The Natural Resource Cure

DEPARTMENT OF ECONOMICS Uppsala University

C-level Thesis

Author: Anna Wiström1

Supervisor: Niklas Bengtsson Spring 2013

The Natural Resource Cure

Abstract

This study explores the natural resource curse and its possible cure via good institutional quality. In theory countries that are resource abundant are said to have slower economic growth than countries that are resource scarce. Earlier studies have shown that resource abundant countries only suffer from the resource curse if the resources are highly appropriable and if the institutional quality is low. If resource abundant countries instead have resources that are highly appropriable and if the institutional quality is high they will benefit from their resources. If a country has resource with low technical appropriability no negative effect on growth is expected. In this study several time periods are studied and it can be concluded that for earlier time periods the resource curse theory in general holds but for later time periods no negative effects of resource abundance on economic growth can be detected.

Keywords: Natural resource curse, economic growth, development, appropriability, institutional quality

JEL classification: N50, O13, O40, O57, P17

Table of Content

1INTRODUCTION ... 3  

2PREVIOUS RESEARCH ... 4  

3THEORY AND DATA ... 6  

3.1THEORY ... 7  

3.2DATA ... 9  

4RESULTS ... 12  

4.1MAIN RESULTS ... 12  

4.2INSTRUMENTING FOR INSTITUTIONAL QUALITY ... 16  

4.3WHY IS THERE NO RESOURCE CURSE BETWEEN 1990 AND 2010? ... 17  

5ROBUSTNESS ... 20  

5.1EXCLUDING OUTLIERS ... 20  

5.2EXCLUDING OR INCLUDING VARIABLES ... 21  

5.3EXCLUDING REGIONS ... 21  

5.4EXCLUDING RICH OR POOR COUNTRIES ... 22  

6DISCUSSION ... 22  

7CONCLUSION ... 24  

REFERENCES ... 26  

APPENDIX ... 28  

1 Introduction

Why have Chile and Liberia had such different rates of economic growth for the last 30 years? And Singapore and Iran? Chile had a growth rate of 3.2 % per year and Liberia -3.2 % per year between 1980 and 2010.2 _{Singapore grew at a rate of 4.6 %}

per year and Iran -1.2 % per year between 1970 and 2000. One thing these four countries have in common is that they are abundant in fuel, ores or metals.3 _{About 64}

% of Chiles and Liberia’s merchandise exports were ores and metals in 1980 and about 25 % of Singapore’s and 90 % of Iran’s merchandise exports were fuels in 1970. What makes these countries experience such different rates of economic growth? One thing that is different between the countries is the institutional quality. The quality of the institutions, governments and legal systems are in Chile and Singapore higher than in Liberia and Iran, making it possible for the two first countries mentioned to benefit from their resources while the two last cannot. To measure institutional quality in this study the International Country Risk Guide (ICRG) index is used. The ICRG index is combined of three components scaled to an o-1 variable where values close to 1 indicate good institutions and values close to 0 indicate bad institutions. The components included in the measure are quality of the bureaucracy, corruption in government and rule of law. The values for the countries mentioned above are 0.57 for Chile, 0.17 for Liberia, 0.89 for Singapore and 0.35 for Iran.

In the previous examples the resources in question had a high level of technical appropriability. What happens with a country’s growth if the resources have a low level of technical appropriability? In both Panama and Paraguay food export was about 70 % of merchandise exports and the ICRG values were 0.22 and 0.14 respectively in 1970. Still both countries grew at a rate of 2.3 % and 2.6 % per year between 1970 and 1990. Less appropriable resources do not always imply a resource curse even if the institutional quality in a country is low.

2 _{For sources and descriptions of all data and variables used in the study see Table 14 in Appendix.} 3 _{The level of resource abundance will in this study be approximated by exports as share of}

The examples above demonstrate three aspects of the theory to be explored in this study. Firstly, the phenomenon that resource abundant countries have slower economic growth than resource scarce countries, known as the natural resource curse. Secondly, the resource curse can only be detected if the resources have high technical appropriability, for example fuel, ores and metals, and not if the resources have low technical appropriability, like food and agricultural products. Thirdly, if a country is abundant in highly appropriable resources it will only be cursed if the institutions in the country have low quality. The theory of the resource curse with these aspects has been explored in earlier research from for example Boschini et al (2007 and 2013) and Isham et al (2005).

In this study different resources with different levels of appropriability will be explored to see if a natural resource curse can be found and if it can be reversed, as it has been in Chile and Singapore, by good institutional quality. Thus the main questions to be answered can be stated as: do resource abundant countries,

depending on the level of appropriability of the resources, suffer from slower economic growth than resource scarce countries? If they do, can the curse be reversed via good institutions quality? This study will with some alterations follow

that of Boschini et al (2007 and 2013). The two studies will be explained in more detail and the differences between this study and those of Boschini et al (2007 and 2013) will be stated in the following parts of the text.

The remaining part of this paper is divided into the following sections: section two covers earlier research, section three explains the theory and data to be used, section four will present the main results, section five checks the robustness of the results, section six discuses the results and future questions to be answered and section seven concludes.

2 Previous research

curse. They explained the negative relationship with what is known as the Dutch

Disease. The basic idea behind the Dutch Disease is that countries that gain a lot of

rents from natural resources will have a crowding-out effect in other economic sectors. Sachs and Warner (1995) argued that positive externalities, for example learning-by-doing, are only present in the manufacturing sector and thus countries with a large natural resource sector and small manufacturing sector will not benefit from these positive externalities. This theory does not explain why Chile and Singapore can have such high growth rates even if they are natural resource exporters.

