The Curse of Natural Resources

(1)

The Curse of Natural Resources

Is economic diversification a way out?

Veronika Mikes S2757958 MSc in International Economics and Business

(2)

(3)

1 Abstract

Natural resource rich nations have a tendency to depend too much on their resources, hence building concentrated economies susceptible to the curse of the natural resources. If this statement is true, does that mean that a more diversified economy is less probable to face the symptoms of the curse? In this paper I ask whether economic diversification is capable of mitigating the effects of the curse. In order to answer this question I incorporate a variable of economic diversification and its interaction term with natural resource abundance to the growth equation applied by Sachs and Warner (1995). The results show that economic diversification is indeed capable of mitigating the curse, however this effect might depend on the level of education and institutional quality.

(4)

2

1. Introduction

Economic growth - one of the biggest objectives of any government around the world. This is no surprise if we consider that economic growth is associated with higher living standards and the overall good performance of the economy. What captured the interest of many economists is the fact that economic growth varies greatly between countries and time periods with different factors involved, presenting an (almost) inextricable puzzle for anyone who is trying to design a proper development plan for a country. Thus the question is what drives economic growth? Samuelson and Nordhaus (2010) summarized the answer calling the four major factors the "four wheels of economic growth"; human capital, natural resources, physical capital and technological change. He pointed out, that no matter what development path a country may take, these four factors, although in varying compositions, always emerge. The topic of this paper relates to one of these "wheels", namely natural resources, and its ambiguous relationship with/to economic growth. Common sense would dictate that natural resources provide beneficial initial conditions for economic growth. For instance selling these resources could raise the wealth of the people of the country, as well as the purchasing power for imports, not to mention it could generate funds for future investments. Unfortunately the experience of resource-rich nations seems to point to a different conclusion. Aside from a few exceptions, most notably Norway, nations with natural resources of considerable size tend to perform worse in terms of economic growth than their resource-poor peers (Auty & Warhurst, 1993; Sachs & Warner, 1995; Gylfason, 2001; Papyrakis & Gerlagh, 2004).

Figure 1. Simple Association between natural resource abundance and economic growth1

1

(6)

4

Empirical tests show that one standard deviation increase in the initial share of primary exports to GDP is associated with an average 1% decrease in GDP per capita per annum (Sachs & Warner, 1995). Based on this negative relationship the question emerged whether the possession of natural resources might actually hinder economic growth, a famous paradox named the resource curse. The opinions regarding this question are divided, there are strong advocates for the theory (Gylfason, 2001; Sachs & Warner, 1995; Hausmann & Rigobon, 2002; Behbudi, Mamipour, & Karami, 2010), as well as adversaries, who believe the curse is a "red herring" (Manzano & Rigobon, 2001; Stijns, 2005; Brunnschweiler, 2008; Brunnschweiler & Bulte, 2008), hence no consensus reached until today. Thus, the question is, under what circumstances natural resources act as a curse and not a blessing. (Sala-i-Martin & Subramanian, 2003).

(7)

5

In this paper I am analysing the relationship of economic diversification to the resource curse and test whether it makes a difference if a country is more diversified. I raise the question: can economic diversification mitigate the effects of the resource curse? In order to answer this question I calculate an entropy measure for economic diversification for 40 nations between 1970-2010 based on sectoral data and introduce it to the growth model commonly applied (Sachs & Warner, 1995; 1997) in the resource curse literature.

(8)

6

2. The natural resource curse

Natural resources play a significant role in economic development, although their role might be different from what we have initially thought.

According to economic theory natural resource abundant nations have better initial conditions than resource poor ones, since they have tremendous amount of physical capital at their disposal. It can be used as inputs to the industry or can be exported generating serious amounts of money inflows to the economy. The accumulating wealth can provide the basis for investments in education, healthcare, innovation and so on. (Nurkse, 1953; Rostow, 1960) The historical cases of Germany and Great Britain support this reasoning - the industrial revolution could not have happened without the coal and ore reserves. However the list of successful examples of resource-based economic growth is rather short. Aside from Norway, and recently strengthened Botswana not many resource rich nations could deliver positive results in the last few decades. It turns that, even though there have been unsuccessful attempts previously as well, decreasing transportation costs and the changing nature of resources (e.g. from coal to oil) created new barriers for resource rich countries to face. (Auty & Warhurst, 1993; Sachs & Warner, 1995) Furthermore, not only resource rich nations tend to perform weakly, but they have been significantly outperformed by resource poor countries in the second half of the 20th century (Auty, 1993). Figure 2 illustrates this observation. Figure 2. GDP per capita growth in resource rich and resource poor countries 1953-20102

2

(9)

7

While resource poor Japan, Korea and Taiwan generated outstanding economic growth, resource rich Bolivia, Mauritius, Nigeria and Venezuela have mostly been stagnating in the last six decades.3 This counterintuitive observation points to the possibility, that natural resources might play a significant and most importantly negative role in driving economic growth. Empirical testing seems to reinforce this assumption. Cross-country regressions show that natural resource wealth is negatively associated with economic growth, even after controlling for variables classified important for growth by the economic development literature (Sachs & Warner, 1995; 1997; Papyrakis & Gerlagh, 2004; Alexeev & Conrad, 2009; Davis, 2013)4. This negative relationship has been named the resource curse paradox. Sachs and Warner (1995) applied ordinary least squared methods and run cross-country regressions for 95 countries. Their study covered 19 years, between 1971-1989, which period was later extended to 1990 (Sachs & Warner, 1997). They used GDP per capita to calculate growth, because it gives a better indication on how natural resources affect welfare. They also assumed exponential growth trends, so they defined growth as the difference of the logarithm of GDP per capita at the end and at the beginning of the time period.5

They chose their control variables from the growth literature, thus they used market openness, bureaucratic efficiency, trade policy, inequality, investment and education. They concluded that higher share of primary exports (SXP) to initial national income at the beginning of the period correlates with a generally lower rate of economic growth in the next twenty years. Nonetheless, it had been argued that this estimation technique might oversimplify the situation and the lower growth can be the result of other factors. For instance, Manzano and Rigobon (2001) found, that the negative relationship becomes ambiguous when panel data methods are applied. Yet Papyrakis and Gerlagh (2007) and Behbudi, Mamipour and Karami (2010) found significant negative relationship, when applying pooled ordinary least squared and fixed effect estimation methods.

3_{For the resource wealth of depicted countries see Appendix 1} 4

See Figure 1 in introducton

(10)

8

2.1 Explanations of the Resource Curse

The theory does not claim that natural resources directly lead to slow economic growth, rather it presses that said resources cause certain distortions in the economy, which in turn lead to low economic performance. Researchers often call these distortions "transmission channels" trying to identify all possible factors and working on models explaining them (Torvik, 2001; Gylfason, 2001; Bulte, Damania, & Deacon, 2005) The three main channels identified by the literature are (1) Dutch disease, (2) rent seeking behaviour and quality of institutions, and (3) volatility in commodity prices (Sala-i-Martin & Subramanian, 2003). Gylfason (2001) adds to these two more: (4) overconfidence and (5) neglect of education and human capital.

