# The spatial spillover effect of environmental regulation on the total factor productivity of pharmaceutical manufacturing industry in China

Based on the panel data of 30 provinces in China from 2004 to 2019, this paper studies the spatial spillover effects of environmental regulation (ER) on total factor productivity of pharmaceutical industry (HTFP) by establishing a spatial econometric model. “Temporal and spatial charact eri stics of ER and H TFP” section discusses the spatiotemporal characteristics of ER and pharmaceutical manufacturing total factor productivity. “Empirical results of panel data regression” section studies the effect of ER on pharmaceutical manufacturing total factor productivity through panel data regression. “Empirical results of spatial econometric model” section studies the spatial spillover effect of ER on pharmaceutical manufacturing total factor productivity through spatial Durbin model. “Further analysis” section further analyzes the intermediary transmission mechanism, regional heterogeneity, endogeneity and robustness of the two.

### Temporal and spatial characteristics of ER and HTFP

The pharmaceutical manufacturing total factor productivity index calculated by the Metafrontier Malmquist-Luenberger index under the mixed distance EBM model is the change rate from *t* to *t* + *1*, and has cumulative characteristics during the study period. Considering the accuracy of temporal-spatial analysis and the stability of the overall change, the annual average value of pharmaceutical manufacturing total factor productivity is used in “Temporal and spatial charact eri stics of ER and H TFP” section, and the cumulative value of pharmaceutical manufacturing total factor productivity is used in the subsequent measurement process.

Figure 5a describes the time trend of annual average change of pharmaceutical manufacturing total factor productivity and the change of cumulative value of pharmaceutical manufacturing total factor productivity in the whole country and the three major regions of east, middle and west. From the images, we can see that the average annual value of pharmaceutical manufacturing total factor productivity from 2004 to 2019 is greater than 1 except 2017, which indicates that pharmaceutical manufacturing total factor productivity is gradually increasing over time during the study period. In addition, pharmaceutical manufacturing total factor productivity reached the maximum (1.3241) in 2010, and then pharmaceutical manufacturing total factor productivity developed steadily. From 2004 to 2019, the annual average of pharmaceutical manufacturing total factor productivity in China was 1.1422, that is, the average growth rate was 14.22{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}, and the cumulative HTFP was 4.1758. The trend of time change in the three regions is similar to that of the whole country, and the overall trend of time change in the three regions is small.

According to Fig. 5b, the spatial distribution of pharmaceutical manufacturing total factor productivity decreased from west to east, and from high to low, it was the west (1.1652), the middle (1.1357) and the east (1.1239). During the study period, the pharmaceutical industry in the western region has developed rapidly, while the eastern region has a strong level of development and is difficult to upgrade, so it is in a stable development trend. Within the region, in the eastern region, Beijing, Shanghai and Liaoning are at a higher level of development, but there is a big gap between their surrounding provinces and central cities; In the central region, the development of each province is relatively average, and the development difference between the surrounding provinces and the central cities is relatively small; In the western region, there is a big gap in the development of various provinces, Ningxia, Gansu and Yunnan are developing faster, while Shaanxi and Guangxi are developing slower.

To sum up, it is important to consider the aggregation of provinces and the heterogeneity of regions in the study of the effect of ER on pharmaceutical manufacturing total factor productivity.

Figure 6a and b describe the temporal and spatial variation characteristics of environmental regulation (ER). Figure 6a describes the change trend and aggregation characteristics of each province over time. Figure 6b describes trends in the spatial dynamics of ER by region and province from 2004 to 2019.

From Fig. 6a, the overall ER of the whole country shows a gradual upward trend over time, indicating that during the study period, the intensity of environmental regulation is increasing. Except for some provinces, most provinces show the spatial agglomeration characteristics of “from agglomeration to decentralization”. Before 2014, the difference of ER in each province is small, but after 2014, the development of ER in each province is more dispersed. The differences in energy utilization, industrial pollution and resource allocation in each province continue to show, which leads to the different intensity of environmental regulation.