More resent research has been trying to deal with the problem that the outcome of natural resource abundance is ambiguous. Isham et al (2005) for example found, in contrast with Sachs and Warner (1995, 1999 and 2001), that institutional quality do play a role in determining if a country will be cursed or not. They also found that it is only what they call point source resources, like fuel, ores, metals and plantation crops, that lead to the resource curse and not diffuse resources, like food and agricultural products.

Mehlum et al (2006) found in line with Isham et al (2005) that a country will be cursed if the institutions are grabber friendly but not if the institutions are producer

friendly. Given that a country is abundant in point source resources it will only be

cursed if the institutions are grabber friendly, undermining the production sector in the economy that lead to long term growth. When instrumenting for institutional quality they found that natural resources only effect economic growth negatively via institutional quality and not directly.

Sala-i-Martin and Subramanian (2003) used instrument variables as well when estimating the effect resource abundance have on growth. As Mehlum et al (2006) they found that resource abundance only effect growth via institutional quality.

explaining why they use net exports instead of exports as the resource abundance approximation they criticize Sachs and Warner (1995) for not being consistent in their measuring. Two countries in the sample used by Sachs and Warner (1995) are measured as net exports instead of exports. The reason Sachs and Warner give is that the two countries, Singapore and Trinidad and Tobago, import a lot of resources and directly export them again and thus these countries will appear to be more resource abundant than they really are. In a study form Lederman and Maloney (2008), the study that Pineda and Rodríguez (2010) is largely based on, the results from Sachs and Warner (1995) were redone and is was show that the negative relationship goes away when using natural resource exports, and not net exports, for all countries. Pineda and Rodríguez (2010) and Lederman and Maloney (2008) therefore argue that net export should be used as the natural resource proxy for all countries to ensure that it is only the resources from within the country that are accounted for. In the studies this paper is largely based on, Boschini et al (2007 and 2013), they found that highly appropriable resources will lead to a resource curse and that good institutions can reverse the curse. In one section in each study they instrumented for institutional quality and found in contrast with Mehlum et al (2006) and Sala-i-Martin and Subramanian (2003) that resource abundance do have a direct negative effect on growth.

3 Theory and data

3.1 Theory

3.1.1 The basic resource curse theory

The basic theory of the resource curse is that countries that are resource abundant have slower economic growth than countries that are resources scarce (Sachs and Warner 1995, 1999 and 2001, Boschini et al 2007 and 2013 and Mehlum et al 2006). This basic theory does not explain the ambiguous outcome of resource abundance.

3.1.2 Appropriability

It is not only to what degree a country is resource abundant that matters when it is determined whether a country will be cursed or not, what type of resources a country has plays an important role as well. If the resources in a country are point source resources, like fuels and minerals, they will more likely cause slower economic growth in a country than if the resource are diffuse, like food and agricultural products (Isham et al 2005). Isham et al (2005, p. 3) describes point source resources as resources that are “…extracted from a narrow geographic or economic base…”. These point source resources will because of their narrow base be easier to control centrally and will thus with high probability lead to rent-seeking, corruption and conflicts that will harm the economic development while diffuse resources most likely will not lead to such behavior (Boschini et al 2007). To use the terminology from Boschini et al (2007 and 2013) point source resources are said to have high technical appropriability and diffuse resources low technical appropriability.

3.1.3 Institutional quality

institutions are bad and a positive relationship when institutions are good. As can be expected from the theory outlined in Boschini et al (2007 and 2013) and Mehlum et al (2006) countries abundant resources will be cursed only if their institutional quality is low.

Figure 1: Graphs showing growth and ores and metals exports

3.1.4 Appropriability and institutional quality combined

Figure 2: Two dimensions: appropriability and institutional quality

3.2 Data

3.2.1 The model

The basic regression model for testing the effects of resource abundance on economic growth used in this study follows that of Boschini et al (2007):

!"_!×!"#$_! is the interaction between natural resources and institutional quality. The index ! represent the different countries included in the study.4

There are some differences between this study and those of Boschini et al (2007 and 2013). One difference is that the regional dummy variables used in Boschini et al (2007) are Sub Saharan Africa and Latin America and in Boschini et al (2013) are Sub Saharan Africa, Latin America, the Middle East and North Africa instead of Africa, the Middle East, Latin America and Asia that are used in this study.5 _{Two other}

differences between the studies are that more resent data will be used and that different time periods will be considered. In the study of Boschini et al (2007) only one time period is studied, 1975-1998, and in the study from 2013 different start years for the time periods studied are used, 1965, 1975 and 1985, but only one end year, 2005. The time periods to be studied in this paper are 1970-2010, 1970-2000, 1970-1990, 1980-2010, 1980-2000 and 1990-2010.6

The proxies for natural resource abundance and the variables in the vector are measured in the beginning of each time period that is studied as to avoid reverse causality, where growth is affecting the explanatory variables.

As in Boschini et al (2007 and 2013) the estimations are made using the Ordinary Least Square (OLS) method. When the measures for natural resources used in the regressions have high technical appropriability !_!will be expected  to be negative,

indicating the resource curse, !_!  to be positive, indicating that good institutions are

beneficial for growth, and finally !!  to be positive. If the curse is to be reversed !_! has

to be larger in absolute value than !_!. When the resources have low technical appropriability we will not expect any significant results for !_! or !_!. We still might se significant and positive results for !!, indicating that good institutions are good for

growth independently of the amount or type of resources a country has.

