The relationship between REN and GTFP has already been studied extensively. Most studies believe that REN can have a positive impact on GTFP by improving energy efficiency and reducing negative environmental externalities (Huang H. et al., 2025; Sun et al., 2025). Studies from developed countries and emerging economies have shown that replacing fossil fuels with renewable energy can improve production efficiency and reduce carbon emissions, thus promoting sustainable economic development (Bashir et al., 2024; Sohag et al., 2021). Some studies also point out that advanced technology, human capital and investment in R&D will significantly affect macroeconomic output (Amin et al., 2025; Šlander-Wostner et al., 2025). However, existing studies have not reached completely consistent conclusions. For example, Tugcu and Tiwari (2016) found that between 1992 and 2012, the development of renewable energy in BRICS countries did not have a significant impact on total factor productivity growth. The disparity could be due to the extent of data coverage, testing techniques, or environmental context. A few studies have also found a non-linear relationship between REN and GTFP (Xie et al., 2021). Xie et al. (2021) found that there is a threshold effect in the relationship between green finance and green total factor productivity, which becomes significant only when green finance development exceeds a certain threshold. It is worth noting that, in the Chinese context, the actual effectiveness of renewable energy applications is also affected by regional transition models and policy changes (Wu et al., 2025a). Despite the widespread attention this type of research has received, there are still disagreements regarding its empirical results. Against the backdrop of energy transition, this study focuses on the impact of REN on GTFP, as well as its potential nonlinear characteristics.

Enhancing GTFP has become crucial for attaining high-quality growth as China’s economy enters a new phase, with GTI playing a key part in this process (Shen et al., 2022). Prior empirical studies have highlighted GTI’s ability to boost productivity, primarily through improvements in efficiency and regulation-induced innovation output (Zhao et al., 2025). Government digitization and green financing strengthen the effectiveness of such technologies by improving governance capability and resource allocation efficiency (Huang X. et al., 2025; Zeng et al., 2025). On the one hand, GTI enhances energy efficiency and adopting cleaner technologies to lessen the reliance on traditional inputs that create negative environmental externalities (Wang J. et al., 2023). Additionally, stricter environmental regulations promote green innovation opportunities, which lead to green productivity-enhancing effects (Feng et al., 2017). Further, there is evidence in the literature that the link between GTI and GTFP has a conditional, nonlinear character. For instance, Yan et al. (2020) found that renewable energy technological innovation contributes to green productivity only after the regional income level reaches a certain threshold. Likewise, following the “appropriateness” theory and the “productivity paradox,” Kong et al. (2015) argued that when technological innovation is inappropriate for the country’s development stage or the structure of factor endowments, its productivity-enhancing impact on GTFP will possibly be very poor. While GTI is considered an essential driver for GTFP, its real effectiveness is affected by the institutional environment in which it operates. However, in context with China’s economic transition, whether the process by which GTI influences GTFP is influenced by finance and government intervention has not yet been further examined.

GF, by reallocating capital to environmentally friendly activities, has become an important mechanism for promoting sustainable economic development. Most macro-level empirical data demonstrates this positive relationship. For instance, Meng et al. (2024) and Lee and Lee (2022) illustrated that GF greatly boosts GTFP through facilitating technological innovation and upgrading the energy structure, with stronger effects in supportive economic and institutional settings. However, the empirical conclusions at the micro level are inconsistent. For example, Piao et al. (2025) discovered that green financial policies adversely effect on GTFP in highly polluting industries, because such policies will aggravate the financing constraints of enterprises and increase the cost of transformation. The relationship between GF and GTFP has nonlinear characteristics besides the linear effect. Shi and Shi (2022) found that there is a threshold effect in the relationship between GF and GTFP, which becomes significant only when green finance development exceeds a certain threshold. The effects of GF on the green economy differ significantly due to institutional and market environments, including fluctuations in carbon prices and economic uncertainty (Wang K. H. et al., 2023; Wu et al., 2025b). The present body of research primarily examines the influence of GF on GTFP from macro or micro perspectives. The synergistic effects of GF, REN, GTI, and GOV on GTFP, along with their regional heterogeneity, and the moderating role of GF in the influence of GTI on GTFP, are still relatively unexplored at the meso-level (provincial).

