Geography of productivity: evidence from China’s manufacturing industries

Abstract

Empirical evidence has confirmed that the locational agglomeration of firms is associated with productivity benefits. Several competing explanations have been proposed for this association: agglomeration externalities effect, sorting effect and selection effect. This work builds on these studies and seeks to make four contributions by using a firm-level dataset on China’s manufacturing industries over the 1999–2007 period. First, it points out that this stand of literature pays insufficient attention to the possibility of adverse sorting. Second, agglomeration externalities have been decomposed to two constituting parts: inter- and intra-industry, which play different roles in the process of sorting and agglomeration. Third, this paper also points out the need to pay attention to firm heterogeneity, in terms of market orientation and ownership type. Finally, the complex role of government policies has been also examined. Econometric results support our hypotheses.

1 Introduction

Space matters, since it is a major determinant of the efficiency of economic and social interactions (Venables 2011). The spatial dimension of inequality in economic activity arises endogenously and often persists. The level, causes and evolution of spatial inequalities have attracted massive inter-disciplinary attention, resulting in a wide range of literature emerging recently that have sought to measure spatial inequality and/or examined its development over time (Castells-Quintana and Royuela 2014; Hudson 2007; Pike and Tomaney 2009). Spatial inequalities have also received interests from policy makers, since regional disparities in economic activity as well as other economic indices boost overall income inequality. Therefore, they are often associated with political and ethnic conflicts and cause social unrest and political instability (Kanbur and Zhang 2005). The question why we observe large inter-regional variations in economic development takes on significance in policy makers’ search for solutions to income inequality and the subsequent social, economic and political issues.

Extant literature tends to emphasize region’s different factor conditions, the quality and quantity of input factors (e.g., capital, labor, technology and infrastructure) as key explanatory factors for regional inequality (Florida and Kenney 1988; Wan et al. 2007), while the importance of factor productivity as a driving force of regional inequality has been largely overlooked (He et al. 2017). Rice et al. (2006) have attributed most of the inter-regional income difference to that in productivity between regions. Regional productivity is further related to its inhabitants—firms, which vary from one another drastically in terms of employment and output size, capital and skill intensity (Bernard and Bradford Jensen 1999; Saito and Gopinath 2009). More importantly, firms in industrial clusters and large cities are often more productive than those located elsewhere (Ciccone and Hall 1996; Henderson 1986). This has long been explained by agglomeration externalities that impact firm performance either through localization (or specialization) economies based on a specialized labor market pooling (matching), the sharing of intermediate inputs and local knowledge spillovers (learning) (Duranton and Puga 2004; Marshall 1920 [1890]), or through Jacobs’s externalities derived from the existence of a large variety of industries in the regional economy (Jacobs 1969). The resulting spatial pattern affects firms’ productivity and leads to spatial inequalities.

Agglomeration externalities, however, are only one reason for regional disparity in productivity. One key component of the agglomeration process is the spatial sorting of heterogeneous firms to locate in regions with specific characteristics. High-productivity firms may ex ante sort into large cities and industrial clusters, since they benefit more from agglomeration than do their low-productivity counterparts (Baldwin and Okubo 2006; Behrens et al. 2014; Forslid and Okubo 2014; Venables 2011). As a result, average firm productivity in large cities and clusters is often higher (Saito and Gopinath 2009). Another mechanism of regional inequalities in productivity is selection. That is, in large cities and clusters, intensification of competition shakes out low-productivity firms, and consequently only high-productivity firms may survive and ex post profitably operate there (Behrens et al. 2014; Melitz 2003; Melitz and Giancarlo 2008; Puga 2010). In this sense, agglomeration improves regional average productivity, though it does not improve productivity of each firm as agglomeration externalities literature expects (Arimoto et al. 2009).

This paper builds on these studies on regional inequality in productivity, by investigating the spatial pattern of China’s new firm formation, and seeks to make four new contributions. First, we point out the possibility of adverse sorting, as high-productivity firms may be less likely to sort into large cities and clusters for some reason, which has been largely overlooked in extant literature on agglomeration, selection and sorting effects. Second, this paper differentiates agglomeration externalities derived from intra-industry agglomeration externalities, and the spread of knowledge and localized capabilities from related industries. Third, we also point out the need to pay more attention to firm heterogeneity, since heterogeneous firms may be affected positively or negatively by agglomeration of different types of firms to different extents. Finally, this paper also attempts to examine the role of government policies.

