The natural resource curse : an empirical paper analyzing the effect of natural resources on growth and inequality and the role of institutions

(1)

The Natural Resource Curse:

An empirical paper analyzing the effect of natural resources

on growth and inequality and the role of institutions.

– Master Thesis Economics –

August 2017

Steven de Jonge

MSc Economics (Development Economics) 10080384

Supervisor:

Prof. Dr. Menno P. Pradhan

Secondary Reviewer: Dr. Adam S. Booij

Abstract

This paper examines the effect of natural resource abundance on economic growth, vertical inequality and horizontal inequality. The main aim is to separate the effect of natural resource stocks from its relation to institutional quality by using recent resource discoveries. Once these endogeneity issues are addressed, the results show that the sole effect of natural resource abundance on economic growth is positive, thus rejecting the existence of a

resource curse. However, I find no evidence that natural resources have a significant effect on vertical or horizontal inequality. The results of all of the three models indicate that

institutional quality has a great influence on the effect that natural resources have on the outcome variables.

(2)

Statement of originality

This document is written by student Steven de Jonge who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

The natural resource curse, or the paradox of the plenty, refers to the widespread belief that the abundance of natural resources in a country negatively influences economic growth, whereas one would expect the opposite to be true (Sachs & Warner, 1995). Economists are fond of paradoxes, and that is probably one of the reasons why a series of studies of Sachs & Warner (1995, 1999, 2001) that showed evidence of the curse, became popular and were accepted as common knowledge. Since so many poorer countries still have abundant natural resources, it is important to better understand the roots of failure in natural resource-led development. This triggered a lot of new studies to be conducted, because if this curse is really true, it is important to understand how it works and how it can be overturned.

Subsequent studies have been inconclusive about the existence and the severity of the curse, and a number of studies have questioned the robustness of the studies by Sachs & Warner and came up with alternative explanations. The debate on the resource curse has been going on for decades, but it is important to keep addressing it in the contemporary world to better understand the mechanisms at hand.

If the curse exists, what exactly is negatively affected by natural resources? The majority of the literature concerning the natural resource curse focusses on GDP growth as outcome variable. This is seen as the ultimate measurement for economic performance, and thus as a proxy for a country’s wellbeing. But there is more than just GDP. For example, while the effect of natural resources on economic growth is important, one should take into account the concept of convergence. A country with a high standard of living might exhibit slow economic growth, but if this country is abundant in natural resources it can hardly be considered ‘cursed’. On top of that, economic growth and well-being do not always go hand in hand. Botswana for example, has been able to use its natural resource revenues to

achieve high growth rates, but inequality in Botswana has been increasing tremendously due to the unfair distribution of resource rents (McFerson, 2009). In this specific example, the importance of institutions plays a role.

Therefore it is important to address the effect of natural resources on other proxies of performance and well-being as well. Stevens (2003) states that the rents that are

associated with natural resources lead to corruption and rent-seeking behavior, and that as a result of this the wealth from these resources is concentrated in the political elite. If these policymakers only care about themselves, and their interests are not aligned with that of the people in their country, natural resource abundance might increase corruption, leading to increased inequality and therefore have a harmful effect on welfare. The term inequality might not directly provide clarity, since different forms of inequality exist. In this paper I will address a measurement of vertical inequality as well as horizontal inequality. Natural

resource abundance combined with poor institutional quality might increase vertical

inequality by means of the just mentioned arguments. However, horizontal inequality might also be affected. The extractive industry is an industry which requires low-skilled physical labor, mainly done by males. This means that a growing extractive industry might decrease school enrolment rates of males compared to females, which influences horizontal

inequality.

Furthermore, an important and large share of the literature (e.g. Bardhan (1997), Gylfason & Zoega (2002), Leite and Weidmann (2002), Sala-i-Martin and Subramanian (2003), Papyrakis

(5)

2

and Gerlagh (2004), Isham et al. (2005), Basedau (2005), Bulte et al. (2005), Pendergast, Clarke & van Kooten (2011)) stresses the crucial role of institutions and institutional quality in relation to natural resources and economic growth and inequality. The importance of institutional quality for economic growth and wellbeing is unambiguous, the way in which to quantify this as well as its relation to natural resources, is not. Many studies have provided proof of the importance of institutions in this debate, but the relation between institutions and natural resources is complicated due to potential reversed causality and interaction between the two variables. Does an abundance in natural resources result in worse institutions and thus negatively influence welfare? Or do poor institutions enable the abundance of natural resources to have a negative effect on welfare? With all these different potential pathways, the effect of natural resources per se, apart from the role of institutions, remains unclear.

An important side note that has to be made, is that there exists compelling evidence that not all kinds of resources may have the same effect on growth and inequality, especially through the interaction with institutions (Gylfason & Zoega (2002), Leite and Weidmann (2002), Isham et al. (2005), Bulte et al. (2005), Wick (2008), Pendergast, Clarke & van Kooten (2011)). The extraction of fossil fuels (especially crude oil), uranium, copper and other valuable metals (e.g. diamonds and gold) are much more prone to rent-seeking behavior. Extraction of these resources is more spatially clustered than that of other resources, like timber and pasture land. They generate higher rents that are also easier to capture because of their spatial concentration, and thus it is far more likely that these resources end up in the hands of a small minority like a military regime or a royal family (Gylfason & Zoega, 2002).

Without property rights, enforcement of property rights and independent courts (or in other words, without well-functioning institutions) rent seeking leads to lawlessness and corruption to the detriment of the economy, while also increasing inequality (de Soto, 2000). Therefore this distinction between different types of resources is crucial when assessing their effect on economic growth and inequality.

In this paper I want to address the resource curse with respect to economic growth as well as inequality, and the research question that I will attempt to answer is; What is the effect of natural resource abundance on economic growth, economic inequality and social inequality? To address this objective I will perform a cross-section analysis, estimating the effect of resource abundance on average growth and differences in inequality over the period 1995-2015. One of the main issues that arises when performing this analysis is the

aforementioned potential endogeneity concerning institutional quality. Therefore, the main aim of this thesis is to disentangle the forces at hand and subsequently isolate the effect of natural resource abundance from its relation with institutions. This is done by approaching the issue as a natural experiment, where recent natural resource discoveries serve as treatment. The contribution of this paper is twofold.

Firstly, to the best of my knowledge, there is no empirical analysis addressing the natural resource curse that attempts to fully isolate the effect of natural resources from its interaction with institutions using data on resource discoveries. This is a new approach in order to tackle endogeneity issues by exploiting the time dimension in the data on natural capital, which will be further discussed in section 4. This extends the studies that have currently been conducted and contributes to our understanding of the resource curse and the role of institutions.

(6)

3

Secondly, the data that is used for this research is the most recent data available, and it covers the period from 1995 to 2015. The original study by Sachs and Warner also covered a time span of two decades, from 1970 to 1990. Subsequent studies that approached the issue from a different angle mostly used the same data for the same time period, to show that their alternative approach altered the results. However, the role of natural resources in these decades is even more complicated due to extreme commodity price booms in the 1970’s and sharp price declines in the 1980’s. Certainly, in the timeframe of this study commodity prices have been far from stable. However, as mentioned before, it is still important to investigate if the curse still exists to this day.

