Impacts of financial development on the energy consumption in China from the perspective of poverty alleviation efficiency

Poverty alleviation and energy saving are two major issues of sustainable development targets. Meanwhile, financial development (FD) is a powerful engine of economic growth, which is regard as a valid approach to contain the demand for energy consumption (EC). However, few studies link the three factors and explore the specific impact mechanism of poverty alleviation efficiency (PE) on the tie between FD and EC. Thus, we employ the mediation and threshold models to evaluate the influence of FD on the EC in China during 2010–2019 from the perspective of PE. We affirm that FD indirectly promotes EC through the channel of PE. The mediating effect of PE is responsible for 15.75% of the total effect of FD on the EC. Moreover, FD generates a significant threshold impact on the EC considering the change of PE. When the PE exceeds 0.524, the role of FD in promoting EC is strengthened. Ultimately, the outcome suggests policymakers need to prominent the trade-off between energy saving and poverty reduction during the fast evolution of financial system.


Introduction
Due to the acceleration of urbanization and industrialization, China has turned into the top energy user in the world in 2010 (Zheng et al. 2019). During 2005-2020, China's energy consumption (EC) undergoes a violent increase with an average annual growth rate of 6.03% year −1 . Since China's economy has achieved recovery after the efficient control of epidemic in 2021, the bounce back of energy consumption demand will probably occur in China (Fang et al. 2018). Moreover, even if the proportion of coal consumption in China's energy structure has decreased, it is responsible for nearly 60% of the total as main energy source in 2020 (National Bureau of Statistics of China 2021). The considerable EC not only leads to ecological problems but also threats China's energy safety and green development (Liu and Diamond 2005;Mi et al. 2017;Cui et al. 2021). Therefore, perusing decarbonization strategies is the urgent mission for China (Mi et al. 2018;Cheng et al. 2020;. China is committed to achieving peak carbon emissions by 2030, increasing the share of non-fossil energy to 20%, and reaching carbon neutrality by 2060 (Duan et al. 2018;Qi et al. 2020). On this premise, exploring the influencing factors of EC and related mechanisms is necessary (Cao and Wang 2022;Ji et al. 2022). Moreover, the valid integration of poverty alleviation and EC reduction is a remarkable factor affecting China's sustainable economic development (Li and Yuan 2023). China has the highest proportion of poor people around the world, and poverty alleviation is inseparable from economic growth that usually requires extensive energy input Guo et al. 2021). Therefore, the promotion of poverty alleviation efficiency (PE) has a two-sided influence on the EC, and the specific impact mechanism is deserved to be explored.
The improvement of financial system, as an available approach of influencing China's economic growth, is helpful to relieve information asymmetry and optimize the distribution of production resources, thereby generating critical implications of sustainable development (Zhang et al. 2019;Yang and Zhang 2020;Wu and Wang 2022). Additionally, financial development (FD) plays a crucial part in the poverty alleviation (Deng et al. 2022), thus generating an impact on the PE. Existing literature have identified FD and poverty as two major driving factors of EC. Also, prior studies have confirmed the connection between FD and poverty reduction. Therefore, it is reasonable to believe that poverty alleviation may play an intermediary role between FD and EC to some extent. Nevertheless, there is no article trying to link the three factors, especially focusing on the specific influence mechanism of PE on the link between FD and EC. Meanwhile, since poverty alleviation has a two-sided influence on the EC, and the effect of FD on the EC may be determined by the indirect influence related to poverty alleviation, FD may generate a threshold effect on the EC in terms of PE. Thus, we should further verify the nonlinear relationship between FD and EC based on PE. However, previous scholars did not focus on the potential mediation and threshold effects of PE on the FD-EC nexus.
Thus, in addition to test the direct influence of FD on the EC, we analyze the impact mechanism of FD on the EC considering PE as a critical transmission channel. Moreover, we explore the possible nonlinear effect of FD on EC using the PE as the threshold variable. Our contribution lies in two aspects. First, previous studies did not conduct the nexus analysis of FD and EC from the perspective of poverty reduction efficiency. We evaluate the PE of China's 30 provinces using a super-efficiency DEA, and integrate FD, PE, and EC into an analytical framework based on the mediation-STIRPAT model. This helps us investigate the mediating effect of PE on the linkage between FD and EC, adding more insights into the role of FD in sustainable development. Additionally, prior studies indicated the linear relationship between FD and EC; however, they did not develop any studies on the nonlinear relationship between FD and EC using the PE as the threshold variable. This may lead to misleading policy implications without considering its nonlinear impact. Thus, we employ a panel threshold STIRPAT model to study whether the impacts of FD on the EC differ with the evolution of PE. This help governments make high-efficiency policy decisions considering multiple sustainable development goals.
Rest of the study is structured as follows. "Literature review" provides literature review. "Methods and data" clarifies the methods and data sources. "Results" presents the results. "Discussion" discusses the results. "Conclusions" concludes and raises policy recommendations.

