Local Industrial Systems and the Location of FDI in Italy

Lisa De Propris, Nigel Driffield and Stefano Menghinello

Abstract

This paper investigates the local determinants of FDI location across Italian manufacturing industries. Specifically it examines the importance of industry-specific local industrial systems as potential catalysts for attracting FDI. The paper develops a model of FDI location choice using a unique FDI database, stratified by industry and province. This extends previous analysis beyond the mere density of activity, to analyse the specific nature of agglomerations and their importance for attracting inward investment. The results also suggest that the importance of agglomeration differs between industries, and offers some explanation for this.

Keywords: Local industrial systems, knowledge sourcing, agglomeration, count data econometrics.

JEL: F23, R12

Lisa DePropris Nigel Driffield Stefano Menghinello

Birmingham Business School The University of Birmingham, Edgbaston, Birmingham B15 2TT l.de_propris@bham.ac.uk

Aston Business School, Aston University, Birmingham B4 7ET n.l.driffield@aston.ac.uk

ISTAT, Italy and Birmingham Business School The University of Birmingham, Birmingham B15 2TT menghine@istat.it

∗Lisa DePropris and Nigel Driffield gratefully acknowledge the financial support of the Nuffield Foundation for the funding of this research (Grant number: SGS/00741/A). Thanks are also due to participants at the EUNIP 2003 annual conference and l’institute-Bimingham Seminar Series for comments on earlier drafts of this paper.

1

1. Introduction

Traditionally, Italy has few policy initiatives designed to attract inward foreign direct

investment (FDI). Many countries in the developed world have come to see the

attraction of inward investment as synonymous with regional development, a position

that is extended by the various EU level policy initiatives designed to attract

internationally mobile capital to certain regions. Recently the stance of the Italian

government appears to have changed, with the strengthening of regional agencies

seeking to promote local development by attracting inward FDI. The extent to which

such policies have been effective is however yet to be tested.

As a result of this apparent ambivalence, Italy has received much smaller

levels of foreign investments compared with the other members of the EU, and Italy

ranks 109th in terms of potential foreign investment attractiveness (UNCTAD, 2003).

Initial analysis suggests that foreign owned firms in Italy are concentrated in

sectors with significant scale economies. Such industries account for 42.8 per cent of

the total FDI. Industries that are more specialised than average attract 24.4%, while

R&D intensive industries attract very little. Foreign firms in R&D intensive industries

are mainly located in metropolitan areas, while foreign-owned firms in older

traditional industries are mostly located in peripheral areas.

In the following analysis we extend this by considering Pavitt’s taxonomy

(1984) in order to examine the determinants of inward investment in Italy across

industries. This therefore extends the work of Piscitello-Mutinelli (1994) and Basile

(2003) who argue that variation in inward FDI in Italy essentially an industry level

phenomenon.

The dominant model of the motivation for a firm to enter a foreign market

through FDI has changed little since the seminal work of Dunning (1958) and Vernon

(1966). The basic framework has been one which envisages the firm generating

certain firm specific assets in its home country, then seeking to exploit these further

by creating income generating assets abroad. The importance of locational or “pull”

factors was very much secondary to this, in explaining which host countries the MNE

chose to locate. There is a relatively large literature that seeks to relate location

specific factors to the determinants of inward investment across countries, regions or

industries, often based on the importance of agglomeration or the possible links

between domestic sector and inward investors. Cantwell (1991), for example, shows

2

that there are significant benefits to both domestic and foreign firms from

agglomeration (see also Shaver, 1998). Location advantages at the local or regional

level could be self perpetuating where further growth of an industry sector makes the

location even more attractive (Head, Ries and Swenson, 1995, see also Krugman,

1991). Under such circumstances random location decisions in the past can result in

the development of specialised support infrastructure and a concentration in a given

industry (Wheeler and Mody, 1992). While the importance of agglomeration for

attracting FDI has been explored, following the work of Coughlin et al. (1991), the

importance of spatial organisation of activity, or clustering for attracting FDI is

largely unexplored. This is particularly important in the Italian context, where Local

Industrial Systems (henceforth LISs) have a long history. Equally, this provides a link

to the relatively recent concept of so called technology (or knowledge) sourcing FDI

that is developing within the international business literature. This is discussed in

more detail in the following section. In this paper, LIS are defined as local

concentrations of firms specialised in one or a few related sectors.

This paper examines the importance of local characteristics and agglomeration

economies for attracting inward FDI in Italy, within the context of LIS and other

forms of spatial organisation. The rest of the paper is set out as follows: section 2

examines the nature of location advantages for MNEs and recent contributions on

technology sourcing, and section 3 provides theoretical background and some

empirical evidence of LISs’ competitive advantages. Section 4 describes a model of

location choice for MNE to test our hypotheses, while data and econometric model

are described in section 5. Section 6 provides some thoughts on policy implications

and presents some concluding remarks.

2. The importance of location theory explaining FDI

The literature on FDI illustrates that the importance of location advantages has

increased, with the emphasis changing from natural and cost-related inputs

endowments to knowledge-based competencies. In particular, as Cantwell and

Santangelo (1999) note, the technological strengths of host countries is a relevant

feature to discriminate between the location options for the multinational firm. In

addition, the localised nature of learning processes has changed the geographical scale

of location patterns from the national to the regional or even local level. For instance,

3

Dicken (1998) and Cantwell and Iammarino (2000) show that foreign R&D activities

in the UK are strongly concentrated in the South-East of England. In a similar vein,

Driffield and Munday (2001) illustrate the importance of agglomeration economies

and spillovers on total factor productivity growth of UK regions, and demonstrate that

a critical level of regional concentration of economic activities, in effect the existence

of significant agglomeration economies, is a necessary condition for spillovers to

occur.

It is clear, however, that the ability of a locality to attract FDI merely

represents the potential for development, and that technology, or knowledge sourcing

is by no means automatic, but depends on the actions of the firms concerned

(Driffield and Love, 2003). In addition, given the nature of public good embodied in

local knowledge, and that the latter is not concentrated in specific firms but embedded

in the local industrial system, MNEs may realise knowledge sourcing as long as their

foreign affiliates undertake cooperative relationships with local firms rather than

engage in a predatory behaviour (Bellandi, 2001).

