RETRACTED: Exploring the influencing factors of environmental deterioration: evidence from China employing ARDL–VECM method with structural breaks

Abstract

Purpose

The environmental deterioration has become one of the most economically consequential and charged topics. Numerous scholars have examined the driving factors failing to consider the structural breaks. This study aims to explore sustainability using the per capita ecological footprints (EF) as an indicator of environmental adversities and controlling the resources rent [(natural resources (NR)], labor capital (LC), urbanization (UR) and per capita economic growth [gross domestic product (GDP)] of China.

Design/methodology/approach

Through the analysis of the long- and short-run effects with an autoregressive distributed lag model (ARDL), structural break based on BP test and Granger causality test based on vector error correction model (VECM), empirical evidence is provided for the policies formulation of sustainable development.

Findings

The long-run equilibrium between the EF and GDP, NR, UR and LC is proved. In the long run, an environmental Kuznets curve (EKC) relationship existed, but China is still in the rising stage of the curve; there is a positive relationship between the EF and NR, indicating a resource curse; the UR is also unsustainable. The LC is the most favorable factor for sustainable development. In the short term, only the lagged GDP has an inhibitory effect on the EF. Besides, all explanatory variables are Granger causes of the EF.

Originality/value

A novel attempt is made to examine the long-term equilibrium and short-term dynamics under the prerequisites that the structural break points with its time and frequencies were examined by BP test and ARDL and VECM framework and the validity of the EKC hypothesis is tested.

Citation

Wang, H. (2024), "RETRACTED: Exploring the influencing factors of environmental deterioration: evidence from China employing ARDL–VECM method with structural breaks", International Journal of Climate Change Strategies and Management, Vol. 16 No. 2, pp. i-xix. https://doi.org/10.1108/IJCCSM-10-2021-0114

1. Introduction

Due to the lack of anticipation of the negative effects of highly developed industry and unfavorable prevention, three major global crises have been caused, namely, resource shortage, environmental pollution and ecological damage. In this context, sustainable development has become a worldwide concern (Zanten and Tulder, 2020). Sustainable development is a theory and strategy frame about the coordinated development of nature, science and technology, economy and society (Croese et al., 2020). Identifying the factors accounting for environmental degradation along the economic growth stream is pertinent to keep the ultimate target of attaining the sustainability into cognizance. In 1987, the World Commission on Environment and Development published the report, namely, Our Common Future, which systematically expounded the idea of sustainable development and defined it as development that can meet the needs of contemporary people and does not harm the ability of future generations to meet their needs (Shan and Zhang, 2020).

Conducting research on China’s sustainable development is pertinent to the world. China is the most populous developing country, the second-largest economy and the largest energy consumption and carbon emission country in the world (Nie et al., 2020). Human economic activities are the greatest threat to environmental sustainability (Pretel et al., 2016).

Economic development in China has long been characterized by resource-driven and labor-intensive characteristics (Deng et al., 2021). Despite flourishing economically, China consumes gigantic natural resources (NR) and simultaneously discharges increasing pollutions persistently (Wang et al., 2020). China’s carbon emissions have topped since 2006 (Deng et al., 2021). Resource depletion, energy shortage and environmental degradation are the key challenges of sustainable development in China (Zhao et al., 2020). The analysis of sustainability in China will provide empirical evidence for the vast number of developing countries. Therefore, in many ways, it is appropriate to examine sustainability with China as a typical example.

Great gap remains in practice and theory of sustainable development paths in existing literature. First, there is little literature on the influencing factors of sustainable development in China, and the research results are multifarious. Second, when coming to literature on sustainable development, CO₂, PM_2.5, etc. are often used to represent environment, which is not enough to express the rich connotation of sustainable development. Third, although there are studies on the influencing factors of sustainable development, because of different development stages, development models, scientific levels and so on, there will be different relationships between variables. Fourth, if the combination of influencing factors is different, the results will vary dramatically.

Consequently, the present work contributes in multiple aspects. First, unlike most of existing literature using a single index as the proxy of environmental sustainability, this paper chooses the comprehensive index, i.e. the ecological footprints (EF), to bridge the abovementioned gap. Second, both human factors and natural factors are considered when selecting influencing factors. Third, the structural breaks of influencing factors that are yet to be extensively considered in the relevant literature are considered in the modeling. Kanjilal and Ghosh (2021) found that when the structural break was not taken into account, the autoregressive distributed lag (ARDL) cointegration test could not validate the cointegration relationship between carbon emission and other variables, but after considering the structural break, the threshold cointegration test verified the cointegration relationship. Fourth, preceding studies have proved that the Granger test based on vector autoregressive model is only suitable when no cointegration exists. When a cointegration relationship exists, adopting the Granger test based on vector error correction model (VECM) comes to more accurate results (Granger, 1987). Fifth, although a small number of scholars have considered the structural break of carbon emission, they have not included the endogenous structural break point in the model. Jian (2018) and Zhou et al. (2015) used Gregory–Hansen cointegration test with endogenous structural break test, but they did not include the structural break point into the model and carried out piecewise research with the break point as the boundary.

