Preliminary Draft Horizontal versus Vertical Multinationals ∗ Kazuhiko Yokota† May, 2005 Abstract A method to break down foreign direct investment (FDI) samples into either horizontal or vertical FDI is proposed. Using the implications of multinational enterprize (MNE) theories, samples of U.S. MNE activities are separated into either horizontal or vertical FDI. The results show that a large share of the U.S. FDIs is horizontal and that the types of FDIs vary with industries. The results of U.S. FDI strategies indicate that the vertical FDI tend to export back to the U.S. market more than the horizontal FDI. Skill difference plays an important role to determine the firm’s strategy. Keywords: Horizontal Multinational; Vertical Multinationals; JEL Classification: F23 ∗ I am grateful to Jim Markusen and Keith Maskus for their helpful comments and suggestions for the earlier draft. All remaining errors are, of course, mine. † The International Centre for the Study of East Asian Development. 11-4 Otemachi, Kokurakita, Kitakyushu, Fukuoka, 803-0814, Japan. Phone: 81-93-583-6202, E-mail: yokota@icsead.or.jp 1 1 Introduction A broad definition of horizontal multinational is that it maintain the whole production process in both home and host countries with headquarters in the home country. On the other hand, vertical multinational is a firm that divide the production process into more than two parts and maintain a single plant in host country keeping headquarter in home country. Furthermore, horizontal multinationals are more likely to be substituted for international trade while vertical multinationals are complement to trade. Horizontal multinationals generally have more job creation effects on host economy than vertical multinationals. The first trade-theoretic model in the line of capital flow in Heckscher-Ohlin model was developed by Mundell (1957). He concludes that trade in factors is a substitute for trade in goods. Substitutability or complementarity of foreign direct investment (FDI) for international trade in goods was analyzed as a location choice problem of MNEs in 1984. Trade-theoretic multinational enterprize (MNE) theories have identified differences in the roles of horizontal and vertical MNEs. The first formal horizontal model originates from Markusen (1984) that shows the roles of firm specific fixed cost and trade costs in the one factor, labor, framework. More recently Markusen and Venables (1998) which is now a standard model of trade-theoretic horizontal model in general equilibrium with two factors of production, skilled and unskilled labor force, show that the similarity in size and endowments in two countries likely to lead more foreign direct investment and MNEs and national firms arise endogenously. On the other hand, the trade-theoretic vertical-MNE model originates from Helpman (1984). He shows in a two-factor framework with monopolistic competition that without trade costs, MNE builds plants in different countries according to the comparative advantages. In other words, the incentive to operate headquarters in one country and production in a other arises from factor price differences across the countries. 2 Combined the horizontal motives together with he vertical motives, Markusen (1997, 2002) models ”knowledge capital” model in which horizontal and vertical MNEs arise endogenously according to the difference in size and relative factor endowments in two countries. In empirical studies, however, little attention have been paid for the different roles of the horizontal and the vertical MNEs. A few exceptions, for example, are Carr, Markusen and Maskus (2001), Aizenman and Marion (2004), and Hanson Mataloni, and Slaughter (2001). Carr et al. (2001) estimate the ”knowledge capital model” and confirm the theory. Aizenman and Miron (2004) focus on the different impacts of supply uncertainty on two types of foreign direct investment (FDI) activities, i.e., horizontal and vertical. Hanson et al. (2001) show that both horizontal and vertical FDI are important and it is difficult to distinguish these two motives from data clearly. However, they categorize three important activities by foreign affiliates; importing intermediates from their U.S. parents for further processing (vertical FDI); selling locally within the host countries (horizontal FDI); exporting from the host countries (either).1 The purpose of this paper is to break the U.S. MNEs into two types of motives, i.e., market oriented (horizontal) and comparative advantage (vertical) motives. Making use of implications of theories, including Helpman (1984), Markusen (1984), Markusen and Venables (1998), and Markusen (2002) as well as empirical findings by Hanson et al. (2001), the sample of U.S. FDI is separated into two groups of FDI and estimate the determinants of FDI strategies and the impacts on the host economies. This paper is organized as follows. Section 2 describes the methodology of separating sample into the horizontal and the vertical FDIs. Section 3 shows the estimation results. Section 4 examines the determinants of U.S. MNE strategies. Section 5 esti1 See Feenstra (2004) for the survey of empirical studies on FDI. 3 mates the spillover effects of U.S. FDI on host economies. Section 6 concludes. 