The remainder of this paper will be structured as follows. Section 2 will give a brief overview of the current state of the literature of the issue at hand, while section 3 describes the data that is used for this research and presents patterns within the data. In section 4 I will

elaborate on the empirical strategy of this study and subsequently in section 5 the results of the analysis will be presented and discussed. Finally, section 6 concludes.

2. Literature review

2.1 The origin of the resource curse

As mentioned before, the subject at hand has been studied extensively. This has been triggered by a series of studies by Sachs & Warner (1995, 1999, 2001) which found a

negative relationship between average GDP per capita growth over a period of 20 years and the level of resource abundance in the base year. This effect remained substantive and statistically significant, even when they controlled for other important drivers of economic growth including initial level of GDP, economic openness, investment and rule of law. Their findings coined the term ‘natural resource curse’ and has been the trigger for many studies addressing this issue from different angles.

2.2 Measuring natural resources

In the resource curse literature the variable most used as a proxy of resource abundance is primary exports as percentage of GDP, which stems from Sachs & Warner (1995). However, Brunnschweiler & Bulte (2008) argue that this explanatory variable suffers from endogeneity issues. Since the denominator of this variable reflects the entire economy, it is unavoidably endogenous, and they argue that this measure captures the level of resource dependence of an economy rather than resource abundance. They repeat the original analysis, but they employ a measure that estimates the total value of natural resources per capita in a country and they show that with this alteration the result is reversed, and that natural resource abundance is in fact associated with higher economic growth.

Furthermore, it is not obvious that resource abundance is related to a high share of resource exports (Brunnschweiler, 2008). Countries like Australia and Norway for example, are rich in natural resources but the income generated by these resources does not make up a large part of their economy, so they are not dependent on them. There are several studies (Norman (2008), Pendergast, Clarke & van Kooten (2011)) in which the results are changed

(7)

4

or even reversed when using resource dependence compared to using resource abundance. This stresses the importance of a distinction between the two.

2.3 The importance of institutions

Politicians in power might be tempted to divert natural resource rents away from growth-enhancing activities and instead use them for so-called ‘white elephant projects’ (i.e. projects that might seem efficient in the short run, but have detrimental effects in the long run) (Basedau, 2005). Furthermore, rent-seeking behavior is considered to have a negative effect on institutional quality as well as on inequality. Rent-recipients might have an incentive to pay bribes to maintain control over the extraction, leading to increased corruption, which has a detrimental effect on economic growth (Bardhan, 1997). Thus it might seem that natural resource abundance impedes growth, while it is in fact the underlying corruption whose effect is magnified by the natural resource abundance.

Moreover, resource abundance alone is not sufficient to explain a possible resource curse, since there are examples of countries (e.g. Norway and Botswana) in which natural resources positively contributed to economic growth. The phenomenon is better explained by other factors, such as political institutions. What determines the direction in which resource abundance affects the economy are the conditions under which those resources are exploited (Basedau, 2005). Firstly, institutional quality determines for a great deal whether resources will be managed to benefit the economy and society. Secondly, the type of resources affect the degree of interaction with institutions. For example, war or domestic conflict over the control of resource-full areas can stop production and negatively affect economic growth. Other papers find that the effect of natural resources on growth is not significant when controlling for institutional quality (Sala-i-Martin and Subramanian, 2003). Natural resources have a negative effect on institutions in the long run, and through this channel it is important to distinguish between different types of resources because they interact differently with institutions. Only certain types of resources are associated with this transmission channel (e.g. blood diamonds or oil fields), and a distinction can be made between point and diffused resources on the basis of whether they are, respectively, spatially clustered or dispersed (Auty, 2001). As mentioned before, this difference plays an important role, especially in the institutional context.

The measurement of institutional quality also differs widely between studies. A common measure is a proxy for rule of law, similar to Sachs & Warner, but often this is combined with other proxies like the protection of property rights for example (Sala-i-Martin and Subramanian, 2003), because only using rule of law as a measurement of potential malfunctioning of institutions is not sufficient.

Brunnschweiler (2008) also controls for institutional quality when looking at the effect of natural resource abundance on economic growth. In her regression two findings emerge. Firstly, her institutional variables always enter the model positive and significant. Secondly, the relative importance of abundance in explaining growth decreases while that of institutional quality increases, once again stressing the importance of institutions.

2.4 Inequality

Variation in the effects of natural resources on well-being cannot only be found between countries, but also within countries. Buccellato and Alessandrini (2009) assess the relation

(8)

5

between natural resources and economic inequality within countries. They find that there is a negative relation, but again, only for certain subsoil resources. However, the role of

institutions remains unmentioned in their paper. Goderis and Malone (2011) find a different but not necessarily contradictory effect. They show that after a resource boom, inequality within countries falls immediately, only to increase steadily until it is back at its original level.

The effect on inequality should definitely be taken into account, since economic growth is only one gauge for a country’s well-being. As mentioned before, Botswana has managed to use its natural resources to boost economic growth, but the emergence of its extractive industry has resulted in increased inequality. It became a rich country with poor people. Furthermore, about half of the population in Venezuela lives in poverty, while this is the Latin-American country with the highest level of natural resources. These are examples of countries in which natural resource abundance has had a detrimental effect on equality. As can be seen, the role of natural resources is ambiguous on various levels, although two conclusions can be drawn from the literature that exists so far. One thing that is clear is that resources might have an effect on growth as well as on inequality, and that institutions play a crucial role in this channel. Secondly, the way in which natural resources are measured has an evident effect on the results, as well as the distinction between different types of natural resources. This paper combines all these findings in an attempt to further unravel the effect of institutional quality and the relation with natural resource abundance.

3. Data

In this section I will give some insight in the data that I obtained for this paper and I will show some first visualizations of patterns that the data seem to exhibit. The majority of the data for this research has been retrieved from the World Bank (2017) database.

3.1 Independent variables

In order to answer the research question, I use three different outcome variables, one indicating economic growth, and two variables that are meant to capture inequality. As mentioned before, this research covers a period of 20 years just as the majority of the

existing literature, since it is logical to assume that the potential effect that natural resources have on a country’s economy, takes time to materialize. However, this paper regards a more recent period compared to the majority of the literature.

3.1.1 Average GDP per capita growth

The main variable of interest in the majority of the literature is the average growth of GDP per capita, for a period of twenty years after a certain base year. The base year in this research is 1995, and the time period that is studied runs from 1996 to 2015. This first dependent variable is calculated as follows:

𝐺𝑟𝑜𝑤𝑡ℎ = ln(𝐺𝐷𝑃2015)− 𝑙𝑛(𝐺𝐷𝑃1996)

(9)

6

Each value is multiplied by one hundred in order to obtain percentage changes. This does not affect results, but it does allow for a more clear interpretation. The data required to calculate this average growth rate is available for 201 of the 207 countries in the sample. If the resource curse is true, the results should indicate that this variable is negatively affected by natural resource abundance.

