The Effects of the Low-Carbon Pilot City Program on Green Innovation: Evidence from China

Abstract

This study examines the effectiveness of the low-carbon pilot city program in promoting green innovation outcomes in China. Using a time-varying difference-in-differences model based on 277 cities from 2003 to 2019, this study finds that the implementation of the low-carbon pilot city program has a positive and significant impact on city-level green innovation outcomes. The policy effect is heterogeneous across different urban infrastructure characteristics, including geographic location, city scale, factor endowment, carbon emission intensity, and ICT infrastructure. This study provides important insights into the effectiveness of low-carbon policies in promoting green innovation and has important implications for policymakers and practitioners who are interested in promoting sustainable development in emerging economies.

1. Introduction

With the rapid processes of urbanization and industrialization leading to extensive energy consumption, cities have emerged as pivotal actors in low-carbon development and climate change mitigation [1,2]. China, as one of the world’s largest emitters of greenhouse gases, has recognized the significance of reducing the carbon footprint and mitigating the impacts of climate change. The low-carbon pilot city program was launched by China’s National Development and Reform Commission (NDRC) to resolve the intricate interplay between economic development and transitioning towards a low-carbon economy paradigm [3]. Low-carbon policies are designed and implemented to accomplish a spectrum of paramount objectives aimed at improving energy efficiency, applying renewable energy, adjusting industrial structure, and increasing carbon sequestration capacity [4]. The low-carbon pilot city program will bring about the technical improvement of energy conservation and emission reduction, prompting relevant entities to upgrade existing production technologies and develop advanced green technologies that fulfill the requirements of low-carbon development [5]. To ensure the attainment of the greenhouse gas emission control target for 2030, China initiated pilot work on low-carbon provinces, regions, and cities in 2010 [6,7]. This pilot program was subsequently expanded in 2012 and 2017 and currently includes more than 100 cities in China. Low-carbon cities encompass a comprehensive decarbonization of urban development by improving energy efficiency, adjusting energy sources, transitioning from high-carbon to low-carbon industries, and adopting more environmentally friendly resource allocation practices [8,9]. The significant attention given at the national level to the low-carbon pilot city program, along with the progressive expansion of the pilot scope, underscores the significance of the program in fostering green and low-carbon development in China. Examining the implementation effectiveness of this pilot policy can provide valuable insights and references for enhancing the effectiveness of the policy and subsequent construction of low-carbon cities in future pilot initiatives.

While previous studies have examined the relationship between low-carbon policies and environmental performance [10,11,12], few studies have focused specifically on the relationship between low-carbon policies and green innovation outcomes. Low-carbon policies play a significant role in transforming the industrial structures and shaping the productive capacities of cities, which could affect the availability of financial resources, technological resources, and human resources needed for green innovation activities [13,14]. For instance, cities with higher productive capacities allow for increased investment in research and development (R&D), enabling the exploration of green technology solutions to environmental challenges and sustainable development [15]. In particular, this paper provides empirical evidence that the low-carbon pilot city policy is an effective tool for promoting green innovation outcomes in China and identifies the underlying mechanism through which the policy affects green innovation outcomes. This empirical study also highlights the importance of policy implementation in achieving the desired outcomes and provides insights into the heterogeneity of the policy effects across different cities.

Specifically, this paper employs a time-varying difference-in-differences (DID) model based on 277 prefecture-level cities in China from 2003 to 2019 to evaluate the policy effect, which provides empirical evidence on the positive impact of the low-carbon pilot city program on green innovation outcomes. In the context of this study, the low-carbon pilot city program serves as a quasi-natural experiment that calls for a DID approach. Therefore, the time-varying DID model is adopted to account for the variation in the timing of the policy implementation across the treatment group and to estimate the policy effect on green innovation outcomes. By comparing the changes in green innovation outcomes between the treatment group and the control group, this paper is able to identify the causal effect of the low-carbon pilot city program on green innovation outcomes.

The main findings of the paper are as follows. First, the implementation of the low-carbon pilot city program in China has a positive and significant impact on city-level green innovation outcomes. Second, the positive impact of the policy on green innovation outcomes is driven by the actual implementation of the low-carbon pilot city policy in time and space instead of by other unobserved factors. Third, the policy effect is heterogeneous across different city characteristics, such as city types and management styles. For example, regarding city scales, the low-carbon policy improves green innovation outcomes better in large cities than in small cities. As for information and communication technology (ICT) infrastructure, the low-carbon policy has a more pronounced impact on green innovation outcomes for cities with better ICT infrastructure.

The major contributions of this paper are threefold. First, this paper fills an important research gap in the literature, which is the lack of empirical evidence on the causal relationship between low-carbon pilot city programs and green innovation activities in China. Our study adds to the literature by providing empirical evidence on the effectiveness of low-carbon policies in promoting green innovation outcomes. Second, the findings of this study provide important insights into the effectiveness of low-carbon policies in promoting green innovation outcomes by identifying the factors that contribute to the success of these policies. Third, based on the major findings, this paper provides important implications for policymakers and practitioners who are interested in promoting sustainable development and green innovation in China and other emerging economies.

The remainder of the paper is organized as follows: Section 2 reviews the related literature and formulates the research hypothesis; Section 3 describes the methodology and data; Section 4 presents the results, including baseline findings, robustness checks, and mechanism analysis; Section 5 conducts the heterogeneity analysis; Section 6 provides relevant discussions of the results; and, finally, Section 7 concludes.

2. Literature Review and Research Hypothesis

This paper is related to two major strands of literature. The first strand of literature examines the impacts of low-carbon pilot city programs, and the second strand focuses on the characteristics and determinants of green innovation.

The low-carbon pilot city program is a policy initiative launched by the Chinese government to promote low-carbon development and reduce carbon emissions in urban areas [6,7,16]. The program provides financial and policy support to selected cities to implement low-carbon measures and develop advanced environmental technologies [8,9]. The program aims to create a model for sustainable urban development that can be replicated in other cities across China [17]. Upon the implementation of the low-carbon pilot city policy, the local government usually benefits from more fiscal and policy support from governments at higher levels, which adds to its willingness and capability to invest in scientific research [18]. Previous studies have explored the impacts of low-carbon pilot city programs, including industrial structure upgrading [19], enhancement of carbon emission efficiency [20], improvement in environmental performance [11], boosting of technological innovation [21,22], and carbon intensity reduction [10,12]. Nevertheless, limited research has employed the time-varying DID estimation strategy to investigate the impact of policies implemented across multiple periods. It has been argued in the literature that low-carbon pilot cities are meant to bring low-carbon practices to both production and consumption, build an environment-protecting and energy-saving society, and establish a benign and sustainable ecological system [23]. Moreover, the low-carbon pilot city program allows local governments to create customized pathways for low-carbon urban development instead of imposing a standardized top–down action plan to which all cities would be mandated to adhere [24]. Although other developing countries may not have a nationwide low-carbon pilot city program similar to China, several regions tend to pursue a low-carbon development path by employing various policies, including environmental regulations, government subsidies, and carbon emission taxes [25,26,27]. Hence, a thorough and systematic evaluation of the low-carbon pilot city program through an appropriate identification scheme not only helps to understand policy formulation and implementation for low-carbon pilot cities in China, but also provides a useful guideline for low-carbon cities in other emerging economies.

