The progress in technology and economy has greatly improved the living standards of human beings. However, many issues involving resources shortage and environmental pollution appear synchronously, threatening the sustainable development of human beings and the earth. Global warming is one of the major environmental problems worldwide. The accumulation of greenhouse gases including CO₂ has aroused widespread concern with the aggravation of global warming. In particular, CO₂ emissions have been recognized as the one of the main challenges of global sustainable development. In the light of the statistical data from World Bank, in the 54 years after 1960, total global CO₂ emissions increased by approximately quadruple, that is, from 939.67million tons in 1960 to 3613.83 million tons in 2014¹. Per capita CO₂ emissions have increased by about 1.6 times, i.e., from 3.10 metric tons in 1960 to 4.97 metric tons in 2014. In the future, CO₂ emissions will continue to increase with the economic development of developing countries^{[1, 2]}. Excessive growth in CO₂ emissions leads to global warming and climate imbalance, resulting in various risks and economic losses including disasters, diseases, and economic damages, which impedes sustainability in economic, society and environment^[3-5]. Mitigating CO₂ emissions is considered as an urgent task in responding to global climate change and sustainability. According to the point view of Stern^[6], if no action is taken to mitigate CO₂ emissions, 5% of the global average GDP will be lost for the overall cost of global warming. Therefore, it is urgent to carry out carbon reduction action on a global scale. To effectively achieve it, it is of great importance to have in-depth investigation on the determinants of CO₂ emissions from cross-country perspective, which may directly affect CO₂ emissions reduction measures, policies and strategies^{[7, 8]}.

¹https://data.worldbank.org.

Many researchers have empirically studied the determinants and key impact factors of CO₂ emissions from different levels including household, industrial, regional, national and global level. For instance, Zhang, et al.^[9] systemically analyzed the influencing factor of household carbon emissions. At the industrial level, Zhao, et al.^[10] explored the major driving factors of CO₂ emissions in China's power industry. Xu and Lin^[11] identified the driving forces of Chinese iron and steel sectors' CO₂ emissions. Ren, et al.^[12] examined the impact factors of CO₂ emissions of China's 19 industry sectors. When it comes to regional level, Wang, et al.^[13] empirically studied what factors affect CO₂ emissions in Beijing. Song, et al.^[14] examined the driving factors of carbon emissions in the Yangtze River Delta, and the findings show that economic scale is the main factor leading to the carbon emissions. Behera and Dash^[15] turned their interest into the South and Southeast Asian region and examine the relationship between urbanization, energy consumption, foreign direct investment and CO₂ emissions. At the national level, Chang^[16] empirically examined the relationship among economic growth, CO₂ emissions and energy consumption in China. Hossain^[17] investigated the influencing factors of CO₂ emissions of newly industrialized countries. Xu, et al.^[18] analyzed the driving factors of CO₂ emissions in different stages in China. Using the panel and time-series data, Shuai, et al.^[19] explored the key impact factors of CO₂ emissions in China among 43 potential impact factors. As for the global level, using multi-national panel data, Sharma^[7] empirically investigated the determining factors of CO₂ emissions. Combining the STIRPAT model and panel data of 125 countries, Shuai, et al.^[8] analyzed the effects of population, affluence and technology on the CO₂ emissions.

Nonmatter what level researchers focus on, it can be concluded that there are two strands about existing literatures. The first strand is to test the causality between variables and CO₂ emissions through Granger causality^[20-22]. The second strand of the empirical studies focus on probing single factor or multiple factors that influencing CO₂ emissions^[23-25]. In the second strand, most empirical works attempts to demonstrate the influencing factors of CO₂ emissions based on IPAT or STIRPIT approach and population, technology and economic growth are recognized as the main driving factors of CO₂ emissions^[26-28]. It is noteworthy that, when identifying the influencing factors of CO₂ emissions, researchers usually take industrial structure into consideration.

Recently, researchers gradually show interest in the impact of industrial structure on CO₂ emissions. Some empirical works conclude that the share of the secondary industry has a significantly positive impact on CO₂ emissions, and the expansion of almost all industries will induce CO₂ emissions, but comparing to the secondary industry, the strength of the tertiary industry is relatively lower^{[29, 30]}. According to this statement, industrial structure change is a significant potential for CO₂ emissions reductions. A majority of empirical studies have demonstrated this point of view^[31-34].

