Reassessing the bank–industry relationship in Italy, 1913–1936: a counterfactual analysis

Until the banking reform in 1936, banks and industrial companies in Italy were strongly intertwined (both in terms of ownership and interlocking directorates). Using Imita.db—a large dataset containing data on over 300,000 directors of Italian joint-stock companies—this paper analyzes what would have happened to the Italian corporate network in the years 1913, 1921, 1927 and 1936 if the German-type universal banks and their directors would have not been there. Our test shows that new centers of the system would have emerged (financial, electricity, and phone companies), confirming the interconnected nature of the Italian capitalism. We also analyze two industries (textiles and iron and steel) characterized by different labor-to-capital intensities to check for sectoral differences. Contrary to conventional wisdom, we find that local banks were important in funding both industries.


Introduction
Financial institutions arise to intermediate between savers in excess of funds and entrepreneurs seeking credit for productive investment. Different economies may face varying problems for capital mobilization, so that financial systems may assume a wide range of forms. Thus, countries such as the UK and the USA developed specialized financial systems in which commercial and investment banking are spread into separate sets of institutions, and stock markets play a paramount role in allocating funds. Conversely, in continental Europe financial systems are often hinged on large jointstock universal banks that provide both commercial banking functions (short-term credit, deposit taking, payments clearing, bill discounting) and investment banking services (underwriting and trading in securities) (Cameron 1967;Fohlin 2012).
Universal banks emerged in a number of continental European countries in the mid-nineteenth century. Germany became the archetype of universal banking, having used this institution to mobilize extensive capital to finance its economy (Fohlin 2007). In the mid-1890s, a consortium of German banks transplanted German-style universal banking in Italy by establishing two joint-stock banks-the Banca Commerciale Italiana and the Credito Italiano-that soon became the two largest credit banks in the country (Cohen 1967). Gerschenkron (1962) singled out the main universal banks as the major drivers of Germany's and Italy's industrial spurts in the pre-World War One years. In his view, in both countries these banks functioned as a ''substitutive factor'' for otherwise missing prerequisites of industrialization, i.e., substantial capital accumulation and a willingness to invest it in industry. According to Gerschenkron, universal banks exerted a considerable influence and control over industrial companies through three major channels: capital participation, sharing of board members and monitoring over day-to-day financial affairs. Benefits of a financial system based on a close relationship between banks and industry included the provision of qualified managerial advice to industrial firms and the alleviation of moral hazard associated with asymmetric information. However, the role of universal banks has been reconsidered by a more recent ''revisionist'' historiography. In particular, Fohlin (1998Fohlin ( , 1999 found evidence that in both Germany and Italy these banks had a limited impact on capital mobilization, industrial investment, and economic growth. In both countries, firms connected to universal banks performed similarly-as regards growth of financial capital, fixed assets and revenues-to ''unattached'' firms.
In-depth research by Fohlin (2007) provides new insights into the structure of the German financial system. Her results show that key characteristics of universal banking emerged late in the industrialization process and its influence came from something other than the formalized control relationship over firms. The importance of universality-the combination of investment and commercial banking-appears mostly in the support that universal banks gave to the development of active securities markets, not in the domination of industry nor in the dramatic alteration of firm behavior or performance. 1 Another major contribution to the revision of the Gerschenkronian paradigm came from Confalonieri's multi-volume work on the history of the Italian banking system (Confalonieri 1974(Confalonieri -1976(Confalonieri , 1982(Confalonieri , 1992(Confalonieri , 1997. This author stressed the continuity between the German-type universal banks and their French-style antecedents as the former hired most of the staff and followed the practices of the latter. Confalonieri argued that the Italian universal banks were more concerned with standard banking activities, i.e., they lent money on a project by project basis, than with planning an overall industrial strategy. Last but not the least, Confalonieri maintained that the universal banks avoided permanent ownership of industrial companies and accepted shares only as a guarantee for loans. This strategy collapsed when the post-First World War crisis made impossible for many companies to repay their debts to banks. Thus, the banks unwillingly became the real owners of much of the ''militaryindustrial complex'' and of several important firms in other industries. When the Great Depression struck, the entire system collapsed and both the banks and their industrial clients were thus bailed out by the State.