Green innovation encompasses the measures adopted by enterprises to address environmental pollution and mitigate the negative impact of their production activities [28,29]. This form of innovation primarily focuses on key environmental aspects, such as environmental protection, energy conservation, materials reduction, and pollution prevention [30,31]. As an environmentally friendly form of innovation, green innovation plays a crucial role in meeting the requirements of sustainable economic development. Consequently, it has emerged as a significant driver for developing economies to achieve competitive advantages that should not be overlooked [32]. However, importantly, innovation encompasses not only technological advancements but also organizational enhancements. Green innovation, in particular, is characterized by lengthy investment cycles and high risks [33]. In comparison to general innovation, green innovation necessitates a larger pool of capital, internal and external knowledge, and advanced technological capabilities. Consequently, green innovation is often regarded as a delicate balance between environmental sustainability and economic interests [34]. Previous studies have extensively explored the driving forces behind green innovation activities [35,36,37]. Specifically, several studies have investigated the determinants within the context of developing countries, providing valuable insights for sustainable development in emerging economies [38,39,40]. Broadly speaking, the driving factors of green innovation can be classified into two categories. First, market factors, such as market pressure and consumers’ demand for green products, play a crucial role [41,42]. Second, policies and regulations have the potential to effectively foster green technology innovation. Government subsidies and preferential tax policies, for instance, can facilitate green innovation during the transition phase [43].

Combining these two strands of literature, we expect that the low-carbon pilot city program can play an important role in facilitating the development of city-level green innovation activities. Therefore, the research hypothesis of this paper is proposed as follows: the low-carbon pilot city program promotes the city-level green innovation outcome in China.

3. Methodology and Data

3.1. Empirical Model Specification

We use a time-varying DID model [44,45] to test our hypotheses on whether China’s low-carbon pilot city program promotes city-level green innovation outcomes. The low-carbon pilot city programs serve as a quasi-natural experiment that calls for a DID approach [46]. The nature of the program in that its implementation timing varies across the treatment group requires the use of the time-varying DID setup to evaluate the policy effect. Specifically, the benchmark DID model used to examine the effects of the low-carbon pilot city program on green innovation is defined as follows:

$${gpa}_{it}=\alpha +{\beta }_1{DID}_{it}+\mathit{\gamma }{\mathit{X}}^{\prime }+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(1)

where subscripts $i$ and $t$ denote city and year, respectively. ${gpa}_{it}$ represents the green innovation outcomes of city $i$ in year $t$, measured by the number of green patent applications. The term ${DID}_{it}$ is an interaction term with varying treatment timing and is defined as ${Treat}_i\times {Post}_t$, where ${Treat}_i$ is a dummy variable that defines the treatment group and ${Post}_t$ is a dummy variable that refers to the period of policy implementation. Specifically, ${Treat}_i$ equals 1 if city $i$ is selected to join the list of cities designated by any round of the low-carbon pilot city programs and 0 otherwise. ${Post}_t$ equals 1 for the year when any round of the low-carbon pilot city programs takes place and the years afterward and 0 otherwise. In our case, there have been three rounds of low-carbon pilot city programs in China thus far, with 2010, 2012, and 2017 as the initial implementation years. Hence, our treatment group contains treated cities with varying treatment timing. Note that the starting time point of the second-round program was in December 2012, which was near the end of the year. Therefore, we set the starting year for the second-round program to 2013 to account for the possible lags in the effectiveness of the policy. $\mathit{X}$ is a vector of a series of control variables, $X_{it}^k$. Detailed information about the control variables is provided in the next subsection. ${\mu }_i$ is the city-fixed effect that controls all time-invariant city characteristics, and ${\delta }_t$ is the year-fixed effect. ${\epsilon }_{it}$ is the error term. Standard errors are clustered at the city level.

In the above benchmark specification, we are interested in the coefficient ${\beta }_1$, which represents the average treatment effect of the low-carbon pilot city program on the green innovation outcomes for treated cities. We expect this coefficient to be positive and significant, supporting our hypotheses on the influence of the low-carbon pilot city program on green innovation.

We further employ a series of robustness tests, including the parallel trend test, propensity score matching, and placebo tests, all of which are aimed at justifying and validating the utilization of the DID model. The parallel trend test detects the pre-treatment parallel trend that might exist between the treatment group and the control group. The propensity score matching technique is employed to eliminate the potential selection bias problem among the sample groups. The placebo tests are then used to rule out unobserved factors that could lead to spurious results by randomly assigning “fake” treatment timing or treatment group. Our main findings remain robust after performing all these robustness tests, and the related results are reported in Section 4.2.

For the rest of the empirical analysis, we conduct a mediation analysis to examine the underlying mechanism through which the low-carbon pilot city policy influences the city-level green innovation. We also perform the heterogeneity analysis regarding different city characteristics such as the cities’ locations, scales, factor endowments (representing different city types), carbon emission situations, and ICT infrastructure development levels (representing different city management styles). These relevant findings are displayed in Section 5.

3.2. Variables

3.2.1. Dependent Variable

The knowledge stock is commonly used to measure innovation [47,48]. In this research, the dependent variable $gpa$ represents city-level green innovation, which is measured by the number of green patent applications per 10,000 people. Generally, the innovation level can be proxied by R&D expenditure, the number of patent applications or patents granted, the number of new products, etc. [49]. Among these measurements, patent-based measurement is usually preferred because it contains actual and practical technical contents that mark the output from the prior R&D stage and, thus, serves as a good proxy for innovation [50]. By virtue of the patent classification system, patents related to green technological progress can be easily identified from general technology [51]. High-quality data on city-level green patents are also more widely available, which is equally important for our analysis.

Moreover, using the number of green patent applications instead of the number of patents granted can reflect the actual innovation intensity because it is probable that the technical content announced in the patent has already contributed to the overall innovation process before the patent is officially published and granted [52]. The per capita measurement (the number of patents per 10,000 people) also alleviates the issue arising from the scale effects of green patent application, possibly brought by the level of economic and technological development of a city.

To further investigate the potential heterogeneous effects of low-carbon policy on different aspects of innovation, including technical content, skill level, originality, and creativity, we divide green patent applications into two categories: green invention patent applications (gipa) and green utility patent applications (gupa). The policy effects are expected to be stronger for green invention patents because the standards of technical content are higher for invention patents than for patents on utility model improvement [53]. In the next section, we will examine the effects of these differentiated categories of green innovation using the benchmark DID model.

