Relatedness economies, absorptive capacity, and economic catch-up: firm-level evidence from China

Abstract

This paper examines the effect of agglomeration economies on firm productivity and the role of absorptive capacity in China. Taking into account the transitioning economy context, I further exploit the gradual and spatially uneven implementation of market-oriented reforms to track the relative importance of firms’ absorptive capacity for economic catch-up. The three main results are as follows. First, firms exhibit higher productivity when located in an area with a denser network of related activities, particularly for firms with higher absorptive capacity. Second, the sources of the productivity gains are driven by externalities that arise from better access to inputs, similar workers and technological-related knowledge spillovers. Third, the role of absorptive capacity becomes more important for boosting productivity following more versus less intensive market-oriented economic reforms. These findings have important policy implications for transitioning economies and suggest that market reforms promote efficiency improvements related to performing actual R&D as well as encourage firms to seek out external sources of knowledge in order to attain market competitiveness.

1. Introduction

Chinese economic reforms led to a growth “miracle” with sustained high growth for several decades, resulting in China’s emergence as a key factor in the global economy.¹ Some evidence suggests, however, that the reform measures led to a one-time level effect on productivity (Zheng et al., 2009), leading many to question the future sustainability of China’s traditional growth model based on low labor costs, exports, and heavy investments. Since the turn of the century, China has placed increasing emphasis on promoting its “indigenous” innovation capabilities in order to shift away from its traditional growth strategy.²

Theories of economic growth support China’s policy shift towards an innovation-led growth strategy. Endogenous growth theory argues that technological progress fostered by investments in innovation is one of the most important determinants of productivity growth (Romer, 1990; Grossman and Helpman, 1991). Similarly in the empirical literature, micro-based evidence confirms that investments in innovation, measured by R&D spending, enhance firm productivity (Griliches, 1973).³ Beyond generating productivity gains by supporting innovation, R&D spending can also be used to increase the firm’s (intangible) knowledge assets by increasing its learning ability or absorptive capacity (Cohen and Levinthal, 1989).⁴

Within an economic catch-up framework, many scholars emphasize that absorptive capacity plays a critical role during the process of economic transition by facilitating international knowledge spillovers and technology transfers (Hobday, 1995; Amsden, 2001; Kim, 2001; Mathews, 2002). Mathews (2002), for instance, was among the first to integrate a resource-based view of the firm within the context of economic catch-up in order to explain how latecomer firms are able to break into knowledge-intensive industries. In support of this view, empirical studies help to confirm that absorptive capacity facilitates international knowledge spillovers and technology transfers (Buckley et al., 2002; Liu and Buck, 2007; Fu, 2008; Dai and Yu, 2013; Scherngell et al., 2014).

Many of the aforementioned studies on economic catch-up focus exclusively, however, on the ability of transitioning economy countries (or firms) to absorb knowledge from international sources of knowledge spillovers like foreign direct investment or international trade. It remains relatively less understood, by contrast, to what extent absorptive capacity potentially influences the ability of transitioning economy firms to successfully exploit agglomeration economies, such as knowledge spillovers. This is surprising since, as Porter (2000) emphasizes, the ability to attain and maintain competitive advantages in a global economy lies, somewhat paradoxically, in location-specific advantages.

In the agglomeration literature, several empirical studies find that the benefits of agglomeration are larger for firms with a higher bundle of internal resources, like absorptive capacity (McCann and Folta, 2008, 2011). While these agglomeration studies generally confirm the absorptive capacity thesis, they typically pertain to advanced economy contexts and the findings are not necessarily generalizable to developing and transitioning economies. A number of agglomeration studies that are applied to transitioning economy contexts typically find evidence in favor of agglomeration economies (Lu and Tao, 2009; Howell et al., 2016, 2018), yet, these studies typically ignore the role of absorptive capacity.

This paper examines the effects of agglomeration economies on firm productivity and the role of absorptive capacity in a transitioning economy context. To this end, I employ a fixed-effects panel estimator using a large panel of Chinese firms over a 10-year period. I rely on an inter-industry, co-proximity measure to capture relatedness economies, or the localized spillovers expected to arise in areas with a denser network of related activities. Exploiting the gradual and spatially uneven economic transitioning process, I further test empirically to what extent the relative importance of absorptive capacity for economic catch-up changes following the implementation of market-oriented economic reforms.

The main findings of the paper are as follows. First, firms exhibit higher productivity when located in an area with a denser network of related activities, particularly for firms with higher absorptive capacity. Second, the sources of the productivity gains are driven by externalities that arise from better access to inputs, similar workers, and technological-related knowledge spillovers. Third, the role of absorptive capacity becomes more important for boosting productivity following more versus less intensive market-oriented economic reforms.

