In order to construct our network analysis, we collect tax data for a sample of 138 countries between the years 2005 and 2012.Footnote 2 Our main source of data on the domestic and international tax system is the IBFD Global Corporate Tax Handbooks for the years 2009–2012 and the IBFD Online Tax Platform. For the countries included in the Global Corporate Tax Handbooks, we collect information on the domestic tax system and, in particular, on taxation of foreign income (including the methods of double tax relief), as well as domestic corporate and withholding tax rates from the respective yearbook.Footnote 3 To the extent that a country is not available in a Global Corporate Tax Handbook, we consult the closest to the missing year data source for the taxation of foreign income, including the IBFD Online Tax Platform, and, unless indicated otherwise, assume the same method of taxation of foreign income for the missing years.
Moreover, we update all domestic corporate and withholding tax rates with the EY (Ernst and Young) Corporate Tax Guides if the IBFD data are not available for a particular year. For instance, for the years 2005–2008, the EY Corporate Tax Guides are our only source of data on domestic corporate and withholding tax rates. We further hand-collect the relevant withholding tax rates and methods of double tax relief from the respective DTTs and applicable protocols. In addition, as the treaty network is subject to four types of changes, we check when new treaties become effective; if treaties have been terminated at a later point in time; if the conditions of the treaties have been changed through protocols in the following years; and if the conditions of the treaties have been altered through amendments in domestic law.Footnote 4 Overall, we consult more than 3000 tax treaties that became effective before 2013 and around 300 accompanying protocols.
We obtain data on bilateral inward FDI stocks between 2005 and 2012 from the UNCTAD (United Nations Conference on Trade and Development) database, and we invert them to measure the investment from the home to the host country. In the presence of FDI diversion via a third country, we would ideally want to observe the indirect investment from the home to the host country via the conduit country. However, the available data report only the immediate home- to host-country FDI stocks. Therefore, we can only estimate the impact of DTTs on these immediate home- to host-country FDI stocks.Footnote 5 Finally, the information on bilateral investment treaties (BITs) is from the Investment Policy Hub of UNCTAD.
Recent contributions by van’t Riet and Lejour (2018) and Hong (2018) employ a network approach to study the centrality of countries in the tax treaties network and, respectively, the structure of tax-minimising (direct and indirect) investment routes. Both studies analyse the tax treaties network for a single year and ignore any changes in the tax treaties network over time. While Hong (2018) uses a simple computational algorithm, van’t Riet and Lejour (2018) use an adapted version of the Floyd–Warshall shortest path algorithm to estimate these tax-minimising investment routes. However, both approaches either underestimate or overestimate the potential for tax treaty shopping.Footnote 6
We take a different approach and develop a Visual Basic Application (VBA) tool to recalculate the tax distance for every possible combination of host, home and intermediate countries.Footnote 7 In this way, we can take into account the actual taxes paid in the jurisdiction before the one receiving the dividends—typically the intermediate jurisdiction—instead of nominal or world-average corporate tax rates. The single limitation of our approach is that we restrict the number of possible intermediate jurisdictions to two in order to avoid long computation time of the analysis. However, this may not be an unrealistic assumption, as Mintz and Weichenrieder (2010) show that only 0.2% of German multinational firms use cross-border group structures with three or more pass-through entities. Moreover, when we analyse our network using the Floyd–Warshall algorithm (allowing for an unlimited number of conduits), we do not find any indirect connection with three or more intermediate jurisdictions that would further reduce the tax distance between any two countries in our sample. For this reason too, we believe that our approach is superior to the Floyd–Warshall algorithm and allows for a more accurate network analysis.
For every year, we update the tax treaties network with all relevant changes. In particular, we account for changes in the provisions of tax treaties through amending protocols and for changes in the provisions of the tax treaties through changes under domestic law (for tax treaties that refer to conditions under domestic law); we add new tax treaties that become effective and remove tax treaties that have been terminated or replaced by new ones in the course of the year being analysed. We assume a fully owned subsidiary engaged in an active course of business and consider only domestic anti-abuse provisions.Footnote 8 Specifically, we account for higher withholding taxes upon dividend distributions to tax havens and clauses that subject the subsidiary or the parent company to a minimum rate of CIT.