Studies already conducted indicate that through regulation, subsidies, and policy coordination, the government, acting as the “visible hand,” greatly raises GTFP (Zhang Z. et al., 2024; Zhou D. et al., 2023). Zhao and Ying (2023) pointed out that the government can promote the transformation of energy structure and improve GTFP by formulating energy policies. Under the “dual carbon” target, the government can provide funds to green enterprises to help them control pollution and reduce emissions (Ma et al., 2025), while also promoting green innovation through subsidies (Ren and Lv, 2014). However, further study has highlighted that excessive or improper government involvement may distort market incentives and resource allocation, thereby reducing GTFP. For example, Wang and Luo (2018) found that government subsidies may give rise to overcapacity and overinvestment in certain industries. Similarly, according to Brandt et al. (2013), local governments are prone to misallocating resources to meet short-term economic growth targets, which distorts factor market prices and eventually makes it more difficult to promote GTFP. Despite the lack of consensus in prior research findings, the nonlinear impact of GOV on GTFP, along with its moderating role in the process through which GTI influences GTFP, has yet to be fully investigated using provincial-level data.

Promoting sustainable development requires the coordinated advancement of economic growth and environmental protection. GTFP has therefore been widely adopted, particularly in China, to evaluate the level of high-quality economic development and ecological efficiency. Since GTFP inherently integrates these two dimensions, identifying its key determinants is crucial. Renewable energy transition plays a key role in promoting economic and environmental sustainability. Research indicates that REN increases energy efficiency and reduces fossil fuel use and environmental damage (Zhao et al., 2021). The energy-growth nexus theory also suggests that energy use and economic growth are closely connected. In this framework, REN contributes to growth and reduces ecological pressure, supporting green and sustainable development (Wang et al., 2024). Therefore, REN is expected to positively affect GTFP, as proposed in Hypothesis 1, i.e., β G T F P i t β R E N i t > 0 .

Endogenous growth theory highlights technological innovation as a primary internal driver of economic growth. The development of clean energy heavily depends on technological advancement, with GTI playing a central role (Jiakui et al., 2023). Previous studies show GTI can enhance energy efficiency, alleviate emission reduction pressures, improve the utilization efficiency of renewable energy, and promote green economic growth (Wang J. et al., 2023). It is evident that GTI primarily contributes to sustainable economic development by enhancing green production efficiency. Thus, GTI is assumed to exert a beneficial influence on GTFP, i.e., β G T F P i t β G T I i t > 0 , as stated in hypothesis 2.

Similarly, GF enhances GTFP primarily through capital allocation and risk aversion effects. On the one hand, GF directs productive resources toward more environmentally efficient sectors. Guided by the ESG (Environmental, Social, and Governance) principles, green finance theory indicates that the financial institutions reallocate funds from high-pollution and energy-intensive industries to green technology and clean energy projects through financial instruments (Xu et al., 2023). By alleviating financing constraints on green investments and optimizing resource allocation through market-based signals, GF contributes to the improvement of GTFP (Behera B. et al., 2024). On the other hand, GF internalizes environmental risks through risk avoidance mechanisms, thereby incentivizing economic entities to accelerate the green transition. Green finance instruments, such as green credit standards, increase financing costs and raise compliance costs for pollution-intensive firms, driving them to internalize environmental externalities and accelerate innovation of green technology (Zhang et al., 2022). This risk-pricing mechanism boosts corporate incentives for green investment, improves resource allocation, and fosters green technical innovation, ultimately leading to the enhancement of GTFP (Jiakui et al., 2023; Lee and Lee, 2022). Based on these theoretical foundations, this study hypothesizes that GF has a positive impact on GTFP, i.e., β G T F P i t β G F i t > 0 , as stated in hypothesis 3.

In addition, the government affects green production efficiency via regulation and market incentives. Institutional distortions will result from inadequate governance and improper policy design when government involvement is limited, such as resource mismatch and crowding out market activity space (Tang and Qin, 2022). In this situation, government actions often replace the market regulation function, weaken the role of price signals, and direct resources such as capital and labor to areas with low efficiency. At the same time, subsidy-based support will also crowd out private investment and reduce enterprises’ initiative for independent innovation (Aragie and Pauw, 2019). Therefore, the GOV in this case is not a supplement to the market, but a substitute for the market, which will hinder the promotion of GTFP.

When the government intervenes more, its role will change. Government behavior is no longer replacing the market but is beginning to help correct its failure. For example, environmental taxes and carbon pricing incentivize enterprises to consider environmental costs when making operational and production decisions. In addition, technical standards and R&D subsidies facilitate companies adopting and spreading green technologies. Together, these initiatives increase energy efficiency while promoting the use of renewable energy sources, which leads to higher green total factor productivity. As illustrated in Figure 1. Thus, hypothesis 4, there is a U-shaped relationship between GOV and GTFP.

