Much of international tax policy rests on judgments regarding

Transcription

1 Do Taxes Influence Where U.S. Corporations Invest? Do Taxes Influence Where U.S. Corporations Invest? Abstract - This paper uses data aggregated from tax returns of more than 5 U.S. multinational corporations (MNCs) to identify the role of host country tax rates in determining the amount of capital invested in 6 potential locations. The empirical results show that average effective tax rates have a significant effect on the choice of a location and the amount of capital invested there. A lower tax rate that increases the after tax return to capital by one percent is associated with about percent more real capital invested if the country has an open trade regime. The attractive power of low tax rates is weakened if the country has a more restrictive trade regime. Approximately 9 percent of U.S. capital abroad would be in a different location in the absence of any effect of taxes. Harry Grubert U.S. Treasury Department, Washington, D.C. John Mutti Grinnell College, Grinnell, IA 5 National Tax Journal Vol. LIII, No. 4, Part INTRODUCTION Much of international tax policy rests on judgments regarding the allocation of capital in response to tax differences among alternative locations. A primary objective of this paper is to assess the responsiveness to taxes of the locational choices abroad of U.S. multinational corporations (MNCs). The empirical analysis is based on data from the U.S. Treasury 99 corporate tax files, which cover the activities of more than 5 major U.S. manufacturing companies in 6 potential foreign locations. The results reported here rest on data aggregated for each country location, and they demonstrate that local average effective tax rates have a significant effect on the amount of capital that U.S. MNCs have in a given location. For a country with an open trading regime, the elasticity of total real capital with respect to a reduction in the host country tax rate that increases the after tax return by percent is about three. A number of papers have analyzed the sensitivity to taxes of foreign direct investment (fdi) into and out of the United States. Early examples of this work based on aggregate data include Hartman (984), Boskin and Gale (987), and Slemrod (99). They have generally found significant tax impacts, but these studies suffer from two major limitations. One is that they relate the annual flow of fdi to the level of current or lagged tax rates. Basic models of firm behavior indicate that there will be a long run equilibrium relationship between See Hines (997) for a convenient summary of work in this area. 85

2 NATIONAL TAX JOURNAL the stock of capital and the level of the cost of capital, not between the annual flow of new investment (or the change in the stock) and the level of the cost of capital. Therefore, a more appropriate analysis would either estimate a stock equation or demonstrate what adjustment costs and tax changes are relevant in determining annual changes in the stock. Furthermore, fdi, which is the change in direct equity abroad, is an inappropriate measure for real investment, because it may simply represent financing or repatriation behavior. Indeed, in their study of U.S. direct investment in Canada, Grubert and Mutti (99b) find that variations in fdi and plant and equipment spending are virtually uncorrelated. Studies based on real investment at the firm level, derived from Standard and Poor s Compustat, have asked somewhat different questions. Harris (99) analyzes the impact of the Tax Reform Act of 986 based on a - categorization of firms expected to face a large increase in their tax burden. Cummins and Hubbard (995) focus on the short run response of U.S. firms investing abroad. Devereux and Griffith (998) are able to avoid some of the ambiguities that Cummins and Hubbard face in measuring changes in capital spending by considering the choice of U.S. MNCs to locate in three European countries over the period From all of these papers, however, it is difficult to derive implications for the way aggregate stocks and flows of capital will be affected by tax policy changes, such as those reported in the opening paragraph. Two cross sectional studies that evaluate the relationship between real capital stocks and local tax rates are most closely related to the present paper. In a study of U.S. manufacturing affiliates in foreign countries based on 98 data from The Benchmark Survey of U.S. Direct Investment Abroad, Grubert and Mutti (99a) find that a percent reduction in the cost of capital increased the stock of U.S. owned net plant and equipment by.5 percent. Hines and Rice (994) use that same data source but also include all non bank affiliates in a larger group of countries. In two different formulations they obtain corresponding elasticities with respect to a percent change in the cost of capital equal to. and 4.5. Because their country sample includes a large number of tax havens with tiny populations, however, there may be many cases in which the real capital owned by an affiliate incorporated in a low tax location is actually used in a branch somewhere else. This paper uses a different and more recent data source to carry out a cross sectional analysis of the relationship between real capita stock and local tax rates. It confirms that the responsiveness of capital to taxes is substantial in many alternative specifications. This result continues to hold when very poor or very low tax countries are excluded from the group of potential locations. Also, it holds if the tax rate is represented by an average for 99 and 99, to avoid the noise of a measure calculated for a single year, or if it is adjusted for possible dependence on the amount of recent investment in a country. Countries with more restrictive trade regimes obtain less of a benefit from lower tax rates in attracting U.S. investment. The tax responsiveness of investment declines. This is consistent with the importance of production for other foreign markets, because more restrictive trade regimes in- Their approach is most closely related to firm level studies of domestic investment behavior, where aggregate analysis has been particularly unsuccessful due to problems of simultaneity in the determination of investment and the cost of capital. See Hassett and Hubbard (996) for a survey of empirical, firm level studies of domestic investment that yield short run estimates of the response of investment to changes in the cost of capital. 86