In recent years, a new strand of literature has studied the relationship between banks and industry in Italy by focusing on corporate governance. Battilossi (2009) argued that unsound practices by the universal banks in the interwar years were permitted and to some extent enhanced by the systematic failure of public and private governance institutions to act as disciplinary devices for banks' risk-taking. In particular, he points out that, in order to reject hostile takeover attempts, the Banca Commerciale Italiana and the Credito Italiano assigned their own equities to ad hoc holding companies. These were owned by the banks themselves and by their ''allied groups,'' i.e., large industrial concerns that were among their largest borrowers. Thus, universal banks entered into close and long-run relationships with financed firms that insulated managers from external controls, favored the elimination of prudential constraints, and increased the potential for conflicts of interest as borrowing industrialists were co-opted among their controlling owners. This degeneration of relationship was epitomized by a pathological escalation of equity stakes, possibly further encouraged by expectations that liquidity mismanagement would have been fixed once again by lack of supervision and unconditional bailouts by monetary authorities when macroeconomic conditions eventually improved. In the same perspective, Brambilla (2012) stresses that poor corporate governance limited information disclosure by borrowing firms. This led to higher grades of opacity as for investment projects quality and industrial undertaking performances, and to higher degrees of risk connected with capital investment by universal banks. 2 Despite the relevance of the relationships between banks and industry in Italian economic historiography, only a few studies have analyzed the sharing of board members-that is the system of interlocks-between banks and industrial companies in the period prior to the Second World War. The pioneering works by Zorzini (1925) and Luzzatto Fegiz (1928) found in the mid-1920s a high presence of directors of the two largest universal banks-Banca Commerciale Italiana and Credito Italiano-on the boards of electric power companies. More generally, they found a high concentration of the whole corporate system, in which 2 % of directors controlled more than one-third of the total share capital of Italian joint-stock companies. Cohen (1967) and Confalonieri (1974Confalonieri ( -1976Confalonieri ( , 1982Confalonieri ( , 1992Confalonieri ( , 1997 stressed that as soon as they began operations, the two largest universal banks formed two networks that involved Italy's two largest railway companies-Strade Ferrate Meridionali and Strade Ferrate del Mediterraneo that after the nationalization of the Italian railways in 1905 turned into holding companies-and firms operating in modern capitalintensive sectors such as electric power, chemicals, iron and steel and shipyards. A study by Vasta and Baccini (1997)-using a large sample of more than 4000 Italian joint-stock companies and formalized network analysis techniques-held that the Italian corporate network does not seem to have been characterized by a strong centrality of banks as it was commonly believed. The location of banks at the center of the network could be detected in 1911 and even more in 1927, but this was no longer the case in 1936, after the collapse of the universal banks. By that time, insurance companies and utility companies had replaced banks at the center of the system (Baccini and Vasta 1995). These authors also find that a highly stable system of interlocks existed in parallel to that centered on the banking system, and remained substantially unchanged over the years. This result is very similar to what has been shown by Fohlin (2007) for Germany, where in the pre-First World War years, there was also a dense web of interlocks among non-financial corporations.
An in-depth analysis of the system of interlocks centered on Italy's largest universal bank, the Banca Commerciale Italiana, is provided by Fohlin (1999). This author shows that 18 % of all Italian joint-stock companies with at least one million lire of share capital in 1911 were affiliated to this bank but accounted for 40 % of total share capital. This bias toward larger firms pervaded the bank's relationships across industries but was more pronounced in some sectors such as transportation, mining, chemicals, food and beverage, and entertainment services.
This article investigates the links between the Italian universal banks and the corporate economy by focusing on cross-memberships in boards of directors of joint-stock companies (interlocking directorates). This approach, which has its origin in modern sociology, gained, in the last decades, significant spaces in all social sciences, including economics and, with a certain delay, economic and business history. 3 The use of the interlocking directorates technique can play a dual role by complementing, on the one hand, the traditional case study method and, on the other hand, by allowing broad overviews of the corporate systems under investigation, which also helps the verification of different theoretical perspectives. The interlocking directorates technique is a powerful tool for the analysis of the governance structures of the corporate system. Nonetheless, a causal relationship between the structure of corporate interlocks and the performance or the patterns of investment of the firms involved as well as the growth of the economy overall is difficult to establish and is beyond the scope of this work. This paper first analyzes the structure of the Italian corporate network in four benchmark years: 1913, 1921, 1927, and 1936. Then it presents a counterfactual experiment, by showing what would happen to the network if the universal banks and all their directors had been eliminated. In order to check for sectoral differences, the analysis is replicated for two industries-textiles and iron and steelcharacterized by different labor-to-capital intensities. We find that a new center of the system would have emerged, confirming the interconnected nature of the Italian capitalism. Moreover, it emerges that local banks were important, particularly in the North of the country, in funding textiles and iron and steel firms.
The paper is organized as follows: After this Introduction, Sect. 2 describes the data and the sources. Section 3 sets out the methodology utilized for this study. Section 4 presents the results of our counterfactual tests, while Sect. 5 focuses on actor centrality. Section 6 replicates the counterfactual exercise for textiles and iron and steel industries. Lastly, Sect. 7 concludes.

