Foreign aid, institutions, and economic growth : evidence from sub-Saharan Africa

(1)

Foreign aid, institutions, and economic growth: evidence from

sub-Saharan Africa

Nathalie Duijvesteijn

6030297

nathalieduijvesteijn@gmail.com

Universiteit van Amsterdam

Thesis MSc Economics

Track: International Economics and Globalization

Final version

July 15, 2015

Supervisor: dr. M. Micevska Scharf

Second reader: dr. D.J.M. Veestraeten

(2)

Statement of Originality

This document is written by Student Nathalie Duijvesteijn who declares to take full responsibility for the contents of this document.

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

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

(3)

Abstract

Theoretically, foreign aid could have positive or negative effects on economic growth, or no effect at all. While some studies find evidence for foreign aid having the same adverse consequences as those predicted by the natural resource curse hypothesis, others find that this is only the case with poor institutions. This thesis empirically investigates how foreign aid relates to economic growth, and how institutional quality affects this relationship. A robust Hausman-Taylor estimator is used to obtain estimates of the varying variables of interest – foreign aid and institutional quality – while controlling for the time-invariant instrumental variables for institutional quality – geography and fractionalization measures – to account for endogeneity problems. The main findings show that foreign aid has a significant positive effect on economic growth, while the influence of institutions is not significant.

(4)

1. Introduction

‘Trade, not aid’ is regarded as an important part of development for the poorest nations as promoted by some nations. Other nations claim that this is an excuse for rich countries to cut back aid. Despite this, Official Development Assistance increased by 6.1% from 2012 to $134.8 billion in 2013, reaching a record high of foreign aid (Provost, 2014). While some nations use excuses like ‘trade, not aid’ to cut back on aid and others clearly see aid as a useful tool to reduce world poverty, as seen in the record levels of foreign aid, the effectiveness of foreign aid remains ambiguous. As Djankov, Montalvo and Reynal-Querol (2006) claim in their paper, foreign aid might be seen as a windfall of resources that may generate the same rent-seeking behaviour as in the curse of natural resources. This thesis seeks to analyze this resemblance, especially with respect to linking political institutions to resources being a blessing or a curse, and how this affects the relationship between foreign aid and economic growth. The focus will be on sub-Saharan Africa, because while many, if not all, of these countries are rich in natural resources and have received substantive amounts of foreign aid, still most of the poorest nations of the world are located in this region.

In 1993 Richard Auty came with his natural resource curse thesis, in which he states that revenues from natural resources have negative political and economic impact on the country rich in natural resources. Many countries in sub-Saharan Africa that are rich in natural resources, such as Congo, Chad and Sierra Leone, are still one of the most poor and underdeveloped countries in the world. In contrast, countries in South-East Asia that are mainly rocky islands or peninsulas, and therefore mainly scarce in natural resources, have been experiencing increased economic growth and development in the past few decades. However, the success story of Botswana, a landlocked sub-Saharan country abundant in natural resources like diamonds that transitioned from Least Developed Country (LDC) status to Middle Income Country (MIC) status in only three decades (United Nations Development Programme, 2012), indicates that natural resource abundance is not necessarily a curse. This success can be attributed to its good institutions, its mature democracy with regularly held elections, and the fact that among African countries Botswana has the best score on the Corruption Perceptions Index 2014 (Acemoglu, Johnson & Robinson, 2003). Also, despite the wealth from natural resources the exchange rate did not get overvalued and the government heavily invested in public goods like health, education, and infrastructure. Botswana’s economic growth and good governance track record is supported by prudent fiscal and macroeconomic management.

The long ongoing debate about the economic future of sub-Saharan Africa used to have two opponents: the pro-aid economist Jeffrey Sachs with his book The End of Poverty: Economic Possibilities

for Our Time (2005) and the aid-skeptic William Easterly with his book The Elusive Quest for Growth: Economists’ Adventures and Misadventures in the Tropics (2001). In 2007 Paul Collier entered as a

middle man in this debate, with his book The Bottom Billion, which is in line with Sachs’ argument that rich countries can really do something for Africa and in line with Easterly’s argument that foreign financial aid alone is not enough. He argues that there are four traps in which poor countries tend to fall.

(6)

The first is the trap of civil conflict, in which he states that poorer countries are more likely to fall into. With his statement “diamonds are a guerrilla’s best friend”, Collier refers to the second trap that entails the curse of natural resources. The third is the trap of landlocked countries with bad neighbors, which makes poor countries economically handicapped. The last is the trap of bad governance, in which he argues that failure of foreign aid effectiveness cannot be blamed on Western imperialists, but stems from weak political institutions in the recipient country. It is the second one, the natural resource curse, to which much attention has been paid by development economists. Researchers have also been very interested in studying the issues of governance causing the fourth poverty trap. More interesting, however, is to combine the second and the fourth to examine the relation between a resource curse and political institutions. Sala-i-Martin and Subramanian (2003) already made such a combination and found indeed that bad governance plays a significant role in the negative impact of natural resources on economic growth. Therefore, this thesis will partly use the framework provided by Sala-i-Martin and Subramanian (2003) in testing to what extent institutional quality matters for the effectiveness of foreign aid.

Much research has been done about the natural resource curse since it was first hypothesized by Auty in 1993, and most studies give evidence for its existence. More recently, the literature examines the similarity between revenues from natural resources and foreign aid. Especially the use of grants as a form of foreign aid can be seen as a similar kind of income to the rents received from natural resources, since grants, unlike loans, do not come with the obligation for repayment. This creates an interesting topic for research on the increased use of grants since the turn of the millennium and the possible negative effects from a political-economy point of view, since an increased inflow of revenues, as in the natural resource curse, might create incentives for corrupt governments to engage in rent-seeking activities rather than using it effectively.

The natural resource curse works through different causal mechanisms: (i) appreciation of the real exchange rate, (ii) rent-seeking behaviour and decreased democratization, and (iii) volatility of revenues from resources causing problems for fiscal policy (Morrison, 2012). As this thesis examines the effects of foreign aid similar to the resource curse combined with the institutional environment, the focus will be on the second causal mechanism in which rent seeking and corruption endangers the effectiveness of foreign aid. The main research question of this thesis is therefore:

How is the relationship between foreign aid and economic growth affected by the quality of institutions in the recipient nation?