From Fig. 6b, the ER of each region and province shows a gradual upward trend over time, which is consistent with the overall change of the whole country. From the regional point of view, the average ER from 2004 to 2019 is the eastern (10.2144), the central (8.6867) and the western (7.8871), and the eastern ER is far more than that of the central and western regions, which shows that the eastern region has strengthened environmental regulation while developing its economy. This is closely related to the eastern industrial structure, pollution prevention and control, environmental input and so on. For example, in the eastern region, Tianjin (14.4987) and Zhejiang (12.5606) are in the leading position, while Fujian (7.1645), Liaoning (7.6985) and Hebei (7.8748) are relatively low; In the central region, Shanxi (11.5473) ranks first, while Jilin (5.5154) ER is not only at the lowest level in the central region, but also relatively backward in the whole country. There is a large range of “high-high” spatial aggregation of ER in the eastern and central regions, while the ER in the western region is generally low, which shows a “low-low” aggregation characteristics.

Therefore, in the follow-up study, we not only need to consider the local impact of ER, but also need to determine whether there is a spatial spillover effect, so as to more comprehensive analysis of the impact of ER on pharmaceutical manufacturing total factor productivity.

### Empirical results of panel data regression

Before studying the spatial spillover effects of ER on pharmaceutical manufacturing total factor productivity, we need to establish a panel data regression model. The effect of ER on pharmaceutical manufacturing total factor productivity was discussed. In this paper, ER is taken as the core explanatory variable and pharmaceutical manufacturing total factor productivity as the explained variable. By constructing mixed panel regression model (OLS), fixed effects model (FE) and random effects model (RE), the optimal regression model is selected. Table 3 is the regression results of the three models. Model-1, Model-2 and Model-3 are the regression results of mixed OLS, fixed effects and random effects, respectively. Model-4 is the result of fixed effect regression with ER squared. In Model-4, the regression coefficient of the square term of ER (ER^2) is not significant, so the impact of environmental regulation on the total factor productivity of pharmaceutical manufacturing industry is not a U-shaped relationship. Through the F test, LM test and Hausman test, the fixed effects model (Model-2) is the best model, and the goodness of fit of Model-2 is the best, which is 0.199, indicating that the explanatory degree of explanatory variables to HTFP is 19.9{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}.

From a fixed effect (Model-2), the regression coefficient of ER to pharmaceutical manufacturing total factor productivity was 16.5, which was significant at the level of 1, pharmaceutical manufacturing total factor productivity will rise by 16.5, which shows that the increase of environmental regulation has significantly promoted the improvement of total factor productivity in the pharmaceutical industry, proving that Hypothesis 1. On the one hand, according to the “compensation effect” and stricter environmental regulations, pharmaceutical enterprises that fail to meet the standards upgrade pollution prevention, they have to control technology, improve energy efficiency, reduce pollution emissions, which will improve the total factor productivity of pharmaceutical. On the other hand, according to the “reverse effect”, pharmaceutical enterprises face the deepening environmental regulation and the increasing cost of pharmaceutical enterprise governance. It will make some enterprises continuously improve production efficiency under the pressure of environmental regulation, and upgrade production technology, industrial structure, resource energy consumption, environmental pollution and other aspects, so as to improve the total factor productivity of pharmaceutical.

From the impact of control variables on pharmaceutical manufacturing total factor productivity, Capital and Profit are significantly positive at the level of 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} and 5{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}, respectively. When Capital and Profit increase by 1, pharmaceutical manufacturing total factor productivity will increase by 0.660 and 4.083. The effect of Income on pharmaceutical manufacturing total factor productivity was significantly negative at 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. This shows that opening to the outside world and the profits of the whole industry are important factors to promote the development of pharmaceutical industry, while business income and the number of employees of industrial enterprises have less effect on improving pharmaceutical manufacturing total factor productivity.

### Empirical results of spatial econometric model

Through the analysis of temporal and spatial characteristics of ER and pharmaceutical manufacturing total factor productivity, this paper needs to further discuss whether there is spatial spillover effect of ER on pharmaceutical manufacturing total factor productivity. This part is divided into two parts. First, build the space weight matrix test and see whether there is spatial autocorrelation between ER and pharmaceutical manufacturing total factor productivity. Second, if there is spatial autocorrelation between ER and pharmaceutical manufacturing total factor productivity, the corresponding spatial econometric model is selected to study the spatial spillover effect of ER on pharmaceutical manufacturing total factor productivity.