The above discussion illustrates why clusters of activity are likely to be

inherently attractive to firms seeking to tap into a pool of specialised knowledge and

competence. Cluster firms are characterised by a high degree of specialisation and

complementarity, that generates dynamic processes of knowledge creation -learning

and innovation -and knowledge transfer- diffusion and synergies. In clusters there are

collective learning processes that generate innovation and thereby competitiveness

also in non high-tech intensive sectors. In fact, clusters can be extremely competitive

in what the literature defines as traditional sectors; for instance, Sassuolo (Italy)

ceramic tile industrial district accounts for one third of the sector world export (De

Propris et al (2003). An innovative and competitive cluster can produce positive

externalities to its entire region: as the cluster grows the extent of vertical and

horizontal product differentiation increases. As a result, the cluster becomes a centre

of accumulated competencies across a range of related industries, and across various

stages of production (the so called production filière). These localised centres of

accumulated knowledge can be very attractive to outside firms.

4

3. The competitive advantages of Local Industrial Systems

The theoretical analysis of the LISs as sources of regional competitive

advantage draws on the concept of territorial competitiveness and the tangible and

intangible factors that drive it. The conceptual starting point of this stream of

literature is the flexible specialisation approach, pioneered by Piore and Sabel (1984),

which considers small firms, and especially firms within LISs, as an alternative model

of industrial development to large vertically integrated firms. This approach

emphasises the specific characteristics of the organisation of production in the LISs

that enable them to cope with uncertainty and to be flexible. The first model of firm

agglomeration was developed by Becattini (1979) who witnessed this phenomenon in

Italy and introduced the concept of industrial district to describe a specific model of

social and industrial development, rather than merely a sector-specific geographical

agglomeration of firms.

As global competitiveness is more and more associated to regional

competitive advantages, LISs in Italy, as anywhere else, have become crucial. Trade

and factor mobility have diminished the importance of traditional national measures

of comparative advantage, while non-tradable knowledge, technology or skills have

become increasingly important as measures of competitiveness (Porter, 1998). The

literature on the knowledge-based economy has emphasised the importance of local

competencies for knowledge creation and learning processes (Prahalad and Hamel,

1990).1 This literature distinguishes two types of knowledge: codified and uncodified

or tacit knowledge (Nonaka, 1991). The former is based on standardised scientific

protocols and then easily tradable on international markets, while the latter is

essentially embedded within single enterprises or specific geographical areas, and

cannot be transferred through standard agreements such as licensing. It follows that

tacit knowledge is limited spatially, and can only be transferred between organisations

with high levels of communication and mutual understanding. This argument

strengthens the link between knowledge creation and the geographical, social and

institutional frameworks supporting firms at the local level. Becattini (1979) defines

the industrial district as “a territorial entity characterised by the active presence of a

group of persons and a population of firms in a given historical and geographical

dimension.” This perspective on local development clearly highlights the strong

1 See Dosi and Malerba (1996) for further discussion on competencies.

5

interplay of social and economic factors as basic conditions for the successful

development of industrial districts.

The role of local technological externalities is important in this context, and

draws on the innovative milieu approach, developed by Aydalot (1986) and Perrin

(1988). Here the organisation of production at the local level is considered a complex

and self-contained micro-system, promoting strong interaction and cumulative

processes. The geographical boundaries of the innovative milieu are defined both by

spatial proximity effects, and by economic and cultural homogeneities within the

milieu. In particular, innovation processes and factors of success are specific to each

context and depend on (a) strong specialisation of the local industry in a filière or

technology, (b) dense interactions and synergies among local firms, (c) collective

learning processes, and (d) a strong sense of belonging to the local community. The

result is that the milieu stimulates “collective learning” (Camagni, 1991 and 1995;

Lawson and Lorenz, 1999). Also Storper (1995) stresses the intangible factors of

regions’ competitiveness and defines the region, and a LIS, as a “nexus of untraded

interdependencies”. His approach outlines how strong competitive regions develop

successful models of production that can not be easily imitated since they are

embedded in the underlying system of shared conventions and norms.

In the literature, definitions of agglomeration have widened to include

networks (Hakansson, 1987), commodity chains (Dicken, 1998), production systems

(Scott and Storper, 1992), and business systems (Whitley, 1992). Recent contributions

qualify LISs as systems of economic and social relations, emphasising the role of the

relational capital (Camagni, 1999) and social capital (Putnam, 1993; World Bank,

2001) for local development. Other contributions have looked at the development of a

specialised service sector as a result of co-operation among local firms (Brusco,

1989), or at the role for local development of business associations (Best, 1990;

Humphrey and Schmitz, 1996; Maskell et al., 1998) or at the targeting of LISs in

wider industrial policy framework (Bennet and McCoshan, 1993).

4. The determinants of the spatial distribution of FDI

Much of the recent work in this area is based on Coughlin et al. (1991), who

develop a model that of MNE location choice based on profit maximisation. Coughlin

et al. (1991) identify the main factors determining the spatial distribution of inward

FDI in the US. They demonstrate that FDI is attracted to regions with high levels of

6

final demand for the output, but also to regions with high densities of manufacturing

activity and extensive transportation infrastructure, whereas higher wages and taxes

deter FDI location.

The analysis of FDI location was recently extended, see for example Basile’s

(2003) analysis of Italy; Crozet et al (2003) of France; Devereux, Griffith and

Simpson (2003) of the UK, and Togo and Arikawa (2002) of Malaysia. Coughlin and

Segev (2000) extend the analysis by including educational attainment as a possible

determinant of FDI attraction. In addition, they show that whilst the condition of the

existing manufacturing base and taxation levels affect location at the state level, urban

regions are more conducive to FDI than rural ones. In his analysis on the location

determinants of FDI across Italian provinces, Basile (2003) considers the indicators of

agglomeration such as public research institutions and the relative concentration of

specialised providers of firm services as potential sources of positive externalities to

attract FDI.

The analysis presented here will extend the literature discussed above in two

ways. Firstly, it focuses on the potential heterogeneity of location determinants across

industries linking the location of the firm to the analysis of FDI, based on the standard

explanations of FDI from the international business literature, viz. (a) resource

seeking, (b) market seeking, (c) efficiency seeking, and (d) strategic asset seeking

(Dunning, 1998). Secondly, it will link the analysis to more complex considerations

of the local organisation of industrial activity, by distinguishing different kinds of

agglomeration economies at the local level.