These findings contribute to the agglomeration and transitioning economies literatures in the following ways. First, the findings show that relatedness economies are associated with productivity-enhancing spillovers. Second, in line with existing firm heterogeneity studies (McCann and Folta, 2008), the findings confirm that efforts to improve internal capabilities via more R&D spending enables firms to more easily seek out, transfer, and convert new ideas and knowledge into their own production function. The findings also contribute to existing transitioning economy studies (Hobday, 1995; Amsden, 2001), and in particular, show that the role of absorptive capacity in promoting economic catch-up depends critically on the stage of economic transition.

The outline of the paper is as follows. The subsequent section outlines the relevant literature review and key hypotheses. Section 3 describes the data and variable development, and econometric strategy. Section 4 presents the main results and Section 5 concludes.

2. Theory and hypotheses

2.1 Agglomeration and firm productivity

In the agglomeration literature, early empirical evidence lends support that co-located industries (and their participating firms) are more productive than their counterparts located in relative isolation (Glaeser et al., 1992; Henderson and Cockburn, 1996).⁵ Building on these early studies, it is increasingly recognized that localized spillovers are more likely to occur among co-located industries that also engage in similar activities and share overlapping knowledge bases (Boschma, 2005; Frenken et al., 2007). This is because higher local industry relatedness helps to facilitate knowledge spillovers between co-located firms due to their close cognitive or technological distance but only partly overlapping knowledge bases.⁶

According to Marshall (1920), co-located firms in the same or related industry benefit from several key cost-saving and productivity-enhancing advantages that arise due to their better access to customer–supplier relationships, labor pooling, and knowledge spillovers. As emphasized in Porter (1990), much knowledge sharing can result from the firm’s transaction-related communication with suppliers and customers. The diffusion of knowledge and knowledge spillovers may also occur as a result of employees moving from one firm to another (Zucker et al., 1998).

Besides Marshallian-based explanations, alternative theories exist that could also be used to explain the positive relationship between agglomeration and regional economic growth. Spatial sorting (Baldwin and Okubo, 2006) and spinoff behavior (Klepper, 2007), for instance, may lead to higher aggregate output among co-located firms even in the absence of any genuine Marshallian-based externalities. These alternative theories make it relatively difficult to know what are the underlying forces that could explain the higher observed productivity among co-located firms.

In an attempt to provide support for the Marshallian-based explanation, several empirical studies exist that attempt to quantify and discriminate between different Marshallian channels and to assess their relative contribution for agglomeration (relatedness) economies. Rigby and Essletzbichler (2002), for instance, measure spillovers exclusively based on geographical proximity, and find that all three Marshallian forces simultaneously have a positive effect on labor productivity in the USA. Ellison et al. (2010) offer the first attempt to study the different (Marshallian) sources of relatedness economies, and find that all three Marshallian forces explain inter-industrial co-agglomeration in the USA. In a related paper, Dauth (2011) find that each Marshallian source of relatedness is positively associated with regional employment growth in western Germany.

In China, existing empirical studies tend to find a positive relationship between clustering (and industry relatedness) and firm productivity (Lu and Tao, 2009; Long and Zhang, 2011; Howell et al., 2016). These positive agglomeration benefits are found to arise due to genuine positive externalities related to the underlying Marshallian explanations (Howell, 2017a), as opposed to being artificially driven due to policy-induced firm re-location behavior (Zheng et al., 2017). Overall, Hu et al. (2015) estimate that regional clustering contributed around 14% of China’s total productivity growth.

Based on the existing literature, it is therefore expected that a positive relationship exists between industry relatedness and firm productivity (H1), and that this relationship is driven at least in part by Marshallian-based explanations (H2).

H1: Firm productivity will be higher in areas with a denser network of related industries.

H2: Firm productivity will be higher in areas that have better access to the Marshallian sources of relatedness economies.

2.2 Absorptive capacity and firm heterogeneity

In the strategy and organization field, the dynamic nature of absorptive capacity is considered to play a central role in driving firm innovation and performance (Cohen and Levinthal, 1990; Kogut and Zander, 1992; Teece and Pisano, 1994; Zahra and George, 2002; Todorova and Durisin, 2007).⁷ In their meta-analysis of absorptive capacity, Zou et al. (2018) conclude that absorptive capacity is a strong predictor of innovation and knowledge transfer, and that its effects on economic performance is fully mediated by innovation. Van Den Bosch et al. (1999) further emphasize that the features of a firm’s absorptive capacity are related to the amount of knowledge available in the firm’s environment.

