Journal of Urban Economics 47, 115᎐135 Ž2000. Article ID juec.1999.2138, available online at http:rrwww.idealibrary.com on
Agglomeration and the Location of Foreign Direct Investment in Portugal1 Paulo Guimaraes ˜ Escola de Economia e Gestao, ˜ Uni¨ ersidade do Minho and CEMPRE, Braga, Portugal
Octavio ´ Figueiredo Faculdade de Economia do Porto, Uni¨ ersidade do Porto and CEMPRE, Porto, Portugal
and Douglas Woodward2 Department of Economics, The Darla Moore School of Business, Uni¨ ersity of South Carolina, Columbia, South Carolina E-mail:
[email protected] Received September 22, 1998; revised May 3, 1999 In urban economics, and more recently in the international economics literature, agglomeration has been offered as a principal determinant of new investment. Yet agglomeration has rarely been subject to precise statistical tests. In this paper, the availability of detailed urban and regional data for Portugal allowed for a close study of the spatial choices for newly created foreign-owned plants. It appears that agglomeration economies are decisive location factors. Service agglomeration has a notably strong effect, while industry-level localization economies and urbanization externalities are verifiable location determinants as well. Distance from the principal cities is statistically significant, but there is no evidence that local labor costs matter. 䊚 2000 Academic Press
I. INTRODUCTION Agglomeration, the spatial externality concept advanced by Alfred Marshall in the 1890s, endures 100 years later as a fundamental explanation of urban growth, productivity, and investment. In a well-known passage from 1 The authors express their appreciation for the support of JNICT, Lisbon, Portugal. An earlier version of this paper was presented at the International Regional Science Association, 37th European Congress, Rome, Italy, August 1997. 2 Author to whom correspondence should be sent.
115 0094-1190r00 $35.00 Copyright 䊚 2000 by Academic Press All rights of reproduction in any form reserved.
116
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
The Principles of Economics, Alfred Marshall w30, pp. 350᎐351x wrote: When an industry has thus chosen a locality for itself, it is likely to stay there long: so great are the advantages which people following the same skilled trade get from near neighbourhood to one another . . . Employers are apt to resort to any place where they are likely to find a good choice of workers with the special skill which they require . . . The advantages of variety of employment are combined with those of localized industries in some of our manufacturing towns, and this is a chief cause of their continued growth.
In Marshall’s view, agglomeration engenders economies that are external to a firm, but internal to a small geographic areaᎏa ‘‘locality.’’ Today, these external economies are known to encompass specialized labor markets and supplier networks as well as knowledge spillovers. In the early 1990s, a ‘‘new economic geography’’ emerged, emphasizing the role of external economies in explaining economic growth ŽKrugman w26, 27x; Porter w38x.. Some writers, known for their work on international trade, claimed to bring more advanced models to urban economics and regional science Žfor critiques, see Isserman w24x and Martin w31x.. Yet a decade earlier, the urban literature had anticipated the theoretical and empirical rediscovery of agglomeration. Important contributions explored spatial external economies through shifts in production functions, assessing the relative impact of localization and urbanization economies on city size and other characteristics ŽHenderson w21, 22x; Nakamura w36x.. Related work in the 1970s considered whether production is more efficient in cities of different sizes ŽSegal w41x; Sveikauskas w43x.. Indeed, agglomeration economies still figured prominently in urban economics ŽDavid and Rosenbloom w12x; Calem and Carlino w7x., even as the ‘‘new economic geography’’ asserted new theoretical claims about its importance in understanding urban systems. Rivera-Batiz’s w40x contribution to the theory of urban service agglomeration, which appeared before the ‘‘new’’ literature, is especially germane to this paper. That is not to say that there is nothing new to say about agglomeration in the 1990s. Economists continue to posit interesting findings, for example, that knowledge spillovers stemming from urban diversity affect urban growth more than industry specialization ŽGlaeser et al. w15x.. We have much more compelling evidence that urban variety, not simply the localization of particular industries, drives economic growth. Thus, Marshall’s legacy spreads as a wider body of research considers agglomeration questions that have been the traditional concern of urban economics. The interest among international economists is curious, given that external economies do not often arise at the international level; they are fundamentally an interregional and interurban phenomenon, prevalent within and across cities, towns, and localitiesᎏrelatively small spaces compared with the larger boundaries covered by a nation-state. The
AGGLOMERATION AND LOCATION IN PORTUGAL
117
centripetal forces of industry attraction are particularly germane to narrowly defined regions or urban locations. Indeed, the Marshallian concept of agglomeration is that of a ‘‘near neighbourhood.’’ While agglomeration economies continue to figure prominently in urban economic theory and empirical research, the concept has not often been subject to precise statistical analysis in the context of industrial location research, the focus of this study. In much of the empirical literature, agglomeration measures are crude; for example, total manufacturing employment is used as a proxy for agglomeration economies that should be at least in part industry-specific ŽCoughlin et al. w10x; Woodward w45x.. A notable exception is Carlton w8x, who made one of the first attempts to model location selection using discrete choice models. Carlton tested the probability of establishing a branch plant in three narrowly defined industries across U.S. metropolitan areas and found that industry-specific agglomeration economies were a statistically significant location determinant. Following a similar methodology, Luger and Shetty w28x tested agglomeration economies at the three-digit SIC level and corroborated the effect on foreign firm location decisions. The study by Smith and Florida w42x also stands out because the paper sharpened the focus, in this case testing whether agglomeration exerted a pull on Japanese automotive manufacturing plant openings among U.S. counties. The results provided evidence that Japanese automotive assemblers preferred the close proximity of suppliers to exploit external spatial economies and the just-in-time inventory system. Typically, empirical research covers a much wider territory. For example, in another study of Japanese-owned plants industry-level agglomeration was tested across U.S. states ŽHead et al. w18x.. U.S. states are large areas that stretch the Marshallian concept of agglomeration. Other empirical research on foreign firm location and agglomeration economies has been based on even more highly aggregated international data ŽBraunerhjelm and Svensson w6x; Micossi and Viesti w35x; Wheeler and Mody w44x.. This paper investigates the location decisions of foreign-owned manufacturing plants in the urban areas and outlying regions of Portugal. In a discrete choice framework, the location choices of newly established foreign-owned plants are tested against a set of industry-specific employment variables and other regional characteristics. The availability of data for small localities in Portugal, coupled with detailed plant establishment information by industry, allowed us to refine our tests to capture the influence of different types of external economies and compare them with other determinants of location. At the outset, a short review of foreign direct investment ŽFDI. in Portugal will prove helpful. Portugal experienced a rapidly increasing inflow of FDI following its 1986 entry into the European Community ŽEC..
