Testing the Role of Financial Development and Urbanization in the Conventional EKC: Evidence from China and India

Greenhouse gases (GHGs) have an extremely critical role in maintaining the earth’s temperature which is important for human life to existThese gases mainly consist of carbon dioxide (), methane, ozone, nitrous oxide, and water vapours. On the heels of industrial revolution, many activities have involved high level of fossil fuel consumption such as urbanization. Expansion of industrialization amplified GHGs and led to global warming and climate change (IPCC, 2007).

Deforestation and the burning of fossil fuels like oil, gas and, coal for generating electricity which are needed for transportation, household and industrial uses have increased GHGs emissions [2]. The increase in GHGs is considered to be the greatest environmental issue faced by the world today [3]. This human induced change in the climate has caused the global annual average temperature of land and ocean to rise by 1.5°F or 0.8°C during the period of 1880 to 2012 [4].

Over the past few years, policymakers, environmentalists, and economists have shifted their attention towards the global environmental issues particularly towards the problems generated due to economic growth or economic development The Environmental Kuznets Curve (EKC) which is based on the work of has very important part in explaining the association among environment and economic growth which suggests that environmental growth has got a reversed U-shaped relation. The EKC exists in certain cases and under certain circumstances.

While studying environment growth relationship, it is significant to study the role of financial development because it plays an extremely important role in economic growth. Countries having efficient financial management system use financial resources effectively, with scarce financial resources. This system would be favourable for the progress of economic development [7].

The financial development-Carbon emission nexus discussed in two strands: One school of thought explains that the carbon emissions increase through financial development. Developed financial markets reduce the financing costs to finance new projects and buy new equipment’s which increased the level of energy hence increase carbon emissions [8]. Zhang et al. [9] argues that financial intermediation increases the buying power of consumer and they can purchase air conditioner, cars, washing machines etc. consuming high energy that leads to affect environment quality negatively. The other school of thought is evident to increase environment quality through financial development. Frankel and Romer [10] argue that financial development attracts foreign direct investment which increase the research and development activities, leads toward better quality of environment. Institutional quality plays an important role in improving environment quality and carbon dioxide emission decreases in the countries where institutions are strong [8, 11].

Before 1978, The People’s Bank of China was the single bank which was operating as central and commercial bank in China but subsequently in 1984; four big state held commercial banks were established who acted as the lender of money to the state owned businesses in construction, rural banking business, foreign trade and national economy, commercial sector and urban industrial activities. As the financial sector began to expand in 1980’s, new banks, financial institutions, and even non-banking financial institutions were established. Moreover, the foreign banks also entered in China in 1990s. In 1990 and 1991 Shanghai Stock Exchange along with Shenzhen Stock Exchange were established. After China’s membership of world health organization during 2001, financial liberalization took place in real terms. Restrictions on interest rates and loans were reduced. China Banking Regularity Commission was established as a supervising body in 2003. By 2006, due to relaxing restrictions on foreign banks, 242 offices were established by 223 foreign banks from 42 countries [12]. In India, financial reforms have taken since 1990’s which have led to the expansion of the financial sector.

Besides these, many other variables are there which have immense effect on environment. Urbanization is a vital aspect which effects carbon emissions. According to the world urbanization prospects (2014) the world has rapidly urbanized. The findings of the report show that 66% population of the world will be living in cities up to 2050. Prediction is that by 2050 larger than 80% countries of the world will be half urban. So the process of urbanization needs to be done in a planned and controlled manner. The problems that can arise due to unplanned and over urbanisation are urban poverty, lack of provision of basic services like access to electricity or sanitation, air pollution, political unrest and violence [13, 14].

The quality of environment decreases by increasing the level of urbanization in different economies with poor environmental concern and awareness. Urbanization still plays an important role and increase across the globe, as people shift from rural areas to urban areas in search of better social environment and social security. Urbanization Promotes productivity, create more wealth and creativity to redesign different structures which would be useful for economic gain. However, the increase in urbanization leads towards consuming high energy leads towards high carbon emissions [15]. In addition, it can be noted that china’s quick urbanization leads to increase high energy demand. The large scale construction of buildings and infrastructures in urban areas need to use large amount of cement, iron and steel, which is hard to import completely, and has to be produced locally. Due to this, it posture great pressure on environmental quality [9]. 

On the bases of population growth rate Zhang et al. [16] explained that from 1978 to 1988 the rate at which urban population grew was much higher than the rate at which the overall population grew. The total population grew at 1.29% per annum while the urban population grew at the rate of 3.79 %. In addition, from 1978- 1999 the overall increase in the rate of total population was 1.44% while the urban population growth reached to 4.97%. From 1978 to 1999 almost 174 million people migrated to urban areas [17].