Several countries in our sample levy a higher withholding tax on dividends when these are distributed to a parent located in a tax haven. Because most of the available domestic tax havens lists are not comprehensive, we adopt a common tax haven list for all countries in our sample across the entire time period (Dyreng and Lindsey 2009).Footnote 9 In accordance with the majority of these domestic provisions, we exclude the anti-abuse treatment when a DTT is in place. Similarly, several countries in our sample adapt a subject-to-a-minimum-tax clause as a condition for claiming the benefits of participation exemption and exemption from withholding tax on dividends.Footnote 10 Since we observe what corporate tax rates the subsidiary and the parent company are subject to, we can easily control for this condition.
We describe the entire international network of double tax treaties with a set of tax treaty dummies depicted in Fig. 1 and a measure of bilateral taxes summarised in Table 2. Our first variable is a dummy that verifies whether a DTT between two countries is present, Treaty. This is the standard variable used in the previous literature. Country pairs that did not conclude a treaty will be our reference category throughout our estimations. For every year in our sample, we then measure the direct tax distance between any two countries taking into account a possible tax treaty between these two countries, DirectTaxDistance.
Measuring the direct tax distance permits us to distinguish between treaties. First, we define RelevantODL tax treaties as tax treaties that reduce the effective tax rate on overseas profits below the one under domestic law of the source and residence countries. For example, in the year 2012, ignoring the bilateral tax treaty, the tax distance between Argentina as the home state and Belgium as the host state is 25% under the domestic law of both countries. However, the applicable DTT reduces the direct tax distance to approximately 1.5%. Hence, the tax treaty is relevant relative to domestic law provisions. We label tax treaties that do not provide for this benefit as IrrelevantADL. Further, we expect RelevantODL DTTs to increase the bilateral FDI and IrrelevantADL to have no effect.
The second innovative element in our analysis is to identify whether an indirect route exists along which the tax distance would be reduced as opposed to the direct route.Footnote 11 For example, in 2012, a South African parent company investing directly in a US subsidiary had to pay 5% tax on distribution of dividends after considering the tax treaty between both countries. However, if the same investment is made through a conduit company in the Netherlands, the tax cost can be reduced to 0%.
Once we estimate the minimum direct and indirect tax cost between any two countries, we ask whether RelevantODL tax treaties remain relevant also considering the entire treaty network. We define RelevantOTN tax treaties as tax treaties that reduce the effective tax rate on overseas profits not only below the one under domestic law, but also to or below the minimum one in the network. In the Argentina–Belgium example above, the lowest possible tax distance when channelling income through the network is 12.5%.Footnote 12 In the absence of the bilateral tax treaty, the MNE has a tax incentive to choose the indirect route over the direct one. However, with a direct tax distance of only 1.5%, the DTT between Argentina and Belgium takes away the advantage of the indirect one and further reduces the minimum tax distance between the two countries by 11 percentage points.
By contrast, the IrrelevantATN dummy indicates tax treaties that reduce the direct tax distance, but not the minimum (indirect) tax distance between the source and residence countries. Consider the case of Argentina as the home country and Germany as the host state. In 2012, the direct tax distance between the two countries is about 26.4% under their domestic law, while the minimum tax distance through the network is 12.5%. Thus, also in this case, we expect the MNE to tax-prefer the indirect route rather than the direct one. Moreover, the DTT between the two countries reduces the direct tax distance to 21.25%, which is still higher than the minimum tax distance through the network. As a result, the tax treaty between Argentina and Germany is irrelevant to the MNE’s decision to invest via a third country. To the extent that MNEs make use of treaty shopping opportunities, we expect RelevantOTN DTTs to have a bigger effect on FDI than IrrelevantATN tax treaties.
We can further decompose the RelevantOTN dummy and differentiate between relevant DTTs that are strictly better than the tax treaties network—StrictlyRelevantOTN—and relevant DTTs that just cut the tax cost of the direct route to the minimum in the network, WeaklyRelevantOTN. In theory, StrictlyRelevantOTN tax treaties should stimulate FDI between two countries for two reasons. First, firms may relocate investments from the indirect route to the direct route or invest directly where they did not invest via a conduit company despite its tax benefit in the absence of the DTT. Second, firms would also benefit from a lower overall tax burden, and this should increase FDI. Presuming non-negligible costs to treaty shopping, WeaklyRelevantOTN tax treaties may increase FDI between the home and host states if firms relocate investments from the indirect route to the direct route.
