The Kyoto Protocol and the Agreement on Climate Change are only two of many programs that have advocated for energy production through sustainable resources—such as wind, solar, and hydro biofuels—to generate electricity, heat homes, and power vehicles. All of this is aimed at mitigating the impact of climate change. The Sustainable Development Goals (SDGs), along with broader environmental, social, and governance objectives, can only be achieved by reducing carbon emissions while simultaneously increasing the economic impact of green energy—reflecting the historical linkage between environmental progress and economic development. Many studies have found a positive correlation between reduced fossil fuel consumption, increasing energy prices, and the use of alternative energy sources. Using panel data from GCC nations, the research team examined the relationship between natural gas use and financial development. The study discovered a favorable correlation between the energy usage in GCC nations and economic development. Their examination of panel data led them to the same conclusion: renewable energy, rather than nuclear power, helps lower carbon emissions and, in the long run, boosts economic development.

Worldwide, severe weather events have been more common since the 1990s, when humans began spewing vast amounts of emissions of combustion byproducts into the air, including carbon dioxide. Therefore, there has been heated discussion and no agreement among scholars worldwide regarding the impact of the fast expansion of renewable energy sources on ecological sustainability (Latif et al., 2021; Aljundi et al., 2024; Makhdum et al., 2022). In addressing climate change, renewable energy sources are considered the most effective option for lowering CO₂ levels in the atmosphere and ensuring long-term ecological sustainability (Ikram et al., 2020; X. Zhou et al., 2020). Nevertheless, the features of the implemented technology, important nations or areas, the scientific process, and financial factors cause the empirical findings to be inconsistent. One example is that using renewable energy sources considerably reduces CO₂ emissions, as shown by S. Liu et al. (2022), who examined the variables impacting this use in 64 countries from 1990 to 2011. By investigating the functional use of green power sources for reducing carbon emissions in the E7 nations, Tawiah et al. (2021) and Sadiq et al. (2022) came to a favorable conclusion. This aligns with the findings of Caglar et al. (2022) regarding China and the BRICS nations; however, the researchers discovered that unexpectedly low levels of renewable energy cause pollution emissions to increase over time. The correlation between renewable energy use and CO₂ emissions in OECD nations from 1970 to 2015 was investigated using the Silva and Machado quantile regression method within a panel data framework. They found a positive correlation between the amount of carbon emissions and the use of renewable energy sources per capita over the long run and also noted that the reverse is also true for these nations. In addition, renewable energy sources greatly reduce carbon dioxide emissions, and the two ideas are interdependent. A further discovery was the existence of a feedback causal relationship between CO₂ emissions and the use of green power sources. For example, when examining the use of renewable energy sources and carbon emissions in West African nations from 1990 to 2018, the authors used the CCEMG and DCCEMG estimation methods to assess the relationship. According to their findings, renewable energy sources and carbon dioxide emissions in West African nations are inversely related. From 1997 to 2017, researchers in the United States examined the 10 most populous states using an NARDL estimating approach to determine the unequal effect of green power’s impact on carbon dioxide emissions.

We use the following empirical model to examine the effect of monetary growth and renewable energy use on CO₂ emissions in China, based on previous research of Yang et al. (2020), Goczek et al. (2021), and Wang and Li (2021): CO 2 t = f GDP t , FD t , REC t , HUC t , AVA t . (1)

In Equation 1, CO₂ represents carbon dioxide emissions, FD stands for financial development, GDP refers to gross domestic product, REC indicates the use of renewable energy, HUC denotes human capital, and AVA represents agricultural production in China. In this study, the econometric model based on time-series linear data is presented in Equation 2: C O 2 t = β 0 + β 1 G D P t + β 2 F D t + β 3 R E C t + β 4 H U C t + β 5 A V A t + μ t . (2)

The time period from 1996 to 2022 is represented by the interval t. β₀ represents the function’s intercept. The variables and the following equations reflect the growth of GDP, expansion of the financial sector, consumption of renewable energy, human capital, and agricultural productivity in the following sequence: β₁, β₂, β₃, β₄, and β₅. Finally, the fluctuating mistake component is represented by µ.