#### Spatial autocorrelation test

Considering the accuracy and feasibility of the follow-up regression results, this paper constructs 0–1 adjacency matrix (W1) and geographical distance matrix (W2) as the basis of the follow-up study. In this paper, we test the spatial autocorrelation of pharmaceutical manufacturing total factor productivity and ER by global Moran index. The value of global Moran index is between − 1 and 1, and any value greater than 0 indicates the existence of positive spatial autocorrelation of the variable, and any value less than 0 indicates the existence of negative spatial autocorrelation of the variable. The global Moran exponents for W1 and W2 matrices are shown in Table 4. The results show that pharmaceutical manufacturing total factor productivity and ER have significant positive spatial autocorrelation.

In order to further analyze the spatial aggregation of each province, this paper calculates the local Moran index, draws the local Moran index scatter plot and the local autocorrelation LISA plot. The Moran scatter plot and LISA cluster plot of 2011 are drawn based on W1 matrix. Figure 7a and b are the Moran scatter plot and LISA of pharmaceutical manufacturing total factor productivity, respectively. Figure 8a and b are Moran scatter plot and LISA cluster plot of ER, respectively.

From the Moran scatter diagram of pharmaceutical manufacturing total factor productivity and ER, it can be seen that most provinces are concentrated in the first quadrant and the third quadrant, which indicates that there are obvious “high-high” or “low-low” aggregation in pharmaceutical manufacturing total factor productivity and ER. From the LISA cluster map of pharmaceutical manufacturing total factor productivity and ER, it can be seen that there is a “high-high” cluster of pharmaceutical manufacturing total factor productivity in Beijing and Jiangsu, a “low-low” cluster in Xinjiang and Chongqing, and a “high-low” cluster in Ningxia. ER in Hebei, Shandong and Jiangsu have “high-high” aggregation, Gansu and Ningxia have “low-low” aggregation, Guangdong has “high-low” aggregation and Hainan has “low–high” aggregation.

From Fig. 7a and b, the Moran scatter plot and LISA aggregation plot are shown that pharmaceutical manufacturing total factor productivity has the spatial agglomeration characteristics of “high-high” or “low-low” aggregation. For example, from Fig. 7a and b, pharmaceutical manufacturing total factor productivity shows “high-high” aggregation of provinces such as Tianjin and Jiangsu, and “low-low” aggregation of provinces such as Xinjiang and Chongqing. China’s pharmaceutical manufacturing industry has a good momentum of development, under the influence of policies, resources and other factors, the overall showing a more obvious regional characteristics. In recent years, remarkable industrial clusters have been formed in the Yangtze River Delta, Dawan District and Bohai Rim, mainly relying on regional innovation-driven, industrial support, economic base and other advantages. The improvement of pharmaceutical manufacturing total factor productivity in the surrounding areas will promote the improvement of local pharmaceutical manufacturing total factor productivity to a certain extent Tianjin and Jiangsu, as representatives of the Bohai Rim and Yangtze River Delta, show the characteristics of “high-high” aggregation.

As an emerging industrial cluster, Sichuan-Chongqing region has a “low-low” aggregation in the results, the main reasons are: (1) there is a short-term effect of R&D investment on the growth of enterprises, but the R&D investment of enterprises needs long-term accumulation; (2) The innovation output cycle of pharmaceutical products is longer than that of other industries, and the innovation achievements may not be obvious in a short time; (3) There is innovation spillover effect in pharmaceutical manufacturing industry, and technology leaders provide technology to transferees involuntarily, which makes technology leaders fail to receive corresponding returns. Xinjiang borders Qinghai, Gansu and Inner Mongolia, and its geographical location is located in the westernmost part of China, far from the industrial cluster cities, forming a “low-low” agglomeration situation.

From Fig. 8a and b, the provinces with “high-high” ER are Hebei, Shandong and Jiangsu, and the provinces with “low-low” ER are Gansu and Ningxia. Hebei and Shandong are traditional provinces with large industrial and resource reserves, and also have severe environmental conditions, which require more stringent environmental regulation. So these industrial clusters have formed a “high- high” cluster.

