The Role of Debt Relief in Sub-Saharan African
Economic Growth: An Empirical Analysis
Msc. Economics Master’s thesis Track: International Economics & Globalisation Faculty of Economics and Business Name: Lisa de Wit Studentnumber: 11419768 Email: lisacarolien@gmail.com Supervisor: D. Veestraeten Second reader: N. Leefmans Nr. of words: 13.571 Date: 15th of August 2017Statement of Originality This document is written by Lisa de Wit who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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.
Abstract This study investigates the effect of debt relief on economic growth in Sub-Saharan Africa in the context of the recent economic growth experienced in that region. A sample of 43 countries is included over a time period of 1995 until 2015 and to control for endogeneity, the system-GMM framework is applied. Previous literature failed to account for the coinciding increase in institutional quality with debt relief under the HIPC initiative, while commitment to structural reforms to better the efficiency of the public sector is a prerequisite under this initiative. This study thus tries to advance on existing knowledge about the relationship between debt relief and economic growth, by also taking institutional quality into account when researching the region of Sub-Saharan Africa specifically. The findings of this study outline that external debt has a significant negative effect on economic growth, where public debt-service-payments are the most important channel of depression. This finding emphasizes the importance of public investment in stimulating economic growth in Sub-Saharan Africa.
Table of Contents 1. Introduction………...5 2. Debt Relief……….7 2.1 Debt Relief and Economic Growth………7 2.2 Debt Relief Initiatives………...8 3. Recent Economic Growth in Sub-Saharan Africa…….…………...10 3.1 Recent Socioeconomic Development in Sub-Saharan Africa……….11 3.2 Drivers of Economic Growth in Sub-Saharan Africa………...12 4. Empirical Literature on Debt Relief and Economic Growth………...14 5. The Empirical Analysis……….17 5.1 The Economic Growth Model ………..17 5.2 The Estimation Method………20 5.3 Data Description………...24 6. Results………26 6.1 Estimation Results………...26 6.2 Robustness………...31 7. Conclusion………..35 8. References………..38 9. Appendix...43
1. Introduction Although every country in the world faces some amount of sovereign debt, not every country faces an actual sovereign debt problem. This is the case when a country experiences problems posed by so-called debt overhang. Krugman (1988) defines debt overhang as the presence of an existing amount of debt, which is sufficiently large that creditors do not expect with confidence to be fully repaid. Especially Sub-Saharan African (SSA) countries seem to struggle with sovereign debt problems1. According to Ndikumana (2004), development aid flows were furthermore diverted into debt service payments, which limited the funds available for actual development projects. This, in turn, hampered the socio-economic development of the SSA countries. Eventually, this led to the involvement of industrialised countries to help alleviate this debt burden. Initially, industrialised countries attempted to do so by non-concessional flow rescheduling, which basically meant that the creditor countries would accept a delay in debt payments and would reschedule them over a given time period. However, this strategy turned out not to be sufficient to alleviate the debt burden of developing countries. New strategies were needed and in 1996, the World Bank (together with the International Monetary Fund (IMF) and other members of the donor community) launched the Heavily Indebted Poor Countries (HIPC) initiative as a global response to this debt problem. In 2005, it was supplemented by the Multilateral Debt Relief Initiative (MDRI) to help to achieve the Millenium Development Goals (MDGs). The aim of these initiatives is to ensure that no poor country faces a debt burden it cannot manage, providing full and irrevocable debt relief. It consists of a two-step process, where the debtor country must first fulfil four conditions regarding the amount of debt and its commitment to poverty reduction strategies, before being considered for the HIPC initiative. The second step, which entails key (policy) reforms, must be concluded before the debtor country receives full debt relief from the World Bank, the IMF, and other multilateral institutions. Since the HIPC initiative thus includes a focus on improving the functioning of the government, the actual debt relief coincides with improvements in the institutional quality of the debtor country (IMF, 2016). 1 Refer to appendix 1 for the development of total external debt in SSA countries.
To date, 36 countries (30 of them located in Sub-Saharan Africa), have received debt relief of 76 billion US$ in total2. By complementing the flows of development aid, the HIPC initiative attempts to make sure that debt sustainability is maintained over time. By reducing the spending on debt service, the additional funds can be spent on health, education, other social services, and pro-growth policies. In this way, the initiative aims to reduce poverty and stimulate economic growth (IMF, 2016). However, the actual extent to which decreasing the external debt stock contributes to economic growth, specifically in the HIPCs situated in SSA, is not clear. Since the HIPC initiative induces improvements in the institutional quality of the debtor country, one could argue that debt relief in itself may not have a significant contribution to economic growth. Instead, debt relief may not be needed to help debtor countries to a higher economic growth path, since the improvements in institutional quality may already be sufficient. It is therefore very important to obtain a clear understanding of this relationship. This paper will investigate this topic further, specifically in the context of the recent economic growth SSA countries have experienced. Following the existing literature, this study will estimate an economic growth model, augmented with several debt variables. To isolate the effect of debt relief more accurately, this paper will also include several variables controlling for the improvements in institutional quality, as proposed by Beny and Cook (2009). They found support for both the “metals” and the “management” hypotheses, which argued that the recent increase in economic growth in Africa has been largely due to increases in the prices of export goods and improvements in institutional quality, respectively. The following section offers the theoretical background behind the relationship between debt relief and economic growth. It also provides a brief review of past debt relief initiatives and the workings of the HIPC initiative. The second section examines the recent economic growth in Sub-Saharan Africa and the contributing factors for this region. The third section reviews the existing empirical literature on the relationship between outstanding debt and economic growth in low-income countries. The fourth section will elaborate on the methodology used in this study, after which the data description can be found. The fifth section comprises the estimation results, followed by the conclusion and implications for future research. 2 Refer to appendix 2 for the development of debt relief in SSA countries.
