Do Smart Cities Restrict the Carbon Emission Intensity of Enterprises? Evidence from a Quasi-Natural Experiment in China

Abstract

The concept of “smart cities” plays a positive role in the overall green and sustainable development of a nation. However, it is still debated whether smart cities can restrain the carbon emission intensity at the micro-level and promote the green transformation of enterprises. To this end, based on China’s smart city policy (SCP) and regional enterprise data from 2008 to 2015, we study the impact of SCP on the carbon emission intensity of local enterprises, using the difference-in-differences method. The results show that SCP significantly reduces the carbon emission intensity of enterprises, and the estimated results remain significant after the propensity score matching. The mechanism analysis finds that digital transformation, innovation by enterprises, and urban green innovation all strengthen the impact of SCP on the carbon emission intensity of enterprises. The conclusions extend the scope of the existing research and provide suggestions for micro-enterprises to take advantage of SCP for better development.

1. Introduction

A smart city is a new, efficient, and technologically advanced city that integrates green and social development [1]. Such cities effectively alleviate the inability to process information related to urban environmental protection and resource utilization efficiency [2]. Through the full application of a new generation of information technology, smart cities can effectively optimize urban services and operations [3], alleviate the information processing capacity contradiction between urban environmental protection and resource utilization efficiency, and achieve the benefits of extensive informatization, industrialization, and urbanization [4]. With the rise of the internet and mobile technology, the smart city is an advanced form of information-based city development.

Owing to its potential advantages brought by digital technologies, the smart city has attracted the attention of many countries, such as the USA, Germany, Japan, and China [5]. However, some studies indicate that environmentally friendly smart cities may exacerbate urban pollution. For example, the large-scale information and communication technology (ICT) and industrial construction brought about by smart city policy (SCP) can lead to an increase in electricity consumption, which, in turn, increases carbon emissions [6]. Although the increases in ICT and internet penetration driven by SCP are not always beneficial to the environment, other studies indicate that, in general, smart cities positively impact the urban environment (e.g., carbon emission reduction) [7].

The above views are quite divergent, and the reason is that many influencing factors contribute to the total carbon emissions in a city. Among these, enterprises, which are the key drivers of an urban economy, generate most of the city’s carbon emissions. To some extent, the carbon emission intensity of enterprises directly determines that of the entire city. Therefore, it is necessary to study the impact of SCP on the carbon emission intensity of enterprises at the micro-level. Furthermore, it is worth noting that a smart city is a recently adopted concept that relies mainly on digital transformation to realize city smartness [8]. Digital technologies are closely associated with the concept of a smart city [9]. We posit that the relationship between smart cities and digitalization can promote the adjustment of the city’s industrial structure, thereby influencing the city’s carbon emissions. This speculation is reasonable from a macro perspective, that is, thinking from the perspective of the whole city. However, does this speculation hold true for micro-enterprises in the city? In other words, does SCP influence the digital transformation of the micro-enterprise and subsequently impact its carbon emission intensity? Most mainstream research uses the construction of information infrastructure in a city to measure its degree of digital transformation. This method, however, does not adequately reflect the degree of digital transformation of urban enterprises.

As mentioned, even if SCP may directly influence some economic and environmental indexes at the city level, their influence at the enterprise level remains ambiguous. There are some specific examples in China showing that SCP has impacted the activity of pollution emission for micro-enterprises. A representative example is the State Grid Corporation of China (SGCC). In 2018, the SGCC completed a total of 38,000 electric energy replacement projects with the support of the “Digital New Infrastructure” policy brought by smart cities, replacing 135.3 billion kWh of electricity, equivalent to reducing coal burning by 75.77 million tons and reducing carbon dioxide emissions by 1.35 billion tons. In November 2020, the SGCC developed China’s first globally empowered industrial internet platform in the energy sector. This platform can provide digital intelligent services in energy production and consumption, which promote the digital transformation of the enterprise.

Thus, to investigate whether this impact is common and infer the causal relationship between SCP and the carbon emission intensity of enterprises, we use a difference-in-differences (DID) method to study a quasi-natural SCP experiment based on China’s SCP and regional enterprise data from 2008 to 2015. Furthermore, we use propensity score matching (PSM) as part of a robustness test. The DID is a useful and popular method to evaluate policies, as it can significantly reduce problems that arise from endogeneity. The specific process of our DID method is described in Section 3.1. Our research period from 2008 to 2015 was selected because 2015 is the latest year that enterprise-level carbon emission intensity data are available (explained in detail in Section 3.1). During this period, SCP significantly reduced the carbon emission intensity of enterprises. Furthermore, research has found that the digital transformation of enterprises, enterprise innovation, and urban green innovation have strengthened the inhibitory effect of SCP on the carbon emission intensity of enterprises.

Our study’s novelty is established in two aspects. First, we directly study the effect of SCP at the micro-level, whereas most of the extant literature focuses only on the effect of SCP at the macro-level, including cities’ pollution emissions, energy consumption, and green development. Thus, we extend the impact of SCP to the micro-enterprise level and study how SCP influences enterprises’ activities and strategies. We believe this is a key contribution to the literature. Second, we analyze the relationship between SCP, digital transformation, and green innovation, looking at enterprise-level data. In the mechanism analysis, we find that SCP significantly improves green innovation performance and enhances the speed of digital transformation for micro-enterprises. Consequently, our results provide new perspectives for further research studying SCP.

Additional contributions are as follows: (1) Using the empirical method, we analyze the impact of SCP on carbon emission intensity at the enterprise level and expand the conclusions of existing studies at the city level; (2) We use digital word frequency to measure the digital transformation of enterprises, which we find to have strengthened the inhibitory effect of SCP on corporate carbon emissions. This finding also verifies Bhujabal et al.’s [7] conclusion regarding reducing carbon emissions through urban information infrastructure construction at the micro-level. As Xu et al. [10] indicate, the positive impact of innovation affects the carbon emission intensity of cities and enterprises through the adjustment of both the industry structure and enterprise production methods. Accordingly, our results verify the positive impact of SCP on enterprise innovation.

The rest of this paper is presented as follows. Section 2 reviews the relevant literature and introduces the implementation of China’s SCP. Section 3 introduces the data sources, variable measurement, and econometric model. Section 4 conducts empirical analyses, robustness tests, a heterogeneity analysis, and a mechanism analysis. Section 5 presents discussions. Section 6 provides conclusions and corresponding suggestions.