Variations in firms’ absorptive capacity, along with other firm-specific capabilities, help to explain how different firms make complex choices such as whether or not to seek out externalities, and how such choices consequently influence the firm’s performance (Kogut and Zander, 1992). A clear distinction exists, for instance, between firms that do and do not possess a sufficient pre-existing knowledge base to benefit from external sources of knowledge. Firms with higher absorptive capacity may be better positioned to seek out, acquire, and benefit from knowledge spillovers, whereas firms without sufficient pre-existing learning capabilities may not benefit at all.

In the agglomeration literature, several empirical studies show that the benefits of agglomeration are larger for firms with a higher bundle of internal resources, proxied by various characteristics, such as age, size, and ownership status, among others (Neffke et al., 2012; Rigby and Brown, 2015). McCann and Folta (2011) rely on a knowledge-based view of the firm to examine how different types of firms benefit from agglomeration. In confirmation of the absorptive capacity thesis, the authors find that biotech firms in the USA that are younger and possess relatively larger knowledge stocks benefit the most from agglomeration benefits.

Based on the above discussion, it is expected that firms with higher levels of learning ability or absorptive capacity, as reflected in higher R&D intensity of the firm, will have higher levels of prior related knowledge. In turn, these firms will be better positioned to absorb and benefit from relatedness economies (H3).

H3: The size of the effect on firm productivity of relatedness economies is mediated positively by firms’ R&D intensity.

2.3 Agglomeration and absorptive capacity during economic transition

A key shortcoming with Hypotheses 1–3 is that they ignore the transitioning economy context. This implies that the effects of industry relatedness and the role of absorptive capacity will be the same irrespective of their stage of economic transitioning. According to micro-foundations of agglomeration theory, however, the effectiveness of the underlying agglomeration mechanisms hinge critically on how well the market is integrated (Duranton and Puga, 2004). It is therefore expected that the agglomeration mechanisms are less effective in pre-reform versus post-reform China due to the greater presence of market imperfection and distortions.

Specifically, the presence of a few large state-owned enterprises (SOEs) that typically dominate local markets prior to market reforms may impede the free flow of the factors of production in the economy and potentially hinder the underlying Marshallian-based externalities. For instance, the size of the local market for independent specialized suppliers may be constrained if SOEs are more likely to source inputs from nonlocal suppliers either via internal supply (vertical integration) or national contracts. Second, labor pooling may be limited if the best workers with the most specialized skill-sets and experience prefer employment in SOEs for prestige and stability. Lastly, a few dominant SOEs in the local economy may hinder the flow of tacit information by reducing social interactions, thereby disrupting the creation of and access to knowledge spillovers in the region.

Following market-oriented reforms, by contrast, a high share of SOEs become dismantled and there is a proliferation of new privately owned enterprises (POE) that enter the market. Of the remaining SOEs left in-tact following reforms, one of their key roles is to serve as conduits of knowledge spillovers, often times achieved by establishing joint ventures with foreign companies (Buckley et al., 2007). The resulting intensification of inter-firm competition and cooperation in post-reform China is, in turn, expected to create the necessary conditions that give rise to positive externalities and generate a net positive effect on firm productivity (H4a).

It is also expected that the returns to R&D will be higher following more intensive market-oriented reforms (H4b). This expectation is based on theoretical and empirical work (Qian and Xu, 1998; Zhang et al., 2003) showing that R&D activities are more efficient in market-oriented economies compared with more centralized ones. By extension, the role of absorptive capacity is expected to be more important following more intensive market-oriented reforms due to the improved efficiencies in carrying out R&D activities combined with the increasing supply of externalities brought about by the reforms. This expectation is largely in line with existing studies in China (Buckley et al., 2002; Liu and Buck, 2007; Fu, 2008; Dai and Yu, 2013), and other transitioning economy contexts (Hobday, 1995; Amsden, 2001; Kim, 2001; Mathews, 2002), that highlight the important role of absorptive capacity in promoting economic catch-up.

H4: The effect of relatedness economies on firm productivity is: (a) larger in areas that experience more versus less intensive market-oriented reforms; and (b) especially for firms with higher absorptive capacity.

3. Empirical framework

3.1 Data

The main data source comes from the Annual Survey of Industrial Firms (ASIF) provided by the National Bureau of Statistics. The ASIF data include all state-owned industrial firms and non-state-owned firms with an annual turnover over five million Renminbi (approximately $600,000), accounting for 90%–95% of industrial output in China during the period 1998–2007. Industries in this dataset include mining, manufacturing, and electricity, gas and water production. The data contains an extensive set of firm characteristics, including information on firm production, sales revenues, employment, geographic location, industry affiliation, R&D expenditures, and so forth.

