The Fight Against Greenhouse Gas Emissions:
Is the Current Global Policy Effective?
MSc in Econometrics
Maarten Peters (11045744)
December 30, 2016
Thesis supervisor: M.J.G. Bun Second reader: J.C.M. van Ophem
Faculty of Economics and Business
Amsterdam School of Economics
Requirements thesis MSc in Econometrics.
1. The thesis should have the nature of a scienti c paper. Consequently the thesis is divided up into a number of sections and contains references. An outline can be something like (this is an example for an empirical thesis, for a theoretical thesis have a look at a relevant paper from the literature):
(a) Front page (requirements see below)
(b) Statement of originality (compulsary, separate page) (c) Introduction (d) Theoretical background (e) Model (f) Data (g) Empirical Analysis (h) Conclusions
(i) References (compulsary)
If preferred you can change the number and order of the sections (but the order you use should be logical) and the heading of the sections. You have a free choice how to list your references but be consistent. References in the text should contain the names of the authors and the year of publication. E.g. Heckman and McFadden (2013). In the case of three or more authors: list all names and year of publication in case of the rst reference and use the rst name and et al and year of publication for the other references. Provide page numbers.
2. As a guideline, the thesis usually contains 25-40 pages using a normal page format. All that actually matters is that your supervisor agrees with your thesis.
3. The front page should contain:
(a) The logo of the UvA, a reference to the Amsterdam School of Economics and the Faculty as in the heading of this document. This combination is provided on Blackboard (in MSc Econometrics Theses & Presentations).
(b) The title of the thesis
(c) Your name and student number (d) Date of submission nal version
(e) MSc in Econometrics
Statement of Originality
This document is written by Student Maarten Peters who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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.
Abstract
This paper uses a panel dataset consisting of the 100 most polluting countries that are part of the United Nations Framework Convention on Climate Change (UNFCCC) to determine the effect of two major global treaties on the reduction of greenhouse gases, namely the forming of the UNFCCC and the ratification of the well known Kyoto Protocol. Assuming these policy decisions are exoge-nous, the results of a panel fixed effect model suggest membership of the UNFCCC has decreased emissions by about 2,5%. When making the distinction between Annex I (developed) countries and non-Annex I (developing) countries, the results imply an adverse effect on emissions for Annex I countries. In specific, membership of the UNFCCC in a certain year increases the emission of
CO2 per capita by 2% for these countries. However, for non-Annex I countries, results indicate the
amount of CO2 emitted per capita decreases by 3,5% if a country is a member of the UNFCCC in
a certain year. For the Kyoto Protocol in general, results show the influence of ratifying this treaty on emissions is not statistically significant. Relaxing the assumption of exogenous policy decisions whilst exploiting an instrumental variables approach, the results suggest neither policy decision has a statistically significant impact on emissions. Concluding, the Kyoto protocol has not proven to be effective in mitigating global emissions, and the impact of the forming of the UNFCCC on emissions remains debatable.
Contents
1 Introduction 9
2 Policy 11
2.1 UNFCCC . . . 11
2.2 Kyoto protocol . . . 12
2.3 The effect of policy . . . 13
3 Literature review 14 3.1 Environmental Kuznets curve . . . 14
3.2 IPAT . . . 16
3.3 Dependent variable . . . 18
3.4 Control variables . . . 19
3.5 Technology . . . 19
3.6 The effect of international trade . . . 20
3.7 Population and population density . . . 21
3.8 U.S. rejection of the Kyoto protocol . . . 21
3.9 Developing countries . . . 22 4 Data description 24 4.1 Sources . . . 24 4.2 Descriptive statistics . . . 25 4.3 Development aid . . . 29 5 Methodology 31 5.1 Baseline model . . . 31 5.2 Panel data . . . 32 5.3 Estimation . . . 33 6 Results 34 6.1 Baseline model . . . 34 6.2 Extensions . . . 37 6.3 Estimation in totals . . . 37 6.4 Static model . . . 38
6.5 Heterogeneity . . . 39
6.6 Endogeneity of policy . . . 40
6.7 Unit roots and co-integration . . . 41
6.8 The non-stationary case . . . 42
6.9 EDGAR data . . . 44
7 Discussion 45 7.1 Conclusion . . . 45
7.2 Limitations and recommendations . . . 46
8 References 48 9 Appendix 54 9.1 Unit root and co-integration tests . . . 54
1
Introduction
Nowadays climate change and pollution are very relevant topics, frequently addressed by activist groups, governments and international lobbies. Global warming is already known to be the main reason for the rise in sea level caused by the warming of oceans and the melting of Arctic ice caps. More specifically, the global average temperature has increased by more than 1,5 degrees Centigrade since the late 1800s. Some regions of the world have even warmed by more than twice this amount. The main reason for this change in temperature has been found to be the build-up of greenhouse gas in the atmosphere of our planet earth.
With a rise in average temperature, greenhouse gas concentration and higher ocean levels, our ecosystem has received a huge shock. The question arises whether the balance in many delicate ecosystems that exist around the world can survive such a shock. Recent studies, such as Hoegh-Guldberg and Bruno (2010) indicate that rapidly rising greenhouse gas concentrations are driving ocean systems towards conditions not seen for millions of years, with an associated risk of funda-mental and irreversible ecological transformation. Further change will continue to create enormous challenges and costs for societies worldwide, particularly those in developing countries. The oceans are, however, not the only part of our ecosystem that are under pressure from global warming. According to Bonan (2008) many forests are also under tremendous pressure from global change.
Another implication of the rising ocean temperatures that could pose a real threat to the balance of our ecosystem is the desertification of very arid areas of our planet. Many studies have shown that the rise in atmosphere and ocean temperatures is causing increasing drought in arid countries around the equator. This desertification can have a direct impact on the food supply in many developing countries that are not adequately equipped to deal with rapid changes in their ecosystem. Other studies, such as Webster (2005) and Knutson (2010), indicate that a rise in ocean temperatures increases the number of tropical storms in addition to their intensity.
Even though, these days, climate change is acknowledged to be a major threat to the preservation of our planet, the climate debate concerning greenhouse warming did not arise as a political issue until the 1990s. Following the publication of the report ”Our Common Future” in 1987 by the World Commission on Environment and Development (1987) the issue was finally addressed internationally. In the decades following this report, the world has seen many changes regarding greenhouse gas emission policy.
of the United Nations Framework Convention on Climate Change (UNFCCC) (United Nations, 1992). The parties to this convention have met annually from 1995 to assess progress in dealing with climate change. During one of these conventions in 1997, the well known Kyoto Protocol was adopted and entered into force in 2005 (United Nations, 1997). As the first legally binding international policy on sustainability, it forced many governments to implement their own policy on greenhouse gas emissions. For instance in 1997 the European Commission adopted the policy ”Energy for the Future: Renewable Sources of Energy” (European Commission, 1997), in which all participating countries agreed to increase the share of renewable energy of the European Unions gross inland energy consumption to 12% by 2010.
