Tax Treaties and Foreign Equity Holding Companies of Multinational Corporations

Transcription

1 Tax Treaties and Foreign Equity Holding Companies of Multinational Corporations Sunghoon Hong March 2018 Abstract Multinational corporations can organize indirect ownership chains with foreign equity holding companies in countries with low taxes and favorable tax treaties. This paper examines the relationship between tax treaty networks and multinational ownership chains by combining ownership data of multinational corporations with a network analysis of tax treaties. Empirical results show that the structure of taxminimizing routes in a treaty network is closely related to the structure of multinational ownership chains. If there is a tax-minimizing direct route from home to source countries, multinational corporations are less likely to use foreign equity holding companies in their ownership chains. If there is no tax-minimizing direct route, as a tax-minimizing indirect route reduces more tax than a direct route does, multinational corporations are more likely to use foreign equity holding companies in their ownership chains. Furthermore, multinational corporations are more likely to locate foreign equity holding companies in countries along tax-minimizing indirect routes. JEL classification: F23, H25, H87 Keywords: multinational corporation, foreign equity holding company, ownership chain, treaty network, tax-minimizing route Korea Institute of Public Finance, 336 Sicheong-daero, Sejong 30147, South Korea, sunghoonhong@kipf.re.kr. 1

2 1 Introduction Multinational corporations often organize complex and indirect ownership structures to operate subsidiary companies around the world. The internal ownership structures of multinational corporations can be influenced by economic and cultural factors, such as tax, trade, language, and legal origin. However, only a few studies have examined how multinational corporations organize their internal ownership structures and why they decide to set up complex and indirect structures with a series of intermediate subsidiaries, such as foreign equity holding companies. Notably, Dyreng et al. (2015) and Lewellen and Robinson (2013) examine internal ownership structures of American multinational corporations. Mintz and Weichenrieder (2010) study the cases of German multinational corporations. Multinational corporations can reduce global tax liabilities, for instance, by choosing transfer pricing rules and setting prices for transactions between related companies; by relocating intangible assets (possibly with cost-sharing agreements) and paying royalties to related companies in countries with low taxes; by using debt instruments (or even debt-equity hybrids) and deducting interest payments from taxable income of related companies in countries with high taxes. 1 It is worthwhile to note that multinational corporations can use these tax planning techniques between any pair of related companies, less subject to the constraints of internal ownership structures. In contrast, dividend payments can be made only to the company that directly owns the shares of the company paying the dividends. 2 In other words, multinational 1 There has been extensive research on tax-motivated profit shifting of multinational corporations. Dyreng and Markle (2016) study how financial constraints of multinational corporations influence their decisions on profit shifting. Dharmapala and Riedel (2013) examine how earnings shocks influence profit shifting. For related studies, see Huizinga et al. (2008) and Desai et al. (2004) on internal debt finance, and see Griffith et al. (2014) and Dischinger and Riedel (2011) on intangible asset location. 2 Huizinga and Voget (2009) show how dividend taxation affects location choices of multinational corporations after cross-border mergers and acquisitions. Voget (2011) examines tax-motivated relocation of multinational headquarters. Barrios et al. (2012) find that corporate taxation of foreign-source dividends in home countries can deter entries into potential source countries. 2

3 corporations need to organize internal ownership structures in advance to receive dividends from subsidiaries in a tax-efficient way. Moreover, after exhausting all the other tax planning techniques, dividend payments through ownership structures may be the last available option for multinational corporations to return profits. In this context, it is important to better understand the factors that determine the internal ownership structures of multinational corporations. Meanwhile, countries coordinate their sovereign taxing rights by using tax treaties as policy instruments. One of the purposes of tax treaties is to prevent double taxation on cross-border economic activity. To this end, tax treaties reduce withholding tax rates on various types of income, such as dividends, interest, and royalties, imposed by source countries. 3 Thus, tax treaties are expected to promote cross-border economic activity, such as foreign direct investment (FDI) of multinational corporations. However, there is little evidence that the existence of a tax treaty significantly increases FDI. Blonigen and Davies (2004) use an indicator variable for the existence of a tax treaty and find no significant relationship between tax treaties and FDI. Louie and Rousslang (2008) also find no significant treaty effect on the rate of return from FDI. Davies et al. (2009) use data on the foreign activities of Swedish multinational corporations and find no significant treaty effect on the level of FDI. 4 Probably, one of the reasons for weak evidence lies in the aggregate nature of FDI data. We cannot distinguish between direct and indirect investment (ownership) structures in aggregate FDI data. However, we may be able to overcome such a limitation by using data on internal ownership structures of multinational corporations. 3 Here the term source refers to a place where (individual or corporate) persons earn income from their activities and home refers to a place where they originally own the rights for their activities. 4 Weyzig (2013) uses Dutch data on special purpose entities and finds positive treaty effects on FDI. Blonigen et al. (2014) use the US BEA data on multinational company operations and find differential effects of tax treaties on foreign affiliate activities, depending on the use of differentiated inputs across industry sectors. They find positive treaty effects on foreign affiliate activities by firms in industries using more differentiated inputs. 3