2 Separating Horizontal and Vertical FDIs There are two steps in separating samples into two types of FDI activities, i.e., horizontal and vertical. In the first step, making use of the broad definition of FDI and empirical findings, sample data on U.S. FDI will be sorted in order of ex post characteristics. Ex post characteristics include the empirical findings that the horizontal MNEs tend to sell their products mainly in local (host) markets while the vertical MNEs tend to import inputs from their parent companies in home for further processing. In the second stage of separating samples, the determinants of real total sales by U.S. FDI are estimated based on the implications of trade-theoretic models. Some important factors of determinants play a crucial or an opposite role on FDI activities. This second stage estimation is also the test of horizontal and vertical FDI theories. Using the broad definitions of horizontal and vertical multinationals and Hanson et al. (2001) findings, a simple indicator to distinguish two groups of FDIs is developed. Since no firm level data on multinationals are available, it is impossible to distinguish between horizontal and vertical MNEs. Hence the results in this paper should be interpreted at the industry level. Hence, to be consistent with the theoretical literature, I will hereafter use the term horizontal (vertical) FDI to refer to the industry which has horizontal (vertical) MNE characteristics. As models by Helpman (1984) and Markusen (1984) and the empirical study by Hanson et al. (2001) indicate that the horizontal multinationals are basically trade substitute and their main concern is to sell the products in the host economy. Hence it is plausible to assume that horizontal multinationals sell their products in the host country’s market with high proportions. In this case, the ratio of domestic sales to the total affiliate’s sales is high. On the other hand, vertical strategy of a firm is likely 4 to more associated with the export of intermediate goods from home (in many cases, from parent company) to host country. Consequently, the ratio of exports from home to host over total affiliate’s sales becomes high. Let D, M , and S be domestic sales, affiliate’s imports from home country, and affiliate’s total sales, respectively. As D/S increases, a firm becomes more horizontal while as M/S increases, a firm becomes more vertical. Combining these two indices into one, D/M indicates the ex post measure of horizontal nature of a firm. As D/M increases, a firm can be said to become more horizontal and vice versa. The first step, thus, is to sort the FDI data by the measure, D/M in descending order. Samples with relatively higher D/M may contain more horizontal FDI nature and samples with relatively lower D/M are expected to show the vertical FDI characteristics. The second step is to estimate the determinants of FDI using sorted data. Let us discuss implications of the models first. Markusen and Venables (1998) have the following testable propositions: proposition 1: multinationals become more important relative to trade as countries become more similar in size. proposition 2: multinationals become more important as world income grows. proposition 3: multinationals become more dominant as host countries become more similar in relative endowments. On the other hand, the model of vertical multinationals by Helpman (1984)2 hy2 See also Feenstra (2004), chapter 11, for the comparison of theoretical models of FDI. 5 pothesizes that vertical FDI is higher when the factor endowments of the countries differ more. Helpman’s model also predicts that neither differences in country size nor the magnitude of the world income affects the FDI outflow. In other words, In Helpman’s vertical multinational model does not indicate Proposition 1 and proposition 2, while it has the opposite prediction of proposition 3. Estimated equation, thus, is: SALEijt = αi + αj + αt + β1 GDP SU Mjt + β2 |M KT DIFijt | + β3 |SKILLDIFjt | + β4 LEN DRAT Ejt + β5 DSP EAKjt + β6 T COSTijt + β7 DADJj + β8 DIST AN CEj + ²ijt (1) SALE represents the affiliate’s real sales (both in the host market and exports to the world). As Feenstra (2004) describes, the dependent variable should be affiliates’ sales in host market for horizontal FDI while affiliates’ export sales for vertical FDI. The first three terms in the right hand side represent the constant dummies, capturing industry, country and year fixed effect respectively. Term GDP SU M represents the world income and should have a positive coefficient if the sample is dominated by horizontal FDI, according to Markusen and Venables (1998). However, as Helpman (1984) predicts, if the sample is dominated by vertical FDI, the coefficient would be zero, because the difference of development stages between two countries plays no role for determining the magnitude of affiliate’s exports in his model. |M KT DIF | represents the similarity of market sizes between the U.S. and host country which is defined as the ratio of host country’s market size to the U.S. market size and the coefficient should be negative.3 |SKILLDIF | represents the difference in skilled labor abundance between the U.S. and host country which is defined as the ratio in absolute 3 Since |M KT DIF | is defined, in absolute term, as the ratio of host economy’s market size to the U.S. market size, the smaller the value means the similarity between the two markets. Market size in turn is defined as the total output minus exports plus imports by country and industry. 