How institutions and natural resources influence the long-run economic growth in Africa

(1)

How institutions and natural resources influence

the long-run economic growth in Africa

Master Thesis MSc International Economics and Business

Author: Joana Valente Ribeiro e Castro Teixeira

Student number: 3505448

Supervisor: Dr. J. Bolt

Co-assessor: Dr. M.A. Papakonstantinou

Submission date: 19th_{June 2018}

Abstract:

The ‘curse of natural resources’ has been widely spread, even though the empirical support for the theory is weak. Nevertheless, previous studies argue that the positive impact of institutions on economic growth helps countries overcome resource dependence. This study focuses on the African continent and corroborates the existence of the ‘curse’, when testing for the relationship between resource dependence and institutions and other determinants of economic growth.

(2)

Table of Figures

Figure 1: Relationship between the average GDP growth rate and the average exports of

natural resources (1970-2015) ... 4

Table 1: Summary statistics ... 13

Table 2: Four models representing the evolution of the regression results... 19

Table 3: Full regression results ... 22

Table 4: Full regression with the Worldwide Governance Indicators (WGI) ... 24

(4)

1

1. Introduction

Over the centuries continents and countries have been growing at different paces. With the increasing globalization in the last decades, the world is becoming more and more interconnected and countries are no longer immune to indirect shocks.

The erratic economic development of Africa has been studied for quite some time now but it still seems to be puzzling, giving room for publications regarding the possible causes and consequences of the slow growth of Africa. Concerning this, researchers studied, among others, the influence of economic and trade policies, historical background, initial conditions, demography and ethnicity, geographical constraints, education and human capital.

The main research question of the paper is how institutions and natural resources influence the long-run economic growth of African countries. This question originates multiple sub-questions such as: Does resource dependence affect growth? If so, is the impact positive or negative? Does the ‘curse of natural resources’1 exist? What is the impact of institutions and do they help to minimize a possible negative effect from being resource dependent? Finally, do resource-dependent countries react differently to external shocks such as the 2008 financial crisis?

There is little consensus about the impact of natural resources richness on economic growth and the drivers of that impact. Havranek, Horvath and Zeynalov (2016) find that from the 43 econometric studies performed in the last 20 years regarding the effect of natural resources on economic growth, 40% of the estimates are negative and statistically significant, 40% are insignificant and 20% are positive and statistically significant.

The impact of institutions is more consensual with a large body of literature supporting the theory that better institutions have a positive effect on economic growth (Efendic, Pugh and Adnett, 2011). With regard to the relationship between institutions and natural resources there are three main views among the academics: the first one suggests that the quality of institutions is hurt by resource abundance and that a decay in the quality of the institutions may drive countries to the ‘resource curse’; the second one indicates that institutions do not play an important role; and finally, the third one where resources interact with the quality of institutions in such a way that resource abundance

1_{Countries with great natural resource wealth tend nevertheless to grow more slowly than resource-poor}

(5)

2 is a blessing when institutions are good and a curse when institutions are bad (Mehlum, Moene and Torvik, 2006a).

This study adds to the already published literature as it focuses on 50 African countries, including area specific variables. It analyzes a large period of time (from 1970 until 2015), which allows for a different approach from previous studies (panel data instead of a cross-section analysis). Regarding the treatment of the main variables, I choose to follow Auty (1990) and compute a natural resource dependence index, which includes not only the share of exports of natural resources, as well as natural resources rents. I use fuel dependence as a proxy for natural resources. Finally, as I am mainly interested in understanding the relationship between institutions and natural resources, I include an interaction term between those variables. By overcoming some of the critics made to previous publications, I intend to obtain more robust results.

Overall, I find no statistical evidence that natural resource dependence has any influence, either positive or negative, on economic growth. The quality of institutions, on the contrary, has a highly significant impact on GDP per capita growth, even when using different measurements of institutions. The interaction term between fuel dependence and institutions is significant and influences growth negatively. One possible explanation might be that fuel dependent countries have bad institutions, opening an opportunity for rent-seeking behavior, or, by contrast, that countries with good institutions are not fuel dependent.

(6)

3

2. Literature Review

There is an extensive amount of literature regarding the factors that affect the long-run economic growth of Africa. The impact of geographical settings (tropical climate, natural resources abundance, landlocked countries), the type of trade policies implemented, the disease burden, the colonial heritage and its different forms, the ethnolinguistic diversity, the differences in education or the quality of institutions have been some of the causes investigated.

Even though the impact of resources has been studied in several papers, Havranek, Horvath and Zeynalov (2016) find in their meta-analysis regarding the impact of natural resources in economic growth that only 34% of the studies include an interaction term between institutions and natural resources. Moreover, the measurement of the natural resources has been criticized by some authors as resource abundance, i.e. the share of natural resources in total exports, might not necessarily mean that natural resources drive growth. Economic development may lean on other factors, and so it is also important to look at natural resources rents to investigate the true impact of natural resources in GDP.

Next, I look into the literature published to date to try to find answers to the questions regarding the potential impact of institutions on economic growth, how resource dependence and the ‘curse of natural resources’ affects the development of a nation, and how both institutions and natural resources interact.

2.1 Natural resources

(7)

4 resources and differentiating between resource dependence and abundance are the main determinants for differences in results across studies.

The contradicting results go far beyond the empirical studies. Gunton (2003) points out two theoretical models regarding the importance of natural resources: the dependence theory and the comparative advantages one. Some arguments from the former include high fixed costs due to capital intensive activities to extract the natural resources; expatriation of profits and dividends, as most firms in the sector are foreign-owned; and impeding the emergence of a strong independent entrepreneurial class. According to the supporters of the comparative advantages theory, foreign capital promotes entrepreneurship, through export-led growth and technology, expediting the development process and improving technological capacity and skills.

Figure 1 shows the relationship between the average real GDP per capita growth rate and the average share of natural resources in total exports for the countries sampled in this study, from 1970 until 2015. There is a negative linkage between both indicators, especially for countries that have high exports of natural resources, such as Angola, Zambia, Gambia, Nigeria and Algeria, which means that the ‘curse of natural resources’ holds when looking specifically at the African continent and within the time period considered. In section 5, I test if the ‘curse’ is empirically confirmed.