2. Literature Review and Hypothesis Development

2.1. Smart Cities and Ecological Environment

The prominent feature of smart cities is the application of ICT and the large-scale construction of related infrastructure. ICT represents the image and expectations of the future [11]. Building a smart city is one significant achievement in the path toward realizing sustainable urban development, which leads to the application of information technology and an increase in the overall competitiveness of cities [12]. Under rapid economic development, the deterioration of the urban environment caused by human activities has become a problem that cannot be ignored [13]. Therefore, smart cities have become a key direction in transforming urban development in many countries [5]. Regions with developed economies and rapid information technology development usually enjoy a high level of social informatization and digitalization. This means they have good basic conditions for smart development and relatively stable planning schemes for smart city construction. Developing a digitally driven smart city is an important way to promote green and sustainable economic growth [5]. Smart cities include the characteristics of urbanization (infrastructure construction) and informatization (digital facility construction and application) because, to a certain extent, they are a product of the combination of these two elements.

Many scholars have researched the impact of urbanization on the environment. Urbanization is seen as an important factor in the soaring global energy consumption and the rapid increase in carbon dioxide emissions [14]. Early studies have found that, since urbanization often accompanies industrialization, there is a close correlation between urbanization and the greenhouse effect [1], which leads to increased urban pollution. Research in developing countries shows that the impact of urbanization on carbon emissions forms an inverted “U”-shaped relationship [6]. However, some scholars posit that environmental pollution will ease with urban expansion, mainly because wealthy cities can transfer part of the environmental cost to other regions [15]. In particular, urbanization measured by different indicators has various impacts on carbon dioxide emissions [14], which explains the differences in research conclusions regarding the impact of urbanization and carbon emissions.

As mentioned, the smart city is the embodiment of in-depth contemporary urban development. Through the combination of the Internet of Things (IoT) and big data, smart cities have greatly promoted the use of professional technology to solve environmental pollution problems [16], which is crucial to urban development and planning [2,16]. Existing research can be divided into two categories. The first explores the factors that promote the construction of smart cities, primarily knowledge [17], IoT systems [5,7,18], and information technology construction [3,9]. The second focuses on the impact of smart cities on urban development. Representative studies have demonstrated the role of smart cities in promoting urban innovation [10,15] and urban competitiveness [12].

Based on the feature of smart cities and their impact on the ecological environment, we propose hypothesis 1.

Hypothesis 1 (H1).

The impact from SCP on the carbon emission intensity of enterprises is significantly negative.

2.2. Smart City Policy and Digitalization

We should consider the potential influence that SCP will bring. The digital transformation of an enterprise refers to the application of digital technology and equipment in the process of business improvement [19,20,21]. Modern society has entered the era of the digital economy [20], and digital technology has significantly changed original production methods, business models, and organizational patterns and has even subverted basic assumptions in many innovation theories [21]. One of the salient features of smart cities is the large-scale application of digital and information technology [22], wherein digital technology and traditional production models are inter-embedded, and production resources are reorganized and optimized with the help of technological innovation [23]. The rapid development of the digital economy has also stimulated research on the digital transformation of enterprises; for example, Yoo et al. [20], Nambisan et al. [21], Li et al. [24], Libert et al. [25], Vial [26], and Nwankpa and Roumani [27]. Thus, we find that the prominent feature for enterprises brought by SCP is to process digital transformation. As mentioned, digital technologies usually lead to less carbon emissions, so we develop hypothesis 2.

Hypothesis 2 (H2).

Digital transformation enhances the negative impact from SCP on the carbon emission intensity of enterprises.

2.3. Smart City Policy in China

It should be noted that the characteristics and effects of SCP may vary from country to country. SCP appears to be the “patent” of developed countries, as few developing countries implement it. China is among the few developing countries that implement smart cities, which provides a reference for in-depth research on the application of smart cities in different countries.

China’s smart city construction began relatively late but has developed rapidly. In January 2013, the Ministry of Housing and Urban-Rural Development of the People’s Republic of China (MOHURD) announced China’s smart city pilot list, which included 37 prefecture-level cities, 50 districts (counties), and 3 towns. A total of 90 cities formed the initial wave of smart city construction in China. In March 2021, the 14th Five-Year Plan for the National Economic and Social Development of the People’s Republic of China and the Outline of Vision 2035, promulgated by the Chinese government, further emphasized the importance of building many smart cities. Since then, local governments have responded positively and have released development plans with the long-term goal of building smart cities by 2035. With the exception of Shaanxi and Jilin, the other 29 provinces have pointed to the importance of promoting smart city construction and improving the level of intelligence in social governance. At present, China is still promoting the construction of smart cities nationwide.

Qian et al. [5] employed the PSM-DID method to study the impact of China’s SCP on economic green growth based on SCP implementation data. They found that smart cities can reduce urban unit gross domestic product (GDP), energy consumption, and waste emissions significantly. Xu et al. [10] and Xin and Qu [19] also used the PSM-DID method to study the impact mechanism of China’s smart city construction on the level of urban innovation and green total factor productivity. Yu and Zhang [4] examined the impact of China’s SCP on city-level energy emission intensity based on the DID method and found that smart city construction had a significant positive impact on city-level energy efficiency. The above studies analyze the impact of China’s smart cities on urban construction and future development from different perspectives and find diverse impacts of China’s SCP. The effects are more significant in China’s megacities and central cities. Based on these studies, we find that SCP can promote green innovation at the city level. However, it is worth studying whether SCP can promote green-related innovation at the enterprise level. Hence, we develop Hypothesis 3.

Hypothesis 3 (H3).

The impact from SCP on green innovation at the enterprise level is significantly positive.

Based on our literature review, we found that most relevant studies use samples and data at the prefecture or city level, whereas research on the impact of smart cities on carbon emission intensity at the enterprise level is still scarce. Based on China’s SCP and the listed companies’ micro-level data, we study the impact of SCP on enterprises’ carbon emission intensity. This information can enrich the micro-level analysis based on existing macro-level research.