3 Do Taxes Influence Where U.S. Corporations Invest? crease the relative cost of exports. If current stocks of investment are used to weight the estimated responses for each trade regime, the corresponding overall elasticity is two. A FRAMEWORK FOR EMPIRICAL ANALYSIS MNC investment generally occurs when a firm expects that locating production abroad will allow it to earn a higher return from some special expertise it possesses. Basic models that demonstrate the determinants of an MNC s optimal capital stock in a given country location are presented by Grubert and Mutti (99a) and by Hines and Rice (994). These models generally depend upon a scale variable that reflects the size of the market to be served and upon the costs of capital, labor, and other variable inputs. For a firm to earn the same after tax return to capital in all locations, the before tax cost of capital is higher in a location where a higher tax rate applies; a higher tax rate results in a smaller optimal capital stock. Other relevant variables include tariffs and transport costs and plant specific fixed costs. Foreign production becomes a more attractive way of serving the foreign market as foreign production costs and plant specific fixed costs are lower and as transport costs and trade barriers are higher (Horstmann and Markusen, 99). The primary issue addressed in this paper is the responsiveness of the amount of real capital invested by MNCs in manufacturing affiliates in a given country to the rate of taxation on the income earned in that country. The data are aggregated over all firms in the country to create a single observation, as if there were a single firm. The choice of this firm in deciding where to invest among several competing locations to serve a given market might be expressed in the following general form: [] K j = f(y, r, t, t,... t N, X, X,... X N ) where K j is amount of real capital located in country j, Y represents the scale of the market served, r is the common after tax return earned in all locations, t j is the tax rate in country j, and X j represents all other variables that vary by location j. There are insufficient degrees of freedom to consider each competing location separately, and a convenient simplification is to consider an average of these alternative rates. Expressing this relationship in a logarithmic form gives the following equation: [] log K i = a + b log[( t i )/( t A )] + c log[( t i )/( t US )] + d log X i + e log N, where t A is the average tax rate over all foreign locations, t US is the U.S. effective tax rate, and N is the number of relevant foreign locations. The role of host country taxation enters in the form ( t i ), which means that the coefficient b is an elasticity indicating the percentage change in capital located in country i in response to a percent change in the after tax return in location i for a given pre tax return, or, equivalently, to a percent change in the cost of capital for a given after tax return. This formulation assumes that the competition offered by other locations, apart from the United States, is fully represented by the number of potential locations, N, and their mean tax rate. If the number of potential foreign locations is large, the average tax rate is the same regardless of which country i is considered. The impor- In a simple model without depreciation or government incentives, the cost of capital will be the after tax rate of return, r, divided by one minus the tax rate, ( t), or r/( t). 87

