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:
1
Table of Contents
Table of Contents ... 1 List of Tables ... 2 List of Abbreviations ... 3 1. Introduction ... 4 2. Literature Review ... 72.1 The ethics of tax avoidance ... 7
2.2 Tax avoidance as a CSR component... 8
2.3 Economic, environmental and social CSR activities ... 9
2.4 Internationalization, CSR and tax avoidance...12
3. Research Design ...14 3.1 Research data ...14 3.2 Matched sample ...14 3.3 Dependent variable...16 3.4 Independent variables ...17 3.5 Moderation terms ...18 3.6 Control variables ...18 3.7 Regression model ...19 4. Empirical results ...21 4.1 Descriptive statistics ...21 4.2 Correlation results ...24
4.3 Logit regression results ...25
4.4 Robustness tests ...29
5. Discussion and conclusion ...35
6. List of References ...38
7. Appendices...43
7.1 Appendix 1 – Paired samples t-test...43
7.2 Appendix 2 – Sign test...45
2
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
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
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).
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.
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).
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
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.
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).
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
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
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
14
3. Research Design
3.1 Research dataThe 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 30th 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).
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.
16 3.3 Dependent variable
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
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
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.
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
21
4. Empirical results
4.1 Descriptive statisticsTable 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 Deviation Median Minimum Maximum Std.
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
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).
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 Deviation Std. 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
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
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)
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.
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).
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
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
30 After the total accruals have been calculated the linear regression is performed with TOTALBTD as the dependent variable and TOTALAC as the independent variable.
𝑇𝑂𝑇𝐴𝐿𝐵𝑇𝐷𝑖𝑡 = 𝑇𝑂𝑇𝐴𝐿𝐴𝐶𝑖𝑡 + 𝜇𝑖𝑡 + 𝜀𝑖𝑡 (3)
Where, 𝜇 is the residual value, 𝜀 is the error term, 𝑖 the specific firm and 𝑡 the corresponding year. The residual value 𝜇 represent the level of tax avoidance and is saved and coded as RSDBTD.
Both proxy measure of tax avoidance are used as the dependent variable in a linear regression. The preformed regression is assessed for first order autocorrelation and shows no signs of dependence of the residuals in both regressions (DW(TOTBTD) = 2.066; DW(RSDBTD) =
2.083). Linearity was assessed by inspecting the partial regression plots between the dependent and independent variables and found that all independent variables are likely to be linear. There was no evidence of multicollinearity, as assessed by VIF values which are not greater than SOCSCORE’s VIF of 4.035 (Hair et Al., 2006).12 Normality of the residuals was assessed by generating a Q-Q plot of the studentized residuals showing potential non-normality. However, as our number of observations is above N=30 the Central Limit Theorem comes into effect one can assume that the means are normally distributed (Brooks, 2014). Finally, the linear regression is statistically significantly for all different models tested with the dependent variable being either TOTALBTD or RSDBTD. The results of the robustness test can be found in tables 6 and 7.
Table 6 reports the linear regression results with the total book-tax difference as the dependent variable. The linear regression model is statistically significant, with F(15,250) = 2.594, ρ < 0.05, with all models adding significantly to the linear regression. The tests show that the model explains very little of the variation in tax avoidance (R²=13,47%) of which the independent variables and interaction terms only add 2.53% on top of the variance explain by the control variables. The results from the robustness test show that social CSR activities are negatively related to tax avoidance when using the total book-tax difference as a proxy measure under the model 4,5 and 6. When the interaction term between the social CSR score and internationalization is added to the model this variable loses it’s statistical significance. This result is not in line with the binominal logit regression, which did show a significant relationship between social CSR and tax avoidance. The results further show that the control variable SIZE is positively associated with tax avoidance (ρ<0.05) Furthermore, the results
31 suggest that MTB is negatively associated with TOTALBTD. Both of these results have not been found in the binominal logit regression and are strong predictors of tax avoidance under the TOTBTD regression model.
