Carbon leakage : the impact of the Kyoto Protocal on the net import of carbon dioxide

(1)

Faculty of Economics and Business

Carbon leakage

The impact of the Kyoto Protocol on the net import

of carbon dioxide

Abstract:

The debate about the effectiveness of policies to reduce CO2 emissions has led to the question if an agreement to reduce carbon emissions is resulting in carbon leakage. This research makes a comparison between the production-based accounting approach and the

consumption-based accounting approach and analyses the impact of the Kyoto Protocol on the difference between these two accounting approaches. By performing a panel data regression on 59 countries over the period of 1995-2011 this study tries to contribute to the discussion about carbon leakage as a consequence of the Kyoto Protocol. Results indicate that the commitment of a country to the Kyoto Protocol show an insignificant effect on the difference between consumption-based and production-based emissions. This implies that there appears to be no direct relationship between carbon leakage and the Kyoto Protocol.

Study programme: BSc Economics and Business

Specialization: Economics and Finance

Name: Alice Bubbers

Student number: 10846131

(2)

Statement of Originality

This document is written by Student Alice Bubbers 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.

(3)

1. Introduction

In order to control climate change, efforts have been made to stabilize atmospheric carbon dioxide emissions. To accomplish stabilization of atmospheric CO2 concentrations, a drastic reduction of global CO2 emissions is vital (Le Quéré, Raupach, Canadell, & Marland, 2009). In spite of several existing restrictions, global carbon dioxide emissions from cement

production and fossil fuel combustion have continued to grow by 2.5% per year over the past decade (Friedlingstein, Andrew, Rogelj, Schaeffer, & Vuuren, 2014). This increase is partly explained by the increased contribution to CO2 emissions from emerging economies, and from the production and international trade of goods and services (Le Quéré, Raupach, Canadell, & Marland, 2009). The debate about climate change and the disastrous consequences of climate change nourish discussion about policy instruments that would reduce the emissions of greenhouse gases (GHG). Many of these policies such as taxes, tradeable permits and regulations attempt to reduce carbon emissions by making the

production of those emissions more expensive (Hoeller, & Wallin, 1991). Climate change is a global problem. Therefore policies to reduce emissions are implemented by as many countries as possible. Nevertheless, such policies include only a section of the countries that actually produce greenhouse gases (Pearce, 1991). When only a subset of countries participate in an agreement, the role of international trade becomes important for it can cause a

misinterpretation of a country’s emissions level. For example, a country’s carbon emissions level can be low when the emissions are measured with a production-based approach. However, this can be caused by importing a significant quantity of carbon embodied in non-energy products. Those emissions will not be counted for when the import is from a country not included in the agreement (Wyckoff, & Roop, 1994). When political or economic gains are associated with artificially keeping the carbon emissions low through the import from countries which are not included in the agreement a problem is originated, called carbon leakage (Wyckoff, & Roop, 1994).

The aim of this study is to discover whether the commitment to the Kyoto Protocol causes carbon leakage. To test this hypothesized relationship the research will analyze the effect of the Kyoto Protocol on the difference between consumption-based emissions and production-based emissions. To do so, this paper uses a panel data regression on the difference in consumption- and production-based emissions of 59 countries for the period 1995-2011.

The paper proceeds as follows. In the next chapter the theoretical context, several previous papers related to the subject and the choice of methodology will be discussed. In

(5)

chapter three, the empirical model, the data collection, a descriptive analysis and the results of the regression are presented. Finally, discussion and conclusions are drawn in chapter four.

2. Literature review

The literature review gives a theoretical framework to set the basis for the empirical research. First, the production-based accounting and consumption-based accounting approach are discussed and a comparison is made between them. In the next section, the concept of carbon leakage is explained. In the third section some insights about the Kyoto Protocol are given. The possible relation between the Kyoto Protocol and carbon leakage is discussed in the fourth section. At last, several related studies are discussed.

2.1 Comparison of Production-based and Consumption-based accounting

Both measurements have advantages and disadvantages. The production-based accounting principle measures the carbon dioxide emissions from coal, oil and gas produced in a country by electricity production, private households and by producing goods and services. An advantage of the production-based accounting principle is that it is easy to calculate and therefore widely used in existing studies (Fan, Hou, Wang, Wang, & Wei, 2016). However, this approach ignores emissions from international air and sea transportation. Those emissions are not produced in a specific country and therefore not included in the accounting approach. Furthermore, the production-based accounting approach ignores international trade and therefore does not consider potential carbon leakage (Franzen, & Mader, 2018). The concept carbon leakage will be explained in more detail in the next section.