We can also narrow down the IrrelevantATN dummy and distinguish between irrelevant tax treaties where the tax-minimising indirect investment route involves one conduit country—IrrelevantATN1—or two conduit countries—IrrelevantATN2. If treaty shopping is costly, MNEs may be less likely to use more complex investment structures. Accordingly, we expect the effects of irrelevant DTTs on direct FDI to increase with the number of intermediate jurisdictions needed to set up the tax-minimising indirect route.
Our sample consists of 138 countries between 2005 and 2012, which corresponds to 18,906 unique country pairs in each sample year.Footnote 13 However, due to missing economic data, the econometric analysis covers only 133 countries. Table 3 summarises the characteristics of the international tax network, while Table 4 gives an overview of the estimation sample summary statistics.
In 2005, more than 44% of unique country pairs exempt foreign dividends under their domestic law—thus ignoring bilateral DTTs. On the other end of the spectrum, 10.5% of unique country pairs give no relief for foreign taxes. 5.7% of all pairs allow foreign taxes to be deducted as a business expense. Remaining country pairs credit the host withholding tax from the domestic tax liability, with almost 11% of all pairs crediting also the underlying corporate tax. Once we include bilateral tax treaties, the shares of no relief and deduction drop to approximately 9% for the former and 5% for the latter; the percentage of countries using the direct credit method remains stable around 29%; indirect credits’ share rises above 11%; while the use of exemption method increases the most to more than 45%.
In terms of the cheapest connection en route, we observe that for more than 53% of all country pairs the direct connection is the cheapest one. Further, 35.5% achieve the minimum tax distance on an indirect route with one conduit company and 10.7% on an indirect route with two conduits. Overall, 7213 out of 18,906 unique country pairs have a zero tax distance, where there are no repatriation taxes on distributed income. Corporate income is taxed thus only once, at the level of the subsidiary, and there is no economic double taxation.Footnote 14 Close to half, 48.5%, of the zero tax distance connections occur on the direct connection, about 42% on an indirect route with one intermediate country and the remaining 9.7% on an indirect route with two intermediates.
In 2005, 3487 country pairs had an effective DTT.Footnote 15 Out of these, 1500 country pairs had a RelevantODL tax treaty to the extent that it reduced the direct tax distance. Among these DTTs, about half, 733, remain relevant once accounting for the possibility of treaty shopping. The overwhelming majority (723 out of 767) of country pairs with an IrrelevantATN dummy is at the disadvantage of a cheaper indirect route that involves only one conduit. Finally, more DTTs, 410 against 323 country pairs, cut the direct tax distance to the minimum one in the network rather than below it.
Moving to the last year in our sample, 2012, and leaving again the effect of tax treaties aside, about 9% of all country pairs have no relief for foreign taxes; 5.8% use deduction as the only relief method; 27.1% apply direct credit; approximately 9.8% offer indirect credit; and 49.4% apply exemption. Taking into account bilateral DTTs, the share of the no relief method drops below 8% and that of the deduction method to 5%. At the same time, the shares of all other methods increase to 27.6% in the case of direct credit; 9.4% in the case of indirect credit; and 50.3% for the exemption method.
Focusing again on the cheapest connections in the network, we see that now only 52.1% of the cheapest connections occur on the direct route. This suggests that treaty shopping has gained in importance over the last decade. The use of indirect routes with one conduit company increases to above 37%, whereas indirect routes with two conduits increases to 10.7%. Overall, 8907 out of 18,906 country pairs have a zero tax distance. Among these, 46.1% country pairs have a direct tax distance of 0%; 43.3% country pairs have a zero tax distance on an indirect route with one intermediate country; while the remaining 10.6% zero tax distances are achieved on an indirect route with two intermediates.