3.1.2 Economic inequality – The Gini coefficient

To address the impact that natural resources have on other indicators of well-being besides economic growth, I employ two measures of inequality. The first concerns the most

widespread measure of economic inequality, namely the Gini coefficient. The Gini coefficient measures the extent to which income of actors within an economy deviates from a perfectly equal distribution. It measures the area between the Lorenz curve (which depicts the actual distribution of income) and the 45° line (which depicts a perfectly equal distribution) as a percentage of the total area under the 45° line. This means that a Gini coefficient of zero represents perfect equality, whereas a Gini coefficient of one hundred represents perfect inequality. In order to be able to see a possible effect of natural resources on economic inequality, the outcome variable equals the difference in the Gini coefficient for the twenty-year time period, i.e.:

Difference in economic inequality = Gini coefficient2015 – Gini coefficient1995 (Eq. 2)

This means that a positive value of this outcome variable represents an increase in economic inequality and thus an impairment of well-being. Unfortunately, the availability of data on the Gini coefficient is very limited, let alone in this case where we need two data points for every country to construct our variable of interest. For convenience, if data for these certain years were not available, a value for one or two years after or before the base and end year was used, in case it was available. After doing this, the variable of interest could still only be constructed for 88 of the 207 countries in the sample. As argued before, natural resource abundance (in combination with poor institutional quality), might increase vertical

inequality. If this is true, the results should show that natural resources have a positive effect on this outcome variable.

3.1.3 Social inequality – Access to education by gender

The second measure of inequality that I shall use is a proxy of social inequality, following Gylfason & Zoega (2002). Since economic growth and economic inequality are closely

related, I think it is important to also address other measures of inequality that might reflect a different harmful pathway supporting the term ‘resource curse’. Gylfason & Zoega (2002) state that natural resource abundance decreases school enrolment rates, since the natural resource sector prefers low skilled labor over high skilled labor, more so than the

manufacturing sector. If this is true, this in itself does not necessarily mean that inequality rises. However, if the enrolment rate for males and females is affected differently as argued before, this mechanism does have an impact on inequality. So ultimately we are interested in the difference of the gender specific difference in enrolment. Therefore I use data on the gross secondary school enrolment by gender, and the outcome variable is constructed as follows:

(10)

7

Difference in social inequality = (Male enr.2015 – Female enr.2015) – (Male enr.1995 – Female enr.1995) (Eq. 3)

In equation 3, as well as in equation 2, a positive value of the outcome variable represents an increase in inequality and thus an impairment of well-being. But, in almost all of the countries in which the enrolment rates for males and females differ, it is higher for males than for females. However, exceptions exist. Nevertheless, as can be derived from equation 3, I follow Gylfason & Zoega (2002) and I employ the arithmetic difference rather than the absolute difference. This means that if for a certain country the enrolment rate in 1995 was 10 percentage points higher for males than that for females, and in 2015 the enrolment rate for males would be 10 percentage points lower than that for females, I consider this to be a

decrease in inequality. This makes interpretation of the variable easier, and I believe that the

few cases in which female enrolment rates are higher will not significantly bias my results. The availability of data on school enrolment rates is limited, so I use the same method that I applied to the construction of the economic inequality variable. After doing this, this variable is still only available for 63 of the 207 countries in the sample. The

arguments brought up earlier state that a growing extractive industry might decrease male enrolment rates compared to female enrolment rates. If this is the case, the results should display that natural resource abundance has a negative effect on this outcome variable and should thus decrease horizontal inequality.

Figure 1. 20-year differences in income inequality and social inequality.

Figure 1 displays the bivariate correlation between the two measurements of inequality. Unfortunately, the countries with missing values differ substantially between the two variables, and as a result there are only 29 countries for which both variables are available, resulting in a very low R2_{. Although the number of observations is modest and one should be}

careful when drawing conclusions, this graph seems to show that a change in economic inequality does not necessarily go hand in hand with a change in social inequality in this sample. y = 0.2197x - 1.4242 R² = 0.0147 N = 29 -25 -20 -15 -10 -5 0 5 10 15 20 -15 -10 -5 0 5 10 Dif fe re n ce e con o m ic in eq u alit y

Difference social inequality

(11)

8

3.2 Natural resource abundance

The explanatory variable in my analysis relates to the natural resources that a country possesses. As mentioned before, there has been a lot of discussion about the proper way to measure natural resource abundance.

As already pointed out, most of the resource curse literature uses primary exports as a share of GDP as a proxy for resource abundance, which is unavoidably endogenous. This variable can better be considered resource dependence rather than resource abundance. Although the two are inseparably connected, since resource abundance might lead to resource dependence because of a comparative advantage, they do absolutely not measure the same thing. The problem is that every proxy for resource abundance that is measured compared to other economic activities, is endogenous. Specifically, every variable measuring flows of natural resources can be considered to be a choice variable and is thus not

exogenous. Subsequently, we should measure resources in absolute terms rather than in relative terms. To validate the general claim of a resource curse it is logical that the analysis should assess the effect of resource stocks per se.

Fortunately, the World Bank provides estimates of the stocks of natural capital in a country for the years 1995 and 2000. This natural capital measure is a compound measure, made up of all types of natural resources. However, as mentioned before, certain clustered, subsoil resources that yield high rents are more prone to rent-seeking and corruption. Therefore, the role of institutional quality seems to apply most to these types of resources. In line with this reasoning, for the remainder of this paper I will focus on the clustered, so-called ‘pointy’ subsoil resources, which exist of the stocks of the main energy resources (oil, gas and coal) as well as the stocks of ten metals and minerals (bauxite, copper, gold, iron ore, lead, nickel, phosphate rock, silver, tin and zinc). These stocks are valued measuring the present discounted value of expected resource rents for a future period of 20 years, denoted in constant 2005 US$. Note that for this method future rents of the resource have to be known. These are extrapolated from rents observed in the year 2000, assuming that resource rents grow at a constant rate.

This level of subsoil assets is scaled by dividing through by population. This

transformation into per capita terms is crucial in determining the extent to which resource rents have the potential to increase average living standards. If resources were measured in aggregate terms, then small countries with small populations would appear resource poor, even if the available natural resources per capita could substantially influence overall standards of living. For a large country with a large population, the reverse would be true. Finally, I take the natural logarithm of the subsoil assets per capita. This logarithmic

transformation is appropriate as the values differ strongly across countries, and it allows for an easier interpretation of the estimated coefficient. Before this logarithmic transformation, I assign a value of 0.01 to the countries that do not have subsoil assets, since it is impossible to take the logarithm of zero.