However, most researchers examine the industrial structure change and CO₂ emissions relationship at the national level^[34-37]. Those studies usually measure industrial structure with industry added value or the ratio of industry added value, which show little policy references. In fact, industrial structure change is an external phenomenon of economic growth, and its internal motivation is the process of specialization and refinement of manufacturing and service industries. Therefore, it is more applicable to measure the industrial structure change with linkage between manufacturing and service sector.

In this paper, we consider the inherent laws of CO₂ emissions with economic development from the perspective of industrial structure change and propose a theoretical framework that how industrial structure impacts CO₂ emissions. Under this framework, we test the impact of increasing linkages between manufacturing sector and service sector on CO₂ emissions. Based on the 41 countries' I-O table data, panel model is performed to obtain the correlation of industrial structure change to CO₂ emissions. The empirical findings are intended to provide a more operational and guiding reference for the formulation of emissions reduction policies.

This paper contributes to the existing literature in two aspects. On one hand, we innovatively introduce the linkages between manufacturing sector and service sector to reveal the inherent motivation of industrial structure change. It proposes a new perspective to understand industrial structure change, which can give theoretical reference on the related studies about industrial structure change. On the other hand, new evidence on the relationship between industrial structure change and CO₂ emissions is provided at multi-national level by using panel data of 41 countries over 2000–2014. The empirical findings provide valuable and applicable policy implications for policymakers.

The reminder of this paper is arranged as follows. In Section 2, we propose a theoretical framework of industrial structure-${\mathrm{C}\mathrm{O}}_2$ emissions. In Section 3, empirical model is introduced. At the same time, we give an overview of data source and descriptive statistics. In Section 4, we make the empirical analysis and discuss the findings. In the last section, we conclude the study and provide policy implications.

Numerous empirical studies investigate the relationship between economic growth and CO₂ emissions, and the mainstream theoretical and empirical works assert that the relationship between economic growth and CO₂ emissions follows an inverted-U environmental Kuznets curve (EKC)^[38-40]. It indicates that, in the beginning, economic development will increase CO₂ emissions. However, after a turning point is reached, the growth of CO₂ emissions gradually decreases with economic development. On basis of the viewpoint of Chenery^[41], economic growth is a set of changes in the production structure of different industries and economic activities. Industrial structure has been valued by many economists given its cumulative linkage mechanism to economic development. It can be concluded that external performance of economic development is the dynamic adjustment and optimization of the industrial structure. Based on this fact and EKC postulation, we established a theoretical framework of economic development, industrial structure and CO₂ emissions as illustrated in Figure 1.

In the early stage, a nation's economic development is driven by traditional manufacturing industries. In the meanwhile, some service industries become independent and specialized in correspond with the development of manufacturing industries. Subsequently, the manufacturing industries of primary processing and resource utilization become an engine for economic development. Therefore, in the initial stage, the expansion of manufacturing undoubtedly leads to a rise in energy consumption, resulting in a rise in CO₂ emissions. However, due to the backward technology and small population size, the growth rate of CO₂ emissions during the period is relatively low.

In the following stage, different industries emerge, and scale of industry is expanded. Some corresponding service functions achieve specialization, forming a more targeted service sector. At this time, the service industry is constantly integrating with the modern service industry on the basis of the traditional service industry, which leads to functional diversity and specialization. At the same time, the production become increasingly complicated, and the industrial chain is continuously refined and extended. With the increase in population size and expansion of industries, CO₂ emissions at this stage increase rapidly.

When economic development approaches the peak, technology level and innovation capacity is significantly enhanced, and high-end manufacturing began to develop. With the growth of the high-end manufacturing industries, service functions become more specialized, independent and targeted. In this stage, the economy enters a mature stage which is dominated by the service sector. In this context, technological progress leads to a significant decline in energy consumption. The stability of energy-intensive industries and population also contribute to the decrease in the growth rate of energy consumption. Those jointly contribute to the slow, even negative growth of CO₂ emissions.

From the theoretical perspective, in different stages of economic development, it differs for the CO₂ emissions and linkage between manufacturing industry and service industry. The CO₂ emissions increase with economic development in the beginning and then decrease in the mature stage. The reason is that, in the mature stage, economic growth is dominated by high-end manufacturing industries and their corresponding service industries, where the producer service sector directly serves the manufacturing industry. That is to say, there is a close linkage between manufacturing sectors and service sectors, which will achieve a slowdown in CO₂ emissions growth.

