Role of low-carbon technology innovation in environmental performance of manufacturing: evidence from OECD countries

Climate change disrupts the balance of natural ecosystems and threatens the sustainable development of human society. As the leading industry in many countries, manufacturing promotes economic growth; unfortunately, it also emits large quantities of greenhouse gases. Thus, it is necessary to transform the production pattern of manufacturing into green production. Although technology innovation is the only way to tackle the issue, different types of technology innovation may lead to various environmental performances. We argue that low-carbon technology innovation (LCTI) is the key to green production. Using data of Economic Co-operation and Development (OECD) countries from 1990 to 2014, we use the patent-stock method to measure LCTI levels and analyze its development trend in OECD countries. Based on the shepherd distance function, we measure carbon efficiency and carbon productivity by the fixed-effect Stochastic Frontier Analysis (SFA) model. Then we investigate the effect of LCTI on carbon emission efficiency in manufacturing by the fixed-effect regression model. After controlling some variables, evidence shows a significant positive influence of LCTI on the environmental performance of manufacturing. The level of LCTI constantly increased in OECD countries during the study period. Among these countries, the level of low-carbon technology in the chemical industry is the highest; in most countries, the low-carbon technology of the production process grows fastest. Policy implications are further discussed.


Introduction
For decades, evidence shows that the global climate is growing warmer, and the cause is greenhouse gas (GHG) emissions (Midilli et al. 2006). Global climate warming disrupts the balance of natural ecosystems and threatens the sustainable development of human society (Tol 2009). Getting to the core of the issue, the Intergovernmental Panel on Climate Change (IPCC) pointed out that human behaviors, especially manufacturing activities, largely contribute to global climate change (IPCC 2007). Manufacturing industry, the leading industry in most countries, not only promotes the social economy but also emits large amounts of GHG, being the main factor contributing to climate change (Fysikopoulos et al. 2014). With the deterioration of the global climate environment, a low-carbon economic development pattern is increasingly attracting attention from countries the world over. GHG reduction of manufacturing has become the key to realizing a low-carbon economy (Chen et al. 2017b). Furthermore, national governments should try to control manufacturing GHG emissions while simultaneously promoting economic growth. In this sense, how to realize the dual goals of GHG reduction and continuous development of manufacturing output has attracted the attention of scholars and policymakers.
Carbon dioxide gas is an important component of greenhouse gases, and carbon efficiency is an essential indicator for measuring a low-carbon economy widely used by scholars (Beinhocker et al. 2008;Du and Li 2019;Lin and Du 2015;Yan et al. 2020). Beinhocker et al. (2008) pointed out that improving carbon efficiency is the key to reducing global Responsible editor: Roula Inglesi-Lotz carbon emissions. Furthermore, we can realize the goal of reducing carbon emissions by 50% in 2050 if carbon efficiency increases by at least tenfold of the 2005 levels. Some scholars think that carbon efficiency is a core criterion of low-carbon economic development (Jiankun and Mingshan 2011;Li et al. 2018). Improving carbon efficiency means that we can realize GHG reduction and the promotion of economic growth of manufacturing simultaneously (Zhang et al. 2018a). Hence, improving the carbon efficiency of manufacturing is a critical issue. It is crucial to measure the carbon emission efficiency of manufacturing sectors and find out its influencing factors, which would not only be beneficial for dealing with the global climate change but also for realizing the lowcarbon transformation of manufacturing sectors.
Today, society is looking for various ways to improve the carbon efficiency of manufacturing, such as institutional and organizational innovation (Zhang et al. 2018b;Sun et al. 2019), residents' demands, and energy use standards (Paksoy and Ozceylan 2014). While technology innovation has been widely accepted as an effective way to reduce GHG emissions, it is also the fundamental way to increase carbon efficiency (Du and Li 2019;Fan et al. 2021;Popp 2012;Yan et al. 2020). Furthermore, because technology innovation was one of the three topics of the United Nation's climate change conference, COP24, policymakers pay special attention to the role of technology innovation in green transformation. The Porter Hypothesis states that stringent environment regulation can stimulate carbon reduction technology innovations and achieve the win-win situation of economic growth and carbon reduction (Porter 1991). However, it is worth noting that different types of technology innovation may demonstrate different environmental performance. Theoretically, environment-friendly technology innovation (or low-carbon technology innovation) would promote carbon efficiency, while environment-unfriendly technology innovation (or high-carbon technology innovation) would inhibit carbon efficiency. Zhang et al. (2017a) believes that lowcarbon technology innovation (LCTI) is the key to reducing GHG emissions. Compared with traditional environmental regulation, LCTI significantly improves carbon productivity by full energy utilization and end-of-pipe treatment technologies, industry structure upgrading, and human capital improvement (Du and Li 2019). In addition, LCTI might reduce the cost of mitigating GHG emissions, which is a crucial potential benefit (Popp 2012). Nevertheless, to the best of our knowledge, current literature primarily studies the LCTI of economies or regions and seldom discusses the LCTI of sectors. Furthermore, very few previous studies investigated the role of LCTI in the green production of manufacturing sectors.
Therefore, based on the above, we raise the following questions: What is the role of LCTI in the environmental performance of manufacturing? Moreover, what is the level of LCTI in manufacturing industry? To solve these issues, the goals of the work are twofold. First, we aim to evaluate the level of LCTI in manufacturing and explore its development trend, which may supplement the referring literature about LCTI indicators. Second, existing studies focus on the relationship between low-carbon technology innovation and environmental performance in regions or economies. This study excluded the interference of industry heterogeneity and investigated the impact of LCTI on the carbon efficiency of manufacturing, providing more stable empirical support for the previous results.
The study is structured as follows: the "Introduction" section introduces the study's background. The "Literature review" section briefly reviews the relevant literature. The "Method and data" section presents the methods of measuring LCTI and carbon efficiency: analysis models and data source. The "Results" section provides descriptive analysis on the development trend of LCTI and discusses the results of the regression. The "Discussion and policy implications" section concludes and provides policy implications to the analyses and future direction of the work.