The rest of the work proceeds as follows: Section 2 summarizes the existing relevant literature; Section 3 states the data sources and reports the model specification and methodology; Section 4 describes the results of each experiment; Section 5 discusses the experimental results; Section 6 concludes the study with some policy recommendations; and Section 7 describes the prospect of future research.

2. Index selection and literature review

First, the EF index is selected as the proxy variable of sustainable development. The proxy of sustainability has always been a topic of intense debate among scholars (Moghadam and Dehbashi, 2018). The EF refers to the area with biological productivity that is needed to maintain the survival of a person, a region or a country or can accommodate the waste discharged by human beings (Nathaniel et al., 2021). Compared with other single indicators (e.g. CO₂ emissions, SO₂ emissions, etc). the EF is a comprehensive index to describe sustainability or environmental degradation (Baz et al., 2020). China surpassed the USA in the total the EF at the beginning of this century. At present, the studies on the EF in China are mostly focusing on the accounting and characterizing, whereas little analysis is focusing on the driving factors and their long-term and short-term effects (Peng et al., 2018). In the analysis of environmental degradation, carbon emissions index is often taken as the proxy variable, which is only one aspect of environmental issues (Peng et al., 2018). Meanwhile, according to the latest report of the Global Footprint Network, China’s per capita EF is 3.619 global hectares (gha), whereas the per capita biological capacity was only 0.948 gha in 2016. This ecological deficit showed that the demand for ecological goods and services in China’s ecosystem was far greater than its supply. A persistent ecological deficit will lead to the degradation of the renewable capacity of renewable resources, which in turn leads to the loss of biodiversity, the degradation and shrinking of natural assets and even the collapse of ecosystems (Lin et al., 2019). Therefore, using the EF index as the proxy variable of environmental degradation or sustainability is suitable for China.

Second, this study selected economic growth [gross domestic product (GDP)] as an explanatory variable of the EF. On the one hand, the existing literature provides multifarious or even contradictory conclusions about the relationship between GDP and environment, such as U-shaped (Charfeddine and Mrabet, 2017), inverted U-shaped (Al-Mulali and Ozturk, 2015) or no relationship (Xiao, 2013). On the other hand, a serious deficiency of existing studies is the excessive reliance on single index [e.g. CO₂ emission (Hussain et al., 2017), SO₂ emission (Khler and Wit, 2019) and PM_2.5 (Dong et al., 2018)] as the agents of environmental degradation. In fact, the EF is the most systematic and holistic indicator to track environmental degradation (Dogan et al., 2020), and a growing number of studies have used the EF as the proxy of environment. For a long time, GDP has been the priority of the Chinese Government and the primary indicator to assess subordinate officials. Hence, the problems of massive consumption of resources have been ignored, which leads to frequent occurrence of environmental disasters. In 2017, the energy consumption in China was 4.49 billion tons of standard coal, accounting for 23.2% of the world’s consumption and 33.6% of the world’s growth. Meanwhile, CO₂ emissions in China accounted for more than 30% of the world’s total emissions (Wan et al., 2020). Resource consumption and environmental problems in China have aroused world concern. Recently, the United Nations released a report saying that extreme climate disasters have increased significantly in the past 20 years, causing serious casualties and economic losses. However, when the GDP reaches a certain level, emissions will be reduced with the economic growth, which is the famous environmental Kuznets curve (EKC) hypothesis (Yang and Xie, 2015). This study will explore whether there is an EKC-shaped relationship between economic growth and the EF in China.

Third, this study selected NR as an influencing variable of the EF. The relationship between the abundance of NR and economic growth is also a controversial topic in academia (Arshad et al., 2020). Balsalobre-Lorente et al. (2018) confirmed that an N-shaped relationship existed between economic growth and CO₂ emissions in European Union’s five countries, and abundant NR were conducive to improving environmental quality. Similarly, Joshua and Bekun (2020), adopting the dynamic ARDL method, came to the same conclusion in terms of nature source based on data of South Africa. However, with the augmented mean group technique and data of Brazil, Russia, India, China and South Africa, Danish et al. (2019) found an insignificant effect of nature sources on emissions. China is vast in land and rich in resources. Nevertheless, on the one hand, the amount of resources per capita is small and uneven; on the other hand, the efficiency of resource utilization is low, especially in the field of energy (Kai et al., 2009). The resource endowment characteristics of “rich in coal, poor in oil and short in gas” have seriously affected the ecological benefits in the process of resources utilization (Wang and Chen, 2020). There has always been a hypothesis, namely, the “resource curse,” in the academic community (Qiu and Chen, 2020). However, the prevailing literature largely ignores the relationship between NR and the EF in China. Therefore, the level of NR was selected as one of the influencing indicators in this study.