Finally, in 2012, 4264 country pairs had an effective DTT. 1903 of them are of the RelevantODL treaties and 2361 of the IrrelevantADL treaties. Similarly to 2005, about half, 1013, of RelevantODL treaties turn irrelevant once accounting for the possibility of treaty shopping. More country pairs are now at a disadvantage of indirect routes involving two intermediate countries (92 out of 1013); the number of country pairs with a WeaklyRelevantOTN treaty increases to 579; and the number of country pairs with a StrictlyRelevantOTN treaty decreases to 311.
The presented statistics show the surprisingly high fraction of irrelevant tax treaties, both against domestic law and against tax treaties network. Even excluding DTTs concluded between EU countries in light of the particular position of the Parent-Subsidiary Directive, we find that more than 70% of IrrelevantADL tax treaties remain in place. This last observation confirms that IrrelevantADL DTTs are not specific to the EU, but a network-wide phenomenon. Moreover, it demonstrates the increasing role of domestic legislation in the avoidance of (economic) double taxation and the scale of domestic law changes.Footnote 16
The high fraction of IrrelevantATN tax treaties (more than half the number of RelevantODL) suggests that a majority of countries does not take into account the entire tax treaty network upon negotiating new DTTs or renegotiating already existing DTTs. Particularly when a more recent tax treaty between two-third countries creates a new tax-minimising indirect route, one could expect more renegotiations of DTTs already in place. In contrast, the once-concluded tax treaties are characterised by a low level of dynamism and infrequent changes.
Tax treaties, treaty benefit and treaty shopping
While the identification of conduit countries is not the main purpose of this paper (in analogy with Van ’t Riet and Lejour 2018, and Hong 2018), we are interested in the channels along which DTTs reduce taxes on repatriation. Until recently, the preamble to the OECD Model Tax Convention stated a single objective of double tax treaties: to eliminate double taxation with respect to taxes on income and capital. In this way, DTTs were supposed to promote international economic activity. At the same time, for more than half (17,899 out of 30,918) of all country pairs in our sample with an effective DTT, the tax treaty does not reduce taxes on repatriation relative to the conditions under domestic law. Given their objective, is there still a rationale behind these tax treaties?
Over the years, DTTs have come to pursue additional goals such as providing legal certainty, preventing tax discrimination in the state of investment and exchange of information for tax matters. Indeed, Ligthart et al. (2011) show in a gravity framework that while countries sign DTTs primarily to reduce international double taxation, providing a legal instrument for the exchange of information in tax matters is another important objective. In support of a broader function of DTTs, Davies et al. (2009) and Blonigen et al. (2014) show positive treaty effects ascribed to giving legal certainty and establishing guidelines for dispute settlement between tax authorities of the signatory countries. Moreover, the current preamble to the OECD Model Tax Convention adds the prevention of tax avoidance and evasion as additional objectives of double tax treaties. Finally, the scope of DTTs stretches beyond taxes on corporate profit and includes determination of fiscal residence and rules allocating taxing rights for individuals. In light of this, even treaties defined as irrelevant in this paper, are far from being fully “irrelevant” and the role of DTTs is unlikely to diminish in the near future.
Having addressed the position of irrelevant tax treaties under domestic law, we focus our attention on RelevantODL DTTs and the driving forces behind their relevance. Table 3 suggests only a modest improvement in the relief methods once the effect of DTTs is taken into account. The development of the host-country dividend withholding taxes under domestic law and under tax treaties can be seen in Fig. 2 for 2005 and in Fig. 3 for 2012. While about 35% of country pairs have zero withholding at source unilaterally, there is a notable shift towards lower withholding tax rates with DTTs in place, especially towards the 5% and 10% withholding tax rates.
We also show the treaty benefit (solid line) for all tax treaties effective in the year 2005 in Fig. 4 and all tax treaties effective in the year 2012 in Fig. 5. We then disentangle this treaty benefit into benefit that can be attributed to lower withholding tax rate (dotted line) and benefit that can be attributed to more generous relief methods (dashed line).
Comparing the treaty benefits of lower withholding taxes and more generous relief methods suggests that the reduction in withholding taxes is the driving force behind treaty relevance. However, while in theory the rate and relief method benefits should add up to the total benefit, they actually slightly overestimate it.Footnote 17 We calculate each channel by holding the other factor constant at its domestic value. As a result, if a treaty operates along both margins—by providing a more generous relief method and lower withholding taxes—we are either not able to achieve the full potential of tax treaties or calculate a given benefit twice.