In Jiangsu Province, which is close to the traditional industrial agglomeration area, the environmental regulation has also appeared the characteristics of “high-high” agglomeration. The main reasons are as follows: (1) Jiangsu is located in the border area of traditional industries, and there may be some enterprises using the layout of other places to avoid supervision, which aggravates the environmental pollution in the adjacent areas, so more stringent environmental regulation is needed; (2) Environmental regulation in Jiangsu has “marginal effect”, and the environmental benefits brought by the same environmental input cost will be lower than other provinces, so more targeted and effective environmental governance measures should be taken to improve the efficiency of environmental regulation. However, Gansu and Ningxia are far away from traditional industrial clusters and heavy industrial clusters, and the intensity of environmental regulation is relatively small, thus forming the “low-low” aggregation characteristics of environmental regulation.

#### Regression results of spatial econometric model

In the above research, due to the existence of “compensation effect” and “inversion effect”, increasing environmental regulation has a significant role in promoting the total factor productivity of the pharmaceutical industry. However, due to the significant spatial autocorrelation between ER and pharmaceutical manufacturing total factor productivity, there are obvious “high-high” or “low-low” aggregation characteristics in the region, so it is very important to study the spatial spillover of ER to pharmaceutical manufacturing total factor productivity. In the process of establishing the spatial econometric model, pharmaceutical manufacturing total factor productivity is taken as the explained variable, ER is taken as the core explanatory variable, and 0–1 adjacency matrix (W1) is taken as the spatial weight matrix to establish the spatial autoregressive model.

Table 5 is the regression results of three spatial econometric models. Model-5, Model-6 and Model-7 are the regression results of SAR, SEM and SDM, respectively. The results of LR test show that the spatial Durbin model (Model-7) is the optimal model. In addition, the goodness of fit of Model-7 was the best, which is 0.19, indicating that the explanatory degree of explanatory variables to HTFP was 19{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}.

In this paper, the results of spatial Durbin model are used as the basis for the follow-up analysis.

First, the coefficient of pharmaceutical manufacturing total factor productivity spatial lag is 0.187, which is significant at 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. This shows that every 1 increase of HTFP in surrounding areas will increase HTFP by 0.187 in this area, which proves Hypothesis 2. The spatial lag term of pharmaceutical manufacturing total factor productivity is significantly positive, which also verifies the existence of spatial autocorrelation of pharmaceutical manufacturing total factor productivity, and the existence of “high-high” or “low-low” aggregation in each province. Under the national strategy of overall development of pharmaceutical industry, governments in various regions have intensified their policy efforts to support the development of pharmaceutical manufacturing industry in their respective regions from the perspectives of capital investment, talent attraction and infrastructure construction. Relying on the local resources to build the pharmaceutical manufacturing city, gradually deepen the degree of cluster industrialization, improve the competitiveness of pharmaceutical manufacturing enterprises in scale and innovation, so as to drive the development of pharmaceutical manufacturing total factor productivity in surrounding cities as a central city.

Second, the regression coefficient of ER’s influence on local pharmaceutical manufacturing total factor productivity is 6.481, which is significant at 5{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. This shows that environmental regulation has a significant role in promoting the improvement of local pharmaceutical manufacturing total factor productivity, which proves that Hypothesis 1. For every 1 increase in local ER, local pharmaceutical manufacturing total factor productivity will increase by 6.481. The regression coefficient of ER on local pharmaceutical manufacturing total factor productivity was 36.96, which was significant at 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. This shows that the ER in the surrounding area has a significant role in improving the local pharmaceutical manufacturing total factor productivity, which proves that Hypothesis 2. For every 1 increase in ER in the surrounding area, the local pharmaceutical manufacturing total factor productivity will increase by 36.96. The role of environmental regulation on the total factor productivity of pharmaceutical manufacturing industry has a significant positive impact on both local and surrounding areas. On the one hand, according to the “backward effect”, strengthening environmental regulation means that polluting pharmaceutical enterprises have high cost, forcing enterprises to carry out technological innovation, so as to improve the total productivity of pharmaceutical factors. On the other hand, environmental regulation will screen out “clean” pharmaceutical enterprises, so that the local formation of “green barriers”. The “backward effect” makes the local green clean technology have a good demonstration effect on the surrounding areas, while the local “green barrier” transfers the non-clean enterprises to the surrounding areas to a certain extent, but because this transfer lags behind the “non-clean transfer” to pharmaceutical manufacturing total factor productivity.