MAR AGO BDI BEN BFA BWA CAF CIV CMR COD COG COM CPV DJI DZA EGY ERI ETH GAB GHA GIN GMB GNB KEN LBR LSO MDG MLI MOZ MRT MUS MWI _NAM NGA RWA_SDN SEN_SLE SWZ SYC TCD TGO TUN TZA UGA ZAF ZMB ZWE -3 -2 -1 0 1 2 3 4 5 6 7 0 10 20 30 40 50 60 70 80 90 100 A v er age R eal G D P G ro w th R at e 1970 -2015

Average Exports of Natural Resources, in percentage of GDP, 1970-2015

Curse of Natural Resources

(8)

5

2.1.1 The Dutch disease and the resource curse thesis

The abundance of natural resources often leads to the so-called the ‘Dutch Disease’. The term was first introduced by ‘The Economist’ to summarize the adverse effects on the Dutch manufacturing sector due to the discovery of natural gas reserves in the 1960s, that lead to the appreciation of the Dutch real exchange rate (Corden, 1984).

Davis (1995) states that the Dutch disease is ‘a rather morbid term’ used to describe the coexistence of booming and lagging sectors in an economy. Additionally, he argues that there is nothing growth-inhibiting in this phenomenon per se; if problems are to arise they are due to resource reallocation and the burden of adjustment at least for the losing factor, and the political pressure this puts on governments to intervene.

Many authors that investigated the performance of mineral economies during the 1970s concluded that autarkic industrialization policies, protection of the shrinking sectors, destabilizing exchange rate policies, and government interference discouraged sectoral factor movements, offsetting the windfall gains (Davis, 1995).

The ‘curse of natural resources’ is, often, mistaken for the ‘Dutch disease’. The ‘Dutch disease’ worsens the competitiveness of the sector that is lagging behind, promotes deindustrialization and distorts the normal market functioning, as governments tend to protect that sector. When institutions are bad, the ‘Dutch disease’ might also give room for rent-seeking behavior, with governments not using the windfall profits from the booming sector to tackle social inequalities. However, the ‘Dutch disease’ is not experienced only in the natural resources sector; it is inherent to any dichotomic sector. The ‘curse of natural resources’ may be seen as a consequence of the ‘Dutch disease’, meaning a consequence of abundance of natural resources. Nonetheless, Davis (1995) argues that it may also be a consequence of one economy becoming a monoeconomy or an undiversified one, hence more subjected to global price shocks, and not a consequence of the type of goods the economy specializes in.

(9)

6 development may lean on other factors, so it is also important to look at natural resources rents to investigate the true impact of natural resources in GDP. Nonetheless, as resource abundance is a component of the measure of resource dependence, I can still draw the following conclusion from the literature reviewed so far:

Hypothesis 1: Resource dependence affects growth negatively.

2.2. Institutions

Institutions are often linked to the colonial experience of the countries. Acemoglu, Johnson and Robinson (2000) divide the colonizer countries into two groups: the “extractive states”, where there was not much protection for private property and their main purpose was to transfer as much resources as possible to the metropolis; and the “Neo-Europes”, where the settlers tried to replicate the European institutions and where private property was protected. To determine whether a colony belongs to one of the referred groups, they use data from the Political Risk Services (PRS) and compute an index of protection against expropriation that attributes 0 to lowest protection and 10 to the highest in order to distinguish between colonial origins. They try to find a relationship between GDP and the average expropriation risk between 1985 and 95 and conclude that there is a strong positive correlation between institutions and economic performance, meaning that different political institutions affect colonies differently.

Bertocchi & Canova (2002) include also an indicator of political institutions, but they link the lower levels of institutions to former Portuguese and Belgian colonies (characterized by extractive states and extreme forms of exploitation) and higher levels of the indicator to British colonies. They confirm that alternative colonial policies may have had an impact on GDP growth rates, once the initial conditions are taken into account. Over the full sample, British colonies were the ones who grew the most, followed by the French and then indistinctly by the others.

(10)

7 The majority of the literature finds that better institutions, i.e. institutions that protect private property and rank higher in either of the indexes, have a positive and significant impact in the long-run economic growth and as such I may conclude that:

Hypothesis 2: Better institutions have a positive impact on economic growth.

2.3 Natural resources and institutions

Although the majority of the countries that are resource abundant experience slower growth when compared to poor-resource countries, exceptions like Norway and Botswana make researchers wonder how these countries escaped the ‘resource curse’. Perhaps more intriguing is what differentiates Botswana from Sierra Leone. Both countries are dependent on diamonds exports, but while Botswana managed to use the revenues from the mining exports to develop the economy and decrease poverty, Sierra Leone’s economic growth is meager, with low human development and armed conflicts. UNCTAD suggests that the improved political process of Botswana, where the parliament oversees the approval of new investment projects and allocates the expenditures on education, vocational training and health services, gave Botswana political stability and allowed the country to develop (UNCTAD, 2017).

However, is there empirical evidence that the quality of institutions is linked to natural resources and that both variables, either by themselves or interacted, affect growth? There are three main views on the topic among the academics: first, where the quality of institutions is hurt by resource abundance and constitutes the intermediate causal link between resources and economic performance; second, where institutions do not play an important role; finally, the one where resources interact with the quality of institutions such that resource abundance is a blessing when institutions are good and a curse when institutions are bad (Mehlum, Moene and Torvik, 2006a).

(11)

8 affects negatively the quality of institutions, contradicting the rent-seeking behavior theory.

Sachs and Warner (1995) include a measurement of bureaucracy to try to explain the resource curse, but they find no statistical significance. Mehlum, Moene and Torvik (2006a) claim that the conclusions from Sachs and Warner do not rule out the possibility that institutions play a role only when institutions are bad, as they only consider institutions as an intermediate causal link.

Finally, the authors that support the dichotomic relationship curse/blessing between resource abundance and institutional quality argue that in countries with good institutions resource abundance attracts entrepreneurs into production, while in countries with weak institutions, entrepreneurs are diverted away from production and into unproductive rent appropriation (Mehlum, Moene and Torvik, 2006a). Brunnschweiler and Bulte (2008) find that the association between resources and institutions runs the other way around: greater abundance leads to better institutions and more rapid growth.

In Mehlum, Moene and Torvik (2006b) the authors start by distinguishing between the cases where rent-seeking activities and production are competing activities and the cases where they are complementary activities. In the first case, the rent-seeking activities and the production activities are different which pays off when institutions are bad and the wealth can be appropriated by political insiders, bureaucrats, and so on. This gives room for ‘grabber-friendly’ institutions. By contrast, when institutions are better, or, as they call them, ‘producer friendly’, rent-seeking activities and production activities coincide. They find that countries rich in natural resources may be either growth ‘losers’ or ‘winners’. It depends on the type of institutions in place. The combination of grabber friendly institutions and resource abundance leads to low growth and countries, eventually, fall in the ‘resource curse’ trap. Producer friendly institutions, however, help countries to take full advantage of their natural resources which, in turn, helps them escape the ‘resource curse’ trap.

I expect to find support for the theory that institutions and natural resources influence each other and that their interaction affects growth, and assuming that the causality runs from institutions to resource dependence:

(12)

9

3. Data

In this section I briefly describe the data and variables included in my analysis and also explain what differentiates my approach from previous publications.