4 Another reason may be that researchers tend to ignore the network effects of tax treaties by considering each treaty separately from the others. Alternatively, we can consider a group of tax treaties as a network between countries. As a navigation device allows us to figure out the shortest routes (to minimize time or distance) in a transportation network, a network analysis of tax treaties can help find the most efficient routes (to minimize taxes) in a tax treaty network. Hong (2018) studies the structure of tax-minimizing routes in treaty networks, computes treaty shopping rates, and identifies common pass-through countries. Petkova et al. (2018) and van t Riet and Lejour (2017) also adopt network methodology to examine the relationship between FDI and tax treaty networks. In this paper, I examine the relationship between tax treaty networks and multinational ownership chains by combining ownership data of multinational corporations with a network analysis of tax treaties in Hong (2018). I find that the structure of tax-minimizing routes in a treaty network is closely related to the structure of multinational ownership chains. If there is a tax-minimizing direct route from home to source countries, multinational corporations are less likely to organize ownership chains with foreign equity holding companies. They also organize less complex ownership chains. If there is no tax-minimizing direct route, as a tax-minimizing indirect route reduces more tax than a direct route does, i.e., as the treaty shopping rate increases, multinational corporations are more likely to organize ownership chains with foreign equity holding companies. They also organize more complex ownership chains. Furthermore, multinational corporations are more likely to locate foreign equity holding companies in pass-through countries along tax-minimizing indirect routes, which can be identified by the treaty network analysis. Notably, the Netherlands, Luxembourg, Hong Kong, and Singapore are frequent locations for foreign equity holding companies. These findings provide evidence that multinational corporations conduct treaty shopping to reduce tax on dividends from foreign subsidiaries. By making direct routes tax-minimizing, countries can remove an incentive for 4

5 multinational corporations to use indirect routes, and can prevent treaty shopping. This paper proceeds as follows. Section 2 provides background information on tax treaty networks and multinational ownership chains. Section 3 develops the research design. Section 4 describes the data on treaty networks and ownership chains. Section 5 presents the results. Section 6 concludes. 2 Background 2.1 Tax treaty networks One of the purposes of tax treaties is to prevent double taxation on crossborder economic activity. To this end, tax treaties reduce withholding tax rates on various types of income, such as dividends, interest, and royalties, imposed by source countries. For instance, the tax treaty between China and the United States requires that the withholding tax rate on dividends should not exceed 10 percent if the recipient of the dividends is a resident of either country. Because tax treaties can reduce taxes on income earned abroad, tax treaties are expected to promote cross-border economic activity, such as foreign direct investment (FDI). A group of tax treaties can be viewed as a network, where each node is a country and each link is given a weight representing a withholding tax (WHT) rate on certain income. Let us consider a tax treaty network between China, the Netherlands, and the United States, illustrated by Figure 1. According to the tax treaty between China and the United States, the WHT rate on dividends is 10 percent. However, the tax treaty between China and the Netherlands sets the WHT rate on dividends at 5 percent if the recipient owns at least 25 percent of the shares of the company paying the dividends. Moreover, the tax treaty between the Netherlands and the United States sets the WHT rate on dividends at 0 percent if the recipient owns at least 80 percent of the shares of the company paying the dividends. In general, the Netherlands imposes no corporate income tax on dividends earned abroad. 5

6 Figure 1. Tax treaty network Netherlands Dividends WHT 5% China Dividends WHT 10% Dividends WHT 0% United States Investment Dividends Now imagine that an American corporation invests in a Chinese subsidiary and that the Chinese subsidiary remits dividends to the American parent. If owned directly, the Chinese subsidiary pays dividends directly to the American parent, and thus, the WHT rate is 10 percent. However, if owned indirectly through a holding company in the Netherlands, the Chinese subsidiary pays dividends to the holding company, which in turn pays dividends to the American parent. By using such an indirect ownership structure with a Dutch holding company, the American parent can reduce the tax on dividends by 5 percentage points. 5 This indirect ownership structure through the Netherlands is an example of tax-minimizing investment structures (or investment routes) in a tax treaty network. In addition, there are a number of locations for holding companies, such as Hong Kong and Singapore, to minimize tax on dividends when an American corporation invests in China. 6 5 When the Chinese subsidiary pays dividends to the Dutch holding company, the WHT rate is 5 percent. The Dutch holding company may pay no corporate income tax on dividends received from the Chinese subsidiary. Moreover, by the tax treaty with the United States, the Netherlands may impose no WHT when the Dutch holding company remits dividends to the American parent. 6 Hong Kong and Singapore have tax treaties with China that reduce the WHT rates on dividends to 5 percent. Hong Kong and Singapore impose no corporate income tax on dividends earned abroad and impose no WHT on dividends. 6

7 In contrast to the case between the United States and China, depending on a pair of home and source countries, there may exist a tax-minimizing direct route in a tax treaty network. For example, the tax treaty between Japan and the United States allows the WHT rate on dividends to be 0 percent if the recipient of the dividends owns at least 50 percent of shares for 12 months. Thus, the direct route from the United States to Japan can be tax-minimizing. Hong (2018) constructs a network of tax treaties between 70 countries and develops computation algorithms to examine the structure of tax-minimizing (direct or indirect) investment routes in the network. Hong (2018) computes treaty network variables and finds significant relationships between FDI and the treaty network variables. The inward FDI stock via a tax-minimizing direct route is estimated to be about 2.14 times larger than the inward FDI stock via a direct route that is not tax-minimizing. However, from bilateral FDI data, we cannot observe the structure of investment routes for FDI. For instance, we cannot distinguish between direct investment from the Netherlands to China and indirect investment through the Netherlands to China, which originates from another country, such as the United States. We may be able to overcome such a limitation by using data on internal ownership structures of multinational corporations. 2.2 Multinational ownership chains Multinational corporations own and operate subsidiary companies in various countries around the world. While some multinational corporations tend to directly own their foreign subsidiaries, many other multinational corporations organize indirect ownership structures with holding companies. Multinational corporations often establish intermediate subsidiaries as holding companies, which own terminal operating subsidiaries. These related companies (parent or subsidiary) may be incorporated in different countries. Such hierarchical ownership structures can be viewed as a set of (direct or indirect) ownership chains from ultimate parents to terminal subsidiaries. 7