3.2.2. Core Explanatory Variable

The $DID$ term, the implementation of the low-carbon pilot city policy, is the core explanatory variable. It is a dummy variable that is equal to 1 for the year when a city starts to adopt the low-carbon pilot city policy and the years after the adoption and is equal to 0 otherwise. The treatment group consists of 119 cities from 2003 to 2019 that adopted the policy in one of the three years, 2010, 2013, and 2017. Note that we exclude from the treatment group two cities, Haikou and Sanya, which are documented as actually starting to adopt the low-carbon policy in 2014, one year after 2013, the time when the majority of the second group of cities adopted the policy. Our main results remain robust if we include these two cities in the treatment group as adopting the policy in 2014 or even reassign their implementation year to 2013.

3.2.3. Control Variables

To more accurately identify the impact of the low-carbon pilot city policy on innovation, we use a series of variables to control for the time-variant city characteristics that may also affect city-level green innovation. The control variables that we incorporate into this research include the following:

Per capita gross domestic product (pgdp). We use the per capita gross domestic product to measure the economic development level of a city [54]. The economic development of a city is the most fundamental driver of knowledge creation and diffusion; the higher the level of economic development, the more support for green innovation [55]. We construct this variable by the ratio of the city gross domestic product (GDP) to the total city population.

Industrial structure (is). The endowment and, hence, the industrial structure of a city play a key role in determining the city’s preference for technology and industrial organization [56]. It is commonly recognized that secondary industry, including mining, manufacturing, and construction, is more energy-intensive and pollution-intensive as compared to primary and tertiary industries [47]. One can expect that the greener the industrial structure, the more technological progress in green technology there will be. Following the existing literature, we measure the industrial structure of a city by the ratio of the added value of secondary industries to the city GDP [47,54].

Urbanization (urban). Urbanization is a process accompanied by the aggregation of production factors, such as capital and human resources, in urban areas, significantly reducing all kinds of costs regarding production and information exchange [57]. In other words, urbanization induces an agglomeration effect that stimulates innovation [58]. We measure urbanization by the ratio of the urban population to the total city population.

Human capital (hc). Human capital refers to the economic value of a worker’s experience and skills, which is particularly important in driving innovation since innovation mainly involves brain work [59]. We measure the human capital of a city using the ratio of student enrollment in higher education to the total city population.

Financial leverage (fin). Financial leverage in this paper is defined as bank loans against substantial production in a city, which represents the city’s financial development level. We measure the financial leverage using the ratio of total loans from financial institutions to the city GDP.

Information and communications technology (ict). The level of development of the information and communication technology sector in a city contributes in a major way to boosting innovation [60]. The activities and intensity of competition in ICT industries can also generate complementary innovations in non-ICT industries, suggesting the contribution of ICT to innovation outcomes [61,62]. Therefore, we measure the development of ICT industries of a city using the ratio of the number of workers in the ICT sector to the total workers.

Expenditure on science and education (exp). The expenditure on science and education of a city reflects how much importance the municipal government attaches to scientific research and human resources. Such government expenditure is associated with technological progress and economic growth [63]. We measure such recognition of science and education by the proportion of expenditure on science and education to total government expenditure.

3.3. Data Source

This paper uses a sample comprising 277 prefecture-level cities in China from 2003 to 2019 for empirical analysis on the impact of low-carbon pilot city policies on innovation. Note that there are 293 prefecture-level cities in China. Due to data availability, cities with missing data in dependent and explanatory variables are excluded from the analysis. Therefore, the unbalanced panel data consist of 277 prefecture-level cities during the sample period. There are 4559 city–year pair observations in our sample. Information about the number and classification of city-level green patents is obtained from the China National Intellectual Property Administration (CNIPA) and the World Intellectual Property Organization (WIPO). The list of cities designated as low-carbon pilot cities is sorted and managed based on open information provided by China’s NDRC. Other information about city characteristics, including the control variables mentioned above and some variables used in the heterogeneity analysis in the next section, comes from the series of China City Statistical Yearbooks released by the National Bureau of Statistics (NBS).

We collect all data from the aforementioned individual databases and place them in a panel dataset after one-to-one merging based on the city name and year. We thoroughly prepare and clean the data following the steps below: (1) removing duplicate or irrelevant observations as well as observations with obvious errors, such as negative values; (2) eliminating extreme outliers by Winsorizing data at the 1st and 99th percentiles; and (3) supplementing missing values in core variables by linear interpolation or comparing data from different external databases. We show the definitions and the descriptive statistics of all major variables in

Table 1. Descriptive statistics.

Variable	Definition	n	Mean	SD	Min	Max
gpa	The number of green patent applications (per 10,000 people)	4559	0.823	1.843	0.004	12.332
gipa	The number of green invention patent applications (per 10,000 people)	4559	0.388	0.951	0	6.296
gupa	The number of green utility patent applications (per 10,000 people)	4559	0.43	0.9	0	5.972
did	A dummy variable equal to 1 for the year when a city adopts the low-carbon pilot city policy and for the years afterward, while equal to 0 otherwise	4559	0.16	0.367	0	1
pgdp	Per capita gross domestic product of a city (10,000 yuan)	4559	3.797	3.014	0.382	15.308
is	The ratio of the added value of secondary industries to the city GDP (%)	4559	47.896	10.605	0.554	77
urban	The ratio of urban population to the total city population	4559	0.352	0.235	0.061	1
hc	The ratio of student enrollment in higher education to the total city population	4559	0.016	0.022	0.001	0.113
fin	The ratio of total loans from financial institutions to the city GDP	4559	0.862	0.489	0.271	3.018
ict	The ratio of the number of workers in the ICT sector to the total workers (%)	4559	1.188	0.746	0.305	5.498
exp	The ratio of expenditure on science and education to total government expenditure	4559	0.196	0.044	0.092	0.304

Data sources: CNIPA, WIPO, NDRC, and China City Statistical Yearbooks issued by the NBS.

.

4. Results

4.1. Baseline Results

Based on the empirical model specified in the previous section, we analyze the impact of a low-carbon pilot city program on a city’s green innovation level.

Table 2. Baseline results.

Models	(1)	(2)	(3)	(4)	(5)	(6)
Dependent Variable	gpa	gpa	gipa	gipa	gupa	gupa
did	0.858 ***	0.474 ***	0.457 ***	0.260 ***	0.384 ***	0.202 ***
	(0.216)	(0.131)	(0.113)	(0.069)	(0.102)	(0.065)
pgdp		0.495 ***		0.246 ***		0.241 ***
		(0.054)		(0.028)		(0.026)
is		−0.025 ***		−0.012 ***		−0.013 ***
		(0.006)		(0.004)		(0.003)
urban		0.535		0.186		0.408
		(0.753)		(0.367)		(0.416)
hc		9.973		3.987		6.473
		(8.154)		(3.981)		(4.129)
fin		0.354 ***		0.159 **		0.200 ***
		(0.122)		(0.064)		(0.059)
ict		0.434 ***		0.267 ***		0.157 ***
		(0.088)		(0.049)		(0.040)
exp		7.033 ***		3.969 ***		2.831 ***
		(1.452)		(0.818)		(0.664)
cons	0.686 ***	−2.483 ***	0.315 ***	−1.378 ***	0.368 ***	−1.053 ***
	(0.035)	(0.470)	(0.018)	(0.264)	(0.016)	(0.226)
City FE	yes	yes	yes	yes	yes	yes
Year FE	yes	yes	yes	yes	yes	yes
n	4559	4559	4559	4559	4559	4559
Adj. R2	0.640	0.795	0.621	0.778	0.647	0.789