The choice of the following variables was motivated by the most influential published literature regarding economic growth, especially economic growth in Africa, although in some cases I will try to find an alternative approach to those variables, as they have received some critics over the years.

3.1 Dependent variable

To measure economic growth (the main focus of this paper when analyzing the impact of resource dependence and institutions), two different indicators are traditionally used: either GDP per capita or GDP per capita growth rate.

Given the heterogeneity of the countries included in the sample and also since I am interested in possible changes in economic growth, I use GDP per capita growth rate as the dependent variable (GDP). However, to control for different initial GDP levels, I include a variable concerning the GDP level in 1970 (GDP70). This allows to identify whether poorer countries grew more rapidly than richer ones, taking advantage of the innovations already developed in advanced countries (Havranek, Horvath and Zeynalov, 2016).

3.2 Independent variables 3.2.1 Natural resource dependence

(13)

10 Natural resources include a vast range of goods with different characteristics and trends (e.g. agricultural, mineral and fuel goods), and although, ideally, I would like to differentiate between all the different types of natural resources, due to lack of data available, I focus my analysis in the fuel sector.

To compute the index, I use data from the World Bank. Fuel exports comprise the commodities in SITC2 section 3 (mineral fuels, lubricants and related materials). The rents consist of the difference between the value of production at world prices and their total costs of production. Additionally, fuel rents were computed as an average of oil, coal and natural gas rents. The dependence is measured in the regression by a dummy (Fuel) that takes the value 1 when one country in a certain year is fuel dependent.

3.2.2 Institutions

The impact of institutions, as stated before, is an often investigated subject and the results are, in general, consensual: better institutions have a positive impact on economic growth, independently of the source used for computing the indicator. Efendic, Pugh and Adnett (2011) support that the positive effect of institutions is on its way to becoming ‘conventional wisdom’ but they point out that evidence for the impact of institutions is not as robust as it could be and suggest that future literature should be clearer about the possible endogeneity of institutions, the specification of the model to be estimated and the choice of dependent variables.

In this paper I follow the Knack and Keefer (1995) approach and use an average of four indicators from the ICRG to assess the quality of institutions. However, this database only covers the period between 1984 and 2011. As such, I also add the World Bank governance indicators to include the remaining period and also as a robustness test (see further details in section 5.2).

The institutional quality index (Institutions) ranks countries from 1 to 6 based on the average of the following indicators: bureaucracy quality, corruption, law and order and investment profile3. According to the ICRG methodology, in the bureaucracy index countries that lack the cushioning effect of a strong bureaucracy receive low points because a change in government tends to be traumatic in terms of policy formulation and day-to-day administrative functions. Corruption includes not only the most common form

(14)

11 of corruption as special payments and bribes but also actual or potential corruption in the form of excessive patronage, nepotism, job reservations, ‘favor-for favors’, secret party funding, and suspiciously close ties between politics and business. Law and order is composed by two separate components, scored from zero to three points each. To assess the “Law” element, the strength and impartiality of the legal system are considered, while the “Order” element is an assessment of popular observance of the law. At last, investment profile covers factors affecting the risk to investment that are not covered by other political, economic and financial risk components. The risk rating assigned is the sum of three subcomponents (contract viability/expropriation, profits repatriation and payment delays), each with a maximum score of four points (Prsgroup.com, 2018).

3.2.3 Interaction between resource dependence and institutions

Furthermore, my research investigates if there is a linkage between institutions and resource dependence. This is an important factor as the causality between the two indicators is not consensual: does resource dependence lead to worse institutions or do bad institutions lead countries to the ‘resource curse’? Put it another way, do good institutions use the profits from the natural resources to develop the economy or do they engage in rent-seeking behavior?

Following Mehlum, Moene and Torvik (2006b), I include an interaction term between fuel dependence and the institutions variable (Institutions_Fuel), in order to capture the joint effect both variables. As I do not test the causality between Institutions and Fuel, I assume that it runs from the institutions to natural dependence. By regressing the variables separately and the interaction term I expect to find how all the variables affect each other. Mehlum, Moene and Torvik (2006b) find a positive and significant coefficient for the interaction between institutions and natural resources and interpret it saying that only when the quality of the institutions is poor, resource abundance harms economic growth.

3.3 Control variables

(15)

12 Havranek, Horvath and Zeynalov (2016) compile the most included macroeconomic characteristics: 79% of the studies included control for initial GDP, 67% for trade openness, 62% for investment and 49% for level of schooling. As I am focusing on African countries, and also following former literature such as Sachs and Warner (1997), I include the beforementioned variables and add some factors that are characteristic of the region.

I, then, control for colonial ruling (with a dummy variable – British - that takes the value 1 for former British colonies and 0 otherwise), the disease burden (proportion of people tested positive for the malaria infection), the level of human capital (based on years of schooling and returns on education), different initial GDP levels (given by the gross GDP per capita in 1970), being landlocked, openness to international trade (exports as a percentage of GDP) and the growth rate of gross capital formation (also known as domestic investment). By including these variables, I can isolate the effect of resources and institutions and better understand their true magnitude. Additionally, as I am looking into a large period of time, I include dummy variables to distinguish between decades. After the 2000s, the periods 2000-2008 and 2009-2015 show the effect pre and post financial crisis. Appendix Table A2 gives a brief description of all the variables included and their sources.

(16)

13 Table 1: Summary statistics

Variable Observations Mean Std. Dev. Minimum Maximum

GDP 2,111 1.435915 7.31185 -50.2301 140.5011 Fuel 1,062 0.1431262 0.3503665 0 1 Institutions 962 2.480255 0.7061621 0.01875 4.25 Institutions_Fuel 562 0.4013197 0.8880738 0 3.125 British 2,208 0.3541667 0.4783684 0 1 Exports_GDP 2,023 30.19446 18.83848 2.524688 124.3932 Landlocked 2,300 0.32 0.4665776 0 1 Human Capital 1,845 1.502853 0.3783858 1.007409 2.809442 GDP70 1,702 1,287.842 1,444.183 265.2292 7,189.076 Malaria 922 1,105.004 3,425.761 0 84,810

Investment growth rate 1,439 6.528763 80.29384 -2,562.384 496.3599

Source: See text

Recapping the expected relationship between the dependent variable, GDP per capita growth rate, and the independent variables follows: Fuel has a negative impact on economic growth, whilst Institutions have a positive influence. As I assume that the causality runs from institutions to resource dependence, I also expect the interaction term between the quality of institutions and fuel dependence to have a positive coefficient. Former colonial experience has a negative influence on economic growth. Trade openness is expected to have a positive effect on growth and so does the Human Capital variable. The three remaining variables, Landlocked, GDP70 and Malaria, are likely to have a negative effect on economic growth. In section 5, I test these predictions.