Third, from the control variables on the impact of pharmaceutical manufacturing total factor productivity point of view, the control variable regression coefficient of the direction and fixed. Open, Capital and Profit have a positive effect on the local pharmaceutical manufacturing total factor productivity, but have a negative effect on the surrounding pharmaceutical manufacturing total factor productivity. Labor and Income have a negative impact on the local pharmaceutical manufacturing total factor productivity, but have a positive impact on the surrounding pharmaceutical manufacturing total factor productivity. The potential reason is that the central city absorbs the resource advantages of the surrounding areas, resulting in a “siphon effect”, which leads to the lack of sufficient resources in the surrounding areas to improve the efficiency level of the pharmaceutical industry in the region.

### Further analysis

#### Mediating transmission mechanism

In this paper, the spatial spillover effect of ER on pharmaceutical manufacturing total factor productivity is studied by establishing a SDM. However, the impact of environmental regulation on the development of pharmaceutical manufacturing industry is not a direct relationship between the two, and there is a complex intermediary effect between them. Therefore, in the further analysis, this paper first studies the mediating effect between the two.

This paper divides the mediating effect path into technical effect and structural effect, and the technical effect is divided into green technology and production technology, so this paper selects Ingrva, RD and Structure as three mediating variables, which correspond to green technology, production technology and structural effect respectively. The results of the mediating effect under the spatial Durbin model are Tables 6, 7 and 8.

Table 6 shows the regression day results with Ingrva as the mediator. ER is the core explanatory variables of Model-8, and the dependent variable is Ingrva. In Model-9, Ingrva was used as the core explanatory variable and pharmaceutical manufacturing total factor productivity was used as the dependent variable. ER and Ingrva were used as the core explanatory variables and pharmaceutical manufacturing total factor productivity as the dependent variable in Model-10.

According to Model-8, the spatial lag coefficient of Ingrva is 0.201, which is significant at 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. For every 1 increase in local Ingrva, the surrounding Ingrva will increase by 0.201. The regression coefficient of ER on local Ingrva was 0.511, which was significant at 5{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. The regression coefficient of the influence of ER in surrounding areas on local ln Ingrva is 0.729, significant at 10{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. This shows that increasing environmental regulation not only improves the local green technology level, but also improves the green technology level of the surrounding areas, and green technology has the effect of diffusion to the surrounding areas.

According to Model-9, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.274, the regression coefficient of Ingrva to local pharmaceutical manufacturing total factor productivity is 0.203, and the regression coefficient of Ingrva to local pharmaceutical manufacturing total factor productivity in surrounding areas is 5.485. The effect of Ingrva on pharmaceutical manufacturing total factor productivity in the local area and the effect on the surrounding area both are positive, which indicates that Ingrva promotes the improvement of pharmaceutical manufacturing total factor productivity.

According to Model-10, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.167, the regression coefficients of ER and Ingrva to local pharmaceutical manufacturing total factor productivity are 5.826 and − 0.00991, respectively, and the regression coefficients of ER and Ingrva to local pharmaceutical manufacturing total factor productivity are 31.68 and 3.17, respectively. Except the effect of Ingrva on local pharmaceutical manufacturing total factor productivity was not significant, the other response coefficients were significant. Both ER and Ingrva act on pharmaceutical manufacturing total factor productivity, that is, peripheral ER has a direct effect on local pharmaceutical manufacturing total factor productivity and an indirect effect through the action of Ingrva.

Through the above analysis, we can get the intermediate transmission path of Ingrva: the increase of environmental regulation in the region and surrounding areas promotes the improvement of local green technology, thus promoting the improvement of pharmaceutical manufacturing total factor productivity in the region and surrounding areas, which proves that Hypothesis 3.

