Repatriation taxes and foreign direct investment: Evidence from tax treaties∗ Michael Smart University of Toronto† June 28, 2010 Abstract This paper examines the effects on foreign direct investment of worldwide versus territorial tax systems. Canada operates a hybrid system of international taxation, under which dividends repatriated from affiliates located in tax treaty partner countries are exempt from domestic taxation, while repatriations from non-treaty countries are subject to a foreign tax credit regime similar to that of the United States. Exploiting the within-country variation in tax rates on repatriations induced by new tax treaties signed by Canada since 1975, I estimate the impact of repatriation taxes on foreign direct investment of Canadian multinational corporations. According to my estimates, outbound FDI rises 80 per cent on average following exemption, and the resulting tax rate elasticity is large and significant. I discuss the implications of the results for optimal tax policy.
1
Introduction
Policies for the taxation of multinational corporation (MNC) profits are currently in a state of ferment, with substantial reforms underway in the tax treatment of foreign-source income in several countries. The United Kingdom and Japan have recently abandoned the principle of worldwide income taxation for resident corporations, choosing instead to exempt much foreign-source business income from tax—in effect moving from a residence basis to a territorial basis for taxation. Meanwhile, the Obama administration in the United States has proposed reforms that would strengthen its system of residence-based taxation, through measures that would limit domestic deductions and foreign tax credits available for certain transactions ∗ Thanks
to Gustavo Bobonis for comments and Andrew Bird for research assistance. of Economics, University of Toronto, and Oxford Centre for Business Taxation,
[email protected] † Department
1
with controlled foreign affiliates. Related reforms are occurring in Canada, Australia, and other countries. These reforms are naturally not without some controversy and, as the preceding two examples suggest, it is still unclear whether the world may be moving towards greater use of territorial taxation, or towards a strengthened version of the residence principle. Understanding the economic effects of taxation of cross-border investments is therefore of considerable interest. In principle, the effects on MNC behaviour of a system of residence-based taxation with foreign tax crediting appear straightforward. Under the credit system, foreign-source income is subjected to taxation at the home country tax rate, while a credit is offered for taxes paid in the host country.1 Consequently, the credit system has the potential to deter investment by MNCs into low-tax host countries. At the same time, however, the credit system should make MNCs less sensitive to tax rate differentials among host countries, and so less inclined to engage in tax planning activities that shift taxable income from affiliates in high-tax host countries to low-tax countries. This paper seeks to provide new evidence on the effects of foreign tax credit regimes of a particularly direct kind, based on changes in the tax treatment of foreign-source income facing Canadian MNCs. Canada operates a hybrid system of international taxation, under which dividends repatriated from affiliates located in tax treaty partner countries are generally exempt from domestic taxation, while repatriations from non-treaty countries are subject to a foreign tax credit regime similar to that of the United States. Moreover, the number of countries covered by the exemption has increased from 11 in 1975 to 91 today. Exploiting the within country-year variation in tax rates on repatriations induced by new tax treaties signed since 1975, I estimate the impact of home country taxes on outward foreign direct investment of Canadian MNCs. Based on both theoretical and empirical considerations, there are reasons to believe that the differences in effect between exemption and credit regimes are smaller than suggested by the simple model sketched above. On theoretical grounds, analysis of residence-based taxes must be modified to reflect the fact that home-country taxes are generally deferred until foreignsource income is repatriated to the parent, rather than due on accrual. Hartman (1985) pointed out that such a tax on repatriations should in principle 1 This ignores the fact that foreign tax credits generally cannot exceed the domestic tax lia-
bility, so that incentives for firms with “excess” foreign tax credits are different. Furthermore, in most cases, the home country tax liability is deferred until income is actually repatriated to the parent which as Hartman (1985) has argued is apt to change the effects of home country taxes substantially. This issue is addressed in detail below.
2
have no effect on MNC decisions about affiliate investments that may be financed with retained earnings, since in choosing affiliate retentions MNCs effectively face a choice between paying repatriation taxes now, or paying repatriation taxes of the same present value later. Furthermore, as pointed out by Grubert (1998) and others, MNCs have access to a number of alternative financial strategies that may permit them to avoid dividend repatriation taxes and which may therefore cause the reach of a system of worldwide income taxation like that of the US to be shorter than it seems. The effects of host country taxes on international investment patterns, and the difference between exemption and credit regimes in particular, has been the subject of much previous research. Perhaps most notable is the work of Hines (1996), who examined how differences in tax rates of US states were related to the pattern of inward FDI from credit and exemption countries in a single year. My approach is similar, but it relies on estimate the difference in investment changes in reforming and non-reforming countries following new treaties, which has the potential to control better for other confounding influences on investment patterns. Section 2 of the paper provides a more detailed of international taxation issues and the previous literature, and Section 3 of the Canadian tax rules for foreign-source income. Section 4 discusses the data and estimation strategy, and Section 5 presents results. The paper concludes with a brief discussion of the implications of the results for the optimal tax treatment of foreign source income.