Literature review
Past studies have concluded that FD can impact EC (Tian et al. 2022), but there is no unified conclusion. Ma and Fu (2020) and Wan and Sheng (2022) pointed out that FD enhances EC in several channels. First, the escalation of financial system provides a convenient approach for consumers to acquire credit supports, making them tend to consume energy intensive products, and thus raising EC (Islam et al. 2013;Yue et al. 2018). Second, FD can help companies more easily obtain financial assistance, which stimulates production expansion and increases energy demand (Saud et al. 2018). Third, FD accelerates economic growth through financial technology innovation and reasonable resource allocation, but economic growth needs to consume massive energy (Wang and Gong 2020). However, Zhong et al. (2019) and Saud et al. (2020) indicated that FD is conducive to declining EC. First, the developed financial system can reduce the financing costs of firms and effectively alleviate information asymmetry through financial intermediaries, such that firms can get enough funds to invest in energy-saving technology projects, thus reducing EC (Liao and Drakeford 2019;Kwakwa et al. 2018). Second, a good business environment with the prosperous financial market can attract substantial foreign capital, which provides additional funds for firms to develop new technologies, raising energy utilization efficiency and therefore reducing EC (Jalil and Feridun 2011;Chang 2015). Finally, FD upgrades EC structure by driving the technology innovation of energy industry, thus effectively reducing EC (Kwakwa 2020;Shahbaz et al. 2018). Other studies investigated the nonlinear relation between FD and EC (Pan et al. 2016;Riti et al. 2017;Yue et al. 2019). These nonlinear models allow to clarify the threshold, regimeswitching, and inverted-U-shaped nexus of FD and EC (Chiu and Lee 2020). They analyzed the threshold impacts of FD on the EC using the FD itself, economic development, and technological advancements as threshold variables (Chang 2015, Sare 2019, Uddin et al. 2022. In terms of the nexus of poverty and energy consumption, some studies suggested that eliminating poverty raises energy demand with consumption pattern shifts (Bruckner et al. 2022;Chen et al. 2022). Hubacek et al. (2017), Anser (2019), and Duarte et al. (2021) proposed that anti-poverty projects boost the increase of EC. However, Duarte et al. (2016) and Fu et al. (2021) believed that extricating people from poverty can enhance the environment consciousness in the society, thus reducing EC. Meanwhile, Jin et al. (2020) highlighted that poverty alleviation and EC can be decoupled, and the reliance of poverty alleviation on the EC declines. Wang et al. (2022b) and Han et al. (2020) proposed that China's photovoltaic poverty alleviation project, aiming at poverty eradication, can drive renewable energy development, decreasing EC.
Moreover, some studies conducted the nexus analysis of poverty and FD. For example, Jiang et al. (2020) and Zhu et al. (2021) indicated that FD is one of the key measures to effectively alleviate poverty. Because that financial institutions can lower the threshold of credit access and relax credit constraints for the poor, which can improve lowincome households' income capacity and their self-hematopoietic function (Yin et al. 2020;Yu et al. 2020). Moreover, a sound financial system can provide credit supports for the establishment of local infrastructure and the cultivation of featured local industries, which are helpful to improve labor productivity and increase the well-beings of low-income residents (Agbodji and Johnson 2019;Yang et al. 2022;Ye et al. 2022). However, some academics held opposite views, arguing that FD cannot help the poor out of their dilemma due to institutional barriers, information asymmetry, and the lack of adequate collateral, as the opportunities of obtaining financial services are unequal between the poor and rich (Seven and Coskun 2016;Kaidi et al. 2019;Mehta and Bhattacharya 2020).