Literature review
A body of literature has investigated the impact of technology innovation on carbon emission efficiency. The current study reviews the relevant literature from three aspects: the measurement of LCTI, the measurement of carbon emission efficiency, and the influencing factors of carbon emission efficiency.
First, existing studies have investigated the measurement methods of technology innovation, while in reality, it is challenging to measure technology innovation levels directly. From an input-output production perspective, three indicators, research and development investment (R&D) data (Zhang et al. 2017b), patent data (Johnstone et al. 2010;Yan et al. 2017), and total-factor productivity data (Keller 2010), are commonly used to estimate technology innovation levels indirectly (Wang 2017). However, no perfect methods exist, and each measurement indicator has both advantages and disadvantages (Popp 2012). Not all inventions can be patented in reality, and the quality of these patented inventions remains uneven, illustrating that patent data are not the perfect measurement of technology innovation (Griliches 1998). Nevertheless, among these indicators, the patent is the only indicator that provides adequate micro-information available for researchers to subdivide the research field in detail (Wang 2017). For this reason, the patent can be an appropriate measurement of LCTI, and it can promote the empirical research of LCTI (Dechezlepretre et al. 2011). In addition, due to the accessibility of patent data and further exploration by researchers, the impact evaluation of technology innovation has gone deeper into different areas, which provides an appropriate indicator for many econometric analyses and especially for a cross-regional comparative analysis; thus, patent data can be compared to the international standard indicator (Haščič et al. 2015).
Second, the majority of the existing studies considered carbon efficiency or carbon productivity as the key indicators of green production (Du and Li 2019;Yan et al. 2020;Zhang et al. 2018a). At present, there are mainly two carbon emission efficiency indicators: single-factor indicator and total-factor indicator. Kaya and Yokobori (1997) initially defined single-factor carbon efficiency as the ratio of GDP to carbon dioxide emissions. However, the single-factor did not consider other factors; besides, it cannot reflect the underlying technology efficiency, energy substitution effects, and other production factors. Under these circumstances, total-factor carbon efficiency has gradually been applied and can effectively overcome the shortcomings of single-factor indicator . Moreover, total-factor carbon efficiency is the static carbon emission performance, while total-factor carbon productivity can reflect the dynamics of carbon emissions performance (Lin and Du 2015). There are currently two measurement methods of total-factor carbon efficiency: Data Envelopment Analysis (DEA) and Stochastic Frontier Analysis (SFA). DEA is a non-parameter method with no restriction in function form and is easily affected by sample data quality . Considering that macroeconomic data tend to have big noise, the SFA method was suggested by some scholars because it divides the decision-making unit (DMU), deviating from the technical frontier into the efficient part and the random error part. Meanwhile, SFA can eliminate data noise while calculating the efficiency value . Wang and Ho (2010) thought that the traditional SFA regarded individual heterogeneity as an inefficient factor, reducing the accuracy of the result, while the improved fixed-effect SFA method not only eliminates data noise but also separates individual heterogeneity from the inefficient part.
Third, there is a string of literature focusing on the influencing factors of carbon efficiency and carbon productivity. Two main methods were utilized to study the influencing factors of carbon emission efficiency. The first type is the decomposition method, namely the production theory method, logarithmic mean division index (LMDI) method, or Malmquist index method (Chen et al. 2017a;Jiankun and Mingshan 2011;Hu and Liu 2016;Li and Cheng 2020;Sueyoshi et al. 2019;Yu et al. 2017;Zhou and Ang 2008). The second method is the econometric analysis method (Cole et al. 2013;Li and Wang 2019;Yan et al. 2020;Yin et al. 2015;Zhang et al. 2018a;Du and Li 2019). Although the methods are different, most of the studies suggested that technology innovation is the critical factor contributing to the environmental performance of production (Yin et al. 2015;Yu et al. 2017;Zhai and An 2020;Zhou et al. 2019). Furthermore, the relationship between climate and environmental innovation has been identified (Aldieri and Vinci 2020; Bai et al. 2019;Du and Li 2019;Peng et al. 2020;Yan et al. 2020). Such that Du and Li (2019) use panel data that included 71 economies to test if the effect of green technology innovation on carbon productivity is significant for economies with high income and not significant for less developed economies. Yan et al. (2020) used partially linear functional-coefficient models to investigate the effect of renewable technology innovation on green productivity and find that the significance of the impact depends on the relative income level of China's provinces. However, most existing studies focus on the role of green technology innovation from an economic or regional view; few studies mention a specific industry, especially manufacturing. Furthermore, the role of LCTI in the green production of manufacturing sectors was seldom discussed in depth. Therefore, this study aims to study the development trend of LCTI and investigate the role of low-carbon technology innovation in the green production of manufacturing and find the paths to improving carbon emissions efficiency in manufacturing.

