Financial Inclusion and CO2 Emissions in Asia: Implications for Environmental Sustainability

We examine the relationship between nancial inclusion and carbon emissions. For this purpose, we develop a composite indicator of nancial inclusion based on a broad set of attributes through principal component analysis (PCA) for 26 countries in the Asia region. Our robust panel regression analysis reveals a signicant positive long-term impact of nancial inclusion on carbon emissions. The pairwise causality test reveals unidirectional long-term causality running from nancial inclusion to carbon emissions. The study suggests that policy makers may design policies that integrate accessible nancial systems into climate change adaptation strategies in order to neutralize the side effect of nancial inclusion deteriorating environmental quality and inclusive sustainable economic growth. the impact of nancial development on CO2 emissions based on dynamic ordinary squares ordinary the run nancial Shahbaz et al., 2013). Hung et al. (2018) examine the impact of the development of nancial systems on CO2 emissions in 25 OECD countries from 1971 to 2007. The empirical results show nonlinearity between nancial development and CO2 emissions. Renzhi and Baek (2020) examine the impact of nancial inclusion on CO2 emissions in an environmental Kuznets curve (EKC) framework based on 103 countries, and support the inverted U-shaped relationship. Zaidi et al. (2019) analyse the link between nancial development and CO2 emissions based on advanced statistical tests such as updated bias-corrected and continuously updated fully modi ﬁ ed estimators in 17 APEC countries from 1990 to 2016 using an EKC framework. The ndings show development of nancial systems to be negatively associated with CO2 emission in the long and short run. casual relationship development nancial system the impact of development of nancial system on CO2 This investigate the non-linear relationship of nancial inclusion with carbon emission in EKC framework