Data and sources
The source we used for this work is Notizie statistiche sulle principali società italiane per azioni, edited and published by Credito Italiano, from 1906 to 1925, and since 1928, by the Associazione fra le Società Italiane per Azioni. The Imita.db database is an electronic version of this source. 4 This dataset contains information regarding companies, boards of directors, and balance sheets of a large sample of Italian joint-stock companies for several benchmark years. The source includes all the joint-stock companies listed on one of the Italian stock exchanges, together with those companies located in Italy whose share capital at the closure of the last balance was higher than a set threshold, which was set at 1 million Italian lire, with the sole exception of 1913, when it amounted to 500,000 lire. Overall, the dataset contains data on more than 38,000 companies, almost 300,000 directors, and more than 100,000 balance sheets. Representativeness, in terms of capital, is very high as the sample covers over 90 % of the total universe of Italian joint-stock companies. As for the directors, we used only data for members of a board of directors in the strict sense, leaving out the members of Collegi sindacali. 5 We have carefully standardized the names of the directors to make them as homogeneous as possible. 4 Imita.db is one of the worldwide largest datasets on joint-stock companies in historical perspective. For details, see: Vasta (2006). The database is available on line: http://imitadb.unisi.it. 5 During the period investigated in this paper, the 1882 Commercial Code regulated corporate governance in Italy. This had designed a two-board system of corporate administration in which the assembly of the shareholders had the legal authority for appointing the two following boards: (1) the board of directors (Consiglio di amministrazione), which was the executive body of the assembly of the shareholders. This usually included both inside and outside directors; (2) the board of syndics (Collegio sindacale), which consisted of three or five regular syndics (sindaci effettivi) and two alternate syndics (sindaci supplenti). Their duty was to exercise control and supervision over the management of the company, to monitor the decision taken by the board of directors (Teti 1999). The board of syndics was fundamentally an auditing board and its function did not coincide with those of the supervisory board in the German system. Thus, similarly to what was done in the two major international research projects on corporate networks in comparative perspective (Stokman et al. 1985;David and Westerhuis 2014a) for our analysis we have selected only members of the board of directors.
Reassessing the bank-industry relationship in Italy, 1913… 187 However, we estimate that the information on boards of directors contained in Imita.db has a margin of error of about 1 %, as is the case with other similar databases (Mintz and Schwartz 1985). These errors are mainly due to cases of homonymy, misprints, or shortcomings in the source.

Methodology
Starting from the actual networks, we analyze and test-for each benchmark yearsome counterfactual hypotheses about the network structure. We study an initial two-way matrix related to the network, and then we analyze a second matrix related to the counterfactual effects. The effects of the change could be distinguished between global and local. Global effects can be referred to the structure of the entire network, whereas local effects are related to the position of a single node in the network. The global effects of the network counterfactuals are related to network stability and network connectedness, whereas the local effects are associated with the stability and robustness to change of the communication channels into the network. In particular, it is possible to analyze how the role of each node changes by passing from the actual to the counterfactual network (Borgatti 2003(Borgatti , 2005(Borgatti , 2006. Analytically, we start from an adjacency matrix A s , of binary elements A s (i, j) = 1, in which we can observe a connection between the firms i and j, and A s (i, j) = 0, in which the two firms are not connected. This matrix refers to the baseline adjacency matrix and is the matrix to which every counterfactual experiment is compared. For this matrix we compute all the different network measures representing the baseline network indexes to compare with the counterfactuals experiments, which are obtained through a perturbation of the original network. A single perturbation could be considered as an elimination of a single node or an elimination of a single vertex. We recomputed all the network measures for each counterfactual experiment, and then we compare the results obtained for the two networks.
The structure of the Italian corporate network at various times in the twentieth century has already been explored in Vasta and Baccini (1997) and Rinaldi and Vasta (2005 by using network analysis techniques. This paper adds to the previous studies a counterfactual experiment for each of the benchmark years considered. Our procedure works as follows: Firstly, we compute statistics for the actual networks in all the benchmark years; secondly, we remove the universal banks-Banca Commerciale Italiana, Credito Italiano, Banco di Roma, Banca Nazionale di Credito and Società Bancaria Italiana 6 -and all the board positions of the directors who sit in at least one universal bank. In this way, if a given director sits in the boards of directors of a universal bank and two industrial companies, we cut the links between the universal bank and each of the two companies as well as the link between the two industrial companies. 