On the contrary, the Indian urban population has increased from 18% to 31.2% from 1960 to 2011 According to 2007’s United Nations State of World Population Report about 40.8% of Indian population will be living in urban areas by 2030.

The outcomes of one research cannot be employed over all countries or for all cases. Therefore, each study is unique and different from the other on the basis of the indicators selected, methodology, and data selection and it gives different results even for the same country [19]. This uniqueness of this research method makes it even broader and more interesting. This study examines the impact of urbanization and economic development on the EKC by looking the cases of China and India. The organization of the paper is as follows. Literature review is presented in Section 2. Data and methodology for the study has been discussed in Section 3. The estimation outcomes are presented in Section 4. Section 5 is the conclusion of the study.

3.1 Model

For the purpose to estimate the impact of economic activities on the environment, Ehrlich and Holdren [50] introduced the IPAT (I=P*A*T) model. This model estimated the impact of population, prosperity and technology on the environment but had certain limitations. Firstly, it was simple mathematical identity so it cannot directly perform hypotheses testing on it for estimation of impact of all factors on the environment. Secondly, it assumed that unitary elasticity was present between I,P,A&T which was in contradiction with the EKC hypothesis [51]. A study by Dietz and Rosa [52] introduced the improved version of IPAT model known as the STIRPAT model. This model is written as

${{I}_{t}}=aP_{t}^{\beta }A_{t}^{\gamma }T_{t}^{\lambda }{{e}_{t}}$

where, I,P,A&T stands for environmental impact, population, prosperity and technology, ‘a’ is constant and ‘e’ is the error term while β, γ, and ƛ are parameters of P,A&T. The subscript ‘t’ symbolizes the unit of analysis. In this model in addition to P and A, any factor that impacts per unit of economic activity can be expressed in terms of T [53].

The benefit of STIRPAT model is that it allows extending the original model by adding social, economic, and technological factors that affect the environment and helps in solving the real environmental problems [54]. For example, many studies have included urbanization as an additional variable in the model like Zhang et al [44]. As per York et al [53], T can be used to express any factor that effects per unit economic activity other than A and P. For example, He et al [47] used the sum number of patents as a proxy for T while Poumanyvong and Kaneko [43] used share of industry and services sector in the GDP as a proxy for T. Furthermore, Shahbaz et al. [40] investigated the role of free trade on the environment as a proxy for T. Likewise, Li and Lin [49] added industrialization in the model to check its impact on the environment.

Following Dietz and Rosa [51] we develop an extended version of the STIRPAT model by adding a component of financial development and urbanization to the original model. Our model is expressed as:

${{I}_{t}}=\alpha P_{t}^{\beta }A_{t}^{\gamma }UP_{t}^{\lambda }FD_{t}^{\delta }EU_{t}^{\eta }{{\mu }_{t}}$     (1)

After converting the variables in the log form our model can be written as:

$\begin{align}  & \ln C{{O}_{2}}_{t}=\alpha +\beta \ln {{P}_{t}}+\gamma \ln {{A}_{t}}+\lambda \ln U{{P}_{t}} \\ & +\delta \ln F{{D}_{t}}+\eta \ln E{{U}_{t}}+{{\mu }_{t}} \\\end{align}$     (2)

Here we have used $\ln \mathrm{CO}_{2 t}$ i.e. CO₂ emissions (metric tons per capita) as a substitute for ‘I_t’ which represents the impact on environment. $\ln P_{r}$ is total population which is described as the total number of legal residents of a country. Moreover, $\ln Y_{t}$ i.e. GDP per capita (constant 2010 US$) is used to represent Affluence (A). We used financial development, urbanization and energy use in order to capture the impact of technology on environment. Therefore, here we used domestic credit to private sector (% of GDP) as a proxy for Financial development ($\ln F D_{t}$). Whereas $\ln U P_{t}$ is urban population (% of total population) and $\ln E U_{t}$ is the total energy consumed which is used to measure energy use. Now model can be written as:

$\begin{align}  & \ln C{{O}_{2}}_{t}=\alpha +\beta \ln {{P}_{t}}+\gamma \ln {{Y}_{t}}+\lambda \ln U{{P}_{t}} \\ & +\delta \ln F{{D}_{t}}+\eta \ln E{{U}_{t}}+{{\mu }_{t}} \\\end{align}$     (3)

In order to test the presence of EKC, another term is introduced in the model that is the square of GDP per capita (constant 2010 US\$) represented as $\ln Y_{t}^{2}$. By adding this term, we analyse the impact of economic growth on environment in the long run.

