The Impact of Urban E-Commerce Transformation on Carbon Emissions in Chinese Cities: An Empirical Analysis Based on the PSM-DID Method

Abstract

To accelerate global green and low-carbon development, China has proposed a “double carbon” target. It is particularly important to explore the carbon reduction effects of e-commerce transformation in cities to achieve sustainable development. Based on the quasi-natural experiment of the National E-Commerce Demonstration City (NEDC) pilot, 263 cities from 2008 to 2017 were selected as samples, and the propensity score matching difference-in-differences (PSM-DID) method was used to investigate the influence of NEDCs on urban carbon emissions in China and its underlying mechanism. The results show that NEDCs can significantly reduce urban carbon emissions; the carbon emission level of pilot cities was reduced by 9.45%. After passing a series of robustness tests, this conclusion remains valid. The policy effects of NEDCs on carbon emissions are heterogeneous across different regions and types of cities, with the policy effect being more significant in central and western cities and in resource-based cities. Further mechanism analysis shows that the NEDC policy reduces urban carbon emissions mainly through two channels, namely, green technology innovation and industrial structure upgrading. This study provides important policy implications for the implementation of e-commerce demonstration city construction according to local conditions and the realization of urban sustainable development under the double carbon goal.

1. Introduction

With the development of the economy, the increasingly serious air pollution problem has gradually aroused great concern and discussion in the international community and become a global problem that needs to be solved. Massive emissions of greenhouse gases, mainly carbon dioxide, have been the main cause of global warming since the mid-20th century. The continued increase in greenhouse gas emissions has had a negative impact on agricultural production, socioeconomic activities, and human life and has ultimately hindered the process of achieving global sustainable development. Sustainable development is a model that focuses on long-term development, and its concept is considered a “historical category” [1], which is achieved by balancing environmental sustainability, social sustainability, and economic sustainability [2]. Carbon emissions and the climate problems that result from them have become a global concern [3,4]. As the world’s largest developing country and the largest emitter of carbon dioxide [5], China attaches great importance to the issue of climate change and has actively taken various measures to promote energy conservation and emission reduction and to facilitate the transition into a low-carbon economy. To control carbon emissions, the Chinese government has proposed a “double-binding” national voluntary contribution target under the Paris Agreement. In terms of total emissions, carbon emissions are expected to peak in 2030, and efforts are being made to reach that peak as soon as possible; in terms of intensity, carbon emissions per unit of GDP are expected to decrease by 60% to 65% in 2030 compared to the levels in 2005. As areas containing concentrations of population, industry, and factor resources, cities constitute a major source of global carbon emissions. According to estimates, urban areas consume approximately 67% of the world’s energy and produce 71% of its carbon dioxide [6]. Therefore, how to reduce urban carbon emissions and achieve sustainable economic, societal, and environmental development while ensuring steady economic growth has become an urgent problem to be solved.

With the wide application of digital technology, the digital economy provides a new way to ease the contradiction between economic growth and a green low-carbon transition. The role of digital technology in achieving carbon neutrality has received increasing attention [7]. As the most active and concentrated manifestation of the digital economy, e-commerce has been a key focus in academia and Chinese government departments [8,9]. E-commerce is a sector in which all business activities are carried out in the trading environment of an information network. The application of e-commerce technology has promoted the deep integration of big data, artificial intelligence, and the real economy, providing new ideas for carbon emission reduction. In addition, e-commerce brings changes in new products and services, which can significantly improve energy output efficiency [10,11] and provide favorable opportunities for the realization of the double carbon goal. In 2009, the Chinese government introduced the “National E-commerce Demonstration City” policy, selecting Shenzhen as the first national e-commerce demonstration pilot city. Since 2011, the status of pilot e-commerce demonstration cities has been gradually extended to other cities, and the scope of the work involved in creating demonstration cities has been expanding, with 70 cities having been approved as national e-commerce demonstration cities by 2017. In particular, the Guidance on the Creation of National E-Commerce Demonstration Cities mentions the importance of creating e-commerce model cities for “reducing the consumption of material resources and energy, reducing environmental pollution and developing a green economy; and enhancing industrial structure and optimizing resource allocation.” The significance and requirements of the construction of NEDCs are in line with China’s goal of developing a low-carbon economy. What must be explored next is whether the NEDC policy will have an impact on China’s urban carbon emissions in practice. What are the characteristics of the impact’s mechanism and effect? Is there any heterogeneity in the carbon emission reduction effect of the construction of NEDCs among different groups? Answering these questions is of great theoretical and practical significance in the search for a new model of urbanization in China and for the achievement of the “30.60” dual carbon goal. Therefore, based on the panel data of Chinese cities from 2008 to 2017, in this study, the national e-commerce demonstration city pilot was taken as a quasi-natural experiment, and the PSM-DID method was used to evaluate the impact of national e-commerce demonstration city construction on urban carbon emissions and its mechanism. Based on the conclusions, the paper provides policy suggestions for the government to promote the reduction in urban carbon emissions while developing e-commerce.

The rest of the paper is organized as follows. The theoretical background is introduced in Section 2. Section 3 provides the policy background and research hypotheses. The data and methodology are presented in Section 4. The results are shown in Section 5. Section 6 presents the discussion and implications. Section 7 provides the limitations and future research.

