efforts in developing countries
Joukje de Vries S2743485
j.de.vries.73@student.rug.nl
MSc International Economics & Business Faculty of Economics and Business
University of Groningen
Supervisor: dr. A. Minasyan Co-assessor: dr. T. Kohl
ABSTRACT
In response to growing concerns about the impacts of climate change, more and larger financial commitments are made by developed countries to assist the developing world in mitigating emissions in greenhouse gases and helping them adapt to the impacts of climate change. Availability of new, real world data on climate-related development aid from the OECD Development Assistance Committee (DAC) allows for empirically testing whether theoretically proposed effects of mitigation and adaptation aid in previous literature hold on the quality of the environment, in terms of greenhouse gas emissions and the Climate Risk Index respectively. Moreover, the suggestion that government effectiveness plays a facilitating role in mitigation and adaptation effectiveness is tested empirically, using an interaction term between government effectiveness and the climate-related aid variables. The results from fixed effects and first differences estimations provide modest statistically significant evidence for a negative correlation between adaptation aid and the Climate Risk Index, supporting the theory that adaptation aid helps recipient countries become more resilient to climate change impacts. The first differences estimation of mitigation-related aid effectiveness provides weak statistically significant results to believe different dynamics are present between greenhouse gas emissions and mitigation aid, such as a possible reverse causal relationship. No robust statistically significant effects are found for a facilitating role of government effectiveness in either mitigation- or adaptation-related aid effectiveness. Implications of these results may be for donor countries to consider the choice between which type of aid to distribute and to reconsider the institutional criteria they handle when choosing the recipient country.
TABLE OF CONTENTS
INTRODUCTION ... 4
CONCEPTUAL FRAMEWORK ... 5
DATA AND METHODS ... 9
Variables and data source ... 9
Sample and dataset ... 12
Estimation method ... 13
Endogeneity concerns ... 15
Model specification ... 16
EMPIRICAL RESULTS ... 17
Descriptive analysis ... 17
Regression results of the fixed effects model ... 20
Alternative estimation method ... 26
Regression results of the first differences model... 28
DISCUSSION AND CONCLUSIONS ... 31
Policy implications ... 32
Limitations ... 33
Conclusion ... 34
REFERENCES ... 35
APPENDIX ... 39
LIST OF TABLES AND FIGURES Table 1. Variables used in the study………..13
Table 2. Descriptive statistics………....18
Table 3. Regression results fixed effects model (mitigation)………....21
Table 4. Regression results fixed effects model (adaptation)………...24
Table 5. Regression results first differences model (mitigation)………28
Table 6. Regression results first differences model (adaptation)………30
Table B1. Hausman test results………..40
Table B2. Wooldridge test results………..40
Table C. Skewness and kurtosis value before and after log transformation……….….41
Figure 1. Mitigation and adaptation finance………..18
Figure 2. Average mitigation aid and greenhouse gas emissions in sample countries…………19
INTRODUCTION
Climate change is no longer a deniable fact. In 1992, the United Nations Framework Convention on Climate Change (UNFCCC) was established, as the world started to realise that measures have to be taken in order to tackle the negative effects of climate change on the habitability of planet earth. The 197 parties to the treaty1 meet each year to analyse the progress in reaching the objective to “stabilize greenhouse gas concentrations in the atmosphere at a level that would prevent dangerous anthropogenic interference with the climate system” (United Nations, 1992:4). Scientific input on adaptation and mitigation options comes from the Intergovernmental Panel on Climate Change (IPCC), which has established that it is undeniable that mankind accelerates climate change. They urge that actions have to be taken to deal with the consequences, such as extreme weather conditions, rising sea levels and global warming (Adger, Huq, Brown, Conqay & Hulme, 2003). The IPCC has found that the ability to cope with the adverse effects of climate change is unevenly distributed throughout the world. Those people who will experience the worst effects of climate change, are also the people that are least equipped to tackle the problems (Barrett, 2013). This specifically refers to developing countries which are vulnerably situated, such as African countries and Small Island Developing States (SIDS) (e.g. Busby, Cook, Vizy, Smith & Bekalo, 2014). This situation is referred to as a ‘double inequality’ when it comes to developing states; although the majority of these countries produce, globally seen, the least amounts of greenhouse gas emissions, they experience a disproportionate amount of the impacts of accelerating climate change (Barrett, 2013).
The vulnerability of the least developed countries (LDCs) in this world is exacerbated by their dependence on income from agriculture (Busby et al., 2014) and political, economic and social conditions which are unfavourable to the security and safety of inhabitants, buildings, the economy and the environment (Murray & Ebi, 2012). The unjust nature of this situation is internationally recognised, as Article 3 and 4 of the UNFCCC assign the largest share of responsibility in allocating resources to climate change adaptation and mitigation to developed nations, including a transfer of financial resources and technology to developing countries for the same objective (United Nations, 1992). In the 2001 Conference of the Parties (COP) in Morocco, funding mechanisms have been formally agreed upon to commit to aiding developing countries adapt to climate change (Adger et al., 2003). The necessity of transferring climate-related development aid is further acted upon with the 2009 Copenhagen Accord (UNFCCC, 2010). Under this accord, developed countries have agreed to increase efforts to assist developing countries with climate adaptation, specifically of an extra $100 billlion per year of climate financing (Eyckmans, Fankhauser & Kverndokk, 2016). Moreover, in 2015, the Paris Agreement was negotiated, in which almost all parties to the UNFCCC have agreed to develop and commit to plans which contribute to the long-term goal of limiting the increase in average global temperature to 1.5 degrees Celsius (UNFCCC, 2015). The agreement includes a reaffirmation of the commitment by developed nations to continue mobilising $100 billion a
year for adaptation in developing countries until 2025. In December 2018, the 24th Conference of the Parties took place in Katowice, Poland, where a number of developed countries have pledged to raise funds on the long term, in order to increase assistance to the most vulnerable developing countries for adapting to climate change (UNFCCC, 2018b).
The extra billions of dollars aimed at climate change adaptation and mitigation in developing countries are on top of the more general development aid which is distributed to developing countries (Eykmans et al., 2016). The goal of this extra development aid is to help the developing world reduce greenhouse gas emissions and also take adaptive measures for future impacts. The need for climate adaptation in developing countries is widely recognised, not just locally, but also on the regional, national and international governmental level as well as by financial donors (Adenle et al., 2017; Bizikova, Parry, Karami & Echeverria, 2015). The United Nations Environment Programme (UNEP) (2016) has estimated that climate adaptation needs in developing countries could rise up to $280-$500 billion per year by 2050. Although it is not enough to fully adapt to climate change, finance through the UNFCCC will help towards gathering these amounts (Fankhauser & Schmidt-Traub, 2011). Notwithstanding the increased financial commitment by the UNFCCC, climate adaptation and mitigation in the developing world present a difficult challenge. The question remains whether this financial contribution is effective in terms of reaching the underlying mitigation and adaptation goals. This thesis will explore that question of whether climate-related mitigation and adaptation finance are effective.
