Introduction

Climate change-related issues have increased enormously recently due to the deterioration of the ES, which has mainly resulted from economic growth and energy utilization1. This has highlighted the critical need for countries to pursue ES. While ensuring ES is important for all countries, it is nevertheless particularly relevant for developing countries that undergo rapid industrialization and urbanization, where the ecological costs of economic progress are often underestimated. This is why the recent literature has focused on the progress of ES across countries, especially developing ones.

In examining country cases, the literature has considered many potentially influential factors. Following up on the seminal study of Grossman and Krueger2various research have focused on the effect of GDP on ES (e.g., Wang et al.3; Magazzino4). Also, considering the leading research of Kraft and Kraft5some other studies have dealt with the energy effect on ES (e.g., Khan et al.6; Mukhtarov7). In this context, previous studies considered mainly fossil fuel (i.e., coal, oil, and gas), whereas much newer ones have focused on clean energy types (i.e., nuclear and renewable). Although both GDP and energy types have been frequently considered in the literature, contemporary literature deals with other factors that have recently emerged. Among all, productive capacity changes of economies have come to the fore (UNCTAD8), which is highly related to the economies’ growth structure. So, productive capacity changes in various areas are highly related to countries’ economic growth, which directly influences ES (Rzayeva and Huseynova9).

Achieving productive capacity changes in economies ensures that ES is critical for all countries. While developed countries have higher productive capacity changes, the case is different for developing countries. Among all, China is one of the most important countries. It is a global industrial hub, has a higher rate of economic growth, consumes a high level of energy, and has a high level of ecological disruption that threatens the global climate. China is responsible for approximately 1/4 of global carbon dioxide (CO2) emissions. Projections suggest that unmitigated climate change may cause a reduction in China’s GDP. For this reason, China has been focusing on the transition of economic sectors while adopting resource-efficient technologies to achieve carbon neutrality (World Bank10).

The interaction between economic growth structure, energy utilization, and ES in China is crucial for policy-oriented research. The ES, which is measured through various indicators like CO2, EFP, and LCF, provides a comprehensive assessment of ecological conditions. Figure 1 demonstrates the progress of the ES in China over the years.

Fig. 1
figure 1

The progress of the ES indicators in China. Source: Global Footprint Network (GFN11). The unit is a billion global hectares (basis point) for Biocapacity and EFP (LCF).

Full size image

As Fig. 1 shows, China has witnessed a steady increase in EFP, reflecting heightened ecological pressures due to rapid industrial growth, urbanization, and population expansion. Conversely, biocapacity, representing ecological resources’ ability to regenerate, has remained relatively constant and created a widening ES gap. This trend signals the urgency of adopting transformative measures for Chinese policymakers to transition toward an eco-friendly economic structure.

One promising avenue to address the challenges related to the ES lies in productive capacity changes, which are proxied by PCI and its subtypes across various areas. Productive capacity refers to the economies’ ability to produce goods and services efficiently (UNCTAD8) while minimizing adverse effects on the ES. Through focusing on PCI indicators, it is possible to understand how productive capacity changes may influence the ES and how they can be leveraged to ensure sustainable development. In this context, Fig. 2 demonstrates the progress of China’s PCI indicators over the years, which helps understand the development trends of productive capacity changes in China.

Fig. 2
figure 2

The progress of PCI indicators in China. Source: UNCTAD12. The unit is the basis point.

Full size image

As shown in Fig. 2, China has successfully ensured productive capacity changes in certain areas. PCI-SC is the leading area, and China has transformed the structure. It is followed by PCI-PS, PCI-EN, PCI-IC, PCI-HC, PCI-IN, and PCI-NC, whereas PCI-TR has the last order. Concerning these subtypes, PCI-T has a relatively slow upward trend.

The literature includes various studies concerning the PCI-ES relationship (Adebayo et al.13; Lin et al.14; Oluc et al.15; Raza and Lin16). Such research has focused on different countries (e.g., OECD countries and Pakistan) and applied conventional analysis approaches (e.g., ARDL and panel regression). On the other hand, the effect of PCI on the ES has been examined, whereas the marginal effects of the factors on the ES have been mainly neglected. Moreover, such empirical examinations have ignored the biggest emitter countries (i.e., China). So, it can be stated that there is a literature gap. Accordingly, this study comprehensively examines the effect of PCI types on the ES by controlling some other critical factors, for the Chinese case as the top emitting country, through using data between 2000 and 2022 and performing a novel KRLS approach for marginal effect analysis.

Consistent with the defined literature gap, the study explores the relationship between productive capacity changes and ES by considering controlling factors. Leveraging a KRLS approach, the study analyzes the marginal effect in China over the period 2000–2022. Applying the KRLS approach for empirical investigation is crucial because it enables researchers to uncover the effects across different percentiles, taking AME and PME into consideration. Also, this approach provides better prediction results than traditional econometric approaches by benefiting from Machine Learning (ML) and derivative calculation characteristics. In this way, a disaggregated level analysis allows researchers to understand how PCI types are effective in ES by reducing ecological pressures. Within this context, the study looks for the answers to the research questions: (i) which type of productive capacity changes are beneficial for China in supporting the ES; (ii) do aggregated or disaggregated level PCI indicators have a same or different effect on the ES; (iii) which type of effect do controlling factors (i.e., GDP, NEC, and REC) have on the ES; (iv) does the effect of the variables are the same or do they vary across percentiles of the factors; (v) are the outcomes robust based on alternative ES proxy (i.e., LCF)?; (vi) does the KRLS approach has a highly successful prediction on the ES by using aforementioned indicators.