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively.

reports the results of the benchmark DID model featuring 3 different dependent variables regarding innovation, gpa, gipa, and gupa. For all specifications, we control the city fixed effects and year fixed effects. The standard errors are clustered at the city level to account for unobserved correlations within clusters. Columns (1), (3), and (5) show the results of the model estimated without any control variables, while Columns (2), (4), and (6) report the results with a full set of control variables. From Columns (1) and (2), we find that the coefficients of the DID term are significantly positive, which supports our research hypothesis. The implementation of the low-carbon pilot city policy has improved city-level green innovation [21,22]. In particular, the policy is estimated to have induced an increase of 0.474 in the number of green patent applications per 10,000 people, and the increase is significant at the 1% level when the full model is estimated. Looking into the different types of green patent applications [21], we learn from Columns (3)–(6) that the coefficients of the DID term are also significantly positive but smaller than those estimated for overall green patent applications. Based on the full model, the impact of the low-carbon pilot city policy on green invention patent applications (0.260) is stronger than that on green utility patent applications (0.202), as expected. The low-carbon pilot city program has generally improved the chosen cities’ green innovation outcomes, including green invention patent applications and green utility patent applications. For control variables, those with a statistically significant coefficient affect a city’s green innovation outcomes in the expected fashion discussed in the previous section. For example, in Column (2), we find that a city’s green innovation may decrease (−0.025, significant at the 1% level) if the city is heavily invested in the less green secondary industries.

4.2. Robustness Checks

4.2.1. Parallel Trend Test

One vital assumption of the DID method is that the trends of both the treatment group and the control group are parallel over time [64] in terms of green innovation before the implementation of the low-carbon pilot city policy. To test the differences between the treatment group and the control group in each year before the policy is implemented, we set up a model containing the interaction terms of the treatment group dummy, ${Treat}_i$, and some years before and after the policy implementation year, as follows.

$${gpa}_{it}=\alpha +{\sum }_{\theta =-5}^4{\beta }_{\theta }{Treat}_i\times {Time}_{t_0+\theta }+\mathit{\gamma }{\mathit{X}}^{\prime }+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(2)

where ${gpa}_{it}$ is the number of green patent applications in city $i$ in year $t$, and $t_0\in \left\{\mathrm{2010,2013,2017}\right\}$ represents the starting year when the low-carbon pilot city policy is implemented in each treated city with related ${Treat}_i$. ${Time}_{t_0+\theta }$ represents a set of dummy variables that equal 1 when the year is $t_0+\theta$ with $\theta \in \left\{x\in Z|-5\le x\le 4\right\}$ and equal 0 otherwise. All other terms remain consistent with those in the benchmark DID model. Standard errors are clustered at the city level. Note that we do not include the interaction term with variable ${Time}_{t_0-5}$ in the actual regression. It follows the standard procedure of using the first period as a base period to prevent collinearity when conducting a parallel trend test [65].

Figure 1 shows the results of the parallel trend test. The x-axis represents the year relative to the implementation year of the policy (e.g., “−1” means 1 year before the implementation year), and the y-axis represents the estimated coefficients ${\beta }_{\theta }$ with a 95% confidence interval for each interaction term. The results show that before $t_0$, the values of the coefficients for the pre-policy years are small and do not significantly differ from 0 (that is, 0 falls into the coefficients’ confidence intervals), suggesting that there are no significant differences in the trend of applications for green patents between the treatment group and the control group. In contrast, after $t_0$, the values of the coefficients for post-policy years are significantly positive and increasing over the years, suggesting an increasingly significant impact of the low-carbon policy on the city’s green innovation outcomes. Notably, we find that in year $t_0$, the coefficient is not significantly greater than 0, which does not affect our main conclusion and implies a time lag in the enhancement effect of the policy. It is reasonable to assume that the effect takes time to manifest because conventionally, green patent application requires time for preparation and follow-up procedures [66]. Therefore, the parallel trend assumption holds, and our results are robust.

4.2.2. Propensity Score Matching

To add to the robustness of the results, we combine propensity score matching and DID, two methods jointly known as PSM-DID, for additional testing to eliminate the potential selection bias problem. Scholars have proposed the use of propensity score as a measure of distance in individual matching [67]. The general idea of PSM-DID is to match cities in the treatment group with those in the control group based on their propensity scores calculated using a vector of observable variables that may affect cities’ probability of being designated as a low-carbon pilot city. The observable variables in this case are the seven control variables used in the benchmark model, with which we calculate the propensity score using the logit model.

Table 3. Robustness test: PSM-DID results.

Models	(1)	(2)	(3)
Dependent Variable	gpa	gpa	gpa
Methods	DID	PSM-DID 4NN	PSM-DID Kernel
did	0.474 ***	0.335 **	0.467 ***
	(0.131)	(0.133)	(0.133)
pgdp	0.495 ***	0.487 ***	0.491 ***
	(0.054)	(0.056)	(0.055)
is	−0.025 ***	−0.022 ***	−0.026 ***
	(0.006)	(0.006)	(0.006)
urban	0.535	0.013	0.242
	(0.753)	(0.753)	(0.743)
hc	9.973	12.643	9.580
	(8.154)	(8.183)	(8.268)
fin	0.354 ***	0.399 ***	0.358 ***
	(0.122)	(0.119)	(0.123)
ict	0.434 ***	0.298 ***	0.349 ***
	(0.088)	(0.080)	(0.083)
exp	7.033 ***	6.371 ***	6.392 ***
	(1.452)	(1.393)	(1.424)
cons	−2.483 ***	−2.191 ***	−2.079 ***
	(0.470)	(0.400)	(0.433)
City FE	yes	yes	yes
Year FE	yes	yes	yes
n	4559	4049	4528
Adj. R2	0.795	0.764	0.785

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively; (3) 4-nearest-neighbor matching and kernel matching are applied in Column (2) and (3), respectively.

reports the results of the PSM-DID method with 4-nearest-neighbor matching and kernel matching. Column (1) is simply the results of the benchmark DID model with gpa as the outcome variable. Columns (2) and (3) show the results with 4-nearest-neighbor matching and kernel matching, respectively. We find that the coefficients of the DID term are still significantly positive (0.335 at the 5% level and 0.467 at the 1% level) and are close to the baseline result of 0.474, indicating that our main results are robust. The fact that the coefficients of PSM-DID are smaller than the baseline also reveals that the effect of the low-carbon pilot city policy on green innovation outcomes will be overestimated if the treatment group and the control group are not properly balanced.

To evaluate the effectiveness of the PSM-DID method, we perform several tests after matching following common practices [45,68]. First, we conduct a balancing test on the distribution of the covariates between the treatment group and control group after 4-nearest-neighbor matching. Figure 2 displays the reduction in bias of covariates between the two groups after matching. We find that the estimated bias of all covariates declines remarkably to less than 10% after matching, which means that we have obtained a well-balanced sample for our empirical analysis.

Second, we compare the distribution of the propensity scores of the treatment group and the control group to check whether the common support assumption holds [69]. Figure 3 shows the distribution of the propensity scores, highlighting the observations that are not supported. We find that most observations (89% of the full sample) are supported; thus, we believe that the PSM method is valid and the results are credible.

Third, we draw graphs based on the kernel density estimates of the propensity scores of the treatment group and the control group to illustrate the matching efficiency. Figure 4 shows the kernel density results before and after 4-nearest-neighbor matching. We can see that cities from the control group are slightly more concentrated toward the treatment group after matching, which is shown by the mean of propensity scores of the control group (the dashed vertical line) moving closer to that of the treatment group (the solid vertical line). This result is consistent with the loss of observations found in the previous step. In sum, the PSM-DID method that we have employed is effective and further confirms the robustness of our main results.

4.2.3. Placebo Test

Placebo tests are usually used to check the soundness of the research design in event studies by ruling out unobserved factors that could cause the same effect as the focal policy setting [70]. To see whether the results examined are robust, a placebo test is carried out by setting up a counterfactual situation where the assignment of a treatment is hypothetical and false. In our case, we simulate several situations where the low-carbon pilot city program takes place in a counterfactual year or in a counterfactual city. Under the simulated circumstances, we expect that the DID results should not be significant or close to the baseline result if our quasi-natural experiment setting is properly designed. That is, the improvement in city-level green innovation outcomes is sufficiently driven by the actual implementation of the low-carbon pilot city policy in time and space instead of by other unobserved factors.

We perform the placebo test in three different ways. First, we arbitrarily set the time of implementation 5 years earlier than the actual time and yield a “fake” DID term ${Treat}_i\times {Post}_{t-5}$.

Table 4. Robustness test: placebo test results.

Model	(1)
Dependent Variable	gpa
did_fake	0.094
	(0.112)
pgdp	0.357 ***
	(0.035)
is	−0.045 ***
	(0.006)
urban	1.045
	(0.767)
hc	3.238
	(8.461)
fin	0.248 **
	(0.108)
ict	0.422 ***
	(0.098)
exp	7.064 ***
	(1.195)
cons	−0.910 **
	(0.433)
City FE	yes
n	4559
Adj. R2	0.770

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively; (3) the implementation year of the low-carbon pilot city policy is set to be 5 years earlier than the actual time for all three rounds of the policy.

shows the results based on this counterfactual specification. The results show that the coefficient of the DID term is not significant and is much smaller than the baseline result (0.474), which is expected.

Second, we further randomly set the time of implementation 1 to 5 years earlier than the actual time by simulation 2000 times and run the regression for each simulation to obtain estimated coefficients. Figure 5 reports the distribution of the coefficients of the random “fake” DID terms ${Treat}_i\times {Post}_{t-t^{random}}$ for the 2000 placebo tests. Although the mean of the coefficients is not around zero (approximately 0.24 marked by the dashed vertical line), the baseline result (0.474) is a small probability event according to the distribution, suggesting that our main results are not accidental.

Third, instead of randomly setting implementation years, we randomly assign some cities as the treated cities. The “fake” treatment group consists of 117 cities randomly selected from the pool of 277 cities in our sample, strictly according to the ratio 7:82:28, the true proportion of cities in each round of the low-carbon pilot city programs. We simulate the random draw of cities 2000 times and run regressions based on the “fake” DID term ${Random\_Treat}_i\times {Post}_t$. Figure 6 shows the distribution of coefficients in kernel density and p values. We find that the distribution mean is approximately zero, with most coefficients having a p value larger than 0.1 (the 10% significance level indicated by the dashed horizontal line). Compared to the estimated coefficients, the baseline coefficient (indicated by the dashed vertical line) is again demonstrated to be a small probability event, indicating that our main results are still robust.

4.3. Mechanism Analysis

After several robustness tests, our main results still demonstrate that compared to non-pilot cities, the pilot cites have a significantly positive and better performance in green innovation outcomes. To examine the underlying mechanism through which the low-carbon pilot city policy affects city-level green innovation, we specify the following model for a mediation analysis [71]:

$$M_{it}={\alpha }_1+a{DID}_{it}+\mathit{\gamma }{\mathit{X}}^{\prime }+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(3)

$${gpa}_{it}={\alpha }_2+c{DID}_{it}+b{M}_{it}+\mathit{\gamma }{\mathit{X}}^{\prime }+{\mu }_i+{\delta }_t+{\epsilon }_{it}$$

(4)

where $M_{it}$ is the mediating variable for city $i$ in year $t$, and other terms remain consistent with those in Equations (1) and (2).

In our research, the mediating variable $M_{it}$ is characterized by the degree of government intervention, govintv, which is measured by the per capita expenditure on science of a city [72]. Upon the implementation of the low-carbon pilot city policy, the local government usually benefits from more fiscal and policy support from governments at higher levels, which adds to its willingness and capability to invest in scientific research [73]. The pilot city’s deeper participation in scientific research through financial support drives the region’s progress in green and sustainable technologies [74]. Unlike exp, which captures the city’s emphasis on scientific activities, per capita expenditure on science reflects the strength and intensity of the city-led investment in science, rendering a comparable proxy for the degree of government intervention [75].

Table 5. Mechanism analysis results.

Models	(1)	(2)
Dependent Variable	govintv	gpa
did	0.006 ***	0.151
	(0.001)	(0.092)
govintv		48.420 ***
		(4.816)
pgdp	0.006 ***	0.230 ***
	(0.001)	(0.047)
is	−0.000 ***	−0.012 ***
	(0.000)	(0.004)
urban	−0.006	0.824
	(0.009)	(0.533)
hc	0.019	8.458
	(0.117)	(5.875)
fin	0.003 *	0.256 ***
	(0.002)	(0.068)
ict	0.003 ***	0.304 ***
	(0.001)	(0.063)
exp	0.140 ***	0.367
	(0.024)	(0.857)
cons	−0.031 ***	−1.169 ***
	(0.008)	(0.300)
City FE	yes	yes
Year FE	yes	yes
n	4300	4300
Adj. R2	0.806	0.883

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and * indicate statistical significance at the 1% and 10% levels, respectively.

reports the regression results of the mediation analysis. Column (1) shows the results of Equation (3), and the coefficient for the DID term, $a$, is positive and significant at the 1% level. In Column (2), the coefficient for the mediator, $b$, is also positive and significant at the 1% level. We find that in Column (2), the effect of the DID term shrinks (from 0.474 to 0.151) upon the addition of the mediator govintv to the model, suggesting a significant indirect effect and an indirect-only mediation [71]. To test the mediated effect ($a\times b$), we use the bootstrap method [76]. By employing bootstrapping 1000 times, we obtain an indirect effect significant at the 1% level. For brevity, we do not report the bootstrap results, and the results are available upon request. Based on the above results and analysis, we find that the underlying mechanism of government intervention exists, which indicates that after the implementation of the low-carbon pilot city policy, the treated cities significantly improve their investment in scientific research, which leads to more green innovation outcomes.