In addition, the research variables’ data are converted to natural logarithms to intentionally decrease the problem of heteroskedasticity. The direct elasticities provided by log-linear variables are particularly helpful for interpretation (Le et al., 2020). Therefore, using a panel specification, the converted log-non-linear models enhanced a CO₂ emission function that is not linear with several variables, as expressed in Equation 3: L C O 2 t = β 0 + β 1 L G D P t + β 2 L F D t + β 3 L R E C t + β 4 L H U C t + β 5 L A V A t + μ i t . (3)

The primary aim of this study is to empirically investigate the long-run and causal relationships between financial development, renewable energy consumption, economic growth, human capital, agricultural productivity, and CO₂ emissions in China, with the overarching goal of identifying pathways toward sustainable development and emission mitigation. To achieve this, the study utilizes annual time-series data spanning from 1996 to 2022, focusing exclusively on the Chinese context. The data were sourced from reputable and consistent statistical databases to ensure accuracy and reliability. In particular, data on CO₂ emissions (measured in kilotons), economic growth (GDP, measured as annual percentage growth), financial development (proxied by domestic credit to the private sector as a percentage of GDP), renewable energy consumption (as a percentage of total final energy use), human capital (measured by the ratio of college students to total population), and agricultural value-added (as a percentage of GDP) were all retrieved from the Chinese Statistical Yearbook and the World Development Indicators (WDI) published by the World Bank. All variables were converted into natural logarithmic form to correct for heteroskedasticity and facilitate elasticity-based interpretation of the coefficients. This data framework allows for an in-depth examination of the dynamic interrelationships among economic, environmental, and structural variables influencing China’s carbon emissions profile.

Table 2 presents the descriptive statistics for the logged variables used in the regression model. Average CO₂ emissions (LCO₂) are 12.85, with a relatively tight dispersion (σ = 0.08), indicating modest year-to-year variation in the sample. Logged GDP (LGDP) shows a broader range—from 22.83 to 32.72—with a standard deviation of 0.57, reflecting China’s rapid economic expansion over the study period. Financial development (LFD) varies between 2.16 and 9.41 (σ = 0.29), while renewable energy consumption (LREC) remains lower in magnitude (mean = 1.76) and less volatile (σ = 0.06). Human capital accumulation (LHUC) averages 14.84 and exhibits moderate variability (σ = 0.53), whereas agricultural value-added (LAVA) displays the widest spread (minimum = 0.44, maximum = 5.43, and σ = 0.91). Skewness values hover near 0, and kurtosis values are below three, indicating that all series are approximately symmetric and platykurtic. Jarque–Bera probabilities exceed 0.05 for each variable, so the null hypothesis of normality cannot be rejected. Collectively, these summary statistics confirm well-behaved distributions, providing a solid foundation for subsequent econometric analysis.

Table 3 displays the pairwise correlation coefficients among the study variables. LCO₂ is positively correlated with LGDP and LFD, suggesting that economic and financial expansion have historically contributed to increased environmental degradation. In contrast, LCO₂ is negatively correlated with LREC, LHUC, and LAVA, indicating that improvements in clean energy use, education, and sustainable agriculture help mitigate emissions. Notably, LAVA shows the strongest negative correlation with LCO₂ (−0.4327), reinforcing the role of green agricultural practices in environmental improvement. Additionally, a strong positive correlation exists between LREC and LHUC (0.6432), suggesting co-movement between renewable energy adoption and human capital development. Multicollinearity concerns appear minimal as no single correlation coefficient exceeds the critical threshold of 0.8, supporting the suitability of these variables for regression analysis.

Unit root tests are essential in time-series econometrics to determine the stationarity of variables, which directly influences the choice of estimation technique. Confirming the integration order is a prerequisite for applying cointegration methods like DOLS as including non-stationary variables without cointegration can lead to spurious regression results. Table 4 presents the outcomes of three unit root tests—ADF, PP, and DF-GLS—applied to the log-transformed variables. The results uniformly indicate that all variables are non-stationary in their levels but become stationary after first differencing, confirming that they are integrated of order one, I(1). For instance, LCO₂ and LREC are not stationary at levels in the ADF and DF-GLS tests but both attain significance at the 1% level upon first differencing. Similarly, LGDP and LAVA also fail to reject the null of a unit root at levels but show strong stationarity after differencing across all three tests. LFD and LHUC follow the same pattern, with significant test statistics in first-differenced form. The consistency of the findings across all tests confirms that the series are suitable for cointegration analysis, supporting the application of the DOLS estimator in the empirical framework.