The rest of this thesis is organised as follows. In the next section, the conceptual framework will be developed based on current literature on the topic of climate aid effectiveness. Afterwards, the data and methods used for the empirical model will be discussed. The results of the study are presented in the section thereafter. The final section of this thesis discusses the results, policy implications and limitations, and concludes this thesis.
CONCEPTUAL FRAMEWORK
Whether development aid is effective in the economic sense, is a widely researched topic with controversial results2. Besides the overall effectiveness, many have studied the conditions under which development aid is distributed. For instance, Hansen and Tarp (2000) find that aid effectiveness varies with the size of the financial transfer. Moreover, institutional quality is frequently discussed in aid effectiveness literature. Burnside and Dollar (2000) asserted for instance that aid is more effective when accompanied by sufficient governance and good policy.
However, research on climate-related development aid effectiveness is distinguished from general aid effectiveness in different ways. Firstly, where aid effectiveness generally focuses on the economic development of a country, climate-related aid is intended to make a country
more resilient to climate change impacts and reduce greenhouse gas emissions. Secondly, development aid has taken many forms over the past decades, equipping researchers with a myriad of data to examine, whereas climate-related aid has only recently become a popular topic, especially after establishing the UNFCCC and an increased urge to tackle climate change. Consequently, climate-related aid, specifically related to climate change adaptation and climate change mitigation, has only been officially registered since the early 2000s, through the OECD Development Assistance Committee (DAC), to monitor the targets set by the UNFCCC. Given the long-term vision of climate adaptation, the short period of time with data to consider may potentially not be enough to draw significant conclusions. Nonetheless, a first indication of potential effectiveness of these relatively new types of aid could emerge.
Despite the academic attention paid to foreign aid effectiveness and its dependencies, less literature is thus available on the effectiveness of climate-related development. Some attempts in this field have been done, but the question whether aid specifically aimed at climate change adaptation and mitigation is effective in reducing greenhouse gas emissions and reducing the recipient country’s vulnerability to impacts from climate change remains insufficiently answered (Eyckmans et al., 2016). Moreover, the studies which do address this topic have not made use of the data which is now available from the OECD DAC. As a few years have passed since the monitoring has been introduced of this type of aid, related to assisting developing countries with climate change adaptation and mitigation, the opportunity presents itself to do a first assessment of the potential effects this aid has on the recipient country, using the data gathered by the OECD DAC.
One introduction of environmental vulnerability as a factor in aid effectiveness research is done by Guillaumont and Chauvet (2001). They explore the possibility that countries which suffer from a combination of high vulnerability to impacts from climate change and slow economic growth, experience more benefits from the aid they receive in terms of economic development. Although they find evidence that effectiveness of development aid is relatively high in countries which suffer from high vulnerability to climate change impacts, it does not address the question of whether this accelerated economic development in turn also affects the resilience to impacts from climate change in these countries. Nonetheless, an implication of their results is that there might be interesting dynamics between aid and effects on climate-related developments.
More recently, Eyckmans et al. (2016) also developed an economic estimation model of the impacts of mitigation and adaptation transfers from rich countries to poor countries. In accordance with the findings from Tol (2005), they also find that these transfers are effective to a certain extent in terms of emission reduction and boosting climate-resilience. However, they also suggest there are barriers to the efficiency of isolated transfers related to climate change mitigation and adaptation, such as the recipient country responding with reallocating their own resources towards other development objectives rather than climate change mitigation or adaptation, which might offset the effects of the foreign transfers. Nonetheless, the modelling suggests that effectiveness of the transfers as they are may still be positive.
One specific topic within the literature on general efforts to tackle the effects of climate change in developing countries is the implementation of Clean Development Mechanism (CDM) projects in Sub-Saharan Africa. Research on the effectiveness of these projects has been done, of which the results imply some indication for the effectiveness of the broader term of climate-related development aid. The CDM was introduced in the Kyoto Protocol, and presents an incentive for developing countries to contribute to combating global climate change. The initiative is done by developed countries which have made commitments under the Kyoto Protocol to limit their emissions. Implementing a CDM project in the development world yields certified emission reduction (CER) credits, which can be used towards meeting the Kyoto targets (UNFCCC, 2018c). Timilsina, De Gouvello, Thioye and Dayo (2010) examine the technical potential of greenhouse gas emissions reduction through implementing CDM projects in Sub-Saharan Africa. Their findings indicate notable greenhouse gas emission reduction potential, which implies that implementing such mitigation projects are effective in terms of reducing emissions. However, the data consists of estimations and is only focused on the energy sector and CDM projects specifically. Other adaptation and mitigation-related aid projects also include investments in different industries and aspects of climate change resilience building, such as infrastructure and disaster risk management. Together with the suggestions by Eyckmans et al. (2016) and Tol (2005), the results of Timilsina et al. (2010) would, however, suggest that aid related to climate change mitigation and adaptation has some degree of desired outcome in terms of fewer greenhouse gas emissions and stronger climate change resilience.
Overall, therefore, in previous literature there is some suggestion which might indicate that mitigation and adaptation transfers are effective in its intended goal, namely reducing greenhouse gas emissions and reducing the recipient country’s vulnerability to impacts from climate change. Therefore, I would expect to find a negative relationship between mitigation aid and greenhouse gas emissions, and a negative relationship between adaptation aid and the overall vulnerability to climate change impacts in the recipient country. This results in the following hypotheses, for mitigation and adaptation efforts respectively:
Hypothesis 1a. Mitigation-related development aid reduces greenhouse gas emissions in the recipient country.
The argument outlined above has so far only considered the independent effect of mitigation and adaptation aid on environmental indicators. However, there are some suggestions in literature that the degree in which climate-related projects are implemented effectively might be in part dependent on the capacity of local institutions to enable the implementation. In general aid effectiveness literature, Burnside and Dollar (2000) have for instance suggested that good governance and policy increase the probability of effective implementation of development aid. This topic is frequently discussed in general aid effectiveness literature, but as the topic of climate financing effectiveness is relatively new, it follows logically that limited research is done on which conditions affect this effectiveness of climate financing as well. Some indications are put forward in existing literature, which provide the theoretical argument to test whether climate-related aid effectiveness, in terms of greenhouse gas emissions reduction and improved climate change resilience, is dependent on the degree of government effectiveness in the recipient country.
One such indication that institutional capacity in the recipient country may be a facilitating factor in the implementation of climate-related development aid, comes from Adenle, Manning and Arbiol (2017). They did a study on the participation of African countries in climate change mitigation programs, consisting of low-carbon development programs which are necessary to abate emissions also in the developing world. Finding that poorer levels of institutional quality correspond to fewer programs being implemented, they argue therefore that building institutional capacity at different levels of the government is necessary to implement low-carbon projects. However, the study does not give explicit insight into whether the institutional quality at the government level also helps the implemented projects become effective. In general, the goal of these projects is to limit emissions and implement technology which makes the local economy more resilient to more extreme impacts from climate change. Therefore, a next step would be to see if this positive effect of higher institutional capacity transfers to the intended effectiveness of these attempts at adaptation or mitigation of climate change effects.
