Industrial clusters and carbon emission reduction: evidence from China

Abstract

In the context of carbon peaking and neutrality goals, this paper examines the impact of industrial clusters on firms’ carbon emissions based on listed manufacturing firms in China from 2008 to 2020. To avoid the endogeneity issue, we constructed an industrial cluster database ten years earlier than the sample period. The results of the empirical analysis suggest that industrial clusters are conducive to decreasing carbon emissions. The stronger the industrial cluster, the lower the carbon intensity. Examining the channel reveals that industrial clusters have an emission reduction effect through digital transformation. The heterogeneity analysis demonstrates that the variations in industrial clusters, including pollutant emissions, energy intensity, and energy consumption structure, significantly affect a firm's carbon emission reduction performance. In addition, environmental policies and the internal and external green development impetus of firms also affect emission reduction behavior. Our findings provide the rationale for governments to promote green transformation and enhance digital competitiveness through industrial cluster development.

Appendix A: The procedure for identifying industrial clusters

1.1 (1) Identify the continuous geographic boundaries of a given industry

For the whole manufacturing industry from 1998 to 2010, we first calculated the kernel density based on the location of each firm in each four-digit industry for each year. Given a four-digit industry A comprising nA firms, we constructed the industry-specific kernel density function of bilateral distances weighted by employment as:

$$\widehat{{K}_{A}}\left(d\right)=\frac{1}{{h}_{A}{\sum }_{i=1}^{{n}_{A}-1}{\sum }_{j=i+1}^{{n}_{A}}{e}_{i}{e}_{j}}\sum_{i=1}^{{n}_{A}-1}\sum_{j=i+1}^{{n}_{A}}{e}_{i}{\cdot e}_{j}\cdot f(\frac{d-{d}_{ij}}{{h}_{A}})$$

where ${d}_{ij}$ is the Euclidean distance between firm i and firm j, $d$ is the median bilateral distance between every pair of all manufacturing firms, ${e}_{i}$ and ${e}_{j}$ are the employment of firm i and firm j, respectively. ${h}_{A}$ is the industry-specific band width, while $f(\cdot )$ is Gaussian distribution function.

We calculated the kernel density at each kilometer for each industry and plotted the actual kernel density distribution. Suppose firms are random distribution. We randomly selected all manufacturing firms’ locations in the sample and simulated the density distribution of the industry with randomly matched firm locations for a specific industry. Finally, compare the actual distribution with the simulated one; if the kernel density is larger than the simulated density, we ascertain the boundaries of the given industries. Figure 

Fig. 2

The kernel density distribution of textile and garment manufacturing industry

2 displays the density distribution of the textile and garment manufacturing industry. The solid line indicates the actual kernel density distribution of the industry, while the dashed lines are the simulated distributions at different levels.

1.2 (2) Identify the firms within industrial clusters

Since we have ascertained localized boundaries at the industry level across the country, we need to identify the actual location of industrial clusters. In particular, each industry firm is considered the center firm of a candidate cluster, and the employment density within a given geographic boundary is calculated based on the center firm. Then, the maximum density is selected as the critical value of the cluster, and firms in the candidate cluster are identified as the first cluster firms. After eliminating firms in the first cluster, we used the same method to calculate the employment density higher than the critical value by narrowing the geographical boundary and obtaining the rest of the clusters from candidate clusters.

See Tables

Table 10 IV regression results as supplementary

10,

Table 11 Robustness regression results as supplementary

11,

Table 12 IV regression results: the effects of digitalization on carbon emission

12,

Table 13 Heterogeneity analysis: industrial cluster heterogeneity as supplementary (carbon density is in level form)

13