With the econometric approach used (i.e., applying a novel empirical methodology), this study provides some contributions. First, this is the leading study that considers aggregated and disaggregated level PCI indicators in uncovering the effect of productive capacity changes on ES. Second, this pioneering study analyzes Chinese cases by benefitting from aggregated and disaggregated level PCI indicators simultaneously. Third, the study applies a marginal effect analysis to uncover how the factors considered (i.e., PCI types, GDP, NEC, and REC) have an effect on the ES across percentiles. Fourth, the study uses EFP as the primary proxy of the ES and considers LCF as the alternative proxy of the ES. So, the study examines both ecological pressure and ecological quality perspectives while applying a marginal effect approach. In this way, based on fresh insights, the study discusses policy implications about how Chinese policymakers can align productive capacity changes with ecological goals while mitigating ecological pressures.

The following sections of the paper provide an analysis of the literature review, methods, empirical outcomes, conclusion, and policy implications to offer a comprehensive understanding of how productive capacity changes influence ES in China. Hence, the research provides important implications for shaping the future direction of China’s economic and ecological policies by considering the productive capacity changes as China seeks to transition towards a more sustainable and eco-friendly growth model.

Theoretical framework and literature review

Theoretical framework

The theoretical framework includes energy-related factors concerning the relationship between energy and economic growth. According to the energy-led growth hypothesis proposed by Kraft and Kraft5economies need more energy as they grow. Hence, there is a critical nexus between energy utilization, economic growth, and ecological disruption. While clean energy utilization further supports economic growth in an eco-friendly manner much more, fossil energy utilization causes economic growth in an nonecological manner.

Also, the environmental Kuznets curve (EKC) hypothesis is a well-established theoretical framework for the relationship between income and ecological progress (especially from a pollution perspective) (Grossman and Krueger2; Panayotou17). EKC theory posits an inverted-U relationship, which implies that emissions increase with rising income to a maximum point, and then emissions begin to decrease after income exceeds a threshold. Depending on the countries and income structure and levels of the countries, the relationship between income and pollution varies, where there may be different curves (e.g., U-shaped, N-shaped, inverted U-shaped), implying a nonlinear relationship. Because the EKC hypothesis is largely examined in the current literature and is commonly employed in empirical research, the study does not elaborate on the EKC too much (Munasinghe18; Andreoni and Levinson19; Dinda20; Stern21).

Both energy-led growth and EKC hypotheses constitute a theoretical background in uncovering the relationship between environment, energy, and economy-related variables as the case in this study. By taking these theoretical background, the study considers income (i.e., GDP) and main clean energy subtypes (i.e., NEC and REC) as the explanatory variables.

Moreover, the most recent literature has been directed to PCI (Lin et al.14; Oluc et al.15), which proposes a new approach that models ecological progress as a function of PCI. Hence, it is important to consider PCI and PCI subtypes in uncovering ecological conditions. Accordingly, the study takes aggregated level PCI as well as PCI subtypes into consideration in the empirical investigation.

Literature on ecological indicators

Preliminary studies (e.g., Zhang and Liu22; Bhattacharya et al.23; Ben Jebli and Hadhri24; Raheem et al.25; Sardianou et al.26; Ulussever et al.27) mainly considered CO2 emissions as the ecological indicator because of the fact that the pollution perspective was highly important in previous studies. These studies uncovered the effect of various factors, including GDP and energy subtypes, on CO2 emissions.

However, the recent literature has shifted also into a new way by proposing new indicators for ecological condition. Among them, EFP and LCF have recently come to the fore of the attention of scholars in empirically examining the ecological condition of countries.

EFP is a much more comprehensive ecological indicator than CO2 emissions because CO2 emissions consider only air pollution, whereas EFP also considers pollution in multiple areas, such as air, soil, and water. Even, LCF is a much more comprehensive ecological indicator than EFP emissions because EFP does not consider ecological capacity by focusing only on the pollution perspective, whereas LCF considers ecological capacity by considering biocapacity in the empirical investigation of ecological conditions.

While some studies have used EFP (e.g., Dai et al.28; Kazemzadeh et al.29,30; Raihan31; Sun et al.32; Çabaş et al.33; Kartal34; Magazzino4; Wang et al.35; Caetano et al.36; Nikou37; Nikou and Sardianou38; Wang et al.39; Zhang et al.40), some studies (e.g., Kartal1; Özcan et al.41; Yang et al.42) have preferred to consider LCF in ecological investigation. Such research has investigated the effect of various factors, including GDP and energy subtypes, on the ecological condition of the countries proxied by either EFP or LCF.

Literature on energy and income

Because the EKC hypothesis constitutes a theoretical background for the relationship between environment and income, numerous studies have examined the relationship between ecological pressure and income by utilizing the EKC framework (e.g., Ben Jebli and Hadhri24; Danish et al.43; Kazemzadeh et al.29; Magazzino4). Such studies have determined various conclusions (e.g., U-shaped, N-shaped, inverted U-shaped relationship) between income and ecological condition, while some studies determine the validity or invalidity of the EKC hypothesis depending on the countries considered, the period examined, and the econometric approach used.