To confirm the existence of a long-run equilibrium relationship among the model variables, panel cointegration tests were conducted using both Pedroni and Kao methodologies. The Pedroni test, following Li and Umair (2023b), allows for heterogeneous intercepts and slope coefficients across cross-sections while accounting for potential cross-sectional dependence. This test evaluates both within-dimension (panel statistics) and between-dimension (group statistics) alternatives. Additionally, the Kao (1999) test, which is based on a pooled regression with fixed effects, was used to validate the homogeneity of cointegrating vectors. These tests are particularly useful in small-sample contexts and provide a robust framework for assessing cointegration in panel datasets.

Table 5 presents the results of the panel cointegration tests. Among the Pedroni test statistics, four out of seven reject the null hypothesis of no cointegration at conventional significance levels. In particular, the panel PP-statistic (p = 0.0007), panel ADF-statistic (p = 0.0271), group PP-statistic (p = 0.0000), and group ADF-statistic (p = 0.0000) offer strong evidence in favor of a cointegrating relationship. Although the panel v-Statistic and rho-statistics fail to reject the null, these are generally considered less reliable in small samples than the ADF-based tests. Supporting these findings, the Kao test further confirms cointegration with an ADF statistic of 1.54321 and a p-value of 0.0286, which allows us to reject the null hypothesis of no cointegration at a 5% significance level. The low residual and HAC variance values further support the model’s stability. Taken together, the results of both Pedroni and Kao tests provide robust evidence that a long-run cointegrating relationship exists among carbon emissions, GDP, financial development, renewable energy, human capital, and agricultural value-added. This validates the suitability of using fully modified OLS or DOLS estimators in subsequent estimation procedures.

The empirical findings of this study align with and diverge from various strands of the existing literature, offering both confirmatory and novel insights. The positive and statistically significant relationship between economic growth (LGDP) and CO₂ emissions is consistent with earlier studies such as Cai et al. (2025) and Yuan et al. (2023), who reported that, in emerging economies like China, growth-led development often intensifies environmental degradation during industrialization phases. Similarly, the observed positive association between LFD and CO₂ emissions supports the findings of Lu et al. (2024), who argued that financial expansion facilitates industrial activities and energy consumption, potentially leading to greater environmental stress in the absence of green regulatory frameworks.

In contrast, the negative and statistically significant impact of LREC on CO₂ emissions corroborates the results of Ul-Haq et al. (2023) and Umair et al. (2023), who highlighted the critical role of renewable sources in reducing carbon intensity. This reinforces the policy relevance of China’s commitment to green energy transitions under its decarbonization roadmap. Additionally, the negative influence of LHUC on emissions is in line with the findings of Umair and Yousuf (2023), who emphasized that education and skill development foster environmental awareness and cleaner technologies. Furthermore, the significant and inverse relationship between LAVA and CO₂ emissions aligns with recent studies of Yu et al. (2023), who recognized the environmental benefits of sustainable agricultural practices and carbon sequestration in land use.

The results from the DOLS estimation, summarized in Table 6, provide compelling evidence of long-run relationships between CO₂ emissions and key development indicators in China. The coefficient of economic growth (LGDP) is positive and statistically significant at a 1% level, indicating that a 1% increase in GDP leads to approximately a 0.45% increase in CO₂ emissions. This finding reaffirms the classical environmental Kuznets hypothesis in its early phase, suggesting that growth in China is still largely driven by carbon-intensive sectors, with environmental costs accompanying economic expansion.

In contrast, LFD is also positively associated with CO₂ emissions, with a coefficient of 1.99, significant at a 5% level. This implies that, although financial growth facilitates capital formation and industrial activity, it may inadvertently contribute to environmental degradation in the absence of strong green finance regulations. LREC, however, exhibits a significant negative impact on emissions, with a coefficient of −0.72 at a 1% level. This clearly highlights the environmental benefits of clean energy deployment, supporting policy directions that prioritize the decarbonization of China’s energy mix.