According to the theoretical analysis above, the industrial structure change is the essence of economic development. The linkages between manufacturing and service sector well reveals the degree of industrial structure change. Thus, it is of great significance to study the relationship between industrial structure change and CO₂ emissions by using the indictor of linkages of manufacturing and service sector.

The main purpose of our study is to examine how industrial structure change affect CO₂ emissions. To estimate this impact, the basic empirical equation is modeled as follows:

where $t$ means the time in years and the subscript $i$ is the countries, $PC{E}_{it}$ denotes CO₂ emissions per capita (in metric tons). ${\mathrm{P}\mathrm{G}\mathrm{D}\mathrm{P}}_{it}$ is the GDP per capita. ${\mathrm{T}\mathrm{E}\mathrm{C}}_{it}$ is the technology level, measured by the number of patents applied by residents. ${\mathrm{E}\mathrm{N}\mathrm{I}\mathrm{N}\mathrm{T}}_{it}$ stands for energy intensity. ${\mathrm{U}\mathrm{R}\mathrm{B}\mathrm{N}}_{it}$ represent urbanization level. ${\mathrm{F}\mathrm{D}\mathrm{I}}_{it}$ and ${\mathrm{T}\mathrm{O}}_{it}$ represents foreign direct investment and trade openness, respectively. $c$ is the intercept and ${\epsilon }_{it}$ indicates error term. ${\mathrm{M}\mathrm{S}\mathrm{L}}_{it}$ is the industrial structure and measured by direct forward linkage of manufacturing sector and service sector, which is calculated by the formula as follows.

where $x_{MS}$ represents the intermediate demand of manufacturing sector' products by service sector, $x_M$ denotes the gross output of manufacturing sector. The definition of all the variables is listed in Table 1.

This study used a balanced panel of 41 countries over the period 2000–2014. All the annual data except for the direct backward linkage indicator is collected from the Database of World Bank. The data to calculate the direct forward linkage of manufacturing sector and service sector is collected from the National Input-Output Table published by the World Input-Output Database in 2016. The World Input-Output Database releases I-O table of 43 countries (regions). We drop Cyprus and Taiwan from the sample due to data integrity and consistency. The descriptive statistics are summarized in Table 2.

As the descriptive statistics show, the mean of per capita CO₂ emissions in 2000–2014 is 8.11 metric tons and the standard deviation is 4.47, indicating large differences among different countries in different years. As for control variables, the high standard deviations indicate large national disparities in terms of social-economic and technological development.

As the descriptive statistics show, the mean of per capita CO₂ emissions in 2000–2014 is 8.11 metric tons and the standard deviation is 4.47, indicating large differences among different countries in different years. As for control variables, the high standard deviations indicate large national disparities in terms of social-economic and technological development.

Before performing the panel model, the LLC^[43] and IPS^[44] unit root test method are executed on data to examine the stationarity of data. Table 3 presents the results of LLC and IPS unit root test. For the variable of per capita GDP and technology, we take their natural logarithm (INPGDP and INTEC) in order to avoid drastic fluctuations of the data. Form the test results, it can be seen that all the variables are stationary and there is no need to change their form when perform the model.

Multicollinearity is one of the most crucial issues should be taken into consideration when performing regression model as it could make the estimate distorted. Considering the potential multicollinearity among the explanatory variables, we calculate the Pearson correlation coefficients of the major variables. The results are reported in Table 4. As can be observed from the correlation coefficient matrix, the biggest correlation coefficients between independent variables is 0.6994 ($p<0.05$). We further calculated the variance inflation factor (VIF), the VIF of each variable is less than 10, indicating multicollinearity is actually not so much of a problem.

According to the basic model, we proceed with regression analysis to examine the effect of industrial structure change on CO₂ emissions based on the panel data of 41 countries from 2000 to 2014. Three models are estimated including the pooled ordinary least squares (pooled OLS), fixed effects regression (FER) model and random effects regression (RER) model. To decide which model should be adopted, a series of tests including $F$ test, $LM$ test and Hausman test are carried out. The test results and comparisons are shown in Table 5. The results of $F$ test and $LM$ test indicate that fixed effects model and random effects model is better than the pooled ordinary least squares, respectively. The Hausman test result shows that the fixed effects model performs better than the random effect model. Therefore, we mainly use the results of the fixed effects model. The empirical results of this study are reported in Table 6.