(1) Dutch disease

One of the most commonly cited theories behind the resource curse is the famous Dutch disease explanation. In order to quickly summarize this model, let us suppose that there are some factors and activities that are driving economic growth, but when a country possesses considerable amount of natural resource, such as oil, these beneficial factors disappear. Most specifically they are "crowded out" by the natural resources. When a boom occurs in raw-material exports the real exchange rate appreciates damaging other activities of the economy. (Corden, 1984; Sachs & Warner, 2001) .

Matsuyama (1992) in his "linkages approach" assumed that there are only two sectors, agriculture and manufacturing. If there are forces pushing the economy away from manufacturing the growth will decline. The problem is that in the "linkages approach" model agriculture directly employs factors which otherwise would be used in manufacturing, but, since oil production requires very few labour, this approach is not applicable to the resource curse in its original form. Thus Sachs and Warner (1995) extended Matsuyama's model introducing three sectors: a tradable natural resource sector, a tradable manufacturing sector, and a non-traded sector

When the natural resource endowment is big, tradable production is concentrated in the natural resource sector, while the demand for non-traded goods increases pulling labour and capital to the non-traded sector. In short, when there is a natural resource boom, the manufacturing sector shrinks while the non-traded sector grows. This "shrinkage" of the manufacturing sector is called the Dutch disease.

(11)

9

structural explanation arguing that the natural resource sector does not possess the necessary backward and forward linkages in the economy, while manufacturing does. Although the exceptional role of the manufacturing sector in economic development is a widely excepted thesis (Lewis, 1954; Rodrik, 2013) there is no actual evidence that other sectors cannot generate learning-by-doing (Stijns, 2005; Hausmann & Rigobon, 2002), for instance Torvik (2001) presents a theoretical model where both traded and non-traded sector have learning-by-doing. It is also possible that the natural resource sector itself can drive learning-by-doing, but it might be dependent on the type of natural resource we are talking about, diffuse or point. Diffuse resources are spread over the place using considerable amount of input, especially labour. A typical diffuse resource is arable land. A point resource is concentrated in one place and do not require much labour, for example crude oil. Point resources are considered to be more harmful than diffuse resources, although the degree of the differences is still debated. (Brunnschweiler, 2008; Sachs & Warner, 2001)

(2) Rent-seeking and quality of institutions

The basic assumption of the rent-seeking model is that natural resource rents are easily appropriable, which causes entrepreneurs to move away from productive activities. It also decreases the incentive to reform or build an effective tax system. The curse of the easy riches pushes economies towards corruption, bribes and mismatched public policies. Many agrees that this effect is the most deleterious out of all, not just for the economy, but also for the society as a whole, although the direction of the causality leaves some doubt behind. (Torvik, 2002; Brunnschweiler & Bulte, 2008; Bardhan, 1997) As a result natural resource-rich nations are more prone to be involved in political violence or armed conflict (Basedau & Lay, 2009; Ross, 2004). On the other hand initial institutional quality can make a difference. Those countries, which had their institutions and property rights established before the discovery of oil or minerals faced much less adverse effects of the curse (Stijns J. C., 2005; Bulte, Damania, & Deacon, 2005; Leite & Weidmann, 2002).

(3) Volatility

(12)

10

sector can exceed 50% of the GDP and other sectors of the economy can also be linked directly or indirectly to the natural resource sector, making the economy rather vulnerable to the "boom and bust" cycles of primary commodities (Gelb, 2010). The external shocks can even cause state budget instability, since most of the government revenues are coming from the export sales (Weinthal & Luong, 2006).

It is argued, that volatility might be the key channel of the resource curse. Van der Ploeg and Poelhekke (2008) found that commodity price volatility results in volatility in GDP per capita growth in resource dependent countries regardless of the type of primary product. The volatile unanticipated output growth affects growth negatively. Figure 3 shows that the more dependent a country becomes on its resources, the more volatile its GDP per capita becomes. Figure 3. The resource curse and volatility6

(4) Overconfidence

This channel describes exactly what its name suggests. The easy income coming from natural resources spoils countries, making them believe that they do not need to invest in other sectors or improve their institutions. This overconfidence pushes them to depend even more on their raw materials. (Gylfason, 2001)

6

(13)

11 (5) Neglect of Education and Human Capital

Gylfason (2001) bases its argument on the fact that the extracting industry does not require a lot of manpower, furthermore mostly low-skilled workers. Consequently, countries highly involved in primary production do not feel the need to invest in education. The role of education in economic development has long been accepted by the scientific community. Many argued that education influences living standards (Mankiw, Romer, & Wile, 1992), makes it easier for people to adapt to change and learn new technologies (Nelson & Phelps, 1966) and increases overall productivity (Lucas, 1988). Ergo, if natural resources are really hindering human capital accumulation, then in turn they would hinder economic growth as well. However it is important to mention that the relationship between natural resources and human capital accumulation is less unequivocal, than the relationship between human capital and economic growth. Stijns (2006) for example finds a positive relationship between natural resource abundance and human capital, while Shao and Yang (2014) conclude that human capital is essential in evading the resource curse. Behbudi et al. (2010) used panel data to test human capital as a transmission channel. They separated their subject countries into two groups based on their resource dependence; major petroleum exporters, whose share for petroleum exports exceeds 50% of all exports and smaller oil exporters. By introducing an interaction term of natural resource abundance and education, using secondary school attained and literacy rates, they found that higher level of human capital seems to offset the negative effects in the case of smaller exporters, but major petroleum exporters do not generate enough human capital to cause a significant change in the curse.

2.2 Economic Diversification and the Resource curse

In order to look at the importance of economic diversification in the resource curse paradox, firstly it is necessary to specify what I mean exactly by the term 'economic diversification' and how is it related to the curse.

(14)

12

There are several arguments for economic diversification. For instance, Todaro and Smith (2009, p. 115) depict economic development as a process of structural change through which economies move away from agriculture and primary sector to manufacturing and service industries. This view is shared by others, to the point that Thompson and Lanier (1987) argued that economic development can be measured by the level of economic diversification. The portfolio arguments are just as popular. Sharing the reasoning with investors, this theory emphasizes the risk of 'putting all our eggs in one basket'. A concentrated economy is more susceptible to external shocks and volatile commodity prices - the major concern of concentrated primary exporters (Weinthal & Luong, 2006). Additionally, it has been highlighted that the changing preferences of consumers can instigate economic diversification (Imbs & Wacziarg, 2003). As income increases the structure of demand follows it and the production has to satisfy more varied needs. However, open economies might behave differently. With decreasing transportation cost and tariffs it is not particularly necessary to produce the goods themselves. Krugman (1991) for instance pointed out that demand externalities make it beneficial for producers to cluster creating regional concentration. The need to locate close to demand is losing significance. Promoters of economic diversification tend to disregard the fact that specialization can have its own benefits as well. Countries with comparative advantages in certain products should parlay this advantage and reap the benefits of their better conditions, as Ricardo argued (Imbs & Wacziarg, 2003).