6 term of skilled labor abundance in host economy to the skilled labor abundance in the U.S. Skilled labor abundance by country is defined as the number of skilled labor over the total labor force. From proposition 3, the more similar the factor endowments, the more horizontal FDI occurs. Hence, the expected sign is negative for the horizontal FDI. However, the coefficient on |SKILLDIF | has two aspects. The vertical FDI reacts the relative costs of production factors and is sensitive to the skilled labor abundance in the host economy. For the vertical FDI, the more dissimilar in skill endowment, the more vertical FDI tends to occur. The expected sign for the vertical FDI is, thus, positive. |SKILLDIF | plays a key role to distinguish the horizontal from the vertical FDI. To sum up, the sign conditions of estimated coefficients are, β1 > 0, β2 < 0, β4 < 0, β5 < 0, β6 > 0, and the sign of β3 is negative for horizontal FDI and positive for vertical FDI. Last four variables are used to control the estimation. LEN DRAT E is a lending interest rate by country which is a proxy for the FDI cost. The higher the LEN DRAT E, the less FDI. Hence, the expected sign is negative both for the horizontal and the vertical FDI. DSP EAK is a dummy variable with value 1 if the host country is a English speaking country, otherwise zero. This is interpreted as another proxy for the FDI cost and the expected sign is negative. T COST is a proxy for trade cost defined as the ratio of CIF value to F OB value of international trade by country and industry. Thus, this variable directly represents freight and insurance costs. The expected sign of this variable is positive for the horizontal FDI (substitute case). However, the sign is ambiguous in the vertical FDI case, because it is generally recognized that the vertical FDI is complement with the international trade. DADJ is the adjacent dummy which takes value 1 if the host country is adjacent to the U.S. It is expected that if the host country is the neighbor of the U.S., FDI flows more. Hence, the expected sign is positive. Last dummy variable DIST AN CE is a distance between the U.S. and the 7 host country. The expected sign is negative. Sample data cover U.S. manufacturing sectors which is composed of six industries.4 Basic data come from the Bureau of Economic Analysis (BEA), Department of Commerce. Detailed descriptions on data are found in appendix. 5 To distinguish the horizontal from the vertical FDI, I estimate the equation (1) for separated samples and check the sign condition of coefficient on |SKILLDIF |. The samples containing mainly the horizontal FDI would have a negative estimated coefficient on |SKILLDIF |, while in the samples containing mainly the vertical FDI a positive sign is expected. At the same time, the magnitude and the significance of the coefficient on T COST will be checked carefully. It is expected that as the separating point of the sample moves from (A) to (G), sign conditions, statistical significance,and the magnitude may vary. First, GDP SU M is expected positive for horizontal type of multinationals but ambiguous for vertical multinationals. Second, and more importantly, |SKILLDIF | is positive for horizontal but turns to be negative for vertical multinationals. This simply reflects the fact that market motive (similarity) is important for horizontal multinationals while comparative advantage motive (dissimilarity) is important for vertical multinationals. Third, trade cost, T COST , is more important for the horizontal FDI than for the vertical FDI. Hence, it is expected to observe that the magnitude of the coefficient is larger in the horizontal FDI case than that in the vertical case. 3 Results of Estimations Table 1 shows the estimation results of estimated equation (1) with various fixed effects, such as industry, year and country. All equations, except for equations (4) and (6), 4 Six industries are, food, chemicals, metals, machinery, electric equipment and transportation, Since there are many missing values in BEA data, available data account for about 40% of total U.S. manufacturing FDI activities. 5 8 satisfy the sign conditions and many of them are highly statistically significant. The finding that the coefficients on GDP SU M are positive and those on |SKILLDIF | are negative for equations (1), (2), (3), and (5) cases suggests that the horizontal FDI dominate the U.S. FDI activities. Comparing equations (1) and (3), and (2) and (5), it is obvious that year dummy plays no role in explaining affiliates’ real sales. Comparing equations (1) and (2), and (3) and (5), it is observed that industry dummy greatly improves results. Estimations (4) and (6) which include country fixed effects have opposite signs of coefficients on |SKILLDIF | and DSP EAK. Since they have the higher adjusted R−squared and relatively low t−statistics, the existence of severe multicolinearity is suspected. So I will adopt only industry dummy in this section. Table 2 shows the estimation results when the sample is divided into two groups, one has higher D/M and the other has lower D/M . Top 20%, for example, is the sample with top 20% which is arranged in descending order along with D/M . It is assumed that top part of the sample contains the horizontal FDI, while the bottom part of the sample contains the vertical FDI. In other words, the horizontal (vertical) motive becomes weak (strong) as horizontal index (D/M ) decreases. For example, all the estimated coefficients in estimation (A) (the first column covers top 20%, and the second column covers bottom 80%) satisfy the expected signs and are highly statistically significant for both samples. Both the coefficients on |SKILLDIF | are negative and highly significant, that means both samples contain a large amount of horizontal FDI observations. The coefficient on GDP SU M in the top sample (left column) remains positive and statistically significant for all equations that is consistent with the horizontal FDI, while that in the bottom sample (right column) is positive in from (A) to (D) losing significance, and turns negative after (E) although that are not significant. This indicates that the lower D/M samples may contain the vertical FDI observations. 