Fourth, this study selected urbanization (UR) as an influencing variable of the EF. UR is a much-debated topic in the environmental field, and views on UR for environmental sustainability vary depending on the time periods and countries. Major changes in the process of UR include production methods, lifestyles and population structure (Wang et al., 2018). These changes would lead to varying land-use patterns, changes in energy consumption and emissions of pollutants, which will have a significant impact on the EF and thus sustainable development. However, whether the impact is insignificant, positive or even negative depends on the different models, processes and pattern of UR (Wang et al., 2018). Al-Mulali et al. (2015) using the ARDL method proved that the impact of UR on emissions is positive. Many other researchers share similar views and believe that UR is a threat to sustainability and will increase energy demand and emissions (Dogan and Turkekul, 2016). However, some other studies did not support, such as Hossain (2011), who claimed that the effect of UR on sustainability varies from country to country. Behera and Dash (2017) argued that UR is unstable in middle- and high-income countries, whereas its role in low-income countries is unclear. Wang et al. (2016) used provincial data to show that the impact of UR varied greatly in China’s provinces. Xu and Lin (2016) claimed that there was an inverted U-shaped relationship between UR and carbon emissions implying that the UR was not conducive to sustainable development in the short term, but is conducive in the long run. Currently, China is promoting the UR at an astonishing speed and scale. Examining the sustainability of UR is crucial not only to China but also to the world. UR rate in China rose rapidly from 17.9% to 59.6% over the period of 1978–2018, and many scholars estimated it would reach about 80% in the future (Zhou, 2020). Such a high-speed inevitably brings a lot of ecological problems. Developed countries have entered the third stage of UR, whereas China is still in the second stage of UR (Chen et al., 2013). Therefore, this study took UR as one of the explanatory indicators.

Fifth, this study also selected labor capital (LC) as an influencing variable of the EF. At present, it is difficult for higher educated college students in China to find jobs, whereas less-educated people (such as migrant workers) find jobs easily (Zhou et al., 2017). This situation is evidence of “labor-intensive” characteristics of the Chinese industry. Although the government has frequently put forward measures and policies to promote the employment of college students, these measures and policies may not be able to fundamentally solve the problem. This is not necessarily a good choice, whether from the perspective of education and policy sustainability or the employment environment sustainability. Zafar et al. (2019) further pointed out that uneducated or unskilled LC are detrimental to the sustainable utilization of NR and good LC can promote the utilization of environmental-friendly technologies, including the technologies of NR. Additionally, China has a large population and the level of LC has improved rapidly. There were 8.34 million college graduates in China in 2019; 8.2 million in 2018; and 7.95 million in 2017 (Yan, 2019). The employment situation of college graduates in China is particularly grim. Therefore, we took the LC factor into the research.

3. Data source and model construction

3.1 Data source

Reform and opening-up is not only the sign of rapid economic development but also the starting point for the market-oriented allocation of all kinds of resources in China. Therefore, the research period is set from 1978 to 2016, and the data sources of specific indicators are shown in Table 1.

3.2 Model and method

3.2.1 Basic model setting.

According to the existing research models (Murshed et al., 2021) and the research requirements, the basic econometric model is set as equation (1):

(1) EFt=α+β1GDPt2+β2GDPt+β3NRt+β4URt+β5LCt+μt

Referring to Table 1 for the meaning of each symbol and its data source, α is a constant term, µ is a random disturbance term and β_i (i =1, 2, …, 5) are the parameters to be estimated. To reduce the impact of abnormal data fluctuations and heteroscedasticity on the model while keeping the relationships between the variables unchanged, the natural logarithms of all variables are taken when the model is set up (Murshed, 2019). The variable GDP and its squared term, respectively, indicate the per capita real GDP, measured in terms of constant 2010 US dollars. GDP per capita is used to assess the dynamic impacts of higher levels of national income on the nation’s EF to comment on the validity of the EKC hypothesis.

3.2.2 BP structural break points test.

In the process of economic activities and environmental changes, the impact and adjustment of some events are likely to lead to changes in economic and environmental structure. Time series under such scenarios are called deterministic nonstationary. To promote the transformation of economic structure and protect the ecological environment, China has continuously introduced new policies related to the variables selected in this study, which may lead to structural breaks of these variables.

The research on structural breaks began with Chow test, but this method can only be used when the structural break point is known and only one structural break point can be detected (Gregory, 1960). Bai and Perron (1998) made an in-depth study on the structural break and proposed the method of endogenous structural break test. Bai (1999) has proved that this method has a better test level and efficiency in the case of small sample, and it is a more accurate and objective method at present.

BP test assumes that the variables transfer n times in time T and produce n structural break points. The model can be expressed as equation (2):

(2) {Yt=X1tTα+X2tTβ1+εt,                 t=1,2,…,T1                     Yt=X1tTα+X2tTβ2+εt,                t=T1+1,T1+2,…,T2 ···Yt=X1tTα+X2tTβn+1+εt,           t=Tn+1,Tn+2,…,T   

where Y_t denotes the dependent variable, X_1t and X_2t are independent variables,  X1tT and X2tT are p-dimensional and q-dimensional column vectors, respectively, α and β_i (j = 1, 2,…, n + 1) are the corresponding parameter vectors to be estimated, ε_t is the random disturbance term, T denotes the total number of samples, n denotes the number of structural break points and T₁, T₂, …, T_n are the time when n structural break points occur.