An interesting feature of economic diversification has been discovered by Imbs and Wacziarg (2003). They examined the process of economic diversification in relation with increasing income and they concluded that the relationship is not a linear one. At first countries indeed tend to diversify, but after a certain level of development - considerably late in the development process- they start to specialize again, forming a U-shape, although not an entirely symmetrical U. This U-shape appears, when examining export concentration as well (Klinger & Lederman, 2006). This finding raises two interesting questions: (1) countries with low levels of initial income are expected to diversify, while (2) countries with high initial income might have concentrated economies as well, without experiencing the symptoms of the resource curse.

(15)

13

the resource curse harms. Thus the feasibility of the diversification strategy in resource rich nations raised serious doubts. (Berezin, Salehizadeh, & Santana, 2002; Weinthal & Luong, 2006)

2.3 Measurement of natural resource abundance and economic

diversification

How can we me measure natural resource abundance? Or how should we measure it? It is a central question to the resource curse literature, since the choice of this variable greatly determines the outcome of the analysis. Sachs and Warner (1995) chose natural resource based exports to GDP as their main measure and called it 'resource intensity'. They also tried three more measures - share of mineral production to GDP, fraction of primary exports in total exports and log of land area per person - later when checking for robustness. This choice of measurement was widely criticised, saying that it measures resource dependence rather than actual abundance. The problem is three-fold. Firstly, it is possible for a country to use its natural resources as inputs for the manufacturing sector. In that case resources are present in both primary and manufacturing export. Secondly, resource dependence can be the result of development failure for many other reasons, than natural resources. Poorly constructed economic policy could lead to an overly concentrated economy. Lastly, the effect of natural resource abundance on economic growth is somewhat determined by the growth model applied. (Stijns, 2005) It is also emphasized, that instead of an initial resource intensity a period average would be a much more reliable measure, mitigating the volatility of primary exports (Brunnschweiler, 2008). Stijns (2005) concludes that Sachs and Warner's results are not robust to changes in the measure of resource abundance.

Instead of Sachs and Warner's trade flow measure, it is suggested to use reserves per capita or production (Stijns, 2005). Brunnschweiler (2008) introduces two new measures developed by the World Bank, per capita mineral and total natural resource wealth.

(16)

14

Thus, making the distinction between diffuse and point resources is not as quantitatively important as people might think.

Moving to the selection of an appropriate measure for economic diversification, there are many to choose from. Most commonly it is measured (1) by the distribution of GDP or by the allocation of other resources, for example labour, to different sectors of the economy or (2) by trade statistics (Imbs & Wacziarg, 2003). But trade statistics largely disregards any other usage natural resources might have aside from export, as well as other economic activities satisfying local consumption. A country with a concentrated export portfolio might have a relatively diversified economy and the negative effects of the resource curse are associated with a concentrated economy. Therefore trade statistics are not suitable for the topic of this thesis. Thus, I prefer to use the first measure.

For the exact measurement method of economic concentration, there are several widely used technique in the literature.

Firstly, the sum of square method, such as the Herfindahl-index, which uses the market shares to measure market concentration. The value ranges from 0 to 1 as the economy is getting more concentrated. There are two potential problems with this method. One is, that the value of the measure depends on the number of participants (N). If N is low, the Herfindahl-index could never reach 0, even if all participants have the same share. Since the number of economic sectors is limited a lower and upper limit is bound to appear narrowing the original range. Two is, that these kind of indices are very sensitive to large shares, causing possible bias. (Ray & Singer, 1973)

Secondly, there are the inequality indices, for instance the GINI coefficient. According to Ray and Singer (1973) the biggest problem with these measures that they only work sufficiently, when N is large.

(17)

15

2.4 Hypothesises

Although the resource curse is supported by several empirical studies, there are also arguments against it, making the issue ambiguous. Since the thesis is based on the original assumption that the resource curse does exist, I need to see whether this assumption holds true in my dataset.

Hypothesis 1. Natural resource abundance expected to negatively correlate with economic growth.

So if the curse is real, then where does it come from? What are the reasons behind it? Based on the previous review of the literature it can be concluded that great part of the theory originates the problems from overly dependence on natural resources. They argue that this extreme dependence makes countries susceptible to volatility, erodes their institutions and makes them neglect their education, thus hindering economic growth. Overdependence means that most of their revenues are coming from their sales of primary resources, thus the structure of their economies are highly concentrated. Consequently, if the symptoms of the resource curse is really caused by overdependence and concentrated economic structure, then countries with more diversified economies should not experience the same effects. In other words higher levels of economic diversification should mitigate the harmful effects of the resource curse.

Hypothesis 2. As economic diversification increases natural resource abundance is expected to correlate positively with economic growth

(18)

16

income diversification can mitigate the effects of the curse, while one increase in labour diversification cannot.

Hypothesis 2a. As economic diversification in terms of income increases natural resource abundance is expected to correlate positively with economic growth

Hypothesis 2b. As economic diversification in terms of labour increases natural resource abundance is expected to correlate negatively with economic growth

However, the previous statements might become void in the case of high income countries, since these nations are expected to behave differently in many ways. Based on the findings of the literature on economic diversification, developed countries are expected to specialize after they reached their level of maximum diversification. Therefore the explained mitigation effect seems illogical in their case. Also, these countries have efficient production technologies, therefore the gap between the labour employed and revenue generated in the agriculture loses its significant. Ergo, economic diversification measured in labour and income should behave similarly.

Hypothesis 3a. Economic diversification is expected to correlate negatively with economic growth in high income countries

Hypothesis 3b. As economic diversification in both labour and income increases natural resource abundance is expected to correlate negatively with economic growth

2.5 Theoretical background

(19)

17

Sachs and Warner (1995) in their "Dynamic Dutch-Disease Endogenous Growth Model" assume that there are three sectors in the economy: a traded manufacturing sector, a non-traded sector and a natural resource sector. The natural resource sector does not employ any capital or labour, but has a constant flow of output, which can be sold at an exogenous world price. Thus they define the production function of the traded (m) and non-traded sectors (n)7 as:

Xm =G(Lm, Km) and Xn =F(Ln, Kn) (1)

Their major assumption is that employment in the manufacturing sector generates the accumulation of knowledge, which raises effective labour in all sectors and leads to economic growth. They introduced variable H, for human capital and argued that it represents "the stock of knowledge in the economy", while ϴ represents the share of labour in the traded sector. So knowledge accumulation can be defined as

Ht = Ht-1(1+ϴt-1) (2)

If we suppose that total labour is 1, the production functions can be rewritten as

Xm =G(ϴH, Km) and Xn =F((1-ϴ)H, Kn) (3)

Following the logic of the Solow model constant return to scale is assumed, thus

xm=g(km), where km=Km/ϴH (4)

and

xn =f(kn), where kn= Kn/(1-ϴ)H (5)

So, the output of the traded and non-traded sector is defined by the amount of capital per effective worker.

When a resource boom occurs the wealth of the people will increase. They can decide to spend this extra wealth on non-traded goods generating demand for non-traded goods. The increasing demand will draw labour from the traded sector to the non-traded sector, thus ϴ will decrease. According to equation (2) it will lead to lower knowledge accumulation, the main driver behind growth.