4 _{See Table 13 in Appendix for a list of all countries included in the study.}

5 _{Boschini et al (2013) states that the results are not sensitive for extra variables, for example regions,}

being included in the model.

6 _{All variables are calculated as five year averages to ensure that no particular year will have a crucial}

3.2.2 The variables

In this study a variation of five measures used in Boschini et al (2013) are used to approximate resource abundance. The measures are taken from The World Bank (2013) and are agricultural, food, fuel and ores and metals exports as share of merchandise exports. Total exports are calculated as the sum of agricultural, food, fuel and ores and metals exports. The data used in Boschini et al (2013) are exports as share of GDP instead of as share of merchandise exports. Looking again at Figure 2 agricultural and food exports are located at the bottom half of the figure, indicating low technical appropriability, while fuel and ores and metals exports are located at the top half of the figure, indicating high technical appropriability. Total exports will capture the whole range of appropriability levels. Table 14 in Appendix describes what products are included in the different exports categories. In the results we will expect to see the resource curse most strongly when using ores and metals exports, some when using total or fuel exports. Agricultural products and food are less technically appropriable so we will not expect the curse to appear when using these two approximations.

Institutional quality in this study is measured by the ICRG index. The three components of the ICRG that are used in this study are quality of the bureaucracy,

corruption in government and rule of law.7

• Quality of the bureaucracy: This component is given a high value if the bureaucracy is somewhat independent of political pressure and change of political rule.

• Corruption in government: A high value is given if corruption is not present. Absence of corruption is essential if people or companies are to be willing to invest within the country.

• Rule of law: This variable measures how well the legal system works and how high the crime rate is in a country. If the legal system is efficient and the crime rate is low the country will get a high value.

7 _{There are more components in the original ICRG index but they are not used in thus study due to}

The ICRG index is calculated as an average of these three components and is scaled to a 0-1 variable (Teorell et al 2011a). High values indicate good institutional quality and low vales indicate bad institutional quality.

There are problems when measuring institutional quality. First, institutional quality is a question of subjectivity. Second, there has been suggested that institutional quality can affect the amount of natural resources found and extracted in a country or that the amount of natural resources can effect the institutional quality. Research has shown that the most likely way institutions and natural resources are correlated is that abundance in natural resources generate institutions with bad quality and that in turn will affect economic growth negatively (Sala-i-Martin and Subramanian 2003 and Mehlum et al 2006). A country with bad institutions might not provide the essential safety for natural resources to be found and extracted while a country with good institutions will provide the framework for people and companies to extract and use the resources efficiently. In this way good institutions will probably lead do more resources being found and extracted while bad institutions will lead to less resources being found and extracted (Sala-i-Martin and Subramanian 2003).

4 Results

In the first part of this section the main results are presented. In the second part results are presented when instrumenting for institutional quality. In the main results no resources curse can be detected in 1980-2000 or 1990-2010 when using ores and metals exports, in the third part possible reasons for this absence are tested.

4.1 Main results

The main results are presented in Table 1. Staring with total exports, if the theory of the resource curse is true we will expect to se some evidence of the curse when using this proxy. Looking at the results in Table 1 we can see that this is the case for only one time period, 1970 to 1990, where there is some indication of a negative relationship between resources and growth.

curse but rather the opposite, abundance in agricultural exports seem to be beneficial for growth when looking at the first three time periods. The interaction term is also significant for the first three time periods and is in absolute value larger than agricultural exports. The results indicate that agricultural export is beneficial for growth but if the institutions are good enough the positive effect of the exports will be reversed and agricultural exports will be negative for growth. We can see that institutional quality is beneficial for growth in all time periods except 1990-2010. As with agricultural exports we do not expect any evidence for a resource curse for food exports. In the results we can se that food exports do not have any influence over growth for any of the time periods. Neither does the interaction term. The only significant variable is institutional quality for all time periods but the last, 1990-2010. When using fuel exports as the resource abundance proxy we will expect some evidence of a resource curse. As with the results for food exports no evidence of a curse can be found in the regressions. Growth appears not to be explained by either fuel exports or the interaction between fuel exports and institutional quality. Good institutions seem to be beneficial for economic growth in all time periods but the last. For countries abundant in ores and metals exports we would expect evidence of a resource curse. Looking at the results we can see that large amounts of exports of ores and metals indeed have a negative effect on growth for the first four time periods. The interaction term is significant for the same time periods and is larger in absolute value. We can thus conclude that ores and metals exports have a negative effect on growth but with good enough institutions the curse will be reversed and ores and metals in combination with good institutions will be beneficial for economic growth. Institutional quality explains some of the growth in the time periods 1970-2000, 1970-1990, 1980-2000 and 1990-2010. In these periods better institutions will lead to faster economic growth.

Looking at the explanatory power, !!_{, in the regressions for each time period we can}

see that for the first five periods the !! _{is around 50 % to 60 % while for the last time}

there is something else explaining economic growth in the period from 1990 to 2010 other than the variables included in the model used in this study.

To summarize the results; evidence of the resource curse can only be detected in the regressions using total exports in 1970-1990 and ores and metals exports in the first four time periods. Looking at food and fuel exports no relationship between exports and growth can be detected in any time period, either positive or negative. When using agricultural exports there appears to exist a resource blessing rather than a resource curse.

