1

Trade and FDI in Services: Complements AND Substitutes!

Carmen Fillat Castejón

University of Zaragoza, Spain

Joseph F. Francois

University of Linz, Austria, and CEPR, London

Julia Woerz

The Vienna Institute for International Economic Studies, Austria

V 0.1

31/08/2007

Abstract:

This paper has tested is there exists a complementary or substitutive effect in the relationship between cross-

border imports and FDI in the services sector, a question rarely analyzed in the literature about the services sector.

We have estimated a static and dynamic model where FDI inward stocks depend on contemporaneous and lagged

services imports, vice versa, and controlling for the usual gravity variables. We found robust contemporaneous

and lagged complementarity from FDI to services imports which is corroborated by a long-run approach. We have

not found a robust static effect from services imports to FDI inward stocks, nor also in the long-run. But an

interesting result is the substitutive effect arising when the dynamic accumulation of FDI is taken into account.

This might explain the absence of a long-run relationship although there exists in the opposite direction. In both

cases such complementary and substitute effect are robust in average but they also depend on the specific group of

countries considered.

Keywords: FDI, imports, services, panel data, substitution and complementary effects.

JEL: F10, F14, F21

2

Introduction

The question whether trade and FDI act as complements or substitutes in delivering goods

across borders is not a new one and has been studied extensively. For instance, Fontagné and

Pajot (1999) provide a comprehensive overview of the rich pool of literature dealing with this

subject. They point out that this relationship depends on the level of analysis: at the firm level

one will expect them to be substitutes, while there are compelling reasons - based on New

Trade Theory arguments - for a complementary relationship at the macro-level (Pfaffermayr

1996). Given these distinctions, which are extended in Egger and Pfaffermayr (2005) to

include further the magnitude of plant set-up costs compared to trade costs, the empirical

findings up to date have remained inconclusive. Fontagné and Pajot (1999) have ascribed this

to a confusion of effects at different levels of the economy (firm, industry and macro level)

and to differences between vertical and horizontal FDI, two points that are both widely

accepted in the literature (Zarotiadis and Mylonidis 2005, Egger and Pfaffermayr (2005),

among others). Reading through the empirical literature suggests that the case for

complementarity between trade and FDI is stronger, which is associated with vertical FDI and

rather low trade costs. This is intuitively compelling given that the majority of FDI takes

place between high developed countries, where vertical FDI is expected to play a greater role

than between partners at different levels of economic development. However, both

relationships are consistent with viewing trade and FDI as two equivalent modes for the

international provision of goods. Thus, like in services trade, these two channels can be seen

as two modes for trade. While this is not as explicitly recognized when talking about

merchandise trade, the GATS explicitly lists even four different modes of delivering services

across international borders, including as the most prominent means of international services

provision cross-border trade (mode 1) and sales through local establishments, i.e. through FDI

(mode 3). Mainly due to data limitations, the questions whether these different modes act as

complements or substitutes in services trade has rarely been dealt with in the literature.

Exceptions are Buch and Lipponer (2004), Hejazi and Safarian (2001) or Fontagné (1999),

among others.

Nevertheless, there are good reasons to question that the relationship between cross-border

trade and FDI in services is the same as in merchandise trade. Banga (2005) points out that

while the determinants for FDI are generally found to be the same for goods producing firms

3

and for services delivering ones, the importance of these determinants differ strongly between

the two sectors. Government regulations, policies, cultural distance and the tradability of

services (influenced by technological progress as well as by economic policy and regulatory

measures) are the prime factors influencing FDI in services. In contrast, market size, barriers

to trade and cost differentials in production are the main determinants for FDI in goods. Thus,

the question whether these two modes of international service delivery act as complements or

substitutes is not only largely unanswered (most existing studies consider either total services

or a particular sector like financial services in Buch and Lipponer, 2004), it is further of great

importance in the present GATS negotiations. Offering schedules are often reluctant to

include mode 3 in the lists. However, when the two modes are acting complementary, this

would act as a backlash on opening up to trade through mode 1 (cross-border trade). In

deriving our theoretical basis for the empirical analysis of this relationship we depart from the

idea of a composite delivery of a service involving different modes of provision.

This paper is intended to fill this gap, using a newly constructed dataset that combines data

for modes 1, 2 and 3 for 28 OECD countries over the period 1994 to 2004, distinguishing

between total services and seven individual service sectors. Our empirical estimations are

based on a Melitz-Krugman-Ethier type model for demand in services, which incorporates

elements of new trade theory. The next section describes the data set in more detail thereby

revealing an important distinction between the long-run relationship and short-run interactions

between cross-border trade and FDI in the service sector. Section 2 derives the theoretical

model. Section 3 looks at the static and dynamic relationship between trade and FDI in

services. This is corroborated in section 4 with the long-run approach analysis where the

heterogeneity by geographic and integrated areas is considered, and the paper finishes with a

preliminary section of conclusions.