8 Figure 2. Multinational ownership chains Pfizer Inc. (United States) Starbucks Corp. (United States) C.P. Pharmaceuticals International CV (Netherlands) Pfizer Japan Inc. (Japan) Starbucks Asia Pacific Investment Holding II (Hong Kong) Starbucks CPG International GK (Japan) Pfizer International Trading Co., Ltd. (China) Starbucks Company Limited (China) Direct Ownership Figure 2 illustrates examples of ownership chains. Each arrow represents direct ownership. Pfizer Inc., an American pharmaceutical corporation, owns C.P. Pharmaceuticals International CV, a Dutch intermediate subsidiary, which owns Pfizer International Trading Co., Ltd., a Chinese terminal subsidiary. However, Pfizer Inc. directly owns Pfizer Japan Inc., a Japanese terminal subsidiary. Starbucks Corp., an American corporation operating a coffee chain store, owns Starbucks Asia Pacific Investment Holding II Limited, an intermediate subsidiary in Hong Kong, which owns Starbucks Company Limited, a Chinese terminal subsidiary. However, Starbucks Corp. directly owns Starbucks CPG International GK, a Japanese terminal subsidiary. Pfizer and Starbucks organize indirect ownership chains with foreign equity holding companies to Chinese terminal subsidiaries. The cross-border length of the ownership chains to the Chinese terminal subsidiaries is 2. However, Pfizer and Starbucks use direct ownership chains to Japanese terminal subsidiaries. The cross-border length of the ownership chains to the Japanese terminal subsidiaries is 1. Multinational corporations may want to minimize tax on dividends from foreign subsidiaries. Together with national tax laws of relevant countries, 8

9 tax treaties determine whether or not it is tax-minimizing for multinational corporations to use direct ownership chains (or direct routes) to terminal subsidiaries. As discussed before, the tax treaty between Japan and the United States allows direct ownership chains to be tax-minimizing between the two countries. Pfizer and Starbucks use direct ownership chains to terminal subsidiaries in Japan. However, the tax treaty between China and the United States does not allow direct ownership chains to be tax-minimizing. Moreover, it is tax-minimizing to use indirect ownership chains with foreign equity holding companies in the Netherlands or Hong Kong. Pfizer and Starbucks use such indirect ownership chains to terminal subsidiaries in China. Dyreng et al. (2015) examine foreign ownership chains of American multinational corporations. They discover that the WHT rate on dividends from source to home countries is positively related to the use of a foreign equity holding company in an ownership chain. In other words, as the WHT rate increases, multinational corporations are more likely to use foreign equity holding companies. However, there is no significant relationship between the existence of a tax treaty and the use of a foreign equity holding company. Lewellen and Robinson (2013) find that tax considerations, such as minimization of WHT imposed abroad, are important factors in organizing foreign ownership chains of American multinational corporations. Mintz and Weichenrieder (2010) find that German multinational corporations set up indirect ownership chains to foreign subsidiaries for tax motives. Previous studies tend to use only tax treaty dummies and WHT rates as explanatory variables. Indeed, the existence of a tax treaty can influence the organization of a multinational ownership chain as the tax treaty reduces the WHT rate on dividends. However, some tax treaties do not reduce WHT rates on dividends at all, and some other treaties do not reduce WHT rates effectively. Such tax treaties may not affect multinational ownership chains. To overcome such a limitation, we can use data generated from a network analysis of tax treaties. 9

10 Regarding the relationship between tax treaty networks and multinational ownership chains, I would like to address the following questions: Are multinational ownership chains less likely to include foreign equity holding companies, and less complex, if there is a tax-minimizing direct route? Are multinational ownership chains more likely to include foreign equity holding companies, and more complex, as a tax-minimizing indirect route reduces more tax than a direct route does? Are multinational ownership chains more likely to include foreign equity holding companies located in countries along tax-minimizing indirect routes? If multinational corporations consider tax as an important factor when organizing ownership chains to subsidiaries, the structure of tax-minimizing routes in a treaty network will be closely related to the structure of multinational ownership chains. 3 Research Design To begin with, I introduce definitions and notations for ownership chains and treaty networks. Within a multinational corporation, a set of ownership chains can describe how the multinational corporation owns its subsidiaries. A corporate entity is called a subsidiary if it is owned directly or indirectly by a multinational corporation. A subsidiary is called terminal if it owns no other subsidiaries. A subsidiary is called intermediate if it is not terminal, i.e., if it owns another subsidiary. It is worthwhile to note that, in an ownership chain, an intermediate subsidiary is often more than an equity holding company. An intermediate subsidiary may earn income from its own operations and from the operations of other subsidiaries it owns. An intermediate subsidiary can be thought of as an equity holding company if it earns income mostly from the operations of other subsidiaries it owns. However, from ownership data, it is not straightforward to determine whether an intermediate subsidiary is simply an equity holding company. Here I use these two terms interchangeably. Each ownership chain k connects a multinational corporation in country 10