Since the curse is almost exclusively present in the regressions using ores and metals exports as measure for natural resources the remaining part of the paper will present results for ores and metals exports only and, without loosing generality and saving space, the time periods 1970 to 1990 and 1990 to 2010.8

Table 1: Main results 1970-2010 1970-2000 1970-1990 1980-2010 1980-2000 1990-2010 Total exports -2.148 (1.714) -3.540 (2.125) -4.206* (2.476) -1.230 (1.309) -2.130 (1.771) -0.186 (1.413) Institutional quality -0.117 (1.510) -0.694 (1.784) -0.455 (1.965) 1.357 (1.188) 1.537 (1.566) 1.979 (1.376) Interaction term 2.215 (1.963) 3.641 (2.356) 3.992 (2.759) 1.305 (1.686) 1.889 (2.140) -0.213 (1.805) N !! 69 0.5415 69 0.5949 69 0.5899 68 0.4894 68 0.5166 67 0.2697 Agricultural exports 4.093* (2.169) 5.274** (2.384) 7.541** (3.013) -0.574 (2.516) -0.216 (2.420) -4.100 (2.648) Institutional quality 2.438*** (0.637) 3.342*** (0.728) 4.621*** (0.906) 2.139** (0.812) 3.103*** (0.973) 1.753 (1.429) Interaction term -5.250** (2.309) -6.988*** (2.494) -11.152*** (3.282) 0.916 (3.936) 0.080 (4.530) 3.444 (5.119) N !! 69 0.5595 69 0.6019 69 0.6068 67 0.4706 67 0.4938 67 0.2854 Food exports 0.360 (1.372) 1.074 (1.496) 2.956 (1.776) 0.344 (1.610) -0.185 (1.975) -0.426 (1.935) Institutional quality 2.235** (0.956) 3.186*** (1.036) 4.523*** (1.244) 2.586*** (0.890) 3.392*** (1.093) 1.924 (1.280) Interaction term -1.152 (1.687) -1.928 (1.842) -3.905 (2.446) -1.151 (1.943) -0.866 (2.290) -0.256 (2.015) N !! 69 0.5304 69 0.5713 69 0.5831 67 0.4748 67 0.5005 67 0.2724 Fuel exports 0.604 (1.196) -0.498 (1.496) -1.148 (2.124) 1.121 (1.266) 1.878 (1.679) 0.532 (2.180) Institutional quality 1.944*** (0.693) 2.473*** (0.752) 3.217*** (0.872) 2.562*** (0.938) 3.519*** (1.229) 1.490 (1.183) Interaction term -0.916 (2.692) 0.202 (3.058) 0.683 (3.989) -1.925 (2.424) -3.696 (3.245) -0.546 (3.458) N !! 66 0.5236 66 0.5706 66 0.5921 65 0.4808 65 0.5060 65 0.2541

4.2 Instrumenting for institutional quality

As suggested above the amount of natural resources in a country can depend on the quality of institutions. Because of this correlation there might be a problem with biased and inconsistent estimates. To deal with this problem we will use the two-stage least square (2SLS) method, instrumenting for institutional quality and the interaction term between the natural resource proxy and institutional quality with variables that are correlated with institutions but not with the error term. The variables used in Boschini et al (2007), and to be used in this study, are latitude, the interaction between the natural resource proxy and latitude and the fraction of people in the country that speaks a western European language.

The results from the first stage regressions are presented in Table 5 in Appendix. We can see that the variables used as instruments are explaining some of the variation in the institutional quality measure and the interaction term. All three instruments are significant in at least one regression.

The results from the second stage regressions are presented in Table 2 below. The main results from Table 1 for the corresponding time periods are also presented for comparison. We see that the results in the first time period do resemble that of the main results to some extent. The sings in front of the estimates are the same when using the two methods but the effects of the variables are larger when using OLS. Since the 2SLS method always gives larger standard errors than the OLS method the variables are not significant when instrumenting for institutional quality and the interaction term.

The results for the second time period do not resemble that of the main results. When using the 2SLS method the estimates for ores and metals exports and the interaction term have the opposite signs from the estimates when using the OLS method, they are larger in absolute value and are not significant. Institutional quality is still positive, larger in absolute value and not significant.

and Subramanian (2003), resource abundance have negative effects on institutional quality and through that on economic growth but abundance in natural resources do not appear to have any direct effect on economic growth. The results are not consistent with that of Boschini et al (2007 and 2013) where they find that the estimates when using the two estimation methods give roughly the same results.

Table 2: Second stage regressions

1970-1990 (2SLS) 1970-1990 (OLS) 1990-2010 (2SLS) 1990-2010 (OLS)

Ores and metals exports -4.636

(3.183) -5.070*** (1.902) -15.703 (15.919) 1.614 (2.064) Institutional quality 7.333 (5.391) 1.759* (0.953) 7.392 (4.876) 2.317* (1.181) Interaction term 9.159 (7.840) 8.819** (3.601) 27.519 (30.250) -4.262 (4.145) N 69 69 66 66

Notes: Dependent variable is Growth. Standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include a constant and control variables listed in the text (not shown). The !! _{values are not shown since they do not have any clear}

statistical meaning when using the 2SLS method.

4.3 Why is there no resource curse between 1990 and 2010?

Figure 3: Graphs showing ores and metals exports and growth

Firstly, looking at the variables growth and ores and metals exports we can see that the minimum growth was much lower between 1970 and 1990 than between 1990 and 2010 and the maximum value of exports was much higher in 1970 than 1990. What implications this can have will be discussed in section 6.