Introduction
The cataclysmic consequences of climate change for human life, health and environmental sustainability have received worldwide attention. Primarily, increased global warming is attributed to anthropogenic greenhouse gas emissions due to massive consumption of fossil fuels and industry discharge (Wawrzyniak and Doryń, 2020). It is widely thought that carbon dioxide is the main cause of environmental degradation among the greenhouse gasses (hereafter GHGs) since it contributes 70% of GHG emissions (Sarkodie et al., 2020). It is widely claimed that the concentration of GHGs in the atmosphere could double from the pre-industrial level by 2035 (Charfeddine and Kahia, 2019). Consequently, worldwide temperature rise may surpass the threshold of 2°C, and if so, the world may experience disastrous consequences of climate change affecting all aspects of life in the form of a rapid rise in sea levels, mass extinction, super droughts, water contamination and health problems (Mujtaba and Shahzad, 2020;Wang et al., 2020;Lu, 2018).
The threat of environmental degradation has directed the attention of scholars to the causal interaction between macroeconomic variables and environmental pollutants. For instance, previous literature ascertains the association between carbon emissions and economic growth , energy consumption (Charfeddine and Kahia, 2019), foreign direct investment (Sarkodie et al., 2020), foreign nance (Alshubiri and Elheddad, 2019), trade openness (Shahbaz et al., 2017), urbanization (Raghutla and Chittedi, 2020), population growth (Yeh and Liao, 2017) and nancial development (Jiang and Ma, 2019;Kayani et al., 2020;Raghutla and Chittedi, 2020). Nevertheless, empirical studies regarding the association between carbon emissions and the newly emerging concept of nancial inclusion are scant. Financial inclusion is considered to be an integral dimension of nancial development since it contributes to the development of the nancial sector and equips nancial intermediaries with sophisticated tools to achieve inclusive sustainable economic growth Dahiya and Kumar, 2020).
Theoretically, opposing views exist in the literature regarding the impact of nancial inclusion on environmental sustainability (Le et al., 2020). On the one hand, nancial inclusion enables the individual as well businesses to avail themselves of useful credit schemes at lower cost which makes the investment in green technology more affordable (Gök, 2020). This way, an inclusive nancial system positively contributes to environmental sustainability by encouraging individuals and businesses to use green technology and adopt better environmental practices, which in turn lead to a reduction in GHG emissions (Jiang and Ma, 2019). On the other hand, an accessible nancial system may damage environmental quality, thereby boosting manufacturing and industrial activity through affordable nancing, which in turn increases CO2 emissions (Charfeddine and Kahia, 2019). Likewise, nancial inclusion also motivates individual consumers to use energy intensive electric appliances (Gök, 2020). The use of these energy intensive goods poses a serious threat to environmental sustainability. So far, two empirical studies, Le et al. (2020) and Renzhi and Baek (2020), establish the relationship between nancial inclusion and CO2 emissions. However, the ndings of these studies are contradictory and inconclusive.
We contribute to the literature in two ways. Firstly, we investigate the impact of nancial inclusion on climate change in Asian countries. We consider Asia as the sample region for the following reasons. Firstly, recent statistics show Asia to have the most dynamic and fastest growing economies . However, despite achieving signi cant economic progress, a lack of access to nancial services and climate change remain two critical challenges to the region's achievement of inclusive economic development Le et al., 2020). Home to two-thirds of the world's poor, almost a billion people have neither a formal bank account nor access to any nancial services (Bhardwaj et al., 2018). Bhardwaj et al. (2018) report that within the developing economies across Asia, about 27% of adults hold a formal bank account and about 33% have utilized either a loan or credit facility. Much effort has been put into including the masses in the nancial sector; however, it is still very challenging to achieve nancial inclusion at a maximum level mainly due to contextual differences across countries, which vary along political, economic, cultural, ethnic and religious lines. Secondly, Asia has been the biggest contributor to global warming over the last decade, contributing one-third of total GHG emissions (see the level of CO2 emissions in Figure 1). Therefore, Asia is very vulnerable to the devastating impacts of climate change (Le et al., 2020).
The second way in which our study contributes to the literature, is that it differs from recent notable studies, such as Le et al. (2020) and Renzhi and Baek (2020), which ignore the issues of long-short causality, pairwise causality, endogeneity, heteroscedasticity and simultaneity problems, in three ways. Firstly, we apply the method of Dumitrescu and Hurlin (hereafter D-H) (2012) to test the pairwise causality and determine the direction of the causal relationship. Secondly, we apply the panel autoregressive distributed lag (ARDL) model to examine the long and short run impact of nancial inclusion on carbon emissions. Thirdly, we test this relationship in a dynamic panel framework through the method of Arellano and Bond (1991) using the difference generalized method of moments (GMM) and system GMM, in order to produce consistent and e cient parameters and thereby overcome the endogeneity, heteroscedasticity and simultaneity problems (Muhammad, 2019), as a robustness check. The results are further validated through seemingly unrelated regression (hereafter SUR) for a second robustness check.
Financial inclusion is an inextricable part of nancial development (Le et al., 2020). Therefore, the theoretical and empirical foundations of nancial development are covered in order to establish the link between nancial inclusion and carbon emissions.
Theoretically, there exist divergent views among researchers regarding the relationship between nancial development and climate change (Jiang and Ma, 2019). On one hand, scholars suggest that nancial development mitigates GHG emissions by facilitating the energy supply sector to upgrade production technology and equipment through the mitigation of nancial constraints by lower borrowing cost (Renzhi and Baek, 2020). Hence, it is evident that the development of the nancial sector, which shows the real availability of capital and funding channels through banks and stock markets, can positively contribute to environmental sustainability via the reduction of GHG emissions (Gök, 2020). From this perspective, nancial development reduces environmental deterioration (Koshta et al., 2020). Several empirical studies acknowledge the positive role of nancial development in combating climate change (Charfeddine and Kahia, 2019;Omri et al., 2015;Shahbaz et al., 2013;Tamazian and Rao, 2010). Saidi and Mbarek (2017) investigate the impact of nancial development on CO2 emissions in a dynamic panel framework setting using time series data on 19 emerging economies over the period 1990-2013.
The empirical results show a negative in uence of the development of nancial systems on CO2 emissions, which shows that nancial development positively contributes to environmental sustainability. Using data on the top 23 renewable energy-using countries from 1985 to 2011, Dogan and Seker (2016) analyse the impact of nancial development on CO2 emissions based on dynamic ordinary least squares (DOLS) and fully modi ed ordinary least square (FMOLS). The study reveals that nancial development and CO2 emissions are cointegrated in the long run and nancial development reduces GHG emissions.
However, other scholars suggest that nancial development exacerbates environmental deterioration via an upsurge in CO2 emissions for the following reasons. Firstly, a well-functioning nancial system lowers the cost of borrowing, which encourages businesses to obtain the capital required to expand production, which in turn increases CO2 emissions (Raghutla and Chittedi, 2020). Secondly, nancial development dramatically promotes social consumption thereby providing better credit utilization, which could facilitate individual consumers purchasing more energy intensive goods such as electrical appliances (Gök, 2020), automobiles and many others, which in turn increases CO2 emissions (Koshta et al., 2020). Thirdly, capital markets are considered important indicators of economic development. The persistent performance of the stock market attracts individual and institutional investors and stimulates the activities of production and consumption, which in turn increase CO2 emissions through massive consumption of fossil fuels (Rajpurohit and Sharma, 2020). The third school of thought supports the nonlinearity of the relationship between nancial development and CO2 emissions. Several studies nd an inverted Ushaped relationship between nancial development and GHG emissions (Omri et al., 2015;Salahuddin et al., 2018;Shahbaz et al., 2013). Hung et al. (2018) examine We can infer from the above literature review that both theoretically and empirically the impact of nancial system development on CO2 emissions is still under debate and the results remain inconclusive. More speci cally, the literature provides only two recent studies of the relationship between the recently emerged concept of nancial inclusion and climate change. These studies offer useful insights and advance the literary work, but the ndings regarding the nancial inclusion-carbon emission nexus are contradictory and inconclusive. Primarily two gaps exist in the literature regarding this topic. Firstly, most previous studies focus on developed and emerging economies, while the region most vulnerable to climate change (Asia) has had the attention of only one empirical study so far. Secondly, the application of different statistical techniques, samples and data create challenges for comparing the research completed by the various scholars. Keeping the above limitations in view, our work relates to the small number of studies that examine the nancial inclusion-carbon emission nexus through advanced statistical techniques such panel ARDL, difference GMM, system GMM and SUR to provide insight for policy makers on this topic.