Table 7 represents the regression result of RD as a mediation variable. Model-11 takes ER as the core explanatory variable and RD as the dependent variable. Model-12 takes RD as the core explanatory variable and pharmaceutical manufacturing total factor productivity as the dependent variable. Model-13 uses ER and RD as the core explanatory variables and pharmaceutical manufacturing total factor productivity as the dependent variable.

According to Model-11, the spatial lag coefficient of RD is 0.275, which is significant at the level of 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. For every 1 increase in local RD, RD in surrounding areas will increase by 0.275. The regression coefficient of ER to local RD was 2.448, which was significant at 5{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. The regression coefficient of ER to local RD in the surrounding area was 3.738, which was significant at 10{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. This shows that increasing environmental regulation not only improves the local technological innovation, but also improves the technological innovation of the surrounding areas, and technological innovation has the effect of diffusion to the surrounding areas.

According to Model-12, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.309, the regression coefficient of RD to local pharmaceutical manufacturing total factor productivity is 0.304, and the regression coefficient of RD to local pharmaceutical manufacturing total factor productivity in surrounding areas is 1.077, all of which are significant. RD has a negative impact on pharmaceutical manufacturing total factor productivity in the region and has a positive the surrounding areas.

According to Model-13, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.190, the regression coefficients of ER and RD to local pharmaceutical manufacturing total factor productivity are 5.935 and 0.334 respectively, and the regression coefficients of ER and RD to local pharmaceutical manufacturing total factor productivity are 32.37 and 0.744 respectively in surrounding areas. All regression coefficients were significant. The peripheral ER not only had a direct impact on local pharmaceutical manufacturing total factor productivity, but also had an indirect impact through the role of RD.

Through the above analysis, we can get the intermediary transmission path of RD: the increase of environmental regulation in the region and surrounding areas promotes the improvement of local production technology, thus promoting the improvement of pharmaceutical manufacturing total factor productivity in the region and surrounding areas, which proves Hypothesis 3.

Table 8 shows the regression results with Structure as the mediating variable. In Model-14, ER is used as the core explanatory variable and Structure is used as the dependent variable. In Model-15, Structure was used as the core explanatory variable, and pharmaceutical manufacturing total factor productivity was used as the dependent variable. In Model-16, ER and Structure were used as core explanatory variables, and pharmaceutical manufacturing total factor productivity was used as dependent variable.

According to Model-14, the spatial lag coefficient of Structure is 0.393, which is significant at the 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add} level. For every 1 increase in the local Structure, the surrounding area Structure will increase by 0.393. The regression coefficient of the effect of ER on local Structure is 2.782. The regression coefficient of the influence of ER in surrounding areas on local Structure is 14.42, which is significant at the level of 5{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}. This shows that the environmental regulation of surrounding areas has a certain degree of impact on the changes of pharmaceutical industrial structure, while the impact of local environmental regulation is not significant.

According to Model-15, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.266, the regression coefficient of Structure to local pharmaceutical manufacturing total factor productivity is 0.178, and the regression coefficient of Structure to local pharmaceutical manufacturing total factor productivity is 0.583. This shows that pharmaceutical manufacturing total factor productivity in this area is not only affected by Structure in this area, but also affected the surrounding areas.

According to Model-16, the spatial lag coefficient of pharmaceutical manufacturing total factor productivity is 0.188, the regression coefficients of ER and Structure to local pharmaceutical manufacturing total factor productivity are 3.366 and 0.173, respectively. The regression coefficients of pharmaceutical manufacturing total factor productivity were 24.08 and 0.473 respectively. The regression coefficients of pharmaceutical manufacturing total factor productivity were significant except for the effect of ER on the local pharmaceutical manufacturing total factor productivity.

Through the above analysis, we can get the intermediary transmission path of Structure: the improvement of environmental regulation in surrounding areas promotes the upgrading of local pharmaceutical industrial structure, thus promoting the improvement of local pharmaceutical manufacturing total factor productivity, which proves that Hypothesis 3.