2 Regional inequality of productivity: agglomeration, sorting and adverse sorting

There has long been concern about regional inequalities in economic activity and income in both developing and developed countries, whether they are increasing or decreasing and the extent to which they can be alleviated and through what measures (Rey and Janikas 2005; Shorrocks and Wan 2005; Sutherland and Yao 2011). This has resulted in various strands of literature, most of which confirm the positive relationship between regional average productivity level and agglomeration of employment, economic and industrial activities (Ciccone and Hall 1996; Henderson 1986; Saito and Gopinath 2009). Common explanations for this empirical finding are agglomeration economies external to firms taking place within regions lead to region-wide increasing returns. This strand of literature tends to distinguish between external economies that are restricted within particular sectors of the economy (i.e., localization externalities), and those that flow across sectors of the economy, urbanization externalities. For Marshall (1920 [1890]), economies from specialization in industry towns derived from local input–output networks, dense local labor pools and knowledge spillovers. Jacobs (1969) was more generally concerned with diversification externalities resulting from interaction, generation, replication, recombination and modification of ideas and applications across different sectors. All these mechanisms share a common prediction: the concentration of firms and workers in space makes them more productive (Andersson and Lööf 2011; Combes et al. 2012).

Apart from the agglomeration externalities story, high-productivity levels of large cities and clusters can also be explained in terms of competition-based selection among firms in such regions, as proposed by the heterogeneous firms literature (Melitz 2003; Melitz and Giancarlo 2008). Firm heterogeneity results in their varying placement across the productivity scale. Severe market competition in large cities and clusters induce low-productivity firms to close down, and only high-productivity firms may survive or profitably operate there, leading to increasing regional average productivity (Forslid and Okubo 2014; Hasan et al. Ottaviano 2012). As argued by Baldwin and Okubo (2006), this strand of literature establishes a new agglomeration force: stronger competition may lower profits and reduce firms’ willingness to stay in large cities and clusters. As a result of a process of competitive selection, a remaining mass of high-productivity firms supply both the domestic and international markets, and low-productivity firms supply either the domestic market or are driven from the market altogether.

A third mechanism is ex ante spatial sorting of heterogeneous firms: high-productivity firms and talented individuals tend to sort themselves to large cities and clusters (Forslid and Okubo 2014; Hasan et al. 2016; Puga 2010; Venables 2011). Based on a theoretical framework combining the “footloose” capital and location model (Martin and Rogers 1995) and the heterogeneous firms model (Melitz 2003), Baldwin and Okubo (2006) shows why spatial sorting occurs, i.e., why firms that choose to locate in large cities and clusters tend to have above-average productivity. In new economic geography model, the agglomeration forces consist of forward and backward linkages, while the dispersion force is associated with local market competition and crowding (Krugman 1991). High-productivity firms are systematically subject to stronger agglomeration forces and weaker dispersion forces than their low-productivity counterparts. On the one hand, high-productivity firms have lower marginal costs, and they tend to sell more. As a result, the forward and backward linkages operating in the large cities and clusters are systematically more attractive to high-productivity firms. On the other hand, high-productivity level of these firms also means that they are systematically less harmed by the stronger local market competition in large cities and clusters, or they are better equipped for coping with the higher degree of local competition in such regions.

The line between sorting, selection and agglomeration effects is, however, very thin and sometimes blurred (Behrens et al. 2014). On the one hand, due to competition-based selection in large cities and clusters, only high-productivity firms and talented individuals will sort into such regions in the first place. In this sense, selection induces sorting. On the other hand, large cities and clusters with talented individuals, where competitive selection is tough, often end up with high-productivity firms that can pay high wages. In turn, this attracts more workers and makes such regions grow bigger, further reinforcing agglomeration externalities. In short, there is a need to acknowledge the complementarities between these three mechanisms. In this paper, we thus see sorting and selection as being interconnected with each other.