The question now arises, whether major policy changes over the past decades have effectively reduced greenhouse gas emissions. Data on yearly emissions of several greenhouse gases, and data on other indicators such as the production and consumption of electricity using different sources have been collected to provide an answer to this question. The goal of this study will be to determine the effect that the two major policy changes, the forming of the UNFCCC and the Kyoto Protocol,
have on carbon dioxide (CO2) emissions.
Using several control variables commonly used in the literature, such as GDP per capita and merchandising imports and exports, this paper will further extend on the literature concerning effects of UNFCCC policy on greenhouse gas emissions by Grunewald and Martinez-Zarzoso (2009) and Kumazawa and Callaghan (2012), by investigating a longer time span and considering various extensions to the models they propose.
This paper is structured as follows. Firstly, the two major policy decisions that are of interest in this paper will be discussed in Section 2. In Section 3 an overview of the literature on modeling environmental degradation with respect to policy and other factors will be given. This Section will asses what methods and which indicators have been used to measure climate change and which control variables have been used to model the behavior of greenhouse gas emissions. In Section 4 the data and its sources will be presented, and the specific characteristics present in the data will be discussed and interpreted. In Section 5 the baseline model will be presented and the corresponding econometric techniques will be discussed. In Section 6 the results for the baseline model will be discussed. In Section 7 several extensions to the baseline model will be discussed as robustness checks. Finally in Section 8, this study will attempt to answer the research question. Also some limitations of this study will be discussed and recommendations for future research will be made.
2
Policy
The two main policy changes to be considered in this paper are the forming of the UNFCCC in 1992 and the signing and ratification of the Kyoto Protocol negotiated in 1997. This section will provide a short review of these two important historical events.
2.1 UNFCCC
Negotiations on what became the UNFCCC were launched in December 1990 by the UN General Assembly. An Intergovernmental Negotiating Committee was convened to conduct these negoti-ations, which were concluded in just 15 months. The Convention was adopted on 9 May 1992, and opened for signature a month later at the UN Conference on Environment and Development in Rio de Janeiro, Brazil. It entered into force on 21 March 1994, after receiving the requisite 50 ratifications. The treaty commits signatories’ governments to reduce atmospheric concentrations of greenhouse gases with the goal of stabilizing greenhouse gas concentrations in the atmosphere at a level that will prevent dangerous human interference with the Earth’s climate system. Currently 197 countries are a member of the UNFCCC, therefore being a globally supported organization. The parties to the UNFCCC are classified as follows:
• Annex I: There are 43 Parties to the UNFCCC listed in Annex I of the Convention, including the European Union. These Parties are classified as industrialized (developed) countries and ”economies in transition”.
• Annex II: Of the Parties listed in Annex I of the Convention, 24 are also listed in Annex II of the Convention, including the European Union. These Parties are made up of members of the Organization for Economic Cooperation and Development (OECD). Annex II Parties are required to provide financial and technical support to the economies in transition and developing countries to assist them in reducing their greenhouse gas emissions (climate change mitigation) and manage the impacts of climate change.
• Annex B: Parties listed in Annex B of the Kyoto Protocol are Annex I Parties with first- or second-round Kyoto greenhouse gas emissions targets. The first-round targets apply over the years 2008-2012. As part of the 2012 Doha climate change talks, an amendment to Annex B was agreed upon containing with a list of Annex I Parties who have second-round Kyoto targets, which apply from 2013-2020. The amendments have not entered into force.
• Least-developed countries (LDCs): 49 Parties are LDCs, and are given special status under the treaty in view of their limited capacity to adapt to the effects of climate change.
• Non-Annex I: Parties to the UNFCCC not listed in Annex I of the Convention are mostly low-income developing countries. Developing countries may volunteer to become Annex I countries when they are sufficiently developed.
2.2 Kyoto protocol
Since the adoption of the convention, parties have continued to negotiate in order to agree on decisions and conclusions that will advance its implementation. Over the years, many so called
climate summits have been held. The first being the 1992 establishment of the party in Rio,
followed by the 1995 summit in Berlin and also the famous 1997 convention in Kyoto, where the Kyoto Protocol was negotiated.
The Kyoto Protocol has had two commitment periods, the first of which lasted from 2008-2012. The second one runs from 2013-2020 and is based on the Doha Amendment to the Protocol, which has not entered into force. The US has not ratified the Kyoto Protocol, while Canada denounced it in 2012. All other Annex I parties have currently ratified the protocol.
Under the Protocol, countries’ actual emissions have to be monitored and precise records have to be kept of the trades carried out. Registry systems track and record transactions by Parties under the mechanisms. The UN Climate Change Secretariat, based in Bonn, Germany, keeps an international transaction log to verify that transactions are consistent with the rules of the Protocol. Reporting is done by Parties by submitting annual emission inventories and national reports under the Protocol at regular intervals. A compliance system ensures that Parties are meeting their commitments and helps them to meet their commitments if they have problems doing so.
The extent to which developing country Parties will effectively implement their commitments under the Convention will depend on the effective implementation by developed country Parties of their commitments under the Convention, related to financial resources and transfer of technology, and will take fully into account that economic and social development and poverty eradication are the first and overriding priorities of the developing country Parties.
2.3 The effect of policy
Over the years, partially induced by the UNFCCC and Kyoto protocol, many countries have adapted different policies which focus on developing renewable energy sources, as well as restricting highly polluting industry sectors, in an attempt to reduce their greenhouse gas emissions. Narayan, Smyth, and Prasad (2007), for instance, find that due to price elasticity of electricity demand there is po-tential to use pricing policies in the G7 countries to curtail residential electricity demand, and thus curb carbon emissions in the long run. The effect of these policies however, still remains ques-tionable. Grubb (1995) emphasizes that neglect of the issue of induced technical change and other adaptive responses may invalidate the policy implications drawn from most integrated assessment models developed to date. This was pointed out again by Harmelink, Voogt, and Cremer (2006), who argued additional policies were needed to achieve the goals set by the European Commission (1997) for their targets in 2010. Furthermore, Gan, Eskeland, and Kolshus (2007) suggested that the applied range of policy instruments may be lacking in providing incentives for the long term development of new technologies concerning green electricity in Europe and the US.
Although evidence suggests some policies do not achieve the intended effect, many countries have seen an increase in transparency concerning pollution information reported by firms. Freedman and Jaggi (2005) for instance, show firms from countries that ratified the Protocol have higher disclosure indexes as compared to firms in other countries. Additionally, larger firms disclose more detailed information on pollution.
It has become increasingly important to keep measuring the impact of the introduction of policies in order to determine the effectiveness of these decisions, making this study a relevant addition to the literature on greenhouse gas emissions.
3
Literature review
This section will attempt to provide a comprehensive overview of the literature on the topic of human impact on environmental degradation, in specific the emission of carbon dioxide. The question is
how one can model the emissions of CO2 in order to isolate and quantify the effects certain policy
changes have on these emissions. This research makes use of a panel data approach to do so. In the following paragraphs, this paper will show why a panel data approach is appropriate for this analysis, and which control variables are often used in the literature.
