Empirical asymmetries in foreign direct investment and taxation

Journal of International Economics 62 (2004) 337 – 358 www.elsevier.com/locate/econbase

Empirical asymmetries in foreign direct investment and taxation John Mutti a,*, Harry Grubert b b

a Meyer Economics, Box A-6, Grinnell College, Grinnell, IA 50112-1670, USA Office of Tax Analysis, US Department of the Treasury, 15th & Pennsylvania Ave. N.W., Washington, DC 202, USA

Received 21 October 2002; received in revised form 1 November 2002; accepted 11 November 2002

Abstract This paper assesses the sensitivity of the operations of multinational corporations (MNCs) to host country taxation. The empirical analysis is based on two different measures of MNC activity by U.S. majority-owned foreign affiliates: panel data for aggregate real gross product in manufacturing that originates in a given host country and micro data for a single year regarding the likelihood of a firm locating in a given host country. The empirical estimates indicate that investment geared toward export markets, rather than the domestic market, is particularly sensitive to host country taxation, that this sensitivity appears to be greater in developing countries than developed countries, and that it is becoming greater over time. D 2003 Published by Elsevier B.V. Keywords: Foreign direct investment; Corporate income taxation JEL classification: F2; H2

1. Introduction Theories of foreign direct investment (FDI) have necessarily been quite eclectic, due to the variety of motivations for such investment to occur. Some draw upon factor endowment explanations, which suggest that endowments and cost conditions in host countries will be particularly important determinants of the pattern of FDI internationally. Vertically integrated multinational corporations (MNCs) are expected to establish separate * Corresponding author. Tel.: +1-641-269-3143; fax: +1-641-269-4985. E-mail addresses: [email protected] (J. Mutti), [email protected] (H. Grubert). 0022-1996/$ - see front matter D 2003 Published by Elsevier B.V. doi:10.1016/S0022-1996(03)00016-3

338

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

production processes in different countries depending upon the inputs required in each process, a strategy demonstrated by MNC outsourcing. Other theories apply to horizontally integrated MNCs that choose to exploit their special expertise by duplicating production operations abroad, thereby carrying out the same processes in many countries in order to avoid transportation costs or trade barriers and gain favorable market access. In a recent empirical test of the relative importance of these insights in explaining FDI, Markusen and Maskus (1999a,b) conclude that the principal motivation of such investment appears to be horizontal integration. A policy implication of such a horizontal investment scenario is that host country taxation may play a limited role in MNC location decisions. When all who produce in the country are subject to the same taxes, and high costs of trade and transportation make production in the host country more attractive rather than importing, there is little competition from producers outside the country who are subject to other tax regimes. If production is destined for export markets, however, as in many vertically-integrated operations, competitors are not subject to those same taxes. As a consequence, host country taxes will play a bigger role. Other authors raise a different distinction over the likely responsiveness of foreign direct investment to taxation (Baldwin and Krugman, 2000): are high-income economies able to impose higher taxes because they offer offsetting benefits in the form of better infrastructure or gains from agglomeration? If host country governments exploit such advantages by imposing higher taxes, estimates that do not take these benefits into account will understate the sensitivity to higher taxes. Research over the past decade generally indicates that foreign direct investment is sensitive to taxation in host countries (see Hines, 1997). One branch of this literature focuses on the real operations of MNCs in host countries. Such research attempts to explain variables such as sales, output, or employment, a contrast to earlier work that focused on measures of financial flows and equity ownership drawn from balance of payments statistics. This paper also addresses the real operations of MNCs, and it presents panel estimates of the determinants of the activity of U.S. manufacturing affiliates abroad. The goal is not to characterize all investment as representative of either the horizontal model or the vertical model of MNC operations, but to demonstrate how distinctions between those two perspectives, as well as differences in a countrys stage of economic development, are important in accurately measuring the importance of taxes and other variables that motivate MNC activity. The paper pays primary attention to U.S. MNC activity shown by the real gross product originating (GPO) from wholly-owned manufacturing affiliates in a given host country, a data series recently created by the Bureau of Economic Analysis (Mataloni, 1997). This series offers an important advantage over other measures that have been used to indicate real activity across countries and over time. Capital stock measures rest on historical book values, which can be misleading in inflationary times. Sales data include the value of imported inputs, which overstates the amount of activity in the host country. Value added in the host country is a better measure of the allocation of production. Estimates based on this framework confirm earlier cross-sectional estimates that taxes are an important determinant of the location of MNC activity. The sensitivity to taxes is smaller, however, when the affiliates operations primarily serve the host market, as a result of trade barriers or other factors that increase the attractiveness of that market, or when

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

339

production occurs in more developed host countries. The panel data come from three different census years, and this framework can control for unmeasured country-specific effects that do not vary over time or for country-specific components of the error term that might otherwise bias the reported importance of coefficient estimates. The panel results are supported by evidence from a different empirical framework. Information on the location decisions of individual companies, while difficult to link over time, can be examined in a single year to assess their sensitivity to a variety of factors. Based on tax return information for 1996, probit analysis of MNC location decisions is used to assess whether these same factors are relevant. Host country size and taxes again are important determinants of parent location choices. Tax sensitivity is lower in highincome countries. Because considerable industry detail is available, the expectation that greater tax sensitivity will be observed in more export-oriented industries can be addressed more directly. The estimates reported here indicate that, in industries more oriented to export markets (electronics, computers, and cars versus food, oil refining, and drugs), location decisions are more sensitive to taxes.

2. A conceptual framework for analysis 2.1. Aggregate MNC production in alternative locations A common empirical framework for analyzing patterns of international trade is the gravity model. A popular formulation applied in Frankel (1997) is to estimate the amount of trade between two countries as a function of their populations, income per capita, distance and adjacency. The larger the two countries, the greater the potential demand in one and the greater the ability to supply in the other. Distance serves as a friction that reduces trade between them, due to higher transportation costs and other costs of acquiring information about the market. Adjacency of two countries is projected to have a further effect in reducing trade frictions, in addition to the shorter distance involved. A theoretical justification for this general approach, and an explicit rationale for the loglinear functional form estimated, has been developed by Anderson (1979), Bergstrand (1985, 1989), and Deardorff (1998) under progressively more general conditions. A recent empirical application by Anderson and van Wincoop (2001) demonstrates, however, that limiting attention to the price relevant for exports from one country to another will result in a misspecification unless comparative prices from competing countries are included. A similar line of reasoning can be applied to derive a comparable framework in the case of foreign direct investment. Because the empirical application here will be for the case of production abroad by U.S. MNCs, there is not a separate focus on the United States versus other home countries. A key concern for a U.S. firm will be the demand it faces in Country A for what it produces there, QAA. For a simplified two-country exposition, let the closest substitute for this good be a somewhat differentiated one produced by the MNC in Country B. In the exposition that follows all relationships are shown in terms of the logarithms of the corresponding variables, but whether a specification linear in the levels of the variables is more appropriate is an empirical issue. For a given size of Country A’s

340

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

market, as indicated by its population and income per capita, the demand in A facing the MNC will be ln QAA ¼ a0 þ a1 ln PopA þ a2 ln GDP=capitaA þ a3 ln PAA þ a4 ln PBA ;

