (Hanlon and Heitzman, 2010). However, there are multiple issues when using financial statements as a primary source of information. Financial statements reflect the economic performance of a firm for external stakeholders and do not inherently disclose tax data, except those items which affect earnings and the balance sheet (Hanlon and Heitzman, 2010).

Even if the tax return data was available it would still be very difficult to measure how much tax was played on the reported earnings, because it is almost impossible to match tax returns to financial statements under different consolidation rules (Mills and Plesko, 2003).

Furthermore, tax laws and enforcement are organized at a national level, thus using U.S. tax returns only provides information about the U.S. portion of the activity of a U.S.

multinational organization (Hanlon and Heitzman, 2010). A recent valid method of combatting some of these issues is the use of a direct measure of tax avoidance based on firm tax disputes. Under this approach tax avoidant firms are defined as firms which have had a tax dispute involving federal, state, local or non-U.S. government authorities, or are involved in controversies over its tax obligations to the community (MSCI, 2016). Previous research has shown that firms accused by the government of engaging in tax shelter activity are a good predictor of tax avoidance (Graham and Tucker, 2006; Lanis and Richardson, 2012; 2015).

Wilson (2009) also finds that book-tax differences are positively associated with incidences where firms were accused by the government of tax sheltering. The main reason for using tax disputes as opposed to BTDs or ETRs is the fact that this direct measure of tax avoidance does not automatically classify firms with a low tax liability as tax avoidance, whereas BTDs and ETRs, due to their continuous measuring levels, do. For example, firms could avoid taxes through investing in government approved tax advantaged assets (Berger, 1993), if so these

(18)

17 firms have lower effective tax rates, but are not behaving unethically. Therefore, only firms which are avoiding taxes not within the spirit of the law, will end up in the sample. This makes tax disputes an ideal measurement of tax avoidance for this study as it is able to more accurately identify tax avoidant firms.

Tax avoidance (TAXD) is denoted as a dummy variable. The variable will take on a 1 if the firm has been involved in a major tax dispute with federal, state, local or non-U.S.

government authorities, or are involved in controversies over its tax obligations to the community. Coded 0 if the firm is not involved in a major tax dispute over the period. The values for this dummy variable are retrieved from the KLD database over the 2004–2009 period.

3.4 Independent variables

The independent and control variables are retrieved from the Thomas Reuters ESG Asset 4 database covering the period 2004–2009. Firms listed in this database are annually reviewed manually on more than 900 data points which are equally weighted to form 250 key performance indicators (KPIs). These KPIs are then combined into 18 categories which form the four pillars of CSR (Cheng, Ioannou, and Serafeim, 2014). This level of detail in the data related to CSR is pivotal to answer what CSR practices make organisations less likely to engage in tax avoidance. Previous studies have supported the use of the Asset4 database as a viable measure of CSR (Cheng, Ioannou, and Serafeim, 2014; Lys, Naughton, and Wang, 2015). Three CSR variables are retrieved from this database, which are aggregated scores on CSR key performance indicators and range from 0 to 1.0. First, the economic performance score (ECNSCORE), which measures ability of the organization to generate sustainable growth and high investment return through the effective use of resources and best management practices (Thomas Reuters, 2016). Theme’s underlying this score are financial performance, client and shareholder loyalty, anti-competitive behaviour and price fixing.

Second, the environmental performance score (ENVSCORE), which measures the impact a firm has on its natural environment. It reflects how a firm tries to avoid environmental risks and capitalizes on environmental opportunities to generate long term shareholder value (Thomas Reuters, 2016). Theme’s underlying this score are resource uses, emissions, waste and innovation. The third and final score is the social performance score (SOCSCORE), which measures the ability of organizations to generate trust and loyalty within its community by treating its employees, customers and society right. This measure reflects an

(19)

18 organizations reputation and perceived license to operate (Thomas Reuters, 2016). Theme’s underlying this score human rights, product responsibility, workforce and community.