3.1 Environmental Kuznets curve
There is a long history of studies that have attempted to determine human impact on environmental degradation, in specific the emissions of different greenhouse gases. Kuznets (1955) was the first to study the relation between economic growth and environmental degradation. In his study he found, in the early stages of economic growth, environmental degradation and pollution increase. Yet beyond some level of income per capita the trend reverses, so that at high-income levels economic growth leads to environmental improvement. This implies that the environmental impact indicator is an inverted U-shaped function of income per capita. This hypothesized relationship is also called the Environmental Kuznets Curve (EKC) named after Kuznets himself.
Nowadays, all econometric studies analyzing human impact on the environment make use of an EKC specification (Cole, Rayner, and Bates (1997); (Holtz-Eakin and Selden, 1995); Suri and Chapman (1998); (Moomaw and Unruh, 1998), Cole and Neumayer (2004); Galeotti, Lanza, and Pauli (2006); Grunewald and Martinez-Zarzoso (2009); Kumazawa and Callaghan (2012)). Some studies such as Stern (2004) agree that in the specification of the EKC model, due to omitted variable bias, there is endogeneity present. In short this implies that statistical inferences may be erroneous if, in addition to the observed variables under study, there exist other relevant variables that are unobserved, but correlated with the observed variables. Using a panel fixed effects model, one can control for these unobserved variables through country and year specific effects.
Indeed, in the previously mentioned literature on environmental impact, panel data is often used for this reason. As an example, Cole et al. (1997) examine the relationship between per capita income and a wide range of environmental indicators using cross-country panel data sets. They argue that meaningful EKCs exist only for local air pollutants whilst indicators with a more global, or indirect, impact either increase monotonically with income or else have predicted turning points
at high per capita income levels with large standard errors, unless they have been subjected to a multilateral policy initiative. Although the Kyoto protocol was adopted in 1997, Cole et al. (1997) still argue that without more multilateral environmental policy, greenhouse gases were unlikely to be controlled in the short term. Consequently, this study will evaluate whether international policy has proven to be sufficient or not.
Similarly, Suri and Chapman (1998) attempt to econometrically quantify the effect of the EKC hypothesis with respect to commercial energy consumption using pooled cross-country and time-series data. They find that both industrializing and industrialized countries have added to their energy requirements by exporting manufactured goods. In this sense, industrialized countries can avoid investing in polluting industries through imports of manufactured goods, which partially explains the turning point in the EKC. This will further be discussed in Section 3.6.
Many studies such as Stern, Common, and Barbier (1996), Harbaugh, Levinson, and Wilson (2002) and Stern (2004) also make use of panel data to argue that the statistical analysis on which the environmental Kuznets curve is based is not robust. They argue that the EKC is dependent on a model of the economy in which there is no feedback from the quality of the environment to production possibilities, and in which trade has a neutral effect on environmental degradation. The actual violation of these assumptions gives rise to fundamental problems in estimating the parameters of an EKC. On the other hand, Cole et al. (1997) suggest that a meaningful EKC does exist for local air pollutants. In general, it can be deduced from the literature on the EKC that there exists some kind of U-shaped relation between environmental degeneration and GDP, although the specific shape and turning points of this curve remains debatable. This study will consider a simple squared relationship for the EKC hypothesis, and give some recommendations for alternative specifications in Section 7.2.
It is important when modeling emissions, to try and capture as much of the unobserved het-erogeneity as possible using several different control variables proposed in the literature. However, Harbaugh et al. (2002) find the specification of and inverted U-shaped EKC differs significantly when including different control variables, which will further be discussed in Section 3.4. Galeotti et al. (2006) also find there is reason to believe that the specification of the EKC is quite different for certain more homogeneous countries. They show there is evidence of an inverted-U pattern for the group of Organization for Economic Co-operation and Development (OECD) countries, with reasonable turning point, regardless of the data set employed. However, for non-OECD countries this is not the case, as the EKC is basically increasing (slowly concave) according to data from the
International Energy Agency and more bell-shaped in the case of data from the Carbon Dioxide Information Analysis Center. One could argue that for most non-OECD countries GDP per capita has not yet achieved the level needed for an inverted-U pattern to be present. Consequently, the segmentation of countries to obtain more homogeneous groups will further be discussed in Section 6.5 as a robustness check.
3.2 IPAT
In the literature, several attempts have been made to analyze the driving forces of greenhouse gas emissions generated from human activities. One of the most groundbreaking studies was conducted by Ehrlich and Holdren (1971). They suggest that environmental impact (I ) is caused by popu-lation (P ), affluence (A) measured in GDP (per capita), and technology (T ) often measured by energy intensity or industrial activity (Cole and Neumayer, 2004). Ehrlich and Holdren (1971) then suggest that the relationship takes on the form I = P AT . An advantage of this model is the simple multiplicative specification of the three hypothesized driving forces causing environmental degradation.
Dietz and Rosa (1997) reformulated the IPAT model into a stochastic equation, which can then be used in order to estimate this relation in a panel setting. Their model is called the Stochastic Impacts by Regression on Population, Affluence, and Technology (STRIPAT) model. This model will be used as a framework to further build on in this research. The model is specified as
Iit= αPitβAγitTitδeit. (1)
Such as for the IPAT model, the STRIPAT model suggests that environmental impact (I) is caused by population (P ), affluence (A), which can be measured in GDP (per capita), and technology (T ) which can be measured by variables such as energy intensity or industrial activity. Notice that there is no specification of the time dimension in this model, as this is a specification that only considers the effects of IPAT for a fixed year. In this case however, one should acknowledge the cross-sectional and time-series nature of the data. In a panel setting, following the examples of Cole and Neumayer (2004), Grunewald and Martinez-Zarzoso (2009) and Kumazawa and Callaghan (2012), one can therefore take the natural logarithm of (1), and rewrite it as
ln(Iit) = ln(α) + βln(Pit) + γln(Ait) + δln(Tit) + it. (2)
In a panel setting one should add country specific effects to capture country specific (time invariant) determinants of I other than P , A and T in addition to an error term, both of which will further
be discussed in Section 5. Important examples for such determinants are climatic differences and geographical factors (Neumayer, 2002). Since taking a logarithm is a monotonic transformation, it is possible to make use of the linear model (2) to estimate the parameters in (1) by making use of techniques suitable for panel data.
Notice this is a static model, which assumes only variables from the current time dimension influence the emissions in the current year. There is growing evidence in the literature showing the pollution-income relationship is a dynamic one, taking Agras and Chapman (1999), Egli and Steger (2007) and Grunewald and Martinez-Zarzoso (2009) as some examples.
For instance, if a highly polluting factory is operative in a certain year, it often is still in use the next year. The same holds for many other polluting factors, such as cars, transportation trucks and airplanes. Therefore it is reasonable to believe the emissions in the previous year are directly related to the emissions in the next year. Following the example of Grunewald and Martinez-Zarzoso (2009) this paper will extend (2) to a dynamic model such as
ln(Iit) = ln(α) + µln(Ii,t−1) + βln(Pit) + γln(Ait) + δln(Tit) + it. (3)
In addition, the specification of a dynamic model can also be a solution if one suspects there is serial correlation in the errors of the static model. For the baseline model in Section 5, this paper will extend (3) by adding several control variables and dummies for the UNFCCC and Kyoto protocol. However, in Section 6.4 this study will also consider a static version of the STRIPAT model as a robustness check.