1. Description of the Data Set and Further Motivation

For the analysis we collected data from different sources (IMF, OCED, World Bank). Our

data for service imports, covering basically modes 1 and 2, comes from published IMF

4

Balance of Payments Statistics, compiled according to BOP Manual 5. FDI stock data, as a

proxy for mode 3 trade, is taken from OECD Source and classified by the OECD’s own

industry classification based on ISIC, revision 3. The time period covered ranges from 1994-

2004. The combination of the two datasets implies that the sample covers 28 OECD

countries.1 The data is mapped to individual service sectors according to the BOP

classification. We left out sectors where the number of missing observations exceeded the

observations that were actually reported. Thus, we focus on the following categories: total

services, transport, travel, communication, construction, finance, insurance and other business

services. We have approximately 200 observations per service category. All other data come

from the World Development Indicators published by the World Bank (i.e. GDP, Value

added, ppps), while distance is taken from CEPII’s distance dataset and exchange rates are

from the IMF International Financial Statistics.

Trade in services has in general risen in the OECD over the past decade. Figure 1 displays the

growth in import volume and FDI inward stocks for total services as well as by the three main

sectors, transport, travel and the sum of the remaining five categories listed above. We shall

call the latter group henceforth “producer services”.2 It becomes evident from Figure 1 that

this category is strongly responsible for the high growth of FDI in the service sector. The

tremendous growth in service sector FDI is almost entirely driven by producer related

services. Also it is the most important category for cross-border trade in services in the

OECD. Growth through modes 1 and 2 has not been as impressive as through FDI, however,

trade flows have doubled over the past decade in all three categories.

[Figure 1 here]

In this paper we focus on the interaction between the two modes of supply, namely across the

border (including here also movement of consumers) and through foreign establishment. We

1 While cross-border trade at the sectoral level (BOP classification) is in principle available for 178 countries in the world, detailed and comparable FDI data by sectors is only available for the OECD members. Consequently our sample contains all OECD countries without Belgium and Luxembourg. 2 This refers to the sum of communication, construction, finance, insurance and other business services. Due to too many missing observations, this group does not reflect all categories usually labelled “producer related services”. Specifically we are missing out here: computer and information services and royalties and license fees.

5

would ideally measure mode 3 trade by the sales of foreign affiliates in the service sector.

However, this type of statistic exists up to date only for very few countries. The U.S. is more

or less the only country which publishes a comprehensive FATS statistic. Thus, we can only

use FDI stocks in the country as a very rough proxy for service supply through foreign

establishment. Implicitly we are therefore assuming that foreign affiliate sales are an invariant

function of the value of foreign direct investment.

Figure 2 plots FDI inward stocks against service imports for all 28 countries for each service

sector separately. The left hand side graph shows the average import flows and FDI stocks in

current US-Dollar over the period 2001-2004. For all service sectors with the exception of

construction services, we see a positive relationship. Thus, more inward FDI in a country is

combined with more service imports in the same sector. This very preliminary look at the data

thus reveals a contemporaneous complementarity between trade and FDI in services.3 In sharp

contrast to this long-run growth rates of both variables are often negatively correlated over the

entire period 1994-2004. Thus, in a dynamic perspective sectors with high FDI growth over

the entire period experienced weaker growth in service imports. This hints towards a

substitutive relationship between the two modes in the long-run. The two exceptions to this

observation are communication and other business services. Figure 3 very impressively

supports this preliminary finding for total services. Here we can see that the share of FDI has

risen in total services trade: the ratio of FDI stocks in services to GDP has increased much

stronger than the ratio of service imports to GDP (which has remained roughly constant).

While in absolute terms, service imports and FDI inward stocks in services were roughly

equal in 1994 (0.84 million USD of service sector FDI stocks and 0.77 million USD of

service imports), by 2004 FDI stocks amounted to 3.3 million USD while service imports

have just about doubled to 1.3 million USD for the OECD in total.

[Figures 2 and 3]

3 For the period 1994-1997, the same positive relationship was observed for all services sectors, also for construction services.

6

2. Theoretical backing of the gravity approach for modelling FDI and trade in the

service sector

Conceptually, cross-border services trade and foreign affiliate sales may be substitutes or

compliments. There are several reasons to expect that they are often gross compliments in

production (i.e. joint inputs) though with some degree of substitution possible. For example,

because services require interaction between provider and consumer (Hill 1977, Francois

1990), it will usually be the case that cross-border trade in services requires some local value

added to facilitate interaction between provider and consumer. In addition, from available

balance of payments and trade data, we observe both trade and FDI across service sectors. If

we are willing to assume that FDI in services is a legitimate measure of affiliate sales in the

service sector, this means we observe both cross-border and affiliate sales.

We start with a general representation of services S as a composite of cross-border inputs T

and affiliate activities F. This may, for example, involve a banking product supported by

headquarter activities but sold and serviced through a local office. Formally, we can represent

total foreign sales of services as in equation (1), where σ=1/(1-ρ) is the Allen-elasticity of

substitution.

