Proximity-Concentration vs. Factor Proportions Explanation: The Case of

Swedish Multinationals in the EU∗

Thomas MATHÄ

The School of Economic and Social Studies University of East Anglia

Norwich NR4 7TJ, United Kingdom Tel.: 0044 1603 593405, Fax: 0044 1603 250434

E-Mail: T.Mathae@uea.ac.uk

and

The European Institute of Japanese Studies Stockholm School of Economics

Box 6501, 113 83 Stockholm, Sweden

Abstract:

Both proximity-concentration trade-off and factor proportions explanations have been

forwarded to explain the existence of multinational enterprises. This paper analyses to what

extent the data supports these different explanations. A first empirical attempt is made to

distinguish explicitly between horizontally and vertically integrated multinationals. The

results indicate that horizontal multinationals are promoted if trade costs are large and R&D

intensity is low, while for vertical multinationals the opposite is the case. Further, both

horizontal and vertical multinationals are explained by small plant-level economies of scale

relative to domestic firm size, similarities in relative factor endowments, and large host

country sizes.

Keywords: Trade Costs, Country Size, Factor Endowments, Horizontal & Vertical

Multinationals

JEL Classification: F12, F21, F23

∗ I would like to thank Stephen Davies, Karolina Ekholm, Bruce Lyons, Erik Mellander, and Roger Svensson

for their helpful suggestions and comments, and to Per Thulin for data assistance. The author wishes to thank the DAAD and SI for financial support of this project and the Industrial Institute for Economic and Social Research for their hospitality.

1

1. Introduction: On both a regional and a global scale the importance of multinational enterprises and

Foreign Direct Investment have grown dramatically in recent decades. The global FDI stock

has increased fourfold between 1982 and 1994, and has doubled as a percentage of world

GDP to 9% (UNCTAD, 1997). Furthermore, the global FDI stock has grown three times the

rate of international trade and four times the rate of world GDP between 1960 and 1993

(European Commission, 1996). In 1995, global sales by affiliates outweighed world exports,

thus being the most dominant mode of servicing foreign markets. This picture is especially

pronounced in developed countries, which account for the predominant share of both FDI

outflows and inflows (UNCTAD, 1997).

Against this background it is thus not surprising that ‘new’ trade theory which

introduced economies of scale and product differentiation into general equilibrium trade

models, is now deemed to be an inappropriate reflection of reality, as multinationals are

excluded from the analysis. Starting with the early approaches of Helpman (1984) and

Markusen (1984), a new literature has emerged that takes account of the increasing

importance of multinational production in servicing foreign markets. Neither approach,

however, considers trade costs. Most of the newer models, which overcome this early

limitation, concentrate on horizontally integrated multinationals (e.g. Brainard, 1993;

Horstmann & Markusen, 1992; Motta, 1992; Markusen & Venables, 1996a,b, 1997, 1998).

One general proposition of these horizontal models is that multinational production is more

likely the smaller plant-specific economies of scale relative to firm-specific economies of

scale and trade costs. Firms must choose to trade-off proximity to customers, which saves

trade costs, with concentration of production at a single location to reap the benefits of plant-

level economies of scale.

Indeed stylised facts point towards horizontal multinationals being empirically far

more important than vertically integrated multinationals (e.g. Markusen, 1995). In the

Swedish context, however, it has been argued that the relationship between Swedish parents

and their European affiliates is mostly of a vertical nature (Andersson et al., 1996).

This distinction is important both theoretically and empirically, as horizontal

multinationals tend to arise if countries are similar in relative and absolute factor endowments

and if trade costs are medium to high, while vertical multinationals mainly arise if countries

are very different with respect to their relative factor endowments and if trade costs between

countries are low (Markusen, Venables, Konan & Zhang, 1996). Recent empirical tests of

2

both proximity-concentration trade-off hypothesis (Brainard, 1997; Ekholm, 1998) and the

factor proportions explanation of multinational sales (Brainard, 1993b; Ekholm, 1995, 1997)

failed to distinguish between the type of integration between parent and affiliate.1

The main objective of this paper is, thus, to make a first explicit attempt to analyse

whether these different theoretical predictions of horizontal vs. vertical multinationals are

supported by the data for EU countries. Secondly, separating horizontally from vertically

integrated MNEs also allows us to test the proximity-concentration trade-off hypothesis. As

only very few empirical tests have been conducted to test these general propositions, this

paper will help to show whether previous empirical findings are robust. This is especially

important as the countries in Brainard (1997) were chosen to “maximise diversity in

geographical coverage, income, production structure, and data coverage” (Brainard, 1997, p.

525). In choosing a relatively homogenous group of countries, as is the case for EU member

states, it will be important to see whether her empirical results are still supported.

In the empirical analysis we use plant and firm level data from Swedish multinationals,

collected on a quadrennial basis at the Research Institute of Industrial Economics in

Stockholm. In doing so we will restrict ourselves to already existing multinationals.2 Despite

its small economic size, Sweden appears to be one of the most multinational countries in the

world when compared to its economic size (Andersson et al., 1996).

The remainder of this paper is organised as follows. Section 2 will explore the

theoretical background. This will lead to the formulation of testable hypotheses. Section 3

presents the data and sources, while section 4 derives the econometric specification. Section 5

discusses the empirical results and the final section concludes.

2. Theoretical aspects: a. OLI Framework

Multinationals need to possess some kind of advantage over national firms, otherwise

they would not find it profitable to enter the foreign market. This is due to additional costs

1 The only paper that makes an attempt to distinguish between horizontal and vertical multinationals is

Brainard (1993b). She distinguishes between multinational sales destined for local sales and multinational sales destined for exports to the home country in an attempt to shed light on the poor performance of the factor proportions explanation of multinationals sales.

2 Firms in the sample may service individual countries in the EU either through exporting or through multinational production, or both simultaneously. However, due to the nature of this database firms are multinational in at least one of the EU member states, but not necessarily in all of them. Thus, a sample selection problem emerges in so far as purely national firms are excluded.

3

that have to be incurred when doing business abroad, such as transport and communication

costs, not to mention different national laws and regulation, language and customs.

A nice point of departure is the OLI approach developed by Dunning (1977, 1981),

which provides a useful organising framework. According to this approach firms need to

possess three advantages to become a multinational enterprise, namely ownership,

localisation, and internalisation advantages.

Firstly, potential multinationals need to possess ownership advantages or intangible

assets, which are firm-specific, such as, managerial expertise, patents, blueprints and

trademarks. These assets are to a large extent internationally mobile and allow firms to spread

production across national boundaries, as they give rise to multi-plant economies of scale (e.g.

Helpman, 1984; Markusen, 1984). Furthermore, these intangible assets often have public good

characteristics, which makes it difficult to appropriate their full return when contracting in the

market place (e.g. Williamson, 1981). As a result, firms may choose to internalise these

transactions costs. It should be stressed, however, that the notion of firm-specific assets is

primarily used to explain the existence of horizontally integrated MNEs. Vertically integrated

MNEs internalise the costs for intermediate products, thereby foregoing ex ante and ex post

contracting costs (e.g. Caves, 1996).

Secondly, firms have to possess some kind of localisation advantages to prefer

multinational production to exporting. For horizontal multinationals these localisation

advantages arise mainly due to savings in transport costs, tariffs and quotas. Given that there

are plant-level economies of scale, trade costs are necessary, as otherwise the firm would

always choose to produce at a single location and export to the foreign market. Thus, the cost

savings from local production in the foreign market have to be traded off against the potential

cost savings from producing in one single location (e.g. Brainard, 1993; Markusen &

Venables, 1996a,b, 1997, 1998). Plant-level economies of scale also give rise to a second

localisation advantage, the host country market size (e.g. Rowthorn, 1992; Brainard, 1993a;

Markusen & Venables, 1996a,b, 1997, 1998). The larger the host country market size, the

more likely it is that the firm will be able to produce at minimum average cost, or differently,

the more likely it is that the Minimum Efficient Scale (MES) of production can be reached.