ð1Þ

where PAA is the price charged in Country A for the good produced there, and PBA is the price charged in Country A for the good produced in Country B.1 For an MNC that already is producing in Countries A and B, it will set PAA and PBA as a function of the market size in A and the marginal cost of production in A and B.2 The principal costs for manufactured goods considered here are labor and capital. Although knowing the wage rate would be sufficient when labor productivity is identical in both locations, an indication of differences in productivity is warranted in the cross section of countries considered here. Without an explicit measure of this variable, GDP per capita may reflect this supply influence in addition to the role it plays in demand. With respect to capital, for a given after-tax return that the firm expects to earn in any location, r, the before-tax cost of capital will be determined by the corporate income tax rate in each location, r=ð1  ti Þ . When marginal cost is constant in each country, the firms solution for PAA and PBA will be a function of market size, the wage rate in each country and the tax rate in each country, regardless of where else output from A or B is sold: ln QAA ¼ b0 þ b1 ln PopA þ b2 ln GDP=capitaA þ b3 ln wA þ b4 ln wB þ b5 ln r=ð1  tA Þ þ b6 ln r=ð1  tB Þ:

ð2Þ

The trade friction variables relevant for trade in goods play a more ambiguous role in the case of foreign direct investment. On the one hand, when local value added accounts for a large share of the products value, greater distance from the United States may encourage affiliate production in Country A, because the cost of importing a final good can be avoided. Similarly, trade barriers that protect the Country A market promote production behind a tariff wall. On the other hand, distance from the United States makes necessary inputs from a U.S. parent more expensive and may raise costs of coordination and management in the affiliate. Trade barriers raise the cost of production in the host country, and for exported goods effective protection becomes negative. If difficulties in managing a foreign affiliate are less related to distance and more related to language and 1 Considering a separate role for population and for income per capita, rather than including GDP alone, allows for potentially different effects on demand for goods that have income elasticities greater than one. That consideration may be especially relevant in the case of MNC goods designed for high-income markets. Frankels gravity model estimates rest on GDP and GDP per capita, two variables that provide the same explanatory power as population and GDP per capita and whose coefficients can be related directly to market size and stage of development. To facilitate comparisons with Frankels coefficients for international trade, the specification for GDP and GDP per capita is used in the empirical analysis. 2 Eaton and Tamura (1996) present a somewhat different argument to justify the gravity model they use to estimate patterns of direct investment abroad by Japanese and U.S. MNCs. Rather than include cost factors such as wages, they regard fixed factor endowments such as population and skilled labor as determinants of the attractiveness of producing in alternative locations.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

341

cultural differences, however, distance may be less appropriate than variables that directly reflect such differences. In sum, the expected signs of the traditional trade friction variables are ambiguous. This framework accounts for sales within Country A, but ignores the fact that some portion of Country A output likely will be exported to Country B. Demand for that production can be represented as ln QAB ¼ c0 þ c1 ln PopB þ c2 ln GDP=capitaB þ c3 ln wA þ c4 ln wB þ c5 ln r=ð1  tA Þ þ c6 ln r=ð1  tB Þ:

ð3Þ

In a world with many countries, the exact destination of affiliate As exports may not be known. Because that drawback is faced in the current study, the market scale variable is population and income per capita in the rest of the world. Additionally, in a world with many countries the cost terms from all competing locations cannot be entered separately without exhausting the available degrees of freedom necessary to identify any responses very accurately. Even treating such variables as weighted averages across the relevant set of competing countries is not straightforward. If all locations are equally substitutable, then separate attention to costs in competing countries in a cross section for a single year is not warranted, because that value will be essentially the same for all countries. Only the variation in that average over time, compared to the variation in cost for an individual country, creates a separate incentive to allocate production differently. When distinctions by region, factor endowments, or income level help define a more limited set of competing alternatives relevant for a given country, then comparisons to others within a given year are necessary. The present analysis is based on a measure of all affiliate production in a country, GPOA, which implies adding together the output sold in the host country, GPOAA, and the output sold abroad, GPOAB . Although the present study does not a priori impose a particular functional form, note that if the loglinear form proves to be most appropriate, the estimated coefficients will be a weighted average of the elasticities in the separate country sources of demand. For example, the percentage increase in GPO for a given percentage reduction in the cost of capital caused by a lower tax rate will be dGPOA =GPOA ¼ ðGPOAA =GPOA Þb5 dtA =ð1  tA Þ þ ðGPOAB =GPOA Þc5 dt=ð1  tA Þ:

ð4Þ

Suppose demand for A’s output is more elastic in export market B than in home market A, which implies c5 greater than b5 , and that exports account for a larger share of A’s output than is true in other locations. In that case the corresponding elasticity of GPO with respect to a change in taxes will be higher in country A than in countries where most of the output is devoted to the domestic market. To assess whether the sensitivity of production to costs does vary with the destination of sales, the empirical estimates include an interaction term between costs and factors that determine the firms reliance on the domestic market.

342

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

2.2. The MNC choice to produce in a given location Past research has examined the firms decision to locate in a given country (for example, see Devereux and Griffith, 1998). This all-or-nothing choice depends upon the demand factors and marginal cost differences identified above, but also upon the importance of firm-specific and plant-specific economies of scale. A horizontal model of MNC operations proposed by Horstmann and Markusen (1992) provides insight regarding what types of firms may choose to establish foreign affiliates and where they might choose to locate them. When there are large firm-specific economies of scale, perhaps from the ability to spread costs of research and development over additional units of output, and there are high costs of serving foreign markets from home country production, establishing foreign affiliates will be desirable. In industries where economies of scale are important at the plant level, the parent firm will be more selective in where it chooses to locate affiliates, and will likely favor host countries with large domestic markets. From a tax perspective, the choice to locate in a country also rests on somewhat different factors than the decision to expand production at existing locations. In particular, the taxation of inframarginal profits may be an important consideration.3 The firm will consider where aftertax profits are greatest, compared to alternative locations, in deciding where to establish an affiliate.

3. The empirical framework for analysis 3.1. Real activity measured by gross product originating (GPO) As an indicator of MNC activity, the BEA measure of real gross product originating in a given host country offers important advantages.4 In a given cross section, it better controls for activity within the country than do affiliate sales, because GPO controls for imported inputs. The BEA measure also allows more consistent consideration of changes in activity over time, due to its more systematic procedure to deflate current measures of activity; the series for manufacturing as a whole is based on an appropriately weighted average of the indices of real output in the seven subsectors of the manufacturing sector. This study relies on real GPO reported for three different benchmark census years of U.S. direct investment abroad: 1982, 1989, and 1994. Economists generally have greater confidence in census year data, as compared to annual surveys, and that distinction is relevant here because annual data frequently must be estimated for smaller affiliates that account for more activity in developing countries. The time intervals between the

3

Devereux and Griffith advocate using a composite tax measure that includes the taxation of infra-marginal rents at the statutory rate as well as determinants of the marginal effective tax rate applicable for expansion at a given location. 4 A disadvantage of the BEA measure is that it is only available for majority-owned affiliates in manufacturing as a whole. No attention can be given to industry detail, which has been important in crosssectional analysis to identify the role of trade barriers, transport costs, and plant-level scale economies (Brainard, 1997). Distinctions over different types of affiliate ownership cannot be explored, either.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