9 Table 1 Deternimants of Real Sales of U.S. Affiliates with Various Dummies GDPSUM (+) |MKTDIF| (-) |SKILLDIF| ( - ,+) LENDRATE (-) DSPEAK (-) TCOST (+) DADJ (+) DISTANCE (-) Obs Adj R-sqd F-value Industry Year Country (1) (2) (3) (4) (5) (6) 0.373 (6.70)** -0.379 (7.82)** -0.227 (2.12)* -0.181 (2.31)* -0.760 (7.92)** 0.564 (5.40)** 1.077 (4.64)** -0.295 (2.34)* 0.427 (8.18)** -0.349 (7.77)** -0.280 (2.81)** -0.194 (2.68)** -0.782 (8.89)** 0.586 (6.12)** 0.776 (3.64)** -0.406 (3.49)** 0.284 (4.40)** -0.473 (8.34)** -0.216 (2.00)* -0.119 (1.41) -0.752 (7.74)** 0.544 (5.19)** 1.134 (4.78)** -0.275 (2.13)* 0.763 (6.71)** -0.509 (9.98)** 0.605 (1.22) -0.094 (0.87) 0.486 (0.54) 0.067 (0.63) 0.831 (8.43)** -0.439 (9.78)** 0.088 (0.21) -0.065 (0.70) 0.147 (0.19) 0.096 (1.04) -1.149 (2.62)** 0.357 (5.84)** -0.422 (7.97)** -0.265 (2.65)** -0.143 (1.83) -0.774 (8.65)** 0.570 (5.92)** 0.823 (3.77)** -0.388 (3.27)** 1108 0.50 139.1 1108 0.58 119.6 1108 0.51 47.7 1108 0.64 56.2 1108 0.59 54.4 1108 0.74 76.1 no no no yes no no no yes no no no yes yes yes no yes no yes Dependent variable is real sales of U.S. affiliates. Absolute value of t-statistics in parentheses * significant at 5%; ** significant at 1% Coefficients of constants and dummies are suppressed. 10 -0.869 (2.30)* 11 (-) ( - ,+) (-) (-) (+) (+) (-) |MKTDIF| |SKILLDIF| LENDRATE DSPEAK TCOST DADJ DISTANCE 0.312 (4.90)** -0.381 (7.09)** -0.313 (2.71)** -0.205 (2.34)* -0.746 (7.46)** 0.612 (5.54)** 0.982 (3.44)** -0.375 (2.87)** 886 0.44 53.88 0.822 (9.48)** -0.100 (2.26)* -0.475 (3.58)** -0.009 (0.10) -0.637 (4.86)** 0.419 (3.89)** -0.437 (1.54) -0.854 (4.79)** 222 0.84 92.47 Dependent variable is real sales of U.S. affiliates. Absolute value of t-statistics in parentheses * significant at 5%; ** significant at 1% Obs Adj R-sqd F-value (+) GDPSUM (A) Top 20% Bot 80% 332 0.76 81.92 0.449 (5.99)** -0.183 (3.94)** -0.622 (5.01)** -0.114 (1.49) -0.220 (1.94) 0.441 (3.57)** 0.612 (2.63)** -0.191 (1.34) 776 0.39 38.48 0.234 (3.39)** -0.423 (7.29)** -0.301 (2.44)* -0.244 (2.51)* -0.762 (7.04)** 0.552 (4.89)** 1.181 (3.42)** -0.397 (2.83)** (B) Top 30% Bot 70% 443 0.7 81.58 0.329 (4.94)** -0.184 (4.38)** -0.728 (6.45)** -0.178 (2.43)* -0.071 (0.67) 0.584 (4.67)** 0.766 (3.41)** -0.211 (1.57) 665 0.39 34.18 0.099 (1.25) -0.52 (7.74)** -0.189 (1.39) -0.278 (2.56)* -0.812 (6.98)** 0.507 (4.30)** 0.911 (2.39)* -0.457 (3.01)** (C) Top 40% Bot 60% Table 2 Estimation by Sepa ation by Separated Samples 554 0.66 83.12 0.348 (5.42)** -0.194 (4.45)** -0.612 (5.68)** -0.162 (2.21)* -0.221 (2.23)* 0.610 (5.13)** 0.862 (3.82)** -0.192 (1.44) 554 0.41 30.12 0.055 (0.62) -0.532 (7.28)** -0.058 (0.39) -0.356 (2.88)** -0.876 (6.85)** 0.481 (3.84)** 1.070 (2.52)* -0.531 (3.22)** (D) Top 50% Bot 50% 12 (-) ( - ,+) (-) (-) (+) (+) (-) |MKTDIF| |SKILLDIF| LENDRATE DSPEAK TCOST DADJ DISTANCE 665 0.6 77.45 0.334 (5.36)** -0.209 (4.65)** -0.562 (5.48)** -0.195 (2.68)** -0.378 (3.89)** 0.677 (5.80)** 0.868 (3.86)** -0.234 (1.77) 443 0.43 26.70 -0.042 (0.44) -0.497 (6.26)** 0.217 (1.25) -0.501 (3.61)** -0.98 (7.14)** 0.416 (3.15)** 1.283 (2.68)** -0.618 (3.44)** 776 0.59 87.18 0.351 (5.95)** -0.239 (5.38)** -0.467 (4.65)** -0.207 (2.85)** -0.522 (5.54)** 0.734 (6.60)** 0.933 (4.20)** -0.199 (1.54) 332 0.43 20.34 -0.139 (1.24) -0.522 (5.77)** 0.529 (2.59)* -0.557 (3.35)** -1.153 (7.49)** 0.319 (2.14)* 1.234 (1.31) -0.668 (3.19)** (F) Top 70% Bot 30% Dependent variable is real sales of U.S. affiliates. Absolute value of t-statistics in parentheses * significant at 5%; ** significant at 1% Coefficients of constants and dummies are suppressed. Obs Adj R-sqd F-value (+) GDPSUM (E) Top 60% Bot 40% 886 0.59 97.78 0.360 (6.72)** -0.267 (6.14)** -0.457 (4.67)** -0.188 (2.67)** -0.477 (5.42)** 0.770 (6.98)** 1.246 (6.00)** -0.051 (0.43) 222 0.47 16.15 -0.111 (0.70) -0.497 (4.31)** 0.533 (2.12)* -0.493 (2.23)* -1.536 (6.84)** 0.217 (1.31) 0.614 (0.46) -0.824 (3.02)** (G) Top 80% Bot 20% Focusing on the coefficients on |SKILLDIF |, it is observed that the coefficients on |SKILLDIF | are all negative and statistically significant (high D/M case) for all cases, while the magnitude of the estimated coefficient becomes greater for all cases as the D/M is decreasing (right column). Especially in cases (F) and (G), coefficients on |SKILLDIF | are statistically significant. This result clearly shows that for all cases (from (A) to (G)), the characteristic of horizontal FDIs appears in higher D/M samples, while that of the vertical FDIs appears in cases (F) and (G). As trade cost increases, the horizontal FDI substitute for trade while vertical FDI prefer low trade cost because of intermediate goods transaction between parents and affiliates. As a result, horizontal FDI tend to have higher coefficient of trade cost variable (T COST ) and vertical multinationals are likely to have lower or even negative coefficient. The results confirm this: the coefficient on T COST in top 20% of (A) is 0.418 with highly statistical significance, while that in bottom 20% in (G) is 0.217 without statistical significance. This is again consistent with the prediction of the previous arguments. Judging from these exercises, I conclude that the horizontal FDI in the U.S. manufacturing sectors dominate vertical FDI and the former accounts for more than 70% of total multinational activities. To look into the results more carefully, I calculate the summary statistics for both the horizontal and the vertical FDI observations. Table 3A shows the summary statistics for total samples, horizontal FDI which account for top 70%, and the vertical FDI which account for bottom 30% of the sample. The first observation is that the real sales, the sum of GDPs, the trade cost variables are higher in the horizontal FDI than in the vertical FDI. The sales of the horizontal FDI are larger than the those of the vertical means that in general the large plants are more likely to be horizontal FDI. The higher sum of GDPs in horizontal FDI indicates that the FDI expands as 13 world income grows for the horizontal case. This is consistent with the proposition 2. That trade cost is also higher in the horizontal than the vertical FDI suggests that the horizontal FDI are more sensitive to the trade cost than the vertical ones. The mean of market difference is larger in the vertical FDI is also consistent with the proposition 1. Although there seems little difference in the factor endowment variable, |SKILLDIF |, the previous regression results clearly show that the factor endowment plays an opposite role in the horizontal and the vertical FDI. Next table (Table 3B) shows the number of observations in two categories, i.e., by host country’s development level and by industry. The upper part of the table shows the numbers of FDI which are included in either developing countries or developed countries. 