Based on equation (2), the BP test is performed using the least square method to estimate the parameter for each possible segmentation. Subsequently, pick up the segmentation with the smallest sum of squared residuals and perform statistical tests on it to get the structural break point. This study employs DM tests [first statistic of DW test with upper limit (UD_max) and second statistic of DW test with different weight (WD_max)] statistics proposed by Bai and Perron (1998) to determine whether the variable has structural breaks, and make use of SupF(l + 1|l) sequence statistics to verify the number and time of structural break points. The null hypothesis of the DM tests statistics is that there is no structural break point, and the alternative hypothesis is that there is a structural break point with an upper limit. The null hypothesis of the SupF(l + 1|l) sequence statistics is that there is one structural break point. The alternative hypothesis is that there are l + 1 structural break points.

3.2.3 Autoregressive distributed lag model.

Following Zhang et al. (2021), to solve the pseudo-regression problem caused by the nonstationarity of time series, cointegration test is conducted to verify whether there is a long-term stable relationship between considered variables. Compared with other cointegration test methods, the bound testing proposed by Pesaran et al. (2010) offers several advantages. First, there are no autocorrelation issues in the ARDL bound testing method, and the endogenous issue can be addressed by selecting the appropriate lag length. Under the premise of determining the optimal lag, ARDL can be used to analyze the long-term relationship of variables no matter the considered variables are I(0), I(1) or both. Second, the requirement for sample size is low, i.e. even if the sample size is small, the test results are still robust (Pesaran et al., 2010). Routinely, the process of building an ARDL model consists of the following two steps:

1. According to equation (1), the following ARDL model is established, and the boundary cointegration test is carried out:

   (3) ΔEFt=Δ0+Δ1EFt−1+Δ2GDPt2+Δ3GDPt+Δ4NRt+Δ5URt+Δ6LCt+∑i=1qα1iΔEFt−i+∑i=0qα2iΔGDPt2+∑i=0qα3iΔGDPt−i+∑i=0qα4iΔNRt−i+∑i=0qα5iΔURt−i+∑i=0qα6iΔLCt−i+ εt

where Δ is the first difference operator. i is the lag order of the difference term, the sign (Σ) denotes the short-run section of equation (3) and α_1i ∼ α_6i articulate the short-term parameters, and ′ϒ₁∼′ϒ₆ are the long-run parameters. ε_t is white noise error term. The null assumption of equation (3) is H0: ′ϒ₁ = ′ϒ₂ = ′ϒ₃ = ′ϒ₄ = ′ϒ₅ = ′ϒ₆ = 0, and the alternative assumption is H1 = ′ϒ₁ ≠ ′ϒ₂ ≠ ′ϒ₃ ≠ ′ϒ₄ ≠ ′ϒ₅ ≠ ′ϒ₆ ≠ 0. If H0 cannot be rejected, it is considered that there is no cointegration relationship among the variables. F-statistics and critical value can be used to judge whether to reject H0 or not.

1. When the cointegration relationship is verified, the long-run cointegration equation could be established to estimate the long-run equilibrium coefficients. The ARDL–VECM model could be obtained through simple linear transformation, and the short-run coefficients of the explanatory variables could be estimated. To avoid the influence of parameter instability on the reliability of the model setting, it is also necessary to do parameter stability test to determine whether the model is reasonable.

3.2.4 Granger causality test.

The cointegration relationship between variables could be verified by ARDL model, but the existence of a cointegration could not prove the causality. It is necessary to carry out Granger causality test. The results of Granger causality test in the form of one-period lag error correction term (ECM_t−1) are more accurate when a cointegration relationship presents. The Granger causality test based on the error correction term (ECM_t−1) could be carried out within the VECM to ascertain the direction of causality between time series, as shown in equation (4):

(4) (1−B)[EFtGDPtGDPt2NRtURtLCt]=[b1b2b3b4b5b6]+∑i=1q(1−B)[C11,iC12,iC13,iC14,iC15,iC16,iC21,iC22,iC23,iC24,iC25,iC26,iC31,iC32,iC33,iC34,iC35,iC36,iC41,iC42,iC43,iC44,iC45,iC46,iC51,iC52,iC53,iC54,iC55,iC56,iC61,iC62,iC63,iC64,iC65,iC66,i][EFt−1GDPt−1GDPt−12NRt−1URt−1LCt−1]+[φ1φ2φ3φ4φ5φ6][ECMt−1][γ1tγ2tγ3tγ4tγ5tγ6t]

where (1−B) is the difference operator, q is the lag order and the ECM_t−1 is the lagged error correction term, pursuing to capture the long-run causality. φ₁ through to φ₆ depicts the speed of adjustment. γ_1t through to γ_6t articulates the random errors explaining the constant variance or homoscedastic property.