2
Taxation and foreign investment: An overview
2.1 International tax rules Investments by multinational corporations are generally subject to tax in the host countries in which the income is earned, but also typically in the home country in which the multinational is headquartered or otherwise deemed to be resident for tax purposes. Thus a US-based corporation operating in Canada must generally pay tax on its Canadian-source income to Canadian governments, but ultimately to the US government as well. This combination of tax on the source and residence principles has the potential to expose cross-border investments to higher rates of tax than purely domestic ones. Recognizing this, most governments have instituted systems for the alleviation of double taxation on international investment. One common such method is simply to exempt foreign-source income from taxation in the home country. This system is used by many OECD countries, including France, Germany and, for most investments, Canada. When
3
exemption is used, the effective rate of tax on an international investment is simply the relevant host country corporate tax rate, plus any withholding taxes due to the host country when and if income from the investment is paid to non-residents of the host. Exemption-based tax systems are often called “territorial taxation” because there is no attempt to tax worldwide income of domestic resident corporations. Other countries do attempt to tax worldwide income but alleviate double taxation by allowing a credit for foreign taxes paid. This system is used by the United States and, until recently, by the United Kingdom and Japan. Under the credit system, the foreign tax credit generally cannot exceed the tax that is due on the income in the home country. As a result, the effective tax rate on foreign source income in the credit system is generally the greater of the tax rates that apply in the host and home countries.2 For example, a Canadian affiliate of a US multinational generally may pay tax in Canada at a statutory rate of 31 per cent, while facing the top tax rate of 39 per cent in the US. Such a corporation then pays 31 per cent of income in taxes to Canada, receives a credit for the full amount and pays tax at the residual rate of 8 per cent (of the pre-tax affiliate income) in the United States. An affiliate of a US corporation operating in Japan, where the top tax rate of 40 per cent exceeds the US rate, would pay no tax at home on affiliate income. The examples make it clear that international tax system can in principle have substantial effects on the allocation of multinational investment between domestic and foreign activities, and among foreign affiliates located in different host countries. Since under the credit system all income is, in the deficient credit case at least, subject to tax at the home country rate, the firm is indifferent based on tax considerations to investing at home or in any host country, a condition that is consistent with capital export neutrality and the maximization of worldwide income. On the other hand, firms resident in high-tax credit countries face higher taxes than purely domestic firms in low-tax host countries, which may distort international cross-ownership of assets (Desai and Hines, 2003). In contrast, international investments of a firm operating under the exemption system are apt to be sensitive to host 2 A firm whose foreign tax credits exceed its home tax liabilities, for which the host country
tax is the final tax, is sometimes said to be in “excess credit” position. A firm whose domestic liabilities exceed foreign tax credits, so that tax is due in the home country, is in “deficient credit” position. In the United States, where these concepts are of greatest importance to the tax system, foreign tax credits may be pooled among affiliates operating in different countries, so that the process of computing repatriation taxes is somewhat more complicated than this simple comparison of home and host statutory tax rates would suggest.
4
country tax differences. Home country taxes due on active business income earned by foreign affiliates can generally be deferred until the income is repatriated to the parent corporation in the form of intercorporate dividends or other payments. The deferral advantage may have profound implications for the effects of home country taxation on international investment. Hartman (1985) argued that deferral makes home country taxation irrelevant to investments financed by retaining earnings in the affiliate, since in choosing affiliate retentions MNCs effectively face a choice between paying repatriation taxes now, or paying repatriation taxes of the same present value later. Consequently, the tax reduces the opportunity cost of investment and its return by the same amount. Repatriation taxes should therefore have no impact on the timing of dividend repatriations, and hence on reinvestment, but instead be capitalized into the value of the firm’s initial investments in the affiliate.3 The Hartman model is an application of the “trapped equity” view (Auerbach, 1979) that asserts the neutrality of dividend taxes for retained earnings more generally. Subsequent theoretical and empirical research has investigated how strategies available for avoiding repatriation taxes may affect firms’ investment strategies and thus overturn the Hartman neutrality result. Nevertheless, there remain reasons to think that the effect on investment patterns of repatriation taxes operating through credit-and deferral systems of worldwide taxation are small, for investments of “mature” foreign affiliates at least. 2.2 Empirical literature While the empirical literature on FDI and taxation (recently surveyed by Feld and Heckemeyer (2009) is voluminous, most of it is concerned with identifying host country effects on investment, rather than home country effects. Devereux and Freeman (1995) however find significant effects on the allocation of outbound FDI flows of a sophisticated measure of the tax rate applying for bilateral flows between pairs of home and host countries, for a panel of OECD countries. Still, their work does not examine the effects of home country repatriation taxes per se, nor the differential effects of credit and exemption methods. Slemrod (1990) appears to be the first to have observed that the sensitivity of foreign investment to host tax rates should be greater for MNCs based in exemption countries (with no repatriation tax) than for those from credit 3 The prospect of repatriation taxes therefore does affect the initial investment in the
affiliate and the speed at which it reaches a point of “maturity” at which further investment is financed through retentions (Hines, 1994).
5
countries (with positive repatriation taxes in the excess credit case). Slemrod’s test based on aggregate inward FDI to the US from seven countries in the 1962-87 period found no significant difference between credit and exemption countries. Hines (1996) applied the same test, but used crosssectional variation in the host tax rate among the 50 US states and inward FDI from the same seven countries in 1987, rather than the intertemporal variation exploited by Slemrod. Hines in contrast found significant differences in investment patterns, with investment from exemption countries disproportionately attracted to low-tax US states, compared to that of credit countries. It is obvious that these results cannot be construed as definitive tests of the theory. The Slemrod and Hines papers in particular rely on data for a very limited set of countries and a limited form of variation (time series or cross-sectional) in repatriation tax rates and regimes that is apt to be correlated with other determinants of investment in the sample. The metaanalysis of de Mooij and Ederveen (2003) suggests that estimated tax rate elasticities are generally higher in cross-sectional studies. But the reliance on cross-sectional variation in many studies is unsurprising, since wholesale reforms to countries’ systems of international taxation are rarely observed. A second branch of the empirical literature examines how the foreign credit regime affects tax planning activities of multinationals, rather than their real investment decisions. Most notably, in a series of thoughtful papers, Rosanne Altshuler, Harry Grubert, Jim Hines and co-authors have examined how real-world aspects of the US deferral-and-credit regime have affected the financial and organizational strategies of US controlled foreign corporations (CFCs), and how this in turn may mitigate the hypothesized effects of repatriation taxes on investment flows. Thus Hines (1994) offers a fully developed model of the dynamics of investment in and repatriations from a foreign affiliate that is based on the Hartman (1985) model, but which is extended to examine how host country borrowing may substitute for equity injections from the home country. Hines then examines correlations between host tax rates and interest expenses of US CFCs, based on 1984 tax returns data, and finds patterns that are consistent with the theory. Related, examining 1990 tax data, Grubert (1998) finds that CFCs facing high effective tax rates on dividend repatriations make greater use of royalty and interest payments in place of dividends to transfer funds back to the parent. Repatriation tax rates however have no impact on the total amount repatriated through all three forms. Altshuler and Grubert (2003) extend the analysis to “triangular” repatriation strategies involving affiliates in a high-tax host country and in a tax haven and 6
again find greater use of repatriation-tax avoidance strategies when repatriation taxes are high, in this case based on a 1996 cross-section of tax returns.