To create the panel, firms are linked over time and assigned a unique numerical ID by following closely the procedure outlined in Brandt et al. (2012). Specifically, the panel data is constructed by first matching firms using the legal person codes, and if there is no match, the firm’s name is used. The remaining firms are matched by using a combination of information on the firm’s legal person code, county code or city code, telephone number, and starting year. Overall, more than 90% of firms can be matched year-to-year using only the firm’s IDs, and the remaining firms are matched using the other information about the firm. An unbalanced panel is constructed by combining the matched and unmatched firms.

While the ASIF data provides the most accurate and comprehensive source of information on Chinese enterprises, the data also suffer from some serious problems, including missing data on key indicators, extreme outliers, vague definition of variables, and measurement errors. To mitigate these issues, the data cleaning procedure closely follows the one outlined in Brandt et al. (2012). First, observations that report missing or negative values for any of the following variables are dropped from the sample: total sales, total revenue, total employment, and fixed capital. Second, to reduce unduly influence from outliers a 1% trim is applied the main variables listed above and R&D expenditures.

The time period is restricted to 2001–2007 as this time period corresponds to China’s post-entry into the World Trade Organization and is the period that witnessed rapid increases in both R&D spending and productivity. To remove potential selection bias of firms likely to be operating at or near the minimum sales threshold, firms that report information for only one time period or that have gaps in their reporting are removed from the sample. The sample is further restricted to domestic Chinese enterprises in the manufacturing sector with at least eight employees.

The resulting sample of firms includes 189,095 Chinese firms located in 255 Chinese prefecture-level cities. Note that due to China’s vast urban expansion and economic growth, some administrative boundaries change over the time period. To avoid any artificial bias due to boundary changes, city IDs that changed during the time period are matched to the initial city code at the beginning of the time period.

3.2 Dependent variable

The dependent variable refers to estimates for firm’s total factor productivity (TFP), derived following the approach in Ackerberg et al. (2006). Ackerberg et al.’s (2006) approach relies on intermediate inputs to proxy unobserved productivity and the system estimator that is used in the estimation allows fixed effects to take into account firms’ (unmeasured) productivity advantages that persist over time. Table 1 reports the annual TFP estimates during the time period of analysis obtained from the Ackerberg et al. (2006) approach. As a benchmark, I also report the results obtained after estimating firm TFP based on two other popular approaches to derive estimates for TFP based on Olley and Pakes (1996) and Levinsohn and Petrin (2003) (See Table A1 in Appendix A for variable description and summary statistics).

Note that in order to obtain the TFP estimates, the firm’s value added, capital stock, and investment had to be developed first developed. Following Brandt et al. (2012), the real value added is constructed by separately deflating output, net of goods purchased for resale and indirect taxes, and material inputs, where the input deflators are calculated using the output deflators and information from China’s 2002 National Input–Output (IO) table.

Next, the real capital stock for 1998 is developed using the perpetual inventory method. A depreciation rate of 9% is assumed and annual investment is deflated using the Brandt–Rawski deflator. Following 1998, the observed change in the firm’s nominal capital stock at original purchase prices is used as the estimate for the nominal fixed investment using the same rate of depreciation and deflator to roll the real capital stock estimates forward.

3.3 Main independent variables

3.3.1 Proxying for industry relatedness

In order to test Hypothesis 1, it is necessary to develop proxies for inter-industry relatedness. Hidalgo et al. (2007) offer a superior proxy to measure relatedness between co-located firms based on the similarity among products of industries (See Appendix B for more details about the construction of the Density measure). The idea here is that firms and industries that use more similar combinations of inputs in their production processes are more likely to rely on the same set of suppliers and clients, and consequently are more likely to be interconnected through skills, technologies, and other common inputs. It follows that firms and industries that produce products with a higher proximity are more likely to interact with one another in various ways, e.g., reliance on similar inputs, similar technologies and research and development, and even the same supply or marketing facilities.

A key advantage of the density measure is its overall comprehensiveness, although it is not capable of discerning among the different underlying mechanisms that drive relatedness. In order to test Hypothesis 2, I rely on three measures that proxy for each main Marshallian channel. As in Li and Zhu (2014); Howell (2017a), China’s 2002 Input–Output tables are used to construct a measure of customer–buyer linkages. The 2004 Chinese Economic Census (CES) data is used to compute the industry’s similarity in skill intensity between industries in terms of their education levels to measure for labor pooling of similar workers. Finally, firm’s co-exporting behavior is used to proxy for technological-related knowledge spillovers (See Appendix B for the variable development procedures).

