MASTER THESIS
An Empirical Study of the
Knowledge-Capital Model of Multinational
Enterprises
Student: Liu Yan 1444115
Supervisor: Dr. Padma Rao Sahib
Co-assessor: Dr Beppo van Leeuwen
Contents
1 INTRODUCTION 5
2 LITERATURE REVIEW 6
2.1 Knowledge-Capital Model . . . 6
2.1.1 Assumptions of the Model . . . 7
2.1.2 Predictions of the Model . . . 9
2.2 Empirical Research . . . 10
3 CONCEPTUAL FRAMEWORK 13 3.1 National Firms, Horizontal MNEs or Vertical MNEs? . . . 13
3.1.1 Equilibrium Equations . . . 13
3.1.2 Horizontal MNEs . . . 15
3.1.3 National firms and Vertical MNEs . . . 18
3.2 Development of Hypotheses: National Firms or MNEs? . . . 18
3.2.1 Total Market Size . . . 19
3.2.2 Difference in Market Size . . . 20
3.2.3 Difference in Factor Price . . . 20
3.2.4 Investment Costs . . . 22
3.2.5 Trade Costs from Country i to j . . . 22
3.2.6 Trade Costs from Country j to i . . . 23
4 METHODOLOGY 23 4.1 Sample . . . 23
4.2 Empirical Model and Methods . . . 24
4.3 Measures of Variables . . . 26
4.3.1 Dependent Variable . . . 26
4.3.2 Market Size . . . 27
4.3.4 Investment Costs . . . 29
4.3.5 Trade Costs . . . 29
4.3.6 Distance . . . 30
4.4 Tests of Assumptions of the Multiple Regression Model . . . 32
5 RESULTS 36 5.1 Positive Skill Difference Part . . . 36
5.1.1 Descriptive Statistics . . . 36
5.1.2 Regression . . . 37
5.2 Negative Skill Difference Part . . . 40
5.2.1 Descriptive Statistics . . . 40
5.2.2 Regression . . . 40
6 CONCLUSIONS 43
7 APPENDIX 45
List of Tables
1 Three firm types . . . 8
2 Review of Previous Research . . . 11
3 Variable, measure, data source and predicted sign . . . 31
4 Normality Test . . . 32
5 Normality after Data Transformation . . . 33
6 Correlations . . . 35
7 Correlations without data for USA . . . 35
8 Descriptive statistics for variables: positive part . . . 37
9 Regression result: positive part . . . 38
10 Descriptive statistics for variables: negative part . . . 41
11 Regression result: negative part . . . 42
12 Countries in the dataset . . . 46
13 Definition of difference professions . . . 47
List of Figures
1 Theoretical Framework . . . 192 Difference in Factor Price . . . 21
1
INTRODUCTION
The Knowledge-Capital Model (KK Model) of the Multinational enterprise (MNE) devel-oped by Carr, Markusen and Maskus (CMM)(2001) combines both horizontal and vertical motivations for foreign direct investment (FDI) and considers them as a function of coun-try characteristics. Horizontal MNEs offer roughly the same product or service in many locations, aiming at taking advantage of the market size effect and avoiding trade costs. Whereas, vertical MNEs geographically fragment production process into stages and place the unskilled-labor-intensive activities in unskilled-labor-abundant countries in order to exploit factor cost differences. KK model creates an theory-driven empirical framework for studying FDI. According to the theory, horizontal MNEs dominate when the total market is large, countries are similar in size and factor price, and trade costs are high. Vertical MNEs dominate when one country is small and skilled-abundant, and the trade costs are not excessive. National firms dominate when one country is large and skilled-abundant, and the trade costs are low. Therefore, market size (total and differences), skill differences (which is short for differences in relative costs of skilled-labor), trade costs are the main country characteristics which determine the volume and direction of FDI.
CMM (2001) test the hypotheses developed from KK model and the results show that the model fits well and gives strong support to the theory. However, the original empirical research is questioned by many other studies, especially with respect to the data coverage and the skill difference variable, which lead to little evidence for vertical FDI. For instance, the variable specification (Blonigen, Davies and Head, 2003), its expected influence on the dependent variable (Davies, 2002) and the choice of measure (Braconier, Norback and Ur-ban, 2005) have elicited considerable debate. In other words, although the theory is widely accepted, more empirical support is needed to get a general consensus.
more direct proxy of skill difference, factor price difference, instead of factor endowment difference in CMM; (c) using the FDI stock data to measure the dependent variable since the sales data in CMM capture the FDI activity in a short time period (one year normally) which might be easily influenced by short run shocks; (d) using an extensive dataset, cov-ering 31 home countries and 57 host countries, in the year 2000 and 2003 to prove the long term validity of the theories.
The organization of this paper is as follows. Section 2 reviews relevant literatures, includ-ing theoretical framework developed by CMM and some other empirical research. Section 3 illustrates the hypotheses. This is followed by the explanation of the methodology. The sample, data collection and measures are introduced. Section 5 discusses the estimation results. The last section draws some conclusions.
2
LITERATURE REVIEW
In this section, the KK model is introduced first. This is followed by a review of some em-pirical research, mainly focusing on three debatable issues, namely data coverage, measure of skill difference variable and its influence on FDI.
2.1
Knowledge-Capital Model
2.1.1 Assumptions of the Model
In the KK model, there are two countries (i and j), and two homogeneous factors of pro-duction (L-unskilled labor and S-skilled labor), which are internationally immobile. Goods X is produced with increasing returns to scale, and is subject to Cournot competition with free entry and exit. Two types of activities are required to produce this product. On the one hand, the firms should undertake some headquarter services, for instance manage-ment, accounting, marketing, innovation and R&D activities, leading to firm-level fixed costs. The headquarter services are skilled-labor intensive and require knowledge-based assets. On the other hand, production and assembling activities which are unskilled-labor intensive take place, generating plant-level fixed costs. Therefore, the firms choose the locations of headquarter services and production activities in order to maximize the profit according to characteristics of both home country and host country1. The different location
combinations of these two activities result in three firm types in the model:
1. National or domestic firm (type di and dj2 firm) locates the headquarter and plant
in the home country and serves foreign market by exports.
2. Horizontal multinational (type hi and hj firm) places the headquarter services in the
home country and duplicates the domestic plant in the host country. As a result, the plants produce similar products in both countries.
3. Vertical multinational (type vi and vj firm) maintains the skilled-labor intensive
headquarter services in the skilled-labor abundant home country, and unskilled-labor intensive production activity in the unskilled-labor abundant host country. If needed, it may re-export the product to home country.