(17)

14

4. Methodology

The aim of this paper is to understand why economic growth is so different across African countries, even if they have similar characteristics, such as resource endowment, location, climate or colonial experience.

I next revise the determinants of growth using an empirical approach, based on the existing literature about the topic and, then, specify the model used in this paper.

4.1 Previous studies

As summarized by Havranek, T., Horvath, R. and Zeynalov, A. (2016), many researchers on the topic of the impact of the natural resources and institutions use a variant of the following model:

𝐺_𝑖𝑡 = 𝛼 + 𝛽𝑁𝐴𝑇_𝑖𝑡+ 𝜑𝐼𝑁𝑆_𝑖𝑡+ 𝛾𝑁𝐴𝑇_𝑖𝑡∗ 𝐼𝑁𝑆_𝑖𝑡+ 𝜃𝑋_𝑖𝑡+ 𝜀_𝑖𝑡, (1)

where i and t denote country and time subscripts; G represents the measure of economic growth; NAT represents a measure of natural resource richness; INS represents the institutional quality of a country and NAT*INS is an interaction term between natural resources and institutional quality; X is a vector of control variables, such as macroeconomic conditions; and ε denotes the error term.

Havranek, Horvath and Zeynalov (2016) conclude that about 80% of the studies regarding natural resources and economic growth use a cross-sectional analysis, where averages of the variables are used and usually smaller periods of time are considered. This methodology might suffer from omitted variable bias and also includes time periods with very different growth rates, but due to the characteristics of averages, it is not possible to spot those dichotomies. As such, I intend to use panel data instead of the traditional approach by a cross-section analysis, in order to minimize these issues. In the next section I look into the benefits of such model. Regarding the estimation model, half of the existing studies use OLS regressions, allowing for endogeneity of the natural resources variables, while one third uses instrumental or lagged variables.

(18)

15

4.2 Present model

Longitudinal (or panel) data collects information from a group of cross-sectional units (i) observed over time (j). In this paper, I include 50 African countries (i=1…i=50) and 46 years of data, from 1970 until 2015 (j=1…j=46). This approach aims to increase the number of observations of a typical cross-section and, thus, obtain more robust results. Panel data models increase the efficiency of the estimates as the researcher has a large number of data points, enlarging the degrees of freedom and reducing collinearity among explanatory variables. Using either the fixed-effects or random-effects approach, can solve the omitted variables bias as invariant characteristics correlated with the observed ones may be included, and both models allow for individual heterogeneity. A pooled OLS is an alternative approach, but groups and time structures are not taken into account as different individuals are pooled together. As such, a pooled OLS does not recognize that different individuals and years are present in the dataset, minimizing the benefits of using a panel dataset.

Focusing on the fixed-effects (within the estimator) and random-effects approach, the former examines the variation within one individual, which means that time-invariant variables are not individually specified as they do not differ across time, being absorbed by the intercept. The random-effects approach states that differences between individuals are random, drawn from a given distribution and all individual differences are captured by the intercept parameters. A critical assumption in the random-effects model is that the random error 𝜀_𝑖𝑗 is uncorrelated with the explanatory variables. As such, if there is no correlation between the error term and the variables, both random-effects and fixed-effects estimators are consistent. However, if that is not the case, the fixed-fixed-effects estimator is consistent but the random-effects one is not.

(19)

16 effects model, meaning that there are significant differences in the panels. The results are presented on the Appendix Table A4.

I additionally tested for endogeneity of both the natural resources dependence and the quality of the institutions variables. Following Brunnschweiler and Bulte (2008), I use 2SLS regressions, where the quality of institutions, the past colonial experience, the land-lockedness of the country and the openness to trade are the instrumental variables for Fuel. For the test of endogeneity of institutions being fuel dependent, the land-lockedness of the country and the past colonial experience are the instrumental variables. The null hypothesis of the Durbin and Wu–Hausman tests is that the variable under consideration can be treated as exogenous. The results are presented in the Appendix Table A5. Both fuel dependence and quality of the institutions are considered exogenous variables, given that I cannot reject the null-hypothesis. This means that I can use a normal regression without instruments and still obtain valid results.

Finally, I tested for the possibility of multicollinearity, using a Variance Inflation Factor (VIF) approach. The results are in Appendix Table A6. In this analysis I excluded the interaction term Institutions_Fuel, as it would bias the results, given that this term is obviously correlated with the two components of the interaction. All variables have a VIF smaller than 5, which means that the variables are not significantly correlated.

To estimate equation (2) I use a random-effects model, with robust standard errors.

𝐺𝐷𝑃𝑖𝑗 = 𝛼0+ 𝛼1𝐹𝑢𝑒𝑙𝑖+ 𝛼2𝐼𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛𝑠𝑖𝑗+ 𝛼3𝐼𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛𝑠_𝐹𝑢𝑒𝑙𝑖𝑗+

𝛼₄𝐵𝑟𝑖𝑡𝑖𝑠ℎ_𝑖+ 𝛼₅𝑀𝑎𝑙𝑎𝑟𝑖𝑎_𝑖𝑗 + 𝛼₆𝐻𝑢𝑚𝑎𝑛𝐶𝑎𝑝𝑖𝑡𝑎𝑙_𝑖𝑗+ 𝛼₇𝐺𝐷𝑃70_𝑖 + 𝛼₈𝐿𝑎𝑛𝑑𝑙𝑜𝑐𝑘𝑒𝑑_𝑖+

(20)

17

5. Results

In this section I first present four models and analyze how controlling for different variables affect the outcome of the regressions, and then I look into more detail to the results of the full regression throughout time. This approach allows me test my hypothesis and conclude about the relationship between institutions and natural resources dependence.

Regarding the evolution of the models, Table 2 presents the four models used, considering the whole time period. I divide the control variables into two groups, differentiating between internal characteristics of the African countries and variables that are known to be determinants of economic growth. Thus, one group contains the colonial heritage, geographical setting and disease burden variables and the other the variables related to trade openness, human capital level, the initial GDP and the domestic investment growth rate. I start by presenting the results for a simplified regression, without control variables, then add separately the two groups of control variables, and finally I present the full regression with both groups. By including the two groups separately I can test which type of control variables have more influence on economic growth.

Although one can claim that there are many variables that are not being controlled for in the simplified regression, some conclusions may already be drawn before including other determinants of growth, especially regarding the sign of the coefficients. In the overall period, only the variable that assesses the quality of the institutions is significant at a 5% level and positive. This means that better institutions promote, in fact, economic growth. The variables for fuel dependence and the interaction term are not significant.