For example, while efficiency seeking may be associated with the availability

of cheap labour, or with high productivity levels and agglomeration economies,

strategic asset seeking behaviour is linked to the sourcing of industry-specific

knowledge that is embedded in specific LISs. In fact, LISs are expected to hold

specific localised competitive advantages (knowledge creation and other locally

embedded intangible assets) absent in other locations with the same industry, and for

this they can attract higher levels of inward investment.

5. Econometric analysis

The basic theoretical model assumes that a firm in a given industry will

choose to locate in a particular region if and only if that choice will provide the

highest return to its investment:

7

(1)

where i denotes the firm and j indicates the locality providing the highest profit

among a set of k regions denoted by the suffix m. Following the study of Carlton

(1983) on the location choice of domestic firms, Coughlin et al. (1991) assume that

the profit earned by the firm i in location j is linked to a set of observable

characteristics of location j.

(2)

where c is the constant term, Xj is a vector of observable characteristics of

location j expressed in log term, β is a vector of unknown coefficients to be estimated

and are the random terms. Given the individual nature of the data and assuming

that the random terms are independent log-Weibull distributed, Coughlin et al. (1991)

were able to use McFadden (1974) conditional logit model that allows to express the

location choice in terms of the probability to locate in a given area, conditional to the

relative characteristics of other location, and to estimate the model using maximum

likelihood.

Coughlin and Segev (2000) outline several alternative econometric

specifications to model location choice. McFadden’s (1974) conditional logit model

provides a consistent framework for the econometric analysis only when individual

data are available. The presence of significant data constraints and the availability of

aggregated data has meant considering alternative solutions, such as the adoption of

model for count data (Coughlin and Segev, 2000). In order to bring together all these

models within a common framework of specification and estimation, we rely on the

contribution of Nelder and Wedderburn (1972) who group all these models under a

single class of models, defined as the generalised linear models (GLM), on the basis

of three common features:

1. The response variable Y has a distribution from the exponential family.

This family of statistical distributions allows heteroscedasticity related to the mean

(expected) value of the distribution. In particular, where is the

dispersion parameter.

2. A link function connects the average value of the response variable to a linear

predictor:

link function (3)

8

linear predictor (4)

where x represents the vector of explanatory variables and µ =E(Y/X) is the mean

(expected) value of the response variable. If we also assume that the response variable

is measured in terms of number of foreign owned firms entering a given location j and

the link function has a logarithmic form, then equations 3 and 4 can be combined in

the following form:

∑= =

n

iiji x

j e 1

β

µ (5)

where the average value of FDI entering a location is assumed to be an exponential

function of the following linear combination of explanatory factors.

The estimation of equation (5) within the framework of generalised linear

model (GLM) requires the specification of both the link function and the theoretical

distribution of the response variable. Given the nature of count data used in this paper

, it is reasonable to assume that the response variable Y, number of occurrences of an

event, has a Poisson distribution given the independent variables X1, X2, X3, ……., XM.

for k = 0,1,2…… (6)

The link function usually assumes the canonical form . So the

logarithm of the mean of the response variable is a linear function of independent

variables. In particular, in the Poisson distribution model the variance of response

variable is assumed to be equal to the mean. If this hypothesis is too restricted,

underdispersion or, more often, overdispersion occurs. In case of overdispersion,

quasi-likelihood estimation can be used. However it is more suitable to adopt an

alternative specification for the distribution of the response variable.

The Negative Binomial distribution provides a more general framework to

model the response variable. In particular, the variance is allowed to follows a

quadratic function of the mean, Var(y) = , where is the dispersion

parameter. The presence of overdispersion in the Poisson model can be detected on

the basis of simple statistic ratios (Scale Deviance or Pearson chi squared deviance

9

divided by degree of freedom), while the choice between Poisson and Negative

Binomial distributions can be performed by testing the null hypothesis H0: : =0 by a

likelihood ratio (LR) test.

Tests for the statistical significance of the parameters in the model can be

performed via the Wald statistic, the likelihood ratio (LR) or score statistic. The Wald

statistic is an efficient statistics for testing the significance of effects while the

likelihood ratio (LR) test requires provides the most asymptotically efficient test

known following Gourieroux et al. (1984a,b).

The goodness of fit of the model can be assessed on the basis of the scaled deviance

or the Person’s chi-square statistic. For a given value of the dispersion parameter φ ,

the scaled deviance is defined as twice the difference between the maximum

achievable log likelihood and the log likelihood at the maximum likelihood estimates

of the regression parameters. If l(y,µ ) is the log-likelihood function expressed as a

function of the predicted mean values µ and the vector y of response values, then the

scaled deviance is defined by

));();((2);( max* ylylyD µµµ −−=

where );( yl µ is the log-likelihood under the model and );( max yl µ is the log-

likelihood under the maximum achievable (saturated) model. For generalised linear

models, the scaled deviance, can be expressed as

);(1);(* µϕ

µ yDyD =

where );( µyD is the residual deviance for the model and is the sum of individual

deviance contributions. The scaled version of both deviance or the Person’s chi-

square statistics, under certain regularity conditions, has a limiting chi-square

distribution, with degrees of freedom equal to the number of observations minus the

number of parameters estimated. The two basic types of residuals used for diagnostic

checking are the so-called Pearson residuals and deviance residuals.

Inter-Industry heterogeneity

10

An obvious, though often ignored problem with this type of work is the issue

of heterogeneity across industries. In order to address this we employ the

classification scheme developed by Pavitt (1984). Starting from a sample of UK

innovating firms, Pavitt defined an empirical classification of industries according to

their technological trajectories. In particular, Pavitt’s taxonomy classifies industries as

characterised either by (i) ‘science based‘ firms; (ii) ‘production intensive‘ firms, or

(iii)‘supplier dominated‘ firms. The second group is further subdivided into the

categories of ‘scale intensive‘ production or ‘specialised suppliers‘. Each typology of

industries is characterised by different profiles in terms of level of R&D expenditure,

knowledge creation and learning behaviour, as well as firm size and main activity. In

particular, science based industries are characterised by high levels of R&D

expenditures. In addition, their knowledge creation and learning processes combine

internal and external sources (from other firms and/or universities and research

centres). In scale intensive industries, R&D expenditures may also be important, but

they are mostly realised internally and focused on process innovation and efficiency

seeking. Specialised suppliers industries are characterised by comparatively small-

and medium-sized firms, focussed on product innovation. Finally, supplier dominated

firms are considered with small innovative capabilities and little R&D oriented.

