5.2. Spatial Correlation Test
Figure 3 shows that the dominant areas of GETI in China are concentrated in the southeastern provinces, while the areas with GETI deficiency are mainly distributed in the northwestern provinces. This initially suggests a spatial correlation of GETI in China, but it needs to be further examined. According to Wang et al. [
70] and Yang et al. [
65], the global Moran’s I index is selected to uncover the spatial autocorrelation of GETI, and the results are displayed in
Table 5. It is clear that there is a positive spatial autocorrelation of GETI between regions, and it is significant at least at the 5% level. The global Moran’s I index examines the overall spatial autocorrelation of GETI, but fails to reflect the spatial autocorrelation of a certain area. Referring to Fan and Xu [
71], we then explore the local spatial autocorrelation of GETI with the Getis-Ord index. The Getis-Ord index divides areas into cold spots and hot spots according to the spatial autocorrelation of variables. The cold spots are the areas where low and low values are clustered, and the hot spots are the areas where high and high values are clustered.
Figure 5 shows the distribution of cold and hot spots for the average GETI from 2001 to 2018. The results revealed that Sichuan province in western China is the cold spot, while Jiangsu, Shanghai, Zhejiang, Fujian, Anhui, Hubei and Jiangxi provinces in eastern and central China are the hot spots. Obviously, GETI also has spatial agglomeration effect in local areas. To demonstrate the spatial autocorrelation more precisely, we also performed Lagrange multiplier (LM) statistical tests [
72]. The results are shown in
Table 6. It can be seen that the LM statistic and the robust LM statistic of the spatial lag test are both significant at the 1% level. The LM statistic of the spatial error test is significant, but the robust LM statistic is not significant. Overall, at least both statistics of the spatial lag test are highly significant, suggesting that it is suitable to adopt the spatial econometric model.
5.3. Baseline Results
In our model, there is no bilateral causal association between environmental regulation and GETI because the measurement indicator of environmental regulation is obtained by text analysis. Meanwhile, the dynamic spatial panel model is adopted in this paper, which can better deal with the possible endogeneity issues by using the generalized moment estimation method.
The empirical results are reported in
Table 7. Model 2 shows the estimation results of DSDM. As a benchmark, the DSAR method is also employed and the results are shown in Model 1. Since DSDM includes DSAR to some extent, the LR test and the Wald test are used to investigate whether DSDM can be degraded into DSAR. The test results in
Table 6 reject the null hypothesis, indicating that DSDM cannot be degraded into DSAR. Therefore, it is reasonable to choose DSDM for analysis.
In model 2, the spatial autocorrelation coefficient ρ is 0.3530, which is highly significant at 1% level, illustrating that the local GETI is significantly affected by the neighboring area. This is mainly because green energy technologies can spread to neighboring areas through such mechanisms as human mobility, investment, and industrial transfer. The positive spatial lag coefficient indicates that the spatial knowledge spillover will narrow the gap of GETI between areas, which is beneficial to the overall improvement of GETI [
73]. As for the core independent variable, the coefficient of environmental regulation is positive, and the coefficient of the square term of environmental regulation is negative, and both are significant at 1% level. It shows that there is a significant inverted U-shaped relationship between environmental regulation and GETI. This is consistent with the results of Shang et al. [
44] and Wang et al. [
74]. The possible explanation is that in the early stage, stronger regulation not only increases the cost of compliance, but also reduces the effectiveness of traditional terminal governance measures [
17]. For companies, GETI has become a more effective way to enhance competitiveness by improving their production processes and productivity. In this sense, the compliance cost induced by environmental regulation is offset by the innovation compensation effect [
75]. Companies are willing to engage in innovations related to green energy technologies and improve their core competitiveness through cleaner production. When the intensity of environmental regulation reaches a certain threshold, environmental regulation will have a negative impact on GETI. On the one hand, under strict supervision, any non-compliance behavior of companies may incur high fines. This will crowd out production and R&D spending, thus hindering GETI activities. Clearly, the compliance cost effect caused by environmental regulation in this period exceeds the innovation compensation effect. On the other hand, the bottleneck effect also exists in GETI [
76]. As the intensity of environmental regulation increases, corporate innovation may shift to other technical fields where it is easier to succeed. More concretely, strict environmental regulation will reduce the risk-bearing capacity of companies, then companies will spare no effort to seek lower-cost green technologies to reduce compliance costs. Therefore, when there is a bottleneck effect in GETI, strict environmental regulation may cause companies to switch to other green innovations with lower costs or less risks, which is not conducive to GETI. We can calculate that the inflection point of the inverted U-shaped curve is 0.3572. In the sample population, 47.96% of the observations are located to the left of the inflection point, and 52.04% of the observations are distributed to the right of the inflection point. Accordingly, provinces should appropriately adjust environmental policies according to their positions to promote GETI. Moreover, the coefficient of L.lnGETI is 0.5379 and the significance level is 1%, suggesting that GETI of the current period will be positively affected by the previous GETI [