2. Theoretical Background

In existing studies, many scholars have explored the influencing factors of carbon emissions. The extant research has found that, as a developing country with a higher carbon emission scale and greater pressure to reduce emissions, China’s energy-saving and emission-reduction measures receive more attention. Hou et al. [12] proposed that promoting the development of energy-saving technologies, producing energy-saving products, and promoting the low-carbon economic development mode is an effective long-term mechanism for China. Guo et al. [13] proposed strategies such as “transforming the economic development model, optimizing the energy structure, improving the technological process and developing low-carbon economy.” Li and Qi [14] adopted static and dynamic panel models to prove that urbanization can promote carbon emission intensity. Yao et al. [15] found that industrial structure, technological progress, and foreign direct investment play a moderating role in the process by which urbanization inhibits carbon intensity. Existing studies mainly explore the influencing factors of urban carbon emissions from the aspects of industrial structure, urbanization, energy structure, etc.

At present, no studies have directly focused on the impact of e-commerce on carbon emissions. Studies similar to the current work mainly include those on the relationship between the digital economy and urban low-carbon development, as well as the relationship between e-commerce and the urban ecological environment. Some scholars have proposed that the promotion of the digital economy helps to promote green and low-carbon urban development. Li et al. [16] conducted fixed-effect regression based on data from 190 countries and empirically tested the influence of the digital economy on air pollutants, and the obtained findings support the EKC assumption. However, whether the Chinese data also support the EKC hypothesis remains questionable. As a typical mode of urban development in the era of the digital economy, Guo et al. [17] found that China’s smart city construction could save energy and reduce per capita carbon dioxide emissions by improving energy efficiency, with an emission reduction effect of approximately 18.42 logarithmic percentage points. However, the intrinsic promoting effect of industrial structure upgrading was ignored in this study. In addition, some scholars have studied the relationship between e-commerce and the environment. Cao et al. [18] used the DID approach based on Chinese urban data to analyze that NEDC policy significantly improved the green total factor productivity of enterprises, which enriched the research on the relationship between e-commerce and green development. Sui and Rejeski [19] used qualitative research methods to conclude that e-commerce can promote ecological and green development. Liu et al. [8] believe that the development of e-commerce promotes the green and high-quality development of Chinese cities. Zhang et al. [20] examined the impact of urban e-commerce on environmental pollution using urban panel data.

In summary, the existing studies have provided important insights for the authors of this paper to explore the impact of urban e-commerce transformation on carbon emissions. However, there are still shortcomings in the existing research. First, academic circles have discussed the influencing factors of carbon emissions and the environmental improvement effect of e-commerce. However, the existing research on the environmental impact of the construction of national e-commerce demonstration cities mainly focuses on water pollution and air pollution, with little reference to the impact of carbon emissions. Second, in the path of urban sustainable development, there are few studies on the impact of policies on carbon emissions from the perspective of industrial structure upgrading and green technology innovation. Therefore, based on data from 263 cities in China, this study adopted the PSM-DID approach to deeply explore the policy effects of NEDCs on urban carbon emissions. This study has the following main marginal contributions. First, in terms of research content, the national e-commerce demonstration city pilot was taken as a quasi-natural experiment to analyze the carbon emission reduction effect of e-commerce. It was found that urban e-commerce transformation can significantly reduce urban carbon emissions, with a more significant effect on central and western cities and resource-based cities. This conclusion provides new evidence of the impact of national e-commerce model city construction on urban carbon emissions and can help the Chinese government formulate more targeted e-commerce policies based on city policies. Second, in terms of theoretical mechanisms, taking green technology innovation and industrial structure as the mechanism variables, the paper discusses the role path of the NEDC policy on urban carbon emissions and provides new ideas for the carbon emission reduction path of e-commerce transformation cities. Third, in terms of research methods, the PSM-DID model did not only effectively avoid the endogeneity problems caused by missing variables but also eliminated the sample selection bias caused by urban heterogeneity. It is a powerful tool to test the effect of the NEDC policy on urban carbon emissions.

3. Policy Background and Research Hypotheses

3.1. Policy Background of the National E-Commerce Demonstration City

To give full play to the role of e-commerce in optimizing resource allocation, reducing energy consumption, and promoting the green development of the urban economy, several departments of the Chinese government have jointly launched the “National E-commerce Demonstration City” program. In September 2009, the Chinese government officially approved the first demonstration city. In March 2011, several ministries and commissions of the Chinese government jointly issued the Guidance on the Establishment of National E-commerce Demonstration Cities, which points out the importance of building an e-commerce demonstration city and defines its overall objectives and main tasks. In the same year, 22 cities in China were selected as e-commerce demonstration cities. In 2014, the Chinese government selected 30 cities, including Dongguan City, as e-commerce demonstration cities in accordance with the principle of “pilot first, gradually promote”. In 2017, 17 cities, including Dalian, were approved to become national e-commerce demonstration cities. To date, there are 70 e-commerce demonstration cities in China.

3.2. Theoretical Analysis and Research Hypotheses

With the development of digital technology, information and communication technology has become an important driving force in the promotion of environmental pollution control [21]. For example, search volume data regarding pollution on network platforms and informal environmental regulations generated by the network haze of public opinion have an effect on improving urban air quality. Moreover, the information spread by digital media can guide the public to embrace the concept of green environmental protection [22], thus reducing carbon emissions. The construction of NEDCs promotes the application and development of new technology by promoting the application and development of e-commerce. Technological progress improves the efficiency of social resource allocation, thus reducing energy consumption and improving the environment [8]. In addition, many scholars have analyzed the impact of e-commerce on air pollution as a reduction in energy consumption. For example, Sullivan and Kim [23] argue that e-commerce can strengthen the relationship between online buyers by enabling consideration of the perception of product quality and product value to increase the sense of trust among online buyers, and thus, reduce overproduction and pollutant emissions. Mangiaracina et al. [24] believe that e-commerce delivery services produce less carbon dioxide. Finally, the significance of the NEDC policy lies in “reducing the consumption of material resources and energy, reducing environmental pollution, and developing a green economy; improve the industrial structure, optimize the allocation of resources”. As an environmentally friendly industry, e-commerce can squeeze the development space of industries with high energy consumption and high emissions through the extrusion effect, optimize urban industrial structure, and promote urban green and high-quality development [8,25,26]. Therefore, research hypothesis H1 is proposed.