While high trade costs have the tendency to encourage the existence of horizontal

multinationals, they discourage the existence of vertical multinationals. Vertical

multinationals fragment the production process, such that the respective production stage is

located in that country which is abundant in the production factor, which is used relatively

4

intensively at that stage (e.g. Helpman, 1984; Helpman & Krugman, 1985; Markusen,

Venables, Konan & Zhang, 1996). Vertical multinationals may, for example, produce

intermediate products, which are exported to the foreign country for further processing. The

final product is then transported back to the home country (e.g. Helpman & Krugman, 1985,

Ch. 13, Zhang & Markusen, 1996). Thus, low trade costs should facilitate the presence of

vertical multinationals. Moreover, the larger the difference of countries is with respect to their

relative factor endowments the more likely it is that vertical multinational will arise

(Helpman, 1984; Helpman & Krugman, 1985; Markusen, Venables, Konan & Zhang, 1996).

Thirdly, internalisation advantages arise due to the public good characteristic of

intangible assets. The firm may choose to transfer these assets within the organisation to avoid

their dissipation, which would otherwise occur due to opportunistic behaviour (e.g.

Williamson, 1981). Such dissipation can occur due to the impossibility of writing complete

contracts (e.g. Ethier, 1986) or due to licensees exploiting acquired knowledge about the

intangible asset after the end of the licensing period (e.g. Rugman, 1985, 1986; Ethier &

Markusen, 1996). Alternatively, licensees are tempted to free ride on the reputation of the firm

to gain short run profits (e.g. Horstmann & Markusen, 1987).3

b. Horizontal vs. Vertical Multinationals

Horizontal multinationals have received more attention in the theoretical literature than

vertical multinationals. This may accord to stylised facts, which point towards horizontal

multinationals being empirically more relevant than vertical multinationals (Markusen, 1995).

In the Swedish case, such a general statement may however be misleading, as the relationship

between Swedish parents and their EU affiliates is to a large extent of a vertical nature. For

example, the share of intermediates in total parent exports to their manufacturing EU affiliates

rose from 53% in 1986 to 75% in 1990 (Andersson et al., 1996). Even when related to total

parent exports to the EU, the share of intermediates is of a considerable magnitude; this share

rose from 12% in 1986 to 18% in 1990.

Hence, it is important to distinguish between horizontally and vertically integrated

multinationals, especially as some of the forces leading to the emergence of horizontal

multinationals are in sharp contrast to the forces leading to the presence of vertical

multinationals. In trying to identify testable hypotheses for the empirical part, I will mainly

draw on the results by Markusen, Venables, Konan & Zhang (1996), which to the best of my

3 For a review see Markusen (1995).

5

knowledge, provide the only theoretical framework that synthesises the emergence of both

horizontal and vertical multinationals.

Their model assumes that there are two countries producing two homogenous

products, with two factors of production, skilled and unskilled labour. Both factors of

production are mobile between sectors but immobile between countries. Sector one is

characterised by skilled labour being used in the firm-specific fixed cost, while a combination

of both production factors is used for the fixed plant set up costs. Unskilled labour is only

used in variable costs and transport costs between countries. In the second sector perfect

competition is assumed to prevail. This sector is taken as the numéraire.

Further assumptions for sector one are firstly that skilled labour requirements for

generation of multi-plant and plant-level economies of scale are identical for national

exporting firms and vertical multinationals. The only difference is that vertical multinationals

draw part of the skilled labour requirement from the host country. Thus, the nature of the

vertical MNE is similar to the treatment in Helpman (1984), i.e. production and trade in

intermediates are not considered.

Secondly, skilled labour requirements in fixed costs for horizontal MNEs is larger than

the skilled labour requirements for vertical MNEs or national firms. This is due to the fact that

horizontal MNEs operate two manufacturing plants, as opposed to vertical multinationals or

national firms, which operate only one plant. However, the skilled labour requirements in

fixed costs by horizontal multinationals are less than twice the requirement of vertical

multinationals or national firms. This reflects the joint-input characteristic of firm-specific

assets.

Thirdly, unskilled labour requirements are assumed to be identical for all plants and

are entirely drawn from the country where the plant is located. Lastly, it is assumed that there

is some small cost to separate headquarters from production.4

The results, as summarised in table 1 below, suggest firstly, that horizontal

multinationals tend to exist if firm-level economies of scale and trade costs are large relative

to plant-level economies of scale. This result appears in numerous horizontal models

(Brainard, 1993a; Horstmann & Markusen (1992); Markusen & Venables (1996a,b, 1997,

1998) and has also been confirmed in recent empirical studies by Brainard (1997) and Ekholm

(1998). The idea is that firms have to trade off the benefits of proximity to customers with the

concentration of production at a single plant. Proximity to customers saves transport costs,

4 This is to avoid degeneracy of the model.

6

while concentration of production allows the firm to reap the benefits of plant-level

economies of scale. It is clear, however, that horizontal multinationals cannot arise if transport

cost are assumed to be negligible. This is due to the fact that by establishing a manufacturing

plant abroad, the horizontal multinational has to incur fixed plant set up costs without being

able to save transport costs by doing so.

Furthermore, horizontal multinationals are associated with similar relative and

absolute factor endowments. Again, these results are consistent with horizontal models of

Brainard (1993a) and Markusen & Venables (1996a,b, 1997, 1998). These predictions are also

supported by recent empirical evidence which comes to the conclusion that multinational

activity is mainly promoted by country size and factor endowment similarities (e.g. Brainard,

1997; Ekholm, 1998).

If countries become dissimilar with respect to relative factor endowments, then the

skilled labour abundant country will have a comparative advantage in producing goods in

sector one. In this case, single plant firms have an advantage over two-plant firms, as

horizontal multinationals locate production in the high cost factor market.

Table 1: Dominant of production regime and country characteristics

Production regime Hypothesis5 Characteristics

Horizontal multinationals

Hyp.1 Hyp.2 Hyp.3 Hyp.4 Hyp.5

• firm-level economies of scale are large relative to plant level economies of scale,

• countries are large, • trade costs are moderate to high, • countries are similar in relative factor endowments, • countries are similar in size.

Vertical multinationals

Hyp.6 Hyp.7

• trade costs are moderate to low, • countries differ significantly in relative factor

endowments. National exporting

firms

Hyp.8 Hyp.9

• trade costs are low and countries are similar in relative factor endowments and size,

• trade costs are moderate and countries are very different in size.

If countries become dissimilar with respect to country size, then the larger country will

have an advantage of producing goods in sector one due to a home market effect. Moreover,

the larger country will have a higher real price of skilled labour, which is due to the general

equilibrium constraint. This is assuming that production of the product in sector one is

relatively skilled labour intensive. In this case, multinational production has two cost

7

advantages over exporting; saving on transport costs and lower prices of skilled labour.

Allowing multinationals to enter raises the demand and price for skilled labour in the smaller

country. In this sense, multinationals reduce factor price differences. If country size

differences become too different, however, production in the smaller country can not be

sustained any more as output levels are too small to recoup fixed plant set up costs. This

explains the association of horizontal multinationals with similarity of absolute and relative

factor endowments.

If factor endowments and transport costs are such that two plant, i.e. horizontal

multinationals cannot be sustained, the question whether equilibrium is dominated by vertical

multinationals or national firms hinges on the fact of whether factor prices are equalised

internationally. If factor prices are virtually equalised then vertical multinationals cannot be

supported due to the cost disadvantage of splitting headquarters services and production. If

factor prices fail to equalise, vertical multinationals can enter as they fragment headquarter

services and production, such that headquarter services are located in the skilled labour

abundant country, while production takes place in the unskilled labour abundant country.

These results are consistent with the work by Helpman (1984), Helpman & Krugman (1985)

and Zhang & Markusen (1996).