Shepard carbon distance function
Based on the study of Zhou et al. (2010), the Shephard carbon distance function is applied in the current study to measure carbon efficiency. Indeed, the Shephard carbon distance function is the type of non-radical directional distance function. Moreover, the Shephard carbon distance function has the advantage that this function does not require adjusting the proportion of desirable and undesirable outputs (Zhou et al. 2012). Hence, compared to the conventional directional distance function, the Shephard carbon distance function has higher discriminating power. The Shepherd carbon distance function can be defined as follows: where K, L, Y, and C denote the capital input, labor input, manufacturing output, and manufacturing GHG emissions, respectively; P is defined as the possible production set: The Shephard carbon distance function describes the deviation of actual GHG emissions from theoretical GHG emissions when capital and labor are kept at the same technology level. Hypothetical GHG emissions can be calculated as C/ D C (K, L, Y, C), which is denoted as C * . The total-factor carbon efficiency formula can be used to estimate static carbon efficiency and is defined as follows: Based on Shepard distance function, Zhou et al. (2010) constructs the dynamic carbon emission model by a Malmquist index, which is also called carbon productivity. Carbon productivity can reflect the dynamics of carbon emission performance and is defined as follows: where i denotes the ith DMU and t represents the period t. The dynamic change of carbon productivity from period t to t + 1 can be estimated with Eq. (4). Carbon efficiency cannot be calculated directly by Eq. (3). Following the study of Lin and Du (2015), the general Shephard carbon distance function in this study is hypothetically represented in Eq. (5); and the translog function was used to construct the specific Shephard carbon distance function form in Eq. (6). The improved model can be expressed as follows: w h e r e Y , and C represent the sample mean of K, L, Y, and C, respectively; α i refers to the individual specific effect, which stands for the unobserved technological heterogeneity; and v it indicates the random error.
According to Eq. (5), the Shephard carbon distance function is linearly homogeneous to carbon output. Therefore, Eq. (6) can be transformed as If μ = 0, u * follows half-normal distribution, and if μ ≠ 0, u * follows non-negative truncated normal distribution, and v it follows the normal distribution. According to the study of Wang and Ho (2010), Eq. (9) can be viewed as a fixedeffect SFA model. Also, according to the characteristics of the sample data in this study, the fixed-effect SFA model was the preferred model to measure total-factor carbon efficiency.