Descriptively, this argument that institutional quality influences aid effectiveness has been suggested in literature relating to CDM implementation as well. Timilsina et al. (2010) find that there are barriers in Sub-Saharan African countries which limit the potential of CDM effectiveness they established, as mentioned earlier. An important barrier they find is the regulatory capacity in the target country. They argue that addressing institutional needs is necessary to increase the effective implementation of CDM projects. These arguments are not directly tested, but findings imply that there might be a reason to investigate whether there is an interaction between the institutional capacity in a country and their performance in terms of adaptation and mitigation.
facilitate implementation of development projects. The suggested relationship between these factors is further explored by Betzold (2015), who argues that in the case of small island developing states (SIDS), local communities need to be empowered by governmental policies which are effective in terms of enabling local projects being implemented. Using data on mitigation- and adaptation-related development finance and interacting it with a measure of government capacity would allow for testing whether this positive interaction exists in terms of climate aid effectiveness.
Overall, the general suggestion is that the recipient country may benefit from strong governmental effectiveness when receiving aid related to climate change mitigation or adaptation, in the sense that the effects on greenhouse gas emissions and vulnerability to impacts from climate change are larger. For that reason, I would expect to find a positive interaction effect between government effectiveness and the different types of aid donated to the recipient country. Applying this argument to mitigation and adaptation efforts results in the following hypotheses:
Hypothesis 2a. Mitigation-related development aid is more effective in reducing greenhouse gas emissions when the recipient country has a higher level of government effectiveness.
Hypothesis 2b. Adaptation-related development aid is more effective in reducing the risk to harmful climate change impacts when the recipient country has a higher level of government effectiveness.
The remainder of this thesis will revolve around empirically testing the hypotheses presented here. In the next section, the data and methods used for these tests will be discussed.
DATA AND METHODS Variables and data source
Due to the difference in the conceptualisation of the outcome of mitigation and adaptation finance effectiveness, I employ two different outcome variables in two respective models. This section will outline the variables which are used in this thesis and discuss the characteristics of these variables.
Greenhouse gas emissions
are included3. These various gases have different concentrations but are compared by computing their ‘global warming potential’, which is then converted to CO2-equivalent values. This allows for adding up emissions of different greenhouse gases based on their potential impact on the climate.
Climate Risk Index
The Climate Risk Index (CRI) will be used as outcome variable to study the effectiveness of adaptation-related development finance. This measure is taken from Germanwatch. The CRI is a comprehensive measure which weighs economic losses and fatalities resulting from extreme meteorological and climatological events4. It is therefore an indication of the level of exposure and vulnerability to climate-change related extreme events. Country- and year-specific CRI values are available for the years 2010-2016. As adaptation-related finance is reported for the same period, these are the years used as basis for analysis.
Climate-related development finance
The main explanatory variable of interest in this study is climate-related development finance. This data is obtained from the OECD Development Assistance Committee (DAC). In 1998, the OECD DAC introduced the monitoring of development finance related to climate change objectives. Since then, bilateral Official Development Aid (ODA)5 flows related to mitigating and adapting to climate change are being reported through DAC’s Creditor Reporting System (CRS), following the aim to tackle global environmental challenges formulated during the Rio conventions in 1992. Several markers (the so-called “Rio markers”) were developed by the UNFCCC, in order to label ODA flows by sub-goal within climate change objectives. The three initial markers are biodiversity, desertification and climate change mitigation, to which a fourth one was added in 2009 to label flows for climate change adaptation. These markers are indicated by the donating country, which applies a scoring system to each development finance activity (ODA flows) to establish whether it is targeted at one of the four markers or not. Of the four markers, two are studied in this paper, separately. The first one is climate change mitigation, the second is climate change adaptation. These two markers are used most frequently and therefore provide a relatively large amount of data, in comparison to the biodiversity and desertification markers, which are not within the scope of this thesis. Of mitigation finance flows, approximately 65% consists of loans, 30% of grants and the rest are different financial instruments6 (OECD DAC, 2018). Of adaptation finance flows, these numbers are 41% and 57% for loans and grants respectively. The definition and criteria for climate change adaptation and mitigation markers are outlined further below.
3 F-gases include hydrofluorocarbons (HFCs), perfluorocarbons (PFCs) and sulfur hexafluoride (SF6).
4 Meteorological events include tropical storms, winter storms, severe weather, hail, tornados and local storms. Climatological events are occurrences such as freezing, wildfires and droughts. Hydrological events such as storm surges, river floods, flash floods and landslides are also included. Source: Germanwatch (2018).
5 ODA is defined as “government aid designed to promote the economic development and welfare of developing countries” (OECD, 2018). ODA includes grants, loans (of which at least 25% is a grant element) and technical assistance provision. It is either provided bilaterally or through a multilateral institution such as the UN or the World Bank.
6 Other financial instruments can be equity and shares in collective investment vehicles, and debt relief. The
Mitigation-related development finance, which is the term for ODA flows that are coded as being aimed at climate change mitigation, is the main explanatory variable in the model assessing the effectiveness of mitigation aid. Aid flows recorded as being significantly or principally related to climate change mitigation are defined as those which contribute to stabilising concentrations of greenhouse gases in the atmosphere. ODA flows are marked with the climate mitigation marker if at least one of the following criteria is met: it contributes to limiting anthropogenic emissions of greenhouse gases; it contributes to protecting or enhancing GHG reservoirs; it contributes to institution building, developing capacity or strengthening policy frameworks to integrate climate change concerns into the recipient’s development objectives; or it contributes to meeting developing countries’ obligations under the UNFCCC. From the recipient perspective, the resulting measure is the total amount of mitigation-related ODA received according to this marker, measured in thousand US dollars. Data on mitigation-related development finance was first recorded in 2000, the most recent observations being from 2016.
Adaptation-related development finance, which is the term for ODA flows that are coded as being aimed at climate change adaptation, is the main explanatory variable in the model which examines the effects of adaptation aid. Defined as aid flows aimed at increasing resilience of human and natural systems to the impacts of climate change and its risks, climate change adaptation has become mandatory to report on for ODA from 2010 onwards. As a result, data on adaptation-related aid is available for the years 2010 to 2016. Criteria for labelling ODA flows with the adaptation markers are: the adaptation objective is explicitly stated in the documentation of the activity; or the activity includes measures which are specifically targeted at increasing the above-mentioned resilience. This could include a broad range of activities, such as investments in infrastructure, enhancement of disaster risk management, and increasing agricultural efficiency. Again from the recipient perspective, the resulting measure is the total amount of adaptation-related ODA received according to this measure, in thousand US dollars.