There is only limited evidence for the factor proportions explanation of multinational

sales. Brainard’s (1993b) results reject the pure factor proportions explanation, although this

explanation seems to support multinational activity to some extent. In an attempt to separate

horizontally and vertically integrated multinationals, Brainard (1993b) finds that affiliate

production destined for exports back to the US differs from that destined for local sales. The

factor proportions hypothesis is generally supported in explaining the share exported back to

the US.6

Ekholm (1998) also finds some support for the factor proportions explanation in that

the share of the net outward foreign production to total foreign production is positively

increasing in capital per capita differences, and especially in human capital differences. Her

results are consistent with earlier obtained results, which indicated that that outward foreign

production is positively affected by the home country’s endowment in overall and human

capital (Ekholm, 1995, 1997).

5 Hypothesis 5 is not testable as the data only contains Swedish outward multinationality. Hypothesis 8 and 9

are not testable either, as the database does not contain purely national firms. 6 More precisely, the share of affiliate sales accounted for by exports back to the US increases in differences in

per capita endowments of capital, and illiterate labour, and decreases in differences in unskilled, literate labour, and transport costs.

8

Another important issue to be considered is the effect that both horizontal and vertical

multinationals have on trade volumes. Horizontal multinationals have the tendency to reduce

trade volumes between countries. This was suggested by models of Brainard (1993a),

Markusen & Venables (1996a,b, 1997, 1998).

For vertical multinationals, such a clear prediction is not possible. Firstly, investment

liberalisation followed by entrance of vertical multinationals may lead to the reversal of trade

flows. The reason is the geographical separation of the production stages, which results in the

home country importing the final product rather than exporting it, as would be the case with

national firms only.

Secondly, investment liberalisation may lead to increases in trade flows. There has

been a considerable debate in the literature as to whether exports and multinational production

are substitutes or complements. In the Swedish case, empirical evidence demonstrates some

complementarity between exports and multinational sales (Swedenborg, 1979, 1982;

Blomström et al., 1988). In contrast to these studies Svensson (1996) finds some evidence of

substitutability, which is especially pronounced for EU countries.

In theory, increases in trade flows are especially likely if one country is small and

skilled labour abundant, whilst the other country is large and unskilled labour abundant.

Indeed, it may be argued that this scenario reflects the case of Sweden. In such a case,

investment liberalisation has the effect of relaxing constraints in both the small and large

country. In the smaller country, production of the final product is constrained due to unskilled

labour scarcity and a small domestic market, while in the large country production of

headquarter services is constrained due to the high price of skilled labour. Thus, investment

liberalisation may lead to relocation of headquarter services to the small skilled labour

abundant country. Production would take place in the larger country, which exports the final

product to the small country, while the small country exports headquarter services to the large

country. As a result the volume of exports may increase.

One can detect surprisingly many stylised Swedish facts as presented by Andersson et

al. (1996) in this scenario. Firstly, Sweden is the home country of relatively many

multinationals, when compared to its country size. Secondly, the internationalisation process

of Swedish multinationals was characterised by expansion through foreign affiliates between

1970 and 1990. Thirdly, the relationship between parent and EU affiliate is to a large extent of

vertical rather than horizontal nature.

9

4. Data The data is taken from the database at the Research Institute of Industrial Economics in

Stockholm. This database contains information about individual Swedish multinationals and

their foreign operations.7 We consider the operations of Swedish multinationals in ten

individual EU member states. Ireland and Luxembourg had to be excluded. This is as

industrial variables were mainly taken from the STAN database (OECD, 1996), which does

not include either country. Data for Germany refers to West Germany prior to 1990 and to

united Germany from 1991 onwards. The variables are defined in million SEK and 1990

prices. The analysis covers the years 1974, 1978, 1986, 1990 and 1994. Earlier years had to be

excluded as the questionnaire prior to 1974 only asked about firms’ exports for countries

where foreign production was established. No survey was conducted in 1982.

The data set is pooled over these years. Interaction variables are used to test for

structural differences between horizontal and vertical multinationals.

5. Econometric specification Dependent variables:

The share of foreign production: AFFSHi,k,t

The dependent variable takes account of exports and multinational production being

simultaneously determined. It is defined as:

AFFSHPROD

PROD EXPi k ti k t

i k t i k t, ,

, ,

, , , ,

=+

The production volume of a multinational i in country k at time t, PRODi k t, , , is

defined as the sum of all affiliates’ total sales in country k at time t minus the sum of their

total imports from their parent company in Sweden at time t. Thus, we make an attempt to

disentangle affiliates’ production from pure resale activities of imported products. EXPi k t, , is

defined as firm i’s total exports to country k at time t. EXPi k t, , includes both exports of

finished products, which may be sold by the affiliate in country k, and intermediate products,

which are used by the affiliate for further processing.

The extent of vertical forward and backward integration: (VFBINTi,k,t)

7 For a complete documentation of the database see Andersson et al. (1996).

10

As alternative dependent variables, we use VFINT, VBINT, and VFBINT. These

variables are specifically designed to analyse the determinants of vertical integration between

parent and affiliate.VFBINT VFINT VBINTi k t i k t i k t, , , , , ,= + , with

VFINTIntermediate Exports to Affiliates

PROD EXPi k ti k t

i k t i k t, ,

, ,

, , , ,=

+

VBINTAffiliate Exports to Sweden

PROD EXPi k ti k t

i k t i k t, ,

, ,

, , , ,=

+

According to hypotheses 6 and 7 vertical multinationals arise if transport costs are low and

countries differ significantly in relative factor endowments. A typical feature of vertical

multinationals is that they engage in intra-firm trade of intermediate products. Furthermore,

the final product is then exported back to the home country, giving rise to trade flow reversals.

These two distinct features for vertical multinationals form part of the numerator

while total foreign sales are used in the denominator.8 This specification has the advantage

that we do not have to rely on an ad hoc distinction between horizontally and vertically

integrated multinationals. This dependent variable will allow us to treat vertical integration as

an entirely endogenous outcome of our determinants.

It is immediately apparent that both dependent variables may be censored. Censoring

of the dependent variable usually refers to a sample where some observations of the dependent

variable are not observed, while the independent variables are observed (e.g. Maddala, 1983;

Greene, 1993). This poses a problem in so far as estimation of the dependent variable by

ordinary least squares (OLS) will result in biased coefficient estimates. This problem can be

avoided by formulating a Tobit regression model, which accounts for censoring at both the

lower and upper tail.

AFFSH is censored at both tails, as firms export to EU countries, but are not

necessarily engaged in multinational production in these countries. Additionally, there are

cases, where the Swedish parent supplies the foreign market entirely via multinational

production, but not via exports from Sweden.9 The specification for AFFSH takes the

following form:

y ß xi i i* = ′ + µ , where yi

* is the real observed variable and

8 Note that both parts of the numerator are embodied in the denominator, as affiliate exports to Sweden is part

of affiliate production, while total exports to country k embodies the exports of intermediate products. 9 Unfortunately, we cannot account for any relationship between affiliates within the EU.

11

y yi i= * if 0 1< <yi*

yi = 0 if yi* ≤ 0

yi = 1 if yi* ≥ 1

As the dependent variables VFINT, VBINT, and VFBINT are only censored at the lower tail of

the distribution, the correction at the upper tail is not necessary.

Explanatory variables

Firm-level economies of scale: Research & Development intensity (R&D)i,t

The R&D to sales ratio is commonly used to explain the existence of multinationals. In

industries with high R&D expenditure, competition is more dependent on the exploitation of

firm-specific assets. Hence, firms with high R&D to sales ratio tend to supply the foreign

market by multinational production (Hypothesis 1). The R&D Intensity is defined as: The

ratio of total R&D expenditure to total sales by the parent company.

Plant level economies of scale and Swedish firm Size: LGj,t FIRMSIZEi,t

We use the above midpoint average plant size at the 3 and 4 digit level of the ISIC

classification, taken from the Swedish industry census of production, as our industry measure

of economies of scale. The years refer to 1975, 1978, 1987, 1990, and 1993. Thus, they

broadly correspond to the observations for the Swedish multinationals. In doing so we attempt

to ensure that the plants operate at or above the Minimum Efficient Scale of production.10

It is clear, however, that plant-level economies of scale on its own is not of much use

empirically. While in theory all respective horizontal and vertical multinationals are equal in

size, it is imminent that, in reality, this is not the case. Thus, plant-level economies of scale

are to be related to domestic firm size, FIRMSIZEi,t which is measured as a firm i’s total sales

in Sweden at time t.11 The nice intuition behind this measure is that it reflects whether the

firm has achieved the necessary economies of scale domestically. The larger the Swedish firm

size is relative to the plant-level economies of scale, the more likely it is that this firm will

turn multinational. The reason is simply that there is no great cost penalty of moving

production abroad, as the firm has exhausted the plant-level economies of scale domestically.