1Home country is defined as the country in which the headquarter activities are located. Consequently,
the other country is host country
2The subscript represents the home country. For instance, d
i firm is the domestic firm which locates
We see that compared with national firms and vertical MNEs which have one headquarter and one plant, horizontal MNEs have two plants in two countries which will incur an extra cost. However, two plants serves both home and foreign markets so that for horizontal MNEs, they do not have trade costs any more.
The KK model is built on the characteristics of knowledge-based assets which are highly Table 1: Three firm types
Firm type i j trade cost?
di H,P i ⇒ j dj H,P j ⇒ i hi H,P1 P2 NO hj P2 H,P1 NO vi H P j ⇒ i vj P H i ⇒ j
related to multinationals, generating three basic assumptions for the KK model (CMM, 2001).
1. Fragmentation: the location of knowledge-based assets may be fragmented from pro-duction.
2. Skilled-labor intensity: knowledge-based assets are skilled-labor-intensive relative to final production.
3. Jointness: the services of knowledge-based assets are joint (“public”) inputs into multiple production facilities.
assets by multiple plants.
2.1.2 Predictions of the Model
To explore the conditions under which one particular firm type dominates the market, CMM(2001) build a general-equilibrium framework. Equilibrium is determined by pricing equation (marginal revenue equals marginal cost) and free-entry conditions (profits are nonpositve). Then they get the profit functions of these three firm types with all the variables. They can see the influence of one variable on the profits by changing it while keeping the others constant. As a result, the relationships between the variables and the choice among the three firm types are clear. Finally, they combine horizontal MNEs and vertical MNEs, the choice is simplified to being MNE or national firm. They rely largely on the simulation results to predict the relationships between the country variables and volume and direction of FDI for the empirical research. The country characteristics are market size, differences in size and skill, and trade and investment costs.
According to CMM (2001), type-di firms3 will be the dominant type active in i if
1. i is both large and skill-abundant. The logic is very simple. Firms locate the head-quarter in the country which is skilled-abundant and plants which is large. Since country i is both large and skilled-abundant, firms put both headquarter and plants there, resulting in type-di firms.
2. i and j are similar in size and relative skill and transport costs are low. When the country are similar, there is no incentive for type-v MNEs. And two-plant type-h MNEs have not advantage when transport costs are low over one-plant type-d firms. 3. Foreign investment barriers in j are high.
Type-hi firms are the dominate type active in country i if
3From here on, I focus on firms headquartered in country i. Analogous rules apply to firms in country
1. Both i and j are large which favors locating plants in both countries.
2. i and j are similar in size and relative skill. If countries are dissimilar in either size or relative skill, one country will be “favored” as a site of plant or headquarters (Markusen, 2002). That is, if countries are similar in relative skill but different in size, then type-d firm will locate in the larger country and export to small country to avoid an extra cost of setting up a plant in the small country. If countries are similar in size but of different relative endowments, then type-v firms headquartered in the skill-abundant countries are preferred.
3. Transport costs are high since type-d firms don’t have transport cost. Type-vi firms will be dominant in i if
1. Country i is small and skill-abundant. Since country i is skill-abundant, firms locate headquarter in i and plant in j because country j is relative large.
2. Transport costs are not high. Sometimes, the products are re-exported from j to i. Therefore, the trade costs should not be too high.
The advantages of the KK model over previous theoretical models are obvious. It allows for simultaneous horizontal and vertical motives for FDI. Furthermore, the model fully indigenizes international trade which is endogenous in generating the predictions of the model as other factors (CMM, 2001).
2.2
Empirical Research
Table 2: Review of Previous Research
Sample Year Influence of
skill variable Measure of skill variable Empirical technique CMM (2001) US inward and outward FDI 1986-1994 linear (positive) proportion of number of skilled labor to total employment WLS & Tobit BDH (2003) 1, US inward and outward FDI 2, OECD data 1982-1992 curvilinear (positive first and negative) OECD data: mean years of education.
OLS & Tobit
Davies (2002)
Ibid Ibid Figure 2b Ibid OLS
BNU (2005) Swedish and US outward FDI 1986, 1990, 1994, 1998
positive relative wage OLS
variable influence, and measure of skill variable in Table 2.
Blonigen, Davies and Head (BDH, 2003) argue that CMM’s empirical framework mis-specifies the terms measuring differences in skilled-labor abundance. Since CMM only use U.S. data, U.S. is almost always the large and skill abundant country. Therefore, the re-search is actually limited to one corner of the simulated diagrams. Pooling U.S inward FDI data from 1984 to 1992 and U.S. outward FDI data from 1983 to 1992, as well as OECD dataset from 1982 to 1992, and allowing for the effects of skill differences on either side of zero, BDH find positive influence of skill difference on FDI when the host country is skill abundant, and negative effect when the home country is skill abundant. For the OECD dataset, the measure of relative skill is the country’s mean years of education.
the relative wage difference is a strongly significant variable explaining vertical FDI.
3
CONCEPTUAL FRAMEWORK
In the theory review section, I summarize the predictions of the KK model in a non-technical way. Herein, I borrow the general- equilibrium framework (CMM, 2001) derived from pricing equation and free-entry condition. Using the partial derivative, we can see the influence of the variables on FDI. Therefore we know when one of the three firm types can dominate the market. This is the first part of this section.
As mentioned, the testable implications for CMM’s empirical model are derived from sim-ulation results. I try, in the second section, to establish the hypotheses directly from the theory which combine the conclusions of type-h and type-v firms.
3.1
National Firms, Horizontal MNEs or Vertical MNEs?
3.1.1 Equilibrium Equations
The following profit equations are borrowed from CMM (2001) and I skip the derivation of the equilibrium equations.
Πhj = β " Mi pi− zic pi 2 + Mj pj− zjc pj 2# − [zj(F + G) + ziG] (4) Πvi = β " Mi pi− zj(c + t) pi 2 + Mj pj − zjc pj 2# − (ziF + zjG) (5) Πvj = β " Mi pi− zic pi 2 + Mj pj− zi(c + t) pj 2# − (zjF + ziG) (6)
In the equations, all the important variables appear.
M is the national income of the country, representing the market size; z denotes the wage rate of skilled labor; c is the constant marginal production cost; t is the transport cost which is measured by the amount of skilled labor needed to transport one unit of good X from one country to another; G is the plant-level and F the firm-level fixed cost, both measured in units of skilled labor.