In Model 1, the estimate result of fixed effect model shows that industrial structure change has a statistically significantly negative effect on per capita CO₂ emissions, indicating that industrial structure change contributes to CO₂ emissions reduction. More specially, 0.1 unit increasing in industrial structure change will lead to a decrease of approximately 0.94 metric tons per capita CO₂ emissions, ceteris paribus. The finding also reveals that the stronger linkage between manufacturing sector and service sector, namely, the more products of manufacturing sector flow to service sector, the less are CO₂ emissions. This result support the findings of Zhou et al.^[37] and Cheng & Liu^[45], which provides evidence on the importance of industrial structure change on the development of a low-carbon economy, showing that an effective way to mitigate CO₂ emissions is adjusting the industrial structure.

We have also found that urbanization, technology and trade openness have statistically significantly negative impacts on the CO₂ emissions. The result is consent with the study of Sharma^[7] and Hossain^[17].What is more, the estimate coefficient of technology, urbanization and trade openness are $-$0.3864, $-$0.0818 and $-$0.3864, respectively, suggesting that technology plays a greater role in mitigating CO₂ emissions compared to urbanization and trade openness. Technical progress is commonly regarded as an important means for the improvement of environmental quality. The estimate result suggests that 1% increase in technology will contribute to 0.38 metric tons per capita CO₂ emissions reduction, ceteris paribus. In terms of urbanization, for each one unit increase in urbanization rate, we found that a nearly 0.08 metric tons decrease in per capita CO₂ emissions, holding fixed all other factors affecting CO₂ emissions. Albeit urbanization has brought tremendous pressure on resource management globally^[46], it tends to form stricter environmental standards and policies with the expansion of urbanization. Thus, the pollution and energy waste are controlled, resulting in less CO₂ emissions, which contribute to sustainability in a long term^[47]. The estimate result presents that trade openness plays an effective role in decreasing environmental degradation as well. The reason might be that, the economic and financial openness aroused by trade openness stimulate enterprises in the action of energy conservation and emissions reduction.

The other findings are that per capita GDP and energy intensity have significantly positive effects on CO₂ emissions, indicating that CO₂ emissions will be stimulated by increased per capita income and energy consumption. The finding is in line with the empirical results of Chang^[16] and Zhou & Liu^[25]. According to Hossain^[17], in a long term, the increased energy consumption would result in more CO₂ emissions, indicating that it is of great significance to improve efficiency of energy utilization. In the meanwhile, Duan et al.^[48] point out, to deal with global CO₂ emissions, energy efficiency improvement is an essential measure. Economic growth is regarded as one of the contributors to environmental pressure. Our empirical results present that 1% increase in per capita income will contribute to approximate increase in 8.6 metric tons per capita CO₂ emissions. These findings imply that itis floundering for government to carry out ambidextrous policies due to the positive relationship between GDP growth and CO₂ emissions. Pursuing economic growth will result in more CO₂ emissions, and the implementation of energy conservation policies will impede economic growth. It is challenging for policymakers to set policies and strategies aiming to balancing the economic development and CO₂ emissions.

Nevertheless, our results clearly show that foreign direct investment has statistically insignificant effect in CO₂ emissions. This finding is inconsistent with the Porter hypothesis, which states that foreign direct investment can reduce CO₂ emissions as it facilitates technical innovation. To further demonstrate the Porter hypothesis, we investigate whether the role of foreign direct investment in CO₂ emissions is moderated by technology. In model 2, we introduced the interaction variable of FDI*INTEC. The estimate result shows that there is no interaction effect between foreign direct investment and technology, implying that our data did not support the Porter hypothesis.

In model 3, we examined the EKC hypothesis by introducing the variable of INPGDP*INP GDP into the empirical model and have found that the EKC hypothesis is supported by our empirical data. In light of the EKC hypothesis, there is an inverted-U relationship between per capita income and environmental degradation^[39]. Thus, the expected sign of the coefficient of per capita GDP is positive and the sign of the coefficient of its square is negative. Our empirical result presents a positive relationship between per capita income and CO₂ emissions, and a negative relationship between the square of per capita income and CO₂ emission. It provides the evidence on the existence of EKC hypothesis.

