Who locates in the central position of a city? In Tokyo, major companies, whose sales reach 10% of Japan as a whole, are located in a small central business area called the Otemachi-Marunouchi-Yurakucho district. All over the world, we can observe that regionally influential firms locate themselves in the most central areas of cities.
Spatial proximity facilitates interactions among economic agents. Ever since Marshall (1920) emphasised three different types of transaction costs – the costs of transporting goods, people, and information and ideas, which can be reduced by location proximity – numerous theoretical models have been developed that examine how these spatial externalities influence the location of firms and households, urban density patterns, and productivity. In the urban economic literature, spatial externalities caused by both market (financial transactions) and non-market interactions (e.g. knowledge transfer, creation, and spillover) among agents are regarded as a crucial factor of urban spatial structure (e.g. Beckmann 1976, Fujita and Ogawa 1982). It is now well recognised that physical proximity between agents is one of the most important determinants of agglomeration. However, most of these studies pay much less attention to the influence of social proximity than that of physical proximity.
Social (relationship-based) distance expresses complex relationships and interdependencies between economic agents. By depicting relationships among a set of agents (e.g., individuals or firms) as graphs, social networks are constructs that are useful in comprehending the social interactions of agents. Helsley and Zenou (2014) propose the first theoretical model that explicitly incorporates the notion of social distance into this issue and examines “how interaction choices depend on the interplay of social and physical distance”. Their model reveals that agents who are more centrally positioned in the social network will position themselves in more central locations in a city.
In recent years, empirical analyses focusing on the relationship between relation-based distance and physical distance among firms have been carried out. There is compelling evidence of the importance of supplier-customer links within a domestic production network and their geographic proximity, in firms' performance (e.g. Bernard et al. 2016). Inter-firm transaction networks are found to be positively correlated both with industry agglomeration (e.g. Nakajima et al. 2013) and with co-agglomeration of industries (e.g. Fujii et al. 2015). These studies pay attention to industry agglomeration at the regional level, while we focus on the locations of firms within a city. Furthermore, effects of intra-firm linkages on locations of their plants are also examined (Behrens and Sharunova 2015).
While there exists empirical evidence of regional agglomeration of industries, turning to locations of firms within a city or within a region, the relationship between the structure of relationship-based networks and geographic configurations is still ambiguous. In our paper (Otazawa et al. 2018), using (non-retail) firm-level transaction data, we estimate the effect of the firms' position in the inter-firm transaction network on spatial location within a city in Japan. Our result empirically supports the theoretical perspective introduced by Helsley and Zenou (2014).
Figure 1 Locations of firms in the target area
Following Helsley and Zenou’s perspective, we assume that the more central a firm in an inter-firm transaction network, the more central and convenient a location it tends to occupy in geographical space to allow for the most efficient transportation of goods and travel to meetings with its business partners and customers. In our analysis, in order to examine the effect of firms' position in the inter-firm transaction network on their geographical location within a city, we estimate a linear regression model where the firms’ positions in the transaction network, which is named its centrality, is the regressor and their geographical locations, named its accessibility, is the response.
As a centrality measure, we employ the PageRank centrality measure devised originally by Google, in order to rank web pages. The entropy-based equation was applied to assess Accessibility that decays with distance to all other firms in the city. In the analysis, we also control some geographical properties of the firms other than accessibility, such as the distance to main railway stations and highway interchanges, in order to discriminate proximity to other firms from the availability of main transport nodes.
Our study aimed to present evidence for the effect of relation-based (social) distance on physical distance of economic agents, but the possibility exists that the reverse effect is true. That is, the spatial proximity of firms may have an impact on transaction networks, as firms that locate close to many other firms in physical space can easily expand their networks by interacting with the surrounding firms. In such cases, traditional approaches (e.g. ordinal linear regression analysis) may yield biased estimates of the impact of centrality on accessibility. In order to account for the reverse causality, we apply an instrumental variable (IV) approach. The IV method is an econometric technique that gives an unbiased estimation of treatment effects in observational studies, and it is used to take such endogeneity into account. For an IV that satisfies these conditions, we create one using information of transactions with firms outside the city in this study. The number of transactions outside the city strongly differs between firms. Hence, we argue that the this variable affects network centrality within the city, but does not directly affect the geographic location of firms within the city, when controlling for distance from main railway stations and highway interchanges.
From the main analysis, we obtained the following results. First, central firms in the transaction network locate closer to other firms in the geographical space of the city. Second, this effect varies depending on industries and it is noticeable for young firms (defined as firms that are eight years old or younger), especially in knowledge-intensive industries such as the information and communications industry and the scientific research, professional, and technical services industry. Third, the effect for single establishment firms is much stronger than multi-establishment firms, as their location decisions are more affected by the location of (potential) transaction partners than multi-unit firms.1 These results suggest that the potential importance of the inter-firm transaction pattern as a determinant of urban spatial configuration.
Marshall, A (1920), Principles of Economics, London: Macmillan.
Beckmann, M J (1976), “Spatial Equilibrium in the Dispersed City”, in G J Papageorgiou (ed.), Mathematical Land Use Theory, MA: Lexington Books, pp.117-125.
Fujita, M, and H Ogawa (1982), “Multiple equilibria and structural transition of non-monocentric urban configurations”, Regional Science and Urban Economics 12: 161-196.
Helsley, R W, and Y Zenou (2014), “Social Networks and Interactions in Cities”, Journal of Economic Theory 150: 426-466.
Bernard, A B, A Moxnes, and Y U Saito (forthcoming), “Production Networks, Geography and Firm Performance”, Journal of Political Economies (see also RIETI Discussion Paper Series, 16-E-055, 2016).
Behrens, K, and V Sharunova (2015), “Inter- and Intra-firm Linkages: Evidence from Microgeographic Location Patterns”, mimeo.
Fujii, D, K Nakajima, and Y Saito (2015), “Determinants of Industrial Coagglomeration and Establishment-level Productivity”, RIETI Discussion Paper Series, 15-E-077.
Nakajima, K, Y U Saito, and I Uesugi (2013), “Role of Inter-firm Transactions on Industrial Agglomeration: Evidence from Japanese Firm-level Data”, RIETI Discussion Paper Series, 13-E-21.
Otazawa T, Y Ohira, and J-V Ommeren (2018), “Inter-firm Transaction Networks and Location in a City”, RIETI Discussion Paper Series,18-E-054.
 Using data of intra-firm linkages, Behrens and Sharunova obtain similar results.