4 NATIONAL TAX JOURNAL tance of this average tax rate and the U.S. tax rate then are subsumed in the constant term, because they are fixed in the cross section. Although the two different relative tax rates are explicitly shown in equation [] to indicate that the capital located in country i may be attracted from other foreign locations and from the United States, the two log( t i ) terms can be grouped together: [] log K i = f + g log( t i ) + d log X i which is the equation estimated by ordinary least squares in this study. 4 Equation [] expresses the desired capital stock in country i as a function of the host country tax rate and the vector of country characteristics. Of course, potential locations may not be regarded as equally likely, due to differences in the proximity to the relevant market or the availability of particular factor endowments. That possibility is considered by introducing a set of regional dummies for North America, Latin America, Asia, and the EC. These dummies are entered as independent variables and are also interacted with the tax rate to assess whether there is greater sensitivity to taxes in neighboring locations with access to the same market. 5 With respect to other country characteristics, gross domestic product (GDP) and GDP per capita are included as possible indicators of the size of the domestic market, where the latter variable is particularly relevant for goods whose demand is income elastic. Higher per capita income may also reflect a more skilled labor force that can be employed in high technology industries, and we attempt to measure that effect independently by including a human capital variable that gives the percentage of the university age cohort who receive higher education. An indicator of openness of the economy is entered as an independent variable and also interacted with the tax variable, in order to assess whether countries with more restrictive trade regimes and less potential to benefit from a more competitive position in export markets get less benefit from lower tax rates. RELEVANT CONCEPTS OF TAXATION The income tax rate used in the empirical analysis is the observed average rate for a given host country. An important consideration is the potential superiority of Hall Jorgenson marginal effective tax rates. Unfortunately, such marginal tax rates are not available for all of the 6 4 OLS estimation assumes that the error terms from each of the country observations are independent. If the error term from investing in one location were correlated with the error term from investing in another location, the precision of the resulting estimates could be overstated. That issue cannot be addressed in the current aggregate framework for a single year. Another issue is that the tax variable may simply be a proxy for unmeasured country specific effects. In a cross section for a single year, this possibility cannot be addressed directly. In a follow up to the present paper, Altshuler, Grubert, and Newlon (998) report a preliminary analysis that looks at aggregate data on capital stocks for two different years, 984 and 99. Over this eight year interval substantial variation in tax rates within individual countries has occurred, which warrants considering how changes in tax rates are related to changes in capital allocations. Such an approach can control for country specific nontax factors that do not change at the same time. The magnitude and significance of the tax variable did not fall in this framework. 5 Including a regional dummy and its interaction with the host country tax variable implies that the observed response to the host country tax rate depends upon the difference between the rate in that country and the mean for the region. Countries not included in the regional dummies are a heterogeneous group that includes South Africa, Nigeria, Israel, and Morocco. Also, see Hines (996) for a related treatment of real capital invested by foreigners across states in the United States. He represents the competing tax in alternative locations by a weighted average rate based on the size of business activity in each state, a measure that ignores regional effects within the United States but does vary for each state considered. 88

5 Do Taxes Influence Where U.S. Corporations Invest? countries in the analysis. Furthermore, the Hall Jorgenson King Fullerton (HJKF) type generally are based on a limited number of features of the tax system, in particular the statutory tax rate, the amount of accelerated depreciation on tangible investments, and investment tax credits. In many host countries companies are offered special ad hoc deals that are difficult to identify simply from basic statutory provisions. In contrast, the country average tax rates computed from the Treasury files reflect all provisions of the tax system as well as special arrangements. Because of the importance of the taxation of infra marginal rents that can be earned from production in a given location compared to their taxation in alternative locations, the average tax rate may even be the preferable indicator of the incentive to locate in a country. Another potentially relevant issue is whether there is a residual tax due in the United States, which in turn depends upon the parent s expected foreign tax credit position. Under U.S. law, active foreign income received by U.S. corporations is taxed when repatriated, with a credit for foreign taxes including both withholding taxes and the underlying foreign corporate tax on equity income if the U.S. company owns at least percent of the foreign company. The credit for foreign tax is limited, however, to what the U.S. tax would be on the equivalent income. If the parent company does not expect to have excess credits, then the residual U.S. tax on potential repatriations may become relevant. Hartman (984) and Sinn (99) have claimed, however, that the repatriation tax should be irrelevant for investment by a mature controlled foreign corporation (CFC), in an analysis that parallels the irrelevance of individual income taxes on dividends in the new view in a domestic context. If residual U.S. taxes are relevant, they will tend to narrow tax differentials between locations because the U.S. repatriation tax would be higher on distributions from low tax countries. If the firm expects to have excess credits, then the marginal effective tax rate on equity income is the host country rate, including withholding taxes on any distributions. Interest and royalties, however, which are generally deductible in the host country, will be exempt in the United States because they are foreign source and can be shielded by the excess credits. A CFC may respond to a high statutory tax rate by increasing its debt to asset ratio, or by paying more royalties to the parent, thereby generating greater deductible expenses in the host country and diluting the deterrent effect of high local taxes. Ignoring residual U.S. taxes and debt finance, and using only the local tax rate, may bias downward the estimated tax responsiveness of MNCs, because the variation in the tax burden across locations is less than implied by the local rate. DATA SOURCES AND MEASUREMENT ISSUES The principal data base was the linked Forms, 547, and 8 for 99. Form is the basic U.S. corporate income tax return and provides information on the parent s income, expenses, and assets. The Form 547 is an information return that MNC parents are required to file for CFC that gives its earnings and profits, balance sheet, sales, foreign taxes paid or accrued, and transactions with affiliates. A CFC is defined as a foreign company, more than 5 percent of which is owned by U.S. shareholders. 6 The Form 8 is used by a parent MNC when claiming a foreign tax credit to reduce its U.S. tax liability. 6 A U.S. shareholder for the purpose of this definition must own percent or more of the foreign company. Only concentrated holdings count. More than 75 percent of the CFCs on the 547 file are percent controlled. The average ownership level is 94 percent. 89

