An industry-specific approach to the relationship between carbon emissions and economic growth of China and India

An industry-specific approach to the relationship

between carbon emissions and economic growth of

China and India

This thesis examines the relationship between CO2 emissions and economic growth by employing annual time series data for the period 1980-2013. Following declining CO2 emissions in a few high income countries, China and India are tested based on the theory of the Environmental Kuznetsk Curve (EKC), which states that in response to more awareness, at a certain point of development, emissions start to decline. More specifically, a new line of

research is followed which states that EKCs are possible on an industry level. Using the ARDL bounds testing approach for co-integration, the separate CO2 emissions for the manufacturing, electricity, transport and residential sectors are analyzed, as well as their Granger causality. Results for the different industries and countries are mixed, as only some

industry-specific results are conclusive for declining CO2 emissions.

Key words: CO2 emissions, EKC, developing countries

Name: Giulia Belardo Student Number: 3135888

Supervisor: Prof. dr. mr. C.J. Jepma Second Supervisor: Prof. dr. B. Los

University of Groningen, Faculty of Economics and Business

Table of Contents

1. Introduction ... 1

2. Literature review ... 1

3. Methodology and Data ... 7

4. Empirical Results and Analysis ... 9

4.1. Descriptive statistics ... 9

4.2. Analysis of unit root ... 12

1. Introduction

development and a replacement of the dirty and obsolete technologies by more clean and efficient ones. The main motivation for testing the relationship between environmental pollution and economic growth is that it allows policy makers to judge the response of the environment to economic growth, and to have knowledge about this response is crucial since one of the objectives of an economy is inter alia to maximize economic growth (Narayan & Narayan, 2010).

2. Literature review

long run bi-directional relationship between the variables. Particularly, they find a positive significant long-run bi-directional relationship between CO2 emissions and economic growth, and between road sector energy consumption and economic growth. Sharif Hossain (2011) used Granger causality tests on newly industrialized countries including also trade openness and urbanization. His conclusions tended towards a unidirectional short run causal relationship from economic growth and trade openness to CO2 emissions; from economic growth to energy consumption; from trade openness to economic growth; from urbanization to economic growth and from trade openness to urbanization. A different study analyzed the EKC hypothesis for Brazil, China, India and Indonesia for the period 1970-2012. The results show relationships of income and energy use with CO2 emissions for all four countries. Indonesia and Brazil showed support for the EKC in both long run and short run, whist China supported the EKC only in the long run, and India showed no support (Alam, Murad, Noman, & Ozturk, 2016).

and co-integrated, especially after controlling for cross-sectional dependence. Moreover, the author uses the system of generalized methods of moments to conduct a panel causality test in a vector error-correction mechanism setting. Unidirectional causality running from real GDP per capita to CO2 emissions per capita are found in both the short run and the long-run. The analysis performed on singular countries provides evidence of an EKC for Greece, Malta, Oman, Portugal and the United Kingdom. However, for the entire panel it can be observed that a 1% increase in GDP generates an increase of 0.68% in CO2 emission in the short run and 0.22% in the long run.

households consume less pollution-intensive goods and services, as the estimated curves are concave. Lastly, they also find that these environmental Engel curves shift down over time, as the households represented in 2002 are responsible for less pollution then their 1984 counterparts with similar real incomes. Significant is that this downward shift does not result from improvements in technology or abatement, as the authors have data of the pollution intensity of production for both years, but this shift echoes a change in consumption due to some combination of changing prices, rules, or other economy-wide effects. The largest portion of the clean up in the USA has come from changes in production technology, but isolating the consumption-related compositional changes in pollution suggests that household-level consumption changes have more than offset the increased pollution from growing household incomes, and this is due in roughly equal parts to a direct effect of income growth via consumer preferences and indirect economy-wide changes such as prices and environmental regulations (Levinson, 2015).

where respectively the quadratic term of the affluence variable is not significant, and none of the affluent terms are significant. The results indicate that not all industrial sectors supported the EKC relationship between GDP per capita and CO2 emissions, as only the electricity sector shows declining emissions as the economy keeps on growing. The authors conclude that the heterogeneity in the EKC relationship across industry sectors implies that there is an urgent need to design more specific policies related to carbon emissions reduction for various industry sectors.

3. Methodology and Data

The analysis is going to firstly focus on the EKC on a total carbon emissions level to then be split into separate industries for the time span 1980-2013. Data on CO2 emissions from fuel combustion per capita in kilograms, the shares of the contribution by the different sectors1 _in total CO2 emissions, and GDP per capita in constant 2010 US$ are collected from the World Development Indicators website (The World Bank Group, 2017). Unfortunately, data on CO2 emissions from the WDI indicators is only available until the year 2013, and due to differences in data from other sources it has therefore been chosen as the only data set. GDP per capita is shown with the 2010 US$ as the base year so to not be influenced by price inflation and measure the true growth of the series. CO_2t, GDP_t, GDP_t2 _{represent the per capita carbon emissions and} the per capita GDP and square of per capita GDP, all after logarithmic transformation. As numerous empirical studies found the presence of a non-linear relationship between CO2 emissions and GDP, the regression form follows:

𝐶𝑂_2𝑡 = 𝛽₀+ 𝛽₁𝐺𝐷𝑃_𝑡+ 𝛽₂𝐺𝐷𝑃_𝑡2+ 𝜀_𝑡.