LHUC also shows a negative and statistically significant coefficient (−0.0865), indicating that investments in education and workforce skills are associated with improved environmental outcomes, potentially by fostering innovation, awareness, and adoption of low-carbon technologies. Additionally, LAVA contributes to emission reductions, with a coefficient of −0.2190, significant at a 1% level. This suggests that improvements in agricultural productivity—likely involving sustainable land use and carbon-absorbing practices—can serve as a meaningful lever in mitigating environmental harm.

The overall model fit is exceptionally high, with an R² of 0.9954 and an adjusted R² of 0.9932, indicating that over 99% of the variation in CO₂ emissions is explained by the selected explanatory variables. The F-statistic is also significant at a 1% level (p = 0.0000), underscoring the joint significance of the model. Collectively, these results confirm that although economic and financial expansion exert upward pressure on emissions, improvements in renewable energy adoption, human capital, and sustainable agriculture play a vital role in offsetting these negative effects and steering China toward a more environmentally sustainable development trajectory.

The empirical results derived from the DOLS model both confirm and challenge the existing literature, offering a nuanced understanding of China’s sustainable development dynamics. The positive and statistically significant relationship between economic growth and CO₂ emissions aligns with the early-stage EKC hypothesis and supports prior findings by Xu et al. (2024), who observed that, in rapidly industrializing economies like China, growth initially coincides with environmental degradation due to the dominance of fossil fuel consumption and carbon-intensive industries. This trend also mirrors the conclusions of Zheng et al. (2025), who reported that economic expansion in developing economies tends to accelerate emissions unless accompanied by structural reforms and green investments.

Interestingly, the study finds that financial development significantly increases emissions, a result that diverges from the argument made by Yiming et al. (2024), who proposed that a more mature financial sector can lead to better environmental quality through improved resource allocation and access to green technologies. However, our findings are consistent with those of Cui et al. (2023), who cautioned that, in emerging markets, financial deepening may initially fund industrial and infrastructure growth with limited environmental oversight, thus exacerbating pollution. This result suggests that without proper regulatory alignment, financial development may not inherently support environmental goals.

In contrast, the strong negative effect of renewable energy consumption on CO₂ emissions reaffirms the conclusions of Wu et al. (2023) and Dilanchiev et al. (2024b), both of whom demonstrated that renewable energy significantly mitigates environmental degradation, particularly when deployed at scale. The significance of human capital in reducing emissions is also consistent with Shi and Umair (2024), who highlighted the role of education and technological knowledge in promoting cleaner production methods and environmental stewardship. This result suggests that investment in human capital may have spillover effects on environmental quality through innovation and efficiency.

Furthermore, the negative impact of agricultural value-added on emissions supports the findings of Li and Umair (2023a), who noted that modernized and sustainable agricultural practices can enhance carbon sequestration and reduce emissions per unit of output. This is particularly relevant in China, where agricultural reforms and afforestation policies have increasingly emphasized environmental conservation.

Table 7 presents the results of FMOLS and CCR estimations, both of which are used to validate the robustness of the long-run relationships among the model variables. The FMOLS results indicate that economic growth (LGDP) has a strong and statistically significant positive effect on CO₂ emissions, with a coefficient of 0.7922, which is higher than that of the DOLS estimates, reinforcing the notion that China’s growth trajectory remains carbon-intensive. LFD also shows a positive and significant impact (0.1536), albeit smaller in magnitude, suggesting that financial expansion contributes to increased emissions, possibly through energy-demanding investments. LREC maintains a negative and significant coefficient (−0.8597), implying that clean energy deployment significantly reduces environmental degradation. LHUC and LAVA also exert negative and statistically significant effects on emissions, further supporting the role of education and sustainable land use in environmental performance.