Based on the above argumentation the general form of the resource curse growth equation is given by

7

(20)

18 Gi=β0+β1 ln(Y i 0)+β2R i +β3Z i +εi (6)

where economic growth (G) depends on initial income (Y0), natural resource abundance (R) and a vector of explanatory variables (Z) in country 'i'. Initial income is included because of the convergence assumption, i.e. richer countries expected to grow slower. Thus initial income and growth is expected to negatively correlate. The explanatory variables account for the indirect effects of the resource curse, aka the transmission channels and other factors identified by the literature as drivers behind economic growth. The resource curse hypothesizes that natural resource abundance is negatively correlated with economic growth, so the curse is proved if β2 is negative and significant.

In order to test my hypothesises I introduce economic diversification (D) as an explanatory variable to the aforementioned equation:

Gi=β0+β1 ln(Y i 0)+β2R i +β3D i +β4 Z i +εi (7)

(21)

19

3. Methodology

The empirical testing of my hypothesises consists of two major parts. In the first part I run ordinary least square regressions (OLS) on the whole dataset to test the original resource curse. Then I separate the countries into two groups: (1) low and middle income countries and (2) high income countries, based on the World Bank classification (World Bank, 2015), and use panel data methods. I hypothesized that high income countries are behaving differently, which I have to control for. Implementing a dummy variable might be a viable option, but there are several reasons why I decided not to. Firstly, there is no single income level at which countries reach their maximum level of diversification and start to specialize, so I cannot introduce a dummy based on that. In other words, I cannot separate the development path into phases based on the direction of the diversification process, even though that would capture the most precisely the effects of diversification. Therefore, I have to assume, based on the works of Imbs and Wacziarg (2003), that the U-shape of diversification applies only to high income countries.

Secondly, a dummy variable for high income countries is possible, but creating two groups gives more flexibility in the analysis. Since the two groups are significantly different in many of their characteristics and expected behaviour, this flexibility might become crucial. Especially, because it makes it possible to use different models if needed, a great strength of panel data sets .

Since my paper is focusing on economic diversification the availability of sectoral data is limiting my choice of countries. I decided to use the 10-sector database of the Groningen Growth and Development Centre (GGDC) (Timmer, de Vries, & de Vries, 2014). This database is available for 42 countries with different income levels. From the group of subject countries West Germany was dropped because it had ceased to exist after the unification of Germany. Taiwan was also dropped because of no available data for natural reserves. Therefore, my group of subject countries consists of 40 nations, out of which 12 has been classified as high income.8 Figure 4 shows the change in economic diversification9 between 1970-2010 in this 12 countries.

8

For the list of countries see Appendix 1

9

(22)

20

Figure 4. Patterns of economic diversification in high income countries (1970-2010)10

Decreasing economic diversification indeed seems to be common, in some cases only in the last few years, while others, for example Hong Kong, left their maximum level of diversification long ago. Comparing it to the low and middle income nations it can be deduced that high income countries indeed seem to specialize at a rate, which is not visible in the case of low and middle income countries.11 This observation supports the separation into two groups.

3.1 Choice of Measurement and Data Description

12

One of the critics mentioned, that economic growth is taking place in the long run and Sachs and Warner's time period is way too short to capture the process (Maloney, 2001). The GGDC database is available between 1970-2010, so I could extend it to forty-one years.

10

Data from GGDC 10-sector database (Timmer, de Vries, & de Vries, 2014), value added. Authors own calculations, for methodology see below

11

For figure on low and middle income countries see Appendix 3

12

(23)

21

I could not find a better database for economic diversification, than the GGDC 10 sector. Other possible databases (ILO, 2014; UNIDO, 2014) are covering shorter time periods or concentrating only on the manufacturing sector, hence not suitable for the purpose of this paper. The GGDC database covers all economic sectors providing data for employment and value added for a diverse group of nations. The group consists of countries from different geographic regions, both low and high income as well as both resource-rich and poor.13 The data for the other variables were taken from the Penn World Tables14 (Heston, Summers, & Aten, 2012; Feenstra, Robert, Inklaar, & Timmer, 2015) and from the World Bank databases (World Bank, 2011; 2014).

First of all, following previous researchers (Sachs & Warner, 1995; Brunnschweiler, 2008) I decided to use real purchasing power parity adjusted GDP per capita (rgdpl) acquired from the Penn World Tables 7.1 (Heston, Summers, & Aten, 2012) for initial income (lrgdpl70, lrgdplini) and for the calculation of economic growth (G7010, rgdpl7010).

Secondly, I decided to use natural reserves per capita (rescap) as a measure for natural resource abundance. I opted for the per capita measure to avoid bias coming from differences in country size. The World Bank (World Bank, 2011) provides estimates for natural capital wealth calculated as the sum of crop, pasture land, timber, non timber forest, protected areas, oil, natural gas, coal, and minerals.

As I have argued before natural resource dependence should appear in the economic diversification variable to some extent, thus for this research reserves are a more viable choice. However data for reserves are limited. The World Bank only provides estimates for 1995, 2000 and 2005. To deal with this issue I calculated a period average (rescapavg), which can serve as an indicator for the overall resource wealth of a nation in the period accounting for deplation. I also introduced an alternative measure for resource abundance; resource rents to GDP (rent) measured in percentages.

Thirdly, I use the GGDC 10-sector database (Timmer, de Vries, & de Vries, 2014) to calculate my economic diversification measures. The database covers agriculture, mining, manufacturing, utilities, construction, trade services, transport services, business services, government services and personal services. I calculated the diversification variable from value

13

The database was constructed using data from national accounts, therefore the differences in data collection methods across countries and time might weaken the reliability of the data, which has to be taken into consideration in the analysis.

14

(24)

22

added in constant prices (divval) as my measure for diversification in income and labour engaged (divlab) for diversification in labour.

Literature on economic diversification favours the entropy index, thus I also opted for it. The entropy index is calculated as follows (Smith & Gibson, 1988):

Where Si is the share of sector i and N is the number of sectors. Fourthly, I chose my four control variables.

For education (educ) the education statistics compiled by Barro and Lee (2010) are highly popular, however the newest version of the Penn World Tables (Feenstra, Robert, Inklaar, & Timmer, 2015) provides a new, more comprehensive index for human capital; the index of human capital per person based on schooling and return on education. Data is available between 1970-2010.

Data for openness (openk) and investment (ki) were taken from the Penn World Tables 7.1 (Heston, Summers, & Aten, 2012). The openness indicator is calculated by summing export and import and divided by constant GDP per capita, while investment is defined as the share of real PPP adjusted GDP per capita at constant prices (2005).

Lastly, finding an appropriate indicator for institutional quality is not easy. I wanted a measure, which accounts for the overall quality of the institution in the country. Simply an indicator for corruption is not enough, since economic diversification depends on the development of the private sector, where agents base their investment decisions on, for instance the protection of property rights. The World Bank Worldwide Governance Indicators (World Bank, 2014) provides an indicator called "Rule of law" (ruleoflaw), which measures the confidence in the enforcement of the rules of society, contract enforcement, property rights, police and courts. It ranges from approximately -2.5 to 2.5. However, the data is only available from 1996 to 2010, and some data points are missing

(25)

23

most of the Asian emerging economies just started their ascension at that time. The same goes for openness, as the Asian high income countries are major exporters, while the European countries, e.g. France is more closed.