Table 7 reports the linear regression results with the residual book-tax difference as the dependent variable. The linear regression model is statistically significant, with F(15,265) = 2.603, ρ < 0.05, with all models adding significantly to the linear regression. The tests show that the model explains very little variance of tax avoidance and just slightly more than for the total book-tax differences regression (R²=13.51%). Consistent with the TOTBTD regression the RSDBTD also did not find any statistically significant relationships between CSR activities and tax avoidance. The results do show that SOCSCORE has a stronger association than when TOTALBTD is the dependent variable, but the result remains insignificant. Furthermore, the results do show that SIZE and MTB are statistically significantly (ρ <0.05). These results have also not been found in the binominal logit regression.
32
33
OLS Regression - TOTBTD
Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
SIZE 0.01 0.02 0.02 0.02 0.02 0.02 0.02 (1.864)* (1.969)* (1.964)* (2.224)** (2.231)** (2.227)** (2.174)** LEV 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (-0.003) (-0.125) (-0.132) (-0.03) (0.017) (0.016) (0.047) CAPINT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (0.211) (0.155) (0.093) (-0.325) (-0.373) (-0.374) (-0.44) INVINT 0.06 0.07 0.08 0.11 0.11 0.11 0.10 (0.464) (0.538) (0.562) (0.833) (0.798) (0.797) (0.711) MTB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (-3.897)*** (-3.931)*** (-3.904)*** (-3.909)*** (-3.907)*** (-3.900)*** (-3.912)*** INSIDST 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (-0.975) (-0.969) (-0.973) (-1.124) (-1.242) (-1.218) (-1.059) ROA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 (-0.573) (-0.383) (-0.427) (-0.606) (-0.594) (-0.587) (-0.616) ECNSCORE 0.00 0.00 0.00 0.00 0.00 0.00 (-0.645) (-0.453) (0.589) (0.51) (0.487) (0.338) ENVSCORE 0.00 0.00 0.00 0.00 0.00 (-0.389) (0.945) (1.024) (1.022) (1.05) SCOSCORE 0.00 0.00 0.00 0.00 (-2.085)** (-2.126)** (-2.085)** (-1.841)* INTERZ_ECN 0.00 0.00 0.00 (-0.986) (-0.856) (-1.37) INTERZ_ENV 0.00 0.00 (0.053) (-0.628) INTER_SOC 0.00 (1.194) YEAR 0.01 0.01 0.01 0.01 0.01 0.01 0.01 (1.747)* (1.768)* (1.807)* (1.818)* (1.688)* (1.646) (1.575) INDUSTRY 0.00 0.00 0.00 -0.01 0.00 0.00 -0.01 (0.067) (-0.002) (-0.017) (-0.457) (-0.361) (-0.346) (-0.446) Constant -21.94 -22.26 -23.42 -23.42 -21.92 -21.79 -20.88 (-1.765)* (-1.787)* (-1.826)* (-1.838)* (-1.708)* (-1.666)* (-1.594) R² 10.94% 11.08% 11.14% 12.64% 12.97% 12.98% 13.47% F-statistic (3.493)*** (3.178)*** (2.893)*** (3.05)*** (2.89)*** (2.673)*** (2.594)*** N 266 266 266 266 266 266 266
See Table 2 for variable definitions Note: t-statistic in parentheses * significance at the 0.1 percent level ** significance at the 0.05 percent level *** significance at the 0.01 percent level
34 Table 7 shows the results from the residual book-tax differences OLS regression model. 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.