Unlike the production-based accounting, the consumption-based accounting approach includes the carbon emissions of international trade. This approach also covers more global emissions with limited participation and does consider potential carbon leakage. However, the downside of this approach is the difficulty and complexity of the measurement (Fan et al., 2016). The consumption-based accounting approach is based on complicated input-output matrices, which involves more assumptions. This can lead to more inaccuracy in the

measurement of CO2 emissions. Furthermore, the consumption-based approach violates the principle of product liability. This principle states that producers are responsible for the safety and quality of their products (Franzen, & Mader, 2018).

(6)

2.2 Carbon leakage

When protocols adopt unilateral emissions abatement by a subset of countries the environmental efficiency is not secured. The consequence of this abatement might be a movement of the greenhouse gas emissions from participating industrialized countries to non-participating developing countries without restrictions. This effect is called carbon leakage (Paltsev, 2001). Carbon leakage is defined as “the increase in emissions outside a region as a direct result of the policy to cap emissions in the region” (Boitier, 2008).

It is called strong carbon leakage when a country with strict emissions controls, taxes or regulations moves its production to countries with fewer restrictions and lower energy costs and imports the products. A carbon leakage is called weak when the cause of the leakage is the encouragement of international specialization to move production of carbon-intensive goods to other countries with lower costs (Franzen, & Mader, 2018).

Carbon leakage has several potential causes. The first source of carbon leakage is due to the change in the demand of global fossil-fuel. As a result of the commitment to the abatement of greenhouse gas emissions the demand may decrease, which in its turn will lead to lower international price for fossil fuels. This may have as a result that the demand for fossil-fuel and emission will increase in non-participating countries. Another reason for carbon leakage is the increase in cost of energy-intensive products in countries that participate in an agreement to abate global carbon dioxide emissions. The increase in the competitiveness of the energy-intensive industries might cause a shift of production from participating to non-participating countries (Paltsev, 2001).

2.3 Kyoto protocol

As a response to climate change, agreements have been designed to abate the growth of the greenhouse gas emissions. The negotiating process associated with this is embodied in the United Nations Framework Convention on Climate Change (UNFCCC) and its Kyoto Protocol (Paltsev, 2001). This protocol aims for the limitation of the aggregate carbon equivalent emissions of the greenhouse gasses by the first part of the 21st_{century. However,} the Kyoto Protocol only covers industrialized countries and economies in transition. Most developing countries have not committed themselves to the Protocol. This is due to the fact that developing countries have made minor contributions to global greenhouse gas

concentrations (Paltsev, 2001). Another possible explanation for the exclusion of developing countries is the economic development of those countries. The Environmental Kuznets Curve

(7)

hypothesis states that the environmental quality worsens with economic growth. However, after reaching a peak point in environmental deterioration a country will improve its

environment. This implies an inverted U-shaped relation between per capita income, which represents the economic growth, and the measured flow of pollution as shown in figure 1 (Huang, Lee, & Wu, 2008)

Figure 1: The Environmental Kuznets Curve hypothesis. Source: (Huang, Lee, & Wu, 2008)

Moreover, the Kyoto Protocol measures the carbon dioxide emissions based on the production of the country (United Nations, 1998). It is measured as the carbon dioxide emissions within national territories and the offshore area of authority of the country. This excludes the emissions associated with the consumption of goods that are produced outside the national territory (Peters, & Hertwich, 2008).

2.4 Relation between Kyoto protocol and carbon leakage

Boitier (2008) states that carbon leakage is the result of the methodology used for the National Inventory Report of the Intergovernmental Panel on Climate Change (IPCC). The methodology uses a production-based greenhouse gas emissions accounting approach and is not unilateral adopted. This framework measures the greenhouse gas emissions as the national emissions coming from domestic production of a country which can lead to a biased view of national GHG emissions. Especially, the international activities, such as international

transportation, pose a problem of allocation (Boitier, 2008). The difference between consumption-based and production-based emissions might be large due to carbon leakage. Countries involved with the Kyoto Protocol have a tendency to import more to look more ‘clean’ when measured with the production-based accounting approach. Hence, low emission countries with the production-based accounting approach might be a high emission country

(8)

with the consumption-based approach due to the difference of emissions in global trade. On the other hand, the opposite takes place when a high emission country is measured in the production-based approach produces a substantial quantity for the living standard of another country (Franzen, & Mader, 2018). The combination of the production-based measurement of carbon emissions and the exclusion of developing countries might result in carbon leakage (Paltsev, 2001).

2.5 Related studies

Carbon leakage is typically measured as the increase of emissions in non-participating countries relative to the reduction of emissions in countries participating in an abatement. Thus, carbon leakage is defined as the increase of emissions in non-abating region divided by the decrease in abating regions. The results of carbon leakage differ depending on

parameterization and the assumptions of the model. The effect of the Kyoto Protocol on the carbon leakage rate is estimated in the range of 5% to 130% (Aichele, & Felbermayr, 2015).