In addition, the theoretical framework has been expanded to include energy-related factors by following the study of Kraft and Kraft5 concerning the energy-growth relationship (e.g., Bhattacharya et al.23; Akram et al.44; Hasanov et al.45; Kartal et al.46; Khan et al.6; Adebayo et al.13; Kazemzadeh et al.30; Hasanov et al.47; Kartal et al.48; Mukhtarov7). While previous studies have focused on fossil fuel-based energy sources, the most recent ones have dealt with clean energy sources.

Literature on PCI and subtypes

UNCTAD developed the PCI. This index uses eight basic components to assess a country’s economic development and growth potential. The PCI measures countries’ productive capacity using an integrated model and helps decision-makers identify essential areas for sustainable development through the indicators below (UNCTAD8). Moreover, this index significantly aids in balancing economic growth with the ES (Oluc et al.17; Lin et al.18).

PCI covers various areas, such as “energy, human capital, information and communication technologies (ICT), institutions, natural capital, the private sector, structural change, and transport” (UNCTAD8). (i) Energy examines the accessibility of dependable and sustainable energy sources in addition to the efficacy of the energy infrastructure. The effective use of energy resources, notably the move towards renewable sources of energy, makes it feasible for countries to accomplish their objectives in sustainable development (Oluc et al.17); (ii) Human capital assesses labor productivity based on the population’s educational achievements, health status, skills, and competencies; (iii) ICT reveal the effect of digital transformation by analyzing the existence of digital infrastructure and the use of technology in society and the economy. The use of ICT can accelerate economic activities, save energy, and reduce CO2 emissions (Oluc et al.17); (iv) Institutions evaluate the nature of institutional frameworks in promoting economic growth by examining the quality of governance and the rule of law. Institutional quality plays a critical role in the regulation of economic activities and the implementation of the ES policies (Bhattacharya et al.23): (v) Natural Capital incorporates access to natural resources and their sustainable usage in establishing a basis for long-term economic activity (Lin et al.14); (vi) Private sector promotes market dynamics and innovation by evaluating the convenience of the business environment, the level of entrepreneurship, and the role of the private sector in the economy; (vii) structural change component contributes to the development of productive capacity by examining the diversification of the economy into different sectors and its transition to higher value-added sectors, which supports long-term economic resilience and growth (UNCTAD8); (viii) Transport component ensures the efficient mobility of goods and services by evaluating the quality and scope of road, rail, air, and sea transportation infrastructure. This component further evaluates the ecological effect resulting from transportation infrastructure.

In contemporary literature, researchers have investigated the effects of some PCI components (e.g., transportation, ICT, institution, energy, human capital). Accordingly, studies that have handled such PCI components have been reviewed. Table 1 summarizes the studies dealing with various PCI components.

Table 1 Summary of the literature on PCI.
Full size table

As seen from Table 1, various research studies have considered various subtypes of PCI. Because the role of energy utilization is highly important for both ecological and economic progress, the studies that have considered productive capacity changes in energy areas have been much more numerous than others. Nevertheless, the literature includes studies that have considered the productive capacity changes in human capital, ICT, institutions, structural change, and transport. In addition, a limited number of studies have used aggregated-level PCI in their empirical analysis.

Evaluation of the literature

In the case of overall evaluation, some common patterns can be seen from the literature examined as follows: (i) a variety of country or country groups have been analyzed in these studies; (ii) some indicators (e.g., CO2 emissions) for ecological condition have been intensively used, whereas the newer ones (e.g., EFP and LCF) have been considered less; (iii) the studies have frequently considered only a few subtypes of PCI, whereas only limited studies have considered aggregated level PCI; (iv) the studies have generally applied traditional econometric approaches; (v) the literature has presented contradictory outcomes for the effect of PCI types on the ES. Hence, some lacking points in the literature can be stated as follows: the present studies have mainly examined the average effect of some PCI types on the ecological condition by applying traditional econometric approaches. This approach, when applied in the studies, caused neglect, where the marginal effect is not considered. Also, China is neglected from this perspective. It is acknowledged that some studies (Zhang and Liu22) have analyzed China, but such studies have not considered the marginal effect and all possible PCI subtypes. Considering the literature gap, this study empirically examines the Chinese case as a leading emitting country by considering aggregated level PCI and disaggregated level PCI subtypes along with some controlling factors (i.e., GDP, NEC, and REC) in terms of their effects on the ES by applying a marginal effect analysis through a novel KRLS approach.

Methods

Data and variables

This research aims to uncover the marginal effects of PCI indicators on the ES, which is proxied by EFP and LCF, by applying a comprehensive analysis. In this line, aggregated level PCI and 8 PCI indicators at disaggregated levels are considered. Moreover, EFP is used as the main proxy of ES (Kazemzadeh et al.29; Çabaş et al.33; Magazzino4), while LCF is used for robustness (Kartal1; Yang et al.42). Table 2 shows the details of the variables.

Table 2 Variables’ description.
Full size table

Using data sources explained above, the study collects yearly between 2000 and 2022. The study uses the logarithmic series to account for elasticities in uncovering the marginal effect of PCI indicators on the ES.

Prediction models

Considering the variables detailed in Tables 2 and 10 prediction models are constructed to analyze the marginal effect of PCI indicators on the ES by controlling some critical factors as well. Table 3 summarizes the prediction models constructed.

Table 3 Details of the prediction models.
Full size table

In the prediction, only control factors (i.e., GDP, NEC, and REC) that are critically important for countries are considered in Model 1 to constitute a base in searching for the marginal effect of the variables on the ES.