Operationalization -Financial Inclusion and Carbon Emission
We categorize the variables into an independent variable ( nancial inclusion), a dependent variable (carbon emission) and control variables (trade openness, energy consumption, industry) for empirical analysis. These are described in Table 2. We collect the data for proxies of nancial inclusion from the Global Findex database. The data for the remaining macro variables are extracted from the World Development Indicators (WDI) for the 26 sample countries: Brunei Darussalam, Iraq, Israel, Japan, Kazakhstan, Republic of Korea, Kuwait, Lebanon, Malaysia, Qatar, Saudi Arabia, Singapore, United Arab Emirates, Bangladesh, Bhutan, Cambodia, India, Indonesia, Jordan, Kyrgyz Republic, Mongolia, Pakistan, Philippines, Sri Lanka, Tajikistan and Vietnam. Table 2 presents the summary statistics of the variables for the 26 selected Asian countries. The means and standard deviations of carbon emission (M = 8.546, SD = 11.108) and energy consumption (M = 3122.33, SD = 4095.72) are comparatively greater than the reported values of Nathaniel and Adeleye (2021) for the African region. The results are in line with idea that Asia as a region is very vulnerable to climate change. The summary of the nancial inclusion proxies reveals that mean values of branches of commercial banks per 100,000 adults and ATMs per 100,000 adults are close to the mean values of these proxies given in Le et al. (2020), while outstanding deposits (% GDP) and loans (%GDP) are consistent with the study of . Among the control variables, the average and standard deviation of energy consumption are greater than those of trade openness (M = 4.489, SD = 0.546) and industry (M = 3.547, SD = 0.420). All the variables are normalized by taking the natural log of each variable for further analysis.
Our variable of interest is the nancial inclusion index. We construct a composite indicator of nancial inclusion based on four widely used proxies: branches of commercial banks per 100,000 adults, number of ATMs per 100,000 adults, the amount of bank deposits and the amount of bank credit (Chatterjee, 2020;Le et al., 2020). These proxies cover two keys aspects of an inclusive nancial system; the two former indicators demonstrate the availability of the banking sector and the two latter indicators demonstrate the use of banking system (Van et al., 2019).
We use the principal component analysis (PCA) technique to estimate the nancial inclusion index. The PCA is the most widely used technique for creating indexes in the literature for several reasons. Firstly, it is a widely known standard technique that extracts hidden features and relationships and removes excess information in order to reduce the dimensionality of data and create a composite indicator (Radovanović, Filipović and Golušin, 2018). Unlike other linear transformation techniques with xed sets of basis vectors, the PCA basis vectors depend on the dataset. We apply two tests, Bartlett's test and the Kaiser-Meyer-Olkin (KMO) test, prior to the PCA estimation in order to determine whether the stated proxies are appropriate for the construction of nancial inclusion. Bartlett's test of sphericity tests whether the correlation matrix is an identity matrix . Factor analysis is considered to be more suitable in the case of a signi cant Bartlett's test (P < 0.05). The KMO test reveals the proportion of common variance that might be caused by underlying factors (Renzhi and Baek, 2020). The KMO index ranges from 0 to 1, with scores larger than 0.5 generally indicating that the factor analysis is suitable. The results of these two tests are reported in Tables 3 and 4. The results of both tests support the use of PCA in this study.
We use PCA to construct the nancial inclusion index. The PCA process is executed in two steps. First, we estimate the various components in order to identify which components account for the variations in the original variable and have the lowest pairwise correlation. In the second step, following Gujarati and Porter (2009), we estimate the index based on the components which account for a portion of variance with eigenvalues greater > 1. The cumulative variations of each component are presented in Table 4.