5. Heterogeneity Analysis

5.1. Heterogeneity in City Types

In our benchmark DID model, heterogeneity in individual prefecture-level cities is controlled by the city-level fixed effect, but the effect of the low-carbon pilot city policy on green innovation may differ in different types of cities. Of the diversified criteria for dividing cities into different groups, we are first interested in some exogenous factors at the macro level that separate cities, such as geographic location, city scale and city factor endowment.

5.1.1. Geographic Location Heterogeneity

To test whether the enhancement effects of the low-carbon pilot city program on green innovation outcomes differ for cities in different geographic locations, we divide cities in the sample into three regions: the eastern region (99 cities), the central region (98 cities), and the western region (80 cities) [48]. Cities from the treatment group and the control group are present in all 3 regions; there are 59, 28, and 32 treated cities in the eastern, central, and western regions, respectively. We employ the benchmark DID model with gpa as the dependent variable on all 3 subsamples separately, and the results are reported in

Table 6. Heterogeneous results by city region.

Models	(1)	(2)	(3)
Dependent Variable	gpa	gpa	gpa
Sample	Eastern	Central	Western
did	0.595 **	0.193	0.316 ***
	(0.230)	(0.179)	(0.113)
pgdp	0.640 ***	0.413 ***	0.129 ***
	(0.069)	(0.092)	(0.041)
is	−0.046 ***	−0.008	−0.000
	(0.014)	(0.006)	(0.004)
urban	−0.555	−0.580	1.160
	(1.370)	(0.916)	(0.888)
hc	18.146	10.980	26.988 ***
	(17.358)	(11.270)	(6.086)
fin	−0.055	0.424 ***	0.518 ***
	(0.287)	(0.142)	(0.108)
ict	0.423 ***	0.348 ***	0.355 ***
	(0.153)	(0.125)	(0.134)
exp	10.539 ***	5.578 ***	4.343 ***
	(2.978)	(2.086)	(0.776)
cons	−2.358 **	−2.162 ***	−2.579 ***
	(1.149)	(0.519)	(0.655)
City FE	yes	yes	yes
Year FE	yes	yes	yes
n	1667	1640	1252
Adj. R2	0.831	0.745	0.786

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively; (3) the dependent variable is gpa and regressed on three subsamples which consist of cities in eastern, central, and western China, respectively.

. We find that the core coefficients for the eastern and western regions are significantly positive, while in the central region, the policy exhibits no significant effect on green patent applications. Specifically, compared to non-pilot cities, pilot cities in eastern China show a 0.595 increase (significant at the 5% level) in the number of green patent applications after the implementation of the policy, slightly higher than the overall increase (0.474). The increase of 0.316 (significant at the 1% level) in western China is smaller. The results suggest that the effects of the low-carbon policy are more likely present in regions where there are high demands for reducing carbon emissions (e.g., eastern China, which is more developed and contributes to more carbon emissions) [77] or regions where there is motivation to achieve green technological progress to follow a green and sustainable development path (e.g., western China, which is conventionally regarded as less developed and advocates an environmentally friendly development pattern) [78]. Therefore, geographic location matters for green innovation, and future low-carbon policies and related climate policies should be targeted at cities following this pattern.

5.1.2. City Scale Heterogeneity

The city scale is another much discussed source of heterogeneity in city-level studies [48]. To explore the city scale effects, we use the city population to separate cities into large cities and small cities. According to the official document issued by China’s State Council in 2014, cities with more than 3 million people are classified as large cities in China. Therefore, we define a city as a large city if its total population is larger than 3 million people in 2010 and others as small cities [21]. We use the data in 2010 because the year 2010 marks the beginning of the low-carbon pilot city program, and the classification of cities will hardly change on a yearly basis given the relatively stable trends observed in the data. We regress gpa on both subsamples using the benchmark DID model, and the results are reported in

Table 7. Heterogeneous results by city scale.

Models	(1)	(2)
Dependent Variable	gpa	gpa
Sample	Small City	Big City
did	0.328	0.510 ***
	(0.206)	(0.158)
pgdp	0.287 ***	0.648 ***
	(0.084)	(0.055)
is	−0.018 *	−0.026 ***
	(0.010)	(0.008)
urban	−1.880 ***	1.069
	(0.664)	(1.123)
hc	25.492	−2.103
	(15.408)	(7.475)
fin	0.371 *	0.350 ***
	(0.208)	(0.129)
ict	0.159	0.480 ***
	(0.161)	(0.092)
exp	8.702 ***	5.525 ***
	(2.183)	(1.775)
cons	−1.177	−2.617 ***
	(0.740)	(0.591)
City FE	yes	yes
Year FE	yes	yes
n	1645	2914
Adj. R2	0.766	0.829

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and * indicate statistical significance at the 1% and 10% levels, respectively; (3) the dependent variable is gpa and is regressed on two subsamples, which consist of small cities and large cities, where the former have populations less than 3 million people and the latter have populations of more than 3 million people.

. We can see from Columns (1) and (2) that the coefficient of the DID term for the subsample of large cities is 0.51, significantly positive at the 1% level and larger than that of small cities (0.328 and not significant) as well as that of the benchmark model. The results suggest that the low-carbon policy improves green innovation outcomes more in large cities than in small cities. The reasons for this phenomenon may include the more developed public administration patterns and the more rigorous regulation and execution in large cities. Note that this finding is contrary to the conclusion reached in existing research, where the effect for small cities is significantly positive and larger than that for large cities [21]. The contrast may arise from the different city samples used in our research and theirs. Our sample includes low-carbon pilot cities in all 3 rounds of the program from 2003 to 2019, while in their research, only the 2nd round (2012) is examined from 2005 to 2016.

5.1.3. Factor Endowment Heterogeneity

To examine whether the impacts of low-carbon pilot city programs on green innovation outcomes vary in cities with heterogeneous factor endowments, we divide cities into resource-based cities and non-resource-based cities according to their natural resource endowments [79,80]. The resource-based cities in China mainly rely on non-renewable resources, such as oil and coal, and, thus, have an employment structure concentrated on extractive industries. Therefore, we calculate the ratio of the number of mine workers to that of total workers for each city in 2010 and divide cities into two groups by the median of the ratio. The regression results are reported in

Table 8. Heterogeneous results by city resource endowment.

