Literature review

Global climate change stands as one of the foremost challenges confronting the contemporary world. The extensive utilization of petroleum and fossil fuels has significantly aggravated the predicaments associated with global warming and environmental deterioration. In recent times, scholars have undertaken investigations into the multifaceted determinants of energy efficiency, with fiscal decentralization emerging as a pivotal factor with discernible implications for matters pertaining to energy and the environment.

Currently, two primary perspectives regarding the influence of fiscal decentralization on energy and environmental matters exist. One perspective posits that fiscal decentralization positively impacts energy and environmental concerns. Early theories on fiscal decentralization argued that, in the context of residents “voting with their feet,” a decentralized governance system promotes the development of a “benevolent government.” This encourages local governments to provide public goods (Tieout Reference Tiebout1956), subsequently improving the ecological environment within their jurisdiction. Furthermore, compared to the central government, local governments inherently possess information advantages (Afonso et al. Reference Afonso, Schuknecht and Tanzi2005), enabling them to gain a deeper understanding of local environmental and energy conditions and to implement targeted environmental management measures (Oates and Schwab Reference Oates and Schwab1988). Additionally, the “environmental federalism” perspective contends that decentralization fosters competition, motivating governments to invest in research and innovation, consequently enhancing regional energy and environmental performance (Hottenrott and Rexhauser Reference Hottenrott and Rexhauser2015; Cheng et al. Reference Cheng, Ren, Dong, Dong and Wang2021).

Conversely, another perspective argues that fiscal decentralization can yield adverse effects on local energy and environmental conditions. Influenced by a “GDP worship” mentality, local governments may partake in a “race to the bottom” in their pursuit of economic growth. They do this by lowering environmental regulatory standards to attract energy-intensive and highly polluting enterprises (Oates and Schwab Reference Oates and Schwab1988; Cui and Liu Reference Cui and Liu2010; Sjoberg and Xu Reference Sjoberg and Xu2018). Consequently, this leads to increased government rent-seeking behavior and a higher tolerance for corporate pollution (Fredriksson and Millimeter Reference Fredriksson and Millimet2002; Lopezópez and Mitra Reference Lopezópez and Mitra2000; Boskovic Reference Boskovic2016). Consequently, this creates a “pollution paradise” that promotes economic prosperity at the expense of the environment, ultimately diminishing energy efficiency within their jurisdiction.

Numerous empirical studies have explored the effects of fiscal decentralization in China on energy and carbon emissions. However, these studies have yielded varying and at times contradictory conclusions. Some investigations propose that fiscal decentralization has the potential to significantly enhance regional energy efficiency and mitigate carbon emissions (Bi et al. Reference Bi, Wang and Fu2022). Conversely, an alternative perspective posits that fiscal decentralization in China can undermine energy efficiency and hinder efforts to reduce carbon emissions and promote sustainable development (Yang et al. Reference Yang, Tang and Zhang2020; Lin and Zhou Reference Lin and Zhou2021). Additionally, a body of research has indicated that the relationship between fiscal decentralization and carbon emissions in China is nonlinear, influenced by the degree of fiscal decentralization (Hao et al. Reference Hao, Chen, Liao and Wei2020; Lv et al. Reference Lv, Pang and Doan2022). Consequently, the influence of fiscal decentralization on energy-related issues in China remains a subject of ongoing and active debate within the scholarly community.

As illustrated in Table 1, a substantial body of recent literature delves into the realm of fiscal decentralization and its interplay with energy-related concerns, rendering valuable insights for our research. It is crucial to acknowledge, however, that this existing corpus of research carries notable constraints. Primarily, many contemporary studies rely on conventional fiscal expenditure or revenue decentralization indicators, typically framed as the ratio of provincial revenues (or expenditure) to central revenues (or expenditure). Regrettably, these metrics primarily reflect the fiscal capacities of local governments, rather than offering a true reflection of the level of decentralization (Lv et al. Reference Lv, Ma and Hu2021). Furthermore, the application of these traditional indicators to gauge fiscal decentralization falls short in addressing endogeneity concerns and is susceptible to estimation biases (Xu et al. Reference Xu, Cai, Zhou and Qi2023). Additionally, the predominant focus of existing literature on provincial-level governments has resulted in a relatively limited sample size, consequently introducing substantial estimation biases. Therefore, this study adopts a fresh approach, scrutinizing fiscal decentralization through the lens of tax sharing and treating the VAT revenue-sharing reform as a quasi-natural experiment. Our objective is to evaluate the impact of fiscal decentralization on the energy intensity of prefecture-level cities, thereby addressing certain limitations observed in prior research.