Hence, the coefficient of FIRMSIZE is expected to be positive, while we expect a negative

coefficient for LG (Hypothesis 1).

10 To reduce heteroskedasticity, variables are expressed in logarithmic form. This is indicated by the prefix L. It

was not possible to express all firm specific variables, such as R&D, in logs, as they contain zero values.

12

Trade Impediments: Industry specific trade costs (LTRADE)j,t

This measure of trade costs is also taken from Swedish industry statistics. The years

and the level of aggregation correspond to the years of our measure for economies of scale

measure. LTRADE reflects industry specific transport and packaging costs and is expressed as

a percentage share of total sales in industry j at time t. This measure has neither a distance nor

a country specific component. However, the measure varies over time. We expect this

measure to be positively related to the share of multinational production in foreign sales if

multinationals are horizontally integrated (Hypothesis 2). We expect a negative or

insignificant sign for vertically integrated multinationals (Hypothesis 6). This is as vertical

multinationals ship intermediates to their foreign production locations, which implies that

transport costs have to be low in the first place for foreign production to take place.

Country size: Gross Domestic Product (LGDP)k,t

GDP values for EU member states are taken from OECD National Accounts yearbook

(OECD, 1997). We expect that larger host country sizes favour multinational production

(Hypothesis 2).

Relative factor endowments: Relative Gross Domestic Product per Capita (LINCDIF)k,t

Similarly, the GDP per capita ratios for countries are taken from OECD National

accounts. It is defined as the absolute difference in the GDP per capita between country k and

Sweden at time t, where GDP per capita is expressed as a logarithmic transformation:

LINCDIFGDP POPULATION

GDP POPULATIONk tk t k t

SWE t SWE t,

, ,

, ,log

//

=

We expect this measure to be negatively related to the affiliate production share if

multinationals are predominantly horizontally integrated (Hypothesis 4) and positively related

to the affiliate production share if the relationship between parent and affiliate is mainly of a

vertical nature. (Hypothesis 7).

11 In the Industrial Organisation literature it is common to relate economies of scale to industry size. As we seek

to explain individual firms’ behaviour, it is more appropriate to relate economies of scale to firm size.

13

Technology differences: Relative Value Added per Employee (LVADIF)j,k,t

Technological differences between countries and industries do not appear in the

theoretical model. Nevertheless, it is clear that one of the reasons for vertical multinationals to

appear is to take advantage of higher productivity levels in the host country. Accordingly, we

will include a measure of relative productivity differences between member state k and

Sweden to account for such a motivation. This measure is calculated from the STAN

database. The measure is defined as the absolute difference in value added per employee in

industry j, country k and time t to the value added per employee in that industry in Sweden at

time t.

LVADIFVALUE ADDED EMPLOYEES

VALUE ADDED EMPLOYEESj k tj k t j k t

j SWE t j SWE t, ,

, , , ,

, , , ,log

//

=

Fixed Time Effects: DUM74, DUM78, DUM86, DUM90, DUM94

As the data set has been pooled over different time periods we include these time

dummies to control for the effect that some companies may appear more often in the sample

than others.

Vertical integration interaction variables: (Suffix -V)

We employ interaction variables to test for structural differences between horizontal

and vertical multinationals. The ultimate aim is to find a variable that discriminates between

horizontal and vertical integration, such that the information about the relationship between

the Swedish parent company and its affiliates in individual EU member states can be used.

The problem, though, is that such a variable is only observed if the parent has established an

affiliate in that country. Including such a variable as an explanatory variable in the regression

would not be appropriate. To avoid this problem we define a parent to be vertically integrated

with its EU affiliates if the share of exports in intermediates from parent to EU affiliates to

total EU affiliate production, labelled VFEU, is in excess of a certain value.12 VFEU is

defined as:

VFEUExport of Intermediates to Affiliates

PRODi t

i,k,tk

i k tk

,, ,

=∑

∑

12 Of course, such discrimination between horizontal and vertical multinationals comes at a cost in terms of lost

information.

14

We use alternative values to distinguish between horizontal and vertical

multinationals. This is done mainly to see whether the obtained results are robust. The most

extreme way is to regard Swedish multinationals as vertical multinationals as soon as

VFEU>0. In the other cases we choose to regard firms to be vertically integrated if either

VFEU exceeds the weighted sample median (VFEU=0.017)13

The variables that are expected to have different impacts on horizontal and vertical

multinationals are transport costs, the country size and the (dis-) similarity of factor

endowments. Large transport costs are expected to be positively significant for horizontal

multinationals, but negatively significant or insignificant in explaining the production share in

total foreign sales of vertical multinationals (Hypothesis 6 and 3). According to the theoretical

model, country size is only expected to have an effect on the foreign production share if firms

are horizontally integrated, but not if they are vertically integrated multinationals (Hypothesis

5). Finally, the existence of horizontal multinationals should be favoured by factor endowment

similarities, while vertical multinationals tend to emerge if countries are different with respect

to their relative factor endowments (Hypothesis 4 and 7).

6. Econometric results: Firstly, I will discuss the results for the specification (I). Thereafter, I report the results

for specification (II) and (III), which include interaction variables to test for structural

differences between horizontal and vertical multinationals. At last, I will report the results for

specifications (IV), (V), and (VI), which is concerned with the extent of vertical integration

between parents and their affiliates in the EU. All results are shown in Appendix 1 in the

respective tables.

Specification (I)

The results in specification (Ia) and (Ib) show firstly, that the share of foreign

production to total foreign sales tends to be high if plant-level economies of scale, trade costs

and host country size are large and if countries are similar in relative factor endowments. The

coefficients of the variables are significant at the 5%, 5%, 1% and 1% level, respectively.

Thus, these results show that the share of affiliate production of Swedish multinationals is

13 The difference between the simple and weighted sample median occurs, as the firms do not have foreign sales

in equally many member states. The simple sample median (VFEU=0.021) is biased towards firms with sales in many EU member states, while the weighted sample median corrects for this.

15

primarily explained by market access and proximity considerations (Hypothesis 2,3,4), and

not by the factor proportions explanation.

The coefficient of our proxy for plant-level economies of scale is positive significant at

the 5% level, which is contrary to the prediction. The sign of this coefficient is however

reversed, and significantly so, if the firm size in the domestic market, LFIRMSIZE, is included

as an additional variable (specification (Ic) and (Id)). This suggests firstly that plant-level

economies of scale variable LG picks up some other effect if the firm size is excluded.

Secondly, these results show clearly that what matters is that firms have achieved and

exhausted the economies of scale at the plant level in their home market in order to become

multinationals.

As a final remark, the coefficients of R&D intensity and sectoral productivity

differences between countries are negative, albeit both being insignificant. This suggest that

on aggregate neither firm-level economies of scale nor productivity differences seem to

significantly affect foreign production shares. However, as we will see in a short while, there

are significant differences between multinationals.

Specifications (II) and (III)

In specifications (II) and (III) we test whether there are structural differences between

horizontally and vertically integrated multinationals. In specifications (II) and (III),

multinationals are regarded as vertically integrated, as soon as VFEU>0 and VFEU>0.017,

respectively. The results suggest that there are some significant differences between

multinationals.

Firstly, we note that large host country sizes, relative factor endowment similarities,

large industry specific trade costs and a large firm size relative to plant-level economies of

scale, promote horizontal multinationals. All these coefficients are either significant at the 5%

or 1% level. Thus, these results nicely confirm the theoretical predictions for the emergence of

horizontal multinationals (Hypothesis 1,2,3,4). Again, the coefficient of the similarity in

productivity differences between industries in the host and home country does not seem to

contribute significantly to the foreign production share of horizontal multinationals.