64% are horizontal and 35% are vertical in developing samples, while 72% are horizontal and 27% are vertical in developed samples. The lower panel of the table shows the number of observations by industry and by country’s development level. Machinery, metals, and electric equipment industries have relatively higher shares of vertical FDI than food, chemicals, and transportation industries. Table 3C shows the detailed information about the number of FDI by industry, by types of FDI, and levels of host economies. For food and chemical industries, there is no clear evidence that horizontal and vertical FDI are different by FDI destinations, i.e., developing and developed countries. However, metal, machinery, electric, and transportation industries have relatively larger number of horizontal FDI than vertical FDI. An interesting finding is found in electric apparatus industry: in horizontal FDI case, 110 industries are operated in developed countries, while only 15 cases are operated in developing countries. On the other hand, in vertical FDI case, 39 industries are in developed, while 33 are in developing countries. This means that electric apparatus industry is very sensitive to the host market and cost advantage motives. In other words, electric industry changes their strategies of FDI to suit the occasion. 14 Tbale 3A Summary Statistics Total Sample HOR.INDEX SALE GDPSUM |MKTDIF| |SKILDIF| LENDRATE DSPEAK TCOST DADJ DIATANCE Mean 7.072 6.613 41.830 3.637 0.635 2.520 0.605 0.081 0.127 8.407 Std. Dev. 2.223 1.771 1.245 1.432 0.420 0.628 0.489 0.377 0.333 0.609 Min 0.000 1.674 38.615 0.021 0.001 0.770 0 -4.576 0.000 6.981 Max 13.339 11.027 44.736 8.633 1.844 4.475 1 0.583 1 9.154 Obs. 1108 1108 1108 1108 1108 1108 1108 1108 1108 1108 Horizontal HOR.INDEX SALE GDPSUM |MKTDIF| |SKILDIF| LENDRATE DSPEAK TCOST DADJ DIATANCE Mean 8.230 7.210 42.291 3.230 0.637 2.522 0.634 0.089 0.179 8.356 Std. Dev. 1.323 1.495 1.054 1.329 0.438 0.650 0.482 0.333 0.384 0.645 Min 6.159 3.618 38.800 0.021 0.001 0.770 0 -2.641 0 6.981 Max 13.339 11.027 44.736 8.633 1.844 4.475 1 0.510 1 9.154 Obs. 776 776 776 776 776 776 776 776 776 776 Vertical HOR.INDEX SALE GDPSUM |MKTDIF| |SKILDIF| LENDRATE DSPEAK TCOST DADJ DIATANCE Mean 4.364 5.216 40.752 4.588 0.630 2.516 0.536 0.060 0.006 8.524 Std. Dev. 1.390 1.569 0.961 1.194 0.377 0.575 0.499 0.463 0.077 0.498 Min 0.000 1.674 38.615 0.755 0.015 1.666 0 -4.576 0 6.981 Max 6.158 8.868 43.746 7.661 1.803 4.097 1 0.583 1 9.148 Obs. 332 332 332 332 332 332 332 332 332 332 15 Tbale 3B Number of FDI DC LDC FOOD DC LDC CHEM DC LDC META DC LDC MACH DC LDC ELEC DC LDC TRAN DC LDC TOTAL HORIZONTAL VERTICAL 323 208 115 64.4% 35.6% 785 568 217 72.4% 27.6% 87 64 23 73.6% 26.4% 82 61 21 74.4% 25.6% 180 146 34 81.1% 18.9% 135 110 25 81.5% 18.5% 117 69 48 59.0% 41.0% 61 40 21 65.6% 34.4% 125 79 46 63.2% 36.8% 20 6 14 30.0% 70.0% 149 110 39 73.8% 26.2% 48 15 33 31.3% 68.8% 73 56 17 76.7% 23.3% 9 9 0 100.0% 0.0% Developing: Low income, lower and Upper middle income countries by World Bank classification Developed: High income coutries by World Bank classification 16 Table 3C Difference in Mean HOR.INDEX SALE GDPSUM |MKTDIF| |SKILDIF| LENDRATE DSPEAK TCOST DADJ DIATANCE TOTAL FOOD CHEM META MACH ELEC TRAN 32.61** 13.18** 22.22** -12.42** 0.60 0.25 6.27** 3.11** 25.11** -6.97** 11.59** 8.41** 11.16** -6.52** 4.35** 1.39 2.84** 0.65 9.53** -3.63** 16.42** 5.22** 13.86** -8.01** 3.70** 1.73 2.20 -1.45 9.95** 1.89 24.72** 9.26** 10.94** -4.27** 1.10 0.48 9.73** 6.97** 11.87** -3.57** 13.75** 8.05** 9.22** -6.29** -6.95** -3.72** 2.25* 4.09** 13.43** -5.51** 16.06** 4.25** 7.68** -4.46** -10.62** -7.81** -3.81** 7.31** 9.41** -2.46** 17.33** 4.28** 11.15** -3.35** 5.34** 2.99** 2.49** -7.04** 6.72** -3.99** ** significant at 1%, and * significant at 5% level. tn1 n2 2 x1 x2 u n1n2 , n1 n2 17 u n1S12 n2S22 n1 n2 2 In the next section, using the results of this section, I estimate the determinants of U.S. FDI strategies, that is either targeting the host market, aiming to export back to the U.S. or export platform strategy. 4 Determinants of U.S. Affiliate Strategies Next exercise is to compare the determinants of the U.S. affiliate strategies, that is, targeting the host country’s market, exporting back to the U.S. market, or exporting to the other countries (export platform strategy). The method used for this purpose is to estimate the following equation; SALE(h)ijt = α + D∗α + β1 M KTijt + β2 U SM KTjt + β3 |SKILLDIFijt | +β4 LEN DRAT Ejt + β5 T COSTjt + γ1 D ∗ M KTijt +γ2 D ∗ U SM KTjt + γ3 D ∗ |SKILLDIFijt | +γ4 D ∗ LEN DRAT Ejt + γ5 D ∗ T COST jt + ²ijt (2) This formulation is slightly different from equation (1). First, dependent variables are U.S. affiliate sales in the host market, exports back to the U.S. market, and the other world market. Second, equation (2) does not include the world income proxy, GDP SU M . Since the purpose of this section is to estimate the determinants of the U.S. affiliate sales in the host market, the U.S. market and the other world market, the world income proxy seems not to play an important role. Hence, I drop that variable and instead add the market size of home and the U.S. separately.6 Third, to detect the significance of the difference between the horizontal and the vertical FDI strategies, I include both constant dummy and slope dummy which are value 1 if it drops in the area of bottom 30% (vertical FDI) and otherwise zero. Thus, I can discuss the 6 Since all variables are expressed in natural log form, ln(M KT /U SM KT ) in equation (1) equals ln(M KT ) − ln(U SM KT ) in equation (2). The interpretation should be the same as before. 