Econometric Approach
We use the theoretical foundation of stochastic impacts by regression on population, a uence and technology (STIRPAT) (Dietz and Rosa, 1997), to examine the impact of nancial inclusion on CO2 emissions in Asia. The proposed STIRPAT model is: where I it stands for environmental effects and the factors included are P it (population), A it (a uence) and T it (technology) for country i at time t. We extend the above STIRPAT model to incorporate nancial inclusion. In light of the existing literature, we also include three relevant control variables, trade openness (Shahbaz et al., 2017), energy consumption (Charfeddine and Kahia, 2019) and industrialization (Zhou et al., 2013). The baseline model is: where Co2 i,t stands for the log of CO2 emissions and FI i,t represents the nancial inclusion index. The control variables are trade openness, energy consumption and industry.
We execute the data analysis in six steps. First, we check the cross-sectional dependency through Pesaran (2004) and Pesaran (2015) tests and then perform the D-H test to examine the pairwise causality of the variables. In the second step, we test the stationarity of the data through widely used unit roots tests (CIPS, IPS, Fisher-ADF, Fisher-PP, Hadri, LLC) (Hasanov et al., 2017;Jiang and Ma, 2019). In the third step, the cointegration among the analysed variables is tested through the methods of Pedroni (2004), Kao (1999) and Westerlund (2005). In the fourth step, we apply the panel ARDL to establish the long-and shortrun exogenous and endogenous relationships. The pooled mean group (hereafter PMG) estimator and mean group (hereafter MG) estimator are two widely used estimators of ARDL. We apply the panel ARDL for the following reasons: (i) the long-and short-run cointegrating vectors are considered heterogeneous across panels in the MG approach, while the PMG assumes the estimating parameters of the cointegrating vector are homogenous across panels which increases the degree of freedom (df); and (ii) the PMG produces e cient and consistent estimations even in the presence of a small number of observation across sections (Da et al., 2018;Mensah et al., 2019;Žiković et al., 2020). Based on the Hausman test results, the PMG is more suitable for data analysis. Hence, the following equation is estimated to empirically test the relationship between nancial inclusion and CO2 emissions: where y it is the dependent variable, x it is a vector of the explanatory variables nancial inclusion index, trade openness, energy consumption and industry, all integrated of order 1, Φ i is the error correction coe cient, βi is the long-run parameter, Ψ ij and δ ij are county-speci c coe cients of the short-term dynamics, pi and qi are lag lengths of the autoregressive distributed lag model, α i represents county-speci c intercepts and ε it is an IID error term.
We test the causal impact of nancial inclusion on CO2 emissions in a dynamic panel framework setting using the SUR method. Following Bashir and Khan (2019), we apply the methods of Arellano and Bond (hereafter AB) (1991) and Blundell and Bond (hereafter BB) (1998) for robustness checks. Dynamic panel models are considered more effective at handling problems of unobserved heterogeneity and simultaneous and dynamic endogeneities in the panel data, and are found to provide consistent and unbiased coe cients. Furthermore, following Bashir (2019), we apply SUR in order to validate the ndings.