One very interesting result is that a high R&D intensity has a negative effect on the

foreign production share of horizontal multinationals. The coefficient is negatively significant

at the 5% level or better in specifications throughout the specifications (II) and (III). This is

contrary to expectation. However, as Norbäck (1998) has demonstrated in the case for

16

horizontal multinationals, the expectation of a positive coefficient on R&D hinges on the

assumption that the technology transfer costs between parent and its affiliates are unrelated to

the magnitude of the R&D intensity. Relaxing this assumption and assuming instead that

technology transfer costs and the R&D intensity are positively related to each other, is

consistent with the expectation of a negative coefficient of R&D intensity on foreign

production shares.

Furthermore, the coefficient of R&DV is positively significant for vertical

multinationals throughout specifications (II) and (III). Thus, there is a significant difference

between predominantly horizontally integrated multinationals and multinationals which

exhibit a substantial degree of vertical integration between parent and affiliate. The coefficient

has to be interpreted relative to the coefficient of horizontal multinationals. Hence, a high

R&D intensity has a relatively larger effect on foreign production share for vertical

multinationals than for horizontal multinationals. The net coefficient is positive, however,

indicating that even in absolute terms high R&D intensity has a positive impact on the foreign

production share of vertical multinationals.14

The positive coefficient of R&D intensity for vertically integrated multinationals may

also indicate that their technology transfer costs are independent of the R&D intensity. This

would be the case, for example, if vertical multinationals embody the technology in the

exports of intermediates to their affiliates. This also enables them to reduce opportunistic

behaviour and may help to fully appropriate the returns of the undertaken research and the

new technology. Hence, for vertical multinationals, the technology transfer costs are

embedded in the trade costs that have to be incurred, while horizontal multinationals’

technology transfer costs may rise with the R&D intensity, as more technical personnel,

scientists and other managerial staff have to be sent to foreign affiliates to supervise

implementation of the firms-specific assets.

Moreover, plant-level economies of scale have a significantly smaller effect on vertical

multinationals than on horizontal multinationals. This result is, however, not robust across

specifications. Firstly, in specification (II) LGV is negatively significant at the 10%, 5% and

5% level in specifications (IIb), (IIc) and (IId), respectively. In specification (IIIa) and (IIIb)

LGV is negatively significant at the 1% level, while it is insignificant in specification (IIIc)

and (IIId). Hence, this seems to suggest that, for a given domestic firm size, larger plant-level

14 The R&D coefficient is positively significant at the 10% or better level in a separate regression, which

includes vertically integrated multinationals only. The results are available from the author upon request.

17

economies of scale have a relatively larger effect of reducing foreign production shares if

multinationals are vertically rather than horizontally integrated.

On the one hand, we would expect that plant-level economies of scale are more

important for horizontal multinationals, which is in accordance with the prediction by

Markusen, Venables, Konan & Zhang (1996). On the other hand, vertically integrated

multinationals may be more dependent on the exploitation of resource-based economies of

scale, which makes it difficult and extremely costly to relocate production geographically.

With regard to trade costs, the results indicate that their effect is significantly smaller

in the case of vertically integrated multinationals. This difference is especially pronounced in

specification (III), where the coefficient of LTRADEV is negatively significant at the 5% level

or better, while the coefficient is only significant at the 10% level for specifications (IIa) and

(IIb). This nicely confirms the expectation that horizontal multinationals emerge to forego

trade costs while vertical multinationals emerge despite incurring trade costs (Hypothesis 3,6).

This is because vertical multinationals engage in intra-firm trade in intermediates. Or

alternatively, their foreign affiliates export the final product to Sweden. In either case, it is

obvious that trade costs are incurred, which tend to erode the locational advantage of the host

country, such as lower factor prices or higher productivity levels. Thus, vertical multinationals

tend to emerge in industries which exhibit moderate to low trade costs.

There is no indication, however, that vertical multinationals arise as a result of factor

endowment differences. The coefficient of the interaction term LINCDIFV is positive except

in specification (IIIb), but it is never significant. Similarly, productivity differences between

respective industries in the host and the home country do not seem to significantly affect

foreign production shares.

Additionally, the host country size seems to play a more important role for vertical

multinationals than for horizontal multinationals. The coefficient of LGDPV is positively

significant at the 5% level or better, except for specification (IIa), where it is only significant

at the 10% level. According to the theoretical model, we would expect host country size to

have a smaller effect on production shares for vertical multinationals.

Specification (IV), (V), and (VI)

18

In this section we report the results for the determinants of vertical forward and/or

backward integration between parent and affiliate, rather than foreign production shares.

Using these alternative dependent variables gives additional insights.

Firstly, the predictions are more clear cut with regard to the extent of vertical

integration between parents and their affiliate than in the case of foreign production shares.

The reason for this is that foreign production shares are expected to be large if the equilibrium

is either dominated by horizontal multinationals or by vertical multinationals. However, the

conditions for the equilibrium regimes to emerge are very different, not to say opposed to each

other. Horizontal multinationals are expected to dominate if trade costs are moderate to high

and countries are similar in size and relative factor endowments, while vertical multinationals

are expected to dominate if trade costs are moderate to low and countries differ significantly

in relative factor endowments. The extent of vertical integration between parent and affiliate,

VFINT, VBINT, and VFBINT, is only expected to be of a high magnitude if trade costs are

moderate to low and if countries differ significantly with respect to the relative factor

endowments. In all other cases, either national exporting or horizontal multinationals

dominate in the equilibrium, which yields low intra-firm trade. This is a much sharper

prediction.

Secondly, we recall that one of the advantages in looking at the production shares was

that differences between horizontal and vertical multinationals could be examined. In doing

so, we were forced, however, to aggregate vertical integration between parents and their

affiliate to the EU level. Hence, vertical integration was not treated as the outcome of

individual country and industry specific characteristics, as desired, but rather as a firm-specific

characteristic. Hence, using the extent of vertical integration as a dependent variable will

allow us to review whether the previous analysis has severe shortcomings. Equally

importantly, vertical integration between parent and affiliate is regarded as an entirely

endogenous outcome of country and industry specific characteristics.

The results show that a high R&D intensity, has a positive effect on the extent of

vertical integration of Swedish multinationals. The coefficient is significantly positive at the

5% level and better for all regressions. This suggests that vertical multinationals embody their

technology and firms-specific assets in the intermediates that are exported to foreign affiliates.

In this way they may be able to reduce the dissipation of new technology. With regard to

backward integrated multinationals, the positive coefficient of R&D corresponds to the classic

Helpman (1984) case, where multinationals fragment production geographically in order to

19

take advantage of different factor endowments and factor prices. The primary task of the home

country is to provide headquarter services, such as R&D, to foreign production units.

The achievement of economies of scale is also an important determinant of vertical

integration. Both the coefficients of LFIRMSIZE and LG are significant at the 1% level,

except in specification (VIa) and (VIb), where the coefficient of LG is significant at the 10%

level only. Again, we notice the reversal of the sign of LG if the domestic firm size is

included.

Moreover, the results indicate very clearly that high trade costs deter vertical

integration of multinationals. The coefficient is negatively significant at the 10% level or

better for specifications (IVc, (Vc), (VIc), (IVd), (Vd), and (VId). This is especially

pronounced for vertical backward integrated multinationals. Thus, this confirms that vertical

multinationals tend to emerge in industries with low transport costs. As the results in

specification (II) and (III) have already indicated very clearly, this is significantly different to

horizontally integrated multinationals.

Large host country sizes have a highly significant effect on vertical integration. The

coefficient of LGDP is positively significant at the 1% level. This result is consistent with the

earlier obtained result in specifications (II) and (III), which suggested that large host country

size may favour vertical multinationals relative to horizontal multinationals.

Additionally, there is a strong suggestion that similarities in relative factor

endowments have a positively significant effect on vertical integration. The coefficient of

LINCDIF is negatively significant at the 1% level throughout specifications (IV), (V), and

(VI). Again, this result is consistent with earlier obtained results, but is contrary to the

theoretical prediction.