This perceived homogeneity of the high income group compared to the low and middle income countries supports that the two groups of countries need different estimation methods . Nonetheless, both groups are quite varying in terms of natural resource wealth, including countries with huge natural reserves, thus the resource curse hypothesis can be tested.

Table 1. Summary Statistics for high income countries

Variable | Obs Mean Std. Dev. Min Max ---+--- rgdpl70 | 97 13491.21 5598.391 816.2062 20436.2 rgdplgrowth | 97 .0279923 .0218927 -.0108403 .1051894 rescapavg | 97 8708.282 6275.25 2.490951 19728.63 divval | 97 1.893052 .1075823 1.602044 2.084798 divlab | 97 1.886865 .1021788 1.415101 2.024881 ---+--- educ | 97 2.709083 .4108115 1.875368 3.596745 ki | 97 .257378 .0764923 .1507414 .4781831 openk | 97 .7283842 .8393765 .1038404 4.189115 ruleoflaw | 38 1.29853 .5697525 -.686711 1.949556

Table 2. Summary statistics for low and middle income countries

Variable | Obs Mean Std. Dev. Min Max ---+--- rgdpl70 | 224 2798.548 2380.311 459.4649 9308.395 rgdplgrowth | 224 .0221336 .0318781 -.0586309 .1883548 rescapavg | 216 7156.382 5639.485 1331.286 29471.44 divval | 224 1.946289 .1958704 1.063803 2.153216 divlab | 223 1.543564 .4355484 .4219254 2.117308 ---+--- educ | 208 2.02138 .4125679 1.165083 2.92111 ki | 224 .2346907 .0956842 .0687386 .6214825 openk | 224 .5380508 .3361757 .0398922 2.049016 ruleoflaw | 86 -.2063205 .608433 -1.500139 1.278001

Looking at the diversification variables a very important thing can be observed. In the case of high income countries the mean of divval and divlab are almost the same, while they differ significantly in low and middle income countries. This seems to be in line with hypothesis 2a, 2b and 3b.

(26)

24

3.2 Estimation method

15

The paper is interested in the interactions between natural resource abundance, economic diversification and economic growth. For that purpose three estimation methods are applied; ordinary least squared (from here on OLS), pooled OLS and fixed effects model. The estimations have two major parts. My investigation starts with the recreation of Sachs and Warner's (1995) basic regression on cross-sectional including all the countries using OLS methods. Based on the growth equation (6) in section 2.5 the following model is estimated: G7010i=α0+ α1lrgdpl70i+ α2rent70i + εi

Where G7010 is the growth rate between 1970 and 2010 for country 'i', calculated as G7010i=(1/T)ln(Yi2010/Y

i

1970), lrgdpl70 is the log of initial income in 1970 for country 'i' and rent70 is share of resource rents to GDP in 1970.

Based on the discussion in section 2.3 I estimate the same regression with an alternative measure for natural resource abundance, in order to test the model for sensitivity in change of measurement. Rescapavg is the period average of natural resource reserves per capita.

G7010i= β0+β1lrgdpl70i +β2rescapavgi + εi

In the second part I turn to the panel dataset, where I separate the countries based on their income status.

Low and middle income countries form a heterogeneous group, thus a model allowing for individual heterogeneity is preferable. Comparing pooled OLS, fixed effects and random effects models, pooled OLS drops out immediately because of the heterogeneity issue. Since I conjecture possible endogeneity of my explanatory variables I run the Hausman test to check the applicability of the random effects model. As expected correlation between the independent variables and the random error term is detected, thus the random effect model is dropped as well. Consequently, the fixed effects model is chosen for estimation in the case of low and middle income countries. However, this model raises two problems. Firstly, fixed effects model cannot handle time invariant variables, while the initial income is a constant variable and a fundamental pillar of the theoretical model. To solve this problem I created eight sub-periods in a similar fashion as Behbudi, Mamipour and Karami (2010).16 Each period covers five years and the GDP per capita of the first year of the period is taken as

15

Estimations have been carried out using STATA 13.1. Empirical specifications are based on the book "Principles of Econometrics" (Hill, Griffiths, & Lim, 2011)

16

(27)

25

initial income. All the other variables are redefined using their period average. This method has an additional benefit, namely it helps dealing with missing data points. On the other hand, structuring the model this way changes the form of the data to "short and wide", which raises the second problem. The number of countries is 28, which is too high to allow for different coefficients for every individual, eliminating the least squares dummy variable model (LSDV) from the choice of possible methods. Hence the assumption is made that all individual heterogeneity is captured be the individual intercepts and the slope coefficients are the same for every individual.

The estimation has two stages. First, I estimate the following regressions with my two measures for economic diversification:

rgdpl7010it=γ1t+γ2 lrgdpliniit+γ3rentit+γ4Dit+γ5kiit+γ6openkit+γ7educit+γ8ruleoflawit+εit

Where rgdpl7010 is the period average of the log of annual growth for country 'i' and time 't', lrgdplini is the initial income of every sub-period and the other variables are the period averages of their respective measures. 'D' stands for economic diversification in value added (divval) and in labour engaged (divlab) in the two regressions, which tests the individual behaviour of the two different diversification measures. The measure for resource abundance is changed to rents, because the fixed effects model cannot handle the time invariant

rescapavg and there are not enough reserves data to provide estimates for every sub-period. In the second stage, an interaction term between resource abundance and economic

diversification is introduced. This interaction term measures whether the negative effect of natural resources decreases in relation with economic diversification.

rgdpl7010it=∂1t+∂2 lrgdpliniit+∂3rentit*Dit+∂5kiit+∂6openkit+∂7educit+∂8ruleoflawit+εit If ∂3 is positive, hypothesis two is accepted.

(28)

26

4. Results

I present my results in three sections. Firstly, I address hypothesis 1 and investigate the original resource curse hypothesis based on the results from my cross-sectional and panel data methods. Secondly, I explain the situation regarding low and middle income countries focusing on hypothesis 2, 2a and 2b. Lastly, I look at the results of the high income country group regarding hypothesis 3 and compare the results to the previous section.

Table 3 summarizes the results. The cross country regressions showed negative relationship between natural resource abundance and economic growth regardless of the measure applied. Figure 5. Simple association between natural reserves per capita and economic growth 1970-2010

(29)

27

that the number of subject countries is not high enough to call the results robust, thus I continue with my panel dataset.