In Model 2, total PCI (i.e., PCI-T) is added to Model 1 to investigate the marginal effect of total PCI on the ES at an aggregated level.

In Model 3–10, each PCI indicator (i.e., PCI-EN, PCI-HC, PCI-IC, PCI-IN, PCI-NC, PCI-PS, PCI-SC, and PCI-TR, in order) is added to Model 1 to search for the marginal effect of relative PCI indicator on the ES at a disaggregated level.

The study uncovers the effect of PCI on the ES comprehensively by considering a total of 10 prediction models constructed.

Empirical procedure

The research follows the empirical procedure shown in Fig. 3 to uncover the marginal effect of PCI indicators on the ES.

Fig. 3
figure 3

Empirical methodology.

Full size image

In the first step, the study examines the main properties of the variables. In the second step, the study analyzes the correlation coefficients. The third step is to control the stationarity of the variables by using the ADF test (Dickey and Fuller64) and the PP test (Phillips and Perron65). The fourth step is to check the nonlinearity structure of the variables through the BDS test (Broock et al.66). In the fifth step, the study first applies the KRLS approach to determine the AME of the factors on the ES. In the sixth step, the study uncovers the PME of the factors on the ES. Lastly, the study uncovers AME and PME on the ES by replacing EFP with LCF.

Considering data characteristics of the variables as an important point in appropriate econometric approach selection (Sinha et al.67; Kartal et al.68,69), which are defined through three leading previous steps, the study runs the KRLS approach (Hainmueller and Hazlett70) for predictions because it is superior to traditional econometric approaches due to the providing better prediction results through benefitting from machine learning and derivative calculation. While the KRLS approach enables researchers to uncover the effects across different percentiles by considering AME and PME (Taşkın et al.71), it does not have any pre-conditions as well.

KRLS approach

The study employs the KRLS approach, which was developed by Hainmueller and Hazlett70based on ML to explore the effect of productive capacity changes on the ES without assuming additive or linear relationships. In this approach, GDP, NEC, and REC variables are used as control variables in the predicted model. As compared to other linear regression approaches that are prone to misspecification bias, the KRLS approach provides superior outcomes (Salvatore et al.72; Warsame et al.73).

The KRLS approach has several additional advantages over other traditional approaches (Hainmueller and Hazlett70). Firstly, through various adjustments, the KRLS approach reduces the variance and susceptibility of predictions. The KRLS approach solves regression and classification problems without relying on linearity or additivity assumptions. By minimizing a complexity-penalized least squares problem, the KRLS approach constructs a flexible hypothesis space by using kernels as radial basis functions (Sarkodie et al.74; Kartal et al.68,69). Consequently, it is well suited to capture the effect of changes in productive capacity on the ES because this approach allows for a generalized linear model interpretation. Also, it allows for more complex interpretations that are useful for studying nonlinearities, interactions, and heterogeneous effects. As reported below, Eq. (1) shows the Gaussian kernel function used in the KRLS approach:

$$\:\text{k}\left({\text{x}}_{\text{h}},{\text{x}}_{s}\right)={\text{e}}^{-\frac{{\left|{\text{x}}_{\text{h}}-{\text{x}}_{s}\right|}^{2}}{{{\upsigma\:}}^{2}}}$$
(1)

.

where \(\:{\text{x}}_{\text{h}}\) and \(\:{\text{x}}_{\text{s}}\) are covariates, and \(\:{{\upsigma\:}}^{2}\) is the bandwidth of this function. The kernel reaches a maximum value when the difference between the covariances is large; if they are close, it reaches zero. As reported below in Eq. (2), the value of the function can be examined for a specific point (\(\:{\text{x}}^{{\upzeta\:}}\)):

$$\:\text{y}=\text{f}\left(\text{x}\right)=\sum\:_{\text{i}=1}^{\text{N}}{\text{w}}_{\text{i}}\text{k}({\text{x}}^{{\upzeta\:}},{\text{x}}_{s})$$
(2)

.

where a linear integration of the kernels can be represented by f(x) while \(\:{\text{w}}_{\text{i}}\) represents the weight for independent variables. Equation (3) shows the vector system of the previous equation:

$$\:\text{y}=\text{K}\text{w}=\left[\begin{array}{cccc}\text{k}({\text{x}}_{1},{\text{x}}_{1})&\:\text{k}({\text{x}}_{1},{\text{x}}_{2})&\:\cdots\:&\:\text{k}({\text{x}}_{1},{\text{x}}_{\text{N}})\\\:\text{k}({\text{x}}_{2},{\text{x}}_{1})&\:\ddots\:&\:&\:\\\:\vdots&\:&\:&\:\\\:\text{k}({\text{x}}_{\text{N}},{\text{x}}_{1})&\:&\:&\:\text{k}({\text{x}}_{\text{N}},{\text{x}}_{\text{N}})\end{array}\right]\left[\begin{array}{c}{\text{w}}_{1}\\\:{\text{w}}_{2}\\\:{\text{w}}_{\text{N}}\end{array}\right]$$
(3)

.

where w shows the scaled weights. The kernel regularized least squares and pointwise MEs are used to predict the final partial derivatives from Eqs. (2) and (3), which is reported in Eq. 4:

$$\:\frac{\widehat{{\delta\:}_{y}}}{\delta\:{x}_{h}^{\left(d\right)}}=\frac{-2}{{{\upsigma\:}}^{2}}\sum\:_{\text{i}}{{\text{w}}_{\text{i}}\text{e}}^{\frac{-{\left\|{\text{x}}_{\text{s}}-{\text{x}}_{\text{h}}\right\|}^{2}}{{{\upsigma\:}}^{2}}}\left({\text{x}}_{\text{s}}^{\left(\text{d}\right)}-{\text{x}}_{\text{h}}^{\left(\text{d}\right)}\right)$$
(4)

.

where \(\:\frac{\widehat{{{\updelta\:}}_{\text{y}}}}{{\updelta\:}{\text{x}}_{\text{h}}^{\left(\text{d}\right)}}\) shows the partial derivatives, and \(\:{{\upsigma\:}}^{2}\) indicates the kernel bandwidth, \(\:\text{a}\text{n}\text{d}\:\text{e}\) represents the exponential parameter.