Secondly, focusing on the variable institutional quality we can see that the minimum value of the ICRG index is larger in 1990 than 19709_{. The fact that the institutions in}

the world have become better can be one reason for the disappearance of the resource curse. But since the institutional measure was instrumented for in the previous section we can conclude that better institutions are not the reason for the lack of the resource curse in 1990-2010.

Thirdly, moving on to the variable initial GDP level we can see that the world in general has become richer. Both the mean and the minimum value have become higher since 1970. The results from running a regression for the time period between 1990 and 2010 but with the initial GDP level from 1970 instead of 1990 can be found in column 1 in Table 4. We can see the there is still no resource curse present in the time period and we can conclude that the fact that the world has become richer has not played a crucial role in the disappearance of the resource curse.

Fourthly, examine the variable average investment we see that the minimum value is larger for the later time period. Highly appropriable resources like ores and metals are dependent on investments and if the investments have become higher in the last 20 years this could be the reason for the absence of the curse. Running a regression with the average investment from 1970 to 1990 instead of 1990-2010 tested this hypothesis. The results are presented in column 2 in Table 4. Looking at the results we see that average investment has not played a role in the disappearance of the resource curse.

Finally there are some countries that only appear in the regression from 1990 to 2010.10 _{To see if there countries are the reason for the absence of the curse we run a}

regression for the time period 1990-2010 and excluding these countries. The results are presented in column 3 in Table 4 and we can see that there is no resource curse present even if the countries listed are excluded.

Table 3: Descriptive statistics

Variable Time period Mean Standard deviation Minimum value Maximum value Observations Growth 1970-1990 1990-2010 1.025 1.909 1.905 1.198 -3.751 -0.230 6.417 6.494 69 66

Ores and metals exports 1970 1990 0.128 0.081 0.209 0.141 0.000 0.000 0.983 0.582 69 66 Institutional quality 1970 1990 0.568 0.606 0.288 0.262 0.087 0.111 1.000 1.000 69 66 Initial GDP level 1970 1990 9.104 13.575 9.547 11.211 0.600 1.335 56.581 48.285 69 66 Average openness 1970-1990 1990-2010 3.957 4.233 0.595 0.530 2.466 3.056 5.773 5.899 69 66 Average investments 1970-1990 1990-2010 23.341 24.332 8.669 6.816 2.358 11.042 49.102 50.811 69 66

10 _{The countries are China, Kuwait, Malta, Oman, Papua New Guinea, Poland, Saudi Arabia and}

Table 4: Results testing why there is no curse between 1990 and 2010

(1) (2) (3)

Ores and metals exports 0.016

(0.020) 0.021 (0.020) 1.402 (1.972) Institutional quality 2.134** (0.952) 2.678** (1.058) 2.943** (1.111) Interaction term -0.029 (0.043) -0.048 (0.042) -0.448 (4.088) N !! 64 0.3104 64 0.3370 58 0.3102

Notes: Notes: Dependent variable is Growth. Robust standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include a constant and control variables listed in the text (not shown). All regressions are for the time period 1990-2010. Kuwait and Saudi Arabia are excluded from the regressions in column 1 and 2 due to missing data.

5 Robustness

The analysis of the main result shows that as a whole the theory about the resource curse holds for ores and metals exports for earlier years but not in the later time periods. When using the OLS method to estimate the equation some assumptions have to be fulfilled. For example the data should not include observations with extreme values or have omitted variables that explain the dependent variable. This section tests if the data used are sensitive to these assumptions being fulfilled. If the results do not change when for example excluding outliers or including new variables we can conclude that the results are robust. If the results are robust they are more reliable for further analysis and can be generalized to other time periods or to be generalized for all the countries in the world and not only the ones included in the regressions. This section largely follows that of Boschini et al (2007).

5.1 Excluding outliers

the DFITS index is larger than 2 !/! where ! is the number of variables, including the intercept, and ! is the number of observations in the regression. Table 6 in Appendix presents the results when running the regressions without outliers.11 _In

general the results follow that of the main results. Ores and metals exports and the interaction term are as in the main results significant for 1970-1990 but not for 1990-2010. For the last time period institutional quality is, as in the main results, significant. The results seem to be robust when excluding outliers.

5.2 Excluding or including variables

The estimated coefficients in front of variables in a regression can alter if different variables are excluded or included from the model. Table 7 and 8 in Appendix presents results when initial GDP, average openness and average investment are excluded from the model in different combinations. Ores and metals exports and the interaction term are significant for 1970-1990 but not 1990-2010 and institutional quality is significant for some of the regressions in both time periods. It seems that the results are not very sensitive to excluding variables.

Collier and Hoeffler (2002) and Boschini et al (2007) suggest that if a country has been in a conflict or civil war it can have had an effect on economic growth. The results when including dummy variables for war or civil war are presented in Table 9 in Appendix. We can see that none of the variables are significantly effecting growth in either time period. The estimates for ores and metals exports, the interaction term and institutional quality are close to those in the main results and we can conclude that the results are not sensitive to adding the variables war or civil war.

5.3 Excluding regions

In the main results countries from the whole world is included but to see if the resource curse is general for the whole world or just a phenomenon located at some

11 _{The countries counted as outliers in 1970-1990 are Bolivia, Brunei, Egypt, Iran, South Korea,}

specific region regressions have been run excluding Africa, the Middle East, Latin America and Asia respectively. The results are presented in Table 10 and 11 in Appendix. Ores and metals exports are not significant in either time period when Africa is excluded and but significant in both time periods when Latin America is excluded. Excluding the Middle East or Asia does not seem to have any effect on the main results. These results indicate that much of the curse is located in Africa.