Table 3. Basic estimations of the resource curse17

Cross-section Low FE High pooled OLS

(1) (2) (3) (5) (6) VARIABLES G7010 G7010 rgdpl7010 rgdpl7010 rgdpl7010 lrgdpl70 -0.00390 0.000162 (0.00234) (0.00271) rent70 -0.000957** (0.000400)

rescapavg -7.04e-07 -1.05e-06***

(5.11e-07) (2.84e-07) lrgdplini -0.0133** -0.0258*** -0.0189*** (0.00568) (0.00302) (0.00338) rent -0.000290 -0.00520*** (0.000445) (0.00113) Constant 0.0570*** 0.0254 0.129*** 0.287*** 0.224*** (0.0199) (0.0204) (0.0453) (0.0301) (0.0327) Observations 37 39 218 96 96 R-squared 0.169 0.068 0.030 0.479 0.443

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Changing the method of empirical testing checks for robustness. The results from the panel dataset show surprising results. The negative relationship presumed by the curse stays, however it is only significant in the high income country group for both measures. This result might be puzzling at first, since most of the negative symptoms are associated with poorer countries, but if we look at the countries investigated an alternative explanation emerges. Figure 6 in Appendix 4 depicts average growth in relation of average reserves per capita in the last 40 years for the high income country group. It shows that countries with the smallest reserves grew the most. But it is also true that these countries had significantly lower initial GDP per capita than any other country18, so the slower growth might rather be explained by the convergence hypothesis. Indeed, the coefficients and t-statistics of initial income are higher indicating a stronger effect on growth. However, the three top performers (Hong Kong, Singapore and Korea) cannot be considered outliers, as a) we are already controlling for initial income, b) according to the resource curse theory there is the possibility that there exceptionally growth rates could be achieved exactly because they are resource poor.

17

Low = low and middle income countries, FE = fixed effects, high=high income countries

18

(30)

28

Controlling for other drivers behind growth (in section 4.3) could help to place the issue in perspective. The ambiguity in the results points to the fact, that cross-sectional studies might really oversimplify the situation, as Manzano and Rigobon (2001) argued, and the resource curse hypothesis loses its strength, when analysing with panel data methods. However the negative relationship remains.

4.2 The resource curse and economic diversification in low and middle

income countries

In the case of low and middle income countries I apply fixed effects panel data methods as described in section 3.2. However, the first results produced relatively high F-statistics indicating some unobserved heterogeneity or heteroskedasticity. In order to deal with this issue I re-estimated the model using cluster-robust standard errors. I present the results from these new estimations.

Now, let us look at what happens to the resource curse, when I add the explanatory variables, including economic diversification. Table 4 summarizes the results.

(31)

29

Table 4. Resource curse in low and middle income countries, FE with cluster-robust standard errors

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLE S rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 lrgdplini -0.0176* -0.0145 -0.0186** -0.0168 -0.0261** -0.0211* -0.0322** -0.0288** 0.0223 0.0190 (0.00953) (0.00957) (0.00901) (0.0105) (0.0108) (0.0115) (0.0126) (0.0116) (0.0137) (0.0118) rent -0.000431 -0.000245 -0.000437 -0.000263 -0.000483 -0.000270 -0.000348 -0.000192 0.000316 0.000793 (0.000536) (0.000498) (0.000441) (0.000404) (0.000410) (0.000383) (0.000322) (0.000332) (0.000628) (0.000722) divval 0.0540 0.0489 0.0433 -0.0260 0.0352 (0.0319) (0.0338) (0.0348) (0.0491) (0.0901) divlab 0.00983 0.0136 -0.00133 -0.0131 0.0745 (0.0265) (0.0269) (0.0270) (0.0273) (0.0467) ki 0.143*** 0.151*** 0.137*** 0.143*** 0.196*** 0.192*** 0.0389 0.0121 (0.0321) (0.0376) (0.0313) (0.0362) (0.0273) (0.0228) (0.0749) (0.0596) openk 0.0343* 0.0402** 0.0261 0.0328* 0.0644** 0.0470 (0.0168) (0.0163) (0.0181) (0.0168) (0.0290) (0.0276) educ 0.0261* 0.0250* -0.0297 -0.0527* (0.0146) (0.0143) (0.0239) (0.0298) ruleoflaw -0.0151 -0.0164 (0.0112) (0.0112) Constant 0.0590 0.123 0.0432 0.0995 0.0967 0.137* 0.220 0.159* -0.217 -0.178** (0.0983) (0.0755) (0.0963) (0.0723) (0.0989) (0.0777) (0.137) (0.0846) (0.207) (0.0724) Observation s 218 217 218 217 218 217 204 203 80 80 R-squared 0.049 0.024 0.129 0.113 0.168 0.163 0.238 0.251 0.275 0.312 Number of id 28 28 28 28 28 28 26 26 26 26

(32)

30

The variables for economic diversification are also insignificant and even though the coefficients for divval are higher, than those of divlab, which points to the possibility that diversification in income has indeed stronger effects, there is no significant difference between the two. Interestingly, after the inclusion of the control variables of investment, openness and education both diversification measures turn negative. All three control variables are correlated positively with economic growth and significant, while rent stays negative. The fact that economic diversification turned negative after these additions implies that this variable might also act as a transmission channel for the curse. Economic diversification is a process, when economies build new sectors. These sectors usually are situated in the private sector. In order to build new activities in the private sector private agents, e.g. entrepreneurs and innovators have to be motivated enough and feel secured to invest. Basic requirements for economic diversification is captured by the other explanatory variables, for example new sectors require educated labour force. The observation that, after controlling for these basic requirements, the diversification turns negative implies that the positive effects of investment, openness and education were captured through the diversification variable and these effects balanced out the negative effect that economic concentration might had. It shows that resource curse might have some indirect effects working through economic concentration. Additionally, the negativity of the diversification also disappears with the addition of institutional quality, which infers that natural resources both directly and indirectly (indirectly through economic diversification) cause institutions to erode. The variables divval and divlab captures dependence on natural resources, thus this result weakly supports the assumption that resource dependence causes the effects of the curse, since the coefficient of the rents remained negative throughout the whole estimation implying other channels. However the lack of significance weakens the results.

(33)

31

Table 5. Mitigation effects of economic diversification in low and middle income countries, FE with cluster-robust standard errors