Furthermore, I exploit the time dimension in the data to detect discoveries of subsoil assets. For all countries, I subtract the value of the per capita subsoil assets in 1995 from the value of the per capita subsoil assets in 2000 in order to obtain the net discoveries of natural resources per capita. If the result is positive, more resources have been discovered than have been extracted:

(12)

9

This is however not an adequate measure of resource discoveries, since resources have been extracted in this timespan as well. Therefore, we need to add the extracted resources during this period to the net discoveries in order to obtain the gross discoveries. Data on actual resource extraction is not available, so data on resource exports is used instead. Since the majority of extracted resources is sold abroad, I believe that this will not significantly alter the results. The data that is available depicts the share of exports of subsoil assets of GDP, and is available for the entire five-year time period. The net discoveries are denoted in per capita terms, so I deflate the level of subsoil assets exports by the size of the population accordingly. The level of resource exports per capita for this period is calculated as follows:

𝑅𝑒𝑠𝑜𝑢𝑟𝑐𝑒 𝑒𝑥𝑝𝑜𝑟𝑡𝑠 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 = ∑ 𝑆𝑆𝐴 𝑒𝑥𝑝𝑜𝑟𝑡 𝑠ℎ𝑎𝑟𝑒𝑖 × 𝐺𝐷𝑃 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎𝑖 1999

𝑖=1995

(Eq. 5)

And subsequently the gross level of resource discoveries per capita is obtained by adding up the results of equation 4 and equation 5. In order to capture the magnitude and therewith the potential effect, I furthermore denote the discoveries in subsoils assets per capita as a percentage of initial GDP per capita:

𝐺𝑟𝑜𝑠𝑠 𝑑𝑖𝑠𝑐𝑜𝑣𝑒𝑟𝑦 = 𝑁𝑒𝑡 𝑑𝑖𝑠𝑐𝑜𝑣𝑒𝑟𝑦 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 + 𝑅𝑒𝑠𝑜𝑢𝑟𝑐𝑒 𝑒𝑥𝑝𝑜𝑟𝑡𝑠 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎

𝐼𝑛𝑖𝑡𝑖𝑎𝑙 𝐺𝐷𝑃 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 × 100 (Eq. 6)

These gross discoveries will be used in order to isolate the effect of natural resources as completely as possible. Although I treat subsoil assets as exogenous, I am aware of the fact that both the extraction of subsoil assets as well as the stocks and the search thereof, are not irresponsive to economic forces. For example, a change in the price of a resource can influence both exploration as well as extraction of this specific resource. Although I acknowledge the potential importance, I will not address this issue further. Data on the subsoil assets per capita is available for 125 countries in the sample. Data on discoveries is available for 112 countries, of which 59 experienced a gross resource discovery. A few countries have zero subsoil assets in 1995 as well as in 2000, but they do have a positive gross discovery. I can only assume that the resources that were discovered in these countries were immediately extracted and sold. For the remainder of this paper, I will use the terms subsoil assets, natural capital, natural resources and resource abundance interchangeably.

3.3 Institutional quality

Another crucial variable in my analysis is institutional quality. Institutions are systems of hierarchical, man-made rules that structure behavior and social interaction (Groenewegen, Spithoven and Van den Berg, 2010). This sounds a bit arbitrary, but there are six World Governance Indicators as conceived by Kaufman et al. (2010) for institutional quality that capture the entirety of this concept. These consist of control of corruption, government effectiveness, political stability, regulatory quality, rule of law, and voice and accountability.

Data on these indicators is available from 1996 onwards for 193 countries, and I will use the data for 1996 since this is closest to the base year. Furthermore, I make a few modifications before I include this variable in my model. The values for all the proxies originally range from -2.5 to 2.5. This range is arbitrary, so for the sake of interpretation I

(13)

10

add a value of 2.5 for every observation so that the values range from 0 to 5, and subsequently I take the natural logarithm of this number so that the coefficient on this variable can be interpreted as an elasticity. Despite the fact that these six proxies all capture a different aspect of institutional quality, they are very closely related, which brings me to my second modification. The interconnectedness of the proxies might lead to

multicollinearity, which in turn results in increased standard errors and subsequent estimation errors. This is a serious threat, as can be seen in table 1 which shows the

correlations of the proxies. Therefore, I combine them and construct one variable, which is the average of the six proxies.

Table 1. Correlation coefficients of the institutional quality proxies.

Corruption Gov.Eff. Pol. Stab. Reg. Qual. Rule of Law Voice

Corruption 1.0000 Gov.Eff. 0.9179 1.0000 Pol. Stab. 0.8762 0.7915 1.0000 Reg. Qual. 0.8526 0.9188 0.8400 1.0000 Rule of Law 0.9100 0.9347 0.8370 0.8838 1.0000 Voice 0.8323 0.8746 0.8713 0.8692 0.8751 1.0000

To summarize the above, the institutional quality is calculated as follows:

𝐼𝑛𝑠𝑡. 𝑄𝑢𝑎𝑙. = 𝑙𝑛 (𝐶+2.5)+(𝐺+2.5)+(𝑃+2.5)+(𝑅𝑒+2.5)+(𝑅𝑢+2.5)+(𝑉+2.5)

6 (Eq. 7)

I believe this measurement of institutional quality to be an improvement compared to proxies used in many previous studies1_{since this variable captures all aspects of what we}

consider to be institutions, and does not limit its scope to one certain aspect.

3.4 Control variables

There are obviously many other factors that affect GDP growth and inequality, and these factors need to be controlled for in the analysis. The choice of control variables is based on the existing literature. Firstly, the logarithm of the initial level of GDP per capita in 1995 is included, denoted in constant 2010 US$, which is available for 183 countries. The economic openness of a country is also included, and is calculated as the average share of imports plus exports of GDP, for the period 1996 – 2015, which is available for 196 countries. Another control variable that recurs frequently in the literature is the level of investment, which serves as a proxy for physical capital. To capture the level of investment, I include the gross capital formation as a percentage of GDP, for the base year 1995, which is available for 174 countries. Furthermore, a proxy for human capital is included as well, namely the literacy rate, which is denoted as the percentage of people older than 15 who can read and write.

1_{E.g.: Sachs & Warner (1995) limit their proxy of institutional quality to the Rule of Law aspect only, whereas} Brunnschweiler & Bulte (2008) only include the Rule of Law and the Government Effectiveness proxies.

(14)

11

This again is measured in the base year 1995, and is available for 135 countries. As a proxy for the health level in a country I include the life expectancy at birth in years, for the base year 1995. This variable is available for a number of 197 countries. Lastly, I include a dummy variable that indicates whether a country is landlocked, which obviously is available for the entire sample. A complete description of all the variables can be found in appendix A1.

3.5 Patterns in the data

Before we turn to the regression analysis, it might be useful to get an idea of what our data looks like and if a first glance shows us potential relations between the variables of interest. At first, let’s have a look at the bivariate relation between resource abundance and

economic growth, which is possible for 125 observations. A scatterplot of these two variables is shown in figure 2. The relation in the graph is negative, and although this depiction does not yield compelling evidence for a natural resource curse, there does definitely not seem to exist a positive relationship between the two variables.

Now let’s do the same for inequality. As mentioned before, the availability of data concerning inequality is limited, and as a result the number of observations for economic inequality and social inequality drop to 60 and 40 respectively. The scatterplots are shown in figure 3 and figure 4, and they notably differ from the trend in figure 2. Once again, recall that this is only a bivariate relation and that the number of observations is very limited, so one should be cautious when drawing conclusions. Having said that, we can conclude that these figures do not provide convincing evidence for a positive relation between natural resource abundance and a decrease in inequality, but it definitely does not provide evidence for the opposite either.

Figure 2. Resource abundance and economic growth.

y = -0.0447x + 2.1325 R² = 0.0216 N = 125 -4 -2 0 2 4 6 8 10 -5 -2,5 0 2,5 5 7,5 10 12,5 Av era ge G DP p c gro w th 1995 -2015

Logarithm of subsoil assets per capita 1995

(15)

12

Figure 3. Resource abundance and economic inequality.