6 NATIONAL TAX JOURNAL The MNC parents in the sample are all U.S. companies in manufacturing with total assets in excess of $5 million that have at least one CFC. Altogether there turned out to be 56 parent companies whose data were available. Sixty potential foreign locations were used in the analysis, which were all the countries with more than five manufacturing CFCs in The analysis was restricted to manufacturing CFCs of manufacturing parents. CFC assets on the Form 547 balance sheet are reported according to general U.S. accounting principles. They are not distorted by host country incentives such as accelerated depreciation. They are historical book values, however, which can be distorted by inflation. This disadvantage of using historical book values may warrant attention to the number of companies that choose to locate in a country, without considering the amount invested there, or to the sales made from that location. Average host country corporate tax rates are derived from the Form 547s by taking total income taxes paid by manufacturing CFCs incorporated in that country divided by their total Earnings and Profits. The latter, defined in the Internal Revenue Code, is intended to reflect net economic income, not host country (or domestic U.S.) taxable income, which is affected by incentives such as accelerated depreciation. In this calculation, only those CFCs with positive income are included, because otherwise the tax measure will be biased upward. This approach yields 6 country average effective tax rates; firm specific rates are not calculated, because they are not known for locations that are not chosen or where negative profits are earned. Data for GDP and population (in 99) are from the World Bank World Development Report 99, supplemented in a few cases by information from The World Factbook 99 published by the Central Intelligence Agency. The human capital variable indicating the share of the country s university age cohort receiving higher education also is from the World Development Report. The measure of the country s trade restrictions, running from zero (for the most open) to three, is based on four degrees of trade policy openness reported in the World Bank World Development Report 987 (page 8). The Report states that the classification is based on: (a) the country s effective rate of protection, (b) its use of direct controls such as quotas, (c) its use of export incentives, and (d) the extent of any overvaluation of its exchange rate. The measure thus reflects more than trade restrictions and in particular may include the effect of any capital controls that allow an over valued exchange rate. EMPIRICAL RESULTS Ordinary least squares estimates of equation [] based on the 6 country locations where at least five MNC manufacturing affiliates operate are reported in Table. Column shows that average effective tax rates have a significant effect on the amount of capital invested in a country. The elasticity with respect to a percent change in the cost of capital, or a percent change in the after tax return for a given pre tax return, is. for countries with the most open trade regime. For those countries the interaction of the tax rate with the trade regime is zero and therefore irrelevant in determin- 7 Puerto Rico was not one of the potential locations in this analysis. The U.S. operations in Puerto Rico under section 96 are incorporated in the United States and not CFCs. The tax rules governing the possessions are unique and more favorable than those governing CFCs. Since the overwhelming share of Puerto Rican output is imported into the United States, 96 corporations differ from most CFCs. Nevertheless, it would be interesting to integrate them into the present data base. 8

7 Do Taxes Influence Where U.S. Corporations Invest? Independent Variables Log (GDP) TABLE THE LOCATION OF CAPITAL BY U.S. MANUFACTURING MNCs IN 99 COUNTRY TOTALS, 6 LOCATIONS.85 (4.58) Dependent Variable, Log of Total Capital.878 (.44).795 (9.5).8 (9.7) Log of Number of Companies in a Location.479 (.4) Log (GDP per capita).48 (.9).6 (.7).49 (.4).6 (.5).4 (.8) Trade Regime.67 (.59).66 (.).64 (.59).75 (.86). (.56) Log ( ETR). (.).9 (.) Trade*Log ( ETR).5 (.44).5 (.5) North America.4 (.95). (4.8).4 (.9). (.89).8 (4.4) Latin America.55 (4.8).5 (5.5).57 (4.77).55 (4.78).86 (4.74) EEC.788 (.7).875 (.8).784 (.5).748 (.9).59 (.98) Asia.64 (.).477 (.58).647 (.4).599 (.9).4 (.9) / (ETR+.).49 (4.) Trade* / (ETR+.).6 (.8) Log ( Adjusted ETR). (.88) Trade*( Adjusted ETR).54 (.7) Log ( Average ETR).65 (.4) Trade*Log( Average ETR).65 (.45) Adjusted ETR is the effective tax rate adjusted for potential endogeneity as explained in the text. Average ETR is the average effective tax rate calculated from 99 and 99. t values are given in parentheses. ing the effect on real capital. The coefficient for the interaction of the trade regime and tax variables shows that more restrictive regimes benefit less from a lower tax rate. Because of this moderating impact of trade restrictions on responsiveness to tax rates, the comprehensive elasticity of real capital with respect to the tax variable, created by weighting each country s trade policy influenced elasticity by the U.S. capital invested there, is two. 8 8 We also use the firm level data to make probit and tobit estimates of the location decisions of individual firms, and we find that they corroborate the aggregate estimates. The combined effect of a lower tax rate increasing 8