H1. 

The NEDC policy can significantly reduce urban carbon emissions.

As a form of the digital economy, e-commerce is the concrete product of the deep integration of digital technology and the real economy. Compared to the traditional business model, e-commerce conducts commercial transactions through the internet, and the technological innovation it brings may have an important impact on environmental pollution and carbon emissions. Grossman et al. [27] first proposed the “technology effect” when explaining the formation mechanism of the EKC, believing that with further economic growth, technological innovation will reduce the input of production factors in the production process, thus reducing the environmental pollution caused by the consumption of production factors. Subsequently, on the basis of the EKC, it has been demonstrated that technological innovation is the necessary condition for the inflection point of the EKC, and technological innovation can promote the upgrading of production technology by improving resource utilization and production efficiency to alleviate the previous extensive production mode. Technological innovation reduces the negative impact of economic growth and promotes the development of a low-carbon economy. Therefore, the green technology innovation brought by the construction of national e-commerce demonstration cities is key to promoting urban low-carbon development. Therefore, research hypothesis H2 is proposed.

H2. 

The NEDC policy can reduce urban carbon emissions through green technology innovation.

Scholars have demonstrated that upgrading industrial structure can effectively reduce carbon emissions through a variety of methods [28]. To realize industrialization, China has given priority to the development of heavy industry, which leads to the overexploitation of natural resources and seriously excessive industrial emissions, resulting in resource waste, environmental pollution, and other problems. The secondary industry concentrates most of the high-energy sectors [29], and there is a significant difference in the energy structure between industry and the service industry, which means that in the process of the gradual upgrading of industrial structure, the energy structure dominated by fossil energy can be greatly improved to reduce urban carbon emissions. Therefore, the question becomes whether the development of e-commerce can change the urban industrial structure from a secondary industry to a tertiary industry. Some scholars have studied the impact of e-commerce development on industrial structure. It has been found that the development of e-commerce drives the development of related industries, such as the logistics industry and the customer service industry. Through the adjustment and optimization of industrial structure, e-commerce can realize the process of “retreating from the secondary industry to the tertiary industry”, promote the green transformation of “service-oriented” systems [8], and thus, reduce carbon emissions. Therefore, research hypothesis H3 is proposed.

H3. 

The NEDC policy can reduce urban carbon emissions through industrial structure upgrading.

4. Data and Methodology

4.1. Data Sources

The research sample of the study included 263 cities in China from 2008 to 2017. During the sample period, a total of 70 cities were approved as national e-commerce demonstration cities. Information about the national e-commerce demonstration cities is mainly available on the websites of the Ministry of Commerce of China, the provincial and municipal governments, etc. Carbon emission data were taken from the China Emission Accounts and Datasets [30], and other data were obtained from the China City Statistical Yearbook. For some missing data, the interpolation method was used for supplementation.

4.2. Research Methodology

For NEDCs, as a pilot policy meant to promote the development of e-commerce, the Chinese government has preferred to select cities with good environments for e-commerce development, wide enterprise application, and high local government attention, which leads to the problem of selectivity bias. Therefore, the PSM-DID model proposed by Heckman et al. [31] was used in this study to evaluate the effect. PSM can effectively address the problem of sample selection bias but cannot avoid the endogeneity problem caused by the omission of variables, while DID can solve the endogeneity problem and obtain the policy treatment effect but cannot solve the sample selection bias. By selecting a model that combines PSM and DID, the policy effect of NEDC construction on carbon emissions could be more accurately evaluated.

4.2.1. PSM

Samples from the treatment group and the control group were matched according to PSM before DID. In this way, a group of non-demonstration cities with characteristics close to those of the e-commerce demonstration cities were selected as the control group.

Specifically, first, based on the covariable $X_i$ of the demonstration city, the probability of city $i$ appearing in the treatment group was estimated, and the propensity score was estimated by the logit function:

$$p\left(X_i\right)=Pr\left(cit{y}_i=1|X_i\right)=\frac{exp\left(\beta X_i\right)}{1+exp\left(\beta X_i\right)}$$

(1)

where $X_i$ is the covariable vector, and $cit{y}_i$ represents the treatment group dummy variable, where $cit{y}_i=1$ denotes that a city is an e-commerce demonstration city, and $cit{y}_i=0$ denotes that a city is a non-demonstration city. $\beta$ is the logit regression coefficient.

Second, after the propensity score value was obtained according to Equation (1), the propensity score was used as the distance function for matching. According to the literature, kernel matching improves the sample utilization rate more than that of nearest neighbor and radius matching, so it is widely used. This study used the kernel matching method in the baseline regression. The average treatment effects on the treated can be written as:

$$ATT=\frac{1}{N^T}{{\displaystyle \sum }}_{i\in T}{Y}_i^T-\frac{1}{N^T}{{\displaystyle \sum }}_{j\in C}w\left(i,j\right)Y_j^C$$

(2)

where $Y_i^T$ is the observation result of city $i$ in the treatment group, and $Y_j^C$ is the observation result of city $j$ in the control group. $N^T$ is the number of individuals in the treatment group after matching, and $w\left(i,j\right)$ is the weight function.

