BSc Thesis Economics and Business Economics
Economic growth and CO
2emissions
A review of different countriesWritten by: Davide van den Bergh 30 july 2016, Amsterdam
Supervised by: Ms Oana Furtuna MSc (Coordinator Dirk Damsma)
2
Table of contents
Abstract
3
1
Introduction
4
2
Literature Review
6
2.1
Support EKC
2.2
Critical statistics EKC
2.3
export and import effects
3
Methodology and data
11
3.1
Ordinary least squares model and assumptions
11
3.2
Problems
13
3.3
Control variables
3.4
Data
14
4
Results and analysis
17
4.1
4.2
5
Conclusion and discussion
19
Appendix
20
Bibliography
22
Statement of Originality
This document is written by Student Davide van den Bergh, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This study aims to research the effect of real GDP per capita on CO2 emissions per capita. It is argued that economic growth leads to environmental pollution in the first stages of growth, but that there is a turning point after which pollution starts to decrease combined with economic growth. At first we find evidence for several rich countries, but after introducing first difference regressions this
evidence disappears, accept for Finland in the period 1991-2011.
Keywords:
4
1
Introduction
The topic of interest evaluated in this thesis is: what is the effect of economic growth on CO2 emissions for different time periods?
The conclusion is that depending on the country and time period there is either a significant positive link or a significant negative link.
The majority of scientists agree that when CO2 emissions and other greenhouse gasses continue to increase the temperature on earth will rise. The exact benefits and problems of such a rise in
temperature are not known, but many scientists agree that it depends on the pace of the increase in temperature, which depends on many natural processes and greenhouse gas emissions
(Schmalensee, Stoker, Judson, 20?).
In an attempt to reduce CO2 emissions world leaders have come together recently in Paris at the Conference of the parties 21 in December 2015 to discuss, agree and find global measures to reduce emissions (conference of the parties 21, 2015). This CO2 reduction is important for our climate, health and global welfare (Aichele & Felbermayr, 2011).
The academic relevance of this paper is to see what the impact of economic growth is on CO2 emissions and we examine how this relation may change over time focusing on rich countries.
Also in the current economic crisis where we are waiting for economic growth and prosperity, it is interesting to ask whether economic growth and the CO2 emission reduction agreement in Paris can coincide. To be more precise, the central research question is : ‘ what is the impact of real GDP growth per capita on CO2 emissions per capita for different time periods and countries, using logarithmic regressions with data ranging from 1971 until 2011’?
Using historic data: regressions will be done for the complete time period 1971 – 2011, as well as for, 1971 – 1991 and 1991 – 2011. The two different periods 1971-1991 and 1991-2011 will be tested against each other. Real oil prices, energy consumption and the import to GDP ratio are used as a control variables, based on the assumption that rising oil prices may induce people to use more efficient products and sustainable energy, thereby reducing CO2 (Sadorsky, 2009). Graphs, tables and scatterplots with CO2 emissions per capita and real GDP per capita will supplement the findings.
The main contribution of this present thesis is that it draws on the most recent reliable data from the World bank, with a common methodology seen in previous papers to research the
5
A comment about previous research is that other papers have investigated a possible Environmental Kuznets curve, which hypothesizes that economic growth and environmental
pollution follow an inverted-U-shaped curve (Jaunky, 2011). This article uses CO2 per capita and real GDP per capita to assure a good comparison with the previous mentioned EKC relation, since the standard EKC regression model uses emissions per capita and real GDP per capita (Stern D. I., 2003).
This will be done for several high income countries, because according to the EKC theory, which will be explained in more detail in the literature review, high income countries may have encountered the turning point after which CO2 emissions per capita start to reduce and Jaunky found in (Jaunky, 2011) already an EKC for several countries. This article focuses on European countries with the addition of three big economies to be named: Australia, Canada and the United states, for which the EKC hypothesis has been tested before in (Jaunky, 2011) and who are rich in terms of GDP per capita according to the GDP per capita rankings in 2015 (GDP per Capita Ranking 2015, 2016). Also we look at Singapore, to compare the results with a rich outlier.
In the next section we describe the findings and main conclusions of past research in the area of economic growth and environment. Section III describes the econometric model used, the assumptions behind this model and the data used for the regressions. Section IIII presents the results and a discussion of our analysis and regressions. The final section describes our conclusion, what we learn from this thesis and my view on the topic.
6
2
Literature Review
This section outlines an overview and discussion of past research regarding environmental pollution and economic growth using journal articles from the year 1990 till the present.
The position of our paper relative to past work and our contributions will be stated. Much attention will be directed towards a possible EKC relation and the main critique on this same
relation. This review will set the beginning of our research and will help to understand, interpret and put the results found in the subsequent result section into context.
2.1 Evidence of an EKC
The Environmental Kuznets curve proposes that the relation between economic growth and environmental pollution follows an inverted U-shape function. This means that during the first stages of economic growth, environmental pollution increases, but after a turning point, pollution decreases (Kuznets, 1955)
(Grossman & Krueger, 1995) have examined the reduced form relationship between four environmental indicators and GDP per capita in their paper Economic growth and the Environment. The four indicators used are: urban air pollution, oxygen regime in river basins, fecal contamination of the river basins and heavy metal contamination of river basins.
An alternative for these reduced form equations would be to relate regulations, technology and industrial composition to GDP per capita and then to relate environmental pollution to
regulations, technology and industrial composition (Grossman & Krueger, 1995).