We build on these arguments and seek to make four contributions. Some studies on market entry and firm heterogeneity have stressed that high-productivity, technology-intensive firms are less, rather than more, willing to locate in large cities and clusters (Alcácer and Chung 2007; Belderbos and Carree 2002; Belderbos et al. 2015; Myles Shaver and Flyer 2000). The explanation for such adverse sorting relates to the role of knowledge spillovers and adverse knowledge spillovers in local agglomerations. High-productivity firms with advanced technologies, innovative capacities, organizational and managerial skills contribute the most to local knowledge spillovers, while simultaneously having the most to lose if locating in large cities and clusters. Competitors may mimic high-productivity firms’ product designs, innovations and organizational approaches, or acquire the latter’s knowledge and expertise through labor poaching. Our first contribution is thus to point out that such an asymmetry in knowledge spillovers could result in a process of adverse sorting, whereby low, rather than high, productivity firms are more likely to sort into large cities and clusters (Belderbos et al. 2015; Greenstone et al. 2010).

As a further step of our analysis, we inspect that whether the sorting and adverse sorting effects of agglomeration is dependent on the specific characteristics of agglomeration and whether firms are affected by their neighbors in different industrial sectors in the same way. Our hypothesis is that high-productivity firms may sort to regions with a concentration of firms in related industries (sorting effect), whereas the presence of a cluster of firms in the same industry sharing the same, narrowly defined product market tends to discourage high-productivity firms to co-locate due to the asymmetry in knowledge spillovers (adverse sorting effect). Here we decompose agglomeration externalities into two constituting parts, by distinguishing intra-industry knowledge spillovers from inter-industry knowledge spillovers that result from the broader pattern of firms in related industries that could share suppliers, customers, and knowledge (Alcácer and Chung 2014; Belderbos et al. 2015; Ellison et al. 2010; Glaeser and Kerr 2009). This is the second contribution of this research.

Hypothesis 1a

High-productivity firms are more likely to sort to regions with a concentration of firms in related industries due to sorting effect.

Hypothesis 1b

High-productivity firms are less likely to sort to regions with a concentration of firms in the same industry due to adverse sorting effect.

As noted above, selection leads to sorting, as only high-productivity firms sort into large cities and clusters given the tough competitive selection in such regions. However, this argument has been questioned by some empirical studies as they have found no evidence of positive selection or sorting effects for firms in metropolitan areas (Belderbos et al. 2015; Combes et al. 2012). This may be due to the fact that in some large cities and clusters, most firms’ markets are international rather than local, such that the degree of market competition is not dependent on local firm density and agglomeration. In addition to market orientation, firms in a locality also vary dramatically in their ownership structures, specific technologies, competences, innovative and productive capabilities, and business models, even though they may all belong to the same industry (Antonelli and Scellato 2015). Since firm-specific competences, organizational routines and knowledge bases affect the ways in which and the extent to which firms interact with each other, it would be problematic to examine the effect of agglomeration externalities and knowledge spillovers without paying attention to firm heterogeneity (Lo Turco and Maggioni 2015; Lööf and Nabavi 2015). Our third contribution is to argue that knowledge spillovers, adverse knowledge spillovers and market competition effects between firms with different ownership types and market orientations may be attenuated if they focus on different niche markets.

Hypothesis 2

Sorting and adverse sorting effects may be shaped by firm heterogeneity.

Finally, our last contribution concerns the impact of regional policy. Most regional policies attempt to alleviate regional inequalities in productivity and increase the share of industrial activities in peripheral areas (Pessoa 2014). This paper focuses specifically on government subsidies. Baldwin and Okubo (2006) and Gaubert (2017) have highlighted that government subsidies at the regional level may attract low-productivity firms to the targeted zone since they have the lowest opportunity cost of leaving the core regions, i.e., large cities and clusters. On the other hand, government subsidies can be distributed in such a way as to increase/decrease market competition (Aghion et al. 2015), thereby strengthening/weakening sorting and adverse sorting effects. China’s decentralization granted local governments more autonomy and allowed them to get involved in shaping the regional economy (He et al. 2008). In China, different local governments have offered different levels of government subsidies in order to entice new entrants, providing a rich context to study the role of regional policies in alleviating or reinforcing regional disparity.

Hypothesis 3a

Government subsidies are likely to attract low-productivity firms.

Hypothesis 3b

Sorting and adverse sorting effects may be dependent on how government subsidies are allocated.