4.2.2. PSM-DID

Considering that PSM does not address the endogeneity problem arising from the omission of unobservable variables, the matched samples were analyzed via DID to assess the policy effect of the construction of NEDCs on carbon emissions. Considering that the NEDC policy has been implemented gradually in batches, the traditional DID model was not applicable. Therefore, this study followed the method of Beck et al. [32] and used a multi-period DID model that can identify policy effects at multiple time points to examine the impact of the NEDC policy on carbon emissions in Chinese cities. Consistent estimates of the ATT were obtained, provided that the assumption of the negligibility of the means was maintained.

$$AT{T}_{PSM-DID}=\frac{1}{N_1}{{\displaystyle \sum }}_{i:i\in I_1{{\displaystyle \cap }}^{ }{S}_P}\left[\left(Y_{i1t}-Y_{i0t^{\prime }}\right)-{{\displaystyle \sum }}_{j:j\in I_0{{\displaystyle \cap }}^{ }{S}_p}w\left(i,j\right)\left(Y_{j0t}-Y_{j0t^{\prime }}\right)\right]$$

(3)

where $\left(Y_{i1t}-Y_{i0t^{\prime }}\right)$ represents the change in carbon emissions of city $i$ in the experimental group before and after the enforcement of the NEDC policy. $\left(Y_{j0t}-Y_{j0t^{\prime }}\right)$ is the change in the carbon emissions of city $j$ in the comparison group before and after NEDC policy enforcement. $N_1$ represents the number of treatment group individuals contained by $I_1{{\displaystyle \cap }}^{ }{S}_p$, $S_p$ is the set of common value ranges, $I_1=\{i:cit{y}_i=1\}$ represents the set of treatment groups, $I_0=\left\{i:cit{y}_i=0\right\}$ represents the set of control groups, and $w\left(i,j\right)$ is the weight corresponding to pairing $\left(i,j\right)$.

The econometric model used to estimate ATT_PSM-DID can be expressed as follows:

$$Y_{it}^{PSM}={\alpha }_0+{\alpha }_{1}cit{y}_i\times yea{r}_{it}+{\alpha }_2{X}_{it}+{\mu }_i+v_t+{\epsilon }_{it}$$

(4)

where $i$ denotes the city, $t$ denotes the year, and $Y_{it}$ is the explained variable, representing the total carbon emissions and carbon intensity of the city. The interaction term ($cit{y}_i\times yea{r}_{it}$) is the core explanatory variable, where $cit{y}_i=1$ if the city has been approved as an e-commerce demonstration city, and $cit{y}_i=0$ otherwise. The time dummy variable $yea{r}_{it}=1$ the year after approval, and $yea{r}_{it}=0$ indicates the year prior to approval. The regression coefficient ${\alpha }_1$ of ($cit{y}_i\times yea{r}_{it}$) reflects the impact of the NEDC policy on urban carbon emissions, which is the focus of this paper. $X_{it}$ is a series of control variables, including the level of economic development, fiscal revenue level, population size, environmental regulation intensity, degree of openness to the outside world, and urbanization level. ${\mu }_i$ represents the city fixed effects. $v_t$ represents the time fixed effects. ${\epsilon }_{it}$ represents a random disturbance term.

4.2.3. Mechanism of Mediation

The mediating effect model proposed by Baron and Kenny [33] was used to test whether industrial structure upgrading and green technology innovation have a significant mediating effect on the relationship between NEDCs and urban carbon emissions. On the basis of Equation (4), Equation (5) was constructed to represent the effect of the NEDC policy on the mechanism variables, and Equation (6) was constructed to test the effect of the NEDC policy and mechanism variables on urban carbon emissions. In this way, the logical relationships among the NEDC policy, mechanism variables, and urban carbon emissions were tested. The specific model is as follows.

$$M_{it}{}^{PSM}={\beta }_0+{\beta }_{1}cit{y}_i\times yea{r}_{it}+{\beta }_2{X}_{it}+{\mu }_i+v_t+{\epsilon }_{it}$$

(5)

$$Y_{it}{}^{PSM}={\gamma }_0+{\gamma }_{1}cit{y}_i\times yea{r}_{it}+{\gamma }_2{M}_{it}+{\gamma }_3{X}_{it}+{\mu }_i+v_t+{\epsilon }_{it}$$

(6)

where $M_{it}$ represents the mediating variables, namely industrial structure upgrading and green technology innovation. The rest of the variables are defined as previously described. The total effect of the policy is ${\alpha }_1$, the direct effect is ${\gamma }_1$ and the mediating effect is ${\beta }_1{\gamma }_2$. If both ${\beta }_1$ and ${\gamma }_2$ are significant in the regression, then the mediating effect of the $M_{it}$ variable can be considered significant. If the coefficient ${\gamma }_1$ is significant and decreases compared to the total effect ${\alpha }_1$, then it indicates a partial mediation effect; if ${\gamma }_1$ is not significant, then it indicates a full mediation effect. If either ${\beta }_1$ or ${\gamma }_2$ is nonsignificant, then the significance of the mediating effect needs to be further determined using the Sobel test.

4.3. Selection of Variables and Descriptive Statistics

(1) Explained variables. The explained variables in this paper are urban carbon emissions intensity $Lncogdp$ and total carbon emissions $Lnco2$. Among them, $Lncogdp$ is used in the baseline regression and $Lnco2$ is used in the robustness test. Both the total carbon emissions and carbon emissions intensity are taken as logarithms. The carbon emission intensity is the ratio of total carbon emissions to urban GDP.

(2) Core explanatory variables. The core explanatory variable is the interaction term $cit{y}_i\times yea{r}_{it}$, which is denoted as $DID$. If a city is approved as a national e-commerce demonstration city in the year of approval and in subsequent years, then $DID=1$; otherwise, $DID=0$.