343

benchmark observations also are sufficiently separated that the observed differences in the explanatory variables are great enough to identify their effects on country output. A panel of observations across 47 countries for these three years is the basis for the empirical analysis. Host country GDP is taken from the World Bank’s World Development Indicators, based on a country’s real GDP converted into 1993 U.S. dollars using the Bank’s purchasing power parity estimate of the value of the host country currency in that year. GDP per capita and population come from this same source. Country average effective corporate income tax rates are calculated from data reported to the U.S. Treasury on Form 5471 by U.S. controlled foreign corporations in manufacturing.5 This measure is used rather than the maximum statutory tax rate imposed by a country, because it is a better indicator of the importance of other provisions in a country’s tax code that affect the attractiveness of producing there. Although marginal effective tax rates (METRs) are a preferable measure to indicate a firm’s incentive to expand output in a given location, such rates are not available for the many developing countries included in this sample. In addition, METRs are most appropriate for analyzing the effect of policy in countries where the rate will be uniformly administered for all firms, a condition that is less likely to be met in developing countries where special incentives and tax holidays are often negotiated. This study uses an average of the current year value and the value two years before, in order to reduce the random noise incorporated in the measure for any single year. This measure of taxation in the host country ignores any residual tax due in the United States when profits are repatriated. As reported by Grubert and Mutti (2001), however, actual taxes paid to the U.S. government on active nonfinancial income are relatively small: in 1996 only $5.2 billion was collected from repatriated income of over $100 billion. Even taking into account the costs of strategies to avoid such a repatriation tax, the combined cost is less than six percent of repatriated income. If the residual tax were more important, then the host country tax would provide a less effective explanation of MNC production choices. The wage variable is based on compensation per worker reported in U.S. dollars in the benchmark survey and deflated by the U.S. consumer price index. Because this value may be determined endogenously together with the amount of output in the country, an instrumental variable to represent the exogenously determined wage value is derived from World Bank data on the percentage of individuals in the relevant age cohort who receive secondary education. This variable serves as a proxy for the stock of human capital of manufacturing production workers. The trade friction variable distance is calculated as the shortest distance between either the United States west coast (Los Angeles) or east coast (Washington, D.C.) and the capital city of the host country. It is calculated based on the great circle method to connect two points on the globe, using a U.S. Geological Survey algorithm and reported at http:// www.indo.com/distance. Two alternative trade policy measures are used. One is based on a World Bank characterization of trade and exchange rate policy, in which countries are assigned to four different classes, depending upon their effective rate of tariff protection, export incentives, direct controls such as quotas, and exchange rate overvaluation. Grubert 5

The series is available in Grubert and Mutti (2000), which reports analysis for a 1992 cross section.

344

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

and Mutti (2000) assigned the most open category a value of zero, and progressively more restrictive categories values of one, two, and three.6 Although such a variable simplifies interpretation of the estimated coefficient, it may impose more structure on the effects of progressively more restrictive trade policy than is appropriate. A second measure of trade policy is based on the average tariff rate in a country as reported by the World Bank and calculated from duties paid divided by imports.7 Attention to the orientation of MNC operations toward the domestic market versus the export market may be necessary to interpret correctly the influence of various cost and demand factors in each country. Because a breakdown of affiliate sales data is incompletely available for the 47 country set, especially in 1982, the study uses the ratio of affiliate sales in the host country market relative to total sales of affiliates in that country averaged over observations for 1989 and 1994. Such a measure may depend upon the extent to which the home market is protected or higher domestic costs of production discourage exports, but also upon the attraction of being close to consumers in a larger host market. The trade restriction and market size variables are used to create an instrumental variable for the domestic market share and alternatively to demonstrate directly their influence on affiliate production. The discussion above suggests that for countries with greater reliance on domestic sales the responsiveness to the host country tax rate will be less. When variables are expressed in logarithmic form, the coefficient estimates are constant elasticities, and the formulation is similar to the common gravity model familiar from studies in international trade in goods. Simply pooling observations across countries and over the three different years, however, may mistakenly treat all observations as independent. Because the error terms for observations from a given country in fact may be related, perhaps due to unmeasured variables that differ across countries, two different estimating strategies control for this possibility. One is to allow for potential country clustering of errors in calculating robust standard errors for the pooled estimates, an approach that also allows for potential correlation of country-specific errors and the included explanatory variables. The other approach is to make random effects estimates using a generalized least squares framework, which is appropriate when the model is specified correctly and there is no correlation of country-specific errors and explanatory variables. 6 The four-value characterization is updated to 1989 and 1994 by using an index of changing trade intensity across time and countries, based on exports plus imports as a share of GDP in the following form: indexðIÞ ¼ ½ActualðIÞLowest=ðHighest  LowestÞ . See Akram (2000) for further discussion, or for a comparable calculation in another context, see UNCTADs human development index. For country i this index will equal 0 when it has the lowest trade ratio at any time across countries and years, and it will equal 1 when the country has the highest trade ratio observed. For a country whose index value doubled within any time interval, the value of its trade restrictiveness rank was reduced by one. Such an adjustment applied to Mexico, the Dominican Republic, Argentina, Turkey, and India. 7 Such a variable may show a higher average tariff rate after a policy of trade liberalization has been carried out, because previously prohibitive tariffs that did not contribute to a higher average as measured initially now result in some trade occurring and an increase in duties paid. For a handful of Latin American countries where that problem appeared, the World Bank series was compared to the zeroone trade openness variable used by Sachs and Warner (1995). If Sachs and Warner reported that a country moved from being closed to open, and the measured average tariff rate nevertheless rose, the initial imputed tariff average for the closed period was raised to the average value reported after the opening.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

345

3.2. Location decisions based on firm-level data Firm-specific information from the tax returns of U.S. MNCs is available for 1996. A cross section in a single year does not allow an examination of whether changes in a variable over time within a country have the same effect as changes across countries in the given year, an important concern in the panel estimates described above. Nevertheless, these data provide the opportunity to observe in which countries within a set of 60 an MNC chooses to operate. Such a dependent variable is free from measurement error across countries, and the probability of a firm locating in a country can be estimated as a function of various characteristics of that country and the parent corporation. This probit analysis supplements the aggregate formulations explained above by considering whether key country characteristics identified there (taxes and trade restrictions) are simply proxies for other conditions that differ by country. An advantage of this more recent cross section is that a wider set of countries and explanatory variables can be included to verify whether the tax and trade policy variables are proxies for other country characteristics. For example, Wei (2000) suggests that host country corruption deters foreign direct investment. Maskus (2000) finds that inadequate intellectual property protection reduces investment in middle- and high-income countries. Whether these various measures have independent and separately identifiable roles to play in the present framework is of interest. The opportunity to include parent characteristics can provide confirmation of insights reported in earlier research on the choice to operate as an MNC (see Caves, 1996). Firms that have more intangible capital, as generated by greater spending on R&D and advertising per dollar of sales, are more likely to invest abroad to exploit that specialized expertise. Larger and older firms are more likely to expand abroad in more locations, as they seek out other opportunities to invest retained earnings abroad. Firms that use physical capital more intensively in their home operations are less likely to invest abroad, given that the economies of scale created by these fixed costs are less likely to be achieved abroad. Conversely, parents that use labor more intensively, all else equal, are more mobile and more likely to invest abroad. The data set to be analyzed includes 728 parent firms, who have the option of locating in 60 different countries. This set of countries represents those where at least five U.S. affiliates operate in the country and it is possible to calculate an average effective tax rate for the country; the tax variable used in the probit analysis is the calculated rate for all affiliates in the manufacturing sector, not a firm-specific rate. The corruption variable, reported in Wei, is a composite from executive surveys reported in The Global Competitiveness Report 1997 of the World Economic Forum and the Harvard Institute for International Development and in the World Bank’s World Development Report; a higher value represents greater corruption. A measure of intellectual property protection is from Ginarte and Park (1997) based on host country legal provisions to protect patent rights; a higher value represents greater protection of intellectual property. The categorization of trade policy in 1996 is from the Index of Economic Freedom of the Heritage Foundation and the Wall Street Journal, which assigns a country score between one and five based on its average tariff rate, non-tariff barriers, and corruption in the customs service; a higher value represents more trade restrictions.