3.5 Moderation terms

A firm’s internationalization is measured by the percentage of foreign sales over total sales⁴. This variable is required to test for the moderation effect of internationalization on the relationship between the dependent variable tax avoidance and the independent variable CSR an interaction term is computed. Three different interaction terms are added to the model to test for the moderation effects of internationalization on the relationship between the three different CSR activities and tax avoidance. The independent variables INTERZ and independent variables ECNSCORE, ENVSCORE, and SOCSCORE are mean centered to account for multicollinearity between the interaction terms and independent variables.

Subsequently, the mean centered variables are multipled with each other to form three interaction terms, namely INTERZ_ECN, INTERZ_ENV and INTERZ_SOC. These moderation terms are added in the model to test how internationalization affects the relationship between the three CSR activities and tax avoidance.

3.6 Control variables

The control variables for the baseline regression are based on the outcome of prior research on tax avoidance. The control variables include proxies for capital structure (LEV), asset mix (CAPINT; INVINT), agency costs (INSIDST), firm size (SIZE), firm performance (ROA), growth (MTB) and industry and year dummies (INDUSTRY; YEAR). Leverage (LEV), measured as long and short term debt divided by total assets, is included as a control variable as research by Gupta and Newberry (1997) and Stickney and McGee (1983) have shown that firms with higher leverage had lower ETRs. Mainly due to interest payments being tax deductible. They also showed that capital intensive firms were associated with lower ETRs due to investment tax credits and long depreciation schedules. Thus, capital intensity (CAPINT), measured as the percentage of capital expenditures over total sales, is included in the regression. Furthermore, inventory intensity (INVINT), measured as total inventory divided by total assets, is included in the model as the opposite of capital intensity based on the study by Lanis and Richardson (2015). The stock ownership of insiders (INSIDST) is also

4 The data is retrieved from the Thomas Reuters Worldscope Database (2016)

(20)

19 (1) included in the regression model. The variable measures the percentage amount of stocks held by insiders as higher stockownership by insiders leads to a greater incentive to boost firm profits (Jensen and Meckling, 1976). Tax avoidance provides the opportunity to insiders to increase profits and subsequently stock prices by lowering the firm’s tax obligations. Firm size (SIZE), measured as the natural logarithm of total assets, is included in the model based on a previous study by (Siegfried, 1974) who found an association between firm size and effective tax rates. He argued that larger firms can influence the political process in their favour and can develop greater tax planning expertise due to their greater resources. The variable measuring profitability is Return-on-assets (ROA), measured as pre-tax income divided by total assets, which has been found by Spooner (1986) to be associated with tax avoidance, however Lanis and Richardson (2015) were unable to support these finding. The last continuous control variable is market-to-book (MTB), which is defined as the market value of equity divided by the book value of equity. It is a proxy of growth prospects and has been associated with tax avoidance in prior studies (Kim and Limpaphayom, 1998; Hoi, Wu, and Zhang, 2013). The last two control variables are dummies to account for differences in firm’s industries (INDUSTRY) and the effect of different years (YEAR) on the model.

3.7 Regression model

The main regression model to test the four hypotheses is a binominal logit regression, mainly due to the dependent variable being a dummy, which takes on the value of 1 if a firm is tax avoidant and 0 if a firm is tax compliant. Besides the obvious argument to use a logit regression because of the dependent variables it is also the most appropriate method when using a matched sample (Maddala, 1991). For the reason being that the sample is not random and exactly half of the sample consists of tax avoidant firms and the other half of tax compliant companies. Maddala (1991) argues that by using a binominal logit regression model the coefficients of the independent variables are unaffected and only the constant term is affected by the matched sampling. This will not impact the result of the main regression as the constant term is of no importance in this study.

