There have been an increasing number of empirical studies on localisation and urbanisation in recent years. Ever since Ellison and Glaeser (1997) presented their index for industrial localisation based on their theoretical model, there has been much subsequent empirical work examining the effects of agglomeration on productivity using their findings. Rosenthal and Strange (2001), for example, use the index to try to find the determinants of agglomeration focusing on knowledge spillovers, labour market pooling, and input sharing for the manufacturing industries.
Many of the existing studies use data tallied regionally by administrative units (in Japan’s case, prefectures and municipalities) when measuring the degree of agglomeration. However, since the same prefecture has an area where the population is concentrated and other areas of lesser concentration, it is easy to imagine that the degree of agglomeration of establishments varies even within the same prefecture. Some establishments exist near the prefecture border. Since existing data assign a single index to all the establishments in the same prefecture in such cases, it is difficult to rule out the possibility of bias in the agglomeration indices.
To overcome this problem, we propose a distance-based index defined for each establishment, rather than one based on data aggregated per administrative unit level, as a measurement of agglomeration using micro-location (latitude and longitude) information from METI’s Census of Manufactures (Konishi and Saito 2012). Our index is designed to reflect locational heterogeneity within a region as an administrative unit and avoids biases caused by spatial correlations and other problems associated with segmenting data based on administrative boundaries. More specifically, the indicators take into account the agglomeration effect of individual establishments in relation to all others. In doing so, we have made an indicator based on distance, assuming that establishments that are located closer to one another and hire more employees will have a larger agglomeration effect. In addition, we have made two different indices – for industrial localisation-type agglomeration and urbanisation-type agglomeration – based on data from the four-digit Japan Standard Industrial Classification code (JSIC).
Using these indices, we measure the impact of these two phenomena on labour productivity for establishments with four or more employees and total factor productivity (TFP) for establishments with 30 or more employees. Figure 1 is a scatter chart showing the relationship of coefficients derived from the regression of labour productivity and the two agglomeration indices for each manufacturing industry according to the two-digit JSIC code. The vertical axis shows the elasticity of the industrial localisation index to labour productivity, while the horizontal axis indicates the elasticity of the urbanisation index to labour productivity. The elasticity tells us the percentage of increase in labour productivity per each 1% rise in the agglomeration indicator. In other words, Figure 1 is a scatter chart of two types of elasticity for each industry, and the blue dots ◆ (1995) and red dots ■ (2005) represent the industries.
Figure 1 shows that establishments with stronger agglomeration tend to have higher productivity in many industries for both localisation and urbanisation, given that the elasticity is positive in almost all industries in both indices. It also suggests that there is an upward-sloping trend in both years. For both years, the straight line represents the linear approximation of the average relationship of elasticity of both indices. This means that industries in which the urbanisation index has a stronger tendency to increase labour productivity also show a more positive effect of the localisation index on productivity (and vice versa). In addition, the fact that the straight line for 2005 lies slightly above that for 1995 suggests that the industrial localisation effect compared with urbanisation effect was slightly stronger in 2005.
Figure 1. Scatter plot of elasticity of industrial localisation and urbanisation indices to labour productivity
Meanwhile, TFP values are positively correlated with the urbanisation index, whereas TFP and the industrial localisation index are uncorrelated in most industries. As an exception, we find some industries for which both the industrial localisation and urbanisation indices are positively correlated with TFP, all of them categorised as declining industries. In some of the industries categorised as growing, no correlation is observed between TFP and agglomeration, whether urbanisation or industrial localisation. Moreover, we observed a negative impact of localisation on TFP in some growth and high-tech industries. The above results show that a promotion of an urbanisation policy is required for growth industries, while policies not only facilitating urbanisation but also deepening localisation are effective for declining industries. These findings imply that certain attributions – establishment size, industry type, etc. – can serve as a guide for the government in identifying the effective target of the cluster promotion policy.
Ciccone, A and RE Hall (1996), “Productivity and the Density of Economic Activity”, American Economic Review, 86(1):54-70.
Ellison, G and E Glaeser (1997), “The Geographic Concentration in U.S. Manufacturing Industries: A Dartboard Approach”, Journal of Political Economy, 105(5):889-927.
Konishi, Y and Y Saito (2012), “Measuring the Impact of Establishment Level Agglomeration on Productivity: Industrial specialization and urbanization effects”, RIETI Discussion Paper, 12-J-006, (in Japanese).
Rosenthal, SS and WC Strange (2001), “The Determinants of Agglomeration”, Journal of Urban Economics, 50(2):191-229.
Sveikaukas, L (1975), “The Productivity of Cities”, The Quarterly Journal of Economics, 89(3):393-413.