multinomial logit (MNL) competing risks model / Complementary log ...

Spinoffs and M&A’s as drivers of spatial clustering. The spatial evolution of the Dutch banking sector in the period 1850-1993 Ron Boschma and Rik Wenting

paper presented at the DIME Final Conference, 6-8 April 2011, Maastricht this is a very first draft version of the paper, please do not quote

Abstract We describe and explain why the Dutch banking cluster clustered in the Amsterdam region. This analysis is based on an unique database of all banks in the Netherlands that existed in the period 1850-1993. We examine the extent to which spinoff dynamics, merger and acquisition activity and the location of Amsterdam had a significant effect on the survival rate of Dutch banks during the last 150 years. Our analyses demonstrate, among other things, that Amsterdam banks were disproportionally active in acquiring other banks, leading to a further concentration of the banking sector in the Amsterdam region. Key words: industrial dynamics, cluster, spinoffs, mergers and acquisitions, banking sector JEL codes: o18, r00, r11

1. Introduction Since long, economic geographers are preoccupied with the question of how to explain the spatial clustering of an industry. Following Marshall (1890), they refer to the importance of location factors and localisation economies, due to a pool of specialised labor, the presence of specialised input suppliers, and access to knowledge about the secrets of the respective trade. However, this view has been challenged recently. Klepper, among others, has claimed that clusters emerge through a self-reinforcing spinoff process, in which incumbent firms give birth to new firms in the same location. In that case, clusters may emerge and persist even when localisation economies are absent or negative (Sorenson and Audia, 2000; Klepper, 2007; Boschma and Wenting, 2007; Wenting, 2008). Recent studies on the spatial evolution of a particular industry tend to support the spinoff argument. Taking stock of the studies conducted so far, Boschma and Frenken (2011) have come to the conclusion that “… clusters emerge as an evolutionary process of spinoff formation, while the role of localization economies in this process is limited, at best” (p. 298). This paper aims to take up this debate further. The first objective is to replicate this type of study for the Dutch banking sector. Whereas most studies on the long-term evolution of an industry have investigated manufacturing industries, our study concerns a knowledgeintensive service industry, which has been understudied so far, exceptions being Fein (1998), Carree (2003), Consoli (2005), Grote (2008), Pratt (1998) and Wenting (2008). We test whether spinoff dynamics and localisation economies can be held responsible for the evolution of the Dutch banking industry, and in particular its spatial concentration in the Amsterdam region. The second objective is to explore more explicitly the role of mergers and acquisitions in our explanatory framework. We argue that mergers and acquisition (M&A) can be regarded as an additional driver of spatial clustering of an industry. To our knowledge, this has not yet been systematically investigated, and the banking sector might be an excellent case to test this, as the number of exits due to acquisition is high, much higher than in some other industries. The third objective of this paper is to present a competing risk model, which allows us to make a distinction between two types of failure. In most of the studies, survival techniques have been employed to calculate the risk of failure of firms. However, an exit of a firm is not necessarily a sign of failure, but may also be caused by a takeover. Since the banking sector includes many takeovers, it is possible to examine the determinants of both types of exits. The analyses are based on an unique database of all entries and exits in the Dutch banking sector for the period 1850-1993 that were collected by the authors. The structure of the paper is as follows. Section 2 develops a perspective on the spatial evolution of industries when discussing localisation economies, spinoff dynamics and M&A activity as possible explanations. Section 3 introduces the data and describes the evolution of the Dutch banking sector, and more in particular its concentration in the Amsterdam region. Section 4 examines which factors can be held responsible for the spatial clustering of the Dutch banking industry. We devote special attention to the merger and acquisition activity of Dutch banks, and of Amsterdam banks in particular. Section 5 discusses the implications for future research. 2. Why do industries concentrate in space? 2.1 localisation economies It is an empirical fact that many industries concentrate in space, especially knowledgeintensive industries. Since the seminal contribution of Marshall (1890), economic geographers have referred to the importance of localization economies to explain the spatial clustering of

an industry. According to Marshall, co-located firms from the same industry will benefit from each other‟s presence because of a pool of specialized labor, the presence of specialized input suppliers, and access to knowledge about the secrets of the respective trade (see e.g. Asheim and Gertler, 2005; Potts and Watts, 2010). In more recent years, scholars have focussed especially on the accumulation of (tacit) knowledge as a driver of spatial clustering and regional specialisation (Storper, 1993, Malmberg and Maskell, 1999). Recently, this primary focus on localisation economies has been challenged more and more. Beaudry and Schiffauerova (2009) have reviewed the vast empirical literature on agglomeration economies, and they show that empirical findings concerning localisation economies are at best inconclusive. In the cluster literature, there is also increasing scepticism whether clusters do impact positively on the performance of cluster firms. There is increasing evidence that clusters may be regarded as local environments that create and attract a lot of new entrants in a particular trade, causing spatial concentration of an industry (Heebels and Boschma, 2011). But at the same time, clusters tend also to produce high numbers of exits, because the selection environment may be extremely tough (Boschma and Wenting, 2007). This may be due to the fact that there are many local competitors around, which decreases the survival rates of entrants, but that may also be caused by the fact that clusters are high cost environments (like high labour costs, high rents, etc.) (see Staber, 2001). Therefore, we expect that most cluster firms will not perform well, with the exception of a few successful cluster firms. As a result, we formulate the following hypothesis: Hypothesis 1: cluster firms show higher failure rates 2.2 spinoff dynamics The spatial formation of an industry has also been described in terms of spinoff dynamics (Arthur, 1994; Cantner et al., 2006; Klepper, 2007; Buenstorf and Klepper, 2009). With spinoffs, we mean new firms that are founded by employees of incumbent firms in the same industry. Arthur (1994) argued that spinoff formation drives clustering of a new industry in a region that is lucky to have many spinoffs in the first stage of the industry life cycle. This is because the probability of a region to give birth to a new spinoff is dependent on the number of incumbent firms in the region. That is, the more spinoffs enter the region, the higher the probability that more spinoffs will be generated. This makes it a path-dependent and selfreinforcing process. Another explanation conceives the spinoff process as a mechanism through which (tacit) knowledge is transferred from parent to offspring, and which positively affects the performance of spinoffs (Helfat and Lieberman, 2002). Klepper (2007) claims that entrepreneurs with previous experience in the same industry (spinoffs) or in related industries (experienced entrepreneurs) perform better than entrants without these types of pre-entry experience. Following this spinoff argument, we come to the following hypotheses: Hypothesis 2a: spinoff show lower failure rates Hypothesis 2b: experienced firms show lower failure rates This implies that the spatial clustering of an industry can be explained without referring to location-specific qualities. In other words, spinoff dynamics and localisation economies provide alternative explanations for why an industry develops and concentrates in space. However, both may also play a complementary role, when, for instance, spinoff activity in a region induces agglomeration forces which further increase the survival rate of firms