Introducing the first group of control variables removes the significance for the variable of institutions obtained in the simplified model. No other variable is significant, which may hint that, in this scenario, controlling for colonial heritage, land-lockedness and malaria infection does not help to explain the evolution of GDP per capita.

(21)

18 coefficient for this variable is too small to be able to draw conclusions. The investment growth rate has a positive impact on economic growth, which is in accordance with the literature. Domestic investment includes, among others, the construction of roads, railways, schools, offices and hospitals. As such, this variable is important to understand to what extent the African countries are investing in long-term assets, but as visible on Table 2, the coefficient is very small, which may hint that although positive and significant, the impact of investment is still insufficient to have a big effect on economic growth.

Finally, looking at the results for the full regression, the coefficient for Institutions is significant at a 1% level and positive. Countries with better institutions grow, on average, more 1.7%. Although the variable for fuel dependence alone is not significant, its coefficient is positive, which means that, bearing in mind that the insignificance might bias the interpretation, natural resources may not be a ‘curse’. However, looking at the interaction between the quality of institutions and fuel dependence, significant at a 10% level, the coefficient is negative. This negative relationship shows that institutions in fuel dependent countries hold back economic grow in 3.4%. As conjectured before in the summary statistics, the quality of institutions in fuel dependent countries is smaller than in the overall African continent, possibly due to more opportunities for rent-seeking behavior or simply because countries with good institutions are not fuel dependent. Regardless of the motivation, it is possible conclude that institutions influence the economic performance of resource dependent countries.

(22)

19 Table 2: Four models representing the evolution of the regression results

Variables Simplified regression First group control variables Second group control variables Full regression Fuel 1.664 5.616 3.935 7.299 (-4.196) (-6.769) (-3.693) (-4.574) Institutions 0.910** 0.992 1.238*** 1.672*** (0.389) (0.731) (0.437) (0.571) Institutions_Fuel -0.685 -2.979 -1.360 -3.416* (-1.756) (-2.941) (-1.511) (-2.074) British -0.0763 -0.466 (0.796) (-1.021) Landlocked 1.022 2.226** (0.761) (0.996) Malaria -0.0000311 -0.0000619 (0.000116) (0.0000801) Exports_GDP 0.00345 0.0320 (0.0231) (0.0389) GDP70 -0.000494*** -0.000682** (0.000164) (0.000346) Human Capital 1.057 4.442*** (0.776) -1.484 Investment growth rate 0.0305** 0.0760*** (0.0130) (0.0133) _cons -0.846 -1.251 -3.638** -11.11*** (-1.217) (-2.017) (-1.577) (-3.026) N 562 314 380 194 R2_within 0.00306 0.0132 0.113 0.299 R2_overall 0.0262 0.0550 0.114 0.264 R2_between 0.252 0.142 0.172 0.361

(23)

20

5.1 Full regression

In this section, I regress equation (2) using dummy variables to differentiate decades and the period before and after the 2008 financial crisis. The purpose of this segment is to assess which variables remain significant throughout the 274 years and which loose or gain importance.

The variable for fuel dependence does not have statistical influence in any period under analysis. These results are not entirely surprising, as Havranek, Horvath and Zeynalov (2016) find that 40% of the studies regarding the impact of natural resources on growth do not find statistical significance for the coefficients measuring resource abundance/dependence. Consequently, I reject Hypothesis 1, that stated that resource dependence has a negative impact on growth.

The positive and significant, at a 1% level, coefficient for institutions in the overall period is in accordance with the literature, which translates that better institutions, i.e. institutions that protect private propriety, with less corruption and bureaucracy, ranking higher in the quality of institutions index, are beneficial for economic growth. The impact of better institutions increases growth in about 1.7%. This positive relationship between institutions and growth allows me to conclude that Hypothesis 2 is confirmed. When breaking the analysis into decades, the Institutions variable is only significance in the 1980s.

The interaction term, as stated before, is negative in the overall period, although both components of the interaction have a positive effect when regressed by themselves. I find no support for the fact that better institutions in resource dependent countries promote growth (Hypothesis 3). On the contrary, the interaction term seems to suggest, if one assumes that the causality runs from the institutions to resource dependence, that institutions in resource dependent countries have an adverse effect, as fuel dependent countries grow 3.4% slower than non-fuel dependent ones.

The contrast from the results of the interaction term and the ones from the previous studies, such as Mehlum, Moene and Torvik (2006b), may derive from the different countries included in the samples, from differences between countries that are abundant in resources but are not dependent on those resources and, finally, from differentiating types of primary goods. The samples from other studies are not restricted to African countries, taking into account South American countries and also Middle-East ones and

(24)

21 even European countries which may skew upwards the results from the interaction, as the quality of the institutions in the former regions is better than in the one in Africa. Regarding the difference between abundance and dependence and distinguishing between mineral and fuel resources, I reproduce the previous literature as robustness tests in section 5.2.

Being an open economy on the 1980s has a highly significant and positive effect, even though the coefficient is small. This result seems reasonable as in the 1970s most countries became independent and so the 1980s were characterized by the takeoff of those nations on international trade. In contrast with the previous literature, the coefficient for being landlocked is positive and significant in the overall period and also between 1990 and 1999. In this sample, the positive impact may be skewed due to the fact that almost one third of the countries included are surrounded only by land, without access to the sea. The human capital variable has a positive and significant impact on growth on the 1990s, after the financial crisis and in the overall period. These results are in accordance with other studies, as higher levels of schooling imply a more qualified labor force and, consequently, the possibility to engage in more complex activities. The coefficient for the initial level of GDP is very small, which may result in a poor interpretation of the results. Finally, the investment rate has a positive and significant impact in almost all periods considered in the analysis, which gives support to the importance of new infrastructures in economic growth.