The following full baseline regression model was used to test the hypotheses:

(21)

20 𝑇𝐴𝑋𝐷𝑖𝑡 = 𝛼0 + 𝛽¹𝐸𝐶𝑁𝑆𝐶𝑂𝑅𝐸𝑖𝑡 + 𝛽²𝐸𝑉𝑁𝑆𝐶𝑂𝑅𝐸𝑖𝑡 + 𝛽³𝑆𝑂𝐶𝑆𝐶𝑂𝑅𝐸𝑖𝑡

+ 𝛽⁴𝐼𝑁𝑇𝐸𝑅_𝐸𝐶𝑁𝑖𝑡 + 𝛽⁵𝐼𝑁𝑇𝐸𝑅_𝐸𝑉𝑁𝑖𝑡 + 𝛽⁶𝐼𝑁𝑇𝐸𝑅_𝑆𝑂𝐶𝑖𝑡 + 𝛽⁷𝑆𝐼𝑍𝐸𝑖𝑡 + 𝛽⁸𝐿𝐸𝑉𝑖𝑡 + 𝛽⁹𝐶𝐴𝑃𝐼𝑁𝑇𝑖𝑡 + 𝛽¹⁰𝐼𝑁𝑉𝐼𝑁𝑇𝑖𝑡 + 𝛽¹¹𝑀𝑇𝐵𝑖𝑡 + 𝛽¹²𝐼𝑁𝑆𝐼𝐷𝑆𝑇𝑖𝑡 + 𝛽¹³𝑅𝑂𝐴𝑖𝑡 + 𝛽¹⁴𝐼𝑁𝐷𝑈𝑆𝑇𝑅𝑌𝑖𝑡 + 𝛽¹⁵𝑌𝐸𝐴𝑅 + 𝜀𝑖𝑡

Where TAXD𝑖𝑡 is a dummy variable coded 1 for every year in which a firm was in a tax dispute with federal, state, local or non-US government authorities and 0 otherwise;

ECNSCORE𝑖𝑡, EVNSCORE𝑖𝑡 and SOCSCORE𝑖𝑡 represent the three different measures of corporate social responsibility on a scale from 0 to 1.0; INTER_ECN𝑖𝑡, INTER_ENV𝑖𝑡 and INTER_SOC𝑖𝑡 are the interaction term of mean centred INTERZ𝑖𝑡 and ECNSCORE𝑖𝑡;

SIZE𝑖𝑡, the natural logarithm of total assets; LEV𝑖𝑡, long term debt divided by total assets;

CAPINT𝑖𝑡, net property, plant and equipment divided by total assets; INVINT𝑖𝑡, inventory divided by total assets; MTB𝑖𝑡, the market value of ordinary equity divided by the balance sheet value of the ordinary equity in the company; INSIDST𝑖𝑡, the percentage of stocks owned by top management; ROA𝑖𝑡, pre-tax accounting income divided by total assets;

INDUSTRY𝑖𝑡⁵; YEAR𝑖𝑡⁶;

5 INDUSTRYit is a dummy variable coded differently for every industry the firm operates in based on Thomas Reuters Wordscope database’s General Industry Classification: 01 Industrial, 02 Utility, 03 Transportation, 04 Bank/Savings & Loan, 05 Insurance, 06 Other financial

6 YEARit is a dummy variable coded 1 if the observations falls in the specific year category.

(22)

21

4. Empirical results

4.1 Descriptive statistics

Table 2 reports the descriptive statistics of the total matched sample. It provides detailed information on the variables mean, standard deviation, median, minimum and maximum of the 266 firm-year observations across the 2004-2009 period. Looking at the results one can see that the mean of the different CSR scores are all between 65 en 69, with median values between 77 and 80. The minimum and maximal values are far apart with values ranging from 3.76 to 98.79. Important to note is that table 2 also shows that firms in the sample have a leverage mean of 50.5 percent, which entails that they are financed with 50.5 percent debt and 49.5 percent equity. Looking at the minimum and maximum values of leverage one firm

Table 2

Descriptive statistics

This table presents the descriptive statistics of all variables in the main regression, excluding the interaction terms. The data sample consists of 266 firm observations of 140 different companies over the period 2004-2009.

Variables Mean Std.