8 NATIONAL TAX JOURNAL The second column of Table allows an assessment of whether there is a magnified impact of low tax rates on capital allocation decisions, rather than a constant proportional effect as assumed with the log ( t) specification. The inverse (. plus the average effective tax rate) is used as the tax variable, where the somewhat arbitrarily chosen value. avoids the extreme values that otherwise would be created at very low tax rates. In this form, taxes are more significant statistically and the overall equation has more explanatory power. The magnified sensitivity at low tax rates may represent a response to the incentive to shift income to low tax countries. Note that the coefficient on the trade openness variable is now positive, but its interaction with the tax variable still shows that restrictive policies weaken the response to low taxes. The regression in column uses the 99 effective tax rate that has been adjusted for differences in the age composition of the companies in each location. Such an adjustment corrects for a potential bias or endogeneity of the average effective tax rate, because that rate is likely to be lower if there has been a large amount of recent investment. Large recent investments that benefit from investment tax credits would be particularly likely to indicate a spurious relationship between low tax rates and a large amount of capital in a location. Although the current analysis deals with the stock of capital and not the yearly change, the book value of capital nevertheless may be affected by large recent investments, too. Examination of the individual CFC data indicates that there is a clear age effect; CFCs incorporated recently have significantly lower effective tax rates than the country average. Although one way to control for endogeneity is to apply an instrumental variable approach that gives a predicted value of taxes based on their relationship to another country level exogenous variable, a preferable approach here, given our access to firm level data, is to make an hedonic adjustment of the tax variable. Based on the coefficients from CFC level regressions of affiliate taxes on the affiliate s age, and using the age distribution of CFCs in each country, we construct an adjusted effective tax rate not distorted by a large flow of recent investments in some countries. The corresponding coefficient estimate of. shows that the estimated tax responsiveness of capital is hardly affected. 9 Column 4 uses a tax variable based on the average of the 99 and 99 effective tax rates, a measure that is less subject to the probability that a firm will locate in a given country and also increasing the amount of capital it chooses to locate there yields a combined elasticity of the expected amount of capital equal to about three. Firm level analysis for separate industries also reveals differences in tax sensitivity that conform to expected differences in mobility: the location of computer and electronics companies is highly responsive to local tax rates, while taxes appear to have no effect on the choice of location in food and drugs, areas where local brand names, regulations, and price controls are likely to have a greater influence on firm location. Finally, micro data provide an opportunity to incorporate the role of the parent s foreign tax credit position. When a dummy variable that indicates whether the parent is in an excess credit position is interacted with the host country tax rate, it has no significant effect on the firm s locational choice. In fact, there are differences across firms in their potential mobility internationally, and the firm s tax sensitivity seems to determine its excess credit position and not the reverse. For example, computer companies, which tend to be in excess limit, are much more sensitive to taxes than petroleum companies, which tend to be in excess credit. 9 The age coefficients derived from the CFC level analysis would be expected to yield much more precise estimates of age effects than simply adding the average age of CFCs as an independent variable in a location regression, because of the large variation of CFC ages within a location but relatively modest differences in average age across locations. As another attempt to control for timing and composition effects that may cause distortions in the average effective tax rate, we replaced that tax variable with the exogenously given statutory tax rate. That procedure is similar to using the statutory tax rate as an instrument for the average effective tax rate. The corresponding coefficient estimate is.9, significant at the 4 percent level. 8

9 Do Taxes Influence Where U.S. Corporations Invest? random influences of a single year and more likely to represent the long run equilibrium relationship sought in cross sectional analysis. The estimated elasticity with respect to ( t) goes up somewhat and is more significant. Host country GDP, GDP per capita, and the regional dummies, which are indicators of the relevant market size, all have significant coefficients. The human capital measure never is significant, and the only effect of including it is to reduce the significance of the GDP per capita coefficient. Therefore, we omit it from the table. Also, the interaction of the regional dummies with the tax term is only significant in the case of the EC, and those results are omitted, too. In the final column of Table the dependent variable is the number of U.S. affiliates located in the country, which demonstrates another aspect of the location decisions of firms and addresses any potential reservation over the measurement of capital. The coefficient of the tax variable is., which indicates that a percent decline in the cost of capital leads to a. percent increase in the number of affiliates located in the country. If the total capital response figure reported in the other columns is thought of as the product of the number of affiliates and the average amount of capital invested per affiliate, taxes can be seen to influence both elements of the total effect. Table indicates how sensitive the Table aggregate results are to the exclusion of certain locations. Apart from our usual concern that the high tax responsiveness observed may be due to a few idiosyncratic countries, this approach also addresses the possible bias from including tax haven countries where firms choose to incorporate even though their capital assets are used outside of that country. Regressions using both the 99 effective tax rate and the 99 9 average are given. The first two columns exclude the countries with the highest and lowest 99 effective tax rates. The results are little different from the comparable ones in Table. The next two columns exclude the five locations with average effective tax rates below 7.5 percent. The tax coefficients Policy Variables Log ( ETR) TABLE SENSITIVITY OF RESULTS TO COUNTRIES AND VARIABLES EXCLUDED Exclude Highest and Lowest ETRs (58). (.6) Exclude ETR<.75 (55).4 (.) Exclude ETR<.75 andetr>.4 (48).7 (.7) Exclude GDPPC< (44). (.6) No Trade Interraction (6).6 (.75) Trade*( ETR).44 (.).88 (.). (.). (.) Log ( Average ETR).9 (.9).9 (.58) 4.9 (.5).49 (.75).79 (.8) Trade*( Average ETR).7 (.4).889 (.4).7 (.99).76 (.4) Trade.579 (.44).7 (.8).9 (.6).46 (.78).59 (.94).68 (.).754 (.6).7 (.74). (.5) The number of countries included is shown in parentheses under the column headings. For coefficient estimates, t values are shown in parentheses..78 (.5) Hines (996) estimates a similar inward investment equation using OLS. For an alternative approach, see Papke s (99) maximum likelihood estimation of a Poisson model. 8

10 NATIONAL TAX JOURNAL decline, but they are significant at the 5 percent level and still substantial in size,.4 with 99 effective tax rate alone, and.9 with the 99 9 average. In addition, if the three locations with population less than one million are excluded, the tax coefficient is not affected. The regressions in the fifth and sixth columns of Table exclude countries with average effective tax rates below 7.5 percent and above 4. percent, leaving 48 in the sample. Surprisingly, the tax responsiveness coefficients go up. The next two columns exclude the 6 locations with GDP per capita less than $,. The tax responsiveness coefficients are hardly affected. The last two columns show that the results are sensitive to the inclusion of the trade regime tax interaction term. The tax responsiveness coefficients decline to less than half their value in Table, and the coefficient based on the 99 effective tax rate alone is significant only at the percent level. The tax coefficient for the 99 9 average, however, remains significant at the 5 percent level as well as the coefficient for the inverse form of the 99 rate used in the fourth column of Table. If a very basic formulation is used, in which the only explanatory variables are GDP and the tax variable, the tax coefficient is three and statistically significant at the percent level. Distinguishing the trade regime becomes important when other country variables, such as GDP per capita and the regional dummies, are included. POTENTIAL EXCLUDED VARIABLES One issue in interpreting the tax results is whether low tax rates are correlated with other policies that promote investment, such as the provision of public goods or non tax subsidies. On a priori grounds, the correlation between low average effective tax rates and other incentives could be either positive or negative. Greater non tax incentives could be associated with high tax rates if a country tries to offset a high tax rate by offering non tax incentives to foreign companies. The 98 Benchmark Survey of Foreign Direct Investment included a set of questions that allow us to address the issue of non tax incentives. Companies were asked to indicate whether they had received various types of incentives (or were subject to various performance requirements). Variables based on the frequency of positive responses by country were never significant and did not affect the tax coefficients. When the percentage of companies in a country stating that they had received non tax subsidies is regressed on the 98 average effective tax rate, the correlation is positive although not statistically significant. (The t value is about.5.) Thus, it appears that including non tax incentives would increase the estimated importance of the tax variable. It might be considered useful to include an indicator of ex ante pre tax profitability in a location, but no valid direct measure is available. One possible indicator of expected profitability is the country s rate of GDP growth in the recent past, on the grounds that corporate profits are higher in more rapidly growing economies. When the rate of growth from 98 to 989, as reported in United Nations statistics, was used as an independent variable, the tax coefficient tended to become slightly larger but the growth coefficient was negative. Finally, we consider more directly the implication of omitting a wage variable, which some researchers have found to be a significant explanatory factor (Wheeler and Mody, 99). If high tax countries also have higher wages, then omitting When the regression is restricted to the 6 countries with open trade regimes, the estimated tax coefficient is.6, which is significant at the 5 percent level. 84

