Environ Dev Sustain (2010) 12:159–177 DOI 10.1007/s10668-009-9187-2
Economic growth and CO2 emissions Ranajoy Bhattacharyya Æ Tapas Ghoshal
Received: 29 October 2008 / Accepted: 9 February 2009 / Published online: 4 March 2009 Ó Springer Science+Business Media B.V. 2009
Abstract This article considers a planner’s optimum control exercise with environmental pollution and derives a testable link between the growth rates of consumption and pollution. The link is then empirically estimated for the case of CO2 emissions for a sample consisting of the union of top 25 countries in terms of CO2 emissions, population and per capita GNP. The analysis suggests that the interrelationship between the growth rates of CO2 emission and economic development is mostly significant for countries that have a high level of CO2 emissions and population. Keywords Environmental pollution Economic growth Optimum control Reduced form Income spectrum
1 Introduction There is a vast amount of research on the relationship between the level of income of a country and environmental pollution. Most of this research is conducted for a cross section of countries (or for panel data). Cross-sectional data is essential in the light of the particular type of question it addresses: whether the relationship follows a bell-shaped pattern or not. The bell shape is explained by the influence of abatement expenditure that becomes (a) affordable and (b) desirable (as ‘environment’ changes from a luxury to a necessity) at higher levels of income. Since a sufficiently wide spectrum of income is never available for time series data on countries, information of income in the cross section becomes Readers should send their comments on this paper to:
[email protected] within 3 months of publication of this issue. R. Bhattacharyya Indian Institute of Foreign Trade, Kolkata, India T. Ghoshal (&) Bureau of Applied Economics & Statistics, Government of West Bengal, New Secretariat Buildings, ‘B’ Block, 3rd Floor, 1, Kiron Sankar Roy Road, Kolkata 700001, West Bengal, India e-mail:
[email protected]
123
160
R. Bhattacharyya, T. Ghoshal
essential.1 Apart from the cross section of countries, some work has also been done for the cross section of states within a country.2 Recently, however, there has been some research on the time series data of particular countries/regions for the case of CO2 emissions. Given the nature of the data and the inability to address questions related to the particular shape of the relationship, this literature has addressed a different set of questions. For example, Lanne and Liski (2003) have used CO2 data from 1870 to 1998 to understand the basic trends and to use them for future projection in the development of carbon intensive energy use. Aldy (2004) has used state-level CO2 data for the US from 1960 to 1999 to estimate pre-trade (production-based) CO2 EKCs and post-trade (consumption-based) CO2 EKCs. Also, Hutington (2004) used CO2 data for the US from 1870 to show that the US will need to find additional policies to curb emissions if it wishes to prevent any further increase in its per capita emissions, and its per capita economy grows by more than 1.8% per year. One clear advantage with time series data is that it can address issues related to economic growth. One of the main objectives of this article is to exploit this advantage to find out the relationship between economic growth and emissions. However, we do not limit ourselves just to the relationship between growths. Once that is done, the second objective of the article is to determine the extent to which there is a long-run stable relationship between the level of economic development and CO2 emissions, and if so, the direction of causality of the variables purely over time. Thus, the article achieves the dual objective of carrying out time series analysis of causal linkage between economic development and environment pollution separately for each member of a group of countries by exploring the relationship between economic development and pollution at both ‘level’ and ‘difference’. The rest of the article is arranged as follows: Sect. 2 presents some background information to carry out exploratory data analysis that would enable us to propose some hypothesis regarding the relationship between the economic growth and growth of CO2 emissions. Section 3 develops theoretical model to derive a testable reduced form relationship supporting this causal empiricism. Section 4 carries out time series analysis on the basis of appropriate data collected over 1950–2000 for a sample of 38 countries. Section 5 considers the long-run relationship and Sect. 6 concludes the paper.
2 The data 2.1 The sample The sample of countries in this article is chosen with a view to highlight those countries, which are either high emitters or clear candidates for high levels of emission: those that have high incomes and/or population. We have thus selected the union of top 25 countries as far as population, per capita income, and CO2 emission are concerned. From this basic 1
Empirical estimates of the EKC conjecture have yielded conflicting results confirming it for some pollutants and rejecting it for others. For example, Shafik and Bandyopadhyay (1992), Panayotou (1993), and Seldon and Song (1994) have confirmed this inverted an U-shaped relationship between economic development and pollution for SO2, NOx and other particles, while Holtz-Eakin and Selden (1995) observed this type of inverted an U-shaped relationship in the case of CO2 emission. However, Grossman and Krueger (1995) and Torras and Boyce (1998) found an n-shaped relationship for SO2, Shafik and Bandyopadhyay (1992) found a monotonically increasing relationship for CO2, while Dinda (2000) obtained an U-shaped relationship for SPM and SO2. 2
See Vincent (1997).