Table 1. Selection of indicators and key findings in previous literature

Theoretical background

The tax revenue-sharing system in China pertains to a fiscal framework wherein tax revenues are apportioned between the central government and local governments. Under this system, tax revenue is bifurcated, with one segment directed to the central government and the other portion designated for local governments. This mechanism serves the dual purpose of maintaining fiscal equilibrium between the central government and local governments while stimulating local authorities to actively engage in greater economic endeavors and tax revenue generation, thereby facilitating local economic growth. The specific proportions and regulations governing tax revenue sharing can exhibit variations based on the type of tax, geographical region, and the prevailing period, and these determinations rest with the central government. Consequently, in contrast to conventional metrics employed for gauging fiscal decentralization, the tax revenue-sharing indicator proves more adept at characterizing the fiscal distribution dynamics between government tiers (Mao et al. Reference Mao, Lv and Chen2018; Zhou and Wu Reference Zhou and Wu2015). This system assumes a pivotal role in the realm of fiscal management in China and exerts a substantial influence on the fiscal strategies and conduct of local governments.

Within the purview of the tax revenue-sharing system, an array of tax categories is encompassed, with the most salient one being the VAT. VAT is the largest tax category in China. According to statistics, in 2021, China’s domestic VAT revenue amounted to 63,519 billion yuan, representing a share of 36.8% of the total tax revenue. Since the inception of the tax revenue-sharing reform in 1994, the ratio governing the sharing of VAT revenue between the central government and local governments has consistently stood at 75:25 and has endured over an extended timeframe. In 2016, China effectively transitioned from business tax to VAT, concomitantly revising the distribution ratio for VAT revenue between the central government and local governments to an equitable 50:50 apportionment. This reform instigated heightened financial incentives for local governments, thereby enhancing their impetus to bolster tax collection efforts and propel economic development. However, it may also be concomitant with an escalation in regional energy intensity.

The impact of VAT revenue-sharing reform on energy intensity manifests through two primary avenues. First, the elevation of revenue-sharing ratios serves to augment financial incentives for local governments, catalyzing their active promotion of business development to bolster tax revenues. Local governments often extend financial subsidies, tax incentives, and increased credit access to bolster business growth (Xu et al. Reference Xu, Cai, Zhou and Qi2023). In response to these incentives, businesses invest in expansion, leading to a corresponding uptick in energy consumption. The acquisition of new production equipment, facility expansions, and daily production operations necessitates a greater energy input, thereby amplifying energy intensity. Second, the escalation of revenue-sharing ratios can compel local governments to relax environmental regulatory standards as a means to attract investments from high-energy-consuming enterprises in their pursuit of tax revenue (Lin and Zhou Reference Lin and Zhou2021). China’s decentralized environmental management system grants local governments’ decision-making and enforcement authority over local environmental matters. Consequently, they possess the autonomy to adjust environmental control measures for businesses independently (Ran et al. Reference Ran, Zhang and Hao2020). The relaxation of environmental regulations often results in the proliferation of high-energy-consuming enterprises, thereby elevating regional energy intensity.

Furthermore, the impact of VAT revenue-sharing reform exhibits variation among different prefecture-level cities due to disparities in the proportion of VAT revenue in their government income. Regions with a higher share of VAT revenue experience more pronounced effects, while those with a lower share encounter comparatively modest repercussions. This divergence presents an opportune avenue for discerning the influence of VAT revenue sharing on energy intensity.