A number of studies have made use of these models in order to measure the impact of human
activity on CO2 emissions over the past decade. For instance, York, Rosa, and Dietz (2003) extend
the STRIPAT model by introducing the concept of ecological elasticity and make use of geological
variables to examine the effects on CO2 emissions and energy footprint. Geological factors have
been omitted from this study, but might prove insightful for further research. Cole and Neumayer
(2004) investigate the impact of demographic factors on CO2 and sulfur dioxide emissions (SO2).
They add urbanization, age groups and household size as demographic factors and find different
results of their impacts on CO2 and SO2 emissions. For CO2 the elasticity of emissions with respect
to population appears to be unity over all population sizes, supporting the view of this research,
whereas for SO2 the population-emissions elasticity seems to be negative for very small population
sizes, but to rise rapidly as population increases.
de-veloped and developing countries to test the theory of the EKC in the context of environmental regulations using a static panel and even extend this model to a dynamic panel model. They find
the Kyoto obligations have a reducing effect on CO2 emissions in developed and developing
coun-tries. Finally, Kumazawa and Callaghan (2012) investigate the impact of the Kyoto Protocol on
world emissions of CO2 using a large unbalanced panel, and find structural breaks in the data that
demonstrate the effects of the Kyoto Protocol. For these studies, data was only available up to 2006, and therefore this study will provide new insights beyond this point.
3.3 Dependent variable
In the literature there are two views on how to model the dependent variable for environmental
impact, in this study taken to be CO2 emissions. The first view takes the total amount of CO2 to
be the dependent variable I in the IPAT or STRIPAT model. The second attempts to model CO2
emissions per capita as I.
The first is more descriptive in nature, as is argued by Shi (2003), and usually attributes
varia-tions in CO2 emissions to changes in population, affluence, and energy intensity (Bongaarts (1992);
Holdren (1991); MacKellar, Lutz, Prinz, and Goujon (1995)). For example, Engelman (1994) plotted long-term trends in global carbon dioxide emissions and population, and found that since 1970 both emissions and population have grown at similar rates, leading him to hypothesize that population growth has been a major force in driving up global emissions over recent decades.
According to Shi (2003), the second category of work has adopted a more empirical approach (Cole et al. (1997); Holtz-Eakin and Selden (1995); Moomaw and Unruh (1998)). These studies
typically focus on the link between economic growth and CO2 emissions, regressing CO2 emissions
per capita on GDP per capita and other predictors. In these studies, population is not treated as a predictor in the model; rather, it is included in the dependent variable, i.e. per capita emissions. These models thus presume that the elasticity of emissions with respect to population change is unitary (Dietz and Rosa (1994);Cole and Neumayer (2004)). This will further be discussed in Section 3.7.
This study will adopt the view of Dietz and Rosa (1994) and Cole and Neumayer (2004), and
take the dependent variable to be CO2 emissions per capita. However, in Section 6.3 the dependent
3.4 Control variables
In order to reduce the effect of omitted variable bias, additional control variables have to be included
in the specification of a model that describes CO2 emissions. Naturally the production of electricity
by fossil fuels has a direct impact on the emission of CO2. Therefore it seems only logical to include
variables for these factors. These variables will be discussed in further detail in Section 4.
There has been a lot of research over the past years on which alternative energy sources can effectively oppress greenhouse gas emissions. Some noteworthy are studies conducted by Menyah and Wolde-Rufael (2010) and Apergis and Payne (2010), who both suggest that nuclear energy
consumption can help to mitigate CO2 emissions, but so far, renewable energy consumption has
not reached a level where it can make a significant contribution to emissions reduction. However, in their research, Marques, Fuinhas, and Manso (2010) conclude the objective of reducing energy dependency appears to stimulate renewable energy use. They also suggest both the lobby of the
traditional energy sources (oil, coal, and natural gas) and CO2 emissions restrain renewable
deploy-ment. For the more developed countries, Sadorsky (2009) argues increases in real GDP per capita
and CO2per capita are found to be major drivers behind per capita renewable energy consumption.
Consequently, renewable energy can help mitigate emissions for the more developed countries in the future. Considering these facts, the production of nuclear and renewable energy can prove to be important control variables.
3.5 Technology
In the IPAT model, technology (T ) is one of the driving forces behind human impact on environ-mental degeneration. Most studies make use of the size of the industry in a specific country to measure the level of technology. Dietz and Rosa (1997) included T in the error term and did not separately estimate the influence of technology on emissions, whereas York et al. (2003) extended the model and introduced T as another explanatory variable.
The production of many different goods, such as food, electronics and household items, will often leave a carbon footprint. In addition, the transportation of these goods also add to the carbon footprint. For example, in two of his papers ((Cole et al., 1997); (Cole, 2003)), Cole uses trade intensity, level of technology and the capital-labor ratio as additional covariates to control for the impact of technology on GHG emissions when examining the EKC.
In addition, Cole and Neumayer (2004) argue energy intensity provides a measure of ”energy productivity” and as such should be directly related to the level and types of technology currently
in place within a country. Similarly, they argue industrial activity or manufacturing share provides a measure of the industrial structure of an economy, an obvious determinant of impact per unit of production. Other aspects of ”technology” not captured by energy intensity and the manufacturing share will finally be picked up by the error term. Since the collected data on energy intensity is quite sparse, it has not been included in this research as a measure of technology. Due to lack of data on other variables, this paper will follow the example of Cole and Neumayer (2004) and use the value added to industrial production as a percentage of the GDP.
3.6 The effect of international trade
As previously discussed in Section 3.1, the flow of pollution through international trade flows has the ability to undermine environmental policies, particularly for global pollutants. For many countries there is a difference between production-based and consumption-based carbon emissions. Specifi-cally, Annex I countries are often net importers and non-Annex I countries are often net exporters
of CO2 emissions. Peters and Hertwich (2008) argue emissions embodied in trade may have a
sig-nificant impact on participation in and effectiveness of global climate policies such as the Kyoto Protocol. They conclude that a better understanding of the role that trade plays in a country’s economic and environmental development will help design more effective and participatory climate post-Kyoto. Davis and Caldeira (2010) also note that emissions associated with the consumption of goods and services differ from traditional, production-based inventories because of imports and
exports of goods and services that, either directly or indirectly, involve CO2 emissions. They show
23% of global CO2 emissions, or 6.2 gigatonnes CO2, were traded internationally in 2004, primarily
as exports from China and other emerging markets to consumers in developed countries.
Similarly, Stretesky and Lynch (2009) make use of panel data to examine the relationship be-tween per capita carbon dioxide emissions and merchandising exports for 169 countries. They argue consumption in the United States (U.S.) is partially responsible for elevated per capita carbon diox-ide emissions in other nations, and carbon dioxdiox-ide trends in other nations are in part driven by U.S. demands for goods. In addition, this also indicates that inefficient production methods among countries that export products to the U.S. may signal a problematic trend in global carbon dioxide emissions.