For the EKC theory to be valid, it is required for β1 to be positive and significant, and for β2 to be negative and significant. This is in order to achieve the downward-sloping U-shaped function, representing the starting phase of economic growth having negative consequences on the environment, but the harm to the environment to diminish after a certain growth point has been reached.

Non-stationary time-series variables should not be used in regression models in order to avoid the problem of spurious regression. The only exception to this rule is if yt and xt are non-stationary I(1) variables, in which case their expected difference, or any linear combination of them is expected to be I(1) as well. In the special case where the linear combination between them is a stationary I(0) process, then y_t and x_t are said to be co-integrated. Co-integration implies that y_t and x_t share similar stochastic trends, and since their difference is stationary, they never diverge too far from each other. A test to check for co-integration is to test the stationarity of the least squares residuals by using a Dickey-Fuller test. If the residuals are non-stationary, then yt and xt are not co-integrated, and any apparent regression relationship between them is said to be spurious. The co-integration relationship is also going to be analysed

with the ARDL-bounds testing approach. Firstly developed by Pesaran and Shin (1999) and with the follow-up extension by Pesaran et al. (2001), this approach has several advantages, in that the ARDL does not require all variables to be integrated of the same order and can be applied with the underlying regressors being integrated of order one, I(1), as well as of order zero, I(0). Moreover, this version of the co-integration test is not linked to a small size sample. The aforementioned ARDL model can be estimated as:

The linear model:

𝐶𝑂₂ = 𝛽₀+ ∑ 𝛾_𝑖 𝑝 𝑖=1 𝐶𝑂_2𝑡−𝑖+ ∑ 𝛿_𝑗 𝑞1 𝑗=1 𝐺𝐷𝑃_𝑡−𝑗+ 𝜀_𝑡 The quadratic model:

𝐶𝑂₂ = 𝛽₀+ ∑ 𝛾_𝑖 𝑝 𝑖=1 𝐶𝑂_2𝑡−𝑖+ ∑ 𝛿_𝑗 𝑞1 𝑗=1 𝐺𝐷𝑃_𝑡−𝑗+ ∑ 𝜑_𝑙 𝑞2 𝑙=1 𝐺𝐷𝑃_𝑡−𝑙2 + 𝜀_𝑡

The Schwarz Criterion (SC) and the Akaike Information Criterion (AIC) were used in the process for the appropriate selection of the lag length of the ARDL model for the respective variables.

The bounds testing approach is based on the F-statistic in a generalized Dickey-Fuller type regression used to test the significance of the lags of the variables under consideration in an unrestricted error correction regression. Under the null hypothesis the test shows that there exists no relationship between the level of the included variables, irrespective of whether the regressors are I(0), I(1) or mutually co-integrated. Pesaran, Shin and Smith (2001) provide one set of critical values which assume that the regressors are I(1), and another set that assumes they are I(0). If the resulting F-statistic falls outside of these critical value bounds, a conclusive inference can be drawn without needing to know whether the regressors are I(1), I(0) or co-integrated between themselves. The rejection of the null hypothesis implies that there is a long-run relationship. If the computed F-statistic is above the values of the upper critical bound, then the H0 is rejected. If the statistic falls between the bounds then the co-integration test is inconclusive, while if the test statistic is below the lower bounds value, the H0 of no co-integration cannot be rejected (Odhiambo, 2009).

hypothesis is that the series are not co-integrated and that the residuals are non-stationary. When series are co-integrated it means that when one variable experiences a change, the other also experiences an adjustment.

In the case that the unit root tests show that the variables are integrated of order 1 and not co-integrated, the first differences have to be used to estimate a vector autoregressive model. Vector autoregressive models (VAR) are used when it it is better to recognize that in many relationships, variables like x and y are simultaneously determined (Carter Hill, Griffiths, & Lim, 2011). This is in order to explore the causal relationship between pairs of time-series variables, and take into account their dynamic properties and interactions. When working with a VAR model, all variables have been transformed to I(0) by taking their first difference. After having estimated a VAR model, a set of Granger causality tests are performed for each equation in the VAR. A variable x is said to Granger-cause a variable y if, given the past values of y, past values of x are useful for predicting y. The common method to test Granger causality is to perform a regression of y on its own lagged values and on lagged values of x and test the null hypothesis that the estimated coefficients on the lagged values of x are jointly zero. Failure to reject the null hypothesis is equivalent to failing to reject the hypothesis that x does not Granger-cause y. For each equation and each endogenous variable that is not the dependent variable in that equation, a series of Wald tests are computed that reports that the coefficients on all the lags of an endogenous variable are jointly zero. For each equation in the VAR, the hypothesis is tested that each of the other endogenous variables does not Granger-cause the dependent variable in that equation.