Panel data is used for IV regression analysis, in which the effect of foreign aid on economic development will be examined with ethnic and language fractionalization and distance from the equator as instrumental variables for institutional quality. Because these instrumental variables are time-invariant and therefore drop out when using fixed-effects regression, this study uses the Hausman-Taylor estimator to obtain estimates for both the time-varying and time-invariant variables. In order to make this

(7)

Hausman-Taylor estimator robust, two modifications are made replacing ordinary least squares estimators with MM-estimators. The panel data consists of data for the 34 countries in sub-Saharan Africa that are eligible for funding from the International Development Association (IDA), for a time span of nine years between 2005 and 2014. For those years the World Bank has scores of its Country Policy and Institutional Assessment (CPIA), which captures the quality of a country’s political institutions and is therefore a good indicator for a country’s ability to effectively use development assistance. Official Development Assistance (ODA) includes all transfers from official sources with a grant element of at least 25% and is used as measure for foreign aid. With these measures for foreign aid and institutions, and GDP per capita growth as measure for economic growth, the main findings show that foreign aid is significantly and positively related to economic growth, and that institutional quality, although not significant, positively affects this relationship.

Chapter 2 will provide an explanation, based on the existing literature, of how abundance of natural resources in a country can lead to slower economic growth and that this is therefore seen as a curse. Some examples will elaborate on whether such curse of natural resources actually exists and what factors play a role in it. In the rest of Chapter 2 the link between the natural resource curse and foreign aid is made, and analyzed in the context of political economy. Chapter 3 explains the empirical model and describes the data used in this model. It continues with a discussion of the obtained results and ends with some robustness checks. Limitations of this study and recommendations for future research are given in Chapter 4, and Chapter 5 ends this thesis with a conclusion.

(8)

2. Literature review

This chapter starts with analyzing the existing literature on the natural resource curse, as this forms the basis from which the link towards similar effects from foreign aid is made. The following section covers the similarities between natural resource rents and foreign aid, and how these affect economic growth. The last part of this literature review covers the existing literature on how institutional quality can work as a causal mechanism between foreign aid and its effectiveness on economic development.

2.1 The curse of natural resources

According to the resource curse thesis (Auty, 1993), revenues from natural resources have negative effects on economic growth and development in resource abundant countries. Sachs and Warner (1995) found evidence for natural-resource-scarce countries growing faster than countries with abundant natural resources. The negative effects can be linked to the resources through several causal mechanisms (Morrison, 2012). The main mechanism, often called the Dutch disease, is the appreciation of the real exchange rate after discovery of natural resources, which leads to a decline in the tradables sector. The Economist (1977, November 26) first mentioned the term Dutch Disease in an article about the struggles of Dutch manufacturers and other exporters after the wake of the discovery and exploitation of natural gas in the Netherlands in the 1970s. Another mechanism is the volatility of revenues from resources causing a problem for fiscal policy. This problem entails the diminishing marginal benefits to social spending, causing that the social gain in spending more in some years does not make up for the social cost of spending less in other years. The last causal mechanism Morrison (2012) mentions is political deterioration – the negative impact on political regimes – meaning that natural resource rents can lead to more corruption, weaker accountability, and less democratization. Some other mechanisms through which natural resource abundance can negatively impact economic development are crowding out of other sectors such as manufacturing, the declining long term trend in world commodity prices, and mineral riches causing a civil war, which is bad for development (Frankel, 2010). These causal mechanisms link the resources to the negative effects as stated in the natural resource curse, but Morrison (2012) argues that those negative effects heavily depend on the institutional environment in place in the countries in question: no negative effects are present in beneficial institutional environments, whereas they are present in countries with bad institutions.

In line with the theory that negative effects caused by resource revenues depend on the institutional environment in place, more research has been done about the natural resource curse combined with the literature on how institutional quality plays a role. Brunnschweiler (2008) examines the relationship between good quality institutions and economic growth in the perspective of the natural resource curse. The estimation results show significant positive outcomes for institutional quality, indicating that institutions matter. Also a study by Mehlum, Moene, and Torvik (2006) confirms that natural resources are only a curse in countries with bad institutions. They made a distinction between producer friendly

(9)

institutions and grabber friendly institutions, in which production and rent seeking are either complementary or competing activities, respectively.

2.1.1 Natural resources, institutions, and growth

In an attempt to address the natural resource curse and its effect in Nigeria, Sala-i-Martin and Subramanian (2003) measured the effect of natural resources on economic growth with instrumental variables regression, taking account for the endogeneity of institutional quality in the effect of natural resources on economic growth. The main instruments used for institutional quality are the fractions of the population speaking English or Western European languages, but also settler mortality rates can be and are used as instrumental variable for institutional quality in the first stage regression. They found that natural resources do not have a significant direct impact on economic growth, but that indirectly through institutional quality the effect is significantly negative. In line with Morrison (2012) and Brunnschweiler (2008), these results show that the quality of institutions is decisive in determining whether natural resources are a curse or not. Bulte, Damania, and Deacon (2005) examine the effect of resources on human development and find, similar to the results of Sala-i-Martin and Subramanian (2003), that there is only an indirect link between resources and human welfare through institutional quality.

Wilson (2013) examines the natural resource curse hypothesis for another particular country, Sierra Leone, for a long timespan, 1930 to 2010, and finds that diamond exploitation has shifted between a resource blessing and a curse in different time periods. In the colonial era and early post-colonial era (until 1967), diamond exploitation was a resource blessing and contributed to economic development. The role of diamonds in economic development declined during the All People’s Congress (APC) era (1968 to 1992), mostly due to patrimonial politics issues, informal diamond mining and smuggling, and corruption, changing diamond exploitation into a resource curse. In the next era diamond exploitation fuelled and prolonged the Sierra Leone civil war, leading to the term ‘conflict diamonds’ that was used by, among others, Paul Collier (2007) when referring to the natural resource curse as one of his four poverty traps. In the post-war era, however, improved governance of the diamond sector led to increased economic development, changing the exploitation of diamonds back to being a resource blessing (Wilson, 2013). It thus can be concluded that the quality of political institutions in the different time periods plays a significant role in determining when diamonds can be used as a curse or a blessing for Sierra Leone: beneficial governance from colonial ties in the colonial era and in the post-war era causes diamond exploitation to be a blessing and improves economic development, while bad governance during the APC era and the following civil war causes diamonds to be a curse.

Not only is institutional capacity to handle sudden windfalls of resource revenues a determining factor of economic growth since the commodity shocks of the 1970s and 1980s, but also institutional capacity itself varies across economies with different sources of export revenue and the export structures influence socioeconomic and political institutions (Isham, Woolcock, Pritchett & Busby 2003). There is

(10)

considerable evidence of rent-seeking behaviour in developing economies, however not in all countries. Baland and Francois (2000) create a model of rent seeking versus entrepreneurship with increased rents caused by e.g. resource booms, trying to solve the differences between economies. They argue that the larger is the number of individuals engaging in rent-seeking activities, the lower are the returns on both rent seeking and entrepreneurship. This possibly increases the marginal effects in production, leading to an economy in which multiple equilibria are possible.

