Thomas Brenner: Local industrial clusters. Existence, emergence and evolution, London, New York 2004. 246 S.

In search for clusters new ideas are often triggered at the interfaces of different disciplines where different pieces of knowledge come together and are combined. The book of Thomas Brenner on the evolution of clusters provides an outstanding example, because, in bringing together insights from evolutionary economics and economic geography, it adds new insights to a quite controversial topic in economic geography. In doing so, Brenner gives a new boost to the extensive literature on clusters that was almost declared dead by leading economic geographers, for being too vague and lacking analytical rigour (e.g. Martin/ Sunley 2003, Taylor 2005). The book is complex, original and thought provoking, presenting a new concept of clusters that is firmly grounded in a general theoretical framework. The book provides a general model on clustering, from which certain propositions are derived, that are tested empirically in regressions analyses, simulations and case studies. The approach is comprehensive, making use of a variety of advanced tools that are disposable to a regional scientist. The empirical work is based on a German study of the distribution of firms in 293 industries among 441 administrative districts during the period 1995-2000. In the final part, some recommendations on cluster policy are deduced from the main findings of the research.
A general approach on clustering The main objective of the book is to develop a theoretical model that can accurately describe the evolution of clusters in general terms, going beyond the peculiarities of each cluster. Following previous attempts (e.g. Sforzi 1990, Paniccia 2002, Maggioni 2002), it succeeds in providing a complementary view on clusters that is largely dominated by case study approaches. Instead of comparing the empirical findings of numerous case studies in search for similarities among clusters, his approach starts from a general theoretical approach that is informed by empirical studies. Brenner claims in this respect that his approach does not assume or take for granted that clusters exist, as many case studies have done when identifying characteristics of clusters, but actually tests their existence empirically. The book takes up three questions that will be dealt with one by one below: why do clusters exist, where do they emerge, and when?
Why does clustering occur? The key question Brenner addresses is why clustering occurs. In providing an answer, his general model draws a sharp analytical distinction between different sources of clustering. On the one hand, there are factors (e.g. access to natural resources, proximity to customers) that are not influenced systematically by changes in the firm population in a region and, therefore, are treated as exogenous factors in the theoretical model. On the other hand, Brenner puts emphasis on local self-augmenting processes that are considered a key cause of clustering. As he explains at length, case studies of clusters have rarely addressed, let alone tested, this issue of self-reinforcing processes empirically. Self-augmenting processes may be caused by many mechanisms, such as information flows and human capital accumulation. In the theoretical model, the concrete underlying mechanisms are not further specified because the importance of each of them is expected to differ significantly between industries. Instead, Brenner adopts a general framework that aims to identify a complete list of clusters in all industries. Consequently, he is not really after explaining why packaging concentrated in the German regions of Mainz-Bingen and Furth, and why optics is clustered in the Lahn-Dill-Kreis, Havelland and Jena regions. In fact, he acknowledges his study should be supplemented by case study approaches to determine which of the concrete mechanisms is responsible for the clustering of an industry in a region, because "... the causes, forms and evolution of clusters might differ significantly between industries. In this respect, case studies are superior to the approach taken here" (p. 121). Based on the analysis of the spatial distribution of firms in each of the 293 German industries in the period 1995-2000, Brenner concludes that the theoretical model describes well for almost all industries (276 out of 293) the firm distributions among the 441 German regions (being clustered or not). This outcome confirms the general approach that "... there is a general level on which clustering in different industries can be treated jointly" (p. 77). Adopting such a general framework of clustering might, however, suggest that it abstracts from industry- and region-specific features. The opposite is true. While the model points out that local selfaugmenting processes are a necessary condition for the existence of local industrial clusters, it is not regarded as a sufficient one. It also depends on sector-specific conditions and regional conditions, as will be explained below. Sector-specific features of clustering The first empirical evidence Brenner presents with respect to the importance of sector-specific features is that clustering does not occur in all German industries: for about half of all the examined industries (148), the cluster distribution describes best the spatial distribution of the firms in an industry. This is not to say that local industrial clusters exist in those