123
Economic growth and CO2 emissions
161
sample, some countries have to be excluded. For instance, in the case of Germany, data for population and per capita income are available from 1970 only. So Germany has been excluded. For similar reasons, Bangladesh (data on CO2 available from 1972), Macao (data on population and per capita income available from 1986), Poland (data on population and per capita income available from 1970 onwards), Ukraine (data on population and per capita income are available from 1989 and data on CO2 emission are available from 1992) and Antigua (data on per capita income available from 1977) have been kept outside this study. After these exclusions, the total number of countries in the sample has turned out to be 38 and the selected countries by stages of development have been furnished in Table 1. 2.2 Data sources The CO2 data have been taken from the Oak Ridge National Laboratory (ORNL 2003). As it is well known, the ORNL emission estimates include CO2 from fossil fuel burning, cement manufacture and gas flaring. Altogether, 38 countries have been taken as classified by the stages of development. The time period considered is 1950–2000 in general. For Indonesia, these data relate to 1960–2000 and for Korea, the time period is 1953–2000, as the data on consumption (our proxy for economic development) for these countries relate to these periods. The data on consumption as well as openness for the corresponding time periods of the selected countries are taken from the Penn World Table, Version 6.1. The data on per capita consumption have been calculated on the basis of per capita income (Real GDP per
Table 1 Selection of countries by stages of development Criteria
High income
A: Top 25 Australia, Canada, France, CO2 emitter Italy, Japan, Spain, UK, USA
Upper middle income
Lower middle income
Brazil, Mexico, Korea, South Africa, Turkey
China, Iran, Thailand India, Indonesia
B: Top 25 by per capita income
Australia, Belgium, Canada, Denmark, Finland, France, Hong Kong, Iceland, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, Norway, New Zealand, Singapore, Spain, Sweden, Switzerland, UK, USA
C: Top 25 by population
France, Italy, Japan, UK, USA
The sample
Australia, Austria, Belgium, Brazil, Korea, Mexico, South Canada, Denmark, Finland, Africa, Turkey France, Hong Kong, Iceland, Ireland, Israel, Italy, Japan, Luxembourg, Netherland, New Zealand, Norway, Singapore, Spain, Sweden, Switzerland, UK, USA
Low income
China, Columbia, Brazil, Korea, Egypt, Iran, Mexico, Turkey, Philippines, South Africa Thailand
Ethiopia, Indonesia, India, Nigeria
China, Columbia, Egypt, Iran, Thailand, Philippines
Ethiopia, Indonesia, India, Nigeria
123
162
R. Bhattacharyya, T. Ghoshal 3.00
Fig. 1 Average per capita CO2 emission by economies
1960
2000
M.T. of Carbon
2.50 2.00 1.50 1.00 0.50 0.00 High Income
Upper Middle Income
Lower Middle Income
Lower income
capita based on Laspyers’ chain index) and consumption as a percentage of income. In order to arrive at the aggregate consumption for the countries in the sample, corresponding population data obtained from the Penn World Table, Version 6.1 have been used. 2.3 Data description 2.3.1 The CO2 data It is well known that CO2 emissions are considerably higher for developed countries (Fig. 1) with United States leading the list not only in terms of emissions but also in terms of the targeted and projected emissions. The extent of CO2 emission by world’s top 20 polluters, both in aggregate and in per capita terms, have been furnished in Table 2. It is observed that the rank in respect of total emission and its per capita differ significantly.3 Further, out of the top 20 polluters, nine are high-income countries, eight are middleincome countries and three are low-income countries. However, the growth rate of pollution is lower for developed countries than for underdeveloped countries4 (Table 3). With a higher base for developed countries, this, however, does not translate to the fact that the total magnitude of emission is higher for less-developed countries. In fact, the extent of additional CO2 emission generated is maximum for the United States (Table 3). The United States with a growth rate of only 1.61% generates almost 10 times more CO2 than Indonesia that has a significantly high rate of growth amongst the reported low-income countries. However, the relatively low-income countries (especially, the lower middleincome countries) are rapidly catching up with the developed countries in terms of emissions.5 For example, China has the second largest average yearly pollution in the 3
The rank correlation between the two series of ranks is 0.281.
4
This has been seen, in general, for almost all the countries considered. However, the results for four countries have been reported here.
5
The average per capita emission for high-income countries is disproportionately high (see Fig. 1) with USA on the lead. Among the rest of the countries, analysis of our raw dataset reveals that the growth rate of emission for the lower middle-income countries (6.86%) have been much higher than that of upper-middle income countries (4.43%) cumulative emission for the lower middle-income countries is about 1.5 times than that of upper-middle income countries. The growth rate of emissions for low incomes countries is almost the same as that of lower middle-income countries, and moreover, it has the lowest standard deviation.
123
Economic growth and CO2 emissions
163
Table 2 World’s top-20 CO2 emitters, 2002 Country
Total CO2 emission
Australia
Rank in total emission
Per capita CO2 emission
Rank in per capita emission
97,096
14
4.94
2
Canada
140,915
8
4.49
3
France
100,358
12
1.69
14
Germany
219,270
6
2.66
6
Italy
117,989
10
2.05
11
Japan
327,939
5
2.57
7
Spain
82,998
20
2.03
12 9
United Kingdom
148,129
7
2.5
1,592,382
1
5.52
1
Brazil
85,492
17
0.49
18
Korea
121,578
9
2.55
8
Mexico
104,543
11
1.01
16
Saudi Arabia
92,794
16
4.22
4
South Africa
94,110
15
2.07
10
957,249
2
0.74
17
98,153
13
1.5
15
Russian Federation
390,439
3
2.69
5
India
332,677
4
0.32
20
Indonesia
83,513
19
0.39
19
Ukraine
83,599
18
1.73
13
United States
China Iran
Source: Oak Ridge National Laboratory (ORNL 2003) Notes: (1) Total Emissions are expressed in thousand metric tons of carbon. National per capita estimates are expressed in metric tons of carbon. (2) The top 20 nations have been classified according to the GNI per capita, 2,000 calculated by World Bank. The groups are: low income (LI) $755; lower middle income (LM) $756–2995; upper middle income (UM) $2996–9265; and high income (HI) $9266 or more
Table 3 Least square growth rates of CO2 emission and extent of average annual additional CO2 emission Country
United States
Growth rate
Annual average additional emission (’000 metric tons of C) Entire period
1981–2000
1991–2000 21,398.30
1.61
16,741.88
13,300.70
10.01
2,454.00
4,119.30
5,071.90
Indonesia
7.08
1,693.38
2,387.50
2,834.80
India
5.95
5,481.88
9,873.95
10,796.90
Korea
Source: Calculated on the basis of estimated emission given by Oak Ridge National Laboratory (2003) Notes: (1) The least square growth rate r is estimated by fitting a linear regression trend line to the logarithmic annual values of the Ln Xt = a ? bt. In this equation X is the variable and t is time. If b* is the least square estimates of b, the average annual growth rate, r, is obtained as [exp(b*) - 1] and is multiplied by 100 to express it as a percentage (World Development Report, 1999–2000; World Bank). (2) Year wise additional emission has been calculated as xt - xt-1, ‘t’ being time point. (3) Simple average over the respective time periods have been calculated to obtain annual average additional emission
123
164
R. Bhattacharyya, T. Ghoshal
Fig. 2 Scatter diagram for income and CO2 emissions by selected countries
second half of the twentieth century (see Table 2). Mexico and Brazil are well within the top 15. India is an exception among the low-income countries. Given its income, its level of emission is disproportionately high (fourth in the world). 2.3.2 CO2 and per capita income The CO2 data has been found to behave differently with respect to economic development (with per capita income as its proxy) compared to other pollutants. For pollutants such as SO2, CO, NOx, and SPM EKC has often been observed with a turning point well within the income sample of the researchers. However, for CO2, most researchers have found a monotonically rising pollution–income relationship or an EKC with a turning point far outside the income sample (see Leib 2004). Since our data are time series with a limited income spectrum, this trend obviously continues here for all countries (Fig. 2 reports six of them). From Fig. 2, it seems that the United States has reached a stage where the curve is about to bend back. Both per capita income and CO2 have grown consistently over time for all countries in the sample except UK and Luxembourg, and thereby data on total consumption and CO2
123
Economic growth and CO2 emissions
165
Table 4 Number countries for which significant correlation exists between CO2 emission and total consumption Item
Significant at 1%
Significant at 5%
Insignificant
Variables at level
35 (35)
1 (2)
2 (1)
Growth of variables
11 (7)
7 (8)
20 (23)
Source: Based on the estimates of emission obtained from ORNL and total consumption obtained from Penn World Table, Version 6.1 Notes: Figures in brackets denote the number of countries for which the correlation coefficient is significant based on the variables in their natural logarithm
emission have also shown increasing trend except for the two countries mentioned above. It is therefore not surprising that they have a high positive correlation among each other.6 The correlation coefficient between CO2 emission and total consumption are fairly high both for original variables and their natural logarithm for almost all the countries except Belgium, France, Luxembourg and UK. For Luxembourg and UK, these correlation coefficients are negative and the absolute values of the correlation coefficient for all the four countries are \0.5. However, high level of correlations is not observed about their growth rates.7 The number of countries for which significant correlation is observed at level as well as the first differences of these variables have been reported in Table 4.