LCTI estimating method
Because the development of LCTI is a cumulative process, patent stock can be a more proper index compared to patent quantity. Thus, the LCTI index in this study is constructed based on patent stock (Yan et al. 2017). The perpetual inventory method was employed to calculate low-carbon technology knowledge stock (Bottazzi and Peri 2007;Verdolini and Galeotti 2011). The estimation methods can be summarized as: where LCT i, t represents the low-carbon technology knowledge stock of economy i in t year, PAT i, t denotes the number of patent applications related to economy i in t year, and δ is the knowledge depreciation rate. We set the initial value of knowledge stock by the following equation: where g s represents the average growth rate of patent application number in the first five years and γ is set as 0.1 according to the existing studies (Bottazzi and Peri 2007;Keller 2002;Verdolini and Galeotti 2011). Based on the data of low-carbon patent technology combining with Eq. (8) and Eq. (9), this study calculated the patent stock of low-carbon technology in manufacturing then obtained the LCTI level in manufacturing industry.
How to accurately measure the LCTI of manufacturing is one of the key processes in this study. Referring to Yan et al. (2017), we employed the newly published "climate change mitigation for production processes" patent code (Y02P), which is in the Cooperative Patent Classification System (CPC), as the identification standard of manufacturing low-carbon technology patents. The intensity of technology innovation activities could be reflected through the patent number, which can fulfill the goal of GHG reduction during the production process. Referring to the secondary classification standard of Y02P, we have selected the Y02P patents mainly applied in the production process of manufacturing to measure the LCTI of manufacturing.