Moreover, GF reduces financing costs and creates an advantageous policy environment that encourages investment in R&D, especially when combined with moderate government intervention, thereby fostering GTI (Akhtaruzzaman et al., 2025; Xu and Lin, 2024), contributing to sustainable development. Given that the interaction effects of GF and GOV with GTI are estimated to promote the improvement of GTFP, the interaction terms’ coefficients are assumed to be positive, i.e., β G T F P i t β G F × G T I i t > 0 , and β G T F P i t β G O V × G T I i t > 0 , in hypothesis 5 and hypothesis 6, respectively. A diagram of the theoretical framework and hypotheses is shown in Figure 2.

For empirical analysis, this study followed Sethi et al. (2024) and formed the baseline equation, which is illustrated in Equation 1. Ln G T F P i t = α + β 1 ⁡ ln ⁡ R E N i t + β 2 ⁡ ln ⁡ G F i t + β 3 ⁡ ln ⁡ G T I i t + β 4 ⁡ ln ⁡ G O V i t + β 5 Z i t + μ i t (1)

Where in Equation 1 the subscripts i and t represent the province and year, respectively. G T F P i t is the dependent variable, measuring green total factor productivity, R E N i t , G F i t , G T I i t , and G O V i t are independent variables, denoting renewable energy consumption, green finance, green technology innovation, and government intervention, respectively. Z i t denotes the vector of control variables, including transportation infrastructure, trade openness, and industrial structure upgrading, and μ i t denotes the residual term.

Furthermore, to examine the potential non-linear effects of REN and GOV on GTFP, Equation 1 is extended by incorporating the squared terms of REN and GOV, as presented in Equations 2, 3, respectively. Ln G T F P i t = θ + γ 1 ⁡ ln ⁡ R E N i t + γ 2 ⁡ ln ⁡ G F i t + γ 3 ⁡ ln ⁡ G T I i t + γ 4 ln ⁡ R E N i t 2 + γ 5 ⁡ ln ⁡ G O V i t + γ 6 Z i t + μ i t (2) Ln G T F P i t = ϑ + δ 1 ⁡ ln ⁡ R E N i t + δ 2 ⁡ ln ⁡ G F i t + δ 3 ⁡ ln ⁡ G T I i t + δ 4 ⁡ ln ⁡ G O V i t + δ 5 ln ⁡ G O V i t 2 + δ 6 Z i t + μ i t (3)

Where ln ⁡ R E N i t 2 and ln ⁡ G O V i t 2 represent the squared term of renewable energy consumption and government intervention in Equations 2, 3, respectively. The remaining variable symbols’ interpretations align with those identified in Equation 1.

Moreover, to assess the moderation impact of GTI with GF and GOV on GTFP, Equation 1 is expanded by incorporating the interaction of ( G F × G T I ) and ( G O V × G T I ) in Equations 4, 5. Ln G T F P i t = ϱ + φ 1 ⁡ ln ⁡ R E N i t + φ 2 ⁡ ln ⁡ G F i t + φ 3 G F × G T I + φ 4 ⁡ ln ⁡ G T I i t + φ 5 ⁡ ln ⁡ G O V i t + φ 6 Z i t + μ i t (4) Ln G T F P i t = ρ + ϕ 1 ⁡ ln ⁡ R E N i t + ϕ 2 ⁡ ln ⁡ G F i t + ϕ 3 ⁡ ln ⁡ G T I i t + ϕ 4 ⁡ ln ⁡ G O V i t + ϕ 5 G O V × G T I + ϕ 6 Z i t + μ i t (5)

Where G F × G T I denotes the interaction term between green finance and green technology innovation in Equation 4, G O V × G T I refers to the relationship between green finance and green technological innovation in Equation 5. All other variables are defined as in Equation 1.

China, the economy that is the second largest in the world, has long had serious issues with environmental pollution. To achieve sustainable economic development, the country actively promotes the establishment of an ecological civilization and participates in global sustainable development governance initiatives. In 2015, the country proposed the goal of “green development” and subsequently formulated and implemented a series of related economic policies around it. Against this backdrop, this study constructed a balanced panel dataset covering 30 provinces in China (Tibet was excluded due to severe data gaps) based on data reliability and availability. The dataset comprises 390 observations, covering the period from 2010 to 2022. This study seeks to examine the major determinants of economic sustainability, employing GTFP as a key metric.