7 In addition to the analysis of the whole networks, we also compare interlocking directorates between banks and industry in textiles and in iron and steel. We expect that banks are considerably more connected with the iron and steel sector because it requires more capital for investment. Sharing board members enhances the ability of the banks to monitor the performance of the companies and therefore the possibility of getting their money back. For both sectors, we performed the counterfactual described above for each benchmark year. 4 The structure of the network Table 1 reports a number of statistics for the actual networks in the four benchmark years and the relevant counterfactuals. The number of vertices (in our case, companies) is related with the number of nodes in the network. This number increases from 1913 to 1921, reaches a peak in 1927, and then decreases in 1936, as a consequence of the reduction in the number of firms included in our sample due to the Great Depression. The number of isolates follows the same pattern; however, it also increases in 1936, possibly because of the bankruptcy of some connecting companies during the Depression. The increase in the number of isolates is higher than the increase in the number of firms, moving from 12 to 21 %. The number of components (which is a subgraph in which any two vertices are connected to each other by paths, and which is not connected to any additional vertices in the supergraph) steadily increases over the years. The largest component in a network is usually very large; in our case its relative size with respect to the number of vertices is 85 and 88 % in the first two benchmark years and then falls to 81 and 71 % in the last two years. The diameter of largest connected component measures the distance between the two most distant nodes: The increase we observe between 1913 and 1921 is sizeable, and given the increase in the number of firms, this means that the network becomes denser in structure. Notably, in 1927 the diameter reduced with respect to 1921, but in 1936 it returned to its 1921 level. Instead, the total adjacency index 8 -which measures the number of edges (interlocks between companies) in the network-increased constantly from 1913 to 1927, but in 1936 it declined. More generally, the structure of the network became more complex from 1913 to 1927, but in 1936 complexity decreased. In fact, the average degree 9 -a 7 Actually, this procedure defines the upper limits of the universal banks' influence in the network. For example, if there is an executive director (for instance the president) of a large Italian steel company who has 10 positions in the network, he creates (10 9 9)/2 = 45 ties in the network. Indeed, he is executive director of the steel company, has been co-opted in the board of a universal bank, and has board position in other eight industrial firms (non-executive director). The sub-network created by this individual is centered on the steel company and is more an indicator of the influence of this company rather than of the universal bank. Thus, in the WP version of this article (Drago et al. 2012), we performed also a less demanding counterfactual experiment in which we eliminated only the universal banks, but retained the board positions that their directors held in any other companies. This allows to define the lower limits of the universal banks' influence in the network. The difference between the actual network and the counterfactual obtained in this way was very small so we eventually decided to drop it. 8 This index is calculated starting from the matrix that contains all ties. 9 The degree of a node is the number of edges connected to it.
Reassessing the bank-industry relationship in Italy, 1913… 189 normalized index unbiased by a change in sample size-also increases from 1913 to 1927 and decreases in 1936. A similar trend is shown by a standardized sample of Italy's top 250 companies by total assets. Here the density-the basic index of the network's connectedness, defined as the ratio between the actual number of interlocks and the number of possible interlocks 10 -rises from 4.77 % in 1913 to a maximum of 8.61 % in 1927 and then drops to 5.44 % in 1936 (Rinaldi and Vasta 2014). 11 Total adjacency index is the number of edges in a network. The average degree measures the average number of companies to which each one is interlocked 10 In the case of a standardized sample, density is an unbiased indicator because comparison concerns samples of the same size. 11 Comparing to other countries, Italy seems to have followed much the same trend as Germany, even if at lower values. Indeed, the density of the network of Germany's top 250 firms rose from 7.34 % in 1914 to 16.19 % in 1928 and then fell to 10.47 % in 1938. The US followed the same trend, even if at values much lower than both Germany and Italy (Windolf 2009(Windolf , 2014. The decline of the density in these countries in the 1930s was the result of the banking laws that prohibited or regulated the financial participation of banks in industries (David and Westerhuis 2014b). A similar trend was observed also in the Netherlands, another bank-centered corporate system. However, here it was not banking laws that put limits on the functioning of banks, but the banks themselves that chose new strategies: after the crisis, Dutch banks returned to trade financing and became reluctant to provide long-term credit to firms (Westerhuis 2014). By contrast, in some other countries, such as France and the UK, the density of the network of the top 250 firms grew steadily from the pre-First World War years to the late 1930s. In the former from 4.90 % in 1911 to 6.30 % in 1937 (François and Lemercier 2014); in the latter from 0.98 % in 1904 to 1.98 % in 1938 (Schnyder and Wilson 2014).
In all benchmark years, we find that firms that were linked to the universal banks were in almost all cases the largest firms in their respective sector. The bias of universal bank's relationship toward largest firms was more pronounced in some sectors than others, especially in capital-intensive industries such as electricity, chemicals, iron and steel, and in the other financial intermediation (finance companies). Conversely, affiliation to large universal banks is lower-both as a proportion of firms and share capital-in some labor-intensive industries such as food, textiles, leather, wood, and furniture. 12 The counterfactual exercise changes the network structure to different extents across benchmark years. The number of isolates is slightly higher than in the actual networks, but their percentage is basically the same. In addition, the percentage of multiple ties is very similar through the benchmark years. The number of components is strongly reduced, but not the size of the largest component. The diameter of the largest connected component is also similar, if anything slightly higher in two years. The dynamics of the indices is quite similar to the one of the actual networks, with an increase in the first three benchmark years and a reduction in the last one. Moreover, their values are quite smaller than in the actual networks. For instance, the average degree sees a 21 % reduction in 1913, then the gap slightly declines to 17 % in 1921, reaches a 25 % peak in 1927 (from 16.8 in the actual network to 12.5 in the counterfactual). Finally, it drops dramatically to a 10 % minimum in 1936 (from 11.0 to 9.9). The average degree in the counterfactuals has rather high values in absolute terms in all benchmarks year, ranging from 9.9 in 1936 to 12.8 in 1921, and is much more stable than that in the actual networks.