S = f F,T( )= A aF F( )ρ + aT T( )ρ( )1

ρ , 0 ≤ ρ ≤ 1 (1)

If sales through affiliates and trade (F and T) are prefect substitutes, then

S = A aF F + aTT( ), ρ = 1 (2)

In more general terms, from the first order conditions for cost-minimization we will have the

following:

7

F = SA−1 aF

PF

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

Pσ = SA− 1+σ( ) aF

PF

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

aFσ PF

1−σ + aTσ PT

1−σ( )σ /(1−σ )

T = SA−1 aT

PT

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

Pσ = SA− 1+σ( ) aT

PT

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

aFσ PF

1−σ + aTσ PT

1−σ( )σ /(1−σ )

(3, 4)

P = A−1 aFσ PF

1−σ + aTσ PT

1−σ( )1/(1−σ ) (5)

From equations (3-5), it is straightforward to link demand for cross-border and local service

sales as a function of changes in the price of cross-border and local affiliate inputs.

dTdPF

= ε + σ( ) Pε +2σ −1aF PF−σ aT

PT

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

Aσ −2PF−1

⎛

⎝ ⎜

⎞

⎠ ⎟

dTdPT

= − Pε +σ aT

PT

⎛ ⎝ ⎜

⎞ ⎠ ⎟

σ

−εaTσ PT

1−σ + σaFσ PF

1−σ( )Aσ −2PT−1

⎛

⎝ ⎜

⎞

⎠ ⎟

(6,7)

A similar set of equations hold for F. In equations (6) and (7), ε<0 is the elasticity of demand

for S. From equation (6), the impact of a drop in the price of providing local affiliate inputs

on cross-border trade depends on the elasticity of substitution between F and T, and the

underlying elasticity of demand for composite services S. If the elasticity of substitution is

relatively low -- in particular if σ < ε -- then they actually serve as gross compliments.

Alternatively, as long as σ > ε , they will serve as gross substitutes.

We have seen dramatic increases in FDI flows in the service industries in the lat 10 years,

along with moves to privatize and deregulate service sectors. Liberalization of service sector

FDI means a reduction in the cost of the cost of running local affiliates. From equations

(3,4) this implies a rising share of local affiliate relative to cross-border sales. Controlling for

overall growth in demand, the theoretical impact on cross-border sales is ambiguous. From

equations (6,7), it will depend on the elasticity of substitution relative to the elasticity of

demand. We can summarize the implications of local service sector liberalization and related

FDI liberalization as follows:

8

• In the cross-section, net complimentarity of F and T means a relatively low technical

degree of substitution

• Over time, increases in total service sales S imply rising both cross-border trade and

FDI

• Controlling for shifts in demand, the impact of FDI growth driven by local market

liberalization over time on cross-border trade is ambiguous

Technical change has a similar set of implications. In our data, we will look at both trade-FDI

interactions in the cross-section, and in a dynamic panel. In the cross-section,

complimentarity will tell us we have a relatively low degree of substitution between cross-

border and local sales of services. In the dynamic panel, we are interested in the relative

evolution of cross-border and affiliate sales.

3. The Static and DynamicView: Complementarity versus Substitutability between

FDI and Imports

In this section we analyze the effect of FDI stock inflows on services cross-border trade and

vice versa for the aggregate of total services, both from a static and a dynamic point of view.

We estimate first a static panel data where we test their complementary or substitute effect

between contemporaneous FDI and services imports and where the most usual gravity

variables are controlled; also, we introduce the lagged right hand side (RHS) variable from

which this effect arises in order to start introducing the possibility of capture this effect along

time, just from one year to another.

So the equation to be estimated are the following:

log servMit = β1* log fdiit-1 + β2 * log (GDPpc)it + β3* log (pop)it + β4 * log(dist)it + εit

log fdiit = β1* log servMi t-1 + β2 * log (GDPpc)it + β3* log (pop)it + β4 * log(dist)it + εit (8)

where:

9

servMit are the total cross-border services imports for country i and year t

fdi i t-1 are the total stock inflows in the services sector in country i and year t-1, one year

lagged but alternatively the contemporaneous FDI effect is tested; the same is tested in the fdi

stocks equation, where both contemporaneous and lagged services imports effects are tested.

GDPpc is the per capita GDP for country i and year t, measured in current PPP US dollars

pop is the population of the host country

dist is the GDP weighted average distance of the host country in order to capture the global

distance to the rest of countries in the sample

ε is the error term, with an unobservable country-specific component and the remainder

disturbance. We estimate the within or fixed effects model where the country-specific effect

and all the regressors are assumed to be independent of the disturbance. The biased of

omitting variables is controlled in this estimation.

Data sources are described in section 1 and in the data panel we have a sample of 24 countries

and 10 years, although there are some missing values in this sample.