After (thorough) research their conclusion is that for various environmental indicators economic growth brings first a phase of deterioration followed by a phase of improvement. The turning point for the previously mentioned pollutants lies in most cases lies around a per capita income of 8000 dollar (in 1985 dollars). This article explores the relation between economic growth and the
environment and not specifically co2 emissions. Still it is interesting to see that an EKC relation exists for several pollutants. In following articles we will see if this EKC relation may also hold for CO2 emissions, which is the topic of interest in our research.
(Panayotou, 1997) explains that at higher levels of income the manufacturing industry of a country may shift to an information and service industry, combined with higher environmental expenditures, improved pollution reducing technology, better enforcement of environmental
7
regulations and more environmental awareness. These outcomes of high income may result in a reduction of CO2 emissions, leading to a negative relation between economic growth and emissions.
The EKC is explained by the following factors: the scale of production, industry composition with different pollution intensities, the input mix may change (using less damaging inputs) and finally improvements in the state of technology leading to less pollution due to production efficiency and emissions changes in the process leading to less pollution. Underlying factors for these variables include environmental awareness, regulation and education (Stern, 2003)
Early EKC studies (Grossman and Krueger (1991), (Shafik &
Bandyopadhyay, 1992)2), (Selden & Song, 1995) suggest that some local pollutants like SO2, fine smoke and suspended particles follow an inverted U shaped curve, whereas carbon dioxide (CO2) does not. According to economic theory this makes sense, because the local pollutants are internalized in the economic function
and result in environmental policies to reduce them, whereas global pollutants like CO2 may be internalized in a (later) stadium (Stern D. I., 2003). With the internalization of local pollutants, we mean that the cost these pollutants give rise to and the cost of removing them is accounted for in the profit function, so that firms optimize their production given they have to pay for these costs.
2.2 Critical towards the EKC
(Stern D. I., 2003) has researched the environmental Kuznets curve in his research paper: The Environmental Kuznets curve. The idea of this curve is that the relation between economic growth and environmental pollution follows an inverted U-shaped function. Meaning that during the first stages of economic growth, environment pollution increases, but after a turning point, pollution decreases. The general tendency of this paper is very critical towards the EKC curve. It refers to (Perman and Stern 2003) that the EKC does not exist when using econometrics to test which relationships are valid and which are correlations taking into account properties such as serial dependence, stochastic trends in time series and testing the adequacy of the models. To put it even stronger, the EKC has never been shown to apply to all pollutants. Contradictory (Stern, 2003) points
8
out that organizations like the World Commission on Environment and Development and the World bank have been promoting the EKC theme arguing that economic growth does not have to hurt the environment in the long term because of a higher demand for and investment in environmental quality as income grows, however Stern in (Stern, 2003) doubts this notion, because of theoretical and statistical critiques.
The article of Stern, 2003 highlights several critiques on theoretical and methodological grounds. The first being that the EKC assumes that environmental damages does not reduce economic activity sufficiently to hurt economic growth. Because if this assumption is not valid, growing too fast in the early stages of development might be unproductive for the long run, thus possibly it is more productive to have slower growth in the first stages of development when environmental degradation is high (Arrow, et al., 1995).
From this we infer it might be an advice, for developing economies, to procrastinate fast growth until sustainable technology and regulation is available.
A second critique is that according to the Hecksher-Ohlin trade theory, under free trade, a country specializes in the production of goods that uses a countries abundant factors intensively and exports these goods. Hence developing countries would specialize in the production of polluting goods, because these countries are abundant in natural resources and labor. Rich countries would specialize in service goods and capital goods because they are abundant in capital. This suggests that a possible EKC relation might be the result of a redistribution of polluting industries due to
international trade (Krugman, Obstfeld, & Melitz, 2011).
However according to (Antweiler, Copeland, & Taylor, 2001) surprisingly free trade seems to be beneficial for the environment.
Stern concluded in (Stern D. I., 2003), that the emissions and income relationship is probably monotonic positive but that the curve shifts down over time due to environmental regulation and more a more efficient use of resources.
It is argued that in the long run the easiest way to solve a (countries) environmental problems is to become rich. (Beckerman, 1992)
According to Rothman (Rothman, 1998), research suggest that the EKC may exist for several cases, but that the environmental measures for which the EKC holds focus on predictors that are local in nature for which abatement is relatively cheap in terms of monetary cost. Whereas the consumption based environmental measures like CO2 emissions, which are easy to externalize and expensive to abate, show no tendency to decline with increasing income.
9
(Rothman, 1998) argues that what may appear as improvements in environmental quality, actually signals the ability of consumers in wealthy nations to distance themselves from
environmental pollution through outsourcing of polluting production.
(Jaunky, 2011) explains that according to the EKC environmental protection will be installed automatically due to economic growth. Also the EKC theory implicitly assumes that causation runs from economic growth to
emissions and not the other way around. Because if causation would run from emissions to economic growth, it would make no sense to research the effect of economic growth on emissions. In that case one might be interested in improving growth using emissions, based on the
assumption that economic growth is wanted, given the general definition of economic growth. Three effects explain the EKC hypothesis and their names are the scale effect, the structure effect and the abatement effect. The scale effect of pollution is monotonically increasing in income and occurs at the start of industrialization ascribable to the setting up of polluting and unproductive factories. The expansion of energy consumption for production results in more waste and
environmental pollution. Secondly, the structure effect refers to an economy undergoing structural changes, for example shifting from a manufacturing based economy to a service based economy. This structure effects follows an inverted U-shaped curve. Finally, at some point countries can afford to use cleaner and cleaning technology, which make use of resources in a more efficient way and offset the environmental effects of certain activities. Also high income countries may ordain environmental laws to minimize pollution levels, this is called the abatement effect and is a monotonic decreasing function of income.