These data show that the weight of the large universal banks in the Italian corporate network varied remarkably over time. It was substantial prior to the Great Depression and especially in the late 1920s, but then it declined dramatically. At the same time, a stable and resilient system of interlocks seems to have existed in parallel to the system centered on the larger universal banks, as claimed by Vasta and Baccini (1997). Remarkably, this result is similar to Germany, where not only the larger Berlin-based universal banks, but also the smaller provincial universal banks were linked by a dense system of interlocks to industrial companies and a web of interlocks also existed among non-financial firms (Fohlin 2007).

Actor centrality
In network analysis, it is assumed that actors that are central have better access to information, better opportunities to spread information and someway a ''power'' to coordinate the whole network. We use nBetweenness as the measure to calculate the centrality of firms. This measure is based on the idea that a firm is more central if it is more important as an intermediary in the communication network. So it calculates for each actor (company) the number of shortest paths between any pairs of actors in the network that pass through this actor (De Nooy et al. 2011). Table 2 shows the mean for the deciles of nBetweenness, across actual and counterfactual networks. Within each benchmark year, the mean of nBetweenness in each decile is quite similar, which can be expected in very large networks as the ones we are analyzing. In all benchmark years, the mean of the three lowest deciles is zero. Such a result is due to the high proportion of isolated firms (i.e., firms that are not interlocked to any other firms) which is not surprising in such large samples. The mean is zero for the four lowest deciles in 1927 and for the five lowest deciles in 1936, in both cases for both the real and counterfactual network. Across benchmark years, the mean of the two highest deciles becomes smaller moving from 1913 to 1936. In particular, we find that the counterfactual network differs from the actual one in the mean of nBetweenness in the three highest deciles, where it is higher in all benchmark years but 1913. Thus, removing universal bank-industry interlockers seems actually to increase centrality of other subjects, making them even more connected than those that were removed.
Tables 3, 4, 5, and 6 show the nBetweenness of the 20 most central companies and the relevant counterfactual in the four benchmark years. The actual networks' statistics show the prominent position occupied by the universal banks, some iron and steel producers, electricity companies, and railway companies.
Looking over time at the real networks, the universal banks strengthen their central position in the ranking. In 1913 we find three of these banks (1st, 5th and 13th) among the top 20, whereas in 1921 we have four of them (1st, 3rd, 8th, and 17th), and in 1927 the same number of banks are ranked 1st, 3rd, 4th, and 6th, taking even more ground. In this sample, 1927 represents once again the apex of the role of these banks, since in 1936 we find only two universal banks among the 20 most central companies, ranked 7th and 20th, respectively.
The counterfactual has some deep effects on the ranking of firms. In 1913, the three universal banks disappear because, by construction, they have been removed from the sample ( Table 3). The top of the ranking marginally changes, with companies moving up by one position because of the removal of Banca Commerciale Italiana. However, in lower positions we observe some serious changes: Unes Unione Esercizi Elettrici moves from the 11th to the 6th position, Società Elettrica della Sicilia Orientale from 18th to 11th, and Officine Meccaniche Italiane from 20th to 14th. Five other firms-in addition to the three universal banks-disappear from the top 20, while eight firms that did not appear in the actual ranking make their entry. The pattern in 1921 in similar: stability in the very first ranks, a number of shifts due to removal of the universal banks, and the changes from the middle to the bottom of the ranking (Table 4). A few companies become more central, strongly improving their rank: Unes Unione Esercizi Elettrici from 16th to 3rd and Banca Bergamasca di Depositi e Conti Correnti from 20th to 11th. Five companies that were not in the ranking of the actual network appear in the top 20 in the counterfactual. Reassessing the bank-industry relationship in Italy, 1913… 193 As noted before, 1927 is the benchmark year in which universal banks had the largest influence in the network. Unsurprisingly, removing the interlockers had its strongest effect. In Table 5 we find that a half of the most central companies-10 out of 20-went out of the top 20 in the counterfactual and were replaced by as many companies.