2. Debt Relief The following section is split up in two parts, the first of which explores the theoretical background on the relationship between debt relief and economic growth. The second part of this section goes into greater detail on the different kind of debt relief initiatives and how the workings of the HIPC initiative. 2.1 Debt Relief and Economic Growth Before exploring the different debt relief initiatives which have been implemented in the past, it is useful to increase the understanding of the relationship between debt relief and economic growth. One of the most important theories with regards to external debt and economic performance, is the debt overhang theory by Krugman (1988). He argues that large external debt can have harmful effects on a country’s economic performance. After the increasing amount of external debt stock reaches a certain threshold, the debtor country can suffer from “debt overhang”. The debt overhang theory proposes that outstanding debt can either be financed (new loans) or forgiven (debt relief). He argues that this choice represents a trade-off for the creditors. Offering debt relief means that the creditors have lost their perspectives on repayment, while refinancing new loans means that the creditors may still receive their repayment if the debtor country turns out to do well in terms of economic performance and is thus actually able to repay its outstanding debt. However, this means that the benefits of strong economic performance will go largely to the debt service payments, rather than to the debtor country itself. This might distort the debtor country’s incentives to use the acquired funds for structural reforms and pro-growth strategies that would strengthen its fiscal position and boost economic performance, since the benefits will go to the creditors anyway. By refinancing new loans, creditors might thus actually diminish the likelihood of repayment, which is why Krugman (1988) suggests (partial) debt forgiveness rather than giving out new loans. This is especially the case when the objective is to stimulate reforms and increase economic growth in the debtor country. There are several other mechanisms through which large external debt can depress economic growth, one of which is so-called capital flight. Eaton (1987) argues that capital flight occurs due to the expectation of increased future tax obligations as the size of the public debt increases. The uncertainty regarding the actions of the government, increases the incentives for agents to place their funds abroad, leaving
limited funds available for investment within the debtor country. Capital flight therefore increases the cost of raising revenue to service debt payments, which in turn raises concerns about the likelihood of repayment for creditors. Investing in an economic environment where the chance of repayment is insecure and the funds are limited, increases the risk associated with investment. Since investors generally tend to be risk averse, such an environment can make it less attractive to engage in investment activities. Investment is generally seen as one of the main drivers of economic growth, which is why the phenomenon of capital flight tends to have a negative impact on economic performance (Alfaro et al. (2000); Mauer and Ott (1995)). When looking at the external debt service payments instead of the total outstanding debt stock, some NGOs (e.g. Oxfam International, 1999) argue that these payments could also potentially affect growth by altering the composition of public spending. The public sector spends money for a variety of reasons, all of which should be aimed to serve the public’s interest. The main objectives are to stimulate economic activity through investment in labour productivity (e.g. education) and supply goods and services that the private sector fails to do (e.g. hospitals, roads, bridges, defence and welfare). High debt service payments could thus squeeze out the amount of resources available for investment in infrastructure, health, and human capital, which are seen as important drivers behind economic growth. Therefore, high external debt could be one of the key obstacles to meeting basic human needs. There are thus several channels through which high sovereign external debt can depress economic growth. It is therefore not surprising that there have been quite a few debt-reducing strategies up until the HIPC initiative, many of which turned out not to be as successful as expected. The following section will go into greater detail about the workings of these initiatives with a focus on the HIPC initiative. 2.2 Debt Relief Initiatives During the past few decades, debt relief has certainly gained importance as a policy tool, where debt relief as a share of total development aid has risen greatly. However, the exact workings of the different initiatives are of great influence on the subsequent achieved success. As has been briefly mentioned before, the process of non-concessional flow rescheduling mainly resulted in a transfer of the burden over time. Despite this
initiative, the debt burden of the debtor countries continued to increase (Johansson, 2007). The increasing debt burden resulted in repayment problems on a systematic basis, especially in the late 1970s and early 1980s. Ever since that moment, official creditors have been providing substantial debt relief to low-income countries (Daseking and Powell, 1999). However, debt relief in itself is far from a new concept. The Paris Club, a group of sovereign creditors, has been providing debt relief to developing countries on a case-by-case basis since the 1950s. Similar to the Paris Club, the London Club consists of a group of private firms who renegotiate the lending of commercial banks to governments. However, the workings of the HIPC initiative distinguishes itself in two important ways. First of all, contrary to the Paris and the London Club, the HIPC initiative offers multilateral debt relief instead of bilateral debt relief, the former of which is quite a new phenomenon. In bilateral debt relief, one country forgives (part of) the outstanding debt of the other country and there are thus only two parties involved. In the case of multilateral debt relief, it has to go through multilateral agencies such as the World Bank and the United Nations. Multilateral aid and debt relief is often preferred in terms of developing purposes due to the fact that there is a higher participation/more resources, political neutrality, and thereby a more efficient allocation of resources (Andreopoulos et al., 2011). Second and most important for this study, the HIPC initiative has a strong focus on improving the institutional quality of the debtor country, in addition to the provision of debt relief. In most HIPCs, over 40 percent of the population live below the poverty line. However, substantial reduction in poverty ratios can only be achieved with sustained per-capita economic growth. Therefore, the HIPC initiative has established an overall policy framework, which allows the debtor country to move to a new path of sustainable growth. At the centre of this framework lies the functioning of the public sector, since substantial public expenditures are needed in social targets such as education and health. Apart from this, a sound macroeconomic framework is needed to create stability (a prerequisite for economic growth), where for example, avoiding high inflation is important. Overall, poor governance will lead to weak implementation of structural reforms and possible misuse of scarce public resources, affecting all other areas required for sustained economic growth (IMF, 1999).