5.4 Excluding rich or poor countries

As with excluding different regions it can be interesting to see if the resource curse is mostly a problem for rich or poor countries. By excluding the richest or poorest 25 % of the countries we can see if the rich countries or the poor countries are the ones mostly cursed. Results can be found in Table 12 in Appendix. When excluding the poorest 25 % of the sample ores and metals exports and the interaction term are not significant in any time period but when excluding the richest 25 % of the sample ores and metals exports and the interaction term are significant in all periods. The interaction term is significant for both time periods when the rich countries are excluded but not significant in either time period when the poor countries are excluded. We can from this conclude that much of the resource curse originates from the poor countries.

6 Discussion

The results from this study are to large extent in line with earlier studies. As in the studies of Boschini et al (2007 and 2013) highly appropriable resources do cause a resource curse while other resources do not. In contrast with these studies the coefficients in this paper are smaller. This could be because the variables used are not exactly the same.

and 2013) did not include the resource abundance proxy in the first stage regression when instrumenting for institutional quality while the proxy is included in the first stage regression in the study of Sala-i-Martin and Subramanian (2003) and this study.

In this study no resource curse could be detected in the regression using fuel exports as a proxy for natural resources even if it is counted as a resource with quite high technical appropriability (Boschini et al 2007 and Figure 2). In Boschini et al (2013) fuel exports are significant for some time periods but not others. The alteration in variables used as natural resource proxies for fuel can be the reason for this difference in results.

While this study has answered some questions it has been the source of new questions as well. A finding that is consistent throughout the study is that no resource curse can be detected for the time period between 1990 and 2010. The tests complete in section 4.3 rejects this lack of significance to be the consequence of better institutions, higher initial GDP levels, larger average investments or the different countries included in the regressions. It can be seen in Table 3 that both resource abundance and growth in the world are less spread in 1990-2010 than 1970-1990 and this lack of variation in the variables could be the reason for the absence of the curse. The lowest growth rates in 1990-2010 are higher than that in 1970-1990 and the most resource abundant country was not as resource abundant in 1990 as in 1970. This means that there are no real resource losers left in the world. In the countries included in this study the countries with the lowest growth in combination with high exports of ores and metals between 1970 and 1990 were Zambia and Liberia with growth rates of -3.1 % and -3.8 % and ores and metals exports of 98 % and 75 % respectively. In 1990-2010 the countries were Bolivia and Papua New Guinea with both having growth rates of 1.4 % and ores and metals exports of 46 % and 58 % respectively. What is the reason for the disappearance of the resources losers? Why are the growth rates higher now than 20 years ago? What happened to the countries that almost exclusively exported natural resources? These are some of the questions still to be answered.

Rodríguez (2010) and Lederman and Maloney (2008). The values used in this study could be misleading and not really show the resource abundance within a country but rather the amount of resources handled at the ports of a country.

One other thing that could be done differently is that the same countries should be used in all the regressions. This to make sure that the difference in results between time periods or types of resources is not dependent on what countries are included in the regressions. To be able to keep as many observations as possible this was not done but as data for more countries gets available with time this might not be a problem in the future.

7 Conclusion

The results from this study can be summarized in the following statements:

• Indications of a resource curse can be detected in 2010, 2000, 1970-1990 and 1980-2010 when using ores and metals exports as an approximation of natural resource abundance.

• No relationship between natural resources and growth can be found when approximating natural resources with total, food or fuel exports.

• A positive relationship can be detected when using agricultural exports as a proxy. Agricultural exports thus seem to bring a resource blessing rather than a resource curse.

• In all regressions where the proxy for natural resources is significant the curse, or blessing, can be reversed with good enough institutions.

• The results seem to be similar to those of Sala-i-martin and Subramanian (2003) when instrumenting for institutional quality and the interaction term with latitude, the interaction between the natural resource proxy and latitude and the fraction of people in a country speaking a western European language. Resource abundance appears to influence institutions negatively and that in turn will influence growth negatively. Natural resources do not seem to have a direct negative effect on growth.

become richer, that the average investment has become higher or that some countries only appear in the regressions for the time period 1990-2010.

• The results seem robust when exclude outliers and include or exclude different variables.

• When excluding Africa or the poorest countries the resource curse disappears indicating that much of the curse is located in Africa and the poorest countries in the world.

The main questions to be answered in this study were: do resource abundant

countries, depending on the level of appropriability of the resources, suffer from slower economic growth than resource scarce countries? If they do, can the curse be reversed via good institutions quality? The answers found in the results of this study

References

Boschini, A. D., Pettersson, J. and Roine, J. (2007). “Resource Curse or Not: A Question of Appropriability”, The Scandinavian Journal of Economics 109(3): 593– 617.

Boschini, A. D., Pettersson, J. and Roine, J. (2013). “The resource curse and its potential reversal”, World Development, 43: 19–41.

CIA (2013) - The World Factbook. Avaliable at: https://www.cia.gov/library/ publications/the-world-factbook/index.html (Accessed June 7, 2013). Collier, P. and Hoeffler, A. (2002). “On the Incidence of Civil War in Africa”, Journal

of Conflict Resolution, 46(1): 13–28.

Themnér, L. and Wallensteen, P. (2012). ”UCDP/PRIO Armed Conflict Dataset”, version 4-2012. Available at: http://www.prio.no/Data/Armed- Conflict/UCDP-PRIO/Armed-Conflicts-Version-X-2009/ (Accessed June 7, 2013).