3 The model Consider a simple model with environmental pollution: Y ¼CþI I¼S S ¼ dY P ¼ aY
ð1Þ
where Y, C, I, S are aggregate income, consumption, investment and saving, respectively. P is the aggregate flow of pollution in the economy. We assume that the flow of pollution is linearly related to the flow of income. Obviously: 0 \ d, a \ 1. In order to derive the relationship between the growth rates of the variables, we recast the above model as a planner’s optimum control problem with the utility function for the economy being U = U(C, P) where, as before, C is aggregate consumption and P the flow of pollution. Following convention, we assume: UC [ 0, UCC \ 0, UP \ 0, UPP \ 0. The literature usually assumes that the value of UCP is 0. However, as it will be obvious, as we progress we need to assume it to be non-zero here. In assigning a sign to UCP , it seems ‘natural’ to assume that UCP is negative (as the flow of pollution increases, the marginal utility from consumption falls). However, a positive sign may not be totally implausible. Pollution is a bad that depresses the overall utility level, but for that very reason, the utility 6
There is also a positive correlation between the ranks of countries arranged according to mean CO2 emissions and mean per capita income (?0.55). 7
The simple correlation between the first difference values of these variables is very weak (the weakest is 0.01).
123
166
R. Bhattacharyya, T. Ghoshal
value of an extra unit of consumption may actually be higher at higher levels of pollution. Also, external effects may play a role here. Pollution may be a byproduct of activities like infrastructure investment or industrial growth that improve the general standard of living of a poor country. This will have a positive impact on the welfare gain from extra consumption. Thus, at least in the initial stages of development (low values of C and P), rise in pollution may be compatible with a rise in marginal utility of consumption. However, as the level of pollution creating activities goes on increasing, rising disutility will overwhelm the beneficial external effects and UCP will eventually turn negative. Thus, the sign should be positive up to a level of these activities and negative thereafter. Assuming a monotonically increasing relationship between pollution creating activities and economic development, we may well expect the sign to change from positive to negative after a level of economic development. As we will see below, in the context of the rather simplistic model that we propose this has an important implication as far as getting an inverted U relationship between the growth rates of C and P is concerned. The production structure, as we have mentioned, follows the AK model. Let the production function of the economy be given by: Y ¼ hK
ð2Þ
We assume the pollution formation equation to be the following: P ¼ aY bA
ða; b [ 0Þ;
ð3Þ
where A is the abatement expenditure and a and b are constants. Given (2), (3) is re-written as: P ¼ ahK bA
ð3aÞ
The rate of change of capital is given by the reformulated equilibrium condition of the economy as in (1), that is:
K ¼Y CA
ð4Þ
using (3) this is reformulated as:
K ¼ hK C A
ð4aÞ
Given the aggregate welfare for the economy U = U(C, P), the planner’s optimum control problem is therefore: Z1 Max
eqt UðC; PÞdt
0
subject to (4) and the initial condition K(0) = K0. This yields the following Hamiltonian function: H ¼ UðC; PÞeqt þ k½hK C A And the following current value Hamiltonian (Hc): Hc ¼ UðC; PÞ þ q1 ½hK C A
ð5Þ
where q1 is the co-state variable, C and A are the control variables and K is the state variable.
123
Economic growth and CO2 emissions
167
From Hc, first order conditions yield: dHc ¼ UC q1 ¼ 0 dC
ð6Þ
dHc ¼ bUP q1 ¼ 0 dA
ð7Þ
The equation of motion for the co-state variable is:
q1 ¼
dHc þ qq1 ¼ ahUP hq1 þ qq1 dK
Using (5) this implies:
q1 ahUP ¼ þ ðq hÞ q1 UC
ð8Þ
Taking log derivative of (5) we have:
q1 UCC C þUCP P ¼ q1 UC
ð9Þ
. Equating q1 q1 from above, and a bit of manipulation yields the following:
g C ah UP q h P ¼ CC þ gCP P gCP C gCP UC Using (5) and (6) this reduces to: gCC C 1 a P h 1 þq ¼ þ gCP b P gCP C
ð10Þ
where gCC and gCP are, respectively, the elasticities of marginal utility with respect to consumption and pollution: gCC ¼
dUC C dC UC
gCP ¼
dUC P dP UC
(10) gives us the relationship between the dynamic paths of P and C that is necessary for the social utility maximization subject to the constraints and initial conditions mentioned above. Note that since gCC \ 0 (as UCC \ 0 and UC [ 0), the relation between P and C depends entirely on gCP. The intuition is simple: the social planner will tolerate higher pollution resulting from an increase in consumption, only if the pollution increases the marginal utility from consumption (UCP [ 0) and thus supports this utility maximization objective in some sense. If pollution on the other hand reduces the marginal utility from consumption, then its increase cannot be justified at all and the utility maximization objective dictates that consumption can be increased only if that reduces pollution. It should be noted that if, as we have argued in the beginning of this section, UCP is positive up to a certain level of economic development and negative thereafter, (10) gives us an inverted U-shaped relationship between P and C. (10) can thus be interpreted as a theoretical support for the inverted U relationship of EKC in a dynamic setup.