118
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
Relative to the size of the economy, these flows placed the country as one of the largest recipients of FDI in Europe. During the 1986᎐1992 period, Portugal received a total of U.S. $12.6 billion. On an average annual basis, this represented 3.0% of Portuguese GDP and 11.4% of gross fixed capital formation. In comparison, the averages for all European Community countries during this period were much lower: 1.6% of GDP and 5.6% of gross fixed capital formation ŽBanco de Portugal w2x.. According to these figures, Portugal received the largest share of inward FDI during this period. A significant share of the FDI inflow was directed to the manufacturing sector, mainly in chemicals, mechanical and electrical machinery and equipment, transportation material, and the food industry.3 An important question concerns the spatial allocation of greenfield Žnew plant. investment. To date, there has never been a location analysis of plant openings by foreign-owned companies in Portugal. Indeed, there is a dearth of empirical location research on foreign direct investment within European countries in general. With a few exceptions ŽHill and Munday w23x; Mariotti and Piscitello w29x., most studies of FDI and regional location have been conducted for the United States ŽLuger and Shetty w28x; Coughlin et al. w10x; Woodward w45x; Smith and Florida w42x; Head et al. w18x; Friedman et al. w14x.. The recent location research in the United States has used micro Žplant-level. data, following Carlton w8x and Bartik’s w4x approach to domestic branch plant location. While there is an emerging consensus on econometric techniques for location studiesᎏnamely, discrete choice modelsᎏseveral difficulties have hampered research thus far. The main obstacle remains data. Proper modeling of plant location decisions requires information on each new investment. At the same time, the modeling requires highly disaggregated industrial and spatial data to account for the specific agglomeration influences and other location factors. Reliable data sets on new plants and disaggregated spatial information are rare. Another problem with previous research on FDI and location has been the lack of information on the entry mode of the new investment. Some studies ŽCoughlin et al. w10x; Mariotti and Piscitello w29x. mix greenfield plant investment with other types of FDI such as mergers, acquisitions, joint ventures, and plant expansions. Greenfield plants clearly require an explicit location decision, while other types of investment may not. Moreover, almost all studies have modeled location choices among highly aggregated regions.4 3
During this period the manufacturing sector accounted for 20.2% of total FDI ŽBanco de Portugal w2x.. 4 Most of the U.S. studies have considered choices among states, a large Žbut varied. geographic unit that encompasses significant differences in regional conditions within the state.
AGGLOMERATION AND LOCATION IN PORTUGAL
119
However, agglomeration economies, labor market conditions, and other factors usually identified as relevant for location decisions apply at a local level. They cannot be adequately accounted for when the models consider only large regions.5 Hence, this study contributes to the existing literature on agglomeration, location, and FDI in two crucial ways. First, it uses precise information at the individual plant level, considering only greenfield plants. Second, the plant location decisions are considered at the level of the concelho, a fairly small administrative region.6 The spatial data, coupled with detailed investment by industry, enable the model to pick up specific agglomeration influences not tested in previous work. The rest of the paper is organized as follows. The next section provides details on the database and presents the econometric methodology. In the third section we discuss the potential determinants of location in Portugal’s regions. We present the empirical results in the fourth section. The final section summarizes the main findings of the paper. II. DATA AND METHODOLOGY Identification of Foreign-Owned Greenfield Plants For over a decade the Ministry of Employment and Social Security of Portugal through the Department of Statistics ŽDEMESS. has collected annual data on all firms operating in Portugal ŽQuadros do Pessoal..7 The DEMESS records consist of firm- and establishment-level information, with special emphasis on the characteristics of the labor force. To our knowledge these data have not been used for studies related to the location of economic activities, even though they include detailed information on plant location at the concelho level. We identified all newly created establishments between March 1985 and March 1992, following a methodology similar to that of Mata et al. w32x. This was possible because DEMESS provides a unique identifier for a firm and its establishments which remains constant through the years.8 By merging the data for all available years Ž1982 to 1992. we identified all newly created establishments in this period. An establishment was identified as new if that was the first time it appeared in the merged data set.9 Woodward w45x was one of the first studies to address this question. The Portuguese concelhos have an average area of 322.5 km2 . 7 This database covers all manufacturing companies operating in Portugal, except family businesses without wage-earning employees. 8 However, in 1991 the DEMESS changed its identifier for establishments. Special care was taken by visually inspecting the merged data for this year. 9 We excluded from our analysis those establishments created before March 1985. First, this was intended to obviate a well-known problem of temporary exits from the merged database ŽMata et al. w32x.. Second, it was intended to lessen the problem of time coherence between some of the explanatory variables and our dependent variables. 5 6
120
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
FIG. 1. Spatial distribution of FDI start-ups in Portuguese concelhos, March 1985 to March 1992. Note: Each dot s 1 unit. Data source: D.E.M.E.S.S., ‘‘Quadros do Pessoal.’’