Influencing factor analysis methods
Three fixed-effect panel models considering time effect are employed in this study to analyze the impact of LCTI on carbon efficiency and carbon productivity, which can be constructed as follows: where TFCE it represents the carbon efficiency in economy i during period t; MCPI it stands for the carbon productivity in economy i from period t to t + 1; lnLCT it indicates LCTI of manufacturing in economy i during period t; L. lnLCT it represents a one-lag period of LCTI; DT t refers to time fixed effect, which defines each period as a dummy variable and t − 1 dummy variables are involved in the model; u i stands for unobservable heterogeneity of manufacturing in economy i; ε it indicates the disturbance that changes with time and individuals; and X represents control variables. The control variables are specified in the following aspects: (1) Environmental regulation intensity We select EPS, EUETS, ETTG, and ENERT, where EPS represents environmental policy intensity and EUETS stands for countries joining the European Emission-Trading Scheme. The value of the participating countries is defined as 1; others are defined as 0. ETTG indicates the ratio of environmental tax to GDP. ENERT is the ratio of energy tax to GDP. Considering that regulation can affect carbon emission efficiency (Porter and Linde 1995), these variables representing environmental regulation intensity are incorporated in the model.
(2) Energy intensity and energy structure Variables MEI and MIS represent energy intensity and structure. MEI is the ratio of energy use to the output of manufacturing, representing energy use intensity. MIS is the ratio of low-energy-density industry output to highenergy-density output in manufacturing, meaning the structure of manufacturing industries. All these variables can reflect the energy consumption of manufacturing, and the coal-based consumption would hinder environmental improvement . Thus, it is necessary to control the impact of energy intensity and energy structure.
(3) Investment Variables INV and GDPK represent an investment; INV is the ratio of gross capital information to GDP, representing investment level; GDPK is the unit capital of GDP, representing capital intensity. The effect of investment on technology development has been identified (Bosetti et al. 2011). Therefore, we incorporate the investment variable in the model.

(4) Human capital
Variable HC represents human capital. Human capital facilitates learning and knowledge sharing among employees. As knowledge is part of innovation, human capital is beneficial for innovation (Ma et al. 2019). Therefore, we incorporate the human capital variable in the model.

(5) Trade level
Variable TRADE stands for the ratio of total importsexports to GDP, which indicated the foreign trade level. Zhang et al. (2018a) pointed out that foreign trade can influence carbon productivity. Therefore, it is necessary to include the trade level variable in the model. Variable lnGDP is the logarithm of the gross domestic product, and GDPK is the unit capital of gross domestic product, representing economic development. Current evidence suggests that the economic development level of regions influences the environmental performance of technology innovation (Yan et al. 2020). Hence, we incorporate the economic development of economies in the model to control the impact.

(7) Government participation (denoted as GOV)
Variable GOV refers to the ratio of government consumption to GDP, representing government participation . As governmental behavior can be a booster for the green transformation of manufacturing industry (Grossman 1988), it is necessary to control the impact of government behavior in the model.

Data sources
The data used in this study are collected from the World Input and Output Database (WIOD), including input and output of manufacturing in 28 OECD countries during the years 1995 through 2014. OECD countries have been among the highest growth economies in the world over the last three decades, rapid economic growth of these countries is associated with the problem of excessive GHG emissions. The selected OECD countries contain the top greenhouse emitters, such as the USA, Germany, and Japan. And the data from CO 2 Emissions from Fuel Combustion (IEA 2020) shows that the greenhouse gas emissions of OECD countries account for 34.7% in world in 2018 (IEA 2020). Hence, the production activities of manufacturing in these countries have played a vital role in greenhouse gas mitigation, and the selected sample in the current study is of research significance. The statistical description of variables is shown in Table 1.
The data of the LCTI patent published by OECD statistics from 1990 to 2014 are obtainable; however, the 2014 specific data of the low-carbon technology patent of the production process are missing. So, for convenience, LCTI data of specific manufacturing from 1990 to 2013 were used for analysis in this study. The data of K, L, and Y in manufacturing used to calculate carbon efficiency were collected from World Input-Output Tables and underlying data. The data of GHG emissions in manufacturing originated from the OECD statistics database. The original patent data are from the OECD database. For control variables, the data of EUETS are from European Emission-Trading Scheme. The data of HC, TRADE, and GOV were collected from PTW90. The data of INV and lnGDP are originally from the WDI database. The data of EPS, ETTG, GDPK, and ENERT were collected from OECD. The data of MEI and MIS were calculated based on the collected data.