As for 1936 (Table 6), the universal banks have already lost their power, and thus the removal of their interlockers has a very limited effect. The first six positions are virtually unchanged, and the reshuffling is usually limited within 3-4 positions. Only four companies that were not in the ranking in the actual network enter among the top 20 in the counterfactual but are placed from 17th to 20th position. We emphasize that Credito Italiano-Italy's second largest universal bank-in the actual network is relegated in 24th position. This exercise has shown that the centrality of the universal banks in the Italian corporate system varied over time. It increased from 1913 to 1927, but then it sharply decreased in 1936, after the universal banks had been severely struck by the Great Depression and had been eventually bailed out by the big state-owned holding Istituto per la ricostruzione industriale (Zamagni 1993;Toninelli and Vasta 2010). Reassessing the bank-industry relationship in Italy, 1913… 195 Nonetheless, the demise of the universal banks did not lead to a disentangling of the network, as new centers with a high value of nBetweenness emerged and played a key role in assuring the cohesion of the system. As said, these results corroborate Vasta and Baccini's (1997) claim that a sizeable and stable system of interlocks existed in parallel, or as a complement, to that hinged on the universal banks. This can be detected in all benchmark years, including 1927, when the influence of the universal banks was higher. Until 1927, such a system of interlocks extraneous to the universal banks was mainly centered on electricity companies, some large steel and trade companies and the Banca d'Italia, at that time a joint-stock bank. In 1936, it still hinged on electricity companies together with some new actors-such as insurance companies and some mandatory syndicates-that had taken up a central role.

Comparing labor-and capital-intensive industries
In this section we analyze the bank-industry relationship in two sectors characterized by different relative intensities of factors of production: textiles and iron and steel. In performing this analysis, for each benchmark year we construct two subnetworks: In the former case we take all the banks and all the textile companies, whereas in the latter case we take the banks mentioned above and all the iron and steel companies. The methodology devised earlier is replicated for these subnetworks. Table 7 reports the statistics of these smaller sub-networks. The pattern in terms of isolates, number, and size of components is similar for the two networks and resembles the one of the whole network. The average degree shows higher values for the textile actual sub-networks than for the iron and steel ones. These results possibly reflect the fact that interlocking directorates were a common practice also in such a labor-intensive industry as textiles, which in a latecomer country as Italy at that time still accounted for a sizeable part of the largest companies. 13 It is remarkable that the average degree, which is a normalized index whose value is not affected by the size of the sample, shows higher values for the textile sub-networks in all the benchmarks considered, even if the gap with respect to the iron and steel networks is narrow. Nonetheless, the counterfactual exercise has stronger effect for iron and steel network, confirming that the links generated by universal banks' directors were more relevant for the capital-intensive industries. Table 8 reports the distribution of mean betweenness for the sub-networks. Compared with the whole networks the picture is somehow different. We find that the 9th decile has a much higher value in these sub-networks than in the main ones, whereas this is not always true for the other. Moreover, the distribution is more stable for textiles when we compare the actual sub-networks with the counterfactual than for iron and steel. The latter, in particular, shows a reduction in the mean nBetweenness in the highest percentile.
We observe four main findings: first, the pattern of the results is similar to those found earlier for the general network, as the effect of a counterfactual shock is increasing from 1913 to 1927 and drops in 1936. Second, and this appears surprising, the connectedness is higher in the banks-textiles sub-network; however-as it was expected-the effect of a counterfactual shock is proportionally higher in the banks-iron and steel one. Third, large universal banks are placed in a paramount position in both sub-networks in benchmarks from 1913 to 1927, but then they are marginalized as a consequence of the Great Depression. Fourth, many small local banks appear among the most central companies for both sub-networks. This feature grows over time from 1913 to 1927 with the same pattern observed for the universal banks, and it is more pronounced in the iron and steel than in the textile industry. 14 In addition, the position of local banks seems to have been 13 It is worth noting that, in 1913, amongst the top 200 Italian manufacturing firms there were 64 textiles companies and that, in 1936, the corresponding figures still amounted to 41. For a detailed account of the structure of Italian big business, see Giannetti and Vasta (2010, table 2.2). 14 Conversely, in Germany local banks were affiliated principally with textile and chemical firms (Fohlin 2007).