Government Effectiveness
Government effectiveness is the measure used to test the hypothesis that the effectiveness of climate-related development finance is dependent on the government effectiveness in the recipient country. Therefore, government effectiveness is included as an interaction term with mitigation-related aid and adaptation-related aid in the two respective models. The government effectiveness indicator is taken from the World Governance Indicators (WGI) reported by World Bank. The WGI consists of six different indicators for institutional quality. Government Effectiveness is one of them7, which is an index representing the perception of the quality of public and civil services, of the independence of political pressures, and of the government’s ability to formulate, implement and credibly commit to qualitative policies. Values range between -2.5 and 2.5, with a higher value indicating better government effectiveness.
Control variables
Based on previous literature, the control variables included in all models are population, gross domestic product (GDP) per capita, value added in agriculture and value added in industry (see e.g. Lin, Omoju, Nwakeze, Okonkwo & Megbowon, 2016; Lin & Xie, 2014; Ray & Ray 2011). In addition, previous literature indicates that energy use is an important factor to control for as well when considering greenhouse gas emissions in particular. (e.g. Omri, 2013). Data for all control variables is taken from World Bank. The expected relationships between the control variables and the two outcome variables are elaborated below.
The variables relating to population, GDP and energy use are in line with theory describing the drivers of climatological change in terms of increased greenhouse gas emissions8. Population is one of the most important drivers of climate change, as a higher amount of people corresponds to more use of resources and land, higher consumption and production levels, and as a result of this, more pollution due to the negative environmental externalities associated with production (e.g. Ray & Ray, 2011). GDP per capita also is shown to have an adverse effect on the environment, as this is usually accompanied by increased industrial production, infrastructure investments, and increased consumption demand, which in turn increases pollution (e.g. Lin and Xie, 2014). Energy use, or the amount of energy consumption needed to produce a certain level of GDP is a measure of energy efficiency, is shown to positively correlate with the level of greenhouse gas emissions, as a higher energy use produces more stress on the environment (Omri, 2013). Value added in the sectors agriculture and industry are included to control for the varying levels of polluting activities between industries (e.g. Stern, 2004), as especially (heavy) industry produces more carbon dioxide emissions, and agriculture includes the presence of high levels of methane emissions. Value added is defined as the net output of a sector, which is all outputs added together, subtracted by all intermediate inputs (World Bank, 2018). Therefore, a positive relationship is expected here as well.
Similar to the expected effect on the level of emissions, population and GDP per capita also are shown to increase general vulnerability to impacts from climate change, or the climate risk index as it is conceptualised as in this thesis (e.g. Lin et al., 2016). Moreover, Germanwatch (2018) argue themselves that the climate risk index is influenced by the population and GDP per capita in the country. Moreover, industries vary in terms of vulnerability to climate change, with specifically agriculture being inherently vulnerable to weather patterns and changes therein (Haile, 2005). For that reason, value added in both agriculture and industry are included as control variables in the models using CRI as the outcome variable.
Sample and dataset
The resulting dataset is a panel consisting of 155 countries. These 155 countries are those which are recorded to have received either type of climate-related development finance (mitigation
and/or adaptation). The countries are specified in appendix A. The most recent data available for total GHG emissions is until 2012. Mitigation-related finance markers have been applied to ODA since 2000. Therefore, the models assessing mitigation-related aid effectiveness consider the period from 2000 to 2012 as source of data. Country- and year-specific CRI values are available for the years 2010-2016. As adaptation-related finance is (mandatorily) reported for the same period, these are the years used as source for analysis of adaptation-related development finance effectiveness. An overview of all variables included in the dataset can be found in table 1.
Estimation method
This paper examines the effectiveness of both mitigation-related and adaptation-related development finance, using GHG emissions and the Climate Risk Index as outcome variables respectively. For these purposes, the econometric method of panel data analysis is employed. This technique has various benefits relative to using time-series or cross-sectional data only. Estimation results are generally superior (Hsiao, 2003) and, according to Baltagi (2005), panel models improve heterogeneity and multicollinearity concerns. Moreover, Baltagi asserts that estimation efficiency is higher in panel models and that they identify effects that time-series models cannot detect.
For panel data models, several estimation techniques are available. The most used are pooled Ordinary Least Squares (OLS), fixed effects and random effects model estimators (Hill, Griffiths & Lim, 2012). Pooled OLS is an estimation technique which pools data on different individuals together, without considering any differences between these individuals which might lead to varying coefficients between individuals in the panel. For pooled OLS coefficient estimations to be consistent, the assumption must be met that all error terms corresponding to the same individual are uncorrelated. However, it is unrealistic to assume that there are no
Table 1. Variables used in the study
Variable Unit of measurement Source*
Greenhouse gas emissions Log Kilotons of CO2 equivalents WB
Climate Risk Index Log Index score GW
Mitigation-related development finance
Log Thousand US$ (constant
2016)
OECD DAC
Adaptation-related development finance
Log Thousand US$ (constant
2016)
OECD DAC
Government Effectiveness Index score WB WGI
Total population Log Number of people WB
GDP per capita Log US$, constant 2010 WB
Energy Use Log KG of oil equivalent per 1000
US$ GDP
WB
Value added in agriculture Log Million US$ (constant 2010) WB
Value added in industry Log Million US$ (constant 2010) WB
unobserved individual characteristics included in the error term, leading to a violation of the assumption that the error terms corresponding to the same individual are uncorrelated. In this study, country-level data is used, for which it is highly plausible to assume the presence of unobservable heterogeneity, such as deeply rooted cultural values and other attributes that may lead to differences between countries that are not included in the empirical model. Therefore, employing a pooled OLS method would most likely render biased results and it is preferred to use a different method.
Panel methods which allow for unobserved heterogeneity across individuals are the random effects and fixed effects estimations (Hill et al., 2012). In the case of random effects estimation, the unobserved heterogeneity between individuals is assumed to be captured in the individual intercept and assumes the random error does not correlate with any of the explanatory variables in the model, implying full exogeneity of the explanatory variables included in the model. Fixed effects estimations also allow for the error term to be correlated with time-invariant heterogeneity between individuals, which, as explained before, is important to allow in the context of assessing country-level data. The difference between random and fixed effects estimations is that with fixed effects, the assumption of full exogeneity is relaxed, such that it takes into account that the error term might include unobserved factors that correlate with the explanatory variables. Since the assumption of the error term not being correlated with the explanatory variables is often violated, using random effects estimations may produce biased and inconsistent coefficient estimations. However, if the assumption is not violated, the random effects method would be preferred as it produces better estimates with smaller standard errors.