Jaunky, 2011) concluded that in the long run co2 emissions have fallen for Greece, Malta, Oman, Portugal, Ireland and the UK, while real GDP has been increasing. Also according to the VECM causality test, there is no causality running from CO2 to income in the long run. Therefore CO2 reducing measures can be installed without much concern for long run income. Their policy advice is to move from energy sources such as coil and oil to renewable energy such as solar energy and that the investment in and production of affordable renewable energy sources may further help to decline CO2 emissions, while at the same time spur growth.
10
So far we note that previous papers have reported an EKC for some countries (Jaunky) while other papers are very critical and argue that there is no EKC (Stern, 2003).
2.3 Domestic emissions versus carbon footprint
Another critical paper written by (Aichele and Felbermayr in 2011), makes a distinction between domestic emissions and the carbon footprint of a nation. They note that because of the trade in goods, there is a wedge between a country’s domestic emissions and a country’s carbon footprint. They continue to argue that the Kyoto agreement has reduced domestic CO2 emissions with 7%, but has not lowered the carbon footprint of nations, to the contrary, it has increased the share of imported over domestic emissions with 14%. The Kyoto protocol has led to a relocation of production. Resulting in the conclusion that the Kyoto protocol, due to its incomplete coverage, possibly has been harmful for the global climate. The agreement has enforced costs on companies and consumers, while not reducing global carbon emissions. The implication of this result is that future global climate deals have to incorporate everybody or should look at the carbon footprint of a nation instead of domestic emissions (Aichele & Felbermayr, 2011).
My opinion is that another way to reduce CO2 emissions is to focus on replacing polluting energy sources with sustainable alternatives that use less CO2 for energy production and to focus on the use of electric cars instead of gasoline cars. Using less polluting alternatives should lead to less emissions, this is similar to the policy implications in (Jaunky, 2011).
The position of my paper relative to past work is to test for individual countries with the most recent dataset available on Worldbank whether an EKC exists for some countries and whether affairs in term of CO2 reduction have become worse or improved compared to the regressions in (Jaunky, 2011).
The choice of countries is for the majority a composition of countries for who an EKC relation has been found in previous papers (jaunky, 2011), but especially Germany is added because of the Energiewende. It is interesting to see whether being a rich countries with a high GDP, results in the demand to reduce CO2 emissions. Likewise it is interesting to view, whether the countries where already an EKC relation has been found, have been able to continue or even improve this situation over the last years. In the result section we will see whether the countries studied are already rich enough to have reached the described turning point, because then we can find a negative relation between growth and emissions.
Energiewende (2011): is a reform and move to solar and wind energy to be able to shut down nuclear facilities in Germany
11
3
Methodology and data
In the first part of this section the econometric model including the various assumptions will be described and the choice of variables will be explained. The second section discusses the possible statistical problems and how we will account for these problems in our research. Also the
hypotheses and statistical tests will be stated. A separate section is dedicated to the control variables, which will be used next to real GDP per capita to explain CO2 emissions.
In the second part of this section the source of the data and the variables used will be described. Transformations to the data/variables are stated and scatter plots will be provided.
In this (research) paper we use ordinary least squares (log-log) regressions to find the income elasticity of CO2 emissions for different rich countries. The main reduced form equation estimated is the same as in (Jaunky, 2011):
LCO2t(1971-2011) = β0 + β1 LGDP t(1971-2001) + ε
3.1 Ordinary least squares estimator and variables
Our analysis begins with a single linear regression model, using the Ordinary least squares estimator, examining a possible relation between CO2 per capita and real GDP per capita including a constant. With CO2 per capita as the dependent variable and real GDP per capita as the independent variable.
The choice for these two variables was made because it seems interesting to me to look how economic growth may impact our climate and especially CO2 emissions. To define economic growth we use the growth in real GDP as dictated by standard economic theory, with the adjustment of using real GDP per capita, because that variable is used in papers exploring the relation between economic growth and the environment (Grossman & Krueger, 1995) (Jaunky, 2011). To view the impact of economic growth on our climate, we have chosen to look at CO2 emissions per capita.
Despite the fact that there are more greenhouse gasses such as methane, we have chosen for CO2 emissions since this is the most well-known greenhouse gas discussed in the news that influences the climate, it is used in previous papers that discuss the EKC relationship and reliable data for these emissions as well as for real GDP per capita is readily available in the world bank data set (Grossman & Krueger, 1995) (Jaunky, 2011) (Stern D. I., 2003) (Databank Microdata Datacatalog, 2016)
12
Using the log log regression has as an advantage compared to a normal regression that we can see the effect of economic growth on CO2 growth/decline percentage wise. Otherwise one would have to calculate this in an extra step. To explain CO2 emissions we could also look at technology and environmental regulation, but exact data for those variables are of questionable validity. Therefore we have chosen to use real GDP per capita. (Grossman & Krueger, 1995). Also by using real GDP, inflation is accounted for, and because of utilizing per capita data, population growth or decline is accounted for.
The Ordinary least squares estimator is based on certain assumptions, which are discussed below. These assumptions are also called the Gauss-Markov assumptions (Stock & Watson, 2012).