18 difference not only between three strategies but also between the horizontal and the vertical FDI. The results of estimation in Table 4 reveal large differences both among strategies and between the horizontal and the vertical FDI. Let us discuss first the differences between the horizontal and the vertical FDI with respect to each strategy. Then, the discussion on the differences among strategies are followed. In the column (1), judging from the magnitude and the significance of the coefficients on constant (Constant and D), there is no significant difference in level of local sales between the horizontal and the vertical FDI. The horizontal FDI has a higher elasticity of market size (MKT) to local sales which is .562 than the vertical FDI which is .393 (.562-.169). The coefficient on |SKILLDIF | of the horizontal FDI has a negative sign (-.619) while a positive (.162=-.619+.781) for the vertical FDI. This means that the similarity in factor endowment promotes local sales of FDI in the horizontal FDI case but deter the local sales in the vertical FDI case. As for the export to the U.S., column (2), large difference between the horizontal and the vertical FDI is a coefficient on constant. The coefficient on constant for the horizontal FDI is negative while that of the vertical FDI is positive and the difference is statistically significant.7 This means that the vertical FDI are more likely to export back their products to the U.S. market than the horizontal FDI in level. This is consistent with Helpman (1984)’s view that the main function of vertical FDI is to export back to the home market. The coefficient on |SKILLDIF | is larger than the horizontal case with 1% statistical significance. The vertical FDI are more sensitive to the factor endowment in the host economy regarding the export back to the U.S. strategy. Interestingly, the 7 The coefficient of the horizontal FDI is -14.25 while the coefficient of the vertical FDI is 7.609 (=-14.25+21.859). The difference between them is thus 21.850 which is statistically significant at 1% level. 19 Table 4 Determinants of U.S. MNE Strategies MKT USMKT |SKILLDIF| LENDRATE TCOST D30_MKT D30_USMKT D30_SKILLDIF D30_LENDRATE D30_TCOST D30 Constant Obs Adj R-sq F-value (1) Local Sales (2) Exports to US (3) Export to Other Countries 0.562 (15.92)** 0.052 (0.82) -0.619 (6.34)** 0.093 (1.22) 0.01 (0.08) -0.169 (2.36)* 0.19 (1.22) 0.781 (3.97)** 0.116 (0.77) -0.006 (0.03) -4.118 (1.39) -3.182 (2.67)** 0.578 (6.02)** 0.488 (2.77)** -1.623 (6.22)** 0.112 (0.52) -0.537 (1.50) -0.562 (2.79)** -0.635 (1.55) 3.534 (6.39)** -1.282 (2.79)** -0.141 (0.20) 21.859 (2.76)** -14.25 (4.18)** 0.26 (2.31)* 0.385 (1.91) -1.03 (3.45)** -0.331 (1.34) 1.479 (3.43)** -0.151 (0.65) 0.269 (0.58) 0.982 (1.55) -2.628 (4.79)** -1.676 (2.01)* 2.549 (0.28) -5.209 (1.32) 1115 0.66 195.44 707 0.17 14.16 684 0.13 10.44 D30s stand for dummy variables with value 1 when it drops in the area of bottom 30% of the sample. Absolute value of t-statistics in parentheses * significant at 5%; ** significant at 1% 20 coefficient on LEN DRAT E is positive in the horizontal FDI while negative in the vertical FDI with statistically significant difference,8 indicating that the vertical FDI are sensitive to the FDI cost in the host country while the FDI cost plays little important roles in horizontal FDI with respect to the export to the U.S. market strategy. Export platform (exporting to the other market) strategy, column (3), shows the almost same trend with the export to the U.S. market strategy but weaker connections between dependent and independent variables. Crucial difference from export to the U.S. strategy is appeared in the coefficient on T COST . The horizontal FDI has a positive coefficient on T COST (1.479) and the vertical has a negative coefficient on T COST (-0.197). The similarity in factor endowment spurs the export to the other market strategy for the horizontal FDI while the similarity deters that strategy in the vertical FDI. Let us discuss the differences among three strategies, that is, local sales, exports to the U.S., and exports to the other countries. The coefficient on M KT is a positive and statistically significant for all strategies but larger in both local sales and export back to the U.S. strategies indicating a large market size is still important for export back to the U.S. strategy. This is consistent with Yeaple (2003) on the U.S. MNE strategy. Only the export to the U.S. strategy has a significant coefficient on U SM KT . This indicates that the larger the U.S. market, the more likely the FDI takes “export back to the U.S.” strategy. As I discussed previously, this is true for both the horizontal and the vertical FDI. The coefficient on |SKILLDIF | is negative and significant for all three strategies in the horizontal FDI while positive in the vertical FDI with respect to local sales and export back to the U.S. strategies. It indicates that both the horizontal and the vertical FDI are more sensitive to the factor endowment similarity of host economy in export strategy than in local sale strategy. 8 The difference is -1.282, and the coefficient of the horizontal FDI is 0.112. Hence the coefficient of the vertical FDI is -1.17. 