11 i to a terminal subsidiary in country j directly or indirectly through intermediate subsidiaries. The multinational corporation is also called the ultimate parent. For each ownership chain k, country i denotes the home country, where the ultimate parent is incorporated. Country j denotes the source country, where the terminal subsidiary is incorporated. Given an ownership chain, a foreign intermediate subsidiary is a subsidiary incorporated in a country other than home and source. For each ownership chain k from the ultimate parent in country i to the terminal subsidiary in country j, F IS ijk denotes an indicator variable for the existence of a foreign intermediate subsidiary (FIS), or equivalently, a foreign equity holding company. Precisely, F IS ijk = 1 if there is a foreign intermediate subsidiary in chain k from country i to country j and F IS ijk = 0 otherwise. The complexity of an ownership chain is measured by its cross-border length, defined as the number of subsidiaries owned by a foreign entity in the chain. For each ownership chain k from the ultimate parent in country i to the terminal subsidiary in country j, LENGT H ijk denotes the cross-border length. If LENGT H ijk = 1, there is no foreign intermediate subsidiary in chain k from country i to country j, i.e., F IS ijk = 0. If LENGT H ijk 2, there is at least one foreign intermediate subsidiary in chain k from country i to country j, i.e., F IS ijk = 1. For each ownership chain k from the ultimate parent in country i to the terminal subsidiary in country j, F ISCO mijk denotes an indicator variable for the existence of a foreign intermediate subsidiary in country m. Precisely, F ISCO mijk = 1 if there is a foreign intermediate subsidiary in country m along chain k from country i to country j and F ISCO mijk = 0 otherwise. Hong (2018) conducts a network analysis of tax treaties. In a tax treaty network, each node is a country and each link has a weight that specifies the withholding tax rate on dividends paid across countries. An investment route is defined as a series of countries from a home country to a source country. When remitting after-tax profits as dividends from a source country to a home country, a multinational investor pays withholding taxes and corporate 11

12 income taxes in countries (in the reverse order) along an investment route. The foreign tax rate of an investment route is defined as a tax rate that combines these taxes in the countries along the route. An investment route is called tax-minimizing if it minimizes the foreign tax rate among all routes from home to source countries. Hong (2018) computes the following treaty network variables. For each of two countries i and j, DIRECT ij denotes an indicator variable for the existence of a tax-minimizing direct route. Precisely, DIRECT ij = 1 if there is a tax-minimizing direct route from country i to country j and DIRECT ij = 0 otherwise. The treaty shopping rate, denoted by T SHOP ij, is defined as the difference between the foreign tax rates of the direct route and a taxminimizing route from country i to country j. The treaty shopping rate shows how much a tax-minimizing route reduces more tax than a direct route does. For each of three countries m, i, and j, P T CO mij denotes an indicator variable for the existence of a tax-minimizing indirect route passing through a country. Precisely, P T CO mij = 1 if there is a tax-minimizing indirect route passing through country m from country i to country j and P T CO mij = 0 otherwise. In addition, T REAT Y ij is an indicator variable for the existence of a tax treaty between countries i and j. W HT ji is the withholding tax rate on dividends paid from an entity in country j to an entity in country i. To examine the factors that determine the inclusion of a foreign intermediate subsidiary in an ownership chain, I use probit regression models with the dependent variable F IS ijk, motivated by Dyreng et al. (2015), as specified in Equation (1). F IS ijk = β 0 + β 1 DIRECT ij + β 2 T SHOP ij + β 3 T REAT Y ij + β 4 W HT ji + γb ij + ηs j + ε ijk (1) Here B ij is a vector of bilateral variables to control for the relationship between countries i and j, S j is a vector of dummy variables for source country 12

13 j, and ε ijk is the error term. Note that the use of source country dummies eliminates the need for source-specific control variables, such as statutory tax rate, corruption perception index, and sovereign risk rating. If there is a tax-minimizing direct route from country i to country j, a multinational corporation in i can minimize tax on dividends by using a direct ownership chain to a terminal subsidiary in j without a foreign intermediate subsidiary. Thus, I expect that the existence of a tax-minimizing direct route is negatively related to the inclusion of a foreign intermediate subsidiary in an ownership chain. As the treaty shopping rate increases, by definition, the difference becomes larger between the foreign tax rates of the direct route and a taxminimizing route from country i to country j. Because a tax-minimizing route reduces more tax than the direct route does, a multinational corporation in i will consider a direct ownership chain to a terminal subsidiary in j less attractive than a tax-minimizing ownership chain with a foreign intermediate subsidiary. Thus, I expect that the treaty shopping rate is positively related to the inclusion of a foreign intermediate subsidiary in an ownership chain. To examine the factors that determine the complexity of an ownership chain, I use Ordinary Least Squares (OLS) regression models with the dependent variable LENGT H ijk, as specified in Equation (2). LENGT H ijk = β 0 + β 1 DIRECT ij + β 2 T SHOP ij + β 3 T REAT Y ij + β 4 W HT ji + γb ij + ηs j + ε ijk (2) Here B ij controls for the relationship between countries i and j, and S j controls for the characteristics of source country j. For the independent variables of interest, I expect that the relationship with LENGT H ijk remains similar to the relationship with F IS ijk. Notably, the existence of a tax-minimizing direct route is negatively related to the complexity of an ownership chain. The treaty shopping rate is positively related to the complexity. 13