4. Empirical Results and Analysis 4.1. Descriptive statistics

China Variable Obs. Mean Std. Dev. Min. Max. GDP per capita (2010 US$) 34 1917.402 1578.193 347.887 5721.694

Ln (GDP) 34 7.221 0. 855 5.852 8.652 CO2 p.c. (All sectors) in kgs 34 3066.9 1578.2 1415.6 6663.3 CO2 p.c. (Manufacturing) 34 1043.9 478.6 606.4 2127.7 CO2 p.c. (Electricity) 34 1398.5 964.8 367.2 3541.9 CO2 p.c. (Transport) 34 215.9 140.6 83.3 558.8 CO2 p.c. (Residential) 34 301.6 50.2 211.1 387.8 CO2 p.c. (Other sectors) 34 106.9 18.4 64.1 138.5 Table 1

India Variable Obs. Mean Std. Dev. Min. Max. GDP per capita (2010 US$) 34 771.4 340.2 393.9 1551.6

Ln (GDP) 34 6.6 0.4 5.9 7.3 CO2 p.c.(All sectors) in kgs 34 859.6 305.7 436.0 1489.2 CO2 p.c. (Manufacturing) 34 213.1 70.2 135.3 399.0 CO2 p.c. (Electricity) 34 430.0 195.5 135.3 795.9 CO2 p.c. (Transport) 34 102.5 32.6 66.6 182.3 CO2 p.c. (Residential) 34 73.5 9.6 55.0 85.1 CO2 p.c. (Other sectors) 34 40.4 11.2 19.3 73.2 Table 2

that year 2000, 𝛽₂ is negative and 𝛽₁is positive for all different industries as well as total CO2 emissions. According to the available data for this study China is on a positive trend towards diminishing CO2 emissions in the future. Predictions of declining CO2 emissions and a continued, even though slowed, growth of GDP per capita tend to agree. In fact, as reported by Greenpeace in February 2017 and by various other sources, China’s CO2 emissions have shown no growth or even a decline for the fourth year in a row (Greenpeace, 2017). The same report states reductions of about 1% in coal consumption over the past year, as well as record numbers in installations of solar panels. For the purpose of this study and a better comparison to India, the entire set of 34 observations from 1980 to 2013 is taken into account.

For India, the total CO2 emissions due to all sectors combined show a downward sloping U-shaped curve when modelling CO2 emissions per capita on the vertical and GDP per capita on the horizontal axis, taking into account the time line. This points towards the finding that India is following a good trend of declining CO2 emissions, although the peak of the curve is not depicted as it probably took place after 2013 or still has to take place. Regarding the separation in different sectors, the sectors that show a down-ward sloping U-shaped curve, with negative values for the 𝛽₂ coefficient are the electricity and heat production, the

residential and the other sectors. Also in the case of India, experts expect CO2 emissions to be on the right path towards a decline. Analogously to China, also in India the use of fossil fuels such as coal has slowed down, and policies towards renewable energy are predicted to prove successful in following a down-ward sloping trend of CO2 emissions over raising GDP per capita. Even though the manufacturing and transport industries tend to show upward-sloping U-shaped curves in this estimation based on the 34 observable years, the same argument as made for the case of China, can be made for India as well, as both these industries are on a downward sloping path if taking into account the years after the turn of the millennium. Perhaps the most interesting results on India is that the electricity, residential and other sectors seem to have reached the peak of the curves already as can be seen in the figures in Appendix 1. The turning point for these three sectors seem to have happened at an income of around 1100-1500US$ per capita, with the electricity sector being at the larger end. This is definitely encouraging information for India, as it actively tries to diminish CO2 emissions and fulfil the Paris agreement established in 2015. Whilst the CO2 emissions from the manufacturing sector do not seem to have reached their maximum yet, declining CO2

4.2. Analysis of unit root

The serial correlation between successive observations, particularly when the sampling interval is very small are indication that typically macroeconomic time series are non stationary. If the presence of unit-root non-stationarity is not accounted for, the study may suffer from an inconsistent and less efficient ordinary least squares parameter estimates unless the variables are co-integrated. The determination of the uni-variate properties of the time series is also the starting point of the analysis of co-integration. The theory of co-integration needs their linear combinations to be stationary and requires the variables to be integrated of the same order. The plots of the different CO2 emissions as well as their first differences are shown in the figures in the Appendix 2. Figure A1 shows that in China CO2 emissions overall have been steadily increasing, but with a much slower growth in the later years. The figures for CO2 emissions caused by electricity and heating, and by transport show a similar pattern to overall CO2 emissions. The CO2 emissions from manufacturing industries show a dip around 2001, the year that China joined the WTO, but with a much faster growth after that as seen in the figure showing the difference. Whilst in the 80’s the manufacturing and construction industries were responsible for almost half of the total of China’s CO2 emissions, and the electricity and heat production sector around 25%, it changed heavily in the last decade as it is the electricity sector that is responsible for more than 50% of all emissions and the manufacturing sector for around 30%. The residential and leftover sector emissions both show a drop at the end of the past millennium, with very fast growth rates after 2000. The difference between years seem to be more or less stable around zero, for the latter two industries, even though they saw large changes during the aforementioned time range.