The equations indicate the profits of different firm types which are calculated by subtracting fixed costs from markup revenues. For instance, equation (1) shows the profit of national firm headquartered in country i. The first part β
Mi pi−zic pi 2 + Mj p j−zi(c+t) pj 2 is the markup revenue of the firm, containing markup revenue from country i βMi
pi−zic
pi 2
and that from country j βMj
p
j−zi(c+t)
pj 2
. Since the products are exported to country j, we can see an extra transport costs is removed from markup revenue from country j. Fixed costs are the sum of firm-level and plant-level fixed costs. Since the headquarter and plant are located in country i, the fixed costs for firm di are zi(G + F ).
From the equations, it is worth noting that at a given level of world income, product prices and factor prices,
2. The fixed costs of type-h MNEs are also higher than the other two firm types since type-h MNEs have multiple plants but the latter two only have single plant;
3. A type-di firm and a type-vj firm have the same markup revenues, according to
the profit equations (1) and (6). Similar story applies to type-dj and type-vi firms.
The difference lies in the fact that both of them have their plants in country i, but headquarters in different countries. Then the wage costs of skilled labor in country i and j determine the profit differences of the two firm types.
Actually, the first two conclusions have already been clear when we have the definitions of the three firm types (Table 1). With the equations, we can do further calculation and see the results directly.
3.1.2 Horizontal MNEs
As to the question that when type-h MNEs dominate, we need some mathematical analysis. I take the partial derivatives of the profit equations with respect to one particular variable to see the influence of that variable on the profits. Suppose initially that the countries are identical, which means that pi = pj, zi = zj, Mi = Mj, Gi = Gj and Fi = Fj.
• Total market size
The size of both country i and country j increases, ∂Mi = ∂Mj ≥ 0
∂Πhi = ∂Πhj ∂Πd i = ∂Π d j = ∂Π v i = ∂Π v j ≥ 0 (7)
Since type-h MNEs don’t bear trade costs, the increase in total market size increases type-h MNEs’ revenues more than other two firm types. Therefore, we conclude that:
Type-h MNEs will tend to dominate when the overall market is large. • Difference in size
Country i becomes larger and j smaller while holding total market size constant4,
∂Mi = −∂Mj ≥ 0
∂Πdi = ∂Πvj ∂Πhi = ∂Πhj = 0 ∂Πdj = ∂Πvi (8) .
Since type-di and type-vj firms focus the sales mainly on country i, now the larger
country, this change results in the increases in the profits of these two firm types. Whereas, type-dj and type-vi firms are worse off from this change. Type-h MNEs
are indifferent to this change since we have assumed that the countries are identical in product and factor prices and total market size dose not change. Although type-h MNEs are not influenced negatively, the change is most favorable to type-di and vj
firms. As a result, we know that:
Type-h MNEs will tend to dominate when the two countries are similar in size. • Difference in factor price
The wage of skilled labor decreases in country i and increases in j while holding total wage constant, ∂zj = −∂zi ≥ 0
∂Πdi ∂Πh i 0 ∂Π h j ∂Π d j (9)
Since type-di firms have their headquarter and plant in country i which now has
cheaper skilled labor, their markup revenues increase because the production costs (zic) decreases and fixed costs (zi(G+F )) decreases. Therefore, they benefit the most
from this change. It is followed by type-hi MNEs whose fixed costs fall and revenues
are kept fixed since they produce in both country and total wage does not change. The revenues of type-hj MNEs are also unaffected but their fixed costs increase.
Type-dj firms have their revenues fall and costs raised. Note that the influence of
this change on the profits of type-v firms is ambiguous since we don’t know the rela-tionship between G and F . Since it is not the type-h firms which benefit the most, we have the conclusion that
Type-h MNEs will tend to dominate when the two countries are similar in factor price.
• Trade costs
Trade costs raise, ∂t 0
∂Πhi = ∂Πhj = 0 ∂Πdi = ∂Πdj = ∂Πvi = ∂Πvj (10)
Since type-h MNEs do not incur trade costs, they are not affected by this change. Whereas other firm types have their revenues fall therefore they are worse off. As a result, horizontal MNEs have the tariff-jumping incentive for FDI. We conclude that
3.1.3 National firms and Vertical MNEs
Opposite to the conclusions of last part, we know that type-d or type-v firms will tend to dominate when total market is small, countries are dissimilar in size or factor prices, or trade costs are low.
From profit equations (1) and (6) we already see that the only difference between a type-di
firm and a type-vj firm is the fixed costs which are determined by the factor prices zi and
zj. Therefore, factor prices are the crucial variables to decide which type is more profitable
between national firms and vertical MNEs. According to inequality (8), we know that when country i is relative large, type di firm or type vj firm could have advantage since
they have the plant in the larger country. Then comparing the profits, Πv
j− Πdi = −zjF − ziG + ziG + ziF = (zi− zj)F ≥ 0 ⇒ zi ≥ zj ⇒ Country j is skilled-labor
abundant
we see that when country j is skilled-labor abundant (and small), it is profitable to locate the skilled-labor intensive headquarter services there, resulting in vj MNEs. Conversely,
when country i is skilled-labor abundant (and large), di is preferable.
Type-v MNEs will tend to dominate when one country is small and skilled-labor abundant. However, if the large country is skilled-labor abundant as well, the factor-price motive for placing the headquarter there plus the market-size motive for locating the plants there will result in type-d firms dominating.
3.2
Development of Hypotheses: National Firms or MNEs?
ideas from other studies to make the model more general and convincing. The country factors which might have effects on FDI from the home country to the host country are the total market size, differences in size and factor prices5, and trade and investment costs. Figure 1 shows the theoretical framework of this paper.
Figure 1: Theoretical Framework
3.2.1 Total Market Size
Inequality (7) shows that type-hi MNEs benefit the most from the increase in the
to-tal market size as they do not bear trade costs. Furthermore, this change also rise the profits of type-vi MNEs. Therefore, the increase in total market size will give rise to the
increase in FDI from the home country to the host country. I test the following hypothesis:
5Throughout the empirical part, country i denote the home country and country j the host country
. The analysis will mainly focus on hi, vi and di firms which locate the headquarter in home country i.
Therefore, FDI from i to j is comprised of two elements, hi and vi. Furthermore, the difference in market
Hypothesis 1: The larger the total market, the greater the FDI from the home coun-try to the host councoun-try.
3.2.2 Difference in Market Size
According to the inequality (8), when country i becomes a larger country (which means difference in size is positive), the profits of type-vi MNEs drop since their plants and main
markets j becomes smaller. On the other hand, when country j becomes larger (which indicates difference in market size becomes negative), type-vi firms will benefit the most.