346

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

If all observations are not independent of each other, estimates based on firm-specific data may erroneously attribute too much significance to the coefficients obtained. All observations for a country might, for example, be subject to a much larger error because of more random government behavior in that country. That potential clustering effect by country is taken into account by reporting adjusted standard errors and by making random effects probit estimates.

4. Empirical results 4.1. Analysis of gross product originating Tables 1 and 2 report the sensitivity of GPO to the various factors presented above. Given the discussion over the potential significance of the host market orientation of

Table 1 Determinants of affiliates real gross product originating; 47 countries, 1982 – 1989 – 1994 Variable

Random effects, host market

Random effects, host market instrumt.

Pooled, World Bank policy variable

Pooled, tariff rate policy variable

Adjust for country cluster

Fixed effects

Random effects, GLS

Log GDP

0.90 (7.25) 0.39 (1.91) 2.49 (2.28) – 0.39 ( – 0.62) 0.56 (4.13) – 0.30 ( – 1.01) – 2.48 ( – 3.18) – 2.16 ( – 1.62)

0.83 (6.38) 0.66 (3.41) 3.90 (2.88) – 0.63 ( – 0.95) 0.50 (3.70) – 0.40 ( – 1.28) – 2.56 ( – 2.43) – 4.27 ( – 2.31)

0.85 (11.12) 0.71 (3.22) 3.42 (4.59) – 0.95 ( – 0.07) 0.28 (1.31) – 0.18 ( – 0.60)

0.84 (10.89) 0.59 (2.67) 3.19 (4.41) – 1.77 ( – 1.38) 0.31 (1.41) – 0.26 ( – 0.88)

0.84 (6.02) 0.59 (2.88) 3.19 (3.02) – 1.77 ( – 1.62) 0.31 (1.10) – 0.26 ( – 0.57)

1.86 (2.02) – 0.39 ( – 0.42) 1.33 (2.65) – 2.09 (1.31) 0.57 (3.76) – 0.17 ( – 0.34)

0.79 (6.73) 0.66 (3.41) 1.59 (3.40) – 0.69 ( – 1.04) 0.49 (3.62) – 0.41 ( – 1.32)

– 0.52 ( – 3.00) – 1.38 ( – 3.15) – 0.52 ( – 2.62) 0.72 (1.32) 0.75

– 0.08 ( – 3.08) – 0.15 ( – 2.98) – 0.48 ( – 2.52) 0.79 (1.47) 0.75

– 0.08 ( – 2.22) – 0.15 ( – 2.57) – 0.48 ( – 1.71) 0.79 (1.50) 0.75

– 0.04 ( – 2.29) – 0.06 (2.18)

– 0.04 ( – 2.69) – 0.07 ( – 2.58) – 0.39 ( – 1.27) 0.89 (1.02) 0.73

Log GDP/cap. Log(1 – tax) Log(1 – alt. tax) Log (real wage) Log (alt. real wage) Host sales Host sales log(1 – tax) Trade policy Trade policy log(1 – tax) Log (distance) Adjacency Adj. R2

– 0.53 ( – 1.77) 0.56 (0.66) 0.77

– 0.37 ( – 1.20) 0.92 (1.03) 0.73

Coefficient divided by standard error given in parentheses.

0.48

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

347

Table 2 Determinants of affiliates real gross product originating; 47 countries, 1982 – 1989 – 1994 Variable

Random effects, instrument for wage

Random effects, market scale and tax

Random effects, language

Random effects, income per cap. and tax

Random effects, nonOECD group

Random effects, time dummy and tax

Log GDP

0.79 (6.77) 0.98 (5.03) 1.36 (2.78) – 0.51 ( – 0.73) – 0.19 ( – 0.93) 0.06 (0.18) – 0.04 ( – 2.63) – 0.08 ( – 2.75) – 0.43 ( – 1.40) 0.75 (0.86)

0.77 (6.46) 0.95 (4.93) 3.29 (3.79) – 0.34 ( – 0.49) – 0.14 ( – 0.69) 0.17 (0.53) – 0.04 ( – 2.27) – 0.07 ( – 2.39) – 0.43 ( – 1.38) 0.79 (0.90) – 0.08 ( – 2.70)

0.78 (6.89) 0.95 (5.03) 3.23 (3.72) – 0.07 ( – 0.10) – 0.16 ( – 0.82) – 0.03 ( – 0.10) – 0.04 ( – 2.40) – 0.07 ( – 2.51) – 0.43 ( – 1.48) 0.33 (0.39) – 0.08 ( – 2.62) 1.12 (2.43)

0.76 (6.54) 0.82 (4.29) 3.99 (4.54) 0.09 (0.13) – 0.28 ( – 1.39) 0.01 (0.03) – 0.04 ( – 2.86) – 0.10 ( – 3.45) – 0.38 ( – 1.28) 0.52 (0.60) – 0.07 ( – 2.46) 0.96 (2.02) – 0.16 ( – 2.72)

0.79 (4.15) 1.72 (7.01) 4.84 (3.59) – 1.67 ( – 1.22) – 0.12 ( – 0.41) 1.26 (1.40) – 0.05 ( – 2.42) – 0.11 ( – 2.91) – 0.45 ( – 1.10) 0.71 (0.63) – 0.08 ( – 1.65)

0.80 (6.87) 0.83 (4.36) 4.74 (4.96) 0.90 (1.22) – 0.01 ( – 0.06) – 0.93 ( – 2.16) – 0.05 ( – 2.18) – 0.10 ( – 3.02) – 0.50 ( – 1.69) 0.19 (0.22) – 0.07 ( – 2.41) 1.25 ( – 2.60) – 2.94 ( – 3.38) – 0.76 ( – 2.32) 1.39 (2.85) 0.76

Log GDP/cap. Log(1 – tax) Log(1 – alt. tax) Log (real wage) Log(alt. real wage) Tariff Tariff log(1 – tax) Log (distance) Adjacency Log(1 – tax) log(GDP) English Log(1 – tax) hi incomea Log(1 – tax) D82 Log(1 – tax) D82hi Adj. R2

0.74

0.73

0.77

0.76

0.67

Coefficient divided by standard error given in parentheses. a Fourth column based on log(1 – tax)  log(GDP/capita).

affiliate production, the first two columns of Table 1 include that variable as well as those shown in Eqs. (2) and (3). Because the market orientation variable is a shorthand summary for other influences, the remaining columns in the two tables consider those factors directly. Table 1 entries particularly address the robustness of the estimates to alternative specifications that allow for possible interdependence of error terms, while Table 2 includes additional explanatory variables that allow a more precise interpretation of the determinants of affiliate production. The results reported are based on a loglinear, constant elasticity formulation. A linear version, with constant slopes and variable elasticities as