The process of the HIPC initiative is split up into four stages, where the first stage concerns the qualification for assistance. The debtor country is required to adopt adjustment and reform programs supported by the IMF and the World Bank, and establish a satisfactory track record. At the end of the first phase (which could take years), the debtor country reaches the decision point, where a debt sustainability analysis will be carried out to determine the current external debt situation of the debtor country. At the decision point, the IMF and the World Bank will formally decide on a country’s eligibility and the international community will commit to provide sufficient assistance. When the debtor country is found eligible, it moves on to the so-called second phase, which is the most interesting in the context of this study. Here the country has to establish a further track record of good performance under the IMF/World Bank-supported programs. The length of this period is not time-bound, but rather depends on the satisfactory implementation of key structural policy reforms, the maintenance of macroeconomic stability, and the implementation of Poverty Reduction Strategies (PRS). During this phase, the debtor country can already receive “interim debt relief”. At the final stage, the debtor country reaches the completion point, where remaining assistance will be provided. This implies further debt relief from bilateral and multilateral creditors, reducing the country’s debt to a sustainable level (IMF, 2001). In 2005, the HIPC initiative was complemented by the Multilateral Debt Relief Initiative (MDRI), where the IMF, the International Development Association, and the African Development Fund cancelled 100 percent of their debt claims on countries that had reached (or would eventually reach) the completion point (IMF, 2016). So, contrary to previous debt relief attempts, the HIPC initiative has a very strong focus on improving the institutional quality of the debtor countries, while at the same time providing multilateral debt relief. Several countries have now reached the completion point since the introduction of the HIPC initiative, most of which are situated in Sub-Saharan Africa. The following chapter will therefore examine the subsequent economic performance in this region. 3. Recent Economic Growth in Sub-Saharan Africa The following section consists out of two parts, the first of which focusses on the recent socioeconomic performance in Sub-Saharan Africa. The second part explores the different drivers of this performance.
3.1 Recent Socioeconomic Development in Sub-Saharan Africa The economic performance of Sub-Saharan African countries has been quite low compared to the rest of the (developed) world, especially during the 1970s and the 1980s. As has been mentioned before, during this time, the debt crises emerged for many developing countries. Sachs (1988) blames the commodity-price booms in the 1970s for the debt crises in the 1980s. He argues that when commodity prices are rising faster than the interest rates on international loans, countries are able to service debt through new loans, while simultaneously reducing the ratio of debt to exports. This opportunity was seized by several developing countries in the 1970s, with the region of SSA being no exception. Many of these countries had little to no access to private capital markets beforehand. When the commodity prices went down again, many debtor countries found themselves unable to service their accumulated debt, experiencing very low or even negative economic growth rates. However, it has been two decades since the introduction of the HIPC initiative and it is therefore very interesting to look at the subsequent performance of the HIPCs situated in Sub-Saharan Africa. Figure 1 shows that before 1996, the average growth rate of the Gross Domestic Product (GDP) was largely negative, while after the introduction of the initiative, the HIPCs have in fact experienced economic growth. Figure 1: Average GDP per Capita Growth in percentages for 30 HIPC SSA Countries. Data source: World Development Indicators, the World Bank. -5 -4 -3 -2 -1 0 1 2 3 4 5 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Even though the recent growth may seem promising, the sobering reality of the situation is that it will take a long time of growth to undo the damage that several decades of stagnation and decline have inflicted upon the region. When looking at the Human Development Index (HDI), which takes more of a human development rather than an economic development focus, this becomes quite clear. The HDI is a summary measure of achievements in three key dimensions of human development; a long and healthy life, being knowledgeable, and having a decent standard of living. Even though the HDI of SSA countries has been rising steadily since the introduction of the HIPC initiative, it still remains among the lowest of the world with an average value of 0.523 compared to 0.701 in the rest of the world (UNDP, 2016). The fact that the HDI is lagging behind is particularly worrisome since human capital is often seen as one of the main drivers behind sustainable economic growth. It has an increasing effect on production through labour productivity and it contributes to a greater competitive advantage via innovation, both of which generate lasting effects upon the economy. (Pelinescu, 2015; Dewan and Hussein, 2001). Both the GDP as the HDI have thus experienced growth ever since the introduction of the HIPC initiative. One could argue, that from the point of view of timing, there may be a link between the provided debt relief under the HIPC initiative and the recently experienced economic growth in SSA. However, as the following section explores, there are several other factors which could have contributed as well. 3.2 Drivers of Economic Growth in Sub-Saharan Africa The discussion about what factors have driven the recent economic growth in Africa, apart from debt relief, can be split up into two dimensions. The first of which entails the composition of exports of the region, while the second dimension entails the recent increase in institutional quality and better economic management. The first dimension is quite unique in Africa, since the continent boasts an abundance of riches. Africa holds for instance a significant portion of the world’s reserves of oil, gold, chromium and the platinum metal groups. It is therefore not surprising that the continent has profited from the rising international demand for these resources, especially from 1990 through 2008. This increase in demand for commodities has given African governments more bargaining power, enabling them to strike better deals that capture more value for the resources (Leke et al., 2010).