The STATA software has been used for the empirical analyses. Also, the study applies specific codes to obtain empirical results for the prediction of EFP and LCF, which are provided in Supplementary Explanations 1 and 2, in order.

Empirical outcomes

Preliminary statistics

This study initially analyzes the descriptive statistics of the variables, which are reported in Supplementary Table 1. GDP and EFP have the highest mean values relative to other remaining variables. Also, PCI indicators have relatively average values, and EC types have relatively lower mean values. Moreover, some variables (e.g., NEC and REC) have a higher variation. In contrast, some others (e.g., GDP, PCI-IC, EFP, and LCF) have intermediate-level variations, while others have much fewer variations. From the distribution point of view, it is seen that all variables have a normal distribution as well.

The study analyzes the correlations between the variables reported in Supplementary Table 2. In the case of EFP, which is the main dependent variable, there are generally positive and high correlations with the explanatory variables. The correlations vary from 0.63 to 0.99. Among all variables, only PCI-NC has a negative correlation with EFP, which is equal to -0.71. On the other hand, in the case of LCF, which is an alternative dependent variable for robustness, there are generally negative and high correlations with the explanatory variables. The correlations vary from − 0.58 to -0.99. Among all variables, only PCI-NC has a positive correlation with LCF, which is equal to 0.66. When considering that EFP proxies the environment from a degradation point of view (Kartal34) and LCF proxies the environment from a quality point of view (Soytaş et al.75; Pata and Kartal76; Özcan et al.41), it can be expected that the signs of the effects should be reverse among EFP and LCF. So, the correlation of EFP with the explanatory variables is consistent with the correlation of LCF with the explanatory variables.

Moreover, the study analyzes the stationarities of the variables, which are reported in Supplementary Table 3. Both the main and alternative dependent variables are stationary at the level. Similarly, four of the PCI indicators are stationary at the level, whereas five of the PCI indicators are stationary at the first difference. Concerning controlling factors, GDP (NEC and REC) is stationary at the level (first difference).

Furthermore, the study analyzes the nonlinearities of the variables, which are reported in Supplementary Table 4. The variables follow a nonlinear structure across all dimensions. When preliminary statistics are evaluated altogether, it can be seen that the variables have a normal distribution, there are high correlations between the variables considered, and all variables have a nonlinear property. Because the use of nonlinear econometric modeling approaches is highly suitable in such a case that variables are mainly nonlinear (Sinha et al.67), the study performs the KRLS approach for empirical analysis to gather prediction coefficients on AME and PME (Kartal et al.68,69).

AME outcomes by KRLS approach

Following preliminary statistics examination, the study performs the KRLS approach to uncover AME within the context of empirical analysis. The outcomes are reported in Table 4.

Table 4 AME outcomes for EFP.
Full size table

The main control variable (i.e., GDP) has an increasing effect on the EFP across all prediction models (i.e., Models 1–10). This has mainly resulted from the use of high natural resource amounts (Li et al.77; Wang et al.78) for producing goods. It has been causing deterioration in the environment by causing ecological degradation and high amounts of CO2 emissions. This finding shows that China’s current economic growth structure has been causing the deterioration of the ES. This finding is also in the same direction as the current literature (Li et al.79; Raihan31; Magazzino4), where the non-eco-friendly structure of the Chinese economy has been defined. Accordingly, this research obtains consistent outcomes with these studies by defining the marginal effect of GDP as negative on the ES.

Also, the other two control variables (i.e., NEC and REC) have a stimulating effect on EFP across all models (i.e., Models 1–10). This determination reveals that the Chinese energy market is highly dependent on fossil fuel energy, and the Chinese economy uses a large amount of non-clean energy to carry out economic activities. This has generally resulted from the high amount of fossil fuel energy utilization in the total energy mix (EI63), while the share of clean EC in the total energy mix has been increasing slowly. A slow increase is not enough to make the energy mix green. That is why clean EC types are not effective in contributing to the ES, and a negative effect of both NEC and REC on EFP is seen in the case of a marginal increase. This determination is compatible with the present literature (e.g., Wang et al.3), where the non-eco-friendly structure of the Chinese energy market has been also determined. So, this investigation presents outcomes compatible with these current studies by showing that the marginal effect of both NEC and REC is not positive on ES.

While Model 1 is used as the base model to uncover only controlling factors, this base model is developed by including each PCI step by step. In this context, Model 2 reveals that total PCI does not contribute to the ES. Similarly, almost all subtypes of PCI, except for PCI-IN and PCI-PS, are ineffective in positively contributing to the ES across Models 3-4-5-7-9-10. By differentiating from these, Model 6 reveals that PCI-IN has a statistically significant effect on ES by ensuring a decline in EFP. Also, Model 8 shows that PCI-PS has a declining but insignificant effect. Overall, it is stated that changes in PCI types are not mainly beneficial in supporting the ES in China, whereas only changes in the PCI of the institutions are undoubtedly helpful. Various programs have been applied in China to transform the economy. However, Model 2–10 outcomes show that the efforts are insufficient in areas except for institutions.