348

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

used in Markusen and Maskus’ analysis of trade flows, was much less appropriate here, as indicated by the Davidson – MacKinnon specification test.8 Begin by considering the estimates from column one. The size of the local economy, holding the stage of development constant, is important. The GDP coefficient is positive, although the value is slightly less than one. Note that in gravity model estimates based on trade data, the general observation of a GDP elasticity equal to 0.7 has been interpreted as a sign that trade does not rise by the same proportion as income because larger countries tend to be more self sufficient (Frankel, 1997). In these estimates of MNC production abroad, the greater the importance of investment designed to serve the domestic market, the greater the expected magnitude of the GDP coefficient. Because part of the affiliates output is exported, where the host country GDP makes no contribution to greater sales, a coefficient less than one is not surprising. GDP in the rest of the world varies little across countries or years in comparison with the host country GDP, and when that aggregate measure was included to reflect the potential scale of export markets, it was not at all significant and did not influence any of the reported coefficients, either. The lack of success in identifying a separate scale effect in export markets suggests that more attention must be paid to the specific markets to which each country exports. The coefficient on GDP per capita also is positive, an indication that, for a given size economy, more MNC production will occur in those countries that are similar to the United States. That result is consistent with the findings of Markusen and Maskus. It may represent the role of demand for goods designed for high-income consumers, or the ability to produce such goods in countries where productivity is higher, or other characteristics of high-income countries that reduce the cost of operating there. The coefficient of the tax variable also is significant, and the reported value of 2.49 indicates that a tax reduction that cuts the cost of capital by one percent results in 2.49 percent greater GPO in the country.9 Note that the overall effect of taxation, however, depends upon another interaction term included in the regression. If the degree to which the U.S. affiliate depends upon the host country domestic market is taken into account, the role of taxes is reduced. Not only does the coefficient for the reliance on the domestic market have a negative effect on the amount of affiliate GPO in the country, but that reliance also reduces its sensitivity to taxes. The 2.49 value applies only if all of the output is exported. Across all countries, the average share of affiliate output sold in the domestic 8 To apply this specification test, first calculate the corresponding fitted values from a linear and a loglinear formulation. Create a variable based on the difference between the fitted values from the linear version and the exponential of the fitted values from the loglinear form. When that variable is included in the loglinear model, it adds no explanatory power. Additionally, create a variable based on the difference between the fitted values from the loglinear version and the log of the fitted values from the linear version. Including that variable in the linear model adds considerable explanatory power. Applying this second step is problematic for a further reason; several of the fitted values from the linear model are negative, casting further doubt on the appropriateness of that model. 9 A possible concern in this framework is that the effective tax rate may be endogenously determined. That could occur if a lower rate is observed in years when there is rapid output expansion and rising investment in physical assets that qualify for favorable foreign treatment. To deal with that possibility in prior work, Grubert and Mutti (2000) created an adjusted tax measure based on micro-level data. The coefficients of an affiliate-level regression of affiliate taxes on the affiliate’s age, plus the age distribution of affiliates in each country, were used to construct an ‘‘adjusted’’ effective tax rate not distorted by a large flow of recent investments. Estimates based on the adjusted measure were nearly identical to those based on the original data.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

349

market is 60 percent, and therefore for that average situation the corresponding responsiveness to a one percent reduction in the cost of capital is 1.19.10 The coefficient of –2.48 for the domestic share variable indicates that greater reliance on the domestic market and fewer sales abroad results in less total output, for a given country’s size. The tax rate in alternative locations has a negative coefficient, as expected if those locations produce substitutes for the output of the country under consideration, but the estimate is not statistically significant.11 With respect to the wage variables, the wage in the host country has an unexpected positive sign that is statistically significant. The potential endogeneity of that variable is addressed in the Table 2 estimates. Given the difficulty of identifying the expected host country effect in column one, not surprisingly the wage in alternative locations is not precisely estimated, either, nor is any interaction with the domestic sales share. The distance variable has an expected negative effect on investment while the adjacency effect is positive, but neither coefficient is statistically significant. The imprecision of these estimates casts doubt on the ability of variables that traditionally are significant in an international trade setting to explain patterns of MNC affiliate production. The Table 2 estimates consider other country characteristics that might better explain MNC production decisions. The second column is based on an instrumental variable for the domestic share of sales, where its exogenous determinants are the average tariff rate and GDP. The primary effect of introducing this instrumental variable is to make the negative interaction term with taxes more significant. From the standpoint of being able to estimate a compact expression, there are advantages to showing directly the effect of greater domestic sales orientation with this variable. However, it provides less guidance with respect to government policy decisions than would a focus on the determinants of this decision. Therefore, the remaining columns of Tables 1 and 2 demonstrate the role of variables that determine this orientation. Column three replaces the domestic share variable with the measure of trade restrictiveness derived from World Bank classifications, as used in Grubert and Mutti (2000). The magnitudes of the estimated coefficients for GDP and taxes generally are quite similar to those shown in columns one and two. With respect to trade restrictions, the estimated coefficient is negative, and the interaction effect with taxes also is negative. The former coefficient suggests that the dominant effect of higher trade barriers is to raise the cost of intermediate inputs and create a disincentive for MNC production, rather than to attract firms to serve a protected market. The interaction effect indicates that a policy of tax reductions will be most successful in promoting more output when the country pursues an open trade policy and does not handicap producers with higher costs. Because the discrete trade policy groupings used in creating that variable and the sequential values assigned to them may misstate the relative degree of restrictiveness 10

The value 1.19 equals 2.49 + (0.60)( – 2.16). A direct test for the tax sensitivity of sales in the domestic versus foreign market is given by Grubert and Mutti (1991) based on benchmark survey data for 1982. Hanson et al. (2001) examine that issue in greater detail using country and industry detail for 1989 and 1994. 11 A suggestive result from a study of affiliate employment by Riker and Brainard (1997) is that employment in a developing country is competitive with employment in other developing countries but complementary with output in developed countries. No distinctions from grouping alternative production sites into high-income and low-income composites were significant in the present study.

350

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

across countries, the average tariff rate is used as an alternative measure of trade policy. The GDP and tax estimates reported in column four again are generally similar to others reported in the table. With respect to the interrelationship of tax and trade policy, just as the most closed economies would find a tax reduction unsuccessful in attracting more MNC activity, a country with an average tariff rate greater than 21 percent would face the same situation; the average tariff rate in the sample is seven percent. Columns three and four are based upon estimates that pool the cross-section data for all three years. Because the error terms may not be independent, especially for a given country, column five reports robust standard errors that take into account potential country clustering of error terms. The significance of the coefficients falls somewhat, but GDP, GDP per capita, and taxes still are significant at the one percent level and the trade policy effects at the five percent level. An alternative issue addressed by the sixth column is that there may be country-specific error terms that are constant over time. The fixed effects estimates reported there make the greatest difference to the importance of GDP and GDP per capita: the biggest changes in affiliate production appear to occur where the country market grows quickly, but areas where GDP per capita is rising most rapidly do not experience an additional increase. Alternatively stated, rising population rather than an increase in the GDP per capita of that population has a larger positive effect on the change in affiliate production. The importance of tax changes, while still significant statistically, is not as large. The seventh column presents generalized least squares estimates of a random effects model. A Hausman test reveals no systematic difference between the coefficient estimates from the fixed effects model and the random effects model, as the test statistic value of 4.76 is much smaller than 15.51, the critical chi square value with eight degrees of freedom. That result adds to our confidence that the random effects model is specified correctly and the explanatory variables are not correlated with country-specific error terms.12 Table 2 is based on random effects estimates that include several additional factors. The first column uses an instrumental variable measure of host country wages. The coefficient estimate now is negative and the wage in competing locations is positive, although neither is statistically significant. With respect to the influence on other coefficients, GDP per capita is now larger. Omitting the wage variable entirely also would lead to a downward bias in the estimated importance of GDP per capita, and including a better wage measure means that GDP per capita no longer picks up the negative effects of paying more productive workers higher wages.13 The second column allows for another determinant of the way greater domestic sales influence the tax sensitivity of domestic output. The larger the home market, the greater the likelihood that production will remain in the host country where demand is likely to be 12 Wheeler and Mody (1992) also work with panel data, but do not report fixed effects estimates because many of the explanatory variables in their data set vary little over the time period analyzed but still provide interesting insights in cross-sectional analysis. The coefficient estimates reported here are more successfully identified over the longer time frame analyzed, although the small number of benchmark years does limit the ability to consider variables that only vary by year and not by country. 13 As an alternative cost measure, unit labor cost was used in place of the wage rate. Its coefficient was negative, while the coefficient of its interaction with the tariff variable was positive. That sign pattern agrees with the results observed for capital costs, but because the labor cost effects were not significant and did not influence other included variables, they are not elaborated here.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