Since the results of Sala-i-Martin and Subramanian (2003) show that resource rents do not have a direct impact on economic growth, this might be seen as a lack of proof for the natural resource curse. Haber and Menaldo (2011) also do not find evidence for the curse when they try to identify any causal relationship between polity and fiscal reliance. Here polity is a standard measure of democracy employed in the resource curse literature and the measure of fiscal reliance on resource revenues is the percentage of government revenues from oil, gas, and/or minerals. The results imply that, no matter how one looks at the relationship between polity and fiscal reliance, there is no evidence for a resource curse. However, more often than not the resource curse has been proven to exist, especially when institutional quality is accounted for.

2.1.2 Political economy and natural resource revenues

As analyzed so far, many studies suggest and confirm that political institutions are decisive in determining when natural resources are a curse or a blessing. It is, however, also interesting to analyze the reversal, thus how political institutions are affected by natural resource rents. Several possible channels through which oil affects political outcomes are rentier effects, delayed modernization effects, and an entrenched inequality effect (Isham et al., 2003). Rentier effects arise when revenues can easily be extracted from a few sources, such that the state has less incentive for taxation and the population less need for civil society. The government can then use the ‘exogenous’ revenues to mollify dissent, and the state has resources to pursue direct repression and violence against dissenters. Delayed modernization is caused by settlers that in high mortality environments concentrated only on rent extraction from high value added products and hence did not invest in the development of high quality government institutions (Acemoglu, Johnson & Robinson, 2001). Also, resource abundance simultaneously strengthens states and weakens societies, and thus yields low levels of development. In areas of geographic space, production is conducive to large-scale production, where relationships tend to bind each person to a social superior. This economic structure is conducive to bad politics and bad governance.

Acknowledging that previous research mainly focused on explaining the resource curse and ways to avoid it, Cabrales and Hauk (2010) incorporate the role of government by building an explicitly political model to explain when natural resources are a blessing and when they are a curse. They do so by including human capital accumulation, as this is widely accepted to be an engine of economic growth, and by allowing simultaneously for elections, political competition, and revolution. The model predicts that

(11)

natural resources are detrimental to human capital accumulation with bad institutions, while no resource curse exists in democracies with good institutions. Also, natural resource abundance harms democracy by making political turnover, i.e. leadership alternation, less likely, especially with bad institutions. In a country where institutions are weak and the opposition cannot manage natural resource revenues as efficiently as the incumbent government even for high levels of education, finding more natural resources increases the chances that the incumbent stays in power. The last prediction this model makes is that the threat of revolution diminishes with good institutions: as good institutions face political competition that forces them to choose to spend the revenues on subsidies for human capital, this creates positive spillovers on society, and thus reduces society’s need for revolution (Cabrales & Hauk, 2010).

Ades and Di Tella (1999) focus more on the causes of corruption and find evidence that both natural rents, such as those from oil, and rents induced by the lack of competition in product markets increase the level of corruption. They examine this more at a business/firm level, but the mechanism through which rents from any source affect corruption levels must be taken into account when examining the effect of foreign aid on political institutions, since higher levels of corruption have been found to lower investment and thereby lowering economic growth (Mauro, 1995).

2.2 Foreign aid

The causal mechanisms linking natural resource rents to the negative effects they might generate as described by Morrison (2012) can work in the same way with revenues from foreign aid. Indeed, foreign aid can be seen as a windfall of resources that may cause the same poor economic performance and rent-seeking behaviour as in the resource curse (Djankov et al., 2006). The political-economy model built for explaining whether natural resources can be seen as a curse or a blessing (Cabrales & Hauk, 2010) can be used to make predictions about the effects of foreign aid on human capital and economic growth.

There are several mechanisms through which aid can enhance economic growth (Boone, 1996). One of them is that capital aid flows can be used for profitable investment projects that are otherwise not undertaken due to the lack of domestic savings in poor countries. Another works through fiscal policy, where aid flows increase revenues for the government and therefore reduces the need for distortionary taxes. This depends, however, on the political regime in the recipient nation, and most critics of aid programs claim that many politicians do not allocate aid as efficient as aimed for with the goals of such programs. This section first analyzes the existing literature about the direct effect through which foreign aid can enhance economic growth. It then continues with how foreign aid can indirectly, through political institutions, affect economic development.

2.2.1 Foreign aid and economic growth

Two mechanisms are often cited through which foreign aid can affect growth (Wright & Winters, 2010). The first one argues that aid can increase capital spending, assuming that the government spends the

(12)

received aid on capital spending and that capital spending promotes growth. In this mechanism political institutions remain exogenous, meaning that aid does not endogenously affect growth through its effect on institutions. For example, Burnside and Dollar (2000) treat institutions as conditional but exogenous, stating that the correct macroeconomic policy is sufficient to spur economic growth. The main flaws of this mechanism are the assumptions made about the role of politics: recipient countries are often governed by corrupt politicians who rather spend the received aid on personal interests instead of on capital spending, and capital spending in itself does not always promote growth. Although Burnside and Dollar (2000) treat institutions as exogenous, their finding that aid increases growth in countries with good institutions but has no effect when bad institutions are in place, has been very influential in policy and academic debates. Good institutions are defined as institutions with sound fiscal, monetary, and trade policies. International aid agencies recognized these findings as important, since they suggest that foreign aid can promote economic growth, and that aid should be selectively distributed to countries with good policies (Easterly, Levine & Roodman, 2004). However, as Easterly et al. (2004) note in their comment on Burnside and Dollar’s paper, the effectiveness of foreign aid in countries with good political institutions should be reconsidered and reexamined with more recent data. Aid effectiveness might not only exogenously depend on political institutions, but aid might also reform policies towards more growth promoting policies.

The second mechanism treats political institutions as endogenous factor in the effect of aid on growth, a relationship that works in two ways. First, inflow of aid creates incentives for increased rent-seeking behaviour, suggesting that aid can be seen as a curse similar to the natural resource curse (Djankov et al., 2006; Moyo, 2009). Second, donors of foreign aid can use aid as a tool for economic or political reform by threatening to withhold foreign aid unless recipient governments pursue the desired reforms (Svensson, 2000). This aid conditionality should therefore stimulate political institutions in their incentive to improve economic growth. In line with this, Sachs, McArthur, Schmidt-Traub, Kruk, Bahadur, Faye, and McCord (2004) argue that unconstrained aid may induce public consumption rather than investment, and this would negatively affect economic growth. It is suggested, therefore, that conditions should be attached for the way of spending the foreign aid when aid is disbursed in the form of grants. On the other hand, Morrissey, Islei and M’Amanja (2006) examine whether concessional loans and grants have different effects on GDP growth and tax revenues, and find rather weak evidence for the negative impact of grants.

