The role of headquarters in multinational profit shifting strategies

The location of funds at geographically distant foreign affiliates likely hampers control and may give rise to agency problems.

Profit shifting costs thus rise convexly in the amount of profit shifted, which may relate to the fact that the tax authority is over-proportionally more likely to identify and fine profit shifting activities the larger the amount of relocated profit. Alternatively, the convex cost function may reflect that it becomes increasingly more costly for the multinational firm to hide profit shifting activities from the tax authorities the larger the amount of relocated income (see e.g. Haufler and Schjelderup 2000).

This particularly holds true for valuable assets and key units like R&D which are, on top, often associated with reputational benefits. R&D is for example well-known to earn higher returns than other investments. From a taxation perspective, one would thus expect that R&D units belong to the internationally most mobile departments within multinational groups. Empirical research, in turn, suggests that the contrary is true and R&D belongs to the least mobile functions within MNEs (see e.g. Abramovsky et al. 2008; Criscuolo et al. 2010).

Note that our data includes information on the parent firm who is the legal owner of the subsidiary. However, the legal owner and the MNE’s home country for tax purposes need not necessarily coincide. Precisely, different countries rely on different criteria to establish “fiscal residence,” the most important being incorporation, seat and central management. These elements are often fragmented. Different national laws imply that a corporation can be resident in multiple jurisdictions at once. As an example, the UK uses two separate tests for residence: firstly, incorporation, i.e. a corporation is resident in the UK if it is incorporated there, and secondly, central management and control, i.e. a corporation is also considered resident where it is managed and controlled, which means where the board of directors meets to make major decisions. Thus, a corporation could in principle incorporate in one country and move its central management and control abroad to avoid residency in the country of incorporation (see Loomer 2008). This is, however, considered to be a risky and therefore rather uncommon strategy as it may be challenged by the tax authorities in the country of incorporation (see also Loomer 2008).

The coverage of firms varies between the countries as Bureau van Dijk draws on different information sources in different countries which vary in their coverage. Nevertheless, profit shifting is expected to be most pronounced for large multinational firms which tend to be captured in all our sample countries.

The information on statutory tax rates is taken from the European Commission. Country data for GDP and GDP per capita are obtained from the IMF World Economic Outlook Database October 2008. The Corruption Perceptions Index is taken from Transparency International and ranks from 0 (extreme level of corruption) to 10 (free of corruption).

Note that the fixed-effects approach is crucial to identify multinational profit shifting behavior. Ideally, the econometrician would want to identify income shifting activities by comparing the affiliate’s observed pre-tax profitability (conditional on the prevailing corporate tax law) with the ‘true’ pre-tax profitability in the absence of tax-motivated income relocations which is unobserved though. Following previous studies, we tackle this problem by estimating a fixed-effects model which determines whether corporate tax rate changes induce changes in the reported pre-tax profitability at the affiliate firms.

We, however, also experimented with other cut-off thresholds, which derives comparable results to the ones reported in this paper.

Note that for presentational convenience, the tax difference variable in the regressions for the parent sample is defined as corporate tax rate at the parent location minus the average corporate tax rate at its foreign subsidiaries. If the parent firm engages in significant profit-shifting activities, we thus again expect a negative coefficient estimate for the tax difference variable.

Note that clustering at the country-year level derives similar results as will be shown below.

Following our description in Sect. 3, the parent dummy variable takes on the value 1 (0) if the parent firm is located in a country with a smaller (higher) corporate tax rate than two thirds of its majority-owned subsidiaries.

Note that this robustness check comes at the cost of a reduced sample size as information on the size of the parent firm is not available for all subsidiaries in our sample. Some subsidiaries, for example, observe non-European parent firms for which accounting information is not contained in the AMADEUS data. For some European parents no unconsolidated accounting information is available.

Again, due to missing parent information, the robustness check comes at the cost of a reduced sample size; see also the previous footnote.

Residence-based taxation according to a tax credit regime implies that income is taxable in the parent country upon repatriation but a tax credit is provided for taxes already paid in the subsidiary country. As income is thus eventually taxed at the parent country tax rate, MNEs may have a diminished incentive to relocate profits to low-tax entities (the disincentive might be small though as repatriation is often deferred). Note further that the main alternative to tax credit regimes is residence-based taxation where repatriated income is exempted from taxation. Information on repatriation taxes was obtained from the Ernst & Young Corporate Tax Guides.

This may especially apply if the size of the absolute tax rate differential between parents and subsidiaries differs systematically between the subgroups of high-tax and low-tax subsidiaries. In that case non-linearities could generate a pattern similar to the one reported in our analysis even in the absence of a parent bias in profit shifting behavior. Note, however, that we consider this to be unlikely as the absolute tax difference is comparable in the subgroups of high-tax and low-tax subsidiaries in our sample (see the descriptive statistics in Table 2).

Augmenting the specifications by polynomials of higher order is rejected by a Wald test for both subsamples.