3
International taxation in Canada
Canada’s tax system for foreign affiliates of domestic corporations is a hybrid of the exemption and credit methods. A foreign affiliate is defined as a foreign corporation that is at least 10 per cent owned by a Canadian corporation or related parties. When a dividend is received from an affiliate out of the active business income earned in a country with which Canada has a tax treaty, the dividend is said to be paid out of “exempt surplus”, and the income is exempt from taxation in Canada. If the income is earned in a country with which no treaty is in force, the dividend is instead from “taxable surplus” and tax is due on the dividend in Canada, and the foreign tax credit system is applied. As discussed, foreign tax credits are limited to the Canadian tax liability on the income, but unused credits may be carried back three years and forward up to seven years. Different rules apply to “passive” income of foreign affiliates, known as Foreign Accrued Property Income (FAPI). Analogous to the US subpart F rules for controlled foreign corporations, in Canada FAPI is taxed on an accrual basis in the hands of the Canadian parent, a device to limit tax avoidance by Canadian corporations through deferral. Similarly, income of unincorporated foreign branches is consolidated with parent income and taxed on accrual.4 In contrast to some other exemption countries, Canada has essentially no rules to govern the allocation of deductible expenses between taxable income of parents and exempt income of affiliates in treaty countries. This has raised a concern that affiliate expenses, especially interest expense, may be shifted to the parent to reduce domestic tax liabilities. Because of the way expense shifting interacts with exemption, the effective tax rate on outbound investments may be substantially below that applying to domestic investments, and even lower than a simple comparison of statutory tax rates would suggest (Mintz, 2004).5 4 In 2007, the exempt surplus policy was amended, with the exemption to be extended also to affiliates in countries with which Canada has concluded a Tax Information Exchange Agreement (TIEA), as well as to treaty countries. At the same time, in a measure aimed at tax havens, if Canada invites a country to sign a TIEA and no agreement has been signed after 60 months, affiliate income is then to be treated as FAPI and taxed on an accrual basis. 5 The federal government in 2007 passed Section 18.2 legislation that would restrict interest deductibility for so-called “double dip” transactions of this kind, but the legislation was repealed by the 2009 Budget, before it became effective.
7
This hybrid system was introduced with effect from 1976,6 replacing an earlier system which exempted essentially all foreign-source income, regardless of the treaty status of the host country. Prior to 1976 Canada had concluded tax treaties with just 11 countries, all of which were major trading partners in the OECD, and most of which had high rates of corporate taxation comparable to Canada’s. By 2009, Canada’s tax treaty network had grown to include 91 countries, several of which levy corporate tax rates substantially lower than Canada’s, including a few acknowledged tax havens (see Table 1 and Figure 1). The main empirical objective of the paper is therefore to examine how the extension of treaties and the resulting fall in repatriation taxes have affected the pattern of outbound foreign direct investment from Canada. Since my research design involves viewing tax treaties as a quasi-experiment in international tax reform, it is important to understand the objectives behind Canada’s hybrid tax system and its treaty negotiations. For this approach to be valid, the decision to sign a treaty must be unrelated to other factors causing the stock of investment to change around the treaty inception date. The 1969 White Paper that promulgated the current approach suggested that goal of exemption was not to provide selective investment subsidies for certain outbound investments but conversely “neither to provide an incentive to Canadians to invest abroad, nor to place a barrier in the way of their doing so.”7 While there is some suggestion that some treaties have been concluded with developing countries as a means of “tax sparing” to encourage investment and development (Arnold, 2002), this idea is disputed. Instead, the main objectives of treaty negotiations appear to be to provide a basis for information sharing with other countries, to facilitate inbound investment, and to achieve comparable treatment among all foreign affiliates of Canadian corporations. Notwithstanding the tax differences, foreign direct investment into both treaty and non-treaty countries is quantitatively important, and investment into non-treaty tax havens has grown substantially in recent decades. Table 2 reports the value and shares of FDI into the top 15 host countries from official data for 2009 and for 1989. The share of investment stocks reported for low-tax treaty countries, for which the exemption applies, has grown substantially, to 16.6 per cent of FDI in 2009 from just 3.0 per cent in 1989.8 Nevertheless, foreign direct investment in non-treaty credit coun6 The policy was enacted in 1972. 7 Quoted in Arnold (2002). 8 The main treaty destinations with top tax rates at least 10 per cent below Canada’s were
8
tries remains important, amounting to about seven per cent of FDI stocks in 2009. While the focus of this paper is on the investment effects of repatriation taxes, the amount of dividends repatriated from active business income in non-treaty countries is in fact small (Department of Finance, 1998). This points to the importance of the deferral advantage in practice, but also to the role of conduit entities in non-treaty countries that may repatriate dividends via transactions with affiliates in treaty countries, rather than directly, which are then treated as exempt surplus of the Canadian parent. Since the home country tax consequences of repatriation differ between treaty and non-treaty countries, Canada’s system may create greater incentives for treaty shopping of foreign affiliates than other countries’.9 As a result, my estimates of the effects of new treaties may reflect reallocation of the assets of foreign affiliates among countries, as well as the effect of the change in the cost of capital on the desired level of total outbound investment.