Model specification

This article employs a standard double difference model to analyze how fiscal decentralization incentives, marked by the VAT revenue-sharing reform, affect urban energy intensity. The model is constructed as follows:

(1) $$E{I_{i,t}} = {\theta _{\rm i}} + {\eta _t} + {\lambda} \times {\rm{Treate}}{{\rm{d}}_i} \times pos{t_t} + \rho \times Contro{l_{i,t}} + {\varepsilon _{{\rm{i}},t}}$$

In this context, EI _i,t represents the energy intensity of city i in year t. Treated _i is a binary variable used to categorize cities into treatment and control groups, with 0 for the control group and 1 for the treatment group. Post _t represents dummy variables for the pre and post-reform years, taking the value of 1 in and after 2016, and 0 otherwise. The interaction term, Treated _i ×Post _t , is the policy treatment effect variable, capturing the impact of the VAT revenue-sharing reform on urban energy intensity. The parameter, λ, is the estimated policy effect parameter of interest in this study. Control _i,t serves as control variables, θ _i represents city-specific fixed effects, and η _i represents time-specific fixed effects.

An essential prerequisite for the effectiveness of the difference-in-differences (DID) model is the requirement of parallel trends. In the context of this article, this implies that before the VAT revenue-sharing reform, the treatment group and the control group should exhibit similar time trends in energy intensity. We construct the following model to examine this:

(2) $${\rm{E}}{{\rm{I}}_{i,t}} = {\theta _i} + {\eta _t} + \sum\limits_{t=2007}^{2020} {{\alpha _t} \times T{\rm{rea}}te{d_i} \times {D_t}} + \rho \times Contro{l_{i,t}} + {\varepsilon _{i,t}}$$

We use the first year of the sample period (i.e., 2006) as the base year, D _t represents time-point dummy variables (with a value of 1 for the respective year and 0 otherwise). Treated _i ×D _t represents the interaction term between the policy variable and the time-point dummy variable. α _t denotes the estimated coefficients for the years 2007–2020, while other variables follow Equation (1). The parameter of primary interest in this section is α _t , which signifies whether there is a significant difference in energy intensity between the treatment and control groups in year t. If α _t is not significant before the policy implementation but becomes significant after the policy, it suggests that the difference in energy intensity between the treatment and control groups is primarily due to the reform.

Data source

This paper selects data from 282 Chinese cities for the years 2006–2020. The data primarily come from various editions of the “China City Statistical Yearbook,” “China Energy Statistical Yearbook,” and “China Taxation Yearbook,” as well as individual city statistical yearbooks and city development statistical bulletins.

The indicators used from the “China City Statistical Yearbook (2007–2021)” include (1) regional GDP, (2) total regional population, (3) total import and export trade volume, (4) the number of regular undergraduate and postgraduate students, (5) the number of nonagricultural population, (6) the value added in the secondary sector, (7) year-end financial institution deposit balance and year-end financial institution loan balance, (8) fiscal scientific expenditure, and (9) annual electricity consumption. From the “China Energy Statistical Yearbook (2007–2021),” the following indicators are used: (1) total natural gas supply and (2) total supply of liquefied petroleum gas. Indicators from the “China Taxation Yearbook (2007–2021)” include (1) government tax revenue and (2) government VAT revenue. The city statistical yearbooks and city development statistical bulletins are primarily utilized to gather information on road mileage between various cities.

Variable description

Dependent variable

The dependent variable in this study is energy intensity, defined as the ratio of energy consumption to GDP. In consideration of data availability, we opted to analyze three primary energy sources: natural gas, liquefied petroleum gas (LPG), and electricity. Data on the consumption of these energy sources were extracted from the China City Statistical Yearbook. Subsequently, we calculated the energy consumption in terms of standard coal using the IPCC 2006 conversion factors. The conversion factor for natural gas was determined to be 1.33 kilograms per cubic meter, while LPG exhibited a conversion factor of 1.7143 kilograms per kilogram, and electricity was standardized at 0.1229 kilograms per 10,000 kilowatt-hours. Finally, we computed the total energy consumption for each city by summing the standardized energy consumption values for natural gas, LPG, and electricity.