(25)

22 Table 3: Full regression results

Variables GDP 1980-1989 GDP 1990-1999 GDP 2000-2008 GDP 2009-2015 GDP 1980-2015 Fuel 14.70 -10.13 -3.150 57.79 7.299 (16.03) (-8.901) (-8.028) (61.22) (-4.574) Institutions 4.222*** -2.779 -0.666 2.687 1.672*** (-1.124) (-1.720) (-1.163) (-2.451) (0.571) Institutions_Fuel -16.14* 0.741 1.499 -25.88 -3.416* (-9.282) (-2.911) (-3.679) (29.25) (-2.074) British 3.092 -1.228 1.892 -1.652 -0.466 (-2.994) (0.912) (-1.247) (-1.564) (-1.021) Exports_GDP 0.326*** 0.0492 -0.0179 0.186 0.0320 (0.125) (0.0638) (0.0417) (0.158) (0.0389) Landlocked 2.563 6.775*** -0.685 0.789 2.226** (-2.717) (-1.932) (0.761) (-1.925) (0.996) Human Capital -4.283 5.721* -3.705 6.497** 4.442*** (-4.394) (-3.101) (-2.261) (-3.269) (-1.484) GDP70 0.00838* 0.000349 0.000110 -0.00526 -0.000682** (0.00500) (0.000362) (0.000323) (0.00437) (0.000346) Malaria -0.000601 -0.000190 0.0000868 0.000590*** -0.0000619 (0.000395) (0.000117) (0.000323) (0.000124) (0.0000801) Investment 0.149 0.0646*** 0.0930*** 0.0631*** 0.0760*** growth rate (0.105) (0.0216) (0.0311) (0.0137) (0.0133) _cons -19.52*** -2.387 7.923** -18.49** -11.11*** (-6.760) (-2.662) (-3.902) (-7.495) (-3.026) N 20 60 83 31 194 R2_within 0.194 0.463 0.285 0.648 0.299 R2_overall 0.523 0.543 0.405 0.598 0.264 R2_between 0.992 0.727 0.585 0.640 0.361

(26)

23

5.2 Robustness test

Due to the incompleteness of data regarding the quality of the institutions, I re-estimate equation (2) using the World Bank governance indicators. Moreover, as mentioned in the previous section, I reproduce the published literature by using natural resource abundance instead of dependence, where abundance is measured using the share of mineral and fuel exports in total exports.

5.2.1 World Bank governance indicators

The Worldwide Governance Indicators (WGI) include the following six dimensions: voice and accountability, political stability and absence of violence/terrorism, government effectiveness, regulatory quality, rule of law and control of corruption. All indicators range from -2.5 until 2.5, being the former attributed to weaker institutions, while the latter to strong ones. These indicators cover a smaller period of time than the ICRG used in as main institutional quality index, from 1996 until 2015, but fill the gap of data availability of the ICRG after 2011.

Table 4 shows the results of equation (2) using WGI to measure the quality of the institutions. The additional variables may be found at the end of the table. Due to insufficient observations for the 1990s decade, I only include the period after 2000. The quality of institutions remains highly significant to explain differences in the economic growth and also with a higher coefficient (2.8%) than the one obtained using the ICRG dataset (1.7%). Fuel remains insignificant. As observed in Table 3, the interaction term between WGI and Fuel is significant and negative. This coefficient is also bigger than the when using the ICRG, meaning it has a bigger adverse effect (-6.1%). One can conclude that regardless of the measurement of the quality of institutions, institutions play an important role explaining economic growth and, moreover, it confirms a relationship between institutions and natural resources.

(27)

24 Table 4: Full regression with the Worldwide Governance Indicators (WGI)

Variables GDP 2000-2008 GDP 2009-2015 GDP 2000-2015 Fuel -4.329 -2.178 -3.545 (-2.991) (-5.769) (-2.969) British 0.771 -2.159*** -0.687 (-1.340) (0.689) (0.635) Malaria -0.000184 0.000483*** 0.000174 (0.000360) (0.000181) (0.000167) Exports_GDP 0.0172 0.0510 0.00782 (0.0401) (0.0364) (0.0240) Human Capital -0.0310 6.575*** 3.108*** (-2.505) (-1.204) (-1.071) GDP70 -0.000186 -0.00252*** -0.000558** (0.000307) (0.000864) (0.000254) Landlocked 0.484 1.346 1.461** (0.967) (0.862) (0.575) Investment 0.0780*** 0.0450*** 0.0532*** growth rate (0.0261) (0.00379) (0.00823) WGI 1.668 3.272*** 2.826*** (-1.146) (-1.114) (0.570) WGI_Fuel -5.295* -6.009 -6.105*** (-3.022) (-4.529) (-2.273) _cons 0.857 -8.069*** -2.886* (-2.952) (-2.396) (-1.710) N 87 71 174 R2_within 0.279 0.536 0.257 R2_overall 0.404 0.598 0.371 R2_between 0.522 0.660 0.502

(28)

25

5.2.2 Resource abundance

In this section I use as indicator for the impact of natural resources the share of mining and fuel on total exports. According to Davis (1995), the World Bank classifies fuel exporters as those countries whose exports of the referred good are 50% or more of the total exports of goods and services. As such, for this robustness test I use a dummy variable that takes the value 1 when the mineral and fuel exports are equal or larger than 50% of total exports, or 0 otherwise as measurement for natural resource abundance.

The results are shown in Table 5. The additional variables can be seen at the end of the table. The variable that assesses the quality of institutions is significant and positive in the overall period, as occurred when using the measurement of resource dependence, with better institutions increasing economic growth by 1.5%. However, the variable

Natural_Resources is not significant in the overall period and neither is the interaction

term between the quality of institutions and the natural resource abundance. In the 1980s both Natural_Resources and Institutions_Natural_Resources are significant. The former has a positive impact on growth, while the interaction term has a negative coefficient. These results confirm what was obtained by regressing equation (2).

Looking at the control variables, human capital, being landlocked and the investment rate remain positive and significant, and the same conclusions referred before regarding the full regression may be drawn for this robustness check.

(29)

26 Table 5: Full regression with natural abundance

Variables GDP 1980-1989 GDP 1990-1999 GDP 2000-2008 GDP 2009-2015 GDP 1980-2015 Institutions 5.862*** -2.501** 0.484 1.132 1.515*** (-1.758) (-1.185) (0.994) (-2.125) (0.382) British -0.0658 -1.752 1.948** -0.0512 -0.321 (-2.325) (-1.300) (0.855) (-1.545) (-1.099) Malaria -0.0000520 -0.0000404 0.000282 0.000484*** -0.0000660 (0.000554) (0.0000948) (0.000400) (0.000122) (0.0000863) Exports_GDP 0.148 -0.00846 -0.0135 0.0784 0.0136 (0.111) (0.0426) (0.0411) (0.0494) (0.0327) Human Capital -12.05 5.299 -4.080** 4.404* 4.090*** (-7.538) (-3.296) (-1.966) (-2.489) (-1.422) GDP70 -0.00530** 0.000374 -0.00000585 -0.00226 -0.000759*** (0.00217) (0.000477) (0.000245) (0.00203) (0.000262) Landlocked -2.070 5.730*** -0.622 2.448** 2.298** (-4.202) (-1.342) (0.698) (0.978) (0.912) Investment 0.0988 0.0599*** 0.0929*** 0.0626*** 0.0803*** growth rate (0.0829) (0.0199) (0.0301) (0.0134) (0.0130) Natural_Resources 24.15* -2.150 -2.287 -0.338 3.879 (13.44) (-6.866) (-6.461) (10.24) (-4.811) Institutions_ -10.60* -1.708 1.782 1.302 -1.345 Natural_Resources (-6.118) (-2.509) (-2.971) (-4.303) (-2.331) _cons 1.577 -1.170 5.037 -11.51** -9.918*** (-6.285) (-2.029) (-3.515) (-5.235) (-2.732) N 21 62 82 31 196 R2_within 0.246 0.420 0.290 0.626 0.310 R2_overall 0.555 0.490 0.421 0.687 0.288 R2_between 0.890 0.588 0.560 0.772 0.410

(30)

27

6. Conclusion and final remarks

After presenting the theoretical models and the empirical results of the impact of institutions, natural resource dependence and the relationship between both variable, I am able answer my research question and sub-questions.