industries, according to Brenner´s definition. This would also require excluding those industries that cluster due to other reasons than selfaugmenting processes. Because such information on each clustered industry is not available, a rather pragmatic approach is followed, bringing the number of real clustered industries down to 97. Irrespective of the precise number coming out of this somewhat ´fuzzy´ identification process, the overall message is clear: local clusters do not exist in all industries. The second empirical proof provided by Brenner in this respect is that clustering industries, as compared to non-clustering industries, have particular characteristics. Logistic regressions on 60 German industries demonstrate that sector-specific mechanisms like intra-industry spillovers and local cooperation with public research institutions and suppliers increase the likelihood of whether an industry clusters or not. Where does clustering occur? Besides sector-specific features, clustering may also be affected by regional conditions. If so, these conditions might determine where clustering emerges. What does Brenner's approach say about this? In his model, three factors might explain the location of clusters. First of all, local self-augmenting processes may be involved, because, for instance, their strength can differ between regions. Having said that, it is understandable but disappointing at the same time that the analyses say very little about this issue in the end, because so little is known about their role: "... we can only make guesses about the differences between regions with respect to these processes" (p. 191). Second, exogenous conditions like the ´attractiveness of regions´ matter a lot in the theoretical model, but these are assumed to influence only the likelihood of the emergence of a local cluster, and, therefore, cannot be considered to be sufficient. Thus, local conditions only influence the likelihood of clustering, and, in this respect, regions may differ. This is much related to the third factor that influences the location of clusters, that is, stochastic events, which makes Brenner´s model a real evolutionary approach. These random events are, for instance, the founding of key firms, the actions of a few leading actors, and historical singularities. Since stochastic events are difficult to capture by modelling, Brenner makes use of a simulation model. In doing, he tests whether the spatial emergence of clusters may be attributed to stochastic processes rather than industry-specific and region-specific characteristics. I will come back to this issue below. In sum, regions are endowed with different local conditions, but these only influence the likelihood of clustering. As a consequence, regional differences can only explain why in certain regions (those that lack the necessary conditions), industrial clusters will not develop, but they are not able to explain the concrete location of clusters. Therefore, it is unpredictable where clustering will take place. This is in line with ideas developed earlier, as laid down in the concept of Windows of Locational Opportunity (see Storper/Walker 1989, Boschma 1997). The achievement of Brenner is that these insights have now been incorporated and formalised in a general theoretical model.
 When discussing the role of space, we have to remind that the principal unit of analysis in Brenner´s model is the industry. Therefore, the books leaves many questions unanswered after having concluded that clusters are not located everywhere in Germany. In that respect, there is much suggested but very little proven. For instance, Brenner repeatedly claims that the spatial pattern of local clusters reflects the history of places in Germany, that economic development in a region and clustering do not go hand in hand, and that the existence of local clusters is more long lasting than their economic effects, but no sound empirical evidence is provided to support these statements. Apart from that, I agree with Brenner that a key research challenge is to compare regions hosting clusters with regions having no clusters, because that would certainly increase our understanding of which regional conditions make a difference. When doing so, it would be a step forward not to take the level of the region as given, but instead account for the fact that various mechanisms behind clustering operate at different spatial levels, and that firms in an industry develop different routines and, thus, respond differently to exogenous changes.
 When do clusters emerge? The theoretical model also accounts for dynamics: local industrial clusters may emerge, stabilise and decline. The basic model describes critical values of the mechanisms causing self-reinforcing processes, below which clustering is ruled out, but above which clustering might occur in the respective region. Depending on certain values of a few critical parameters, it analyses whether a change in an exogenous condition (e.g. a change in demand) results in a significant increase or decrease in the number of firms in the respective industry in a region (causing a shift between the two stable states of clustering and no clustering), or whether it causes only a small change in the size of the local firm population (leaving the stable state unchanged). Based on these basic ideas, Brenner proposes a four-stage model of the evolution of local industrial clusters, each of which is characterised by specific processes.