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLE S rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 lrgdplini -0.0190* -0.0134 -0.0198** -0.0151 -0.0274** -0.0194 -0.0328** -0.0291** 0.0257* 0.0190 (0.00980) (0.0102) (0.00934) (0.0109) (0.0108) (0.0119) (0.0130) (0.0117) (0.0144) (0.0120) rent -0.00590*** 0.00167 -0.00525** 0.00299 -0.00545*** 0.00282 -0.00488 -0.00106 -0.0275 0.000926 (0.00206) (0.00200) (0.00199) (0.00214) (0.00184) (0.00205) (0.00769) (0.00199) (0.0177) (0.00288) divval 0.0357 0.0329 0.0268 -0.0431 -0.0618 (0.0318) (0.0336) (0.0342) (0.0614) (0.116) c.rent#c.divv al 0.00294** 0.00259** 0.00267** 0.00226 0.0134 (0.00119) (0.00114) (0.00104) (0.00388) (0.00848) divlab 0.0161 0.0245 0.00935 -0.0154 0.0750 (0.0243) (0.0230) (0.0229) (0.0270) (0.0519) c.rent#c.divla b -0.00125 -0.00212 -0.00202 0.000531 -7.58e-05 (0.00152) (0.00159) (0.00153) (0.00121) (0.00149) ki 0.139*** 0.163*** 0.133*** 0.155*** 0.197*** 0.191*** 0.0150 0.0123 (0.0337) (0.0284) (0.0333) (0.0274) (0.0290) (0.0233) (0.0736) (0.0586) openk 0.0347** 0.0395** 0.0265 0.0331* 0.0726** 0.0470 (0.0165) (0.0168) (0.0180) (0.0168) (0.0298) (0.0282) educ 0.0263* 0.0247* -0.0452 -0.0525* (0.0147) (0.0144) (0.0271) (0.0297) ruleoflaw -0.0167 -0.0164 (0.0110) (0.0113) Constant 0.106 0.104 0.0851 0.0662 0.140 0.105 0.258 0.167* -0.00723 -0.179** (0.102) (0.0855) (0.101) (0.0792) (0.0994) (0.0833) (0.171) (0.0872) (0.251) (0.0749) Observations 218 217 218 217 218 217 204 203 80 80 R-squared 0.068 0.030 0.143 0.132 0.183 0.180 0.240 0.251 0.317 0.312 Number of id 28 28 28 28 28 28 26 26 26 26

(34)

32

The first column presents the results of including the interaction term between income diversification and resource rents. The findings are in line with hypothesis 2, that diversification can mitigate the effects of the curse. In the first column the coefficient of rents is significantly negative, while the interaction term is positive and significant. Based on this I came to the following conclusions.

Economic diversification is positively correlated with growth, but not significant enough. That infers that countries simply with higher economic diversification probably will not experience significantly higher growth, in other words economic diversification in itself is not a significant driver behind growth - as the previous regression showed. On the other hand, the significant and negative resource abundance coefficient implies that the resource curse appears in countries without sufficient economic diversification, lowering economic growth. The joint effect of the two variable is positive indicating that the dynamics of economic diversification differs in resource rich and resource poor countries. It gains importance in resource abundant nations, where it can mitigate the negative effects of the curse. This positive relationship stays even after controlling for investment and openness. However, after adding education and institutional quality the mitigation effect loses its significance, although remains positive. Thus I can conclude that economic diversification is indeed important for escaping the resource curse, however building human capital is probably more crucial. First human capital has to be increased in order to have a more diversified economy, and for this good institutional quality is necessary. However the results regarding the rule of law is not significant, thus the inferences about them are not as robust as about education.

(35)

33

4.3 The resource curse and economic diversification in high income

countries

Findings in section 4.1 indicated that there is no significant difference regardless whether I use rents or reserves in panel data sets. The R-statistics also indicated that rents might explain economic growth more reliably. Based on these and for better comparability with the results from the previous sector I keep rents as my measure for resource abundance. I use reserves to test for robustness.

In section 2.4 I hypothesized that high income countries behave differently. Indeed, Table 6 shows that both divval and divlab are negative and significant indicating the presence of the specializing tendency in high income countries. However, in this group labour diversification seems to be more relevant, also specialization is more visible there. Diversification in value added quickly loses its significance after the inclusion of control variables, while labour mostly stays significant and negative. Thus, the difference between the different diversification measures remains in the high income country group as well, unlike my hypothesis stated .

Initial income and resource rents stay negative and significant even after including all the control variables, but we should not jump to conclusions too quickly. As I have argued before the three top performers in terms of economic growth happens to have both the lowest initial incomes and the lowest amount of natural reserves. It might be true, that they could grew this fast because there were no natural resources to hinder them. But it is also possible that their growth performance has nothing to do with their lack of resources. When dropping these observations from the sample, the effect of resource rents loses significance, while initial incomes stays significant.19 The convergence assumption and the control variables seem to explain the tendencies of economic growth more accurately, than the resource curse. Looking at the findings from the second estimation might help to clear this question.

19

(36)

34 Table 6. Resource curse in high income countries, Pooled OLS

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) VARIABLES rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 rgdpl7010 lrgdplini -0.0232*** -0.0190*** -0.0207*** -0.0172*** -0.0251*** -0.0214*** -0.0288*** -0.0258*** -0.0492*** -0.0400*** (0.00304) (0.00354) (0.00307) (0.00346) (0.00369) (0.00440) (0.00450) (0.00644) (0.0105) (0.0109) rent -0.00399*** -0.00586*** -0.00258** -0.00377*** -0.00336*** -0.00370*** -0.00372*** -0.00374*** -0.00749*** -0.0116*** (0.00117) (0.00110) (0.00124) (0.00127) (0.00127) (0.00126) (0.00129) (0.00126) (0.00219) (0.00258) divval -0.0452*** -0.0263 0.00315 0.00626 -0.0383 (0.0158) (0.0167) (0.0217) (0.0217) (0.0378) divlab -0.0611*** -0.0451** -0.0258 -0.0129 -0.0943** (0.0187) (0.0188) (0.0225) (0.0265) (0.0426) ki 0.0707*** 0.0707*** 0.0518* 0.0526** 0.0477* 0.0487* -0.0519 -0.0509 (0.0255) (0.0239) (0.0267) (0.0265) (0.0267) (0.0268) (0.0490) (0.0460) openk 0.00564** 0.00380 0.00667** 0.00518* 0.00458 0.00301 (0.00273) (0.00247) (0.00282) (0.00289) (0.00443) (0.00336) educ 0.00642 0.00491 0.00453 -0.00134 (0.00456) (0.00534) (0.00975) (0.00941) ruleoflaw 0.0255*** 0.0217*** (0.00846) (0.00701) Constant 0.346*** 0.336*** 0.266*** 0.268*** 0.256*** 0.275*** 0.269*** 0.281*** 0.563*** 0.601*** (0.0355) (0.0324) (0.0449) (0.0387) (0.0444) (0.0387) (0.0452) (0.0391) (0.122) (0.107) Observations 96 96 96 96 96 96 96 96 37 37 R-squared 0.522 0.533 0.559 0.574 0.579 0.585 0.588 0.589 0.550 0.601

(37)

35

Table 1 in Appendix 6 summarizes the results of the second regression. I had two hypothesises regarding high income countries. One, because of their development level there should not be any significant difference between the effects and relationships of the two types of diversifications. Two, because of their expected specializing tendencies the mitigation effect should not hold. The results seem to support the first one, but go against the second hypothesis. Both the divval and divlab variables stay negative even after controlling for education, investment, openness and institutional quality, although divlab is more significant, as argued before. It indicates that, unlike in the low and middle income group, economic diversification in itself is very important, but with different reasoning. Specialization alone can drive growth and not just as a mitigation effect. Interestingly the interaction terms are all positively correlated with growth. It seems that the mitigation effect of economic diversification holds in high income countries as well, but there are two differences with low and middle income countries. Firstly, the interaction term with the value added diversification variable loses its significance quickly the same time when resource rents do. Secondly, labour diversification seems to have an even stronger mitigation effect here. Since this result raises some questions I checked the reliability of my results and also estimated the regressions for high income countries using average reserves per capita. 20 Three differences emerged, which influence the interpretations of the results. Firstly, the coefficient of reserves per capita loses its significance after introducing the control variables and even turn positive at a point. Secondly, labour diversification also becomes insignificant, while income diversification gains power. Lastly, the interaction term of reserves and labour diversification turns negative. This indicates that the choice of measure might have created the bias, that labour diversification is more significant. After this sensitivity test it can be concluded that both diversification measures are negatively correlated with economic growth as Hypothesis 3a predicted. Thus the high income group and the low and middle income group are moving to different directions in terms of diversification.