Reverse causation bias makes it difficult to interpret the effects of the relatedness measures. To help address this concern, the relatedness variables enter into the model as initial values. Using initial values may help to reduce the problem of unobserved local productivity shocks but only if the shocks that are relevant to current productivity are relatively recent.

3.3.2 Proxying for absorptive capacity

In order to test Hypothesis 3, it is necessary to develop a proxy for firms’ absorptive capacity. In line with the literature, the firm’s absorptive capacity is measured by the firm’s innovation effort, expressed as one plus the logarithm of the ratio of R&D expenditures to total sales. It is acknowledged that in many transitioning economies investments in research projects may be of low quality, and therefore may not be a suitable proxy for absorptive capacity. However, this is not likely to be the case for China. Since 2001, R&D investments made by both private and public enterprises not only rapidly increased in the amount but also improved substantially in terms of quality, leading to significant productivity gains at the firm level (Howell, 2015, 2017b, 2019; Boeing et al., 2016).

A key concern that does arise by using R&D spending as a proxy for absorptive capacity is that firms that engage in R&D are likely to systematically differ from those that do not, potentially leading to selection bias. In order to mitigate this concern, a two-step approach is taken to derive estimates for firm’s R&D intensity based on a Heckman-style sample selection model. After correcting for selection effects, the predicted values obtained from the parameters recovered in the second stage of the selection model are used to proxy for the firm’s absorptive capacity, denoted as R&D* (See Appendix C for more details).

3.3.3 Proxying for local market-oriented reforms

In order to test Hypothesis 4, it is necessary to develop a proxy for the intensity of market-oriented reforms. A suitable candidate for the proxy is the share of state enterprise employees in the local economy. As in Howell et al. (2016), I rely on the ASIF data to calculate the proportion of state enterprise employees across industries for each city in each year to proxy for China’s market reforms. Interaction terms are then created between the market reform proxy and the relatedness measures to see how the effects of (overall) relatedness, and each Marshallian source, impact firm productivity in areas that undergo more intense market-oriented reforms.

The designation of which areas and sectors were chosen to undergo more intensive market reforms was not at random, however, which makes it difficult to claim that the reforms represent an exogenous shock. To address this issue, propensity score matching combined with difference-in-differences (PSM DID) estimation is used as an identification strategy. The PSM DID approach consists of two stages. In the first stage, a nonlinear matching technique is employed to ensure the compatibility between the treatment group and the control group (See Appendix D for more details on the matching procedure).

PSM is sensitive to unobserved heterogeneity, which is why in the second stage, the DID estimator is subsequently employed to remove the time-invariant unobserved heterogeneity across firms, such as such as managerial skill. The dependent variable is first differenced by calculating the difference between the post-treatment and pre-treatment means. That is, only a balanced sample of firms that are observed both before and after the treatment are kept in the sample.

3.4 Estimation strategy

X_it includes a vector of firm and industry controls, along with year-industry and year-region fixed effects, and α_i represents a firm-fixed effect. In line with the literature, firm controls include firm size (ln), and a non-linear term, age (ln), and ownership structure where information on both the firm’s legal status and the structure of paid-in capital is used to determine whether the firm is a SOE or a non-SOE. In addition, industry entry rate (ln), industry size (ln), and industry sales growth (ln) are also included to control for industry competition and lifecycle effects (See Table A1 in Appendix A for variable description and summary statistics).

4. Results

Table 2 reports the model results after estimating different forms of equation (1). In each specification, the dependent variable is estimates of firm TFP derived using the Ackerberg et al. (2006) approach. The standard errors reported in parentheses are robust and corrections have been made for potential correlation of errors between firms operating within the same city. Column (1) includes only the density measure, followed by the addition of firm and industry level controls in Columns (2) and (3), respectively. In column (4), the density measure is replaced with proxies for each Marshallian source of relatedness. Each proxy is standardized for comparison purposes. Due to the high correlation (>0.8) between input linkages and output linkages, the latter is removed from the model.

In confirmation of Hypothesis 1, the density measure is positively related to firm productivity across Columns (1)–(3). The size of the coefficients range between 0.093 and 0.178 and are highly statistically significant. Holding other factors constant, an increase of half a standard deviation over the mean in the average density measure (leading to an increase of 39% [(0.11/0.14)/2] in density) raises firm TFP by 3.5–6% depending on specification.