3. Materials and Methods

According to a previous analysis, we found that SCP was not enacted in all cities in China and that smart cities remain a small proportion of all cities. For convenience, the cities that have become smart city pilot cities based on SCP were the treatment group, and the rest were the control group. As many studies have shown, enterprises’ carbon emission intensity remains steady over time; that is, the trends of carbon emission intensity in the treatment and control groups would be parallel, in theory (we test this in Section 4.2). Hence, China’s SCP can be an ideal quasi-natural experiment, suitable for the DID method to infer causality between SCP and the carbon emission intensity of enterprises.

The specific estimation model is shown in Section 2.3. First, we conduct a baseline regression to test whether there is a causal relationship between SCP and carbon emission intensity. Second, if the estimated result is significant and passes the parallel trend test, the DID method will verify whether a causal relationship exists. Third, we conduct placebo tests to check whether the relationship is fake, as the existence of a relationship does not equate to robustness. Fourth, we use the PSM-DID method to alleviate the self-selection problem. Fifth, the heterogeneity analysis is needed to test whether the former relationship was influenced by the degree of industrialization, the enterprise scale, and the enterprise age, all of which are important. Finally, through the mechanism analysis, we explain why SCP could affect the carbon emission intensity of enterprises from the perspective of digital transformation and innovation. Our framework and steps are shown in Figure 1.

3.1. Data Resource

Our data consist of three parts. The first part is the Chinese National Tax Survey Database (CNTSD) from 2008 to 2015. Each enterprise needs to pay an energy tax if it consumes energy. This database contains the consumption of three types of fossil fuels by energy-consuming enterprises. In addition, it should be noted that 2015 is the latest year in this database. The second part is the official website of the MOHURD; 2008–2015 SCP data for China were obtained through manual sorting. Finally, the other data of the listed companies and other prefecture-level cities come from the China Stock Market and Accounting Research Database (CSMAR) and various statistical yearbooks of the provinces in China. Links to the above data are given in the Data Availability Statement at the end of the article.

3.2. Variables Selection

3.2.1. Enterprise Carbon Emission Intensity

Referring to Cui et al. [28], the dependent variable in this study is enterprise carbon emission intensity (lnCarbonEff). The current method for measuring carbon emissions (CO₂) is based on the Guidelines for National Greenhouse Gas Inventories, published by the United Nations Intergovernmental Panel on Climate Change (IPCC) in 2006. The calculation method is proposed as

$${CO}_2={\sum }_{i=1}^n{E}_i\times NC{V}_i\times CE{F}_i\times CO{F}_{\dot{i}}\times \frac{44}{12}$$

(1)

where E represents final energy consumption; NCV is the net calorific value of energy (called the average low calorific value in the Chinese national standard GB/T2589-2008); CEF is the carbon emission factor per unit of calorific value equivalent; COF is the carbon oxidation factor (99–100% of the carbon in fossil fuels is oxidized; so, according to IPCC, the default value of COF is set to 1); 44 and 12 are the carbon dioxide and carbon molecular weights, respectively; and $i$ represents the category of various energy sources. According to the formula provided by the IPCC, the carbon dioxide emissions of enterprises can be calculated when their energy consumption is known.

However, owing to the difficulty in obtaining enterprise power consumption data and the large spatial and temporal differences between carbon emissions generated by power production and consumption, we focus only on the direct carbon emissions generated by the direct fossil energy consumption of enterprises. By dividing fossil fuel carbon emissions by the enterprises’ gross output value (GOV), the carbon emission intensity of industrial enterprises is measured by the logarithm of carbon emissions per unit of output value. The calculation method is presented as

$$\ln CarbonEff=\ln \frac{C{O}_2}{GOV}$$

(2)

3.2.2. Smart City Policy

The independent variable in our study is SCP (Policy), and it is a dummy variable. If the city $c$ becomes a smart city in year $t$, it is assigned a value of 1. Otherwise, it is assigned a value of 0.

3.2.3. Control Variables

To study the precise effect of SCP on the carbon emission intensity of enterprises, we need to control some factors at the enterprise and city level. We now describe the added control variables and selection bases. Large-scale enterprises are subject to more social supervision, and more shareholders pay attention to enterprise environmental information to evaluate their environmental performance [29]. Therefore, we control the size of the enterprise (Size), measured by the logarithm of the total assets of the enterprise. Generally, enterprises with better profitability can afford the expenditures required to reduce carbon emissions and optimize enterprise carbon emission intensity [30,31]. Therefore, we control the return on assets (ROA) and debt-to-asset ratio (DAR). Furthermore, fixed assets, such as plants, machines, and equipment, can reflect capital intensity, which is usually related to polluting activities [32]. Therefore, we control both enterprise fixed assets (lnFX) and regional fixed asset investment (lnFI). Considering that property rights will significantly affect the enterprise’s business performance and strategy formulation [33], which will potentially affect the enterprise’s carbon emissions and pollution indicators, the company attribute (Equity) is controlled as the last control variable at the enterprise level. Considering that the level of economic development will significantly affect the business activities of local enterprises, and a large number of studies have shown that the level of economic development is also the main factor affecting urban carbon emissions [6,34,35,36,37], we add the logarithm of prefecture-level city GDP (lnGDP) to control differences in the levels of regional economic development.

Table 1. Primary variables and explanations.

Variable Type	Symbol	Variable Name	Explanations
Dependent variable	lnCarbonEff	Enterprise carbon emission intensity	See Section 3.2.1.
Independent variable	Policy	Smart city policy	Dummy variable. See Section 3.2.2.
Enterprise-level variables	Size	Enterprise scale	Total assets of the enterprise.Logarithmic value
	ROA	Return on assets	Net profit/total assets of the enterprise
	DAR	Debt-to-asset ratio	Total liabilities/total assets of the enterprise
	Equity	Enterprise property	Stated-owned, non-stated-owned, overseas-funded enterprise, and so on.
	lnFX	Fixed assets	Fixed asset logarithm of the enterprise.Logarithmic value.
	DTR	Digital transformation ratio	See Section 4.6.1.
	RDSSR	R&D sum spend ratio	See Section 4.6.2.
	PQS	Patent quote sum	See Section 4.6.2.
City-level variables	lnGDP	Gross regional product	Logarithmic value. See Section 4.4.
	lnGDP1	Gross regional product of primary industry	Logarithmic value. See Section 4.4.
	lnGDP2	Gross regional product of secondary industry	Logarithmic value. See Section 4.4.
	lnGDP3	Gross regional product of tertiary industry	Logarithmic value. See Section 4.4.
	GDPR1	Proportion of GDP of primary industry	Gross regional product of primary industry/GDP
	GDPR2	Proportion of GDP of secondary industry	Gross regional product of secondary industry/GDP
	GDPR3	Proportion of GDP of tertiary industry	Gross regional product of tertiary industry/GDP
	GreenPatent	Green patent applications	See Section 4.6.2.
	GreenUtility	Green utility model applications	See Section 4.6.2.
	lnFI	Fixed asset investment at the city level	Logarithmic value.

presents the variables used in our study, and

Table 2. Descriptive statistics.