AGGLOMERATION AND LOCATION IN PORTUGAL
121
Of these, we identified 758 newly created establishments that were totally or partially participated in by foreign capital. The spatial distribution of these greenfield plants is shown in Fig. 1. As can be seen, these investments are concentrated on the more urban coastal side of the country, especially around the largest cities of Lisbon and Porto. Methodology Once a foreign investor decides to open a manufacturing plant in Portugal, he or she is faced with a set of 275 spatial choices consisting of all existing concelhos.10 Following McFadden w33x we consider that the investor i, if it locates in concelho j, will derive a profit of i j . This latter value is composed of a deterministic and a stochastic term. More formally,
i j s Ui j q i j ,
Ž 1.
where i j stands for a random variable. Alternative j will be preferred by investor i if
i j ) ik ,
᭙k, k / j.
Ž 2.
The stochastic nature of the profit function implies that the probability that location j is selected by decision maker i equals Pi j s Prob Ž i j ) i k . ,
᭙k, k / j.
Ž 3.
Additionally, if we assume the error terms to be distributed independently and according to a Weibull distribution, we can rewrite the probability of locating at j as Pi j s
exp Ž Ui j . 275
Ý
.
Ž 4.
exp Ž Ui k .
ks1
The above equation expresses the conditional logit formulation.11 If we further assume that the systematic part of profit is affected by a set of m regressors, we can estimate the effects these have on location decisions. Typically, it is assumed that Ui j is a linear combination of the explanatory variables, Ui j s  1 X i1j q  2 X i2j q ⭈⭈⭈ qm X imj .
Ž 5.
10 We exclude from the analysis the islands of the Azores and Madeira. The number of new foreign investments in these islands is quite small. 11 The high number of choices precludes the use of the multinomial probit model.
122
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
The estimation of the above model is complicated by the magnitude of the choice set.12 Fortunately, McFadden w34x has established that consistent estimators of the parameters could still be obtained from a random sample of choices taken from the full choice set. Likewise, we consider that the choice set for individual i consists of its actual choice and 39 other randomly selected choices. In the next section, we present the independent variables that are hypothesized to explain the choices. III. LOCATION FACTORS As the previous section explained, the probability that a foreign firm locates in a particular concelho depends on how the characteristics of that concelho affect profits relative to the characteristics of all other concelhos. Therefore, the explanatory variables to be considered are those specific to the concelho believed to have an impact on the expected profits of the firm. These local characteristics can affect profits from both the cost and revenue sides. Table 1 shows a list of the independent variables and their expected signs. An important factor affecting the location decision, and the focus of this research, is the existence of agglomeration economies. From a theoretical point of view, agglomeration is justified once space is introduced into the framework of the traditional profit model. In general, agglomeration economies consist simply of those positive externalities resulting from the spatial concentration of existing economic activity. As mentioned earlier, agglomeration economies result from industry-specific localization, obtained when firms in the same industry draw on a shared pool of skilled labor and specialized input suppliers; more general economies are garnered by firms across all industries. Essentially, there are two major types of externalities. First, localization economies are realized through the size of a particular industry in an area. Additionally, service agglomeration and urbanization economies Žsometimes measured by total manufacturing activity or the population size of the area. potentially bolster productivity and attract more firms to a locality. In the past, different types of agglomeration economies were sometimes lumped together and measured using a single variable, typically one related to total manufacturing activity. Head et al. w18x stressed the inadequacies of this approach, arguing that the impact of agglomeration economies on location decisions can be better measured if one accounts for the different types of agglomeration. This requires a set of specific variables for each location. 12 One reviewer of an earlier version of this paper raised a question about the stability of the coefficients for different sizes of the randomly selected choice sets. Thus, starting from a choice set of 10, we increased it in steps of 2 until we were sure that the coefficients became practically unchanged with additional runs.
AGGLOMERATION AND LOCATION IN PORTUGAL
123
TABLE 1 Independent Variables: Definitions, Expected Effects, and Sources
Variables
Definition
Expected effect on location
Source
Total Log of total manufacturing manufacturing employment per km2 agglomeration
q
DEMESS, Lisbon, Portugal, yearly data from 1985 to 1991
Industry-specific Share of manufacturing agglomeration employment in the same three-digit SIC as the investor
q
DEMESS, Lisbon, Portugal, yearly data from 1985 to 1991
Foreign-specific Share of employment in agglomeration foreign plants
q
DEMESS, Lisbon, Portugal, yearly data from 1985 to 1991
Service Share of total employment in agglomeration the tertiary sector Žbusiness and financial services, excluding nonprofit-related activities.
q
National Institute of Statistics ŽINE., Lisbon, Portugal, 1991
Labor costs
Index of concelho manufacturing wage Žbase s national average.