Babiker (2000) used a MIT Emissions Prediction and Policy Analysis (EPPA) Model to analyze the carbon leakage resulting from the Kyoto Protocol. EPPA model is a general equilibrium model of the world economy. This model uses a dataset called GTAP-E. This dataset accommodates a representation of energy markets in physical units, comprehensive accounts of regional production and bilateral trade flows. Babiker (2000) found generally favorable movement in terms of trade for energy importers and adverse movement for energy exporters for countries committed to the Kyoto Protocol, which would imply carbon leakage due to the Kyoto Protocol.

Aichele and Felbermayr (2015) derived a theoretical gravity equation. In this equation greenhouse gas emission are included in trade. The structure of the panel database allows controlling for the endogenous selection of countries into the Kyoto Protocol. They found an increase of 8% in embodiment of carbon in import of participating countries and an increase of 3% in the emission intensity of the import. Paltsev (2001) adopted a decomposition technique which attributes increases in GHG emission by non-participating countries to specific sectors in participating countries. Paltsev (2001) found a carbon leakage rate of ten percent as a result of Kyoto Protocol.

Davis and Caldeira (2010) use a multiregional input-output (MRIO) analysis that is based on monetary flows between regions and industrial sectors. It considers the consumption of one region produced in another region, the total economic output of each sector in each region, and a matrix of intermediate consumption. Furthermore, they aim to explain the

(9)

difference between production emissions and consumption emissions with the population, GDP per capita, carbon intensity of energy consumption and the energy intensity of world GDP. They found that substantial CO2 emissions are traded and therefore excluded in traditional production-based national emissions inventories. They also found that the difference between consumption-based and production-based accounting is greatest when carbon intensity of GDP where high and trade imbalances are large (Davis, & Caldeira, 2010).

2.6 Choice methodology

This paper will examine the carbon leakage by comparing the production-based carbon dioxide emissions with the consumption-based carbon dioxide emissions of different countries over a period of time and using the independent variable participation in Kyoto Protocol. The difference will be examined with a panel data analysis. According to Stock and Watson (2007) a panel data analysis allows the researcher to make a comparison of the values of the dependent variable over time. The dependent variable will consist of the difference between production-based carbon dioxide emissions and the consumption-based carbon dioxide emissions divided by the production-based carbon dioxide emissions. Using both accounting approaches of emissions enables to identify whether and to what extent

improvements in decoupling and in productivity are a result of the Kyoto Protocol or due to the outsourcing of production. Accordingly, it will help to estimate carbon leakage

(Wiedmann, 2009).

3 Methodology and Empirical Analysis

The purpose of the methodological and the empirical analysis section is to set out the

framework through which the research question is executed. In the first section a description of the model used in the panel data regression and the assumptions of the model are stated. In the second section the dependent and the independent variables are discussed. It covers the used databases and discusses the reliability. The third section is the descriptive analysis. This section will provide insights in the observations of the variables used. The last section

presents the outputs of the different regression with the difference between consumption-based emissions and production-consumption-based emissions as the dependent variable.

(10)

3.1 Empirical model

In order to test the hypothesized relationship between the Kyoto Protocol and carbon leakage a panel data regression will be applied. A panel data regression allows for yearly observations within the selected countries (Stock, & Watson, 2007). The regression models will be based on panel data for the period of 1995 until 2011 for 59 countries using Stata. The panel regression model is based on yearly data and therefore 17 years are counted for. The

dependent variable of the three models is the difference between consumption-based carbon dioxide emissions and based carbon dioxide emissions divided by the production-based carbon dioxide emissions.

The first model shows the regression of a random effect (RE) panel regression in which the dependent variable is regressed on the dummy variable Kyoto. The second model adds some control variables to control for omitted variable bias (Stock, & Watson, 2007). This model uses a fixed effects (FE) panel regression in which the dependent variable is regressed on the involvement with the Kyoto Protocol (Kyoto), population in million (P), energy per unit GDP (E) and emissions per unit energy (K). The GDP per capita (GDPPC) is left out of this model because a correlation is expected between the control variable and the Kyoto Protocol. This is expected by reason of the above mentioned relation between the economic growth of a country and the willingness of a country to improve its environment. The third model is a fixed effect (FE) panel regression, where GDP per capita is added to the regression equation. Therefore, the dependent variable is regressed on the involvement in the Kyoto Protocol (Kyoto), population in million (P), energy per unit GDP (E), emissions per unit energy (K) and GDP per capita (GDPPC).