#### Heterogeneity analysis

Due to the differences of temporal and spatial variation characteristics of ER and pharmaceutical manufacturing total factor productivity in the eastern, central and western regions, the spatial spillover effects of ER on pharmaceutical manufacturing total factor productivity will also be different in different regions. Therefore, this paper takes into account the spatial agglomeration and geographic location heterogeneity of provinces in different regions, establishes a spatial Durbin model, and discusses the impact of ER on pharmaceutical manufacturing total factor productivity in the eastern, central and western regions.

Table 9 is the regression result of heterogeneity analysis. Model-7 is the regression result of the whole country, which is used as the control group of heterogeneity analysis. Model-17, Model-18 and Model-19 are the regression results of the eastern, central and western, respectively.

According to the spatial lag coefficient of pharmaceutical manufacturing total factor productivity, the influence degree of pharmaceutical manufacturing total factor productivity in the surrounding areas on local pharmaceutical manufacturing total factor productivity is only significant in the central.

According to the regression coefficient of the impact of local ER on local pharmaceutical manufacturing total factor productivity, three regions are all not significant. From the regression coefficient of the impact of ER on local pharmaceutical manufacturing total factor productivity in the surrounding areas, only the impact of eastern (53.63) is positive and significant at the level of 1{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}.

To sum up, the spatial spillover effect of the three regions is not obvious by analyzing the heterogeneous characteristics of the three regions alone. This further verifies the necessity of taking 30 provinces in China as a whole as the research object.

#### Endogenous discussion

Although the use of spatial econometric model can better study the spatial spillover effect of ER and pharmaceutical manufacturing total factor productivity, there may be endogenous problems caused by the omission of variables and the results of bias, so this paper uses GS2SLS spatial econometric tool variable method to alleviate the endogenous problems that may exist in the model.

Based on Hering and Poncet (2014), this paper uses *Ventilation* as the instrumental variable of environmental regulation^{79}. According to Jacobson (2003), the air flow coefficient is equal to the product of the boundary layer height and the wind speed^{80}. In this paper, based on the global network of ten meters wind speed and boundary layer height data in the ERA-Interim database of the European Center for Medium-Range Weather Forecasts, the air circulation coefficient of each network in the corresponding year is calculated, and then the air circulation coefficient of each province is obtained according to the longitude and latitude matching of each provincial capital city.

When air pollutant emissions are the same, cities with low air ventilation coefficient tend to use more stringent environmental regulation tools. The calculation process of environmental regulation itself includes environmental pollution, so it can be considered that there is a correlation between environmental regulation and air circulation coefficient. Moreover, the air circulation coefficient only depends on natural phenomena such as climate conditions, and there is no other mechanism with the total factor productivity of the pharmaceutical industry, so the air circulation coefficient as an instrumental variable, which has exogeneity.

Appendix Table 5 is the result of the GS2SLS instrumental variable method. From the results, Ventilation coefficient and environmental regulation (ER) are significantly negative at the level of 10{18fa003f91e59da06650ea58ab756635467abbb80a253ef708fe12b10efb8add}, with a coefficient of − 3.165. The results are in agreement with the theoretical expectation.

The spatial lag coefficient of ER is 0.0842, but the result is not significant, which indicates that local environmental regulation is endogenous, while the environmental regulation of surrounding areas is not endogenous, indicating that there is no two-way causal relationship between ER of surrounding areas and local pharmaceutical manufacturing total factor productivity.

#### Robustness test

In order to test the robustness of the model established in this paper, the following methods are used to compare the robustness of the results by replacing the control variables and the spatial weight matrix. Appendix Table 6 shows the results of the robustness test. Model-7 was the control group, Model-21 replaced the control variable Open with Open2, Model-22 replaced the control variable Capital with Capital2, and Model-23 replaced the 0–1 adjacency matrix (W1) with the geographical distance matrix (W2). As a result, the magnitude and significance of the regression coefficients changed only slightly, but not in direction. Therefore, the spatial econometric regression results obtained in this paper are robust.

In addition, the results reported in this article are the absolute form of all variables. In order to better verify the robustness of the model, the results of all logarithmic and mixed variables are reported in the appendix (Appendix Tables 1–5). These results have great consistency in model selection and the direction of main variables, but there are some differences in the significance of variables.