Regarding the negative impact of resource abundance or dependence, I find no empirical support and so I reject Hypothesis 1. This suggest that the ‘curse of natural resources’ does not have empirical support and, so, there are other factors that may explain the slow growth of the African countries.

For measuring the impact of institutions on economic growth, the results are quite robust as even using different sources for the institutions’ quality (the ICRG indicators or the World Bank indicators), the associated coefficients are positive and significant. These results are in accordance with a great part of the literature and so they confirm Hypothesis 2, meaning that better institutions promote economic growth.

Concerning the interaction terms between natural resource dependence and institutions, I find a negative and significant impact of institutions in fuel dependent countries. This result is contrary to the one found by Mehlum, Moene and Torvik (2006a), where the coefficient of the interaction term is positive, meaning that resource abundance is harmful to growth only when the institutions are bad. One possible explanation for the contradicting results is that fuel dependent countries have simply worse institutions or that countries with good institutions are not fuel dependent. In addition, one may infer that as both Fuel5_{and Institutions have a positive coefficient, the negative impact of the}

interaction term arises from misallocation of the windfall gains of resource dependence, confirming the ‘resource curse’ theory.

This study adds to previous literature due to the method used (panel data) and also by including more recent data, which helps to find if the African countries are improving (or not) their economic conditions. Testing for fuel dependence gives different results than the usual abundance variable used in 70% of the studies (Havranek, Horvath and Zeynalov, 2016), and suggests that natural resources rents are also important to study resource dependence. The interaction term also gives insight to the relationship between institutions and natural resources, and, as I find a negative impact on economic growth, the paper adds more controversy to the academic debate.

(31)

28 Nonetheless, the study has some limitations: although I am using panel data, due to the lack of data available and the high number of control variables introduced, the number of observations in each decade is small. As such, the conclusions drawn from those coefficients may be biased to the countries that are included in the sample. This problem can, probably, be tackled in future studies as more data will be collected. The use of a random-effects model and of a lin-lin regression may not be the most suitable model, even if other studies use these approaches. A more complex model may help to fully appropriate the impacts of natural resources and institutions. Finally, the African countries are very heterogeneous among themselves, and generalizing the conclusions might distort the reality of some countries.

The analysis of the presumed ‘curse of natural resources’ and the impact of institutions in overcoming that ‘curse’ in the African countries has still a lot of possibilities for future researchers to delve into. New control variables can be introduced such as domestic savings rates, following Ploef (2011), or ethnic and language fractionalization within countries, as in Mehlum, H., Moene, K. and Torvik, R. (2006b). New measurements for natural resource dependence may be computed, as in this study I use fuel dependence as a proxy for an overall dependence of natural resources, disregarding mineral and agricultural dependence. As stated before, using a more complex econometric model might give different results when testing the relationship between natural resource dependence and the quality of the institutions. Introducing lagged variables allows to capture prolonged effects of the dependence of natural resources and, also, of the evolution of the quality of institutions. Gathering more data regarding different types of natural resources may be beneficial for more robust results. And, finally, performing a critical analysis between countries that are known to have escaped the dependence of natural resources, such as Botswana, and other peer countries can help recommend policy implications to overcome low economic growth.

(32)

29

References

Auty, R. (1990). Resource-based industrialization. Oxford: Clarendon Press, p.200.

Acemoglu, D., Johnson, S. and Robinson, J. (2000). The Colonial Origins of Comparative Development: An Empirical Investigation. SSRN Electronic Journal.

Bertocchi, G. and Canova, F. (2002). Did colonization matter for growth?. European Economic Review, 46(10), pp.1851-1871.

Brunnschweiler, C. and Bulte, E. (2008). The resource curse revisited and revised: A tale of paradoxes and red herrings. Journal of Environmental Economics and Management, 55(3), pp.248-264.

CORDEN, W. (1984). Booming Sector And Dutch Disease Economics: Survey And Consolidation *. Oxford Economic Papers, 36(3), pp.359-380.

Davis, G. (1995). Learning to love the Dutch disease: Evidence from the mineral economies. World Development, 23(10), pp.1765-1779.

Efendic, A., Pugh, G. and Adnett, N. (2011). Institutions and economic performance: A meta-regression analysis. European Journal of Political Economy, 27(3), pp.586-599.

Gunton, T. (2003). Natural Resources and Regional Development: An Assessment of Dependency and Comparative Advantage Paradigms. Economic Geography, 79(1), pp.67-94.

Havranek, T., Horvath, R. and Zeynalov, A. (2016). Natural Resources and Economic Growth: A Meta-Analysis. World Development, 88, pp.134-151.

(33)

30 Mehlum, H., Moene, K. and Torvik, R. (2006a). Cursed by Resources or Institutions?. The World Economy, 29(8), pp.1117-1131.

Mehlum, H., Moene, K. and Torvik, R. (2006b). Institutions and the Resource Curse*. The Economic Journal, 116(508), pp.1-20.

Ploeg, F. (2011). Natural Resources: Curse or Blessing?. Journal of Economic Literature, 49(2), pp.366-420.

Prsgroup.com. (2018). [online] Available at: https://www.prsgroup.com/wp-content/uploads/2012/11/icrgmethodology.pdf [Accessed 18 May 2018].

Ross, M. (2001). Does Oil Hinder Democracy?. World Politics, 53(03), pp.325-361.

Sachs, J. and Warner, A. (1995). Natural Resource Abundance and Economic Growth. NBER, Working Paper no 5398.

Sachs, J. and Warner, A. (1997). Sources of Slow Growth in African Economies. Journal of African Economies, 6(3), pp.335-376.

Sachs, J. and Warner, A. (2001). The curse of natural resources. European Economic Review, 45(4-6), pp.827-838.