Gauss-Markov assumptions:
The first assumption is that the expected error term is zero, given the independent variables. If this is the case, other factors contained in the error term are unrelated to the independent variable. In other words: the mean of these other factors is zero.
µi is a random variable with E (εi│Xi) =0
The second assumption is a statement about how the sample is drawn. The sample should be drawn randomly, so that (Yi and Xi) are independently and identically distributed.
(Yi Xi) are independently and identically distributed (i.i.d.)
In our research we use data about variables that are observed over time. This is called time series data and violates the above assumption. Also the point with time series data is that a value observed in the present is probably correlated with the values from the previous period, which violates assumption five of no autocorrelation in the errors. The next section will explain how we will account for these problems related to time series data.
The third assumption is that ‘Large outliers are unlikely’, because large outliers may have a significant impact on the regression results, which might make them misleading. A scatter plot of the data reveals that there are a couple of outliers for some countries. An example is a peak in CO2 emissions in 1996 for Denmark, but according to (Greenhouse gas emission statistics, 2016) there was a peak in emissions in 1996 due to a cold winter, that increased heating demands. Overall we find no large outliers that need to be removed from the dataset.
A fourth assumption is homoscedasticity, this means that the variance of the error term given the independent variables is constant. Variance (µi│Xi) = σ2
Otherwise the error term is heteroskedastic. A plot of the data shows there may be some heteroskedastic, but the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity, as well as White’s test does not reject the H0 of constant variance, which leads us to believe that
13
The last and fifth assumption is that there should not be autocorrelation in the error term. Covariance (µi, µj) = 0 i ≠j The Durbin-Watson test rejects the hypothesis of no autocorrelation for our dataset. Thus this fifth assumption is violated, but when we do the regressions with the variables in first differences than the Durbin-Watson test does not reject the hypothesis of no autocorrelation.
If assumptions 1,2 and 3 are valid then the OLS estimator of the betas (β0 β1 … βn) are unbiased and consistent
If assumptions 1,2,3,4 and 5 are valid then the OLS estimator is efficient within the class of unbiased and linear estimators. An efficient estimator is the most precise estimator, in other words, it is the estimator with the smallest variance.
The OLS estimator presents us with several regression statistics to be named: The coefficients (betas)
t-statistics p-values F-statistics R squared
In our case, the coefficients of real GDP, oil prices, energy consumption and import to GDP ratio tell us the percentage change in CO2 emissions when the corresponding variable increases with one percent. It is also important to look at the t-statistics and p-values of every coefficient, because these tell us if the coefficient is statistically significant or not. When using a multiple regression model, the F-test can be used to test whether two or more variables together have coefficients that statistically significant explain the dependent variable. Finally we also look at the R squared to view how much of the variance in CO2 emissions is explained by the independent variables in the model.
My hypothesis is that there is a negative relation between economic growth and CO2 emissions for the rich Northern countries such as Norway. Because sustainable energy in the Netherlands is imported from Norway (Relations the Netherlands-Norway, 2016). In Germany there has been die energiewende starting in 2011, to replace coal mines and nuclear reactors with green energy. Unfortunately our dataset for CO2 emissions runs until 2011, so the effect of the
14
3.2 Time series, omitted variable bias, multicollinearity and causality
This section explains how we account for the fact that we are using time series data and discusses the problems associated with a multiple regression model, such as omitted variable bias and multicollinearity.
In this research to be able to use time series data we declare the data to be time series data using the tsset command. Time series data probably have a unit root, this means that today’s value of a variable is explained by the previous value plus an error term Yt = Yt-1 + ε . To test if our variables have an unit root, we have used the augmented Dickey-Fuller test, and found that all variables have an unit root. To account for this we perform the regressions using the variables in first differences, which means that the difference between the current and the previous value is used for the regression instead of the current value. Δ Yt = Yt - Yt-1 After performing the regressions in first differences we tested again for first order auto-correlation using the Durbin-Watson test and found no evidence of first order auto-correlation for all countries studied.
Two possible problems are omitted variable bias and multicollinearity. Omitted variable bias occurs when a variable, that is correlated with one of the independent variables and an explanatory variable of the dependent variable is left out. This results in a positive or negative bias of the independent variable that is correlated with the omitted variable. To account for this problem we have searched in previous literature for other variables that are important to consider when exploring the growth pollution link. As a result we found that it might be useful to include an antecedent average of real GDP per capita, real oil prices, energy use per capita and the import to GDP ratio. These control variables are more elaborately discussed in the section control variables.
There can be perfect multicollinearity and imperfect multicollinearity. Perfect
multicollinearity means that one of the variables is perfectly correlated with the other variables, in this case we cannot run the OLS regression and have to adapt the regression to make it work again. Imperfect multicollinearity means that two or more of the independent variables are highly
correlated, this does not mean that we have to change the regression, but we should note that it may result in less precisely estimated coefficients (larger sampling variance) (Stock & Watson, 2012).
Viewing the correlations table of our data in the appendix we see correlations between 0.69 and 0.87 for real GDP per capita and oil prices and. ‘This may be the result of the use of oil in the production of goods, which might have increased real GDP’ (Zou & Chau, 2006).