The two first columns in Table 1 show the estimations for services imports, where 1a

corresponds to the static and contemporaneous estimation of fdi effect on services imports,

and column 1b considers the one period lagged effect. Both of them yield a positive

significant effect of per capita GDP and negative from weighted distance, as expected, and a

complementary effect arising from both contemporaneous and lagged FDI inflows. Anyway,

we might expect FDI is not completely exogenous and is predetermined; in this case the

accurate way of estimating this effect also with fixed effects, is the Arellano and Bond(1991)

GMM instrumental estimation, which considers the past information about FDI4. Columns 2a

and 2b in Table 1 present this estimation and also in this case the complementary effect is

confirmed, and it is higher than in the within estimation, probably indicating that it is

convenient to take into account the previous years information about FDI stock inflows in

order to capture the complete effect on services imports.

[Tables 1 and 2 here]

4 All possible lags are taken in order to predict the value of FDI. Several lag structures have been considered and the result is robust. As a rule of thumb, it is suggested to maintain the number of instruments below the number of groups in order to avoid the overestimation of the Sargan test.

10

Table 2 presents the static estimation for FDI equation, also in the two first columns 1a and

1b. In this case there is no evidence of any effect of FDI on import services, nor

contemporaneous or previous. Also we might expect that FDI is not exogenous, moreover

because we have a really high adjusted R2 which can indicates the autocorrelation possibility.

So, again we test the instrumental consistent estimation by using GMM, and only when the

past information about contemporaneous services imports is considered we can check there is

some complementary effect on FDI5. So, although there seems to be a clear complementary

effect of FDI on services imports, this effect is not robust in the opposite case, that is an

increase in services imports does not necessary imply an increase on FDI stock inflows, from

the static point of view.

As we said, both FDI and services imports can be predetermined and exhibit some dynamics,

as the extensive literature about FDI determinants underlines agglomeration, and here the past

information about imports services seems to be relevant. So that we estimate now a dynamic

model for both equations, which now turn to be as follows:

log servMit = α* log servMi t-1 + β1* log fdiit-1 + β2 * log (GDPpc)it + β3* log(pop)it + β4 * log(dist)it + εit

log fdiit = α* log fdiit-1 + β1* log servMi t-1 + β2 * log (GDPpc)it+ β3* log(pop)it + β4 * log(dist)it + εit

(9)

Columns 3a and 3b in Table 2 show the estimation for services imports equation. Arellano and

Bond GMM method is the accurate method of estimation for dynamic panels, where the

variables are taken in differences in order to get rid of the fixed effects and so eliminate the

possible correlation between them and the error term; as instruments, all the possible lagged

endogenous variable are taken6. Also here the effect of FDI is considered as predetermined,

with all possible past information taken into account. In these cases, we confirm a dynamic

process in services imports, which seems to be more important when the contemporaneous

5 Again, all possible lags have been taken, which means in this case about five years lagged. Several possibilities for this have been considered, with the same conclusion. 6 Several sets of lags are used as instruments with the same result than the one presented. Because of the sensitivity of estimation to the instruments considered the coefficients vary a little but not their statistical significance.

11

effect of FDI is considered, with a very high dynamic coefficient of 0.77. In any case, we find

again a complementary effect of FDI on imports. But the high dynamic coefficient might be

indicating the possibility of persistency in this estimation, in which case the autorregresive

process dominates the model and regressors would turn to be almost independent of the

endogenous variable; in this case, the instruments are weak and the coefficients biased.

Blundell and Bond (1998) show that in this case, and especially when the sample is small, it is

possible to improve the efficiency and precision of dynamic estimations when a system of

differenced and levels equations is estimated, by using lagged and differenced variables as

instruments respectively. For this reason, we have tested the hypothesis of persistency, where

the alpha dynamic coefficient equals one, and we find that this cannot be rejected when

contemporaneous FDI is used. To control the possible bias for small samples and persistent

processes we have also performed the SYSTEM-GMM Blundell and Bond estimation, which

results are shown in columns 4a and 4b. Here again, the dynamic process is very relevant and

significant, and with this more precise estimation we cannot reject the hypothesis of

persistency. Although we loose most of the gravity variables, we still can check the robustness

of a complementary effect of FDI on imports services, with a 0.23 elasticity for

contemporaneous FDI and one of 0.09 for lagged FDI.

If we turn to FDI equation and the effects coming from imports services (Table 2), we also see

a significant dynamic process, but where the dynamic coefficient is not so high than the one for

imports equation. The possibility of persistence in this process is tested, and it can not be

rejected when the effect from lagged services imports is considered7. In order to avoid the

possible inefficiency and bias coming from this persistency we perform the SYSTEM-GMM

estimation, confirming the dominance of the dynamic process, although never so important in

the FDI process as it is in the imports one. And what is more relevant for our inquiries here is

that, when contemporaneous services imports are considered, the persistency could be rejected

and a complementary effect of imports on FDI is found. But this disappears, an so all the

gravity variables, when the SYSTEM-GMM estimation is performed. But when the lagged

import services effect is considered, we cannot reject persistency and a SYSTEM-GMM

estimation is recommended. In this case, the dynamic process turns to have a more precise 7 In this case, this lagged effect is not significant. So, apparently, the dynamic process dominates.