OLS Regression - RSDBTD
Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
SIZE 0.108 0.129 0.143 0.161 0.162 0.162 0.158 (1.884)* (2.103)** (2.141)** (2.413)** (2.432)** (2.428)** (2.373)** LEV 0.000 0.000 0.000 0.000 0.000 0.000 0.000 (0.13) (-0.053) (-0.062) (0.044) (0.119) (0.118) (0.149) CAPINT 0.000 0.000 -0.001 -0.003 -0.003 -0.003 -0.003 (0.048) (-0.032) (-0.113) (-0.547) (-0.625) (-0.626) (-0.693) INVINT 0.236 0.353 0.389 0.679 0.622 0.629 0.539 (0.234) (0.348) (0.383) (0.667) (0.613) (0.615) (0.527) MTB -0.017 -0.018 -0.017 -0.017 -0.017 -0.017 -0.017 (-3.597)*** (-3.659)*** (-3.629)*** (-3.634)*** (-3.641)*** (-3.635)*** (-3.647)*** INSIDST -0.006 -0.005 -0.006 -0.006 -0.007 -0.007 -0.006 (-1.362) (-1.354) (-1.36) (-1.52) (-1.712)* (-1.68)* (-1.513) ROA -0.004 -0.002 -0.002 -0.004 -0.004 -0.003 -0.004 (-0.594) (-0.322) (-0.383) (-0.57) (-0.552) (-0.545) (-0.575) ECNSCORE -0.003 -0.002 0.001 0.001 0.001 0.000 (-0.958) (-0.693) (0.42) (0.298) (0.277) (0.125) ENVSCORE -0.001 0.003 0.004 0.004 0.004 (-0.525) (0.89) (1.019) (1.018) (1.046) SCOSCORE -0.010 -0.010 -0.010 -0.009 (-2.176)** (-2.247)** (-2.202)** (-1.95)* INTERZ_ECN 0.000 0.000 0.000 (-1.555) (-1.34) (-1.796)* INTERZ_ENV 0.000 0.000 (0.066) (-0.638) INTER_SOC 0.000 (1.231) YEAR 0.067 0.068 0.074 0.074 0.065 0.065 0.061 (1.437) (1.472) (1.551) (1.561) (1.37) (1.332) (1.259) INDUSTRY 0.006 -0.003 -0.005 -0.046 -0.033 -0.031 -0.041 (0.07) (-0.032) (-0.054) (-0.512) (-0.363) (-0.345) (-0.449) Constant -135.719 -139.276 -151.044 -151.040 -133.360 -132.165 -125.133 (-1.454) (-1.491) (-1.57) (-1.582) (-1.391) (-1.352) (-1.279) R² 10.08% 10.40% 10.50% 12.15% 12.98% 12.98% 13.51% F-statistic (3.189)*** (2.961)*** (2.709)*** (2.915)*** (2.892)*** (2.675)*** (2.603)*** N 266 266 266 266 266 266 266
35
5. Discussion and conclusion
This study aimed to find whether different types of CSR activities are associated with tax avoidance. A novel direct measurement of tax avoidance was used based on previous research by Lanis and Richardson (2015). The logit regression results, based on a sample of the S&P500, showed that firms who engage in more social CSR activities are more likely to avoid taxes. The results also showed that internationalization moderates the relationship between social CSR activities and tax avoidance.
First, the binominal logit regression results showed that social CSR activities are positively associated with tax avoidance. This supports the third hypothesis and the risk management theory on CSR, which argues that CSR is being used to lift a firm’s positive reputation and thereby mitigating negative corporate publicity and events. However, under this theory it is surprising to find that environmental CSR is not associated with tax avoidance in similar vein. It could be that social CSR activities are more visible to the public than environmental CSR as social CSR activities focus on the workforce, consumers and society. Thus, corporate reputational cost can be more effectively combatted by engaging in social CSR activities, resulting in a significant positive association between social CSR and tax avoidance. The results also demonstrate that internationalization moderates the relationship between tax avoidance and social CSR. However, the results from the binominal logit regression do not indicate that the relationship between economic and environmental CSR is contingent on a firm’s internationalization. One possible explanation is that international governmental agencies such as the UN or OECD put less emphasis on economic and environmental policies, thus not finding larger differences between national and international organizations on these CSR scores. A second explanation could be that economic and environmental CSR policies are easier to transfer across borders, whereas social CSR might be more country specific. However, the predictive capability of the three types of CSR is rather low13 as was also found in previous studies (Lanis and Richardson, 2015). Hence, the concept CSR might not be a good predictor for tax avoidance in general.