In India total CO2 emissions and CO2 emissions from manufacturing were steady until around 2005, when they saw a very fast growth, whilst again electricity and heat production emissions have been increasing steadily over the time span. In India, carbon dioxide emissions from transport have seen a much faster increase since 2005, which were probably due to the beginnings of heavy FDI flows into the country. Residential CO2 emissions as well as the remaining sectors have showed ups and downs over the years as shown in figures A4 and A5 in the Appendix 2.

a unit root in a time series, which is an augmented Dickey-Fuller test, except that the time series is transformed via a generalized least squares regression before performing the test. The authors and other studies have shown that this test has significantly greater power than the previous versions of the augmented Dickey-Fuller test. The null hypothesis of the test is that the dependent variable is a random walk, possibly with drift, whilst the alternative hypothesis states that the dependent variable is stationary about a linear time trend. The results for the DF-GLS, where the number of lags was set to 9 by the Schwert (1989) criterion, are used in the standard Augmented Dickey-Fuller test shown in the tables.

Table 3 shows that at level the variables for China show no significance (apart of the Ln GDP). At first difference all variables show significance. This shows that the variables are stationary at level I(1) and that their first difference helped to stabilize the mean of the time series, by eliminating deviations in the level, and so eliminating any possible trend.

For India the results from the unit root tests show that at level for the CO2 emissions from all sectors combined, from electricity, from residential buildings and other sectors, it is possible to reject the null hypothesis of a unit root at different levels of significance. At first difference all variables have shown ability to reject the null hypothesis of a random walk, and the regression used to obtain the test statistic included a constant term and zero lags for all variables. Therefore, the relevant variables for India are also integrated of order 1 and the series has made adapt for the following co-integration tests.

China

Variables ADF test (at level) ADF test (at first difference) CO2 all sectors -1.892 (1) trend -3.177** (0) CO2 manufacturing -1.758 (1) trend -23.275** (0)

CO2 electricity -2.703 (1) trend -4.981*** (0) CO2 transport -1.367 (4) trend -5.955*** (0) CO2 residential -1.932 (2) trend -6.005*** (0) CO2 other sectors -1.951 (1) trend -6.196 *** (0) GDP -3.662** (1) trend -3.360** (0) GDP2 _{-1.639 (1) trend -3.066*** (0)} Table 3: (1) = at first lag. All have been regressed with a trend term. *, **, *** indicate significance at

India

Variables ADF test (at level) ADF test (at first difference) CO2 all sectors -3.938** (4) trend -5.037*** (0) CO2 manufacturing -2.093 (4) trend -4.445*** (0) CO2 electricity -3.233** (2) -5.747*** (0) CO2 transport -1.702 (3) trend -4.530*** (0) CO2 residential -4.219*** (7) -6.309*** (0) CO2 other sectors 3.304* (0) trend -6.064*** (0) GDP -0.947 (1) trend -4.250*** (0) GDP2 _{-0.643 (1) trend -3.805*** (0)} Table 4: *, **, *** indicate significance at 10%, 5% and 1% respectively.

4.3. Analysis of Co-integration

China

Equation t-stat Outcome Equation t-stat Outcome CO2 (all sectors), GDP -1.99 _integration No co- CO_{GDP, GDP}2 (all sectors), 2 -3.29

Co-integration at 10% level CO2 manufacturing, GDP -1.81 _integration No co- CO2_{GDP, GDP} manufacturing, 2 -3.13 Co-integration _{at 10% level}

CO2 electricity, GDP -2.92 _integration No co- CO_{GDP, GDP}2 electricity, 2 -2.94 _integration No co-CO2 transport, GDP -1.77 _integration No co- CO_{GDP, GDP}2 transport, ₂ -3.25 Co-integration _{at 10% level} CO2 residential, GDP -1.19 _integration No co- CO_{GDP, GDP}2 residential, 2 -1.76

No co-integration CO2 other sectors, GDP -2.06 _integration No co- _{GDP, GDP}CO2 others, 2 -2.34

No co-integration Critical values for a co-integration regression containing an intercept