4
Estimation and data
4.1 Model and estimation The foregoing discussion suggested a simple, testable model of how the interaction of host and home country taxation determines the effective tax rate that is applied to the income of foreign affiliates, and how the effective tax rate in turn is hypothesized to influence outbound investment. Consider a direct equity injection into a foreign affiliate in a host country i , which pays income to the parent in the form of a dividend. The host country corporate tax rate is τhi and the home country tax rate, applicable in the case of residence taxation is τr . In addition, the host country levies a withholding tax w i on dividends repatriated. If the home country applies the exemption or territorial taxation method—as is the case in Canada where a bilateral tax treaty is in force—then there is no tax due in the home country on repatriation, and the effective tax rate on pre-tax income earned by the foreign affiliate is the sum of the host country source tax rate and the net withholding tax rate, say τei = τhi + w i (1 − τhi ). If the home country uses the deferral-and-credit system of worldwide taxation—applicable in Canada when no treaty is in place—then the dividend is grossed up by the amount of tax paid to the host and the full amount included in home country taxable income on repatriation of the dividend. However, the credit cannot exceed Barbados, Ireland, Hungary, and Switzerland in 2009, and Barbados and Singapore in 1989. 9 For a discussion of treaty shopping more particularly in the context of inbound investments, see Duff (2010).
9
the home rate of tax τr . Consequently the effective rate of tax on affiliate pre-tax income in the no-treaty case is τei = max{τr , τhi + w i (1 − τhi )}. For the empirical work, it is useful to work in terms of the host tax rate h τi and the repatriation tax τ∗i = τei −τhi . Combining the preceding cases gives the repatriation tax rate (
τ∗i
=
w i (1 − τhi ) © ª max τr − τhi , w i (1 − τhi )
if i has a tax treaty otherwise.
(1)
To examine the effect of host and repatriation taxes on investment, consider a (mildly structural) estimating equation of the form log K i t = α + βτhit + γτ∗i t + θx i t + u i t
(2)
where K i t is the natural logarithm of the stock of outbound foreign direct investment from Canada into i in year t , x i t a vector of control variables, u i t an error term, and the tax rates τh and τ∗ are now indexed by the subscripts i t , as they are permitted to vary across countries and over time within the panel data set. In this framework, the canonical theories of tax effects can be represented as testable parametric restrictions on the estimated semi-elasticities of investment, β and γ (for host and repatriation taxes respectively). If investment responds to the combined effective tax rate τei without regard to the potential for deferral of repatriation taxes, then β = γ. In contrast, under the Hartman model, for foreign investment financed through retained earnings, repatriation taxes are neutral, and γ = 0. Note in particular the symmetry of the two hypotheses. In the former case, investment allocation is held to be unaffected by host country taxes, to the extent that host taxes are lower than home taxes, so that firms are in “deficient credit” status. In the latter case, in contrast, investment is held to be unaffected by home country taxes, but host country tax differences do matter. Estimating tax elasticities from models like (2) raises a number of empirical issues. In my panel data set, the variation in tax rates is between countries and over time, due to differences in local tax rates between host countries, and variation in Canadian and host tax rates over time. Moreover, since the repatriation tax is positive only for low-tax countries (for which τhit + w i t (1 − τhit ) < τrt ), there is independent variation in the two tax rates. Crucially, because the two tax rates vary over time within a country, (2) can be estimated while allowing unobservable fixed effects in the stock of investment for each host country i and year t into the error term. Including country effects means that the tax elasticity estimates will not be 10
influenced by cross-sectional differences between countries—for example, high-tax countries tend to be attractive destinations for investment because of non-tax factors that are difficult to measure. Including year effects in turn purges the influence of aggregate correlation between declining tax rates and rising investment over time. Apparently, most previous studies of home country effects on investment have not controlled for a full set of unobservable time and geographic fixed effects. Notwithstanding the panel data approach, least squares estimates of tax rate elasticities from (2) are apt to be biased. This may be due to more subtle forms of omitted variable bias than the unobservable fixed effects, or to simple measurement error. For example, reductions in host country tax rates within a country may occur contemporaneously with other reforms and other economic developments that influence inward foreign direct investments from all countries, regardless of how they tax repatriations. The marked reduction in corporate tax rates in Ireland of a decade ago, and the coincident rise in foreign investment, is a notable example of this problem. Problems of measurement error are likewise endemic in aggregate tax rate regressions of this kind. In the empirical work that follows, I calculate the tax rate variables using the top statutory corporate tax rates applying in the relevant year in the host country and in Canada. While this approach is common, it is understood that investment patterns may be better explained in some cases by marginal effective tax rates on capital, or perhaps by effective average tax rates (Devereux and Griffith, 2003), but these alternatives are in general difficult to measure correctly for the representative investment project.10 Even where income shifting considerations suggest that it is the difference in statutory tax rates that is most relevant to the analysis, the host country’s top statutory rate may be less relevant than special rates that apply to certain industries, to holding company structures, or to certain regions of the country. The direction of bias in estimated tax rates elasticities is in principle unknown. The conventional result on attenuation bias in the presence of a single variable measured with error does not apply in this case, since measurement error in the host tax rate and the repatriation tax rate are apt to be negatively correlated. Furthermore, omitted variables may lead to upward bias in elasticity estimates, at least in the simple example of related changes to tax and non-tax investment policies discussed above. 10 Even more commonly in the literature, the effective tax rate is defined to be the ratio of tax revenues to a measure of pre-tax income, which raises endogeneity issues that parallel closely the problems we discuss here.