Variable Type	Symbol	Sample Size	Mean	Standard Deviation	Min.	Max.
Dependent variable	lnCarbonEff	8362	0.10	0.42	0.00	8.69
Independent variable	Policy	8362	0.30	0.46	0.00	1.00
Enterprise-level variables	Size	8362	14.45	1.72	0.00	23.55
	ROA	8362	0.04	0.09	−3.20	0.21
	DAR	8362	0.44	0.24	0.00	1.46
	Equity	8362	4.63	2.50	1.00	8.00
	lnFX	8362	11.80	2.57	0.00	20.25
	DTR	8362	0.39	1.13	0.00	32.92
	RDSSR	4897	4.76	5.25	0.00	58.85
	PQS	5980	110.30	300.30	0.00	2274
City-level variables	lnGDP	8362	8.49	1.12	3.85	10.20
	lnGDP1	8362	14.18	0.77	9.73	16.18
	lnGDP2	8362	16.82	1.06	11.91	18.27
	lnGDP3	8362	16.97	1.41	12.39	19.12
	GDPR1	8362	4.95	5.36	0.03	48.64
	GDPR2	8362	45.03	11.45	17.02	91.00
	GDPR3	8362	50.05	13.80	8.50	86.60
	GreenPatent	11,682	1829.62	2759.47	0.00	13,971.00
	GreenUtility	11,682	1252.67	1541.71	0.00	7268.00
	lnFI	8362	17.00	0.89	13.48	18.85

Note: This table includes all the variables used in this study, some of which are added as control variables in the regression. Variables lnGDP1 to GDPR3 are used as covariates in propensity score matching. Explanations of the rest of the important abbreviations: SCP: smart city policy; ICT: information and communication technology; DID: difference-in-differences method; PSM: propensity score matching method; IPCC: Intergovernmental Panel on Climate Change.

presents the descriptive statistics of these variables.

As shown in Table 2, the sample size for our main variables is 8376, which is sufficient to ensure the validity of our results. For the main variables, the standard deviations of lnCarbonEff and Policy are comparatively low, which means they are fewer extreme values. Other variables have similar features. Therefore, the data we use are comparatively suitable, and our later research results are credible.

3.3. Econometric Model

Referring to Yu and Zhang [4], we use the DID method to examine the impact of China’s SCP on the carbon emission intensity of enterprises. By adding time, individual, regional, and industry fixed effects and clustering to cities, the designed estimation model is shown as

$$\ln CarbonEf{f}_{i,t,j,c}={\alpha }_0+{\alpha }_{1}Polic{y}_{t,c}+{\sum }_k{\alpha }_k{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(3)

where subscripts i, t, j, and c represent the enterprise, year, industry, and (prefecture-level) city, respectively. lnCarbonEff is the carbon emission intensity of the enterprise; Policy is SCP; X_k is a series of control variables; γ_i, λ_t, μ_c, ν_j represent the individual (enterprise), time (year), region (city), and industry fixed effects; ε_i,t,j,c is a random disturbance term; and ${\alpha }_0$ is a constant. The core coefficient we focus on is ${\alpha }_1$, whose economic implication is the impact rate of SCP on the carbon emission intensity of enterprises.

4. Results and Analysis

4.1. Baseline Regression

According to model (3), the estimation results of SCP on the carbon emission intensity of enterprises are shown in

Table 3. Regression results of smart city policies and carbon emission intensity.

Variable	lnCarbonEff
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Policy	−0.052 **	−0.052 **	−0.052 **	−0.052 **	−0.057 **	−0.058 **	−0.070 **	−0.078 **
	(−2.02)	(−2.02)	(−2.02)	(−2.04)	(−2.19)	(−2.24)	(−2.46)	(−2.02)
Size		−0.005	−0.005	−0.006	−0.006	−0.011	−0.010	−0.010
		(−0.43)	(−0.47)	(−0.54)	(−0.52)	(−0.94)	(−0.86)	(−0.79)
ROA			−0.089	−0.069	−0.076	−0.082	−0.101	−0.100
			(−1.09)	(−0.83)	(−0.83)	(−0.88)	(−1.01)	(−0.79)
DAR				0.071	0.092	0.084	0.102	0.100
				(1.19)	(1.52)	(1.37)	(1.56)	(1.33)
Equity					0.023 **	0.024 **	0.019	0.019
					(1.99)	(2.06)	(1.50)	(0.99)
lnFX						0.008	0.006	0.006
						(1.47)	(1.04)	(0.89)
lnGDP							0.131	0.166
							(0.73)	(0.82)
lnFI								−0.067
								(−0.52)
Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.115 ***	0.180	0.189	0.171	0.048	0.024	−1.023	−0.169
	(15.39)	(1.18)	(1.24)	(1.12)	(0.29)	(0.15)	(−0.69)	(−0.09)
Observations	7871	7871	7869	7869	7797	7783	7281	7247
Adj. R-squared	0.132	0.132	0.132	0.132	0.138	0.140	0.143	0.106
F Statistics	4.076	2.129	1.816	1.717	2.353	2.351	1.914	0.958

Note: t-values are reported in parentheses. The robust standard errors are clustered at the city level. *** and ** represent significance at the levels of 1% and 5%, respectively. Policy: SCP; lnCarbonEff: carbon emission intensity of enterprises; Size: enterprise scale; ROA: return on assets; DAR: debt-to-asset ratio; Equity: enterprise property; lnFX: fixed assets; lnGDP: gross regional product; lnFI: fixed assets investment at the city level; Adj.: adjusted.