1% 5% 10%

-3.96 -3.37 -3.07

Table 5: Engle-Granger co-integration test

India

Equation t-stat Outcome Equation t-stat Outcome CO2 (all sectors), GDP -1.63 _integration No co- CO_{GDP, GDP}2 (all sectors), 2 -1.94 _integration No co-CO2 manufacturing, GDP -1.17 _integration No co- CO2_{GDP, GDP} manufacturing, 2 -2.32

No co-integration CO2 electricity, GDP -1.33 _integration No co- CO_{GDP, GDP}2 electricity, 2 -1.88

No co-integration CO2 transport, GDP -3.42 Co-integration _{at 5% level} CO_{GDP, GDP}2 transport, 2 -3.64

Co-integration at 5% level CO2 residential, GDP -1.40 _integration No co- CO_{GDP, GDP}2 residential, 2 -2.09

No co-integration CO2 other sectors, GDP -2.89 _integration No co- _{GDP, GDP}CO2 others, 2 -4.35

Co-integration at 1% level Critical values for a co-integration regression containing an intercept

1% 5% 10%

-3.96 -3.37 -3.07

Table 6: Engle-Granger co-integration test

India, a change would also be observable in the CO2 emissions due to transportation and other sectors, which contain commercial activities, and also agriculture, forestry and fishing. Based on the results of the previous quadratic modelling it can be seen that transport sector follows a positive relationship with GDP, as with the continued increase of GDP, also the CO2 emissions are increasing. For the other sectors that relationship is negative, as with increasing GDP per capita, it seems that CO2 emissions decline.

The results for the bounds F-test for co-integration for the estimated ARDL specification show the lags of Ln GDP in the linear model and the lags of Ln GDP as well as the lags of Ln GDP2 in the quadratic model. Similarly to the Engle-Granger co-integration test for the linear model for China, where none of the variables showed co-integration, the bounds F-test for CO2 emissions and GDP per capita showed no co-integration. The values of the F-statistic for all sectors as well as the total CO2 emissions were smaller than the upper critical bounds for 10% significance in the quadratic estimation (Table 7), which illustrates no co-integration, as the null hypothesis of no co-integration can not be rejected. This points towards the non-existence of a linear relationship between CO2 emissions and GDP. The only variable that shows co-integration in the quadratic estimation is the total CO2 emissions. This points to a long-run relationship with GDP per capita. The overall estimation from 1980 to 2013 seems to suggest that the trend is positive, as an increase in GDP also sees an increase in CO2 emissions. Based on the aforementioned notion of only analyzing the observations beginning at 2000, it would suggest that an increase in GDP would be related to a decrease in CO2 emissions overall in China.

For India the bounds F-test approach of Pesaran, Shin and Smith (2001), showed the same result as the previous test regarding the linear model apart of the transport industry where the bounds F-test found no co-integration. Additionally to the transport and other sectors in the Engle-Granger test, the bounds F-test also shows a co-integrating relationship between the CO2 emissions in the manufacturing and residential sectors in the quadratic estimation. This means that a relationship has been found between these four industries’ CO2 emissions from fuel combustion per capita and GDP per capita.

CO2 manufacturing 2,0 2.220 2,0,0 4.020 CO2 electricity 1,0 4.634 1,0,0 2.900 CO2 transport 3,0 2.232 3,0,0 3.191 CO2 residential 3,0 1.795 3,0,0 3.699 CO2 other sectors 1,0 2.045 1,0,0 3.059 Table 7: Bounds F-test

India Model F-statistic Outcome Model F-statistic Outcome CO2 all sectors 1,2 3.278 1,2,2 1.710

CO2 manufacturing 1,2 2.816 1,0,0 4.785* Co-integration CO2 electricity 1,0 4.259 1,0,0 2.466

CO2 transport 1,2 1.845 3,2,2 7.894*** Co-integration CO2 residential 1,0 1.752 1,0,0 4.917* Co-integration CO2 other sectors 1,0 3.725 1,0,0 5.625** Co-integration Table 8: Bounds F-test

After having chosen the most adapt lag length of the variables subsequently to an ARDL model selection based on Akaike Information criterion and the Schwarz criterion, the significance of the lags has been noted. The regression of the best fitting model has been given in tables 9 (China) and 10 (India). When estimating the ARDL model for the quadratic relationship for China the results show the same outcomes as the ARDL model used in the bounds F-tests, as the squared value of GDP does not show any significance in any of the variables. This indicates that whilst the scatter plots together with the quadratic fitting show a relationship, the quadratic ARDL model seems to suggest that the lags of GDP per capita are not relevant to predict future values of CO2 emissions. Therefore, the CO2 emissions of all sectors as analyzed through the years 1980 to 2013 show no importance of the GDP variable and denies the apparent EKC relationship as the lags of the quadratic form of GDP per capita are not relevant.