14 To examine the factors that determine the location of a foreign intermediate subsidiary, I use probit regression models with the dependent variable F ISCO mijk, motivated by Dyreng et al. (2015), as specified in Equation (3). F ISCO mijk = β 0 + β 1 P T CO mij + β 2 T REAT Y im + β 3 W HT mi + β 4 T REAT Y mj + β 5 W HT jm + γb im + δb mj + ηp k + ε mijk (3) Here B im is a vector of bilateral variables to control for the relationship between countries i and m, and B mj for the relationship between countries m and j, where m is an intermediate country other than home and source. P k is a vector of dummy variables for ultimate parents. ε mijk is the error term. Note that the use of ultimate parent firm dummies eliminates the need for firm-specific control variables, such as firm size, profitability, and R&D spending. If there is a tax-minimizing indirect route passing through country m from country i to country j, a multinational corporation in i can minimize tax on dividends by using an indirect ownership chain to a terminal subsidiary in j with a foreign intermediate subsidiary in m. Thus, I expect that the existence of a tax-minimizing indirect route passing through a country is positively related to the location of a foreign intermediate subsidiary in the country. 4 Data To construct data on ownership chains, I use the Orbis database of Bureau van Dijk. 7 I focus on a sample of corporations that are constituents of the S&P 500 index but not classified as banks in the database. I also focus on cross-border ownership chains, i.e., ownership chains with ultimate parents and terminal subsidiaries incorporated in two distinct countries, to study the effect of tax treaties. 7 Accessed in October 2017 at orbis.bvdinfo.com 14

15 I construct data on ownership chains as follows: (i) For each corporation, or for each ultimate parent, find and list all terminal subsidiaries. (ii) For each terminal subsidiary, identify a corporate shareholder. If this corporate shareholder is an ultimate parent, complete an ownership chain. (iii) If not, the corporate shareholder is an intermediate subsidiary. Identify a corporate shareholder of this intermediate subsidiary and check whether or not it is an ultimate parent. (iv) Repeat the previous step until we reach an ultimate patent and complete an ownership chain. Overall, the data include 402 ultimate parents and 265,076 subsidiaries. Among these subsidiaries, I find 86,517 terminal subsidiaries and construct the same number of cross-border ownership chains. Table 1 shows the numbers of ultimate parents, intermediate subsidiaries, and terminal subsidiaries by countries of incorporation. For instance, there are 375 ultimate parents, 47,433 intermediate subsidiaries, and 5,600 terminal subsidiaries incorporated in the United States. Outside the United States, the United Kingdom and Japan host the largest numbers of intermediate and terminal subsidiaries. Except for these countries, Ireland, Bermuda, the Netherlands, Singapore, and Luxembourg tend to attract more intermediate subsidiaries than they attract terminal subsidiaries. In contrast, China, Mexico, Russia, Poland, and Malaysia tend to attract less intermediate subsidiaries than they attract terminal subsidiaries. For each of the 86,517 ownership chains, I measure the complexity of an ownership chain by using its cross-border length, denoted by LEN GT H and defined as the number of subsidiaries owned by a foreign entity in the chain. For each ownership chain, I also compute F IS, an indicator variable for the existence of a foreign intermediate subsidiary. Table 2 shows the numbers of ownership chains by F IS and LENGT H. There are 56,032 ownership chains (about 64.8 percent) with LEN GT H equal to one. Because these ownership chains cross a national border only once, they include no foreign intermediate subsidiaries, and thus, F IS = 0. There are 30,485 ownership chains (about 35.2 percent) with LEN GT H 15

16 Table 1. Numbers of firms by countries of incorporation Country Ultimate Parent Intermediate Subsidiary Terminal Subsidiary United States ,433 5,600 Ireland 11 5,015 2,067 United Kingdom 4 23,249 10,395 Bermuda 4 2, Switzerland 3 3,667 2,058 Netherlands 2 18,920 3,756 Singapore 1 2,600 1,143 British Virgin Islands Curacao Canada 0 2,378 5,103 Japan 0 22,032 4,295 China 0 1,111 4,163 Germany 0 4,384 3,992 France 0 6,151 3,277 Brazil 0 1,390 2,587 Italy 0 3,398 2,161 Australia 0 3,080 2,150 India 0 2,094 1,980 Mexico ,946 Spain 0 2,921 1,762 Luxembourg 0 11,211 1,316 Sweden 0 1,705 1,196 Russia ,172 Hong Kong 0 1,511 1,149 Poland ,070 Belgium 0 1,159 1,046 South Africa Cayman Islands 0 1, Malaysia New Zealand Total ,559 86,517 16

17 Table 2. Numbers of ownership chains by F IS and LENGT H F IS Ownership Chain Percent 0 56, , Total 86, LENGT H Ownership Chain Percent 1 56, , , , , Total 86, greater than one. Because these ownership chains cross a national border at least twice, they include at least one foreign intermediate subsidiary, and thus, F IS = 1. Some ownership chains exhibit long and complex structures. There are 4,919 ownership chains (about 5.7 percent) with LEN GT H greater than three. Each of these ownership chains includes at least three foreign intermediate subsidiaries from an ultimate parent to a terminal subsidiary. For each of the 30,485 indirect ownership chains, I find the countries where foreign intermediate subsidiaries are incorporated, to compute F ISCO, an indicator variable for the existence of a foreign intermediate subsidiary in a specific country. Table 3 presents the numbers of indirect ownership chains by countries of incorporation. For instance, there are 1,148 ownership chains with foreign intermediate subsidiaries in the Netherlands when American ultimate parents invest in British terminal subsidiaries. Also, there are 1,068 ownership chains through Luxembourg from the United States to the United Kingdom. Overall, the Netherlands and Luxembourg are the most frequent locations for foreign intermediate subsidiaries. Notably, when American ultimate parents invest in Chinese terminal subsidiaries, they frequently organize ownership chains with foreign intermediate subsidiaries in Hong Kong, the Netherlands, and Singapore. As mentioned in 17