The robustness test, based proxies of tax avoidance, is not consistent with the binominal logit regression’s results when examining the independent variables. The logit regression results showed that social CSR is positively associated with tax avoidance, while the OLS regressions did find a statistically significant relationship under models 4-6, when
36 the interaction term of social CSR and internationalization was added the variable became insignificant. Moreover, the OLS regression results do not show that internationalization has a moderating effect on the relationship between social CSR and tax avoidance. Furthermore, there are large differences when examining the control variables. Unlike the main regression the OLS regressions do show that firm size and a firm’s market-to-book ratio are statistically significant with tax avoidance. Furthermore, the robustness tests do not find a significant relationship between leverage and capital intensity. Even though, previous studies have shown that BTD and tax shelters or tax disputes as measurement results in comparable results (Graham and Tucker, 2006; Lanis and Richardson, 2015). However, one must not be too brusquely to dismiss the results from the main regression. One must take into account the completely different method used to measure tax avoidance and the subsequent different types of analysis. Both values stem from a completely different source as opposed to ETRs and BTDs for example. Furthermore, tax disputes are a direct dichotomous measure of tax avoidance whereas BTD are measured on a continuous scale. Which entails that that the analysis is done with either a binominal logit regression or an OLS regression. Both methods are very different from a statistical and computational standpoint. Thus, for example, a larger sample size would not alleviate the differences between the logit and OLS regressions as the dependent variable is completely different. Neither would dividing the BTDs variables into a tax avoidant and tax compliant category result in effects similar to that of the binominal logit regression. It is important to note that both approaches have its shortcomings. The direct measurement is be a less developed approach and you lose the continuous measurement of tax avoidance. However, it provides a more clear distinction between tax avoidant and tax compliant firms as current proxy measures cannot distinguish between tax compliant firms with low ETRs or high BTD and tax avoidant firms. Thus, tax disputes are a less arbitrary approach to the measurement of tax avoidance and ensures that the sample consist of accurately identified tax avoiding firms. Therefore, the difference in results between the binominal logit regression and the robustness test stems from the difference in measurement of the dependent variable and is not attributable to errors in the model.
37 multiple causes; First, the firms in the sample were taken from the S&P 500 index which limits the amount of observations drastically and introduces selection bias. It does provide insights for large multinationals in the United States, but might not be representative of the population. Second, due to difficulty with data availability it was not possible to include some variables in the model which have shown to be significant in previous studies. Third, the sample consists of strongly internationalized firms with a mean of 35.1 percent of foreign sales over total sales and is therefore most likely not representative of the total country firm internationalization sample. Furthermore, the direct measure of tax disputes is subject to a selection bias as with tax disputes it is likely that only firms are selected on the upper level of the tax avoidance scale. These firms are taking more risk with their tax planning strategies and are therefore more likely to be under governmental scrutiny. This leaves a more polarized matched sample than a proxy measure of tax avoidance would. A second problem which arises when using tax disputes is that firms which are able to better conceal their tax avoidance will not be investigated by tax authorities and thus not end up in the sample. This referred to as an adverse selection bias or opposite survivorship bias and adds potential problems. For example, as was pointed out larger firms have more political and economic power and are therefore more likely to avoid taxes (Zimmerman, 1983). Due to this research design these larger firms would not end up in the sample and thus could change the regression results.
38
6. List of References
Aid, C., 2009. False profits: robbing the poor to keep the rich tax-free. A Christian Aid Report.
Armstrong, C. S., Blouin, J. L., Larcker, D. F., 2012. The incentives for tax planning. Journal of Accounting and Economics 53(1), 391-411.