(𝐹CR− 𝐹PR)it = 𝛽0 + 1𝐾𝑦𝑜𝑡𝑜it + 𝛼i + 𝜈it (1)

(𝐹CR− 𝐹PR)it = 𝛽0 + 1𝐾𝑦𝑜𝑡𝑜it + 2 𝑃it + 𝛽3 𝐸it + 𝛽4 𝐾it + 𝛼i + 𝛿t + 𝜈it (2)

(𝐹CR− 𝐹PR)it = 𝛽0 + 1𝐾𝑦𝑜𝑡𝑜it + 2 𝐺𝐷𝑃𝑃𝐶it + 3 𝑃it + 𝛽4 𝐸it + 𝛽5 𝐾it + 𝛼i + 𝛿t + 𝜈it. (3)

The first assumption of the model regressions is that the error term has conditional mean zero, given the independent variables of all time periods. This implies that there is no omitted variable bias. The second assumption is that the variable for one country is distributed independently of and identically to the variables for another country. Specifically, the

(11)

variables are independently and identically distributed across countries for I = 1,…, 59. This is achieved when the countries are selected by simple random sampling. The third assumption holds that large outliers are unlikely. The last assumption is that no perfect multicollinearity is allowed (Stock, & Watson, 2007).

Wooldridge (2006) states that the difference between the fixed effects and random effects model is that the random effects regression model has an additional assumption that the unobserved effect, 𝛼i, is uncorrelated with each explanatory variable. This implies that random effects model is more efficient, when the unobserved effect is uncorrelated with each explanatory variable. The fixed effect model allows for correlation between the unobserved effect and the explanatory variables (Wooldridge, 2006).

The decision to use a fixed or random effects model is based on the Hausman test. The Hausman test is designed to find the model with the highest efficiency (Wooldridge, 2006). The detailed outcomes of the test can be found in the appendix section A. The Hausman test indicates the use of random effects regression model for model 1 and fixed effects model regression for model 2 and model 3. This implies in model 1 that the country specific effects are uncorrelated with the engagement to the Kyoto Protocol. While in model 2 and 3 the country specific effects do correlate with the explaining variables.

In order to test the above-mentioned assumptions two tests are performed on serial correlation and heteroskedasticity. Serial correlation is likely because the values of the independent variables of one year tend to have a relation with the values of the independent variables of the next year. For example the Kyoto Protocol; when a country is committed to the Kyoto Protocol in 2000 it is very likely that the same country is still committed in 2001. Similarly, it is reasonable to expect 𝜈it to be serial correlated, because it consists of time-varying factors that are determinants of the difference between consumption-based emissions and production-based emissions, but are not included in the regression and some of those omitted factors might be serial correlated (Stock, & Watson, 2007). To test for serial correlation between residuals in the panel data, a Wooldridge test will be performed. The results of the test are for the three models that serial correlation is present and can be found in the appendix section B. A test is performed on homoskedasticity because heteroskedasticity is expected due to the expectation that countries will have different errors over time. To test if the models are homoskedasticity or heteroskedasticity a Wald test is performed. The Wald test is a F-statistic that tests the null hypothesis of the errors being stable. The outcomes of the tests are heteroskedasticity for the three models and therefore the Wald test confirms that the errors differ over time. The detailed analysis can be found in the appendix section C.

(12)

In order to deal with the problems of heteroskedasticity and serial correlation,

clustered standard errors are used in Stata. This way the regression errors are allowed to have an arbitrary correlation within the country, but assumes that the regression errors are

uncorrelated across the countries (Stock, & Watson, 2007).

3.2 Data

The dependent variable (FCR-FPR) of the linear regression is the difference between the consumption-based carbon dioxide emissions (CR) and the production-based carbon dioxide emissions (PR) divided by the production-based carbon dioxide emissions. This discrepancy represents the net effect of emissions embodied in trade scaled by the production-based emissions. The purpose of the scaling of the differences to the production-based carbon emissions is to control for big difference due to the size of the amount of carbon emissions of a country to another. The dependent variable measures the surplus of consumption-based emissions over production-based emissions in a percentage of production-based emissions. The consumption-based carbon dioxide emissions will be extracted from the Organization for Economic Cooperation and Development (OECD) iLibrary database. This database is the online library of the OECD and contains all the available data published by the OECD (OECD, 2015). The OECD has developed models to estimate CO2 emissions embodied in final demand to make a contribution to a better understanding of how CO2 emissions are driven by global consumption patterns since the early 1990s (Wiebe, & Yamano, 2016). The data used for this research is extracted from a model constructed by the OECD. The model uses a combination of the 2015 edition of the OECD Inter-Country Input-Output (ICIO) tables and detailed International Energy Agency (IEA) CO2 emissions from fuel combustion data (Wiebe, & Yamano, 2016). The consumption-based emissions can be found under CO2 embodied in domestic final demand. The production-based carbon dioxide emissions will also be extracted from the OECD iLibrary. To estimate the production-based emissions, the OECD allocated the IEA carbon dioxide to the 34 target industries in OECD ICIO and to final

demand (OECD, 2015). The production-based emissions can be found under CO2 emissions based on production.

The variable Kyoto Protocol is a dummy variable, which will be one when a country ratified the Kyoto Protocol and zero when the country did not accepted or confirmed the protocol. The data about the year of acceptation of the Kyoto Protocol per country can be found in the United Nations treaty collection.