(34)

31

Appendix

Appendix Table A1: Countries included in the analysis

Country Name Country Code Country Name Country Code

Algeria DZA Malawi MWI

Angola AGO Mali MLI

Benin BEN Mauritania MRT

Botswana BWA Mauritius MUS

Burkina Faso BFA Morocco MAR

Burundi BDI Mozambique MOZ

Cabo Verde CPV Namibia NAM

Cameroon CMR Niger NER

Central African Republic CAF Nigeria NGA

Chad TCD Rwanda RWA

Comoros COM Senegal SEN

Congo, Dem. Rep. COD Seychelles SYC

Congo, Rep. COG Sierra Leone SLE

Cote d'Ivoire CIV South Africa ZAF

Djibouti DJI Sudan SDN

Egypt, Arab Rep. EGY Swaziland SWZ

Equatorial Guinea GNQ Tanzania TZA

Eritrea ERI Togo TGO

Ethiopia ETH Tunisia TUN

Gabon GAB Uganda UGA

Gambia, The GMB Zambia ZMB

Ghana GHA Zimbabwe ZWE

(35)

32 Appendix Table A2: Variables description

Variable Variable description Source

GDP Annual percentage growth rate of GDP per capita based on constant local currency, based on constant 2010 US dollars.

World Bank

Fuel Indicator variable if a country in a specific year is considered to be fuel dependent. It is composed by an average of oil, coal and natural gas rents as a percentage of GDP and the share of fuel exports in total exports.

Author (based on Auty, 1990; data from World Bank)

Institutions Computed arithmetic average of the indicators bureaucracy quality, corruption, law and order and investment profile.

Author (computed with the ICRG 2011 database)

Institutions_Fuel Interaction term between the variables

Institutions and Fuel.

Author

British Indicator variable for different origins of the metropolitan power; if a colony was under British influence, the dummy takes the value 1.

Author

Malaria Plasmodium Falciparum parasite rate (the proportion of persons positive for malaria infection among those examined) from 1970 until 2015.

Snow et al., (2017)

Human Capital Human capital index, based on years of schooling and returns on education, from 1970 until 2014.

Penn World Table

Exports_GDP Share of exports in GDP from 1970 until 2015 World Bank

GDP70 GDP per capita, based on constant 2010 US dollars.

World Bank

Landlocked Indicator variable if the country is completely landlocked, i.e., countries that are entirely surrounded by land.

Sachs and Warner (1997)

Investment growth rate

Annual growth rate of gross capital formation (formerly gross domestic investment). This metric includes additions to the fixed assets of the economy plus net changes in the level of

(36)

33

inventories. Fixed assets include land improvements (fences, ditches, drains, and so on); plant, machinery, and equipment purchases; and the construction of roads, railways, and the like, including schools, offices, hospitals, private residential dwellings, and commercial and industrial buildings. Inventories are stocks of goods held by firms to meet temporary or unexpected fluctuations in production or sales, and "work in progress." According to the 1993 SNA, net acquisitions of valuables are also considered capital formation.

WGI Computed arithmetic average of the indicators voice and accountability, political stability and absence of violence/terrorism, government effectiveness, regulatory quality, rule of law and control of corruption

World Bank

WGI_Mineral Interaction term between the variables WGI and

Mineral

Author

WGI_Fuel Interaction term between the variables WGI and

Fuel

Author

Natural_Resources Indicator variable that takes the value 1 if a country in a specific year is resource abundant; otherwise, it is 0.

Author

Institutions_Natural _Resources

Interaction term between the variables

Institutions and Natural_Resources

(37)

34 Appendix Table A3: Hausman Test

Coefficients (b) (B) (b-B) sqrt(diag(V_b-V_B)) FE RE Difference S.E. Fuel 8.733474 7.298955 1.43452 1.622231 Instit 2.102661 1.671793 0.4308678 0.7737616 Instit_Fuel -4.074832 -3.416194 -0.6586387 0.5731266 Exp_GDP 0.0879907 0.0319716 0.0560191 0.0282452 HC 5.847904 4.441589 1.406315 1.237066 Malaria -0.0000722 -0.0000619 -0.0000103 0.0000212 Inv_grate 0.0730172 0.0759844 -0.0029673 0.001153

b = consistent under Ho and Ha; obtained from xtreg B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test: Ho: difference in coefficients not systematic

chi2(7) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 8.73

Prob>chi2 = 0.1893

(38)

35 Appendix Table A4: Breusch and Pagan Lagrangian multiplier test for random

effects

Breusch and Pagan Lagrangian multiplier test for random effects

GDP[ID,t] = Xb + u[ID] + e[ID,t]

(39)

36 Appendix Table A5: Endogeneity tests

Test for endogeneity of Fuel

Instrumental variables (2SLS) regression Number of obs = 194 Wald chi2(5) = 66.86 Prob > chi2 = 0.0000 R-squared = 0.2552 Root MSE = 3.4448

GDP Coef. Std. Err. z P>|z| [95% Conf. Interval]

Fuel .5071962 1.006407 0.50 0.614 -1.465324 2.479717

Human Capital 1.114938 0,9225449 1.21 0.227 -0.6932165 2.923093

Malaria -0.0000403 0,000086 -0.47 0.639 -0.000209 0.0001283

GDP70 -0.000524 0.0002296 -2,28 0.022 -0.000974 -0.000074

Investment growth rate 0.0815399 0.0110104 7.41 0.000 0.05996 0.1031199

_cons -0.8609489 01.341606 -0.64 0.521 -3.490449 1.768551

Instrumented: Fuel

Instruments: Human Capital GDP70 Investment growth rate Export_GDP Landlocked Instit

Tests of endogeneity Ho: variables are exogenous

(40)

37 Test for endogeneity of Institutions

Instrumental variables (2SLS) regression Number of obs = 194 Wald chi2(6) = 66.73 Prob > chi2 = 0.0000 R-squared = 0.2538 Root MSE = 3.4479

GDP Coef. Std. Err. z P>|z| [95% Conf. Interval]

Institutions -0.4021809 1.113378 -0.36 0.718 -2.584362 1.780001 Exports_GDP 0.004861 0.030035 0.14 0.892 -0.547813 0.0629535 Human Capital 1.160369 1.051976 1.10 0.270 -0.9014662 3.222205 GDP70 -0.0004858 0.0002147 -2.26 0.024 -0.0009066 -0.0000651 Malaria -0.0000474 0.0000863 -0.55 0.583 -0.0002167 0.0001218 Investiment growth rate 0.0810242 0.113032 7.17 0.000 0.0588704 0.1031781 _cons 0.287333 2.692637 0.01 0.991 -5.248738 5.306204 Instrumented: Institutions

Instruments: Exports_GDP Human Capital GDP70 Malaria Investment growth rate Fuel British Landlocked

Tests of endogeneity Ho: variables are exogenous

(41)

38 Appendix Table A6: VIF test

Variable VIF 1/VIF

How institutions and natural resources influence the long-run economic growth in Africa