These findings also affect the theory behind the environmental Kuznets curve. Some stud-ies claim the U-shaped relation between environmental degeneration and GDP is determined by changing trade patterns rather than growth-induced pollution abatement, and these trade patterns
have typically been neglected by EKC studies. However, Cole (2003) argues the inclusion of ad-ditional covariates has little impact on the income-emissions relationship, and that the impact of trade patterns on pollution emissions is small. Considering all these facts, merchandising imports and exports will be used as additional variables to control for the effect of international trade.
3.7 Population and population density
As has been discussed in Section 3.3, there are two approaches to modeling emissions. When taking the total amount of emissions as the dependent variable, the population of a country can function as the variable for P in the IPAT model. Countries with a large population will obviously emit more greenhouse gases in total than low population countries.
When modeling the emissions per capita, the elasticity of emissions with respect to population change is assumed to be unitary (Dietz and Rosa (1994); Cole and Neumayer (2004)). Therefore the variable P is included in the dependent and control variables. However, Shi (2003) analyzes
the effects of population growth on CO2 emissions for 93 countries, and demonstrates that global
population changes over the last two decades are more than proportionally associated with growth in carbon dioxide emissions, and that the impact of population change on emissions is much more pronounced in developing countries than in developed countries.
To control for the disproportional impact of population change on emissions, this study will make use of another factor closely linked to population, namely population density. This control variable has frequently been used in the literature on the EKC (Grossman and Krueger (1995);
Hilton and Levinson (1995)). Population dense countries often have a much higher per capita CO2
emission, since big cities tend to pollute more than several small cities or villages. Also the transport of goods, for instance food and household items, from farms and factories contribute way more to
the emission of CO2 in population dense areas.
Therefore when considering the total emission CO2 emissions per country, total population has
to be included as a control variable in the model. Similarly when considering the per capita emission of greenhouse gases, population density can be included as a control variable in the model.
3.8 U.S. rejection of the Kyoto protocol
Although the U.S. signed the Kyoto Protocol in 1988, the treaty has never been ratified by the Senate. Since the U.S. accounted for 36% of emissions in 1990, without U.S. ratification only a coalition of the European Union, Russia, Japan and another small party could place the treaty into
legal effect, which was done in 2001. According to Manne and Richels (2004) this is likely to affect compliance costs for the remaining non-Annex I countries during the first commitment period up to 2012, perhaps rendering the membership of the UNFCCC and the Kyoto protocol less effective. In the case of other OECD countries, compliance costs may decline, but perhaps not as much as some have suggested. In the case of the economies in transition, compliance costs are likely to increase. In addition, Canada has withdrawn their support of the Kyoto protocol in 2011. Since the time dimension of the data used in this study only ranges to 2011, this will not have an impact on the results. Nevertheless, since the U.S. already accounts for a large share of the total emissions in the world, the exclusion of the U.S. from the model will also be considered as a robustness check in Section 6.2.
3.9 Developing countries
Many developing countries still strongly rely on polluting sources for their energy supply and do not yet have the means to reduce emissions effectively. Therefore one of the goals of the Kyoto protocol is to encourage the richer and developed countries (Annex I) to provide the necessary financial and technological support to the developing countries that have signed the treaty (non-Annex I). Babiker, Reilly, and Jacoby (2000) have investigated the effects of trade on greenhouse gas emissions of developing countries. They find adverse effects fall mainly on energy-exporting countries, for some even greater than on countries that are assuming commitments. Removing existing fuel taxes and subsidies and using international permit trading would greatly reduce the adverse impacts and also reduce economic impacts on the countries taking on commitments. Moreover, this study provides more evidence that there are significant differences between these two groups of countries, supporting the views of Galeotti et al. (2006), Grunewald and Martinez-Zarzoso (2009) and Kumazawa and Callaghan (2012) that segmenting Annex I and non-Annex I countries will lead to more homogeneous groups of countries. In addition, Shi (2003) shows there is a significant difference in the elasticity of emissions with respect to population change between these segments of countries, as discussed in Section 3.7. Therefore, this study will consider the segmentation of Annex I and non-Annex I countries as a robustness check in Section 6.5.
Since the introduction of the UNFCCC all Annex I countries have taken on an obligation to provide the non-Annex I countries with financial and technological support for the development of sustainable energy. Indeed, Michaelowa and Michaelowa (2007) analyze the interaction of climate and development policy that has taken place since the early 1990s. They conclude that, while
there are valid reasons for long-term collaboration with emerging economies on greenhouse gas mitigation, there should be a separate budget line for such activities to avoid ”obfuscation” of a decline of resources aimed at poverty alleviation. Nevertheless, mitigation will remain attractive for donors because it ensures quick disbursements and relatively simple measures of success. Moreover, mitigation activities in developing countries provide politicians in industrialized countries with a welcome strategy to divert the attention of their constituencies from the lack of success in reducing greenhouse gas emissions domestically. Consequently, this study will consider using the amount of development aid destined for the development of renewable energy as a control variable in Section 4.3.
4
Data description
The data collected for this research consists of a panel of the 197 countries that were a member of the UNFCCC as of December 2015. The time dimension of the panel ranges from 1960 up to 2011. Below the different variables that have been used and the sources of the data will be presented.
4.1 Sources
Country specific data on several variables listed below have been taken from the website of the
World Data Bank1 (WDB). These variables include emissions for all types of greenhouse gases,
namely Carbon Dioxide (CO2), Methane gas (CH4), Nitrogen Oxide (NO2) and other greenhouse
gases (HFC, PFC and SF6) in metric tons (Mt). However, for other gases than CO2, the data is
very sparse and therefore not usable for this study. Real GDP and GDP per capita in $, total
population and population density (people per sq. km of land area) have also been included.
Similarly, the percentage of total electricity production from six different sources, namely Coal, Oil, Gas, Nuclear, Hydroelectric and other renewable sources, have been included. Data on the total production (in kWh) is not available from the WDB. Data on imports and exports of merchandise, including electronics and fuels have been obtained. Lastly, value added as a percentage of GDP to the industry of a country has been taken from the WDB as an indicator for technology (T ).
It is interesting to note however, that the CO2 emission data as can be obtained from the WDB
is slightly different to the CO2 emission data from Emissions Database for Global Atmospheric
Research2 (EDGAR). In the sensitivity analysis, this data then will be used to check whether there
is a significant difference in results when using an alternative source for the data. Note that the WDB has emissions data from 1960 until 2011 available, whereas EDGAR has data from 1970 to 2014 available.
As has been discussed in Section 2, the participating countries of the UNFCCC are divided into three categories, Annex I, Annex II and non-Annex I parties. The Annex I & II parties have higher emission targets and requirements, the Annex II parties are obliged to provide financial resources for the non-Annex I parties, which are classified as developing countries. A list of all countries that are members of UNFCCC, including membership date, date of ratification and whether they are
Annex I or Annex II can be found on the website of the UNFCCC3. Indicator variables are added
1
http://data.worldbank.org
2
http://edgar.jrc.ec.europa.eu/
to each country from the year they have ratified the membership of the UNFCCC up to the present day. This has also been done for the ratification of the Kyoto Protocol. These indicators can then be used to evaluate whether these two important global policy agreements have indeed led to a decrease in greenhouse gas emissions.