. Without any control variables, the estimated coefficient of ${\alpha }_1$ is −0.052, as shown in column (1). This indicates the originally negative and significant relationship between SCP and the carbon emission intensity of enterprises. The control variables are gradually added to columns (1) to (8). We can see in column (8) that, after adding control variables, fixed effects, and clustering to cities, the estimated coefficient becomes −0.078 and is significant at the 5% level. That is, SCP has effectively reduced the carbon emission intensity of enterprises. Hence, this verifies the relationship between SCP and the carbon emission intensity of enterprises, and the hypothesis H1 is accepted.

4.2. Parallel Trend Test

The premise of the validity of the DID method is to pass the parallel trend test. Referring to Beck et al. [28], we conduct a parallel trend test on the treatment group (Policy = 1) and the control group (Policy = 0). The design estimation model is shown in formula (4), and the results are shown in Figure 2. The horizontal axis of Figure 2 is the periods before and after the establishment of SCP, and the vertical axis is the percentage change of the estimated coefficient of the test statistic $a_m$. We can see that, before the SCP took place, the 95% confidence intervals of the estimated coefficients of the treatment group and the control group were estimated to include the 0 in the vertical axis; that is, the difference between the treatment and the control group was not significant. Thus, the parallel trend test passed.

$$\ln CarbonEf{f}_{i,j,t,c}=a_0+{\sum }_{m=1}^4{a}_{m}Polic{y}_{t,c}^{-m}+{\sum }_{n=1}^3{a}_{n}Polic{y}_{t,c}^{+n}+{\sum }_k{a}_k{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(4)

4.3. Placebo Test

The placebo test is a robustness test that changes the timing of policy shocks artificially [38]. To further test the robustness of the results, we refer to the method of Lu et al. [39] and use the placebo test. The specific method is to select the known cities without SCP as the treatment group for regression within the research interval. Repeating this random sampling process 1000 times, the normal distribution kernel density of the 1000 “pseudo-policy” estimated coefficients is shown in Figure 3, and the estimated coefficient −0.078 of the baseline regression in this study is marked with a dashed line in the figure. We can see in Figure 3 that the placebo estimation results are significantly different from the baseline regression results, indicating that there is no false association between SCP and the carbon emission intensity of enterprises.

4.4. PSM-DID Estimation

As mentioned above, China’s SCP reflects obvious regional heterogeneity, mainly determined by the level of local economic development. Therefore, we refer to the ideas of Xin and Qu [1], Yu and Zhang [4], Qian et al. [5], and Xu et al. [10] and re-estimate the model (4) by PSM to reduce the estimation error caused by potential sample selection bias. Then, the matched samples undergo the DID estimation; the results can better reflect the real impact of SCP on the carbon emission intensity of enterprises.

As covariates, we select the gross regional product (GDP), the gross regional product of each industry sample (GDP1, GDP2, GDP3), and their ratio to the GDP (GDPR1, GDPR2, GDPR3) in the study interval. We use the 1:1 nearest neighbor method to match the samples. PSM is conducted, and economic development indicators after matching are shown in Figure 4.

Part (a) of Figure 4 shows that the standard deviation of each indicator is reduced significantly after matching. Specifically, all covariate errors are reduced to within 20%, which indicates that the PSM method is suitable for our research. Part (b) of Figure 4 represents the propensity score of the treatment and control groups. In each score interval, the density of both the treatment and control groups is moderate. Parts (c) and (d) of Figure 4 show that, after the PSM, the treatment and control groups are approximately gathered at the intervals [0.5, 0.7].

The matched samples are re-estimated using DID, and the results are presented in

Table 4. Regression results of the PSM-DID estimation.

Variable	lnCarbonEff
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Policy	−0.080 *	−0.082 *	−0.082 *	−0.081 *	−0.091 *	−0.093 *	−0.124 **	−0.133 *
	(−1.69)	(−1.72)	(−1.72)	(−1.71)	(−1.91)	(−1.95)	(−2.35)	(−1.89)
Size		−0.026	−0.027	−0.028	−0.027	−0.028	−0.013	−0.013
		(−1.27)	(−1.31)	(−1.36)	(−1.32)	(−1.26)	(−0.55)	(−0.41)
ROA			−0.103	−0.084	−0.100	−0.194	−0.323	−0.320
			(−0.85)	(−0.67)	(−0.67)	(−0.97)	(−1.20)	(−0.96)
DAR				0.068	0.086	0.058	0.096	0.102
				(0.66)	(0.81)	(0.53)	(0.73)	(0.70)
Equity					0.009	0.010	0.003	0.003
					(0.52)	(0.55)	(0.15)	(0.17)
lnFX						0.000	−0.011	−0.010
						(0.01)	(−1.07)	(−1.30)
lnGDP							0.590 *	0.633 *
							(1.73)	(1.72)
lnFI								−0.134
								(−0.94)
Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Industry fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Constant	0.136 ***	0.508 *	0.524 *	0.511 *	0.450	0.474	−4.244	−2.403
	(12.44)	(1.73)	(1.78)	(1.73)	(1.44)	(1.51)	(−1.56)	(−0.80)
Observations	3557	3557	3557	3557	3520	3509	2999	2976
Adj. R-squared	0.131	0.131	0.131	0.131	0.141	0.140	0.123	0.033
F Statistics	2.871	2.238	1.732	1.407	1.315	1.180	1.798	0.842

Note: t-values are reported in parentheses. The robust standard errors are clustered at the city level. ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively. Policy: SCP; lnCarbonEff: carbon emission intensity of enterprises; Size: enterprise scale; ROA: return on assets; DAR: debt-to-asset ratio; Equity: enterprise property; lnFX: fixed assets; lnGDP: gross regional product; lnFI: fixed assets investment at the city level; Adj.: adjusted.

. The results show that, in the samples matched by the economic development level of prefecture-level cities, the influence of SCP on the carbon emission intensity of enterprises remains significantly negative. After adding control variables, the coefficient estimated by PSM-DID is −0.133 (shown in column (8) of Table 4), which is higher than the baseline regression value of −0.078. Hence, the robustness of the results is strong regarding the influence of SCP on the carbon emission intensity of enterprises.