Attig, N., Boubakri, N., El Ghoul, S., Guedhami, O., 2016. Firm internationalization and corporate social responsibility. Journal of Business Ethics 134(2), 171-197.
Avi-Yonah, R. S., 2008. Corporate social responsibility and strategic tax behavior. In: Tax and corporate governance. Springer Berlin Heidelberg, pp. 183-198.
Babiak, K., Trendafilova, S., 2011. CSR and environmental responsibility: motives and pressures to adopt green management practices. Corporate social responsibility and environmental management 18(1), 11-24.
Liang, H., Marquis, C., Renneboog, L., & Sun, S. L., 2014. Speaking of corporate social responsibility. Harvard Business School Organizational Behavior Unit Working Paper,
(14-082).
Berger, P. G., 1993. Explicit and implicit tax effects of the R & D tax credit. Journal of Accounting Research, 131-171.
Ballas, A. A., Tsoukas, H., 1998. Consequences of distrust: the vicious circle of tax evasion in Greece. Journal of Accounting, Ethics & Public Policy 1(4), 572-596.
Vonwil, M., & Wreschniok, R., 2009. The CSR myth: true beauty comes from within. In Reputation Capital (pp. 83-99). Springer Berlin Heidelberg.
Bowen, H. R., Johnson, F. E., 1953. Social responsibility of the businessman. Harper. Brooks, C. 2014. Introductory econometrics for finance. Cambridge university press.
Carroll, A. B., 1999. Corporate social responsibility evolution of a definitional construct. Business & society, 38(3), 268-295.
Carroll, A. B., 2006. Corporate social responsibility: A historical perspective. In: Epstein, M., Hanson, K. O. (Eds.), The Accountable Corporation Vol. 3: Corporate Social Responsibility. Praeger Publishers, Westport CT, pp. 3-30.
39 Cheng, B., Ioannou, I., Serafeim, G., 2014. Corporate social responsibility and access to
finance. Strategic Management Journal, 35(1), 1-23.
Christensen, J., Murphy, R., 2004. The social irresponsibility of corporate tax avoidance: Taking CSR to the bottom line. Development 47(3), 37-44.
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S., 2003. Alternative Regression Models: Logistic, Poisson Regression, and the Generalized Linear Model. Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). London.
Crocker, K. J., Slemrod, J., 2005. Corporate tax evasion with agency costs. Journal of Public Economics 89(9), 1593-1610.
Crowe, M. T., 1944. The moral obligation of paying just taxes (No. 84). Catholic University of America Press.
Davis, A. K., Guenther, D. A., Krull, L. K., Williams, B. M., 2013. Taxes and corporate sustainability reporting: Is paying taxes viewed as socially responsible.
Deepa Gokulsing, R., 2011. CSR matters in the development of Mauritius. Social Responsibility Journal 7(2), 218-233.
Desai, M. A., Dharmapala, D., 2006. Corporate tax avoidance and high-powered incentives. Journal of Financial Economics 79(1), 145-179.
Dharmapala, Dhammika, 2014, “What Do We Know About Base Erosion and Profit Shifting? A Review of the Empirical Literature,” Illinois Public Law and Legal Theory Research Papers Series No. 14-23 (Champaign: University of Illinois College of Law).
Dillard, J., Murray, A., 2013. Deciphering the domain of corporate social responsibility. In: Haynes, K., Alan, M., Dillard, J. (Eds.), Corporate Social Responsibility: A research handbook. Routledge, New York, pp. 10-27.
Frank, M. M., Lynch, L. J., Rego, S. O., 2009. Tax reporting aggressiveness and its relation to aggressive financial reporting. The Accounting Review 84(2), 467-496.
Freedman, J., 2003. Tax and corporate responsibility. Tax Journal 695(2), 1-4.
Friedman, M., 1970. The social responsibility of business is to increase its profits. New York, 122-124.