Extended STIRPAT model
The IPAT model (Ehrlich and Holdren 1971) assumes that the driving factors of environmental pressure (I) is mainly composed of population (P), affluence (A), and technology level (T), and the three factors show an equal proportional change relationship with environmental pressure, i.e., I = PAT. Dietz and Rosa (1997) improved the IPAT model to nonlinear random regression STIRPAT model, which allows to add random items into the model, such that multiple control variables can be involved in the empirical analysis based on various research targets. Here, we use the extended STIR-PAT model to explore the impact of FD on EC. The standard STIRPAT model is shown as follows: where I is environmental pressure. α represent the coefficient of the model. P, A, and T represent the population, affluence, and technological level respectively. β 1 , β 2 , and β 3 are the estimated parameters for P, A, and T, respectively. i, t, and ε are the region, time, and random error terms, respectively. We take logarithms of all variables to eliminate the effect of heteroscedasticity. The model is as follows: To accurately evaluate the effect of FD on the EC, we construct an extended STIRPAT model learning from the research of Wang and Gong (2020). In detail, the environmental pressure (I) is represented by energy consumption (ec). Population size (pop) and economic development level (gdp), respectively, reflect the two elements of population (P) and affluence (A). The technical level (T) is substituted (2) ln I it = ln + 1 ln P it + 2 ln A it + 3 ln T it + ln it for the FD level (fd), as FD is an effective means to reasonably allocate resources, which unavoidably contributes to technological progress. Meanwhile, according to previous literature (Yu et al. 2022;Huo et al. 2021b;Zhao et al. 2021), we set industrial structure (is), energy intensity (ei), and energy poverty (ep) as control variables related to EC. We build the basic model as follows: where lnec represents EC, lnfd represents FD, lnpop represents population size, lngdp represents economic development level, lnis represents industrial structure, lnei represents energy intensity, and lnep represents energy poverty.

Mediation-STIRPAT model
To explore the mediating influence of PE on the association between FD and EC, we build a mediation-STIRPAT model referring to the hierarchical step-based multiple regression analysis (Huo et al. 2021a). First, we verify the influence of FD on the EC, that is the significance of coefficient c in Eqs. (4). If c is significant, we next check the significance of coefficient a in Eq. (5) and coefficient b in Eq. (6). If both a and b are significant, the PE, as a mediating variable, affects the relation of the FD and EC. Finally, we look at the effects of FD and PE on the EC, and here, we mainly focus on the significance of coefficient c′ in Eq. (7). If c′ is significant, PE affects the association between FD and EC with the estimated share of mediating effect in the total effect as ρ = ab′/c × 100%. If the coefficient c′ is not significant, PE plays a full intermediary role between FD and EC.

Panel threshold-STIRPAT model
To investigate the possible nonlinear effects of FD on the EC from the perspective of PE, we combine the extended STIRPAT model with the panel threshold model (Hansen 1999) to construct a panel threshold-STIRPAT model considering PE as the threshold variable, which is shown as: where I(⋅) is an indicative function that identifies the value 1 if the parentheses requirement is met and 0 otherwise. γ is the threshold value. X it are a series of control variables. Equation (8) reflects the single threshold case, and we can extend it to investigate multi-threshold effects.

Explained and explanatory variable
Energy consumption (ec) is the explained variable. Based on China's provincial energy consumption inventory from the CEADs database, we estimate the total EC in 30 provinces. Financial development (fd) is the explanatory variable. China does not have a unified standard for measuring provincial FD level. Previous studies often construct a comprehensive index involving various financial related indicators (e.g., credit scale, securities market development, and FDI) to evaluate FD. However, considering the insufficient impacts of capital market and FDI on the poverty reduction, we do not focus on the indicators related with stock market and FDI. Moreover, most of China's financial assets are absorbed by banks. Therefore, we select FD scale, which is reflected by the ratio of total deposits and loans to the total GDP, for assessing FD level. A relatively larger value for the FD scale indicator means larger impacts of financial system on the economic development, thereby reflecting a higher FD level.