21 5 Spillover Effects of FDI In this section, I estimate the effects of spillovers by FDI on the host economy. The following equation is estimated: T F Pitj = α + β1 SKILLjt + β2 SIZEijt + β3 EM P LOY EEijt + ²ijt . (3) All variables are in natural logarithmic forms. The dependent variable is the level of total factor productivity by industry and country, which is defined as ln T F Pijt = ln GDPijt − sijt ln Kijt − (1 − sijt ) ln Lijt , where Kijt is capital stock and Lijt is the number of labor force in country i and industry j in time t. sijt is the capital expenditure share. All necessary data for T F P calculation are obtained from the World Bank. Independent variable SKILL is defined as the share of skilled labor to the total labor force and SIZE is the average output per firms (including both local and MNEs), which is in turn defined as ln SIZE = ln(output) − ln(number of f irms). EM P LOY EE is the number of employees (in the industry) who are working for MNEs. Table 5 shows the estimation results for two samples, top 70% and bottom 30%. As discussed, the former sample is likely contain more horizontal FDI and the latter more vertical FDI. Estimation group (1) excludes EM P LOY EE as an explanatory variable, while group (2) includes EM P LOY EE. The reason is twofold: first to check the multicollinearity between EM P LOY EE and other variables, and the second to check the robustness of estimation. Judging from the results, it seems that there is no multicollinearity and estimates are robust. The estimation results for group (1) reveal that skill abundance positively affects the levels of productivity in both horizontal and vertical cases. This is consistent with 22 Table 5 Spillover Effects of MNEs (1) Horizontal MNEs Top 70% Vertical MNEs Bot 30% 1.061 (6.15)*** 0.218 (4.21)*** 1.656 (8.72)*** -0.204 (2.93)*** Constant 3.849 (6.51)*** Observations Adj. R-sq F-value 469 0.11 28.91*** SKILL SIZE (2) Horizontal MNEs Top 70% Vertical MNEs Bot 30% 8.329 (11.97)*** 1.012 (5.71)*** 0.185 (3.33)*** 0.118 (2.01)** 3.011 (4.07)*** 1.751 (8.24)*** -0.234 (3.03)*** 0.098 (1.75)* 8.009 (11.33)*** 229 0.26 42.08*** 457 0.11 19.92*** 218 0.25 25.45*** EMPLOYEE Dependent variable is the level of TFP. See text for the definitions of variables. Absolute value of t-statistics in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% 23 the previous literatures9 ; the host country benefits more from FDI when it is more skilled labor abundant. However, one striking result is that horizontal and vertical samples have the opposite signs of estimated coefficients on SIZE. This means that as output per firm increases, the level of productivity increases in that industry for the horizontal case. The reverse is true for the vertical sample. This is also consistent with Aitkin and Harrison (1999) finding, that the host economy benefits more from vertical FDI if the industry is relatively small. These results do not change if the EM P LOY EE variable is added as an independent variable in the right hand side as shown in group (2). The results of group (2) show that the number of workers employed by MNEs positively affects the level of productivity. This finding in turn may support the formulation of spillover. As the number of workers hired by MNEs increases, the possibility that those workers move to local firms increases, thereby increasing the total productivity of the industry. 6 10 Conclusions This paper proposed a methodology to distinguish the horizontal FDI from the vertical FDIs, making use of implications of trade-theoretic models as well as empirical studies. Estimated results clearly showed the differences in characters between horizontal and vertical FDI. In separating the sample into two groups of FDI, factor endowment played a crucial role. Basic idea to distinguish two types of FDI is that factor endowment similarity spurs horizontal FDI while dissimilarity spurs vertical FDI.11 One of this paper’s novel contributions to the literature is to reveal that the horizontal and vertical 9 See Yokota (2004) for a detailed survey. Since the data on skilled labor employed by MNEs are not available, total number of employees employed by MNEs is used. 11 It should be noted again that this is an implication from trade-theoretic FDI models, not the result of my exercise. 10 24 FDI coexist in U.S. manufacturing sector, which is consistent with the findings of Hanson, Mataloni, and Slaughter (2001). And the horizontal FDI dominates U.S. MNE activities, which in turn is consistent with Brainard (1993). This paper also identified that the U.S. horizontal FDI tend to go developed countries with 72% share while the U.S. vertical FDI go developed countries with 64% share. These figures indicate that the vertical FDI do not necessarily a phenomenon of the North-South only while a horizontal production integration occurs even between dissimilar endowment countries. Another exercise revealed that the vertical FDI are more likely to export back to the U.S. market. This is consistent with the definition of vertical FDI by Helpman (1984). On the other hand, the horizontal FDI are more likely to target the local market of the host country depending on the magnitudes of market size and factor endowment of the host economy. These results are almost consistent with the fact findings by Yeaple (2003), although he did not estimated the horizontal and the vertical FDI separately. As I have shown, in determining the MNE strategy, U.S. horizontal FDI have quite different motives from the vertical FDI. Then, the determinants of the level of productivity in the host economy was identified. Interesting results are: firstly, skill abundance in both the horizontal and the vertical FDI positively affect host country’s productivity. This is consistent with the previous literatures. Secondly, the number of workers in MNEs also positively affects the productivity. Thirdly, average output per firm positively affects productivity in the horizontal FDI case, but negatively affects productivity in the vertical FDI case. This is also consistent with the previous finding discussed earlier. As shown, the horizontal and the vertical FDI behave quite differently. This is a reason why the separation of groups of FDI is necessary to understand the MNE issues empirically. 