Are there really high-tech clusters? The geographic concentration of German manufacturing industries and its determinants

In this paper the terms “geographic concentration” and “agglomeration” are used synonymously.

For a discussion of various measures of geographic concentration see, for example, Devereux et al. (2004) and Combes and Overman (2003).

One may worry that EG’s model does not capture the firms’ true location decision process. In Alecke et al. (2005) we therefore included a simpler measure of geographic concentration which also controls for the plant size distribution for comparison. The Spearman rank correlation with the EG index is almost 0.9 both at the county and the more aggregate planning region level.

For these industries the difference between G and G ₀ is larger than 1.96 times its standard deviation.

The classification used is common and due to Grupp et al. (2000). See Table 8 in the Appendix for more details.

One would also like to examine the coagglomeration of industries with strong upstream-downstream ties as has been done in Ellison and Glaeser (1997). Unfortunately, input/output dependencies are available only at the two-digit industry level.

Maurel and Sédillot (1999) proposed a descriptive way to quantify the relative strength of industry-specific and group-specific agglomeration. They note that the concentration of a whole industry group measured by the “simple” γ of the group, can be written as the weighted average of the γ’s of the industries (“intra-industry concentration”) and some group-specific component (“inter-industry concentration”) which is not to be confused with EG’s γ ^c . Then it is possible to express intra-industry agglomeration (the weighted γ’s) as a fraction of the group’s total concentration. This ratio ranges from as low as −2 to 134% (see column 3 in Table 3). A fraction of intra-industry concentration greater than 100% corresponds to a negative contribution of the inter-industry component. Communications Engineering on rank 22, for example, is a group whose industries by themselves are more concentrated than the group which amounts to a dispersing inter-industry component. The same is true for Manufacture of Electrical Equipment on rank 21; here the negative sign reflects that the group as a whole is even dispersed while the intra-industry component is positive.

Spatial correlation means that there is a tendency of neighboring regions to have the same characteristics.

Not much can be said about the error with which γ is measured. See the footnote in Ellison and Glaeser (1997), p. 908.

Precise estimates of the range of spillovers are rare. Anselin et al. (1997) and Funke and Niebuhr (2000) suggest that the range is at least 50–75 miles.

In Alecke et al. (2005) as a measure of labour specificity an industry’s deviation from the average labour composition (across all types of occupations) is additional used. The results are very similar.

Marshall (1920) argued that higher transportation costs induce firms to locate closer to suppliers and customers. This results in the colocation of trade partners and has to be distinguished from pure localization economies because it can make a single industry either agglomerated or dispersed.

The transportation cost per unit of weight are assumed to be constant across industries so that the portion of total transportation cost in output (the importance of transportation cost), \(\tfrac{c}{t}\tfrac{{{\text{weight}}}}{{{\text{output}}}}\), is proportional to the inverse unit value.

Three extreme outliers (Watches, Jewellery and Fish Processing) are excluded. All are very small (0.08, 0.23 and 0.1% of manufacturing employment); Watches and Jewellery are characterised by family-owned, small-scale handcrafts for which the location decision is presumably determined by family tradition. Naturally, Fish processing is located only at Germany’s short coastline in the North. For eight non-extractive industries no transportation cost are available and they are assigned its average value. Seven industries are excluded because data for more than one variable is missing.

This result seems contrary to Rosenthal and Strange (2001) who find that manufactured inputs significantly increase agglomeration while “non-manufactured” ones, containing financial, legal, repair etc. services, decrease it. The crucial question is whether manufactured or non-manufactured inputs are more industry-specific, more difficult to transport over distance and more likely to be produced under increasing returns. The theory is about increasing returns and extremely high transportation costs which applies foremost to services which drive the inputs-sharing argument more than physical goods. One explanation for the different results is the subtle but important difference in what the variables measure. Our service variable takes only technical and industrial services as opposed to Rosenthal and Strange’s variable which also contains items such as legal and financial services which may indeed be “available everywhere” (ibid, p. 17). In the end our manufactured inputs variable corresponds to their “non-manufactured” one and it is all about naming variables.

We thank an anonymous referee for this remark.

The value is chosen so that 43% of the observations are censored. In EG 43% of the industries are “not much” concentrated.

Unreported results with 8, 9, 10 quantiles are very similar. The thresholds proposed by EG (0.02 and 0.05) cannot be used because there would be too few observations in the high- and medium-γ category.