Therefore, for vertical FDI from country i to j, it is increasing in size difference when host country is larger while it is decreasing in the size difference when the home country is larger. From theoretical part, we know that similarity in market size favor the horizontal FDI, which follows the similar patten as vertical FDI. As a result, we conclude that FDI from the home country to the host country has an inverted U-shaped relationship to dif-ference in market size.
Hypothesis 2: When the market size difference is negative, FDI from the home country to the host country increases; When the market size difference is positive, FDI decreases.
3.2.3 Difference in Factor Price
Skill difference variable has brought about a lot of discussions in two aspects. One is about the relationship between this variable and FDI, the other is about the measurement. I discuss the first one here, and the second aspect later in the methodology section.
head-quartered in the skilled-abundant country. However, as some authors (Blonigen, Davies and Head, 2003; Davies, 2002) mentioned in their studies that KK model actually predicts a non-monotonic relationship between skill difference and FDI.
As shown in Figure 2a, on the one hand, type-hi FDI will decrease as the divergence of
Figure 2: Difference in Factor Price
skill difference and it reaches its maximum when the two countries are the same in factor prices. On the other hand, for type-vi FDI, it only flows from the skill-abundant country
to unskilled, which means type-vi FDI only occurs on the positive side of x-axes, where
indicates country i is relatively skilled-labor-abundant. Furthermore the amount will in-crease in the skill difference. Putting them together, we have Figure 2b, which represents the relationship between this variable and FDI from country i to country j. Therefore, I test the following hypothesis:
3.2.4 Investment Costs
The effect of investment costs is straightforward. When the investment barriers are high for instance in terms of some restricted policies, FDI decreases to avoid the extra costs and risks. Whereas, if FDI is encouraged by some favorable policies in the host country, then FDI to this country will increase. Therefore, it is hypothesized that:
Hypothesis 4: The higher the investment costs, the less the FDI from the home country to the host country.
3.2.5 Trade Costs from Country i to j
As far as trade costs are concerned, the model I explained in the last section is somewhat simplified. It does not distinguish the trade costs form country i to j and that from country j to i. In fact, trade costs from country i to j is related to the tariff-jumping incentive for horizontal FDI (hi). When it comes to type-vi FDI, since the firms may re-export the
product from country j where they locate the plants to country i, the trade costs from j to i matters in this case.
Therefore, for trade costs from i to j, type-hi MNEs benefit from the increasing trade costs
since they avoid these costs but di firms do not.
3.2.6 Trade Costs from Country j to i
As mentioned, trade costs from j to i only affect vi firms. Obviously, they are worse off
when these trade costs rise since they have their markup revenue drop. Therefore, I test the hypothesis:
Hypothesis 6: The higher the trade costs from country j to country i, the less the FDI from country i to j.
4
METHODOLOGY
In this section, I mainly describe the sample, empirical model, measures of variables and data sources.
4.1
Sample
In the previous studies, three sets of data are used widely. CMM (2001) use the two-way FDI data with the U.S. over the period from 1986 to 1994, and the U.S. is either country i (home country) or j (host country) in every observation. Blonigen, Davies and Head (2003) use U.S. inward FDI data from 1984 to 1992 and U.S. outward FDI data from 1983 to 1992. OECD data set from 1982 to 1992 are also used by several other authors (Bloni-gen, Davies and Head, 2003; Braconier, Norback and Urban, 2005). As to the sample, I make changes in terms of data coverage, research year and sample separation.
Firstly, I use the data from FDI Country profiles provided by UNCIAD WID(United Na-tions Conference on Trade and Development World Investment Directory)6. 112 economies’
FDI information in the year 1990 to 2003 are available, including two way FDI flows and stocks, by industry and geographical origin (for inward FDI) or destination (for outward
FDI); activities of transnational corporations (TNCs); and the legal framework within each country/economy. In this paper, I use FDI stock data to measure the dependent variable. I choose FDI stock data instead of sales (as CMM) or flow which is also in the country profile for the following consideration. Sales and flow reflect FDI activity in a short time period (one year normally) which may be easily influenced by a short-run fluctuation. Whereas stock represents the accumulated quantity of FDI till the period in question which might indicate the long-term strategy of investment. Totally there are 31 home countries and 57 host countries in the dataset of this paper (See Appendix).
Secondly, previous empirical research are on the basis of data from the middle of 1980s to the middle of 1990s. I use the updated date for the year 2000 and 2003 to see whether the theory reflect a long run trend and has general application.
Last but not least, since the variable skill difference has a non-monotonic effect on FDI, I separate the data into two subsets as Davies (2002), “positive” part and “negative” part in which the skill differences are positive and negative respectively. From Figure 2b we see that the negative part only represents horizontal FDI and the skill differences affect FDI positively. And the positive part contains both horizontal and vertical FDI and shows a U-shape relationship between skill differences and FDI.
4.2
Empirical Model and Methods
Ordinary least squares (OLS) regression is used to estimate the empirical model in this paper7. Although CMM (2001) use weighted least squares (WLS) and Tobit regression,
the following studies find the OLS results are qualitatively similar to those of WLS and Tobit procedures and many of them (BDH, 2003; Davies, 2002; Braconier, Norback and Urban, 2005) use OLS regression to estimate the model.
The dependent variable F DIij is the FDI stock from the home country i to the host country
j at a given year (2000 or 2003). The independent variables are the country characteristics
expected to have influence on the dependent variable, including total market size, differ-ences in market size and factor prices, investment and trade costs, and the geographical distance between the countries. Therefore, the basic estimation equation is:
F DIij = β0 +β1∗ SU M GDP + β2∗ GDP DIF SQ + β3∗ SKDIF F
+ β4∗ SKDIF SQ + β5∗ (GDP DIF F ∗ SKDIF F )
+ β6∗ IN V C + β7∗ T CIJ + β8(T CIJ ∗ SKDIF SQ)
+ β9∗ T CJI + β10∗ DIST +
Total market size is denoted by SU M GDP = GDPi+ GDPj. I anticipate its sign to be
positive.
Difference in market size is GDP DIF F = GDPi − GDPj, since we expect an inverted
U-shape relationship between difference in size and FDI, we test the square form of size difference, GDP DIF SQ = (GDPi− GDPj)2, which I expect to have a negative sign.
Difference in factor prices and its square are given by SKDIF F = SKi − SKj and
SKDIF SQ = (SKi − SKj)2. For the negative part, as shown in Figure 2b, β3 is
ex-pected to be positive and β4 to be 0. As to the positive part, I expect β3 to be negative
and β4 to be positive.