Deviation Median Minimum Maximum

ECNSCORE 69.71 26.07 78.09 3.76 98.79

ENVSCORE 65.23 31.08 80.45 9.99 97.27

SCOSCORE 67.69 27.11 77.33 3.93 98.59

SIZE 16.92 1.25 16.98 13.29 20.83

LEV 51.5 115.87 37.44 0 1669.37

CAPINT 10.88 14.57 5.73 0.39 115.54

INVINT 0.07 0.07 0.07 0 0.49

MTB 3.58 14.04 2.66 -118.08 102.96

INSIDST 10.79 16.5 1.04 0 84.03

ROA 0.12 0.12 0.11 -0.61 0.84

TAXDit is a dummy variable coded 1 for every year in which a firm was in a tax dispute with federal, state, local or non-US government authorities and 0 otherwise; ECNSCOREit,

EVNSCOREit and SOCSCOREit represent the three different measures of corporate social responsibility on a scale from 0 to 1.0; INTERZit denotes the empirical measure of

internationalization; SIZEit, the natural logarithm of total assets; LEVit, long term debt divided by total assets; CAPINTit, which is net property, plant and equipment divided by total assets; INVINTit, inventory divided by total assets; MTBit, the market value of ordinary equity divided by the balance sheet value of the ordinary equity in the company; INSIDSTit is the percentage of stocks owned by top management; ROAit, pre-tax accounting income divided by total assets

(23)

22 is highly leveraged with a value of 1669.37 percent. However, there is no reason not to include this

variable in the model as it is not identified as an outlier by looking at the studentized residuals (Hair et Al., 2006) nor does the data input seem incorrect⁷. There are no other remarkable anomalies among the variables in table 2.

Table 3 shows the descriptive statistics of the firms that have been in a tax dispute with the government and firms who have not. The sample is equally split with 133 observations for each group. A paired samples t-test was used to determine whether the mean difference between paired observations is statistically significantly different from zero. It shows that the between the two groups in the sample LEV, CAPINT, INVINT, INSIDST are all statistically different from zero (ρ < 0.05). A sign test was also carried out to determine whether the median difference between the matched observations is significantly different from zero. The test shows that the medians of LEV, INVINT and INSIDST are statistically different between the two groups (ρ < 0.05).

7 The firm associated with this leverage is DNB (Dun and Bradstreet) and the data shows high leverage values for DNB in years surrounding the maximum value of 1669.37%.

(24)

23

Table 3

Matched sample descriptive statistics

This table presents the matched sample descriptive statistics of all variables in the main regression, excluding the interaction terms. The data sample consists of 266 firm observations of 140 different companies over the period 2004-2009.

Tax Avoidant TAXD = 0 TAXD = 1

N Mean Std.

Deviation Minimum Median Maximum N Mean Std.

Deviation Minimum Median Maximum

ECNSCORE 133 70.29 26.9 4.74 78.56 98.79 133 69.14 25.3 3.76 77.6 98.68

ENVSCORE 133 65.78 29.68 10.14 77.08 97.27 133 64.68 32.52 9.99 84.79 96.89

SCOSCORE 133 68.01 25.56 3.93 75.92 98.59 133 67.37 28.66 8.57 78.19 97.49

SIZE 133 16.85 1.2 13.29 16.88 20.5 133 16.99 1.29 14.04 17.05 20.83

LEV 133 0.36 0.25 0 0.34 1.21 133 0.67 1.61 0 0.4 16.69

CAPINT 133 0.08 0.08 0.01 0.04 0.42 133 0.14 0.19 0 0.08 1.16

INVINT 133 0.08 0.07 0 0.08 0.27 133 0.06 0.07 0 0.05 0.49

MTB 133 2.56 11.25 -118.08 3.07 24.6 133 4.6 16.34 -51.74 2.18 102.96

INSIDST 133 0.13 0.16 0 0.05 0.76 133 0.08 0.16 0 0.01 0.84

ROA 133 0.11 0.08 -0.22 0.1 0.41 133 0.13 0.15 -0.61 0.11 0.84

See Table 2 for variable definitions

(25)

24 4.2 Correlation results

Table 4

Correlation matrix and multicollinearity statistics

This table provides the correlations among the dependent and independent variables as well as the multicollinearit y (VIF) of each variable.

Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 VIF

1 TAXD 1.000

2 ECNSCORE -0.022 1.000 1.687

3 ENVSCORE -0.018 .489** 1.000 1.188

4 SCOSCORE -0.012 .661** .766** 1.000 1.198

5 INTERZ_ECN 0.056 -0.103 0.007 -0.062 1.000 1.240

6 INTERZ_ENV 0.090 0.005 -0.070 -.124* .513** 1.000 1.094

7 INTERZ_SOC .178** -0.070 -.124* -.213** .690** .727** 1.000 1.136

8 SIZE 0.057 .339** .504** .455** 0.039 -0.039 -0.017 1.000 1.400

9 LEV .131* -.230** -.197** -.186** 0.066 0.043 0.039 -.248** 1.000 2.146

10 CAPINT .223** -.146* -.174** -.276** -0.019 0.083 0.108 -0.015 -0.082 1.000 2.804 11 INVINT -.158** .133* .135* .238** -0.088 -0.114 -0.104 -0.067 -0.112 -.192** 1.000 4.035 12 MTB 0.073 -0.030 -0.046 -0.046 0.003 -0.004 0.003 -0.105 -0.093 -0.046 -0.017 1.000 2.167 13 INSIDST -.149* -0.084 -0.105 -.136* -.128* -.164** -.191** -.223** 0.035 -0.048 0.000 0.044 1.000 2.283 14 ROA 0.085 .161** -.126* -0.031 -0.012 0.015 0.007 -.233** .156* -.173** -0.078 .222** -0.006 1.000 3.325 Notes: Sample size: N=266, See Table 2 for variable definitions

**. Correlation is significant at the 0.01 level (2-tailed).

*. Correlation is significant at the 0.05 level (2-tailed).

(26)

25 Table 4 shows the Pearson correlation results which displays that tax avoidance is not significantly correlated with either the economic (ECNSCORE), environmental (ENVSCORE) or social (SOCSCORE) scores of CSR. This indicates that there is no significant association between tax avoidance and these CSR scores. The results do show that tax avoidance is significantly correlated with the interaction term of internationalization and social CSR activities (ρ < 0,01). Tax avoidance is positively correlated with leverage (LEV) (ρ < 0,05) and negatively correlated with inventory intensity (INVINT) (ρ < 0,01), capital intensity (CAPINT) (ρ < 0,01) and management stock ownership (INSIDST) (ρ < 0,05). It should be noted that some of the other explanatory variables are also correlated between them. The highest correlation reported is between environmental CSR (ENVSCORE) and social CSR (SOCSCORE) with a Peason correlation of 0,766 (p >0,01). Therefore, it is important to test for multicollinearity by calculating variance inflation factors (VIF). The tests show that no single VIF was greater than 4,108 (VIF < 5), thus the effects of multicollinearity are not significantly enough to affect the binominal model (Hair et Al., 2006).

4.3 Logit regression results

First, tax avoidance is regressed on the control variables which have been associated with tax avoidance in previous studies. The second step was to consecutively add the three different CSR variables to the regression to test H1, H2 and H3. The last step involved bringing the interaction terms between the three CSR variables and internationalization into the analysis to test H4a-c. The assumptions underlying a standard multiple regression, such as independence of observations, homoscedasticity and normality do not adhere to a binominal logit regression. However, the assumptions regarding multicollinearity, linearity and outliers have to be considered in a logit regression. It has been shown that there is no multicollinearity between the continuous independent variables⁸. Linearity was assessed by the Box-Tidwell (1962) procedure. A Bonferroni correction was applied by dividing the alpha (α) by the number of 23 terms in the model which resulted in the linearity assumption to be rejected when ρ < .00217 (Tabachnick & Fidell, 2007). The results show that all continuous independent variables were found to be linearly related to the logit of the dependent variable⁹

8 See 4.2 Correlation results – Table (4)

9 See 7.3 Appendix 3 – Box Tidwell linearity test

(27)

26 Lastly, the sample has been checked for outliers and three observations with their corresponding match have been omitted from the regression¹⁰

A binomial logistic regression was performed to ascertain the effects of corporate social responsibility and the moderation effect of internationalization on the likelihood of organizational tax avoidance. The logistic regression model was statistically significant, χ²=

67.221, ρ < .0005 with 21 degrees of freedom. The model explained 29.77% (Nagelkerke R2) of the variance in tax avoidance and correctly classified 68,0% of cases.