Another issue is the question of causality. (Perman & Stern, 2003) tested for granger
causality between income and CO2 emissions and found that causality runs from income to emissions in developing countries, but from emissions to income in developed countries. This would mean that
15
our regressions should be the other way around or that we might have a simultaneity problem. However according to a more recent paper (Jaunky, 20011): ‘Unidirectional causality, running from GDP to CO2 emissions, is detected in both the short and the long run, since this is also the interest of our research we choose to follow the methodology of (Jaunky, 2011).
3.3 Control variables
To incorporate the effect of previous income on the environment we have chosen to include a control variable of the antecedent three year average of real GDP per capita, as done in (Grossman & Krueger, 1995). When this three year average resulted in such an amount of collinearity that the variable was omitted, we changed this variable into a four or five year average.
A second control variable is real oil prices, because in (Sadorsky, 2009) is mentioned that rising oil prices may promote a switch to renewable energy usage, a reduction in consumption and the acquisition of more efficient products. Therefore they use real oil prices in their model to estimate renewable energy usage. The previous mentioned effects of changing real oil prices also impact CO2 emissions, as these emissions are especially emitted when fossil fuels are used as an energy source for production (Sadorsky, 2009). For energy consumption we expect the coefficient to be positive, since our CO2 emissions are for the majority based on the burning of fossil fuels for energy consumption (Schmalensee, Stoker, & Judson, 1998). ‘’ A hypothesis in the debate about the environment is that polluting goods are imported from abroad instead of produced domestically ‘’. Therefore we expect the coefficient of the import to GDP ratio to be negative (Friedl & Getzner, 2003).
3.4 Data
This article uses the most recent data from the World bank data set for specific countries. CO2 emissions per capita are those stemming from the burning of fossil fuels and the production of cement, they include carbon dioxide produced during consumption of solid, liquid and gas fuels. Real GDP per capita is computed using 2010 US dollars as a constant. Oil prices are inflation adjusted oil prices (World Bank) (Historical Crude Oil Prices, 2016).
For Germany we have CO2 data starting from 1991, this is due to west and east Germany, for all the other countries we use data from 1971 until 2011. The variables CO2 and real GDP are declared time variables and transformed into natural logarithmic variables (ln) to allow an income elasticity of CO2 comparison.
16
Since this article uses real GDP instead of nominal GDP, the elasticity between GDP and CO2 becomes stronger, because assuming positive inflation nominal GDP grows faster than real GDP. The difference in growth rates is the amount of inflation. 𝐷𝑒𝑙𝑡𝑎 𝑟𝑒𝑎𝑙 𝐺𝐷𝑃𝐷𝑒𝑙𝑡𝑎 𝐶𝑂2 If Delta real GDP is smaller than nominal GDP, the income elasticity of CO2 emissions becomes larger than when nominal GDP was used.
Before we continue with our regressions we first present several scatter plots. A scatter plot of logarithmic GDP and CO2 for Germany suggests a negative relation between the two variables, but when providing a scatter plot using the variables in first differences, this negative relation disappears, as shown in the appendix.
A scatter plot of natural logarithmic CO2 per capita and real GDP per capita in Germany for the period 1991-2011 before using first differences.
-. 2 -. 1 0 .1 C O 2 p e r C a p it a -.04 -.02 0 .02 .04
Real GDP per Capita
Own calculation Stata (first difference)
Figure 3: Scatter Plot Netherlands 1971-2011
2 .2 2 .3 2 .4 2 .5 C O 2 p e r C a p it a 10.4 10.5 10.6 10.7
Real GDP per Capita
Own calculation stata natural logarithms
17 -. 1 -. 0 5 0 .0 5 C O 2 -.1 -.05 0 .05 Energy consumption Own calculation stata (first differences)
18
4
Results and analysis
The main equation to be estimated to test a possible EKC relation is: LCO2 = β0 + β1 LGDP + ε
Where β0 represents a constant, β1 is the coefficient of interest and ε is the error term. A coefficient of -0.5 would mean that for a one percentage increase in GDP, a CO2 reduction of 0.5 percent would be achieved. This is done for fourteen countries for the time periods 1971-2011, 1971-1991 and 1991-2011. A schedule with the varying results before doing first differenced regressions is presented below:
Table 1
Income elasticities of CO2 emissions per capita
Countries 1971-2011 1971-1990 1991-2011
LGDP LGDP LGDP
Coefficient t-statistics Coefficient t-st. Coefficient t-statistics
Australia Canada Denmark -0.37 -4.72 -0.31 -2.21 -1.06 -4.17 Finland 0.17 3.22 0.12 0.86 018 1.56 Germany - - -0.80 -9.27 Greece 1.26 10.70 2.50 9.11 0.44 7.43 Ireland 0.26 10.41 0.33 4.67 0.14 1.99 Italy 0.35 10.95 0.28 5.78 0.39 2.10 Netherlands -0.13 -2.69 -0.39 -1.56 -0.13 -3.16 Norway 0.25 4.54 0.40 5.02 1.03 7.31 Singapore -0.32 -2.77 0.33 3.88 -2.42 -9.31 Spain 0.61 11.26 0.63 3.68 0.70 4.21 Sweden -1.00 -11.70 -1.67 -9.61 -0.30 -3.32 U.S. 41 observations
19
The positive significant betas range from 0.17 for Finland to 2.5 for Greece and the negative significant betas range from -0.13 for the Netherlands to -2.42 for Singapore.
A statistically significant negative relation between CO2 and real GDP in the long run is found for Denmark, Netherlands, Singapore and Sweden. In the most recent time period from 1991 till 2011 a statistically significant negative relation is estimated for Denmark, Germany, Netherlands, Singapore and Sweden.