(3) Mechanism variables. According to the above theoretical analysis, green technology innovation and industrial structure upgrading may be the transmission mechanisms by which the NEDC policy reduces urban carbon emissions. Green technology innovation was measured on the basis of the research of Zhang et al. [25] by the natural logarithm of the number of green patents in each city, which is denoted as $GTI$. The percentage of the secondary sector added value in GDP indicates the industrial structure upgrading [34], which is denoted as $ISU$.

(4) Control variables (covariates). In the PSM stage, the propensity score values were obtained by conditional probability estimation based on the observable variables, which were selected according to their relation to urban carbon emissions. Considering the close correlation between urban carbon emissions and economic development and the factors affecting the environment, and referring to the studies of Grossman and Krueger [27], Auffhammer et al. [35], and Xu et al. [36], six variables, including the economic development level, fiscal revenue level, population size, environmental regulation intensity, openness to the outside world, and urbanization level, were selected as control variables for this study. The economic development level is represented by the logarithm of per capita GDP, which is denoted as $Lnpgdp$, and the level of fiscal revenue is expressed by the proportion of the local government’s fiscal revenue in GDP, which is denoted as $Fiscal$. The population size is represented by the logarithm of the population size of the city at the end of the year, which is denoted as $Lnpop$. The proportion of industrial pollution regulation inputs in GDP represents the environmental regulation intensity, which is denoted as $Regu$. The share of FDI in GDP indicates the degree of openness to the outside world, which is denoted as $Open$. The share of the actual total urban population in the registered population represents the urbanization level, which is denoted as $Urban$. Descriptive statistics of the variables are shown in

Table 1. Descriptive statistics of variables.

Variables	Observations	Mean	Std. Dev.	Min	Max
$Lncogdp$	2862	8.0751	1.1096	4.6535	12.7001
$Lnco2$	2862	5.9722	1.1550	2.1168	9.5078
$DID$	2862	0.0999	0.2999	0.0000	1.0000
$GTI$	2862	7.4907	1.6099	1.7918	11.5801
$ISU$	2862	0.4871	0.1044	0.1490	0.851
$Lnpgdp$	2862	10.4550	0.6554	4.5951	13.0557
$Fiscal$	2862	0.0369	0.0372	0.0003	0.5000
$Lnpop$	2862	15.0854	0.6866	12.1335	17.3395
$Regu$	2862	0.5337	0.1506	0.0010	0.9557
$Open$	2862	0.0828	0.1152	0.0015	0.9637
$Urban$	2862	0.3238	0.2059	0.0105	2.4436

.

5. Results

5.1. PSM Results

Propensity score matching began with the estimation of propensity score values on the basis of the logit model to estimate the propensity score values of individuals entering the treatment group and then proceeded to match based on the propensity score values. The NEDC policy was first implemented in Shenzhen in 2009 and then in batches of cities in 2011, 2014, and 2017. Drawing on Blundell and Dias [37], a year-by-year matching approach was used to match the treatment group sample with the control group sample for multiple periods. Among them, 2009 and 2011 comprised the first batch of pilots, with 23 treatment groups; 2014 was the second batch of pilots, with 30 treatment groups; and 2017 was the third batch of pilot projects, with 17 treatment groups. To avoid the influence of the policy effect on the matching results, we used the matching time node before the implementation of the policy. At the same time, to avoid the impact of random events in the intermediate period due to the excessive advance of the matching time, we set the matching time in the year before the implementation of the NEDC policy, that is, the matching years of multi-period PSM were set as 2010, 2013, and 2016 in this study. Referring to the practice of Zhou and Wang [38], the total control group samples were determined based on the intersection idea, that is, the common control group samples matched several times were included in the total control group samples.

To test the matching effect, a balance test was first performed. The purpose of the balance test was to test whether the means of each covariate remained significantly different after matching. We used the balance test results of 2014 as an example. A total of three matches were conducted in this study. In the three matches, most variables were nonsignificant at the level of 10%, except for a few variables that were significant in individual years, which all satisfied the balance hypothesis and common support hypothesis, and the matching effect was good. The results are shown in

Table 2. Balance test results in 2014.

Variables	Sample	Mean		Standard Bias (%)	Standard Bias Reduction (%)	t-Statistic	p > t
		Treatment Group	Control Group
$lnpgdp$	Unmatched	10.8550	10.3330	83.9	92.0	19.16	0.000
	Matched	10.8140	10.8560	−6.7		−1.25	0.211
$Fiscal$	Unmatched	0.0207	0.0418	−70.1	94.6	−13.27	0.000
	Matched	0.0214	0.0226	−3.8		−1.52	0.128
$lnpop$	Unmatched	15.4720	14.9670	79.4	95.5	17.50	0.000
	Matched	15.4410	15.4180	3.6		0.75	0.454
$Regu$	Unmatched	0.5539	0.5275	18.6	62.4	3.99	0.000
	Matched	0.5529	0.5429	7.0		1.20	0.230
$Open$	Unmatched	0.1389	0.0656	58.1	80.3	14.99	0.000
	Matched	0.1286	0.1430	−11.4		−1.86	0.063
$Urban$	Unmatched	0.4334	0.2903	64.8	93.4	16.47	0.000
	Matched	0.3990	0.4086	−4.3		−0.93	0.354

Note: “Unmatched” represents samples before matching, while “Matched” represents samples after matching.

. The mean values of all covariates were significantly different before matching. After matching, except for the $Open$ variable, the t-statistics of the other covariates were not significant, the absolute value of standard bias was all less than 10%, and the degree of bias was greatly reduced. This indicates that the matched sample satisfies the balance assumption.