In the case of India, the optimal lag lengths based on the model selection through AIC and SC, are a bit different than the preferred models in the bounds F-test co-integration approach. The case of India provides positive results towards the importance of the lags of the squared GDP per capita variable. The total CO2 emissions sectors provides the best result towards answering the research question of the existence of the EKC. In the total emissions it can be seen from Table 10 that both linear and quadratic GDP variables are significant in the estimation of this least squares regression. This ARDL model though points to an upward-sloping U-shaped curve as the lag of GDP squared is positive and not negative. The same also holds for the electricity and transport sectors, but whilst this result contradicts the results found in the quadratic fitting for the electricity sector it does confirm it for the transport sector as the latter also shows an upward-sloping trend. The results for the manufacturing sector also show that the ARDL(1,1,1) model with one lag of every variable is the best fitting one, however it does not show any significance for the lags of the GDP variable. Therefore, whilst a quadratic relationship has been found in the case of some sectors in India, these still don’t show a declining relationship between CO2 emissions and GDP, and the EKC hypothesis has to be declined when basing the analysis on least squares regressions and lags of the variables.

4.4 Analysis of Granger-causality

The different series have also been tested for Granger-causality in order to check if given the past values of the dependent variable, the past values of the independent variable are useful to predict values of the dependent variable. After having tested for the optimal lag length to take during the vector autoregression estimation, and having estimated the appropriate VAR model, the Granger-causality test has been applied on the first differences of the variables (Tables 11 and 12).

The results for China show that in the case of the manufacturing sector and the CO2 emissions stemming from all industries, both the log of GDP and the squared log of GDP Granger-cause the CO2 emissions, as the null hypothesis of no Granger-causality can be rejected at the 10% and 5% level, respectively. Whilst for manufacturing and total emissions the optimal lag length of the variables in the regression is 4 for the electricity and transport sector the optimal lag lengths are 1 and 2, respectively. For both Ln GDP and Ln GDP2 _{the null hypothesis of no} Granger-causality can not be rejected for neither electricity or transport, or any of the other industries. This acts as a follow up test towards the importance of the lags of GDP in prediction of CO2 emissions and is in contrast with the above ARDL estimation as there no significance has been found of past values of GDP per capita. These results seem to suggest that in fact past values of the linear and quadratic terms of GDP contain information that helps predicting CO2 emissions for all and manufacturing sectors, beyond the information that is alone contained in past values of CO2 emission.

China

Dependent variable Relationship F-statistic CO2 (all sectors) 4 th _{lag GDP -> CO} 2 4.35** 4th _{lag GDP}2 _{-> CO} 2 3.69** CO2 manufacturing 4 th _{lag GDP -> CO} 2 4.22* 4th _{lag GDP}2 _{-> CO} 2 3.59* CO2 electricity 1 st _{lag GDP - CO} 2 0.56 1st _{lag GDP}2 _{- CO} 2 0.22 CO2 transport 2nd _{lag GDP -> CO} 2 1.82 2nd _{lag GDP}2 _{-> CO} 2 2.03 CO2 residential 4 th _{lag GDP - CO} 2 2.45 4th _{lag GDP}2 _{- CO} 2 1.81 CO2 other sectors 4 th _{lag GDP - CO} 2 1.21 4th _{lag GDP}2 _{- CO} 2 0.77

Table 11: Granger-causality test

India

Dependent variable Relationship F-statistic CO2 (all sectors) 1 st _{lag GDP - CO} 2 0.16 1st _{lag GDP}2 _{- CO} 2 0.19 CO2 manufacturing 1 st _{lag GDP - CO} 2 0.83 1st _{lag GDP}2 _{- CO} 2 1.23 CO2 electricity 2 nd _{lag GDP -> CO} 2 3.11* 2nd _{lag GDP}2 _{-> CO} 2 2.99* CO2 transport 1 st _{lag GDP -> CO} 2 3.85* 1st _{lag GDP}2 _{-> CO} 2 4.54** CO2 residential 2nd _{lag GDP -> CO} 2 3.35* 2nd _{lag GDP}2 _{-> CO} 2 3.15* CO2 other sectors 2 nd _{lag GDP - CO} 2 2.25 2nd _{lag GDP}2 _{- CO} 2 2.09

Table 12: Granger-causality test

5. Conclusion

through annual differences in GDP per capita on CO2 emissions per capita on a total national level and in the most polluting sectors.

The bounds F-test for co-integration show no evidence in neither China nor India that the model is significant at the linear level, whilst only the total emissions in China showed a quadratic relationship of co-integration. In India four sectors showed a co-integration relationship at the quadratic level, which are the manufacturing, residential, transport and other sectors. The estimation for the ARDL models do not show any significance for the income variable for the Chinese industry emissions, only past lags of emissions are relevant in inferring over future values of emissions. In the case of India, GDP and the squared value of GDP play a significant role in total emissions and emissions from the electricity and transport sectors. This represents a positive finding towards the purpose of this thesis, as it has been proven in various ways that a relationship between GDP and CO2 emissions is present. Even though the results for India are more illustrative than the ones for China for the total data set. As aforementioned, it seems that China is somewhat on a path of a downward-sloping EKC curve if one starts the analysis around 2000. Although after having done further tests for co-integration and having estimated the ARDL model with the latter 14 observations, the results don’t show much dissimilarity to the ones with the full set of observations. Moreover, the Granger-causality test was performed in order to test whether lagged values of economic growth have improving capabilities on forecasts of CO2 emissions. In this case the results show that in the case of China, total CO2 emissions and CO2 emissions from the manufacturing sector are somewhat Granger-caused by GDP, even at the squared value. For India, the electricity, transport and residential sectors showed rejection of no Granger-causality. This indicates that past values of GDP contain information that helps predict CO2 emissions beyond the significance of past values of CO2 emissions alone.