Lin et al.14 conclude that total PCI benefits the ES, while PCI subtypes have mixed effects in 40 Belt and Road host countries. Similarly, Oluc et al.15 conclude that PCI develops the ES by decreasing CO2 emissions in 38 OECD countries. Thus, the outcomes of this analysis differ from those of the aforementioned studies. The difference may be caused by the fact that this study focuses on the Chinese case, while previous studies have dealt with a panel of selected countries. Hence, the outcomes for a specific country (i.e., China in this case) may differ from others due to scope differences. On the other hand, the outcomes of this analysis in terms of PCI subtypes are consistent with Lin et al.14where it is concluded that subtypes of PCI have a mixed effect on the ES, as is the case here.

PME outcomes by KRLS approach

Following the examination of AME, the study also uses the KRLS approach to uncover PME, where the outcomes are reported in Fig. 4.

Fig. 4
figure 4figure 4figure 4

PME outcomes for EFP. The x-axis denotes the ES indicator (EFP) and the y-axis denotes the marginal effect of the variable.

Full size image

GDP has a positive PME across all levels of EFP in Model 1, which is the main model. While GDP has a higher increasing marginal effect at the middle level of EFP, the increasing effect is a bit less at lower and higher levels of EFP. Nevertheless, a unit incremental increase in GDP certainly causes a unit increase in EFP. Moreover, GDP has a similar effect across all models (i.e., Model 1–10). Thus, GDP follows an inverted U-shape structure regarding its effect on the ES in China, which is consistent with the findings of Dai et al.28 and Li80representing a non-eco-friendly structure.

Both NEC and REC have a positive PME across all levels of EFP in Model 1, which is used as the main model. They have a lower increasing PME at lower and higher levels of EFP, whereas there is a much higher increasing PME at the middle level of EFP. However, there are still positive PMEs of both NEC and REC on EFP, which is also valid for all models, including Models 1–10. In this way, both EC types have an inverted U-shape structure on the ES in China, which is compatible with the findings of Sun et al.32 and Xiong and Mo81.

For the PME of PCI indicators, nearly all subtypes of PCI, excluding PCI-IN and PCI-PS, have a positive PME, which implies that those types of PCI do not contribute to the progress of the ES in China. Unlike those, Model 6 demonstrates that PCI-IN has a negative PME on EFP. While the decreasing PME is much stronger at the middle level of EFP, it is a bit weaker at the lower and higher levels of EFP. Nevertheless, PCI-IN has indeed declined PME on EFP, which implies support for the progress of the ES in China. On the other hand, Model 8 presents that PCI-PS has a negative PME on EFP. Although PME is much stronger at lower and higher levels of EFP, it is a bit weaker at the middle level of EFP. Therefore, among all PCI indicators, only the PME of PCI-IN helps decline EFP and contributes to the progress of the ES in China. While China has been putting various initiatives into effect to preserve ES, the outcomes of the models demonstrate that there is a long way and room for growth in this respect because of changes in PCI indicators, which imply the transformation of various areas (e.g., energy, human capital, ICT, natural capital, structural change, transport), are not helpful yet to ensure the ES in China. On the other hand, the outcomes of PME are compatible with those of Lin et al.14who define the mixed effect of PCI subtypes.

Robustness check

In what follows, the KRLS approach is run by replacing the main variable (i.e., EFP) with an alternative one (i.e., LCF). In this context, AME outcomes are reported in Table 5.

Table 5 AME outcomes for LCF.
Full size table

Because EFP represents ecological progress from the point of view of degradation (Kartal34) and LCF denotes ecological progress from the point of view of quality instead of either degradation or pollution (Özcan et al.41), it can be naturally expected that the signs of the effects of the variables should be vice versa according to the use of either EFP or LCF. From this perspective, when the outcomes between the main indicator (i.e., EFP) and alternative indicator (i.e., LCF) are evaluated together, it can be seen that the outcomes are entirely reversed, which empirically proves the robustness of the outcomes. Also, when the explanatory powers of the prediction models across the main and alternative indicators are examined, it can be defined that the explanatory powers are almost the same, which is 99.59%, reflecting a high prediction capacity.

Lastly, the study performs the KRLS approach to uncover PME for the alternative indicator (i.e., LCF), where the outcomes are reported in Fig. 5.

Fig. 5
figure 5figure 5figure 5

PME outcomes for LCF. The x-axis denotes the ES indicator (LCF), and the y-axis denotes the marginal effect of the variable.

Full size image

As is the case for AME outcomes between EFP and LCF, similarly, PME outcomes are also expected to be reversed. By considering these points, it can be seen that the PME outcomes between EFP and LCF are completely reversed, which empirically validates the robustness of the outcomes.

Summary of the outcomes

Following the application of a comprehensive empirical examination to uncover the ES in China, Fig. 6 summarizes the empirical outcomes of this research.

Fig. 6
figure 6

Summary of the empirical outcomes. +,–, and x denote the increasing, decreasing, and insignificant effect on the ES, respectively.