12

coefficient of 0.83, lower than unit, all the gravity variables are significant and with the

expected sign and, on the top of this, a significant substitutive effect of lagged services imports

on FDI inflows is found. So, by considering the dynamic process of agglomeration in FDI

inflows where FDI seems to follow market pull factors as agglomeration, market size and

distance, we find a negative effect from recent past services imports, which could be indicating

that the tradability of services reduces the probability of providing them by investing in the

market country (Banga, 2005). In this case, liberalizing services could difficult of deepening

the channel through FDI, which follows market incentives but the final FDI stock inflow

would be lower if imports are promoted.

To sum up, there is a robust complementary effect from contemporaneous or recently past FDI

stock inlows on services cross-border imports, which is according to the literature for

manufactures and some specific services like financial ones in some case studies, and here it is

found both in a static and a dynamic analysis. But the complementary effect of cross-border

imports on FDI in services is not found in the static analysis, and furthermore, it is found a

substitution effect coming from recent past imports when the dynamic process of FDI growth

is considered.

5. The long-run approach and the geographic heterogeneity

This finding of robust dynamic processes and of the complementary effect from FDI to

imports, but a substitution one from imports to FDI, we think it might be relevant to test the

existence of a long-run relationship between FDI and imports where they keep being

complements or substitutes. For a period of 10 years we can adopt the approach by Pesaran and

Smith (1995) and Pesaran et al. (1999) which estimates the long-run coefficients of interest.

We use a simple heterogeneous partial adjustment panel model as:

log (Yit) = αi + βi log(Xit) + λi log(Yit-1) + uit uit~IN(0,σi2) (10)

where i=1…N is the group of countries over t=1…10 years, Xit denotes the variable from

which we want to test the complementary of substitutive effect. The associated long-run

13

coefficients can be derived as θi=βi/(1-λi). The country-specific intercept picks up all omitted

factors that vary across countries. A convenient reparameterisation of (10) is:

Δlog (Yit) = αi - (1-λi)[log(Yit-1) - βi/(1-λi) * log(Xit) ] + uit (11)

= αi - (γi)[log(Yit-1) - θi log(Xit) ] + uit (12)

this

This non-linear equation allows estimate the long-run parameters of interest θ and γ. If the

individuals are homogeneous the estimators are consistent, but they are not if there is

heterogeneity in the sample, even if the time dimension would be high. There are serveral

options for avoiding this heterogeneity bias, usually consisting in estimating each country

effect separately and then compute the Mean Group estimator (Pesaran and Smith, 1995), just

averaging the individual effects. Because this method is not particularly efficient, Pesaran et al.

(1999) suggest the Pooled Mean Group estimator, which allows the intercepts, the short-run

coefficients and the error variances to vary across panel groups, but imposes common long-run

coefficients8. So equation (5) becomes:

Δlog (Yit) = αi - (γi)[log(Yit-1) - θ log(Xit) ] + uit (13)

Equation (13) is estimated in Table 3, first by pooling both long-run parameters and assuming

country-specific effects. But because there is not any reason for believing there is

homogeneity, we have estimated two alternative models: on one hand the mean group for the

dynamic parameter (γi) and the pooled mean group for the long-run parameter (θ); on the other

hand, the pooled mean group for the dynamic parameter (γ) and the mean group for the long-

run parameter (θi)9.

8 Pesaran et al. (1999) also argue that short-time coefficients are more likely to vary across countries than the long-run parameters. 9 This two alternatives yield almost identical estimators, so that we don not estimate all the individual effects for all the parameters to avoid loosing degrees of freedom. This is also the reason for not including here the gravity controls as we did in the previous section. Anyway both estimations are being compared excluding these controls.

14

Table 3 shows how the long-run effect of import services in FDI inflows is never significant,

so we can keep in mind the dynamic but short-run estimation of this substitution effect we

found in the previous section. But we find a positive long-run effect of FDI inflows on import

services, again indicating a complementary effect like the one found for the short-run static and

dynamic estimation. The dynamic parameter (γi) is identical both in the pooled, pooled mean

long-run effect or mean long-run effect (θ), with the small value of 0.006-0.007. But the long-

run effect of FDI on services is quite different is we assume an homogeneous or heterogeneous

sample. So that, we can estimate the dynamic parameter – with the mean group or the pooled

mean group because it yields the same value here – and then estimate the mean group

parameter (θ) in order to compare. Both estimations yield again an identical long-run

parameter, with a value of 9.27, and its correspondent short-run value equal to 0.068 in

average. This mean is computed from the estimates for the different integrated areas in the

sample: the European Union (EU), NAFTA, Australia and New Zealand (AUNZ) and the rest

of countries in the sample (OTHER), where only NAFTA and OTHER show a significant

complementary effect, with a long(short)-run parameter of 4.95(0.04) and 1.62(0.01)

respectively.