Following Tone and Sahoo (2003) and Park et al. (2024), GTFP is assessed utilizing the Global Malmquist-Luenberger (GML) index, which is obtained through a non-radial, non-oriented super-efficiency SBM method. The GML indices for the 30 provinces were calculated using Max-DEA Ultra software, incorporating non-desired outputs. Input indicators include labor, capital, and energy: The employee number serves as a proxy for labor; the perpetual inventory approach is used to construct capital stock, which serves as a proxy, with real gross fixed capital formation as investment input and a constant depreciation rate of 9.6%, and energy is represented by the total energy consumption. Output indicators are divided into desirable and undesirable outcomes. The desired output is real GDP (base year 2006), adjusted for pricing considerations to reflect the level of economic development. Undesired outputs include emissions of industrial wastewater, sulfur dioxide, smoke, dust, and carbon dioxide, which represent environmental pollution levels. Table 1 displays the definitions for each variable.

Furthermore, REN, GTI, GOV, and GF are included as the explanatory variables. A majority of studies use the ratio of electricity generation from renewable sources to total electricity generation to evaluate the development of renewable energy (Song et al., 2022; Zhu et al., 2022). Drawing on Yu et al. (2021), the ratio of green patents to total patents is used to calculate GTI, according to the National Intellectual Property Administration of China. GOV is measured by fiscal expenditure as a percentage of regional GDP (Lee et al., 2024). Following the study of Liu et al. (2019), this study measures green finance across four dimensions: green credit, investment, securities, and insurance. To refine the indicator system, this study proposes adding an additional dimension that incorporates environmental protection support, considering fiscal policy (see Table 2). Furthermore, following Lee and Lee, 2022, this study calculates a comprehensive index using the entropy method for objective weighting.

Control variables include transportation infrastructure (TI), openness (OP), and industrial structure upgrading (IS), which is in line with previous research on green productivity and sustainable development (Hosan et al., 2022; Liu et al., 2024). Enhancements to road infrastructure can mitigate traffic congestion, thus reducing energy consumption and pollution emissions, which benefits GTFP (Zhao et al., 2022). To quantify infrastructure, this study uses the logarithm of highway mileage in each province. OP has an impact on a country’s GTFP by optimizing its industrial structure and influencing technological spillovers (Zhang, 2021), it is measured by the share of goods and services traded in GDP. Following Zhao et al. (2024), IS is the value addition ratio between the secondary and tertiary sectors. To reduce data variability and ensure consistency, all variables are transformed into their logarithmic forms. Minor missing values are handled using interpolation techniques. Table 3 provides descriptions of the variables and the data sources.

To ensure the reliability of the panel estimations, several standard diagnostic tests were conducted prior to model estimation. Specifically, cross-sectional dependence was assessed employing the CSD test as developed by Friedman (1937) and Pesaran (2004); heteroskedasticity was evaluated using the Breusch–Pagan test (Breusch and Pagan, 1979); serial correlation was examined using (Wooldridge, 2010); the Fisher-ADF and IPS panel unit root tests were used to check for stationarity; cointegration long-run relationships were examined through cointegration tests proposed by Kao (1999) and Westerlund (2005); and multicollinearity was diagnosed using variance inflation factors (VIF). Moreover, the appropriate specification between fixed-effects and random-effects models was determined using the Hausman test (Hausman, 1978).

Based on the diagnosis results, this study uses the Generalized Least Squares (GLS) method as the main estimation approach. Further, serial correlation and cross-sectional dependence in the error terms are also addressed using Driscoll-Kraay (D-K) standard errors with a fixed-effects model. This study employs the two-step System GMM method established by Blundell and Bond (1998) to enhance the robustness of the results and reduce potential endogeneity in a dynamic panel scenario. Using the System GMM method, the lagged dependent variable and the main explanatory variables (REN, GTI, GF, and GOV) are set as endogenous, and the other control variables are set as exogenous. The instruments are built from lagged values starting from period t-2, and the instrument set is collapsed to limit the number of instruments. Windmeijer-corrected standard errors are used for the two-step estimates.

This study conducts empirical analysis with STATA 18. As a first step, Table 4 displays the descriptive statistics and correlation matrix. The mean value of GTFP is 0.695, and its standard deviation is 0.358. The average value of REN is 5.584, and the mean value of GTI is 9.553. Among all variables, GTI and REN have the largest standard deviations, which shows that these variables vary greatly across provinces. On the other hand, the standard deviations of GF and GOV are 0.123 and 0.102, respectively, while their mean values are 0.320 and 0.245. Furthermore, the correlation results indicate generally positive relationships between GTFP and the explanatory variables, except for GOV, IT, and IS.

