Is Corporate Social Responsibility Performance Associated with Tax Avoidance?

In line with existing empirical business research (e.g., Rego 2003; Frank et al. 2009; Chen et al. 2010), we define corporate tax avoidance as the downward management of taxable income through tax-planning activities. We specifically define a tax avoidant firm as one that has had a tax dispute involving federal, state, local or non-U.S. government authorities, or was involved in a controversy over its tax obligations which raised public concern during the period (MSCI 2012). Hence, tax avoidance may include tax-planning activities that are legal or that may fall into the gray area. This differentiates tax avoidance from tax evasion which only relates to illegal activities.

The KLD/MSCI STATS database includes over 90 indicators in seven stakeholder or social issue areas, including: community relations, corporate governance, diversity, employee relations, environment, human rights, and issues related to firm’s products and core business. The data set also includes over 20 indicators in the following controversial business issues: alcohol, firearms, gambling, military, nuclear power, and tobacco. All indicators are rated positive, negative or neutral. STATS data is published once at the end of each calendar year. The data is a snapshot of a firm’s social and environmental performance at that moment in time. It includes observations for each year (beginning with 1991) and provides a table of data with a collection of around 650 firms that comprise the FTSE KLD 400 Social Index and S&P 500 with one record for each firm. Beginning in 2001, KLD expanded its coverage to include the largest 1,000 US firms by market capitalization. In 2003, KLD again expanded the coverage to the largest 3,000 US firms by market capitalization (MSCI 2012).

As a sensitivity analysis of our main results, we also employ several indirect proxy measures of tax avoidance based on book-tax differences (BTDs) in our study. Prior research finds that BTDs are a more precise measure of tax avoidance compared with ETRs (see, e.g., Frank et al. 2009; Wilson 2009; Lisowsky 2010).

Huseynov and Klamm (2012) used only three (from a possible seven) components (measures) of CSR performance from the KLD database: (1) corporate governance; (2) community; and (3) diversity.

For instance, community and political involvement, environmental protection, social and community development and investment, promoting staff welfare and development, and having policies in place that maintain a good relationship with customers, suppliers and government bodies (Lanis and Richardson 2012).

The CEC (2001, p. 4) claims that CSR: “is essentially a concept whereby companies decide voluntarily to contribute to a better society and a cleaner environment.” Hence, there should be wide variation in the CSR performance of firms, which logically leads researchers to ask whether such variation impacts other corporate attributes and policies.

See the widely reported cases in the world media of Apple, Google and Starbucks, for example.

Specifically, Google cut its worldwide income tax bill by approximately $3.1 billion over 3 years using the “Double Irish Dutch Sandwich” technique which moved profits through units in Ireland, the Netherlands and Bermuda. This tax-cutting strategy helped cut the firm’s income-tax rate to around 2.4 percent on the profits it attributed to its foreign subsidiaries during the 3-year period. However, the statutory corporate income tax rate in the US is 35 percent (Womack and Drucker 2011).

The KLD data are only available up to 2009 (at the time of the research study) and certain research control variables were not available prior to 2003. Hence, the sample period of 2003–2009 was selected.

Although our comparison sample’s cut-off point of ±40 percent appears to be large, it is consistent with Kaplan and Reishus (1990). In addition, most of the tax-avoidant firms and non-tax-avoidant firms in the sample are similar within a range of ±40 percent. As the mean market value of common stock of the tax-avoidant firms in our sample is $34.2 million, the related matched firm size may range from $17.1 to $51.3 million. There is no reason to believe that such a range has a significant impact on our results.

Moreover, Beasley (1996) also used a direct measure of accounting fraud in his research by studying firms that were in dispute with the Securities Exchange Commission (SEC) over restatements of financials.

Callan and Thomas (2009) provide detailed discussion of the various ways in which KLD data are used in CSR-related research. They recommend a simple summation or averaging method instead of the more arbitrary weighting methods used in some studies.

We note that our corporate governance control variables are consistent with those used in other accounting research (e.g., Beasley 1996). They are different to the corporate governance ratings in the KLD index itself, which consists of limited compensation, ownership strength (which refers to ownership of firms in which KLD has cited as having an area of social strength), transparency (which relates to reporting on a wide range of social and environmental issues), political accountability, and other issues (e.g., the culture of the firm). Thus, the KLD index ratings relating to corporate governance are different to our corporate governance control variables.

According to Hair et al. (2006), a correlation coefficient for a pair of explanatory variables between ±0.25 and ±0.75 indicates a moderate level of collinearity between the two variables.

Hair et al. (2006) suggest that a VIF value above the threshold of ten corresponds to a high level of multi-collinearity amongst the explanatory variables.

For instance, depreciation expense can lead to temporary BTDs, whereas R&D tax credits can result in permanent BTDs.

A description of the method developed by Desai and Dharmapala (2006) for calculating the BTDs residual is provided in “Appendix”.

Industry-sector (INDSEC) dummy variables defined at the two-digit GICS code level are also included as control variables in our OLS regression models as it is possible for tax avoidance to fluctuate across different industry sectors (Omer et al. 1993). We include nine INDSEC dummy variables in our OLS regression models: energy; materials; industries; consumer discretionary; consumer staples; health care; information technology; telecommunications; and utilities (omitted sector). No sign predictions are made for the INDSEC dummies.