77]. Clearly, there is a path-dependent effect in GETI. That is, areas with a larger stock of green energy knowledge are more likely to pursue similar technologies in the future. In terms of control variables, industrial structure plays a significant role in promoting GETI. This is mainly because the companies related to energy are concentrated in the secondary industry. For this reason, a higher share of the secondary industry means more energy companies and more likely to obtain relevant innovation patents. The effects of other control variables on GETI are not significant. Regarding the spatial interaction term, the spatial interaction coefficients of ER and ER
2 are not all significant, indicating that spatial spillover effect is not significant. As for the spatial interaction term, the coefficient of W × ER is negative and the coefficient of W × ER
2 is positive, indicating a U-shaped relationship between environmental regulations in adjacent areas and local GETI. That is to say, the environmental regulation of neighboring areas first hinders and then promotes local GETI. However, the U-shaped spatial spillover effect of environmental regulation is not significant because the coefficient of ER is not significant. This may be due to the two-way influence of environmental regulation in neighboring areas. On the one hand, the strict environmental regulation in neighboring areas may lead to the transfer of polluting industries to the local area, which will adversely affect the local GETI [
78]. On the other hand, environmental regulations in neighboring areas may have a demonstration effect, which will make local governments adopt similar policies to promote GETI [
40]. The insignificant U-shaped relationship may be caused by the fact that neither effect is dominant. Among the control variables, only the coefficient of trade openness is significantly negative, implying that a higher level of trade openness in adjacent areas is not conducive to local GETI. It indicates that the higher concentration of foreign companies in neighboring areas has a siphon effect on the local area.
The results of Model 1 are similar to those of Model 2. The spatial autoregressive coefficient of GETI is positive, demonstrating that local GETI will be comprehensively affected by adjacent areas. The coefficient of ER is positive and the quadratic coefficient of ER is negative, indicating that an inverted U-shaped correlation existed between environmental regulation and GETI. Moreover, GETI has a significant path-dependent effect.
To further explore the reliability of model 2, the following robustness tests are performed. (1) Replace the spatial weight matrix. In this method, three spatial weight matrices are adopted, namely, adjacency weight matrix, economic distance weight matrix, and geographical economic distance weight matrix. Specifically, the setting of adjacency weight matrix is as follows.
if area
i and area
j are adjacent areas, and
if area
i and area
j are not adjacent areas. The estimation results of DSAR model and DSDM model of adjacency weight matrix are shown in Model 3 and Model 4 respectively. The economic distance weight matrix is set as follows.
if
i ≠ j, and
if
i = j, where
agdpi is the GDP per capita of area
i. The estimation results of DSAR model and DSDM model of economic distance weight matrix are shown in Model 5 and Model 6 respectively. The geographical economic distance weight matrix is set as follows.
, where
represents the geographical distance weight matrix in Equation (2),
represents the economic distance weight matrix. The estimation results of DSAR model and DSDM model of geographical economic distance weight matrix are shown in Model 7 and Model 8 respectively. (2) Replace the measurement index of GETI. First, in line with Cheng et al. [
79], the number of invention patents is taken as the proxy variable of GETI. Then, the estimation methods of model 1 and model 2 are repeated to obtain model 9 and model 10. Second, referring to Wu et al. [
80], the number of patents granted is taken as an indicator of GETI, and the estimation results are shown in Model 11 and Model 12. (3) Education is added to the model as an additional control variable. Education level, to some extent, determines the quality of employees in a region and thus can be considered as one of the factors affecting GETI. The level of education is measured by the number of college students per 10,000, and the data can be obtained from China Statistical Yearbook (2002–2019). The estimation methods of DSAR and DSDM are repeated to obtain model 13 and model 14. As shown in
Table 7 and
Table 8, the results of the three robustness tests are similar to model 2. Namely, GETI has significant spatial spillover effect and path-dependent effect, and there is a significant inverted U-shaped correlation between environmental regulation and GETI. Furthermore, both spatial spillover effect and path dependence effect exist in GETI. Therefore, our empirical results are both reliable and robust. It is worth noting that in model 4, the LR test passes, whereas the Wald test fails. This does not affect the conclusion, because the main conclusions of Model 3 and Model 4 are consistent.