Figure 4. Resource abundance and social inequality.

y = 0.0993x - 0.909 R² = 0.0073 N = 60 -15 -10 -5 0 5 10 15 20 25 -5 -2,5 0 2,5 5 7,5 10 12,5 Dif fe re n ce e con o m ic in eq u alit y

Fitted values y = 0.1198x - 3.6399 R² = 0.0054 N = 40 -30 -20 -10 0 10 20 -5 -2,5 0 2,5 5 7,5 10 12,5 Dif fe re n ce s o cial in eq u alit y

(16)

13

4. Empirical strategy

In this paper, the effect of natural resource abundance on three different outcome variables is estimated with a cross-section analysis using standard OLS estimation. However, the ambiguous role of institutions complicates the analysis. A number of different models is applied in order to address this problem.

4.1 Endogeneity of institutions

Theoretically, there are many ways in which institutions affect the relation between natural resources and the outcome variables. One potential mechanism is that the quality of institutions at the moment of extraction determines whether the revenue generated from this extraction is beneficial or detrimental for growth and inequality. As mentioned before, Basedau (2005) pleads that what determines the direction in which resource abundance affects the economy are the conditions under which those resources are exploited. In his opinion it is the institutional quality that determines for a great deal whether resources will be managed to benefit the economy and society. Furthermore, Norman (2008) states that there exists a possibility that politicians managing natural resources systematically make faulty decisions concerning growth-related policies through excessive borrowing and irrational optimism. If this is true, it might not be the case that resource abundance negatively influences institutions, but that existing weaknesses are exacerbated by a large extractive industry. Butkiewicz & Yanikkaya (2010) also state that weak institutions enable corruption and result in low growth rates, but that natural resources might promote growth if existing institutions are strong. Brunnschweiler & Bulte (2008) and Hammond (2011) concur with this view and argue that the political conditions under which resources are exploited determine the effect it has on the economy. Thus resources can be a blessing for an institutionally strong country, but a curse for an institutionally weak country. In this line of reasoning, institutional quality operates as a moderator variable, and there exists a qualitative interaction between institutional quality and natural resource abundance. This means that both the direction and the magnitude of the effect of natural resources on the outcome variable is determined by the level of institutional quality.

Another pathway might be that natural resources negatively affect institutional quality, and that institutional quality affects the outcome variables. In this case institutional quality operates as a mediator variable. Natural capital of a country belongs to its citizens, but the politicians in power are entitled to exercise ownership over these resources on their behalf (Carbonnier, 2011). The existence of rents might prevent alignment of the interests of the majority of the population and the interests of the political elite. This might cause myopia on the part of the politicians in power, resulting in damaging rent-seeking behavior which undermines growth-promoting, welfare-enhancing policies. Rents from the extractive sector are more easily captured than revenues from taxation and the public purse is less closely watched, thus a large stock of natural capital could also lead to high levels of corruption. Moreover, natural resources could increase the value of being in power, resulting in an increased chance of conflict as well as more frequent changes in political power, with subsequent positive or negative effects on income and inequality (Wick, 2008).

(17)

14

Lastly there exists a relation between institutional quality and the share of natural resource extraction (i.e. natural resource dependence) of an economy. Poor development policies might simultaneously lead to a higher share of natural resource exports and low growth rates. The extractive industry might be the default sector in the absence of productive investment due to poor institutions. Furthermore, poor institutions may be the cause of a lack of industrialization and subsequently a high dependence on natural resources.

Figure 5. Endogeneity issues concerning natural resources and institutions.

All these channels that affect the estimation results are depicted in figure 5. Endogenous channel 1 is easily avoided by carefully choosing an explanatory variable. As mentioned before, natural resource abundance should be measured in stocks, since this is not a choice variable, and the level of natural resources cannot be influenced by institutional quality. However, endogeneity issues still exist if institutional quality functions as a moderator variable (via channel 3) or if it functions as a mediator variable (via channel 2 and channel B), or both. Since we are ultimately interested in channel A, it is necessary to find a way to isolate this effect.

4.2 The model

A number of different models is employed in order to empirically address the endogeneity issues mentioned above. It is logical to assume that the effects of natural resources might take some time to accrue, and therefore the timespan of twenty years is incorporated in the model. Ultimately, we want to find out what the effect of natural resource abundance in the base year is on GDP growth and inequality in the subsequent twenty-year period, controlling for initial values of other influencing factors in the base year. I start off with a simple

regression model of the following form:

(18)

15

In this equation Y is either economic growth or one of the two inequality measures, NR indicates the natural logarithm of the subsoil assets per capita, INS indicates the natural logarithm of the institutional quality measure, and X indicates a vector of control variables. Furthermore, 𝛽0 is a constant, the other 𝛽’s are the estimated coefficients, and εi is the error

term. To justify the choice of control variables not only by theory and former research but also empirically, the controls used in the current analysis are added consecutively into the equation. By studying the effect of adding controls consecutively on the coefficient of interest and on the coefficient of determination (R²), this step-by-step procedure demonstrates the reduction of bias in assessing the effect of natural resources on the outcome variable.

However, even if all the controls are included, the possible endogeneity channels 2 and 3 from figure 5 might still be present. In an attempt to tackle endogeneity channel 3, I extend the model by including an interaction term in the equation. An interaction term is included if the effect of one explanatory variable on the outcome variable depends on the value of another explanatory variable. The model then gets the following form:

Y = β0 + β1 NRi + β2 INSi + β3 Xi + β4 NRi × INSi + εi (Eq. 9)

This model only differs from the previous one concerning the interaction term. In this model, the effect of a change in natural resources on the outcome variable equals:

𝜕 𝑌

𝜕 𝑁𝑅= 𝛽1+ 𝛽4 𝐼𝑁𝑆𝑖 (Eq. 10)

By including the interaction term, the effect of natural resources on the outcome variable now depends on the value of the institutional quality variable. In this way, we correct for channel 3 in figure 5.

In an attempt to correct for channel 2 in figure 5, I include the gross discovery variable from equation 6 in the model. The intuition behind this is as follows. The effect that natural resources have on an economy takes time to accrue, but so does the potential effect it has on institutional quality. However, we can assume that the latter takes much more time than the former. Brunnschweiler & Bulte (2008) label institutions as ‘roots of a society’ that only change gradually, and according to Alexeev & Conrad (2009), the existence of channel 2 in figure 5 does not make sense intuitively, since institutional quality seems to be quite persistent over time. They state that institutional quality changes only slowly and that relatively recent discoveries of natural ‘pointy’ resources should not have a radical effect on institutions.

