**Hypothesis 2 (H2).** *Governance transformation can promote GTFP improvement.*

### 2.2.3. The Theoretical Mechanism of Governance Transformation on the Porter Hypothesis

The previous research showed that the market-oriented governance transformation was helpful for improving the market competition and promoting corporate technological innovation [20]. Whether the governance transformation has a compound effect on the Porter hypothesis depends largely on the nature of the innovation effect of the governance transformation. Ettlie et al. [55] demonstrated that technological innovation can be divided into incremental innovation and radical innovation. Radical innovation refers to a kind of innovation that can produce a significant impact on market rules, the competition situation, and industrial layout, and may even lead to an industrial reshuffle [56,57]. Therefore, strengthening radical innovations and promoting the transformation of corporate technological innovations from incremental innovations to radical innovations are not only an important way for enterprises to break through technological bottlenecks and achieve an innovation catch-up, but they are also the key to the compound effect of governance transformation on the Porter hypothesis [45,58].

From the perspective of corporate governance, state-owned enterprises still contain more administrative governance factors because the reform is not in place, while private enterprises have more economic governance factors. Therefore, the innovation characteristics of enterprises with different ownerships will show differences due to different governance mechanisms. That is, state-owned enterprises prefer incremental innovations, while private enterprises hope to obtain radical innovations for subverting the existing market structure. Specifically, in the early stages of environmental regulation, radical innovation is not easy to realize. State-owned enterprises can achieve incremental innovations because they can obtain many subsidies from the government. Thus, the green competitiveness of state-owned enterprises is stronger than that of private enterprises. In the later stages of environmental regulation, the radical innovations of private enterprises are gradually realized, while the radical innovations of state-owned enterprises are slow due to the difficulty of governance transformation. At this time, the green competitiveness of private enterprises will exceed that of state-owned enterprises. Therefore, the governance transformation can accelerate the realization of radical innovations and can promote the improvement of GTFP. Based on this, the following hypothesis is proposed:

**Hypothesis 3 (H3).** *Governance transformation can accelerate the realization of the Porter hypothesis; namely, governance transformation contributes to the achievement of the improvement of environmental regulation on GTFP.*

In summary, the theoretical mechanisms of the hypotheses constructed in this paper can be summarized in Figure 2. As can be seen in Figure 2, environmental regulation will affect GTFP through three channels: the innovation compensation effect, the compliance cost effect, and the crowding out effect. Thus, the research Hypothesis 1 (H1) is obtained. Meanwhile, governance transformation can not only improve the internal production efficiency of enterprises, but it can also enhance the allocation efficiency among enterprises, which can promote GTFP. Therefore, the research Hypothesis 2 (H2) is formed. In addition, the innovation characteristics of enterprises with different ownerships will exhibit significant differences due to different governance mechanisms; namely, state-owned enterprises prefer incremental innovation, while private enterprises hope to obtain radical innovation for subverting the existing market structure. Accordingly, we can put forward the research Hypothesis 3 (H3).

**Figure 2.** The theoretical mechanism diagrams of hypotheses.

#### **3. Econometric Methodology and Data**

#### *3.1. Econometric Model Specification*

Under the new growth theory, economic growth is mainly promoted by capital, labor, and technological progress. In addition, previous studies have shown that environmental regulation, the optimization of resource allocation, and R&D investment can give rise to technological progress. Similarly, FDI and export, can bring about technology spillovers under the condition of an open economy [8,59]. Accordingly, the production function can be described as follows:

$$Y(y\_{desire}, y\_{undesire}) = A(er, gt, rd, fdi, ex, t) \cdot F(K, L) \tag{1}$$

In Equation (1), *Y* represents the total output, consisting of the desired output and the undesired output; *A* stands for the green total factor productivity (GTFP) considering undesired output; *er* denotes environmental regulation; and *gt*, *rd*, *fdi*, and *ex* represent governance transformation, R&D investment, foreign direct investment, and export, respectively. *K* and *L* indicate capital and labor, respectively. In addition, *A* indicates the Hicks neutral technology progress function. For simplicity, in reference to the study by Zhang [22], we assume that *A* in Equation (1) includes a variety of components, as follows:

$$A(er, \text{gt}, rd, fdi, ex\_{\text{t}}\text{t}) = A\_{i0}e^{\lambda\_i t} er\_{it}^{a\_i} \text{gt}\_{it}^{\beta\_i} r d\_{it}^{\gamma\_i} fdi\_{it}^{\delta\_i} e \text{x}\_{it}^{\tau\_i} \tag{2}$$