4. Experimental results

4.1 BP test results

All variables are tested by the BP test with Eviews 10.0 software. The test results show that there are structural break points in the EF series, but no structural break points in other variables. Table 2 demonstrates the results of the BP test on the EF. At a 5% significance level, UD_max statistics, WD_max statistics and SupF(1) are all significant, indicating that there is at least one structural break point in the EF series. According to the suggestion of Perron (Pesaran et al., 2010), the number and time of break of the EF series are judged by the sequence statistics of SupF(1). The results demonstrate that there are two structural break points in the EF series and the time was between 1992 and 2003, respectively.

The virtual variables are introduced as the bounds, and the whole time series is divided into three subintervals. Then equation (1) is transformed into equation (5):

(5) EFt=α+β1GDPt2+β2GDPt+β3NRt+β4URt+β5LCt+β6DU1992+β7DU2003+μt

where DU₁₉₉₂ and DU₂₀₀₃ are the virtual variables in 1992 and 2003, respectively. Its meaning is shown in equation (6):

(6) DU1992={0,  t<19921,  t≥1992， DU2003={0,  t<20031,  t≥2003

Subsequently, these two dummy variables will be included when the bound testing was conducted and the ARDL model was built.

4.2 Unit root test

To avoid the spurious regression phenomenon, the empirical analysis started with the test for unit root by applying the augmented Dickey–Fuller and Phillips–Perron tests, and the results are shown in Table 3.

The results in Table 3 reveal nonstationarity in levels, except for UR. Hence, the first-order differencing is conducted, and all the series show stationarity. Therefore, all the series (EF, GDP², GDP, NR, LC, UR) are stationary at their first difference or they are I(1) variables.

4.3 Cointegration relationship testing

According to the results of the unit root test, the bound cointegration test can be applied to examine the cointegration relationship (see Table 4). The F-statistic of bound cointegration in Table 4 is 4.899, which exceeds the critical value of I(1) under the 1% significance level. Therefore, during the sample period, there is a cointegration relationship between the EF and other considered variables.

4.4 Autoregressive distributed lag model estimation results

Eviews 10.0 software is used to estimate the above ARDL model. Considering the limitation of the sample size, the maximum lag order of the variable is set to 2. The estimated results are shown in Table 5. Combined with double maximum (DW), Akaike information criterion (AIC) and Schwartz derived a criterion (SBC), the estimation result of ARDL (0,1,2,1,1,1,1,1) without trend item is the best choice. After diagnosing and testing the ARDL (0,1,2,1,1,1,1,1) (see the lower half of Table 5), the model is set properly. Therefore, based on ARDL (0,1,2,1,1,1,1,1), the long-run cointegration model and ARDL–ECM model are constructed, and the long- and short-run effects of each variable on the EF could be analyzed.

The estimation results of ARDL (0,1,2,1,1,1,1) are shown in Table 6. From the results, GDP, UR, NR, LC and two structural break points are all significantly correlated with the EF. In the long run, the EF will increase by 1.083% for every 1% increase in GDP. When the square term of GDP increases by 1%, the EF will decrease by 0.016%. This verifies the presence of the EKC hypothesis in China. However, China is still in the upward phase of the inverted U-shaped curve, and it will take time to reach the turning point.

The NR have a significant positive impact on the EF and the coefficient of elasticity is 0.352. For every 1% increase in the NR, China’s EF will increase by 0.352%, indicating that there is a resource curse in China.

There is a significant positive correlation between UR and the EF with the coefficient of 1.958 implying the improvement of UR in China has greatly increased the EF. Routinely, in the long run, with the improvement of UR, the agglomeration effect of cities can effectively improve the efficiency of resource allocation and optimize the industrial structure. Simultaneously, the centralization of population and industry is conducive to technology spillovers such as energy conservation and environmental protection, which will improve the production efficiency and resource utilization on a large scale and reduce the EF. However, if the pathway of UR is unsustainable or the resource utilization is unscientific, it will lead to the opposite effect. However, such an unscientific and unsustainable situation exists in pathways of UR in China.

At the 5% significance level, the LC have a negative impact on the EF and the coefficient of elasticity is −2.541 implying that for every 1% increase in the LC, the EF would reduce by 2.541%. The LC can not only reduce the EF by adopting new technologies and new lifestyles but also reduce the EF by using new production technologies and transforming economic growth.

The two structural break points in 1992 and 2003 also have a significant impact on the EF in China indicating the relevant policies in 1992 are conducive to the decline of the EF, whereas the rapid economic growth in 2003 lead to an increase in the EF.

4.5 Error correction model estimation

Furthermore, an ARDL–VECM model is established to analyze the short-run relationship between variables. The results are shown in Table 7.