$\begin{align}  & \ln C{{O}_{2}}_{t}=\alpha +\beta \ln {{P}_{t}}+\gamma \ln {{Y}_{t}}+\psi \ln {{Y}^{2}}_{t} \\ & +\lambda \ln U{{P}_{t}}+\delta \ln F{{D}_{t}}+\eta \ln E{{U}_{t}}+{{\mu }_{t}} \\\end{align}$     (4)

The theory of EKC is that economic growth leads to environmental degradation when an economy has lower income level or economic growth (shown by Y_t), but once a particular amount of income level or growth is achieved any addition in the growth rates will lead to an improvement in the environmental quality (shown by $Y_{t}^{2}$), thus forming an inverted U shaped relation between economic growth and environmental quality.

3.2 Data and variable

In this paper, we test the presence of EKC for India and China as both are great emitters in the world and both tend to have achieved high GDP rates over the past few years [55]. We also test the impact of urbanization and financial development on carbon emissions in India and China. A time series analysis is done for both countries from 1980 to 2014. The entire data is collected from World Development Indicators (WDI) database.

Impact on Environment (I): This study uses the level of CO₂ emissions measured in metric tons per capita produced through burning of solid, liquid and gas fuels to represent the impact of various factors on the environment. Enhancement of carbon discharges leads to degradation of environment while decreases in CO₂ emissions are considered as an improvement in environmental quality. This variable is represented as CO₂.

Population (P): The variable of population is measured by total population which is described as the total number of legal residents of a country. This variable is represented as P in the model.

Affluence (A): For measuring affluence we have used GDP per capita (constant 2010 US\$) which is calculated as the GDP divided by mid-year population. Moreover, as we also aim to test the presence of EKC so we also introduce a square term of GDP per capita (constant 2010 US\$). The variable Y and Y² are utilized in the model to test EKC hypothesis.

Urbanization (UP): For measuring urbanization data was collected of urban population (% of total population) which is calculated as total population living in urban areas. UP in the model represents urbanization.

Financial development (FD): It has taken household loans to private sector (% of GDP) which includes all the financial services given to the private sector by financial cooperation like loans, trade credits etc.as a proxy of financial development. The variable DCPS in the model represents financial development.

Energy use (EU): Energy use (kg of oil equivalent per capita) is the total energy consumed is used to measure energy use and is represent by EU in the model.

3.3 Estimation procedure

3.3.1 Unit root test

Many macroeconomics time series variables tend to be non-stationary, especially while the sample size is minute. Testing unit root is considered to be a necessary condition because if the non-stationary variables are used in estimations it will give meaningless results and would be considered as spurious [56]. Moreover, in the existence of unit root the OLS parameters estimated are also inconsistent and less efficient until and unless co-integration exists between the variables [57]. One way of making the variables stationary is to take the first difference but this approach is not commonly used because it prevents from analysing the long term association involving variables [19, 58]. Therefore, in this study we used Augmented Dickey fuller test (ADF) and Philips-Perron (PP) test to test the presence of unit root.

3.3.2 Autoregressive distributed lag (ARDL) model

Pesaran M.H. and Pesaran, B. [59] introduced the ARDL Bound Testing Approach. This technique is quite useful as it can be applied if the pattern of integration of the variables is I (0), I (1) or I (0) / I (1). It can also catch the data generation process in common to definite modelling framework because it includes sufficient number of lags. ARDL can also be used to get Error Correction Model (ECM) through linear transformation so it becomes possible to assess short term & long term parameters at the same time [60]. This model is best suited for small samples as the results are more reliable in comparison to erstwhile methods like Engle and Granger. Moreover, due to assortment of the suitable lags the problem due to endogeniety and serial correlation is also eliminated in this model and even if the explanatory variables are endogenous estimations can still be carried out Pesaran et al. [61].

As the variables included in this study are either integrated at I (0) or I (1), the best approach for estimating the co-integration between the variables in this study is ARDL. So now we will write model given in Eq. (4) as ARDL framework:

$\begin{align}  & \Delta \ln C{{O}_{2t}}={{\alpha }_{0}}+\sum\limits_{i=1}^{p}{{{\alpha }_{i}}\Delta \ln C{{O}_{2t-i}}} \\ & +\sum\limits_{i=1}^{p}{{{\beta }_{i}}\Delta \ln {{P}_{t-i}}}+\sum\limits_{i=1}^{p}{{{\gamma }_{i}}\Delta \ln {{Y}_{t-i}}}+\sum\limits_{i=1}^{p}{{{\psi }_{i}}\Delta \ln Y_{t-i}^{2}} \\ & +\sum\limits_{i=1}^{p}{{{\lambda }_{i}}}\Delta \ln U{{P}_{t-i}}+\sum\limits_{i=1}^{p}{{{\delta }_{i}}\Delta \ln DCP{{S}_{t-i}}} \\ & +\sum\limits_{i=1}^{p}{{{\eta }_{i}}\Delta \ln E{{U}_{t-i}}}+{{\theta }_{1}}{{\left( \ln C{{O}_{2}} \right)}_{t-1}}+{{\theta }_{2}}{{\left( \ln P \right)}_{t-1}} \\ & +{{\theta }_{3}}{{\left( \ln {{Y}^{{}}} \right)}_{t-1}}+{{\theta }_{4}}{{\left( \ln Y_{t}^{2} \right)}_{t-1}}+{{\theta }_{5}}{{\left( \ln UP \right)}_{t-1}} \\ & +{{\theta }_{6}}{{\left( \ln DCPS \right)}_{t-1}}+{{\theta }_{6}}{{\left( \ln EU \right)}_{t-1}}+{{\mu }_{t}} \\\end{align}$      (5)