Model 1 includes all the control variables and serves a comparison to the other models. The results show that the control variables size (SIZE), leverage (LEV), capital intensity (CAPINT) and managerial stockownership (INSIDST) are statistically significant (ρ

< 0.05), which is in line with previous research. Size is positively associated with tax avoidance, which supports previous studies suggesting that larger firms are more tax avoidant as they have greater political and economic power (Zimmerman, 1983; Lanis and Richardson, 2015). Leverage is also positively associated with tax avoidance, which is supports the idea that more leveraged firms use tax deductible interest payments to avoid taxes. Capital intensity is also positively associated with tax avoidance, which supports the study by Stickney and McGee (1983) who argue that firms with large capital assets are more likely to avoid taxes. Both control variables are both very powerful predictors with Wald (LEV)

=10.104 and Wald(CAPINT) = 15.662. Model 2 only explains 0.02% (R²) more variance than offered by model 1. H1, which predicted that economic CSR activities would be negatively related to tax avoidance, is therefore not supported (B = 0.001, ρ > 0.05). Model 3 tries to identify if environmental CSR activities are positively related to tax avoidance. The addition of ENVSCORE to the model only added 0.04% to the overall R² and the coefficient (B = 0.001) is not significant (ρ > 0.05), thus H2 is not supported. Model 4 sees the addition of social CSR. The unstandardized coefficient and Wald-statistic are larger than the other two CSR variables it does only adds 0.47% to the explained variance of the model. Therefore, H3, which predicted that social CSR activities would be negatively related to tax avoidance is not supported (B = 0.011, ρ > 0.05). With models 5-7 the main effect of the different CSR components and the moderating effect of internationalization are added to the model. Model 5 test H4a, which predicts that the relationship between tax avoidance and economic CSR activities is moderated by a firm’s internationalization. However, this hypothesis is not supported as the model is able to only slightly improve the overall predictability of the model.

10 See 3.2 Matched sample

(28)

27 The moderating effect of internationalization on environmental CSR is added under model 6.

The results are not significant and thus H4b cannot be supported. The last tested model includes the moderating variable of internationalization on social CSR activities. The regression results show a sharp increase in R² of 4.64% compared to the addition of the other antecedent variables in model 6 and a 5.56% increase of variance explained than offered by the control variables in model 1. Furthermore, the results show that model 7 is statistically significant (B = 0.001, ρ > 0.05) and thus provides support for H4c, which predicts that the relationship between tax avoidance and social CSR activities is moderated by a firm’s internationalization. Thus, showing that social CSR activities have a more positive effect on tax avoidance when firms are more internationalized. The addition of this interaction term does also change the unstandardized coefficients and t-statistics of the other independent variables. Some with such a degree that they become significant, such as the score reflecting social CSR activities (SOCSCORE).

Table 5 shows the binominal logit regression model and the different models tested.

Model 1 is used to test for the effects of the control variables on tax avoidance. Models 2-4 pertain to the three different CSR activities, economic, environmental and social. Models 5-7 include the interaction terms between the different CSR activities and internationalization.

(29)

28

Table 5

Binominal Logit Regression

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

SIZE .257 .246 .231 .214 .212 .213 .193

(3.998)** (3.251)* (2.456)* (2.0933) -2.030 (2.046) -1.624

LEV .019 .019 .019 .019 .019 .019 .019

(10.454)*** (10.519)*** (10.572)*** (10.465)*** (10.515)*** (10.510)*** (11.097)***

CAPINT .066 .066 .066 .068 .068 .068 .072

(11.617)*** (11.722)*** (11.881)*** (12.531)*** (12.884)*** (12.988)*** (14.683)***