In confirmation of Hypothesis 2, Column (4) reveals that the proxies for each Marshallian source of relatedness economies are positively associated with firm productivity, and are highlight statistically significant. In terms of magnitudes, the coefficient on input linkage is between 2 and 7 times larger than the coefficients for labor pooling and knowledge spillovers. Aside from true knowledge spillovers, one reason for the larger coefficient on input linkages is because as an industry grows, it increases demand for its inputs, which in turn, spurs growth of its suppliers.

The results on the remaining firm and industry control variables generally fall in line with expectations. Across each specification, the results show that firms’ R&D intensity has a positive effect on firm productivity. SOEs are less productive, on average, compared with non-SOEs. Smaller and younger firms tend to be less productive than their respective counterparts, although the effect of firm size is nonlinear. Firms have higher productivity if they are in industries that are larger, have higher sales growth and have higher entry rates.

As a sensitivity check, Table 3 reports the results with respect to the relatedness variables using alternative estimation procedures to calculate TFP based on the OP method in Columns (1)–(2) and the LP method in Columns (3)–(4). The density measure is reported in Columns (1) and (3), while the proxies for Marshallian sources of relatedness economies are reported in Columns (2) and (4). Each specification includes the same set of firm and industry controls as the fully specified model in Table 2. Irrespective of estimation strategy to derive TFP estimates, the results remain qualitatively the same as the initial findings.

4.1 Absorptive capacity and relatedness

In Table 4, interaction terms are added between R&D* intensity and density, and each Marshallian proxy, to explore the role of absorptive capacity on firm productivity. In each specification, the dependent variable is firm TFP estimates derived using the Ackerberg et al. (2006) approach. The standard errors reported in parentheses are robust and corrections have been made for potential correlation of errors between firms operating within the same city. Each specification includes the same set of firm and industry controls as the fully specified model in Table 2.

In Column (1), the interaction term between R&D* intensity and Density is positive and highly statistically significant. In confirmation of Hypothesis 3, this result indicates that firms with higher levels of R&D intensity benefit more from relatedness than their counterparts with lower levels of R&D intensity. In Columns (2)–(4), interaction terms between R&D* intensity and each Marshallian proxy enter separately into the model as well as enter in jointly in Column (5). The positive and statistically significant coefficients on the interaction terms further confirm Hypothesis 3. Looking at the coefficients on the interaction terms in Column (5), however, the size of the coefficient on the interaction term between R&D intensity and input linkages is significantly smaller compared with the coefficients on the other interaction terms.

One potential explanation for the relatively small size of the coefficient on the interaction term with input linkages is that absorptive capacity plays a less important role for firms to benefit from customer–buyer linkages compared with labor pooling or pure technological-related knowledge spillovers. Combined with the results found above on the primary effects, one possible explanation for this finding is that better access to customer–buyer linkages generates additional externalities besides true knowledge spillovers, which benefit all firms more equally. By contrast, externalities on labor and ideas may be more limited to embodying pure knowledge spillovers, as emphasized, respectively, in Zucker et al. (1998) and Glaeser and Gottlieb (2009), in which case the firm’s pre-existing capabilities play a more important mediating role.

4.2 Absorptive capacity and relatedness during economic transitioning

Based on theories related to agglomeration (Duranton and Puga, 2004) and economic transitioning (Qian and Xu, 1998), a key concern that arises with the previous empirical framework is that it ignores the fact that China’s gradual implementation of market-oriented reforms may influence the effects of relatedness and the role of absorptive capacity. To address this issue, an empirical test is offered to study the relative importance of absorptive capacity in areas that have underwent more versus less intense market-oriented reforms, proxied by changes in the local share of SOE employment. To take into account the non-random implementation of SOE reforms, the PSM DID approach is used as an identification strategy (See Appendix D for a discussion).

The results are presented in Table 5. Panel A includes the density measure, while Panel B substitutes the density measure with the proxies for each (Marshallian) source of relatedness economies. In each panel, Column (1) carries out the estimation procedure for all firms in the matched sample. The remaining columns check for parameter heterogeneity in firms’ absorptive capacity. That is, Columns (2) and (3) split the sample into firms with lower and higher absorptive capacity base on the median value of firms’ R&D* intensity.

In Column (1) of Panels A and B, the coefficients on the primary effects for market reforms and the relatedness measures are, respectively, positive and statistically significant. The results in Columns (2) and (3) show that the size of the primary effects are larger for firms’ with comparatively high RD intensity. In general, these findings on the primary effects of the relatedness measures confirm earlier findings that support Hypotheses 1–3.