Full size image

As shown in Fig. 6, PCI-IN has a statistically significant increasing effect on the ES by decreasing EFP. Partially, PCI-PS has a curbing effect on EFP, which is beneficial for the ES, while the effect of PCI-PS is insignificant. Also, the remaining PCI subtypes do not contribute to the ES. Controlling factors, which include GDP, NEC, and REC, also do not have a good effect on the ES, which causes an increase.

Conclusions, policy implications, and future research

Conclusions and discussion

The public concern for ES has been developing worldwide due to the adverse effects of climate change. Human-induced activities have mainly caused this. Consequently, there has been a negative trend over recent years that threatened the ES. Among all countries, China is specifically relevant because it has the highest emissions in the world. For this reason, the ecological problems in both China and the world have been increasing. This condition requires a special focus on the Chinese case to preserve the ES in China and the world.

As indicated in Fig. 1, there is a rapid increase in EFP in China. As a result of this condition, the ES, which is proxied by LCF, has been declining over the years. Therefore, examining the causes and potential solutions against this negative trend is necessary. Consistent with this requirement, this research examines the Chinese case, uses novel indicators for the ES (i.e., EFP as main and LCF as robustness indicator), focuses on the effect of productive capacity changes in different areas along with the controlling factors (e.g., GDP, NEC, and REC), and applies a novel KRLS approach for the period of 2000–2022.

The outcomes of the KRLS approach empirically show that aggregated level PCI and all PCI indicators, except for PCI-IN and PCI-PS, are inefficient in ensuring ES. On the other hand, PCI-IN has a supporting effect on the ES, while PCI-PS has an increasing but insignificant effect. As PCI indicators, either at the aggregated level or disaggregated level, show the productive capacity change in respective areas (e.g., energy, institutions, private sector, transport) of economies, it can be argued that the economic transformation in these areas in China has not been enough to make contributions to the ES. Yet, it is acknowledged that China has been making a great effort to transform its economy into an eco-friendly one through a variety of policy measures across different sectors and locations. For example, China has been investing in clean energy. However, the main energy source is still fossil-based. This is the reason why PCI-EN is not beneficial in supporting the ES in China. This is similar to most of the other PCI subtypes. On the other hand, China successfully transformed its institutions to make them eco-friendly, while it has been on the right pathway for private sectors but has not reached a sufficient level. Among all, some institutional policies, such as environmental protection law, national carbon emission trading system, central environmental inspection system, green finance regulations, and environmental tax reform, can be mentioned. With this help, China can benefit from PCI-IN by developing a governance structure and regulatory effectiveness of institutions in ensuring the ES.

Besides, the outcomes demonstrate that GDP, NEC, and REC use have a decreasing effect on the ES. This determination implies that China does not have an eco-friendly income structure. Even its current energy mix consists mainly of fossil fuel sources, whereas China has been trying to increase the share of clean energy in the total energy mix, where it has been successful in increasing the share of clean energy from 6.1 to 18.2% from 2000 to 2022 (EI63). Nevertheless, fossil fuel (clean) energy has an 18.2% (81.8%) share of the total energy mix for the year 2022 (EI63). On the other hand, Chinese policymakers’ policy sets may be inefficient in ensuring a beneficial effect on the ES due to infrastructure challenges. Accordingly, a much better and more comprehensive energy transition policy, which also considers policy efficiency and infrastructure challenges, is highly required so that Chinese policymakers can ensure that China can benefit from clean energy utilization in supporting the ES.

Moreover, the outcomes are robust in the case of alternative ES indicator use. Furthermore, the econometric approach applied presents high prediction outcomes, around 99.7% among the main and robustness models. Overall, the effects of PCI types marginally differentiate across prediction models. Based on these summarized outcomes, the study is partially consistent with the literature (e.g., Lin et al.14; Oluc et al.15). For instance, Lin et al.14 determine the reducing effect of PCI on CO2 emissions in 40 BRI countries by running a system generalized method of moments and Feasible generalized least squares. Similarly, Oluc et al.15 determine the decreasing effect of PCI on CO2 emissions in 38 OECD countries by applying the PMG-ARDL approach. However, these studies do not consider disaggregated level PCI indicators. Instead, they only consider aggregated level PCI. Different from these studies, this research considers both aggregated and disaggregated level PCI indicators. Hence, this study determines that aggregated level PCI is not beneficial for China’s ecological condition, whereas PCI-IN has a supporting effect and PCI-PS has an increasing but insignificant effect on the ES. While this study demonstrates partially consistent results with the current literature, it achieves to present fresh outcomes by considering the marginal effect of the variables on the ecological condition that varies across PCI subtypes and percentiles.

Policy implications

China should enhance the capacity of institutions to formulate and implement policies that promote ES because of the supporting effect of PCI-IN. This point includes improving regulatory frameworks, enforcement mechanisms, and transparency in decision-making processes. In this way, China can develop an institutional framework and benefit from the positive effect of institutions in ensuring the ES directly as well as enabling positive progress in PCI subtypes. China cannot benefit from aggregated-level PCI indicators and disaggregated-level PCI indicators except for PCI-IN in supporting ES. Accordingly, Chinese policymakers should re-evaluate the actions taken related to these indicators. Hence, Chinese policymakers can determine the appropriate areas where corrective actions should be taken, so that China can begin to benefit from these PCI types. In addition, China should foster collaboration between public institutions and private enterprises to drive ES initiatives to enable positive contributions to ES. Public-private partnerships can facilitate the sharing of resources, knowledge, and expertise to achieve common environmental sustainability goals. China should build specific policies to make other PCI indicators (e.g., PCI-SC) efficient to contribute positively to ES.