Models	(1)	(2)
Dependent Variable	gpa	gpa
Sample	Non-Resource-Based City	Resource-Based City
did	0.589 ***	−0.002
	(0.192)	(0.104)
pgdp	0.629 ***	0.233 ***
	(0.066)	(0.053)
is	−0.052 ***	−0.007
	(0.010)	(0.007)
urban	−0.627	0.976
	(1.006)	(1.059)
hc	−7.209	28.199 ***
	(11.787)	(8.878)
fin	0.111	0.325 ***
	(0.175)	(0.103)
ict	0.442 ***	0.142 **
	(0.108)	(0.063)
exp	9.123 ***	1.869
	(2.150)	(1.201)
cons	−1.261 *	−1.484 ***
	(0.659)	(0.321)
City FE	yes	yes
Year FE	yes	yes
n	2236	2207
Adj. R2	0.840	0.714

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively; (3) the dependent variable is gpa and is regressed on two subsamples, which consist of non-resource-based cities and resource-based cities, where the former have a below-median ratio of mine workers to total workers and the latter have an above-median ratio.

. We can see from Column (1) that for non-resource-based cities, the effect of low-carbon policy on green innovation outcomes is significantly positive and stronger than the baseline result. Column (2) shows that the effect for resource-based cities is even negative and not significant. Intuitively, non-resource-based cities would attach more importance to improving green innovation when provided the chance to be a low-carbon pilot city, possibly because they lack natural resources for long-term economic development. On the other hand, resource-based cities may lack motivation to boost green innovation because they are rich in resources and do not need to enhance resource utilization efficiency. Nevertheless, policymakers should bear in mind the role that a city’s factor endowment plays in promoting green innovation and duly shift the focus to resource-based cities when natural resources begin to be depleted.

5.2. Heterogeneity in City Management

In addition to differences in exogenous factors, including the geographic location, size, and factor endowment of a city, heterogeneity in other dimensions of city characteristics can affect the impacts of the low-carbon pilot city program and even reduce its effectiveness.

5.2.1. Carbon Emission Heterogeneity

The effect of the low-carbon policy on green innovation depends on the city’s carbon emissions given that the main purpose of green innovation is to improve energy utilization efficiency and, eventually, reduce carbon emissions [22]. We divide the cities in our sample into two groups, high carbon emission cities and low carbon emission cities, by the cities’ per capita carbon dioxide emissions in 2010. Cities with per capita carbon dioxide emissions above the median are defined as high carbon emission cities, and vice versa. The data on city-level carbon emissions are retrieved from China’s Carbon Emission Accounts & Datasets (CEADs). We expect the effect of the low-carbon policy to be stronger for cities with high carbon emissions because they have more incentives to make green technological progress to respond to the appeal for carbon emission reduction made by the Chinese government over the past 20 years [81].

Table 9. Heterogeneity results by city carbon emissions.

Models	(1)	(2)
Dependent Variable	gpa	gpa
Sample	Low Carbon Emissions	High Carbon Emissions
did	−0.055	0.636 ***
	(0.072)	(0.219)
pgdp	0.274 ***	0.514 ***
	(0.059)	(0.069)
is	−0.002	−0.033 ***
	(0.003)	(0.012)
urban	0.162	1.451
	(0.257)	(1.368)
hc	8.485	9.221
	(6.012)	(9.110)
fin	0.309 ***	0.096
	(0.100)	(0.210)
ict	−0.007	0.447 ***
	(0.020)	(0.115)
exp	0.173	11.488 ***
	(0.444)	(2.349)
cons	−0.653 ***	−3.405 ***
	(0.210)	(0.875)
City FE	yes	yes
Year FE	yes	yes
n	2242	2238
Adj. R2	0.702	0.814

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** indicates statistical significance at the 1% level; (3) the dependent variable is gpa and regressed on two subsamples, which consist of cities with low carbon emission and high carbon emission, where the former have below-median per capita carbon dioxide emissions and the latter have above-median per capita carbon dioxide emissions.

reports the results based on the two subsamples by carbon emissions. Comparing Columns (1) and (2), we find that the coefficient for low carbon emission cities is negative and not significant, while high carbon emission cities experience an increase of 0.636 (significant at the 1% level) in green patent applications if the low-carbon policy is implemented, which supports our expectation. These results indicate that the efforts made in energy conservation and emission reduction matter for city-level green innovation. Therefore, cities with high carbon emissions should be given due consideration when designating pilot cities or designing other climate policies.

5.2.2. ICT Infrastructure Heterogeneity

A city’s infrastructure, especially the ICT or digital infrastructure in the more modern ages (including 5G internet, Big Data, cloud computing, data hub, IoTs) is believed to be imperative to all economic and social activities and development in the city [82]. Therefore, we test the potentially different effects of the low-carbon pilot city program arising from the heterogeneity in such infrastructure across cities. Note that here, the ICT infrastructure is a fundamental factor of a city that matters in all city operations and that differs substantially from the control variable ict, which represents the strength and competitiveness of a city’s ICT industries. We measure the ICT infrastructure of a city by the ratio of internet users to the city total population in 2010 because the more internet users there are, the more ICT infrastructure is needed [83]. We regard cities with a ratio above the median as having high-quality ICT infrastructure, and vice versa.

Table 10. Heterogeneous results by city ICT infrastructure.

Models	(1)	(2)
Dependent Variable	gpa	gpa
Sample	Low-Quality	High-Quality
did	−0.033	0.595 ***
	(0.051)	(0.206)
pgdp	0.174 ***	0.543 ***
	(0.043)	(0.064)
is	0.002	−0.038 ***
	(0.003)	(0.011)
urban	0.153	0.430
	(0.276)	(1.163)
hc	13.098 **	6.646
	(6.015)	(8.409)
fin	0.254 ***	0.149
	(0.089)	(0.196)
ict	0.020	0.415 ***
	(0.017)	(0.107)
exp	0.895 **	11.721 ***
	(0.381)	(2.337)
cons	−0.795 ***	−2.876 ***
	(0.173)	(0.830)
City FE	yes	yes
Year FE	yes	yes
n	2228	2266
Adj. R2	0.703	0.815

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively; (3) the dependent variable is gpa and regressed on two subsamples, which consist of cities with low-quality ICT infrastructure and high-quality ICT infrastructure, where the former have a below-median ratio of internet users to city total population and the latter have an above-median ratio.

reports the results of the two subsamples. Columns (1) and (2) reveal that cities with high-quality ICT infrastructure are significantly affected by the low-carbon pilot city policy, and green patent applications increase by 0.595 (significant at the 1% level). In contrast, cities with relatively poor ICT infrastructure are not significantly affected by the policy (the coefficient is negative and not significant). Hence, acquiring advanced and integrated ICT infrastructure is conducive to the effectiveness of low-carbon policies on city-level green innovation. Local governments are encouraged to provide firms and society as a whole with better ICT infrastructure to enhance the role of innovation-oriented climate policies.

To further investigate the role of ICT infrastructure, we turn to the smart city initiative in China, a digital infrastructure-based policy, as an alternative proxy to measure the development status of city-level ICT infrastructure. The concept of constructing a “smart city” in China was first introduced by IBM in 2009 and was adopted by the Chinese authorities [84]. Since 2012, various pilot city programs on smart city construction have been launched by the Chinese authorities with focuses including but not limited to housing, environment, industry, and technological innovation, which are all equally important. As a new pattern of city development, smart city construction is well documented to have contributed to altering city governance modes, improving resource allocation and utilization efficiency, and fostering a green and sustainable city-environment relationship [60,85].

