Agglomeration economies in manufacturing industries: the case of Spain

Abstract

This study analyses the extent of geographical concentration of Spanish industry between 1993 and 1999, and studies the agglomeration economies that could underlie that concentration. The results confirm that there is major geographic concentration in a number of industries with widely varying characteristics, including high-tech businesses and those linked to the provision of natural resources as well as traditional industries. The analysis of the scope of spillovers behind this agglomeration supports the idea that transportation costs may induce plants in some industries to locate near their customers and suppliers. However, one cannot conclude that this is a common fact for all industries. This study also shows that the higher the technological level of an industry, the higher the agglomeration it experiences. This result implies the importance of the labour market, informational spillovers and producer services location for the agglomeration of these industries.

Acknowledgements

Financial support from the Spanish Ministry of Science and Technology via grants BEC2002-04102-C02-01 and BEC2002-02527, from FEDER, and from the Regional Government of Galicia via grant PGIDT03PXI 30002PN are gratefully acknowledged.

Notes

See, for example, Glaeser et al. (1992), Henderson et al. (1995) and Ellison and Glaeser (1997) in the American case; Haaland et al. (1999) in the European case; or Maurel and Sédillot (1999) in the French case.

Marshall (1890), Christaller (1933), Lösch (1940) and Pred (1966), among others.

Important theoretical contributions in the field are Krugman (1991) and Venables (1996), among others. A review of this literature can be seen in Schmutzler (1999). Also, Fujita et al. (2000) offer a thorough analysis of the main contributions.

This topic has been analysed in a formal model by Venables (1996).

These three reasons had already been identified by Marshall (1890) at the end of the 19th century.

In contrast to other works that study the geographic concentration of industry in Spain (Callejón and Costa, 1995, Paluzie et al., 2000, and Viladecans, 2000), not only the Gini and Ellison and Glaeser (1997) indices are used but also that proposed by Maurel and Sédillot (1999).

When using the term spillovers one actually mean agglomeration economies in production.

Another distinction is drawn in the literature between localization economies (across businesses in the same industry) and urbanization economies (across businesses in different industries). See Henderson et al. (1995) and Glaeser et al. (1992) among others.

This model does not discriminate between these two possible causes of the location decision.

This 2-dimensional random variable is made up of two non-independent Bernouilli variables.

This probability depends on the size of the location, measured in terms of aggregate industrial employment there, so that if one location has twice as much employment as another, the probability of a plant in the industry analysed choosing to locate there is twice as high as at the other location. In other words, x _i is the proportion of industrial aggregate employment at i.

This probability, p, is precisely

, where j, k ∈ i denotes the plants in the industry that choose to locate at location i.

At empirical level the location may be a natural district, department, province, region, state, etc.

Note that the fraction of employment in the sector at a location can be written in terms of random variables, and hence the primary indices can also be considered as random variables.

.

The Gini index is calculated by ordering the various units of territory in accordance with the Hoover-Balassa index, which measures the ratio s_i /x_i . The x-axis represents the cumulative proportions of industrial employment as a whole, and the y-axis the cumulative proportions for the industry under study. The Gini index measures the quotient between the area within the corresponding Lorenz curve and the 45-degree line and the area below this line. Specifically, the Gini index would take the form

where q _i denotes the cumulative proportion of employment in the sector and p _i the cumulative proportion in the industry as a whole for the first i units of territory in the ranking obtained via the Hoover-Balassa index. A particular industry which is distributed in a way similar to the industry as a whole gives a value for the Gini index of zero. The Gini index has also been calculated, taking population distribution as a reference, and the results are very similar. Correlation between the two indices is high.

The correlation of the array obtained from the M-S index with that obtained from the E-G index in the Spanish case is around 56%. With the Gini index it is 68%. Using French data, the paper by M-S finds higher correlations for the first two indices (90%).

The EI covers all population for firms over 20 workers, while for smaller firms an estimation based on a representative sample is undertaken.

In 1993 the survey was modified in two important points: the survey unit changed from establishments to firms, and the CNAE-93 industrial classification was adopted. The period analysed begins in 1993 so that homogeneous data are available for the full period.

The INE provides no information on industries in a localization (province or CCAA) when there are less than four plants.

Industries 11, 12, 13, 16 and 23 have been eliminated. These industries cover part of the mining and extraction industry, tobacco and coke plants/oil refineries/nuclear waste treatment. In industries 11 and 12 the INE provides no information for the industry. In industries 13, 16 and 23 the number of CCAA in which data are not available is 9, 11 and 13 out of 17, respectively.

A deeper analysis of the concentration in 1999 can be seen in Alonso-Villar et al. (2003).

In this case the Herfindahl index informs us that employment is distributed across many plants, but in spite of this a high spatial concentration is observed.

It should be noted that the Gini index does not classify industries 34, 40 and 41 among the lowest. This is probably due to the high degree of concentration of output at a small number of plants, as can be deduced from the Herfindahl index.

The Gini index in this case does not show very high figures.

Callejón (1997) also shows that the E-G index does not change significantly between 1981 and 1992.

See the Appendix.

This classification excludes industries 10–14 and 40–41.

Both the industrial classification given by the input-output matrix and the OECD include 3-digit industries.

A deeper analysis on the different sub-industries of Textiles cannot be undertaken, since the input-output matrix does not allow it. However, it seems that input-output relationships between the different sub-industries can cause the high degree of inter-industry spillovers in this sector, see Alonso-Villar et al., 2003.

Percentage that the main customer represents.

Percentage that the main seller represents.

Most high-technology industries are located in Catalonia and Madrid. In fact, Pharmaceutical goods (244), which represents 44.8% of total employment in the first group, has 54% of its output in Catalonia and 28% in Madrid.

The Chemical industry (24) also shows a high degree of inter-relation between plants belonging to different sub-industries, as shown in Alonso-Villar et al. (2003).

High technology industries strongly depend on producer services. As Hansen (1994) comments, only 10–15% of the value of an IBM computer comes from the manufacturing process, the rest coming from services such as research, design, engineering, maintenance, or sales.

Coffey and Polèse (1989) support evidence of the centralization in producer services in countries such as Canada, the UK, France and the USA.

Callejón (1997) also shows that the E-G index does not change significantly between 1981 and 1992.

By j, k ∈ i we mean that j and k locate in the same geographic area i.

Analogous expressions can be found when j ∈ l, k ∈ l′, l and l ^′ being two subsectors in sector r, l≠ l′. In this case, we denote by γ₀ = corr(U_ij , U_ik ). Otherwise, that is, if j, k ∈ l, we denote by γ_l = corr(U_ij , U_ik ).

In order to simplify notation, by j and k two plants belonging to the same sector r are meant without making it explicit in the equation. By j ∈ l, k ∈ l′ it is meant that j belongs to subsector l, while k does to subsector l ^′, l and l ^′ being two subsectors of sector r.