Investment cost variable is represented by IN V C and trade costs from country i to j and from country j to i are T CIJ and T CJ I respectively. I expect their coefficients to be negative, positive and negative.
CMM (2001) add two terms to interact the skill difference with GDP difference and trade costs from i to j.
For the variable GDP DIF F ∗ SKDIF F , CMM are not sure about the sign, but suggest a weak negative influence. According to them, FDI is highest when the country is small and skilled-labor abundant. This is about type-vi MNEs which lose the most from increase
MNEs dominate the market when countries are similar in size and relative skill, the differ-ence in size and skill affect horizontal FDI in the same direction. Therefore, I anticipate a negative influence as CMM, but may expect different results.
In the same way, we can have the predicted sign of variable T CIJ ∗ SKDIF SQ since trade costs from country i to j encourage type-hi MNEs but it is increasing in skill similarity.
Therefore, this interaction term is also negatively affect the dependent variable.
The last variable is DIST , geographic distance between countries. According to CMM (2001), it is not clear theoretically what the sign of the distance effect should be because it is an element in both export costs and investment and monitoring costs. Since most of the previous research (Brainard, 1997; CMM, 2001 and 2002; Davies, 2002) find a negative effect, I also expect the sign to be negative in this paper.
4.3
Measures of Variables
There are six main variables in this paper, namely FDI, GDP, factor price, investment costs, trade costs, and distance. Then some calculations are needed based on the original data for GDP and factor price. In this part, I introduce the measures of the variables and data source.
4.3.1 Dependent Variable
The dependent variable F DIij, as just mentioned, is measured by the FDI stock obtained
4.3.2 Market Size
Market size is proxied by the country’s gross domestic product (GDP) derived from pur-chasing power parity (PPP) calculation for the year 2000 and 2003. The data is collected from the online database “The CIA World Factbook”8 The Central Intelligence Agency
(CIA) publish the survey annually which contains the information about economic, politi-cal and geographic conditions of 267 countries and economies. Key data are grouped under the headings of geography, people, government, economy, communications, transportation, military, and transnational issues. The GDP data is obtained from the economy category. The data of SU M GDP and GDP DIF SQ is calculated based on the definitions. Further-more, I also use the data of official exchange rate in the world factbook to convert all the FDI data to US dollars.
4.3.3 Relative Skill Difference
As I mentioned, skill difference is a debatable variable. Besides the adaption I made con-cerning the direction of the influence of this variable on FDI activity, herein, I make a modification on the proxy of this variable. CMM (2001) use the proportion of skilled labor to total employment as a measure of skilled-labor abundance. However, relative skill abun-dance could not capture the skill variable in the theory directly for the following reasons. The motivation for vertical FDI is the factor price differences among the countries. Fur-thermore, in the general-equilibrium framework, factor price is also used by CMM. Factor price and factor endowment seem to be equivalent to them. However, the relative factor price is not highly correlated to the relative factor endowment according to the correlation test by Braconier, Norback and Urban (2005). They explain the reason why the link be-tween relative factor endowment and relative factor price is so weak is because of the labor market distortions, taxes, non-homothetic preferences and measurement errors. Therefore,
8Date for 2000 is obtained from: http://www.umsl.edu/services/govdocs/wofact2001/index.html and
the relative factor price is more theoretically suitable for this paper.
The data of factor price is collected from the survey Prices and Earnings provided by Union Bank of Switzerland (UBS). In every three years, the surveys are conducted by employees of foreign UBS offices, correspondent banks, consumer organizations, chambers of commerce and universities in 70 cities around the world. For the ‘earning’ part, the data is taken from the questionnaires with 8 questions each on wages, wage deductions and working hours across 13 different professions (UBS, 2000 and 2003). The skill data in CMM (2001) is collected from Yearbook of Labor Statistics by the International Labor Organization. They define the skilled labor as workers in occupational categories 0/1 (pro-fessional, technical, and kindred workers) and 2 (administrative workers). In the similar way, I select product manager, department head, engineer and skilled industrial worker in this survey as the skilled labor and calculate the average as the measure of the skilled wage, and building labor9 as the unskilled labor.
There are also some problems of the data. Firstly, the survey is carried out in the lim-ited industries, for instance, metalworking industry, pharmaceuticals, chemicals and food industry. However, they are highly relevant for manufacturing (Braconier, Norback and Urban, 2003). Furthermore, the data is for a particular city, mostly the capital of the country or the most economically developed city of the country, but not for the country. However, it suits the paper well since FDI appears to be concentrated on the economic centers of the country.
The relative price of skilled labor is represented by the ratio of the wage of skilled labor to the wage of unskilled labor.
SKi =
wi s
wi l
A low relative price corresponds to the skilled-labor abundance of the country. Therefore, the skill difference is measured by
SKDIF Fij = SKj− SKi = wj s wlj − wi s wi l . 4.3.4 Investment Costs
CMM (2001) use the index of impediments to investment, reported in the World Competi-tiveness Report of the World Economic Forum, to measure the investment costs. However, the data are proprietary and I cannot get them. Therefore, I turn to the World Develop-ment Indicators, annually provided by the World Bank10. Indicators are organized in 6
sections: world view, people, environment, economy, states and markets, and global links. The data for this paper falls in the category of economy, the investment climate. The author believe that lack of access to credit is one of the biggest barriers entrepreneurs face in starting and operating a business. The PRS Group’s composite International Country Risk Guide (ICRG) rating is a widely used index for investor to assess risks and costs. The PRS Group collects data on 22 components of risk in three subcategories, political, financial and economic risks. Then the risk ratings for these subcategories are combined based on a formula to the country’s composite rating from 0 to 100. The highest rating means the lowest risk, therefore the lower investment costs. According to the hypothesis, low investment costs encourage FDI. Then the measure, ICRG rating should be positively related to FDI. The data for December of 2000 and 2003 are used in this paper.