The amount of development aid destined for the development of renewable or cleaner conven-tional electricity generation received by all the non-Annex I countries can also prove to be very
insightful. This data can be found on the website of the OECD4. Additionally, by looking at the
effect that this aid has had on the mitigation of greenhouse gas emissions, the effectiveness of de-velopment aid can be investigated. However, this data is only available from the year 1995 and onwards. Moreover, the data is very sparse for the developing countries, and therefore it is not possible to include this variable into the model. However, this study will briefly discuss the trend of development aid in this Section, and compare it to the number of countries that have signed the Kyoto Protocol.
4.2 Descriptive statistics
Different fuels emit different amounts of carbon dioxide in relation to the energy they produce when
burned. In Table 1 the emissions across fuels are compared by the amount of CO2 emitted per unit
of energy output or heat content, in order to show which fuels are more polluting5. It is obvious
that of these fossil fuels coal is the most polluting source, followed by different types of oil and lastly natural gas. It is therefore important to make the distinction between the different non renewable sources of electricity production if they are included as control variables in the model.
Table 1: Pounds of CO2 emitted per million British thermal units (Btu) of energy.
Coal (anthracite) 228.6 Diesel fuel and heating oil 161.3
Coal (bituminous) 205.7 Gasoline 157.2
Coal (lignite) 215.4 Propane 139.0
Coal (sub-bituminous) 214.3 Natural gas 117.0
A source of energy being renewable does not mean it produces little or no greenhouse gas emissions. It can be toxic, hazardous, or environmentally catastrophic. A renewable source is basically something that can’t be used up because it can be recreated somewhat quickly. However,
4
http://www.oecd.org/development/stats/idsonline.htm
not all renewable sources are classified as entirely emissions neutral, since some sources require energy-dependent water to cool them or to run steam turbines. While we need water to generate electricity, we also use electricity to generate water. In figure 1, all different sources energy are classified in order to get an idea how sustainable each source actually is.
Figure 1: How the different energy sources can be classified.
A huge drawback of the panel at hand is the sparsity of the data set. Data on CO2 emissions
only reaches from the year 1960 to 2011 for most countries. However, for some countries there is fewer data available on emissions. A noteworthy example is Germany, for which there is only data from 1990 to 2011, and Serbia for which there is only data from 2006 to 2011. Moreover, it turns out that for many less developed countries there is very few data at hand concerning other variables such as electricity production, GDP and population density. On the one hand considering the effect most of these countries have on global emissions and on the other hand considering the sparsity of the data, the decision has been made to only include the 100 most polluting countries in the panel.
These 100 countries account for 99,25% of the total CO2 emissions over the entire time span of this
panel (from 1960 up to 2015). Similarly, only considering the year 2011, it turns out that these 100 countries still account for 99,12% of all emissions. Therefore it seems appropriate to only use this selection of countries.
Finally, what remains is a panel consisting of 100 countries with a time dimension ranging from 1960 to 2011. In Table 2, averages of all relevant variables for the year 2011 are shown for the different segments that have been used. Note that in the remainder of this paper, pc will stand for per capita.
Country segment Countries CO2 (Mt) CO2 pc GDP ($) GDP pc Industry (%GDP) Density
Annex I & II 37 364,962 8.32 1.27e12 37,310 27.37 116
non-Annex I 63 295,742 6.39 3.96e11 11,091 36.34 384
Total 100 321,353 7.10 7.31e11 21,092 32.77 284
Country segment Population % Coal % Oil % Gas % Hydro % Nuclear % Renewables
Annex I & II 3.48e7 23.37 1.94 26.83 18.58 17.18 8.56
non-Annex I 7.81e7 18.99 16.90 38.53 21.33 0.90 1.89
Total 6.21e8 20.61 11.37 34.20 20.31 6.92 4.36
Table 2: Averages for different segments in 2011
As we are dealing with several different time series within this panel, this subsection will investi-gate these series separately for characteristics such as trends and similarities between the dependent variable and the control variables. In Section 6.7 the possibility of unit roots and co-integration in the data will also be discussed.
Firstly, the distribution of CO2 emissions in total and per capita, renewable energy production
as a percentage of total energy production, and GDP per capita will be presented in figures to get a visual overview of which regions are more polluting. These graphical representations of the data
can also give us a more intuitive insight in the situation at hand. In figure 2 the total CO2emissions
and the percentage of total energy production by renewable sources are depicted on the world map.
(a) Total CO2 emissions. (b) % of total energy production by renewables.
Figure 2: Source: Data world bank
Notice that countries which are highly polluting in general seem to have a lower percentage of total energy production by renewables than those who pollute less. Intuitively this makes sense, since
a great deal of the greenhouse gas emissions in many countries stem directly from the generation of electricity. In 2015 for example, in the US about 37% of all greenhouse gas emissions came from
the production of electricity6. This is also one of the main reasons to make use of control variables
for the production of electricity through non-renewable sources. Next consider CO2 emission per
capita and GDP per capita, which are depicted on the world map in figure 3.
(a) CO2 per capita emissions. (b) GDP per capita.
Figure 3: Source: Data world bank
It is immediately clear the per capita emissions and the per capita GDP are related. This concedes with the literature on the relation between emissions and GDP, such as the EKC discussed in Section 3.1 and the IPAT model discussed in Section 3.2.
To investigate trends and other specific behavior over time, this subsection will also display cross sectional averages of the most important series. In figure 4 the cross sectional average of
CO2 emissions per capita are displayed for three different groups: all countries, only the Annex I
countries and only the non-Annex I countries.
3 4 5 6 7 8 9 10 11 1960 1980 2000 2020
Cross sectional mean all countries
CO2 per capita
Year Graphs by country_id 3 4 5 6 7 8 9 10 11 1960 1980 2000 2020
Cross sectional mean Annex I
CO2 per capita
Year Graphs by country_id 3 4 5 6 7 8 9 10 11 1960 1980 2000 2020
Cross sectional mean non-Annex
CO2 per capita
Year
Graphs by country_id
Figure 4: Cross sectional averages for CO2 emissions per capita.
Interestingly, it can be seen from the graphs that for the more developed countries, the emissions per capita are slowly decreasing after 1980, although they are still much higher than those of the lesser developed countries, for which the emissions per capita are still steadily increasing. In total it looks like the emissions per capita are stabilizing. Next the GDP per capita for the three different segments is shown in figure 5.
0 10000 20000 30000 40000 1960 1980 2000 2020
Cross sectional mean all countries
GDP per capita Year Graphs by country_id 0 10000 20000 30000 40000 1960 1980 2000 2020
Cross sectional mean Annex I
GDP per capita Year Graphs by country_id 0 10000 20000 30000 40000 1960 1980 2000 2020
Cross sectional mean non-Annex I
GDP per capita
Year
Graphs by country_id
Figure 5: Cross sectional averages for GDP per capita.