Whether random or fixed effects is more appropriate, can be tested by performing the Hausman test. This test assesses whether there are systematic differences between the estimated parameters of the random and fixed effects models. The null hypothesis for the Hausman test is that the difference in coefficients between the fixed and random effects model is not systematic. At a statistical significance level of 5%, this hypothesis is rejected, and it is assumed that the difference in coefficients is systematic, in which case random effects estimations may be biased and the fixed effects model is preferred. The results of the Hausman test are displayed in appendix B1. The results indicate that for the models involving mitigation finance effectiveness (hypothesis 1a and 2a), the fixed effects model is preferred as there are systematic differences between the estimated parameters of both techniques. For the models involving adaptation finance effectiveness (hypothesis 1b and 2b), the null hypothesis is not rejected, and the test implies that estimations of the random effects model are consistent and unbiased. Nonetheless, some caution with this outcome must be noted. It may not be realistic to assume that in the models including adaptation aid no unobserved heterogeneity is present, as there are most likely unobserved county characteristics in the error term that may correlate with the explanatory variables and which can result in biased estimations when using the random effects method. For these reasons, fixed effects will be employed for all hypotheses. Random effects estimations are also included in the results, but caution is required for interpretation thereof.
observations for an individual are uncorrelated. However, if this assumption is violated, serial correlation would be present in the error term (Wooldridge, 2002). Since the dataset contains a time-series for each country, a test for such autocorrelation in the error term is performed. The Wooldridge test tests the null hypothesis that there is no first-order autocorrelation in the error term and produces a Wald statistic. The results of the Wooldridge test can be found in appendix B2. The results suggest that for the models including adaptation finance (hypothesis 1b and 2b), there is no statistically significant evidence at the 10% statistical significance level for presence of autocorrelation in the error term. However, at the same time, these models only consider a small number of periods (2010 to 2016), which may decrease the power of the Wooldridge test (Drukker, 2003). Therefore, I take into account that the assumption that there is no autocorrelation in the error term may be violated when assessing the fixed effects models.
One potential remedy for autocorrelation is to estimate a dynamic model. This means that the lagged value of the outcome variable is added as an explanatory variable in the equation, which allows to study the impact of previous values of the outcome variable on subsequent values of that same variable (Hill et al., 2012). However, due to the limited number of time periods in the dataset considered in this thesis, employing this technique leads to biased estimations of the coefficient of the lagged outcome variable, as argued by Nickell (1981). Therefore, this is not a suitable remedy for autocorrelation.
Alternatively, re-estimating the fixed effects models using first differences presents a better solution. The first differences method takes the difference between the observation in time t and the observation in time t-1 for all variables and runs the regressions on these differences. In addition to allowing for the error term to be correlated with time-invariant heterogeneity between individuals, using first differences also controls for serial correlation in the error term. In order to take the possibility of autocorrelation being present in the data, as demonstrated by the Wooldridge test results, the models will be re-estimated using the first differences method.
Endogeneity concerns
value of the explanatory variable has a statistically significant effect on the future values of the outcome variable, then the relationship is determined as Granger-causality. The theory was first introduced by Granger (1969). Both models with and without lagging the explanatory variables of interest will be shown, in order to assess the presence of reverse causality.
Another concern is that of omitted variable bias. Both unobservable variant and time-invariant variables may cause problems in interpreting the results of the estimations. Unobservable time-invariant heterogeneity between different recipient countries is controlled for by using the fixed effects model, as it captures all country-specific characteristics that do not generally vary over time, such as culture, language, or geographical location. To control for aggregate time-varying effects that affect all countries in the sample, year fixed effects are added to each regression, capturing aggregate trends such as global economic shocks or other events which are common to all countries in the sample. Year fixed effects are included by means of a dummy variable for each year included in the analysis. Moreover, a second estimation using first differences, besides also controlling for unobservable time-invariant heterogeneity, further controls for potential serial correlation. Lastly, in each model the coefficients are estimated using country-clustered standard errors, relaxing the assumption that observations within groups (recipient countries) are independent, which means that the standard errors allow for intra-country correlation.
Model specification
As outlined above, the model of interest includes lagged values of the explanatory variable of interest. Therefore, the baseline fixed effects model to test the hypothesis for mitigation-related finance effectiveness is constructed as the following:
𝐺𝐻𝐺𝑖,𝑡 = 𝛽0+ 𝛽1𝑀𝐹𝑖,𝑡−1+ 𝑋′𝑖,𝑡𝜃 + 𝜇𝑖+ 𝛾𝑡+ 𝜀𝑖,𝑡 (1)
In which GHGit represents the level of greenhouse gas emissions in country i in year t, MFi,t-1
represents the amount of mitigation-related development finance received by country i in year t-1, and thus the lagged value of mitigation aid, X’i,t represents the vector of control variables
for country i in year t as specified earlier, μi includes country fixed effects, γt includes year
fixed effects, and εi,t denotes the error term.
As explained before, the model for adaptation-related development finance effectiveness includes a different outcome variable. Similar to the model in equation (1), the main model of interest includes the lagged value of the explanatory variable of interest. Therefore, the baseline model for the adaptation-related finance effectiveness model is the following:
𝐶𝑅𝐼𝑖,𝑡 = 𝛽0+ 𝛽1𝐴𝐹𝑖,𝑡−1+ 𝑍′𝑖,𝑡𝜑 + 𝜇𝑖+ 𝛾𝑡+ 𝜀𝑖,𝑡 (2)
In which CRIi,t represents the Climate Risk Index for country i in year t, AFi,t-1 represents the
in year t as specified earlier, µ1 includes country fixed effects, γt includes year fixed effects, and
εi,t denotes the error term.
The second hypothesis to be tested is that government effectiveness influences the effectiveness of mitigation- and adaptation-related development finance, namely that when government effectiveness is higher, effectiveness of climate-related aid is higher in terms of reducing impacts from climate change. Therefore, government effectiveness is included in the model as an interaction effect with the variables for climate-related development finance. Due to concerns about reverse causality, also in this case the models including lagged values of the explanatory variables of interest (climate aid and its interaction with government effectiveness) are of main interest. When including the lagged effect of government effectiveness and the interaction term into the mitigation-related finance effectiveness model, the following equation results:
𝐺𝐻𝐺𝑖,𝑡= 𝛽0+ 𝛽1𝑀𝐹𝑖,𝑡−1+ 𝛽2𝐺𝐸𝑖,𝑡−1+ 𝛽3(𝑀𝐹𝑖,𝑡−1× 𝐺𝐸𝑖,𝑡−1) + 𝑋′𝑖,𝑡𝜃 + 𝜇𝑖+ 𝛾𝑡+ 𝜀𝑖,𝑡 (3)
In which GHGi,t, MFi,t-1, X’i,t, μi, γt and εi,t denote the same as in equation (1), GEi,t-1 represents
government effectiveness for country i in year t-1, and MFi,t-1×GEi,t-1 represents the interaction
between mitigation-related finance and government effectiveness for country i in year t-1.