Following this argument, I will incorporate the gross discovery variable in my model in order to further isolate channel A in figure 5. As mentioned before, the institutional quality measure is an average for the 6 proxies in the year 1996. In order to invigorate the current argumentation, for each country I constructed this variable for every subsequent year up until 2015. I then calculated the country-specific standard deviation of the institutional quality for this twenty-year period. Recall that this variable (before the

logarithmic transformation) reaches from 0 to 5. The average standard deviation for all 193 countries is 0.09, with only 15 countries that have a standard deviation that is larger than 0.2 (which is 4% of the total range). This supports the claim that institutional quality, despite

(19)

16

being possibly affected by resource abundance, does not change radically, and can be considered rather time-invariant for this twenty-year period. Therefore, by including the resource discovery variable in the model while keeping everything else equal, allows me to further isolate channel A in figure 5. A counterargument that one might think of, is that the discovery of natural resources might not be a truly exogenous ‘treatment’, and that selection bias might arise if we assume that in countries that are rich in natural resources, people are more likely to search for, and thus discover, natural resources. Since in this case the random assignment of treatment is based on the level of resources, it is crucial to still control for this resource abundance. After inclusion of the discovery variable, which is also interacted with the institutional quality variable, the model then becomes one of the following form:

Y = β0 + β1 NRi + β2 INSi + β3 Xi + β4 NRi × INSi + β5 DISCi + β6 DISCi × INSi + εi (Eq. 11)

One issue concerning this model is that, as mentioned before, one could expect the abundance of natural resources and the discovery of natural resources to be highly correlated. But as a matter of fact, the correlation coefficient between the level of subsoil assets per capita and the level of gross resource discoveries per capita is only 0.6477. Therefore I conclude that no serious problems of multicollinearity will arise.

If the resource curse is true after addressing the endogeneity issues, we expect β5 to

have a negative sign in the growth model. If the arguments concerning the effect of natural resources apply, then we expect β5 to have a positive sign in the economic inequality model,

leading to an increase in inequality, and a negative sign in the social inequality model, resulting in decreasing social inequality.

5. Results

In this section the findings of the analysis will be discussed. The results are obtained using the statistical package Stata version 10 and will be presented separately for each of the three outcome variables.

5.1 Natural resources and economic growth

Table 2 shows the results of the economic growth model. In column (1) GDP growth is regressed on natural resource abundance and institutional quality only. In column (2) to (7) the control variables are added consecutively, column (8) show the results for the model including the interaction term between institutional quality and subsoil assets, and column (9) shows the results of the model that also includes resource discoveries as well as

discoveries interacted with institutional quality. In the simplest model in column (1), subsoil assets have a negative, statistically significant effect at the 10% level. Despite the significant effect, the model is clearly misspecified as it does not include any control variables.

Note that the coefficient of determination, R², increases throughout column (1) to (7) when more controls are added, which is supported by former research. Despite the fact that most of the point estimates are statistically insignificant, the specification of column (7) is the preferred one, as the controls are in line with controls used in past research and due to the increasing R². Consecutively adding controls results in an insignificant negative effect of subsoil assets on economic growth, and when the interaction term of subsoil assets and

(20)

17

institutions is included the effect even becomes positive. Since the sign of the coefficient on subsoil assets changes as the interaction term is included, the crucial role of institutions is confirmed.

Following the earlier line of reasoning, we are ultimately interested in the results in column (9), and the large increase of the R2_{between column (8) and (9) contributes to the}

belief that this is the most relevant specification. The coefficient on the discovery variable, as well as its interaction with institutional quality, are the main coefficients of interest since these indicates the sole effect of natural resources on economic growth, which reflects channel A in figure 5. The coefficients of these variables are positive and significant at the 5% and 10% level, indicating that a discovery of natural resources of the magnitude of 1% of GDP leads to a percentage increase of average GDP growth of 0.0484 + 0.0546 ×

institutional quality. This indicates that the effect of natural resources increases if the institutional quality increases, and this confirms the belief that the effect that natural resources have on GDP growth depends on the institutional quality. Moreover, the coefficient on the interaction term is larger than the individual coefficient, once again

stressing the importance of institutional quality in the natural resource debate. The direction of the effect depends on the chosen measurement of institutional quality. As mentioned before, the proxies used for institutional quality initially had a value ranging from -2.5 to 2.5, implying that a negative value of institutional quality would result in less economic growth if natural resource abundance increased. Although the crucial role of institutions in relation to natural resources seems to be confirmed, it is striking that in most of the specifications, its individual effect on GDP growth is highly insignificant. Furthermore it is worth mentioning that initial GDP is significant with a negative sign in every specification, indicating

convergence. As for life expectancy, it is striking that the effect on growth is negative, although insignificant. One could argue that, as life expectancy is usually correlated with economic prosperity, a higher initial expected age would indicate a higher level of initial income, and a higher initial income would mean lower marginal productivity, thus a slower growth path according to classical development theory (Solow, 1956).

The main insight from the results in table 2 is that the coefficients on the discovery variable as well as its interaction with institutional quality both have a positive sign and are both significant, rejecting the existence of a resource curse. However, it should be taken into account that the number of observations in column (9) is limited, therefore one should be cautious when drawing conclusions from these results.

5.2 Natural resources and economic inequality

Table 3 shows the results of the economic inequality model. Again, in column (1) the difference in economic inequality is regressed on natural resource abundance and

institutional quality only. In column (2) to (7) the control variables are added consecutively, column (8) show the results for the model including the interaction term between

institutional quality and subsoil assets, and column (9) shows the results of the model that also includes resource discoveries as well as discoveries interacted with institutional quality. Just as in the previous model, the R² increases throughout column (1) to (7) when more controls are added. In column (1), subsoil assets have a positive, although statistically insignificant effect on the outcome variable. Consecutively adding controls results in a negative effect of subsoil assets on the outcome variable. This implies that natural resource abundance decreases economic inequality, although the coefficient is highly insignificant.

(21)

18

When the interaction term between subsoil assets and institutional quality enters the model the coefficient on subsoil assets increases quite radically. This again shows that institutional quality plays a crucial role in the mechanism at hand. It is striking that the coefficient on the subsoil assets is negative in the model of column (9), while the coefficient on the discovery variable is positive. Following the same line of reasoning as in the economic growth model, the two coefficients relating to resource discoveries reflect the sole effect of natural resources on economic inequality. This implies that a discovery of natural resources of the magnitude of 1% of GDP leads to an increase of the outcome variable of 0.0290 + 0.1080 × institutional quality, and thus to an increase in inequality. Again it is striking that the

coefficient of the interaction term is much larger than the individual coefficient, once more demonstrating the importance of institutions. Although this is the results that was expected, both coefficients are highly insignificant and thus we can neither confirm nor reject the hypothesis that natural resource abundance increases economic inequality. Furthermore it is remarkable that in the final specification, the sign of the institutional quality coefficient is positive. One would expect that an increase in institutional quality would lead to a decrease in economic inequality, but these results imply the opposite, although this coefficient is insignificant as well.

It is important to recognize that the number of observations in this model is very limited. The low R2_{in the different specifications, as well as the fact that only 1 coefficient is}

significant at the 10% level in column (9), leads me to believe that no valuable conclusions can be drawn from these results. Therefore, the effect that natural resources have on economic inequality remains inconclusive.

5.3 Natural resources and social inequality

In table 4 the results of the social inequality model are presented in the same way as in the previous two models. Again the R², increases throughout column (1) to (7) when more controls are added.