Here μ_t is white noise, α₀ is the drift component, the Σ’s represents the error correction dynamics and the θ’s represents the long run relationship between variables.

3.3.3 Co-integration: ARDL bound test

Autoregressive distributed lag (ARDL) bound test is a set of two stage method. Firstly, it determines the most advantageous numbers of lags required. Here, number of lags for every variable ARDL estimated as (p+1) ^k quantity of regressions. While p denotes the utmost number of lags and K represents the amount of variables in an equation. The selection of lag order is based on either AIC or SBC, AIC decides ultimate relevant lag length whereas SBC gives slightest possible lag length. Then, ARDL restricted testing method assesses the ARDL equation by ‘ordinary least square (OLS)’ method for the purpose to examine long term relationship between the variables and it is calculated through ‘F test’. The Null hypothesis states that there is no association among the variables in long run. whereas the alternate hypothesis states the existence of a long term association among the variables.

$\mathrm{H}_{0}: \theta_{1}=\theta_{2}=\theta_{3}=\theta_{4}=\theta_{5}=\theta_{6}=0$

$\mathrm{H}_{1}: \theta_{1} \neq 0, \theta_{2} \neq 0, \theta_{3} \neq 0, \theta_{4} \neq 0, \theta_{4} \neq 0, \theta_{5} \neq 0, \theta_{6} \neq 0$      (6)

The two sets of upper & lower bound key results given by Pesaran et al. [61] are then compared to the calculated F statistics in order to admit or refuse the null hypothesis. The ‘Upper critical bound value (UCB)’ assumes that all variables are I (1) while ‘Lower critical bound value (LCB)’ supposes that all variables are I (0). If the calculated F statistic is greater than the UPB then the Null hypothesis will be rejected meaning that there is a long run relationship between variables, whereas if the calculated F statistic is lower than the LCB then the null hypothesis will be accepted which means that there is no long run relation between the variables. However, if already estimated F value falls among LCB & UCB then the outcomes obtained are uncertain. Once selection of the best lag order is made we move towards the second part of the ARDL bound test which is to estimate the long term relationship among the variables.

3.3.4 Error correction model (ECM)

After the short run relationship is confirmed from the bound test amid the variables the subsequent measure is to calculate the ECM. In this model the term ϕ ECM_t-1 helps in measuring the speed or time taken to shift reverse to the long term equilibrium after the short term shocks. The sign of ECM term is observed which should be negative and significant [28]. The ECM equation can be written as:

$\begin{align}  & \Delta \ln C{{O}_{2t}}={{\alpha }_{0}}+\sum\limits_{i=1}^{p}{{{\alpha }_{i}}\Delta \ln C{{O}_{2t-i}}} \\ & +\sum\limits_{i=1}^{p}{{{\beta }_{i}}\Delta \ln {{P}_{t-i}}}+\sum\limits_{i=1}^{p}{{{\gamma }_{i}}\Delta \ln {{Y}_{t-i}}}+\sum\limits_{i=1}^{p}{{{\psi }_{i}}\Delta \ln Y_{t-i}^{2}} \\ & +\sum\limits_{i=1}^{p}{{{\lambda }_{i}}}\Delta \ln U{{P}_{t-i}}+\sum\limits_{i=1}^{p}{{{\delta }_{i}}\Delta \ln DCP{{S}_{t-i}}} \\ & +\sum\limits_{i=1}^{p}{{{\eta }_{i}}\Delta \ln E{{U}_{t-i}}}+{{\theta }_{1}}{{\left( \ln C{{O}_{2}} \right)}_{t-1}}+\phi EC{{M}_{t-1}}+{{\mu }_{t}} \\\end{align}$      (7)

3.3.5 Stability and diagnostic tests

Pesaran and Pesaran [59] suggested using cumulative (CUSUM) test for conducting the diagnostic test. These tests are used for checking the model in support of serial correlation, heteroscedasticity and non-normality and to ensure goodness of fit. CUSUM statistics are recursively updated and designed alongside break points. If plot falls between the essential bound i.e. 5% significance level in that case ‘null hypotheses of coefficients of regression is stable being not to be rejected.