123
168
R. Bhattacharyya, T. Ghoshal
4 Estimating the relationship between the growth rates of consumption and CO2 emission In line with the above theoretical analysis, we use an aggregate consumption of the economy rather than per capita income as an indicator of economic development in the rest of the article. In estimating Eq. 10, we assume that the elasticities of marginal utility with respect to consumption and pollution are constant so that (10) can be written down in the conventional form:
P C ¼ u þ n þ et P C
ð10aÞ
where u¼
gCC gCP
and,
1 a þq n¼ h 1 gCP b
. with n and u as constants. For empirical purposes, we define x x as Dlnxt. DLCE and DLTC (where CE stands for CO2 emission and TC aggregate consumption with L denoting their natural log) are (0) for all countries in the sample. We therefore use simple OLS for estimating the parameters in (10a) for all these countries. The results are reported in Table 58, where the countries are arranged according to their income. It is revealed from the significance of n that though no clear pattern immerges as far as income categories are concerned, the relationship is relatively strong in the case of high- and middle-income countries. Out of 23 high-income countries in the sample of 38 countries, it is significant for 10 countries. The relationship between the variables is strongest (statistically significant at the 5% level) for United States, Spain, Switzerland, Japan, Iceland, France, Finland, Canada and Belgium. For upper middle-income countries, the relationship is significant for four out of five selected countries. It is significant at 5% level for Korea, Mexico and Turkey, and at 10% level for Brazil. For lower middle-income countries, out of a total of six selected countries the relationship is significant at 5% level for China and Philippines, while it is at the 10% level for Iran and Thailand. The relationship between the growth rates is almost absent for the low-income countries (Indonesia is an exception).9 Thus, clearly the results for the growth rates are different in spirit from the EKC literature. It is weaker for lower income countries than for higher income countries.10
8
It can be seen from Table 5 that the error terms in the OLS regressions are white noise for all countries except Italy.
9 Interestingly, United Kingdom that does not have a significant growth rate of CO2 emissions does not have a significant relationship between the growth rate of emissions and consumption as well. However, for Luxembourg that again does not have any growth in CO2 emission during the period under study, the correlation between growth in emission and growth in consumption (in their natural log) is negative and also n is negative for this country (as seen in Table 5). In fact, for the five countries in the sample (Denmark, Luxembourg, Norway, Hong Kong and China) that have negative correlation between consumption and emission also, as expected, have negative values of n though these negative values are not statistically significant (except China, see Table 5). 10
India with its high level of emission has an insignificant relationship between the growth rates.
123
Economic growth and CO2 emissions
169
Table 5 Emission growth and consumption growth in (aggregate) Country
Without openness a
DLTC
b
LM
With openness c
DF
DLTCa
DLOPNa
LMb
DFc
High income Australia
0.012
0.00001*
-7.240*
0.180
Austria
0.238
0.799*
-7.978*
0.151
-0.182 0.472*
0.521*
-5.477*
1.580*
-8.393*
Belgium
1.098*
0.099*
-7.525*
0.657
0.551*
0.537*
-6.736*
Canada
0.696*
0.006*
-7.062*
0.678*
0.343**
0.103*
-7.260*
Denmark
-0.019
Finland
1.059*
France Hong Kong
0.676* -0.106
0.001*
-6.911*
0.021*
-6.809*
0.009*
-7.226*
0.017*
-5.279*
-0.004 0.909* 0.657* -0.123
-0.131 0.716* 0.402* -0.384
0.009*
-6.981*
0.0002*
-6.686*
0.008*
-7.123*
0.084*
-5.129*
Iceland
0.473*
1.210*
-6.674*
0.376**
0.307
1.132*
-6.510*
Ireland
0.198
1.500*
-7.464*
0.17
-0.161
3.521*
-7.880* -7.898*
Israel
0.297
0.832*
-6.305*
0.176
-0.376
0.797*
Italy
0.4196
2.280*
-2.940
0.466*
0.605
1.024*
-5.317*
Japan
1.103*
0.008*
-6.974*
0.945*
0.586*
1.955*
-5.748*
Luxembourg
-0.192
Netherlands
0.424**
New Zealand Norway
0.217 -0.16
2.626*
-5.926*
0.228*
-6.279*
-0.261 0.394**
0.744*
1.568*
-5.798*
0.669*
0.002*
-6.694*
2.714*
-8.968*
0.217
-0.047
2.790*
-8.912*
0.002*
-6.678*
-1.186
0.918
0.010*
-6.796*
Singapore
0.764
2.357*
-7.666*
0.453
-0.451
2.111*
-7.497*
Switzerland
2.244*
2.619*
-9.631*
25.255*
-0.028
2.615*
-9.629*
Sweden
0.07
0.266*
-7.374*
0.092
0.522
0.484*
-7.633*
Spain
0.441*
0.017*
-6.942*
0.474*
0.066
0.096*
-6.726*
United Kingdom
0.188
3.828*
-9.651*
0.329
United States
1.163*
0.003*
-4.511*
1.001*
-0.41 0.368
3.375*
-9.344*
0.024*
-7.243*
Upper middle income Brazil
0.329**
0.784*
-6.100*
0.242**
0.055
0.928*
-6.020*
South Africa
0.228
0.440*
-7.543*
0.203
0.053
0.397*
-7.433*
Turkey
0.316*
0.515
-7.699*
0.319*
0.02
0.521*
-7.659* -6.498*
Lower middle income -0.986*
8.184
-4.593*
-0.384
0.092
0.553*
Columbia
China
0.373
4.454
-9.494*
0.239
0.177
2.308*
-8.621*
Egypt
0.212
0.597*
-7.666*
0.22
0.157
0.789*
-7.737*
-0.029
Iran
0.487**
1.076*
-8.491*
2.980*
-5.614*
Thailand
0.468**
1.097*
-5.935*
0.456**
-0.071 0.443*
0.027*
-6.640*
Philippines
0.891*
0.564*
-6.800*
0.876*
0.271*
0.077*
-7.396*
Lower income Ethiopia
0.564
0.129*
-7.598*
0.934
0.494
0.138*
-7.227*
India
0.054
1.222*
-7.801*
0.056
0.006
1.328*
-7.846*
Indonesia
0.557*
0.212*
-6.371*
0.579*
0.156
0.615*
-6.652*
123
170
R. Bhattacharyya, T. Ghoshal
Table 5 continued Country
Without openness a
DLTC Nigeria
0.016
b
With openness c
LM
DF
0.332*
-6.410*
DLTCa 0.063
DLOPNa 0.315
LMb
DFc
0.250*
-6.520*
Notes: (1) Serial correlation in OLS regression has been taken care of by taking appropriate lag. (2) Unit root exists for the residuals of the regression equation without openness; thereby the coefficient has been taken as insignificant ‘*’ and ‘**’ denote the coefficients are significant at 5% and 10%, respectively. In the case of LM and DF, ‘*’ denotes absence of serial correlation and unit root for residuals respectively a OLS estimate b in the regression equation DLCE = a ? b DLTC ? et for regression without openness and OLS estimate for b1 and b2 for the regression equation DLCE = a ? b1 DLTC ? b2 DLOPN ? et with openness b
LM values for testing serial correlation
c
Dickey–Fuller statistics for testing unit root test for residuals with 95% critical values
The pattern of the results between the growth rates of the variables can also be looked upon in terms of the criteria by which the sample of countries has been chosen (Table 8). The most significant result is that the countries with a high level of emission have a high level of association between the growth rates of CO2 emission and economic development (14 out of top 20).