(Boschma and Frenken, 2003). Recent empirical studies show that the spatial clustering of an industry is more due to spinoff formation, while the role of localisation economies is limited, at best (De Vaan et al., 2010; Boschma and Frenken, 2011). Some studies have demonstrated that clusters emerge and persist when localisation economies are absent or even negative (Sorenson and Audia, 2000; Klepper, 2007; Boschma and Wenting, 2007; Wenting, 2008). As explained in hypothesis 1, we expect a negative effect of being located in a cluster on firm‟s survival. By contrast, we expect that cluster firms that are spinoffs companies will perform better. There are two main reasons for that. One is that spinoffs have better capabilities to cope with the strong selection environment of clusters, and are better capable of exploiting the benefits that accrue to clusters at the same time (Heebels and Boschma, 2011). Another reason is put forward by Klepper (2007) who claims that successful parents generate many spinoffs, but also successful spinoffs. So, spinoff firms in a cluster may be more successful because they are offspring from successful parents in the cluster. Unfortunately, in our empirical study we cannot disentangle these two effects, because we have only limited information on the parents of spinoff banks. However, what we can test is the following: Hypothesis 3: spinoffs in clusters show lower failure rates 2.3 M&A activity Little attention has been drawn so far to the role of mergers and acquisition in the spatial clustering of industries, although there are a few exceptions (e.g. Markusen, 1985; Chapman, 1991). We claim that M&A activity is likely to impact on the spatial evolution of an industry, especially during the shake-out phase of its life cycle, when M&A activity is most intense in most industries (De Jong, 1981; Klepper, 1997). We argue there are two main reasons why M&A activity should be taken seriously when explaining the spatial clustering of an industry. The first reason is that exits of firms may not only be caused by failure, but may also occur because firms are acquired by other firms. In survival analysis, there is a tendency not to make a distinction between these two types of exits, as if both are signs of failure. If exits due to acquisition would concern weakly performing firms, then there would indeed be no reason to treat them separately from exits due to failure. However, studies have shown that exits due to acquisition may concern very successful firms, and therefore, are anything but a sign of weakness (e.g. Fein, 1998; De Vaan et al., 2010; Weterings and Marsili, 2010). There is also evidence that in the banking sector, this is indeed the case (see e.g. ). This implies that the determinants of exits due to failure may be different from the ones explaining exits due to acquisition, which we explore below. A crucial question is whether cluster firms will have a higher probability to fail due to acquisition. To our knowledge, there is no such empirical study. As M&A activity is a way for firms to expand and grow, in clusters, firms have more opportunities to take over other firms, because there are more candidates around from the same sector. In addition, when the acquiring and the acquired firm share the same location, they know each other well, which will reduce uncertainty and possibly M&A failure. There is some evidence that geographical proximity is a driver of M&A within countries (see e.g. Rodriguez-Pose and Zademach, 2003), but we do not know whether that also applies to clusters. And M&A activity might also be induced by strategies to enlarge the firm‟s market geographically, by taking over distant firms in markets the acquiring firm was not yet active. Cluster firms might have a strong incentive to take over non-cluster firms, because the competitive pressure is much higher in clusters, as outlined in section 2.1. This may be reinforced by that fact that clustering leads to the local emergence of specialized services like consultants and lawyers

specialized in M&A, that may further boost M&A by cluster firms. As a result, we expect that cluster firms are disproportionally more active in taking over other firms (both in their own region and outside), as compared to non-cluster firms. But where we expect that cluster firms show lower exit rates due to failure (hypothesis 1), we do not expect that cluster firms have lower failures rates due to acquisition. This is because, on the one hand, cluster firms are likely to be targets of acquisitions, but on the other hand, we expect that cluster firms are also active in taking over non-cluster firms. We formulate two hypotheses: Hypothesis 4a: cluster firms are disproportionally more active in taking over other firms Hypothesis 4b: cluster firms do not show higher exit rates due to acquisition No study so far has tested whether spinoffs are more likely to be initiator or target of acquisitions. At this stage, we do not have theoretical expectations. However, there is a tendency of successful firms to be more active in taking over other firms (Fein, 1998). And since our hypothesis 2a states that spinoffs are more likely to be successful, one could possibly argue that spinoffs are expected to be disproportionally more active in taking over other firms. And since there is no tendency of acquired firms to be more successful (Fein, 1998), we could expect that spinoff companies are not acquired relatively more often than other firms, especially with respect to inexperienced firms. This leads to two hypotheses. Hypothesis 5a: spinoffs are disproportionally more active in taking over other firms Hypothesis 5b: spinoffs do not show lower exit rates due to acquisition The second reason to account for the role of M&A activity is that the acquisition of a firm may be considered a form of post-entry learning, in which acquiring firms get access to the knowledge of acquired firms (Ahuja and Katila, 2001; Piscitello, 2004; Cassiman et al., 2005). In that respect, acquiring firms may increase their capabilities and get better routines over time. In addition to that, M&A activity is full of uncertainty and may lead to failure (Fein, 1998). Having experience in acquiring other firms might be crucial to ensure that the acquisition is well implemented and brings value to the acquiring firm. Therefore, we expect that the more experienced a firm is in doing acquisitions, the better it will perform. There is no reason to believe this is different in the case of more experienced firms in M&A located in clusters. However, we expect that more experienced firms in M&A are not necessarily more often targets of takeovers. This follows a similar line of reasoning as in hypothesis 5b, in which spinoffs are expected not to have lower exit rates due to acquisition. We formulate the following hypotheses: Hypothesis 6a: more experienced firms in M&A show lower failure rates Hypotheses 6a: more experienced firms in M&A in clusters show lower failure rates Hypothesis 7a: more experienced firms in M&A do not show lower exit rates due to acquisition Hypotheses 7b: more experienced firms in M&A in clusters do not show lower exit rates due to acquisition