In addition, China should continue to support public R&D initiatives to encourage relevant public institutions and universities to prioritize sustainable innovations by focusing on the circular economy and resource efficiency. In this context, China should provide capacity-building programs for institution staff to help them gain a deeper understanding of ecological principles and sustainable practices to improve the quality of their work. Allocation of financial sources and public policy support to public R&D initiatives is critical to provide progress in R&D initiatives, which may provide a positive contribution to the ES.

Besides, China should minimize the negative effect of GDP on the ES by supporting a circular economy and green growth strategies by encouraging companies to recycle resources, design products with longer life cycles, and support industries transitioning to sustainable production methods, particularly energy-intensive and polluting sectors, to reduce waste. Moreover, China cannot benefit from using clean energy to support the ES. Similarly, PCI-EN, which implies the energy transition, is not beneficial for China. Those findings together imply the inefficient energy transition policy of China. Accordingly, Chinese policymakers should re-review their energy transition policy and focus on the potential displacement effect between clean energy sources.

Furthermore, because the outcomes empirically reveal that the effects of the variables considered vary across percentiles of the variables, Chinese policymakers should consider the marginally varying effect in reshaping ES policies, which are related to productive capacity changes, income, and clean energy. Thus, Chinese policymakers can have the opportunity to support the ES by benefiting from productive capacity changes in the relevant areas, as well as income structure and clean energy utilization.

China has implemented various policies (e.g., the environmental protection law revised in 2015 and the national carbon emission trading system in 2021) in recent years. Such initiatives have been aiming at ensuring ES. So, the policy implications proposed in this study align with China’s recent approach to ensure ES. Hence, China’s current ecological policy framework can be extended by integrating the policy implications proposed by this study. It empowers the current policy structure about ES in China.

Although this study proposes some policy implications, there may be some challenges in implementing those. Weaker governance structures may cause challenges, where applying ES-related policies can be much harder if governance structures become weaker due to weaker governing bodies. Another challenge may come from political instability, which is also related to a weaker governance structure. The next challenge may be corruption. In case of weaker governance structure and political instability, corruption may be seen commonly, which threatens the development, implementation, and enforcement of ES-related decisions. Corruption may cause variations in ES-related decisions, which may be harmful to some economic actors because they can be affected wrongly due to corruption. Moreover, a weaker governance structure may cause misallocation of sources, making it hard to organize public awareness campaigns, causing a trade-off between economic development and ecological conservation, disabling to providing of necessary data and monitoring structures due to the transparency problems, being climate-vulnerable due to their reactive approach instead of proactive ones to the ecological issues, and short-term point of view in implementing initiatives supported either by the country or international actors. Those things can be stated as challenges that weaker governance structures can cause because policymakers cannot make necessary ES-related decisions on time, with the correct scope, in case of weaker governance structures, political instability, and corruption. Hence, the consideration of such challenges by policymakers is critical in terms of the achievement of the ES.

Future research

This study tries to apply a comprehensive econometric approach to find answers to the research questions it addresses. Against this backdrop, the study naturally has some limitations that scholars can consider in the design of future research.

First, because the study focuses on examining the Chinese case, new studies can either include many more developing countries or include more countries that are both developed and developing countries, comparatively. Hence, new studies may present comparative outcomes for those countries.

Second, because the study uses aggregated level data for EFP and LCF, new studies can work with disaggregated level data. Hence, many more details (e.g., sectors, cities, or other subtypes) can be examined in new studies in this way.

Third, since the study considers EFP and LCF as ecological indicators, new studies can consider using other indicators, such as biodiversity. Hence, new studies can include other perspectives on ES.

Fourth, due to the inclusion of the main controlling factors (i.e., GDP, NEC, and REC), new studies can also consider other factors (e.g., R&D investments, patents, artificial intelligence, and population) that may be effective on the ES. Even, economic, socio-political, and human geography drivers (Silva et al.82) as well as environmental technologies, financial policies, uncertainties (e.g., energy, climate, geopolitical), and natural resources (Dao et al.83) or indices reflecting various issues (e.g., green growth, human development, financial development) (Kazemzadeh et al.84), or the role of weaker governance structure proxied by various indicators (e.g., political instability and corruption) can be considered in designing new research. Hence, a wide range of perspectives for controlling factors can be included in new research.

Fifth, due to yearly data use and KRLS approach application, new studies can work with either much newer data, if possible, or much higher-frequency data (e.g., quarterly, monthly, and daily), or perform other novel econometric techniques. Also, new studies can apply pure ML approaches and use cross-validation or out-of-sample testing within the scope of the application of pure ML approaches. Besides, new research can perform other approaches (e.g., WTC, WLMC, dynamic ARDL simulations) so that they can consider time and frequency-varying effects, autoregressive effects, lagged effects, and the effects in short and long-run distinction.

Sixth, because the study does not take causality between ecological indicators and explanatory variables, new studies can have the possibility to consider causality between the variables. Even, new studies can make empirical analysis by considering the reverse causality between the variables.

Lastly, new studies can consider applying other techniques that enable researchers to perform different empirical approaches for the main analysis and robustness checks, which is not the case in this study. Also, a set of alternative econometric approaches can be applied within the context of robustness checks. Besides, new studies can be considered to uncover the relationship across various periods as well as different prediction model specifications. Hence, new studies focusing on other perspectives through new econometric approaches can include the most recent developments.