Looking at the positive and statistically significant coefficient on the interaction terms, reveals that firms benefit more from industry relatedness, and each Marshallian source, following the implementation of more versus less intense market-oriented reforms. Checking for parameter heterogeneity in Columns (2) and (3), the results show that firms with comparatively higher amounts of R&D intensity benefit from technological-related spillovers in areas that experience more versus less intensive market reforms. As indicated in Column (4), the size of the coefficient on the interaction term between market reforms and density is significant larger, in both statistical and economic sense, for high R&D intensive firms versus less R&D intensive firms.

These results confirm both parts of Hypothesis 4, suggesting that the effects of relatedness and the role of absorptive capacity become more pronounced in post-reform China. From the perspective of agglomeration theory (Duranton and Puga, 2004), market mechanisms supplant the role of the state in post-reform periods, leading to a better integrated market with fewer distortions and imperfections, which in this case, appear to facilitate externalities arising on customer–buyer linkages, labor pooling, and knowledge spillovers. According to theories about transitioning economies (Qian and Xu, 1998), the role of absorptive capacity is more pronounced following market reforms likely due to efficiency improvements related to performing R&D (Zhang et al., 2003), combined with the increasing competition and higher incentives to seek out external sources of knowledge to improve overall efficiency.

5. Conclusion

In order to promote sustainable growth, China has rapidly vamped up its investments in R&D and dramatically transformed its spatial economy by actively encouraging firm and industry agglomeration. This paper examines the effects on firm productivity of spillovers expected to arise due to the dense concentration of related industries, and the role of absorptive capacity in explaining those effects. Taking into account the transitioning Chinese economy context, further emphasis is placed on investigating to what extent the relative importance of absorptive capacity for economic catch-up changes following an intense period of market-oriented economic reforms.

The main results show that the effect of relatedness economies enhances firm productivity. The source of these productivity gains arise at least partially due to Marshallian channels such as better access to inputs, people, and ideas. The size of the productivity gains are larger for firms with relatively high levels of absorptive capacity. The results also show that the role of absorptive capacity becomes even more important for promoting economic catch-up following the implementation of market-oriented economic reforms.

These findings contribute to the agglomeration and transitioning economy literatures in the following ways. In line with previous agglomeration studies applied to advanced market economies, this paper provides evidence in favor of not only agglomeration economies but also helps to confirm the absorptive capacity thesis in a transitioning economy context. This paper also contributes directly to the transitioning economies literature by showing empirically that the role of absorptive capacity for promoting economic catch-up depends critically on the stage of economic transition.

Funding

This project received funding support from the Natural Science Foundation of China No. 71603009.

Footnotes

References

Appendix A Variable summary information

Appendix B Relatedness measures

Density measure

Input–output linkages

Firms that have better access to customers and suppliers reduce transportation costs. Firms are therefore expected to enjoy higher productivity gains when they share closer input–output relationships with local leading industries. China’s 2002 I–O Tables published by the National Bureau of Statistics (NBS) are used to measure supplier-buyer linkages. The I–O Tables provide input–output relationships between 122 sectors. A concordance between the I–O sectors and the CIC industries published by the NBS is used to map the I–O sectors to the 4-digit CIC industry from the ASIF. Input_hj is defined as the share of industry $h′s$ inputs that come from industry j and Output_hj as the share of industry $h′s$ outputs that are sold to industry j.

Labor pooling

Firms that enjoy better access to labor pooling of suitable workers are expected to minimize costs including risk sharing and better matching. Firms are expected to enjoy higher productivity gains when they use similar types of labor as the local predominant industries. As a proxy for labour pooling of similar workers, the 2004 CES data is used to compute the similarity in skill intensity between industries in terms of their education levels.⁸ In the 2004 CES, there are five education categories: junior high school or below, senior high school, diploma, undergraduate/college, and postgraduate. An important weakness of this proxy is that there are few worker classes to base the metric and also consideration is only given to similarity in education levels but not in fields of education.

The industry skill intensity is calculated at the 4-digit CIC industry level. Share_ho is defined as the fraction of industry $h′s$ employment with education level o. Similarity of workers in industries h and j is measured through the correlation of Share_ho and Share_jo across education levels. as it only considers similarity in education levels but not in fields of education.

Knowledge spillovers

Firms may also co-locate to benefit from knowledge spillovers—the direct or indirect transfer of information or ideas. Firms that share closer technological proximity to the leading industries are expected to enjoy larger productivity gains. It is acknowledged that it is extremely difficult to find a suitable measure that captures the exchange of knowledge between workers and firms. While patent citation information is a more commonly used proxy in the literature, such information is not yet available for China.