 Whereas the general model is well thought over, the empirical part is, by and large, not always convincing, although some interesting results are obtained from the simulations. The weaker points are mainly caused by data problems, which is fully recognised and brought to the attention by Brenner himself. For example, it is not very convincing to conduct an analysis on the evolution of clusters that covers a period of only 5 years (1995-2000) when one expects their evolution to take at least a few decades to grow and mature, as many case studies have demonstrated. The data on German industries also do not allow the identification of new industries, which makes it quite tricky to draw conclusions on emerging clusters, such as the one that states that clusters emerge during a very short period of time. Moreover, when the analyses show that a substantial number of industries switch between stable states in a period of 5 years, due to, what is called, ´statistical artefacts´, it is hard to imagine what are the consequences for the findings concerning the dynamics of clustering. Having said this, the analyses show that most industries show a stable spatial distribution of firms during the period examined. What I found a much more intriguing and thought provoking outcome is that the forces leading to the emergence of clusters differ from the ones that cause their existence and relative stability. The model shows that self-augmenting processes are a prerequisite for the emergence of local clusters, but that their existence does not require that these processes are still active or at work. So, a local industrial cluster may still survive while the reasons for its existence are no longer there. This is verified in the German study showing that the emergence of clusters, but not their persistence, is explained by industrial characteristics. This implies, most interestingly, that one should be careful explaining the existence of clusters by looking at the current characteristics of industries and regions. Indeed, it suggests that path dependency rather than clustering forces explains why most clusters do not disappear. As noticed before, a simulation model has been developed to estimate the extent to which stochastic events matter in the emergence of clustering, meaning that the spatial outcome cannot be predicted. Simulations are an appropriate technique to analyse stochastic processes, because they can be run repeatedly, a possibility that analyses based on the use of empirical data cannot offer. Some mechanisms behind clustering (e.g. local spillovers) have been included in the simulation model. In these simulations, ranges for all parameters have carefully been chosen with the help of empirical studies on the characteristics of all industries, in order to reflect differences between industries and regions observed in reality as good as possible (e.g. with respect to the spatial range of spillovers). The results confirm that the emergence of clusters is, by and large, a stochastic process, meaning that the characteristics of industries only determine the likelihood of the emergence of local clusters, but do not determine whether they emerge or not. In addition, the simulations show that early dynamics in an industry determine the location of the clusters, followed by a stable pattern soon after that, in which clustering stabilises. To be more precise, the simulations demonstrate that in the first three years, it is unpredictable where clustering occurs, that is, the outcome is independent of the number of firms in the regions. After five years, however, the location of clusters is completely determined. This outcome suggests that the emergence of clustering is characterised by a process in which more and more potential regions fall behind until a few regions emerge as the leading ones. Policy implications
 In the final part of the book, some implications for cluster policies are discussed. These recommendations are much welcomed, because regional policy makers in many countries have recently embraced the idea of promoting clusters in their regions. In that respect, it is hard to avoid the impression that a lot of public spending is currently wasted on cluster policies (see, e.g., Martin/Sunley 2003), although the first real policy evaluations will only become available in the coming years. I agree on almost all arguments made by Brenner, but I think he should have made a more firm statement about the impossibilities of cluster policy in the end. Although Brenner does not push it that far, the main message I got out of his study is that cluster policies are almost bound to fail, because policy makers find themselves in an awkward and almost impossible position. On the one hand, timing is considered essential because it affects the likelihood of policy success. To be more precise, the possibilities for policy