3. Evolution of the Dutch banking sector Since localisation economies, spinoff dynamics and M&A activity may each play their role in the spatial evolution of an industry, it is up to empirical research to disentangle these and assess their importance. We analyse the Dutch banking sector during the period 1850-1993 by means of a survival method, following Klepper, among others. For that purpose, we collected data on the years of entry and exit of each bank that entered the industry in the Netherlands during that period, the location of the head office, merger and acquisition activity in the banking sector, and the pre-entry industrial background of the entrepreneur. We collected data from a number of sources (see for more details Boschma and Ledder, 2010). We used the so-called Nederlandse financiële instellingen in de twintigste eeuw: balansreeksen en naamlijst van handelsbanken published by the Dutch Central Bank to compile a list of banks that were active in the period 1850 to 1993. This source lists every bank in the Netherlands for this period, the years they were in business, the location of their headquarters, changes in ownership structure and reorganisations. Other sources we used were the Nationale Vereniging van Banken and the Nederlandsch Economisch-Historisch Archief. Information on the pre-entry industrial background of the entrepreneurs was acquired from Geschiedenis van de Algemene Banken in Nederland 1860-1914 (Kymmell 1992, 1996), Geschiedenis van de Nederlandsche Bank (De Jong, 1967), the online databank on Dutch entrepreneurs of the Internationaal Instituut voor Sociale Geschiedenis, trade journals, archives of cities and regions, and chronicles on the history of particular banks. Our data sources cover the period 1850-1993. This implies we cannot analyse the full life cycle of the Dutch banking sector, which is, of course, much older. In fact, Amsterdam was a leading international financial centre in the seventeenth century (Israel, 1995). Consequently, our study covers only part of the life cycle of the Dutch banking industry but nevertheless a most interesting part. Before 1860, there did not exist a modern banking sector in the Netherlands, although there certainly was a money and stock market. This changed in the early 1860s, when the first banks with a juridical structure of a limited liability company were created. Having large sums of capital to invest was new to the Dutch banking system. In our database, we have information on headquarters of banks, not branches. This means we deal with the most knowledge-intensive part of this service industry where firmspecific routines are formed. Our database counts 906 banks that entered the Dutch banking sector in the period 1850-1993. For 112 banks, the year of entry is unknown. Of these 906 banks, 779 banks had to exit the banking sector in the period 1850-1993, 119 banks were still active in 1993, and for 8 banks, the year of exit is unknown. About half of the exits (i.e. 394 exits) were caused by bankruptcy, closure, diversification into other activities than banking, et cetera. The remaining part of the exits in the Dutch banking sector (385 banks) could be attributed to merger and acquisition activity. Figure 1 presents the evolution of the Dutch banking sector in terms of total numbers of entries, exits and firms in the period 1850-1993 (for details, see Boschma and Ledder, 2010). Before describing this pattern, we have to remind that the number of firms and number of exits are underestimated in the first decades after 1850, because we do not have information on banks that were founded before 1850. Figure 1 shows that, except for a short intermezzo during the First World War, the total number of banks increased till 1929, when a maximum of 478 banks was reached. What is remarkable is that the number of exits was extremely low in the second half of the nineteenth century. Entry levels were a bit higher but also remained low till the 1890s. This has been attributed to, among other reasons, the low tendency of firms to lend money from banks, because in the second half of the nineteenth century, that was considered a sign of weakness (Nierop, 1972). Since the 1890s, however,

there has been a sharp and steady increase in the number of entrants, until the 1930s, when entry levels dropped sharply and remained low ever since. The number of exits also started to increase around the turn of the century, but especially in the 1920s and early 1930s. At the turn of the century, the industry was already dominated by five banks: Nederlandsche-Handelmaatschappij, Twentsche Bank, Rotterdamsche Bank, Amsterdamsche Bank and Incasso Bank. In 1900, their total market share was 35 per cent, which rose further to 48 percent in 1918, but fell down again to 38 per cent in 1928. In 1930, the number of exits overtook the number of entrants and the shakeout of the industry started. In 1940, the market share of the big five had risen to 52 per cent (Kymmell, 1996). The declining trend in the number of firms decelerates in the 1970s. In the 1970s, there is a short increase of exit levels, after which the number of exits stabilizes at a low level. In 1993, there were 119 banks left, still a considerable number. By that time, the Dutch banking sector had evolved into an oligopoly dominated by three banks (ABN-AMRO, ING Group and Rabobank), which had a market share of 80 per cent (Van der Lugt, 1999; Bos, 2004). Figure 1. The number of firms, entrants and exits in the Dutch banking sector, 1850-1993

Figure 2 describes the number of exits due to M&A activity and their share in the total number of exits in the Dutch banking sector in the period 1850-1993. As explained earlier, about half of all exits were caused by mergers and acquisitions over the whole period, which is extremely high in comparison to other industries, like automobiles, where it is only 5 per cent. M&A activity was highest in the period 1914-1929, during which numerous small, mostly regional banks were taken over (Bosman, 1989). M&A activity slowed down after that, until a second wave of M&A occurred in the 1960s. Figure 2. The number of M&A exits in the Dutch banking sector, 1850-1993

In order to sketch the spatial evolution of the Dutch banking industry, we have assigned the location (municipality) of all banks to one of the 40 labour markets (COROP regions) in the Netherlands. In the very exceptional case that a bank moved from one region to another, we assigned the bank to the region where it had been active for most of its time. In Figure 3, we have depicted the evolution of the number of Amsterdam-based banks and banks located outside the Amsterdam region for the period 1850-1993. In Figure 4, we show the share of the four major bank regions of the Netherlands (i.e. the Amsterdam region, Rotterdam region, Utrecht region and The Hague region)1 in the national total for that same period. Figure 3. Number of banks in and outside Amsterdam, 1850-1993

1

Regions are so-called COROP regions which correspond to labour market areas in the Netherlands. For instance, COROP region Groot-Amsterdam includes the city of Amsterdam and surrounding municipalities like Aalsmeer, Amstelveen, Diemen, Edam-Volendam, Haarlemmermeer, Purmerend and Uithoorn.

1990

1980

1970

1960

1950

1940

1930

1920

1910

1900

1890

1880

1870

1860

1850

Number of firms s

350 300 250 200 150 100 50 0

Time (in years) No. of Amsterdam firms

No. of non-Amsterdam firms

Figure 4. Shares of the four major urban Dutch regions in total number of banks, 1850-1993 100% 80% 60% 40% 20%

Amsterdam

The Hague

Rotterdam

Utrecht

1985

1976

1967

1958

1949

1940

1931

1922

1913

1904

1895

1886

1877

1868

1859

1850

0%

Elsewhere

As mentioned before, both figures include banks that entered only after 1850, so the findings in the first decade after 1850 should be treated with caution. Figure 3 shows a steady increase of the number of banks in the Amsterdam region. Around 1900, as mentioned before, the Dutch banking sector was dominated by five large banks, of which four were based in the Amsterdam region. The increase in the number of banks in Amsterdam accelerated in the 1920s. After reaching a peak in 1930, a decline set in until the late 1950s, after which the number of banks stabilized at a level of about 60-80 banks till 1993. In relative terms, as shown in Figure 4, the share of the Amsterdam region in the total number of banks dropped from 38 per cent in the early 1860s to a mere 19 per cent in 1915. This was not so much caused by exits of banks located in Amsterdam, but by a relative increase of the shares of the Rotterdam region (in the late 1860s) and the Hague region (in the 1900s). Till far in the 1910s, the majority of Dutch banks was still located outside the 4 major urban areas of the Netherlands. This picture changed after 1915, when 116 banks were founded in the Amsterdam region in just a period of 15 years, which increased its share to almost 35 percent in 1930. This share stabilized for almost forty years, till foreign banks