To give an example of how the t-statistic, the F statistic and the R2 behave for different regressions we use Singapore as an example. The standard regression results in a t-value of -2.77, an F-value of 7.66 and a R-squared of 0.1643. When we include inflation adjusted oil prices we note that all values improve, suggesting a better model. Also oil prices have a significant negative effect on CO2 emissions, the coefficient for oil prices is -0.0081. Finally a split of the time periods around 1991 results in remarkably higher values for the t and F test, for the period after 1990 the R-squared is estimated to be 0.82. All results and summary statistics can be found in the appendix. Also
interesting to note is that (Jaunky, 2011) found a negative relation between real GDP and CO2 emissions for Greece in the long run, whereas we estimate a positive relation.
But now we perform the regressions again using the variables in first differences to account for the unit root problem, which was indicated by the dicky fuller test. Except for energy consumption in Denmark and the Netherlands where no unit root problem was found.
A schedule with the varying results after doing first differenced regressions shows a
completely different picture compared with before: The significant negative coefficients of Denmark, Germany, Singapore and Sweden have disappeared. Instead now only Finland has a statistically significant negative coefficients for real GDP per capita in the period 1991-2011. Greece has a significant negative coefficient for average income and the hypothesised negative relation between imports to GDP and CO2 emissions hold for Greece as well. For Spain this import relation is the opposite of expected, also we find a negative relation for real GDP per capita, oil and average real GDP per capita, but only the last one is significant at the 5% level for Spain. Finally, energy
consumption is a significant factor explaining CO2 emissions for all countries, this effect on CO2 emissions is mostly positive, except for the Netherlands and Sweden.
20 Table 2 first difference regressions with control variables Income elasticities of CO2 emissions per capita
Countries 1991-2011 1991-2011 1991-2011 R2
LGDP LOil LEnergy LImport Average GDP
Coefficient Coefficient Coefficient
Australia 0.027 -0.00031 0.00017** -0.068 -0.014 0.78 Canada 0.426 -0.016 0.847** -0.164 -0.086 0.74 Denmark -1.596 0.0001 1.436** 0.0046 -0.028 0.37 Finland -0.521** -0.003 2.128** 0.017 0.93 Germany 0.350 0.0025 0.827** -0.040 0.045 0.88 Greece -0.211 -0.0004 0.410** 0.0027 -0.323** 0.69 Ireland 0.070 -0.0001 1.102** -0.094 -0.00075 0.94 Italy 0.775** 0.000 0.762** -0.048 0.032 0.95 Netherlands -0.523 0.0085 -0.0003** 0.142 0.192** 0.58 Norway 1.240 -0.098 1.141** -0.242 0.245 0.84 Spain -1.304 -0.0628 0.682** 0.571** -0.569** 0.79 Sweden -0.361 0.0284 1.264** 0.0855 0.035 0.72 U.S. 0.138 -0.0002 1.018** 0.0015 -0.003 0.94
Note: ** when significant at 5% level 41 observations
21
5
Conclusion
Conclusion
A lot of research has been done regarding economic growth and the environment. Several papers argue there is an EKC relation, while other papers are very critical about the statistical truth/validity of this relation. Even when the income elasticity of CO2 emissions is negative for individual
countries, this might be the result of the shifting of polluting industries to other countries and a higher import of CO2 emissions relative to domestic emissions. According to our research at first the EKC hypothesis does seem to hold for several rich countries. To be named Denmark, Germany, Netherlands and Singapore. But after the use of our variables in first differences and the inclusion of average real GDP, real oil prices, Energy consumption and the import to GDP ratio as control
variables. We solely find a statistically significant negative relation between real GDP per capita and CO2 emissions per capita for Finland in the period 1991-2011.
Discussion
Limitations are the number of observations, because we are working with annual data we have a limited number of observations (around 41).
For the countries where we do not find a negative relationship this might mean that the turning point is not reached yet, it might be the case that there is no turning point solely based on economic growth, or there might be a measurement error present in our research.
A suggestion for further research is to inspect the results before and after shocks in GDP per capita CO2 emissions per capita, to inspect other control variables more thoroughly that influence CO2 emissions, such as deviations from the long term mean temperature and dummy variables for the signing of climate deals. And to simply use more recent data when this is available. Especially it would be interesting to view how our regressions would change in the forthcoming years after the climate top in Paris.
22
Appendix
Scatter plots of first differenced real GDP per capita and CO2 emissions per capita for all countries studied in the period 1971-2011.