Although Pavitt’s taxonomy was defined in a rather eclectic manner, it has

been extremely influential, providing a sound basis for the comparison of firm

behaviour across industries.

In addition to addressing heterogeneity, such a classification provides a link to

the motives for FDI discussed above, following Dunning (1998), where location

choice within industries will vary by type of industry. For example, resource seeking

FDI is associated with the local endowment of assets and resources, including:

infrastructure, skilled labour and business services supply. Market Seeking FDI

includes variables connected to the exploitation of the potential of local input or

output markets, identify by the local level of income per capita and by rates of overall

and young unemployment, respectively. Efficiency seeking FDI include all factors

that affect the costs and efficiency of local production such as investment incentives,

manufacturing labour productivity and presence of static or dynamic agglomeration

economies. Finally, Strategic asset seeking FDI focuses on knowledge-related assets,

and in particular on public institutions as well as local industrial systems potentially

promoting learning processes. Clearly the nature of the industry will determine the

11

type of FDI that is attracted, which will in turn determine the location. As such,

imposing uniform coefficients across these types of industries is likely to be invalid.

Further details of the classification scheme are presented in Appendix B, while

the breakdown of inward FDI across these sectors is presented in table 1. This

classification identifies four broad typologies of manufacturing sector: science-based

industries, scale-intensive industries, specialised-suppliers industries and supplier

dominated industries.

Table 1 Distribution of firms in Italy under foreign control by Pavitt sector -1996-1999

No. foreign firms

Share in % of the total

No. foreign firms located in

metropolitan areas

Share of foreign firms in metropolitan areas over the

total of foreign firms Science-based industries 426 13.8 261 61.3

Scale-intensive industries 1324 42.8 662 50.0

Specialised suppliers industries 754 24.4 396 52.5 Supplier-dominated industries 589 19.0 208 35.3

Total 3093 100.0 1527 49.4

6 Data

The territorial unit of analysis used in this paper is the Italian province.

Provinces represent a further administrative disaggregation from the 20 Italian

standard planning regions, and thus provide a more suitable level of analysis of

industrial location. There are currently 103 such provinces in Italy. The reference

period of the analysis is 1996-1999.

Data on the number of enterprises under foreign control at the provincial level

were provided by CNEL-ICE- Politecnico di Milano. These data are stratified by

province and industry, providing unique and hitherto unexplored data on firm

location. Data used to generate the independent variables at the provincial level were

provided by ISTAT (Italian Office of National Statistics), and mostly harmonised

within the national account conceptual framework, with the relevant exception of the

provincial index of infrastructure endowment, elaborated by the Italian industrial

12

confederation. In addition, the data used to identify LISs refer to the employment in

manufacturing plants, stratified by industry (three digit level of NACE) and province,

and are provided by ISTAT, Census of Industry and Service statistics, reference year

1996.

Industry level data, including FDI and the input data to identify LISs, were

aggregated according to Pavitt’s (1984) taxonomy, described in the previous section.

In addition, an attempt to classify independent variables on the basis of their relative

importance in terms of different typologies of location choice behaviour (resource,

market, efficiency and strategic assets seeking) was also performed and included in

table 2.

A regional dummy variable, MILAN was introduced to capture specific local

phenomenon. FDI may be attracted to Milano by the high concentration of firm Head

Offices in Milan, as well as the main office of the National Stock Exchange Market.

The reference period of the analysis is 1996-1999. The list of explanatory

variables, their sources and definitions are provided in table 2. In particular, the

definitions of agglomeration-related variables are discussed at length in the next

section.

13

Table 2 Data Summary Code Definition Mean Standard

Deviation Min-Max range

Sources

FDI No. firms under foreign control, including greenfield investment and M&A (1996-1999)

30.03 94.86 0-940 CNEL-ICE- Politecnico di Milano

Breakdown of FDI data according to Pavitt’’s industry classification

FDI_RD No. firms under foreign control in science-based industries

4,14 16,05 0-154 CNEL-ICE- Politecnico di Milano

FDI_SE No. firms under foreign control in scale-intensive industries

12,85 41,67 0-411 CNEL-ICE- Politecnico di Milano

FDI_SS No..firms under foreign control in specialised-suppliers industries

7,32 24,92 0-244 CNEL-ICE- Politecnico di Milano

FDI_TS No.firms under foreign control in supplier-dominated industries

5,72 13,60 0-131

CNEL-ICE- Politecnico di Milano

Location variables related to resource seeking FDI INFRA

Infrastructure endowment, measured by a composite index of economic infrastructure endowment at the provincial level, Italy=100 (1997)

100 37.66 15.40-195.40 Italian Industrial Union

MEDU

Average level of educational attainment, measured as % of provincial population aged 6 and over with an high school diploma (1991)

17.83 2.46 12.20-26.30 ISTAT

HEDU High level of educational attainment, measured as the percent of graduate by province (1991)

3.39 0.87 1.54-6.96 ISTAT

SOCENV

Risk of the social environment, measured as the no. extortions per 100.000 units of the resident population (1996)

6.52 6.83 1.16-50.60 ISTAT

LABENV Degree of economic conflict in the local labour market, measured as the no. labour disputes per 100.000 units of the resident labour-force (1996)

193.46 190.68 8.16-1268 ISTAT

SERV

Supply of real business services, measured by the employment in non-financial business services (K section of NACE) as % of the local labour force (1996)

9.25 2.07 5.67-17.37 ISTAT

Location variables related to market seeking FDI DEM Potential market share, measure as the ratio of total

personal income relative to manufacturing employee in the province (Wheat, 1986 and Duffy, 1994). Not gravity adjusted (1996)

218.03 104.56 89.09-593.87 ISTAT

UNEMP Unemployment rate (1996) 11.21 7.29 2.52-31.83 ISTAT YUNMP Unemployment rate for young people, aged 14-24 , 1996 17.90 6.87 5.44-41.11 ISTAT Location variables related to efficiency seeking FDI PROD Manufacturing labour productivity, measured as value

added (in euro) per labour unit in manufacturing production (1996)

41552 6390 25784-71575 ISTAT

AGGL

Manufacturing base, measured by the employment in manufacturing (D section of NACE) as % of total employment (1996)

22.90 10.44 7.14-46.41 ISTAT

SOUTH Dummy variable that detects provinces with potential national or EU investment incentives (all Southern provinces, the central provinces of Lazio and provinces that received financial benefits from the “Cassa del Mezzogiorno” )

Administrative data

Location variables related strategic asset seeking FDI METROPOL Dummy variable for provinces that include large metropolitan areas, defined by ISTAT (11 areas) ISTAT LIS Dummy variable that detect LIS by industry and province. In particular, LIS identifies combination of

province and sector with a LQ above 1 ISTAT

Classification variables related to different types of province with the same industry LIS Dummy variable for two types of LIS by industry and province: (a) strongly specialised LIS are

identified by a LQ above 1,5; (b) weakly-specialised LIS are identified by a LQ in the region 1-1,5 Authors’ elaboration with ISTAT census data

ID-like provinces

Dummy variable for two additional types of province in supplier-dominated or specialised-suppliers industries, characterised by a significant presence of industrial districts (ID).