Slow growth in developing economies is commonly associated with the absence of strong political and legal institutions, and the presence of multiple powerful ethnic groups in society. Tornell and Lane (1999) present a dynamic model of the economic growth process containing these features. When there is a windfall of resources, like increased foreign aid transfers, these multiple power groups extract the transfers from the government. The government has to increase tax rates to finance the more than proportional increase in redistributive transfers, causing firms to move to the shadow sector, where productivity is low and no taxes are paid, and thus declining the growth rate of the economy. Also, when

(13)

power groups have access to the transfers, they can increase their consumption by performing rent-seeking activities to appropriate these transfers. Due to these rent-seeking activities, an increase in foreign aid transfers does not necessarily lead to increased economic growth for the recipient nation (Svensson, 2000).

2.2.2 Foreign aid and political economy

Already before the notion of the natural resource curse, Grossman (1992) came with a game theory model on how foreign aid causes a reallocation of resources from production to an intensified struggle over distributive shares, explaining the mechanism through which foreign aid can induce rent-seeking behaviour. He argues that when a country receives more foreign aid, the ruler of that country faces a tradeoff between distributing the extra resources among the ruler’s clientele, or using it for good policy in order to keep the population from insurrection. Here insurrection is defined as a generic term for any activity that challenges the established system of property rights and taxation, and thus forms a threat for the incumbent government. This model shows some resemblance with the political model conducted by Cabrales and Hauk (2010), in which natural resources being a curse or a blessing depend on the behaviour of politicians and their incentives to pursue good policies.

Institutions are the humanly devised constraints that structure political, economic and social interaction (North, 1991). They consist of both informal constraints (sanctions, taboos, customs, traditions, and codes of conduct), and formal rules (constitutions, laws, property rights). Glaeser, La Porta, Lopez-de-Silanes, and Shleifer (2004) examine the relationship between institutions and economic growth. They find that human capital is a more basic source of growth than are institutions, that poor countries grow out of poverty through good policies, often pursued by dictators, and subsequently improve their political institutions.

Already in 1996, Boone examines the effectiveness of foreign aid while taking account of politics, but rather than testing aid’s impact on economic growth, he studies how aid is used. Besides this main difference in his study and this thesis, Boone also makes two interesting assumptions. The first assumption is that aid is fungible, meaning that the recipient government can allocate the funds as needed. The second assumption is that aid is not conditional, meaning that political regimes in the recipient nation and policy choices are not directly affected by aid flows. The main finding is that aid flows are almost all spend on consumption, of which around a quarter went to private consumption and three quarters to public consumption, and not on public or private investment. Also, the size of the government increases due to the increase in public consumption, and aid has no significant impact on poverty indicators, such as infant mortality, primary schooling, and life expectancy. When using instruments for aid, acknowledging that aid flows can be endogenously determined by for instance political determinants and population size, the results do not significantly change. Only when small countries are included that receive aid for large investment projects (>50% of GNP), which are mostly not fungible, the impact of aid on investment

(14)

becomes significantly larger than zero. The lack of impact on infant mortality and other human development indicators proves the failure of government in the recipient countries. Governments could improve human development indicators either directly by spending aid flows on increased consumption for its poorest citizens, or indirectly by increasing the provision of public services that promote the well being of the poor. As in his study different political regimes are pooled together, Boone (1996) also tests for possible differences in how aid is used under different political regimes, but finds that the results are not significantly different. However, the impact on human development indicators becomes significant: liberal political regimes use received aid to lower infant mortality rates by roughly 30%.

The causal mechanisms through which resources can cause adverse consequences for economic development might be the reason for the fact that some studies, like the one by Boone (1996), find no effectiveness of foreign aid, and that others find a positive effect, but only when governments have sound policies. Boone (1996) tests the effectiveness of aid with a linear regression model, while others model the aid-growth relationship as non-linear, either by including aid squared as regressor (Lensink & White, 1990), or by including an interaction term between aid and policy (Burnside & Dollar, 2000).

Djankov et al. (2006) examine the relationship between foreign aid and institutions, and show that foreign aid negatively affects democracy in developing countries and reduces investment, while increasing consumption, causing lower economic growth. This leads to the hypothesis that easy resources from foreign aid may induce corruption and rent-seeking behaviour among parties in power. They use ODA as a measure for foreign aid, so all transfers from official sources with a grant element of at least 25%, but a distinction is made between the grant and loan components. The results show that aid has a positive effect on investment when the ratio of grants to ODA is small enough, indicating that ODA given in the form of loans increases investment. When the ratio of grants to ODA becomes large enough, ODA will negatively affect investment, suggesting that the shift from loans to grants actually decreases the effectiveness of foreign aid on economic development.

Djankov et al. (2006) make this distinction in the ratio of grants to ODA, comparable to the distinction between IDA loans and grants, because the discussion on loans versus grants saw a re-awakening at the turn of the millennium. The publication of the report by the International Financial Institution Advisory Commission (IFIAC) in 2000, the so-called Meltzer-report, which suggested increasing the use of outright grants of the IDA to the poorest recipient nations, contributed largely to this re-awakening of the discussion. It is important to look at the way in which aid is disbursed, but the disbursement of foreign aid is mostly excluded from research on the effectiveness of aid on development. One of the recommendations in the UN Millennium Project (2005) is to provide grants rather than loans and that debt sustainability should be redefined as “the level of debt consistent with achieving the Millennium Development Goals”. However, the lack of evidence in the favour of grants made Djankov et al. (2006) question this recommendation and also why the G-7 and Sachs et al. (2004) called for an increased use of grants rather than loans.

(15)

Gupta, Clements, Pivovarsky, and Tiongson (2003) examine whether the impact of aid on the revenue effort of recipient countries depends on the composition of aid and find that only grants (and not loans) adversely affect revenue collection. This is in contrast with the results from Morrissey et al. (2006), who find that the negative impact of grants was rather weak.

As Morrison (2012) argues that the negative effects of natural resource rents depend on the institutional environment in place, and as foreign aid might generate the same negative effects, it can be argued that institutions also matter for the effectiveness of foreign aid in a recipient nation. In line with this, Dutta, Leeson and Williamson (2013) compare the pessimistic view on foreign aid, which states that more aid leads to more dictatorial institutions, with the optimistic view, which states that aid can make institutions in developing countries more democratic. Hypothesized is that foreign aid has a more modest impact on political institutions, namely that aid amplifies recipients’ existing institutional orientations. Put simply, this means that aid makes dictatorships more dictatorial and democracies more democratic, which is then called the amplification effect. The results show that both the pessimistic and optimistic views on foreign aid ascribe too much power to aid’s ability to influence recipients’ political institutions: aid does not alter their institutional orientations, it just amplifies their existing ones.