started to enter the Netherlands. In combination with exits that occurred mainly in the rest of the Netherlands, Amsterdam increased its share in terms of number of firms to around 56 percent in 1993, but in terms of market share, the concentration of bank activity in the Amsterdam region was much higher than that (Sluyterman et al., 1998). In the next section, we investigate more in detail this clustering process of the Dutch banking sector in the Amsterdam region. Among others, we test whether this spatial clustering process has been caused by (1) a high intensity of spinoff activity (e.g. did Amsterdam banks give birth to many and successful spinoffs in the Amsterdam region?), (2) a high intensity of M&A activity (e.g. were Amsterdam banks extremely active in taking over banks elsewhere in the Netherlands?), or (3) location-specific qualities of the Amsterdam region (e.g. did clustering bring benefits to Amsterdam banks over time?). 4. Empirical findings 4.1 method We follow Klepper (2007) and others and employ duration analysis to estimate the effects of the location (like localisation economies), the pre-entry background of entrepreneurs (like spinoffs) and the experience of firms in M&A activity (cumulative experience in doing acquisitions) on the probabilities of firms to exit. We make use of a competing risk model. This allows us to distinguish between the effects of the independent variables on the probability of exit when exit can assume alternative modalities. As described earlier, we found that about half of the Dutch banks exited due to failure, and the other half exited due to acquisition by another bank in the period 1850-1993. In the first case, the bank disappears from the market either by the actual closure of a bank or by the termination of banking activities by a firm. In the second case, the bank is partially or wholly integrated in another bank. In the previous section, we formulated hypotheses that differentiate between the two types of exits. By means of a competing risk model, we can actually test whether different factors affect the two types of exits. The duration variable is the length of time from the year of entry of the bank to the year of exit. The dependent variable distinguishes between three categories: exit by failure, exit by acquisition, and when a firm has survived. The survival time is measured in years and is right-censored for banks that have survived until 1993. We allow for the independent variables to vary during the time period of observation. Each year a bank exists is considered as an observation (at time period t), and we estimate the probability the bank survives the next year (at time period t+1), based on characteristics of the bank and its location at time period t. Therefore, we take into account changes in regional characteristics (such as number of inhabitants, or number of local banks) and changes in firm characteristics (such as firm age, or the cumulative number of acquisitions undertaken by a firm). Other covariates do not change over time, such as the entrepreneurial background of the firm (e.g. being a spinoff or not). We run the estimations with 716 banks for which we had information on all variables. Because we have a large number of banks with a considerable average age, the total number of observations is 27,222. Because of the discrete nature of the survival time in our data, we applied the discretetime method proposed by Allison (1982) and extended by Jenkins (1995; 2004) to estimate the parameters of the model. Jenkins (2004) proves that the multinomial logit model represents an approximation in discrete time of a duration model. Specifically, the model assumes that each type of event takes place in continuous time between discrete observations,

with specific hazard rates that remain constant within intervals. As such, we apply a discretetime model with a multinomial logit formulation2. 4.2 main findings The findings of the multinomial logistic regression analysis are presented in Table 1. In the different models, we test our hypotheses by comparing the probability of exit due to failure and the probability of exit due to acquisition in a year with the group of firms that survived a year. In all our estimations, we controlled for firm age denoted as ln(t) in Table 1. Concerning this variable, our findings show that firm age appears to lower failure rates of banks, but it increases the probability to exit due to acquisition. In other words, younger firms have higher failure rates, whereas older firms have a higher probability to become acquired by another bank. Apart from the fact that these results are interesting by itself, they also how relevant it is to distinguish between the two types of exits. First, we tested whether location matters for survival in the banking sector. As explained earlier, we have only data on headquarters of banks, not on their branches. We included two location-specific variables in our estimations. The first variable takes up the effect of localisation economies (Loc.econ), which has been measured on a logarithmic scale as the number of banks in each region3 for each year in the period 1850-1993. Since we have no data on branches of banks, this variables does not take up the effect of local competition. The second variable estimates the cluster effect of the Amsterdam region. A dummy variable Amsterdam has been included to see whether being located in the Amsterdam banking cluster affects negatively the survival of banks. Table 1 (part I). Estimates of the multinomial logistic regression analysis (outcome “survived” is the comparison group) Model 1 Model 2 Dependent Dependent Dependent Dependent variable: variable: variable: variable: Probability Probability Probability Probability to exit due to exit due to exit due to exit due to failure to acquisition to failure to acquisition Amsterdam Loc.econ. Spinoff Experienced Foreign

0.440* (0.208) -0.026 (0.080) -0.574** (0.154) -1.078** (0.136) -0.998 (0.514)

0.156 (0.242) -0.405** (0.081) -0.247 (0.174) -0.752** (0.140) -0.057 (0.381)

0.771** (0.223) -0.208* (0.083) -0.855** (0.159) -1.177** (0.138) -1.183* (0.517)

0.560* (0.255) -0.572** (0.082) -0.359* (0.177) -0.710** (0.142) 0.026 (0.382)

2

Weterings and Marsili (2010) have applied a similar model in their empirical study of the spatial concentration of industries and the probability for firms to exit due to failure or acquisitions. 3

We have used the so-called COROP regions in the Netherlands which correspond more or less to labour market areas at the NUTS3 level.

Cohort 1 (1850-1913) Cohort 2 (1914-1929) Cohort 3 (1930-1945) Amsterdam *Spinoff Amsterdam *Experienced Amsterdam *Cohort 1 Amsterdam *Cohort 2 Ln_cumM&A

-1.561** (0.200) -0.943** (0.193) 0.010 (0.144)

ln(t)

-0.150** 0.258** (0.057) (0.068) -2.378** -3.217** (0.302) (0.330) 463.05** -3088.324 27222

0.068 0.435** (0.053) (0.065) Constant -4.065** -4.459** (0.248) (0.283) LR chi2 194.32** Log likelihood -3222.692 N 27222 ** significant at 0.01 level * significant at 0.05 level

-2.488** (0.325) 0.458** (0.151) -0.651* (0.197)

Table 1 (part II). Estimates of the multinomial logistic regression analysis (outcome “survived” is the comparison group) Model 3 Model 4 Dependent Dependent Dependent Dependent variable: variable: variable: variable: Probability Probability Probability Probability to exit due to exit due to exit due to exit due to failure to acquisition to failure to acquisition Amsterdam Loc.econ. Spinoff Experienced Foreign Cohort 1 (1850-1913) Cohort 2 (1914-1929)

1.430** (0.314) -0.232** (0.084) -0.585** (0.205) -1.020** (0.165) -1.198* (0.517) -1.372** (0.225) -0.759* (0.219)