By substituting Equation (2) into Equation (1), we get the following equation:

$$Y\_{it}(y\_{desir\bar{\imath}\_{it}}, y\_{undesir\bar{\imath}\_{it}}) = A\_{i0} e^{\lambda\_i t} er\_{it}^{a\_i} g t\_{it}^{\beta\_i} r d\_{it}^{\gamma\_i} f \,\mathrm{d}\_{it}^{\delta\_i} e \mathbf{x}\_{it}^{\tau\_i} \cdot \mathbf{F}(\mathbb{K}\_{it\prime} \mathbf{L}\_{it}) \tag{3}$$

where *i* and *t* denote the region and time, respectively; *Ai0* represents the initial productivity level; *λit* reflects exogenous productivity changes; and *αi*, *βi*, *γi*, *δi*, and *τ<sup>i</sup>* represent the parameters of environmental regulation, governance transformation, R&D investment, foreign direct investment, and export trade on GTFP, respectively.

In terms of the definition of GTFP, we divide the two sides of Equation (3) by using *F*(*Kit*,*Lit*) to obtain the following formula:

$$GTPP\_{it} = \frac{Y\_{it}(y\_{desired}, y\_{undesire\_{it}})}{F(K\_{it'}L\_{it})} = A\_{i0}e^{\lambda\_{i}t} \text{er}\_{it}^{\kappa\_i} \text{gt}\_{it}^{\delta\_i} \text{rd}\_{it}^{\gamma\_i} f \text{di}\_{it}^{\delta\_i} \text{ec}\_{it}^{\tau\_i} \tag{4}$$

By taking the natural logarithm of both sides of Formula (4), we obtain the theoretical framework describing the impact factors of GTFP, as follows:

$$\ln GTFP\_{it} = \ln A\_{i0} + \lambda\_i t + a\_i \ln e r\_{it} + \beta\_i \ln g t\_{it} + \gamma\_i \ln r d\_{it} + \delta\_i \ln f d\_{it} + \tau\_i \ln e x\_{it} \tag{5}$$

To correctly explore the effect of environmental regulation, governance transformation, and their interaction terms on GTFP, based on the above analysis framework, the linear term and the quadratic term of environmental regulation are both introduced into the econometric model to investigate the possible nonlinear relationship between environmental regulation and GTFP. Since the current productivity growth may be affected by past productivity, the first-order lag term of GTFP is also included in the model as an explanatory variable to examine the dynamic cumulative effect of GTFP growth. Meanwhile, the estimation method, the system generalized method of moments (GMM), is used in this paper mainly because the system GMM can effectively overcome the endogenous problem by introducing instrumental variables. Based on the above considerations, the dynamic panel regression model constructed in this study is as follows:

$$\ln G TFP\_{it} = a\_0 + a\_1 \ln G TFP\_{it-1} + a\_2 \ln E R\_{it} + a \varepsilon (\ln E R\_{it})^2 + a\_4 \ln G T\_{it} + \beta X\_{it} + V\_i + \varepsilon\_{it} \tag{6}$$

where *i* and *t* represent the province and year, respectively (*i* = 1, 2, 3, ... , 30, *t* = 2003, 2004, ... , 2017); *Vi* indicates the provincial fixed effect; ε*it* is the random error term; and

*GTFPit* denotes the green total factor productivity for each province; *α<sup>1</sup>* is a hysteresis multiplier, indicating the effect of the last period of GTFP on the current GTFP; *ERit* denotes environmental regulation; *GTit* represents governance transformation; and *Xit* is the control variables vector, including R&D investment, export trade, foreign direct investment, and the factor endowment structure, respectively. To be specific, the coefficient of the quadratic term of ER illustrates the nonlinear effect of environmental regulation on GTFP. When the estimated value of parameter *α<sup>3</sup>* is significantly more than 0, Hypothesis 1 is confirmed. Simultaneously, when the result of the parameter *α<sup>4</sup>* is significantly greater than 0, Hypothesis 2 is confirmed. In addition, the interaction term of *ER* and *GT* is included in the model to verify Hypothesis 3. The dynamic panel regression model with the cross-term is established as follows:

$$\begin{aligned} \ln G T F P\_{it} &= a\_0 + a\_1 \ln G T F P\_{it-1} + a\_2 \ln E R\_{it} + a\_3 (\ln E R\_{it})^2 + a\_4 \ln G T\_{it} \\ &+ a\_5 \ln E R\_{it} \times \ln G T\_{it} + \beta X\_{it} + V\_i + \varepsilon\_{it} \end{aligned} \tag{7}$$

In Equation (7), ln*ERit*×ln*GTit* indicates the cross-term of environmental regulation and governance transformation. When the estimated result of the coefficient of the interaction term is significantly greater than 0, Hypothesis 3 is confirmed.