351

less elastic. To represent this possibility, GDP is interacted with the tax variable. The estimated coefficient on the tax term rises to 3.29, while the coefficient on the interaction term is –0.08. The mean value of the log of GDP is 25.6, and therefore for the average country the response to a one percent reduction in the cost of capital will be 1.24. For the smallest country in the sample, for which the log of GDP is 22.5, the corresponding tax elasticity is 1.49. The third column includes a dummy variable for countries where English is the native language, presumably an indicator of lower costs of communication and similar legal systems, which are additional reasons that production there will be more attractive. The coefficient of 1.12 implies that production in those countries will be three times greater (the exponential of 1.12) than in comparable non-English speaking countries. The suggestion that production in high-income countries may be less sensitive to taxes than elsewhere is addressed in column four. Such an effect is observed if an OECD dummy variable is interacted with the tax variable, or as shown in the estimates reported, when GDP per capita for those countries is interacted with the tax variable. Not only do high-income countries as a group appear to be less sensitive to taxes, but within that group, the higher the GDP per capita, the less the sensitivity. That negative interaction effect is obtained if the interaction is extended to countries with GDP per capita greater than $7500. Such a pattern is similar in some respects to the Kuznets curve for environmental quality, where higher income results in less pollution, but only for countries with income per capita greater than $5000 to $8000, depending upon the pollutant examined (Dasgupta et al., 2002). Another way of demonstrating the difference in sensitivity to taxes is to omit the highincome OECD countries from the sample. The result from doing so is given in column five. The tax coefficient, 4.84, is greater than in the full sample, 3.23. Also, the coefficient for GDP per capita is substantially larger, an indication that outside of the OECD the beneficiaries of greater MNC affiliate production have been middle-income countries, not the least developed countries with the lowest wages. These results demonstrate a pattern but do not explain the reasons it is observed. One possibility is that countries with higher GDP per capita make more expenditures for public goods and infrastructure, which in turn raise the productivity of private firms. Measures of such expenditures cannot be easily identified in the IMF’s Government Finance Statistics Yearbook. Based on items included in the categories transportation and communication and economic affairs and services, those budget shares fall as GDP per capita rises, although expenditure per capita rises. If those explanatory variables are included, greater expenditures per capita have a negative effect on affiliate GPO, but if this term is interacted with the tax term, the negative coefficient indicates that output in countries that make higher expenditures is less sensitive to taxes. Specific examples of available infrastructure are telephone lines or electricity production per person as reported in the World Bank’s World Development Indicators. Interacting such an infrastructure variable with the tax variable also suggests that taxes are less of a deterrent when more infrastructure is available. Nevertheless, none of these effects is measured very precisely, as the interaction coefficients are not statistically significant. Potential benefits from agglomeration of activity in high-income countries cannot be demonstrated easily for manufacturing as a whole, because the effects of agglomeration on

352

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

foreign direct investment are observed most clearly at an industry level (Head et al., 1999). While further exploration of these effects will require different data, the current studys reliance on the interaction variable above is an appropriate step to avoid a downward bias in the estimated role of taxes that otherwise would occur for countries where these benefits are not available. A final issue addressed in Table 2 is the likelihood that MNC operations are becoming more sensitive to taxation. To assess this tendency, interact a dummy variable for 1982 with the tax term, an approach used by Altshuler et al. (1998) in their analysis of the difference in affiliate capital stocks between 1984 and 1992. Their result, that the sensitivity of investment to host country taxation is rising, appears most strongly here if the high-income effect is treated as a dummy variable. The greater responsiveness particularly applies outside of the high-income OECD countries. In spite of vocal concern expressed by some nations within Europe over the need for greater tax harmonization, the consequences of tax competition with respect to the location of MNC activity appear to be felt to a greater extent outside of Europe. Of course, concerns over tax competition also rest on changes in the tax base and the amount of taxable income declared in a country, regardless of the amount of real activity located there, and those issues are not addressed here. 4.2. Micro-analysis of firm location decisions Table 3 reports probit estimates that show the probability of an individual firm choosing to locate in a given country as a function of parent and country characteristics. If a parent firm locates an affiliate in a given country, the location variable takes a value of one, whereas it equals zero for country locations it does not choose. Focusing on this measure of MNC activity means that no issues of measurement of capital in countries with different inflation rates and capital price distortions arise. Tobit analysis of the firms stock of real capital in the locations chosen yields similar results to those reflected in the probit analysis, with respect to the sign and significance of the explanatory variables. Variables that might more likely be relevant for a marginal increase in the amount invested did not prove to be significant, and because the Tobit estimates do not provide additional insight, Table 3 reports only the probit analysis. Again, because of the potential overstatement of the amount of independent information contained in observations for the 728 different parent companies, the potential clustering of the error terms for observations within a given country is taken into account. Also, the estimated coefficients have been transformed to show the corresponding marginal effects on the probability of choosing a given country location, evaluated at the mean values of the independent variables. Characteristics of the parent firm are reported at the top of the table. The results confirm several insights reported in Caves. Direct investment abroad is carried out by firms with special expertise or intangibles (more R&D and advertising per dollar of parent sales), and also by those parents that are larger (log of parent capital) and older (age from date of incorporation). In addition, parents whose operations are capital intensive (depreciable capital per dollar of sales) are less likely to find that foreign production allows them to achieve economies of scale, and conversely, those that are labor intensive (greater employee compensation per dollar of sales) are likely to be more mobile.

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

353

Table 3 Choice of manufacturing location for 728 parent corporations. Marginal effects for probits of affiliates in 60 countries, 1996 Variable

Probit

Parent characteristics R&D/sales 0.587 (11.65) Advertising/sales 0.38 (12.58) Capital/sales 0.090 (15.33) Labor cost/sales 0.088 (6.90) Age 0.002 (18.83) Log operating 0.034 assets (31.71) Country characteristics Log GDP 0.044 (42.35) Log GDP/capita 0.001 (0.99) Adjacency 0.089 (17.68) Log distance 0.00002 (2.16) English 0.049 (13.72) Policy variables Log(1 – tax) Log(1 – tax) high incomea Trade barriers Trade barriers log(1 – tax) Corruption

Random effects

Random effects

Random effects

Random effects

Random effects

0.573 (11.43) 0.362 (11.99) 0.093 (13.85) 0.089 (5.79) 0.002 (18.32) 0.033 (26.03)

0.575 (11.49) 0.363 (12.08) 0.093 (13.89) 0.089 (5.78) 0.002 (18.39) 0.034 (26.91)

0.575 (11.45) 0.363 (12.05) 0.093 (13.87) 0.089 (5.78) 0.002 (18.36) 0.034 (26.55)

0.572 (11.44) 0.364 (12.19) 0.092 (13.95) 0.089 (5.81) 0.002 (18.43) 0.034 (26.95)

0.567 (11.96) 0.378 (12.64) 0.087 (15.36) 0.087 (6.90) 0.002 (18.88) 0.034 (31.72)