3 Data and research design

This paper uses the China’s Annual Survey of Industrial Firms (ASIF) (1999–2007). The ASIF is administered by the National Bureau of Statistics of China and covers all Chinese industrial state-owned enterprises (SOEs) and non-SOEs with annual sales of five million Yuan (around USD 600,000 at the exchange rate of 2000) or more. The database provides firm-level data on firm structure and operation. A comparison with the 2004 full census of industrial firms reveals that these firms generated 90% of output and 98% of exports. This study focuses on firms in 424 4-digit manufacturing industrial sectors and their entry into China’s 337 prefecture-level cities.

The following equation is estimated using the OLS model, to examine whether high-productivity firms are more or less likely to enter cities with a concentration of firms in the same or other related industrial sectors.

$$ \begin{aligned} New\;Entrant\_TFP_{f,c,t} & = \beta_{0} + \beta_{1} Location\;Quotient_{c,i,t - 1} \\ &\quad + \beta_{2} Inter{ - }industry\;relatedness_{c,i,t - 1} + \beta_{3} X + \varepsilon_{f,c,t}\end{aligned} $$

(1)

The dependent variable is the total factor productivity (TFP) of a new entrant (firm f) in city c and year t (New Entrant_TFP_f,c,t). The ASIF 1999–2007 dataset includes each firm’s start year. In year t’s dataset (t = 1999, 2000, …, 2007), if a firm f’s start year equals t, it is seen as a new entrant in year t. We calculate TFP for each new entrant using the semi-parametric algorithm developed by Olley and Pakes (1996). X represents control variables. Lagged terms of independent and control variables are adopted. We pool all the observations during the period of 1999–2007. Given the multi-level structure of Eq. (1), standard errors are clustered at the 4-digit industry, city and year level.

To investigate whether the sorting and/or adverse sorting effects of agglomeration is dependent on the specific characteristics of agglomeration and whether firms are affected by their neighbors in different industries in different ways, we need to find indicators to measure intra-industry and inter-industry agglomeration externalities, respectively. We use the location quotient (LQ) of industry i in city c as a proxy of the agglomeration of the same industry, which is calculated as the ratio of the share of industry i in city c’s total employment to the share of this industry in the national total.

$$ {\text{LQ}}_{c,i} = {\raise0.7ex\hbox{${Employment_{c,i} /\mathop \sum \nolimits_{i} Employment_{c,i} }$} \!\mathord{\left/ {\vphantom {{Employment_{c,i} /\mathop \sum \nolimits_{i} Employment_{c,i} } {\left( {\mathop \sum \nolimits_{c} Employment_{c,i} /\mathop \sum \nolimits_{c,i} Employment_{c,i} } \right)}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\left( {\mathop \sum \nolimits_{c} Employment_{c,i} /\mathop \sum \nolimits_{c,i} Employment_{c,i} } \right)}$}} $$

(2)

where Employment_c,i is the number of employees in industry i and city c, which is calculated by adding up the number of employees of all firms in industry i and city c in the ASIF.

To examine the impact of the agglomeration of firms in related industries, an industry relatedness index is required. We appeal to the industry space representation developed by Hidalgo et al. (2007). The industry space is a network that formalizes the notion of relatedness between industries. Hidalgo et al. (2007) assume that two industries are related with each other if regions tend to have revealed comparative advantage (RCA) in both. The RCA indicator measures export output specialization in international trade. In this research, the LQ is used to measure domestic output specialization (Isard 1960). A region has a revealed locational advantage in an industry when the share of this industry in the region’s total employment is larger than the share of this industry in the national total. The relatedness (ϕ) between industry i and industry j is calculated as:

$$ \phi_{i,j} = \hbox{min} \left\{ {P\left( {{\text{LQ}}_{c,i} > 1|{\text{LQ}}_{c,j} > 1} \right),\left. {P\left( {{\text{LQ}}_{c,j} > 1|{\text{LQ}}_{c,i} > 1} \right)} \right\}} \right. $$

(3)

where LQ_c,i is the location quotient of industry i in city c. City c is considered as having a revealed locational advantage in industry i, if LQ_c,i is above 1. The relatedness between industry i and industry j is the minimum of the conditional probability of specializing in industry i, given that city c has an industrial specialization in industry j (i.e., P (LQ_c,i> 1|LQ_c,j> 1)), and the conditional probability of specializing in industry j, given a revealed locational advantage in industry i (i.e., P (LQ_c,j> 1|LQ_c,i> 1)). The rationale behind this relatedness indicator is that if two industries are related with each other, they probably demand similar factor inputs, resources, capabilities and technology and are likely to be produced together. Based on Eq. (3), this bilateral relatedness indicator between industry i and j is calculated for China’s 424 4-digit industries. The matrix of these relatedness indicators characterizes the industry space.¹