As an alternative measure, information on firms’ co-exporting patterns to infer the knowledge spillovers between industries. The rationale behind this measure is based on the idea that the firms exporting in different industries are more likely to integrate with each other if there are more information exchange between these industries. Since the market for information is quite incomplete, vertical or horizontal integration between firms can be used as a means to internalize these externalities.

China’s General Administration of Customs (GAC) data is used to obtain information on the firm’s co-exporting patterns, which includes the universe of all Chinese trade transactions by importing and by exporting firms at the HS 8-digit level. Because the trade customs data do not contain information about domestic production or characteristics of the firms, it is not possible to identify if the firm is a manufacturer or a wholesaler, distributer and/or intermediary. Intermediary firms are therefore indirectly identified, and subsequently removed, by searching for the presence of specific phrases in the company names. The specific phrases include importer, exporter, and trading, with the corresponding Chinese (pinyin) variants being “mao yi,” “wai mao,” “wai jing,” “jin chu kou,” “jing mao,” “gong mao,” and “ke mao.”

It is acknowledged that the co-exporting patterns capture anything that would give rise to economies of scope at the firm level, including, but not limited to, knowledge spillovers. Once the other two Marshallian channels for labor pooling and input–output linkages are included, firms’ co-exporting behavior is likely to be a better proxy for knowledge spillovers. Strictly speaking, however, this proxy more accurately captures externalities that are unaccounted for by the other measures.

Appendix C Endogenous absorptive capacity

Equation (C.1) is estimated using a random effects probit model given the panel structure and the binary character of the dependent variable. The random effects structure assume that the errors are not correlated with the regressors, an unrealistic assumption in this case. To address this issue, the Mundlak specification is used by including a vector of means of the time-varying regressors as control variables to allow for some correlation between the random effects and the regressors.

Because the inverse mills ratio is used, it is required to find an exogenous variable to satisfy the exclusion restriction. In this case, it is decided to use firm size to satisfy the identification assumption. The size of the firm is expected to influence the decision to invest in innovation, but it does not affect the intensity of that investment once the decision to invest has been taken. This claim is based on the existing empirical literature that tends to find that in transitioning economy contexts large firms invest more in R&D in level but not proportionally more once the decision to invest has been taken (Crespi and Zuniga, 2012).

The results show that firms that are more capital intensive, older, have higher sales growth, and are in industries that have faster growth, higher entry rates, are more concentrated and more export intensive tend to be more likely to invest in R&D, and invest more intensively. POEs are more likely to invest and invest more intensively compared with SOEs, but not priv. SOEs.

Appendix D PSM-DID identification strategy

To study the impact of technological relatedness (and its sources) on firm productivity in regions where more intensive market reforms have been carried out, a key assumption is that market reforms are an exogenous shock. Such an assumption is unlikely in this case due to the non-random assignment of market reforms that initially favored coastal provinces. To deal with the identification problem, a PSM DID strategy is employed to help reduce concerns about selection bias. A two-stage approach is developed.

In the first stage, a nonlinear propensity score matching technique is used to construct a control group of firms that match most closely to treated firms based on observable characteristics. A list of covariates is linked to SOE reforms, a main component of China’s overall market reforms, in order to identify the most appropriate control group. The covariates include both firm- and industry- level variables to take into account various indicators of firm’s internal capabilities and productivity while also controlling for the conditions of the local operating environment. Firms in the control group are matched to the treatment group on the basis of the pre-treatment (less intense SOE reforms) mean of these variables.

Table A2 shows the results from the first step’s estimation of the probit model. The dependent variable is a dummy variable that takes a value of 1 if the firm is in the treatment group, and 0 otherwise. The objective here is to first check whether the selected covariates are important determinants of the policy treatment. All covariates are measured by the mean before the policy treatment. The results show that the covariates are indeed statistically significant. Firms that receive policy treatment tend to be less capital intensive, larger, younger, have lower growth rates, have a smaller share of state-backed capital, and are in industries that have lower growth rates, are less concentrated, have higher new entry rates and that are more export intensive.

Based on the above determinants of policy treatment, a matched control group is constructed to compare with treated firms. To assess the credibility of the matching procedure, a formal pairwise t-test comparison between the treated and matched control firms is carried out to see whether there are any significant differences. Table A3 presents the balancing tests for the propensity score matching, comparing the pre-treatment mean of the SOE reform determinants between the treated and matched groups. The large P-values indicate that there is no significant difference in the selected covariates between the treated and matched samples. The results thus show that the matching procedure provides an appropriate foundation for the DID estimation. Following the common support condition, focus is placed on the matched firms that fall within the support of the propensity score distribution of the treated group.