5 Extensions: robustness, cointegration and causality The above analysis was done with total domestic consumption as a proxy for economic development. The reduced form dictated this. However, the analysis can readily be extended in several directions both in terms of the nature of the variables considered as well as in terms of method. In this section, we report three such extensions: first, we test for the robustness of the results in terms of per capita rather than aggregates. Second, we introduce openness as a new variable. The logic for introducing this is that though the reduced form dictates aggregate consumption as the appropriate variable, the regressions in the previous sections were all in terms of domestic consumption. Third, we have extended the analysis to test for the existence of a long-run relationship between the variables. The results for the first two extensions also appear in Tables 5 and 6 where, as usual, openness is defined as Exports plus Imports divided by real GDP (denoted by OPN).11 It is clear from those tables that the results are robust to the inclusion of openness and redefinition of the variables in terms of their per capita levels. For openness, the two exceptions are Belgium and Iran, where the coefficient for total consumption became insignificant after considering the variable. For per capita levels, the pattern of the relationship does not change significantly.12 For cointegration of the aggregate variables, the Johansen procedure has been applied for 27 countries [for the remaining 11 countries, at least one of the variables under consideration is not I (1)]. This test suggests the existence of stable long-run link between 11
DLOPN is I(0) for all the 38 countries considered.
12
The only exception is Netherlands, where with the inclusion of openness the coefficient of per capita consumption became insignificant.
123
Economic growth and CO2 emissions
171
Table 6 Emission growth and consumption growth (per capita) Country
Without openness a
DLCPC
b
LM
Without openness c
DF
DLCPCa
DLOPNa
-0.137
LMb
DFc
High income Australia
-0.019
0.050*
-7.062*
0.130
0.224
0.889*
-8.041*
0.125
Austria
0.486*
0.184*
-6.504*
1.900*
-8.558*
Belgium
1.004*
0.120*
-7.470*
0.559
0.572*
0.534*
-6.707*
Canada
0.643*
0.024*
-7.124*
0.551**
0.378
0.198*
-7.374*
Denmark
-0.051
Finland
1.032*
France Hong Kong
0.618* -0.046
0.017*
-7.003*
0.024*
-6.803*
0.011*
-7.248*
0.00002*
-5.450*
-0.037 0.870* 0.595* -0.019
-0.133 0.722* 0.407* -0.302
0.038*
-7.076*
0.0008*
-6.673*
0.011*
-7.143*
0.029*
-5.308*
Iceland
0.463*
0.575*
-6.758*
0.362**
0.322
0.473*
-6.624*
Ireland
0.216
1.289*
-7.407*
0.194
-0.169
2.819*
-7.752* -8.182*
Israel
0.169
0.554*
-8.664*
0.044
-0.403**
0.450*
Italy6
0.368
2.060*
-3.023
0.424
0.589
0.916*
-3.64
Japan
1.115*
0.005*
-6.992*
0.962*
0.584*
1.965*
-5.748*
Luxembourg
-0.257
Netherlands
0.375**
New Zealand Norway
0.158 -0.162
2.643*
-5.937*
-0.363
0.162*
-6.348*
0.357
2.809*
-9.056*
0.0009*
-6.689*
0.188 -0.16
0.759*
1.506*
-5.819*
0.648**
0.0004*
-6.751*
-0.023
2.868*
-9.020*
0.901
0.010*
-6.796*
Singapore
0.785
2.434*
-7.696*
0.507
-0.411
2.223*
-7.543*
Switzerland
2.174*
2.043*
-9.342*
2.262*
-0.12
2.082*
-9.390*
Sweden
0.044
0.388*
-7.490*
0.066
0.525
0.671*
-7.768*
Spain
0.416*
0.01
-6.963*
0.446*
0.058
0.067*
-6.774*
United Kingdom
0.369
1.227*
-7.293*
0.271
United States
1.125*
0.015*
-4.502*
0.952*
-0.4 0.377*
3.587*
-9.420*
0.027*
-7.231*
Upper middle income Brazil
0.19
0.708*
-6.142*
0.19
0.068
0.781*
-6.087*
Korea
0.510*
2.577*
-4.949*
0.518*
0.055
2.794*
-5.383*
Mexico
0.786*
1.089*
-8.035*
0.851*
-0.124
1.446*
-8.222*
South Africa
0.786*
1.089*
-8.035*
0.154
0.058
0.466*
-7.499*
Turkey
0.301*
0.624*
-7.772*
0.305*
0.021
0.635*
-7.729*
Lower middle income China Columbia
-0.634
1.059*
-6.407*
-0.575
0.111
1.590*
-6.353*
0.304
0.103*
-6.742*
0.046
0.216
2.610*
-8.745*
Egypt
0.213
0.561*
-7.636*
0.218
0.16
0.700*
-7.728*
Iran
0.635*
0.918*
-8.395*
0.539
0.072
1.567*
-8.574*
Thailand
0.403
0.819*
-6.059*
0.383
0.449*
0.003*
-6.885*
Philippines
0.933*
0.648*
-7.939*
0.885*
0.273*
0.105*
-7.448*
Lower income Ethiopia
0.759
0.181*
-7.694*
0.712
0.514
0.556*
-8.180*
India
0.064
1.206*
-7.809*
0.068
0.01
1.363*
-7.877*
Indonesia
0.547*
0.247*
-6.404*
0.570*
0.152
0.664*
-6.678*
123
172
R. Bhattacharyya, T. Ghoshal
Table 6 continued Country
Without openness a
DLCPC Nigeria
0.018
b
Without openness c
LM
DF
0.351*
-6.444*
DLCPCa
DLOPNa
0.065
0.321
LMb
DFc
0.261*
-6.508*
Notes: (1) Serial correlation in OLS regression has been taken care of by taking appropriate lag. (2) Unit root exists for the residuals of the regression equations in both cases, thereby the coefficients have been considered to be insignificant ‘*’ and ‘**’ denote the coefficients are significant at 5% and 10% respectively. In case of LM and DF, ‘*’ denotes absence of serial correlation and unit root for residuals respectively a OLS estimate b in the regression equation DLCE = a ? b DLTC ? et for regression without openness and OLS estimate for b1 and b2 for the regression equation DLCE = a ? b1 DLTC ? b2 DLOPN ? et with openness b
LM values for testing serial correlation and ‘c’ denotes the Dickey-Fuller statistics for testing unit root test for residuals with 95% critical values