11
The solution to these problems explored in this paper is to construct instrumental variables estimates of tax rate elasticities, based on the differencein-differences in repatriation tax rates, and in investment, that are associated with changes over time in Canada’s tax treaty network. Let D i t be an indicator equal to one if there is a tax treaty between Canada and country i in force in year t . Given Canada’s double taxation rules, D i t is certainly correlated with τ∗i t . As I will argue below, controlling for tax rates and for country and year fixed effects, there is little reason to believe that D i t is correlated with the unobservable determinants of investment u i t , or with measurement error in τrt − τhit . Isolating the effects of variation in taxes due to treaties alone is therefore a useful step towards better inferences about the effects of repatriation taxes. 4.2 Data In the empirical exercise I look at how outbound FDI from Canada to other countries changed around the in-force date of bilateral tax treaties, and how this change is related to the magnitude of the repatriation tax on dividends from Canadian affiliates prior to the treaty. The principal data requirements are therefore on outbound FDI stocks, tax treaties, and host and home statutory tax rates. The FDI data are from official Canadian government sources, and cover investments into 95 countries over the 1976–2009 period. The start date of the FDI series vary among countries, and some annual observations are missing due to confidentiality restrictions; on average, there are 21 annual observations per country. The imbalance in the panel raises the spectre of selection bias, but incorporating country fixed effects and additional control variables into regressions should help. The records on in-force dates of bilateral tax treaties and the associated withholding tax rates on related-party dividends are taken from ibfd.com. As noted, the number of treaties rose from 11 to 91 over the sample period; all treaties are listed in Table 2. Data on statutory tax rates are from a variety of sources. Top tax rates in OECD countries are available for the 1981–2009 period from the OECD. Rates for non-OECD countries are from tax guides published by the accounting firms KPMG and Ernst and Young, and are supplemented where missing with information obtained through online searches of various sources. In addition to these main variables of interest, I control for other possibly confounding determinants of FDI patterns based on a panel-data version of the “gravity” model. Information on national GDP, population, and nominal exchange rates are obtained from the Penn World Tables, and distances between national capital cities from data 12
at cepii.fr. Overlapping data from the different sources results in a usable sample that is somewhat smaller. Thus there are 1738 observations in the sample with controls, but 1243 in the sample with tax rates, since reliable information on taxes in the earlier years of the sample is difficult to obtain. In my analysis, the treatment group is those countries that signed a treaty during the sample period (labelled “reformers”), and the potential control group consists of both countries that were never treated (“nontreaty”) and those whose treaties predate the FDI sample (“treaty”). The summary statistics in Table 3 are therefore presented separately for each of the three groups. The table highlights some important differences among the three groups. Reformers on average have per capita GDP and Canadian inbound FDI that is similar to non-treaty countries, and substantially lower than the always-treated countries, which are predominantly in the rich OECD. On the other hand, non-treaty countries have much lower host tax rates than the treatment group. Still, host tax rates in the treatment group are well below Canadian top statutory tax rates: the repatriation tax rate, defined here as the difference of statutory income tax rates, is 9.2 per cent in the reforming countries in pre-treaty years. Based on these comparisons, it seems prudent to include all non-reforming countries in the sample to control for other time-varying determinants of Canadian FDI, but to investigate the implications of using alternative control groups as well.
5
Results
5.1 Effects of tax treaties To begin with, I examine the reduced-form effect of new tax treaties on investment, and set aside the tax rate changes for the next section. At its simplest, this corresponds to a basic difference-in-difference specification of the form log K i t = a i + d t + ηD i t + e i t (3) where the estimator for η is the mean change in FDI to reforming countries relative to non-reforming countries at the same date. An illustration of the method and the data is presented in Figure 2, which graphs the mean of log real FDI stocks for reforming countries as a function of the number of years since (or until) the respective treaty in-force date, say t − Ti where Ti is the in-force year of the treaty with country i . While there is no discernible trend in FDI in pre-treaty years, there is pronounced upward movement beginning in the first treaty year and continuing until the fourth year after the in-force date; FDI appears to continue growing after that at a decreased rate. To ver13
ify that the apparent effect of treaties does not merely reflect confounding trends in aggregate Canadian FDI, I also graph in Figure 2 the corresponding evolution of FDI in the control group. To map the control group onto the post-treatment time variable, I compute for each observation in the treatment group the average log FDI for all countries in the control group in the same calendar year. The Figure then shows the unweighted average of these matched control averages for each value of t −Ti . The control group exhibits a slight upward trend in pre-treaty years but not afterwards. Thus there is some evidence of a “dip” in pre-treaty FDI for the treatment group, relative to the control, but the effect is small. The slow growth in FDI after treaty events in consistent with a model of lagged or partial adjustment of FDI stocks to tax rates, or with persistence in unobservable innovations to FDI. Because of this possibility, in the regression analysis that follows I report estimated standard errors that are robust to correlation among the errors, clustered by country, as suggested by Bertrand et al. (2004). I report some dynamic specifications of the model as well. The basic reduced form estimates are in Table 4. The first column of numbers corresponds exactly to the difference-in-difference specification (3) and shows that FDI rose on average 82 per cent (according to the log approximation) in reforming countries relative to control countries after treaty events. The standard error of the estimate is 39 per cent. The next two columns add controls for other determinants of bilateral investment that might be correlated with treaty events. The four main variables I use are log real GDP in the host and home countries, log population of the host, and an indicator for presence of a free-trade agreement between Canada and the host country. A popular model in international trade suggests that the magnitude of GDP and population effects (which proxy for market size) should be inversely related to the distance between the two countries. I proxy for this gravity effect in two ways. In column (2), I interact the GDP and population variables with an indicator for countries in Europe and the Middle East and the United States (labelled “Europe” for brevity in the tables), which effectively allows these variables to exert a different influence in Europe compared to the rest of the world.11 In column (3), I include interactions between the GDP and population variables and the log distance between countries. The estimated treatment effect in both specifications is very similar to the simple difference-indifference estimate. On the basis of the t statistics and coefficient estimates, the model of column (2) seems to 11 The direct effect of Canadian GDP is not in the regressions, since it is collinear with the year effects.