In the theoretical framework, we postulate that technical progress triggers the emergency of high-end manufacturing and a higher linkage between manufacturing sector and service sector. Thus, the effect of industrial structure change on CO₂ emissions might be moderated by technology level. To further investigate the mechanism of industrial structure change affecting carbon emission, we examine the interaction effect of technology and industrial structure on CO₂ emission by including the interaction of MSL and INTEC. The estimate result is reported in Table 7 as Model 4. The coefficient of interaction term is statistically significant and positive, indicating that reductive effect of industrial structure change on CO₂ emissions declines as the technology level improves. The advanced technology contributes to CO₂ emissions reduction improves economic productivity and leads to industrial transformation. However, the effect of technology is limited and determined by its orientation. Despite it is highly endorsed that environmental issues can be alleviated by development green technology related to new renewables and sustainability, information and communications technologies are dominated in the present era. Our moderation effect of technology is constant with the finding of Moyer and Hughes^[49], which argue that there is a downward impact of the information and communications technologies on environment degradation on the whole. Moreover, we examined the moderation effect of energy intensity on the relationship between urbanization and CO₂ emissions. The urbanization is found to have a positive impact on CO₂ emissions reduction. Nevertheless, the process of urbanization would lead to energy consumption. In model 5, we introduce the interaction term RURB*ENINT and the empirical result presents a statistically significant and positive coefficient. Combining the negative coefficient of urbanization, the finding suggests that with the energy use increase, the effect of urbanization on environment degradation will decrease. Economic growth and energy consumption are regards two main CO₂ emissions contributors. In model 6, we investigate the interaction effect of per capita income and energy intensity on CO₂ emissions by adding variable of INPGDP*ENINT on to basic model. The result suggests that energy use strengthens the effect of income on CO₂ emissions.

To check the robustness, we add the control variables into the basic model in turn and replace the variable of industrial structure (MSL) with the variable INSTR (the ratio of output of tertiary industry to that of secondary industry) as well. The robustness check results are reported in Table 7. In addition, we check the robustness of interaction effects by repetitive random sampling. Table 8 reports the check results of selection on 80% of the total sample. All the results suggest that the robustness is verified. Therefore, we conclude that the effect of industrial structure change is robust, indicating the adjustment and upgrading of industrial structure is an effective way to realize economic restructuring and CO₂ emissions reduction, promoting the sustainable development of the region.

This paper investigates the impact of industrial structure change on CO₂ emissions based on a panel data of 41 countries from 2000 to 2014. The empirical study shows new evidence on the relationship of industrial structure change and CO₂ emissions. Our main findings are that industrial structure change significantly negatively affects CO₂ emissions, which implies the upgrading of industrial structure contributes to CO₂ emissions reduction and sustainable development. Additionally, we examine the impacts of per capita income, urbanization, energy use, technology, trade openness and foreign direct investment on CO₂ emissions. It is noted that per capita income and energy use have significantly positive effects on CO₂ emissions; on the contrary, technology and trade openness play significantly negative roles. The following are the main policy implications that could be raised from our empirical study.

First, we have found that industrial structure change contributes to less CO₂ emissions, which implies that there is a need to implement more industrial structure adjustment policies so as to restructure the energy system and reduce CO₂ emissions in the long term. To achieve industrial structure change, it is necessary to encourage and incentive the development of service sector, which will pull the transition of manufacturing sector from traditional manufacturing to high-end manufacturing. Second, the increasing of urbanization contributes to less CO₂ emissions, which indicates that country should consider accelerating the process of urbanization. Urbanization can improve the public infrastructure utilization and promote the formation of stricter environmental standards. However, duo to the complicated determents of environment quality, sterner environmental and energy policies should be paralleled.

We provide theoretical analysis and new empirical evidence on the effect of industrial structure change on CO₂ emissions. However, this study has some shortcomings which worth further and more sophisticated studies in the future. Firstly, restricted by the input and output table derived from World I-O Database, we cannot investigate the effect of industrial structure change on CO₂ emissions from different income levels. Secondly, given the availability of sectoral data, further research can make in-depth analysis on the evolution of manufacturing and service sector, discussing its contributions to CO₂ emissions to deeply understand the nexus of industrial structure change and CO₂ emissions.