0.848 (0.482) -0.569** (0.082) -0.202 (0.191) -0.764** (0.153) -0.022 (0.382) -3.036** (0.466) 0.601** (0.161)

1.425** (0.314) -0.228** (0.084) -0.590** (0.205) -1.011** (0.165) -1.199* (0.517) -1.378** (0.224) -0.756** (0.220)

0.854 (0.484) -0.581** (0.083) -0.197 (0.191) -0.776** (0.154) -0.047 (0.383) -3.032** (0.466) 0.601** (0.162)

Cohort 3 (1930-1945) Amsterdam *Spinoff Amsterdam *Experienced Amsterdam *Cohort 1 Amsterdam *Cohort 2 Ln_cumM&A ln(t)

0.002 (0.145) -0.752* (0.314) -0.577* (0.293) -0.678 (0.408) -0.611 (0.379)

-0.648** (0.197) -0.565 (0.492) 0.175 (0.455) 1.520* (0.638) -1.140** (0.439)

-0.146* 0.270** (0.057) (0.069) Constant -2.501** -3.310** (0.308) (0.334) LR chi2 492.81** Log likelihood -3073.447 N 27222 ** significant at 0.01 level * significant at 0.05 level

-0.017 -0.630** (0.145) (0.197) -0.719* -0.606 (0.314) (0.494) -0.508 0.101 (0.293) (0.458) -0.715 1.591* (0.408) (0.640) -0.625 -1.089* (0.379) (0.441) -0.555* 0.272* (0.264) (0.122) -0.127* 0.249** (0.058) (0.070) -2.550** -3.234** (0.309) (0.335) 503.57** -3068.067 27222

Our estimations seem to confirm hypothesis 1. The estimations show that localisation economies has a negative effect on the hazard rate, while the Amsterdam dummy has a positive effect. In other words, the more local banks are around in a region in general, the higher the survival rate of banks, with the exception of the Amsterdam region, where the survival rate of banks is lower45. So, instead of increasing survival rates of banks because of being located in the main banking cluster, it actually increased failure rates of banks. And while our estimations show that Amsterdam banks show higher exit rates due to failure, they have not higher exit rates due to acquisitions. This confirms hypothesis 4b. Our second set of independent variables measures the effect of having a pre-entry background as entrepreneur in the same or related industries. Following Klepper (2007), firms were classified as spinoffs when at least one of the founders had worked for or had founded a bank previously. When an entrepreneur had previous experience in several firms, the last firm he or she worked for was considered the parent of the spinoff. Some spinoffs had multiple founders that had worked for different firms. In that case, the parent of the spinoff was assigned to the founder that was described as the most influential in the new spinoff company. Of all 716 banks that entered the banking sector in the period 1850-1993, 204 firms were defined as spinoffs. A second category of firms has been defined as experienced firms when 4

We also included dummies for the other major cities in the Netherlands, that is Utrecht, Rotterdam and The Hague, which also showed relatively high concentrations of banks, and these location-dummies turned out to show the same results as for the Amsterdam cluster. 5

We also constructed a variable to assess the effect of urbanization economies, which captures all the effects of being located in an urban environment. This might especially benefit services firms like banks since they may be more dependent on a big local market than manufacturing firms. This variable has been measured as the log of the number of inhabitants per squared km in a region, which is provided for all COROP regions in the Netherlands by the Central Bureau of Statistics for each year in the period 1850-1993. Due to very high correlation with the variable localization economies, we decided to leave out this variable in all estimations.

they had prior experience in related activities. Related activities have been defined as financial services in general. In the nineteenth century, these related activities concerned cashiers, bankers and stock-brokers (Kymmell, 1992). Cash could be obtained from a cashier, banker or stock-broker through the selling of not-due claims of bills of exchange and/or promissory notes, or through making a loan with securities or personal properties. None of these occupations had as their main activity the provision of credit (see for details, Boschma and Ledder, 2010). We counted a total of 288 experienced firms that entered banking activities in the period 1850-1993. These include both diversifiers and entrants that set up de novo banks founded by heads of firms in related activities. The third type of firms concerns 224 inexperienced entrepreneurs with no prior experience in banking and no experience in related industries6. In Figure 5, the evolution of the number of the three types of banks are shown for the period 1850-1993. Figure 5. Evolution of the accumulated number of the three types of banks, 1850-1993

200 150 100 50

No. of spinoff firms No. of inexperienced firms

1993

1982

1971

1960

1949

1938

1927

1916

1905

1894

1883

1872

1861

1850

0

No. of experienced firms

In model 1, we have included the dummy variables spinoffs and experienced firms, with inexperienced firms as the omitted reference category. Our findings confirm our expectations laid down in hypotheses 2a and 2b: spinoffs and experienced banks do indeed show lower failure rates. In other words, having a pre-entry entrepreneurial background in the same industry (banking) and related industries (financial services) increases the survival of banks. We also included a Foreign Bank dummy variable, because this might be considered a particular type of spinoff companies, whose survival (next to pre-entry learning) also depends on the strategy of the foreign multinational bank located elsewhere. Our estimations show that foreign banks have indeed lower exit rates due to failure: the coefficient for Foreign Bank is negative and significant. If we compare these results for the two types of exits, we show remarkable differences. While being a spinoff or a foreign bank lowers exit rates due to 6

For the 898 entrants with years of entry and exit known, we could find information on the pre-entry background of the founder for 736 banks. So, the group of entrants with an unknown entrepreneurial background (162 banks) is relatively small, in comparison to other survival studies. This group of entrants has been excluded from the analyses. We found that these had a shorter life span than the entrants with a known background. This makes sense, because one expects little information is available for banks that existed only a few years. On 3 of these 736 banks with a known entrepreneurial background, data on the year of exit were missing.