Figure 1 presents the average energy intensity across different Chinese cities spanning the period from 2006 to 2020. The graphical representation highlights notable disparities in energy intensity observed among these cities. Regions characterized by higher energy intensity levels include the northern cities, where heavy industry is prevalent, certain cities within the coal and mineral-rich central region, and urban centers within the densely populated and highly urbanized Yangtze River Delta and Pearl River Delta areas. In summary, a substantial degree of variation in energy intensity is evident across cities throughout China.

Figure 1. Average energy intensity by cities (2006–2020).

Control variables

Referring to Chen et al. (Reference Chen, Du and Huang2019), Guang et al. (Reference Guang, He, Wen and Sharp2019), and Ma et al. (Reference Ma, Zhao, Jia, Wang and Zhang2022), the following variables are chosen to act as the control variables:

1. (1) Per capita GDP (PGDP): Per capita GDP refers to the logarithmic representation of the percentage of GDP to the total population of residents.

2. (2) Technology expenditure (TECH): Technology expenditure is measured by the proportion of technology spending to the regional GDP.

3. (3) Urbanization (URBAN): The level of urbanization is evaluated by the proportion of urban residents to the country’s overall population.

4. (4) Human capital (HC): HC is represented in this article by the ratio of the total number of regular undergraduate and graduate students to the total population at the end of the year.

5. (5) Openness (OPEN): Openness is measured by the ratio of the sum of imports and exports to total GDP, the amount of imports and exports multiplied by the RMB/US dollar exchange rate for that year.

6. (6) Transport infrastructure (TRANS): Measured by the logarithm of the road mileage.

7. (7) Financial development (FINANCE): Financial development is measured by the proportion of the year-end balance of financial institutions’ loans and deposits to the regional GDP.

8. (8) Industrial structure (INDUS): We measure the industrial structure of a region with the value-added of secondary industry to GDP.

Table 2 provides a comprehensive overview of the summary statistics for the pivotal variables employed in the regression analysis, encompassing the mean values and disparities within the “treatment group” and “control group” subsamples. Additionally, we assess the legitimacy of the incorporated control variables. Significantly divergent results emerge from mean difference tests conducted on an array of indicators between the treatment and control groups, thus substantiating the rationale for incorporating these variables into the econometric model.

Table 2. Descriptive analysis

Note: ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively. The entire sample includes the total sample size, average values, standard deviations, maximum values, and minimum values for each variable. Subsamples present sample sizes, mean values, and mean differences for the treatment group and the control group, respectively.

Baseline results

The outcomes of the baseline regression analysis are presented in Table 3. In the initial column (1), the coefficient associated with the interaction term (Treated×Post) is 0.028, demonstrating statistical significance at the 1% level. In the subsequent column (2), accounting for additional variables, the coefficient pertaining to the interaction term registers at 0.019, retaining statistical significance at the 1% level. This observation suggests a favorable influence on urban energy intensity resulting from the fiscal decentralization incentive introduced by the VAT sharing reform.

Table 3. Baseline results

Note: ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively.

Parallel trends test

The investigation entails estimating coefficients and 95% confidence intervals for the interaction term, as per Formula (2), and presenting them graphically (refer to Figure 2). The outcomes reveal an absence of a substantial disparity in energy intensity changes between the treatment and control groups preceding the initiation of the VAT sharing reform. This observation implies the absence of a noteworthy pre-existing trend divergence in energy intensity between the treatment and control group cities prior to the reform, thereby supporting the validity of the parallel trends test. Furthermore, the interaction term coefficient lacks significance in the year of reform implementation but attains significance in the subsequent year. This suggests a distinct and enduring positive impact of the VAT sharing reform on energy intensity.

Figure 2. Parallel trends test.

Robustness checks

Placebo test

This study employs a placebo test methodology by manipulating the timing of events. In order to prevent ascribing changes in energy intensity between the treatment and control groups to temporal variations, the investigation conducts additional regressions, advancing the policy shock points to 2011 and 2013. The policy dummy variables are denoted as post2011 and post2013, respectively. The results presented in Table 4 indicate that the estimated coefficients of the interaction term (Treated×Post2011 and Treated×Post2013) do not achieve significance in the 10% level. This implies that the positive effect of the VAT sharing reform on energy intensity was not observable before the reform’s implementation.