Beny and Cook (2009) tested the first dimension in their study via the “metals” hypothesis, which they measured through several variables indicating the boom in export prices and the change in terms of trade. They did find significant support for the effect of “metals” upon economic growth in their cross-country growth regressions, but this might only explain part of Africa’s reversal of fortunes in recent decades. The second hypothesis for which they found significant support in their analysis is the so-called “management” hypothesis, which was measured through several indicators of macroeconomic policy and institutional quality. The “management” hypothesis thus captures the second dimension of drivers of African economic growth. The improvement of institutional quality may even be more important to sustainable economic growth in the region than the surge in export prices. As Beny and Cook (2009) argue in their paper, commodity booms are typically followed by a decrease in commodity prices. When the commodity prices eventually fall, better economic management may become especially important in maintaining the recently experienced growth rates and the subsequent achieved higher living standards. As Rodrik (1997) argues, the fundamentals for long-term growth are mainly human capital, macroeconomic stability, and the rule of law. He even goes as far to say that too much focus on “openness to trade” can be counterproductive, since it diverts policy makers’ attention away from the aforementioned fundamentals. Leke et al. (2010) agree and argue that the key reasons behind the increase in economic growth in Africa included government action to improve macroeconomic conditions, and undertake microeconomic reforms to create a better business climate. These actions included the ending of armed conflicts, creating the political stability needed for economic growth, and reducing the average inflation rate. These structural changes in institutional quality might ensure that Africa’s economic growth is not a one-time event, but rather give rise to an economic take-off. As should be clear from the analysis above, when studying the impact of debt relief upon economic growth in the region of Sub-Saharan Africa, there are many other factors which could have had a significant impact on economic growth as well. The most important contributing factor may be the improvement of the structure of the public sector and institutional quality, since this will be more important in ensuring sustainable economic growth than the often temporary effect of an increase in export prices. This is especially interesting in context of the HIPC initiative, since one could
argue that not debt relief but rather the coinciding improvement in institutional quality is the main driver of economic growth. In the following section, the empirical literature on the relationship between debt relief and economic growth will be examined, focussing on how the literature has attempted to isolate the effect of debt relief. 4. Empirical Literature on Debt Relief and Economic Growth Even though the HIPC initiative may seem promising in theory, the empirical literature has found mixed results regarding the effects of debt relief. Shortly after the HIPC initiative was launched, Elbadawi et al. (1997) used the fixed effects method to estimate the effect external debt on economic growth in developing countries. Like most of the existing empirical literature, the paper focusses on the effect of changes in the level of outstanding debt, rather than looking at debt relief specifically. The authors identified four channels through which high external debt may depress economic growth. The first two direct channels refer to a so-called debt Laffer curve, which shows that there is a limit at which debt accumulation stimulates growth3. This limit exists at point *A, after which the expected debt repayment goes down again. The first channel dominates the second channel at the left of point *A, where current debt inflows stimulate economic growth through the increasing availability of funds for investment in pro-growth strategies. As the amount of external debt increases, the expected debt repayment increases as well, since the current debt inflows stimulate growth, thus depicting a positive relationship. However, at a certain limit (*A), the second channel starts to dominate the first channel, where the accumulated debt from the past starts to affect economic growth negatively. Debt service payments increase as the external debt stock increases, decreasing the funds available for investment in pro-growth strategies. Once this limit is reached, the expected debt repayment goes down again, since the economy is no longer stimulated (debt overhang) and the probability increases that the debtor country will default on their debt. So, at low levels of debt, the first channel stimulates economic growth, but after a certain threshold, high levels of debt tend to reduce economic growth. The first channel is captured by including the ratio of debt stock to GDP, while the second channel is captured by including the squared ratio of debt stock to GDP. 3 Refer to appendix 3 for a visual representation of the “debt Laffer curve”.
The third channel used by Elbadawi et al. (1997) captures the direct negative effect of external debt on economic growth through a liquidity constraint, as high debt service payments could cause the “crowding out” of public investment. The crowding out hypothesis states that when a debtor country faces high debt service payments, these obligatory payments leave less funds available for the public sector to invest in activities which could actually stimulate economic growth. In this way, high debt service “crowds out” public investment, preventing the public sector to devote resources to productive activities, but instead forces the transfer of (part of) the available funds to foreign creditors. They capture this by including a debt service to export ratio. The fourth and final channel captures the indirect effect of all of the mechanisms described above on public expenditure, and to capture this effect they included a fiscal balance variable. They do find a negative relationship between the level of outstanding debt and economic growth, but their simulation results suggest that even though the HIPC initiative would make a positive contribution, it might still be inadequate to help HIPCs to move towards a higher growth path. This study was quite influential in the sense that the four channels which they distinguished in their growth model, have been used throughout much of the literature. When looking at more recent studies, one of the most influential papers is that of Pattillo et al. (2002). Following the earlier literature, they augmented a standard growth specification by adding several debt variables. They use the same variables to control for the four channels through which external debt influences economic growth as proposed by Elbadawi et al. (1997). First, they estimated a fixed effects model to account for the presence of country-specific effects. However, since they include a lagged income variable, the results may be biased when using this specific estimation method. When including a lagged variable in a fixed effects specification, the error term is very likely to influence the lagged variable (depending on the time horizon), which in turn could result in a (downwards) biased coefficient. Therefore, they also estimated the model using the system-GMM method, since this method yields unbiased results, and simultaneously addresses the endogeneity problems for some of the explanatory variables. Pattillo et al. (2002) decided to estimate the model including and excluding a variable for gross investment, since theory suggests that the influence of external debt on economic growth may also take place through changes in the quality of investment rather than the volume of investment per se. This can for example occur because the
same nominal expenditure on investment is not allocated to the most productive activities. Here the volume of investment can thus be high, but the quality can still be low, diminishing the positive effect of investment on economic growth. By excluding the gross investment variable, they test whether the negative effect of external debt on growth is stronger through the quality of investment or the volume of investment. However, when they eliminated gross investment from the regression, the negative impact of debt on growth is only slightly stronger. This finding suggests that the main channel through which substantial external debt affects growth negatively, is through a lower quality of investment rather than a lower volume of investment. Many studies have implemented similar methods to research the effect of external debt on economic growth. For example, Clements et al. (2003) also used both the system-GMM method and the fixed effects method to estimate the effect of external debt on economic growth. As a measure for external debt stock, they estimate the model using the face value of debt and the Net Present Value (NPV) of debt. In theory, the NPV should reflect the degree of concessionality of loans, which is a measure of the “softness” of loans and reflects the benefit to the borrower compared to a loan at market rate (OECD, 2011). By taking this concessionality into account, it tends to measure the expected burden of future debt service payments more accurately. Contrary to the previously discussed literature, Clements et al. (2003) looked at private investment and public investment separately, rather than gross domestic investment as a whole. They decided to do so, since they there is limited research regarding the specific channels through which debt affects economic growth. Interestingly enough, they find that only public investment has a strong significant effect on growth. When exploring the public investment channel further by estimating the effect of external debt on public investment, they do find support for the “crowding out” effect. The relationship is nonlinear, which implies that the crowding-out effect intensifies as the ratio of debt service to GDP rises. However, the size of the coefficient is very modest, thus implying that debt relief per se cannot be expected to lead to large increases in public investment and thereby have a significant indirect effect on growth via this specific channel. They do find a significant direct effect of external debt on economic growth, projecting that once most of the HIPCs have reached their completion point in 2005, the substantial reduction in debt would add 0.8-1.1 percent to their GDP per capita growth rates.