4.5. Heterogeneity Analysis

4.5.1. Degree of Industrialization

Directly correlated with carbon intensity, the production structure of enterprises is related to the industry in which the enterprise operates. Generally, the carbon emission intensity of heavy industrial enterprises should be greater than that of light industrial enterprises. To verify whether the impact of SCP on the carbon emissions of enterprises is heterogeneous due to the degree of industrialization, we divide all the enterprise samples into two groups, heavy-industry and light-industry, according to the degree of industrialization and in accordance with the Chinese industry classification standards. The regression results show that SCP has a more significant effect on the carbon emission intensity of heavy-industry enterprises. This indicates that the high requirements of SCP for environmental protection have forced heavy-industry enterprises to adjust their production structure to promote carbon emission reduction. The regression results are shown in columns (1) and (2) in

Table 5. Regression results of the heterogeneity analysis.

Variable	lnCarbonEff
	Light-Industry	Heavy-Industry	Large-Scale	Small-Scale	Young Enterprise	Old Enterprise
	(1)	(2)	(3)	(4)	(5)	(6)
Policy	−0.157	−0.102 *	−0.152 **	0.018	−0.025	−0.062 *
	(−1.61)	(−1.93)	(−2.49)	(1.14)	(−0.55)	(−1.69)
Control	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Industry fixed effect	Yes	Yes	Yes	Yes	Yes	Yes
Constant	9.241	5.014 **	−3.150	0.454	3.698	−0.658
	(1.26)	(1.96)	(−0.73)	(0.27)	(0.91)	(−0.24)
Observations	887	2933	3961	2972	2436	4414
Adj. R-squared	0.150	0.191	0.121	0.207	0.245	0.095
F Statistics	1.015	1.517	1.425	0.744	0.361	1.045

Note: t-values are reported in parentheses. The robust standard errors are clustered at the city level. ** and * represent significance at the levels of 5% and 10%, respectively. Policy: SCP; lnCarbonEff: carbon emission intensity of enterprises; Control: control variables, including Size, ROA, DAR, Equity, lnFX, lnGDP, and lnFI; Adj.: adjusted.

.

4.5.2. Enterprise Scale and Age

There is also a direct relationship between the carbon emission intensity of an enterprise and its scale and age, so we group the sample according to enterprise size and age. The enterprises with a larger-than-average size, namely, Size > mean (Size), are large-scale enterprises, and those with a smaller-than-average size are called small-scale enterprises. Columns (3) and (4) in Table 5 present the regression results and show that small-scale enterprises have a more significant effect on carbon emission intensity than large-scale enterprises. Moreover, as the inherent production capacity of large-scale enterprises often leads to greater carbon emissions, SCP has generated more digital and information transformation opportunities for such enterprises. This stimulates them to reduce production costs and improve production efficiency.

Similarly, an enterprise with a higher-than-average age, namely, Age > mean (Age), is considered an old enterprise, and that with a lower-than-average age is considered a young enterprise. The regression results are shown in columns (5) and (6) in Table 5. The results indicate that SCP has a more significant effect on the carbon intensity of old enterprises, indicating that enterprises with higher age (or a longer history of development) are more inclined to change their development methods and use more digital production methods to maintain their core competitiveness, under the impact of SCP.

We also categorized a sample of Chinese cities by location but did not find significant geographic heterogeneity in the SCP impact on the carbon emission intensity of enterprises.

4.6. Mechanism Analysis

To analyze the potential mechanism behind SCP’s impact on the carbon emission efficiency of enterprises, we try to explain it from the perspective of digital transformations and innovations. The research concept is shown in Figure 5.

4.6.1. Digital Transformation

The construction of digital infrastructure mandated by SCP provides enterprises with the conditions for digital transformation. Using digital technology, enterprises can combine their own advantages with smart cities to effectively transform production methods and strategies [24]. Enterprises can use this information to serve their production decisions, market-oriented tracking, and production process optimization [40]. Therefore, digital transformation and platform-based enterprise development are inevitable requirements based on strategic iterative upgrades and represent the general direction of the future development of the environmental industry. With the help of digital transformation, enterprises can fully tap the value of data and facilitate product and service innovation and upgrading, thereby generating new digital businesses. Digitalization can improve operational intensity internally and improve customer satisfaction externally; moreover, digital transformation can bring more core competitiveness to enterprises under the derivation of SCP.

More importantly, in an era of a digital economy, enterprises adhering to the traditional operation model will face pressure from competitors’ digital transformation. The traditional production and operation mode of enterprises may significantly limit the development of enterprises [21]. The revolutionary impact of digital transformation leads to the elimination of outdated production models. Consequently, high-pollution and high-emission production models will no longer be the main force for urban economic development driven by SCP.

Based on the above analysis, we find that SCP may further promote the digital transformation of enterprises in the cities they operate in; that is, digital transformation enhances the impact of SCP on the carbon emission intensity of enterprises. To test that impact, we refer to Xue et al. [41] to select the digital-related word frequency to measure the level of digital transformation of enterprises. Formula (5) represents the model used to estimate the impact of digital transformation on enterprises.

$$\ln CarbonEf{f}_{i,t,j,c}={\beta }_0+{\beta }_{1}Polic{y}_{t,c}\ast DTR+{{\displaystyle \sum }}_k{\beta }_k{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(5)

$$DTR=\frac{\mathrm{Digital} \mathrm{related} \mathrm{words} \mathrm{in} \mathrm{enterprise} \mathrm{annual} \mathrm{report}}{\mathrm{Total} \mathrm{words} \mathrm{in} \mathrm{enterprise} \mathrm{annual} \mathrm{report}}\ast 1000$$

(6)

where DTR is the degree of the digital transformation of enterprises, and the calculation formula is shown in Equation (6). Because digital words are rare in the annual report, which leads to an extremely low coefficient, we multiply 1000 by the digital-related words in the enterprise annual report in Equation (6). We measured the degree of enterprises’ digital transformation by using Python to mine and count the digital vocabulary (cloud computing, big data, Internet of Things, etc.) in the annual report of the enterprise, and we obtain the digitized word frequency for each enterprise during the year. The digital words we selected are shown in

Table 6. Types of digitization categories and specific vocabularies.