(13)

The first control variables is Gross Domestic Product per capita (GDPPC) because global trade is influenced by GDP. For instance, when a financial crisis occurs, the GDP per capita will decrease and global trade will decrease consequently. This will affect the

dependent variable and therefore should be included in the regression. The data for GDP will be contracted from the database Penn World Table (PWT) version 9.0. This database provides information on relative levels of income, output, input and productivity, covering 182

countries between 1950 and 2014. The expenditure-side real GDP at chained purchasing power parity in million 2011 US$ (RGDPe_{) will be used. This GDP is used to compare} relative living standards across countries and over time (Feenstra, Inklaar, & Timmer, 2015). For the regression model GDP per capita is used to control for differences in magnitude of the population. To obtain the GDP per capita, the GDP must be divided with the population of the country in the year of interest.

The second control variable is the population in million (P). When the population increases, global trade will increase and therefore the dependent variable of the regression will be influenced (Raupach et al., 2007). The data for this control variable, population (in millions), is extracted from the database Penn World Table.

The third control variable is energy per unit GDP (E). This control variable represents commercial primary energy: fossil fuels, nuclear, and renewables when used to generate electricity. It refers to the energy from production plus imports and stock changes minus exports and energy use due to planes and ships in international transport. Hence, it presents the commercial primary energy measured with a consumption-based accounting approach. This measures the energy efficiency of a nation’s economy (World Bank, 2018). The data is extracted from the database World Bank Open Data. The database World Bank Open Data is a database that contains collections of time series data on a variety of topics. The control variable energy per unit GDP can be found under Energy use (kg of oil equivalent) per $1,000 GDP (The World Bank Group, 2018).

The last control variable is emissions per unit energy (K). The variable, emissions per unit energy, is taken on in the regression because when emissions rises per unit energy the difference between consumption-based and production-based emissions will automatically be influenced (Raupach et al., 2007). The data for this variable is extracted from database World Bank Open Data and can be found under CO2 intensity (kg per kg of oil equivalent energy use). This data represent the ratio of carbon dioxide per unit of energy, or the amount of carbon dioxide emitted as a result of using one unit of energy in production (The World Bank Group, 2018).

(14)

To control for time-specific effects the fixed effects model includes a time-indicator variable denoted as 𝛿t in the model specification. The variable 𝛼i controls for the country-specific effects. i refers to the country and t to the time of the observation.

Although the available data is comprehensive, it is not flawless. There is missing data for various variables. Especially the variables of developing countries not involved in the Kyoto Protocol are excluded from the databases. Consequently, the principle of data selection emerges to keep a balanced dataset. The research consists data of 59 countries, where 35 countries are a part of the OECD and the remainder are non-OECD countries. A summary of the countries is given in the appendix section D.

3.3 Descriptive analysis

In the appendix section E, a summary of the descriptive statistics of the data for the entire period of 1995 until 2011 is presented. This period represents a period with a commitment to the Kyoto Protocol and without the commitment for the countries. The mean difference between consumption-based emissions and the production-based emissions divided by the production-based emissions is 0.124 for the 59 countries. This would imply that on average the consumption-based emission are 12.4% higher compared to the production-based emissions. The averages for the entire period and the 59 countries investigated are for gross domestic product per capita 24,578.07, for population 78.095 million, for energy per unit GDP the average is 134.708 and for the emissions per unit energy 2.329.

In the appendix section F and G, a distinction is made between the years a country is committed to the Kyoto Protocol and the years a country is not part of the Kyoto Protocol. The mean of the discrepancy between consumption- and production-based emissions of a country committed to the Kyoto Protocol is 0.140. This suggest that consumption-based emissions are 14.0% higher compared to the production-based emissions on average for countries involved with the Kyoto Protocol. Hence, when a country is involved with the Kyoto Protocol on average the consumption-based emissions are 1.14 times production-based emissions. This implies a positive net effect of emissions embodied in trade whereas the average of the discrepancy when there is no commitment is 0.107 on average. Which implies that the consumption-based emissions are 1.107 times production-based emissions. Hence, there is a higher mean when countries are committed to the Kyoto Protocol than when they are not committed.

(15)

The table in the appendix section H presents the correlations between the variables. The correlations between the various variables are low. This means that the relationship between the variables is weak.

Graphic 1 and 2 present a graphical illustration of how the difference between consumption-based emissions and based emissions divided by the production-based emissions behave between 1995 and 2011. Graph 1 represents the discrepancy without the Kyoto Protocol and graph 2 with the Kyoto Protocol. In graph 1 only a few lines continue to the end of the time reeks, this suggest that only a few countries in the sample have never agreed to the Kyoto Protocol. In graph 1 there is some more variation over time in

comparison with graph 2. This suggest that countries with the Kyoto Protocol have more variation over time than countries without the Kyoto Protocol. But overall, small variations are detected over time.