11 Do Taxes Influence Where U.S. Corporations Invest? wages might cause the importance of taxes that we report to be overstated. Ideally we would specify this relationship more fully and consider not only the role of wages but also of labor productivity in developing an appropriate labor cost variable. Because such information is not available, we instead regressed the average effective country tax rate on the measure of manufacturing wage rates reported by Wheeler and Mody. We found no statistically significant relationship. Also, when we included the wage variable as an independent variable to explain the amount of capital located in a country, it appeared to be collinear with GDP per capita; neither variable was significant, and the remaining coefficient estimates and their significance were not affected. IMPLICATIONS FOR CAPITAL ALLOCATION The empirical analysis has identified the distribution of MNC real capital among high and low tax countries. Can the estimated tax coefficients be interpreted as elasticities on the margin, given the increase of capital from U.S. MNCs a jurisdiction can expect if it lowers its average effective tax rate? Yes, if we assume the capital allocation structure in equation [], in which K j = f( t j, t A, t US, N, X j ) and there are a large number of countries, N, where an MNC can invest. Consider a country making a small cut in its tax rate so that it is just equal to the next lowest country in the tax distribution. It can expect the same investment as its neighbor now has, because the only difference between country j, which has just lowered its tax rate, and its neighbor is that country j faces a slightly lower average tax rate. That is, country j will not have competition like its former self with a tax rate just slightly above its own. But if there are many potential competitors, this effect, the change in t A, would be insignificant. The 85 total derivative of K j with respect to t j includes a dt A /dt j term that becomes zero as N becomes very large. As noted earlier, the actual demand for capital function may have a more complex structure, with some groups of locations being much better substitutes than others because of geographic or factor endowment proximity. Only in the case of the EC did we find that special attention to heightened tax competitiveness within a region was warranted, and therefore, we do not pursue that distinction. If the simple choice function applies, we can make a summary calculation of the effect of differences in foreign tax rates on the allocation of U.S. MNC capital abroad. In fact, our use of a cross section of countries in a single year makes it impossible to estimate the effect of a change in t A or t US. Nevertheless, if the U.S. tax rate is held constant while all foreign tax rates converge on the mean (leaving t A unchanged), then the estimated coefficient for t i should give the effect of each country s shift toward the mean. Using the tax response coefficient reported in Table, we calculate that adjusting each tax rate to the mean host country value yields absolute differences equal to 7.6 percent of total investment. Because any reallocated capital is accounted twice, this figure suggests that almost 9 percent of capital is reallocated for tax considerations. CONCLUSIONS Host country average effective tax rates appear to have a highly significant effect on the location and investment decisions of U.S. manufacturing companies. This conclusion is based on country level analysis of the international operations of more than 5 U.S. companies in 6 potential locations. The results appear to be quite robust. Tax responses remain significant when tax havens or very poor countries are excluded from the sample.