5.4. Further Analysis
As mentioned earlier, green energy technologies include AEPTs and ECTs. Does environmental regulation have the same impact on these two types of technologies? To answer this question, we further examine the effects of environmental regulation on the innovation of AEPTs and ECTs, respectively, and the results are displayed in
Table 9. We can see that the dependent variables of both Model 15 and Model 16 are alternative energy production technology innovation (AEPTI). The difference is that the estimation method of Model 15 is DSAR, while the estimation method of Model 16 is DSDM. When the dependent variables of model 15 and model 16 are replaced by energy conservation technology innovation (ECTI), then obtain model 17 and model 18 can be obtained. It not difficult to find that the spatial autocorrelation coefficients of all models are positively significant, indicating that both AEPTI and ECTI are affected by similar innovations in neighboring areas. For environmental regulation, the outcomes of all models are similar to model 2. That is, as the intensity of environmental regulation increases, both AEPTI and ECTI increase first and then decrease. Although the inverted U-shaped correlation exists between environmental regulation and AEPTI as well as between environmental regulation and ECTI, the location of the inflection point is different. Using the mathematical calculation of the parabola, it can be known that the inflection points of environmental regulation on AEPTI and ECTI are 0.3985 and 0.3337, respectively. As shown in
Figure 6a, ECTI is more likely to reach the inflection point as environmental regulation increases. The possible reason is that in a certain period, with the enhancement of environmental regulation, the direction of GETI in companies is more focused on AEPTI. This is reflected in
Figure 6a as AEPTI continues to rise when ECT reaches the inflection point. Finally, it can also be found that the lag terms of all dependent variables are significantly positive, suggesting the existence of path dependence in both AEPTI and ECTI.
Considering the unbalanced characteristics of China’s economy, we also wanted to know whether the role of environmental regulation on GETI is heterogeneous in different regions. As a result, we further explore the influence of environmental regulation on GETI in eastern China as well as in central and western China. According to the National Bureau of Statistics of China, eastern China consists of 11 provinces (cities) including Beijing, Tianjin, Hebei, Liaoning, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, and Hainan. The other 19 provinces (cities) belong to central and western China. The results are reported in
Table 10. Specifically, the areas in Model 19 and Model 20 are both eastern China. Model 19 is a benchmark model based on DSAR, and Model 20 shows the estimation results of DSDM. By replacing the eastern areas in model 19 and model 20 with the central and western areas, model 21 and model 22 are obtained. Clearly, the spatial spillover effect and path-dependent effect of GETI are significant in both eastern China and central and western China. The inverted U-shaped relationship is also significant, suggesting that the role of environmental regulation on GETI in different regions are similar. Further investigation of the inverted U-shaped relationship shows that the inflection point in eastern China is quite different from that in central and western China, as the inflection point values are 0.5315 and 0.314 respectively. From
Figure 6b, we can see that, with the strengthening of environmental regulation, eastern China reaches the inflection point later than central and western China. This may be due to the fact that eastern China has gathered a lot of capital and R&D resources, and has higher affordability in cost compliance and greater advantages in innovation.
Table 10 also shows that the spatial lag coefficient of GETI in eastern China is smaller than that in central and western China, implying that the spatial spillover effect of GETI in eastern China is smaller. The possible explanation is that the higher innovation level makes companies in eastern China pay more attention to internal improvement mechanisms. That is, rather than absorbing knowledge spillovers from neighboring areas, local companies are more inclined to increase internal R&D expenditures to improve their overall innovation level. In contrast, due to the weak capacity and the poor risk bearing capacity, companies in central and western China are more willing to absorb knowledge spillovers from neighboring areas to improve their innovation level.