Graph 1: Line plot without Kyoto Protocol

Graph 2: Line plot with Kyoto Protocol

Graph 3 shows the movement of the discrepancy of consumption-based and production-based emission as a percentage of the production-based emissions per country

(16)

over the years 1995 until 2011. The movements differ substantially per country. This movement suggest that fixed effects for countries is appropriate.

Graph 3: The movement of the discrepancy of consumption-based and production-based emission per country over the years 1995 until 2011

3.4 Results

The outputs of the first model in table 1 below shows the regression result of a random effects panel regression in which the difference between consumption-based emissions and

production-based emissions as a percentage of production-based emissions is regressed on the dummy variable Kyoto. The coefficient is 0.046 and is significant at a significance level of 1%. The coefficient is positive which implies a mild indication that as countries commit themselves to the Kyoto Protocol the difference between consumption-based and production-based emissions of a country increases. In the appendix section I, the R-squares of the models are presented. The model has a low overall R-squared of 0.0043, which implies that only 0.43% of the dependent variable is explained by the regression model. This is understandable because the regression model consists of only one explanatory variable.

The second model extends model 1 by including three control variables, population in millions (P), energy per unit GDP (E) and the emissions per unit energy (K). The outputs of the fixed effects panel regression can be found in table 1. There is no significant effect found on the dummy variable Kyoto and on the emissions per unit energy. There is a significant negative effect of energy per unit GDP at a significance level of 1%. This would imply that

(17)

when the energy per unit GDP increases the difference between the consumption-based emissions and the production-based emissions will decrease. Hence, the carbon dioxide import will decrease when energy per unit GDP increases. As stated in section 3.2 energy per unit GDP measures the energy efficiency. If the energy per unit GDP goes up the energy efficiency of a country decreases. This means that the dependent variable of the model and thus the net import of carbon emissions increases as the energy efficiency increases. When using a significance level of 10%, there is a small significant negative effect of population in million on the dependent variable. This would mean that if the population goes up by 1 million the difference between consumption-based and production-based emissions per 1 million population goes down. Hence, the bigger the population the smaller the difference between consumption-based and production-based emissions. Again, the overall R-squared is low.

The outputs of the last model in table 1 show the regression results of the fixed effects panel regression. The difference between consumption-based emissions and production-based emissions as a percentage of the production-based emissions is regressed on the dummy variable Kyoto, population in million, energy per unit GDP, the emissions per unit energy and GDP per capita. The output shows no significant effect of the dummy Kyoto on the dependent variable. This implies that there is no effect on the discrepancy between consumption-based and production-based carbon dioxide emissions due to the commitment of a country to the Kyoto Protocol. Again, there is a significant negative effect of energy efficiency on the dependent variable of the model at a significance level of 1% and a small significant effect of population on the dependent variable. The other variables included in the model do not show a significant effect on the difference between consumption-based and production-based emissions.

(18)

Model 1 (RE) Model 2 (FE) Model 3 (FE) Kyoto 0.046*** 0.015 0.011 (0.01) (0.01) (0.01) P -0.001* -0.001* (0.00) (0.00) E -0.002*** -0.002*** (0.00) (0.00) K 0.013 0.015 (0.03) (0.03) GDPPC 0.000 (0.00) constant 0.100** 0.361*** 0.335* (0.03) (0.08) (0.14)

Table 6: regression results. Note: *** denotes significance at 1%, ** denotes significance at 5% and * denotes significance at 10

4. Discussion and conclusion

The discussion and conclusion section is dedicated to the discussion of the outcomes and suggests further research. Firstly, a conclusion will be drawn from the performed research. The second part will provide a discussion about the research and suggestion about further research will be given. At the end a recommendation is given for the accounting approach to use for protocols in the future.

The main goal of this study was to examine the impact of the Kyoto Protocol on the carbon leakage. This was done by examining the impact of the Kyoto Protocol on the difference between consumption-based carbon emissions and production-based carbon emissions. This is appropriate to investigate the carbon leakage because the discrepancy represents the net import of carbon dioxide. When the net import of carbon dioxide increases, the emissions produced inside a region will decrease. When this would increase as a

consequence of the commitment to the Kyoto Protocol, it would suggest carbon leakage, as carbon leakage is defined as the increase in emissions outside a region as a direct result of the policy to cap emissions in the region (Boitier, 2008). The research involves a panel data regression of 59 countries over the period of 1995 until 2011.

(19)

The analysis of the discrepancy between the consumption-based and production-based carbon emissions reveals no empirical evidence for an impact of the Kyoto Protocol. This implies that the net import of carbon emissions is not effected by the commitment of a country to the Kyoto Protocol. This would mean that the Kyoto Protocol did not result in carbon leakage from countries committed to the Kyoto Protocol to countries that did not agree on the Kyoto Protocol. The research did obtain a significant effect of energy intensity on the net import of carbon emissions.