[Table 4 here]

It might be interesting to estimate the different short-run parameters for the different integrated

areas by using the dynamic panel data approach. This is shown in Table 4. In the first column

the dynamic and trade effect are controlled by the gravity variables like in the previous

estimations. We can confirm the significance of these effects and again a complementary effect

from recent past FDI on services imports for EU, NAFTA and OTHER, but not for AUNZ,

where it is insignificant. Also we can confirm a dynamic effect for FDI and a substitutive effect

arising from recent past services imports for the same areas: EU, NAFTA and OTHER. Just for

better comparison with the long-run approach, we estimate the same dynamic panel but

excluding the gravity controls, and then we see the coefficients are not very different to the

short-run ones computed from the Pesaran approach. FDI keeps the short substitution effect for

EU and OTHER. And services imports maintain the complementary effect from FDI for EU,

NAFTA and OTHER, with coefficients of 0.06, 0.06 and 0.01 respectively, and that are quite

15

similar to the short-run ones estimated in the Pesaran approach: 0.02, 0.04 and 0.01,

respectively.

This evidence suggests that, although more research has to be done, it seems to be a robust

complementary effect from FDI towards services imports in the short and long-run, and also a

substitutive effect from recent past services imports towards FDI in the short-run with no

evidence in the long run, all of them heterogeneous depending on the EU, NAFTA, AUNZ or

OTHER areas in the OECD countries are considered.

Conclusions

This paper has tested is there exists a complementary or substitutive effect in the relationship

between cross-border imports and FDI in the services sector, a question often studied for

manufactures but rarely analyzed for services in the literature.

For the sample of OECD countries, we have estimated a static and dynamic model where FDI

inward stocks depend on contemporaneous and lagged services imports, vice versa, and

controlling for the usual gravity variables. We found robust contemporaneous and lagged

complementarity from FDI to services imports, a result very usual in the literature on

manufactures and also in some country studies on services trade. This result is corroborated by

a long-run approach, where a robust dynamic and long-run complementary effect is again

found but heterogeneous dependent on specific groups of countries in the sample, with

associated short-run parameters which are similar to the dynamic panel estimations.

We have not found a robust static effect from services imports to FDI inward stocks, nor also

in the long-run. But an interesting result is the substitutive effect arising when the dynamic

accumulation of FDI is taken into account. This might explain the absence of a long-run

relationship although there exists in the opposite direction.

16

References

Arellano, M. and SR Bond (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58, 277-297.

Bagchi-Sen, S. 1995. FDI in US Producer Services: A Temporal Analysis of Foreign Direct Investment in the Finance, Insurance, and Real Estate Sectors, Regional Studies, Vol. 29, pp. 159-170.

Banga, R. (2005), “Foreign Direct Investment in Services: Implications for Developing Countries”; Asia-Pacific Trade and Investment Review 1(2), 55-72.

Blundell, R. and SR Bond (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87, 115-143. Buch, C. and A. Lipponer (2004), “FDI versus cross-border financial services: The

globalisation of German banks”, Deutsche Bundesbank Discussion Paper series 1 05/ 2004.

Egger, P. and M. Pfaffermayr (2005). Trade, Multinational Sales, and FDI in a Three-factor Model. Review of International Economics, 13(4), 659-675. Fontagné, L. (1999). Foreign Direct Investment and International Trade: Complements or Substitutes?. STI Working Papers 1999/3. Fontagné, L. and M. Pajot (1999). Investissement direct à l’etranger et échanges extérieurs : un impact plus fort aux Etats-Unis qu’en France. Economie et Statistique, (326-327) : 31-52. Hejazi, W. and A.E. Safarian (2001), The complementarity between U.S. foreign direct

investment stock and trade. Atlantic Economic Journal 29 (4), 420-437.

Hill, T.P. 1977: “On goods and services,” The Review of Income and Wealth, 23, 315-338.

Francois, J.F 1990: “Trade in Nontradables: Proximity Requirements and the Pattern of

Trade in Services,” Journal of Economic Integration.

Pain, N and D van Welsum (2004). International Production Relocation and Exports of Services. OECD Economic Studies nº 38, 2004/1.

Pesaran, MH and RP Smith (1995). Estimating Long-Run Relationships from Dynamic

Heterogeneous Panels. Journal of Econometrics, 68, 70-113. Pesaran, MH, Y SHIN and RP Smith (1999). Pooled Mean Group Estimator of Dynamic

Heterogeneous Panels. Journal of the American Statistical Association, 94, 621-634.

17

Pfaffermayr, M. (1996). Foreign Outward Direct Investment and Exports in Austrian Manufacturing: Substitutes or Complements? Weltwirtschaftliches-Archiv. 132(3): 501-22

Windmeijer (2005). A Finite Sample Correction for the Variance of Linear Two-Step GMM

Estimators. Journal of Econometrics 126(1), 25-51. Zarotiadis , G. and N. Mylonidis (2005), “FDI and Trade in the UK: Substitutes or

Complements?”, ETSG Conference, Dublin.