18 Table 3. Numbers of indirect ownership chains by countries of incorporation Ultimate Parent Country Intermediate Subsidiary Country Terminal Subsidiary Country Indirect Ownership Chain United States Netherlands United Kingdom 1,148 United States Luxembourg United Kingdom 1,068 United States Netherlands Germany 598 United States Hong Kong China 588 United States Luxembourg Germany 474 United States Luxembourg France 418 United States Luxembourg Netherlands 413 United States Netherlands Italy 399 United States Netherlands France 397 United States United Kingdom Germany 389 Ireland Bermuda United States 380 Bermuda Ireland United States 374 United States Luxembourg Italy 373 United States Netherlands Brazil 363 United States Germany United Kingdom 303 Ireland Luxembourg United States 295 Bermuda Cayman Islands United States 292 United States Luxembourg Brazil 288 United States Netherlands Spain 286 United States Netherlands Poland 285 United States United Kingdom Netherlands 276 United States Switzerland United Kingdom 272 United States France United Kingdom 267 United States Netherlands China 254 United States Netherlands Russia 238 United States Netherlands Sweden 237 United States France Italy 235 United States Singapore China 231 United States Australia New Zealand 226 United States Bermuda United Kingdom Total 30,485 18

19 Section 2.1 and analyzed by Hong (2018), there are tax-minimizing indirect routes passing through Hong Kong, the Netherlands, and Singapore from the United States to China. From the database of Hong (2018), I obtain the treaty network variables DIRECT, T SHOP, and P T CO. In addition, I construct T REAT Y and W HT by using the information on tax treaties and national tax laws from Deloitte International Tax Source and PwC Worldwide Tax Summaries. 8 I use six bilateral variables to control for the relationship between a pair of countries. CONT IG, COMLANG, COMLEG, and COLONY are indicator variables for a shared border, a common official language, a common legal origin, and a colonial relationship, respectively. DIST is the populationweighted distance in thousands of kilometers. Head et al. (2010) use these bilateral variables to examine the patterns of international trade. I obtain the data for these variables from the CEPII. 9 In addition, RT A is an indicator variable for the existence of a regional trade agreement (RTA) between two countries. Egger and Larch (2008) examine the determinants of a regional trade agreement. I obtain the RTA data from Mario Larch s database. 10 Table 4 presents summary statistics for the variables in Equations (1) and (2). For about 35.2 percent of the 86,517 ownership chains, there is a foreign intermediate subsidiary, i.e., the mean of F IS is about The mean LEN GT H of ownership chains is about However, the median LENGT H is 1, as Table 2 implies. For about 74.5 percent of 82,274 ownership chains, there is a tax-minimizing direct route from home to source countries, i.e., the mean of DIRECT is about For about 86.1 percent of 82,274 ownership chains, there is a tax treaty between home and source countries, i.e., the mean of T REAT Y is about The mean W HT is about percent while the mean T SHOP is about percentage points. Thus, on average, a multinational corporation can reduce the tax rate on foreign-source dividends to percent (= ) by using 8 Accessed in March 2016 at dits.deloitte.com and taxsummaries.pwc.com 9 Accessed in September 2017 at cepii.fr/cepii/en/bdd modele/bdd.asp 10 Accessed in September 2017 at ewf.uni-bayreuth.de/en/research/rta-data 19

20 a tax-minimizing ownership chain. Note that the number of observations for DIRECT, T SHOP, T REAT Y, and W HT is less than the total number of ownership chains because these variables are constructed for pairs of 70 countries, as in Hong (2018). The number of observations for bilateral control variables is also less than the total number of ownership chains. Table 5 provides correlation coefficients between F IS, LEN GT H, and treaty variables. The inclusion of a foreign intermediate subsidiary in an ownership chain is negatively correlated with the existence of a tax-minimizing direct route and positively correlated with the treaty shopping rate. The complexity of an ownership chain, measured by the cross-border length, is negatively correlated with the existence of a tax-minimizing direct route and positively correlated with the treaty shopping rate. Note that DIRECT, T SHOP, and W HT are highly correlated with each other. Table 6 presents summary statistics for the variables in Equation (3). For each of the 30,485 ownership chains with foreign intermediate subsidiaries, I find the countries where the foreign intermediate subsidiaries are incorporated. Note that an ownership chain can include foreign intermediate subsidiaries in more than one country. There are 46,638 distinct combinations of ownership chains and countries with foreign intermediate subsidiaries. For each of such combinations, I consider 70 countries for possible locations of foreign intermediate subsidiaries and construct about 68 observations. 11 Among these 68 observations, one is for the actual country m where a foreign intermediate subsidiary is incorporated, i.e., F ISCO m = 1, while the others are for the countries where foreign intermediate subsidiaries could have been incorporated but were not, i.e., F ISCO m = 0. Consistent with this construction, Table 6 shows that the mean of F ISCO is about 0.015, which is approximately equal to 1/ Precisely, for each combination of an ownership chain and a country, if both home and source countries belong to the set of the 70 countries (about 93.2 percent of the 46,638 combinations), to avoid repetition, 68 observations are created. If one is in the 70 countries but the other is not (about 6.6 percent), 69 observations are created. If none of home and source countries belong to the set of the 70 countries (about 0.2 percent), 70 observations are created. 20