the levels of aggregate consumption and aggregate CO2 emission for 19 countries (see Table 7). The countries for which there was no relationship are: Austria, India, Indonesia, Israel, Philippines, Sweden, Thailand, Turkey and United Kingdom. More or less similar results were obtained while applying the test on the variables in per capita terms.13 United Kingdom is a special case, as far as the relationship between emission and its regressors are concerned. The reason for United Kingdom being an exception is obvious from a casual look at the emissions data: for United Kingdom CO2 emission did not grow at all within the period of study (1950–2000). On conducting an analysis similar to Table 8 (in Table 9) for the cointegration results, we find that the long-run relationship is stronger with respect to the level of economic development rather than emissions, as was the case in the case of the relationship between the growth rates. Turning to causality, we attempt to determine causality only for those countries for which cointegration (balanced or unbalanced) exists.14 It can be seen from Table 10 that there is causality from consumption to CO2 for only eight countries. Most of these are high-income countries: Australia, China, Mexico, Netherlands, New Zealand, Singapore, Spain and Switzerland. When the variables are considered in per capita terms, the causality in this direction exists for 10 countries; France, Israel, and United Kingdom are the new inclusions (China does not have causality in this direction at the per capita level). On the other hand, reverse causality (either at aggregate or at per capita level) exists for six countries: China, Hong Kong, India, Spain, Turkey and United Kingdom. Given the absence of any logical link from emissions to development, the reverse causality result 13
Results of 11 unbalanced equations are also reported in Table 7. See Banerjee et al. (1993) for an interpretation of unbalanced equations.
14 Granger (1988) pointed out that if a pair of series is cointegrated, there must be Granger-causation in at least one direction. This approach to find whether x causes y is to see how much of the current y can be explained by past values of y, and then to see whether adding lagged values of x can improve the explanation. y is said to be Granger-caused by x if x helps in the prediction of y, or equivalently if the coefficients on the lagged x’s are statistically significant. The Granger-causality test is based on the bivariate regressions of the form yt ¼ a0 þ a1 yt1 þ a2 yt2 þ þ al ytl þ b1 xt1 þ b2 xt2 þ bl xtl xt ¼ a0 þ a1 xt1 þ a2 xt2 þ þ al xtl þ b1 yt1 þ b2 yt2 þ bl ytl for all possible pairs of (x, y) series in the group. The reported F-statistics are the Wald statistics for the joint hypothesis b1 ¼ b2 ¼ ¼ bl ¼ 0
123
Economic growth and CO2 emissions
173
Table 7 Cointegrating vectors between CO2 emission and consumption (for both aggregate level and per capita) Country
LTC
LPCC
Country Switzerland
0.482
1.006, 2.071
Australia
0.591
0.416
United Kingdom
–
(-)0.1714
Austria
–
–
United States
0.443
0.167
Belgium
0.124
(-)0.008
Upper middle income
High income
LTC
LPCC
Canada
0.27
0.001
Brazil
0.880
0.798
Denmark
(-)3.275
(-)3.195
Korea
1.4644
1.5604
Finland
1.583, 4.151
1.527, 3.381
Mexico
1.140, 1.235
1.488, 2.127
France
0.857, 0.293
0.590, 0.036
South Africa
1.444, 0.906
1.391, 0.757
Hong Kong
0.8334
–
Turkey
–
2.128d
4
Iceland
0.455
0.196
Lower middle income
Ireland
0.8754
–
China
Israel Italy
4
–
0.916 4
1.003
4
4
0.994
4
Columbia Egypt
1.2214
1.2194
4
0.8964
4
1.3344
0.961 1.174
Japan
0.850
0.825
Iran
1.029, 4.962
1.002
Luxembourg
(-)0.804
–
Philippines
–
2.3194
Netherlands
0.311
0.098
Thailand
–
–
New Zealand
1.368
1.956
Lower Income
Norway
1.310
1.318
Ethiopia
2.7494
–
4
4
Singapore
1.531
1.784
India
–
0.700
Spain
(-)0.402, 1.056
0.606, 1.112
Indonesia
–
–
Sweden
–
–
Nigeria
3.396
–
Notes: (1) The order of VAR has been taken as 10 for all the countries. (2) The 95% critical values for the LR test based on maximal eigenvalues of the stochastic matrix are 14.880 for the null r = 0 against the alternative r = 1 and 8.070 for r B 1 against r = 2. The 90% critical values for the said test are 12.980 and 6.500, respectively. (3) ‘–’’ denotes absence of cointegration between the variables. (4) Estimation with unbalanced equations (see Banerjee et al. 1993 for interpretation)
appears to be spurious. However, let us note that Granger causality has a thematic implication that is not always appreciated, whereas interpreting results are derived from it. By its very statistical nature, Granger causality is a tool that comments on the extent to which a series can forecast the values of another series.15 To this extent, we can say that the emission data is a good indicator of the level of economic development of the countries for which the reverse causality exists. Now to summarize the results of the causality analysis in the light of Tables 8 and 9 above, it can be seen from Table 11 that causality from consumption to CO2 emission is stronger with respect to economic development. However, the reverse causality (from CO2 emission to consumption) is relatively stronger when the countries are arranged with respect to CO2 emission.