Lastly, the sensitivity test also showed that the mitigation effect works the same way in both income groups. Even though high income countries are generally moving towards economic concentration the effects of the resource curse can be mitigated by the same strategy: by diversifying the revenues.

20

(38)

36

5. Conclusion

In this paper I was interested to see if economic diversification plays a role in deciding whether natural resources turn to be a blessing or a curse. Since most of the negative symptoms of the curse is said to be originating from too high dependence on natural resources and concentrated economies, I raised the question whether economic diversification can mitigate the effects of the resource curse. I divided my research question into four parts. Firstly, I asked whether the original resource curse holds its water in my dataset. Secondly, I hypothesized my main question, that diversification can turn resource curse into resource blessing. Thirdly, I formed the hypothesis that economic diversification in income and in labour have different effects on the curse. Lastly, I assumed that high income countries are started to specialize again, therefore I presupposed that the mitigation effects do not emerge in the case of high income countries.

Applying both cross-sectional and panel data methods I found that the resource curse hypothesis is sensitive to changes in measurement and empirical methods. On the other hand, the mitigation effect did appear in the case of all income groups, indicating that economic diversification might indeed be a viable escape route for countries suffering from the resource curse. However the mitigation effect loses its power when controlling for education and institutional quality, therefore these two can be considered more important requirements for the curse, although diversification should not be neglected. It also turned out that only income diversification can mitigate the curse in the case of both country groups, even though high income countries indeed seem to be specializing.

Therefore I can conclude that economic diversification is a possible way out of the resource curse, however there are some limitations to my study. There is no sufficient data for natural reserves for longer time periods, so alternative measures have to be used, which can cause biases in the results. Also, I calculated from my diversification measures from 10 sectors for only 40 country. Extending these parameters could result in some meaningful changes in the results. Lastly, the level of income and diversity when countries start to specialize again is heterogeneous, thus I separated the sample into two groups, but this method might be a bit rough and might have caused changes in my results.

(39)

37

6. References

Hawaii Economic Issues. (2011). Measuring Economic Diversification in Hawaii. State of Hawaii - Department of Business, Economic Development & Tourism.

Alexeev, M., & Conrad, R. (2009). The Elusive Curse of Oil. The Review of Economics and Statistics , 91 (3), 586-598.

Auty, R. M. (1993). Sustaining Development in Mineral Economies - The Resource Curse Thesis. London: Routledge Publ.

Auty, R. M., & Warhurst, A. (1993). Sustainable Development in Mineral Exporting Economies. Resources Policy , 19 (1), 14-29.

Bardhan, P. (1997). Corruption and Development: A Review of Issues. Journal of Economic Literature , 35, 1320-1346.

Barro, R., & Lee, J. (2010). A New Data Set of Educational Attainment in the World. 1950-2010. Journal of Development Economics , 104, 184-198.

Basedau, M., & Lay, J. (2009). Resource Curse or Rentier Peace? The Ambiguous Effects of Oil Wealth and Oil Dependence on Violent Conflict. Journal of Peace Research , 46 (6), 757-776. Behbudi, D., Mamipour, S., & Karami, A. (2010). Natural Resource Abundance, Human Capital and Economic Growth in the Petroleum Exporting Countries. Journal of economic Development , 35 (3), 81-102.

Berezin, P., Salehizadeh, A., & Santana, E. (2002). The Challange of Diversification in the Caribbean. IMF Working Paper 02/196, .

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, 248–264. Bulte, E. H., Damania, R., & Deacon, R. T. (2005). Resource Intensity, Institutions, and Development. World Development , Vol. 33, No. 7, 1029-1044.

Corden, W. M. (1984). Booming sector and Dutch disease economics: Survey and consolidation. Oxford Economic Papers , 36, 359-380.

Davis, G. A. (2013). Replicating Sachs and Warner’s Working Papers on the Resource Curse. The Journal of Development Studies , 49 (12), 1615-1630.

European Commission. (2015, April 10). European Union, Trade in goods with Norway. Directorate-general for trade.

(40)

38

Gelb, A. (2010, November 4-5). Economic Diversification in Resource Rich Countries. Lecture at Natural resources, finance, and development: Confronting Old and New Challenges Seminar . Algeria: Central Bank of Algeria and the IMF Institute in Algiers.

Gelb, A. (1988). Oil Windfalls: Blessing or Curse? In (A World Bank research publication). New York: Oxford University Press .

Gylfason, T. (2001). Natural resources, education, and economic development. European Economic Review , Vol. 45. No. 4-6, 847-859.

Hausmann, R., & Rigobon, R. (2002). An Alternative interpretation of the 'Resource Curse': Theory and Policy Implications. NBER Working Paper Series , Working Paper 9424.

Heston, A., Summers, R., & Aten, B. (2012, nov). Penn World Table Version 7.1. Income and Prices at the University of Pennsylvania: Center for International Comparisons of Production.

Hill, R. C., Griffiths, W. E., & Lim, G. C. (2011). Principles of Econometrics (4th Edition ed.). John Wiley & Sons, Inc.

ILO. (2014). Yearbook of labor statistics. Geneva: International Labour Office Press.

Imbs, J., & Wacziarg, R. (2003). Stages of Diversification. The American Economic Review , 63-94. James, A. G., & James, R. G. (2011). Do resource dependent regions grow slower than they should? Economics Letters , 111, 194-196.

Klinger, B., & Lederman, D. (2006). Diversification, Innovation, and Imitation Inside the Global Technological Frontier. World Bank Policy Research Workin Paper , No. 3872.

Krugman, P. (1991). Geography and trade. Cambridge, MA: MIT Press.

Leite, C., & Weidmann, J. (2002). Does mother nature corrupt? Natural resources, corruption and economic growth. In G. A. (Eds.), Governance, corruption, and economic performance (pp. 156–169). Washington DC: International Monetary Fund.

Lewis, W. A. (1954). Economic Development with Unlimited Supply of Labour. Manchester Schoo , 22 (2), 139-91.

Lucas, R. E. (1988). On the Mechanincs of Economic Development. Journal of Monetary Economics , 22, 3-42.

Mankiw, G. N., Romer, D., & Wile, D. N. (1992). A Contribution to the Empirics of Economic Growth. The Quarterly Journal of Economics , 107 (2), 407-437.

Manzano, O., & Rigobon, R. (2001). Resource Curse or Debt Overhang? NBER Working Paper Series , Working Paper No. 8390.

Matsuyama, K. (1992). Agricultural productivity, Comparative advantage, and economic growth. Journal of Economic Theory , 58, 317-334.

The Curse of Natural Resources