To assess multicollinearity, a variance inflation factor test was performed (see Table 5). There is no evident multicollinearity among the variables, as shown by the average VIF value of 2.84, which is below the normal standard of 5. Following this, the IPS and ADF tests were used to check whether the variables were stationary.

The outcomes of unit root tests in Table 6 indicate that all the variables except for GTI, TI, and IS are stationary at “first difference, stationary [I (0)]”. Following this, this study assesses the existence of long-run relationships using the Westerlund cointegration, Pedroni, and Kao tests In Table 7. The significant p-values from the Westerlund panel cointegration test indicate a long-term link among the variables. Additionally, a series of diagnostic tests are conducted, as presented in Table 8. These comprise the Hausman test, the Wooldridge autocorrelation test, and the Breusch-Pagan heteroskedasticity test.

The Hausman test rejects the null hypothesis at the 5% significance level, thus the fixed effects model is used in Table 8. Furthermore, the Wooldridge test for autocorrelation and the Breusch-Pagan test for heteroskedasticity are both statistically significant, suggesting autocorrelation and heteroskedasticity in the model, respectively.

Additionally, Table 9 shows the results of the Friedman (1937) and Pesaran (2004) cross-sectional dependence tests, indicating that cross-sectional dependence exists in the dataset. To identify these issues, this study estimates Equations 1–5 (illustrated with Models I to V) using the GLS and Driscoll–Kraay (D-K) estimators, with the results reported in Tables 10, 11.

This study re-estimates the baseline models using the two-step System GMM method to address potential endogeneity and verify the robustness of the results. Table 13 presents the outcomes.

Model-I in Table 13, both REN and GTI have positive effects on GTFP. GTFP improves by 0.142% and 0.084% per unit increase in REN and GTI, respectively. GF also has a positive but statistically insignificant influence on GTFP. An explanation might exist in institutional and policy frictions, including weak policy coordination, a regulatory preference for low-risk projects, and a green credit orientation that emphasizes scale over efficiency, all of which might hinder the crucial role of green finance in promoting productivity-driven green innovation (Zhang and Vigne, 2021). However, the results show that GOV reduces GTFP in China, with an estimated effect of −1.823%.

In Model II, the quadratic term of renewable energy exhibits a significant positive influence on GTFP (coefficient = 0.063, p < 0.05). Model III shows a U-shaped relationship between GOV and GTFP, with a nonlinear term coefficient of 0.144 and an inflection point of 0.39. This shows that when the level of intervention is low, its impact is significantly negative; only when the intervention intensity exceeds a certain level can it have a beneficial effect. In Model IV, while both GF and GTI boost GTFP, the coefficient of their interaction term is positive but not statistically significant. One possible explanation is that the institutional and structural constraints of China’s green financial system have weakened its role in promoting green technological progress and productivity improvement. In Model V, both GOV and GTI show positive direct effects on GTFP. Furthermore, their interaction term is found to provide a significant boost to GTFP.

The model diagnostics confirm estimation reliability. The Arellano-Bond AR (1) test identifies significant first-order autocorrelation, whereas the AR (2) test fails to reject the null hypothesis of no second-order serial correlation, confirming proper model specification. The Hansen J-test shows that the overidentifying restrictions are valid, confirming instrument appropriateness.

Although the explanatory variables exhibit significant correlations with the dependent variable, correlation alone does not imply structural causality. Therefore, this study utilizes the Dumitrescu-Hurlin (D-H) panel causality test to investigate Granger causality among variables, specifically whether the lagged values of one variable can predict the current values of another. The D-H test allows cross-sectional dependence and is appropriate for heterogeneous panel data, and it is applicable when the variables are stationary in levels or first differences (Sethi et al., 2024). The lag length is selected based on standard information criteria, and alternative lag specifications are examined as robustness checks. The results indicate that the detected Granger-type relationships are primarily evident at one-period lags, while they weaken under longer lag structures.

The results reported in Table 14 provide evidence of short-run Granger-type predictability between REN, GOV, IS, and GTFP, suggesting mutually reinforcing temporal linkages. In addition, GOV is found to Granger-cause REN, GTI, OP, and IS, while exhibiting a unidirectional predictive relationship with GF and TI. The control variables TI, OP, and IS also display bidirectional Granger-causal links with GTFP.

Figure 4 shows a visual summary of the results, which are consistent across the different estimating methodologies employed.