Figure 1. Growth rates of imports (left) and FDI (right) in the services sector

OECD

0

500000

1000000

1500000

2000000

2500000

3000000

3500000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

EU-15

0

500000

1000000

1500000

2000000

2500000

total

trans

port

trave

l

produ

cer

total

trans

port

trave

l

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Figure 1. Growth rates of imports (left) and FDI (right) in the services sector (cont.1)

EFTA

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

NAFTA

0

200000

400000

600000

800000

1000000

1200000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Figure 1. Growth rates of imports (left) and FDI (right) in the services sector (cont.2)

AU-NZ

0

20000

40000

60000

80000

100000

120000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

JPN-KOR-TUR

0

50000

100000

150000

200000

250000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Figure 1. Growth rates of imports (left) and FDI (right) in the services sector (y cont.3)

NMS

05000

100001500020000250003000035000400004500050000

total

trans

porttra

vel

produ

cer

total

trans

porttra

vel

produ

cer

Mio

USD

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Figure 2: Trade flows and growth in trade flows by sectors (2001-2004)

trade flow s 01-04, total

y = 0,7132x + 2,3674

0

2

4

6

8

10

12

14

0 5 10 15

trade f low s 01-04, transport

y = 0,5839x + 5,2165

0

2

4

6

8

10

12

0 2 4 6 8 10 12

trade f low s 01-04, travel

y = 0,5915x + 5,0523

0

2

4

6

8

10

12

0 2 4 6 8 10 12

trade f low s 01-04, communication

y = 0,6938x + 0,5281

0

1

2

3

4

5

6

7

8

9

0 2 4 6 8 10 12

trade f low s 01-04, construction

y = -0,0548x + 5,5161

0

1

2

3

4

5

6

7

8

9

-2 0 2 4 6 8 10

trade f low s 01-04, f inance

y = 0,7032x - 0,5638

0

1

2

3

4

5

6

7

8

9

10

0 2 4 6 8 10 12 14

trade f low s 01-04, insurance

y = 0,7562x + 0,2264

0

2

4

6

8

10

12

0 2 4 6 8 10 12 14

trade f low s 01-04, other business services

y = 0,5695x + 5,0316

0

2

4

6

8

10

12

14

0 2 4 6 8 10 12

log-run grow th, total

y = -0,0397x + 0,0642

0

0,05

0,1

0,15

0,2

0,25

0 0,1 0,2 0,3 0,4

Note: Data are in logs, long run growth is calculated as the trend over 1994-2004.

Figure 3. Share of FDI and imports in total services trade

0 500000

1000000 1500000 2000000 2500000 3000000 3500000 4000000 4500000 5000000

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

Mio USD

ImportsFDI

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

per cent

FDI Imports

Table 1: Static and Dynamic Estimation of the Services Imports Equation SERVICES M 1a 1b 2a 2b 3a 3b 4a 4b dep.var. static static static static dynamic dynamic dynamic Dynamic within within predetermined predetermined diff-gmm diff-gmm sys-gmm sys.gmm log servM(-1) 0.7768 0.4735 1.3319 1.0342 5.76 2.76 5.37 8.78log fdi 0.0865 0.348 0.1846 0.2295 2.02 2.49 2.85 1.85 log(GDPpc) 0.7804 0.7125 -0.1804 0.0928 -0.4626 -0.0286 -0.9296 -0.1854 3.60 4.03 -0.31 0.18 -1.67 -0.13 -1.57 -0.81log(pop) 0.2514 0.1218 -0.6934 0.0757 -0.3644 0.0012 -0.506 -0.1246 0.71 0.32 -0.96 0.19 -1.31 0.01 -1.68 -1.10log(dist) -1.7038 -2.2697 -0.9138 -1.746 -1.4262 -2.0544 0.5888 0.1155 -4.10 -6.36 -1.44 -3.41 -3.16 -5.44 1.49 0.79log fdi(-1) 0.1075 0.2649 0.1507 0.0902 3.11 1.86 1.82 2.08country dummies yes yes Sargan p-value 0.0761 0.0237 0.359 0.188 0.1151 0.1593

ar1p-value 0.2014 0.8422 0.1567 0.2951 0.1963 0.597 ar2p-value 0.2136 0.0558 0.1797 0.0203 0.1221 0.0919

obs. 173 164 157 164 182 190n.groups 23 23 23 23 23 24

instruments 13 12 22 21 18 18adjR2 0.99 0.99

Wald test 0.11 0.01 0.19 0.77dynamic = 1

NOTE: t-statistics in italics. All possible lags as instruments and 5 for sys-gmm. Bold means significant. Two-step dynamic estimation is shown, with the Windmeijer (2005) correction for heteroskedastic disturbances.