As can be seen from Table 7, the model is properly set and the coefficient of the error correction item is negative, which is in line with the reverse correction mechanism, indicating that after deviating from the long-run equilibrium, the EF will be revised by 82.6% in the future. In the short run, the EKC hypothesis between economic growth and the EF cannot be verified. The impact of changes in NR on the EF is still significantly positive and its short-run effect (0.203) is slightly lower than the long-run effect (0.352). Therefore, if the utilization model of nature resource is not transformed in time, the accumulation of this negative effect would have a serious impact on sustainable development. In the short run, the relationship between UR and the EF is still significantly positive; the short-run changes in LC are negatively correlated with the changes in the EF, but this negative effect is not obvious. At present, China is promoting the construction of a new type of UR, hoping that the new path of UR can contain the EF. In addition, the cultivation of LC would take a certain amount of time, signifying that LC cannot effectively restrain the EF in the short run.

4.6 Granger causality test

To verify the economic rationality of long-run and short-run effects, it is necessary to conduct a Granger causality test. The Granger causality test based on VECM is used to verify the causal relationship between variables. The results are shown in Table 8.

From Table 8, all the explanatory variables are the Granger causes of the EF in the short run. Simultaneously, the EF is also the Granger cause of GDP and NR, which means that the current changes in the EF will have an impact on future GDP and NR. The bidirectional Granger causalities between the EF, GDP and NR enlighten that to realize the sustainable development, the Chinese Government should arrange the goal of environmental improvement reasonably. For convenience, Figure 1 briefly describes the relationship in Table 8.

4.7 The parameter stability testing

Cointegration estimation using time series might affect the stability of parameter estimation results. Therefore, it is necessary to verify the stability of the estimated parameters. Referring to the methods in the work of Boutabba (2019), we verified the stability of the estimated parameters, and the results are shown in Figure 2. Figure 2 shows that the values of cumulative sum (CUSUM) and CUSUM of squares are within the two critical lines, indicating that the parameters between the EF and the explanatory variables are stable and reliable.

5. Results and discussion

The results of in-depth empirical analysis have brought some clear practical knowledge. The results of the empirical analysis based on China’s national situation will be further explained, analyzed and discussed below.

First, the connections between GDP and environmental protection have been a much-debated topic in academics which is often examined in the frame of the ECK hypothesis. However, a plethora of existing literature quantify environmental degradation in terms of CO₂, the awareness of the EF propensities to represent environmental degradation has largely been ignored in the EKC narrative. This study employs the EF to bridge this gap via empirically examining the existence of the EKC association in terms of the EF in the context of China. The results of empirical analysis show that there is an inverted U-shaped relationship, in the long run, between economic growth and the EF, which supports the conclusion of Wu et al. (2020). Relevant studies prove that when the economic growth reaches a certain level, the technical and structural effects of economic growth on the EF exceed the scale effect, which makes economic growth conducive to curb emissions. However, China is still in the rising stage of the EKC postulation currently exposing that economic growth in China is still at the scale effect stage in which both economic growth and the EF are rising simultaneously. It will take time to reach the inflection point, and the inverted U-shaped relationship between the two variables has not been established in the short run. This is probably because, since the reform and opening-up, China’s economic growth has largely depended on high energy consumption. Simultaneously, the rapid development of UR and industrialization in recent years has increased the demand for resources and intensified the EF.

Second, Zafar et al. (2019) argued that NR curbed the EF in the USA. However, this finding contradicts Hassan et al. (2020) for Pakistan. China uses NR intensively and the efficiency requires being improved. Remarkable achievements of economic development have been made in China with great contributions stemming from NR. Some studies believe that since the reform and opening-up, the average contribution of NR to China’s economic development has reached about 6% (Danish et al., 2019). The positive contribution of the NR to the EF in China implies the occurrence of the resource curse, further confirming the conclusion of Wang et al. (2016). Improving resource efficiency is an important way to change this situation. While it is difficult to upgrade technology in the short run, improving management levels has become a top priority.

Third, in the long run, UR will significantly increase the EF, meaning that China’s path to UR is also unsustainable. From 1978 to 2018, China’s UR rate rose sharply from 17.9% to 59.6%, and many scholars estimate that China’s UR level will reach 80% in the future (Zhou, 2020). The rapid UR lacks sustainable planning, and some government officials even regard UR indicators as political achievements to expand blindly. This is one of the main reasons for the rapid rise of the EF brought about by UR in China. In the short run, UR has also significantly increased the EF (significant at 10% level), which further proves that the current UR needs to be improved, in terms of resource allocation efficiency, management level, etc.