0.042 (28.31) 0.000 (0.037) 0.092 (7.72) 0.00001 (0.593) 0.049 (9.56)

0.145 (13.66)

0.143 (9.34)

0.012 (4.68) 0.0011 (3.28)

0.012 (3.54) 0.0012 (0.92)

0.042 (28.98) 0.009 (3.07) 0.091 (8.05) 0.00003 (1.54) 0.046 (8.89)

0.042 (28.73) 0.008 (2.36) 0.091 (7.81) 0.00002 (1.32) 0.049 (9.31)

0.041 (27.68) 0.003 (0.85) 0.088 (8.53) 0.00002 (1.03) 0.055 (10.75)

0.042 (36.07) 0.001 (0.44) 0.082 (16.05) 0.00001 (0.68) 0.057 (14.84)

0.208 (9.24) 0.103 (4.23) 0.015 (4.45) 0.020 (1.64)

0.195 (8.41) 0.026 (3.23) 0.015 (4.20) 0.019 (1.49)

0.247 (11.82) 0.138 (6.10) 0.017 (4.92) 0.030 (2.43) 0.012 (5.14) 0.000005 (4.05)

0.234 (13.36) 0.131 (7.15) 0.020 (7.10) 0.046 (4.22) 0.009 (5.50) 0.000003 (0.35) 0.00002 (1.09) 0.0065 (4.94)

Intellect. prop. protection IPPhigh income IPPmed income Coefficient/standard error given in parentheses. a Column four based on log(1 – tax)  log(GDP/cap.).

354

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

With respect to country characteristics, the size of the host market, given by its GDP, again is highly significant, and the probability of choosing a country with one percent greater GDP than average increases by 0.04.14 The role of GDP per capita depends upon the other variables entered in the equation estimated, although in no case is its influence large. It does not appear to play a separate role beyond raising the income of a given population, and a firm is not more likely to locate affiliates in countries with a standard of living similar to the United States. As in the aggregate analysis, the distance variable does not seem to allow very precise predictions of affiliate location choices; the positive but insignificant coefficient may indicate that establishing an affiliate abroad is a preferred way of serving a distant foreign market. In contrast to the aggregate analysis, adjacency does appear to confer a particular advantage on Canada and Mexico relative to other potential locations. In both the aggregate and the micro-analysis the English language dummy is a significant and positive influence. Of the policy variables, lower taxes make a country location more attractive. The tax coefficient of 0.14 indicates that a one percent decline in the cost of capital due to a tax reduction results in an increase of 0.14 in the probability that location will be chosen. Compared to the influence of a one percent higher GDP, the tax effect is important. Again, trade barriers discourage MNC entry to a country. If the tax variable is interacted with the measure of barriers to trade, it generally shows that taxes have a smaller effect the less open the economy is. Allowing for potential differences in the role of taxation at different levels of host country income supports the view that location in high-income host countries appears less sensitive to host country taxation. In this analysis, the high-income group includes countries with GDP per capita greater than $10,500, and as shown by the dummy variable treatment in column three, being a member of this group results in taxes having half the influence on location that they do elsewhere. In column four, where the log of GDP per capita is interacted with the tax variable for those countries, the negative coefficient implies that for a country with an average standard of living in this group, the tax effect drops from 0.195 to 0.120. The final two columns include variables that represent host country corruption and intellectual property protection. Including those variables does not allow a convincing interpretation of their role. More important, they generally have little effect on the other country variables. Table 4 reports probit estimates for several individual industries. A smaller number of parent firms in each industry results in less precisely estimated effects. Parent character-

14 The marginal effects reported are calculated from estimates where the coefficient for GDP, say, indicates how large a change in the index z of a standard cumulative normal distribution will result from a one percent increase in GDP. The change is measured in terms of standard deviations of the cumulative normal distribution. The initial probability calculated at the mean of the independent variables is 0.0638, which corresponds to a z value of – 1.525. The untransformed coefficient on GDP is 0.345; for a tenth of a percent increase in GDP the new z value would be – 1.525 plus 0.0345, or – 1.4905. The corresponding probability would be 00.0681. Thus, the change in the probability would be 0.0043, which closely approximates one-tenth of the coefficient on GDP in the transformed representation. (The actual coefficient is calculated from the slope of the cumulative distribution at the initial value and is not based on a discrete change in GDP.)

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

355

Table 4 Choice of manufacturing location by industries. Marginal effects for probits of individual affiliates, 1996 Variable Parent characteristics R&D/sales Advertising/sales Capital/sales Labor cost/sales Age Log operating assets Country characteristics Log GDP Log GDP/capita Adjacency Log distance English

Policy variables Log(1 – tax) Trade barriers Trade barriers log(1 – tax)

Electronic 103

Cars 33

Computer 26

Food 56

Drug 35

Oil 30

0.39 (0.57) 1.60 (5.13) 0.04 (4.34) 0.12 (4.20) 0.001 (3.60) 0.02 (8.67)

2.75 (5.46) 0.28 (0.35) 0.09 (1.75) 0.06 (1.79) 0.002 (4.82) 0.02 (5.42)

0.03 (0.15) 0.68 (2.74) 0.11 (5.62) 0.29 (6.14) 0.001 (1.35) 0.05 (11.34)

3.15 (6.41) 0.55 (8.01) 0.17 (4.27) 0.43 (7.10) 0.004 (6.66) 0.06 (11.18)

1.30 (4.82) 0.28 (2.45) 0.12 (2.36) 0.69 (4.81) 0.01 (11.98) 0.05 (8.33)

1.61 (1.19) 0.02 (0.25) 0.09 (7.34) 0.49 (6.54) 0.0004 (2.55) 0.02 (7.90)

0.05 (16.00) 0.01 (1.45) 0.11 (4.40) (0.061 (2.84) 0.030 (1.80)

0.06 (11.83) 0.01 (1.32) 0.11 (2.47) 0.04 (3.16) 0.03 (1.70)

0.02 (5.65) 0.01 (2.37) 0.06 (1.79) 0.01 (0.73) 0.03 (2.67)

0.04 (6.68) 0.01 (2.13) 0.26 (1.88) 00.03 (2.56) 0.08 (4.65)

0.07 (11.276) 0.01 (1.60) 0.06 (2.08) 0.06 (4.06) 0.12 (4.20)

0.01 (4.89) 0.00 (0.35) 0.01 (.64) 0.01 (1.57) 0.01 (1.82)

0.197 (8.34) 0.033 (4.89) 0.077 (3.06)

0.236 (5.41) 0.013 (1.21) 0.053 (1.41)

0.096 (3.13) 0.024 (2.77) 0.068 (2.17)

0.056 (1.22) 0.004 (0.41) 0.018 (0.50)

0.075 (1.39) 0.032 (2.65) 0.067 (1.58)

0.018 (1.75) 0.005 (2.08) 0.011 (1.21)

Coefficient/standard error given in parentheses, number of parents below industry.

istics still show that within each of these industries, it is generally older and especially larger firms, those with less capital-intensive production at home, and those with more labor-intensive operations that are more likely to locate abroad. The relative intensity of R&D effort and advertising within an industry are not as consistently important factors in determining which parents establish affiliates abroad. The tax coefficients vary in accord with a ranking across industries constructed from data in the 1994 Benchmark Survey that shows the local market orientation of affiliate sales by industry. For all manufacturing, local sales as a share of total affiliate sales have an average value of 60 percent. Industries with above-average reliance on the local market are food (79 percent), oil refining (80 percent), and drugs (68 percent), while industries with below-average reliance on the local market are computer equipment (41 percent), motor vehicles (48 percent), and electronic

356

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

components (53 percent). Industries with above-average export orientation are more sensitive to taxation, as shown by the larger marginal tax effects and the greater significance of the estimates in the left half of the table compared to the right half. In addition, the interaction of the tax and trade policy variables tends to be more important for cost-sensitive export industries. The size of the host country market indicated by the log of GDP is quite significant in all case, but surprisingly not more so for those industries where a larger share of sales are local.