The methodology is quite similar across the studies which have been discussed, using approximately the same control variables to isolate the effect of the level of external debt on economic growth. These control variables contain several measures for “metals” and human capital, such as openness, terms of trade, and school enrolment. However, the third factor which has been highlighted in this paper before, has not been controlled for in the previous literature. Since the importance of the “management” hypothesis has been highlighted in the region Africa in previous studies (Beny and Cook, 2009) and under the HIPC initiative, it is very important to include variables which measure the impact of institutional quality on economic growth. In the following section, the economic growth model as implemented in the previous literature will be augmented by including several variables which capture the effect of institutional quality and governance on economic growth. By doing so, this study attempts to isolate the effect of debt relief more accurately. 5. The Empirical Analysis In this section, the first part will elaborate on the economic growth model used in this study. The second part will explain the estimation method which will be used in this analysis after which the final part will provide a descriptive analysis of the data used. 5.1 The Economic Growth Model Following the existing literature, the familiar economic growth model is augmented with three variables accounting for external debt. Since the focus of this paper is to study the effect of debt relief on economic growth, the focus will lie upon the effect of a decrease in the level of external debt. Three indicators will be used in the regression, namely the face value of the total stock of external debt as a share of Gross National Income (GNI), the face value of total external debt as a share of exports of goods and services4, and the public debt service to exports ratio. Following Elbadawi et al. (1997), the debt to GNI ratio is meant to capture the debt overhang effect (first two channels) and the debt service to exports ratio is meant to capture the influence of the crowding out effect (third channel). To capture the effects of the fourth channel, which concerns the effect of external debt on public expenditure, a fiscal balance variable should be 4 The public external debt ratio is preferred to the total external debt ratio. Unfortunately due to lack of data for SSA, the total external debt will be used in this analysis.
included. However, for the region of Sub-Saharan Africa, the fiscal balance data contained a lot of gaps. The following regressions were initially run including this variable, but the missing data had a negative impact on the empirical analysis and as such, it has been decided to exclude the fiscal balance variable from this analysis. The variables used were chosen by taking the main models as discussed in the previous empirical literature into consideration (Elbadawi et al. (1997); Pattillo et al. (2002); Clements et al. (2003)). Contrary to the existing literature, this model includes several variables to control for the effect of the (improved) institutional quality on the recent economic growth in SSA, as inspired by Beny and Cook (2009). This leads to the following empirical model:
(1) GRYPCit = a0 + a1GRYPCit-1 + a2POPGRit + a3GROIVit + a4OPENit + a5BTOTit +
a6EDUCit + a7INFLit + a8PROPIit + a9DEBTSit + a10EXTDEBTit + a11EXTDEBT2it +
µit
where µit is the error term, consisting out of the country-specific fixed effects ni and the
idiosyncratic shocks uit (µit = ni + uit). The subscript (it) refers to the country and the
time period.
The dependent variable GRYPCit represents the annual growth rate of GDP per
capita and its one-period lagged value GRYPCit-1 is included on the right-hand side. The
lagged value needs to be included to account for the dynamic process of economic growth, where a higher initial level of GDP per capita generally results in a lower growth rate. One explanation for this phenomenon could be the convergence hypothesis, which concerns the tendency of poorer economies to experience higher economic growth than richer economies (de la Fuente, 1997). We therefore expect a negative sign for this variable. Several control variables are included, POPGRit controls for the effect of the population growth rate and is measured in annual percentages. The previously discussed empirical literature has found a negative sign for this variable, but theoretically, the population growth rate could take any sign. It could have a positive effect on economic growth, since “growing economies need growing populations”, increasing the supply of both workers and consumers (Berry, 2014). However, it could also have a negative effect on economic growth through the availability of resources. A rapid population growth rate in poor countries could divert scarce capital away from
savings and investment, thereby hampering economic growth. One could go even further and argue that given a fixed quantity of land, population growth would reduce the amount of resources available for each individual, ultimately resulting in disease, starvation, and war (Fox and Dyson, 2015). GROIVit controls for the effect of gross domestic investment, measured as a percentage of GDP. As has been mentioned before, domestic investment is generally seen as a driver of economic growth and we therefore expect a positive sign. OPENit reflects the aforementioned effect of exports on economic growth in Sub-Saharan Africa, measuring the degree of openness (trade as a percentage of GDP). BTOTit reflect the
terms of trade (net barter terms of trade index, where 2000=100), which has been used as a control variable for the “metals” hypothesis as well by Beny and Cook (2009). Following chapter 3.2 on the effect of trade and export prices on economic growth in Africa, we expect a positive sign for these two variables. To account for the effect of human capital on growth, a measure of education is included in the model. EDUCit reflects the gross primary school enrolment rate and we expect a positive sign for this variable, since investment in human capital increases the labour productivity and thereby exerts a positive influence on economic growth. DEBTSit reflects the public debt service payments as a share of exports and it is included in the model to control for the aforementioned “crowding out” of public investment. We therefore expect a negative sign for this variable, since it diminishes the positive effects of public expenditure. EXTDEBTit and EXTDEBT2it comprise one of the two alternative measures of external
debt, where the quadratic term is included to control for the nonlinearity of the relationship between external debt and economic growth (Elbadawi et al., 1997). The two alternative measures are the aforementioned total external debt as a share of GNI (FVGNIit) and total external debt as a share of exports (FVEXPit). Following the
theoretical negative effect of external debt on economic growth as discussed in chapter 2.1, we expect a negative sign for these debt variables.