Table 2: Static and Dynamic Estimation of the FDI Equation FDI dep.var. 1a 1b 2a 2b 3a 3b 4a 4b static static static static dynamic dynamic dynamic Dynamic within within predetermined predetermined diff-gmm diff-gmm sys-gmm sys.gmm log fdi(-1) 0.2957 0.5828 1.0338 0.8337 1.65 2.54 11.57 6.46log servM 0.4405 0.6803 0.5463 -0.2570 1.59 2.17 3.1 -1.17 log(GDPpc) 3.3375 3.9123 2.3792 3.8418 1.9761 2.2224 0.3978 1.4239 9.76 12.48 2.98 4.76 2.78 2.77 0.88 2.59log(pop) 0.7507 1.1024 3.1251 4.0413 2.2518 2.2284 0.145 0.6955 0.72 1.23 3.7 5.45 3.06 3.49 0.62 2.15log(dist) -1.8344 -2.545 -2.0167 -3.5472 -1.3263 -1.7257 -0.2298 -0.8588 -1.53 -2.41 -2.41 -2.18 -0.99 -1.00 -0.80 -2.47log servM(-1) -0.0258 -0.1264 -0.5909 -0.6925 -0.11 -0.19 -1.34 -2.61country dummies yes yes Sargan p-value 0.1035 0.0365 0.2583 0.1869 0.2888 0.2043

ar1p-value 0.1398 0.0663 0.2461 0.0885 0.0371 0.0451 ar2p-value 0.9395 0.6695 0.9862 0.2531 0.9622 0.2445

obs. 200 182 173 157 149 149 173 173n.groups 23 23 22 22 23 23

instruments 12 11 20 19 22 21adjR2 0.99 0.99

Wald test 0.007 0.083 0.71 0.21dynamic = 1

NOTE: t-statistics in italics. 10 lags as instruments. Bold means significant. Two-step dynamic estimation is shown, with the Windmeijer (2005) correction for heteroskedastic disturbances.

Table 3: long-run FDI-services imports relationship FDI Services imports pooled Pooled Mean group pooled Pooled Mean group

Long-run/ Mean group

dynamic

Long-run/ Mean group

dynamic θ γ θ γi θi γ θ γ θ γi θi γ long-run parameters (θ) log FDI 2.0665 9.2708 9.2709 2.52 1.60 log FDI-EU 1.8661 1.59log FDI-NAFTA 4.9533 2.11log FDI-AUNZ 28.6396 1.26log FDI-OTHER 1.6244 1.69log ServM -32.8315 47.1554 47.1554 -0.18 0.91 log servM-EU 8.147 1.62 logservM-NAFTA 3.2714

0.51 log servM-AUNZ 152.7559 0.79 logservM-OTHER 24.4472 1.43 dynamic parameters equilibrium correction parameter (γ) 0.0009 -0.0050 -0.0050 0.006 0.0074 0.0074 0.21 -1.04 1.65 1.84 γ-EU -0.0009 0.0015 -0.62 0.73 γ-NAFTA -0.0003 0.0039 -0.33 1.03 γ-AUNZ -0.0162 0.0228 -1.10 2.17 γ-OTHER -0.0026 0.0013 -0.75 0.78

adjR2 0.52 0.53 0.53 0.30 0.30 0.3Obs 173 173 173 190 190 190

NOTE: t-statistics in italics. All variables with 1 lag. Bold means significant.

Table 4: Short run FDI-services imports relationship by area SERVICES IMPORTS FDI SYS-GMM SYS-GMM SYS-GMM SYS-GMM log servM(-1) 0.9216 0.9429 log fdi(-1) 0.8339 1.1358 4.72 30.75 7.03 25.2log fdi(-1) EU 0.0878 0.0605 log servM(-1) EU -0.4226 -0.117 1.79 2.10 -1.89 -2.65log fdi(-1) NAFTA 0.0906 0.0593 log servM(-1) NAFTA -0.4508 -0.0909 1.35 2.09 -1.79 -0.93log fdi(-1) AUNZ 0.0972 0.0546 log servM(-1) AUNZ -0.284 -0.4901 2.65 1.35 -0.88 -0.72log fdi(-1) OTHER 0.1008 0.0072 log servM(-1) OTHER -0.4548 -0.114 2.23 2.49 -2.16 -2.18log(GDPpc) 0.0255 log(GDPpc) 1.1417 0.09 2.16 log(pop) -0.0334 log(pop) 0.4739 -0.24 1.36 log(dist) -0.0291 log(dist) -0.8045 -0.18 -1.91

Sargan p-value 0.54 0.35 sarganp 0.44 0.93

ar1p-value 0.63 0.65 ar1p 0.04 0.04ar2p-value 0.18 0.09 ar2p 0.42 0.77

obs. 190 190 N 173 173n.groups 24 24 N_g 23 23

instruments 24 24 instruments 23 21Wald test 0.69 0.07 Wald test 0.28 0.01

dynamic = 1 dynamic = 1

NOTE: t-statistics in italics. Bold means significant.