y
DEMESS, Lisbon, Portugal, yearly data from 1985 to 1991
Elementary education
Proportion of the labor force with elementary education level
y
DEMESS, yearly data from 1985 to 1991
Secondary education
Proportion of the labor force with secondary education level
y
DEMESS, Lisbon, Portugal, yearly data from 1985 to 1991
Population density
Log of population density
y
National Institute of Statistics ŽINE., Lisbon, Portugal, yearly data from 1985 to 1991
Distance to Porto and Lisbon
Log of average distance in time to Porto and Lisbon
y
C.C.R.C., Lisbon, Portugal, 1986
Porto
Dummy: 1 if the concelho belongs to the Porto distrito; 0 otherwise
q
Lisbon
Dummy: 1 if the concelho belongs to the Lisbon distrito; 0 otherwise
q
124
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
Moreover, agglomeration economies may play a special role in foreign investors’ greenfield site selection decisions. Generally, information and search costs weigh higher for foreign investors’ decisions than for those of domestic investors ŽCaves w9x.. FDI also often involves substantial risk and coordination costs, especially for greenfield investments ŽAharoni w1x; Caves w9x.. There are potential fixed and variable administrative costs which increase when a plant is being managed across borders. Even for other European investors, Portugal represents an environment where the particular laws, customs, and so forth are less familiar. Agglomeration economies can offset these costs. It would seem likely that the presence of other firms in the industry, other foreign firms, and services can help service communications, transport, and other needs and will be important location considerations. In our study, we test four types of agglomeration economies in concelhos and separately for urbanization effects in the major cities. The first type of agglomeration effect is industry-specific localization economies. The existence of Marshallian industrial clusters in a sector signals a set of favorable conditions for an investor in that sector. Those conditions may include the existence of intermediate suppliers, natural resources, and a pool of specialized workers. Carlton w8x, as noted earlier, tested for the probability of establishing a branch plant in three narrowly defined industries across U.S. metropolitan areas and detected that industry-specific localization economies were statistically significant. Similarly, we test localization economies as the share of manufacturing employment in the same three-digit standard industrial classification ŽSIC. as the investor. The second type of agglomeration tested here relates to the concentration of business services. Rivera-Batiz’s w40x theoretical model offers a sound basis for understanding agglomeration economies in the often-neglected urban service sector. Urban service agglomeration economies may be particularly relevant to foreign firm location. As argued by Woodward w45x, foreign firms often prefer the availability of local professional services. This effect Žperhaps best considered a measure of urbanization economies. is measured by the concelho’s share of total employment in the tertiary sector Žexcluding nonprofit-related activities.. Third, agglomeration may be foreign-specific. Foreign investors may be attracted to areas with existing concentrations of foreign-owned firms. Being less knowledgeable as to the general conditions of the region, investors may emulate the decisions of other foreign firms to reduce uncertainty. Mariotti and Piscitello w29x argue that there are spillovers from the local foreign agglomeration to the pool of potential international investors. Finally, we have included a direct measure of total manufacturing activity to account for other types of agglomeration effects not considered
AGGLOMERATION AND LOCATION IN PORTUGAL
125
above. Previous empirical location research has included this variable, along with industry-specific localization economies, as a measure of urbanization economies ŽHansen w17x.. The general presence of manufacturing activity Žmeasured by manufacturing employment per square kilometer. may be attractive to firms that have a less specific demand for specialized labor and other inputs, but seek to locate in areas with a heritage of industrial activity. Again, this is sometimes the only agglomeration measure included in previous work. While agglomeration factors represent a particular focus of the analysis, they must be considered along with other relevant effects on the site selection decision. On the cost side of the profit function, one usually considers production factors such as the cost of land, labor, and capital. The evidence concerning the impact of labor costs on location is mixed. Some authors have found that higher wages deter foreign investment ŽLuger and Shetty w28x; Coughlin et al. w10x., while others did not find a significant relationship ŽWoodward w45x.. The latter emphasized the need to account for unit labor costs, instead of nominal wage rates, to control for the level of qualification and skill qualities of the work force. To address these concerns we use an index of concelho manufacturing wage rates.13 To test work force quality, we enter two variables into the model: the proportion of workers with an elementary education level and the proportion of workers with a secondary education level. The first, Elementary Education, includes workers that at most completed primary school. Secondary Education includes workers that at most completed high school and more than primary school. The excluded category are all workers that have education beyond high school Žthat have undergraduate and graduate degrees, technical courses, and so forth.. Information on land costs is difficult to obtain. Following Bartik w4x we use population density as a rough proxy for industrial land costs. This author stressed that the population density should reflect land costs because residential and industrial users compete for land. This argument is even more relevant for the present study, given the small regional level employed. Alternatively, population density could be viewed as capturing certain urbanization economiesᎏalthough with controls for Porto and Lisbon it is unlikely that the population size of small localities considered here really measures urbanization. Given the unique characteristics of Portugal’s principal cities ŽPorto and Lisbon., the distance to these urban centers in travel time is entered in the regressions. This variable is essential to any empirical study that purports 13 All the variables for which we had annual information are lagged by 1 yr Žsee Table 2.. A few concelhos did not have any manufacturing activity. For these cases the national manufacturing wage average was used.
126
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
to account for both agglomeration and accessibility to the major urban areas. Porto and Lisbon have the most developed transport infrastructure in the country, with the only international airports and the two largest seaports. Also, Porto and Lisbon contain the head offices for the major Portuguese financial institutions and the main governmental departments. They offer a diverse range of professional services and cultural amenities not found elsewhere in the country that may be especially attractive to foreign investors. It has previously been asserted that the uncertainty of new investment tends to favor locations in highly urbanized areas to minimize risk ŽRichardson w39x.. Previous research on FDI in Europe has suggested that foreign firms Žin particular. seek urbanization economies ŽBlackbourn w5x.. Dummy variables were included in the model to account for the additional, unobserved urbanization economies potentially found in the distritos14 of Porto and Lisbon. Capital costs, with the interest rate serving as a proxy, are usually invariant across local choices. For this reason they are generally not included as an explanatory variable in location models. Some papers investigating U.S. states have found that tax differentials help explain the location decisions of foreign investors ŽCoughlin et al. w10x; Friedman et al. w14x; Woodward w45x.. However, taxes are not relevant for Portugal because those that can affect capital costs are set at the national level. On the revenue side, independent variables commonly used to account for differences in market demand are regional per capita or total income ŽCoughlin et al. w10x; Woodward w45x; Mariotti and Piscitello w29x.. However, as suggested by Coughlin et al. w10x, the explanatory power of these variables tends to be low for FDI because it is unlikely that the market served by the foreign firm coincides with the boundaries of the regions considered. This problem is even more relevant if the analysis is performed at a small regional level. As pointed out by Mariotti and Piscitello w29x, a firm can easily gain market access to a neighboring region. These observations are particularly pertinent for our study for two reasons. First, there is empirical evidence that foreign manufacturers located in Portugal are serving international marketsᎏmostly the rest of the European Community.15 Second, the dimension of the concelho market is too small to render it attractive for a foreign firm. For these reasons this study does not include local demand variables.
14 The distrito is a higher administrative region level composed of several adjacent concelhos. Portugal Žmainland. is divided into 18 distritos. 15 Around 80% of the foreign firms operating in Portugal are export-oriented. These firms export more than 50% of their total sales. See Banco de Portugal w3, p. 75x.