Type	Digital-Related Words
Artificial intelligence	Business intelligence, image understanding, investment decision assistance systems, intelligent data analysis, intelligent robots, machine learning, deep learning, semantic search, biometrics, face recognition, speech recognition, identity verification, autonomous driving, natural language processing.
Blockchain technology	Digital currency, distributed computing, differential privacy technology, smart financial contracts.
Cloud computing technology	Stream computing, graph computing, memory computing, multi-party secure computing, brain-like computing, green computing, cognitive computing, fusion architecture, billion-level concurrency, EB-level storage, Internet of Things, cyber-physical systems.
Big data technology	Data mining, text mining, data visualization, heterogeneous data, credit investigation, augmented reality, mixed reality, virtual reality.
Digital technology applications	Mobile internet, industrial internet, internet medical, e-commerce, mobile payment, third party payment, NFC payment, smart energy, B2B, B2C, C2B, C2C, O2O, network connection, smart wear, smart agriculture, smart transportation, smart medical care, smart customer service, smart home, smart investment, smart tourism, smart environmental protection, smart grid, smart marketing, digital marketing, unmanned retail, internet finance, digital finance, fintech, financial technology, quantitative finance, open banking.

Note: the original digital-related words are in Chinese, and this table lists their English translation. We should point out that most of the existing research on digital transformation is at the qualitative level, and few studies adopt quantitative methods. Although the indicator used in this study (i.e., $DTR$) does not accurately reflect the degree of the digital transformation of enterprises, it can still reflect enterprises’ perceptions of digital transformation to a certain extent [42]. Therefore, it has certain rationality and reference values.

. We added the interactive items (i.e., DTR*Policy) and their original items (i.e., Policy and DTR) to the model. If the interactive item was significant, the moderating effect of digital transformation was established. The estimated results of this model are shown in column (5) of

Table 7. Regression results of SCP on green innovation and digital transformation.

Variable	Enterprise Level		City Level		Digital Transformation
	RDSSR	PQS	GreenPatent	GreenUtility	lnCarbonEff
	(1)	(2)	(3)	(4)	(5)
Policy	0.225 **	39.657 ***	223.90 8 ***	127.787 ***	−0.198
	(2.22)	(4.76)	(14.44)	(11.13)	(−0.67)
Policy*DTR					−22.449 *
					(1.74)
Control	Yes	Yes	Yes	Yes	Yes
Year fixed effect	Yes	Yes	Yes	Yes	Yes
City fixed effect	Yes	Yes	Yes	Yes	Yes
Enterprise fixed effect	Yes	Yes	Yes	Yes	Yes
Industry fixed effect	Yes	Yes	Yes	Yes	Yes
Constant	12.417	1701.156 **	6725.162 ***	5773.045 ***	0.472
	(0.82)	(2.10)	(5.30)	(4.81)	(−0.08)
Observations	4897	5980	11682	11682	4328
Adj. R-squared	0.873	0.740	0.995	0.990	0.003
F Statistics	4.229	4.375	49.387	35.201	1.18

Note: t-values are reported in parentheses. The robust standard errors are clustered at the city level. ***, **, and * represent significance at the levels of 1%, 5%, and 10%, respectively. The regression results of SCP to green innovations at the enterprise level and city level are shown in column (1) to (2) and column (3) to column (4), respectively. The regression result of the moderating effect of digital transformation between SCP on the carbon emission intensity of enterprises is shown in column (5). Policy: SCP; lnCarbonEff: carbon emission intensity of enterprises; DTR: the degree of the digital transformation of enterprises; RDSSR: R&D spend sum ratio; PQS: the number of patent citations; GreenPatent: the number of green patent applications at the city level; GreenUtility: the number of green utility model applications at the city level; Control: control variables, including Size, ROA, DAR, Equity, lnFX, lnGDP, and lnFI; Adj.: adjusted.

. The coefficient of the interactive item is −22.45, with the same direction and significance of coefficient −0.078 in column (8) of Table 3. Therefore, the digital transformation at the enterprise level enhanced the reducing effect of SCP on the carbon emission intensity of enterprises. Hence, the hypothesis H2 is accepted.

4.6.2. Green Innovation

Another distinctive feature of the digital economy is innovation, which includes technological innovation at the enterprise level and green innovation at the city level. At the enterprise level, the digital transformation generated by the SCP can significantly enhance the enterprise’s operational intensity. This enables the enterprise to achieve greater output performance under the resource boundary of the original research and development (R&D) innovation [43]. Generally, the R&D spend sum reflects the importance that enterprises place on innovation. In addition, the number of patent citations, compared with the number of patents, can reflect the enterprise patent quality and better reflect the level of enterprise innovation.

At the city level, SCP has produced many innovations, especially green innovation patents, which play a significant role in urban environmental improvement and ecological governance [10]. For example, “the digital intelligent operation system for rural sewage treatment terminal facilities” developed by the China State Construction Engineering Group Co., Ltd. is based on the basic network built by the IoT. It uses the cloud platform as a data center and is supplemented by inspections of intelligent equipment, the automatic transformation of on-site sewage treatment terminals, and on-site safety protection. This system represents a case of digital innovation in rural revitalization, the ecological environment, and the intelligent operation of rural pollution control, and it reflects the main characteristics of smart cities.

Based on the above information, referring to the analyses of Hall and Jaffe [44], Zahra and Bogner [45], and Meuleman and De Maeseneire [46], we select the R&D spend sum ratio (RDSSR) and the number of patent citations (PQS) as proxy variables for innovation at the enterprise level. Referring to Xu et al. [10], we select the number of green patent applications (GreenPatent) and the number of green utility model applications at the city level (GreenUtility) as proxy variables for green innovation at the city level. We select green patents because they can better reflect the innovation level of urban environmental protection and ecological governance and are more relevant to SCP. The estimated results are shown in columns (1) to (4) of Table 7.