Heston, A., Summers, R. and Aten, B. (2012). Penn World Table Version 7.1, Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania. Available at: https://pwt.sas.upenn.edu/ php_site/pwt71/pwt71_form_test.php (Accessed June 7, 2013).

Isham, J., Woolcock, M., Pritchett, l. and Busby, G. (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. Lederman, D. and Maloney, W. F. (2008). “In search of the Missing Resource Curse”,

Policy Research Working Paper 4766. Available at:

https://openknowledge.worldbank.org/bitstream/handle/10986/6901/ WPS4766.pdf?sequence=1 (Accessed June 7, 2013).

Mehlum, H., Moene, K. and Torvik, R. (2006). “Institutions and the Resource Curse”,

The Economic Journal 116(508): 1–20.

Pineda, J. and Rodríguez, F. (2010). “Curse Or Blessing? Natural Resources and Human Development”, United Nations Development Programme, New York. Available at: http://hdr.undp.org/es/informes/mundial/idh2010/ trabajos/HDRP_2010_04.pdf (Accessed June 7, 2013).

Sachs, J. D. and Warner, A. M. (1995). “Natural resource abundance and economic growth”, National Bureau of Economic Research Working Paper 5398. Available at: http://www.nber.org/papers/w5398 (Accessed June 7, 2013).

Sachs, J. D. and Warner, A. M. (1999). “The big push, natural resource booms and growth”, Journal of development economics 59(1): 43–76.

Sachs, J. D. and Warner, A. M. (2001). “The curse of natural resources”, European

economic review 45(4): 827–838.

Sala-i-Martin, X. and Subramanian, A. (2003). “Addressing the natural resource curse: An illustration from Nigeria”, National Bureau of Economic Research Working Paper 9804. Available at: http://www.nber.org/ papers/w9804 (Accessed June 7, 2013).

Teorell, J., Samanni, M., Holmberg, S. and Rothstein, B. (2011a). “The QoG Codebook”, Version 6Apr11. University of Gothenburg: The Quality of Government Institute. Available at: http://www.qog.pol.gu.se/data/ datadownloads/qogstandarddata/ (Accessed June 7, 2013).

Teorell, J., Samanni, M., Holmberg, S. and Rothstein, B. (2011b). “The QoG Standard Dataset”, Version 6Apr11. University of Gothenburg: The Quality of Government Institute. Available at: http://www.qog.pol.gu.se/data/ datadownloads/qogstandarddata/ (Accessed June 7, 2013).

The World Bank (2013) - World Development Indicators Online Database. Available at: http://databank.worldbank.org/data/views/variableselection/

Appendix

Table 5: First stage regressions

1970-1990 (1) 1990-2010 (2) 1970-1990 (3) 1990-2010 (4) Latitude 0.490** (0.225) 0.441*** (0.136) -0.107** (0.041) -0.062 (0.038)

Interaction between latitude and ores and metals exports -2.645** (1.294) -1.476** (0.597) 1.321*** (0.170) 0.510 (0.371)

Fraction speaking European language 0.073

(0.048) 0.120*** (0.040) 0.002 (0.007) -0.011 (0.009) N !! 69 0.8033 66 0.8357 69 0.9097 66 0.9055 Notes: Dependent variable in column 1 and 2 is institutional quality and the dependent variable in column 3 and 4 is the interaction between ores and metals exports and institutional quality. Robust standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include a constant, ores and metals exports and control variables listed in the text (not shown).

Table 6: Results without outliers

1970-1990 1990-2010

Ores and metals exports -6.529***

(1.126) 2.060 (2.029) Institutional quality 1.313 (0.906) 1.933* (1.011) Interaction term 11.887*** (2.272) -3.619 (3.600) N !! 62 0.7677 58 0.3938

Table 7: Results when excluding different variables, 1970-1990 Ores and metals exports -5.605*** (1.889) -4.612** (1.767) -5.810*** (1.951) -5.320*** (1.883) -5.028*** (1.900) -5.413*** (1.941) Institutional quality -0.384 (1.047) 1.875** (0.909) -0.510 (1.009) -0.542 (1.045) 1.718* (0.884) -0.585 (1.007) Interaction term 11.037*** (3.818) 8.026** (3.402) 11.422*** (3.871) 10.363*** (3.650) 8.738** (3.639) 10.544*** (3.680) Initial GDP level -0.104*** (0.013) -0.110*** (0.013) Average Openness 0.167 (0.274) 0.387* (0.209) 0.066 (0.303) Average Investment 0.029 (0.027) 0.028 (0.030) N !! 69 0.4942 69 0.6159 69 0.4967 69 0.5085 69 0.6291 69 0.5089 Notes: Notes: Dependent variable is Growth. Robust standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include control variables for Africa, The Middle East, Latin America and Asia (not shown).

Table 8: Results when excluding different variables, 1990-2010

Table 9: Results when including variables for war or civil war

1970-1990 1970-1990 1990-2010 1990-2010

Ores and metals exports -5.357***

(1.861) -6.028*** (1.848) 1.836 (2.094) 1.452 (2.271) Institutional quality 1.599 (0.964) 1.176 (0.991) 2.433* (1.330) 2.219 (1.327) Interaction term 9.442*** (3.540) 11.050*** (3.471) -4.648 (4.346) -3.997 (4.432) War -0.209 (0.382) 0.121 (0.446) Civil War -0.695 (0.482) -0.152 (0.496) N !! 69 0.6309 69 0.6473 66 0.2768 66 0.2774

Notes: Dependent variable is Growth. Robust standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include a constant and control variables listed in the text (not shown).