AGGLOMERATION AND LOCATION IN PORTUGAL
127
IV. EMPIRICAL RESULTS Econometric Estimates The results of the model are shown in Table 2. This table displays four different specifications. Overall, the model performed well, as evidenced by the 2 and t tests. In specification 1, all variables that are statistically significant have the expected signs, with the exception of population density and labor costs. Population density may pick up a number of influences besides relative land costs, even with the other urban variables held constant. Labor costs, often a fraction of overall plant operating costs, may not weigh heavily in the local decision. Instead, as Hill and Munday w23x suggest, labor costs may be interpreted as a proxy for qualifications and skills of the work force. To test this argument we ran specification 2, which includes two additional variables that control for the education level of the work force. As we can see in Table 2, the impact of the labor costs declined dramatically, and this variable became insignificant. In specifications 3 and 4 we included two dummies to account for the distritos of Lisbon and Porto. The inclusion of these two variables for the large urban areas of Portugal did not change the estimates for the agglomeration and other independent variables, except Žnot surprisingly. population density. The Porto and Lisbon dummy variables are statistically significant, suggesting that once local agglomeration economies are taken into account, there is evidence that the two main urban areas exert an additional unobserved effect on foreign investors. Furthermore, the tradeoff between agglomeration and distance can be seen from the results of all specifications. A locality’s distance from Portugal’s principal cities has negative implications for the location of foreign plants. The high level of highway investment in more recent years, cofinanced by the European Community, may reduce travel time and disperse investment in the future. Above all, the results presented in Table 2 provide evidence that agglomeration economies are a determinant factor in the location of foreign investors. Comparing the coefficients of industry-level Žlocalization. economies and service agglomeration Žan urbanization measure. indicates that the latter has the highest impact. Total manufacturing employment Žan urbanization measure often tested in location studies. also appears to be significant. Moreover, observe that additional urbanization economies of Lisbon and Porto exert a detectable pull on foreign firm location beyond the higher level of services in the concelhos that comprise these cities. On the other hand, foreign-specific agglomeration does not seem to matter once services and the locational pull of the major citiesᎏincluding the ports, specialized labor, and higher order urban servicesᎏis accounted for in the model.
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
128
TABLE 2 Regression Results for All Investments Ž N s 758. Variables Total manufacturing agglomeration
Specification 1 0.38089* Ž7.118.
Specification 2 0.46078* Ž8.258.
Specification 3 0.36333* Ž6.140.
Specification 4 0.43550* Ž7.107.
Industry-specific agglomeration
3.4102* Ž13.121.
3.3874* Ž12.990.
3.3854* Ž13.029.
3.3695* Ž12.925.
Foreign-specific agglomeration
y0.28000 Žy0.864.
y0.18359 Žy0.567.
y0.39337 Žy1.179.
y0.26029 Žy0.786.
Service agglomeration
4.3073* Ž6.036.
4.1654* Ž6.036.
4.1285* Ž5.617.
3.9859* Ž5.582.
Labor costs
0.01264* Ž5.312.
0.00058 Ž0.185.
0.01494* Ž5.635.
0.00257 Ž0.786.
Elementary education Secondary education Population density
y3.6810* Žy5.006.
0.14830 Ž0.149.
0.43728 Ž0.451.
0.05158 Ž0.760.
0.09275 Ž1.255.
0.00879 Ž0.118.
y0.33226*** Žy1.803.
y0.35304*** Žy1.867.
y0.34109*** Žy1.795.
Porto
0.45398* Ž3.775.
0.50992* Ž4.202.
Lisbon
0.24631*** Ž1.846.
0.24746*** Ž1.841.
Distance to Porto and Lisbon
Log-likelihood
2
0.13731** Ž2.048.
y3.7470* Žy5.034.
y0.36425** Žy1.977.
y1976.8515
y1959.5918
y1969.4369
1638.6382
1673.1576
1653.4674
y1950.6006 1691.34
Notes. t-values are in parentheses. The symbols *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.
Besides agglomeration variables Žespecially the urbanization measures., the distance to Portugal’s urban core cities, implying higher travel costs, proved to be statistically significant across all specifications. These findings uphold the basic tenets of urban and regional location theory. However, the concelhos’ population density and labor costs were not statistically significant. The lack of significance of population density was not expected since it is assumed to reflect land costs, a prominent factor in neoclassical location analysis. But this result could arise because population density is
AGGLOMERATION AND LOCATION IN PORTUGAL
129
not a good proxy for land cost variations.16 Of course, the alternative explanation for this variableᎏthat it is another measure of urbanization economiesᎏis also not supported by the evidence, as these effects are picked up by other variables. The insignificance of labor costs corroborates the findings of other authors who examined location in small districts ŽHansen w17x; Woodward w45x.. It is possible that all else being equal, higher wages within a country are unlikely to deter investment at the local level. Wages may well be a determinant of the decision to locate in Portugal as opposed to other countries in the EC, but not a part of the decision to pick a concelho.17 Robustness Tests In this section we test the robustness of our specifications. This is accomplished by applying the specifications in Table 2 to subsamples of our data set. The first test concerns the possible existence of a different logic in the location of foreign investment according to the ownership of capital. In our original sample, we included all investments with foreign participation Žsee above.. For our first test of the results’ robustness, we selected only those investments which had a majority Žmore than 50%. of the capital controlled by foreign investors. A second argument often posed in the literature is that regional incentives can affect the location decision. If true, this could distort our empirical findings. The Portuguese central government does not have an established policy of directing foreign investment to underdeveloped areas. However, regions sometimes compete for larger investments. To test if the location determinants are different for smaller firms, we constructed a subsample of those new plants that had fewer than 100 employees. Finally, we constructed a subsample consisting only of those foreign investments performed after 1989. This regression tests the temporal stability of our empirical findings. The results for the regressions computed on the three subsamples are shown in Tables 3᎐5. Despite the smaller dimension of the samples, the coefficient estimates are remarkably stable. All the agglomeration variables that were significant in Table 2 are still statistically significant in the regressions shown in these tables. However, note that the distance in time to Portugal’s urban core cities loses some significance in these new regressions, particularly in those shown in Table 5ᎏthe regression for investments performed in the more recent period. In this table the impact of urbanization on large cities Žthe dummy for Lisbon and the distance 16 The unavailability of alternative measures for land costs prevented us from exploring this issue in more depth. However, note that previous work on interurban location has failed to detect land price as a statistically significant effect. See Hansen w17x. 17 For a study of wage effects and the location of foreign direct investment among countries, see Cushman w11x.