INVINT -2.997 -3.042 -3.075 -3.407 -3.325 -3.127 -4.225

(1.839) (1.881) (1.916) (2.273) (2.169) (1.893) (3.041)*

MTB .022 .023 .023 .022 .023 .023 .022

(3.127)* (3.164)* (3.124)* (3.219)* (3.292)* (3.252)* (3.447)*

INSIDST -.017 -.017 -.017 -.017 -.016 -.015 -.010

(3.893)** (3.923)** (3.931)** (3.575)* (3.287)* (2.849)* -1.259

ROA .017 .016 .016 .018 .018 .019 .019

(1.264) (1.048) (1.093) -1.296 -1.240 -1.348 -1.387

ECNSCORE .001 .001 -.003 -.002 -.003 -.009

(0.048) (.011) (0.156) (0.084) (0.152) (1.144)

ENVSCORE .002 -.003 -.003 -.003 -.004

(0.078) (0.144) -0.173 (0.181) (0.325)

SCOSCORE .010 .010 .011 .024

(0.950) (0.988) (1.194) (3.936)**

INTERZ_ECN .000 .000 -.001

(0.482) (0.079) (3.204)*

INTERZ_ENV .000 .000

-.407 -1.626

INTER_SOC .001

(10.845)**

Constant -7.120 -7.023 -6.807 -6.465 -6.594 -6.657 -7.156

(6.946)*** (6.576)*** (5.741)*** (5.136)*** (5.272)*** (5.339)*** (5.955)***

YEAR No No No No No No No

INDUSTRY No No No No No No No

R² 24.21% 24.23% 24.26% 24.66% 24.85% 25.02% 29.77%

Chi-square (53.294)*** (53.343)*** (53.421)*** (54.388)*** (54.871)*** (55.279)*** (67.221)***

Hosmer and

Lemeshow test .521 .280 .408 .505 .322 .589 .904

N 266 266 266 266 266 266 266

See Table 2 for variable definitions Note: Wald statistic in parentheses

* significance at the 0.1 percent level

** significance at the 0.05 percent level

*** significance at the 0.01 percent level

(30)

29 4.4 Robustness tests

In this study a direct measure of tax avoidance is used by using tax disputes as the qualitative dependent variable. However, the measurement difficulty of tax avoidance leads to many different methods and discussions on what methods to use (Wilson, 2009; Lisowsky, 2010;

Hanlon and Heitzman, 2010). To add to this discussion and test if the results are generalizable across multiple methods a robustness test is performed with two different measures of tax avoidance based on book-tax differences.

As mentioned above the most prevalent proxy measures of tax avoidance are the effective tax rate or the book-tax difference. For this robustness test the book-tax difference (BTD) is used as a proxy measure of tax avoidance as prior research has identified that BTDs can more accurately estimate the level of tax avoidance than the previously mentioned effective tax rate methods (Wilson, 2009; Lisowsky, 2010). As the name suggests, BTDs measure the difference between the book income and taxable income, wherein a higher reported book income than reported taxable income results in a deferred tax liability.

Research by Wilson (2009) shows that firms with larger book-tax differences are more likely of being accused of tax avoidance. Following previous research two different BTD methods are used (Lanis and Richardson, 2015). The first proxy measure of tax avoidance is the total book-tax difference (TOTALBTD) measured as the pre-tax accounting income minus the income tax expense divided by the corporate tax rate¹¹ and scaled by total assets. The second measure is based on book-tax differences residuals (RSDBTD). This method was created to remove earning management activities from the proxy measure of tax avoidance based on BTD. Desai and Dharmapala (2006) argued that BTD also reflect earnings management activities with no link to tax avoidance. The method uses total accruals (TOTALAC) as a measure of earnings management activities, measured as follows:

𝑇𝑂𝑇𝐴𝐿𝐴𝐶𝑖𝑡 = 𝐸𝐵𝐼𝑇𝑖𝑡 − 𝐶𝐹𝑂𝑖𝑡 (2)

,where; EBIT represents the pre-tax accounting income, CFO the cash flow from operations,𝑖 the specific firm and 𝑡 the corresponding year. The total accruals are scaled by lagged total assets.

11 As all companies in the sample are based in the United States, thus the corporate statutory tax rate is used to calculate the taxable income is 35% (OECD, 2016. Retrieved from:

https://stats.oecd.org/Index.aspx?DataSetCode=TABLE_II1)