Notice for both segments there seems to be a very similar upward trend in GDP per capita, although the level in both segments is quite different. Obviously the level of GDP per capita is higher in the more developed countries, as can be seen from both graphs. On the first sight the emissions per capita and GDP per capita series do not seem very alike, which implies there are
factors influencing CO2 emissions other than GDP. Subsequently, the percentage of total electricity
production from all possible sources is depicted below. Again this is done for the three segments previously discussed.
4.3 Development aid
As mentioned before, it is interesting to investigate the effect of wealth transfers from the Annex I countries to the non-Annex I countries on carbon dioxide emissions. However, data on these transfers are very limited. Including these variables will only make the data more sparse than it already is, and therefore will not be included in the model. In the figure 6, the number of countries that have ratified the membership of the UNFCCC and the Kyoto Protocol are depicted. Also the total amount of development aid paid and received for the development of sustainable energy is depicted. There seems to be quite a discrepancy between the amount paid and received in each year. This is mostly due to bad registration and lack of data. However, it can be seen that there is a steady increase in the amounts paid and received as the amount of countries that have ratified these treaties increases.
(a) Number of countries and ratification
(b) Total development aid for renewable energy
5
Methodology
In this section the econometric framework used to answer the research question will be discussed: have the major policy changes, namely the forming of the UNFCCC and the ratification of the Kyoto protocol effectively reduced greenhouse gas emissions?
5.1 Baseline model
This paper will reformulate (3) to obtain the baseline model for this research. This extended version of the dynamic STRIPAT model is given by
ln(Iit) = α + βln(Ii,t−1) + γ1ln(Ait) + γ2ln(Ait)2+ ψUit+ κKit+ δ0ln(Zit) + ηit, (4)
with i the country index and t the year index. In this specification, notice the composite error term
is given by ηit= αi+ ωt+ it, where αi are the time invariant country specific fixed effects, ωt are
country invariant year fixed effects and it is a time varying idiosyncratic error term. The other
variables in the model are listed below.
Iit (CO2 pc) Environmental impact (I) measured in CO2 emissions per capita
Ait (GDP pc) Affluence (A) measured in gross domestic product (GDP) per capita in $
Uit (UNFCCC) Dummy if country i has (1) or has not (0) ratified membership to the UNFCCC in year t
Kit (Kyoto) Dummy if country i has (1) or has not (0) ratified the Kyoto protocol in year t
Zit Vector containing control variables, which are listed below:
Renewables Percentage of total energy production by renewable sources
Hydro-power Percentage of total energy production by hydro power
Oil Percentage of total energy production by coal
Gas Percentage of total energy production by natural gas
Coal Percentage of total energy production by oil
Population Total population (P ) (only for the regression in totals)
Density Population density (P ) (only fro the regression per capita)
Industry Technology (T ). The value added to industrial production as a percentage of GDP
Imports pc Merchandising imports in $ per capita
Exports pc Merchandising exports in $ per capita
Notice, since the variables for % or total energy production by source add up to 100%, this study has excluded the variable for % of energy production by Nuclear power (since this variable has the
least observations of all energy source variables). If one includes all variables in the regression there will be perfect multicollinearity, since the last variable can be predicted using all the other variables. This is also known as the dummy trap, which also occurs when dealing with percentages.
The added value of the inclusion of these additional control variables will be thoroughly inves-tigated in section 6, as Harbaugh et al. (2002) find the specification of and inverted U-shaped EKC differs significantly when including different control variables. In order to compare models using different variables this paper will make use of the Akaike Information Criterion (AIC) developed by Akaike (1974) and the Bayesian Information Criterion (BIC) developed by Schwartz (1978). Note that these ”model selection” criteria can only be used to compare different models that use the exact same data for their estimation.
5.2 Panel data
A great advantage of panel data over a cross-sectional sample is that it is possible to control for the
country specific effects αi. If these effects are present, failure to do so leads to biased estimates if
these fixed or latent effects are correlated with the explanatory variables, as is likely to be the case. Equation (4) will be estimated using fixed effects (FE) OLS to account for unobserved hetero-geneity across countries that does not change over time. Previous research has alluded to country-specific differences such as the types of governments and political institutions that could play a role (Kumazawa and Callaghan, 2012). Congleton (1992), for instance, show authoritarian governments are more likely to enact less severe environmental standards than those of democratic governments. Such political variables are hypothesized to have effects that may not change over time within a country and contribute to the presence of the fixed effect.
In the ideal case the error term it is assumed to be i.i.d. (independently and identically
dis-tributed). This is however a strong assumption to impose. Since there is a high chance that the observations for certain groups of countries are correlated, the i.i.d. assumption is probably violated in this case. As a generalization, one can assume ”clustered standard errors”. When making use of clustered standard errors, one keeps the assumption of zero correlation across countries, but allows the within-country correlation to be nonzero. This implies that observations for country i are
al-lowed to be correlated for all t in some unknown way, inducing correlation in it within i, but that
countries i and j do not have correlated errors. In the presence of clustered errors, given that the other assumptions that are imposed imposed hold, OLS estimates are still consistent but standard errors may be inconsistent.
A problem that can arise with the use of panel data is that one or more of the series of variables turn out to be non-stationary. If variables are non-stationary then any correlation between the dependent and the explanatory variables could be due to the trending in both variables that is caused by a third variable not included in the model. This is called a spurious relationship, and is commonly noticed when performing regressions using non-stationary data. When the regression results turn out to be spurious, the resulting coefficients and their standard errors cannot be trusted. It is therefore important to investigate the stationarity of the different series of variables that are
used in this research. Panel unit root tests have been conducted for the data on emissions of CO2
and GDP. However, no decisive conclusions can be drawn from these tests. Therefore, for the baseline model, this study will assume that the data is stationary and there is no co-integration present in the data. In section 6.7, these unit root tests and the possibility of having non-stationary data will be discussed to further extent.
5.3 Estimation
As mentioned previously, this paper will assume none of the series contain a unit root and there is no co-integration for the baseline model. When performing a regression on a dynamic panel model such as (4), a difficulty arises with the fixed effects (FE) OLS estimation when the time dimension T is small, even if the number of countries N is large.
As Nickell (1981) shows, this arises because there is a correlation between the demeaned variable and the error term. The resulting correlation creates a bias in the estimate of the coefficient of the lagged dependent variable which is not mitigated by increasing N , in this case the number of countries. However, Judson and Owen (1999) find that in dynamic panel models, such as (4), the bias of FE OLS estimation (also known as least squares dummy variables or LSDV estimation) for dynamic panel data models can be sizeable, even when T = 20. Furthermore, when T = 30 they show that FE performs just as well or better than the viable alternatives.
The maximum time dimension of the panel used in this study is 52 years. However, due to a lack of data on emissions and GDP, this time dimension is smaller for some countries. The top 100 most polluting countries (except Myanmar, North Korea and Serbia) all have data available for more than 20 years. Therefore it can be concluded that the Nickell bias in this study will probably be of minor concern. It follows that, given the exogeneity of the regressors, a simple FE OLS estimation can used. Consequently, in section 6 the baseline model will be estimated making use of FE OLS regression using clustered standard errors.
6
Results
In the first part of this section the results for the baseline model will be summarized. Assuming stationarity, the dynamic version of the STRIPAT model (4) discussed in Section 5 will be considered first.