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
130
TABLE 3 Foreign Majority-Owned Investments Ž N s 518. Variables Total manufacturing agglomeration
Specification 1 0.36802* Ž5.756.
Specification 2
Specification 3
0.44858* Ž6.730.
0.34937* Ž4.937.
Specification 4 0.42605* Ž5.812.
Industry-specific agglomeration
3.3331* Ž10.891.
3.2923* Ž10.750.
3.3022* Ž10.794.
3.2684* Ž10.681.
Foreign-specific agglomeration
y0.14554 Žy0.377.
y0.06719 Žy0.174.
y0.26918 Žy0.676.
y0.15546 Žy0.393.
Service agglomeration
4.0775* Ž4.899.
3.8766* Ž4.805.
3.8965* Ž4.534.
3.6773* Ž4.387.
Labor costs
0.01265* Ž4.461.
0.00120 Ž0.323.
0.01504* Ž4.788.
0.00306 Ž0.795.
Elementary education
y3.9776* Žy5.164.
y3.9227* Žy5.167.
Secondary education
y0.37092 Žy0.347.
y0.04119 Žy0.040.
Population density Distance to Porto and Lisbon
0.13006 Ž1.626.
0.04752 Ž0.587.
0.07785 Ž0.885.
y0.00856 Žy0.096.
y0.30104 Žy1.374.
y0.27627 Žy1.256.
y0.28584 Žy1.266.
y0.28021 Žy1.229.
Porto
0.52268* Ž3.588.
0.58193* Ž3.954.
Lisbon
0.29293*** Ž1.821.
0.31166*** Ž1.922.
Log-likelihood
2
y1380.8775
y1367.4328
y1374.1025
y1359.3140
1059.9242
1086.8132
1073.4742
1103.0512
Notes. t-values are in parentheses. The symbols *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.
variable. is clearly less important. Further research on location in Portugal could help determine whether this tendency toward deconcentration continues over time as new European Community᎐sponsored infrastructure seeks to open up traditionally rural areas of the country for new investment. V. CONCLUSION This paper provides new evidence concerning the factors that influence the location of foreign direct investment. The most important finding to surface in the research concerns agglomeration economies. The Marshal-
AGGLOMERATION AND LOCATION IN PORTUGAL
131
TABLE 4 All Investments ŽFewer than 100 Employees. Ž N s 688. Variables Total manufacturing agglomeration
Specification 1 0.35919* Ž6.452.
Specification 2 0.44850* Ž7.707.
Industry-specific agglomeration
3.1938* Ž11.271.
3.1598* Ž11.092.
Foreign-specific agglomeration
y0.51020 Žy1.447.
y0.41385 Žy1.173.
Service agglomeration
4.4545* Ž5.802.
4.3399* Ž5.860.
Labor costs
0.01231* Ž4.886.
Elementary education Secondary education Population density
0.16539** Ž2.341.
Distance to Porto and Lisbon
y0.32516*** Žy1.703.
0.00085 Žy0.257.
Specification 3 0.33953* Ž5.500. 3.1655* Ž11.178. y0.64328*** Žy1.765.
Specification 4 0.42000* Ž6.562. 3.1435* Ž11.038. y0.50502 Žy1.392.
4.2608* Ž5.388.
4.1441* Ž5.409.
0.01510* Ž5.392.
0.00170 Ž0.496.
y3.8912* Žy5.287.
y3.8031* Žy5.242.
0.33650 Ž0.344.
0.61692 Ž0.647.
0.07129 Ž0.998.
0.11581 Ž1.485.
0.02527 Ž0.321.
y0.28616 Žy1.497.
y0.31034 Žy1.579.
y0.29338 Žy1.483.
Porto
0.50290* Ž3.986.
0.55620* Ž4.375.
Lisbon
0.26379*** Ž1.869.
0.26100*** Ž1.833.
Log-likelihood
2
y1810.4104
y1791.7880
y1802.2317
y1782.1109
1455.0774
1492.3222
1471.4348
1511.6764
Notes. t-values are in parentheses. The symbols *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.
lian notion of agglomeration emphasizes that a ‘‘locality’’ may benefit from industrial clustering based on economies that are external to a firm but internal to the area. The influence of external economies carries clear implications for industrial location ŽHenderson w20, p. 81x.. Past location research has strongly suggested that new investments by both foreign and domestic firms are in fact attracted by agglomeration, but empirical tests often have been crude. In many analyses, total manufacturing activity in large regions serves as a rough proxy for agglomeration economies.
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
132
TABLE 5 All Investments ŽMarch 88᎐March 92. Ž N s 529. Variables Total manufacturing agglomeration
Specification 1
Specification 2
Specification 3
Specification 4
0.33387* Ž5.283.
0.41696* Ž6.279.
0.32134* Ž4.611.
0.39932* Ž5.490.
Industry-specific agglomeration
3.3830* Ž11.122.
3.3395* Ž10.958.
3.3686* Ž11.082.
3.3307* Ž10.943.
Foreign-specific agglomeration
y0.38163 Žy0.965.
y0.34965 Žy0.887.
y0.49719 Žy1.226.
y0.43996 Žy1.093.
Service agglomeration
4.0669* Ž5.180.
3.8240* Ž4.979.
3.9298* Ž4.884.
3.6777* Ž4.648.
Labor costs
0.01489* Ž5.369.
0.00364 Ž1.011.
0.01665* Ž5.408.
0.00512 Ž1.370.