25 A Definitions and Data Sources of Variables* Variable SALE Dimension i×j×t GDP SU M j×t |SKILLDIF | j×t |M KT DIF | i×j×t LEN DRAT E DSP EAK j×t j×t T COST j×t DADJ DIST AN CE GDP P C j j j×t SIZE i×j×t DF OOD i DCHEM i DM ET A i DM ACH i DELEC i D i×j×t TFP i×j×t EM P LOY EE i×j×t Definition sales by majority-owned non-bank U.S. affiliate in host country (BEA), measured in 1995 constant price using GDP deflator (IFS, IMF). sum of U.S. real GDP and host country’s real GDP, constant prices, Laspeyles. (PWT) absolute value of host country’s skilled labor share over U.S. skilled labor share. Skilled labor is defined as the sum of occupational categories 0/1 (professional, technical, and kindred workers) and 2 (administrative workers) divided by total number of workers. The nearest values are used for fulfill the missing values for avoiding the loss of degree of freedom. (ILO)** absolute value of market size difference between host and U.S. market. Market size (total output minus exports plus imports) in host country minus market size in U.S. (WB) proxy for investment cost defined as lending rate of host country. (IMF) another proxy for investment cost defined as a dummy variable with 1 for non-English speaking host countries. proxy for trade cost defined the freight costs by country and industry, CIF/FOB. (Feenstra) dummy variable with value 1 if the host country is adjacent to the U.S. distance between U.S. and the host country. real GDP per capita in host countries. constant prices, Laspeyles. (PWT) proxy for scale economy. Output per firm by industry in the host country. (WB) dummy variable with value 1 if the industry is manufacturing foods. ISIC code 311 and 313. dummy variable with value 1 if the industry is manufacturing chemicals and products. ISIC code 351 and 352. dummy variable with value 1 if the industry is manufacturing metal and metal products. ISIC code 372 and 381. dummy variable with value 1 if the industry is manufacturing machinery. ISIC code 382. dummy variable with value 1 if the industry is manufacturing electric equipments and electronics. ISIC code 383. dummy variable with value 1 when the observation drops in the area in bottom 30% in the sample, i.e., it is vertical industry. level of total factor productivity, proxy for the industry productivity. See text for the definition. the number of workers hired by MNEs. (BEA) *All variables are in natural logarithmic forms. ** The number of skilled labor in U.S. affiliates by industry and country is available only for 1989, 1994, and 1999 (these are benchmark survey conducted by BEA), I used 1989 data for the periods between 1983 and 1991, 1994 data for between 1992 and 1995, and 1999 data for between 1996 and 2000. 26 BEA: (http://www.bea.gov/bea/di/di1usdop.htm) PWT: Penn World Table (http://pwt.econ.upenn.edu/php site/pwt index.php) ILO: International Labor Organization (http://laborsta.ilo.org/) WB: World Bank data base, “Trade and Production, 1976-1999.” IMF: International Financial Statistics data base. Feenstra: World Trade Flows, 1980-1997 B Country, Industry and Data Periods Data have three dimensions, i.e., country, industry and time. Data includes 44 countries, 6 industries and 17 years (1983 - 1999): Countries: Argentina, Australia, Austria, Canada, Chile, China, Colombia, Costa Rica, Denmark, Ecuador, Egypt, Finland, France, Germany, Greece, Guatemala, Honduras, Hong Kong, India, Indonesia, Ireland, Italy, Japan, Korea, Malaysia, Mexico, Netherlands, New Zealand, Norway, Panama, Peru, Philippines, Portugal, Singapore, South Africa, Spain, Sweden, Taiwan, Thailand, Trinidad and Tobago, Turkey, United Kingdom, Venezuela. Industries: Food, Chemicals, Metal, Machinery, Electric Machinery, and Transportation. Time Periods: Data availability varies greatly depending on country. So panel is unbalanced. 27 References [1] Aitkin, Brian and Ann E. Harrison (1999)“Do Domestic Firms Benefit from Foreign Investment? Evidence from Venezuella,” American Economic Review, 89, 605-618. [2] Aizenman, Joshua, and Nancy Marion (2004), “The Merits of Horizontal versus Vertical FDI in the Presence of Uncertainty,” Journal of International Economics, 62, 125-148. [3] Brainard, S. Lael (1993), “An Empirical Assessment of the Factor Proportions Theory,” NBER working paper no. 4269 [4] Carr, David, James Markusen, and Keith Maskus (2001), “Estimating the Knowledge-Capital Model of the Multinational Enterprise,” American Economic Review, 91, 691-708. [5] Feenstra, Robert C. (2004), Advanced International Trade, Princeton University Press, Princeton. [6] Hanson, Gordon, Raymond Mataloni, and Mathew Slaughter (2001), “Expansion Strategies of U.S. Maultinational Firms,” NBER working paper no. 8433. [7] Helpman, Elhanan (1984), “A imple Theory of Trade with Multinational Corporations,” Journal of Political Economy, 92, 451-471. [8] Markusen, James (1984) “Multinationals, Multi-plant Economies, and the Gains from Trade,” Journal of International Economics, 16, 205-226. [9] Markusen, James (2002), Multinational Firms and the Theory of International Trade, MIT Press, Cambridge. [10] Markusen, James and Anthony Venables (1998), “Multinational Firms and the New Trade Theory,” Journal of International Economics, 46, 183-204. [11] Yeaple, Stephen (2003), “The Role of Skill Endowments in the Structure of U.S. Outward Foreign Direct Investment,” Review of Economics and Statistics, 85(3), 726-734. [12] Yokota, Kazuhiko (2004), “Comparative Advantage and Vertical Multinaitonals,” ICSEAD Working Paper series, 2004-34. 28