Cross-Country Evidence on the Role of Independent Media in Constraining Corporate Tax Aggressiveness

These include but are not limited to editorials in leading news outlets such as Bloomberg’s “The Great Corporate Tax Dodge,” the New York Times’ “But Nobody Pays That,” The Times’ “Secrets of Tax Avoiders,” and the Guardian’s “Tax Gap.”

For example, a recent report by Confederation of Indian Industry (CII) and global services firm Ernst & Young (EY) indicates that countries like India and China are looking at tax base erosion.

In this paper, we follow Hanlon and Heitzman (2010) and define tax avoidance broadly as the reduction in explicit taxes paid. Hanlon and Heitzman (2010) state that “if tax avoidance represents a continuum of tax planning strategies where something like municipal bond investments are at one end (lower explicit tax, perfectly legal), then terms such as “noncompliance,” “evasion,” “aggressiveness,” and “sheltering” would be closer to the other end of the continuum. Therefore, we define tax aggressiveness as tax planning strategies at the more aggressive end of the tax avoidance spectrum that are more likely to push the envelope of tax law and to include the more aggressive tax-related activities that the media presumably are more concerned about in their coverage.

Although state-owned media may have higher incentives to expose tax avoidance practices, prior literature shows that state ownership of media is associated with higher corruption, thus reducing or completely reversing these incentives (Houston et al. 2011).

Within two days of announcement of the merger between Pfizer and Allergen, The New York Times carried an editorial headlined, “Pfizer’s Big Tax Avoidance Play” (The New York Times, November 24, 2015). In it, the editorial states, “This merger is a tax-dodging maneuver that enriches shareholders and executives while short changing the public and robbing the Treasury …” This is an example of tax aggressive activities becoming headline grabbing newsworthy events in major media outlets.

Potential savings from aggressive tax avoidance strategies can be economically large (Scholes et al. 2014). For example, Bloomberg News reports that Google avoided $2 billion in worldwide income taxes in 2011 by channeling $10 billion of revenue into a Bermuda shell company.

Dowling (2014) characterizes the fair share of tax as the statutory tax rate times a reasonable estimate of the firm’s taxable profits (that is, tax base).

Anecdotal and empirical evidence indicates that the direct costs alone can be quite substantial. For example, in 14 cases of tax sheltering, Wilson (2009) finds that the interest charges paid by firms to tax authorities amounted to 40 % of the tax savings originally generated by the tax shelter transactions. Graham and Tucker (2006) report the case of GlaxoSmithKline P.L.C. in 2004, which owes the IRS $5.2 billion in back taxes and penalties related to a transfer pricing strategy dating back to 1989.

Supporting this notion, Austin and Wilson (2015) find evidence that firms with valuable customer brands engage in lower levels of tax avoidance, due to the threat of reputational damages associated with tax avoidance. In an experimental study, Hardeck and Hertl (2014) also document that media coverage of aggressive corporate tax strategies can affect corporate reputation, and consumers punish tax aggressive companies by lowering their willingness to pay for companies’ products.

For instance, Citizens for Tax Justice reported in February 2012 on General Electric’s low tax rate of two percent over ten years. Despite this negative coverage, General Electric continues to be among the top tax avoiders in another study published in October 2015 (“The use of offshore tax havens by Fortune 500 companies”).

Lanis and Richardson (2014) employ an alternate measure of tax avoidance based on firm tax disputes, in the U.S. This measure is difficult to implement, since our sample covers 32 countries.

The importance of a private and competitive media is widely recognized; it is often called “the fourth estate,” along with the executive, the legislature, and the courts (Djankov et al. 2003).

For more details of the Panama Papers controversy, see ICIJ’s website: https://​panamapapers.​icij.​org/​ and Süddeutsche Zeitung’s website: http://​panamapapers.​sueddeutsche.​de/​articles/​56febff0a1bb8d3c​3495adf4/​.

Dyck et al. (2010, case summaries) highlight the case of Sprint Corporation, where an article by the New York Times alerted the IRS to the existence of four tax shelters promoted by Ernst and Young, and subsequent IRS investigations charged Sprint’s top executives with personal tax evasion via a mechanism that allowed them to cash out options without incurring tax liability for up to 30 years. This case illustrates that the media can conduct independent inquiry into tax avoidance activity.

Many firms claim that they are transparent in their corporate values; however, Paine et al. (2005) document that most corporate codes of conduct rarely discuss tax obligations.

As discussed in Dyck et al. (2010), even though journalists might be less specialized, they benefit from revealing complex issues, because high profile stories might help establish their career and reputation.

The discussion in this and the following paragraph is largely based on Houston el al. (2011).

Djankov et al. (2003) also provide data for the market share of circulation of private newspapers. In our sample, all countries (except the Philippines) have a market share of 100 %. We therefore do not use this variable because of its lack of variation.

Industry competition is commonly measured based on the market concentration ratio. Following Houston et al. (2011), we define the media industry as less competitive if the aggregate market share of the top five largest television stations or daily newspapers is high. Conversely, we define the media industry as more competitive if the aggregate market share of the non-top five largest television stations or daily newspapers is high.