Including the same factors for the adaptation-related development finance model, in order to test hypothesis 2b, gives the following equation:
𝐶𝑅𝐼𝑖,𝑡 = 𝛽0+ 𝛽1𝐴𝐹𝑖,𝑡−1+ 𝛽2𝐺𝐸𝑖,𝑡−1+ 𝛽3(𝐴𝐹𝑖,𝑡−1× 𝐺𝐸𝑖,𝑡−1) + 𝑍′𝑖,𝑡𝜑 + 𝜇𝑖+ 𝛾𝑡+ 𝜀𝑖,𝑡 (4)
In which CRIi,t, AFi,t-1, Z’i,t, µ1, γt , and εi,t denote the same as in equation (3), GEi,t-1 represents
government effectiveness for country i in year t, and AFi,t-1×GEi,t-1 represents the interaction
between adaptation-related finance and government effectiveness for country i in year t-1.
Equations (1), (2), (3) and (4) all represent fixed effects regression models to test hypotheses 1a, 1b, 2a and 2b respectively. In the empirical results section, a descriptive analysis of the data will be done first. Afterwards, the results of the models will be shown. Moreover, the results without lagging the variables mitigation finance, adaptation finance and government effectiveness are also shown in order to explicitly address the reverse causality concerns.
EMPIRICAL RESULTS Descriptive analysis
Table 2. Descriptive statistics
Variable* Mean St. Deviation Minimum Maximum
Greenhouse gas emissions 204,055.00 823,771.5 4.90 12,500,000
Climate Risk Index 76.10 33.07 2.17 126.17
Mitigation aid 91,723.39 308,355.60 0.157 5,410,978 Adaptation aid 77,186.81 145,411.60 1.089 1,241,892 Government effectiveness -0.49 0.67 -2.45 1.88 Population 38,900,000 151,000,000 9420 1,390,000,000 GDP per capita 4,449.19 5,178.20 193.87 43,942.94 Energy use 157.65 118.42 4.55 824.23
Value added in agriculture 14,341.94 57,260.10 5.97 762,307
Value added in industry 55,337.74 264,399.90 1.32 4,715,860
*See table 1 for units of measurement
the aid variables, this could imply that some countries have received hardly any development finance whereas some countries have received large sums. Another, very likely, possibility is that over the years, more development finance has been allocated to developing countries with relation to climate mitigation or adaptation. This development is displayed in figure 1. In the figure it seems that the year 2009 is a turning point; from then onwards, a surge in mitigation-related finance materialises. The adaptation marker has only been applied from 2010, but fast growth in adaptation aid is clear. This development could potentially be explained by the implementation of the Copenhagen Accords in 2009, when developed countries agreed to increase their efforts in supporting developing countries financially with mitigating and adapting to climate change impacts (UNFCCC, 2010). Given the data it seems that countries have indeed increased their efforts in allocating development aid for climate mitigation and adaptation purposes.
Figure 2 displays yearly average mitigation aid in the sample countries again and adds the average greenhouse gas emissions per year in the sample countries on the right-side axis. Comparing both lines, one can see that both mitigation aid and greenhouse gas emissions on
5 10 15 20 25 30 35 40 Millio n USD
Figure 1. Mitigation and adaptation finance
average are generally increasing. After 2007, there is a one-time decrease in average greenhouse gas emissions. Mitigation aid also increases every year with the exception of one year, which in this case is 2011. At first sight, there is no clear indication that mitigation aid leads to a decrease in the (growth rate of) greenhouse gas emissions. Moreover, as data on greenhouse gas emissions is not up to date yet, it is unclear what the development after 2012 in greenhouse gas emissions is. The lack of graphical indication that mitigation aid has a reducing effect on greenhouse gas emissions may have several implications. Firstly, it could point to a confirmation of reverse causality concerns, as indicated in the previous section. The possibility that mitigation aid is distributed according to whether a recipient country needs it (in terms of high levels of emissions) is explored in the results section too, by showing the models with and without lagging the explanatory variables of interest. Secondly, it might be needed to take a more long-term approach to the question whether mitigation aid is effective in reducing greenhouse gas emissions. However, this requires a longer span of data, which is unavailable at this time, as efforts to transfer climate-related development finance to developing countries are relatively recently introduced after multiple UNFCCC conferences, as outlined in the introduction.
In figure 3, adaptation-related aid is displayed again and on the right-side axis, the average Climate Risk Index in the sample countries is displayed. The average CRI does not show a clear trend over the six years in the graph. Adaptation aid has been quite steadily increasing since 2010. In this case too, it is not immediately clear if there should be a relationship between both variables. The hypotheses will be tested empirically in the next section, which will shed further light on whether there is a relationship between mitigation- and adaptation-related development finance and greenhouse gas emissions and the CRI respectively.
After testing for the level of normal distribution of each of the variables in terms of skewness and kurtosis, and taking into account the nature of the data, a log transformation is performed for the variables GHG emissions, CRI, mitigation finance, adaptation finance, population, GDP
0 50 100 150 200 250 300 0 50 100 150 200 250 300 T h o u san d k ilo to n s o f C O2 eq u iv alen ts Millio n USD
Figure 2. Average mitigation aid and greenhouse gas emissions in sample countries
per capita, energy use, value added in agriculture and value added in industry. In appendix C, a table with skewness and kurtosis levels is displayed. The results show that the log transformation has improved the distribution of the data generally, with scores on skewness and kurtosis closer to the desired 0 and 3 respectively.
In the next section, the results for the fixed effects regressions are displayed and discussed.
Regression results of the fixed effects model
Mitigation-related development aid effectiveness (hypothesis 1a and 2a)
Table 3 shows the results of the fixed effects and random effects model for hypothesis 1a and 2a, for mitigation-related development finance with total greenhouse gas emissions as outcome variable. There is no consistent statistical significance for the effect of mitigation aid on greenhouse gas emissions across all six regressions in the table. In other words, these regressions do not provide robust support for the hypothesis that mitigation aid has a reducing effect on the amount of greenhouse gas emissions. Across the six models, the explanatory power does not vary much, as seen from the within adjusted R2. The values range between 0.196 and 0.234, meaning that in for instance column 1, approximately 23.4% of the variance in the outcome variable is explained by the variance in the explanatory variables included in the model.
Firstly, discussing the control variables, it is shown that energy use has a statistically significant positive coefficient across all models, at the 5% statistical significance level at least. This implies that countries that use more energy to produce a unit of GDP generally have higher levels of greenhouse gas emissions. This is in line with expectations, as for instance argued in a study by Omri (2013), that more intense use of energy produces more stress on the environment in terms of pollution. Moreover, GDP per capita is displayed to have a positive and statistically significant coefficient in columns 2, 4, 5 and 6. This finding is not robust across all models, but it does imply that higher levels of GDP are on average associated with higher
0 10 20 30 40 50 60 70 80 90 0 20 40 60 80 100 120 140 2009 2010 2011 2012 2013 2014 2015 2016 In d ex Millio n USD
Figure 3. Average adaptation aid and climate risk index in sample countries
levels of greenhouse gas emissions. This is also according to expectations, as higher levels of GDP per capita are generally associated with relatively higher demand for production and consumption, which are inherently associated with more pollution (Lin and Xie, 2014). In the last two columns, which display the results of the random effects estimations, population also has a positive and statistically significant effect on greenhouse gas emissions. As argued by Ray & Ray (2011), population is an important driver of emissions, as more people together use more resources which are associated with impact on the environment in terms of pollution.
