Internal and external scale economies are important in trade and urban economics respectively. I argue that internal scale economies mainly drive intra-industry trade (IIT) between similar countries. However, some empirical studies reveal that traditional Ricardian comparative advantage theory may be more important in explaining trade patterns. External economies of scale will then be used to show agglomeration of firms in cities. I elaborated on this based on the theoretical foundations of system of cities by Henderson (1974) and the empirical results of Ellison et al. (2010). Finally, I emphasize the role of scale economies in urban economics theory.
Internal scale economies are essential for IIT and specialization (Balassa and Bauwens, 1988). In New Trade Theory, Krugman (1979) argues that firms specialize in a limited subset of products in order to reap increasing returns to scale. This model has autarky equilibrium with firms producing differentiated products (Clark, 2010). This specialisation of producers satisfies consumers’ diverse tastes and demand for varieties of horizontally differentiated goods (Lloyd and Lee, 2002). This demand for product variety manifests in cross-hauling, as seen in the automobile industry. The US produces Ford cars and Japan produces Hondas, but both countries still trade cars with each other. Moreover, scale economies causes firms to enter deeper specialization in production components and stages, so intermediate good inputs from similar economies have to come together to form and finalize the product. For example, Apple’s iPhone is designed in California but sources its manufacturing components from Geneva, Korea, Japan and Taiwan before it is assembled in China. As a result, exports of differentiated goods with increasing returns to scale often go to countries whose firms produce similar goods, because there is a large market for additional product variety and intermediate good inputs.
However, in Leamer and Levinson’s (1995) overview of empirical literature they report little evidence that scale economies provide an important determination of trade. Davis (1995) argued that IIT might occur even with constant returns to scale and perfectly competitive markets, attributing IIT to traditional Ricardian determinants of trade where small technical differences induce specialisation and trade instead. In two studies, Davis and Weinstein (1996, 1999) included internal economies of scale into the H-O framework to assess the relative importance of comparative advantage and increasing returns to scale for production structure and trade. In the first study, they find that 90 per cent of the explainable OECD trade specialisation arises from endowment differences, and scale economies account for at most 10 percent. In the second study, they analysed the regional production structure of nineteen Japanese industries, and eight of these showed significant evidence for sizeable home market effects. They deduced that scale economies were more important in the regional, and not international, structure of production. However, Harrigan (1994) argues that most empirical evaluations of IIT models are misleading because of incorrectly regressing the Grubel and Lloyd index on various proxies for scale economies. He finds some evidence that the volume of trade (not proportion of IIT) is higher in sectors with high internal economies of scale. Schmitt and Yu (2001) also find that an increase in scale economies leads to a higher share of trade relative to total production.
I now turn my attention to external economies of scale. Sources of urban increasing returns to scale are understood in terms of agglomeration economies. Marshall (1920) identifies the microfoundations of these agglomeration economies – specialized suppliers, skilled labour pools and a common knowledge base. Firms gain agglomeration economies when these factors are geographically concentrated.
Firstly, co-locating firms can exploit the scale economies in the production of the intermediate good, and can reduce costs when they share these intermediate inputs. Vernon (1972) raises the example of clustering dressmakers around a button-maker. Button-makers use indivisible inputs and specialized labor, so cost per button decreases as quantity increases. Buttons are not standardised inputs, so dressmakers incur modification costs to interact with the button-makers to produce the perfect button.
However, a cluster of several dressmakers can generate sufficient button demand to allow button-makers to exploit scale economies, leading to lower button prices. Larger total demand for buttons will allow the
button-maker to specialise in different varieties, thus reducing modification costs. Duranton and Puga (2003) further find that an increase in final production by virtue of sharing a wider variety of intermediate suppliers requires a less-than-proportional increase in primary factors.
Secondly, firms cluster because they can reap scale economies by sharing a labour pool, manifested through many agglomeration mechanisms. Marshall (1920) emphasized the risk-sharing properties of a large labour market. As individual firms become more or less productive, workers can shift across employers, thereby maximising productivity. Krugman’s (2001) model explains how this reduces the variance of worker wages, and firms in a cluster yield higher profits. Helsey and Strange (1990) note that larger markets improve the chances of matching between firms and workers, as well as the average quality of matches. Rotemberg and Saloner (2000) provide a model where firms cluster together so that workers will come and invest in human capital, since they do not face ex post appropriation. Combes and Duranton (2006) emphasizes employment sharing, where new startups locate near older firms so that they can hire away their workers.
Finally, external scale economies cause knowledge spillovers, where co-locating firms may facilitate innovations and can accordingly result in a productivity increase for firms (Ogawa and Fujita, 1980; Fujita and Ogawa, 1982). Proximity results in information sharing on new technologies and therefore leads to faster technology adoption (Griliches, 1958). Latest knowledge about technological developments is often valuable to firms only for a short period, and the reciprocal exchange of information among co-located firms engaged in innovation can decrease uncertainty (Feldman, 1994). Thus, it benefits innovative firms to locate near sources of information and each other (Carlino and Kerr, 2014). Saxenian (1994) argues that this is one cause of industrial concentration in Silicon Valley. Glaeser and Khan (2001) also argue that the urbanisation of high human capital industries like finance highlights the role that density plays in accelerating the idea flows. Telecommunication advancements may dismiss the importance of external scale economies in the geographical proximity of small and specialised firms (Heijman, 2007). However, high knowledge-intensive activities, like the development of new products, need both “face-to-face” contact and “face-to-product” contact. Therefore, only standardised products may become footloose (Desrochers, 1998; Heijman et al., 2004).