Authors’ elaboration

Location variables related to the “dartboard” effect LAND Geographical extension of the province, measured in hectares ISTAT

14

Identifying different sources of agglomeration economies Despite the importance of agglomeration economies in attracting FDI there is

only limited study in the literature of the different types of agglomerative forces. The

“new economic geography” approach (Krugman 1991, Ottaviano 2003) focuses on

agglomeration effects driven by pecuniary or technological externalities and

cumulative processes. In contrast industrial and urban economics focuses on the

specific externalities generated by strongly specialised local industrial systems (LIS)

and metropolitan areas, respectively. A relatively restricted literature has developed

along these lines following Brusco (1990) and Glaeser at al (1992).

In the spirit of Coughlin et al. (1991), agglomeration economies are measured

in terms of local density of manufacturing activities, given by the share of

manufacturing employment over total employment at the provincial level. This

variable captures both statistic agglomeration economies and potential dynamic

effects, since it provides also a measure of the level of industrial development of the

local manufacturing industry. In addition, in contrast to some empirical studies that

consider the number of manufacturing firms as a measure of agglomeration

economies of the local industry, it removes the problem of spurious correlation with

the geographical size of the province.

LISs are identified at the provincial and sector level on the basis of location

quotients (LQ), calculated with respect to census of industry and service data for the

year 1996.

LQ is defined as follows:

tot

i

j

ij

ij

empempempemp

LQ =

Where Emp represents employment in local production units and i and j

denote respectively sector and province. Values of LQit ranging from 1 to 1.5 identify

weakly specialised LIS, while values above 1.5 denote strongly specialised LIS. This

approach generates a set of industry-specific dummy variables that vary across

industries, classified on the basis of Pavitt (1984) taxonomy. A specific typology of

LIS, the industrial district (ID), is also identified for the supplier-dominated and

15

specialised-suppliers industries only, on the basis of the ID-like provinces taxonomy

proposed by Becattini and Menghinello (1998). This defines a set of indicators that

measure the concentration of IDs at the provincial level and across all industries.

Following Coughlin and Segev (2000), urban areas are identified at the provincial

level on the basis of the inclusion of metropolitan areas, officially defined by ISTAT.

7. Results

The results from the estimation of equation (5) are presented in Table 3. The

presence of a high degree of correlation among explanatory variables is one of the

major problems in regional models, especially when social and economic covariates

are combined. In addition, location choice models require controlling for the

“dartboard effects”2..

In order to identify a set of independent covariates, a process of a priori

variable selection was performed. The essential problem was one of correlation

between two sets of variables, the education and unemployment variables.

Specifically HEDU was found to be highly correlated with SERV due to the fact that

business services require higher standard of education than manufacturing.

Furthermore, AGGL is strongly and negatively correlated with UNEMP, DEM and,

indirectly, with LABENV; a reason for this could be that the intensity of

manufacturing activity is also a good index of local development.

As a result, only a limited set of not strongly correlated variables enter the

model: SERV, AGGLOM, INFRA and PROD. In particular, the variable SERV

captures the concentration of business services, as well as the local standard of high

education. The AGGL variable measures the intensity of agglomeration economies

derived from the concentration of manufacturing activities at the local level; it also

captures negative conditions demand for final products and labour, as well as a low

conflict rate in the local labour market. The territorial extension of each province

(measured in hectares) was used in the model to control for potential “dartboard

effects”.

The hypothesis of the variable having a Poisson distribution was rejected in

favour of a Negative Binomial by standard LR test on the dispersion parameter. . 2 The "dartboard" effect in location model occurs as a result of variations in size of region. While it is possible to address this by using normalised or relative measures, a strong correlation between these variables and the geographical size of the region may occur. It is important therefore to control for these effect by including a log variable that captures the size of each region.

16

Table 3. Negative Binomial Regression Results by Pavitt Industries (1996-1999)

Parameter

All industries

Science-based industries

Scale- intensive industries

Specialised- suppliers industries

Supplier-dominated sectors industries

Intercept

-17.015 (483.52)***

-22. 179 (140.96)***

-17.187 (203.42)***

-16.726 (155.49) ***

-18.125 (109.48)***

AGGL

0.068 (71.67)***

0.103 (32.71)***

0.065 (35.40)***

0.066 (26.06)***

0.065 (26.30)***

SERV

0.145 (8.22)*** 0.217 (4.05)** 0.109 (2.40) 0.169 (5.94)** 0.109

(2.25)

INFRA

0.016 (25.93)*** 0.011 (2.53) 0.018

(16.57)*** 0.019

(14.16)*** 0.011

(4.01)**

PROD

0.000 (16.31)***

0.000 (5.68)**

0.000 (3.67)**

0.000 (0.01)

0.000 (6.95)***

MILANO

1.408 (7.89)***

0.597 (0.44)

1.607 (4.96)**

1.442 (5.18)**

1.63 (6.17)**

METROPOL -0.232 (0.71) -0.470 (0.56) -0.284 (0.49) 0.004 (0.00) -0.459

(0.92)

Industry-specific LIS 1.450 (15.35)***

0.340 (2.51)

0.561 (5.43)**

0.259 (0.83)

Main Typologies of LIS

De-specialised provinces -1.842 (11.56)*** -0.758

(6.81)***

Weakly specialised provinces -0.540 (1.04) -0.314 (1.44)

Strongly specialised provinces 0 0

Typologies of LIS and ID

De-specialised and non ID intensive provinces -0.908 (4.15)**

Weakly specialised and non ID intensive provinces -0.440 (1.66)