Based on the relatedness indicator, we calculate the average relatedness between industry i that firm f belongs to and city c’s industrial structure. The inter-industry relatedness indicator has been developed to measure to what extent industry i that firm f belongs to is related to the existing industrial structure of city c. Recent studies suggest that if an industry is related to a number of industries in which a region is specialized, the inter-industry relatedness of this industry in the region is high, and agglomeration externalities derived from firms in related industries should be strong. The inter-industry relatedness indicator is thus measured as follows:

$$ Inter{ - }industry\;relatedness_{c,i} = \frac{{\mathop \sum \nolimits_{j} x_{c,j} \phi_{i,j} }}{{\mathop \sum \nolimits_{j} \phi_{i,j} }} $$

(4)

where x_j,c,t takes the value of 1 if city c has a revealed locational advantage in industry j, and zero otherwise. Inter-industry relatedness around an industry will be high if a region is specialized in most of the industries related to the industry under consideration and is thus used as a proxy of inter-industry agglomeration externalities.

Three control variables are included. The first two variables correspond to factors with respect to localized capabilities, which play critical roles in indigenous knowledge creation and local knowledge spillovers (Maskell and Malmberg 1999; Zhu and Fu 2013). One is an index of regional research and development capacity, measured as the number of universities in city c (University). The other is the length of highway divided by the number of population in city c as a proxy of physical capital and infrastructure (Highway). GDP per capita is added to control the impact of regional economic development on new firm formation. Data on these three variables are derived from China’s City Statistical Yearbooks.

4 Geography of productivity in China

The impressive economic growth of China has been driven to a large extent by an export-oriented industrialization model and China’s integration into the global economy. The resulting geographies of manufacturing employment and production were significantly shaped by global sourcing for export markets (Fig. 1). Production and employment have become heavily concentrated in the coastal regions of East and Southeast China, with primary concentrations in Shandong, Jiangsu, Zhejiang and Guangdong provinces, and some outliers in regional centers, such as those in central China along the Yangtze River. Figure 1 shows the progressive regional agglomeration of China’s manufacturing industries during 1999–2007.

Fig. 1

Spatial distribution of employment in China’s manufacturing industries by city

The top 25% firms in terms of TFP are classified as high-productivity firms, whereas the bottom 25% are considered as low-productivity firms. Figures 2 and 3 show the spatial distribution of new high- and low-productivity firm formation in China’s manufacturing industries by city in 1999 and 2007. First, the numbers of both high- and low-productivity, new entrants have increased dramatically in most Chinese cities from 1999 to 2007. Second, high-productivity firms tended to enter China’s coastal regions and some regional economic centers where a large number of establishments were located (Figs. 1, 2). This is in line with the theoretical predication that high-productivity firms sort into large cities and industrial clusters, since they benefit more from agglomeration than low-productivity firms do (i.e., spatial sorting). Third and more interestingly, low-productivity firms seem to be more likely to enter industrial clusters along China’s coastal line and the Yangtze River as well (Fig. 3), indicating such firms somehow also managed to benefit from agglomeration and tended to enter large cities. In other words, spatial sorting alone does not fully explain the geography of productivity in China. One possible explanation for the third finding could be adverse spatial sorting effect. To better understand this conundrum, we further analyze the relationship between each new entrant’s TFP in a city and the degree of the city’s agglomeration of firms, and distinguish agglomeration externalities derived from different types of firms, to which we now turn.

Fig. 2

Number of high-productivity new entrants by city (1999 and 2007)

Fig. 3

Number of low-productivity new entrants by city (1999 and 2007)

5 Empirical results

In our estimation, the logarithm of GDP per capita is used. The geographical unit of analysis is China’s prefectural level city. Table 1 reports the econometric results. A multi-collinearity diagnostic test tells that the highest variance inflation factor (VIF) is 2.49, below the critical point of 10, suggesting that multi-collinearity is quite trivial. In Model 1, we focus on all Chinese manufacturing firms. The sign of Highway is positive, indicating that high-productivity firms tend to enter cities with advanced infrastructure. High-productivity firms also prefer cities with a presence of a large number of universities, which not only represent the city’s R&D capacity but also serve as a source of skilled labor and help promote the city’s human capital. As a result, investments in infrastructure, R&D capacity and human capital would foster the formulation of a favorable environment for entrepreneurial activities and particularly attractive to high-productivity firms (Lo Turco and Maggioni 2015). The sign of GDP per capita’s coefficient suggests that more developed regions are more likely to entice high-productivity firms, due probably to high-productivity firms’ desire for large markets.