15 This ability to forecast may well translate into causality if economic logic supports it. If economic logic dictates something which is quite contrary to what the Granger causality analysis suggests, then all we can say is that the series contains ‘the’ market’s best information as to where (the explained series) might be headed’’ (Hamilton 1994, p. 307).
123
174
R. Bhattacharyya, T. Ghoshal
Table 8 Number of Countries for which relationship is significant Item
Variables at level
Growth of variables
Name of the variables
Within top 10
Within top 20
Without openness
With openness
Without openness
With openness 13 (9)
CO2 emission
7 (7)
7 (7)
14 (12)
Per capita income
3 (3)
3 (3)
9 (9)
9 (7)
Population
7 (6)
7 (6)
12 (10)
11 (9)
CO2 emission
5 (4)
4 (3)
11 (9)
11 (8)
Per capita income
5 (4)
5 (4)
11 (9)
11 (8)
Population
5 (4)
4 (3)
9 (7)
8 (6)
Source: Compiled on the basis of regression results given in Tables 5 and 6 Notes: (1) Significance of relationship is based on the significance of b (b1) coefficients in the regression equation: Yt = a ? bXt ? et, where y denotes growth in CO2 emission and X denotes growth in consumption (Yt = a ? b1X1t ? b2X2t ? et, where y denotes growth in CO2 emission, X1 denotes growth in consumption and X2 denotes openness). (2) All variables have been considered in their natural logarithm. (3) Figures in brackets denote the number of countries for which the relationship is significant based on regression equation as stated in 1 with the respective variables (except openness) in per capita terms
Table 9 Number of countries for which cointegration exists between CO2 emission and consumption (balanced cases) Item
Name of the variables
Within top 10
Within top 20
Variables at level
CO2 emission
4 (5)
10 (11)
Per capita income
7 (6)
14 (12)
Population
4 (4)
6 (7)
CO2 emission
1 (2)
6 (5)
Per capita income
1 (1)
8 (6)
Population
5 (4)
7 (7)
Growth of variables
Source: Compiled on the basis of results on cointegration given in Table 8 Notes: (1) Significance of relationship is based on the existence of cointegration with balanced equations between CO2 emission and consumption both in aggregate and per capita level. (2) All variables have been considered in their natural logarithm. (3) Figures in brackets denote the number of countries for which the cointegration exists with the respective variables in per capita terms
Table 10 F-statistics for Granger-causality estimated by consumption and CO2 emission regressions Country
Aggregate variables in Log
Per capita emission variables in Log
LTC does not cause LCE F-values (probability)
LCE does not cause LTC F-values (probability)
LPCC does not cause LPCE F-values (probability)
LPCE does not cause LPCC F-values (probability)
Australia
5.119* (0.004)
0.121 (0.927)
4.962* (0.005)
0.084 (0.968)
Belgium
1.950 (0.137)
1.093 (0.363)
2.092 (0.116)
1.248 (0.305)
Brazil
2.333 (0.088)
1.994 (0.130)
2.750 (0.055)
1.592 (0.206)
123
Economic growth and CO2 emissions
175
Table 10 continued Country
Aggregate variables in Log
Per capita emission variables in Log
LTC does not cause LCE F-values (probability)
LPCC does not cause LPCE F-values (probability)
LCE does not cause LTC F-values (probability)
LPCE does not cause LPCC F-values (probability)
Canada
0.368 (0.777)
0.290 (0.832)
0.446 (0.721)
0.220 (0.882)
China
2.676** (0.060)
3.308* (0.030)
2.112 (0.114)
3.532* (0.023)
Columbia
0.486 (0.694)
1.697 (0.183)
0.378 (0.770)
1.313 (0.283)
Denmark
0.493 (0.689)
0.312 (0.816)
0.555 (0.648)
0.319 (0.812)
Egypt
1.498 (0.229)
1.751 (0.172)
1.553 (0.215)
1.580 (0.209)
Ethiopia
1.998 (0.129)
10.404 (0.751)
na
na
Finland
0.993 (0.433)
0.858 (0.470)
1.058 (0.377)
1.036 (0.386)
France
1.781 (0.166)
1.210 (0.318)
2.199** (0.103)
1.089 (0.364)
Hong Kong
0.675 (0.574)
3.650* (0.023)
na
na
Iceland
1.277 (0.295)
0.775 (0.515)
0.678 (0.570)
0.760 (0.523)
India
na
na
0.671 (0.575)
2.609** (0.064)
Iran
0.608 (0.614)
1.836 (0.158)
0.780 (0.513)
1.716 (0.181)
Ireland
2.436 (0.078)
0.203 (0.894)
na
na
Israel
na
na
3.125* (0.036)
1.639 (0.195)
Italy
0.608 (0.614)
1.115 (0.354)
0.618 (0.607)
1.007 (0.400)
Japan
1.266 (0.299)
0.944 (0.428)
1.260 (0.301)
0.456 (0.714)
Korea
0.867 (0.466)
1.909 (0.144)
0.936 (0.432)
2.414 (0.082)
Luxembourg
2.045 (0.122)
0.405 (0.750)
na
na
Mexico
7.404* (0.0004)
1.657 (0.191)
6.881* (0.001)
1.145 (0.342)
Netherlands
3.043* (0.039)
0.647 (0.589)
2.929* (0.045)
0.574 (0.635)
New Zealand
2.289** (0.093)
0.539 (0.658)
2.324** (0.089)
0.341 (0.796)
Nigeria
0.907 (0.446)
2.606 (0.064)
na
na
Norway
1.301 (0.287)
0.934 (0.433)
1.271 (0.297)
0.834 (0.483)
Philippines
na
na
1.146 (0.342)
1.303 (0.286)
Singapore
8.642* (0.0004)
0.134 (0.939)
8.452* (0.0004)
0.287 (0.834)
South Africa
1.855 (0.152)
0.268 (0.848)
1.918 (0.142)
0.231 (0.874)
Spain
3.806* (0.017)
2.564** (0.068)
4.100* (0.012)
3.379* (0.027)
Switzerland
4.010* (0.014)
0.220 (0.882)
3.853* (0.016)
1.092 (0.363)
Turkey
na
na
0.264 (0.851)
6.714* (0.001)
United Kingdom
na
na
2.962* (0.043)
2.674** (0.060)
United States
1.439 (0.245)
0.770 (0.518)
1.172 (0.332)
0.956 (0.423)
Notes: (1) Three periods lag has been considered in the Granger-causality regression equations. (2) The countries for which cointegration (both balanced and unbalanced cases) is absent for aggregate variable as well as their per capita have not been considered at all. ‘na’ denotes not applicable due to the absence of cointegration (both balanced and unbalanced cases) for a specified set of variables. In this article, we do not determine the causality between the variables unless they have a long-run relationship between them ‘*’ and ‘**’ denote the significance at 5% and 10%, respectively