failure, no such effect shows up when the probability to exit due to acquisition is concerned. In other words, spinoffs and foreign firms are not more likely to be targets of acquisitions. This confirms hypothesis 5b. The third set of independent variables concerns time of entry which is included as control variable. According to Klepper (2007), early entrants in a new industry outperform late entrants because of weaker selection. Although we do not cover the early stage of the life cycle of the banking sector, we also expect banks that entered in the second half of the nineteenth century to have a lower hazard rate at every age, because the banking sector was by then characterized by small banks and low levels of competition, which allowed banks to grow and expand. To account for this early entry effect, we defined a first cohort of banks that entered the period 1850-1913. Then, we defined a second cohort of firms that entered the banking sector during a very turbulent phase in the Dutch banking sector in the period 19141929, during which scale economies grew in importance, entry barriers rose, and M&A activity was very intense. The third cohort captures firms that entered Dutch banking in the period of the Great Depression and World War II. Cohort 4 covers entrants in the post-WWII period. We treated Cohort 4 as the reference category. From model 2 in Table 1 onwards, we have included dummy variables for Cohorts 1, 2 and 3, with Cohort with entrants 1945-1993 as the reference category. The estimations show that banks belonging to the Cohorts 1 and 2 show lower failure rates: their coefficients are highly significant with a negative sign. Our findings also show that banks from Cohort 2 had a higher probability to exit due to acquisition, while Cohorts 1 and 3 had lower probabilities to become victim of takeovers. In model 3, we investigated more in detail the effects of the pre-entry background of entrepreneurs and the time of entry in the case of Amsterdam banks. The estimations demonstrate that the interaction variables Amsterdam*Spinoffs and Amsterdam*Experienced are highly significant with a negative sign, meaning that spinoffs and (to a lesser extent) experienced firms in the Amsterdam cluster showed lower failure rates, while banks in the Amsterdam cluster in general showed higher failure rates. This finding is in line with hypothesis 3. These results indicate that not only do spinoffs outperform other types of entrants, but also that spinoffs in the Amsterdam cluster performed better. In addition, we can read in Table 2 that the Amsterdam cluster also was very successful in generating many spinoffs: 53 per cent of all spinoffs located in the Amsterdam region, and 80 per cent of all foreign banks located in the Amsterdam region. Thus, the Amsterdam cluster did not only generate more spinoffs than any other region in the Netherlands, spinoffs in the Amsterdam cluster also performed better. Table 2.Type of pre-entry background in Amsterdam and elsewhere Type of background No. of firms Share of firms Spinoff 204 28.5% - in Amsterdam 108 52.9% - elsewhere 96 47.1% Experienced firm 288 40.2% - in Amsterdam 95 33.0% - elsewhere 193 67.0% Inexperienced firm 224 31.3% - in Amsterdam 34 15.2% - elsewhere 190 84.8% Total 716 100.0%

We measured the number of M&A done by each bank per year. This allows us to proxy the effect of firm learning by doing M&A. For each year a bank exists we counted the number of banks it acquired, including all banks it acquired in previous years (cumM&A). This cumulative number is transformed on a logarithmic scale to take diminishing returns to scale into account. Amsterdam firms appear to be disproportionally more active in taking over other firms 5. Conclusions This paper has made an attempt to describe and explain the spatial evolution of the banking industry in the Netherlands, and more in particular, the spatial clustering of Dutch banking in the Amsterdam region, for a period of almost 150 years. We found strong evidence that spinoff companies and experienced firms had a higher survival rate. This is in line with previous studies that attach importance to the pre-entry experience of entrepreneurs in the same industry (banking) and related industries (financial services). In addition, we found evidence that the clustering process in the Amsterdam region might have been caused by a high intensity of spinoff activity: Amsterdam banks did not only give birth to a disproportionally high number of spinoffs in the Amsterdam region, but these Amsterdambased spinoff companies also performed better in the long run. And the Amsterdam region also attracted a disproportionally high number of foreign banks, which, in general, had higher survival rates than domestic banks. We also found evidence that the banking sector showed an unprecedented high number of exits caused by M&A activity. This may provide an additional explanation for the spatial clustering of an industry that has not been taken up in existing studies. Our study could indeed demonstrate that Amsterdam banks were extremely active (disproportionally so) in taking over banks elsewhere in the Netherlands. More in particular, we could show that only a small number of banks were responsible for the lion share of all acquisitions, and these banks were almost without exception Amsterdam-based. This was especially true in the later part of the industry life cycle when the Amsterdam region rapidly increased its share in the Dutch banking industry. Based on these findings, it seems plausible that M&A activity contributed to the further spatial clustering of banking in the Amsterdam region in the twentieth century. In that respect, M&A activity delineates a lineage structure between firms that cross regional boundaries and lowers the number of firms in an industry over time. This is opposite to the lineage structure caused by the spinoff process which is mainly intra-regional and leads to an increasing number of firms over time, but which also adds to spatial clustering. Overall, these results tend to show that entry levels are affected positively in the Amsterdam cluster, but exits levels as well. In other words, the Amsterdam cluster attracted high numbers of entrants, but simultaneously, it was also a harsh environment for banks in general to survive. A key finding was that the Amsterdam region was able to attract a disproportionate number of spinoff companies and late entrants that performed relatively well in general. But we found weak evidence that Amsterdam-based spinoffs performed better than spinoffs located elsewhere, and no evidence that Amsterdam-based late entrants performed better than late entrants located elsewhere. Above all, Amsterdam-based banks were very successful in taking over other banks, both locally and non-locally. It is not unlikely that the Amsterdam location may have contributed to this disproportionate M&A activity, due to the local presence of services specialized in M&A and other financial organizations like the Dutch Central Bank.

This project opens up many research challenges. First of all, there is a need to replicate this study in other countries where banking is spatially concentrated, like the US (New York), the UK (London) and Germany (Frankfurt). In this context, one should also account for the internationalization of the banking sector, and how that has affected the evolution of the Amsterdam banking cluster (see e.g. Engelen and Grote, 2009). Given the overall importance of M&A activity as driver of spatial clustering, future research should concentrate more on how routines are transferred from acquired to acquiring firm, because this has remained a black box in our study, due to a lack of data. This is likely to differ between firms, depending on their M&A strategy. For instance, it might be the case that M&A in banking is much less driven by getting access to successful routines of other firms, as in high-tech sectors, for instance, and much more by conquering market shares from competitors. And last but not least, it would also be interesting to include network effects on the performance of banks. Such a study could look at the evolution of interlocks between banks over time, and determine what are the main drivers (like geographical proximity) behind the network dynamics concerning interlocks. Interlocks between banks are also probably good predictors of which banks will be acquired by which bank, and therefore may affect the long-term survival of banks. Such an approach would include network dynamics in the study of the life cycle of clusters, which is a challenging topic by itself, and which would add to the explanatory framework that analyzes the spatial evolution of industries and the life cycle of clusters from an evolutionary perspective (Ter Wal and Boschma, 2010). This study has also implications for the study of cluster life cycles, which is the central focus of this special issue. Since Porter (1990), clusters have attracted widespread attention in economic geography. This literature is, however, not unproblematic (see e.g. Martin and Sunley 2003). One of the critiques is that clusters have been treated as static entities, as if clusters do not evolve, and as if their existence can be explained by looking at their current features. This issue of the dynamics of cluster evolution has been taken up by scholars recently (Audretsch and Feldman, 1996; Feldman and Schreuder, 1996; Staber, 1997, 2001; Maggioni, 2002; Brenner, 2004; Feldman et al., 2005; Iammarino and McCann, 2006; Staber, 2009; Menzel and Fornahl, 2009; Ter Wal and Boschma, 2010). Although this emerging literature on cluster life cycles has brought new and valuable insights, broadly speaking, it still suffers from a number of weaknesses: (1) this literature has remained rather conceptual; (2) it does not depart from a common theoretical framework; (3) case studies on life cycles of clusters have remained rather descriptive; and (4) when explaining the long-term evolution of clusters, they do not test for explanations other than Marshallian externalities. This paper aims to take up these issues when explaining the dynamics of the Amsterdam banking cluster since 1850. Unfortunately, our study could not trace the parents for any spinoff companies because of a lack of information. This a major drawback of our study which should be included in any study on this topic. In our study, we could show some evidence of the role of mergers and acquisition, but we could not determine whether that is caused by a size effect (economies of scale) or a postentry learning effect, in which acquiring firms get access to the knowledge of acquired firms (Ahuja and Katila, 2001).