We decide to follow this methodology because it is not known on theoretical grounds whether a certain benefit should be attributed to more generous relief methods or lower withholding taxes. In particular, as long as the home corporate tax rate is higher than the host withholding tax rate, relief by credit method would take away the entire advantage of lower withholding taxes. If a subsequent treaty further reduces the withholding tax rate and at the same time exempts foreign dividends, it cannot be established which effect happens first.
We are also interested in which countries are more likely to have RelevantODL and RelevantOTN treaties. Logit estimations suggest that relevant tax treaties are positively correlated with the GDP per capita ratio of the home country and negatively correlated with the GDP per capita ratio of the host country.Footnote 18 Not surprisingly, relevant DTTs are negatively correlated with EU membership of home and host countries. However, the correlation is positive with the year of signature of the tax treaty. This might be indicative of countries’ learning effects over the years as well as the tendency over the years to impose higher source withholding taxes and grant less generous unilateral relief methods towards the low-tax jurisdictions.
In terms of the relief methods, we observe different effects depending on whether we look at the domestic method of double tax relief or the one specified in the DTT. Whereas more generous relief methods under domestic law are negatively correlated with relevant tax treaties, the opposite is true for more generous relief methods under DTTs. The underlying mechanism is evident: the more generous the domestic relief method, the lower the domestic tax distance between the host and home states and the less room there is for the treaty to lower the distance even further. A more generous treaty relief method has the potential to reduce home taxation of foreign dividends beyond the unilateral conditions and secures the benefits of lower withholding taxes at source.
We observe an analogous mechanism in the case of source withholding tax rates of dividends. The higher domestic withholding tax rates are, the more likely it is that the treaty will lower these rates. The higher the source withholding tax rate is, the less likely it is that the treaty provides a benefit beyond the conditions of domestic law. In addition, it appears that the marginal effect of domestic and treaty withholding tax rates is stronger than that of domestic and treaty relief methods. This supports the notion presented in Figs. 4 and 5 that the reduction in withholding taxes is the driving force behind treaty relevance.
In the case of RelevantOTN treaties, the results are the same except for two variables: the GDP per capita ratio of the home country—which is only weakly significant for RelevantODL treaties—and the signature year of the treaty are no longer significant. The significance level and the sign of coefficients of all other variables remain intact. However, the much lower magnitude of domestic and treaty relief methods as well as domestic and treaty dividend withholding taxes suggests that interactions with other treaties in the network are less prone to domestic tax policy choices.
Finally, Fig. 6 describes the economic consequences of treaty shopping. On the horizontal axis, we plot for every single observation in our panel the direct tax distance in the absence of treaty shopping, DirectTaxDistance. On the vertical axis, we show effective taxes paid if instead an indirect route via one or two conduits is chosen. Points along the diagonal exhibit no gains of treaty shopping. All countries where the direct distance is the cheapest route will be along this line. The greater the vertical distance from the diagonal, the bigger the saving due to treaty shopping. We show ample possibilities for treaty shopping, in many cases reducing the actual tax burden to zero.
Figure 6 reveals two interesting patterns. We find a series of vertical lines, which typically reflect individual country pairs, where neither domestic tax regulation nor the DTT have changed, and hence, the tax burden along the direct route remains unchanged. However, subsequent treaties signed with or between third countries have reduced the tax burden along the indirect route, demonstrating how the international DTT network undermines national policy. We also observe that a great deal of our observations occurs along the 5%, 10% and 15% effective tax rates, which reflect the withholding tax rates usually agreed on in DTTs. These are represented by the horizontal lines. Under the exemption system applied by the majority of countries in our sample, the actual tax burden is brought back from the level of domestic withholding tax rates to these common treaty withholding tax rates. Moreover, a significant number of observations are concentrated along the 25%, 30% and 35% direct tax distance, which coincides with the corporate tax rates of many counties. These points comprise all instances where the home country unilaterally offers a foreign tax credit—thereby setting the direct tax distance equal to the domestic corporate tax rate—but the MNEs benefit from tax treaties with a more generous method of double tax relief.