21 Table 4. Summary statistics for F IS, LENGT H, and other variables Variable Description Obs Mean SD Min Max F IS 1 if there is a foreign 86, intermediate subsidiary in the chain LEN GT H number of subsidiaries 86, owned by a foreign entity in the chain DIRECT ij 1 if there is a tax-minimizing 82, direct route from i to j T SHOP ij difference between the foreign 82, tax rates of the direct route and a tax-minimizing route T REAT Y ij 1 if there is a tax treaty 82, between i and j W HT ji withholding tax rate on 82, dividends paid from j to i CONT IG ij 1 if i and j share a national 82, border COMLANG ij 1 if i and j share a common 82, official language COMLEG ij 1 if i and j share a common 82, legal origin COLONY ij 1 if i and j experience 82, a colonial relationship DIST ij population-weighted distance 82, between i and j in 1,000 km RT A ij 1 if there is a regional trade 82, agreement between i and j Table 5. Correlations between F IS, LENGT H, and treaty variables F IS LENGT H DIRECT ij T SHOP ij T REAT Y ij W HT ji F IS *** *** 0.112*** *** 0.072*** LENGT H 0.976*** *** 0.173*** *** 0.126*** DIRECT ij *** *** *** *** *** T SHOP ij 0.076*** 0.090*** *** *** 0.815*** T REAT Y ij *** *** *** 0.009*** *** W HT ji 0.032*** 0.045*** *** 0.768*** 0.047*** - Note: Pearson coefficients are presented above the diagonal and Spearman coefficients below. ***, **, and * denote significance at 1%, 5%, and 10%, respectively. 21

22 Table 6. Summary statistics for F ISCO, P T CO, and other variables Variable Description Obs Mean SD Min Max F ISCO m 1 if there is a foreign 3,174, intermediate subsidiary in country m in the chain P T CO mij 1 if there is a tax-minimizing 3,174, indirect route passing through country m from i to j T REAT Y im 1 if there is a tax treaty 3,015, between i and m W HT mi withholding tax rate on 3,015, dividends paid from m to i CONT IG im 1 if i and m share a national 3,015, border COMLANG im 1 if i and m share a common 3,015, official language COMLEG im 1 if i and m share a common 3,015, legal origin COLONY im 1 if i and m experience 3,015, a colonial relationship DIST im population-weighted distance 3,015, between i and m in 1,000 km RT A im 1 if there is a regional trade 3,015, agreement between i and m T REAT Y mj 1 if there is a tax treaty 2,927, between m and j W HT jm withholding tax rate on 2,927, dividends paid from j to m CONT IG mj 1 if m and j share a national 2,927, border COMLANG mj 1 if m and j share a common 2,927, official language COMLEG mj 1 if m and j share a common 2,927, legal origin COLONY mj 1 if m and j experience 2,927, a colonial relationship DIST mj population-weighted distance 2,927, between m and j in 1,000 km RT A mj 1 if there is a regional trade 2,927, agreement between m and j 22

23 Table 7. Correlations between F ISCO, P T CO, and treaty variables F ISCO m P T CO mij T REAT Y im W HT mi T REAT Y mj W HT jm F ISCO m *** 0.032*** *** 0.028*** *** P T CO mij 0.043*** *** *** 0.124*** *** T REAT Y im 0.032*** 0.077*** *** 0.397*** *** W HT mi *** *** *** *** 0.035*** T REAT Y mj 0.028*** 0.126*** 0.397*** 0.061*** *** W HT jm *** *** *** 0.066*** *** - Note: Pearson coefficients are presented above the diagonal and Spearman coefficients below. ***, **, and * denote significance at 1%, 5%, and 10%, respectively. For each observation with country m, I use the database of tax-minimizing routes computed by Hong (2018) and check the existence of a tax-minimizing indirect route passing through country m from home to source countries. Mostly, tax-minimizing indirect routes pass through only one country. On average, there are about 4 tax-minimizing indirect routes from home to source countries. Table 6 shows that the mean of P T CO is about 0.059, which is approximately equal to 4/68. T REAT Y im, W HT mi, and bilateral variables, such as CONT IG im, are defined for a pair of home country i and intermediate country m. There is a tax treaty between about 70.6 percent of home and intermediate countries. On average, the withholding tax rate on dividends paid from intermediate to home countries is about percent. T REAT Y mj, W HT jm, and bilateral variables, such as CONT IG mj, are defined for a pair of intermediate country m and source country j. There is a tax treaty between about 78.2 percent of intermediate and source countries. On average, the withholding tax rate on dividends paid from source to intermediate countries is about percent. Table 7 provides correlation coefficients between F ISCO, P T CO, and treaty variables. The location of a foreign intermediate subsidiary in a country is positively correlated with the existence of a tax-minimizing indirect route passing through the country. 23