Intermediary variable and threshold variable
Poverty alleviation efficiency (pe) is not only an intermediary variable but also a threshold variable here. Data envelopment analysis (DEA) is often used for efficiency evaluation by calculating the deviation level of decision-making units (DMUs) relative to the optimal DMUs, involving various inputs or outputs. However, the efficiency score is limited in the range from 0 to 1 using the traditional DEA model (e.g., CCR model). Thus, the super-efficiency DEA (SE-DEA) model, which permits the efficiency value to exceed 1 (Andersen and Petersen 1993;Matthias and Maik 2005), is employed here with the MaxDEA 8 Ultra software. This means that we can rank efficient DMUs, while the efficiency values of inefficient DMUs are in line with those of CCR. Mathematically, we estimate the provincial PE as: where γ * is the optimal efficiency value, θ is a scalar that represents the share of the jth DMU's input vector, which is used to generate the jth DMU's output vector with the help of reference technology. X j is an m-dimensional input vector, and Y j is an s-dimensional output vector for the jth DMU. λ k is the relative weight of input factors. If γ * = 1, the DMU is efficient. Table 1 shows specific input and output indexes for estimating the PE.

Control variables
To eliminate possible interference of other factors, we refer to previous literature and set the following variables as control variables.
(1) Economic development level (gdp). Previous studies indicated that economic development enables more resources flow into the technological innovation filed of energy industry, expands the market demand of energy industry, and intensifies EC (Sun and Chen 2022). Meanwhile, some studies believed that economic growth would make people pay more attention to environmental protection, which will help reduce EC (Zhao et al. 2021). We here employ GDP per capita to reflect economic development level.
(2) Population size (pop). Past studies have confirmed that the expansion of population drives the growth of EC by scale effects (Wang et al. 2016). Moreover, population aggregation can stimulate the development of public service projects that are conducive to improving energy efficiency, which effectively reduces EC (Huang et al. 2021a). Therefore, we employ the total population to measure population size. (3) Industrial structure (is). The secondary industry still contributes a large part of China's EC (Xiong et al. 2019). Generally, the increase of the share of the secondary industry in the total GDP contributes to the growth of EC Huang et al. 2021b). Therefore, we use the share of industry in the total GDP to describe industrial structure. (4) Energy intensity (ei). Energy intensity, that is the energy consumption per unit of output, reflects the energy efficiency. Generally, a higher energy intensity indicates a lower energy efficiency, thereby inducing the growth of EC (He et al. 2017;Cao et al. 2021).
Here, we use the ratio of energy consumption to the total GDP to assess energy intensity. (9) (5) Energy poverty (ep). Previous studies have probed the nexus between energy poverty and EC (Lin and Wang 2020;Dong et al. 2021). We here employ a narrow concept to reflect energy poverty, and thus, it is only measured by the gap in per capita demand of natural gas and electricity in China's provinces referring to Bu et al. (2022). Mathematically, energy poverty is estimated as: where NE i is the per capita demand of natural gas and electricity in province i. ε is the harmonic coefficient, ε = 0.01. A larger EI i means that the province i suffers larger difficulty to employ natural gas and electricity; thus, the province uses more fossil fuels.

Data sources
We use a panel data of China's 30 provinces during 2010-2019. Taiwan, Macao, Tibet, and Hong Kong are not involved in our study due to lacking data. The EC data are from the China Emission Accounts and Datasets (CEADs). (10)

Preliminary analysis
We first perform a cross-sectional dependence (CD) test for the method selection, as ignoring the CD test can lead to inconsistent outcomes. We use Pesaran's CD test to perform cross-sectional correlation (see Table 4). The results indicate that except for the variable of FD, all variables are significant at the 1% level, which means that they have cross-sectional dependence. We then perform a stationary panel unit root test to evaluate the stability of variables. Given the existence  Table 5 show that all variables are significant at 1% after first-order difference. This implies that they are all stable only after the first-order difference. Finally, we conduct panel cointegration analysis to test the presence of long-run relationship among variables. The ADF panel cointegration test proposed by Kao (1999) allows for individual fixed effects and evaluate the homogeneous cointegration linkage by the pooled regression, which is available when the timescale of the panel series is small (Gutierrez 2003); thus, we use this method. The panel cointegration analysis in Table 6 shows that the existence of four statistics is significant, which implies that there is a strong long-run linkage between FD and EC.