6.1 Baseline model
Assuming that all variables of interest are stationary over time, and that the policy variables are predetermined, one can make use of panel FE regression with clustered standard errors to estimate the baseline model (4).
As has been discussed in Section 5, the effect that including different control variables has on the outcome will be investigated. In Table 3, column I represents a basic EKC model that only incorporates GDP and the policy variables. In column II the factor population (P ) is added to the model in the form of population density. In column III the factor technology (T ) is added in the form of value added to industry as a % of GDP. In column IV the full model as described in (4) is used. Next to the estimation results, the AIC and the BIC have also been included in order to be able to compare the models using different variables.
Note that these coefficients describe the short term impact of these variables on the emissions per capita. In Table 4, these coefficients (and their standard deviations) have been transformed to describe the long term effects of these variables, since the implications of the long term effects are much more interesting for this research.
Fixed effects OLS estimation of specification (4) I II III IV log CO2(-1) pc .836∗∗ (.016) .835∗∗ (.016) .835∗∗ (.020) .773∗∗(.037) log GDP pc .314∗∗ (.034) .316∗∗ (.034) .270∗∗ (.032) .188∗∗(.040) log GDP2 pc -.016∗∗ (.002) -.016∗∗(.002) -.014∗∗ (.002) -.012∗∗ (.002) UNFCCC -.024∗∗ (.008) -.024∗∗(.008) -.016 (.010) -.019+ (.011) Kyoto .009 (.010) .008 (.009) .017 (.011) .013 (.011) log Density -.017 (.018) -.055 (.036) -.032 (.035) Industry .003∗∗ (.001) .004∗∗(.001) Renewables -.003∗∗ (.001) Hydro-power -.000 (.001) Oil .001∗ (.001) Gas .001 (.001) Coal .002∗∗(.001) log Imports pc .051∗∗(.018) log Exports pc .014 (.015) AIC -5817.149 -5642.200 -5109.144 -5011.911 BIC -5472.411 -5292.327 -4769.247 -4670.715
Table 3: Coefficients of country specific effects, time dummies and constant not reported.
Coefficients indicate short term effects. + p<0.10, ∗ p<0.05, ∗∗ p<0.01
FE OLS estimation with implied long term effects of specification (4)
I II III IV log GDP pc 1.915 (1.264) 1.915 (1.249) 1.175 (.828) .828 (.776) log GDP2 pc -.098 (.074) -.097 (.073) -.085 (.073) -.053 (.039) UNFCCC -.146 (.297) -.145 (.294) -.097 (.367) -.084 (.213) Kyoto .055 (.372) .048 (.331) .103 (.404) .057 (.213) log Density -.103 (.661) -.333 (1.322) -.141 (.679) Industry .018 (.037) .018 (.019) Renewables -.013 (.019) Hydro-power -.002 (.019) Oil .004 (.019) Gas .004 (.019) Coal .009 (.019) log Imports pc .225 (.349) log Exports pc .062 (.291) AIC -5817.149 -5642.200 -5109.144 -5011.911 BIC -5472.411 -5292.327 -4769.247 -4670.715
Table 4: Coefficients of country specific effects, time dummies and constant not reported.
+ p<0.10, ∗ p<0.05, ∗∗ p<0.01
After transforming the coefficients of the dynamic model to obtain the long term impact, it can be seen that the standard errors are quite large. Therefore these results do not provide any
meaningful insight.
Notice that the lagged term for emissions is always smaller than unity and significant at the 1% level. Also the terms for GDP per capita always have the same sign, and are significant at the 1% level. As more control variables are added to the model, the magnitude of the coefficients for the lagged term of emissions and that of GDP per capita decreases. It follows, as more variables are added to the model, these variables capture additional information previously contained in the error terms or in the GDP per capita variable.
As for the effect of the policy variables, it seems the Kyoto Protocol has not had a significant influence on the emissions per capita. However, the forming of the UNFCCC does seem to have a significant negative impact on emissions per capita. Adding more variables to the model does however reduce the statistical significance of this policy variable. For the simple EKC model, and the model incorporating population density, the UNFCCC variable is significant at the 1% level. Adding the factor for technology (T ) to the model in column III results in the coefficient for the policy variable to become statistically insignificant. In the full model in column IV, the UNFCCC variable is significant at the 10% level. When considering the short term impact, given the log-level specification it can be interpreted that the impact of being part of the UNFCCC in a certain year
reduces the CO2 emissions of that country by 2.5%, given the models in column I and II of Table 3.
When looking at the coefficients for the control variables, it is interesting to note the population density is never significant in each of these models. It can be seen that the value added to industry does have a significant effect on emissions. The production of renewable energy has significant decreasing impact on emissions, whereas of the conventional energy sources, Oil and Coal have a significant increasing impact. Lastly, notice only imports and not exports of merchandising goods have a significant impact on emissions. Notice however, that the sign of this coefficient is positive. This does not coincide with the literature found in Section 3, which suggests emissions are expected to rise with exports and fall with imports of merchandising goods.
When comparing the models in the four different columns based on the AIC and BIC it can be seen that the inclusion of more control variables leads to significantly higher values of the AIC and BIC. This implies that the simple model in column I is preferred over the more complex models in columns II, III and IV. After regressing an AR model on the residuals of this regression, for the full model in column IV, there is no evidence for serial correlation in the error terms.
Summarizing, it turns out that other variables actually capture some information of the binary policy variables when they are included in the model. Statistically, the Kyoto variable is not
signif-icant and the UNFCCC variable loses its significance when more variables are added to the model. However, before drawing any conclusions, it is important to consider the possible implications when some of the assumptions that have been made turn out to be invalid.
6.2 Extensions
In the next part of this Section, 6 sensitivity checks for the baseline model will be considered.
1. Firstly, equation 4 will be estimated using the total CO2 emissions as dependent variable. All
variables previously defined per capita will also be included as totals.
2. Similarly to equation (4), the static version of the baseline model as discussed in Section 3.2 will be estimated, after which these results will be compared to the dynamic case.
3. As discussed in Section 3, there is reasonable evidence that segmenting countries based on Annex I status reduces the heterogeneity that is present in the data. Therefore the baseline model (4) will also be estimated for these two groups of countries separately.
4. Next, in addition to the standard FE OLS regression, these models will be estimated by IV to control for the possibility of endogeneity in the policy variables.
5. Assuming that emissions and GDP are co-integrated, we will use Dynamic OLS to estimate the co-integrating relationship between the outcome and control variables.
6. Finally, as a robustness check, the baseline model will be estimated using emissions data obtained from another source, namely from the EDGAR.
All results for the extensions are shown in the Appendix. Next to these extensions, as a result of their rejection of the Kyoto Protocol, the exclusion of the U.S. from the model has been investigated. There is little to no difference when comparing these results to those of the baseline model, therefore it can be concluded that the results are robust to the exclusion of the U.S. from the model. In favor of space, these estimation results have been omitted.
6.3 Estimation in totals
In the literature there are two popular views on modeling emissions, as has been discusses in Section 3.3. In the baseline model this research has taken the dependent variable to be emissions per capita.