5.2. Results of PSM-DID Analysis

On the basis of PSM, the urban carbon reduction effect of the NEDC was assessed using the DID method according to Equation (4). By incorporating control variables into the baseline regression model one at a time, this study examined the impact of the NEDC policy on urban carbon emissions.

Table 3. Results of baseline regression.

Variables	$\mathit{L}\mathit{n}\mathit{c}\mathit{o}\mathit{g}\mathit{d}\mathit{p}$
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
$DID$	−0.0958 *** (0.0195)	−0.0969 *** (0.0191)	−0.0954 *** (0.0191)	−0.0957 *** (0.0194)	−0.0958 *** (0.0194)	−0.0942 *** (0.0197)	−0.0945 *** (0.0198)
$lnpgdp$		−0.0585 *** (0.0213)	−0.0733 *** (0.0248)	−0.0742 *** (0.0260)	−0.0743 *** (0.0259)	−0.0759 *** (0.0257)	−0.0766 *** (0.0257)
$Fiscal$			−0.1160 * (0.1104)	−0.130 * (0.1152)	−0.1292 * (0.1141)	−0.1261 * (0.1151)	−0.112 * (0.1126)
$Lnpop$				−0.0148 * (0.0728)	−0.0146 * (0.0730)	−0.0225 * (0.0694)	−0.0123 * (0.0667)
$Regu$					−0.0146 ** (0.0291)	−0.0291 ** (0.1410)	−0.0144 ** (0.0290)
$Open$						−0.0743 * (0.0976)	−0.0758 * (0.0963)
$Urban$							−0.0496 ** (0.0620)
YFE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
CFE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.9693 *** (0.0080)	1.5670 *** (0.2178)	1.5057 *** (0.2063)	1.2836 (1.1291)	1.2902 * (1.1346)	1.1689 * (1.0828)	1.3535 * (1.0431)
Observations	2690	2690	2690	2690	2690	2690	2690
R2	0.6406	0.6443	0.6453	0.6453	0.6454	0.6457	0.6459

Note: ***, **, * denote significance levels at 1%, 5%, and 10% respectively. Robust standard errors in parentheses are clustered at the city level. YFE denotes year fixed effects. CFE denotes city fixed effects.

shows all of the results of the baseline regression when different control variables were included. In column (1), the coefficient of the interaction term $DID$ is negative at the 1% significance level when controlling only for city fixed effects and time fixed effects, without adding other control variables. When control variables affecting urban carbon emissions are gradually added into columns (2) to (7) of Table 3, the interaction term coefficient remains negative at the significance level of 1%. According to the results in column (7), if the other conditions remain unchanged, then the carbon emission level of a city after it becomes a national e-commerce demonstration city decreases by 9.45% on average compared to that of a non-demonstration city. E-commerce gives full play to its advantages of information transmission across time and space and cost reduction, which can improve the economic effect while taking into account the environmental effect and reducing urban carbon emissions. This result preliminarily verifies the research hypothesis H1.

5.3. Parallel Trend Test and Dynamic Effects Analysis

As an effective method to evaluate the policy effect, the precondition of DID was that the samples of the demonstration cities and non-demonstration cities showed the same trend before the policy was carried out, i.e., parallel trends. After a city was approved as an e-commerce demonstration city, the dynamic change of the policy effect needed to be further tested. Therefore, this study used the event study method to conduct parallel trend tests and dynamic effect tests, setting up the following regression model.

$$Y_{it}={\delta }_0+{{\displaystyle \sum }}_{k=-9}^6{\rho }_{k}Polic{y}_{i,t+k}+{\delta }_1{X}_{it}+{\mu }_i+v_t+{\epsilon }_{it}$$

(7)

where $Polic{y}_{i,t+k}$ is a dummy variable. A negative value for $k$ represents $k$ years before the implementation of the NEDC policy, and a positive value represents $k$ years after the implementation of the NEDC policy. ${\rho }_k$ is the coefficient of focus, representing the difference in carbon emission levels between the demonstration cities and non-demonstration cities before or after $k$ years of NEDC implementation. The other variables are defined as in Equation (4). The test results are shown in Figure 1.

As shown in Figure 1, before the implementation of the NEDC policy, the estimated coefficients of the $Policy$ are not significant. This indicates that before the implementation of NEDCs, the change trends of the carbon emission levels in the demonstration cities and non-demonstration cities were basically parallel. Thus, the sample passes the parallel trend test. In addition, after the implementation of the NEDC policy, the policy effect was significant and negative. This indicates that NEDCs have a long-term carbon reduction effect. That is, the policy effect becomes increasingly stronger with the passage of time.

5.4. Robustness Tests

To further examine whether the results of reducing urban carbon emissions by NEDCs are robust, this study conducted robustness tests by replacing the explained variables and replacing the PSM method. The results of the robustness tests are presented in

Table 4. Robustness tests.

Variables	Replaced Explained Variable	Replaced PSM Method
	$\mathit{L}\mathit{n}\mathit{c}\mathit{o}2$	$\mathit{L}\mathit{n}\mathit{c}\mathit{o}\mathit{g}\mathit{d}\mathit{p}$
	(1) Kernel Matching	(2) Local Linear Regression Matching	(3) Radius Matching	(4) Mahalanobis Matching
$DID$	−0.0505 *** (0.0146)	−0.084 *** (0.016)	−0.059 *** (0.024)	−0.095 *** (0.016)
Controls	Yes	Yes	Yes	Yes
YFE	Yes	Yes	Yes	Yes
CFE	Yes	Yes	Yes	Yes
Constant	−1.0872 * (0.8532)	−0.9165 ** (0.870)	−1.0861 ** (0.953)	−1.1652 * (0.9376)
Observations	2690	2705	2643	2620
R2	0.7225	0.8654	0.8623	0.8702

Note: ***, **, * denote significance levels at 1%, 5%, and 10% respectively. Robust standard errors in parentheses are clustered at the city level. YFE denotes year fixed effects. CFE denotes city fixed effects.