4.3.5 Trade Costs
The trade cost index from CMM (2001) is taken from the same source as investment cost index which is defined as a measure of national protectionism, or efforts to prevent importation of competitive products. As mentioned, the data are proprietary, I use the
data from Economic Freedom of the World 2005 Annual Report (Gwartney, Lawson and Gartzke, 2005), including 127 nations for 2003 and the most recent years. The indices measure the degree to which the policies and institutions of countries are supportive of economic freedom. Thirty-eight components and sub-components are used to construct a summary index (ranging from 0 to 10) and to measure the degree of economic freedom in five areas: size of government, legal structure and protection of property rights, access to sound money, freedom to exchange with foreigners, and regulation of credit, labor and business. The indices which are chosen for this paper to measure the country’s trade costs are from the fourth category. The first two subcategories, taxes on international trade (comprising three elements, taxes as percentage of exports and imports, mean tariff rate, and variability of tariff rates) and regulatory trade barriers (with two elements, hidden im-port barriers, and costs of imim-porting) exactly represent the country’s tariff and non-tariff barriers to trade. I calculate the average of the two indices for the year 2000 and 2003 to measure the country’s trade costs in this paper. The lower index indicates higher trade costs. According to the theory, higher trade costs to the host country (T CIJ ) encourage horizontal FDI which is stimulated by tariff-jumping incentive. Therefore, the coefficient of T CIJ variable is expected to be negative. Consequently, it is expected that the coefficient of the interaction term T CIJ ∗ SKDIF SQ is positive. In the same way, we can derive that the predicted sign of coefficient of T CJ I is positive.
4.3.6 Distance
The measure of distance is the number of kilometers between the capitals of the countries, which is collected from online City Distance Tool11 provided by Geobytes, inc. Entering the names of origin and destination cities, you can get the distance between the two cities in kilometers.
I summarize the measure, data source and predicted sign of the variables in Table 3.
Table 3: Variable, measure, data source and predicted sign
Variable Predicted
sign
Measure Data source
F DIij
Dependent Variable
The FDI stock from country i to j
UNCIAD WID country profile
SU M GDP + GDP (PPP) The CIA World
Factbook
GDP DIF SQ - Ibid Ibid
SKDIF F negative part: +
positive part:
-SK: relative wage of skilled labor
Union Bank of Switzerland
SKDIF SQ negative part: 0
positive part: + Ibid Ibid GDP DIF F *SKDIF F - (GDPi− GDPj) (SKi− SKj) Ibid IN V C + Index of investment costs of entering country j World Development Indicator, ICRG rating
T CIJ
-Index of trade costs of exporting from country i to j Economic Freedom of the World T CIJ *SKDIF SQ + T CJ (SKi− SKj)2 Ibid T CJ I +
Index of trade costs of exporting from country j to i
Ibid
DIST - kilometers between
capitals
4.4
Tests of Assumptions of the Multiple Regression Model
The multiple regression model relies on several assumptions. In order to ensure the validity of the results and conclusions, I discuss and test the assumptions next.
According to Harnett (1982), four assumptions of multiple regression should always be tested, namely normality, linearity, homoscedasticity and nonmulticollinearity.
First of all, regression assumes that variables are normally distributed. I use skewness and kurtosis test to inspect the normality of the variables. The values of skewness and kurtosis should be within the interval of [-2,2]. However, according to Table 4, we see that the two values for dependent variable, F DIij, are much larger than 2 which indicates a big
problem about the normality of this variable. In order to improve normality, I take the natural logarithm of the variable and Table 5 shows the change. Both values are reduced to the range of (-1,1).
Table 4: Normality Test
N Skewness Kurtosis
Statistic Statistic Std. Error Statistic Std. Error
F DIij 1233 6.768 0.070 54.651 0.139 SU M GDP 1233 1.843 0.070 1.80 0.139 GDP DIF SQ 1233 1.385 0.070 1.988 0.139 SKDIF F 1233 1.073 0.070 1.851 0.139 IN V C 1233 -0.556 0.070 -0.288 0.139 T CIJ 1233 -0.853 0.070 0.269 0.139 T CJ I 1233 -1.255 0.070 1.363 0.139 DIST 1233 0.617 0.070 -0.498 0.139
Table 5: Normality after Data Transformation
N Skewness Kurtosis
Statistic Statistic Std. Error Statistic Std. Error
LnF DIij 1233 -0.826 0.070 0.450 0.139 SU M GDP 1233 1.843 0.070 1.80 0.139 GDP DIF SQ 1233 1.385 0.070 1.988 0.139 SKDIF F 1233 1.073 0.070 1.851 0.139 IN V C 1233 -0.556 0.070 -0.288 0.139 T CIJ 1233 -0.853 0.070 0.269 0.139 T CJ I 1233 -1.255 0.070 1.363 0.139 DIST 1233 0.617 0.070 -0.498 0.139
constant across all the independent variables, which is called homoscedasticity. These two assumptions are checked by examining the residual plots, plots of the standardized resid-uals against standardized predicted values. Linearity is shown when points form a line. Homoscedasticity is shown when the residuals are evenly scattered around the horizontal line. From Figure 3, we see the data generally meet the two assumptions.
Figure 3: Linearity and Homoscedasticity Tests
difference are very large. Therefore, I delete the U.S. data and get the correlation shown in Table 7. Without U.S., the SU M GDP and GDP DIF F SQ are not highly related and the problem is solved. However, we can not study the world economy without the U.S., so we include the data of the U.S. in the regression.
Table 6: Correlations
SU M GDP GDP DIF SQ SKDIF F IN V C T CIJ T CJ I Distance
SU M GDP 1 GDP DIF SQ 0.916** 1 SKDIF F 0.035 0.025 1 IN V C -0.013 -0.025 -0.320** 1 T CIJ -0.088** -0.046 -0.213** 0.777** 1 T CJ I -0.162** -0.159** 0.170** 0.024 -0.002 1 DIST 0.208** 0.176** 0.215** -0.142** 0.147** -0.135** 1 N=1233
**:Correlation is significant at the 0.01 level(2-tailed).
Table 7: Correlations without data for USA
SU M GDP GDP DIF SQ SKDIF F IN V C T CIJ T CJ I Distance
SU M GDP 1 GDP DIF SQ 0.700** 1 SKDIF F 0.055 0.054 1 IN V C 0.031 0.042 -0.316** 1 T CIJ -0.109** -0.092 -0.210** 0.780** 1 T CJ I -0.041** -0.038** 0.189** 0.024 -0.010 1 DIST 0.136** 0.102** 0.208** -0.137** -0.140** -0.120** 1 N=1069
5
RESULTS
The estimation results are reported and discussed in this section. Since the data is divided into two parts, positive and negative part, with positive skill difference and negative skill difference, the results are also represented in two parts.