are worth noticing and taking into account when providing policies. It is to acknowledge that technologies to help achieve the abatement effect are very costly, and that only with perseverance and patience will countries such as China and India be more environmentally friendly. Relevant for future studies is the notion that it is of high importance to split CO2 emissions stemming from different sectors, as it has definitely been shown that results differ if total or specific CO2 emissions are taken into account. Moreover, future studies who will benefit of more up-to-date data, are possibly going to find relevant results, as it is documented that emissions of the past few years are stagnating an even on a declining trend.

Bibliography

Alam, M., Murad, W., Noman, A., & Ozturk, I. (2016). Relationships among carbon emissions, economic growth, energy consumption and population growth: Testing Environmental Kuznets Curve hypothesis for Brazil, China, India and Indonesia.

Ecological Indicators , 70, 466-479.

Ang, J. B. (2007). CO2 emissions, energy consumption, and output in France. Energy Policy ,

35 (10), 4772-4778.

Apergis, N., & Payne, J. E. (2010). The emissions, energy consumption, and growth nexus: evidence from the commonwealth of independent states. Energy Policy , 38 (1), 650-655.

Carson, R., Jeon, Y., & McCubbin, D. (1997). The relationship between air pollution emissions and income: US Data. Environment and Development Economics , 2 (4), 433-450.

Carter Hill, R., Griffiths, W., & Lim, G. (2011). Principle of Econometrics (4th Edition ed.). USA: Wiley.

Elliott, G. R., Rothenberg, T. J., & Stock, J. H. (1996). Efficient tests for an autoregressive unit root. Econometrica , 64 (4), 813-836.

Fujii, H., & Managi, S. (2013). Which industry is greener? An empirical study of nine industries in OECD countries. Energy Policy , 57, 381-388.

Greenpeace. (2017, February 28). China forecasts fourth year of stable or declining CO2

emissions as world awaits Trump climate action. Retrieved from Greenpeace - Press

Releases 2017:

http://www.greenpeace.org/international/en/press/releases/2017/China-forecasts- fourth-year-of-stable-or-declining-CO2-emissions-as-world-awaits-Trump-climate-action---Greenpeace/

Hamit-Haggar, M. (2012). Greenhouse gas emissions, energy consumption and economic growth: A panel cointegration analysis from Canadian industrial sector perspective .

Energy Economics , 34 (1), 358-364.

Jaunky, V. C. (2011). The CO2 emissions-income nexus: Evidence from rich countries.

Energy Policy , 39 (3), 1228-1240.

Jayanthakumaran, K., Verma, R., & Liu, Y. (2012). CO2 emissions, energy consumption, trade and income: A comparative analysis of China and India. Energy Policy , 42, 450-460.

Levinson, A. (2015, March 11). Does economic growth reduce pollution? Retrieved from World Economic Forum - Environment:

Levinson, A., & O'Brien, J. (2015). Environmental Engel Curves. NBER Working Paper No.

20914 . Cambridge, MA: National Bureau of Economic Research.

Lin, B., Moubarak, M., & Ouyang, X. (2014). Carbon dioxide emissions and growth of the manufacturing sector: Evidence for China. Energy , 76, 830-837.

Liu, G. (2006). A causality analysis on GDP and air emissions in Norway. Discussion Papers

No. 447 .

Markandya, A., Golub, A., & Pedroso-Galinato, S. (2006). Empirical Analysis of National Income and SO2 Emissions in Selected European Countries. Environmental and

Resource Economics , 35 (3), 221-257.

Narayan, P. K., & Narayan, S. (2010). Carbon dioxide emissions and economic growth: Panel data evidence from developing countries. Energy Policy , 38 (1), 661-666.

Odhiambo, N. (2009). Energy consumption and economic growth nexus in Tanzania: An ARDL bounds testing approach. Energy Policy , 37 (2), 617-622.

Pao, H.-T., & Tsai, C.-M. (2010). CO2 emissions, energy consumption and economic growth in BRIC countries. Energy Policy , 38 (12), 7850-7860.

Pesaran, M., & Shin, Y. (1999). An autoregressive distributed lag modeling approach to cointegration analysis. In S. Strom, Econometrics and Economic Theory in the 20th

Century (pp. 371-413). UK: Cambridge University Press.