Australia Canada Denmark -. 1 -. 05 0 .0 5 .1 D .lco 2a us -.04 -.02 0 .02 .04 D.lgdpaus -.1 -.0 5 0 .0 5 D .lco 2ca n -.05 0 .05 D.lgdpcan -. 2 -. 1 0 .1 .2 D .lco 2d en -.05 0 .05 D.lgdpden
23
Finland
Germany
Greece
-. 2 -. 1 0 .1 .2 D .l co 2 f -.1 -.05 0 .05 .1 D.lgdpf -. 1 -. 0 5 0 .0 5 D .lco 2 g -.06 -.04 -.02 0 .02 .04 D.lgdpg -. 1 -. 05 0 .0 5 .1 .1 5 D .lco 2 gr -.1 -.05 0 .05 .1 D.lgdpgr24
Ireland
Italy
Netherlands
-. 1 -. 05 0 .0 5 .1 .1 5 D .lco 2i r -.1 -.05 0 .05 .1 D.lgdpir -. 1 -. 05 0 .0 5 .1 D .lco 2i t -.05 0 .05 .1 D.lgdpit -.2 -.1 0 .1 D .lco 2n e -.04 -.02 0 .02 .04 D.lgdpne25
Norway
Spain
Singapore (not first differenced)
-.3 -.2 -.1 0 .1 D .lco 2n o -.04 -.02 0 .02 .04 .06 D.lgdpno -.6 -.4 -.2 0 .2 .4 D .lco 2s -.05 0 .05 .1 .15 D.lgdps 1 1. 5 2 2. 5 3 lco 2s 9 9.5 10 10.5 11 lgdps
26
Singapore
Sweden
United States
-.2 -.1 0 .1 .2 D .lco 2sp -.04 -.02 0 .02 .04 .06 D.lgdpsp -.2 -.1 0 .1 .2 D .lco 2sw -.05 0 .05 D.lgdpsw -. 1 -. 05 0 .0 5 D .lco 2u s -.04 -.02 0 .02 .04 .06 D.lgdpus27 Correlations gdpsw -0.9442 0.9531 -0.8886 0.9775 -0.8154 0.9585 0.3000 co2sw 0.6295 -0.6221 0.6754 -0.6414 0.6696 -0.5441 0.2770 gdpsp -0.8982 0.8889 -0.8548 0.9344 -0.7664 0.9824 0.3594 co2sp -0.4208 0.3636 -0.3285 0.4369 -0.3053 0.7109 0.5030 gdpno -0.8483 0.9012 -0.8816 0.9202 -0.7101 0.9914 0.3786 co2no -0.8797 0.8402 -0.8549 0.8617 -0.7676 0.8158 0.1946 gdpne -0.9017 0.9298 -0.9077 0.9730 -0.7984 0.9756 0.2753 co2ne 0.4365 -0.5102 0.6177 -0.5900 0.6893 -0.5511 0.2610 gdpit -0.7837 0.7940 -0.7419 0.8565 -0.6387 0.9773 0.4572 co2it 0.0083 -0.1039 0.1761 -0.0334 0.1605 0.2961 0.5142 gdpir -0.8776 0.8876 -0.8579 0.9263 -0.7493 0.9891 0.3842 co2ir -0.0053 -0.0383 0.0059 0.0708 0.0752 0.3960 0.4954 gdpgr -0.8964 0.8541 -0.8056 0.8776 -0.7156 0.9319 0.3791 co2gr -0.6138 0.5711 -0.5772 0.6392 -0.4822 0.8595 0.4554 gdpf -0.9214 0.9288 -0.8747 0.9682 -0.7934 0.9747 0.3202 co2f -0.2143 0.2427 -0.0969 0.2041 0.1815 0.3908 1.0000 gdpden -0.8442 0.8901 -0.8457 0.9218 -0.7012 1.0000 co2den 0.8265 -0.7900 0.8471 -0.8353 1.0000 gdpg -0.9202 0.9613 -0.9051 1.0000 co2g 0.8632 -0.9314 1.0000 gdps -0.9150 1.0000 co2s 1.0000 co2s gdps co2g gdpg co2den gdpden co2f
gdpsw -0.5595 0.9816 0.8741 0.9521 0.5681 0.9775 -0.6230 co2sw 0.6826 -0.6388 -0.7048 -0.5590 -0.1290 -0.5871 1.0000 gdpsp -0.5493 0.9821 0.8646 0.9766 0.6930 1.0000 co2sp -0.3336 0.5781 0.3630 0.6620 1.0000 gdpno -0.5271 0.9787 0.8432 1.0000 co2no -0.4319 0.8904 1.0000 gdpne -0.5902 1.0000 co2ne 1.0000 co2ne gdpne co2no gdpno co2sp gdpsp co2sw gdpsw 0.9942 0.7427 0.9438 0.1743 0.9698 0.1127 0.9102 co2sw -0.6185 -0.3832 -0.6311 0.1910 -0.5474 0.2738 -0.4292 gdpsp 0.9905 0.8595 0.9665 0.3535 0.9954 0.2603 0.9640 co2sp 0.6190 0.9139 0.6803 0.8112 0.7082 0.8522 0.8167 gdpno 0.9694 0.8459 0.9263 0.3695 0.9853 0.2358 0.9574 co2no 0.8741 0.6250 0.8551 0.1125 0.8404 -0.0681 0.7417 gdpne 0.9901 0.7792 0.9270 0.2582 0.9799 0.1190 0.9304 co2ne -0.5601 -0.4117 -0.4716 -0.0488 -0.5544 0.0434 -0.4983 gdpit 0.9381 0.9157 0.9047 0.5480 0.9741 0.4595 1.0000 co2it 0.1732 0.6382 0.2616 0.8590 0.2837 1.0000 gdpir 0.9838 0.8631 0.9453 0.3879 1.0000 co2ir 0.2526 0.7060 0.2820 1.0000 gdpgr 0.9553 0.8537 1.0000 co2gr 0.7962 1.0000 gdpf 1.0000 gdpf co2gr gdpgr co2ir gdpir co2it gdpit
28 Summary statistics gdpsw 41 33081.16 7147.897 22863.82 46036.79 co2sw 41 7.165678 1.839237 4.704369 10.74016 gdpsp 41 19626.48 4982.353 11944.15 27661.02 co2sp 41 5.984428 1.074897 3.763558 8.097058 gdpno 41 48694.06 13946.96 25549.16 69094.74 co2no 41 8.44602 1.126292 6.918398 11.61619 gdpne 41 32352.55 7486.296 21710.52 45043.25 co2ne 41 10.88574 .8415204 9.384833 13.37862 gdpit 41 25539.19 5463.194 15359.56 32830.73 co2it 41 7.107515 .6563498 5.762318 8.216487 gdpir 41 28850.08 14409.57 12314.05 53918.14 co2ir 41 8.753438 1.350506 6.885428 11.38728 gdpgr 41 17204.61 3441.086 11708.26 24470.27 co2gr 41 6.766975 1.691306 3.149596 8.98084 gdpf 41 28393.09 7797.263 16139.15 42415.09 co2f 41 10.71166 1.093063 8.562246 13.2611 gdpden 41 38233.94 8036.614 25720.91 50695.08 co2den 41 10.59141 1.33216 7.248329 13.71465 oil 41 49.03488 24.11217 17.26 107.36 gdps 41 18262.34 9158.46 5344.379 36154.04 co2s 41 11.39503 3.918064 2.655247 19.11902 Variable Obs Mean Std. Dev. Min Max
29
Bibliografie
(2015). Opgeroepen op juni 10, 2016, van COP21: cop21.gouv.fr
Aichele, R., & Felbermayr, G. (2011). Kyoto and the carbon footprint of nations. Journal of
Environmental Economics and Management, 336-354.