Development level of LCTI in OECD manufacturing
Based on the patent stock method, we calculated the aggregated LCTI and the LCTI of specific manufacturing in OECD countries in this study. Following the standard of YO2P firstlevel technology classification, low-carbon technology is mainly divided into eight types: metal processing, the chemical industry, the petrochemical industry, mineral processing, agricultural produce, the production process, integrated application, and potential emission reduction. Figure 1A shows the development level of LCTI in OECD manufacturing and its growth rate. Patent stock represents the level of LCTI. We can see that the LCTI level in OECD manufacturing is almost on the rise from 1990 to 2014, increasing by 323% in 2014. The growth rate of patent stock is rapidly growing from 1992 through 2000, possibly due to the signing of the United Nations Framework Convention on Climate Change in 1992, which aims to reduce GHG emissions. The growth rate declines rapidly from 2001, mainly affected by Internet Economic Dot. The growth rate is negative in 2014, mainly due to the missing data of low-carbon technology patents of the production process. As technology innovation is conducive to economic development, we calculated the low-carbon patent intensity indicator by the gross product per patent in manufacturing, representing the LCTI quality. Figure 1B illustrates the variation of low-carbon patent intensity in OECD manufacturing and its growth rate. The results show that the LCTI quality in OECD manufacturing increased by 55% in 2014 compared with the quality exhibited in 1995. Moreover, the growth rate of patent intensity fluctuates around zero during the study period, reflecting an uncoordinated development of innovation and production in manufacturing. In 2009, the growth rate of patent intensity was abnormally high, mainly affected by the financial crisis of 2008, which has hysteresis effects on manufacturing production. Figure 2A illustrates the development of low-carbon innovation in manufacturing in seven countries: Japan (JPN), America (USA), Germany (DEU), Korea (KOR), France year. In addition, the LCTI levels of LVA, ISL, and LTU are 5.0, 7.0, and 6.2, respectively, in the year 2014, while the LCTI level of USA is 11180.6. It is an indication that there is a vast gap between the highest countries and the lowest countries. Figure 3 shows the development change of low-carbon patent intensity of manufacturing in seven countries, Finland (FIN), Luxembourg (LUX), the UK (GBR), Germany (DEU), the Netherlands (NLD), Austria (AUT), and France (FRA). These countries have the highest level of low-carbon patent intensity in OECD countries from 1995 to 2014. Moreover, the aggregated patent intensity of LCTI of these seven countries accounts for more than 50% of the total OECD countries during the years 1995 to 2014. Among them, LUX shows the most significant increase in low-carbon technology patent intensity, followed by Poland and GBR. The patent intensity in LUX increased by 404% from 1995 to 2014, mainly caused by the rapid development of its steel industry. In OECD countries, KOR, Hungary (HUN), and Czech Republic (CZE) have the lowest low-carbon patent intensity in manufacturing, and in the year 2014, the value of patent intensity in these three countries is 0.0017, 0.0015, and 0.0155, respectively. In contrast, the patent intensity of LUX is 3.6138 in the year 2014, which is much higher than in the lowest countries.
In this study, the patent stock indicator represents the LCTI level, while the patent intensity indicator represents the LCTI quality in manufacturing. Figure 2 and Figure 3 indicate that the ranking of countries that have the highest LCTI quality differs from the countries that have the highest LCTI level. It can be seen that DEU, GBR, and FRA demonstrate both high patent intensity and high patent stock, which implies that the level and the quality of LCTI in these countries are relatively high. It is noteworthy that although the USA and JPN have high patent stock, these countries have relatively low patent intensity, suggesting that the LCTI quality in these countries is relatively low. Figure 4 shows the level of specific LCTI in OECD countries that have the highest patent stock. It can be seen that the highest patent stock of specific low-carbon technology in OECD countries are low-carbon technologies of the chemical industry, the production process, metal processing, and potential emission reduction. In addition, patent stock of the four low-carbon technologies is consistently increasing during the study period, accounting for about 80% of the total lowcarbon technology patent stock. Specifically, the proportion of the patent stock of low-carbon technology of production process grew the fastest, from 5.5 in 1990 to 31.0% in 2013, while the proportions of other technologies decreased during the study period. Also noteworthy, most OECD countries have a high level of LCTI in the chemical industry and a rapid development speed of LCTI in the production process. It is an indication that the chemical industry is the primary area in manufacturing to apply low-carbon technology from 1990 to 2013, and the production process has the most potential to achieve low-carbon manufacturing in most OECD countries.