As has been argued before, the aforementioned control variables might not be enough to accurately estimate the effect of the level of external debt on economic growth. Therefore, INFLit and PROPIit are included in the model, both of which account
for the changes in institutional quality, as based upon Beny and Cook (2009). The first variable which reflects institutional quality, is the inflation rate, taken as the GDP deflator in annual percentages. INFLit is expected to have a negative sign, since a low
inflation rate is generally seen as an indicator of macroeconomic stability. Finally, the property rights index is included in PROPIit. As used by Beny and Cook (2009), the property rights index assesses the extent to which a country’s legal framework functions efficiently for individuals to freely accumulate property. This means that the legal framework needs to be secured by clear laws, which are enforced effectively by the public sector. The more effective the legal protection of property, the higher a country’s score will be (ranging from 0 to 100) (Miller and Kim, 2017). This means that we expect a positive correlation between the property rights index and economic growth. However, the estimation of this model might result in some problems. One problem which arises with model is that of reversed causality. A high level of external debt stock may hamper economic growth, but low economic growth may also cause the level of external debt stock to rise. This reverse causation may also apply to other explanatory variables in the model (e.g. gross investment). To overcome this issue, the model will be estimated using the system Generalized Method of Moments (GMM) methodology as proposed by Blundell and Bond (1998). 5.2 The Estimation Method The main problem with estimating a dynamic panel data model, as the one used in this study, is an econometric one. As has been briefly mentioned before, the estimated coefficients are known to be biased when the model includes both fixed effects and lagged dependent variables. The lagged income variable in this model (GRYPCit-1) is
correlated with the error term µit, due to its correlation with the time-invariant component of the error term5. This correlation between an explanatory variable and the error term violates an assumption necessary for consistent estimates by OLS and thus gives rise to bias (Leefmans, 2017). Applying the Generalized Method of Moments (GMM) can solve this problem in two different ways. Before discussing the system-GMM framework, it is very useful to understand the difference-GMM framework first. When facing endogeneity problems as discussed above, one solution involves taking first differences of the original model. 5 The following analysis closely follows Leefmans (2017).
When looking at equation (1), this would then become:
(2) DGRYPCit = a1DGRYPCit-1 + a2DPOPGRit + a3DGROIVit + a4DOPENit + a5DBTOTit
+ a6DEDUCit + a7DINFLit + a8DPROPIit + a9DDEBTSit + a10DEXTDEBTit +
a11DEXTDEBT2it + Dµit
The first difference transformation removes both the constant a0
and the country-specific effects within the error term ni. However, the right-hand side still contains
variables which are endogenous. For example, the lagged dependent income variable GRYPCit-1 still correlates with the error term. To be more specific, the GRYPCit-1 in
DGRYPCit-1 now correlates with the uit-1 in Dµit. The only way to overcome this problem,
is to find a suitable instrument for DGRYPCit-1 (Leefmans, 2017)
Arellano and Bond (1991) developed the difference-GMM framework, where instruments from within the dataset are used. As suggested by Andersen and Hsiao (1981), the natural candidate instruments for GRYPCit-1 would be GRYPCit-2 and, if the
data is transformed by first-differencing, DGRYPCit-2 as well. In the case of first
differencing, both of these variables are mathematically related to DGRYPCit-1 =
GRYPCit-1 - GRYPCit-2, but not to the error term Dµit = uit - uit-1. When assuming that the
error term is not serially correlated, using the second (or further) lag of a variable as an instrument is acceptable. However, a potential weakness in the difference-GMM framework was revealed in later work by Blundell and Bond (1998). It turned out, that the lagged levels were actually often rather poor instruments for first differenced variables, especially if these variables are close to a random walk (Roodman, 2006). The difference-GMM framework might thus not be the right estimation method for this study. However, Blundell and Bond (1998) developed a new approach to overcome these shortcomings. This new framework first-differences the instruments instead of the regressors, in order to get rid of the fixed effects and to make them exogenous. This means that there is no need to transform our original equation (1) into (2), but we simply instrument the lagged dependent variable (and any other possible non-exogenous variable) with its first-differenced lag. For example, GRYPCit-1 from the
original equation (1) is instrumented with DGRYPCit-1. This method is valid, as long as it
the fixed effects. In sum, the first approach (difference-GMM) instruments the first-differenced endogenous variables with lagged levels, whereas the second approach instruments the levelled endogenous variables with differences (Leefmans, 2017). The system-GMM framework, as proposed by Blundell and Bond (1998) combines the two approaches as described above. In this way, the positive aspects of both approaches are maximized (e.g. in system-GMM, time-invariant regressors can be included, while these would disappear in difference-GMM (Roodman, 2006)). Concretely, the system-GMM framework builds up a system of two equations; the original equation (1) and the transformed one (2). It thus comprises a differenced equation (used in the difference-GMM framework) and its lagged level instruments and a level equation with its lagged differenced instruments. The system-GMM estimator still treats this system of two equations as a single-equation estimation problem, seeing as the same linear relationship is believed to apply in both the transformed and the original variables (Leefmans, 2017). When applying the system-GMM framework, there is another choice which has to be made. Namely, the aforementioned Arellano-Bond estimators have one- and two- step variants, where the difference lies