Antweiler, W., Copeland, B., & Taylor, M. (2001). Is free trade good for the environment? American
Economic Review.
Arrow, K., Bolin, B., Costanza, R., Dasgupta, P., Folke, C., Holling, C., et al. (1995). Economic growth, carrying capacity, and the environment. Science, 520-521.
Beckerman. (1992). Economic growth and the environment: whose growth? whose environment?
World Development, 481-496.
Databank Microdata Datacatalog. (2016, July). Opgeroepen op 2016, van The World Bank:
data.worldbank.org
Felder, S., & Rutherford, T. (1992). Unilateral CO2 Reductions and Carbon Leakage: The
Consequences of International Trade in Oil and Basic Materials. Journal of Environmental
Economics and Management, 162-176.
Friedl, L., & Getzner, M. (2003). Determinants of CO 2 emissions in a small open economy. Ecological
economics, 133-148.
GDP per Capita Ranking 2015. (2016, July). Opgeroepen op July 2016, van Knoema:
https://knoema.com/sijweyg/gdp-per-capita-ranking-2015-data-and-charts
Greenhouse gas emission statistics. (2016, June). Opgeroepen op July 2016, van Eurostat statistics
explained:
http://ec.europa.eu/eurostat/statistics-explained/index.php/Greenhouse_gas_emission_statistics
Grossman, G. M., & Krueger, A. B. (1995). ECONOMIC GROWTH AND THE ENVIRONMENT. The
Quarterly Journal of Economics, 353-377.
Historical Crude Oil Prices. (2016, June). Opgeroepen op June 2016, van InflationData.com:
inflationdata.com/Inflation/Inflation_Rate/Historical_Oil_Prices_Table.asp
Jaunky, V. C. (2011). The CO2 emissions-income nexus: Evidence from rich countries. Energy Policy, 1228-1240.
Krugman, P., Obstfeld, M., & Melitz, J. (2011). International Economics. Pearson International Edition.
Kuznets, S. (1955). ECONOMIC GROWTH AND INCOME INEQUALITY. The american economic review, 1-28.
Narayan, P., & Narayan, S. (2010). Carbon dioxide and economic growth: panel data evidence from developing countries. Energy Policy, 661-666.
30
Panayotou. (1997). Demystifying the environmental Kuznets curve: turning a black box into a policy tool. Environment and Development Economics, 465-484.
Perman, R., & Stern, D. (2003). Evidence from panel unit root and cointegration tests that the environmental Kuznets curve does not exist. Australian Journal of Agricultural and Resource
Economics.
Relations the Netherlands-Norway. (2016, July). Opgeroepen op July 2016, van Government of the
Netherlands: https://www.government.nl/topics/international-relations/contents/norway Rothman, D. S. (1998). Environmental Kuznets curves real progress or passing the buck A case for
consumption-based approaches. Ecological Economics, 177-194.
Sadorsky, P. (2009). Renewable energy consumption, CO2 emissions and oil prices in the G7 countries. Energy Economics, 456-462.
Schmalensee, R., Stoker, T., & Judson, R. (1998). WORLD CARBON DIOXIDE EMISSIONS: 1950–2050.
MIT press journal.
Selden , T., & Song, D. (1995). Neoclassical growth, the J curve for abatement and the inverted U curve for pollution. Journal of Environmental Economics and Environmental Management, 162-168.
Shafik, N., & Bandyopadhyay, S. (1992). Economic Growth and Environmental: Time Series and
Cross-Country Evidence. Washington DC: The World Bank.
Stern, D. I. (2003). The Environmental Kuznets Curve. International Society for Ecological Economics. Stern, N. (2008). The Economics of Climate Change. The American Economic Review, 1-37.
Stock, J., & Watson, M. (2012). Introduction to Econometrics. Harlow: Pearson Education. Zou, G., & Chau, K. (2006). Short- and long-run effects between oil consumption. Energy Policy,
3644-3655.