Pesaran, M., Shin, Y., & Smith, R. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics , 16 (3), 289–326.

Ren, S., Yuan, B., Ma, X., & Chen, X. (2014). International trade, FDI (foreign direct investment) and embodied CO2 emissions: A case study of Chinas industrial sectors.

China Economic Review , 28, 123-134.

Robalino-Lopez, A., Mena-Nieto, A., Garcia-Ramos, J.-E., & Golpe, A. (2015). Studying the relationship between economic growth, CO2 emissions, and the environmental Kuznets curve in Venezuela (1980–2025). Renewable and Sustainable Energy

Reviews , 41, 602-614.

Saboori, B., & Sulaiman, J. (2013). CO2 emissions, energy consumption and economic growth in Association of Southeast Asian Nations (ASEAN) countries: a cointegration approach. Energy , 55, 813-822.

Saboori, B., Sapri, M., & bin Baba, M. (2014). Economic growth, energy consumption and CO2 emissions in OECD (Organization for Economic Co-operation and

Development)'s transport sector: A fully modified bi-directional relationship approach. Energy , 66, 150-161.

Schwert, G. W. (1989). Tests for unit roots: A Monte Carlo investigation. Journal of Business

Sharif Hossain, M. (2011). Panel estimation for CO2 emissions, energy consumption, economic growth, trade openness and urbanization of newly industrialized countries.

Energy Policy , 47 (1), 6991-6999.

Sica, E. (2014). Economic dualism and air quality in Italy: testing the Environmental Kuznets Curve hypothesis. International Journal of Environmental Studies , 71 (4), 463-480. Soytas, U., Sari, R., & Ewing, B. T. (2007). Energy consumption, income, and carbon

emissions in the United States. Ecological Economics , 62 (3-4), 482-489.

The World Bank Group. (2017, April 27). World Development Indicators. Retrieved from The World Bank: http://data.worldbank.org/data-catalog/world-development-indicators

Wang, S., Li, Q., Fang, C., & Zhou, C. (2016). The relationship between economic growth, energy consumption, and CO2 emissions: Empirical evidence from China. Science of

The Total Environment , 542 (A), 360-371.

Wang, Y., Zhang, C., Lu, A., Li, L., He, Y., ToJo, J., et al. (2017). A disaggregated analysis of the environmental Kuznets curvefor industrial CO2emissions in China. Applied

Energy , 190, 172-180.

Xu, B., & Lin, B. (2015). Factors affecting carbon dioxide (CO2) emissions in China's transport sector: a dynamic nonparamteric additive regression model. Journal of

Cleaner Production , 101, 311-322.

Yang, X., Lou, F., Sun, M., Wang, R., & Wang, Y. (2017). Study of the relationship between greenhouse gas emissions and the economic growth of Russia based on the

Environmental Kuznets Curve. Applied Energy , 193, 162-173.

Zambrano-Monserrate, M. A., & Fernandez, M. (2017). An Environmental Kuznets Curve for N2O emissions in Germany: an ARDL approach. Natural Resources Forum , 41 (2), 119-127.

Appendix 1: China:

𝐶𝑂2(𝑎𝑙𝑙) = 10.66 − 1.29 ∗ 𝐺𝐷𝑃 + 0.13 ∗ 𝐺𝐷𝑃2

𝐶𝑂2(𝑚𝑎𝑛𝑢𝑓) = 15.16 − 2.73 ∗ 𝐺𝐷𝑃 + 0.22 ∗ 𝐺𝐷𝑃2

China:

𝐶𝑂₂(𝑡𝑟𝑎𝑛𝑠𝑝) = 9.47 − 1.88 ∗ 𝐺𝐷𝑃 + 0.18 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑟𝑒𝑠𝑖𝑑) = 9.63 − 1.06 ∗ 𝐺𝐷𝑃 + 0.07 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑜𝑡ℎ𝑒𝑟) = 7.62 − 0.86 ∗ 𝐺𝐷𝑃 + 0.06 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑎𝑙𝑙) = −14.16 + 5.47 ∗ 𝐺𝐷𝑃 − 0.35 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑚𝑎𝑛𝑢𝑓) = 19.81 − 5.01 ∗ 𝐺𝐷𝑃 + 0.43 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑒𝑙𝑒𝑐𝑡𝑟) = −47.96 + 15.13 ∗ 𝐺𝐷𝑃 − 1.05 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑡𝑟𝑎𝑛𝑠𝑝) = 19.88 − 5.25 ∗ 𝐺𝐷𝑃 + 0.44 ∗ 𝐺𝐷𝑃2

𝐶𝑂₂(𝑟𝑒𝑠𝑖𝑑) = −14.01 + 5.26 ∗ 𝐺𝐷𝑃 − 0.38 ∗ 𝐺𝐷𝑃2

Appendix 2: China:

Figure A1

India:

Figure A3