Impact of LCTI on carbon efficiency
In order to investigate the role of LCTI in green production manufacturing, a two-way fixed model is employed in this study to investigate the impact of LCTI on carbon efficiency. The F values of the joint significance test on a time dummy variable are 12.51, 5.70, and 5.82; the P values are all zero; and there are time-fixed effects, refusing the null hypothesis at the significance level of 1%. Model 1 and model 2 were applied in this section to investigate the impact of the first-order lag LCTI on carbon efficiency. As shown in Table 2, the coefficient of L. ln LCT in model 1 is 0.027 at the significant level of 10%, and the coefficient of L. ln LCT in model 2 is 0.023 at the significant level of 1%. Model 3 was employed to investigate the impact of the current LCTI on carbon efficiency, and the coefficient of L. ln LCT is 0.019 with a significant level of 5%. The results of Table 2 suggest that LCTI promotes carbon efficiency of manufacturing, and a lag effect of LCTI exists. The mean value of the coefficients of lnLCT and L. ln LCT is 0.02, which indicates that a 1% improvement of LCTI increases manufacturing carbon efficiency by 0.02 units.
It can be seen that the coefficient lnGDP is negative at the significance level of 1%, which means that economic development is not conducive to the improvement of carbon efficiency, possibly because countries pursue economic growth at the expense of environmental interests. The coefficients of EUETS are positive in the three models at the significance level of 1%, which denotes that joining the European Union Carbon Trade markets increases the carbon efficiency. The coefficients of EPS are positive in models 2 and 3 at the significance level of 1% and 5%, respectively, which means that increasing environmental regulation stringency improves carbon efficiency in manufacturing. The coefficients of HC are positive in the three models. The significance level of models 2 and 3 are 1%, and model 1 is 5%, which shows that improving human capital helps increase carbon efficiency. The coefficients of the control variable GOV in models 2 and 3 are negative at the significance level of 1%, which shows that government participation in the market would hinder the improvement of carbon efficiency in manufacturing, possibly by distorting market competition.

Impact of LCTI on carbon productivity
Further investigations are carried out in this section to analyze the impact of LCTI on carbon productivity in manufacturing. Table 3 shows that the coefficient of first-order lag LCTI is The low-carbon patent intensity of manufacturing in OECD countries OECD others low-carbon technology of potential emissions reduction low-carbon technology of metal processing low-carbon technology of production process low-carbon technology of chemical industry Fig. 4 Level of specific LCTI in OECD countries distinctly positive, and the significance level is 5%, which indicates that LCTI is beneficial for increasing the carbon productivity in manufacturing, and there exists a first-order lag effect. The coefficient of LCTI is 0.008, which shows that a 1% increase in the current LCTI leads to a 0.008 unit increase in carbon productivity in manufacturing in the next period. As shown in Table 3, the coefficient of control variable L. IVN is positive at the significant level of 10%, meaning capital investment tends to increase the carbon productivity of manufacturing. The coefficient of variable MIS is positive at a significant level of 1%, which shows that optimizing industrial structure is beneficial for improving carbon efficiency in manufacturing.