5.1
Positive Skill Difference Part
5.1.1 Descriptive Statistics
Table 8 shows the descriptive statistics of the variables. The maximum FDI stock is be-tween UK (home country) and the Netherlands (host country) in the year 2003, indicating a horizontal FDI since this occurs between two advanced countries which are both large, and similar in size and relative factor prices. The GDP summation of U.S. and China in the year 2003 is the largest one in the dataset. As to the size difference, Sweden and Switzerland in 2003 are the most similar countries in size. Whereas, the difference in size between U.S and Estonia in 2003 are the largest of all. According to the original data, there are a couple of countries with same relative factor price, for instance, Italy and Japan (SKDIF Fij=1.94), Hong Kong China and Finland (2.09) in the year 2000, Ireland and
Table 8: Descriptive statistics for variables: positive part
N Minimum Maximum Mean Std. Deviation Unit of
measurement
F DIij 780 0.1 284678.0 6915.15 22246.76 millions of US dollars
LnF DIij 780 -2.74 12.56 6.58 2.78
SU M GDP 780 84.2 17429 2823.10 3444.71 billions of US dollars
GDP DIF SQ 780 3 1.2E+08 1.4E+07 32902037.27
SKDIF F 780 0 13.72 2.23 2.51 SKDIF SQ 780 0 188.23 11.26 25.62 GDP DIF F ∗ SKDIF F 780 -18384.99 139535.5 2783.15 12704.24 IN V C 780 54.8 91.0 76.09 8.35 T CIJ 780 4.65 9.70 7.99 1.21 T CIJ ∗ SKDIF SQ 780 0 1554.12 83.4152 200.46 T CJ I 780 6.3 9.5 8.71 0.61 DIST 780 55 19633 6514.12 4798.23 km 5.1.2 Regression
Regression results shown in Table 9 are encouraging, especially with respect to skill dif-ference variable which is the main concern of this paper. The R square shows that 37.7% of variation in the FDI from home country to host country could be explained by this estimation regression model. Moreover, from F test which capture the overall significance of the model, we see that the model is significant at the level 0.01.
inter-action terms.
Size Variables
Both total GDP and the square of GDP difference consistently affect FDI with the hypoth-esized sign. SU M GDP exerts a positive effect on FDI, confirming that increasing total market encourages FDI. Furthermore, the negative coefficient of GDP DIF SQ verifies that FDI increases in size difference when host country is larger and decreases in size difference when home country is larger.
Table 9: Regression result: positive part
Variable Coefficient t-Statistics Sig Sign as predicted/Significant?
Constant -11.063 -6.677 0.000
SU M GDP 7.580E-04 -8.538 0.000 Yes/Yes
GDP DIF SQ -5.24E-08 -8.538 0.000 Yes/Yes
SKDIF F -0.432 -4.496 0.000 Yes/Yes
SKDIF SQ 0.146 5.182 0.000 Yes/Yes
GDP DIF F ∗ SKDIF F 4.218E-05 5.204 0.000 No/Yes
IN V C 1.643E-02 1.017 0.310 Yes/No
T CIJ 0.381 1.461 0.076 No/No
T CIJ ∗ SKDIF SQ 1.54E-02 -4.653 0.000 Yes/Yes
T CJ I 1.455 10.678 0.000 Yes/Yes
DIST -3.59E-05 -2.028 0.043 Yes/Yes
N=780 R2=0.377
F=46.615 (Sig=0.000)
Skill Variables
coefficients of these two variables, β3 and β4, I calculate the estimated inflection point
(point i in Figure 2b), which is 1.4812. It means that FDI starts to rise when the skill
difference is 1.48. Since the skill difference ranges from 0 to 13.72 with the average of 2.23 (see Table 8) and the positive slope part characterizes vertical FDI, we can see the vertical FDI in this model which obtains few empirical supports previously.
Cost Variables
Among the four cost variables, only the hypotheses of T CJ I and DIST are significantly supported. Trade costs from country j to i negatively13 affect vertical FDI since type vi
need to re-export the products from j where they place the plants to home country i. Geographical distance is negatively related to FDI, which is consistent with other research. As mentioned by CMM (2001), the sign of this variable is ambiguous in theory. It is elements of trade costs from i to j, trade costs from j to i, as well as investment costs. On the one hand, distance may favor (horizontal) FDI which tends to avoid the transport costs, indicating a positive relationship. Notice that this is similar to hypothesis 5. On the other, transport costs for vertical MNEs who export products back to the home country increases because of distance. In addition, distance might also result in larger investment and monitoring costs. Then we would expect a negative effects of this variable on FDI, which is highly related to hypothesis 4 and 6. The estimation results support hypothesis 5 to some extent.
However, T CIJ ’s tariff-jumping incentive for horizontal FDI is not supported. Although investment costs influence FDI negatively as predicted, the relationship is not significant. I will discuss these two variables later.
Interaction Terms
The sign of interaction term between difference in size and skill is significantly positive.
12i = −β3 2β4 = −
−0.432 2∗0.146 = 1.48
13Please note that the positive coefficient indicates a negative relationship because of the measurement.
The result does not support the hypothesis but is not unexpected. As a matter of fact, the sign of this variable is theoretically unclear. As mentioned, for vertical FDI, the effect should be negative, while positive for horizontal FDI. Since horizontal FDI is the highest when countries are similar in size and skill. The estimation results show the relationship for horizontal FDI.
The second interaction term affect FDI negatively and significantly as predicted. Since trade costs from i to j is only related to horizontal FDI positively, while skill difference disencourage type-hi FDI.
5.2
Negative Skill Difference Part
5.2.1 Descriptive Statistics
Table 10 shows the descriptive statistics for variables. It is generally consistent with those in the positive part. We find the average FDI stock is around 1/3 larger than that in the positive part. The largest FDI stock is from country pair UK (home country) and the Netherlands14 (host country) in 2000. By definitions of the two parts, we already know
that in the positive part, skill difference data as well as average of course is all positive and in the negative part, data is negative. The largest skill differences are 13.72 and 8.4 in two parts. Besides, two subsets are very similar.
5.2.2 Regression
According to the estimation results in Table 11, the model explain 53.1% of the variation of the FDI activitiy. Furthermore, F test shows that the model as a whole is significant.
14In 2000, the skill difference between UK and the Netherlands is -0.11, resulting in the maximum FDI
Table 10: Descriptive statistics for variables: negative part
N Minimum Maximum Mean Std. Deviation
F DIij 453 0.1 277356.0 9952.26 31438.00
LnF DIij 453 -2.38 12.53 5.99 3.42
SU M GDP 453 31.3 14547 2518.90 34
GDP DIF SQ 453 3 1.2E+08 1.2E+07 30771169.28
SKDIF F 453 -8.4 0 -1.2859 1.57 SKDIF SQ 453 0 70.61 4.11 10.13 GDP DIF F ∗ SKDIF F 453 -18467.44 63666.04 -122.92 6042.34 IN V C 453 53.8 91.0 78.251 8.37 T CIJ 453 3.50 9.65 8.13 1.26 T CIJ ∗ SKDIF SQ 453 0 508.37 32.46 80.39 T CJ I 453 6.3 9.5 8.56 0.85 DIST 453 55 19633 4884.42 4769.32
The skill difference variable affect the FDI positively as predicted. The result is consistent with the negative part of Figure 2b which is only related to horizontal FDI, indicating that horizontal FDI dominate the market when countries are similar in relative skill.
