I. Introduction
Since the reforms and the opening up of China’s economy, the country has maintained rapid growth and recorded worldclass achievements. However, with China’s rapid development, the inefficient use of energy has led to increasingly severe phenomena, such as climate warming and environmental pollution (Xia & Xu, 2020). According to Chinese emission accounts and datasets (https://www.ceads.net.cn/), from 2000 to 2018, China’s carbon dioxide emissions increased from 906 tons to 2,667 tons. In this context, many studies have conducted research on energy efficiency measurements and influencing factors. There are currently three main methods for the measurement of energy efficiency in academia. First, the most common measurement is the ratio of economic output to energy input (Lee et al., 2020; Yu, 2017). Second, energy efficiency measurement models that do not consider undesired output, such as the models of Charnes et al. (1978, henceforth CCR) and Banker et al. (1984, henceforth BCC), are also widely used because of their relative simplicity. Third, in addition to considering the expected output, measurement models of undesired output under resource constraints are also considered, because these models consider resource constraints and the measurement results are more in line with reality (Tone, 2001). This paper employs the third method to measure China’s energy efficiency.
This paper examines the effect of marketoriented reforms (MRs) on energy efficiency and the mechanism of action based on the panel data of 30 Chinese provinces (including municipalities directly under the central government, but excluding Tibet, Hong Kong, Macao, and Taiwan) from 2010 to 2018, using a slackbased measure (SBM) model with fixed effects estimation. The contributions of this paper are summarized as follows. First, this research incorporates MRs, industrial structure upgrading, and energy efficiency into the research framework, which broadens the research field of energy economics to a certain extent. Second, we use the superSBM model, which takes into account undesired output, to measure energy efficiency, which is more realistic.
Scholars have found that the influencing factors of energy efficiency are very complex, including financial development (Chiu & Lee, 2020), industrial structure (Zhou et al., 2013), technological progress (FisherVanden et al., 2004), environmental policies (Bongers, 2020; Gorus, 2020), and market factors (Liu & Lee, 2020; Wang et al., 2017). However, research on market factors is rare, and most of it is focused on economic growth and environmental pollution (Hsieh & Klenow, 2007; Yin et al., 2018).
Theoretically, MR has a possible impact on energy efficiency, and the impact paths include three main aspects. First, MR can reduce the dependence effect of the extensive growth model. MR also can eliminate companies in the market that use resources inefficiently, as well as promote research and development (R&D) and investment in lowinput, lowenergy production technologies (Fan et al., 2007). In other words, marketization could unlock the crude growth model by promoting the upgrading of industrial structure, which, in turn, will improve energy efficiency.
Second, MR can reduce the rentseeking behavior of enterprises. Due to China’s natural resources being owned by the state, local officials have absolute power to allocate resources. Since the supervision system of Chinese officials is not perfect, the situation is prone to enterprises’ rentseeking behavior, which means that, even if some enterprises have low productivity, they can still obtain production factors at prices below market through rentseeking, and the MR of factors could alleviate this situation.
Finally, MR can alleviate the phenomenon of market segmentation. Due to the GDP Championship, local officials prioritize the allocation of production factors to enterprises in their jurisdictions for political performance and discriminate against enterprises in other regions in terms of price (Dai & Cheng, 2016). However, do the above three impact pathways really exist? What are the mechanisms of action? These questions need to be verified by scientific empirical methods.
