Cross-jurisdictional income shifting and tax enforcement: evidence from public versus private multinationals

Under a territorial tax regime, countries exempt foreign income from additional home country tax also when the foreign tax rate is below the domestic rate. By contrast, under a worldwide regime, governments additionally tax foreign income but allow for tax credits for the foreign tax already paid. We treat the subtleties of both systems in more detail in the additional analyses in Sect. 5.3.2.

An example of high disparities in national statutory tax rates is the 28 % tax rate charged on profitable UK firms in 2009 when firms located in a neighboring country, Ireland, were taxed at only 12.5 %. This large disparity has raised serious concerns, and politicians are taking actions in an attempt to avoid a large “tax exodus” (Houlder 2009).

Some countries were only included in the Ernst & Young Transfer Pricing Guide after 2000, so we cannot track the full historical record of tax rulings and authority rigorousness. For those countries, we consider the situation that existed in the first year that information became available in the guide and assume the same situation was applicable for the earlier years as well. Because this method of classifying countries is conservative, it would only work against the findings presented in the empirical sections below.

Goncharov and Jacob (2012) study differences in accrual accounting for tax purposes among 33 OECD countries and cover all the countries under analysis in the current study. They report very little to no change in a tax accrual index from 2001 to 2005. We therefore presume that the 2005 tax environment is close to the pre-2005 years. In addition, we find in unreported sub-period tests that our main results hold both for the pre-2005 and post-2005 period, although public versus private MNC differences become stronger for the later years.

The Belgian coordination centers regime dates back to 1983 and was introduced to attract multinational companies’ activities. The European Commission ordered Belgium to repeal the special tax schedule for coordination centers, due to noncompliance with EU state aid rules. The special tax regime phased out at the end of 2010.

Klassen and Laplante (2012b) show empirically that firms with low foreign tax rates relative to US tax rates shift significantly more income out of the United States when foreign reinvestment-related incentives are high.

Note that Huizinga and Laeven (2008) who investigate family-level income shifting in subsidiary and parent companies of private and public European MNCs are an exception to this. However, while these authors describe that their sample contains both public and private company observations, they do not build tests on a firm’s listing status.

The claim that management compensation incentives shape a firm’s tax avoidance is confirmed by both theoretical and empirical work. Desai and Dharmapala (2006) model tax evasion while controlling for agency costs and show that increases in variable pay are positively related to corporate tax sheltering. This result is consistent with survey findings from Cloyd et al. (1996), who suggest that managers avoid reporting lower book performance because this is perceived as lowering the market value of a firm (and hence their variable compensation component). For a large international sample, Atwood et al. (2012) also find that corporate tax avoidance is shaped by management compensation, including stock options.

For the EU sample under analysis, we observe about 40 % of worldwide (i.e., tax credit) regimes and for 10 on 19 of the sample countries, no clear CFC rules are put in place (in alphabetical order: Austria, Belgium, Greece, Luxembourg, Ireland, the Netherlands, Spain, the Slovak Republic, Switzerland, and Sweden). When CFC rules apply, local tax authorities can include the undistributed income of corporations in foreign countries in the corporate tax base of resident parent companies, which decouples the financial reporting impact from the repatriation decision.

Note that we only observe income shifting in the affiliates in EU countries. We cannot include information on offshore income shifting because financial data on non-EU subsidiary fiscal entities are not captured by the database, hindering us from fully capturing structured tax evasion schemes with tax haven countries (such as for instance Double Irish-Dutch Sandwich constructions—O’Carroll 2011). We acknowledge that this is a potential shortcoming of our study but argue that excluding data from these tax haven countries would work against finding evidence in support of our main hypotheses. To mitigate this bias, we select a subsample of MNCs for which the observed subsidiaries represent at least 50 % of the MNC consolidated sales. Results for this subsample provide evidence that is largely in line with the findings for our full sample. Additional information on these results is reported in Table 5.

All consolidated variables are denoted with the subscript ‘MNC.’

Note that all firm-level variables are winsorized at the top/bottom 1 % to minimize the impact of outliers.

Note that a negative (positive) number, by definition, corresponds to lower (higher) foreign statutory tax rates.

In our primary tests, we use the cluster2gen Stata command to operationalize the two-way cluster-robust standard errors. We decide to cluster standard errors by MNC/year to control for potential unobserved year-correlations in MNC income shifting decisions across different subsidiaries. We thank Dan Taylor for providing the cluster2gen Stata command to calculate two-way cluster-robust standard errors for OLS with multiple fixed effects.

Note that 4.77 % is the standard deviation in STRDIFF for the full sample of public and private MNC observations. Table 3 reports a standard deviation of 4.33 % for private MNC observations and 5.32 % for public MNC observations.

It happens in 630 cases (or 1.1 % of all observations) that the statutory tax rates (STRs) in subsidiary and parent countries are identical. When verifying the potential impact on regressions (1) and (2), we observe very similar results as for the ones originally reported in the paper.

Note that in all regressions—except for regression (8), which compares public versus private MNC income shifting—the joint effect marked in gray captures the income shifting coefficient for weak enforcement setting observations.

Note that, while this matching procedure retains most of the private MNC subsidiary observations (78.6 % or 28,860/36,700), less than 50 % (49.2 % or 11,926/24,258) of the public MNC observations remain in the sample.

A solution that may reduce the difference in all treatment and control variables may go via the requirement of a stricter match of nearest neighbors, e.g., a one-to-one match. However, when requiring a unique match, the number of public MNC observations that are withheld drops further to 27.2 %, and the balancing property of the matching algorithm is no longer satisfied. We therefore withhold the 5:1 match in this sensitivity check. Even though the matching procedure cannot fully eliminate the growth difference in both samples, the sample means are much closer compared to before the matching procedure.

In summary, this measure summarizes all information about income shifting incentives (differences in statutory tax rates [t] across all EU-based MNC subsidiary countries) and income shifting opportunities (the scale of the firm’s operations B across all countries). Another interpretation for this measure is that MNC subsidiary-country affiliates’ tax reporting incentives can be more easily achieved if larger activity levels are realized in affiliate countries with lower STRs (i.e., more sales are located in low-tax countries). In the presence of tax incentives, this composite tax measure is expected to relate negatively to the reported income level in country i. For more information on the composite tax measure, see Huizinga and Laeven (2008), p. 1166 onward.

Note that, for interpretational reasons and for consistency with our main analyses on income shifting incentives, we recoded the C and the C_parent score into a dummy variable equal to one if satisfying the condition in the column head and zero elsewhere and multiplying by minus one in the case of a negative C in regression models 22–29.

The link between audit rates and tax compliance has been studied by for instance Dubin et al. (1990). The authors find that a continual decline in the audit rate from 1977 to 1986 caused a significant decline in IRS collections, indicating that tax audit likelihoods are positively related to tax compliance.

Univariate statistics show a Pearson correlation between our tax enforcement variable and the natural logarithm of population density of 0.54 (p < 0.01) and GDP per capita of 0.09 (p < 0.01).