Strongly specialised and non ID intensive provinces -0.225 (0.29)

Low-intensive ID provinces -0.576 (2.74)

High-intensive ID provinces 0

Dispersion 0.181 0.540 0.377 0.258 0.24

Log Likelihood 10762.792 898.283 3562.815 1808.383 953.390

Scaled Deviance 68.189 67.400 74.109 85.204 87.480

Scaled Pearson Chi-Square 71.095 67.789 70.640 73.616 77.591

DF 59 58 58 58 58

17

Sample size 66 66 66 66 66

The results of the first set of regressions show that agglomeration economies,

calculated in terms of local concentration of manufacturing activities, have a positive

impact in attracting FDI across all Pavitt industries and this is particularly true for

science-based industries. The local offer of business services, which is strongly

correlated with high levels of education, positively affect FDI location in science-

based industries, and in specialised suppliers industries, but is below the national

average and not statistically significant for the other two types of industry. This result

seems to be consistent with the argument that these two industries rely on a network

of specialised business service firms and require a highly skilled local labour market.

Infrastructure endowment is relatively more important for the localisation of foreign

firms in the specialised-suppliers industries and scale-intensive industries, while it is

below the national average for supplier-dominated and science-based industries; in

particular, it does not appear to be statistically significant. The relevance of the

infrastructure endowment for specialised-suppliers and scale-intensive industries is

consistent with the complex nature of these productions that require the minimisation

of transportation costs and the adoption of optimal logistic solutions. Productivity is

positive and statistically significant across all sectors, except for specialised-suppliers

industries. The dummy variable MILANO is positive and statistically significant for

all sectors, except for science-based industries. The METROPOL variable is mainly

negative but not statistically significant across all sectors, thus showing that urban

economies are not relevant for FDI location, besides the effects accounted for

infrastructure endowment and the local offer of business services..

Industry-specific LIS effects are strongly positive and statistically significant

for science-based and specialised-suppliers industries. This illustrates that the LIS

effect is distinct from the agglomeration and urban economies effects. One possible

explanation for this is that LIS effects are associated with technology sourcing FDI.

In science-based industries, foreign firms would benefit from the externalities

generated in specialised LISs through their engagement in formal and informal

linkages with local high-tech firms or institutions. On the other hand, in specialised-

suppliers industries, foreign firms are very likely to benefit from location spillovers

through learning-by-interacting processes, mainly realised via user-producer linkages

with other local firms along the local production filière. In contrast, the insignificance

18

of LIS effects in supplier-dominated sectors may be explained by the fact that LISs in

these industries present entry barriers, such that entry by FDI is viewed as very risky

and possibly unprofitable..

The second set of results refers to the link between FDI and types of LISs. The

analysis of the impact of different types of LIS in attracting FDI within the same

industry provides some interesting insights. The analysis is carried out by considering

one type of LIS as a “benchmark” that is set equal to zero: in our case this is

represented by “strongly specialised provinces”. In both science-based and

specialised-suppliers industries, weakly specialised provinces and, in particular, de-

specialised provinces have a relatively negative impacts on FDI attraction at the local

level: this strengthens previous evidence suggesting the key role of specialised LIS as

local catalysts of FDI.

In the last set of regressions, we extend the analysis to include industrial

district as a specific type of LIS. We find that in specialised-suppliers industries

highly intensive ID provinces overperform types of LIS in terms of FDI attraction .

6. Policy implications and concluding remarks

The potential for agglomeration economies appears to be important for MNEs

comparing location, and this is particularly important for science-based industries. In

contrast, urban economies are not important for the attraction of FDI, unless they are

specific to a particular urban centre, i.e. Milan. Focussing on the LIS effect, FDI in

science-based and specialised industries is attracted to strongly specialised provinces,

while industrial districts attract specialised-suppliers industries. Finally, FDI is not

attracted by industrial districts in supplier-dominated industries or in scale-intensive

industries. Such, industries are often very competitive in the world markets,

demonstrating high degrees learning and innovation, so in Dunnings (1979)

terminology, potential inward investors possess no ownership advantages here in

order to encourage FDI..

These findings have important policy implications for the attraction and

retention of FDI. On the one hand, one of the main competitive advantages of LISs is

the intangibility of the learning processes and innovation, with this varying across

industries according to their reliance on tacit and/or codified knowledge. Our results

seem to suggest that FDI is attracted to LISs in industries where knowledge is mostly

codified and, therefore, more easily transferable from the host to the foreign firms.

19

This happens through foreign firms’ engagement in R&D activities with local firms

and institutions (science-based industries) and where inter-firm relations are

structured in user-producer linkages. An important mechanism here is vertical

cooperation creating a channel for knowledge transfer, particularly in specialised-

suppliers industries. It appears that those industries where tacit knowledge constitutes

an important component of their embedded knowledge and learning are less attractive

for FDI, in that the tacitness of knowledge and the embedded nature of firms’

interaction act as barriers to entry.

While the contribution of FDI to local development per se is beyond the scope

of this paper, the results here provide an important distinction when seeking to

evaluate the potential impacts of FDI. It is important to determine whether potential

inward investors will seek to become embeded locally or adopt predatory behaviours.

A MNE is locally embedded when it promotes cooperation and development of trust

with local firms, as well as it is committed to a locality with long-term investments in

both physical and human capital. This decision is determined by the analysis of the

firm concerning the potential gains from long-term cooperation compared with the

short term gain from predatory behaviour (Bellandi, 2001). The strengthening of the

degree of local or regional embeddedness of the MNE may produce strong benefits

for the local industry and community, although the long-terms effects on local

governance should be carefully evaluated.

An important question for future research therefore concerns the distinction

between the short and long run in this respect. Drawing on the results of our paper, it

can be argued that it questionable whether FDI in science-based and specialised-

suppliers industries are likely to be associated with long-term commitments, while the

lack of foreign firms’ interest in supplier-dominated sectors can also be due to the

fact that the appropriation of benefits would entail monetary and non-monetary

investments.

To conclude, we would suggest that policy-makers haveto be aware of the

need to strike a balance between attracting FDI, without falling into the trap of

adopting short-term measures by targeting and opening up sectors palatable to foreign

investors, and ensuring that foreign firms commit themselves to localities since only

embedded FDI benefits both foreign and host firms.