Considering the coefficient estimations of interest in more detail, columns 1 and 2 show the model including mitigation aid as explanatory variable of interest, without and with lagging mitigation aid respectively. In neither of the models a statistically significant effect of mitigation aid on greenhouse gas emissions is shown. This implies that there is no statistically significant support for hypothesis 1a.
Table 3. Regression results fixed effects model Outcome variable:
Total GHG emissions†
Fixed effects Random effects
(1) (2) (3) (4) (5) (6)
Mitigation finance† -0.005 [0.003]
-0.001
[0.005]
Lagged Mitigation finance†
0.005 [0.004] 0.009** [0.004] 0.005 [0.004] 0.012*** [0.004] Government effectiveness -0.121 [0.098]
Lagged Government effectiveness
-0.137** [0.059]
-0.177*** [0.063] Mitigation finance† × Government
effectiveness
0.007
[0.008]
Lagged Mitigation finance† ×
Lagged Government effectiveness
0.005 [0.006] 0.011* [0.006] GDP per capita† 0.296 [0.205] 0.453* [0.234] 0.336 [0.217] 0.540** [0.240] 0.390* [0.201] 0.457** [0.211] Population† -0.257 [0.323] -0.068 [0.384] -0.238 [0.351] -0.041 [0.405] 0.925*** [0.196] 0.957*** [0.205] Energy use† 0.244** [0.094] 0.341*** [0.117] 0.193** [0.088] 0.330*** [0.117] 0.426*** [0.091] 0.415*** [0.091] Value added in agriculture† -0.012
[0.095] -0.057 [0.094] -0.043 [0.103] -0.039 [0.100] -0.019 [0.082] -0.010 [0.085] Value added in industry† 0.091
[0.124] 0.000 [0.163] 0.059 [0.126] -0.040 [0.171] 0.100 [0.146] 0.066 [0.154] Constant 9.918** [4.366] 8.065 [5.136] 10.853** [4.761] 7.433 [5.281] -11.291*** [1.147] -11.766*** [1.118] No. of observations 958 881 900 823 881 823 No. of countries 106 104 106 104 104 104 No. of periods 12 12 11 11 12 11
Country fixed effects Yes Yes Yes Yes No No
Year fixed effects Yes Yes Yes Yes Yes Yes
Within R2 0.234 0.230 0.200 0.224 0.200 0.196
Within Adjusted R2 0.219 0.215 0.183 0.207 n/a n/a
†These variables are logged.
Next, columns 3 and 4 show the models which test the second hypothesis (2a), namely that mitigation aid is more effective when government effectiveness in the recipient country is higher, with and without lagging the variables of interest respectively. From column 3, no statistically significant evidence of a relationship between mitigation aid, government effectiveness and greenhouse gas emissions are shown. However, in column 4, both the lagged mitigation aid and government effectiveness variables display statistically significant coefficients. Mitigation aid shows a positive coefficient of 0.009 at the 5% statistical significance level. This implies that on average, increasing aid by one percent in one year corresponds to a 0.009 percent increase in greenhouse gas emissions the following year, ceteris paribus. At the 5% statistical significance level, the coefficient of -0.137 for government effectiveness is negative and implies that increased government effectiveness is associated with lowering levels of greenhouse gas emissions. Although both mitigation aid and government effectiveness have an independent effect in this model, there is no statistical evidence of a joint effect, as shown by the statistically insignificant coefficient for the interaction term between the two variables.
Contrary to the hypothesised relationship, the result for mitigation aid implies that mitigation aid has an aggravating effect on greenhouse gas emissions in the recipient country. A possible explanation would be to consider the presence of reverse causality. As suggested before, giving aid is not random. A myriad of literature is available on the allocation of development aid, and on what the considerations are by donor countries to allocate relatively more or less aid to certain countries. In the context of climate change mitigation, Halimanjaya (2015) finds that, among other determinants, CO2 intensity in a developing country is a statistically significant determinant of the amount of mitigation finance allocated there by donor countries. The level of CO2 emissions, however, is not specifically found to be a determinant of aid allocation. Nonetheless, as energy use is included in the models here as a control variable for energy intensity, this allows for seeing whether energy intensity (measured in terms of energy needed to produce a unit of GDP, which corresponds to more emission-prone economic activity) is associated with higher levels of greenhouse gas emissions. Energy use displays a statistically significant positive coefficient of 0.330 at the 1% statistical significance level, implying that countries that use more energy to produce a unit of GDP, on average have higher levels of greenhouse gas emissions. The positive association between levels of greenhouse gas emissions and energy intensity found here, and the positive association between emission intensity and aid allocated to the country, as found by Halimanjaya (2015), might be reason to consider that the result found points to a reverse relationship. Possibly, donor countries allocate mitigation-related development aid according to which countries display prospects of high levels of greenhouse gas emissions. However, the results do not directly test these relationships, thus one can only speculate based on the results above. In future research, it could be interesting to see whether such a relationship exists between predicted levels of emissions, energy intensity and aid allocated for emission mitigation.
country to reallocate their own resources so as to achieve their own preferred balance in mitigation and other objectives such as adaptation or increased consumption. Especially in the latter case, receiving mitigation aid may result in the recipient (developing) country responding with domestic economic measures, e.g. making sure that subsistence levels of income for inhabitants are met, which increase the consumption of goods that may in turn have negative external effects on the environment. However, this dynamic is also not directly tested in the model and one can only speculate within the limits of this study.
As noted, the coefficient for government effectiveness is negative and statistically significant. This implies that an increase in government effectiveness corresponds to a decrease in greenhouse gas emissions, on average. This relationship was not the main interest in this research, as the main coefficient of interest is the interaction term between government effectiveness and mitigation-related aid. Nonetheless, the independent effect of government effectiveness has interesting implications. It implies that countries with stronger institutions, among which is government effectiveness, produce fewer greenhouse gas emissions. This is in line with findings by for instance Tamazian and Rao (2010), who find evidence of institutional quality having a positive effect on the environment in terms of lower per capita CO2 emissions. Although in this model both mitigation aid and government effectiveness have an independent effect, the interaction term between these two variables and therefore their joint effect is not statistically significant. In that sense, the hypothesis 2a is not supported by the results in this model.
mechanism of the interaction term with government effectiveness may also need to be reconsidered.
In summary, no robust statistically significant effects of mitigation aid on greenhouse gas emissions emerge. This implies that there is no support for hypothesis 1a, which says that increased mitigation-related development finance reduces the amount of greenhouse gas emissions in the recipient country. Moreover, hypothesis 2a is also not supported. There is no statistically significant evidence that when aid is increased in countries with higher government effectiveness, the impact on greenhouse gas emissions is larger.