On average, countries do have a higher discrepancy between the two measurements of carbon emissions when committed to the Kyoto Protocol than when the countries are not committed to the Protocol. This is partly explained by the energy per GDP of different countries. The rest of this phenomenon must be explained by other factors not taken into consideration in this research. Further research is recommended to explain the movement of net import of carbon emissions.

It should be stated that the selection of countries is not completely random, as data from developing countries that are not included in the Kyoto Protocol is not available and therefore not included in the research. The consequence of this is that the sample primarily consists of countries that agreed on the Kyoto Protocol at one time. Therefore, it is possible that the sample only consists of countries that are willing to improve their environment and therefore willing to abate their consumption and production-based carbon emissions. This can cause a sample of only the ‘good’ countries and can give an inaccurate outcome. Therefore, further research is recommended with a comparison between countries committed and countries not yet committed to the Kyoto Protocol.

Even though the consumption-based accounting approach has the advantage of

including the carbon emissions of international trade and therefore considers potential carbon leakage, it is recommended to stay with the production-based accounting approach (Fan et al., 2016). When investigating the difference between the production-based CO2 emissions and the consumption-based CO2 emissions small differences are detected. Thereby, the

production-based emission accounting approach is wide-spread used, easily to calculate and has lower uncertainty.

(20)

Reference list

Aichele, R., & Felbermayr, G. (2015). Kyoto and Carbon Leakage: An Empirical Analysis of the Carbon Content of Bilateral Trade. Review of Economics and Statistics, 97(1), 104-115.

Babiker, M., Reilly, J., & Jacoby, H.D. (2000). The Kyoto Protocol and developing countries. Energy Policy, 28(8), 525-536.

Boitier B. (2012). CO2 emissions production-based accounting vs consumption: Insights from the WIOD databased, WIOD Conference paper, April 2012.

Davis, S.J., & Caldeira, K. (2010). Consumption-based accounting of CO2 emissions.

Proceedings of the National Academy of Sciences, 107 (12), 5687-5692.

Fan, J., Hou, Y., Wang, Q., Wang, C., & Wei, Y. (2016). Exploring the characteristics of production-based and consumption-based carbon emissions of major economies: A multiple-dimension comparison. Applied Energy, 184(C), 790-799.

Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015). The Next Generation of the Penn World Table. American Economic Review, 105(10), 3150-3182.

Franzen, A., & Mader, S. (2018). Consumption-based versus production-based accounting of CO2 emissions: Is there evidence for carbon leakage? Environmental Science and

Policy, 84, 34-40.

Friedlingstein, P., Andrew, R.M., Rogelj, J., Schaeffer, M., & Vuuren, D.P. (2014). Persistent growth of CO2 emissions and implications for reaching climate targets. Nature Geoscience, 7, 709-715.

Hoeller, P., & Wallin, M. (1991). Energy Prices, Taxes and Carbon Dioxide Emissions.

Paris: OECD Publishing, 106.

Huang, W.M., Lee, G.W.M., & Wu, C.C. (2008). GHG emissions, GDP growth and the Kyoto Protocol: A revisit of Environmental Kuznets Curve hypothesis. Energy

Policy, 36(1), 239-247.

Le Quéré C., Raupach M.R., Canadell J.G., & Marland G. (2009). Trends in the sources and sinks of carbon dioxide. Nature Geoscience, 2(12), 831-836.

Paltsev, S. (2001). The Kyoto Protocol : Regional and sectoral contributions to the carbon leakage. The Energy Journal, 22(4), 53-79.

Pearce, D. (1991). The role of carbon taxes in adjusting to global warming, The Economic

(21)

Peters, G., & Hertwich, P. (2008). Post-Kyoto greenhouse gas inventories: Production versus consumption. Climatic Change, 86(1), 51-66.

OECD. (2015). OECD production- and consumption-based CO2 emissions estimates. Retrieved from https://stats.oecd.org/Index.aspx?DataSetCode=IO_GHG_2015 Raupach, M.R., Marland, G., Ciais, P., Le Quéré, C., Canadell, J.G., Klepper, G., & Field,

C.B. (2007). Global and regional drivers of accelerating CO2 emissions. Proceedings

of the National Academy of Sciences, 104(24), 10288.

Stock, J. H., & Watson, M. W. (2007). Introduction to econometrics. Boston: Pearson/Addison Wesley.