In the long run, the improvement of LC has significantly reduced the EF. This shows that China’s education has indeed made a great contribution to sustainable development. In 1949, it was found in China that 540 million Chinese were illiterate, accounting for 80% of the total population. In some remote and rural areas, the illiteracy rate was as high as more than 90%. In 2001, China achieved the strategic goal of basically popularizing nine-year compulsory education and eliminating illiteracy among young and middle-aged people. In 2018, there were 28.31 million general junior college students and more than 16.7 million full-time teachers at all levels and types. The proportion of financial education funds to GDP remains above 4% for many years. On the one hand, the rise in the level of LC has improved the level of resource utilization and the overall technological level; on the other hand, it has also enhanced the awareness of sustainable development and popularized the environment-friendly technology. Although it is insignificant statistically, China’s LC would also curb the rise of the EF in the short run.

In addition, the two structural break points in 1992 and 2003 also had a significant impact on the EF. Since 1992, the market sector has been participating in the allocation of resources, which optimizes the allocation of resources, improves the efficiency of the use of resources and effectively suppresses the EF. Since 2003, China has entered a new era of an investment boom, with excessive investment scale and rapid economic growth. The average growth rate of GDP from 2003 to 2011 was more than 9% and even reached 14.2% in 2007. Excessive economic growth leaded to a shortage of energy supply, which is unfavorable to the decline of the EF. Also, during this period, the rapid development of industrialization and UR made the problem of high energy consumption and high pollution more prominent, resulting in the soared EF.

6. Conclusion and policy recommendation

The main conclusions of this paper include the following aspects:

• In the long run, the EKC hypothesis between the EF and economic growth has been verified indicating that the win–win goal could be achieved. However, China is still in the rising phase of the EKC and it would take time to reach the inflection point. NR and UR have adverse effects on the EF, and LC help to restrain the EF. The two structural break points have a significant impact on the EF implying that the economic and environmental policies in 1992 were favorable to restraint of the EF, whereas the rapid economic development since 2003 has led to a significant increase in the EF.

• In the short run, the EKC hypothesis did not hold. NR had a negative impact on the EF indicating that the utilization of NR was in an unsustainable way. The UR and the EF were positively and significantly correlated with each other. The LC could not effectively reduce the EF.

• The results of the Granger causality testing show that, in the short run, all the explanatory variables were the Granger causes of the EF, and EF was the Granger cause of economic growth and NR indicating that there was a reverse mechanism between the EF and economic growth and NR. In the short run, the excessive decline of the EF was likely to be detrimental to the economic growth and the NR. At the same time, the LC were the Granger cause of the UR and the NR, and the continuous promotion of LC was beneficial for the sustainable development.

Combined with the research conclusions, the following policy recommendations are put forward:

• In view of the significant effect of GDP on the EF, China should transform the way of economic development and implement the strategy of sustainable development. In the long run, China should speed up the upgrading of industrial structure and build a green and circular economy. China should adopt a two-pronged approach that improves the efficiency of resource utilization and transforms the way of development simultaneously to promote the coming of the inflection point of the EKC and realize the win–win goal. In the short term, on the premise of maintaining economic growth, China needed to appropriately control the speed of the EF reduction and minimize the adverse impact on economic growth caused by the rise of the EF.

• Strive to improve the efficiency of resource. The World Bank report has pointed out that China’s resource utilization efficiency was lower than the international levels. Although this lower efficiency was partly caused by China’s current economic structure, there is still much room for improvement in resource utilization. At the same time, increasing consumption of raw materials and low resource utilization efficiency had also led to insufficient investment in waste disposal and large amounts of waste. The absence or failure of management system and policy is not only a crucial cause of environmental pollution and low efficiency of resource utilization but also a vital breakthrough to improve the efficiency of resource utilization in China.

• Take the new way of UR and implement sustainable UR strategy. Compared with developed countries, although the level of UR in China has been improved quickly, there is still much room for improvement. In the process of UR, China should severely keep away from the mass demolition, mass construction and thousands of cities with the same style, comprehensively consider the economic development, employment, public infrastructure and other aspects, design the UR scientifically and avoid the waste of resources.

• Adhere to the strategy of giving priority to the development of education. LC can not only improve the technical level and resource utilization but also increase the adoption of new and environment-friendly technologies. Keeping the proportion of financial education funds to 4% of GDP is an important measure for the sustainable development. It is necessary to increase investment in education in economically backward and remote areas, make up for the shortcomings and improve the overall level of China’s LC. The second is to improve the quality of education, respect individual differences and reform the current one-size-fits-all system in the field of education in China. Finally, China should give full play to the important role of LC in sustainable development and be people oriented in economic development, UR, environmental governance and resource utilization.

7. Prospects for the future

The findings have certain practical implications for the realization of sustainability, but some limitations still need to be further explored. First, sustainable development is a worldwide systemic problem that requires global systematic change and coordination. The new perspectives, such as economic globalization, world education level according to the planning in the 2030 agenda of the United Nations, should be added to further researches. Second, different countries or regions may have different interrelationships among variables. Therefore, it is necessary to strengthen the in-depth study of different regions and countries. Combined with the characteristics of different factors, different suggestions for improving sustainable development should be put forward. Finally, the symmetrical ARDL model is built in this work to get symmetrical relationships. In the future the asymmetric ARDL model should be built for further asymmetric results.