in the weight matrices used. The GMM-estimation ususally contains a larger number of instruments than regressors, increasing the likelihood of ‘overidentification’. This makes it impossible to solve the system of moment conditions, that is, ensuring that the inner products of all regressors are equal to zero. By using a weighting matrix (either one-step or two-step), GMM minimizes the magnititude of the moment conditions. Generally, the two-step system-GMM is preferred to the one-step system-GMM, since the former tends to be more efficient. However, the two-step estimates of the standard errors are often severely downward biased (Blundell and Bond, 1998). However, Windmeijer (2005) developed a command for Stata which circumvents this problem, namely xtabond2. Contrary to xtabond, it implements a finite-sample correction to the two-step covariance matrix. This is actually able to make two-step robust more efficient than one-step robust, especially in the case for system-GMM (Roodman, 2006). Having discussed the appropriate estimation method, it becomes useful to include all valid lags of the variables as instruments. The specification of the amount of lags depends upon the kind of regressor one is facing. Regressors can generally be differentiated into three specifics; ‘pre-determined’, ‘exogenous’ and ‘endogenous’
regressors. Endogenous regressors are correlated with the past and present error terms, while pre-determined regressors are only correlated with the past error terms. Exogenous regressors are not correlated with either. In this analysis, there are only three exogenous variables, namely the property rights index, the rate of inflation and the degree of openness. These three variables are not correlated with the present or the past error terms, since they are considered to be mainly influenced by factors outside of this model. For example, the property rights index is dependent on physical and intellectual property rights, which are not present in the model. The lagged value of the GDP per capita growth rate, the population growth rate, the barter terms of trade, and the primary school enrolment rate on the other hand, are considered to be pre-determined variables. They are determined prior to the current period and thus rely heavily on the past, but are not considered to be affected by the current values. Finally, gross domestic investment and every debt variable are considered to be endogenous. This means that they are influenced by the whole model, whether it be the past or the current period. For example, gross domestic investment could influence economic growth, as has been explained before, but this relationship could also work the other way around6. In system-GMM estimation, it is customary to instrument the endogenous variables with lagged terms of t-2 periods and back, while for pre-determined variables, lagged terms of t-1 periods and back tend to be sufficient (Roodman, 2006). In order to verify the results generated by system-GMM, two tests will be implemented. The first is the so-called Arellano-Bond test, which checks for autocorrelation in the error terms. By construction, the error terms of the differenced equation should display autocorrelation of order one, since Dµit = uit - uit-1 and Dµit-1 =
uit-1 - uit-2 share a common term (namely, uit-1). However, in the original equation, serial
independence in the level error term is warranted. This means that autocorrelation in the original level equation can be detected by observing second order autocorrelation in the differenced residuals. If a significant AR(2) statistic is encountered, it means that the second lags of endogenous variables are not appropriate as instruments. A useful feature of the xtabond2 command is the ability to restrict the amount of lags included in the model. Especially in samples that are fairly large in terms of time periods, leaving the amount of instruments used unrestricted, will lead to a huge number of 6 Refer to appendix 4 for a short overview of the classification of the regressors.
instruments. Numerous instruments can ‘overfit’ endogenous variables, which increases the difficulty of estimating a large matrix of fourth moments. The estimated variance of the coefficients can then become ‘too big’, which can result in biased coefficients. Since this study uses a time period of 21 years, the minimum number of instruments will be considered for each variable (Roodman, 2009; Baum, 2013). The null hypothesis in this test states that the error term is serially uncorrelated, thus only when we reject this hypothesis in the AR(2), there is cause for concern. The second test which shall be implemented is the Sargan-Hansen test, which is used to test for over-identifying restrictions in the model. It tests the crucial assumption within system-GMM framework that the instruments together are exogenous and thus not correlated with the error term. The null hypothesis in this test is that the over-identifying restrictions are valid, meaning that we strive for a model in which the null hypothesis cannot be rejected (Roodman, 2006). 5.3 Data Description The analysis uses panel data for 43 countries situated in Sub-Saharan Africa over the time period from 1995 up until 2015. This specific time period has been chosen, because it considers time before, during, and after the implementation of the HIPC initiative. A list of the countries included can be found in appendix 5. Somalia, South Sudan, Seychelles, Namibia, and Sao Tomé en Principe have been excluded from the sample, due to a lack of data. All data for the dependent and the independent variables has been collected from the databases of the World Bank, specifically from the World Development Indicators. The only variable for which the data has been collected from another source is the PROPIit variable. The data for this variable has been collected from the Index of Economic Freedom. Table 1 comprises the summary statistics for all the variables used in this regression, over 21 years7. One thing that is quite remarkable, is the cross-country variation of all the variables. For example, the annual growth rate of GDP per capita varies from -36.83% towards 140.50%. The minimum refers to the recent economic growth (or lack thereof) in the Central African Republic (2013). 7 Additional data descriptions can be found in appendix 6.