Moreover, vertical forward integrated multinationals are attracted by differences in

productivity levels. The coefficient of LVADIF is positively significant at the 5% level or

better in specification (IV). This confirms that geographical fragmentation of production

occurs to take advantage of industry-specific productivity, technology or labour cost

differences between the host and home country.15 To our surprise, the opposite seems to be

the case for backward integrated multinationals, where the coefficient of LVADIF is

negatively significant at the 10% level in specification (Vd). Reasons that may account for this

is that more and more Foreign Direct Investment is driven by acquisitions of independent

15 We chose to concentrate on absolute differences in value added per employee. As value added per employee

is highly correlated with the wage per employee, a positive coefficient of LVADIF may either result from

20

foreign firms, which in turn export to Sweden. Hence, a negative coefficient may be consistent

with acquisitions, which are mainly directed to competitors with similar products, and which

give the Swedish parent access to new markets and to new distribution networks.

Finally, as the results for specification (IV), (V), and (VI) are consistent with the

results obtained in specifications (II) and (III), we have reason to believe that using the

variable VFEU to discriminate between horizontal and vertical multinationals, did not

significantly affect the results.

Concluding Remarks

Recent developments in the new trade and investment literature suggest that there are

fundamental differences between horizontal and vertical multinationals. Horizontal

multinationals tend to emerge due to market proximity considerations, which have to be

traded off against concentration of production designed to reap the benefits of economies of

scale. Horizontal multinationals are favoured by similarities in country size, and relative factor

endowments. On the contrary, vertical multinationals fragment production into geographically

separate stages to exploit factor price, and thus relative factor endowment differences or

technologies. Due to resulting intra-firm trade in intermediates, vertical multinationals tend to

emerge if transport costs are low and relative factor endowment differences are large.

Previous empirical results suggest foreign production is mainly promoted by the

proximity-concentration trade-off considerations and to a minor extent by the factor

proportions explanation (Brainard, 1993b, 1997; Ekholm, 1995, 1997, 1998). These studies

fail, however, to distinguish between horizontal and vertical multinationals.

This paper makes a first explicit attempt to distinguish between horizontal and vertical

multinationals. In doing so we choose to focus on the EU affiliates of Swedish multinationals.

Importantly, and in contrast to stylised facts presented by Markusen (1995), Andersson et al.

(1996) have argued and shown that the relationship between Swedish parent and EU affiliates

is mainly of a vertical nature, which makes a distinction between horizontal and vertical

multinationals and their respective determinants relevant.

The results show firstly that there are significant differences between horizontally and

vertically integrated multinationals. There is some evidence that suggests that R&D intensity

has a positive effect on the foreign production share in foreign sales for vertical

multinationals, while there are strong signs that the opposite is true for horizontal

locating production abroad to take advantage of higher productivity levels or lower labour costs in the host

21

multinationals. Further, R&D intensity has a clear positive effect on the extent of vertical

integration of Swedish multinationals. The negative coefficient for horizontal multinationals

is consistent with the explanation forwarded by Norbäck (1998), who demonstrated that the

sign of the coefficient depends on the assumption of whether the transfer costs between parent

and affiliate are related or unrelated to the size of R&D undertaken. The results for vertical

multinationals suggest that technology and firm-specific assets are embodied in the

intermediates that are exported to the affiliates. In doing so, new firm-specific knowledge and

new technologies are kept in the home country, thereby reducing the possibility of their

dissipation to foreign competitors.

Moreover, a high share of multinational production is more likely if the firm has

achieved and exhausted the economies of scale on the plant-level in the home country. This is

consistent with the theoretical prediction, and with earlier results by Brainard (1997) and

Ekholm (1998). The exploitation of economies of scale is more important for vertical

multinationals, however. This may suggest that these multinationals depend on the availability

of cheap domestic resources, such hydro-electric energy, timber or ore. Hence, relocation of

production imposes relatively greater cost penalty for these firms.

Moreover, high trade costs tend to promote horizontal multinationals, while they have

a significantly smaller effect on vertical multinationals. The extent of vertical integration

between parents and their affiliates is significantly affected by the trade costs are that have to

be incurred, when shipping intermediates to foreign production units and/or shipping final

products back to Sweden.

With regard to country size and relative factor endowments, we note that both

horizontal and vertical multinationals tend to arise as if countries are large in size and similar

in their relative factor endowments. Host country seems also to be a relatively more important

determinant for vertical multinationals. In addition, the extent of vertical integration is also

entirely explained by factor endowment similarities. This suggest that the factor proportions

explanation of multinational sales is irrelevant for explaining Swedish multinationals in the

European Union. This reinforces the results obtained by Brainard (1993b, 1997) and Ekholm

(1998). Additionally, while Brainard (1993b) found some evidence that the share of foreign

production, which is exported back to the home country, is explained by the factor proportions

explanations, we could not find any such effect for Swedish multinationals in the EU.

country.

22

Lastly, the extent of vertical forward integration between parent and their affiliates is

increasing in sector-specific productivity differences between the host country and the home

country, while vertical backward integration is increasing by similarities therein.

8. Bibliography

Andersson, Thomas, Fredriksson, Torbjörn and Svensson, Roger (1996): Multinational Restructuring, Internationalization and Small Economies: The Swedish Case, (London: Routledge)

Blomström, Magnus, Lipsey, Robert E. and Kulchycky, Ksenia (1988): U.S. and Swedish Direct Investments and Exports, in Baldwin, R. (Ed.): Trade Policy Issues and Empirical Analysis, (Chicago: Chicago University Press)

Brainard, Lael. S. (1993a): A Simple Theory of Multinational Corporations and Trade with Trade-Off Between Proximity and Concentration, in NBER Working Papers Series, No. 4269

Brainard, Lael. S. (1993b): An Empirical Assessment of The Factor Proportions Explanation of Multinational Sales, in NBER Working Papers Series, No. 4583

Brainard, Lael. S. (1997): An Empirical Assessment of the Proximity-Concentration Trade-off Between Multinational Sales and Trade, in American Economic Review, Vol. 87, No. 4, pp. 520-544

Caves, Richard (1996): Multinational Enterprise and Economic Analysis; 2nd edition, (Cambridge: Cambridge University Press)

Dunning, John H. (1977): Trade, Location of Economic Activity and MNE: A Search for an Eclectic Approach, in Ohlin, B., Hesselborn, P.-O., and Wijkman, P.-M. (Eds.): The International Allocation of Economic Activity: Proceedings of a Nobel Symposium at Stockholm, (London: Macmillan)

Dunning, John H. (1981): Explaining the International Direct Investment Position of Countries: Towards a Dynamic and Developmental Approach, in Weltwirtschaftliches Archiv, Vol. 117, pp. 30-64

Ekholm, Karolina (1995): Multinational Production and Trade in Technological Knowledge, PhD Thesis, Lund Economic Studies No. 58

Ekholm, Karolina (1997): Factor Endowments and the Pattern of Affiliate Production by Multinational Enterprises, CREDIT Research Paper, No. 97/19, University of Nottingham

Ekholm, Karolina (1998): Proximity Advantages, Scale Economies, and the Location of Production, pp. 59-76, in Braunerhjelm, Pontus and Ekholm Karolina (Eds.): The Geography of Multinational Firms, (Boston: Kluwer Academic)

Ethier, Wilfred J. (1986): The Multinational Firm, in Quarterly Journal of Economics, Vol. 101, pp. 805-833

Ethier, Wilfred J. and Markusen, James R. (1996): Multinational Firms, Technology Diffusion and Trade, in International Journal of Economics, Vol. 41, pp. 1-28

European Commission (1996): European Economy, Economic Evaluations of the Internal Market, No. 4, (European Commission)

Greene, William (1993): Econometric Analysis, 2nd edition, (New York: Macmillan) Helpman, Elhanan (1984): A Simple Theory of International Trade With Multinational

Corporations, in Journal of Political Economy, Vol. 92, No. 3, pp. 451-471

23

Helpman, Elhanan and Krugman, Paul (1985): Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy, (Cambridge Mass.: MIT Press)