. In column (1), total carbon emissions are used as a proxy variable for the level of carbon emissions in a city and then retested using the PSM-DID method. In column (1), the coefficient of the interaction term $DID$ is −0.0505, which is consistent in sign and significance with the baseline regression. In columns (2), (3), and (4) of Table 4, we replaced the PSM method in the baseline regression with local linear regression matching, radius matching, and Mahalanobis matching and conducted robustness tests. The coefficients of the $DID$ were all significantly negative. The robustness tests indicate that the conclusions obtained from the baseline regression model are robust and dependable. The NEDC policy can significantly reduce urban carbon emissions. Therefore, the research hypothesis H1 is supported.

5.5. Mechanism Analysis

In the theoretical analysis section above, we analyzed the transmission mechanism of industrial structure upgrading and green technology innovation on the national e-commerce demonstration city policy affecting urban carbon emissions. In this study, empirical analysis was conducted using Equations (4) to (6) to test the transmission mechanism hypothesis. The test results are shown in

Table 5. Mechanism test results of the NEDC policy impact on urban carbon emissions.

	$\left(1\right) \mathit{L}\mathit{n}\mathit{c}\mathit{o}\mathit{g}\mathit{d}\mathit{p}$	$\left(2\right) \mathit{I}\mathit{S}\mathit{U}$	$\left(3\right) \mathit{L}\mathit{n}\mathit{c}\mathit{o}\mathit{g}\mathit{d}\mathit{p}$	$\left(4\right) \mathit{G}\mathit{T}\mathit{I}$	$\left(5\right) \mathit{L}\mathit{n}\mathit{c}\mathit{o}\mathit{g}\mathit{d}\mathit{p}$
$DID$	−0.0945 *** (0.0198)	0.1351 *** (0.0208)	−0.0682 *** (0.0102)	0.1079 ** (0.0143)	−0.0755 *** (0.0162)
$GTI$			−0.1953 *** (0.2169)
$ISU$					−0.1758 ** (0.016)
Controls	Yes	Yes	Yes	Yes	Yes
YFE	Yes	Yes	Yes	Yes	Yes
CFE	Yes	Yes	Yes	Yes	Yes
Constant	1.3535 * (1.0431)	0.8537 * (0.9087)	1.0638 * (1.0435)	0.7546 * (0.5463)	1.0925 * (1.3042)
Observations	2690	2690	2690	2690	2690
R2	0.6459	0.7275	0.8109	0.9375	0.9278

Note: ***, **, * denote significance levels at 1%, 5%, and 10% respectively. Robust standard errors in parentheses are clustered at the city level. YFE denotes year fixed effects. CFE denotes city fixed effects.

. According to the concept behind the step test, model (1) confirms that the NEDC policy can significantly reduce urban carbon emissions, and model (2) verifies the impact of the NEDC policy on the industrial structure upgrading of the mechanism variable. The results show that the coefficient of $DID$ was 0.1351 and significantly positive at the 1% level, indicating that the NEDC policy had a positive effect on the industrial structure. In model (3), after adding the industrial structure, the coefficient of $DID$ was −0.0682, which was somewhat decreased compared to the benchmark regression coefficient, indicating that the industrial structure had a partial mediating effect in the path of carbon emission reduction by NEDC policies. Its mediating effect was 0.0264, accounting for 27.93% of the total effect. This shows that industrial structure upgrading is an important mechanism for the NEDC policy in reducing urban carbon emissions, and research hypothesis H2 is supported.

Similarly, on the basis of the results of model (1), model (4) verifies that the NEDC policy had a significant impact on the mechanism variable of green technology innovation. In model (5), after adding green technology innovation, the policy effect coefficient of e-commerce demonstration city construction on urban carbon emissions was −0.0755, which is lower than the benchmark regression coefficient, indicating that green technology innovation had a partial intermediary effect on the path of NEDC policy to reduce carbon emissions. Its intermediate effect was 0.0189, accounting for 20.07% of the total effect. This indicates that green technology innovation is an important mechanism by which the NEDC policy reduces urban carbon emissions. In summary, research hypothesis H3 is supported.

5.6. Heterogeneity Test

There are great differences among cities in geographical location, resource endowment, and other aspects, which may lead to heterogeneity of the carbon emission reduction effect of the NEDC policy across different cities. This study further analyzed the heterogeneity of the carbon emission reduction effects of the NEDC policy from the perspectives of different urban areas and different urban types.

First, the entire sample was divided into three subsamples according to the regionality of the city, and the resulting subsamples were tested. Models (1) to (3) of

Table 6. Results of heterogeneity test.

	Urban Regions			Urban Types
	(1) Eastern Region	(2) Central Region	(3) Western Region	(4) Non-Resource-Based	(5) Resource-Based
$DID$	−0.0847 (0.1763)	−0.124 ** (0.8754)	−0.0735 *** (0.0865)	−0.1093 (0.0865)	−0.039 *** (0.0323)
Controls	Yes	Yes	Yes	Yes	Yes
YFE	Yes	Yes	Yes	Yes	Yes
CFE	Yes	Yes	Yes	Yes	Yes
Constant	1.0236 (0.1364)	1.0376 * (0.1073)	0.9865 (0.1420)	1.3470 ** (0.1171)	1.2982 * (0.1271)
Observations	886	1072	732	1005	1685
R2	0.7052	0.8161	0.6936	0.6743	0.7105