Under the CCR model, all variables remain statistically significant, and the signs of the coefficients are consistent with FMOLS. However, the magnitudes differ slightly: LGDP’s coefficient is smaller (0.3517), suggesting a more moderate influence of growth when controlling for endogeneity and serial correlation. LFD increases in effect size under CCR (0.4965), while LREC’s negative coefficient remains significant, albeit with a smaller magnitude (−0.0289). The consistent significance and direction of coefficients across both models underscore the robustness of the core findings. Both FMOLS and CCR models report high R-squared values (0.9935 and 0.9945, respectively), confirming a strong explanatory power of the regressors. These outcomes reinforce the validity of the long-run associations established in the DOLS results and provide strong empirical support for policy interventions aimed at decarbonizing growth through financial, educational, and energy system reforms.

The diagnostic inspection presented in Table 8 validates the reliability and statistical soundness of the estimated model. First, the Jarque–Bera test for normality yields a test statistic of 1.15794, with a p-value of 0.5512, indicating that the residuals are normally distributed and do not deviate from the Gaussian assumption—an essential prerequisite for inference in linear regression models. Second, the Lagrange multiplier (LM) test for serial correlation shows a p-value of 0.4956, which fails to reject the null hypothesis of no autocorrelation. This result ensures that the error terms are not systematically related over time, affirming the temporal independence required for consistent estimators.

Additionally, the Breusch–Pagan–Godfrey test for heteroscedasticity reveals a p-value of 0.6214, confirming that the residuals have constant variance over time and across observations. This homoscedasticity assumption is fundamental for the validity of standard errors, confidence intervals, and hypothesis testing. Beyond these statistical tests, the structural stability of the model was further assessed using the CUSUM and CUSUM of squares (CUSUMQ) tests. The plots from both procedures confirm that the cumulative sum of recursive residuals and their squared values remain within the 5% confidence bounds, implying no structural break or instability in the model throughout the estimation period. Collectively, these results provide robust evidence that the model is well specified, the estimators are efficient, and the inferences drawn from the regression results are statistically reliable and valid for policy analysis.

The Granger causality test results offer critical insights into the dynamic interrelationships among the key variables affecting (Figure 1) environmental performance (CO₂ emissions) in China. As shown in Table 9, a series of statistically significant F-statistics allows us to reject the null hypothesis of no causality in several key pairwise cases. Notably, LGDP is found to Granger-cause CO₂ emissions, confirming the unidirectional causal pathway from economic expansion to environmental degradation. Similarly, LFD also Granger-causes CO₂ emissions, indicating that financial sector expansion may contribute to environmental pressures through increased industrial and energy activities. However, no causal relationship is observed between LREC and CO₂ emissions, suggesting that the transition toward clean energy has not yet reached a critical mass sufficient to exert a detectable impact on emission reduction.

Furthermore, LHUC shows a bidirectional causal relationship with both CO₂ emissions and economic growth, highlighting the central role of human capital not only in fostering growth but also in influencing environmental quality. The finding that LHUC Granger-causes LCO₂ and vice versa suggests complex feedback loops potentially driven by education, awareness, and sustainable practices. Meanwhile, LAVA does not Granger-cause CO₂ emissions, indicating that its role in environmental sustainability might be mediated through other channels, such as carbon absorption, rather than direct emission linkage.

In terms of interactions among explanatory variables, the analysis reveals that financial development is Granger-caused by agricultural productivity, pointing to the role of rural and agricultural investment in driving financial inclusion and credit expansion. Renewable energy consumption is causally affected by financial development and agricultural value-added, further supporting the hypothesis that enabling sectors—finance and agriculture—are critical to renewable energy adoption. Finally, the bidirectional causality between human capital and economic growth reinforces the developmental paradigm, where skill accumulation and knowledge diffusion are mutually reinforcing with GDP growth. These results substantiate a web of interconnected dynamics that must be carefully considered in designing coherent, evidence-based policies for sustainable development in China.

To ensure the robustness and reliability of the empirical findings, several diagnostic and validation tests were performed. First, multicollinearity among the explanatory variables was assessed using the variance inflation factor (VIF). As presented in Table 10, all VIF values were well below the conventional threshold of 5, indicating that multicollinearity does not pose a serious threat to the validity of the estimated coefficients. In particular, VIF values ranged from 1.88 for LREC to 3.22 for LFD, confirming that the explanatory variables are sufficiently independent and the model is well-specified for long-run estimation.