Horstmann, Ignatius J. and Markusen, James R. (1987): Licensing Versus Direct Investment: A Model of Internalization by the Multinational Enterprise, Canadian journal of Economics, Vol. 20, No. 3, pp. 464-481

Horstmann, Ignatius J. and Markusen, James R. (1992): Endogenous Market Structure in International Trade (Natura Facit Saltum), in Journal of International Economics, Vol. 32, pp. 109-129

Maddala, G. S. (1983): Limited-Dependent and Qualitative Variables in Econometrics, (Cambridge: Cambridge University Press)

Markusen, James R. (1984): Multinationals, Multi-plant Economies, and the Gains From Trade, in Journal of International Economics, Vol. 16, pp. 205-226

Markusen, James R. and Venables Anthony J. (1996a): The Increased Importance of Direct Investment in North Atlantic Economic Relationships: A Convergence Hypothesis, pp. 169-189, in Canzoneri, Mattew B., Ethier, Wilfred J. and Grilli, Vittorio (Eds.) The New Transatlantic Economy, (Cambridge: Cambridge University Press)

Markusen, James R. and Venables Anthony J. (1996b): The Theory of Endowment, Intra-industry, and Multinational Trade, NBER Working Papers Series, No. 5529

Markusen, James R. and Venables Anthony J. (1997): The Role of Multinational Firms in the Wage-Gap Debate, in Review of International Economics, vol. 5, pp. 435-451

Markusen, James R. and Venables Anthony J. (1998): Multinational Firms and New Trade Theory, in Journal International Economics, Vol. 46, pp. 183-203

Markusen, James R., Venables, Anthony J., Konan; Denise Eby and Zhang, Kevin H. (1996): A Unified Treatment of Horizontal Direct Investment, Vertical Direct Investment, and the Pattern of Trade in Goods and Services, Working Paper, No. 465, The Research Institute of Industrial Economics, Stockholm

Motta, Massimo (1992): Multinational firms and the Tariff-jumping Argument: A Game Theoretic Analysis with some Unconventional Conclusions, in European Economic Review, Vol. 36, pp. 1557-1571

Norbäck, Pehr-Johan (1998): Multinational Firms, Technology and Location, Dissertations In Economics, 1998:2, Department of Economics, Stockholm University, (Edsbruk: Akademitryck)

Nordström, Kjell A. (1991): The Internationalization Process of the Firm - Searching for New Patterns and Explanations, (Stockholm: Stockholm School of Economics)

OECD (1996): International Direct Investment Statistics Yearbook 1996, (OECD: Paris) OECD (1997): National Accounts, Main Aggregates, Volume 1, 1960-1995, (OECD: Paris) Rowthorn, R. E. (1992): Intra-Industry Trade and Investment under Oligopoly: The Role of

Market Size, in Economic Journal, Vol. 102, pp. 402-414 Rugman, Alan M. (1985): Internalization is Still a General Theory of Foreign Direct

Investment, in Weltwirtschaftliches Archiv, Vol. 121, No.3, pp. 570-575 Rugman, Alan M. (1986): New Theories of the Multinational Enterprise: An Assessment of

the Internalization Theory, in Bulletin of Economic Research, Vol. 38, pp. 101-118 Svensson, Roger (1996): Effects of Overseas Production on Home Country Exports:

Evidence Based on Swedish Multinationals, in Weltwirtschaftliches Archiv, Vol. 132, No. 2, pp. 304-329

Swedenborg, Birgitta, (1979): The Multinational Operations of Swedish Firms: An Analysis of Determinants and Effects, (Stockholm: Almquist & Wicksell)

24

Swedenborg, Birgitta, (1982): Svensk Industri i Utlandet: En Analys av Drivkrafter och Effekter, (Stockholm, The Industrial Institute of Economic and Social Research)

UNCTAD (1993): From the Common Market to EC 92: Regional Economic Integration in the European Community and Transnational Corporations, (New York: United Nations)

UNCTAD (1995): World Investment Report: Transnational Corporations and Competitiveness, (United Nations: New York)

UNCTAD (1997): World Investment Report: Transnational Corporations, Market Structure and Competition Policy, (United Nations: New York)

Williamson, Oliver E. (1981): The Modern Corporation: Origins, Evolution, Attributes, in Journal of Economic Perspectives, Vol. 19, pp. 1537-1568

Zhang, Kevin H. and Markusen, James R. (1996): Vertical Multinationals and Host Country Characteristics, NBER Working Paper Series, No. 6203

25 Appendix 1:

Tobit Estimates for Foreign Production Share without industry dummies

Dep. Var. AFFSH No. of obs. 1698 Threshold value Lower=0, Upper=1 Specification Ia Ib Ic Id Log likelihood -1347.5 -1347.5 -1273.4 -1273.0

Constant -4 608 *** -4 597 *** -3 902 *** -3 862 *** (-6,834) (-6.817) (-6.686) (-6.617)

R&D -0,425 -0.439 -0.429 -0.455 (-0,555) (-0.572) (-0.571) (-0.609)

LFIRMSIZE 0.269 *** 0.269 *** (10.585) (10.608)

LG 0,077 ** 0.077 ** -0.276 *** -0.278 *** (2,204) (2.202) (-6.566) (-6.622)

LTRADE 0,103 ** 0.103 ** 0.108 ** 0.107 ** (2,359) (2.326) (2.513) (2.489)

LGDP 0,270 *** 0.269 *** 0.228 *** 0.226 *** (6,504) (6.491) (6.439) (6.403)

LINCDIF -0,408 *** -0.389 *** -0.400 *** -0.347 *** (-3,512) (-3.014) (-3.923) (-3.103)

LVADIF -0.036 -0.104 (-0.321) (-0.987)

PERIOD74 0,191 * 0.194 * 0.319 *** 0.329 *** (1,869) (1.882) (3.146) (3.203)

PERIOD78 0,014 ** 0.014 0.069 0.068 (0,183) (0.179) (0.955) (0.953)

PERIOD90 -0,148 -0.147 ** -0.087 -0.082 (-2,107) (-2.079) (-1.360) (-1.292)

PERIOD94 -0,069 *** -0.063 -0.113 -0.096 (-0,867) (-0.758) (-1.548) (-1.286)

T-statistic in brackets. Estimates are heteroskedastic consistent. *, **, *** denotes significance at the 10%, 5%, and 1% level, respectively.

26

Tobit Estimates on Differences Between Horizontal and Vertical Multinationals: CUT-OFF POINT VFEU>0 CUT-OFF POINT VFEU>0 017

Dep. Var. AFFSH AFFSH No. of obs. 1698 1698 Threshold value Lower=0, Upper=1 Lower=0, Upper=1 Specification IIa IIb IIc IId IIIa IIIb IIIc IIId Log likelihood -1338.8 -1338.4 -1266.6 -1266.1 -1328.6 -1328.5 -1254.1 -1253.6

Constant -4 471 *** -4 408 *** -3 876 *** -3 828 *** -4 431 *** -4 433 *** -3 920 *** -3 892 *** (-6.723) (-6.612) (-6.789) (-6.657) (-6.872) (-6.845) (-6.946) (-6.859)

R&D -4.091 *** -4.045 *** -4.053 ** -4.036 ** -5.746 *** -5.743 *** -3.451 ** -3.441 ** (-2.821) (-2.807) (-2.182) (-2.164) (-4.195) (-4.186) (-2.151) (-2.123)

LFIRMSIZE 0.317 *** 0.319 *** 0.277 *** 0.277 *** (6.843) (6.856) (8.146) (8.114)

LG 0.135 ** 0.139 ** -0.152 ** -0.153 ** 0.188 *** 0.187 *** -0.179 *** -0.182 *** (2.388) (2.452) (-2.314) (-2.330) (4.010) (3.981) (-3.136) (-3.196)

LTRADE 0.219 *** 0.226 *** 0.214 ** 0.212 ** 0.253 *** 0.252 *** 0.185 *** 0.180 *** (3.023) (3.091) (2.425) (2.400) (3.996) (3.970) (2.837) (2.754)