In Table 7, we can see that the estimated coefficients of SCP at the enterprise level and city level are all significant. At the enterprise level, implementing SCP will bring about a 25.5% incremental increase in the R&D spend sum ratio of enterprises and an incremental increase of 39.66 in the sum of patent quotes. At the city level, implementing SCP will increase the number of cities’ green patents by about 223.91 and the number of green utilities by about 123.79. These all reflect the significant and enormous relationship between SCP and green innovation. Based on this, we can find an influencing path from SCP to green innovation and finally reduce the carbon emission intensity of enterprises. That is, the hypothesis H3 is verified.

$$RDSS{R}_{i,t,j,c}={\varphi }_0^R+{\varphi }_1^{R}Polic{y}_{t,c}+{\sum }_k{\varphi }_k^R{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(7)

$$PQ{S}_{i,t,j,c}={\varphi }_0^P+{\varphi }_1^{P}Polic{y}_{t,c}+{\sum }_k{\varphi }_k^P{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(8)

$$GreenPaten{t}_{t,c}={\phi }_0^P+{\phi }_1^{P}Polic{y}_{t,c}+{\sum }_k{\phi }_k^P{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(9)

$$GreenUtilit{y}_{t,c}={\phi }_0^U+{\phi }_1^{U}Polic{y}_{t,c}+{\sum }_k{\phi }_k^U{X}_k+{\gamma }_i+{\lambda }_t+{\mu }_c+{\nu }_j+{\epsilon }_{i,t,j,c}$$

(10)

5. Discussion

5.1. Implications for Research

We examine China’s SCP and data on the carbon emission intensity of regional enterprises from 2008 to 2015 and adopt the DID method to empirically analyze the impact of China’s SCP on these emissions. The results show that SCP has a significant negative impact; that is, SCP inhibits the carbon emissions of enterprises. Specifically, SCP can reduce the carbon emission intensity of enterprises by about 7.8% in our research samples and periods. However, we should emphasize that the absolute number is only for reference, because the value may change differently based on different calculation methods, samples, and periods. However, we have confidence in this number because the value of 7.8% is not much different from the estimated results obtained by Yu and Zhang [4] and Cui et al. [28].

Our research results are consistent with the conclusion of Qian et al. [5] and extend the conclusion of Yu and Zhang [4] on a micro level. Simultaneously, we find that SCP affects the carbon emission intensity of enterprises by increasing R&D investment, promoting innovation, and enhancing green innovation at the city level. These results provide evidence that SCP can stimulate green total factor productivity [10] and green innovation at the city level [19] and further indicate that SCP can stimulate innovation at the enterprise level. In addition, our heterogeneity analysis finds that the impact of SCP on the carbon emission intensity of enterprises is heterogeneous in terms of industrialization, enterprise size, and enterprise age, whereas the heterogeneity of the urban geographic location is not obvious. Our mechanism analysis finds that digital transformation strengthens the inhibitory effect of SCP on the carbon emission intensity of enterprises. That is, digital transformation stimulated by SCP is an important method to adjust production and operation processes for enterprises. This result provides evidence that smart cities are the large-scale application of digital and information technology, which corresponds to the conclusions of van den Buuse [22].

However, we should note that SCP may have a different impact on carbon or pollution emissions due to the differences in regions and countries. For instance, Yigitcanlar and Kamruzzaman [47], based on the data in the United Kingdom from 2005 to 2013, found that the impact of SCP on the carbon emission levels in a city is not a simply linear relation. Considering the characteristics of the UK, Cavada et al. [48] propose the interdependent triptych philosophy for “smart city” resilience to offer tailored solutions for cities, communities, and individuals. There are many similar cases, which indicates that it is important to focus on the differences in regions or countries and provide suitable SCP to better implement in order to reduce carbon emissions.

5.2. Marginal Implications and Limitations

We believe that our research can inform government SCP policies by providing a new perspective and reference on achieving success. Environmental sustainability is vital for humans’ wellbeing. SCP is one of the pathways to achieving that objective. SCP can significantly reduce the carbon emission intensity of enterprises, increase energy efficiency, make our cities more livable, less polluted, and energy efficient, and, with innovative planning, make cities physically and psychologically healthier for inhabitants.

This study has a few limitations that should be mentioned. First, the research period in this paper is 2008–2015, which is limited by the CNTSD and may not reflect recent trends. However, our results show a significant negative relationship between SCP and the carbon emission intensity of enterprises. We believe that this effect is still present. Second, we use digital word frequency in the annual reports of listed companies to measure the digital transformation of enterprises. In our case, this serves to verify the moderating role of digital transformation between SCP and enterprise carbon efficiency. However, although useful, there are limitations to such an analysis of digital transformation indicators. We hope to analyze the internal impact mechanism of enterprises’ digital transformation using better indicators. Third, we use green patent-related indicators to measure the level of innovation at the city level. Compared with other patents, green patents can reflect smart city characteristics to a certain extent, but not fully. Fourth, we only focus on the direct fossil fuel consumption and exclude the electricity in the carbon emissions. In our research period, electricity is not a huge part of enterprises’ carbon emissions. In recent years, however, it has become a major proportion of enterprises’ carbon emissions, which means that future research that focuses on this field should note the electricity consumption in carbon emissions. Fifth, the most estimated coefficients obtained by PSM-DID only represent a significance level of 0.10. To test the extreme situation, we select the strictest method for the propensity score matching (i.e., we use the 1:1 nearest neighbor method to match the samples), which ensures that the matched samples are more representative. However, the process of matching may eliminate many samples. In this case, the significance of estimated coefficients may increase (from 0.05 to 0.1), but it is only to show that the robustness of our estimated results in the baseline regression is strengthened.

6. Conclusions and Recommendations

The main objective of this study is to verify the causal relationship between SCP and the carbon emission intensity of enterprises by using the DID method. Based on a quasi-natural experiment in China’s SCP from 2008 to 2015, we find that SCP significantly reduces the carbon emission intensity of enterprises, which indicates that hypothesis H1 is verified positively. Furthermore, we find that green innovation and digital transformation are important mechanism pathways, indicating that hypotheses H2 and H3 are verified positively.

Based on the above conclusions and discussions, we conclude with the following suggestions. First, at the city level, we believe that SCP is beneficial to improving the urban environment. Among businesses, however, the benefits are only more pronounced for heavy-industry, large-scale, and older firms. Such enterprises are among the main driving forces of urban economic development. Therefore, when formulating SCP rules, local governments should focus on analyzing the type and structure of their local enterprises, supporting their digital transformation in the creation of solid digital transformation. Second, as innovation is an important factor affecting carbon emission intensity, local governments should emphasize auxiliary policies related to their SCP to promote innovation among local enterprises. Reasonable financial subsidies should also be provided to encourage innovation. Third, the selected empirical proxy variables for digital transformation and innovation enrich the existing literature and provide pointers for research. We conclude that, when governments formulate green policies, they should consider the impacts on smart city construction, digital transformation, and innovation equally, as they are interconnected.