Table 10: Results when excluding different regions, 1970-1990

Excluding Africa Excluding Middle East Excluding Latin America Excluding Asia

Ores and metals exports -2.899

(1.764) -5.258*** (1.939) -6.666*** (1.590) -5.007** (1.922) Institutional quality 1.963* (1.074) 0.977 (0.878) 2.454** (1.218) 1.953* (1.144) Interaction term 5.313* (2.875) 9.194** (3.699) 11.319*** (3.706) 9.128** (3.644) N !! 55 0.6459 64 0.6479 51 0.6997 59 0.5138

Table 11: Results when excluding different regions, 1990-2010 Excluding Africa Excluding the Middle East Excluding Latin America Excluding Asia

Ores and metals exports 1.587

(2.055) 2.341 (2.031) -15.320*** (4.336) 1.336 (2.613) Institutional quality 2.328* (1.187) 2.972** (1.194) 1.052 (1.294) 1.417 (1.223) Interaction term -4.325 (4.045) -4.650 (5.735) 19.323*** (6.309) -0.515 (5.214) N !! 62 0.2883 59 0.3362 46 0.5380 54 0.1265

Notes: Dependent variable is Growth. Robust standard errors in parenthesis. * significant at 10 % level, ** significant at 5 % level and *** significant at 1 % level. All regressions include a constant and control variables listed in the text (not shown).

Table 12: Results when excluding rich or poor countries

1970-1990 (1) 1990-2010 (2) 1970-1990 (3) 1990-2010 (4)

Ores and metals exports -4.919**

(2.160) 3.754* (2.032) -2.762 (1.913) 1.374 (2.854) Institutional quality 2.558* (1.379) 3.169** (1.436) 1.981 (1.278) 1.146 (1.175) Interaction term 8.187* (4.616) 8.452* (4.738) 5.113 (3.088) -0.892 (5.404) N !! 52 0.6252 50 0.3298 52 0.6002 50 0.3516

Table 13: Countries included in the study Algeria Angola Argentina Australia Austria Bahrain Bangladesh Belgium Bolivia Brazil Brunei Cameroon Canada Chile China Colombia Costa Rica Denmark Dominican Republic Ecuador Egypt El Salvador Finland France Gabon Ghana Greece Guatemala Guyana Haiti Honduras Hungary Iceland India Indonesia Iran Ireland Israel Italy Ivory Coast Jamaica Japan Jordan Kenya Kuwait Liberia Malawi Malaysia Malta Mexico Morocco Netherlands New Zealand Nicaragua Nigeria Norway Oman Pakistan Panama

Papua New Guinea Paraguay Peru Philippines Poland Portugal Saudi Arabia Senegal Singapore South Africa South Korea Spain Sri Lanka Sudan Suriname Sweden Switzerland Syria Thailand Togo

Trinidad & Tobago Tunisia Turkey United Kingdom United States Uruguay Venezuela Zambia

Table 14: Sources and definitions of variables

Variable Source and definition

Growth Source: Heston et al (2012). Growth is the average yearly growth rate of

GDP in percent calculated as (ln  (!"#_!/!"#_!)/!)×100 where !"#_! is the GDP level at the end year and !"#_! is the GDP level at the start year. The variable is originally called rgdpch and is the PPP adjusted GDP per capita at 2005 constant prices.

Total exports Source: The World Bank (2013). Total exports are measured at the start

year and is calculated as the sum of agricultural, food, fuel and ores and metals exports.

Agricultural exports Source: The World Bank (2013). Agricultural raw material exports are measured at the start year and include Standard International Trade Classification (SITC) section 2 excluding divisions 22, 27 and 28.

Food exports Source: The World Bank (2013). Food exports are measured at the start

year and include SITC section 0, 1, 4 and division 22.

Fuel exports Source: The World Bank (2013). Fuels exports are measured at the start

year and include SITC section 3.

Ores and metals exports Source: The World Bank (2013). Ores and metals exports are measured at the start year and include SITC division 27, 28 and 68.

Institutions Source: Teorell et al (2011b). The ICRG index includes measures on

quality of the bureaucracy, corruption in government and rule of law.

Initial GDP level Source: Heston et al (2012). The variable is originally called rgdpch and is the PPP adjusted GDP per capita at 2005 constant prices. GDP is measured at the start year. Measured in thousands.

Average openness Source: Heston et al (2012). The variable openness is the sum of

imports and exports as share of GDP and is originally called openc. Average openness is then calculated as the sum of openness from the start year to the end year divided by the number of years in the period.

Average investment Source: Heston et al (2012). Investment is the share investment to GDP

and is originally called ki. Average investment is then calculated as the sum of investment from the start year to the end year divided by the number of years in the period.

Regions Source: CIA (2013). Regions used are Africa, the Middle East, Latin

America and Asia. Dummy variables are used; they take the value 1 if a country is located in the region and 0 otherwise.

Latitude Source: CIA (2013). The absolute value of latitude converged to a 0-1

scale.

Fraction of people speaking a western European language

Source: Lewis et al (2013). The fraction of people in a country that speaks English, German, French, Spanish or Portuguese as first language.

War Source: Themnér and Wallensteen 2012. A country is said to be at war a specific year if there has been at lest 25 battle-related deaths during that year.

Civil War Source: Themnér and Wallensteen 2012. A country is said to be in a