Also in the literature on foreign aid causing a resource curse, some studies do not find evidence for such a curse. One study considers the causality underlying the political aid-curse (Altincekic & Bearce, 2014). The results show that foreign aid does not cause a political curse, because it should not be seen as a revenue source that can be used for other expenses than those intended by foreign donors, relating to the infungibility of aid. However, as Altincekic and Bearce (2014) already note, their results come from a large sample of potential aid recipients and can therefore differ among individual countries with different policy regimes.

The increased amount of academic literature about the role of institutions and policies in the effectiveness of aid on growth has had its impact on international aid agencies. The view of aid agencies has changed over time towards increased awareness of which policies and institutions are truly improving development (Easterly, 2007). As noted before, the Burnside and Dollar paper (2000) was of great influence for international aid agencies in the decision to increase foreign aid flows to developing nations. At the same time, the United Nations, the IMF, the World Bank, and national aid agencies initiated the Millennium Development Project in order to reduce poverty rates, infant mortality, and other indicators of low development by the year 2015. Now the deadline of reaching the Millennium Development Goals in 2015 is approaching fast, the international aid agencies propose the global Sustainable Development Goals (SDGs) that should guide policy and funding for the next 15 years. However, in order for the SDGs to work, the effectiveness of aid should become less ambiguous. The next chapter seeks to contribute to this by empirically analyze the effect of foreign aid on economic growth, while controlling for aid’s effect on institutional quality.

(16)

3. Data and empirical analysis

This chapter starts with describing the data and model that are used in testing whether foreign aid has an effect on economic growth and how the quality of political institutions affects this relationship. It then continues with the results obtained from the estimation procedure and a discussion about whether these results are in line with expectations from the existing literature. In order to test whether the results are robust, some robustness checks follow the results section.

3.1 Methodology

As discussed before, several mechanisms through which resource revenues can have an effect on economic growth have been identified in the literature: (i) appreciation of the real exchange rate, (ii) rent-seeking behaviour and decreased democratization, and (iii) volatility of revenues from resources causing problems for fiscal policy (Morrison, 2012). This thesis examines only how the relationship between economic growth and foreign aid can be affected by the second mechanism of rent seeking and corruption. While Sala-i-Martin and Subramanian (2003) capture all three effects in their empirical specification for testing the relationship between a natural resource curse and political institutions, a simplified version of their framework is used as the baseline model in this thesis. To allow for control for unobserved heterogeneity across countries, panel data is used instead of cross-sectional data. The simple version of the baseline model captures the effects of foreign aid and institutional quality on economic growth for country

i in year t:

Growthit = α + β1InstitutionalQualityit + β2Aidit + εit (1)

where Growthit is the growth rate of GDP per capita, Institutional Qualityit is measured by the scores of

the Country Policy and Institutional Assessment (CPIA), Aidit is net official development assistance

(ODA) per capita, and εit is the random error term.

3.1.1 Endogeneity problems

Several issues in this simple specification arise. As aid may flow to countries whose economic growth rates are getting worse, this might cause endogeneity problems for the aid variable (Burnside & Dollar, 2000; Easterly et al., 2004). A simple way to avoid the endogeneity of aid is to use the lag of aid, because the growth rate of GDP per capita in year t does not affect the inflows of ODA in year t-1. Another way of controlling for the endogeneity of aid is to use instrumental variables. Boone (1996) uses several instruments for aid, such as the logarithm of population (small economies tend to get more aid than large ones due to political and structural reasons), twice-lagged aid (which is uncorrelated with emergencies and business cycle factors), and political determinants of foreign aid flows (bilateral strategic relationships between particular countries). The results of his study when using instruments for aid do not differ

(17)

significantly from the results when no instruments are used, suggesting that using lagged aid instead of using instruments is sufficient enough to control for endogeneity of aid.

A bigger issue in the simple model specification above is that institutions are endogenous to economic performance, because causality runs both ways, causing omitted variable bias. Richer countries typically choose or can afford to install better institutions, causing endogeneity of institutional quality in the simple model (Acemoglu et al., 2001). Endogeneity of institutions can also be explained by policy endogeneity, where agents understand that different policies will map into different outcomes: “If state policy making is a purposeful action, responsive to economic and political conditions within the state, then it may be necessary to identify and control for the forces that lead policies to change if one wishes to obtain unbiased estimates of a policy’s incidence” (Besley & Case, 2000, p.672). Since agents understand that different institutions lead to different policies and outcomes, the political economy approach suggests that the same reasons for not treating policies as exogenous are relevant for institutions. A possible solution for endogeneity of institutions is to extend the simple model to an instrumental variables (IV) regression model and use instruments for institutional quality. The instruments must be correlated with the CPIA scores, but not with economic growth and foreign aid. This is in line with previous studies where instruments were used to control for endogeneity of institutions (e.g. Acemoglu et al., 2001; Hall & Jones, 1999).

3.1.2 Instruments for institutional quality

Appropriate instruments for institutional quality can be found when looking at several centuries of world history. During the ‘Scramble for Africa’ between 1881 and 1914 African territory was colonized by European powers, all bringing different kinds of influence to the continent. Acemoglu et al. (2001) claim that not the identity of the colonizer matters, but the conditions in the colonies. There were different types of colonization policies, leading to different sets of institutions. Colonies were either extractive states, where no protection of private property and no checks and balances against government existed, or neo-Europes, like New Zealand and Australia, where protection of private property needed to be introduced. Also the feasibility of settlements affected the probability of different sets of institutions, and the early institutions affect the institutions in place today. Therefore, Acemoglu et al. use the mortality rate of colonial settlers, called settler mortality, as instrument for institutional quality. Although it seems logical that the colonizers’ mortality experience determined their colonization strategy and influenced the creation of either weak or strong institutions, the study by Acemoglu et al. started a settler mortality debate going back and forth between Acemoglu et al. (2001; 2005; 2006) and Albouy (2004; 2006; 2008). Albouy’s biggest critique on the Acemoglu et al. (2001) paper is that it contains many inconsistencies in how the dataset is put together and that European settler mortality is a weak instrument for institutional quality. In reaction to this debate, Subramanian (2007) comes with alternative/better measures of settler mortality. However, the most complete available dataset on settler mortality rates, whether or not it is accepted as a

(18)

valid instrument, only has data for 16 of the 34 IDA eligible countries in sub-Saharan Africa (see Appendix A). Therefore, another instrument must be found.

In line with the theory that early institutions created by colonial settlers affect current institutions in the former colonies, Hall and Jones (1999) use various correlates of the extent of Western European influence as instruments for institutions. They use characteristics of geography such as distance from the equator, because economies with a greater distance to the equator have more similar climates to Western European countries and are in general more successful in terms of per capita income. Also the extent to which primary Western European languages are spoken as first languages today serves as a proxy for colonial settler influence and is used as instrument for institutional quality. Since the dataset of Hall and Jones is somewhat outdated (though this does not matter for distance to the equator) new data sources have to be used for the fraction of population speaking English or other Western European languages.

Relating to the fraction of population speaking primary Western European languages, an important determinant of the quality of government is ethnolinguistic fractionalization (La Porta, Lopez-de-Silanes, Shleifer & Vishny, 1999). La Porta et al. (1999) find that countries that are more ethnolinguistically homogeneous have better governments than ethnolinguistically heterogeneous countries. The theory behind this finding is that in more fragmented societies, one ethnic group can impose restrictions on political liberty to impose control on the other ethnic groups. Alesina et al. (2003) also find evidence for this theory, namely that the democracy index is inversely related to ethnic fractionalization. This is consistent with the fact that Canada, New Zealand, Australia and the U.S., countries that were relatively homogeneous settler colonies, did a better job in establishing democratic institutions than more ethnically diverse former colonies in Africa. Not only ethnic fractionalization, but also language differences can affect efficient interaction and coordination in networks of production, trade and knowledge (Desmet et al., 2012). Although ethnic and language fractionalization might change over long time periods and therefore lead to endogeneity and invalidity of this instrument, such change is unlikely to have significant impact in the time period studied in this thesis. Therefore, next to distance to the equator, ethnic and language fractionalizations are used as instruments for institutional quality.

3.1.3 Adjusted empirical model specification

Controlling for endogeneity of some variables in the baseline model leads to the following specification of the IV regression model:

Growthit = α + β1InstitutionalQualityit + β2Aidit-1 + εit (2)

as the second stage regression, which is the same equation as equation (1), but now using the one-period lag of Aidi ; and

(19)

as the first stage regression for institutional quality, where EFraci is the measure of ethnic

fractionalization, LFraci is the measure of language fractionalization, Disti is the measure of distance to

the equator, and ηi is the error term.

Estimation of the above model using panel data brings some difficulties, since most of the variables on the right-hand side in equation (3) are time-invariant (Jacob & Osang, 2011). The first-difference estimator usually used to address omitted variables problems in panel data removes country-specific effects, but it would also cause all time-invariant variables such as geography measures and other instruments for institutions to be excluded from the regression. When first-differencing in the fixed-effects models cannot be used, one can use the random effects model. However, this model assumes independence between the explanatory variables and individual error terms, something that is often not the case in economic applications and is unlikely to hold in the model tested here. The estimation method proposed by Hausman and Taylor (1981) offers a solution to these issues, because it allows to obtain estimates of the observed time-invariant variables and to control for unobserved, time-invariant country-specific effects.

3.1.4 The Hausman-Taylor estimator

The Hausman and Taylor (1981) estimator can be represented by the following specification:

yit = β1X1it + β2X2it + γ1Z1i + γ2Z2i + µi + εit (4)

where X1it is a vector of exogenous, time-varying variables assumed to be uncorrelated with µi and εit; X2it

is a vector of endogenous, time-varying variables assumed to be possibly correlated with µi but orthogonal

to εit; Z1i is a vector of exogenous, time-invariant variables assumed to be uncorrelated with µi and εit; and

Z2i is a vector of endogenous, time-invariant variables assumed to be possibly correlated with µi but

orthogonal to εit. Further, µi is the unobserved, panel-level random error term, and εit is the idiosyncratic

error term. Due to the possible correlation of X2it and Z2i with µi, the simple random-effects estimators

would lead to biased results. By mean-differencing the data before estimating β1 and β2, the within

estimator using fixed effects regression removes the unobserved panel-level random error term µi but it

also eliminates the time-invariant vectors Z1i and Z2i. The Hausman and Taylor (1981) estimation method

resolves this problem.

First, estimation using fixed effects regression consistently estimates β1 and β2, but eliminates Z1i

and Z2i. Then the within residuals (𝑑!) obtained from this regresion are regressed on Z1i and Z2i to obtain

intermediate, but consistent, estimates of γ1 and γ2, called 𝛾! and 𝛾!, using X1it and Z1i as instruments. Sets

of within and overall residuals can be obtained by using the within estimates of β1 and β2 and the

intermediate estimates 𝛾! and 𝛾!. The variance components estimated using these sets can then be used for generalized least squares (GLS) transformation on each of the variables. When using 𝑿!!" as notation for the GLS transform of X1it, 𝑿!! as notation for the within-panel mean of X1it, and 𝑿!!" as notation for the

(20)

within transform of X1it and use similar notations for the other variables X2it, Z1i, Z2i, µi and εit, the

Hausman and Taylor (1981) estimator can obtain the coefficients of interest with the following instrumental variables regression:

𝑦!" = 𝛽!𝑿!!"+ 𝛽𝟐𝑿!!"+ 𝛾𝟏𝒁!!+ 𝛾𝟐𝒁!!+ 𝜇𝒊+ 𝜖𝒊𝒕 (5) where 𝑿!!", 𝑿!!", 𝑿!!, 𝑿!! and Z1i are used as instruments. The identifying condition for the Hausman and

Taylor (1981) model is that the number of X1 variables is at least as large as the number of Z2 variables.

The steps described above can be found in formula form in the first part of Figure 1 in Appendix A. In this figure the notation for the GLS transform is with an asterix (yit*) instead of a breve (𝑦!"), because this figure is taken from the paper by Baltagi and Bresson (2012).

3.1.5 Robust Hausman-Taylor estimator

The Hausman-Taylor estimator as first proposed in 1981 is a bit outdated and has received some criticism over the past few decades. The main shortcoming of this model is that it uses ordinary least squares (LS) estimators in the different steps, which in the presence of outlying observations leads to unreliable inference due to their sum of squared residuals. Because the LS-estimator awards huge importance to observations with large residuals, this distorts the estimation when outliers are present. Three types of outliers influence the LS-estimator in regression analysis: good leverage points, bad leverage points, and vertical outliers. To control for vertical outliers, an M-estimator can be used that has high Gaussian efficiency and is robust for vertical outliers, but not for bad leverage points. Another way to obtain robust estimates is to use an S-estimator, which resists to a contamination of up to 50% of outliers (also called having a high breakdown point of 50%), but has much lower Gaussian efficiency (Verardi & Croux, 2009).

In order to obtain a robust Hausman-Taylor estimator, Baltagi and Bresson (2012) propose to use two MS-estimators instead of ordinary least squares (LS) estimators as used in the ‘classic’ Hausman-Taylor estimator. An MS-estimator alternates an S-estimator (for continuous variables) and an M-estimator (for dummy variables) until convergence. The ordinary LS-M-estimator in the first part of step 1 to obtain the within residuals (𝑑_!) should be replaced with the robust within MS-estimator as proposed by Bramati and Croux (2007), which leads to a very small efficiency loss in the absence of outliers but yields large gains in the presence of outliers. The second part of step 1 of the ‘classic’ Hausman-Taylor estimator, the 2SLS-estimator, should be replaced with the two stage generalized MS-estimator (2SGMS) inspired from Wagenvoort and Waldmann (2002). These two modifications of the ‘classic’ Hausman-Taylor estimator are visible in the second part of Figure 1 in Appendix A.

An MS-estimator can be particularly helpful in fixed-effects panel data models (Bramati & Croux, 2007), and would therefore be a good alternative of the LS-estimators in the ‘classic’ Hausman-Taylor estimator. However, since this study does not have dummy variables specified, another alternative is used.

(21)

An MM-estimator also combines the S-estimator, capturing its robustness and resistance to outliers, with the M-estimator, gaining its high efficiency. Therefore, instead of using within MS-estimators in the first part of step 1 and the 2SGMS in the second part as proposed by Baltagi and Bresson (2012), MM-estimators replace the LS-MM-estimators to get a robust Hausman-Taylor estimator.

3.2 Data

The empirical model derived in the previous section will be tested using panel data covering IDA eligible countries in sub-Saharan Africa during a time span of nine years, between 2005 and 2014. Although South Sudan is one of the 34 IDA eligible countries in sub-Saharan Africa and is therefore interesting to include in the dataset of this thesis, the country only exists since the year 2011, leading to incomplete data for this country to include in the study. Therefore, the dataset is reduced to 33 countries in total. This section describes the data that is used, beginning with the dependent variable, economic growth, followed by the independent variables and in the end the instrumental variables used to account for endogeneity problems.

3.2.1 Economic growth

The dependent variable in the model is measured as the annual growth rate of gross domestic product (GDP) per capita, based on constant local currency. Data is retrieved from the World Bank’s World Development Indicators.

3.2.2 Foreign aid

The primary independent variable of interest in testing the effect of foreign aid on economic growth is, obviously, foreign aid, measured as net official development assistance (ODA). Practically all aid targets and assessments of aid performance use ODA as the key measure of aid (OECD, 2015). ODA consists of all transfers from official sources and only includes loans made on concessional terms with a grant element of at least 25 percent (calculated at a discount rate of 10 percent), and grants. Military assistance is excluded, because foreign aid only aims at improving economic or human welfare. Since in practice this consists of virtually all grants due to a very high average grant component (Boone, 1996), the increased use of grants as form of foreign aid since the turn of the millennium is captured in this measure. Table 6 in Appendix A gives an overview of the average grant component in each country of the dataset. The average grant component of all countries together is 90.7%, so indeed very high. Also, most data for ODA is only available as net ODA, so ODA net of repayments of principal.

Theory argues that foreign aid causes economic growth only with good quality institutions, where governments spend aid on investment rather than consumption (Morrison, 2012; Dutta et al., 2013). Aid must be distributed over the national economy in order to be effective, which makes the size of the national economy an important part of the aid-growth relationship (Bearce & Tirone, 2010). Therefore, net ODA as measure of aid is deflated by the population size, giving net ODA per capita as measure for aid

(22)

used in the empirical model. Data for net ODA per capita is also retrieved from the World Bank’s World Development Indicators, and is denoted in current US dollars.

Table 1 - Descriptive statistics growth and foreign aid variables

Variable Mean Std. Dev. Min Max Observations

GDP per capita growth (annual %)

overall 2.32 4.10 -37.26 13.24 N = 288

between 2.24 -1.58 7.82 n = 32

within 3.46 -33.89 12.74 T = 9

Aid (net ODA per

capita in current US$) overall 68.98 51.28 5.24 395.19 N = 297 between 44.09 24.81 258.18 n = 33 within 27.17 -44.81 246.44 T = 9 3.2.3 Institutional quality

As mentioned in the previous section, the independent variable institutional quality is measured by the ratings of the Country Policy and Institutional Assessment (CPIA). The CPIA captures the quality of a country’s political institutions and is therefore a good indicator for a country’s ability to effectively use development assistance. The IDA eligible countries are rated against a set of sixteen criteria grouped in four clusters: (1) Economic management, (2) Structural policies, (3) Policies for social inclusion and equity, and (4) Public sector management and institutions. These ratings form the annual CPIA scores on which the World Bank’s IDA Resource Allocation Index (IRAI) is based. The IDA uses this index as a key element in the country performance rating, i.e. a country’s ability to effectively use official development assistance, when deciding on providing funding for these countries. The IRAI is the average of the averaged scores for each cluster, of which each are rated on a scale of 1 (low) to 6 (high).

As main variable for institutional quality, this study uses the CPIA public sector management and institutions cluster, which is based on a country’s performance on rule-based governance and property rights, the quality of financial and budgetary management, the efficiency of revenue mobilization, the quality of public administration, and transparency, accountability, and corruption in the public sector (The World Bank Group, 2015). Therefore, especially this cluster of the CPIA ratings matches the definition of institutional quality used throughout this thesis in the way that it covers a government’s ability or disability to foster economic growth with received foreign aid.

Especially one criterion included in the CPIA public sector management and institutions cluster reflects some theories covered in chapter 2 about how the quality of institutions affect the use of aid, namely the CPIA rating of transparency, accountability, and corruption in the public sector. This rating assesses the extent to which parties in power can be held accountable for the use of foreign aid and other resources by the voting public and by law. It also assesses the extent to which employees in the public sector need to account for the use of resources, administrative decision-making, and obtained results (The World Bank Group, 2015). This rating is therefore used in the robustness checks in section 3.4.

(23)

Because the CPIA public sector management and institutions cluster is part of the overall IRAI, the IRAI replaces this measure for institutional quality when checking for robustness of the estimation results. The other clusters on which the IRAI is based cover structural policies, economic management, and policies for social inclusion and equity. The CPIA structural policies cluster is based on a country’s performance on the financial sector, trade, and the business regulatory environment (The World Bank Group, 2015). The financial sector rating includes the policies installed by government to regulate the financial sector. The trade rating is based on the government’s policies promoting or hindering trade in goods. The extent to which policy, legal, and regulatory environments hinder or help private business in fostering economic growth is covered by the business regulatory environment rating. The CPIA economic management cluster is the average of the ratings of macroeconomic management, debt policy and fiscal policy (The World Bank Group, 2015). This covers the extent to which the government is able to effectively use monetary and fiscal policy measures and control their impact on economic growth. The CPIA policies for social inclusion and equity cluster rates countries on their efforts for labor, social protection and human resource creation, gender equality, and environmental sustainability (The World Bank Group, 2015).

In order to test whether the time-varying endogenous variables have sufficient within-panel variation to serve as their own instruments means and standard deviations are measured for the overall, between, and within estimators (see Table 2). The numbers show sufficient within-panel variation for all measures of institutional quality.

Table 2 – Descriptive statistics endogenous time-varying variables (on a scale of 1 to 6)

Variable Mean Std. Dev. Min Max Observations

CPIA public sector management overall 2.99 0.47 2.2 3.9 N = 284 between 0.45 2.26 3.78 n = 32 within 0.13 2.38 3.48 T-bar = 8.875 CPIA transparency overall 2.78 0.59 1.5 4 N = 284 between 0.55 1.72 3.94 n = 32 within 0.22 1.87 3.59 T-bar = 8.875 IRAI overall 3.21 0.48 1.99 3.95 N = 284 between 0.46 2.26 3.86 n = 32 within 0.23 2.82 3.64 T-bar = 8.875

3.2.4 Instruments for institutional quality

The exogenous time-invariant variables used as instruments for institutional quality are the distance from the equator, ethnic fractionalization, and language fractionalization. Distance from the equator is included as a geography measure, because this determined whether early institutions were extractive or productive (Acemoglu et al., 2001). Distance from equator is measured in degrees of latitude of the central point in a country, ranging from 0 to 64, and divided by 100 to be able to see the effect size (Laitin, Moortgat & Robinson, 2012).

(24)

Fractionalization increases when the number of different groups increase, meaning that a higher value of fractionalization indicates a higher probability that two randomly selected individuals in a country belong to different ethnic or language groups (Montalvo & Reynal-Querol, 2005). Different sources are used for the fractionalization measures. Alesina et al. (2003) used Encyclopedia Britannica (2001) as a main source to compute data for ethnic and language fractionalization. Encyclopedia Britannica reports the shares of languages spoken as mother tongues and is generally based on national census data. For ethnic fractionalization they used other sources when needed, often dating back a few more years. Table 7 in Appendix A lists the data for fractionalization and indicates when Alesina et al. used another source, and the different source dates.

Montalvo and Reynal-Querol (2005) computed data for ethnic fractionalization using different sources. They compared their constructed dataset with the one of Alesina et al. (2003), among others, and found a strong correlation. Since the one by Montalvo and Reynal-Querol (2005) is more up to date, this one is used in this study. There are three gaps, of which two are filled in with the data from Alesina et al. (2003). This means that for Burkina Faso the value obtained from Encyclopedia Britannica 1983 is taken, and for Eritrea the value from Levinson 1998. For Sao Tome and Principe no data for ethnic fractionalization is available.

The endogenous time-invariant variables used as instruments for institutional quality are rule of law and control of corruption. Data is retrieved from the World Bank’s Worldwide Governance Indicators (Kaufmann, Kraay & Mastruzzi, 2010). The governance estimates range from approximately -2.5 (weak) to 2.5 (strong). Rule of law reflects agents’ perceived confidence in the rules of society, and especially the quality of contract enforcement, the police, courts, property rights, and the likelihood of violence and crime. Higher positive values indicate that the social environment in a country performs better. Control of corruption reflects the perceived extent to which public power is used for private gain, and to which elites and private interests capture the state. It measures the ability of a country’s policies and institutions to fight and prevent corruption, so a higher value means that corruption is better controlled for. Values don’t differ much across the years covered in this thesis, so averages are computed in order to have time-invariant measures of these variables.

The exogenous variables X1it and Z1i must be sufficiently correlated with the time-invariant

endogenous variable Z2i to be valid instruments for this variable (see Table 3).

Table 3 - Correlations exogenous variables with endogenous time-invariant variable

Aid, lagged Distance Efrac Lfrac RoL CoC

X1it Aid, lagged 1.0000

Z1i Distance 0.0937 1.0000

Efrac 0.0650 -0.4896 1.0000

Lfrac 0.0244 -0.4223 0.8884 1.0000

Z2i Rule of Law 0.1306 0.2609 0.1098 0.1228 1.0000

(25)

All correlations between the exogenous variables and the time-invariant endogenous are significantly different from zero, indicating that the exogenous variables can serve as instruments. Almost all correlations with the time-invariant variable Control of Corruption are stronger than with Rule of Law, suggesting that the validity of the instruments is higher when Control of Corruption is used as endogenous time-invariant variable in the model.

3.3 Results and discussion

This section elaborates on the results of the robust Hausman-Taylor estimation, see Table 4. In this model,

X1it and X2it are the exogenous and endogenous time-varying variables, respectively, and Z1i and Z2i are the exogenous and endogenous time-invariant variables, respectively. The dependent variable is the annual growth rate of GDP per capita, and the CPIA public sector management and institutions cluster is the variable for institutional quality. The variable for foreign aid – net ODA per capita – is lagged with one period and values for the year 2004 are used to fill in the missing values after creating these lags. Due to the identifying condition for the HT-estimator that the number of X1 variables is at least as large as the

number of Z2 variables, and the panel data used in this thesis only includes one X1 variable, only one Z2

variable can be used. Therefore, two panel regressions are performed, one using Rule of Law (RoL) as endogenous time-invariant variable and the other using Control of Corruption (CoC). The results show that this does not lead to very different estimates.

Table 4 - Robust Hausman-Taylor estimation results

Dependent variable: GDP per capita growth rate (annual %)

(1) (2)

Model RoL Model CoC

X1it Aid, lagged (net ODA per capita

in current US$)

0.0304*** 0.0328***

(0.00657) (0.00997)

X2it Institutional quality (CPIA

public management cluster)

3.840 4.415

(5.232) (3.498)

Z1i Distance from equator -6.580 -2.711

(4.361) (9.105) Ethnic fractionalization -8.106*** -8.939*** (2.427) (2.909) Language fractionalization 9.235*** 9.135*** (2.338) (1.627) Z2i Rule of Law -1.420 (5.684) Control of Corruption -3.029 (5.680) Observations 257 257 R-squared 0.378 0.347

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

Foreign aid, institutions, and economic growth : evidence from sub-Saharan Africa