Benchmark results
Because our model is cointegrated according to the panel cointegration analysis, which might lead to the problem of endogeneity and efficiency loss, we use the AMG method to evaluate the long-term cointegration nexus of FD and EC. Following the principles in Pesaran's (2006), the AMG method considers the cross-sectional dependence and country-specific heterogeneity. Moreover, the estimator can apprehend unobservable common elements by integrating temporal dummy variables into the model. According to the result of model (1) (see table 7), FD encourages the increase of EC at the significance level of 10%. Danish and Ulucak (2021) and Komal and Abbas (2015) found analogous conclusions, showing that a sound financial market can provide perfect financial supporting services for enterprises, thereby improving their vitality and increasing EC. Moreover, FD can provide individuals more diversified ways to manage their own finance, and thus increasing income, which expands the demand for EC. In terms of control variables, economic development has a strong role in promoting the growth of EC, because economic development needs the resources inputs in the productive activities, which will promote EC (Meng et al. 2022;Gu et al. 2019). The expansion of population leads to more EC, as population growth increases energy demand. The reinforcement of industrial process insignificantly increases EC, because industrial EC takes up the vast majority of EC layout (Li and Lin 2015). The growth of energy intensity enhances EC, which is the same as the outcome of Liu et al. (2019). The growth of energy poverty reduces EC, but its effect is not obvious. This is because during the process of overcoming energy poverty difficulty, people tend to increase the total quantity of EC due to government subsidies for natural gas and electricity. However, because the local developed energy infrastructure is still insufficient, people have limited access to modern energy, which insignificantly increases EC. Due to possible endogenous problems in our study (Dieudonné et al. 2021, Wu et al. 2022, we reassessed the effect of FD on the EC using the instrumental variable

Mediating effect of poverty alleviation efficiency
To test the mechanism of FD on the EC, we build a mediation-STIRPAT model to test the mediating impact of PE. The test results indicate that the p-values are all smaller than 0.1 in the models (1) to (4) (see Table 8), which means that the models (1) to (4) are suitable for fixed effect models. In the light of the results of model (1), FD positively intertwined with EC at the 5% significance level, which is consistent with the obtained results of AMG estimator. The result of model (2) indicates that the improvement of financial system helps to improve PE.
Previous studies have proved that the improvement of financial system through the banking propagation can help alleviate regional income inequality (Beck et al. 2010) and reduce rural poverty (Wang et al. 2022a). Moreover, Burgess and Pande (2005) have highlighted that eradicating poverty requires the structural change in the economy (e.g., financial sector development) (Burgess and Pande 2005). Specifically, FD through banking expansion could contribute to the increase of PE by creating job opportunities and increasing the investments in the public infrastructure. Also, FD can facility poor households obtain credits, which can increase the assets of the poor. According to the result of model (3), the improvement of PE significantly stimulates the increase of EC. This is line with the tradeoff hypothesis of anti-poverty efforts and environmental quality improvement, which highlighted that the achievement of poverty alleviation is at the expense of serious environmental problems (Walelign et al. 2021;Yang et al. 2021;Li and Yuan 2023). For the model (4), FD significantly drives the growth of EC at the significance level of 10%, while the positive impact of PE on the EC is not obvious. The intermediary effect of PE in the total effect of FD on the EC is 15.75%. This shows that the PE has an active partial intermediary effect on the linkage between FD and EC. Thus, the ameliorate of financial system can indirectly increase EC by improving PE. Moreover, to further verify the robustness of mediating effects, we reassessed the results of models (1) to (4) using the 2SLS method (Ye et al. 2021) (see Table S1). These estimates are basically identical with those of Table 8. Therefore, the results of mediating effect model are credible.