Elementary education Secondary education Population density Distance to Porto and Lisbon
0.15940** Ž2.046. y0.20854 Žy0.983.
y3.6410* Žy4.713.
y3.6086* Žy4.734.
0.01444 Ž0.014.
0.25763 Ž0.252.
0.074495 Ž0.937. y0.17821 Žy0.837.
0.11640 Ž1.355.
0.02959 Ž0.338.
y0.19576 Žy0.899.
y0.17916 Žy0.815.
Porto
0.41448* Ž2.802.
0.47669* Ž3.193.
Lisbon
0.23595 Ž1.480.
0.24808 Ž1.544.
Log-likelihood
2
y1420.4591
y1407.7480
y1416.2975
y1402.4766
1061.9164
1087.3384
1070.2394
1097.8812
Notes. t-values are in parentheses. The symbols *, **, and *** denote significance at the 1%, 5%, and 10% levels, respectively.
In Portugal, the availability of precise local data, coupled with detailed investment information by industry, led to new findings about specific locational influences on new plant investment. The results allow for a comparison of different types of external economies as factors in the local selection decisions of foreign investors. Service agglomeration economies apparently have the strongest effect of the agglomeration factors, followed by industry-level Žlocalization. economies. A possible extension of this paper would link these empirical results to Rivera-Batiz’s w40x theoretical model of the urban service. Overall, external economies are statistically
AGGLOMERATION AND LOCATION IN PORTUGAL
133
significant and robust even when the regression includes total manufacturing agglomeration as a controlᎏthe single variable often used in past research. Furthermore, even with controls for the unobserved advantages of large-scale urban economies in Porto and Lisbon, specific agglomeration influences are statistically significant. Our analysis suggests that urbanization economies Žespecially for large cities. far outweigh industryspecific localization economies. With Porto and Lisbon dummy variables, distance Žtravel time. to the major Portuguese cities still apparently deters new plant location. As urban diversity economies, not industry localization, drives economic growth according to the ‘‘new economic geography’’ ŽGlaeser w15x., it also appears to exert a strong pull on location in Portugal. The relative importance of service agglomeration indicates that developed urban areas will continue to attract most FDI. This result is consistent with the widely accepted industrial organizationrmarket imperfections theory of foreign direct investment ŽCaves w9x.. Multinational enterprises must manage their proprietary assets over a greater distance than domestic investors. Moreover, foreign companies may be unfamiliar with operating in the local environment, i.e., the local laws, customs, taxes, and so forth. A range of high-level services may help overcome the distance and other transaction barriers facing foreign firms. According to neoclassical location theory, labor and other differences in costs across space could influence the investment decision beyond agglomeration. For concelhos, the small regions considered in this paper, there is no evidence that labor cost differences influence location. Rather, public policy initiatives to promote highway infrastructure may have an influence. The regressions show that the concelhos’ distance Žtravel time. to Portugal’s principal cities still has negative implications for new plant location by foreign-owned firms. At the same time, the strong urban orientation, particularly toward the core port cities of Portugal, may wither over time, as the results in Table 5 suggest. A trend toward deconcentration of foreign investment was also found in earlier studies of European Community countries like the United Kingdom and the Netherlands. When these countries experienced a surge of investment in the 1960s and 1970s, similar to Portugal’s after it joined the EC in 1986, the investment spread away from the core urban areas ŽDicken and Lloyd w13x; Kemper and De Smidt w25x.. The large-scale public infrastructure investment in Portugal beginning in the 1980s may reduce travel time and attract private investment to new localities, helping to disperse economic growth and development. Given that many projects are just beginning to open faster road connections from once-isolated concelhos to the rest of Western Europe, the concentration in FDI in Portugal ŽFig. 1. may diminish in the future.
134
˜ GUIMARAES, FIGUEIREDO, AND WOODWARD
REFERENCES 1. Y. Aharoni, ‘‘The Foreign Investment Decision Process,’’ Division of Research, Graduate School of Business Administration, Harvard University, Boston Ž1966.. 2. Banco de Portugal, O investimento directo estrangeiro em Portugal no perıodo de 1986 a ´ 1992, Boletim Trimestral, 15, 103᎐124 Ž1993.. 3. Banco de Portugal, Comparative export behavior of foreign and domestic firms in Portugal, Boletim Economico, Marc¸o, 73᎐83 Ž1995.. ´ 4. T. Bartik, Business location decisions in the United States: Estimates of the effects of unionization, taxes, and other characteristics of states, Journal of Business Economic Statistics, 3, 14᎐22 Ž1985.. 5. A. Blackbourn, The impact of multinational corporations on the spatial organization of developed nations: A review, in ‘‘The Geography of Multinationals’’ ŽM. Taylor and N. Thrift, Eds.., pp. 147᎐157, St. Martin’s Press, New York Ž1982.. 6. P. Braunerhjelm and R. Svensson, Host country characteristics and agglomeration in foreign direct investment, Applied Economics, 28, 833᎐840 Ž1996.. 7. P. Calem and G. Carlino, Urban agglomeration and economies in the presence of technical change, Journal of Urban Economics, 29, 82᎐95 Ž1991.. 8. D. Carlton, The location and employment choices of new firms: An econometric model with discrete and continuous endogenous variables, The Re¨ iew of Economics and Statistics, 65, 440᎐449 Ž1983.. 9. R. E. Caves, ‘‘Multinational Enterprise and Economic Analysis,’’ Cambridge Univ. Press, Cambridge, UK Ž1996.. 10. C. Coughlin, V. Terza, and V. Arromdee, State characteristics and the location of foreign direct investment within the United States, The Re¨ iew of Economics and Statistics, 73, 675᎐683 Ž1991.. 11. D. O. Cushman, The effects of real wages and labor productivity on foreign direct investment, Southern Economic Journal, 54, 174᎐185 Ž1987.. 12. P. David and J. Rosenbloom, Marshallian factor market externalities and the dynamics of industrial location, Journal of Urban Economics, 28, 349᎐370 Ž1990.. 13. P. Dicken and P. Lloyd, Geographical perspectives on United States investment in the United Kindgom, En¨ ironment and Planning A, 8, 685᎐705 Ž1976.. 14. J. Friedman, H. Fung, D. Gerlowski, and J. Silberman, A note on ‘‘State Characteristics and the Location of Foreign Direct Investment Within the United States,’’ Re¨ iew of Economics and Statistics, 78, 367᎐368 Ž1996.. 15. E. L. Glaeser, H. D. Kallal, J. A. Scheinkman, and A. Shleifer, Growth in cities, Journal of Political Economy, 100, 1126᎐1152 Ž1992.. 16. P. Guimaraes, ˜ R. Rolfe, and D. Woodward, Regional incentives and industrial location in Puerto Rico, International Regional Science Re¨ iew, 21, No. 2, 119᎐138 Ž1998.. 17. E. R. Hansen, Industrial location choice in Sao ˜ Paulo, Brazil: A nested logit model, Regional Science and Urban Economics, 17, 89᎐108 Ž1987.. 18. K. Head, J. Ries, and D. Swenson, Agglomeration benefits and location choice: Evidence from Japanese manufacturing investments in the United States, Journal of International Economics, 38, 223᎐247 Ž1995.. 19. J. V. Henderson, Ways to think about urban concentration: Neoclassical urban systems versus the new economic geography, International Regional Science Re¨ iew, 19, 31᎐38 Ž1996.. 20. J. V. Henderson, ‘‘Urban Development: Theory, Fact, and Illusion,’’ Oxford Univ. Press, New York Ž1988.. 21. J. V. Henderson, Efficiency of resource usage and city size, Journal of Urban Economics, 19, 47᎐70 Ž1986..