20

Appendix A

Pavitt’s classification of manufacturing industries (1992) Pavitt’s ’sectors NACE three digits codes Science-based industries

244, 300, 321, 322, 323, 331, 332, 333, 334, 335, 353

Scale-intensive Industries

211, 212, 221, 222, 223, 231, 232, 233, 241, 242, 243, 245, 246, 247, 251, 252, 261, 264, 265, 266, 267, 268, 271, 272, 273, 274, 275, 283, 284, 285, 296, 297, 341, 342, 343, 354

Specialised Suppliers Industries

291, 292, 293, 294, 295, 311, 312, 313, 314, 316, 351, 352, 355

Supplier-dominated Industries

151, 152, 153, 154, 155, 156, 157, 158, 159, 160, 171, 172, 173, 174, 175, 176, 177, 181, 182, 183, 191, 192, 193, 201, 202, 203, 204, 205, 262, 263, 281, 282, 286, 287, 315, 361, 362, 363, 364, 365, 366, 371, 372

21

References

Aydalot, P. (1986) Milieux innovateurs en Europe, Paris, Groupe de Recherche Européen sur les Milieux Innovateurs.

Basile, R. Castellani, D. and Zanfei, A. (2003) Location choices of multinational firms in Europe: the role of national boundaries and EU policy, Working Paper Series on Economics, Mathematics and Statistics, N.78, University of Urbino, Italy.

Basile, R. (2002) The Locational Determinants of Foreign-Owned Manufacturing Plants in Italy: Preliminary Results, Rivista di Politica economica, July-August.

Becattini G. (1979) Dal “Settore” Industriale Al “Distretto” Industriale. Alcune Considerazioni Sull’unità Di Indagine Dell’economia Industriale, Rivista Di Economia E Politica Industriale, (1).

Bellandi M. (2001) Local development and embedded large firms, Entrepreneurship & Regional development, (13): 189-210.

Bennett R, McCoshan A and Coulson. A (1993) Enterprise and human resource development - local capacity building, Local government studies, 19 (4), p.599.

Best M. (1990) The new competition, Cambridge, Policy Press.

Brusco S. (1990) The idea of the industrial district. Its genesis, in Pyke F., Becattini G. and Sengenberger W. (Eds.) Industrial districts and inter-firm co-operation in Italy, International Institute for Labour Studies, Geneva, pp.10-19.

Camagni, R.P. , Ed. , (1991) Innovation networks - spatial perspectives, Belhaven Press, London and New York, 247p.

Camagni, R.P. (1995) ‘The concept of innovative milieu and its relevance for public policies in European lagging regions’, Papers in regional science,74 (4), pp. 317-40

Camagni, R.P. (1999) “The city as a milieu: applying the GREMI approach to urban evolution”, Revue d'économie régionale et urbaine, 3: 591-606

Cantwell, J. A. (1991) The international agglomeration of R&D, In M.C. Casson, (Ed.), Global research strategy and international competitiveness, Oxford, Blackwell.

Cantwell, John .A. (1991) ‘The international agglomeration of R&D.’ In M.C.Casson, editor, Global research strategy and international competitiveness. Oxford: Blackwell.

Cantwell, J.A. and Janne, O. (1999) ‘Technological globalisation and innovative centres: The role of corporate technological leadership and location hierarchy’. Research Policy, Vol. 28, pp119-144.

22

Cantwell, J and S. Iammarino (2000) ‘Multinational corporations and the location of technological innovation in the UK regions’. Regional Studies, 34(4), pp. 317-332.

Cantwell and Santangelo, GD (1999) ‘The frontier of international technology networks: sourcing abroad the most highly tacit capabilities’, Information economics and policy, 11 (1): 101-123

Carlton, D.W. (1983) ‘The location and employment choices of new firms: an econometric model with discrete and continuous endogenous variables’, Review of Economics and Statistics, 65(3): 440-449

Chung, W. and Alcàcer, J (2002) Knowledge Seeking and Location Choice of Foreign Direct Investment in the United States, Management Science, 48 (12): 1534-54

Coughlin, C et al. (1991) State Characteristics and the Location of Foreign Direct Investment within the United States, The Review of Economics and Statistics, 73(4): 675-683.

Coughlin, C.C. and Segev, E. (2000) Location determinants of new foreign-owned manufacturing plants, Journal of Regional Science, 40 (2): 323-51

Crozet, M. Mayer, T. and Mucchielli, J. (2003) 'How do Firms Agglomerate? A Study of FDI in France', CEPR Discussion Paper no. 3873. London, Centre for Economic Policy Research

De Propris L and Driffield N (2003) FDI, Clusters and Knowledge Sourcing, Birmingham Business School Working Paper 2003-14.

De Propris L, Driffield N. and Menghinello S (2003) Localised Competitive Advantages and Local Industrial Systems in Italy, Birmingham Business School Working Paper 2003-29.

Devereux, Griffith and Simpson (2004) "The Geographic Distribution of Production Activity in the UK"(2003) Regional Science and Urban Economics, forthcoming

Dicken, P. (1998) Global Shifts: Transforming the World Economy, London, Paul Chapman Publishing.

Dosi, G. and Malerba, F., (1996) Organisation And Strategies In The Evolution Of The Enterprise, London, Macmillan.

Driffield, N.L. and Love, J.H. ‘FDI, (2003) Technology Sourcing and Reverse Spillovers.’ The Manchester School, vol. 71 (6) pp. 659-672.

Driffield, N.L. and Love, J.H. (2003a) Does The Motivation for Foreign Direct Investment Affect Productivity Spillovers to The Domestic Sector? Available at: http://research.abs.aston.ac.uk/working_papers/0201.pdf

23

Driffield, N.L. and Munday, M.C. (2001) ‘Technical Efficiency in UK Industries: A Stochastic Production Frontier Approach.’ Regional Studies, Vol. 35(5) pp. 391-99.

Dunning, J.H. (1998) Location and the multinational enterprise: a neglected factor?”, Journal of International Business Studies, 29 (1)

Fosfuri, A. and Motta, M. (1999) Multinationals without advantages, Scandinavian Journal of Economics, vol. 101, 617-30.

Friedman et al. (1992) What attracts foreign multinational corporations? Evidence from branch plant location in the United States, Journal of Regional Science,32 (4): 403-18.

Friedman, J. et al. (1996) A Note On “State Characteristics and the Location of Foreign Direct Investment within the United States”, The Review of Economics and Statistics, 78 (2): 367-368.