14
be the preferred specification, and I use this set of control variables in all subsequent regressions. The regression reported in column (4) includes the lagged dependent variable as a regressor, but not country fixed effects. This specfication can be rationalized by a structural model of gradual adjustment to desired longrun capital stocks (Bobonis and Shatz, 2007). Angrist and Pischke (2009) observe that the estimator in column (4) underestimates the treatment effect if the true model is that of column (2), and conversely column (2) overestimates the treatment effect if column (4) is correctly specified. The two estimates therefore provide robust bounds on the treatment effect. According to column (4), the short-run response to a treaty event is a 16 per cent rise in FDI (standard error 7 per cent), and the coefficient on lagged log FDI is 0.81, which implies a long-run response to the shock of 87 per cent, of which 70 per cent occurs in the first eight years. Thus the two estimates of the treatment effect are quite similar. An alternative approach to estimating the dynamic response is a semiparametric one, which regresses log K i t on a set of indicators the number of years since (or until) the reform D i × (t − Ti ) and the other controls, where D i is an indicator variable for countries in the treatment group. (To keep sample sizes in each bin large enough, the variable is top coded at 10 and bottom coded at -5.) The resulting year-by-year treatment effects and a 95% confidence interval around them are graphed in Figure 3, where the reference category is the observations exactly one year prior to the treaty date. Consistent with Figure 2, the estimated treatment effect rises quickly and becomes significantly positive in the third year after the treaty, and then rises to a maximum of 159 per cent in the ninth year. The estimated treatment effect would be biased if treaty events were endogenously determined. This would occur, for example, if Canadian firms successfully lobbied the Canadian government to conclude a treaty after they had established initial investments in a country, or if host governments sought treaties in response to changes in FDI, say to mitigate a decline in inbound investment from Canadian firms. If either of these possibilities were relevant, then there would be a differential trend in reforming and control countries in the years leading up to treaty events. No such trend is observed in Figure3, and the F -statistic on joint significance of the 4 estimated pre-treatment effects is an insignificant 0.59 (p = 0.67). There remains the possibility that treaty events respond to anticipated future changes in FDI, but testing that hypothesis is not possible. One possible objection to this approach is that the control group, comprising both high-income countries that were always treated and low-tax 15
countries that were never treated, is not an appropriate comparison group for reformers. Table 4 reports difference-in-difference estimates for alternative control groups. In this table, although only the treatment effect estimate is reported for concision, all specification include the same control groups as in column (2) of Table 3, which is repeated in column (1) of Table 4 for comparison. In column (2), only observations on reformers are included in the sample, so that identification of the treatment effect relies on the variation in treaty in-force dates among reformers. The point estimate is smaller in this case at 54 per cent, which is not significant. In column (3), the full sample is used, but separate year effects are estimated for each of six regions of the world,12 so that the treatment effect is estimated from differences in differences within each region. The estimate is similar to column (1) but again insignificant. In column (4), to provide a basis for the tax rate elasticities estimated below, the regression sample includes only host countries and years for which statutory tax rates are observed. In this case the effect is larger and significant at 113 per cent. As a last check against the possibility of confounding influences, I report in column (5) results of a “placebo” regression, in which the dependent variable is the log of outbound FDI from the United States into country i in year t , in place of log K i t , and the independent variables are the same as in column (1).13 US FDI should be unrelated to Canadian tax treaties, so that if the model is correctly specified the predicted treatment effect is zero, but US FDI might be affected by the same unobservable economic shocks as those facing Canadian firms. Moreover, Blonigen and Davies (2004) have shown that US FDI is unaffected by US tax treaties, which is nevertheless consistent with my findings, since US treaties do not lead to exemption of foreign-source income as in the Canadian case. Using the placebo data, the estimated change in US FDI is -16 per cent, which is not significantly different from zero. Thus there is no evidence in the US data that the treaty effect is driven by contemporaneous economic changes. 5.2 Tax rate elasticities I turn next to estimating the effects of taxes on FDI directly. The starting point is ordinary least squares estimates of the parameters in equation (2) based on ordinary least squares, incorporating once again fixed effects for country and year and the preferred set of controls. Because I do not have reliable data on the withholding tax rates that would apply to dividend repa12 The six are Europe, Middle East, Australasia, Latin America, and Africa. 13 The US data are from Bruce Blonigen and cover the 1976-2000 period.