LGDP 0.232 *** 0.225 *** 0.160 *** 0.157 *** 0.220 *** 0.220 *** 0.195 *** 0.196 *** (5.206) (5.043) (4.141) (4.012) (5.542) (5.520) (5.664) (5.626)

LINCDIF -0.531 *** -0.648 *** -0.606 *** -0.590 ** -0.404 *** -0.380 ** -0.428 *** -0.350 ** (-2.866) (-2.601) (-3.196) (-2.346) (-2.846) (-2.191) (-3.323) (-2.209)

LVADIF 0.156 -0.033 -0.037 -0.135 (0.803) (-0.171) (-0.238) (-0.927)

R&DV 4.637 *** 4.545 *** 4.254 ** 4.188 ** 7.841 *** 7.841 *** 4.686 ** 4.648 ** (2.677) (2.631) (2.091) (2.054) (4.522) (4.482) (2.541) (2.505)

LFIRMSIZEV -0.052 -0.054 -0.045 -0.044 (-1.130) (-1.169) (-1.171) (-1.152)

LGV -0.101 -0.105 * -0.162 ** -0.160 ** -0.185 *** -0.184 *** -0.107 -0.104 (-1.579) (-1.640) (-2.109) (-2.091) (-3.088) (-3.067) (-1.519) (-1.475)

LTRADEV -0.141 * -0.150 * -0.145 -0.143 -0.293 *** -0.291 *** -0.198 ** -0.193 ** (-1.629) (-1.719) (-1.458) (-1.441) (-3.445) (-3.420) (-2.305) (-2.244)

LGDPV 0.048 * 0.051 ** 0.081 *** 0.082 *** 0.067 *** 0.066 *** 0.054 ** 0.052 ** (1.878) (1.982) (3.071) (3.099) (2.735) (2.702) (2.356) (2.267)

LINCDIFV 0.175 0.336 0.285 0.321 0.016 -0.010 0.074 0.023 (0.890) (1.247) (1.490) (1.219) (0.094) (-0.047) (0.490) (0.121)

LVADIFV -0.237 -0.076 0.040 0.072 (-1.012) (-0.336) (0.189) (0.366)

PERIOD74 0.186 * 0.189 * 0.333 *** 0.342 *** 0.223 ** 0.224 ** 0.312 *** 0.320 *** (1.817) (1.833) (3.345) (3.394) (2.230) (2.226) (3.225) (3.278)

PERIOD78 0.018 0.017 0.092 0.091 0.058 0.059 0.090 0.090 (0.230) (0.216) (1.289) (1.279) (0.746) (0.747) (1.248) (1.244)

PERIOD90 -0.126 * -0.123 * -0.073 -0.068 -0.206 *** -0.206 *** -0.139 ** -0.135 ** (-1.800) (-1.744) (-1.153) (-1.080) (-2.880) (-2.861) (-2.140) (-2.076)

PERIOD94 -0.057 -0.056 -0.098 -0.082 -0.084 -0.081 -0.084 -0.068 (-0.705) (-0.669) (-1.334) (-1.098) (-1.051) (-0.977) (-1.137) (-0.907)

T-statistic in brackets. Estimates are heteroskedastic consistent. *,**, *** denotes significance at the 10%, 5%, and 1% level, respectively.

27

Tobit Estimates: The Extent of Vertical Integration Between Parent and Affiliate Vertical Forward Integration Vertical Backward Integration Vertical Forward and Backward Integration

Dep. Var. VFINT VBINT VFBINT No. of obs. 1698 1693 1693 Threshold value Lower=0, Upper=+infinity Lower=0, Upper=+infinity Lower=0, Upper=+infinity Specification IVa IVb IVc IVd Va Vb Vc Vd VIa VIb VIc VIdLog likelihood -370.6 -367.7 -310.6 -308.8 -259.8 -258.5 -136.7 -135.0 -445.4 -444.3 -349.8 -349.7

Constant -1 273 *** -1 304 *** -0 922 *** -0 933 *** -1 020 *** -1 023 *** -0 540 *** -0 539 *** -1 381 *** -1 375 *** -0 805 *** -0 805 *** (-6.361) (-6.490) (-6.038) (-6.082) (-7.408) (-7.428) (-5.491) (-5.495) (-7.211) (-7.087) (-5.626) (-5.591)

R&D 0.673 *** 0.714 *** 0.822 *** 0.860 *** 0.369 *** 0.352 ** 0.283 ** 0.265 ** 0.683 *** 0.703 *** 0.874 *** 0.881 *** (3.478) (3.660) (4.704) (4.875) (2.564) (2.432) (2.189) (1.994) (3.489) (3.569) (5.205) (5.229)

LFIRMSIZE 0.070 *** 0.070 *** 0.061 *** 0.060 *** 0.071 *** 0.071 *** (8.017) (8.072) (12.877) (12.770) (10.419) (10.426)

LG 0.024 *** 0.025 *** -0.067 *** -0.068 *** 0.019 *** 0.020 *** -0.068 *** -0.067 *** 0.016 * 0.016 * -0.079 *** -0.079 *** (2.742) (2.783) (-4.554) (-4.604) (2.591) (2.674) (-8.303) (-8.248) (1.874) (1.867) (-6.524) (-6.530)

LTRADE -0.018 -0.017 -0.020 * -0.020 * -0.020 ** -0.020 ** -0.023 *** -0.023 *** -0.015 -0.015 -0.023 ** -0.023 ** (-1.360) (-1.235) (-1.814) (-1.802) (-2.275) (-2.303) (-3.186) (-3.165) (-1.193) (-1.142) (-2.170) (-2.172)

LGDP 0.068 *** 0.069 *** 0.044 *** 0.045 *** 0.056 *** 0.056 *** 0.028 *** 0.028 *** 0.081 *** 0.081 *** 0.045 *** 0.045 *** (5.371) (5.485) (4.683) (4.743) (6.523) (6.523) (4.759) (4.743) (6.980) (6.827) (5.176) (5.139)

LINCDIF -0.121 *** -0.163 *** -0.098 *** -0.124 *** -0.099 *** -0.082 *** -0.088 *** -0.073 *** -0.146 *** -0.173 *** -0.130 *** -0.136 *** (-3.764) (-4.514) (-3.712) (-4.337) (-3.990) (-3.054) (-5.448) (-4.200) (-4.185) (-4.580) (-5.109) (-5.065)

LVADIF 0.076 *** 0.052 ** -0.038 -0.032 * 0.048 (*) 0.012 (2.580) (2.022) (-1.516) (-1.859) (1.575) (0.482)

PERIOD74 0.039 0.033 0.050 * 0.046 (*) 0.022 0.025 0.044 ** 0.046 *** 0.035 0.031 0.051 * 0.050 * (1.370) (1.169) (1.711) (1.584) (1.107) (1.259) (2.526) (2.667) (1.216) (1.055) (1.782) (1.728)

PERIOD78 0.023 0.023 0.024 0.025 0.003 0.003 0.011 0.012 0.006 0.006 0.010 0.010 (1.093) (1.117) (1.318) (1.348) (0.173) (0.196) (0.983) (1.022) (0.314) (0.309) (0.575) (0.571)

PERIOD90 -0.055 *** -0.059 *** -0.036 ** -0.039 ** -0.038 *** -0.036 *** -0.030 *** -0.028 *** -0.054 *** -0.056 *** -0.036 ** -0.037 ** (-2.856) (-3.020) (-2.221) (-2.372) (-2.868) (-2.710) (-3.014) (-2.821) (-2.988) (-3.109) (-2.360) (-2.403)

PERIOD94 -0.045 ** -0.058 ** -0.035 * -0.044 ** -0.025 * -0.018 -0.021 ** -0.015 -0.043 ** -0.051 ** -0.028 * -0.030 * (-1.994) (-2.546) (-1.872) (-2.295) (-1.688) (-1.222) (-2.052) (-1.481) (-2.100) (-2.464) (-1.728) (-1.787)

T-statistic in brackets. Estimates are heteroskedastic consistent. *, **, *** denotes significance at the 10%, 5%, and 1% level, respectively.