12 NATIONAL TAX JOURNAL Countries with more restrictive trade policies appear to attract less U.S. investment, perhaps because trade restrictions are indicators of restrictions on business in general. Countries with restrictive trade regimes are also less able to use low taxes to attract investment. Presumably this reflects the fact that much of the output that might potentially be attracted would be sold in other markets. For countries with open trade regimes, the combined tax response elasticity, based on a higher probability of choosing to locate in a country and a larger amount of capital invested there, is approximately three. This figure is not applicable to all host countries, but if current stocks of investment are used to weight responses for each trade regime, the corresponding overall elasticity is two. Thus, most U.S. MNC capital appears to be located where foreign tax changes can substantially affect the amount of investment there. Acknowledgments We thank Gordon Wilson and Paul Dobbins for providing us with the data base and advising on its use. Without implicating, we are grateful to Len Burman, Rachel Griffith, Bill Randolph, and two referees for helpful advice. Nothing in this paper should be construed as the views and policy of the U.S. Treasury Department. REFERENCES Altshuler, Rosanne, Harry Grubert, and Scott Newlon. Has U.S. Investment Abroad Become More Sensitive to Tax Rates? NBER Working Paper No. 68. Cambridge, MA: National Bureau of Economic Research, 998. Boskin, Michael, and William Gale. New Results on the Effects of Tax Policy on the International Location of Investment. In The Effects of Taxation on Capital Accumulation, edited by M. Feldstein, 9. Chicago: University of Chicago Press, 987. Cummins, J.G., and R.G. Hubbard. The Tax Sensitivity of Foreign Direct Investment: Evidence from Firm Level Panel Data. In The Effects of Taxation on Multinational Corporations, edited by M. Feldstein, 47. Chicago: University of Chicago Press, 995. Devereux, Michael, and Rachel Griffith. Taxes and the Location of Production: Evidence from a Panel of U.S. Multinationals. Journal of Public Economics 68 No. (June, 998): Grubert, Harry, and John Mutti. Taxes, Tariffs and Transfer Pricing in Multinational Corporate Decision Making. Review of Economics and Statistics 7 No. (May, 99a): Grubert, Harry, and John Mutti. Financial Flows versus Capital Spending: Alternative Measure of U.S. Canadian Investment and Trade in the Analysis of Taxes. In International Economic Transactions, Issues in Measurement and Empirical Research, edited by P. Hooper and J.D. Richardson, 9 7. Chicago: University of Chicago Press, 99b. Harris, David G. The Impact of U.S. Tax Law Revision on Multinational Corporations Capital Location and Income-Shifting Decisions. Journal of Accounting Research supplement (99): 4. Hartman, David G. Tax Policy and Foreign Direct Investment in the United States. National Tax Journal 7 No. 4 (December, 984): Hasset, K.A., and R. G. Hubbard. Tax Policy and Investment. In Fiscal Policy: Lessons from Economic Research, edited by A. Auerbach, Cambridge: MIT Press, 997. Hines, James. Altered States: Taxes and the Location of Foreign Direct Investment in America. American Economic Review 86 No. 5 (December, 996):

13 Do Taxes Influence Where U.S. Corporations Invest? Hines, James. Tax Policy and the Activities of Multinational Corporations. In Fiscal Policy: Lessons from Economic Research, edited by A. Auerbach, Cambridge: MIT Press, 997. Hines, James, and Eric Rice. Fiscal Paradise: Foreign Tax Havens and AmericanBusiness. Quarterly Journal of Economics 9 No. (February, 994): Horstmann, Ignatius, and James Markusen. Endogenous Market Structures in International Trade. Journal of International Economics (99):9 9. Papke, Leslie. Interstate Business Tax Differential and New Firm Location. Journal of Public Economics 45 No. (99): Sinn, Hans Werner. Taxation and the Birth of Foreign Subsidiaries. In Trade Welfare and Economic Policies, Essays in Honor of Murray C. Kemp, edited by H. Herberg and N. Long, 5 5. Ann Arbor: University of Michigan Press, 99. Slemrod, Joel. Tax Effects of Foreign Direct Investment in the United States: Evidence from a Cross Country Comparison. In Taxation in the Global Economy, edited by A. Razin and Joel Slemrod, Chicago: University of Chicago Press, 99. Wheeler, David, and Ashoka Mody. International Investment Location Decisions, Journal of International Economics No. (August, 99):

14 NATIONAL TAX JOURNAL 88 Trade Regime TABLE A APPENDIX OF COUNTRY-LEVEL DATA Country Average Effective Tax Rate Adjusted Tax Rate Number of CFCs GDP per capita Real Capital Stock, 99 Population Canada Mexico Costa Rica El Salvador Guatemala Honduras Panama Cayman Islands Dominican Republic Jamaica Argentina Brazil Chile Columbia Ecuador Peru Uruguay Venezuela Bermuda Belgium Denmark France Ireland Italy Luxemburg Netherlands United Kingdom Portugal Spain Germany

15 Do Taxes Influence Where U.S. Corporations Invest? 89 Trade Regime TABLE A (continued) APPENDIX OF COUNTRY-LEVEL DATA Country Average Effective Tax Rate Adjusted Tax Rate Number of CFCs GDP per capita Real Capital Stock, 99 Population Greece Austria Finland Norway Sweden Switzerland Turkey Egypt Morocco Kenya Nigeria Zimbabwe South Africa Zambia Israel India Indonesia Malaysia Pakistan Philippines Singapore Sri Lanka Thailand China Taiwan Hong Kong Japan Korea Australia New Zealand

16