24 5 Results Table 8 presents the probit regression results for Equation (1) to examine the factors that determine the inclusion of a foreign intermediate subsidiary in an ownership chain. Each regression contains one of the four treaty variables as the main independent variable. 12 Each regression also contains the bilateral variables and the source country dummies. Standard errors clustered by source country are reported in parentheses below each estimate. 13 In column (1) of Table 8, as expected, the existence of a tax-minimizing direct route is negatively and significantly related to the inclusion of a foreign intermediate subsidiary in an ownership chain. The existence of a taxminimizing direct route (i.e., moving DIRECT ij from 0 to 1) is estimated to decrease the probability of using a foreign intermediate subsidiary in an ownership chain by about 10.4 percentage points. In column (2), as expected, the treaty shopping rate is positively and significantly related to the inclusion of a foreign intermediate subsidiary in an ownership chain. Moving T SHOP ij from the 25th percentile to the 75th percentile of its distribution (i.e., moving T SHOP ij from 0 to 4) is estimated to increase the probability of using a foreign intermediate subsidiary in an ownership chain by about 6.4 percentage points. In column (3), the existence of a tax treaty is negatively and significantly related to the inclusion of a foreign intermediate subsidiary in an ownership chain. This finding is rather surprising because Dyreng et al. (2015) report no significant relationship between tax treaties and foreign intermediate subsidiaries. Consistent with the findings of Dyreng et al. (2015), in column (4), the withholding tax rate on dividends paid from source to home countries is positively and significantly related to the inclusion of a foreign intermediate subsidiary in an ownership chain. 12 Due to the high correlations between the treaty variables noted in Table 5, I do not use them together as independent variables. 13 It does not change the results qualitatively to adjust standard errors for clustering at the ultimate parent firm level. 24

25 Table 8. Tax treaties and the inclusion of FIS (1) (2) (3) (4) F IS F IS F IS F IS DIRECT ij ** (0.155) T SHOP ij 0.050*** (0.004) T REAT Y ij *** (0.193) W HT ji 0.048*** (0.004) CONT IG ij ** ** * ** (0.474) (0.469) (0.504) (0.466) COMLANG ij 0.445* (0.253) (0.236) (0.260) (0.233) COMLEG ij (0.174) (0.160) (0.183) (0.160) COLONY ij (0.106) (0.091) (0.073) (0.097) DIST ij ** ** ** ** (0.042) (0.029) (0.034) (0.029) RT A ij (0.246) (0.200) (0.217) (0.200) Constant *** (0.465) (0.330) (0.468) (0.340) Observations 82,178 82,178 82,178 82,178 Pseudo R Note: All regressions include source country dummies. Standard errors clustered by source country are in parentheses. ***, **, and * denote significance at 1%, 5%, and 10%, respectively. 25

26 Table 9. Tax treaties and the complexity of an ownership chain (1) (2) (3) (4) LENGT H LENGT H LENGT H LENGT H DIRECT ij *** (0.082) T SHOP ij 0.059*** (0.003) T REAT Y ij *** (0.186) W HT ji 0.057*** (0.003) CONT IG ij *** *** *** *** (0.181) (0.164) (0.206) (0.161) COMLANG ij 0.281*** 0.122* * (0.095) (0.071) (0.115) (0.069) COMLEG ij (0.078) (0.060) (0.102) (0.060) COLONY ij ** (0.065) (0.066) (0.046) (0.078) DIST ij * ** *** ** (0.041) (0.025) (0.028) (0.025) RT A ij (0.224) (0.160) (0.168) (0.160) Constant 3.058*** 2.591*** 3.825*** 2.601*** (0.337) (0.194) (0.362) (0.195) Observations 82,178 82,178 82,178 82,178 R Note: All regressions include source country dummies. Standard errors clustered by source country are in parentheses. ***, **, and * denote significance at 1%, 5%, and 10%, respectively. 26

27 Table 10. Tax treaties and the location of FIS (1) (2) (3) (4) F ISCO m F ISCO m F ISCO m F ISCO m P T CO mij 0.223*** 0.098*** (0.024) (0.023) T REAT Y im 0.236** 0.226* (0.120) (0.115) W HT mi ** ** (0.008) (0.008) CONT IG im * (0.181) (0.226) (0.199) (0.231) COMLANG im 0.252*** 0.263*** 0.293*** 0.285*** (0.045) (0.045) (0.043) (0.045) COMLEG im 0.153*** 0.232*** 0.136*** 0.202*** (0.044) (0.040) (0.050) (0.049) COLONY im 0.391*** 0.349*** 0.396*** 0.338*** (0.066) (0.067) (0.069) (0.070) DIST im *** *** *** *** (0.009) (0.011) (0.011) (0.013) RT A im *** *** *** *** (0.096) (0.101) (0.096) (0.096) T REAT Y mj 0.144** 0.115** (0.064) (0.058) W HT jm *** *** (0.003) (0.003) CONT IG mj 0.111*** 0.101*** 0.105*** 0.101*** (0.030) (0.031) (0.030) (0.030) COMLANG mj 0.325*** 0.353*** 0.317*** 0.345*** (0.049) (0.046) (0.051) (0.046) COMLEG mj 0.107*** 0.111*** 0.108*** 0.116*** (0.025) (0.026) (0.027) (0.026) COLONY mj * (0.077) (0.076) (0.086) (0.084) DIST mj *** *** *** *** (0.005) (0.007) (0.005) (0.006) RT A mj ** * (0.040) (0.040) (0.037) (0.034) Constant *** *** *** *** (0.096) (0.082) (0.102) (0.096) Observations 2,867,964 2,867,964 2,867,964 2,867,964 Pseudo R Note: All regressions include ultimate parent firm dummies. Standard errors clustered by ultimate parent firm are in parentheses. ***, **, and * denote significance at 1%, 5%, and 10%, respectively. 27