As the results in the positive part, hypotheses regarding three variables are not supported either because of the sign of the variable coefficient or the insignificance of the relationship. These three variables are GDP DIF F ∗ SKDIF F , IN V C and T CIJ .
Table 11: Regression result: negative part
Variable Coefficient t-Statistics Sig Sign as predicted/Significant?
Constant -15.621 -7.898 0.000
SU M GDP 1.073E-03 11.200 0.000 Yes/Yes
GDP DIF SQ -7.17E-08 -7.093 0.000 Yes/Yes
SKDIF F 0.498 2.421 0.016 Yes/Yes
SKDIF SQ 0.255 3.105 0.002 No/Yes
GDP DIF F ∗ SKDIF F -3.54E-05 -1.801 0.072 Yes/No
IN V C 0.112 4.631 0.000 Yes/No
T CIJ 1.255E-02 0.80 0.937 No/No
T CIJ ∗ SKDIF SQ 2.30E-02 -2.125 0.034 Yes/Yes
T CJ I 1.389 8.307 0.000 Yes/Yes
DIST -1.32E-04 -5.239 0.000 Yes/Yes
N=453 R2=0.531
F=50.090 (Sig=0.000)
The estimation results for trade costs from country i to j is a little bit unexpected. In both positive and negative parts, the influence is opposite to hypothesis 5, increasing trade costs increases the tariff-jumping incentive for (horizontal) FDI. However, we find some empirical supports of this hypothesis from the negative effects of distance on FDI. Since the relationship is not significant in both parts, maybe more data is needed to get a more convincing result.
6
CONCLUSIONS
The purpose of this paper is to estimate the Knowledge-Capital Model of the multinational firms established by CMM (2001) which incorporates both horizontal and vertical moti-vations for foreign direct investment. According to the theories, the volume and direction of FDI is a function of characteristics of both home country and host country, namely total market size, difference in market size and relative skilled-labor prices, trade costs in two directions, investment costs of host country and geographical distance between two countries.
Among the variables, skill difference is the most debatable one especially with respect to its specification, its measures and its influence on FDI. Therefore, I make some changes which correspond to these problems. First of all, based on the theories, this variable has a very non-monotonic influence (see Figure 2) on FDI but not the positive effect in CMM (2001) which is derived from simulation results. Besides, in order to test this non-monotonic re-lationship, I separate the data into two subsets, one with positive skill difference and the other negative skill difference. Furthermore, a more direct specification of relative skill, relative factor price, is used instead of relative endowment. Therefore, the measures is changed to the relative skilled-labor wage.
Table 12: Countries in the dataset
Home countries Host countries
Argentina, Australia, Austria, Argentina, Australia, Austria,
Brazil, Bulgaria, Canada, Belgium, Brazil, Bulgaria,
Chile, Colombia, Czech republic, Canada, Chile, China,
Denmark, Estonia, Finland, HongKong China, Colombia,
France, Germany, Ireland, Czech republic, Denmark,
Italy, Luxembourg, Japan, Egypt, Estonia, Finland,
Mexico, Netherlands, New Zealand, France, Germany, Greece,
Norway, Panama, Poland, Hungary, India, Indonesia,
Portugal, Slovakia, Sweden, Ireland, Israel, Italy,
Switzerland, UK, USA, Japan, Kenya, Latvia,
Venezuela Malaysia,Mexico, Netherlands,
Table 13: Definition of difference professions
Professions Definitions in Prices and Earnings
Product manager
Employed in the pharmaceuticals, chemicals or food industry, middle-management position, university or technical college graduate with at least 5 years’ experience in the field; about 35 years old, married, no children.
Department head
Operational head of a production department with a staff of over 100 in a sizable company in the metalworking industry; completed vocational training and many years’ experience in the field; about 40 years old, married, two children.
Engineer
Employed by an industrial firm in the electrical engineering sector, university or technical college graduate with at least 5 years’ work experience; about 35 years old, married, two children.
Skilled industrial worker
Skilled worker with vocational training and about 10 years’ experience in the metalworking industry; approximately 35 years old, married, two children.
8
REFERENCE
Blonigen, Bruce A.; Davies, Ronald B.; Head, Keith. “Estimating the knowledge-capital model of the multinational enterprise: comment”. American Economic Review, June 2003, 93(3), pp. 980-994.
Braconier, Henrik; Norback, Pehr-johan; Urban, Dieter. “Multinational Enter-prises and Wage Costs: Vertical FDI Revisited”. Journal of International Economics, December 2005, 67(2), pp. 446-470.
Braconier, Henrik; Norback, Pehr-johan; Urban, Dieter. “Reconciling the Evi-dence on the Knowledge-Capital Model”. Journal of International Economics, September 2005, 13(4), pp. 770-786.
Brainard, Lael S. “An Empirical Assessment of the Proximity-Concentration Tradeoff between Multinational Sales and Trade”. American Economic Review, September 1997, 87, pp. 520-544.
Carr, David L.; Markusen, James R.; Maskus, Keith E. “Estimating the Knowledge-Capital Model of the Multinational Enterprise. American Economic Review, June 2001, 91(3), pp. 693-708
Davies, Ronald B. “Hunting high and low for vertical FDI”. University of Oregon Eco-nomic Department Working Papers 2002-12.
Gwartney, James; Lawson, Robert; Gartzke, Erik. “Economic Freedom of the World, 2005 Annual Report”, 2005, The Fraser Institute.
Harnett, Donald L. “Introduction to Statistical methods”, 3rd edition. 1982, Addison-Wesley.
Helpman, Elhanan. “A Simple Theory of International Trade with Multinational Cor-porations”. Journal of Political Economy, June 1984, 94(3), pp. 451-471
Markusen, James R. “Multinationals, Multi-plant Economies, and the Gains from Trade”. Journal of International Economics, May 1984, 16(3/4), pp. 205-226
Union Bank of Switzerland. “Prices and earnings around the globe: 2000 edition.” 2000.