Table 4. Time placebo test: Altering the timing of reform implementation

Additionally, in order to mitigate the possibility of policy impact effects arising from random occurrences, this study adopts a methodology inspired by Moon (Reference Moon2022). It involves the random generation of a treatment group to derive coefficient estimates for the reform’s impact on “pseudo-treated” urban energy intensity. This process is iterated 500 times, resulting in 500 estimated coefficients and their corresponding p-values. As illustrated in Figure 3, the distribution of estimated coefficients from these random simulations converges around 0, while the baseline regression coefficient (0.019) lies completely outside this distribution. This underscores that the empirical findings of this study are not contingent on randomness or incidental factors.

Figure 3. Distribution of estimated coefficients from randomly generated treatment groups.

Adjusting the setting standards for the treatment and control groups

In the baseline regression, this study categorizes cities based on their local governments’ reliance on VAT in the year preceding the VAT sharing reform, designating the top and bottom halves as the treatment and control groups, respectively. To enhance result robustness, we redefine the treatment and control groups to include cities in the top and bottom thirds of local government reliance on VAT. Given that the newly defined treatment and control groups display a more substantial disparity in local government reliance on VAT prior to the reform, the coefficient of the interaction term in the DID model is expected to be larger (Campello and Larrain Reference Campello and Larrain2016). As illustrated in the initial column of Table 5, the estimated coefficient of the VAT sharing reform variable (Treated×Post) is 0.028, surpassing the baseline regression’s 0.019 and significantly positive at the 1% level. This suggests that the VAT sharing reform has a significant impact on increasing urban energy intensity (see Column [1] of Table 5).

Table 5. Robustness checks

Note: ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively.

Heterogeneity of geographical location

China’s extensive territory showcases substantial disparities across its regions concerning resource endowment, levels of technological development, and managerial proficiency. In the realm of energy consumption, a noteworthy divergence is evident among the eastern, central, and western regions concerning energy acquisition costs, methods of energy utilization, and various other factors. This dissimilarity gives rise to distinct levels of energy intensity within the eastern, central, and western regions. The eastern region includes Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, and Hainan; the central region includes Shanxi, Inner Mongolia, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, Hunan, and Guangxi; and the western region includes Sichuan, Guizhou, Yunnan, Xizang, Shaanxi, Gansu, Ningxia, Qinghai and Xinjiang. Consequently, this study stratifies the entire sample based on geographical location, specifically into eastern, central, and western regions, and conducts a comprehensive analysis of the regional heterogeneity characteristics associated with the impact of VAT reform on energy intensity.

The findings, as presented in the first three columns of Table 6, reveal that the coefficient of the interaction term (Treated×Post) is not statistically significant in the eastern region, whereas in the central and western regions, this coefficient is significant. This implies that the influence of the VAT reform is more prominent in the central and western regions. The observed discrepancy can be attributed to the advanced economic development in the eastern region, characterized by a preference for cleaner industrial growth, along with substantial local fiscal resources. As a result, the impact of the VAT apportionment reform on energy intensity remains inconspicuous in the eastern region. In contrast, the central and western regions, marked by less developed economies and heightened fiscal pressures compared to the eastern region, exhibit a substantial effect on energy intensity. This is driven by government efforts, incentivized by tax reforms, to boost tax revenue through the introduction of a significant number of energy-intensive industries.

Table 6. The results of heterogeneous effects

Note: ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively.