Qualitative studies have identified the role played by technological progress and innovation in ICT and digital technologies in developing smart cities through the application of such technologies in the form of urban infrastructure [86,87]. In this sense, we can view a city as possessing a high level of ICT infrastructure if it is designated a “smart city” by a smart city program during the sample period. The data on smart city programs are obtained from the Ministry of Housing and Urban–Rural Development of China. There are 38 cities selected into the smart city programs in our sample, covering only 616 observations. Due to sample limitations, we perform a triple difference regression using an interaction term of the already specified DID term and a dummy variable ${sc}_i$, which equals 1 if the city has undertaken smart city construction and equals 0 otherwise.

Table 11. Heterogeneous results by the smart city program.

Model	(1)
Dependent Variable	gpa
did	0.347 **
	(0.136)
did × sc	0.921 **
	(0.439)
pgdp	0.479 ***
	(0.053)
is	−0.024 ***
	(0.006)
urban	0.505
	(0.770)
hc	9.804
	(8.052)
fin	0.358 ***
	(0.119)
ict	0.422 ***
	(0.089)
exp	7.182 ***
	(1.458)
cons	−2.471 ***
	(0.461)
City FE	yes
Year FE	yes
n	4559
Adj. R2	0.798

Note: (1) Standard errors are clustered at the city level and reported in parentheses; (2) *** and ** indicate statistical significance at the 1% and 5% levels, respectively; (3) the smart city dummy is excluded to prevent collinearity.

reports the regression results. We find that the coefficients of the DID term and the interaction term are both significantly positive. For non-smart cities, the effect of low-carbon policy on green innovation is slightly smaller (0.347) than the baseline result. For smart cities, the implementation of the low-carbon policy induces an increase of 1.268 in green patent applications, nearly 2.7 times as many as the baseline result. These results reveal that the low-carbon policy is more effective if cities are supported by a high level of ICT infrastructure construction, which is demonstrated by the status of a city being incorporated into smart city programs [48]. Therefore, decision-makers can formulate and introduce similar policies to jointly reinforce the effectiveness of climate policies and urbanization policies alike in the future.

6. Discussion

6.1. Result Discussion

Based on the empirical analysis of city-level panel data from 277 prefecture-level cities in China from 2003 to 2019, this study finds that the low-carbon pilot city program plays a crucial role in promoting the development of city-level green innovation activities. The implementation of the low-carbon pilot city program has a positive and significant impact on city-level green innovation outcomes. The results remain robust through a series of robustness tests. Our paper differs from previous studies in two important aspects. On one hand, although the relationship between low-carbon policies and environmental performance has been discussed in the previous literature [10,11,12], the direct causal evidence linking the low-carbon pilot city program and green innovation is still limited in the existing literature. Taking the low-carbon pilot city policy as an exogenous shock, we employ the time-varying DID model to identify the causal effect of the low-carbon pilot city program on green innovation outcomes. The identification strategy allows us to exploit the variations across cities and over time, which can provide convincing evidence of the causal relationship. On the other hand, while the economic and environmental impacts of the low-carbon pilot city program have been investigated in the existing literature [16,17,18,19], the heterogeneity of the policy effect tends to be ignored in these studies. We conduct a comprehensive set of heterogeneity tests based on a variety of city characteristics, including geographic location, city scale, factor endowment, carbon emission intensity, and ICT infrastructure. Specifically, the heterogeneity analysis shows that the policy effect is more pronounced in cities with better ICT infrastructure, suggesting that the low-carbon pilot city program could be more effective in promoting city-level green innovation outcomes if it is accompanied by ICT-related policies such as the smart city program. Therefore, the systematic heterogeneity analysis reveals a better direction for designing and implementing relevant policies aiming at fostering green innovation in emerging economies.

6.2. Policy Implications

The policy implications of this paper are as follows. First, as shown in the empirical analysis, the low-carbon pilot city program is an effective tool for promoting green innovation outcomes in China [21,22]. Policymakers can use this evidence to design and implement similar policies in other cities and regions to foster sustainable development and green innovation. Second, the effectiveness of the low-carbon pilot city policy is driven by the actual implementation of the policy in time and space. It is found that government support through increasing investment in scientific research plays a vital role in transforming the low-carbon policy into desired green innovation outcomes [74]. Policymakers should pay attention to the implementation details of the policy, such as the timing, the location, and the cities’ budgets for scientific research to ensure that the policy is effective in achieving the desired outcomes. Third, this study provides insights into the heterogeneity of the policy effect across different cities. Policymakers should consider the city scale and resource endowment when designing and implementing low-carbon policies and tailor the policies to the specific needs and characteristics of different cities, such as whether the city suffers from high carbon emission [46]. Fourth, this study highlights the importance of ICT infrastructure in promoting sustainable development and green innovation. The effectiveness of the low-carbon pilot city policy is enhanced by a high level of ICT infrastructure construction. Policymakers should invest in the development of ICT infrastructure to support the implementation of low-carbon policies and enhance the effectiveness of such policies [48,84].

6.3. Limitations and Future Directions

We acknowledge several potential limitations of our study. One limitation is that this paper focuses only on the impact of the low-carbon pilot city program on green innovation outcomes and does not consider other potential outcomes, such as carbon emissions reduction or economic development. Another limitation is that our study covers only a limited period and may not capture the long-term effects of the policy. Additionally, this paper focuses on the impact of the policy on the treatment group and does not consider the spillover effects on other cities or regions.

Several future research directions are worth further exploration. First, this study focuses on the effects of the policy on the pilot cities, but it is important to understand the spillover effects of the policy on neighboring cities and regions and how these effects may vary across different spatial scales in future research. Second, this study identifies the actual implementation of the policy as an important mechanism, but further research is needed to understand the specific implementation details that are most effective in promoting green innovation outcomes. Third, this study provides important insights into the effectiveness of low-carbon policies in promoting green innovation outcomes. Further studies are required to fully understand the long-term effects of these policies, as well as the role of other factors, such as institutional quality and social capital, in promoting sustainable development and green innovation.

7. Conclusions

This study provides causal empirical evidence and shows that the low-carbon pilot city program in China has a positive and significant impact on green innovation outcomes, as evidenced by the empirical analysis using a time-varying DID model. This indicates that the positive impact of the policy on green innovation outcomes is driven by the actual implementation of the low-carbon pilot city policy in time and space instead of by some other unobserved factors. This study also finds that the policy effect is heterogeneous across different urban infrastructure characteristics, including geographic location, city scale, factor endowment, carbon emission intensity, and ICT infrastructure. This paper highlights the importance of policy implementation in achieving the desired outcomes and suggests that future research should consider the long-term effects of the policy and the potential spillover effects on other cities or regions. In sum, this paper provides empirical evidence on the effectiveness of the low-carbon pilot city program in promoting green innovation outcomes in China. Policymakers and practitioners can use the findings to design and implement effective low-carbon policies that facilitate sustainable development in emerging economies.