There are many other factors of agglomeration besides scale economies, including home market effects, urban consumption opportunities, and rent-seeking (Duranton and Puga, 2004). Kim (1995, 1999) and Ellison and Glaeser (1999) have also argued that natural advantages are very important in determining agglomeration. This suggests a smaller role for scale economies in the process of agglomeration. To shed light on this assertion, I proceed to examine exactly what is the role of scale economies in the system of cities, based on the theoretical underpinnings in Henderson (1974).
Henderson (1974) states that to provide an incentive for adding residents and creating a city, we must “add centripetal forces so that the agglomeration of people and firms generates benefits”. There are three stylized economies of scale models for doing this, and in each model, the agglomeration results in an optimal city size: “utility increases as residents are added until it peaks at some optimal city size when the centrifugal force of commuting cost and the centripetal agglomeration force are balanced at the margin”.
Ellison et al. (2010)’s work is relevant in showing these initial centripetal agglomeration forces. They studied the extent to which US manufacturing industries co-agglomerated based on an index of co-agglomeration, regressing it on measures of proximity between the two industries in terms of Marshall’s (1920) agglomeration economies: input-output linkages, labor pooling, and knowledge spillovers. Results reveal that sectors buying similar intermediates tended to co-agglomerate the most, followed by sectors that employ similar workers. To mitigate the possibility that certain industry pairs buy intensively from each other precisely because they are agglomerated, they used UK measures to instrument for US industry characteristics. However, as the authors concede, these instruments will only mitigate simultaneity bias if there are similarities in the ways in which natural advantage drives industry co-location in the US and in the UK. Nonetheless, similarity between OLS and IV estimation results for the determinants of agglomeration alleviates concern regarding simultaneity.
In analysing knowledge spillovers, Ellison et al. (2010) used two data sources: Scherer’s (1984) technology flow matrix, designed to capture the extent to which R&D activity in one industry flows out to benefit another industry, and the NBER Patent Database, using data on patent citations. However, using patent citations as a proxy for knowledge spillovers is problematic. In their own study, Jaffe et al. (2000) found that almost one-half of citations “do not correspond to any apparent spillover”. Moreover, the technology matrix and patent database capture only official exchanges of technology, omitting informal or accidental technology exchanges (Howard et al., 2015). These affect the accuracy of analysing knowledge spillovers as an agglomerative force. Krugman (1991) further questions the validity of using patent citations as spillovers "leave no paper trail by which they may be measured and tracked".
Unexplored gaps in Ellison et al. (2010) include the relative importance of the different agglomeration mechanisms within a narrow geographical scope. Some authors find that knowledge spillovers may be relevant but only at a very local level (Rosenthal and Strange, 2001; Jofre-Monseny et al., 2011). This is consistent with the study of Rosenthal and Strange (2008) who find that human capital spillovers are important and attenuate sharply with distance.
Moreover, it is unclear how these results would generalize to non-manufacturing industries. Services involve face-to-face interaction and are more costly to transport, thus we might think that input-output relationships are particularly important in service sectors (Kolko, 1999). In a similar study, Billings and Johnson (2013) included service sectors and found considerably less co-localisation effects. Ellison et al. (2010) further concede “ideas may be more important in more innovative sectors, so idea flows might be more important elsewhere”.
While Ellison et al. (2010) examined interdependencies among manufacturing industries, it would perhaps have been interesting to group industries into clusters to look at alternative measures of relatedness among industries. Previous studies have focused on science-based clusters (Porter, 2001, 2003; Feldman and Audretsch, 1999) and input-output-based clusters (Feser and Bergman, 2000). Nonetheless, directly measuring complementarity in economic activity in a consistent and unbiased manner is challenging (Delgado et al., 2012).
In sum, Ellison et al. (2010) focuses on explaining industry localization and co-agglomeration patterns of industries, notably input-output linkages, technology linkages, and occupational labor requirements. However, it does not consider agglomeration mechanisms in narrow geographical scopes, and fails to address the broader array of interdependencies associated with clusters of related industries and service industries.
We proceed to consider how scale economies fit into broader urban economics theory. Despite urban agglomeration economies, all production in most economies does not occur in just one mega-city (Paddison, 2000). City sizes are limited because of other diseconomies to sizes of cities, which are based on escalating commuting, congestion and land rent costs (Henderson, 1974; Mills, 1967). Different sizes of cities depend on different degrees of agglomeration (Henderson 1974, 1977); and different types of cities depend on which force of agglomeration is dominating in a city (Abdel-Rahman, 1990). We can also look beyond these theories to analyse the spatial distribution of cities (Christaller and Baskin, 1966; Lӧsch, 1954). Ellison et al. (2010)’s work is hence only relevant in explaining the centrifugal agglomeration forces of city size and type determination.
In conclusion, this essay discussed the importance of internal scale economies in intra-industry trade, but also showed how neoclassical trade theory may better account for empirical reality. This essay also suggests perhaps scale economies are important in regional, and not international, structure of production. External scale economies are then highlighted in urban economics, based on specialized suppliers, skilled labour pools and a common knowledge base. Finally, Ellison et al. (2010)’s work is used to examine the role of scale economies in agglomeration and broader urban economics theory.