No study so far has tested whether spinoffs are more likely to be initiator or victim of acquisitions. Therefore, we did not have any theoretical expectations at this stage, but our study showed interesting findings. This should be theorized more carefully. What is also left to empirical studies is the relationship between parent and spinoff in terms of M&A activity. Are parents more inclined to take over their spinoff firms, and./or do spinoff firms take over their parent? This could be expected because of their social proximity. This could also play a role in the de/branching process outlined above. That is, many spinoffs are generated by a small number of successful parents, while these successful parents acquire these spinoff companies again at a later stage. A perfect mirror image arises here. We could not test this here, because of limited information on the parents of spinoffs. From such an industry life cycle perspective, M&A activity can be viewed as a de-branching process in which firm-specific routines merge, and the number of firm-specific routines in the industry decreases. This is depicted on the right side of Figure. Through M&A activity, a lineage structure between firm-specific routines across space is formed as time goes by, as knowledge and routines are transferred from acquired to acquiring firms. The spinoff process also contributes to the evolution of this lineage structure, as shown on the left side of Figure. However, the spinoff process sets into motion a branching process in which routines are transferred from parents to spinoff firms, and which makes the number of firm-specific routines increase over time. As explained above, both knowledge transfer mechanisms are likely to contribute to the spatial concentration of an industry. This is because the spinoff process is a self-reinforcing and path-dependent process that occurs at the regional level, in which a relatively small number of parent organizations give birth to a relatively large number of (successful) spinoffs. With respect to M&A activity, this is because intra-regional M&A‟s will primarily occur within clusters, while inter-regional M&A‟s will concern mainly cluster firms that acquire non-cluster firms. Spatial clustering is further reinforced by the fact that only a small number of cluster firms will do most of the acquisitions because of cumulative learning and internal scale economies. Figure Branching (through spinoff process) and de-branching (through M&A activity) of organizational routines

branching

time

de-branching

firm

Bibliography Ahuja, G. and R. Katila (2001), Technological acquisitions and the innovation performance of acquiring firms: a longitudinal study, Strategic Management Journal 22, pp. 197–220 Allison, P.D. (1982) Discrete-Time Methods for the Analysis of Event Histories, Sociological Methodology 13: 61-98. Arthur, W.B. (1994), Increasing Returns and Path Dependence in the Economy, Ann Arbor: The University of Michigan Press. Audretsch, D. B. and M.P. Feldman (1996), Innovative clusters and the industry life cycle, Review of Industrial Organization, 11, 253-273. Bos, J.W.B. (2004), Does market power affect performance in the Dutch banking market? A comparison of reduced form market structure models, De Economist, 152 (4), 491-512. Boschma, R.A. and K. Frenken (2003), Evolutionary economics and industry location, International Review for Regional Research, 23, 183-200. Boschma, R.A. and J.G. Lambooy (1999), Evolutionary economics and economic geography, Journal of Evolutionary Economics, 9, 411-429. Boschma and Ledder (2010), The evolution of the banking cluster of Amsterdam 1850-1993. A survival analysis, in: D. Fornahl, S. Henn and M.P. Menzel (eds.), The emergence of

clusters. Theoretical, empirical and political perspectives on the initial stage of cluster evolution, Cheltenham: Edward Elgar. Boschma and Wenting (2007), The spatial evolution of the British automobile industry: Does location matter? Industrial and Corporate Change, 16 (2), 213-238. Bosman, H.W.J. (1989), Het Nederlandse Bankwezen. Serie bank- en effectenbedrijf nr. 1. Amsterdam: Nederlands Instituut voor het Bank- en Effectenbedrijf Brenner, T. (2004), Local Industrial Clusters. Existence, Emergence and Evolution. London and New York: Routledge. Buenstorf, G. and S. Klepper (2009), Heritage and agglomeration. The Akron tyre cluster revisited, The Economic Journal, 119 (537), 705-733. Cantner, U., K. Dressler and J.J. Krueger (2006), Firm survival in the German automobile industry, Empirica, 33, 49-60. Carree, M.A. (2003) A hazard rate analysis of Russian commercial banks in the period 19941997, Economic Systems, 27, 255-269. Cassiman, B., M.G. Colombo, P. Garrone and R. Veugelers (2005), The impact of M&A on the R&D process. An empirical analysis of the role of technological and market relatedness, Research Policy, 34 (2), pp. 195-220. Chapman, K. (1991) The International Petrochemical Industry. Evolution and Location, Oxford: Basil Blackwell. Consoli, D. (2005) The dynamics of technological change in UK retail banking services. An evolutionary perspective, Research Policy 34: 461-480. Dahl, M.S., C. R. Pedersen and B. Dalum (2003), Entry by spinoff in a high-tech cluster, DRUID working paper, 3-11. Engelen, E. and M.H. Grote (2009), Stock exchange virtualisation and the decline of secondtier financial centres. The cases of Amsterdam and Frankfurt, Journal of Economic Geography, 9: 679-696. Fein, A.J. (1998), Understanding evolutionary processes in non-manufacturing industries: Empirical insights from the shake-out in pharmaceutical wholesaling, Journal of Evolutionary Economics, 8 (3), 231-270. Feldman, M.P., J. Francis, and J. Bercovitz. (2005), Creating a cluster while building a firm. Entrepreneurs and the formation of industrial clusters, Regional Studies, 39, 129-141. Feldman, M. P. and Y. Schreuder (1996), Initial advantage. The origins of the geographic concentration of the pharmaceutical industry in the Mid-Atlantic region, Industrial and Corporate Change, 5, 839-862. Grote, M.H. (2008), Foreign banks‟ attraction to the financial centre Frankfurt. An inverted U-shaped relationship, Journal of Economic Geography 8: 239-258.. Helfat, C.E. and M.B. Lieberman (2002), The birth of capabilities. Market entry and the importance of pre-history, Industrial and Corporate Change, 11 (4), 725-760. Iammarino, S., McCann, P. (2006) The structure and evolution of industrial clusters. Transactions, technology and knowledge spillovers, Research Policy, 35 (7): 10181036.