Table 1 Estimation results on spatial sorting and adverse sorting

Moving onto the results connected more closely with the central arguments, as clarified in Sect. 3, we distinguish intra-industry agglomeration externalities (Location quotient in Table 1) from inter-industry agglomeration externalities that result from the broader pattern of firms in related industries that could share suppliers, customers, infrastructure and knowledge (Inter-industry relatedness in Table 1). The parameter of Inter-industry relatedness is positive and significant in Model 1, which is consistent with the theoretical prediction proposed by the spatial sorting effect theory (H1a). High-productivity firms sort into cities with a concentration of firms in related industries, since they not only benefit more from intensive forward and backward linkages with firms in related industries in large cities but also target at relatively different niche markets from those of firms in related industries (Forslid and Okubo 2014; Hasan et al. 2016). However, as is shown in Model 1, industrial specialization (Location quotient) has a negative and significant impact on high-productivity firms’ entry, supporting the existence of adverse sorting effect (H1b). Competitors in the same industry sharing the same, narrowly defined product market may mimic high-productivity firms’ product design, innovations, organizational and managerial approach, or ‘steal’ the latter’s know-how through labor poaching (Alcácer and Chung 2007; Belderbos and Carree 2002; Myles Shaver and Flyer 2000). Meanwhile, high-productivity firms that already possess advanced technologies and innovative capacities benefit the least from and contribute the most to the cluster of firms in the same industry. Such an asymmetry result in a process of adverse sorting, whereby high-productivity firms are less, rather than more likely to sort into large cities crowded by firms in the same industry. In short, our empirical results indicate that spatial sorting and adverse sorting co-exist, as inter- and intra-industry agglomeration externalities play completely different roles in luring high/low-productivity firms.

This paper also seeks to examine the impact of government policies that have the potential to influence the locational choice of firms, by paying particular attention to the role of subsidies provided by Chinese local governments. Two independent variables are added in Model 2 in Table 1 accordingly. Subsidy is simply defined as the ratio of subsidies to industrial output in city c. Furthermore, we follow Aghion et al. (2015), and calculate the sectoral dispersion of subsidies (CompHerf_subsidy). First, we use the Herfindahl index to measure the concentration of subsidies within sector i in city c and year t.

$$ Herf\_subsidy_{c,i} = \mathop \sum \limits_{ci} \left( {\frac{{Subsidy_{c,f,i} }}{{\mathop \sum \nolimits_{ci} Subsidy_{c,f,i} }}} \right)^{2} $$

(5)

where Subsidy_c,f,i denote the amounts of subsidies received by firm f in industry i and city c from governments. As with standard Herfindahl indices, a smaller number indicates a higher degree of the dispersion of subsidies, that is, a more equitable allocation of those across firms in the sector. We then calculate the sectoral dispersion of subsidies, CompHerf_subsidy, as follows,

$$ CompHerf\_subsidy_{c,i} = 1 - Herf\_subsidy_{c,i} $$

(6)

The higher the sectoral dispersion of subsidies, the more competitive business environment local governments seek to formulate.

On the one hand, the parameter of Subsidy is negative and significant, confirming our hypothesis that government subsidies at the city level designed to entice new entrants tend to entice the least efficient ones (H3a). The reason is that the most inefficient firms are the ones that have the least to lose from not locating in large cities but chasing for government subsidies. This may explain why some regional government subsidies are ineffective in improving regional competitiveness. On the other hand, the coefficient of CompHerf_subsidy is positive and significant, implying that high-productivity firms can be lured to certain regions with higher levels of dispersion of subsidies. The higher the sectoral dispersion of subsidies, the more competitive business environment local governments seek to formulate. High-productivity firms are systematically less harmed by severe market competition, and/or are more capable of dealing with higher levels of local competition, whereas low-productivity firms are often frightened off by severe competition. More importantly, in Model 3, we add two interaction terms, CompHerf_subsidy*Location quotient and CompHerf_subsidy*Inter-industry relatedness. The former’s coefficient is negative and significant, while the latter’s is positive and significant. This implies that both sorting effects derived from inter-industry knowledge spillovers and adverse sorting effects stemming from intra-industry knowledge spillovers are reinforced in regions with higher levels of dispersion of subsidies. Government subsidies can be used as a tool to boost market competition (Aghion et al. 2015), further strengthening intra- and inter-industry knowledge spillovers. Sorting and adverse sorting effects could thus be shaped by how government subsidies are allocated (H3b).