AGGLOMERATION AND LOCATION IN PORTUGAL
135
22. J. V. Henderson, The impact of government policies on urban concentration, Journal of Urban Economics, 12, 280᎐303 Ž1982.. 23. S. Hill and M. Munday, The determinants of inward investment: A Welsh analysis, Applied Economics, 54, 258᎐266 Ž1991.. 24. A. M. Isserman, It’s obvious, it’s wrong, and anyway they said it years ago? Paul Krugman on large cities, International Regional Science Re¨ iew, 19, 37᎐48 Ž1996.. 25. N. J. Kemper and M. De Smidt, Foreign manufacturing establishments in the Netherlands, Tijdscrift ¨ oor Economische en Sociale Geografie, 71, 21᎐44 Ž1980.. 26. P. Krugman, ‘‘Increasing Returns and Economic Geography,’’ MIT Press, Cambridge, MA Ž1991.. 27. P. Krugman, Increasing returns and economic geography, Journal of Political Economy, 99, 483᎐499 Ž1991.. 28. M. Luger and S. Shetty, Determinants of foreign plant start-ups in the United States: Lessons for policymakers in the Southeast, Vanderbilt Journal of Transnational Law, 18, 223᎐245 Ž1985.. 29. S. Mariotti and L. Piscitello, Information costs and location of FDIs within the host country: Empirical evidence from Italy, Journal of International Business Studies, 26, 815᎐841 Ž1995.. 30. A. Marshall, ‘‘Principles of Economics,’’ Vol. 1, Macmillan, London Ž1898.. 31. R. Martin, The new geographical turn in economics: Some critical reflections, Cambridge Journal of Economics, 23, 65᎐91 Ž1999.. 32. J. Mata, P. Portugal, and P. Guimaraes, ˜ The survival of new-plants: Start-up conditions and post-entry evolution, International Journal of Industrial Organization, 13, 459᎐481 Ž1995.. 33. D. McFadden, Conditional logit analysis of qualitative choice behavior, in ‘‘Frontiers in Econometrics’’ ŽP. Zarembka, Ed.., pp. 102᎐142, Academic Press, New York Ž1974.. 34. D. McFadden, Modelling the choice of residential location, in ‘‘Spatial Interaction Theory and Planning Models’’ ŽA. Karquist, Ed.., North-Holland, Amsterdam Ž1978.. 35. S. Micossi and G. Viesti, Japanese direct manufacturing investment in Europe, in ‘‘European Integration: Trade and Structure’’ ŽA. Winters and A. Venables, Eds.., Cambridge Univ. Press, New York Ž1991.. 36. R. Nakamura, Agglomeration economies in urban manufacturing industries: A case of Japanese cities, Journal of Urban Economics, 17, 108᎐124 Ž1985.. 37. Organization for Economic Cooperation and Development ŽOECD., ‘‘Reviews on Foreign Investment,’’ OECD, PortugalrParis Ž1994.. 38. M. Porter, Competitive advantage, agglomeration economics, and regional policy, International Regional Science Re¨ iew, 19, 85᎐91 Ž1996.. 39. H. W. Richardson, ‘‘Regional Economics,’’ University of Illinois Press, Urbana, IL Ž1979.. 40. F. Rivera-Batiz, Increasing returns, monopolistic competition, and agglomeration economies in consumption and production, Regional Science and Urban Economics, 18, 125᎐153 Ž1988.. 41. D. Segal, Are there returns of scale in city size? Re¨ iew of Economics and Statistics, 58, 339᎐350 Ž1976.. 42. D. Smith and R. Florida, Agglomeration and industrial location: An econometric analysis of Japanese-affiliated manufacturing establishments in automotive-related industries, Journal of Urban Economics, 36, 23᎐41 Ž1994.. 43. L. Sveikauskas, The productivity of cities, Quarterly Journal of Economics, 89, 393᎐413 Ž1975.. 44. D. Wheeler and A. Mody, International investment location decisionsᎏThe case of U.S. firms, Journal of International Economics, 33, 57᎐76 Ž1992.. 45. D. Woodward, Locational determinants of Japanese manufacturing start-ups in the United States, Southern Economic Journal, 58, 261᎐273 Ž1992..