Israel, J. (1995), The Dutch Republic: Its Rise, Greatness and Fall, 1477-1806, Oxford: Oxford University Press. Jenkins, S.P. (1995) Easy estimation methods for discrete-time duration models. Oxford Bulletin of Economics and Statistics 57(1): 129-138. Jenkins, S.P. (2004) Survival analysis, unpublished manuscript. Institute for social and economic research, University of Essex, Colchester, UK. Downloadable from: http://www.iser.essex.ac.uk/teaching/degree/stephenj/ec968/pdfs/ec968lnotesv6.pdf Jong, A.M.de (1967), Geschiedenis van de Nederlandsche Bank: van 1814 tot 1964. Part I-V, Haarlem: Joh. Enschede and Zonen. Jong, H.W. de (1981) Dynamische markttheorie, 2nd revised version, Leiden/Antwerp: Stenfert Kroese. Klein, J.P. and M.L. Moeschberg (1997), Survival Analysis: Techniques for Censored and Truncated Data, New York: Springer-Verlag. Klepper, S. (1997), Industry life-cycles, Industrial and Corporate Change, 6 (1), 145-182. Klepper, S. (2007), Disagreements, spinoffs and the evolution of Detroit as the capital of the U.S. automobile industry, Management Science, 53 (4), 616-631. Koster, S. (2006), Whose child? How existing firms foster new firm formation: individual start-ups, spin-outs and spin-offs, dissertation, University of Groningen, Faculty of Spatial Sciences: Groningen. Kymmell, J. (1992), Geschiedenis van de Algemene Banken in Nederland 1860-1914. part I. Amsterdam: Nederlands Instituut voor het Bank- en Effectenbedrijf Kymmell, J. (1996), Geschiedenis van de Algemene Banken in Nederland 1860-1914. part II. Amsterdam: Nederlands Instituut voor het Bank- en Effectenbedrijf. Lugt, Van der, J.A. (1999), Het Commerciële Bankwezen in Nederland in de Twintigste Eeuw, Een Historiografisch Overzicht, NEHA Jaarboek. Maggioni, M.A. (2002), Clustering Dynamics and the Location of High-Tech-Firms, Heidelberg: Springer Verlag. Markusen, A. (1985), Profit Cycles, Oligopoly and Regional Development, Cambridge: MIT Press. Martin, R. and Sunley, P. (2003) Deconstructing clusters: chaotic concept or policy panacea?, Journal of Economic Geography, 3 (1): 5-35. Menzel, M.P. and D. Fornahl (2009), Cluster life cycles. Dimensions and rationales oc cluster evolution, Industrial and Corporate Change, doi: 10.1093/icc/dtp036. Myrdal, G. (1957), Economic Theory and Underdeveloped Regions, London: Duckworth. Nierop, H.A. (1972), Schets van het Bankwezen, Haarlem: De Erven F. Bohn N.V. Norton, R.D. (1979), City Life-Cycles and American Urban Policy, New York: Academic Press. Norton, R.D. and J. Rees (1979) The product cycle and the spatial decentralization of American manufacturing, Regional Studies, 13: 141-51. Otto, A. and S. Kohler (2008), The contribution of new and young firms to the economic development of clusters in Germany. Comparative analysis of a growing, a mature and a

declining cluster, in U. Blien and G. Maier (eds), The Economics of Regional Clusters. Networks, Technology and Policy, Cheltenham, Edward Elgar, pp. 171-189. Piscitello (2004), Corporate diversification, coherence and economic performance, Industrial and Corporate Change 13 (5): 757–787. Porter, M.E. (1990), The Competitive Advantage of Nations, London: MacMillan Press. Pratt, D.J. (1998), Re-placing money. The evolution of branch banking in Britain, Environment and Planning A, 30 (12): 2211-2226. Rodriguez-Pose, A. and H.M. Zademach (2003), Rising metropolis. The geography of mergers and acquisitions in Germany, Urban Studies, 40 (10): 895-923. Scott, A.J. (1988), New Industrial Spaces. Flexible Production Organization and Regional Development in North America and Western Europe, London: Pion. Scott, A.J. and M. Storper (1987), High technology industry and regional development. A theoretical critique and reconstruction, International Social Science Journal, 112: 21532. Sluyterman, K., J. Danker, J. Van der Linden and J. Luiten van Zanden (1998), Het Coöperatieve Alternatief: Honderd Jaar Rabobank 1889-1998, Den Haag: Sdu Uitgever. Staber, U. (1997), An ecological perspective on entrepreneurship in industrial districts, Entrepreneurship and Regional Development 9, 45-64. Staber, U. (2001), Spatial proximity and firm survival in a declining industrial district. The case of the knitwear firms in Baden-Wurttemberg, Regional Studies 35: 329-341. Storper, M. and R. Walker (1989), The Capitalist Imperative. Territory, Technology and Industrial Growth. New York: Basil Blackwell. Ter Wal, A.L.J. and R.A. Boschma (2010), Co-evolution of firms, industries and networks in space, Regional Studies, in press. Wenting, R. (2008), Spinoff dynamics and the spatial formation of the fashion design industry, 1858-2005, Journal of Economic Geography 8: 593–614. Weterings, A, and Marsili, O. (2010) Spatial concentration of industries and new firm exits: Does this relationship differ between exits by failure and by M&A?, Working paper presented at the DIME Workshop „Industrial Dynamics and Economic Geography‟, Utrecht, September 2010.

Table. Descriptive statistics of variables used in the regression analysis. Amsterdam Loc.econ. Spinoff Experienced Foreign Cohort 1 Cohort 2 Cohort 3 M&A ln(t)

Mean Std.Dev. Min 0.333 0.471 2.934 1.232 0.251 0.434 0.533 0.499 0.031 0.173 0.294 0.456 0.197 0.398 0.198 0.399 0.078 0.373 3.024 1.092

Max 0 0 0 0 0 0 0 0 0 0

1 4.883 1 1 1 1 1 1 3.664 4.963

Table. Correlation matrix of independent variables Amsterdam Loc.econ. Spinoff ExperiencedForeign Cohort 1 Cohort 2 Cohort 3 M&A ln(t) Amsterdam 1.000 Loc.econ. 0.801 1.000 Spinoff 0.217 0.236 1.000 Experienced -0.025 -0.057 -0.619 1.000 Foreign 0.035 0.052 0.309 -0.191 1.000 Cohort 1 -0.074 -0.261 -0.134 0.038 -0.045 1.000 Cohort 2 -0.058 0.058 -0.034 -0.051 -0.034 -0.319 1.000 Cohort 3 0.032 0.173 0.056 -0.026 -0.026 -0.321 -0.246 1.000 M&A 0.125 0.121 -0.026 0.086 -0.004 -0.122 -0.028 -0.013 1.000 ln(t) 0.027 0.018 -0.078 0.170 -0.057 -0.245 -0.171 0.030 0.203 1.000