Threshold effect of poverty alleviation efficiency
We apply the panel threshold-STIRPAT model to analyze the potential nonlinear impact of FD on EC under different PE levels. We conduct the threshold effect test to check the number of thresholds and corresponding thresholds. We check the significance of threshold effects by the bootstrap method (Hansen 1999) (see Table 9 and Fig. 1). The significance test allows the single threshold at the 1% significance level, while the double threshold failed, indicating that there is only one threshold effect in the model. The threshold outcome suggests that the FD affects EC differently as the change of PE, which can be decomposed into two sections (Table 10). Specifically, the expansion of financial system has an obvious active impact on the EC increase for the provinces with lower PE. A percentage increase in the FD raises the growth in the EC by 0.012%. Meanwhile, the effect of FD on the EC is significantly enhanced in the provinces with higher PE. A percentage growth in the FD causes the rise in the EC by 0.071%. Generally, with the gradual achievement of poverty alleviation, FD plays an increasing role in promoting EC is strengthened. Zameer et al. (2020a) have confirmed that lower poverty results in higher FD, as the improvement in the education levels through anti-poverty projects enhances the participation of financial markets from residents. As the improvement of financial market is positively intertwined with the economic revitalization in the poor areas, perfecting financial system in the poor areas is conducive to increasing EC. Due to the disparity in the economic development and resource endowments among various provinces, the influence of FD on the EC is characteristic of regional heterogeneity. Therefore, we further analyze the areal distribution of

Impacts of financial development on the energy consumption
We conclude that FD increases EC. This is accordance with prior conclusion that FD drives the increase of EC via economic development (Komal and Abbas 2015;Ha et al 2022). From the consumption perspective, Liu et al, (2021), Coscia andRusso (2018), and Ji and Wang (2020) highlighted that FD broadens the channels for consumers to access financial services and effectively breaks capital restrictions for enterprises, thereby simulating large-scale production activities and increasing EC. However, some studies provide opposite results, as they consider the depth and structure of FD to measure FD, while we measure FD based on the FD scale. For example, considering the depth of FD, Zhang et al., (2016) and He et al., (2019) showed that FD can guide resources allocation away from inefficient projects, which contributes to the expansion of renewable energy industry and thus reduces EC. From the point of view of FD efficiency, Anton and Nucu (2020) concluded that FD can effectively decline the production cost of companies by reducing investment deficits, which can promote energy-intensive industries develop clean energy technologies and update facilities, thereby reducing EC. These differences in results indicate that the selection of FD indicator has heterogeneous influences on the EC, and it is crucial to analyze the influence of FD on the EC considering various characteristics of financial systems. However, noted that as previously we have clarified that our study focused on the energy impact of FD considering the role of PE; thus, we only select the FD scale to measure FD level. This is decided by research aim, but this is a main limitation for our study, and thus, future research should further explore the impact of FD on the EC employing a more comprehensive indictor. However, our results still provide critical implications. We found that FD scale dominated by the banking expansion increases EC, which means that the exploitation of banking sector is not enough to effectively improve the ability to use energy (Yu et al 2022, Zameer et al. 2020b, Xu and Tan 2020. Thus, instead of promoting the expansion of banking sector, the government should actively improve the resource allocation ability of financial services by creating a good market environment.

Mediating and threshold impacts of poverty alleviation efficiency
According to the results of the mediation model, PE plays an important mediating role between FD and EC. First, FD is conducive to improving PE, as a stable financial system can provide credit supports for the government in the poor areas to invest in infrastructure and public projects, improving the living environment in the poor areas and reducing poverty level (Sehrawat andGiri 2016, Abdul andMaurizio 2017). Additionally, FD can further deepen financial technology innovation (e.g., decentralized finance, open banking, and artificial intelligence), which can help increase the opportunity for low-income population to financial products and services, such that the financial demands of poor areas can be satisfied, and the PE can be improved (Wang and He 2020;Ye et al 2022). We also found that the enhancement of PE increases EC. This is because that the consolidation of the foundation for poverty alleviation means the growth of average household income level and the perfection of local infrastructure, which contributes the furtherance in the EC . Meanwhile, with the national emphasis on the poverty alleviation, the government continues to popularize targeted poverty alleviation projects (e.g., photovoltaic poverty alleviation), which promotes production activities by enabling characteristic industries in the poverty-stricken zones, and thereby increasing EC (He et al 2022, Zameer et al. 2020a). Generally, PE plays a key intermediary role between FD and EC. FD improves PE, which mediately causes the rise in the EC. We also suggested that the effect of FD on the EC increases with the enhancement of PE. This is because with the improvement of PE, the poor can access diversified financial services through convenient financial channels, which expands their disposable income, and then upgrading the "quality" and "quantity" of their EC (Danish and Ulucak 2021, Wang and Tan 2020, Wang et al 2021. However, China still belongs to developing countries, and the poor here prefer to increase the quantity of consumption rather than its quality. This means that while being lifted out of poverty, households are not interested in investing in reconstruction or energy efficiency initiatives with high costs and risk uncertainty (Schueftan et al. 2021). Thus, it is difficult to realize consumption structure upgrading towards clean products, which inhibits the reduction of EC. Therefore, at this stage, it is critical issue that how to reach the targets of anti-poverty, FD, and sustainable development simultaneously by updating consumption structure and increasing low-carbon awareness of households.