Adaptation-related development aid effectiveness (hypothesis 1b and 2b)
The results of the fixed effects models to test hypotheses 1b and 2b, for adaptation-related development finance with the outcome variable Climate Risk Index, are shown in table 4. Across all columns, adaptation aid has a statistically significant effect on the climate risk index, albeit it weak at the 10% statistical significance level and with varying coefficient signs, depending on whether adaptation aid is lagged or not. In other words, there is some weak Table 4. Regression results fixed effects model
Outcome variable:
Climate Risk Index†
Fixed effects Random effects
(1) (2) (3) (4) (5) (6)
Adaptation finance† 0.022*
[0.013]
0.024*
[0.014]
Lagged Adaptation finance†
-0.025* [0.014] -0.028* [0.015] -0.040*** [0.011] -0.038*** [0.012] Government effectiveness -0.007 [0.218]
Lagged Government effectiveness
0.241
[0.228]
-0.107 [0.118] Adaptation finance† × Government
effectiveness
0.006
[0.018]
Lagged Adaptation finance† ×
Lagged Government effectiveness
-0.016 [0.019] 0.006 [0.017] GDP per capita† -0.176 [0.390] -0.074 [0.430] -0.223 [0.389] -0.101 [0.433] -0.164 [0.111] -0.123 [0.103] Population† -0.061 [0.777] -1.243 [0.981] -0.023 [0.776] -1.279 [0.982] -0.261** [0.126] -0.234** [0.119] Value added in agriculture† 0.316
[0.250] 0.364 [0.273] 0.316 [0.252] 0.352 [0.269] 0.033 [0.052] 0.027 [0.052] Value added in industry† -0.031
[0.160] 0.028 [0.183] -0.019 [0.158] 0.022 [0.188] 0.125 [0.087] 0.106 [0.082] Constant 0.185 [12.369] 15.813 [15.906] -0.274 [12.363] 17.060 [16.022] 6.114*** [0.587] 5.875*** [0.560] No. of observations 844 721 844 721 721 721 No. of countries 133 128 133 128 128 128 No. of periods 7 7 7 7 7 7
Country fixed effects Yes Yes Yes Yes No No
Year fixed effects Yes Yes Yes Yes Yes Yes
Within R2 0.026 0.037 0.027 0.038 0.029 0.028
Within Adjusted R2 0.013 0.022 0.011 0.021 n/a n/a
†These variables are logged.
support for the hypothesis that adaptation aid has a reducing effect on the climate risk index of the recipient country, but it matters whether the adaptation aid variable is lagged or not.
Column 1 shows the model which includes adaptation aid as the explanatory variable of interest. Adaptation aid displays a positive coefficient of 0.022 at the 10% statistical significance level. This implies that a 1% increase in adaptation aid corresponds to a 0.022% increase in the climate risk index of the recipient country. The finding of this positive coefficient implies that reverse causality concerns could be realistic, such that there is a possibility that donor countries allocate more aid if the risk to climate change impacts is more severe. This relationship has been suggested before in literature on aid allocation, by for instance Betzold and Weiler (2017). They find evidence that countries with higher levels of vulnerability to impacts from climate change receive relatively higher levels of aid. This implies once more that giving aid is not random and that in the context of climate change adaptation, donor countries may well consider the degree of need for funds in the potential recipient to determine where their resources are allocated.
The lagged value of adaptation aid is introduced in column 2, to see whether adaptation aid has an effect on the climate risk index in the following year. The result is that the sign of the coefficient changes when compared to the model with the non-lagged value of adaptation aid in column 1. In this model, a 1% increase in adaptation aid one year corresponds to a 0.025% decrease in the climate risk index the following year on average, at the 10% statistical significance level. This implies there is some weak support for the hypothesis that increased adaptation aid has a reducing effect on the climate risk index of the recipient country. A further implication of this could be that the projects on which the adaptation aid is spent, have a positive effect on the resilience to climate change impacts of the country. According to the criteria for the adaptation marker according the OECD DAC, as outlined earlier in this thesis, the projects labelled as adaptation-related may fall under for instance infrastructure investments, disaster risk enhancement and improvement of agricultural efficiency. These are all examples of projects that contribute to the economic development of the recipient country. The implication is that these findings complement the conclusions by for instance Tol (2005) and Millner and Dietz (2015), in the sense that economic development and therefore increased access to resources, which increases adaptive capacity, reduce the vulnerability of a country to impacts from climate change.
However, the coefficient estimations of government effectiveness and the interaction term of government effectiveness and adaptation aid are not found to be statistically significant in columns 3 and 4. Therefore, there is no statistical evidence in this model to support the hypothesis that adaptation aid allocated to countries with higher government effectiveness has a larger effect on the climate risk index than aid allocated to countries with lower government effectiveness. Based on these results, one might conclude that this aspect in the recipient country is not of great importance for the effectiveness of climate-related development finance. In general aid effectiveness literature, government effectiveness is sometimes found to be of influence in implementing aid (e.g. Burnside & Dollar, 2000). It appears that effects that have been found in general aid effectiveness literature do not transfer to climate-related development aid effectiveness in this study, or that other country-level factors and developments may be more important in the climate change resilience context, which have not been included in this study but do have an effect on the climate risk index.
Columns 5 and 6 re-estimate hypotheses 1b and 2b using the random effects model, including lags for the explanatory variables of interest. Using random effects confirms the negative effect of lagged adaptation aid on the climate risk index, with negative coefficients at the 1% statistical significance level of -0.040 and -0.038 respectively. In comparison to the fixed effects estimations, the coefficients from the random effects model are larger in absolute terms and have a higher level of statistical significance (1% as opposed to only 10% in the fixed effects model). Next, similar to the fixed effects models, there is no statistically significant evidence of an interaction between government effectiveness and adaptation aid, which implies there is no support for hypothesis 2b. Nonetheless, as mentioned before, the results from the random effects model must be interpreted with more caution than the results from the fixed effects model, as the random effects estimations do not consider the possibility of country fixed effects playing a role in the estimation of the model.
Alternative estimation method
In the previous section, the fixed effects models do not conclusively determine whether there is statistical support for all the hypotheses stated in the beginning of this paper. For that reason, an alternative method is employed to re-estimate the models and determine whether this is a better fit. In this section, first difference models are employed to test whether mitigation and adaptation aid are effective, as well as whether government effectiveness plays a moderating role in this relationship. First difference panel models take the difference of each variable between time t and time t-1’s value, which are then used to estimate the model. The benefit of employing this method is that it controls for autocorrelation. Specifying the equations (1) to (4) again according to this method yields the following equations respectively:
∆𝐺𝐻𝐺𝑖,𝑡 = 𝛽0+ 𝛽1∆𝑀𝐹𝑖,𝑡−1+ ∆𝑋′𝑖,𝑡𝜃 + 𝜇𝑖+ 𝛾𝑡+ 𝜀𝑖,𝑡 (5)
This equation represents the first differences model for hypothesis 1a, in which ∆GHGi,t