Heterogeneity of energy dependence

Due to differences in industrial structure, economic development levels, and other factors, cities vary in their degree of energy dependence. Some cities have developed by relying on energy-intensive industries, such as heavy industry and chemical manufacturing, while others have a more diversified industrial structure that includes a greater share of services and high-tech industries. To examine whether the impact of the VAT sharing reform varies across cities with different levels of energy dependence, this paper divides the sample into two groups: high-energy-consuming cities and low-energy-consuming cities, based on the differences in energy consumption per unit of GDP before the VAT sharing reform. The median of the average energy consumption per unit of GDP for the sample cities from 2006 to 2015 is used as the dividing point. Cities with values above this median are classified as the high-energy-consuming cities group, while those with values below are classified as the low-energy-consuming cities group. As shown in the last two columns of Table 6, the impact of the VAT sharing reform on energy intensity is more pronounced in the high-energy-consuming cities group, with the coefficient of the interaction term (Treated×Post) being higher in these cities compared to the low-energy-consuming group. This may be because, compared to high-energy-dependent cities, the low-energy-consuming cities tend to have a more diversified industrial structure and higher energy efficiency, resulting in a relatively smaller impact of the VAT reform on their energy intensity.

Government competition

In the face of both fiscal constraints and promotion incentives, local government competition serves as an enduring driver for rapid economic growth within the region. The persistent introduction of high-energy-consuming industries, spurred by governmental competition, further amplifies urban energy intensity. Consequently, the augmentation of VAT sharing intensifies the endeavors of local governments in business attraction due to heightened governmental competition. This posits that governmental competition may magnify the impact of VAT sharing reforms on energy intensity.

To empirically examine this proposition, the study assesses the moderating effect of governmental competition on the correlation between VAT sharing reforms and energy intensity. The degree of governmental competition is gauged by the logarithm of per capita actual utilization of foreign capital in each city. The regression analysis includes the interaction term (Treated×Post×GC), symbolizing the reform’s interaction with governmental competition. The results, presented in the initial column of Table 7, reveal a regression coefficient of 0.010 for the interaction term, signifying statistical significance at the 1% level. This implies that, in regions characterized by heightened local governmental competition, the implementation of VAT sharing reforms corresponds to a more substantial increase in energy intensity.

Table 7. Moderating effects

Note: ***, **, and *indicate significance at the 1%, 5%, and 10% levels, respectively.

Environmental regulation

Environmental regulations can concurrently diminish energy consumption and foster economic growth, facilitating a sustainable, green transformation of economic development (Murshed et al. Reference Murshed, Rahman and Alam2021). In locales characterized by more stringent environmental regulations, diverse policies may have been enacted to curtail energy intensity. Consequently, when juxtaposed with regions featuring less stringent regulations, areas that enforce more rigorous environmental control measures experience a mitigated incentive effect regarding tax sharing linked to energy intensity.

To empirically assess this hypothesis, this study introduces a variable representing the intensity of environmental regulation (ER) at the prefecture-level city government, derived from the frequency of “environmental protection”-related terms in government work reports. There are 15 environmentally related terms: low-carbon, environmental protection, air, green, PM2.5, chemical oxygen demand, carbon dioxide, PM10, ecological, pollution discharge, emission reduction, pollution, environmental protection, sulfur dioxide, and energy consumption. The calculation involves determining the proportion of environmental vocabulary relative to the total vocabulary in the report. Regression results, presented in column (2) of Table 7, demonstrate a significantly negative coefficient associated with the interaction term between environmental regulations and the reform of VAT revenue sharing (Treated×Post×ER) at the 5% significance level. This suggests that, in regions with more rigorous environmental regulations, the influence of VAT sharing on energy intensity is attenuated.

Technological innovation

Existing literature suggests that innovation has the potential to decrease energy intensity, phase out obsolete production capacity, and facilitate industrial transformation and upgrading (Cheng et al. Reference Cheng, Ren, Dong, Dong and Wang2021; Lin and Zhou Reference Lin and Zhou2021). Consequently, in regions characterized by robust innovation capabilities, the anticipated effect of VAT revenue-sharing reform on energy intensity is expected to be mitigated.

To empirically scrutinize this hypothesis, this study quantifies the degree of technological innovation in each city by taking the logarithm of the number of patent applications and subsequently incorporating this measure into the regression analysis model. The regression outcomes presented in column (3) of Table 7 reveal that the coefficient associated with the interaction term (Treated×Post×TEC) is significantly negative at the 1% significance level. This suggests that with an escalation in the level of technological innovation within cities, the influence of VAT revenue-sharing reform on energy consumption intensity diminishes.