In the specification in column (1) subsoil assets have a positive but insignificant effect on the outcome variable. Consecutively adding controls results in an increasing effect of subsoil assets, and when the interaction term of subsoil assets and institutions is included the effect becomes negative. Since the sign of the coefficient on subsoil assets changes as the interaction term is included, the crucial role of institutions is confirmed in this model as well. Once again we look at the coefficients relating to the resource discoveries which are negative as expected. The results imply that a discovery of natural resources of the

magnitude of 1% of GDP leads to a decrease of the outcome variable of 0.8898 + 1.1227 × institutional quality, and thus to a decrease in inequality. Similar to the previous models, the coefficient of the interaction term is larger than the individual coefficient. Although the coefficients are insignificant, the R2_{of the model increases remarkably when discoveries are}

included in the model, indicating the relevance of this variable. Once again however, the limited number of observations is a serious issue for this model. Just as in the previous model, all variables related to natural resources are insignificant, indicating that we cannot draw unambiguous conclusions from these results and the effect of natural resources on social inequality remains unclear. The hypothesis that natural resource abundance decreases social inequality is neither accepted nor rejected.

(22)

19

Table 2. The effect of natural resources on economic growth. Dependent variable: Average GDP per capita growth (1995 – 2015).

(1) (2) (3) (4) (5) (6) (7) (8) (9) Subsoil assets -0.0450 * -0.0251 -0.0201 -0.0289 -0.0395 -0.0355 -0.0342 0.0318 0.1389 (0.0255) (0.0264) (0.0274) (0.0275) (0.0366) (0.0370) (0.0374) (0.1139) (0.1251) Institutional quality 0.1436 1.1045 ** 1.0318 * 0.2267 0.1427 0.2450 0.2295 0.4397 0.4732 (0.4002) (0.5373) (0.5307) (0.5428) (0.9049) (0.8424) (0.8818) (0.9801) (0.9662) Initial GDP -0.4361 *** -0.4486 *** -0.2643 ** -0.6538 *** -0.4946 ** -0.5019 ** -0.4694 * -0.5278 * (0.1170) (0.1154) (0.1041) (0.1962) (0.2332) (0.2392) (0.2511) (0.2797) Openness 0.0019 0.0001 0.0003 0.0001 0.0001 0.0007 0.0042 (0.0017) (0.0019) (0.0030) (0.0030) (0.0030) (0.0031) (0.0032) Investment 0.1035 *** 0.1271 *** 0.1418 *** 0.1412 *** 0.1394 *** 0.1985 *** (0.0341) (0.0475) (0.049) (0.0504) (0.0511) (0.0453) Literacy rate 0.0088 0.0183 0.0175 0.0177 0.0149 (0.0079) (0.0118) (0.0125) (0.0127) (0.0105) Life expectancy -0.0512 -0.0447 -0.0475 -0.0345 (0.0407) (0.0514) (0.0521) (0.0491) Landlocked 0.1919 0.2680 -0.2254 (0.5927) (0.5936) (0.5470) SSA x Institutions 0.0804 0.2063 (0.1297) (0.1364) Discovery 0.0484 ** (0.0235) Discovery x Institutions 0.0546 * (0.0300 R² 0.0223 0.1314 0.1348 0.2403 0.2904 0.3070 0.3083 0.3124 0.4805 N 123 121 121 117 81 81 81 81 74

Note: Robust standard errors in parentheses.

(23)

20

Table 3. The effect of natural resources on economic inequality. Dependent variable: The change of the Gini coefficient between 1995 and 2015.

(1) (2) (3) (4) (5) (6) (7) (8) (9) Subsoil assets 0.1004 0.0591 0.0615 0.0722 -0.0348 -0.0237 -0.0194 -0.4948 -0.0259 (0.1410) (0.1498) (0.1438) (0.1517) (0.1490) (0.1477) (0.1402) (0.3879) (0.4056) Institutional quality 0.9775 -0.4753 1.7274 0.2788 -0.7615 1.2913 1.2801 -0.2971 1.8527 (3.1775) (3.9575) (4.4211) (4.5050) (4.8954) (4.1609) (4.1154) (4.3852) (4.7025) Initial GDP 1.1542 0.4277 0.5384 -0.7451 0.8766 0.5149 0.6113 -0.2953 (0.8665) (0.9942) (0.9986) (2.2385) (2.2883) (2.3237) (2.3376) (2.4152) Openness -0.0361 * -0.0314 -0.0477 ** -0.0479 * -0.0519 ** -0.0544 ** -0.0342 (0.0200) (0.0213) (0.0237) (0.0246) (0.0255) (0.0246) (0.0258) Investment 0.1459 0.1552 0.2106 0.2011 0.2200 0.2036 (0.1448) (0.1585) (0.1693) (0.1726) (0.1689) (0.1987) Literacy rate 0.0908 ** 0.1281 ** 0.1226 ** 0.1197 ** 0.0860 * (0.0415) (0.0593) (0.0586) (0.0579) (0.0464) Life expectancy -0.2489 -0.1878 -0.1903 (-0.0929) (0.2005) (0.1922) (0.1904) (0.1390) Landlocked 1.6777 0.9622 1.2675 (2.1501) (2.1771) (2.9243) SSA x Institutions 0.6545 0.1754 (0.4961) (0.4809) Discovery 0.0290 (0.1371) Discovery x Institutions 0.1080 (0.1370) R² 0.0138 0.0317 0.0645 0.0686 0.1457 0.1874 0.1989 0.2215 0.2369 N 59 59 59 56 50 50 50 50 44

(24)

21

Table 4. The effect of natural resources on social inequality. Dependent variable: The change of the difference between the average secondary-school enrolment rates of males and females between 1995 and 2015.

(1) (2) (3) (4) (5) (6) (7) (8) (9) Subsoil assets 0.1125 0.1790 0.1652 0.2708 0.3546 0.3338 0.3409 -1.7778 -4.7236 (0.3056) (0.3154) (0.3290) (0.3444) (0.4066) (0.4420) (0.4588) (1.9529) (4.1970) Institutional quality 4.3881 7.5973 9.0567 12.2071 * 15.8123 * 18.3727 * 17.2171 * 4.1241 -16.0605 (3.2800) (5.9804) (7.7014) (7.3489) 9.5240 10.1053 10.2248 17.8003 31.4636 Initial GDP -0.8135 -0.9527 -1.7304 -4.2899 * -2.7397 -2.6263 -2.4138 -0.4474 1.1122 1.2848 1.2739 2.5221 3.1235 3.2711 3.3167 3.0072 Openness -0.0143 -0.0168 -0.0392 -0.0346 -0.0338 -0.0233 0.0294 0.0371 0.0369 0.0414 0.0428 0.0440 0.0455 0.0603 Investment -0.3479 * -0.6841 *** -0.5912 ** -0.5640 ** -0.4204 -0.7940 ** 0.2077 0.2136 0.2505 0.2831 0.3038 0.4002 Literacy rate 0.2005 0.3058 0.2985 0.3572 * 0.1264 0.1299 0.1789 0.1817 0.2101 0.2344 Life expectancy -0.4369 -0.3757 -0.3405 0.0628 0.4218 0.4566 0.4276 0.4268 Landlocked 1.8180 -0.9942 -11.0429 6.2597 7.3207 7.1031 SSA x Institutions 2.0347 5.4285 1.7932 4.0247 Discovery -0.8898 0.6452 Discovery x Institutions -1.1227 0.6972 R² 0.0386 0.0505 0.0554 0.1090 0.3726 0.4137 0.4176 0.4372 0.5557 N 39 39 39 39 26 26 26 26 23

(25)

22

6. Conclusion

This thesis about the natural resource curse analyzed the effect of natural resources on economic growth, economic inequality and social inequality using cross-sectional data. Various steps were taken to tackle the endogeneity issues that arise due to the

interconnectedness of natural resources and institutional quality. Firstly an interaction term was included in the model, interacting subsoil assets with institutional quality. This was done in order to correct for a possible role of institutional quality as a moderator variable.