Cross-Sectional Dependence and the Dumitrescu and Hurlin Causality Test
We used the Pesaran (2004) and Pesaran (2015) tests for cross-sectional dependence (CD). These tests are suitable for larger cross-sections and smaller time series data such as those of the current research (N = 26 > T = 12). Table 5 presents the results of both tests. Under the null hypothesis, there is cross sectional independence. The results of the CD tests reject the null hypothesis of cross-sectional independence and accept the alternate hypothesis of cross-sectional dependence, since all the stated variables are statistically signi cant at 1.00%. These results are in line with the ndings of Le et al. (2020). Hence, the results support the presence of cross-sectional dependence among the variable across all the Asian countries.
We examine the pairwise causality between nancial inclusion and carbon emissions in a panel data framework setting through the D-H test, and the results are presented in Table 6

Panel Unit Root Tests and Cointegration
We test the stationary of the data through commonly used panel unit root tests (hereafter PURT) CIPS, IPS, Fisher-ADF, Fisher-PP, Hadri, LLC (Hasanov et al, 2017;Jiang and Ma, 2019) to determine whether the variables are integrated at the level I(0) or rst difference I(1) and avoid the problem of spurious regression (Granger and Newbold, 2014). Table 7 presents the results of the PURTs. It can be seen that all the variables are integrated at I(1) with few exceptions, implying that all the variables are stationary at order one.
The method of Pedroni (2004) is applied to test the cointegration among the variables. This test is suitable in cases of large N and small T (N>T or 26). Table  8 present the results. Of the seven tests of cointegration, four support the existence of long-run cointegration among the variables. The methods of Kao (1999) and Westerlund (2005) are also used as a robustness check. Both tests support the existence of long-run relationships among the variables. The ndings of the robustness tests provide stronger proof of cointegration amongst the analysed variables. Hence, we draw a conclusion based on the results that there is a long-run relationship among the variables in the Asia region.

Regression Results
We test the long-and short-run relationships between nancial inclusion and carbon emissions in a panel ARDL framework setting. We decide between the PMG and MG estimators using Hausman's test of slope homogeneity (Mensah et al., 2019;Žiković et al., 2020). The results presented in Table 9 reveal that that the PMG estimator is more suitable for the analysis. We apply the PMG estimator using an ARDL(1,1,1,1,1) based on the Schwarz information criteria (SIC).
The PMG estimator reveals that nancial inclusion signi cantly and positively in uences carbon emissions in the long run. The short-term coe cient is also statistically signi cant, which implies that nancial inclusion positively affects CO2 emissions. The positive signs of the long-and short-run coe cients show that the activities related to nancial inclusion and environmental deterioration are closely associated with each other. Our ndings support the theoretical idea that an accessible nancial system offers a well-developed mechanism of risk mitigation, e cient resource pooling and strong corporate governance, as well as boosting industrial activities for better economic prospects which lead to further environmental degradation. However, our results contrast with the theoretical proposition that the development of nancial systems positively contributes to environmental sustainability via a reduction in GHG emissions, facilitating individuals and businesses making investments in upgraded green technology by mitigating nancial constraints and lowering borrowing costs.
Our ndings are similar to Jiang and Ma (2019) and Le et al. (2020), whose empirical results also support the claim that accessible nancial systems exacerbate environmental quality.

Robustness Checks
Our previous ndings based on the PMG estimator con rm the positive in uence of nancial inclusion on carbon emissions. However, we apply two further robustness checks to verify the reliability of the empirical results. We examine the nancial inclusion-carbon emissions relationship in a panel data framework. Hsiao (1986) argues that panel regression estimation has several bene ts. Firstly, the panel estimation allows us to account for unobserved heterogeneity.
Secondly, large numbers of observations provide more degrees of freedom. Thirdly, it allows us to control for unobservable county-speci c characteristics which may be correlated with our exogenous variables (Renzhi and Baek, 2020). Hence, following Bashir and Khan (2019) and Bashir (2019), we test the relationship using dynamic panel regression estimations and SUR in order to obtain consistent and e cient regression coe cients. Table 10 presents the estimation results of the dynamic panel models. The ndings of the post-estimation tests support the appropriateness of the GMM specification. The coe cients of nancial inclusion in the AB and BB models are positive and statistically signi cant at the 1% level, which implies that nancial inclusion may further exacerbate environmental degradation. The results are consistent with our ndings. Similarly, Table 11 shows the results of the SUR model, which further validate our previous ndings.
The overall results support the claim that a developed nancial system accumulates capital through pooling and mobilization of savings, reduces the problem of asymmetric information regarding investment opportunities and ensures the optimal allocation of available nancial resources (Kayani et al., 2020), which in turn lubricates the wheels of the economy and accelerates economic growth , consequently increasing CO2 emissions (Raghutla and Chittedi, 2020). This is consistent with the theoretical stance that nancial development leads to economic prosperity which further deteriorates environmental sustainability. Our ndings share common ground with several empirical studies such Al-Mulali, Ozturk and Lean (2015) and Kayani et al. (2020). These studies support the idea that nancial development leads and environmental degradation follows.