The remainder of the paper is organized as follows. Section II introduces the model construction, indicator selection, and data sources. Section III presents the empirical results. Section IV draws the conclusions of the study.
II. Research Design
A. Models
A1. Measure of Energy Efficiency
Suppose there are n decision making units, m input factors, q desired outputs, and p undesired outputs. Define \(X=\left[x_{1}, \cdots, x_{n}\right]\in R_{m\times n}\), \(Y^{a}=\left[y_{1}^{a}, \cdots, y_{n}^{a}\right] \in R_{q \times n}\), and \(Y^{b}=\left[y_{1}^{b}, \cdots, y_{n}^{b}\right] \in R_{p \times n}\). Then the set of production possibilities can be state as
\[\begin{align}p\left(x_{o}, y_{o}^{a}, y_{o}^{b}\right)=&\left\{\left(\bar{x}, \bar{y}^{a}, \bar{y}^{b}\right) \mid \bar{x}\geq \sum_{j=1}^{n} \lambda_{j} x_{j}, \right.\\ &\bar{y}^{a} \left.\leq \sum_{j=1}^{n} \lambda_{j} y_{j}^{a} \bar{y}^{a}\leq \sum_{j=1}^{n} \lambda_{j} x_{j}^{b}, \lambda \geq 0\right\}\end{align}\tag{1}\]
The linear programming equation for the superSBM model containing nondesired outputs is
\[\begin{align} &\rho^{*}=\min \frac{\frac{1}{m} \sum_{i=1}^{n} \frac{\bar{x}_{i}}{x_{i o}}}{\frac{1}{q+p}\left(\sum_{r=1}^{q} \frac{\bar{y}_{r}^{a}}{y_{r o}^{a}}+\sum_{u=1}^{p} \frac{\bar{y}_{u}^{b}}{y_{uo}^{b}}\right)}\\ &\text { s.t. }\left\{\begin{array}{l} \bar{x} \geq \sum_{j=1, \neq 0}^{n} \lambda_{j} x_{j}, j=1, \cdots, m \\ \bar{y}^{a} \leq \sum_{r=1, \neq 0}^{n} \lambda_{j} y_{j}^{a}, r=1, \cdots, q \\ \bar{y}^{b} \leq \sum_{u=1, \neq 0}^{n} \lambda_{j} y_{j}^{b}, u=1, \cdots, p \\ \bar{x} \geq x_{0}, \bar{y}^{a} \leq y_{o}^{a}, \bar{y}^{b} \geq y_{o}^{b} \\ \lambda \geq 0, \sum_{h=1, \neq 0}^{n} \lambda_{h}=1 \end{array}\right. \end{align}\tag{2}\]
where \(\rho^{*}\) represents the energy efficiency value, where the higher the value, the greater the energy efficiency; \(X\) is the input factor vector, \(Y^{a}\) is the desired output vector, \(Y^{b}\) is the nondesired output vector, \(\lambda\) is the weight vector, \(o\) is a decision unit, and the input and output variables plus the horizontal line represent the corresponding projection values.
A2. Fixed Effects Model
Since we are using panel data, we need to choose either a fixed effects model or a random effects model. The pvalue of the Hausman test is 0.005, so we choose a fixed effects model for parameter estimation. The model is as follows:
\[E E_{i t}= \beta_{0}+\beta_{1} M R_{i t}+\beta_{i} X_{i t}+u_{i}+\eta_{t}+\varepsilon_{i t}\tag{3}\]
where \(EE\) stands for energy efficiency, \(MR\) stands for MR, \(X\) is the impact control variable, \(u_{i}\) is individual effects, \(\eta_{t}\) is time effects, and \(\varepsilon_{it}\) is an error term.
A3. Intermediary Effect Model
To further clarify the effect of using the classical mediating effect model, as follows:
\[E E_{i t}=\beta_{0}+\beta_{1} M R_{i t}+\beta_{i} X_{i t}+u_{i}+\eta_{t}+\varepsilon_{i t}\tag{4}\]
\[I S U P_{\dot{i} t}=\delta_{0}+\delta_{1} M R_{i t}+\delta_{i} X_{\dot{i} t}+\phi_{i}+\varphi_{t}+\gamma_{it}\tag{5}\]
\[\begin{align}E E_{i t}=&\alpha_{0}+\alpha_{1} M R_{i t}+\alpha_{2} I S U P_{i t}\\ &+\alpha_{i} X_{i t}+\mu_{i}+\vartheta_{t}+v_{i t}\end{align}\tag{6}\]
where \(ISUP\) represents the level of the industrial structure upgrading, and the other variables are similar to those in Equation (3). If both \(\delta_{1}\) and \(\alpha_{2}\) are significant, the mediating effect exists. At this point, if \(\alpha_{1}\) is still significant, \(MR\) has only a partial mediating effect.
B. Variables and Data Sources
The dataset used in this paper is the panel data of 30 Chinese provinces from 2010 to 2018, and the variable data are obtained from the China Statistical Yearbook.
B1. Explained Variable
Our explanatory variable is energy efficiency (EE1), measured using the superSBM model, and the input–output variables are described as follows:

Capital input. The capital stock, calculated by the perpetual inventory method, is in millions of yuan.

Labor input. The number of people employed at the end of the year, in millions.

Energy input. Twelve types of energy enduse consumption, in millions of tons of standard coal.

Desired output. The real gross domestic product (GDP), in billions of yuan.

Nondesired output. Carbon dioxide emissions and sulfur dioxide emissions, in millions of tons.
B2. Core Explanatory Variable
Our core explanatory variable is MR, which refers to the entire process of China’s transformation from a planned economy to a market economy, involving many aspects of economic, social, and legal regulations in the process of economic system reform, such that a single indicator cannot measure MR. Therefore, we use the findings of the China Marketization Index Research Group (http://www.ce.cn/cysc/) as a proxy for marketization reform.
B3. Mediating Variable
The mediating variable is industrial structural upgrading (ISUP). Theoretically, as a country’s industries move from lower to higher levels of development, the share of resourceintensive industries will gradually decline and the share of knowledge or technologyintensive industries will increase, which can enhance the efficiency of energy use. Therefore, we choose the output value of tertiary industry divided by the output value of secondary industry as a proxy indicator of industrial structure upgrading.
B4. Control Variables
We control for several variables, including the technology level (TEC), which uses the share of R&D investment in the GDP as a proxy; foreign direct investment (FDI), in millions of US dollars; industrial structure (IS), which we proxy with the share of secondary industry output in the GDP; and the energy structure (ES), which we define as the proportion of coal consumption in energy consumption. The results of the descriptive statistics for each variable are presented in Table 1.
III. Empirical Results
Table 2 reports the results of the estimation of the effect of MR on energy efficiency, without the inclusion of control variables in column (1) and with the inclusion of control variables in column 2, and the explanatory variables in columns (1) and (2) are energy efficiency (EE1) as measured by the superSBM model. To ensure the robustness of the regression results, we use the GDP output per unit of energy consumption (EE2) as another measure of energy efficiency and add it to the regression model. The regression results are shown in columns (3) and (4).
As shown in columns (1) and (2) of Table 2, MR is able to promote energy efficiency with and without the inclusion of control variables; all these results show that MR can promote energy efficiency, and all are significant at the 1% level. This result shows that marketization can use the price mechanism to eliminate the high energy consumption and emission of enterprises and to promote energy efficiency, consistent with other research results (e.g., Han et al., 2021). Columns (3) and (4) similarly show that MR is positively correlated with energy efficiency at the 1% level, which validates the robustness of the results to some extent.
The previous analysis verifies the facilitative effect of MR on energy efficiency, but the mechanism of action is unclear. According to the previous theoretical analysis, MR can direct the flow of various production factors to more productive sectors and improve energy efficiency by improving resource allocation efficiency and promoting technological innovation. Therefore, we use a mediating effects model to verify this effect mechanism. The regression results are presented in Table 3.
Column (2) in Table 3 shows that MR is positively significant at the 1% level, indicating that a higher degree of marketization promotes the upgrading of industrial structure. Column (3) shows that ISUP is still significantly and positively correlated with energy efficiency, indicating that industrial structure upgrading can promote energy efficiency. Thus, industrial structure upgrading has a significant mediating effect on energy efficiency in the MR process, consistent with the findings of Jun Lan et al. (2012). The upgrading of industrial structure is usually accompanied by an improvement in production efficiency and technology levels, improving energy use efficiency. In addition, MR is still positively significant at this point, but the coefficient decreases from 0.547 to 0.480, indicating that ISUP only has a partial mediating effect, and other mechanisms could still be mediating the effect between marketization and energy efficiency, which is the direction of our subsequent research efforts.
IV. Conclusion
Based on panel data on 30 Chinese provinces from 2010 to 2018, this paper investigates the relations between marketization and energy efficiency and the mechanism of action, using a fixed effects model and a mediating effects model. The main findings include the following. On the one hand, the higher the degree of marketization, the more it can promote energy efficiency. On the other hand, industrial structure upgrading plays a significant mediating effect in the process of marketization’s impact on energy efficiency. Based on the above conclusions, this research suggests that China still needs to accelerate MR to achieve the dual goals of industrial structure upgrading and energy efficiency improvement.