Thereafter, the time dimension in the data was exploited in order to construct a variable for resource discoveries. The goal of including this new variable, was to correct for a potential role of institutional quality as a mediator variable. Combining these two steps and including them both in the model, allowed for an isolation of the true effect of natural resources on GDP per capita growth, economic inequality and social inequality.

The results of the economic growth model show that the effect of natural resources on economic growth apart from its relation with institutions (thus channel A in figure 5) is positive and significant, thereby rejecting the existence of a curse. The results show that the effect of natural resources per se is positive, but that the role of institutions is crucial. A positive and significant coefficient on the interaction term shows evidence for the belief that the effect that natural resources have on economic growth depends on the level of

institutional quality. Therefore, institutional quality determines if natural resource

abundance turns out to be a curse or a blessing for a country in terms of economic growth. However, economic growth is only one of many indicators of a country’s wellbeing.

Therefore the effect of natural resources on inequality is addressed as well. Firstly an analysis is done using the change in economic inequality, measured by the Gini coefficient, as outcome variable. But, due to limited availability of this variable, the number of

observations in this model is much lower. Although the estimated coefficients have the expected sign, they are insignificant in this model. But while the effect of natural resource abundance on economic inequality remains inconclusive, the results do show that the role of institutional quality is crucial.

Finally, a model is employed that uses social inequality as outcome variable, which is measured by the change in the difference in school enrolment rates for males and females. Unfortunately the same availability issues as in the economic inequality model arise, and again the estimated coefficients have the expected sign but are statistically insignificant. Therefore the effect that natural resources have on this proxy of social inequality remains unclear as well.

Furthermore I have shown that institutional quality plays a crucial role in the effect that natural resources have on growth and inequality. Especially in the growth model, the results show evidence that institutional quality for a large part determines whether natural

resources are a blessing or a curse. Although this study makes the assumption that

institutional quality is constant over a time period of twenty years, this is not true for longer periods of time. Institutions are persistent, but not completely time-invariant (Rodrik et al., 2004). This means that countries that experience negative effects of natural resource

(26)

23

extraction due to poor institutions can still attempt to overturn this. These countries should attempt to develop policies that enhance institutional quality.

There are still some last side notes that have to be made. One that is already mentioned is the fact that institutional quality is an arbitrary term, and that this variable is difficult to quantify. This should be kept in mind when looking at the results and drawing conclusions from them.

Lastly, another important point is that every observation in this analysis comprises a country. Not one country is the same, and therefore the many different, complicated

mechanisms at hand might apply differently to each country. There exists no ‘one size fits all’ solution in this context, and although this study contributes to the overall understanding of the natural resource curse, every case should be examined separately.

Although the current research did contribute to a better understanding of the forces surrounding the natural resource curse, further research is certainly required. In order for future research to further help understand this issue and avoid problems that are

encountered in this paper, it is crucial that extensive data keep getting collected and made available. Further research, based on this additional data, can provide recommendations for solid policy implications. Since not all results obtained in this research were conclusive, it is not possible to draw such implications from it.

(27)

24

References

Alexeev, M., & Conrad, R. (2009). The elusive curse of oil. Review of Economics and

Statistics, 91(3), 586-598. doi:10.1162/rest.91.3.586.

Auty, R. (2001). The political economy of resource-driven growth. European economic

review, 45 (4), 839-846.

Bardhan, P. (1997). Corruption and development: a review of issues. Journal of Economic

Literature , 35 (3), 1320-1346.

Basedau, M. (2005). Context matters - Rethinking the resource curse in sub-Saharan Africa.

SSRN Electronic Journal. doi:10.2139/ssrn.906983.

Brunnschweiler, C. N. (2008). Cursing the blessings? Natural resource abundance, institutions, and economic growth. World Development, 36(3), 399-419.

Brunnschweiler, C. N., & Bulte, E. H. (2008). The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and

Management, 55(3), 248-264. doi:10.1016/j.jeem.2007.08.004.

Bulte, E. H., Damania, R., & Deacon, R. T. (2005). Resource intensity, institutions, and development. World Development, 33(7), 1029-1044.

Buccellato, T. & M. Alessandrini (2009), Natural resources: A blessing or a curse? The role of inequality, Discussion Paper 98, SOAS, University of London.

Butkiewicz, J. L., & Yanikkaya, H. (2010). Minerals, institutions, openness, and growth: An empirical analysis. Land Economics, 86(2), 313-328. doi:10.3368/le.86.2.313. Carbonnier, G. (2011) Introduction: The global and local governance of extractive

resources. Global Governance: A Review of Multilateralism and International

Organizations: April-June 2011, Vol. 17, No. 2, pp. 135-147.

Carbonnier, G., Brugger, F., & Krause, J. (2011). Global and local policy responses to the resource trap. Global Governance, 17 (2), The Governance of Extractive Resources

(Apr.-June 2011), 247-264.

Frankel, J. (2010). The natural resource curse: A survey. NBER Working Paper Series 15836. http://www.nber.org/papers/w15836

Goderis, B., & Malone, S. W. (2011). Natural resource booms and inequality: Theory and evidence. Scandinavian Journal of Economics, 113(2), 388-417. doi:10.1111/j.1467-9442.2011.01659.x.

Groenewegen, J., Spithoven, A. H., & Van den Berg, A. (2010). Institutional economics: An

introduction. Basingstoke, England: Palgrave Macmillan.

Gylfason, T., & Zoega, G. (2002). Inequality and economic growth: Do natural resources matter?, CESifo Working Paper Series 712, CESifo Group Munich.

Hammond, J. L. (2011). The resource curse and oil revenues in Angola and Venezuela.

Science & Society, 75(3), 348-378. doi:10.1521/siso.2011.75.3.348.

Isham, J. (2005). The varieties of resource experience: Natural resource export structures and the political economy of economic growth. The World Bank Economic Review,

19(2), 141-174. doi:10.1093/wber/lhi010.

Kaufmann, D., Kraay, A., & Mastruzzi, M. (2010). The worldwide governance indicators: A summary of methodology. Data and Analytical Issues, World Bank Policy Research

Working Paper 5430.

Leite, C., & Weidmann, J. (2002). Does mother nature corrupt? Natural resources, corruption, and economic growth. In G. Abed & S. Gupta (Eds.), Governance,

The natural resource curse : an empirical paper analyzing the effect of natural resources on growth and inequality and the role of institutions