Note: ***, **, * denote significance levels at 1%, 5%, and 10% respectively. Robust standard errors in parentheses are clustered at the city level. YFE denotes year fixed effects. CFE denotes city fixed effects.

are the results of regional heterogeneity tests, which show that the policy effect of NEDC construction on urban carbon emissions is significantly positive in the central and western regions, indicating that the NEDC policy can significantly reduce the carbon emissions of central and western cities, while the policy effect on the relatively developed eastern regions is nonsignificant. The reason may be that the relatively developed eastern cities tend to have a higher economic aggregate, greater population density, and higher energy consumption, which leads to higher carbon emission levels. Therefore, such cities form strong carbon emission dependence and carbon-locking effects. In contrast, the carbon emission levels of cities in central and western China are lower, the carbon lock-in effect is weaker, and the response to the NEDC policy is more sensitive and rapid. Therefore, cities in western China can exert the emission reduction effect of the policy more quickly, and thus, the policy is more effective.

Second, all sample cities were divided into resource-based and non-resource-based cities, and regression estimation was carried out. Models (4) and (5) of Table 6 are the results of urban-type heterogeneity tests. In model (4), the coefficient of the interaction term $DID$ was −0.1093, but not significant. In model (5), the coefficient of the interaction term $DID$ was −0.039, and significant. In conclusion, the NEDC policy is more effective in resource-based cities. The reason may be that resource-based cities are rich in natural resources and mainly rely on the resource industry for further development. For a long time, the urban development mode has tended to form carbon emission dependence and carbon-locking effects [39], and economic development has faced the dilemma of the “resource curse”. The establishment of national e-commerce demonstration cities promotes the application and development of the e-commerce industry, combining the traditional advantages of resource-based cities with IT-based e-commerce. In addition, the development of e-commerce transfers a section of labor and capital from the resource sector to the more diversified service industry sector. This helps to adjust and optimize the industrial structure and achieve green and low-carbon city development.

6. Discussion and Implications

6.1. Discussion

In this study, the NEDC policy was used as a quasi-natural experiment, and the causal relationship between the NEDC policy and urban carbon emissions in China was examined using the PSM-DID method. The findings indicate that the NEDC policy significantly reduces carbon emissions in Chinese cities. Most current studies on urban carbon emissions in China have focused on studying the factors that influence carbon emissions. These factors include industrial structure [10,11], energy structure [15], urban transportation [40,41], urban space and morphology [42,43], the digital economy [44,45], smart city construction [46,47], and low-carbon city policies [48,49]. More in line with this study, some scholars have explored the impact of e-commerce on environmental pollution [8,18,19,20], mainly focusing on air pollution and green development. To date, no study has included urban e-commerce development and carbon emissions in the same research framework. This study is the first to examine the impact of urban e-commerce transformation on urban carbon emissions from the perspective of e-commerce development. In doing so, this paper fills the gap in the research on the carbon emission reduction effect of e-commerce development, presenting a new way to achieve carbon emission reductions. In addition, the NEDC policy effect is significantly heterogeneous across city samples. The mechanism test reveals that industrial structure upgrading and green technology innovation are important channels through which the NEDC policy can reduce carbon emissions in Chinese cities. This finding provides strong support for clarifying the underlying mechanism of the impact of urban e-commerce development on carbon emissions and for the Chinese government to formulate policies for the parallel development of carbon emission reduction and urban e-commerce.

6.2. Implications

As a pilot policy for the promotion of the transformation of urban e-commerce, the NEDC policy plays an important role in improving the environment and promoting the process of urban green development. However, there is still a lack of empirical studies on the relationship between the NEDC policy and urban carbon emissions. In view of this, the study uses the PSM-DID model to explore the carbon reduction effect of the NEDC policy based on the data from 263 cities in China. The study shows that the NEDC policy can significantly reduce urban carbon emissions. The heterogeneity analysis showed that NEDC is more effective in central, western, and resource-based cities in China. Finally, through mechanism analysis, it was found that the NEDC policy can reduce carbon emissions through industrial structure upgrading and green technology innovation.

These findings have several implications for China and other developing countries. First, the NEDC policy is conducive to urban low-carbon development, and the transformation of urban e-commerce should be further strengthened. The government should appropriately increase investment in information and communication technology research and development to lay a solid digital foundation for the construction of national e-commerce demonstration cities. In this paper, the dynamic effects show that the carbon emission reduction effect of the NEDC policy becomes increasingly stronger over time. Therefore, the Chinese government should focus on building a long-term mechanism for the carbon reduction effect of NEDC construction to ensure the sustainability of the effect of low-carbon city construction. Second, the Chinese government should attach importance to the role of industrial structure upgrading in urban low-carbon development, promote regional industrial structure upgrading to advanced development, limit the development of enterprises with high energy consumption and high carbon emissions enterprises, and give full play to the role of market mechanisms to promote the low-carbon industrial structure. Third, we need to step up innovation in green technologies, accelerate industrialization with technological progress, eliminate outdated production capacity, improve demand structure, and promote the development of emerging industries. Fourth, attention should be given to regional differences. On the one hand, the government should develop a long-term mechanism for urban e-commerce development to promote energy conservation and emission reduction. On the other hand, the mandatory “one size fits all” policy should be avoided, and the NEDC policy should be promoted according to local conditions based on the characteristics of a city, its resource endowment, and its economic development status.

7. Limitations and Future Research

This study has several limitations. First, due to current data availability, the sample period of this paper was from 2008 to 2017. Thus, it was difficult to accurately assess the effect of the NEDC policy implemented in 2017, and the sample years of this paper need to be further supplemented. Second, whether the carbon reduction effect of the NEDC has an impact on neighboring cities and its spatial spillover effect need to be explored in depth in future studies.