United Nations Treaty Collection. (2007). Kyoto Protocol to the United Nations Framework Convention on Climate Change. Retrieved from

https://treaties.un.org/doc/Publication/MTDSG/Volume%20II/Chapter%20XXVII/XX VII-7-a.en.pdf

United Nations. (1998). Kyoto Protocol To The United Nations Framework Convention On Climate Change. Retrieved from https://unfccc.int/sites/default/files/kpeng.pdf Wiebe, K., & Yamano, N., (2016). Estimating CO2 Emissions Embodied in Final Demand

and Trade Using the OECD ICIO 2015: Methodology and Results. OECD Publishing.

Wiedmann, T. (2009). A review of recent multi-region input–output models used for consumption-based emission and resource accounting, Ecological Economics, 69(2) Wooldridge, J. M. (2006). Introductory econometrics: A modern approach. Mason, OH:

Thomson/South-Western.

World Bank. (2018). World Bank Open Data. Retrieved from

https://data.worldbank.org/indicator/EN.ATM.CO2E.EG.ZS?view=chart

Wyckoff, A.W., & Roop, J.M. (1994). The embodiment of carbon in imports of manufactured products: Implications for international agreements on greenhouse gas

(22)

Appendix

Chi-square test value Probability Indication

Model 1 0.215 0.6426 Random effects

Model 2 36.775 0.0000 Fixed effects

Model 3 36.004 0.0000 Fixed effects

Appendix A: Hausman test on Random or Fixed effects

F Statistic Probability Indication

Model 1 34.613 0.0000 Serial correlation

Appendix B: Test on serial correlation occurrence

Chi-square test value Probability Indication

Model 1 62.27 0.0000 Heteroskedasticity

(23)

OECD countries

Non-OECD countries

Australia Korea Argentina

Austria Latvia Bulgaria

Belgium Luxembourg Brazil

Canada Mexico

Brunei Darussalam

Chile Netherlands China

Czech Republic New Zeeland Colombia

Denmark Norway Costa Rica

Estonia Poland Cyprus

Finland Portugal Croatia

France

Slovak

Republic Indonesia

Germany Slovenia India

Greece Spain Cambodia

Hungary Sweden Lithuania

Iceland Switzerland Malta

Ireland Turkey Malaysia

Israel

United

Kingdom Philippines Italy United States Romania

Japan Russia Saudi Arabia Singapore Thailand Tunisia Viet Nam South Africa

(24)

Variable Observations Mean Std. Dev. Min Max (FCR-FPR) 1003 0.1240737 0.255364 -0.552713 1.389966 Population (in million) 1003 78.09457 218.4047 .267454 1348.174 Energy per unit GDP (E) 1003 134.7077 57.6861 57.05072 458.2714 Emissions per unit energy (K) 1003 2.328551 0.61867 0.3247582 3.553257 GDP per capita 1003 24578.07 15766.05 1098.134 90444.99

Appendix E: Descriptive statistics for the entire sample

Variable Observations Mean Std. Dev. Min Max

(FCR-FPR) 526 0.139983 0.2523149 -0.4277074 1.389966 Population (in million) 526 81.29019 236.5318 0.286865 1348.174 Energy per unit GDP (E) 526 124.2862 52.42605 57.05072 458.2714 Emissions per unit energy (K) 526 2.331268 0.6020515 0.3247582 3.517773 GDP per capita 526 26375.28 15727.69 2167.246 90444.99

(25)

Variable Observations Mean Std. Dev. Min Max (FCR-FPR) 477 0.1065301 0.2578118 -0.552713 1.085431 Population (in million) 477 74.57069 196.676 0.267454 1277.189 Energy per unit GDP (E) 477 146.1997 61.0011 66.28729 343.7818 Emissions per unit energy (K) 477 2.325555 .6371142 .5108376 3.553257 GDP per capita 477 22596.25 15585.82 1098.134 84288.27

Appendix G: Descriptive data without Kyoto agreement

Variable s (1) (2) (3) (4) (5) (6) (1) (FCR-FPR) 1.000 (2) Kyoto 0.0655 1.000 (3) P -0.2184 0.0154 1.000 (4) E -0.3250 --0.1898 0.2177 1.000 (5) K -0.4781 0.0046 0.1455 -0.0920 1.000 (6) GDPPC 0.1787 0.1198 -0.2678 -0.2059 -0.0893 1.000

(26)

Model 1 (RE) Model 2 (FE) Model 3 (FE) Kyoto 0.046*** 0.015 0.011 (0.01) (0.01) (0.01) P -0.000* -0.000* (0.00) (0.00) E -0.002*** -0.002*** (0.00) (0.00) K 0.013 0.015 (0.03) (0.03) GDPPC 0.000 (0.00) constant 0.100** 0.361*** 0.335* (0.03) (0.08) (0.14) 𝑅2_(within) _0.0621 _0.2033 _0.2043 𝑅2_(between) _0.0022 _0.0906 _0.0912 𝑅2_(overall) _0.0043 _0.0948 _0.0956

Appendix I: STATA output fixed and random regression models * p<0.05, ** p<0.01, *** p<0.001