5. Implications and conclusions The primary goal of this paper is to assess the importance of policy variables, especially the role of taxation, on the direct investment abroad. Conceptually, distinctions between horizontal versus vertical integration of a firm’s operations can be expected to affect the role that host country taxation will play in MNC location decisions. Also, differences between the apparent tax responsiveness in high-income versus low-income countries may create an important distinction to be made in estimating empirical relationships and drawing policy inferences. Empirically, these relationships have not been explored very thoroughly. Nor has the robustness of more general indications of the tax sensitivity of foreign investment been demonstrated both over time and across countries. The present study uses the newly available BEA measure of real gross product originating to address these issues. In addition, parent location decisions that can be inferred from corporate tax returns for 1996 provide another check on these characterizations. Estimates confirm that export-oriented production is particularly sensitive to host country taxation, as shown when the market orientation variable is entered directly in the analysis and when its determinants are included instead. From the final column of Table 2, a reduction in the host country tax rate that decreases the cost of capital by one percent is projected to increase real value added by 2.97 percent for an average sized country outside of the OECD that pursues an open trade policy. Responsiveness to host country taxation is lower in high-income OECD countries; as income per capita rises by one percent, the responsiveness to taxation falls by 0.16. Finally, as suggested by Altshuler et al., the tax elasticity has grown over time; the corresponding tax elasticity for countries outside of the OECD would have been 0.76 smaller in 1982. The importance of a country’s trade policy and the export orientation of affiliate production is confirmed in firm-specific estimates based on a single year cross section where distinctions across individual industries are possible. Attempts to include other policy variables did not yield clear inferences. Measures of corruption and intellectual property protection were not closely correlated with taxation, however, and therefore they did not alter the patterns reported when those measures were omitted. The differences found here regarding the degree of MNC sensitivity to host country taxation suggest why not all countries are likely to adopt the same policies toward FDI. Large countries with attractive domestic markets have less incentive to offer special tax concessions to prospective entrants. A lower tax rate is less important in attracting MNC activity geared to serve that market, especially when trade barriers protect it. Over this time period at least, taxes are less of a deterrent to location in high-income economies,

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

357

perhaps because they offer better infrastructure, agglomeration benefits, or a uniquely attractive market opportunity. Those high-income countries are better able to ignore competitive pressures to cut their own taxes. An important element in the success of lowand middle-income countries seeking to attract export-oriented industries appears to have been offering lower tax rates.

Acknowledgements This work was carried out while Mutti was on leave at the U.S. Treasury Department and at the Institute for International Economics. The authors thank the editor, the referees and seminar participants at both institutions for helpful comments, but the views expressed do not necessarily reflect the position of either institution. All errors are our responsibility.

References Akram, T., 2000. A technical note on the methodology of the globalization index. A.T. Kearney. Altshuler, R., Grubert, H., Newlon, S., 1998. Has U.S. investment abroad become more sensitive to tax rates? NBER Working Paper No. 6383. Anderson, J., 1979. A theoretical foundation of the gravity equation. American Economic Review 69 (1), 106 – 116. Anderson, J., van Wincoop, E., 2001. Gravity with gravitas: a solution to the border puzzle. NBER Working Paper 8079. Baldwin, R., Krugman, P., 2000. Agglomeration, integration and tax harmonization. CEPR Working Paper No. 2630. Bergstrand, J., 1985. The gravity equation in international trade: some microeconomic foundations and empirical evidence. Review of Economics and Statistics 67 (3), 474 – 481. Bergstrand, J., 1989. The generalized gravity equation, monopolistic competition, and the factor-proportions theory of international trade. Review of Economics and Statistics 71 (1), 143 – 153. Brainard, L., 1997. An empirical assessment of the proximity – concentration trade-off between multinational sales and trade. American Economic Review 87, 520 – 544. Caves, R., 1996. Multinational Enterprises and Economic Analysis, 2nd Edition. Cambridge University Press, New York. Dasgupta, S., Laplante, B., Wang, H., Wheeler, D., 2002. Confronting the environmental Kuznets curve. Journal of Economic Perspectives 16 (1), 147 – 168. Deardorff, A., 1998. Determinants of bilateral trade: does gravity work in a neoclassical world. In: Frankel, J. (Ed.), Regionalization in the World Economy. University of Chicago Press, Chicago, pp. 7 – 22. Devereux, M., Griffith, R., 1998. Taxes and the location of production: evidence from a panel of US multinationals. Journal of Public Economics 68 (3), 335 – 367. Eaton, J., Tamura, A., 1996. Japanese and U.S. exports and investment as conduits of growth. In: Ito, T., Krueger, A. (Eds.), Financial Deregulation and Integration in East Asia. University of Chicago Press, Chicago, pp. 51 – 72. Frankel, J., 1997. Regional Trading Blocs. Institute for International Economics, Washington, DC. Ginarte, J.C., Park, W., 1997. Determinants of patent rights: a cross-national study. Research Policy 26, 283 – 301. Grubert, H., Mutti, J., 1991. Taxes, tariffs and transfer pricing in multinational corporate decision making. Review of Economics and Statistics 73 (2), 285 – 293. Grubert, H., Mutti, J., 2000. Do taxes influence where US multinational corporations invest? National Tax Journal 53 (4), 825 – 839. Grubert, H., Mutti, J., 2001. Taxing International Business Income: Dividend Exemption versus the Current System. American Enterprise Institute, Washington, DC.

358

J. Mutti, H. Grubert / Journal of International Economics 62 (2004) 337–358

Hanson, G., Mataloni, R., Slaughter, M., 2001. Expansion strategies of U.S. multinational firms. Working paper. Head, K., Ries, J., Swenson, D., 1999. Attracting foreign manufacturing: investment promotion and agglomeration. Regional Science and Urban Economics 29, 197 – 218. Hines, J., 1997. Tax policy and the activities of multinational corporations. In: Auerbach, A. (Ed.), Fiscal Policy: Lessons from Economic Research. MIT Press, Cambridge, pp. 401 – 444. Horstmann, I., Markusen, J., 1992. Endogenous market structures in international trade. Journal of International Economics 32, 109 – 129. Mataloni, R., 1997. Real gross product of U.S. companies’ majority-owned affiliates in manufacturing. Survey of Current Business April, 8 – 17. Markusen, J., Maskus, K., 1999a. Discriminating among alternative theories of the multinational enterprise. NBER Working Paper 7164. Markusen, J., Maskus, K., 1999b. Multinational firms: reconciling theory and evidence. NBER Working Paper 7163. Maskus, K., 2000. Intellectual Property Rights in a Global Economy. Institute for International Economics, Washington, DC. Riker, D., Brainard, S.L., 1997. U.S. multinationals and competition from low wage countries. NBER Working paper 5959 Sachs, J., Warner, A., 1995. Economic reform and the process of globalization. Brookings Papers on Economic Activity 1, 1 – 95. Wei, S.-J., 2000. Local corruption and global capital flows. Brookings Papers on Economic Activity 2, 303 – 346. Wheeler, D., Mody, A., 1992. International investment location decisions, the case of US firms. Journal of International Economics 33, 57 – 76.