Firms in different industries are not equally affected by agglomeration externalities. Based on learning processes in different industrial sectors, Pavitt (1984) has developed a taxonomy of industries. At the extreme ends of Pavitt’s spectrum are supplier-dominated and science-based industries. Supplier-dominated industries include firms in traditional, labor-intensive sectors. The technological intensity of these industries is low due to their narrow knowledge base. In contrast, science-based industries are characterized by high levels of investment in R&D and innovation aimed at generating new products. Models 4–7 in Table 1 report the econometric results on two supplier-dominated industries (textile and furniture) and two science-based industries (general-purpose and electrical machinery). Inter-industry relatedness has a positive and significant coefficient in all four models, indicating that strong inter-industry agglomeration externalities can attract high-productivity firms in all four industries (i.e., sorting effect). Adverse sorting effect among firms in the same industry is, however, only witnessed in science-based, technology-intensive industries, since imitation, adverse knowledge spillovers and labor poaching between firms have relatively more negative impacts on high-productivity entrants in such industries than in traditional, labor-intensive industries.

Table 2 reports the empirical results on whether the effects of sorting and adverse sorting on new high-productivity firm formation are contingent on firm heterogeneity. We distinguish agglomeration externalities derived from different types of firms: exporters and non-exporters; state-owned enterprises, privately owned enterprises and foreign-owned enterprises. While the Location quotient indicator in Table 1 is calculated based on all firms, in Table 2 we use only the city’s exporters, non-exporters, state-owned firms, private firms, or foreign firms bundle to calculate Location quotient_exporter, Location quotient_non-exporter, Location quotient_state, Location quotient_private, and Location quotient_foreign, respectively. Likewise, we calculate Inter-industry relatedness_exporter, Inter-industry relatedness_non-exporter, Inter-industry relatedness_state, Inter-industry relatedness_private, and Inter-industry relatedness_private, according to whether the city’s exporters, non-exporters, state-owned enterprises, private firms, or foreign firms bundle is used as the references to identify revealed locational advantage in Eq. (4) (i.e., x_c,j).

Table 2 Estimation results on sorting, adverse sorting and firm heterogeneity

The highest VIF value, associated with the variable Inter-industry relatedness_non-exporter, is 6.06, indicating that multi-collinearity is not severe. The key findings are as follows. First, we found no evidence of adverse sorting effects between exporters in the same industry (Location quotient_exporter in Model 1). Rather, a cluster of exporters in the same industry attract high-productivity exporters, due probably to the fact that their markets are international rather than local, such that the degree of market competition is not dependent on local firm agglomeration. Similarly, high-productivity domestic-oriented entrants are reluctant to co-locate with other domestic-oriented firms (Inter-industry relatedness_non-exporter in Model 2), due to the former’s concern of adverse agglomeration externalities and labor poaching. Meanwhile, both domestic- and export-oriented, new entrants value agglomeration externalities derived from the other type of firms in related industries (Inter-industry relatedness_non-exporter in Model 1 and Inter-industry relatedness_exporter in Model 2), since they do not compete directly with one another and can potentially establish intensive forward and backward linkages.

Second, high-productivity new entrants, especially private firms, are likely to enter regions with a concentration of state-owned and foreign enterprises in related industries (Model 3–5), confirming sorting effects between firms with different ownership types. However, high-productivity, foreign and private firms are unwilling to enter regions crowded by private firms in both the same industry and related industries (Model 4–5), suggesting adverse sorting effect stemming from both intra- and inter-industry knowledge spillovers. One possible explanation is that, in China where the protection of intellectual property rights is weak, private firms with relatively weak R&D capacities often focus on imitating their neighbors’ technologies rather than innovation (Fan et al. 2013). The presence of a cluster of private firms may thus frighten off new, high-productivity foreign and private firms in the same industry and related industries. In short, sorting and adverse sorting effects are also dependent on firm heterogeneity (H2).