Reassessing the bank-industry relationship in Italy, 1913… 197 weakened in 1936 with regard to the previous benchmark year. Nonetheless, their ties with industry seem to have stood the Great Depression better than those of the large universal banks. 15 The overall result that emerges from the analysis of the bank-industry relationship in the textile and iron and steel sectors is that, contrary to conventional wisdom, interlocking directorates were not limited to the larger banks (especially universal banks) and to the larger firms operating in capital-intensive industries. Instead, we have found that the sharing of board members was a common practice in Italian capitalism, which involved to a large extent also small local banks and small firms operating in labor-intensive industries. This suggests that one of the main peculiarities of Italian capitalism, i.e., the dwarfism of its entrepreneurial base, could at least in part be reconsidered. In fact, as interlocking directorates are an element that makes it possible to broaden a firm's boundaries, we can assert that at least some of the Italian small and medium-sized firms, although remaining small, Table 7 Networks (actual and counterfactual) characteristics, textiles, and iron and steel industries  Textiles Iron and steel 1913Iron and steel 1921Iron and steel 1927Iron and steel 1936Iron and steel 1913Iron and steel 1921Iron and steel 1927Iron and steel 1936 Actual network are able to be connected to other firms and banks, and act as a bigger player. This can be seen as the ''only way to grow'' for firms operating in a latecomer country or, at least, as the easiest way to reach a reasonable size. 16 Interestingly enough, this result is very similar to Germany, where the largest firms were usually linked to the great universal banks headquartered in Berlin, whereas small firms were more commonly affiliated to local banks operating in the same region (Fohlin 2007). In particular, our counterfactual experiments make us reconsider the respective roles of larger and smaller banks in Italian industrialization. Actually, the formation of the Italian banking system in the nineteenth century was characterized by the creation of a few larger banks (first French-style and then, after the 1893 banking crisis, German-style) that operated at the national level and of some thousand smaller banks, credit societies and saving banks that operated at the local level. The conventional wisdom in Italian historiography (Gerschenkron 1962;Cohen 1967;Mori 1992) maintains that, until the Second World War, the funding of industrial ventures, especially in capital-intensive industries in the northwest of the country, was carried out principally, if not exclusively, by the larger banks, especially the German-style universal banks. According to this view, smaller local banks started to play a relevant role in the financing of industry only in the 1950s during the ''Golden Age,'' as a part of a set of policies devised to promote small manufacturing Textile 1913  Reassessing the bank-industry relationship in Italy, 1913… 199 firms. The Bank of Italy played a decisive role in this respect, as after the Second World War it prompted a restructuring of the banking system to strengthen the local banks that funded small firms clustered in the industrial districts in the northeast of the country. The growth of the former was strongly associated with that of the latter. In fact, throughout the post-Second World War period the national banks within industrial districts had a much lower share of the local credit market than elsewhere. Lower assessment, monitoring, enforcement costs, and social connections between local banks' managers and local entrepreneurs gave local banks a competitive edge within industrial districts. Bank competition was restricted to prevent an increase in industrial concentration since it was felt that if the small firms were deprived of necessary credit they would be forced to merge with the larger firms (Conti and Ferri 1997;Conti 1999;Carnevali 2005). Instead, our analysis has shown that prior to the Second World War large universal banks were not the only relevant actor that supported Italian industrialization, and in several areas of the country (especially Lombardy and some other regions of the North), small local banks had established strong ties with industrial companies already in the 1910s. The strength of these ties grew in the interwar years. A dense web of ties between local banks and industrial companies can be detected for both a labor-intensive industry as textiles (and this could be expected) and for a capital-intensive industry as iron and steel (and this appears more surprising).
It is possible that local banks were linked to the larger universal banks in the sense that the former might function as branches of the latter. In fact, it was possible for small banks located in peripheral areas to collect funds regionally and make deposits with the universal banks in their base cities. However, our evidence shows little support for this claim. In fact, in both the textiles and the iron and steel subnetworks only a minority of local banks in the top 20 were interlocked to a universal bank, the highest proportion being one out of three in the textile sub-network in 1913. 17 Thus it seems that the links between local banks and industry were largely part of a system of interlocks that existed independently on the one centered on the larger universal banks. In their respective territories, such local banks acted as small universal banks that provided firms with both short-and long-term loans and sometimes also participated in their share capital. The sharing of board members became also widely superimposed to credit relationships and cross-shareholdings between banks and industry at the local level (Piluso 2009).
Lombardy stands out as the region that accounted for the highest number of local banks among central companies in both the textiles and the iron and steel subnetworks throughout the period investigated, with a proportion ranging from 70 % to the totality of local banks included in the top 20. Small universal banking existed in Lombardy already in the 1910s, but it boomed in the 1920s with the setting-up of several banks explicitly aiming at funding industrial investment (i.e., the Banca Industriale di Busto Arsizio, the Banca Industriale di Bergamo, and the Banca 17 Banking concentration in Italy was lower than in the England and Germany, but higher than in the USA. Thus, on the eve of the First World War, the four largest Italian credit banks held 15 % of total bank assets. At the same time, the corresponding figure for the top five banks was 36 % in England, 33 % in Germany, and only 4.5 % in the USA (Fohlin 2012).
Industriale Gallaratese). At the same time, several other banks that had previously invested principally in agriculture (i.e., the Credito Agrario Bresciano, the Banca di Legnano, and the Banca Agricola Mantovana) turned also to industrial investment (Zamagni 1995). In 1927, some of these banks appear in the top 2 in at least one of our sub-networks, together with some long-established banks. Among the latter, two local banks located in the small industrial town of Bergamo stand out: Banca Piccolo Credito Bergamasco and Banca Mutua Popolare di Bergamo. 18 The system of interlocks centered on small universal banks in Lombardy was partially destroyed by the Great Depression. However, most Lombard local banks overcame the crisis and were still present in our sample in 1936. In that year some of them appeared among the top 20 in either the textiles or the iron and steel subnetwork and two of them-the Banca Belinzaghi and the Banca Unione-in both.