We do not compute this measure over longer windows, such as five-year or ten-year windows (e.g., Dyreng et al. 2008), to avoid limiting our sample size. As noted by Dyreng et al. (2008), tax avoidance measures that are estimated over shorter periods of time may be imperfect because they include payments to (and refunds from) the tax authorities upon settling of tax disputes that arose years ago. Tax avoidance measures that are estimated over longer periods mitigate this concern because the income to which these taxes relate will more likely be included in the same ratio as the taxes. As a sensitivity check, we use a longer horizon of five years to compute tax avoidance, and find qualitatively unchanged results (untabulated). We also use two other alternate proxies of corporate tax avoidance. First, we compute tax avoidance based on the difference between the taxes on pre-tax income computed at the home-country statutory corporate tax rate and the tax expense recognized instead of the taxes actually paid. This measure is more closely related to the concept of GAAP effective tax rate, because it measures tax avoidance based on tax expense recognized in the financial statements rather than on cash tax actually paid. To capture the alternate measure of tax aggressiveness, we use an indicator variable that equals one if the country-industry tax avoidance based on tax expense recognized is within the top quartile, and zero otherwise (TAXAGGR_ALT). Second, we use the continuous measure of tax avoidance as the dependent variable. Our untabulated results are robust with these two alternate proxies of tax measures.

Other measures of tax avoidance used in the extant literature include DTAX (Frank et al. 2009), tax shelter prediction score (Wilson 2009), and unrecognized tax benefit (UTB) prediction score (Rego and Wilson 2012). However, because we use an international sample of firms from Compustat Global, many of the variables required to compute these measures of tax avoidance are either not available or not applicable in a setting outside the U.S. (e.g., tax shelter prediction score and UTB prediction score).

Industries are defined following the classification in Frankel et al. (2002), which is based on the following SIC codes: agriculture (0100–0999), mining and construction (1000–1999, excluding 1300–1399), food (2000–2111), textiles and printing/publishing (2200–2799), chemicals (2800–2824, 2840–2899), pharmaceuticals (2830–2836), extractive (2900–2999, 1300–1399), durable manufacturers (3000–3999, excluding 3570–3579 and 3670–3679), transportation (4000–4899), utilities (4900–4999), retail (5000–5999), services (7000–8999, excluding 7370–7379), and computers (3570–3579, 3670–3679, 7370–7379).

We control for the statutory tax rate to avoid the potential mechanical relation that may result from the tax avoidance measure computation including the statutory tax rate.

We hand-collect each country’s annual statutory corporate tax rate and whether the tax system is worldwide or territorial from various sources, including Ernst and Young’s Worldwide Corporate Tax Guide, KPMG’s Corporate and Indirect Tax Rate Survey, PwC’s Worldwide Tax Summaries, and PwC’s “Evolution of Territorial Tax Systems in the OECD” report.

Including higher order interaction terms in the model may cause the coefficient on the main effect of the conditioning variable to change sign unexpectedly, because the inclusion of MEDIA and its interaction with the conditioning factor can introduce multicollinearity among the interaction terms. To alleviate this concern, we mean-center MEDIA and the conditioning variables in our regression analyses (Aiken and West 1991; Neter et al. 1989).

Ai and Norton (2003) argue that the interaction effect in a non-linear model, such as the logistic specification of Eq. (3), cannot be evaluated and interpreted simply by looking at the sign, magnitude, and statistical significance of the coefficient on the interaction term. Rather, interpreting the interaction effect requires computation of modified statistics based on cross-derivatives or cross-differences. However, Greene (2010) contends that the modified statistics proposed by Ai and Norton (2003) do not provide meaningful interpretations and statistical inferences. In addition, Kolasinski and Siegel (2010) draw on the extant statistics literature (e.g., Le 1998) and show that the interaction coefficient and test statistic in a standard logistic specification are appropriate for research dealing with non-extreme probabilities and are economically meaningful. Therefore, we continue to estimate and interpret the interaction effects in Eq. (3).

Our sample period ends in 2007 because we obtain the requisite data from the Legacy Global Compustat database. The last year for which data are available in this database is 2007. The new Global Compustat database, which has the more recent data, does not include pre-tax exceptional items (data item 57 in the old database) and foreign income taxes (data item 51 in the old database). Therefore, we are unable to compute the variable TAXAVOID, which requires data item 57 as an input and the variable BTAXC, which requires data item 51 as an input, using the new Global Compustat database.

Our results are robust when these three countries are included in the sample.

In a robustness test, we employ weighted least squares to control for variation in sample country composition and re-estimate our models by country-year, so that each country-year observation receives equal weight in the regression (see “Additional Robustness Checks” section).

Our results are similar when we use an alternative proxy for democracy based on a democracy index obtained from the Polity IV dataset of Marshall and Jaggers (2007). This index is derived from coding the competitiveness of political participation, the openness and competitiveness of executive recruitment, and constraints on the executive, with higher values representing a more democratic environment. We do not include both proxies as instruments because both variables are very highly positively correlated (ρ = 0.83).

We also conduct two additional robustness checks. First, we control for country-industry fixed effects in the regression. Second, we employ Fama–MacBeth (1973) estimation method based on yearly regression. Our untabulated results indicate that the coefficient on Comp_TV and Comp_Press is still negative and significant at the 1 % level; however, Private_TV is now insignificant. Overall these additional tests suggest that media competition, rather than media ownership has a more pronounced effect on tax aggressiveness.

We do not use these governance variables in our main regression because complete data is only available for 24 countries.