The potential issue of reverse causality must be addressed. First, this issue is not serious in our research, since our samples only include new entrants. It is unlikely for intra- and inter-industry agglomeration externalities in year t − 1 to be affected by the TFP of new entrants in year t. Second, as a robustness check, we re-estimate all models by using 2-year intervals (see Tables 3, 4 in “Appendix”). Compared with the results presented above, these changes produce only minor effects. Finally, the dependent variable is the TFP of a new entrant in year t. Our data only capture the new entrants’ TFP at the end of the year. For instance, if a firm enters a city in January, there will be almost one whole year from the time point of new firm entry to the end of year, that is, the time point of data collection. The firm may be able to boost its productivity during this period by taking advantage of intra- and/or inter-industry agglomeration externalities. In this case, we are not sure if it is high-productivity firms sorting into regions with high/low levels of agglomeration or new entrants improving their productivity during the period from the moment of new firm formation to the time point of data collection. Although this period is no longer than one year, it may still have some impacts. Consequently, as a robustness check, we only include firms founded after June. The estimated parameters of all variables are mostly unaltered, except significance changes in a few cases (see Tables 5, 6).

Table 3 Estimation results on sorting and adverse sorting (two-year lag)

Table 4 Estimation results on sorting, adverse sorting and firm heterogeneity (two-year lag)

Table 5 Estimation results on sorting and adverse sorting (new entrants from July to December)

Table 6 Estimation results on sorting, adverse sorting and firm heterogeneity (new entrants from July to December)

6 Conclusion

Recent empirical studies have confirmed that the locational agglomeration of firms is associated with productivity benefits. Several competing explanations have been put forward for this correlation. The predominant one is the notion of Marshallian and Jacobsian agglomeration externalities, which contends that firms can enjoy positive externalities derived from geographical industrial clustering. A second explanation is based on the selection effect associated with increased competition within clusters. Specifically, co-location of firms in local markets leads to tougher competition, forcing the exit of weak firms with low-productivity. Finally, high-productivity firms may sort into industrial clusters since agglomeration benefits may be more pronounced for such firms.

The idea that there are greater advantages and chances of survival in highly agglomerated locations for high-productivity firms would imply that these firms self-select into large cities and industrial clusters. However, the possibility of adverse sorting, whereby high-productivity firms may be less likely to sort into large cities and clusters, has been largely overlooked by extant literature. In addition, we differentiate intra- and inter-industry agglomeration externalities. By using firm-level data of China’s manufacturing industries during 1999–2007, this study shows that in the case of inter-industry agglomeration externalities, spatial sorting effect dominates while adverse sorting effect is much more prevalent in the case of intra-industry agglomeration externalities. Spatial sorting and adverse sorting are not mutually exclusive, but often co-exist, as inter- and intra-industry agglomeration externalities play completely different roles in luring high/low-productivity firms.

Regional government subsidies may attract low-productivity firms since they have the least to lose from not locating in large cities but chasing for government subsidies, whereas high-productivity firms can be lured to regions with higher levels of dispersion of subsidies. Furthermore, spatial sorting effect prevails in most industries, while adverse sorting effect is much more dominant in technology-intensive industries. Finally, the effects of sorting and adverse sorting on new high-productivity firm formation are contingent on firm heterogeneity in terms of market orientation and ownership. In this sense, this research has adopted a more complicated view that envisages regional inequality in productivity as contingent not only on spatial sorting and adverse sorting effects, but also on a variety of factors, such as firm heterogeneity and government intervention.

Several policy implications can be drawn from the empirical findings. First, it is much easier for regional governments to attract high-productivity firms that are related to regional preexisting industrial structure, particularly in technology-intensive industries, due to sorting effects. Second, regional governments should also factor firm heterogeneity into their industrial policies, since high-productivity firms often value agglomeration externalities derived from other types of firms in terms of market-orientation and ownership structure, with which they could potentially establish forward and backward linkages. Finally, government subsidies are a double-edged sword. Subsidies at the regional level are only attractive to least efficient firms, whereas more evenly distributed subsidies may be used to entice high-productivity firms.

Appendix

See Tables 3, 4, 5 and 6.