123
176
R. Bhattacharyya, T. Ghoshal
Table 11 Number of countries for which causality exists between CO2 emission and consumption Item
Variables at level
Growth of variables
Name of the variables
Within top 10
Within top 20
Consumption to CO2 Emission CO2 emission to consumption
Consumption to CO2 emission
CO2 Emission to consumption
CO2 emission
1 (3)
3 (3)
4 (4)
5 (5)
Per capita income
2 (2)
1 (1)
4 (5)
2 (2)
Population
1 (0)
3 (3)
1 (1)
4 (4)
CO2 emission
2 (1)
2 (2)
3 (3)
5 (5)
Per capita income
4 (2)
3 (3)
3 (4)
4 (4)
Population
0 (0)
1 (1)
3 (3)
4 (4)
Source: Compiled on the basis of results of causality analysis given in Table 9 Notes: (1) Significance of relationship is based on the existence of causality between variables CO2 emission and consumption both in aggregate and per capita level. (2) All variables have been considered in their natural logarithm. (3) Figures in brackets denote the number of countries for which the causality exists with the respective variables in per capita terms
6 Summary and conclusions The main conclusion of this article is that the link between growth rates of aggregate consumption and emissions is relatively stronger for developed countries. The cointegration relationships also point to the same conclusion as all 10 of the top 10 per capita income countries have a stable long-run relationship between emissions and economic developments at the level. This result is robust to per capita levels as well. Also, the countries for which there exists causality from consumption to CO2 emission (both in aggregate as well as per capita) are mostly high-income countries. There are at least two major implications of these conclusions. First, the results for the growth rates are different in spirit from the EKC literature in the sense that there is no evidence of a weakened link between economic growth and CO2 emissions as the level of income rises. Second, it strengthens the Kyoto protocol’s contention that developed countries should take proactive steps in reducing their own green house gas emissions rather than complaining about the leniencies shown to the less developed countries. Acknowledgements We are deeply indebted to Manas Ranjan Gupta, Saumen Sikdar, Sarmila Banerji and Dipankar Coondoo for their valuable comments in preparing this draft. Errors are ours.
References Aldy, J. E. (2004). An environmental Kuznets curve analysis of US state-level carbon dioxide emission. Harvard University: Depart of Economics (Mimeo). Banerjee, A., Dolado, J. J., Galbraith, J. W., & Hendry, D. F. (1993). Cointegration, error correction, and the econometric analysis of nonstationary data. Oxford: Oxford University Press. Dinda, S., Coondo, D., & Pal, M. (2000). Air quality and economic growth: An empirical study. Ecological Economics,34, 409–423. doi:10.1016/S0921-8009(00)00179-8. Granger, C. W. J., & Newbold, P. (1988). Some recent developments in a concept of causality. Journal of Econometrics, 39, 199–211.
123
Economic growth and CO2 emissions
177
Grossman, G. M., & Krueger, A. B. (1995). Economic growth and the environment. The Quarterly Journal of Economics,112, 353–378. doi:10.2307/2118443. Hamilton, J. D. (1994). Time series analysis. Princeton, New Jersey: Princeton University Press. Holtz-Eakin, D., & Selden, T. M. (1995). Stoking the fire? CO2 emission and economic growth. Journal of Public Economics,57, 85–101. doi:10.1016/0047-2727(94)01449-X. Hutington, H. G. (2004). U.S. carbon emissions, technological progress and economic growth since 1870. Energy Modeling Forum, Stanford University, EMF OP 53. Lanne, M. L., & Liski, M. (2003). Trends and breaks in per capita carbon dioxide emissions, 1870–2028. Finland: Department of Economics, University of Helsinki (Mimeo). Leib, C. M. (2004). The environmental Kuznets curve and flow verses stock pollution: The neglect of future damages. Environmental & Resource Economics,29(4), 483–506. doi:10.1007/s10640-004-1046-x. Oak Ridge National Laboratory. (2003). Estimates of CO2emissions from fossil fuel burning, cement manufacturing and gas flaring. Carbon Dioxide Information Analysis Centre. Panayotou, T. (1993). Empirical tests and policy analysis of environmental degradation at different stages of economic development. Discussion paper 1, International Labor Office, Geneva. Seldon, T. M., & Song, D. (1994). Environmental quality and development: Is there a Kuznet curve for air pollution emissions? Journal of Environmental Economics and Management,27, 147–162. doi: 10.1006/jeem.1994.1031. Shafik, N., & Bandyopadhyay, S. (1992). Economic growth and environmental quality: Time series and cross-country evidence, World Bank Background papers, Washington, DC. Summers, R., Heston, A., & Aten, B. (2002). Penn world table, version 6.1. Center for International Comparisons at University of Pennsylvania. Torras, M., & Boyce, J. K. (1998). Income inequality and pollution: A reassessment of the environmental Kuznet curve. Ecological Economics,25, 195–208. doi:10.1016/S0921-8009(97)00177-8. Vincent, J. R. (1997). Testing for environmental Kuznets curves within a developing country. Environment and Development Economics,2, 417–431. doi:10.1017/S1355770X97000223. World Bank. (2000). World development report, 1999–2000; emerging the 21stcentury. Oxford, New York: Oxford University Press.
123