Discussion and policy implications
In this paper, we have investigated the role of LCTI in the environmental performance of manufacturing. The results show that the impact of LCTI on the carbon efficiency and carbon productivity of manufacturing is significantly positive in OECD countries, indicating that LCTI is conducive to the green production of manufacturing. In addition, there exists a lag effect of LCTI carbon performance of manufacturing. The results of the current study are similar to the existing results (Du and Li 2019;Yan et al. 2020), and the results support the view that green technology is beneficial for green production. Unlike previous studies, this study excluded the disturbance of heterogeneity on industries and focused on the manufacturing industry to investigate the environmental performance of LCTI, providing more stable empirical support.
The results of the current study indicate that it is necessary to applicate low-carbon technology in manufacturing industry to realize the green production transformation in OECD countries. It is possible to improve LCTI by increasing R&D in low-carbon technologies, stimulating related low-carbon technology inventions, and introducing foreign advanced technology in OECD countries. And it is worth noting that both the quality and level of low-carbon technology patent in manufacturing distribute unevenly in OECD countries. Note: There is a standard error in the parenthesis. *, **, and *** represent the significance of 1%, 5%, and 10% respectively. In addition, the results in this study show that the chemical industry is the primary area in manufacturing to apply lowcarbon technology, while the production process has the most potential to achieve low-carbon manufacturing in most OECD countries; thus, we should primarily focus on these two areas to increase investment in technology research and development. Finally, the findings of the current study also reveal that enhancing environmental policy stringency, increasing human capital, and improving industry structure is beneficial for the environmental performance of manufacturing in OECD countries. Consequently, governments should intensify environmental supervision over manufacturing, such as raising taxes, constructing a carbon trade market, and fining polluting. Additionally, it is essential to increase the human capital level in manufacturing sectors. For example, we can encourage and support schools, scientific research institutions, and enterprises to establish technology innovation training bases, create talent incentive systems, and improve talent flow mechanisms. Allocating qualified personnel to overseas training programs and stepping up efforts to attract more talent to the manufacturing sector are also recommended. Still, there are some limitations in our current study. First, for more accurate results, further investigation needed to be conducted to analyze the role of low-carbon technology in the green production of segmented manufacturing industries, excluding the heterogeneity of manufacturing industries. Therefore, new methods need to be explored to measure the LCTI of segmented manufacturing in the future. Furthermore, the influencing mechanism of LCTI should be studied in more detail in the future. For example, the application and use of technology entail a sound technology foundation of receivers, which could be a considerable cost for a country, which may be difficult for developing countries to absorb the LCTI and promote the green transformation of manufacturing industry, and heterogeneous effect of LCTI may exist when considering the economic development level.

Conclusions
Manufacturing plays a vital role in economic growth and social development. It is necessary to shift the production model of manufacturing toward green production to reduce GHG emissions and improve the quality of the atmospheric environment. To investigate the issues mentioned above, the current study focuses on the development trend of LCTI in manufacturing using annual data from 1990 to 2014. Moreover, the role of LCTI on the environmental performance of manufacturing in OECD countries was investigated as well.
For key indicators, the LCTI level was measured by patent stock, and the environmental performance of manufacturing was measured by carbon efficiency and carbon productivity.
As a recap, the contributions of this study are twofold: First, compared with general technology innovation, we investigate the role of LCTI in the environmental performance of manufacturing, focusing on specific industries and technology. Theoretically, this study can provide stable empirical support that low-carbon technology improves the environmental performance of manufacturing by excluding the disturbance of heterogeneity on industries and technologies. Second, we measure the specific LCTI of manufacturing sectors, supplementing the referenced literature concerning LCTI indicators and providing policymakers with data and indicators about the LCTI of manufacturing.
Through empirical analysis, we obtained four main conclusions. First, LCTI is conducive to improving the environmental performance of manufacturing in OECD countries. Improving LCTI can increase carbon efficiency and carbon productivity in manufacturing. Second, LCTI of manufacturing in OECD countries increased during the years 1990 through 2014. LCTI of the chemical industry demonstrated the highest level, and LCTI of production showed the fastest development in most OECD countries from 1990 to 2013. Third, some countries that possess high low-carbon patent stock have relatively low patent intensity in manufacturing, and the gap of the quality and level of LCTI is relatively huge between OECD countries. Finally, increasing environmental regulation stringency and human capital, joining carbon trade markets, and optimizing industrial structure can improve the environmental performance of manufacturing.