Policy implications
In the light of above analysis results, we propose some policy suggestions. First, the effect of financial markets in the EC is critical in China, which can provide references to emerging economies those heavily depend on fossil fuels. We found that the expansion of financial market measured by banking sector scale leads to the substantial increase in EC demand. Thus, China should take advantage of banking development to minimize EC by promoting investments in clean energy projects. Specifically, local governments should attach importance to the efficiency of financial resources allocation by designing an evaluation system with comprehensive indicates and ensuring sustainable development as pivotal components of official performance assessment. Moreover, considering China's FD characteristics, our finding indicates that current financial growth pattern dominated by banking system development may be not sustainable. Thus, China should actively optimize financial infrastructure and improve the openness of financial system by encouraging financial reform and upgrading financial structure, such that the decoupling mechanism of finance and technology can push the reduction of EC. Moreover, poverty alleviation and energy reduction are major targets for the sustainable development of China. Our finding shows that FD drives the amelioration of PE, which further promotes the increase of EC. This indicates that there is a trade-off between efforts to relieve poverty and reduce EC, and it is difficult to realize win-win regarding the two important sustainable targets, especially with the rapid development of finance. Thus, while the poor are lifted out of poverty, the integrated development of "green finance" and "green consumption" should be highlighted. This can be realized through environmentally friendly awareness creation and persistent education investments, which promotes consumption shifts towards energy-efficiency products. Moreover, governments should voluntarily improve the access to effective "green credits" with a preferential interest rate for low-income households and enhance the utilization of renewable energy technologies for poverty reduction, thereby contributing to lifting the quality of EC during the evolution of anti-poverty.
Finally, differentiate policies for FD should be implemented based on the PE in different areas. For regions with high PE, the government should think highly of the profound combination of financial supports for poverty mitigation with energy reduction due to a larger active influence of FD on the EC. Specifically, the quality of FD should be improved and the function of green finance in the energy consumption mitigation should be emphasized, which help drive the fitness of FD level with the PE and EC quantity. For regions with low PE, the government should carry out education and publicity work for local financial institutions combining with the local conditions of PE, FD, and EC, such that they recognize the significance of renewable energy development for the green development of the whole society.

Limitations
There are still some limitations that need to be further discussed. First, based on the research purpose, we only focused on the impact of FD scale on the EC. However, in addition to the FD scale measured by the banking scale, the roles of stock market, FDI, and insurance market are also necessary to the sustainable development of finance resulting in different viewpoints of finance-energy nexus. Therefore, future research can establish a more all-round FD indicator to study the influence of FD on the EC. Also, based on the availability of data, our empirical research data only involves the provincial data of China. If we can collect data at the municipal and county levels, or even at the micro business level, the research conclusions will be further enriched. Finally, China's EC has the feature of spatial agglomeration and spatial correlation. Thus, we can further involve spatial factors into future analysis.

Conclusions
Applying the panel data of 30 provinces in China during 2010-2019, we integrated FD, PE, and EC into an analytical framework by building the extended STIRPAT model. We started our study through examining the immediate influence of FD on the EC. Next, we explored the mediating impact of PE on the relationship of FD and EC. Finally, we use PE as the threshold variable to test the nonlinear impact of FD on the EC. We suggest that FD directly encourages the growth in EC. Also, FD can indirectly affect EC by the channel of PE. Specifically, FD is conducive to improving PE, while the enhancement of PE tends to boost EC. The mediating effect of PE is responsible for 15.75% of the total effect of FD on the EC. Additionally, FD and EC present a nonlinear tie considering varied PE. When the PE exceeds 0.524, the role of FD in promoting EC is strengthened.