16
triations in the the years when no treaty was in force, I set w i = 0 in equation (1) and so define the repatriation tax as the difference between the home and host statutory rates when the difference is positive and no treaty is in effect, and zero otherwise. In this case, the tax rate elasticities are identified from within-country variation in host country tax rates and in the home tax rate (for non-treaty observations) as well as from the variation in repatriation taxes due to tax treaties. Estimates of the semi-elasticities are in column (1) of Table 5; parameter estimates for the control variables are suppressed in the table. The estimates imply a semi-elasticity of 3.0 for the host tax and 1.6 for the repatriation tax. That is, an increase in the host tax of one percentage point is estimated to cause a 3 per cent reduction in the stock of FDI, and a one percentage point increase in the repatriation tax a 1.6 per cent reduction in FDI. As discussed in Section 4, estimates that are identified from year-to-year variation in statutory tax rates are likely unreliable, because it is difficult to measure accurately the tax rates that are relevant to foreign direct investment decisions. The alternative is to rely on the decline in repatriation taxes (to zero) resulting from treaty events, since the resulting change tends to be large and can be measured more accurately. (Like the previous regressions, this is a difference-in-difference estimator, but for purposes of comparison to the OLS estimates it is implemented as an instrumental variables procedure, in which the excluded instrument for the repatriation tax rate τ∗i t is the treaty variable D i t . Lacking a suitable instrument for the host tax rate, I treat it as exogenous.) The estimated semi-elasticities, reported in column (2) of Table 5, are now much larger, equal to 4.8 for the host tax and 13.3 for the repatriation tax. The last column in Table 5 examines the effects of withholding tax rates based on changes in the rates for related-party dividends induced by amendments to the existing bilateral treaty network. There are 27 such treaty amendments in all, which reduced the applicable withholding tax rate from 15 or 10 per cent to 10 or 5 per cent. Regressing log FDI on the withholding tax rate and host tax rate and the usual controls, in the case for the smaller sample of countries and years with an existing tax treaty, yields and estimated withholding tax rate semi-elasticity of 2.4, which in this case is insignificantly different from zero. The estimated elasticities are in the range found in the previous literature. The survey by de Mooij and Ederveen (2003) reports a mean semielasticity of host country tax rates equal to 3.3. Comparable estimates of the repatriation tax elasticity are difficult to find in the literature. However, Hines (1996) reports that a one per cent difference in state tax rates is asso17
ciated with a 9-11 per cent difference in the shares of inbound FDI owned by investors from exemption countries relative to credit countries, with is consistent with a similarly large effect of repatriation taxes. Altshuler and Grubert (2001) present mixed evidence on the effects of repatriation taxes, reporting differences in the cross-sectional pattern of US and Canadian outbound investment that is consistent with a large effect, but much smaller effects based on differences between US firms in excess credit and deficient credit positions. Certainly, a larger effect of repatriation tax reforms is to be expected for Canada, where the possibility of treaty shopping should make investment more sensitive to repatriation taxes than when worldwide taxation is applied more generally.
6
Conclusion
This paper provides robust evidence that home-country taxes on foreignsource income exert a large and significant negative effect on foreign direct investment. Canadian tax treaties that involve a reform from residencebased taxation to exemption have increased investment into the affected countries by 80 per cent on average for a sample of tax treaties implemented in the last 35 years. Based on the observed levels of home country taxes applied to repatriations from these countries before the treaty events, the implied elasticity of investment stocks with respect to repatriation taxes is large. A full welfare analysis of the Canadian tax system would require estimates of several parameters outside the scope of this paper, notably the domestic tax revenues foregone due to exemption, and the extent to which the increase in investment in the reforming countries was accompanied by changes in home country investment by parents of the foreign affiliates, or reallocations of other foreign assets due to treaty shopping. Given the large estimated elasticities, however, the most reasonable conjecture is that Canada’s hybrid system of international taxation has distorted investment decision and entailed a substantial loss of profitability to Canadian MNCs and associated deadweight loss to the domestic economy.
References Altshuler, R., and H. Grubert (2001) ‘Where will they go if we go territorial? Dividend exemption and the location decisions of US multinational corporations.’ National Tax Journal 54(4), 787–810 (2003) ‘Repatriation taxes, repatriation strategies and multinational financial policy.’ Journal of Public Economics 87(1), 73–107 Bertrand, M., E. Duflo, and S. Mullainathan (2004) ‘How Much Should
18