makers to influence the clustering process decrease with time, and turn almost to zero after clustering in an industry has become stabilised. This implies policy may only have an impact on the location of clusters before they start to emerge. On the other hand, Brenner states that policy can only become effective when applied to the right industry and in the right region. As he recognises, this demands almost the impossible from policy makers. It first requires that policy makers have to identify the right industry, which means new industries at their time of emerging when only very few firms will exist, which will, therefore, remain unnoticed for some period almost by definition. The policy maker has also to assess whether this industry has any growth potential, which is a complicated task, as ´picking the winner´ policies in the past have demonstrated. In addition, the policy maker needs to identify the specific characteristics of the industry, to estimate whether the industry has a capacity to cluster, and determine whether and which local conditions might favour its emergence. At the same time, the policy maker has to assess whether, and to what degree, these conditions are locally present, and compare these with the situation in other regions. Finally, it has to develop a strategy to strengthen these local conditions, and finally implement the most effective policy measures. Needless to say, policy makers do not have this capability, as Brenner acknowledges himself. Even when they would have such powers, the previous arguments makes it inevitable that policy efforts will be implemented much too late, long after the clustering of the industry has already been established and stabilised somewhere in space. But let us assume policy makers can manage these problems, without any delay. Then, the policy maker is still confronted with another highly unpleasant implication of Brenner´s study. That is, policy makers can only influence the likelihood of becoming a cluster, but cannot determine the actual place of local clusters, meaning there is no guarantee of success (see, e.g., Lambooy/Boschma 2001). What Brenner´s study has taught us is that the very essence of clustering (i.e. the maximum number of clusters is very limited in each industry) already implies that most policy makers at the regional level will fail to develop any cluster in their home region. In this respect, as Brenner proposes, coordination of cluster policy at a higher (e.g. national) level may indeed be a step forward, but it does not take away the arguments made above: it still needs the identification of the right industries in the right regions. The question then is: would it not be better to abandon the idea that policy makers can make a difference in cluster development? Is this not the lesson learnt from Brenner´s study? In sum
 There is no doubt Brenner has made an important contribution to the current literature on clustering. Having said that, throughout the book, he underlines the importance of combining general approaches and case study approaches, because that would further increase our understanding of what clustering is really about, and what role local self-reinforcing processes play in the evolution of clusters. This recommendation of bringing together of what are currently two quite separate traditions of methodology is certainly one of the greatest merits of Brenner´s book. There is still much learning from clustering ahead of us.
References
Boschma, R.A. (1997): New industries and windows of locational opportunity. A long-term analysis of Belgium, Erdkunde 51 (1), pp. 1-19.
Lambooy, J.G. and Boschma, R. A. (2001): Evolutionary economics and regional policy. The Annals of Regional Science 35, pp. 113-131.
Maggioni, M.A. (2002): Clustering dynamics and the location of high-tech firms, Heidelberg/New York: Physica-Verlag.
Martin, R. and Sunley, P. (2003): Deconstructing clusters: chaotic concept or policy panacea?. Journal of Economic Geography 3, pp. 5-35.
Paniccia, I. (2002): Industrial districts. Evolution and competitiveness in Italian firms. Cheltenham: Edward Elgar.
Sforzi, F. (1990): The quantitative importance of Marshallian industrial districts in the Italian economy. F. Pyke, G. Becattini and W. Spengenberger (eds.): Industrial districts and inter-firm co-operation in Italy, Geneva: International Institute for Labour Studies, pp. 75-107.
Storper, M. and Walker, R. (1989): The capitalist imperative. Territory, technology and industrial growth. New York: Basil Blackwell.
Taylor, M. (2005): Embedded local growth: a theory taken too far?. R. A. Boschma and R.C. Kloosterman (eds.): Learning from clusters. A critical assessment from an economic-geographical perspective, Dordrecht: Springer, pp. 69-88.  
Autor: Ron Boschma

Quelle: Geographische Zeitschrift, 92. Jahrgang, 2004, Heft 4, Seite 249-253

Kommentar schreiben