Figures

Figure 1.

Causality between variables

Figure 2.

Plot of cumulative sum (CUSUM) and cumulative sum of squares (CUSUMSQ)

Table 1.

Variables and data sources

Variables	Symbol	Measurement	Data source
Ecological footprint	EF	Ecological footprint (global hectares per capita)	Global Footprint Network
Natural resources	NR	Total natural resource rent (percentage of GDP)	World Development Index
Economic growth	GDP	GDP per capita (constant 2010 US dollars)	World Development Index
Labor capital	LC	Labor capital Index	Penn World Tables
Urbanization	UR	Urban population percentage of the total population	World Development Index

Table 2.

Test results of structural break points of BP

Statistics	UDmax	WDmax	SupF(1)	SupF(2|1)	SupF(3|2)
The value of the statistic	51.82**	107.18**	2.82**	19.11**	1.51
Break time	1992	2003

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 3.

Unit root test results of each variable

Variables	The original sequence	First-order difference sequence
EF	1.085	−0.952***
GDP2	−1.652	−3.685***
GDP	1.985	−1.632**
NR	−2.365	−3.124***
UR	−1.235**	−3.215***
LC	1.652	−1.373***

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 4.

Results of bond cointegration testing

Equation (5)	F-value
	4.899***
The critical value	P-value	I(0)	I(1)
	10%	1.89	2.52
	5%	2.01	2.91
	1%	2.99	3.85

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 5.

Estimated results of ARDL (0,1,2,1,1,1,1,1)

Variables	Estimated coefficient	T-statistic	P-value
α	2.154	0.892	0.210
GDP2	−0.154	−0.951	0.154
GDP2 (−1)	0.295	1.098	0.082
GDP	2.891	0.954	0.185
GDP (−1)	−4.857	−1.024	0.052
GDP (−2)	1.092	3.854	0.001
NR	0.607	4.185	0.001
NR (−1)	0.124	3.251	0.003
UR	0.621	0.528	0.430
UR (−1)	−3.287	−2.982	0.011
LC	0.024	1.096	0.108
LC (−1)	−0.385	−3.854	0.000
DU1992	−0.054	−0.981	0.080
DU1992 (−1)	−0.092	−3.652	0.003
DU2003	0.012	0.684	0.351
DU2003 (−1)	0.029	2.036	0.031
	R2 = 0.99871
	AIC = 52.1746	SBC = 42.6851	DW = 2.1688
Model diagnosis result
χSC2(1)	9.842***	χF2(1)	2.016**
χSC2(1)	12.854***	χSC2(1)	5.361**

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 6.

ARDL results of long-term cointegration estimation

Variables	Estimated coefficient	Standard error	T-statistic
α	4.058	5.021	0.821
GDP	1.083	0.562	−3.021***
GDP2	−0.016	0.032	−1.362*
NR	0.352	0.101	5.385***
UR	1.958	0.384	−4.854***
LC	−2.541	0.032	−1.873**
DU1992	−0.254	0.084	−3.541***
DU2003	0.065	0.085	2.201***

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 7.

Estimated results of ARDL–VECM model

Variables	Estimated coefficient	Standard error	T-statistic
ΔGDP	2.001	1.921	1.982
ΔGDP (−1)	−0.982	0.543	−3.799***
ΔGDP2	−0.185	0.521	−0.967
ΔNR	0.203	0.095	3.254***
ΔUR	0.682	0.985	0.658*
ΔLC	−0.025	0.088	1.638
ΔDU1992	−0.052	0.073	−0.952
ΔDU2003	0.186	0.114	0.697
Δα	2.365	2.964	0.259
ECMt−1	−0.826	0.001
R2	0.986	Adjusted R2
F	9.873***	Maximum likelihood	79.652
DW	2.609	RSS	0.004

Note:

Corner marks

*,

** and

***, respectively, represent significance levels of 10, 5 and 1%

Table 8

Granger causality test based on VECM

Variables	ΔEF	ΔGDP	ΔGDP2	ΔNR	ΔUR	ΔLC
ΔEF		5.032（0.13）	4.930（0.06）	12.352（0.003）	5.361（0.32）	0.215（0.47）
ΔGDP	26.237（0.00）		0.921（0.68）	10.251（0.53）	10.385（0.01）	0.617（0.33）
ΔGDP2	27.064（0.00）	0.891（0.53）		8.320（0.02）	2.691（0.02）	0.951（0.28）
ΔNR	19.352（0.00）	1.076（0.43）	3.581（0.29）		0.251（0.66）	0.517（0.01）
ΔUR	32.854（0.00）	2.650（0.26）	1.286（0.43）	6.574（0.20）		0.214（0.51）
ΔLC	15.051（0.00）	3.051（0.19）	3.075（0.16）	6.215（0.02）	2.014（0.02）

Note:

The degree of freedom of the χ² test statistic of a single explanatory variable is 1; the data in () are the p-values