Conclusion
Massive unmoderated development has created several challenges including environmental deterioration and climate change. Global warming is a serious threat to the existence of human beings and has thus been considered a primary concern over the last couple of decades. The literature provides theoretical and empirical support regarding the major macroeconomic determinants of climate change such as economic growth, energy consumption, population and nancial development. Recently, nancial inclusion has attracted worldwide attention in the ght to combat climate change. Therefore, we examine the impact of nancial inclusion on carbon emissions in 26 Asian countries through series of advanced statistical techniques in order to provide insight for policy makers. Our ndings reveal a positive relationship between nancial inclusion and CO2 emissions. The results are consistent with idea that nancial inclusion, which facilitates individuals and businesses obtaining easy access to nancial services, motivates businesses to expand the scale of production and increases the power of individual consumers to purchase energy intensive electric appliances which accelerate the use of energy from fossil fuels, resulting in higher CO2 emissions. Our ndings are qualitatively robust across econometric models such as PMG, ARDL, AB, BB and SUR.
Our ndings have far reaching implications for Asian countries, providing insights for policy makers regarding the linear positive relationship between nancial inclusion and carbon emissions. Policy makers need to interpret the results of this research with great caution. They may incorporate accessible nancial systems into climate change adaptation strategies in such a way as to considerably reduce carbon emissions. For instance, under the widely acknowledged green credit guidelines of China, the nancial system provides adequate support through green credit nancial services to reduce carbon emissions and contribute to the green economy and sustainable development. Secondly, regulatory authorities may align initiatives for nancial inclusion with environmental protection policies. Governments may uplift marginalized segments of society by expanding inclusive nance in a way that addresses environmental deterioration. Moreover, small and medium enterprises (SMEs) could be encouraged to adopt carbon mitigation practices through easy access to green nancial services. Investments based on public-private partnerships that work towards environmental sustainability are part of the sustainable development goals (SDGs). Thirdly, businesses tend to expand their production through nancial loans instead of investment in green technology projects, since developing countries are under extreme pressure to develop economically, which leads to further climate change. This shows the unavoidable contradiction of economic development and environmental sustainability. However, we suggest that sustainable inclusive growth can be achieved with the help of policies that lead to synergies between growing nancial inclusion and mitigating CO2 emissions.
Our study covers the Asia region and subsamples to establish the relationship between nancial inclusion and carbon emissions. However, future studies could consider individual countries, which have various political, economic, cultural, ethnic and religious attributes. Secondly, future research could consider a larger number of countries across developed, emerging and developing markets in order to provide insights for policy makers, foreign investors, fund managers and other stakeholders. Thirdly, future studies could explore the role of levels of governance on the relationship between nancial inclusion and various proxies of climate change. Fourthly, another interesting area for research would be the construction of nancial inclusion based on a broader set of fundamental attributes such as penetration, availability and usage of nancial systems in each country or region in order to broaden the understanding of policy makers related to the non-linear relationship between nancial inclusion and climate change.

Declarations
Ethics approval and consent to participate Not applicable Consent for publication Not applicable

Availability of data and materials
The date used in this study is available on the o cial website of the OECD and Bloomberg.

Competing interests
The authors declare that they have no competing interests Funding Not applicable Authors' contributions SJ and TA conceptualized the idea, did the analysis and wrote the initial draft. SJHS reviewed and proof-read, and also supervised the work.