Glaeser, E. et al. (1992) Growth in Cities, Journal of Political Economy, 100 (6): 126-1152

Görg, H. & E. Ströbl (2001) Multinational companies and productivity spillovers: a meta-analysis, Economic Journal, vol. 111, pp. F723-39.

Gourieroux, C., Monfort, A. and Trognon, A., (1984a) Pseudo maximum likelihood methods : Theory. Econometrica, 52 , 681-700.

Gourieroux, C., Monfort, A. and Trognon, A., (1984b) Pseudo maximum likelihood methods :applications to Poisson models. Econometrica, 52 , 701-720.

Hakansson, H. (1987). Industrial Technological Development: A Network Approach, London, Croom Helm.

Head, K., Ries, J. and Swenson D. (1995) Agglomeration benefits and location choice: Evidence from Japanese manufacturing investments in the United States, Journal of International Economics, vol. 38, 223-247.

Hines, J.R.jr (1996) Altered States: Taxes and the Location of Foreign Direct Investment in America, The American Economic Review, 86 (5): 1076-1094.

Humphrey, J and Schmitz, H (1996) ‘The Triple C Approach to Local Industrial Policy’, World Development 24 (12): 1859-1877.

Kogut, B. and Chang, S. J. (1991) Technological capabilities and Japanese foreign direct investment in the United States, Review of Economics and Statistics, vol. 73, 401-13.

Krugman, P. (1991) ‘Increasing returns and economic geography’, Journal of political economy, 99(3): 483-499

24

Kuemmerle, W. (1999), ‘Foreign direct investment in industrial research in the pharmaceuticals and electronics industries – results of a survey of multinational firms’, Research Policy, Vol. 28, pp. 179-93.

Lawson, C. and Lorenz, E. (1999) Collective learning, tacit knowledge and regional innovative capacity, Regional studies, 33 (4): 305-17

Makino, S. et al. (2002) Asset-Exploitation Versus Asset-Seeking: Implications for Location Choice of Foreign Direct Investment from Newly Industrialised Economies, Journal of International Business Studies, 33 (3): 403-421

Maskell P et al. (1998) Competitiveness, localised learning and regional development: specialisation and prosperity in small open economies, Environment and planning, 16, (4), 503.

McFadden, D. (1974) ‘Conditional Logit Analysis of Qualitative Choice Behavior,’ in P.Zarembka, ed. Frontiers in Econometrics, pp105-142, New York: Academic Press.

Nelder, J.A. and Wedderburn R.W.M. (1972) ‘Generalized linear models’, Journal of the Royal Statistical Society, A 135(3): 370-384.

Neven, D. and Siotis, G. (1993) Foreign Direct Investment in the European Community: Some Policy Issues. Oxford Review of Economic Policy, Vol. 9 (2), pp 72-93.

Nonaka, I. (1991) ‘The knowledge-creating company’, Harvard business review, 69 (6) p.96-

Pavitt, K. (1984) Sectoral Patterns of Technical Change: Towards a Taxonomy and a Theory, Research Policy 13, 343-373.

Pearce, R. D. (1999) ‘Decentralised R&D and strategic competitiveness: globalised approaches to generation and use of technology in multinational enterprises (MNEs)’ Research Policy, Vol. 28, pp. 157-78.

PERRIN J.-C. (1988) Nouvelles technologies et développement régional: l'analyse des milieux innovateurs. Aix-en-Provence, CER, Research paper no 94-95.

Piore, M.J. and Sabel, C.F. (1984) The Second Industrial Divide, Basic Books, New York.

Mariotti, S. – Piscitello, L. (1994) ‘Le determinanti dei differenziali di attrazione territoriale degli investimenti diretti esteri’, L’Industria, n. 2.

Porter M. (1998) Clusters and The New Economics of Competition, The Harvard Business Review, Nov-Dec.

Prahalad, C. and Hamel G. (1990) ‘The Core Competence Of Corporations’, Harvard Business Review, 68 (3).

25

Putnam R. (1993) Making Democracy Work, Princeton, Princeton University Press.

Scott, A.J. and Storper, M. (1992) ‘Regional development reconsidered’ in Regional development and contemporary industrial response - extending flexible specialisation, Belhaven Press, London and New York, pp.3-24

Sforzi, F. (1990). ‘The Quantitative Importance Of Marshallian Local industrial system In The Italian Economy’ In Pyke F. - Becattini G. - Sengenberger W. (ed.), Local industrial system and Inter-Firm Co-Operation In Italy, ILO, Ginevra.

Shaver, Myles J. 1998. Do foreign owned and US owned establishments exhibit the same location pattern in US manufacturing industries? Journal of International Business Studies. 29 (3): 469-492.

Siotis, G. (1999) Foreign direct investment strategies and firms’ capabilities, Journal of Economics and Management Strategy, vol.8, 251-70

Storper, M. (1995) The resurgence of regional economies, ten years later, European Urban and Regional Studies, 2: 191-221.

Togo, K. and Arikawa, Y. (2002) Agglomeration Effects versus Policy Effects: The case of the Electronics Industry in Malaysia, Asian Economic Journal, 16 (3): 229-245.

UNCTAD (2001) World Investment Report 2001: Promoting Linkages. United Nations.

UNCTAD (2003) World Investment Report 2003. United Nations.

Van Pottelsberghe de la Potterie B. and Lichtenberg, F. (2001) ‘Does Foreign Direct Investment Transfer Technology Across Borders?’, The Review of Economics & Statistics, 83 (3): 490-497.

Vernon, R. (1966) ‘International investment and international trade in the product cycle’, Quarterly Journal of Economics, Vol.80, pp.190-207.

Wheeler, David. & Ashoka Mody. 1992. International investment location decisions: The case of US firms. Journal of International Economics, 33: 57-76.

Whitley, R. Ed (1992) European business systems - firms and markets in their national contexts, SAGE Publications, London, 291p.

Woodward, D. (1992) “Locational Determinants of Japanese Manufacturing Start-ups in the United-States” Southern Economic Journal, Vol. 58 Issue 3, pp. 690-709

World Bank (2001) World Development Report: Building institutions for markets, World Bank, Washington DC.

26

Zanfei, A. (2000) Transnational Firms and the Changing Organisation of Innovative Activities.. Cambridge Journal of Economics. Vol. 24 (5). p 515-42

27