In addition to multicollinearity diagnostics, impulse response-like simulations were conducted to evaluate the stability and direction of environmental performance (EP) in response to shocks in each explanatory variable. Figure 2 illustrates the dynamic EP responses following one-standard deviation shocks in GDP, financial development, renewable energy, human capital, agricultural value-added, and CO₂ emissions. The plots demonstrate consistent and theoretically coherent patterns: economic and financial shocks positively influence EP over time, while a shock to CO₂ emissions causes a deterioration in EP, especially in the short term. These patterns provide strong validation of the long-run elasticity estimates obtained via the DOLS model.

To further reinforce these findings, Figure 3 presents the disaggregated response panels under varying shocks. These multi-panel visualizations confirm that the impact of each variable on environmental performance is not only statistically robust but also directionally consistent across the forecast horizon. Overall, the robustness checks—combining multicollinearity diagnostics and dynamic response analysis—support the credibility of the model and ensure that the policy implications derived from the empirical results are grounded in a well-tested econometric framework.

Based on the empirical findings of this study, several targeted policy recommendations emerge that align directly with the statistical relationships observed between key variables and CO₂ emissions in China over the 1996–2022 period. First, since economic growth was found to significantly increase emissions, it is essential for Chinese policymakers to adopt a growth model that is decoupled from carbon-intensive industrialization. This includes redirecting public and private investments toward low-emission sectors and implementing stricter environmental regulations for high-polluting industries, particularly those that expand rapidly during periods of accelerated GDP growth.

Second, the study confirms that financial development contributes positively to CO₂ emissions, suggesting that the financial sector currently supports carbon-intensive economic activity. To address this, financial regulators should mandate the integration of environmental risk assessments in credit allocation and establish green finance taxonomies that differentiate and prioritize environmentally sustainable investments. The government should also expand incentives for banks and investment firms to fund renewable energy, energy efficiency, and low-carbon technologies through green bonds, concessional loans, and risk-mitigation instruments.

Third, the significant negative effect of renewable energy consumption on emissions highlights the importance of scaling up clean energy deployment. Policy should focus on accelerating the integration of renewables into the national grid by improving storage capacity, enhancing smart-grid infrastructure, and eliminating fossil fuel subsidies that disincentivize clean energy uptake. The empirical results further support the introduction of performance-based subsidies and feed-in tariffs tailored to wind, solar, and hydro sectors to maximize emission reduction potential.

Fourth, given the inverse relationship between human capital and CO₂ emissions, the government should increase investment in education, particularly in green skills, environmental engineering, and climate-focused research and development. Environmental literacy should be embedded into all educational levels to raise awareness and promote sustainable consumption behavior across society. This also implies incentivizing the private sector to engage in employee upskilling programs focused on energy-saving practices and green technologies.

Finally, the finding that agricultural value-added reduces emissions suggests that modernizing and decarbonizing the agricultural sector is an effective mitigation strategy. Policymakers should promote climate-smart agriculture by providing technical training, access to low-emission technologies, and financial support for practices that enhance carbon sequestration, such as organic farming, no-till cultivation, and efficient irrigation. Targeted rural investment programs and subsidies for low-carbon agricultural equipment can amplify these benefits while boosting productivity. These sector-specific recommendations, grounded in empirical evidence, are crucial for guiding China’s path toward an environmentally sustainable and economically resilient future.

Although this study provides valuable insights into the long-run dynamics between financial development, renewable energy consumption, and CO₂ emissions in China, it has several limitations. The analysis is confined to a single country and relies solely on macro-level time-series data, which may mask regional disparities and sector-specific variations. Additionally, the study focuses only on carbon dioxide emissions as the indicator of environmental degradation, omitting other important pollutants such as methane, nitrous oxide, and particulate matter. The econometric framework, although robust, does not account for structural breaks or nonlinear dynamics that may influence the variables over time. Future research should extend the analysis by incorporating a broader set of environmental indicators, applying panel data across multiple emerging economies, and exploring nonlinear models such as threshold regressions or asymmetric ARDL to better capture complex relationships. Moreover, disaggregated sectoral or provincial data could provide more granular policy guidance tailored to regional environmental and economic realities.