We Trust Differences-in-Differences Estimates?’ Quarterly Journal of Economics 119(1), 249–275 Blonigen, B.A., and R.B. Davies (2004) ‘The effects of bilateral tax treaties on US FDI activity.’ International Tax and Public Finance 11(5), 601–622 Bobonis, Gustavo J., and Howard J. Shatz (2007) ‘Agglomeration, adjustment, and state policies in the location of foreign direct investment in the united states.’ Review of Economics and Statistics 89, 30–43 De Mooij, R.A., and S. Ederveen (2003) ‘Taxation and foreign direct investment: A synthesis of empirical research.’ International Tax and Public Finance 10(6), 673–693 Department of Finance (1998) Report of the Technical Committee on Business Taxation (Ottawa: Department of Finance) Desai, M.A., C.F. Foley, and J.R. Hines (2001) ‘Repatriation taxes and dividend distortions.’ National Tax Journal 54, 829–851 Desai, Mihir A., and James R. Hines (2003) ‘Evaluating international tax reform.’ National Tax Journal 56, 487–502 Devereux, Michael P., and H. Freeman (1995) ‘The impact of tax on foreign direct investment: Empirical evidence and the implications for tax integration schemes.’ International Tax and Public Finance 2(1), 85–106 Devereux, Michael P., and Rachel Griffith (2003) ‘Evaluating tax policy for location decisions.’ International Tax and Public Finance 10, 107–126 Feld, Lars, and Jost Heckemeyer (2009) ‘FDI and taxation: A meta-study.’ Technical Report 2540, CESifo Grubert, H. (1998) ‘Taxes and the division of foreign operating income among royalties, interest, dividends and retained earnings.’ Journal of Public Economics 68(2), 269–290 Hartman, D.G. (1985) ‘Tax policy and foreign direct investment.’ Journal of Public Economics 26(1), 107–121 Hines, J.R. et al. (1994) ‘Credit and deferral as international investment incentives.’ Journal of Public Economics 55(2), 323–347 Hines Jr, J.R. (1996) ‘Altered states: Taxes and the location of foreign direct investment in America.’ American Economic Review 86(5), 1076–1094 Mintz, J., and A.J. Weichenrieder (2008) The Indirect Side of Direct Investment: Multinational Company Finance and Taxation (Forthcoming) Mintz, Jack (2004) ‘Conduit entities: Implications of indirect tax-efficient financing structures for real investment.’ International Tax and Public Finance 11, 419–434 Slemrod, J. (1990) ‘Tax effects on foreign direct investment in the United States: Evidence from a cross-country comparison.’ In Taxation in the Global Economy, ed. Assaf Razin and Joel Slemrod (University Of Chicago 19
Press) pp. 79–117
20
Table 1: Outbound foreign direct investment and tax treaties
Country United States United Kingdom Australia Bahamas France Brazil Bermuda Singapore Barbados Ireland Indonesia Netherlands Switzerland Germany Spain
Foreign direct investment, 1989 Millions of Percentage of 2008 dollars total Treaty? 84,485.0 63.7 Yes 16,552.6 12.5 Yes 3,467.3 2.6 Yes 2,717.7 2.0 No 2,643.0 2.0 Yes 2,507.2 1.9 Yes 2,232.4 1.7 No 2,217.5 1.7 Yes 1,708.3 1.3 Yes 1,668.0 1.3 Yes 1,645.6 1.2 Yes 1,529.1 1.2 Yes 1,527.6 1.2 Yes 1,299.1 1.0 Yes 851.2 0.6 Yes
Country United States United Kingdom Barbados Ireland Cayman Islands Bermuda France Australia Hungary Bahamas Brazil Netherlands Germany Chile Switzerland
Foreign direct investment, 2009 Millions of Percentage of 2008 dollars total Treaty? 256,393.9 44.5 Yes 64,181.4 11.1 Yes 40,022.1 6.9 Yes 22,296.1 3.9 Yes 19,044.7 3.3 No 17,816.3 3.1 No 15,556.7 2.7 Yes 12,545.7 2.2 Yes 12,016.8 2.1 Yes 11,478.2 2.0 No 11,227.0 1.9 Yes 9,617.0 1.7 Yes 9,394.3 1.6 Yes 8,188.5 1.4 Yes 6,620.6 1.1 Yes
21
Table 2: Canada’s bilateral tax treaties 1941-1975: United States Germany Netherlands Denmark
(1941) (1954) (1954) (1956)
Finland Australia Japan United Kingdom
(1958) (1959) (1965) (1965)
Norway Trinidad and Tobago Ireland
(1966) (1967) (1968)
1976-1985: Belgium Barbados Israel New Zealand Switzerland France Dominican Republic Philippines
(1976) (1976) (1976) (1976) (1976) (1976) (1977) (1977)
Singapore Jamaica Pakistan Romania Morocco Indonesia Spain South Korea
(1977) (1977) (1977) (1978) (1978) (1980) (1980) (1980)
Austria Italy Malaysia Bangladesh Thailand Cyprus Sweden Egypt
(1980) (1980) (1980) (1982) (1985) (1985) (1985) (1985)
1986-95: Tunisia Ivory Coast Brazil Sri Lanka Tajikistan Kazakhstan Georgia Kyrgyzstan Moldova Guyana Turkmenistan
(1985) (1986) (1986) (1986) (1987) (1987) (1987) (1987) (1987) (1987) (1987)
Azerbaijan Belarus Ukraine Kenya India Armenia Uzbekistan China Russia Cameroon Malta
(1987) (1987) (1987) (1987) (1987) (1987) (1987) (1987) (1987) (1988) (1988)
Zambia Poland Papua New Guinea Luxembourg Mexico Slovak Republic Czech Republic Argentina Zimbabwe Hungary
(1989) (1989) (1990) (1991) (1992) (1992) (1992) (1995) (1995) (1995)
1996-2009: Latvia Estonia South Africa Iceland Tanzania Lithuania Vietnam Chile
(1996) (1996) (1997) (1998) (1998) (1998) (1999) (2000)
Algeria Nigeria Croatia Jordan Bulgaria Portugal Ecuador Slovenia
(2000) (2000) (2000) (2001) (2002) (2002) (2002) (2003)
Mongolia Kuwait Peru United Arab Emirates Senegal Venezuela Oman Gabon
(2003) (2003) (2004) (2004) (2004) (2005) (2006) (2007)
22
Figure 1: The geographic distribution of tax treaties
23
Figure 2: The dynamics of FDI around treaty dates
8 7 6 mean log FDI
5 4 3 2
Treatment countries Control countries
1 0 -10
-5
0 5 Years since treaty in force
24
10
15
20
Table 3: Summary statistics
Real FDI (2008 $ millions) Treaty Host country tax rate Repatriation tax rate Real GDP (2008 $ millions) population (millions) Distance (km) Observations
Non-treaty Reformers 1,367.4 1,216.7 0 0.6 19.1 32.2 22.4 9.2 126,559.1 566,138.2 10.6 69.8 6,669.0 8,819.0 439 729
25
Treaty 10,454.2 1 34.4 1,430,214.1 76.4 7,412.5 662
Table 4: Reduced form estimates of the effect of tax treaties
Tax treaty
(1)
(2)
(3)
(4)
0.82** (0.39)
0.80** (0.39)
0.87** (0.40)
0.09 (0.30) 2.09*** (0.75) 0.00 (0.04) 0.06 (1.26) 0.42** (0.16) -0.69 (1.27)
-0.17 (0.33) -13.72 (8.92) -0.55 (1.18)
0.16** (0.07) 0.81*** (0.01) -0.09 (0.12) 0.39*** (0.11) 0.00 (0.01) -0.04 (0.26) 0.13*** (0.04) 0.37 (0.28)
Lagged log real FDI Free trade agreement Log host real GDP Log exchange rate Europe*Log host real GDP Europe*Log exchange rate Europe*Log home GDP Log distance*log real host GDP
1.77* (1.02) 0.07 (0.13) -0.60 (1.01) -4.25** (1.94)
Log distance*log exchange rate Log distance*log real home GDP Log distance*log population
Observations 1,822 1,738 1,738 1,589 R-squared 0.82 0.85 0.85 0.96 Country fixed effects Yes Yes Yes No *** p