#### *3.2. Variable Selection and Description*

#### 3.2.1. Dependent Variable

The dependent variable used here is green total factor productivity (GTFP). In general, productivity indexes used in prior studies focused on measuring marketable outputs relative to the paid factors of production. However, the generation of by-products, such as environmental pollutants, was not taken into consideration. In view of this, Chung et al. [60] proposed the Malmquist-Luenberger index, based on the directional distance function, which extends the original Malmquist index to account for undesired outputs, such as industrial sulfur dioxide, industrial smoke and dust, and industrial sewage. Therefore, to overcome the drawback of lacking environmental factors in the traditional productivity index, and to inherit the advantages of the Malmquist-Luenberger index following Chung et al. [60] and Oh [61], this paper uses the global Malmquist-Luenberger index (GML), based on the directional distance function, to measure the GTFP growth for each province in China. The GML index used in this study is defined as follows:

$$GML^{t,t+1}(\mathbf{x}^t, \mathbf{y}^t, \mathbf{b}^t, \mathbf{x}^{t+1}, \mathbf{y}^{t+1}, \mathbf{b}^{t+1}) = \frac{1 + D^G(\mathbf{x}^t, \mathbf{y}^t, \mathbf{b}^t)}{1 + D^G(\mathbf{x}^{t+1}, \mathbf{y}^{t+1}, \mathbf{b}^{t+1})} \tag{8}$$

In Equation (8), *<sup>D</sup>G*(•) represents the directional distance function; *<sup>x</sup><sup>t</sup>* represents the input vector constructed by *N* kind of input factors; *yt* represents the desirable output vector constructed by *M* kind of desired outputs; and *bt* represents the undesirable output vector constructed by *I* kind of undesired outputs. Specifically, the input factors include: (1) capital input, where the capital stock of each province is calculated by employing the perpetual inventory method [9]; (2) labor input, measured by the annual average number of people employed for each province; and (3) energy consumption, expressed by the provincial primary energy consumption, which is converted to standard coal. In addition, the output factors include: (1) the desirable output, denoted by gross industrial output; and (2) the undesirable output, measured by a variety of environmental pollutants, including industrial sewage emissions, industrial sulfur dioxide, and carbon dioxide emissions.

3.2.2. Core Independent Variables

• Environmental regulation (ER). According to the Porter hypothesis, reasonable environmental regulation can effectively stimulate enterprise enthusiasm for green innovation, promote pollution emissions reduction, and improve green productivity. In previous studies, there were no clear and unified criteria to represent environmental regulation, which can mainly be measured by pollutant emission reductions, environmental pollution control investments, pollution reduction expenditures, regulatory enforcement stringencies, and pollution sewage charges [4,39,62,63]. For the consideration of data integrity and availability, this paper selects the ratio of the total investment of provincial industrial pollution control to the gross industrial output as a proxy variable for the provincial environmental regulation level. Usually, the higher the proportion of the investment, the greater the environmental regulation intensity, and the better the effect of regional green development. In addition, we also select the ratio of the total investment of industrial pollution control to the operating cost of industrial enterprises, as a substitute variable for the robustness analysis.

• Governance transformation (GT). China's corporate governance, which is deeply rooted in the transitional economy, is gradually transforming from administrative governance to economic governance. In essence, governance transformation reflects the persistent improvement of the degree of marketization for a region. Generally speaking, private enterprises have better economic governance, while state-owned enterprises have stronger administrative governance in China. Therefore, the governance transformation, which is measured by the ratio of the main business income of private industrial enterprises to the sum of the main business income of state-owned and private-owned industrial enterprises for each region, is selected here. The larger the ratio of governance transformation, the higher the degree of marketization, and the better the resource allocation and productivity of enterprises.

### 3.2.3. Control Variables

To alleviate the endogenous problem caused by the omitted variables, and to improve the estimation accuracy, a series of control variables are selected in the model by referring to the previous literature. The variables included are as follows:

