African Sovereign Wealth Funds and their effect on financial development, economic development and poverty alleviation

Bachelor Thesis in Economics

African Sovereign Wealth Funds and their effect on financial

development, economic development and poverty alleviation

Y. M. Brouwer

Thesis supervisor: Lucyna Górnicka

A thesis submitted for the degree of Bachelor of Science in Economics and Business

Faculty of Economics and Business

Amsterdam, the Netherlands and

Frankfurt am Main, Germany June, 28, 2015

2

Abstract

Despite large resource wealth in many African countries, extreme poverty still persists on a large scale. In this thesis, an attempt is made to answer the question if Sovereign Wealth Funds (SWFs) have a positive effect on financial development, economic development and poverty alleviation. Existing literature was reviewed and empirical research was carried out using a multiple regression model including a Difference-in-Differences (DID) estimator. The main finding is that a largely significant, positive relationship exists between SWFs and two measures of financial development taken from the Global Financial Development Database (GFDD). The literature review shows that financial development potentially has a positive effect on economic development and in turn on poverty alleviation. This effect is dependent on a country’s legal system, institutional quality, income distribution and other macro-economic factors. The main channel through which SWFs have their effects is the conversion of resource revenue into sustainable forms of wealth for future generations. This is the way in which SWFs can reduce poverty in the present and the future generations.

3

Content

1 Introduction p. 4

1.1 Research Aim and Question p. 4

1.2 Literature Overview p. 7 1.3 Hypothesis p. 10 2 Method p. 11 2.1 Data p. 11 2.2 Analysis p. 13 3 Findings p. 16 3.1 Results p. 16 3.2 Assumptions p. 17 3.3 Summary of Findings p. 19

4 Conclusion and Recommendations p. 20

References p. 22

Appendix 1 – Derivation of the DID Estimator p. 24

Appendix 2 – Heteroscedasticity of the Error Term p. 26 Appendix 3 – Results of the Model Specification Tests p. 27 Appendix 4 – Correlations Error Term and Independent Variables p. 28

4

1 Introduction

Many African countries possess valuable natural resources. Large parts of their national revenue come from the sale of these resources. Nigeria, for example, made $350 billion in oil revenue between independence in 1960 and 2007. Yet, the country’s economy has shrunk and the proportion of people living on less than $1 per day rose by approximately 34 percentage points (Asfaha, 2007). Asfaha relates this to the so-called resource curse, which he describes as a strong negative relation between economic growth and natural resource wealth. The intuition behind this statement is, firstly, the fact that high commodity sales to other countries increase the real exchange rate, making other sectors less competitive. Secondly, the discovery of natural resources tends to increase corruption in government institutions. Nigerian governance and politics have changed fundamentally due to military dictatorships plundering oil wealth for their own use (van der Ploeg, 2011). In this way, revenues are not used to stimulate the economy and allow it to grow.

To ensure that present and future generations can benefit from Africa’s rich soil, resource revenues must be turned into sustainable forms of wealth, fostering financial and economic development. Financial development is defined as the mitigation of the effects of imperfect competition, limited enforcement and transaction costs by financial systems (Čihák, Demirgüç-Kunt, Feyen, & Levine, 2012). The question is, how is resource revenue turned into sustainable forms of wealth? In the remainder of chapter 1, a framework is constructed to work towards answering this question. In chapter 2, a multiple regression model is built to test the related hypothesis. The data used for the model and further method specifications are also explained. In chapter 3, the results of the regression analysis are presented. In chapter 4, conclusions are drawn from the findings and recommendations for future data- and model improvements are made.

1.1 Research Aim and Question

The aim of this thesis is to see if, and in what way, African home-based sovereign wealth funds (SWFs) can contribute to financial and economic development in Africa, with poverty alleviation as the final goal. The Organization for Economic Cooperation and Development (OECD) defines SWFs as “pools of assets owned and managed directly or indirectly by

5

state that commodity-based SWFs are an increasingly popular and potentially powerful tool for managing resource revenues and dealing with the resource curse. Africa’s first SWFs were established by Botswana and Ghana in 1994 with many oil exporting countries such as Chad, Libya and Nigeria to follow (Triki and Faye, 2011). According to the same article, there are currently 14 African countries with a SWF. Eight of these SWFs are stabilization funds and seven are development funds, consisting of revenues from commodity sales. These revenues mainly come from the sale of oil, diamonds and other minerals.

A stabilization fund’s main objective is to insulate the fiscal budget and the economy from commodity price swings (Allen & Caruana, 2008). Another driver of the resource curse mentioned by Asfaha (2007) is volatile commodity prices. Oil and gas prices especially are known to be volatile, causing boom and bust cycles in resource revenues with periods of low revenue exceeding periods of high revenue. According to Collier, van der Ploeg, Spence and Venables (2009), for a country to achieve sustainable growth, it has to stay on a consistent path of investment expenditure that maintains maximum returns. To stay on this path, a portion of the revenue from commodity booms has to be saved to keep up investment during periods of declining commodity revenue. This is what a stabilization fund does: it spreads out resource revenue over time. This function satisfies one of the SWFs’ “national objectives” mentioned by the OECD and is the first step for SWFs in promoting African economic development.

Another prevailing form of SWFs in Africa is the development fund. A development fund invests in socio-economic projects such as infrastructure, education and supporting industrial development (Allen & Caruana, 2008). This use of SWFs is important, given that many developing countries with natural resources lack economic growth due to underinvestment in the public sector (Collier et al., 2009). In a research paper for the Deutsche Bank, Reisen (2008) also recognizes the importance of investment. He puts forward the Hartwick Rule for investment in resource-driven economies. This rule defines the amount of investment in capital needed to offset the declining stock in non-renewable resources. Reisen (2008) states that many countries do not follow the Hartwick Rule, meaning that they have negative genuine savings (net savings minus value of resource depletion plus value of investment in capital) and become poorer each year. Development funds aimed at investing (parts of) their wealth in the public sector can help countries achieve positive net savings, hereby making them richer, not poorer.

6

The importance of a well-functioning financial system, regarding financial and economic development, was put forward by Joseph A. Schumpeter in 1912 (Levine, 1997). Schumpeter says that financial development leads to technological innovation and fosters economic growth while badly performing financial systems tend to hinder economic growth. Existing literature on SWFs offers insights into the ways in which these funds can affect financial systems. The International Monetary and Financial Committee (IMFC), for example, comments on “(…) the growing importance of Sovereign Wealth Funds in international financial markets” (Truman, 2008, p. 5). Dixon and Monk (2011) suggest that development funds support development in financial systems, helping improve investor climate over time and crowding-in both domestic and international investment. These systems are represented by financial institutions and intermediaries, defined as coalitions of agents that combine to provide financial services. Furthermore, the NEPAD-OECD Africa Investment Initiative (2008) points out that there is a growing trend towards investment diversification and non-traditional resource allocation to public sector investment projects such as infrastructure. Increasing and diversifying investment by governments and non-government organizations can lead to scale advantages, decreasing imperfect competition and transaction costs in the allocation of financial resources, which exactly fits the definition of financial development.

One important comment has to be made regarding the management and governance of the funds. There are many factors that influence the effectiveness of SWFs on financial development. Truman (2008) describes some concerns about the fact that the increased importance of SWFs means an increase in the share of wealth that is controlled by the government. He speaks of large-scale corruption and pursuit of political objectives with investments, which have been a problem in Africa even before the establishment of SWFs. In addition, conflict of interest in management between international- and national assets is stipulated. To overcome these concerns, Truman creates a “blueprint for SWF best practice”. This document contains codes of conduct or standards for these funds regarding structure, accountability and transparency, governance and behavior. The model used for testing the effects of SWFs on financial development that is set up in section 2 controls for problems regarding governance.

The question to be answered in this thesis is: Can African home-based SWFs affect financial development, economic development and poverty alleviation in Africa? Figure 1

7

shows a schematic representation of the structural reasoning to answer this question. In order to find an answer, first a literature overview was done to determine if financial development has an impact on economic development and if, in turn, economic development can positively affect poverty alleviation. The literature overview confirmed that these relationships may indeed be assumed to exist. This justifies studying the relationships between SWFs and financial development empirically.

1.2 Literature Overview

The study of the relationship between financial development and economic growth goes back to 1781, when Alexander Hamilton highlighted banks’ positive effects on economic growth. Adam Smith, on the contrary, stated in 1819 that banks harm the morality, tranquility

and wealth of nations (Levine, Loayza & Beck, 2000). As stated earlier, another important contributor to this debate was Joseph A. Schumpeter, who argues that financial intermediaries are an essential factor in spurring technological progress and economic development. Schumpeter’s view corresponds to current beliefs, which have been shortly stipulated in the introduction and will be further examined in this chapter.

Before moving on, a short comment on the use of terminology has to be made. So far, I have mainly used the term “economic development”, as opposed to many papers in which the term “economic growth” is used. Economic growth is part of development, but economic development is much more. There is no clear-cut definition of economic growth and its relation to economic development. Soubbotina (2004) explains that on the one hand, economic growth supports the reduction of poverty and other social problems by enhancing a nation’s wealth and on the other hand can have severe negative effects on development such as greater income inequality, weakened democracy and overconsumption of natural resources needed for future generations. In her book Beyond economic growth: An

introduction to sustainable development (2004), Soubbotina attempts to explain the

differences between economic growth and economic development. This distinction, though, Figure 1 – Structure Tree

This figure schematically represents the structure of this thesis’ general reasoning and research method. The arrows represent possible links. The right column shows the type of research. SWFs Empirical Financial Development Literature Economic Development Literature Poverty Alleviation

8

is beyond the scope of this thesis. Furthermore, the two concepts are found to be highly correlated. For these reasons, the two terms are used interchangeably throughout this text.

To understand in what ways financial development affects growth, Levine (1997) has created a framework that explains how financial intermediaries deal with market frictions such as information- and transaction costs. This framework shows that two of the financial intermediaries’ primary functions are allocation of resources and mobilization of savings. Next, capital accumulation and technological innovation are named to be channels through which these functions affect growth. The link between financial development and technological innovation has already been acknowledged in 1912 by Joseph Schumpeter. According to Holden and Prokopenko (2001), financial intermediaries lower transaction costs through economies of scale regarding the collection of information on firms’ future investments and profitability, simultaneously reducing information asymmetry between borrowers and lenders. This makes gathering information more efficient and less costly and helps funds to reach the most promising firms. “Improved capital allocation efficiency fosters faster growth” (Holden & Prokopenko, 2001, p. 4). Since financial intermediaries are part of the financial system, their improvement and development is a big driver of financial development.

Levine et al. (2000) empirically examine the (causal) relationship between financial intermediation and economic growth. By using new data and new econometric procedures, they build upon and improve existing research, attempting to resolve issues regarding potential biases encountered in previous research. Their results support the view that financial intermediation, and therefore financial development, has a positive effect on economic growth. A major part in the effectiveness of financial development as driver of economic growth, is played by a country’s legal system. A legal system which prioritizes creditors for receiving their full claim on corporations is found to have a positive influence on the functioning of financial systems and promotes economic growth to a higher degree (Levine et al., 2000).

The next step in our framework is the question if, and in what way, economic growth affects poverty. The relationship between economic growth and the alleviation of poverty has been studied empirically by Roemer and Gugerty (1997). Their study shows that the average income of the poorest 40% in the world increases with a one-to-one relationship with the rate of GDP growth. More specifically, they present the results of a study on the

9

effect of growth-oriented strategies on poverty headcounts in six African countries. The results show that for five out of these six countries, poverty fell during a period of economic growth. This result, though, is strongly dependent on the distribution of income and other demographic characteristics in the countries.

There have long been doubts about the effect of economic growth on the reduction of poverty. Growth rates in the developing world have exceeded those of most western, developed countries. Still, the fruits of this growth have reached the poor to a limited degree. Ahluwalia, Carter and Chenery (1979) argue that the main reason for this is the fact that, due to an unfair and unequal income distribution, the poor are being left out of the benefits from economic growth. Especially in resource rich economies, large parts of national wealth, mainly from commodities, are in the hands of a selected group of people. Poverty alleviation can be reached best through the channeling of commodity revenues to the poor via investment in socio-economic projects like infrastructure, healthcare and education or intensifying production in labor-abundant industries (Roemer & Gugerty, 1997). This is where the difference between economic growth and economic development plays an important role. In this text, I assume these concepts to be equal, but in reality there is a big difference between the two. I will make recommendations for further research on the difference between economic growth and development and their effect on poverty alleviation in the last chapter.

From the reviewed literature, I can conclude that there is a visible relation between financial development, mainly represented by progress in financial intermediation, and economic growth. In turn, under the right conditions, economic growth is found to stimulate poverty alleviation. Many factors, however, play a role in the magnitude of the effects of financial development on growth and in turn, of growth on poverty alleviation. These factors are, among others, macro-economic policy, income distribution between the rich and poor, the strength of the legal system and control of corruption. These factors have a relation to the difference between economic growth and economic development, which incorporates much more economic, and more importantly, socio-economic elements. But as stated before, growth is assumed to be equal to development, as this distinction goes beyond the scope of this thesis.

10

1.3 Hypothesis

The existence of positive relations between financial development, economic development and poverty alleviation makes the first link in the structure of reasoning, the link between SWFs and financial development, worth investigating. To do this, I set up the following hypothesis:

African home-based SWFs have a positive effect on financial development in their respective countries.

To test if this hypothesis is true, I set up a multiple regression model with financial development as the dependent variable. The regression model will provide an estimator of the effect of SWFs on financial development. The coefficient of this estimator will be tested to have a significantly positive effect on financial development. Further specifications of the model and the testing procedure will be given in section 2.2.

11

2 Method

I will now proceed to the empirical part of this thesis. In section 2.1, the data are described and in section 2.2, the regression model and method are explained.

2.1 Data

The first two questions that need to be answered before building the regression model are: What exactly is financial development and how is it measurable? The definition of financial development in the introduction corresponds to its most conceptual and basic form. To further understand and quantify financial development, Čihák et al (2012) constructed four characteristics of financial systems, hereby contributing to the Global Financial Development Database (GFDD), a collection of financial system characteristics of more than 200 countries. Its data run from 1960 to 2011. The database contains measures on four characteristics: depth, access, efficiency and stability. Depth reflects the size of financial intermediaries, access represents the degree to which people use them, efficiency primarily concerns the cost of intermediating credit and stability characterizes systematic risk, stress tests and other tools for measuring stability (Čihák et al 2012).

After reviewing existing literature and accounting for data coverage, the characteristic that I will use to represent financial development is financial depth. Depth has been used in similar research on financial development. The GFDD contains different measures to quantify financial depth. I will use two of those measures as dependent variables in the regression model. The first one is “Bank Private Sector Credit to GDP” (credit), which captures financial resources provided to the private sector by deposit money banks as a share of GDP. Deposit money banks are a collection of commercial banks and other financial institutions that accept transferable deposits. From articles on financial development by Čihák et al (2012), Beck et al (2008) and De Gregorio and Guidotti (1995), it follows that this is the best measure of financial depth. This measure directly represents the role of financial intermediaries in channeling funds to the private market participants, i.e. resource allocation. Bank private sector credit to GDP has been linked to economic growth and poverty alleviation (Čihák et al, 2012). As an alternative, I use “Deposit Money Bank Assets to GDP” (deps), which is defined as total assets held by deposit money banks as a share of GDP. These two measures are highly correlated (correlation coefficient of 0.9484)

12

but measure different aspects of financial depth. Credit contains only financial resources provided to the private sector, whereas deps also includes financial resources provided to the government and bank assets other than credit.

To control for effects on financial development not related to SWFs, the following independent variables are added to the regression model:

Rule of law rule

Control of corruption corr

Population density (people per km2 of land area) dens

Personal remittances received as % of GDP rem

Net foreign direct investment (FDI) inflows as % of GDP fdi

GDP per capita in current US$ gdpcap

Log of GDP with constant 2005 US$ lgdp

Inflation (GDP deflator, annual %) inf

Rule of law and control of corruption are important controls as the institutional quality of the SWFs host country weighs in on the effectiveness of the SWF regarding financial development. Van der Ploeg (2011) states that low rule of law and low institutional quality have a negative effect on both financial development and economic growth. A SWF’s governance tends to reflect its government’s institutional quality and norms of governance (Aizenman & Glick, 2009). Rule of law and control of corruption are measured on a scale from -2.5 to 2.5 and are part of the World Governance Indicators. Data are available from 1996 to 2011 (Kaufmann, Kraay & Mastruzzi, 2010). The years 1997, 1999 and 2001 are not reported. I assigned values to these years equal to those of the preceding year, as the paper says that small year-to-year differences are insignificant and one has to look at trends. Beck et al (2008) control for population density as denser population means deeper financial systems and thus higher financial development.

Most of the controlling variables included in my regression stem from Aggarwal, Demirgüç-Kunt and Pería (2011). To control for country size they add the log of GDP to their model. A bigger country should mean higher financial development. GDP per capita is used to control for economic growth, which, according to other literature, has a strong positive effect on financial development. The same article also includes inflation in its model. Inflation is said to distort economic agents’ decision-making on nominal magnitudes, discouraging financial intermediation and promoting saving, which are both part of financial development. The article states that FDI inflows are the largest source of external finance

13

for developing countries, significantly affecting financial development in a positive way. They find all these variables to be significant at 1% in their regressions on financial development, using data of 109 countries from 1975 to 2007. Personal remittances, funds received from migrants working abroad, are identified as the second largest external source of finance for the developing world. Remittances were also found to have a positive effect on financial development at 1% significance.

Tables 1 and 2 below show some descriptive statistics on the SWFs and comparative statistics between the treatment and control group, which are to be specified in more detail in the next section.

2.2 Analysis

Another control variable would be the size of the country’s SWF. But since, due to low transparency in African SWFs’ structure, adequate yearly data on the exact size and investment activities is lacking, I will use the method of Difference-in-Differences (DID) with panel data to estimate the effect of SWFs on financial development. I base my work on theory about DID by Stock and Watson (2012) and Cameron and Trivedi (2005). The basic idea of DID estimation is to estimate differences in change of the dependent variable over time for a treatment and a control group separately. Both groups are assumed to develop according to similar patterns. This is called the “parallel-trend assumption”. The DID estimator calculates the difference between the average change over time in the dependent variable for a treatment group and the average change over time in the dependent variable for a control group. This difference is the difference-in-differences and represents the Table 1 – Description of SWFs in Treatment Group1

This table summarizes the characteristics of the SWFs in the treatment group of the DID estimation.

SWF Name Country Year of Establishment

Funding Source Fund Type Assets Under Management2 Reserve Fund

for Oil Angola 2004 Oil Stabilization Fund 0.2

Fonds de Stabilisation des Recettes Budgétaires

Chad 2006 Oil Stabilization Fund 0.003

Excess Crude

Fund (Account) Nigeria 2004 Oil and Gas Stabilization Fund 3

National Oil Account

São Tomé and

Príncipe 2004 Oil Development Fund 0.010

1

Source: Triki and Faye (2011). 2

14

policy’s estimated effect. The policy in this case is the establishment of a SWF, the dependent variable is financial development, the treatment group is a group of countries which in some point in time have established a SWF and the control group is a group of countries with similar characteristics as the treatment group, but have not established a SWF.

The treatment group consists of Angola, Chad, Nigeria and São Tomé and Príncipe. Libya and Mauritania, originally included in the dataset, were left out due to lack of data. The treatment group countries were picked based on the year of SWF establishment. The control group consists of Cameroon, the Democratic Republic of Congo, Côte d'Ivoire, Liberia, Mali, Niger, Zambia and Zimbabwe. These countries do not have a SWF and are, just as the treatment group, specified by the International Monetary Fund (2012) as resource-rich developing countries. This minimizes the influence of possible cross-country differences not controlled for by the variables included in the regression model.

The differences in effect from the policy change (treatment) are taken between one time period before and one time period after the implementation of the policy. Except for Chad which has had a SWF since 2006, all of the treatment group’s SWFs have been established in 2004 (Triki & Faye, 2011). Adequate data are available from 1996 to 2011. I therefore choose to set the pre-treatment period (t = 0) from 1996 to 2004 and the post-treatment period from 2004 to 2011 (t = 1).

Table 2 – Comparative Statistics

In this table, the means of some of the variables used in the regression for both the countries in the treatment- and the control group are compared. All values are the mean of all observations over the entire sample period (1996-2011).

Treatment Group Control Group

Country credit deps corr dens gdpcap Country credit deps corr dens gdpcap

Angola 6.62 10.44 -1.31 12.78 1996.01 Cameroon 8.77 11.72 -1.04 37.14 866.90

Chad 3.57 5.57 -1.14 7.57 502.56 Dem. Rep.

of Congo 1.84 2.25 -1.54 23.11 235.36 Nigeria 16.46 22.34 -1.11 148.85 861.18 Côte d'Ivoire 15.27 19.06 -0.78 53.54 956.21 São Tomé and Príncipe 20.06 20.58 -0.39 159.05 861.36 Liberia 9.31 11.69 -1.10 32.99 193.59 Mali 16.03 17.32 -0.53 9.47 432.11 Niger 6.80 7.76 -0.88 10.01 254.12 Zambia 6.42 13.15 -0.75 15.00 787.72 Zimbabwe 21.05 23.20 -1.05 32.62 550.59

15

As part of the DID estimation method, the regression model includes two dummy variables, treati and postt, and one interaction variable between the two, Xit. Treati equals 1

if country i is in the treatment group and 0 if country i is in the control group. Postt equals 0

in the years before the establishment of a SWF (t = 0) and 1 in the years after (t = 1). Xit

takes the value of 1 if country i is in the treatment group in the period after the establishment of a SWF, it takes the value 0 for all other cases.

The regression model is specified as follows:

Yit = αi + β1 · Xit + β2 · treati + β3 · postt + β4 · ruleit + β5 · corrit + β6 · densit +

β7 · remit + β8 · fdiit + β9 · gdpcapit + β10 · lgdpit + β11 · infit + ɛit

where Yit is a measure of financial development. β1 is the DID estimator and captures the

average change in financial development for the treatment group, less the average change in financial development in the control group. It is the change in financial development over time that is caused by the establishment of a SWF. See Appendix 1 for a formal derivation of the DID estimator. β2 shows the initial difference in financial development between the

treatment and the control group. β3 captures differences in financial development between

the treatment and the control group that are not related to the treatment. β4, β5, β6, β9 and

β10 are percentage point changes in Yit due to a unit change in one of the corresponding

independent variables. β7, β8 and β11 are percentage point changes in Yit due to a one

percentage point change in the corresponding independent variables. ɛit is an unobserved

estimation error with an expected value of 0 and a variance that is assumed to be constant over time.

To run the regression, I make use of the xi: reg command in Stata. This command allows for the use of dummies without creating them by hand. It includes the dependent variable, the independent variables plus a dummy for the treatment group and a dummy for the two time periods t = 0 and t = 1 (Barron & Basurto, 2013). I use robust standard errors in the estimation of the regression coefficients. In Appendix 2, it is explained why.

16

3 Findings

In this section, the results from the linear regression of the model specified in section 2.2 will be presented, tested and interpreted. The aim of this chapter is to test if SWFs have a positive effect on financial development. The results of the regressions on financial development are presented in section 3.1. The assumptions behind the regression method are tested and discussed in section 3.2. Section 3.3 is a summary of the findings.

3.1 Results

Table 3 summarizes the results from both the regression on credit and deps. The first thing that stands out is the fact that the variable of interest, X, has a significantly positive effect on both credit and deps. The establishment of a SWF in a country is estimated to raise bank private sector credit as a share of GDP by 10.738 and deposit money bank assets as a share of GDP by 10.537 percentage points. Both relationships are found to hold at the significance level of 1%.

Many of the controlling variables are shown to have significant effects on financial development. An increase of 1 in the measure of control of corruption raises credit by 4.543 percentage point, where it raises deps by 6.252 percentage points. A one-unit rise in the number of people per square kilometer of land area, increases credit by 0.0802 and deps by 0.0929 percentage points. The natural logarithm of GDP appears to have a positive effect on both measures of financial development. A one-unit increase in the log of the GDP raises

credit by 2.090 and deps by 4.360 percentage points. The measure for rule of law, FDI and

the GDP per capita all have positive, significant effects on deps, but not on credit. Remittances received as percentage of GDP and inflation are estimated to have no effect on both credit and deps.

The coefficients for the two dummy variables treat and post show similar outcomes for credit and deps. At a 1% significance level, the coefficient on treat is -8.690 and -10.369 for credit and deps respectively. This means that for both measures of financial development, the treatment group had a higher initial value than the control group. The coefficient on post is insignificant for both credit and deps, meaning that for both measures of financial development, there are no factors affecting the dependent variable other than the independent variables included in the regression.

17

Table 3 – Results of the Regressions on Financial Development

The coefficients of the independent variables are shown in the first line of each result. The respective robust standard errors are presented between brackets below the coefficients. * and ** denote significance at the 5% (p < 0.05) and 1% (p < 0.01) level, respectively.

Results credit deps

rule 3.277 (1.928) 3.901* (1.605) corr 4.543* (1.536) 6.252** (1.495) dens .0802* (.0196) .0929** (.0214) rem .0929 (.1829) .1041 (.1892) fdi .0910 (.0565) .1300* (.0607) gdpcap .0015 (.0011) .0044** (.0009) lgdp 2.090* (.8285) 4.360** (0.8285) inf .0008 (.0006) .0007 (.0006) treat -8.690** (2.542) -10.369** (2.785) post -.6678 (.8467) -1.450 (.7378) X 10.738** (2.8780) 10.537** (2.809) constant (α) -6.913 (7.781) -26.288 (7.828) R2 0.6412 0.7825 Observations (Df) 116 (104) 114 (102) 3.2 Assumptions

For the estimators, and especially that of the difference in differences, to be a good estimation of the real world they have to be unbiased. That is, the expected value of the estimators have to, on average, equal their true value. In mathematical terms this means:

= where ∈ [1, 2, … , 11]

For the estimators to be unbiased, the following assumptions need to hold: I. The regression model is correctly specified.

18

II. The error term has an average value of 0: [ ] = 0.

III. The error term is not correlated with the other variables included in the regression.

Here, the parallel-trend assumption of no correlation between the error term and the DID estimator, _{, = 0, is the most important. It means that the treatment and the} control group have the same average growth in financial development over time. This is growth that is not related to any of the independent variables in the regression model.

To test the first assumption, I make use of the model specification link test (link test). The method and results of this test are presented in Tables 4 and 5 in Appendix 3. From the link test on both credit and deps, it is clearly visible that there are no omitted variables and thus the model is correctly specified. Assumption 1 is satisfied.

To test if the other two assumptions hold, I use the outcomes of the regressions and predict the residuals, which are estimates of the error term. To test if the expected value of the error term is zero, I take the mean of the predicted residuals and see if this value equals zero. For credit and deps, the means of the residuals are 7.87-10 and -3.66-9 respectively. This outcome leads me to assume that, with 95% probability, the expected values of the error terms for credit and deps are both zero, meaning that the second assumption is also satisfied.

Checking the last assumption, the assumption that the error term is not correlated with other variables included in the regression, means checking if the correlation coefficients between the estimated residuals and the independents variables are equal to zero. Table 6 (included in Appendix 4) shows that none of the independent variables are correlated to the residuals, meaning that the third assumption, including the parallel-trend assumption, also holds.

The outcomes of the tests show that all assumptions underlying the DID estimation method are satisfied. This means that all outcomes are assumed to be unbiased and on average to represent their true values.

19

3.3 Summary of Findings

The first thing that I can conclude from this chapter is that the regression model set up in section 2.2 is correctly specified. Most of the controlling independent variables that were added have the effects on financial development that were derived from the literature. This is backed by the outcome of the test on model specification. The second thing that I can conclude is that the hypothesis formulated in section 1.3 is found to be true. The DID estimator has a significantly different value from 0. This means that African home-based SWFs do have a positive effect on financial development. Through this, the aim of this chapter is reached and the structure of reasoning is completed. I can therefore now move on to answering the research question.

20

4 Conclusion and Recommendations

Given persevering poverty, great natural resource wealth and growing popularity of SWFs in Africa, the aim of this thesis was to examine if these funds can affect poverty alleviation. A structure of reasoning was set up to answer the following research question: In what way can African home-based SWFs affect financial development, economic development and poverty alleviation in Africa? The study of the relationship between SWFs and financial development was carried out empirically. Theoretical and empirical literature was studied to provide insights in links between financial development and economic development and in turn those between economic development and poverty alleviation. In this section, I summarize the results of my study, answer the research question, discuss limitations of the study and make suggestions for future research and data improvements.

SWFs’ stabilizing function spreads out resource revenue booms over time, filling fiscal gaps in times of declining revenue, without hindering economic growth. Their development function supports industrial development and spurs the growth of investment diversification and non-traditional investment. This change in investor climate leads to increased economies of scale in the allocation of financial resources. These scale advantages increase the efficiency of the collection of information on firms’ potential and their future profitability and therefore reduce asymmetric information between borrowers and lenders. Both of these effects mitigate market frictions and lower transaction costs in financial intermediation, which exactly is the definition of financial development.

The empirical part of this thesis aimed at determining quantitatively if SWFs have a positive effect on financial development. The main finding that resulted from the regression on the two measures of financial development was that SWFs do indeed have this positive effect. This outcome corresponds to current beliefs and predictions about SWFs, namely the belief that they currently play, and in the future will play, a major role in financial development. This was the first step in the structure of reasoning and allowed me to move on to the literature review.

The main result from the literature review was that financial development has a positive effect on economic growth and economic growth has poverty alleviating effects. For economic growth to have a poverty alleviating function in resource-rich countries, though, resource revenues have to be invested in better infrastructure, healthcare, education and other socio-economic projects. This is what the SWFs’ development function

21

entails and this is the way in which SWFs can transform resource revenues into sustainable forms of wealth for present and future generations to benefit from.

One important issue arises here, which simultaneously is a limitation of this research. I assumed economic development to be equal to economic growth. In reality, though, economic growth alone is not enough to alleviate poverty. Growth has to affect everyone in the economy, not just the rich and the powerful. Here the distinction between economic growth and development plays an important role.

Other limitations lie on the data front. Firstly, the lack of data on other measures of characteristics of financial systems prevented me from checking if the results of the research also hold for financial access, efficiency and stability. All these features need to be taken into account to capture the full image of the financial system. Secondly, data on SWF-specific characteristics such as size or investment portfolio and a more elaborate measure of governance are missing. Instead of including these characteristics in the regression model, the DID estimator was used, which only gives average differences in financial development between the treatment and control group over time due to the establishment of a SWF. This could give a biased image of the effects of SWFs in specific countries. Lastly, there is the issue of possible endogeneity. For example, it has been shown that financial development has a positive effect on economic growth, which is represented in the model by GDP growth. This means that the dependent variable affects an independent variable. This reversed causality leads to a biased estimate of the effect of SWFs on financial development. A way to resolve this problem is to include instrumental variables for the independent variables that possibly cause endogeneity.

To deal with the issue of the differences between economic growth and economic development, I recommend further research on the (empirical) relationships between these two concepts and clear conditions under which economic growth has a poverty alleviating effect. To get a more complete picture of financial systems, there is a need for a more comprehensive measure of financial development and a continuing improvement in and expansion of the GFDD as well as more transparency regarding SWF-specific characteristics such as size and investment portfolio.

While taking the limitations into account, it can be concluded that African home-based SWFs, through their stabilizing and developmental function, promote financial development, spur economic development and alleviate poverty.

22 References

Aggarwal, R., Demirgüç-Kunt, A., & Pería, M. S. M. (2011). Do remittances promote financial development? Journal of Development Economics, 96(2), 255-264.

Ahluwalia, M. S., Carter, N. G., & Chenery, H. B. (1979). Growth and poverty in developing countries. Journal of Development Economics, 6(3), 299-341.

Aizenman, J., & Glick, R. (2009). Sovereign wealth funds: Stylized facts about their determinants and governance. International Finance, 12(3), 351-386.

Allen, M., & Caruana, J. (2008). Sovereign wealth funds: A work agenda. International

Monetary Fund, 29.

Asfaha, S. (2007). National revenue funds: Their efficacy for fiscal stability and intergenerational equity.

Balding, C. (2008). A portfolio analysis of sovereign wealth funds. Available at SSRN

1141531.

Barron, M., & Basurto, P. 2013. Stata Training. AGRODEP Technical Note 08. Washington, DC: International Food Policy Research Institute.

Beck, T., Feyen, E., Ize, A., & Moizeszowicz, F. (2008). Benchmarking financial development. World Bank Policy Research Working Paper Series.

Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: Methods and applications. Cambridge: Cambridge University Press.

Čihák, M., Demirgüç-Kunt, A., Feyen, E., & Levine, R. (2012). Benchmarking financial systems around the world.

Collier, P., Ploeg, F. V. D., Spence, M., & Venables, A. J. (2009). Managing resource revenues in developing economies.

De Gregorio, J., & Guidotti, P. E. (1995). Financial development and economic growth. World

Development, 23(3), 433-448.

Dixon, A. D., & Monk, A. H. (2011). What role for sovereign wealth Funds in Africa’s development? Center for Global Development,‘‘Oil-to-Cash Initiative Background

Paper.’’ Available at http://www. cgdev.

org/doc/Initiatives/Oil2Cash/Final_Formatted_Monk_and_Dixon_SWF_10, 11.

Holden, P., & Prokopenko, V. (2001). Financial development and poverty alleviation: issues and policy implications for developing and transition countries.

23

International Monetary Fund (2012). Macroeconomic policy frameworks for resource-rich

developing countries. Policy paper.

Kaufmann, D., Kraay, A., & Mastruzzi, M. (2010). The worldwide governance indicators: a summary of methodology, data and analytical issues. World Bank Policy Research

Working Paper, 5430.

Levine, R. (1997). Financial development and economic growth: views and agenda. Journal

of economic literature, 688-726.

Levine, R., Loayza, N., & Beck, T. (2000). Financial intermediation and growth: Causality and causes. Journal of monetary Economics, 46(1), 31-77.

NEPAD-OECD Africa Investment Initiative. (2008). Emerging public and sovereign fund

investors in Africa’s infrastructure: Challenges and Perspectives. Retrieved from

http://www.oecd.org/investment/investmentfordevelopment/41775935.pdf Reisen, H. (2008). How to spend it: Commodity and non-commodity sovereign wealth

funds (No. 28). Research notes working paper series//Deutsche Bank Research.

Roemer, M., & Gugerty, M. K. (1997). Does economic growth reduce poverty. CAER II, 499. Soubbotina, T. P. (2004). Beyond economic growth: An introduction to sustainable

development. World Bank Publications.

Stock, J. H., & Watson, M. W. (2012). Introduction to econometrics (3rd, Global ed.). Upper Saddle River, NJ: Pearson.

Triki, T., & Faye, I. (2011). Africa's quest for development: Can sovereign wealth funds help? International Finance Review, 12, 263-290.

Truman, E. M. (2008). A blueprint for sovereign wealth fund best practices. Revue

d'économie financière (English ed.), 9(1), 429-451.

Van der Ploeg, F. (2011). Natural resources: Curse or blessing? Journal of Economic

24

Appendix 1 – Derivation of the DID Estimator

This appendix provides a formal derivation of the DID estimator.

The regression model:

Yit = αi + β1 · Xit + β2 · treati + β3 · postt + β4 · ruleit + β5 · corrit + β6 · densit +

β7 · remit + β8 · fdiit + β9 · gdpcapit + β10 · lgdpit + β11 · infit + ɛit

Dummy variables:

postt = 0 before the treatment. postt = 1 after the treatment.

treati = 1 if country i is in the treatment group. treati = 0 if country i is not in the treatment group.

Step 1: calculate the difference in financial development between treati = 1 and treati = 0 at

time t = 1:

| " = 1 − | " = 0 =

1 1 − 0 + %{ '1

|

|

where δ = β4 · ruleit + β5 · corrit + β6 · densit + β7 · remit + β8 · fdiit +β9 · gdpcapit + β10 · lgdpit +

β11 · infit

→ ₁ + %{ '1

|

|

and where OVB stand for the omitted variable bias.

Step 2: calculate the difference in financial development between treati = 1 and treati = 0 at

time t = 0:

. | " = 1 − . | " = 0 =

25

given that _{". | " = 1 − ". | " = 0 = 0 − 0 = 0 as no one is treated yet at t = 0.}

Step 3: calculate the difference in differences of financial development between time t = 1

and t = 0:

1 + %

{

}

{

}

+ %{*+, - = 1 − *+, - = 0 }

given the key assumption of DID estimation that the OVB is constant over time:

*+, - = 1 − *+, - = 0 = 0

then:

0'11 - = 1 − 0'11 - = 0 =

→ { | " = 1 − | " = 0 } − { . | " = 1 − . | " = 0 } =

26

Appendix 2 – Heteroscedasticity of the Error Term

In this appendix, the rationale for using robust standard errors in the regressions that are run on financial development is given.

One of the assumptions underlying OLS estimation is that the error terms have a constant variance over time. To test this, the error term’s estimated values, the residuals, are plotted against the fitted values from the regression. A constant variance would mean a constant pattern of data points in the plot. In both figure 2 and 3 it is visible that the pattern of data points gets wider towards the right end, indicating a non-constant variance of the residuals and thus heteroscedasticity. Using robust standard errors in the regression controls for heteroscedasticity.

Figure 2 – Plot Residuals of Regression on credit and deps

This figure shows a plot of the residuals versus fitted values from the regression on credit and deps.

Credit Deps -1 5 -1 0 -5 0 5 1 0 R e s id u a ls 0 10 20 30 Fitted values -1 5 -1 0 -5 0 5 1 0 R e s id u a ls 0 10 20 30 40 Fitted values

27

Appendix 3 – Results of the Model Specification Tests

In this appendix, the way in which the model specification link test (link test) works is explained and the results of the test for the regressions on credit and deps are presented separately.

The link test assumes that if the regression model is correctly specified, one should not be able to find any additional significant independent variables, except by chance. The regression model is refit using two new variables. These variables are _hat, the variable of prediction, and _hatsq, the variable of the squared prediction. The coefficient of _hat must be significant and that of _hatsq must not be significant for the model to be correctly specified.

The results of the tests are presented in the tables below:

Table 4 – Results link test on credit

This table shows the results from the link test on credit performed in Stata. The values displayed in brackets are standard errors. Coefficient 2 > |4| _hat 1.1908 (0.3316) 0.000 _hatsq -0.0059 (0.0100) 0.557

Table 5 – Results link test on deps

This table shows the results from the link test on deps performed in Stata. The values displayed in brackets are standard errors. Coefficient 2 > |4| _hat 1.0977 (0.2020) 0.000 _hatsq -0.0025 (0.0050) 0.691

Tables 4 and 5 show that _hat is significant and _hatsq is not significant for both regressions. This means that the regression model is correctly specified for both measures of financial development.

28

Appendix 4 – Correlations Error Term and Independent Variables

This appendix contains the table with values of the correlation coefficients between the residuals and independent variables for the regression on credit and deps separately.

Table 6 – Correlations Residuals and Independent Variables

This table shows the correlation coefficients between the predicted residuals and the independent variables from the two regressions.

credit deps Residuals Residuals rule 0.000 rule 0.000 corr 0.000 corr 0.000 dens -0.000 dens -0.000 rem -0.000 rem -0.000 fdi 0.000 fdi 0.000 gdpcap 0.000 gdpcap 0.000 lgdp -0.000 lgdp -0.000 inf -0.000 inf -0.000 treat -0.000 treat -0.000 post -0.000 post -0.000 X -0.000 X -0.000