These findings can perhaps help to shed light on Fohlin's (1998Fohlin's ( , 1999 finding that investment in fixed capital did not relate to attachment to Italy's largest universal bank (Banca Commerciale Italiana), firms affiliated to this bank invested less on average relative to their size, and fastest-growing firms were not interlocked to it. Fohlin's evidence may at least in part be consequent on the fact that credit was available also to firms that were not linked to the Banca Commerciale Italiana or, more generally, to the four larger universal banks, since universal banking was a practice largely used also by small local banks, as it is indirectly shown by their propensity to generate interlocking directorates with industrial firms.

Conclusions
This paper has investigated, by the use of new data, the links between the universal banks and the corporate economy in Italy by focusing, with updated methodologies, on cross-memberships in boards of directors of joint-stock companies (interlocking directorates). It has first analyzed the structure of the Italian corporate network in four benchmark years: 1913, 1921, 1927 and 1936. Then it has presented a counterfactual experiment, by showing what would happen to the network if the universal banks and all their directors had been removed. This analysis has been replicated for the textile and iron and steel industries, characterized by different labor-to-capital intensities, to check for sectoral differences. 18 The Banca Mutua Popolare di Bergamo was founded in 1869 as a credit cooperative to provide funds to the local economy. Members were mainly tradesmen and craftsmen, and some protestant Swiss textile entrepreneurs stood out among them. Nonetheless, the bank showed a preference for smaller deposits and lending to small enterprises. Four years after the Banca Bergamasca di Depositi e Conti Correnti was founded and became the main source of funds to the local cotton and silk industries. Both Italian and Swiss entrepreneurs were amongst the bank's shareholders, directors and borrowers. In the 1920s the bank pursued rapid and risky expansion by lending and assuming large equity stakes in connected industrial borrowers which eventually led to its bankruptcy in 1932. In 1891 the catholic bank Piccolo Credito Bergamasco was founded and it also became an important source of funds to the local economy. In 1925 a fourth local bank, the Banca Industriale di Bergamo, was created and soon developed strong ties particularly with Italcementi, Italy's largest cement producer headquartered in Bergamo (Romani 2011).
Reassessing the bank-industry relationship in Italy, 1913… 201 The main results of this paper support part of the criticism of recent historiography (Confalonieri 1974(Confalonieri -1976(Confalonieri , 1982(Confalonieri , 1992(Confalonieri , 1997Fohlin 1998Fohlin , 1999 toward Gerschenkron's view about the centrality of German-style universal banks in fostering Italian industrialization, especially in capital-intensive industries. Our counterfactual exercise has shown that the centrality of the universal banks in Italian corporate system varied over time. It increased from 1913 to 1927, but it decreased sharply in 1936, as a consequence of the Great depression. At the same time, we have shown that Italian capitalism seems to be structured to a remarkable extent on a sizeable and stable system of corporate interlocks that existed in parallel to that centered on the universal banks. This latter system can be detected in all benchmark years, including 1927, which marked the apex of the influence of the universal banks. Remarkably, this result is similar to what Fohlin (2007) found for Germany, where not only the larger Berlin-based universal banks, but also the smaller provincial universal banks were linked by a dense system of interlocks to industrial companies and a web of interlocks existed also among non-financial firms.
Finally, to test Gerschenkron's hypothesis that universal banks fostered principally capital-intensive industries, this paper used the same methodology to analyze the bank-industry relationship in two sectors characterized by different relative intensities of factors of production: textiles (labor-intensive) and iron and steel (capital-intensive). This exercise provides four main outcomes: (1) the pattern found for the two sectors is similar to those found for the general network, as the effect of a counterfactual shock is increasing from 1913 to 1927 and drops in 1936; (2) surprisingly, the connectedness is higher in the banks-textiles sub-network even if, the effect of a counterfactual shock is proportionally higher in the banks-iron and steel one; (3) large universal banks are central in both sub-networks until 1927, but then their role fades; (4) many small local banks appear among the most central companies in both sub-networks, and this feature varies over time with the same pattern of universal banks. Nonetheless, in 1936 the ties of local banks with industry seem to have stood the Great Depression better than those of the large universal banks. This happened also in Germany, where the largest firms were usually linked to the great universal banks headquartered in Berlin, whereas small firms were more commonly affiliated to local banks operating in the same region (Fohlin 2007). However, local banks in Germany were affiliated principally with textile and chemical firms and not so much with iron and steel ones as we have found in Italy.
Thus, on the one hand, this paper finds support for Gerschenkron's view that large German-style universal banks were linked principally to larger firms in Italy. On the other hand, it shows that Gerschenkron failed to grasp that the creation of interlocking directorates was not limited to a few large banks, but was a common practice (at least in Lombardy and some other areas of the North) that involved many local banks which in turn developed a dense web of ties with industrial firms. Only a minority of such local banks were linked to a large a universal bank, whereas most of them seem rather part of systems of interlocks that existed independently on the one centered on the larger universal banks.    Table 9 Reassessing the bank-industry relationship in Italy, 1913… 205