Variable Mean Standard Deviation Minimum Maximum GDP Growth Rate per Capita (annual %) 2.412872 8.072255 -36.8299 140.5011 Population Growth Rate (annual %) 2.607863 .905966 -1.290823 7.917892 Gross Domestic Investment/GDP (%) 21.5252 17.18875 -2.424358 219.0694 Trade/GDP (%) 75.2998 47.50559 14.77247 531.7374 Terms of Trade Index 113.9434 36.92143 21.39672 257.3389 Primary School Enrolment (annual %) 93.44216 25.07878 28.97797 152.2515 Inflation Rate (annual %) 23.57067 213.1779 -31.56591 5399.507 Property Rights Index 36.3723 15.78594 5 75 Public Debt Service/Exports (%) 2.323804 4.671669 0 109.6493 Face Value Debt/GNI (%) 80.30495 115.772 3.89922 1380.766 Face Value Debt/Exports (%) 314.4479 452.4013 3.89922 3789.766 Nr. Of Countries 43 43 43 43 Table 1: Summary Statistics, 1995-2015. This is probably due to the outbreak of conflict in that specific year, which consists of an ongoing civil war between the Seleka (coalition of armed, primarily Muslim groups) and
the so-called anti-balaka coalitions (primarily Christian fighters). The government has only maintained control of the capital, Bangui, while lawlessness allows armed groups to thrive in the rest of the country (Global Conflict Tracker, 2017). On the other extreme, Equatorial Guinea can be found, which has experienced incredible growth from 1995 until 1997 due to the discovery of off-shore oil fields (Oil & Gas Journal, 1995). This heterogeneity in observations does not remain limited to the GDP growth rate. Our variables of interest, namely the debt variables, tend to vary considerably as well. For example, the highest debt to GNI ratio of 1380.77% belongs to Liberia in the period 2004-2006. Fortunately, with the help of the HIPC initiative, Liberia has managed to bring this ratio down to 41.17% in 2015. This is also the case for Sierra Leone, whose debt to exports ratio of 3789.07% has diminished through the HIPC initiative to a ratio of 82.88% in 2014. When looking at the property rights index, the mean is quite worrisome, as has been said before. However, some countries at the maximum are actually comparable to high-income countries. Middle-income country Cabo Verde lays very close to the United Kingdom in terms of their Property Rights Index varying between 70 and 75. Unfortunately, countries as Zimbabwe still remain among the lowest of the world with a Property Rights Index varying between 5 and 10. This large variation in observations within Sub-Saharan Africa might suggest the existence of outliers. Most parametric statistics, such as standard deviations and correlations, are sensitive to outliers. Since the assumptions of most common statistical procedures are based on these statistics, outliers can in fact influence the analysis. At the end of the following chapter, checks with regards to outliers will be conducted to examine the robustness of the obtained results. 6. Results The following section is divided into two parts, where the first explores the obtained results for the system-GMM estimation of several variations of equation 1 considering the whole sample and data over 21 years. The second part will perform the robustness checks with regards to the effects of outliers and adding time dummies. 6.1 Estimation Results Table 2 summarizes the results obtained for the relationship between debt relief and economic growth in 43 SSA countries between the years 1995 and 2015.
Variable (1.1) (1.2) (1.3) (1.4) (1.5) (1.6) GRYPCit-1 0.079* (0.040) 0.168*** (0.062) 0.142** (0.064) 0.130 (0.102) 0.093 (0.065) 0.127 (0.094) POPGRit 4.473** (1.802) 1.086** (1.654) 0.987 (1.351) 0.641 (2.151) 0.904 (1.221) 2.207 (1.856) GROIVit 0.257*** (0.089) 0.136** (0.065) 0.012 (0.055) 0.056 (0.080) 0.033 (0.058) 0.036 (0.080) OPENit 0.013 (0.037) 0.014 (0.028) 0.039*** (0.014) 0.028 (0.029) 0.052*** (0.016) 0.101** (0.039) BTOTit 0.016 (0.018) 0.018 (0.017) 0.007 (0.017) 0.016 (0.020) 0.031** (0.014) 0.043** (0.018) EDUCit -0.121** (0.048) -0.102** (0.045) -0.095** (0.036) -0.071 (0.044) -0.121*** (0.036) -0.137** (0.057) INFLit -0.003*** (0.001) -0.003*** (0.001) 0.009 (0.032) PROPIit -0.051* (0.029) -0.030 (0.028) -0.058 (0.044) DEBTSit -0.238* (0.118) -0.233* (0.046) -0.223 (0.041) -0.256*** (0.066) FVGNIit -0.010 (0.008) 0.003 (0.018) FVGNI2it 0.000 (0.000) -0.000 (0.000) FVEXPit 0.002 (0.003) 0.005 (0.004) FVEXP2it -0.000 (0.000) 0.000 (0.000) Constant -6.414 (5.800) 3.960 (6.970) 5.870 (3.027) 3.050 (6.886) 2.941 (5.102) -2.367 (8.632) Observations 654 583 615 547 516 468 Instruments 12 14 18 20 18 20 AR(2) 1.64 -0.33 1.87 1.89 2.01 2.23 P-value of AR(2) 0.101 0.741 0.062* 0.059* 0.045** 0.026** Hansen J statistic 1.68 5.78 8.21 14.38 5.14 6.06 P-Value of Hansen 0.891 0.328 0.413 0.072* 0.743 0.641 Standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1) Table 2: Regression results of economic growth and debt relief in SSA (full sample, 1995-2015) Regression (1.1) in the second column estimated a traditional economic growth model, without the inclusion of any variables measuring debt or any variables measuring
institutional quality and macroeconomic stability. Interestingly enough, the lagged value of the GDP per capita growth rate has a positive coefficient instead of the expected negative coefficient. This finding continues throughout the different regressions, having a significantly positive effect in half of them. Apparently in Sub-Saharan Africa, countries which have a higher initial GDP per capita growth rate, experience higher current GDP per capita growth rates. Gross domestic investment has a significantly positive effect on economic growth in regression (1.1), as was expected, and the same goes for both of the trade variables (although not at a significant level). Strangely enough, the primary school enrolment rate has a significantly negative effect on economic growth. This finding seems theoretically quite implausible, however, empirically the impact of education on economic growth tends to vary. For example, Clements et al. (2003) also find a negative effect of their education variable upon economic growth when using system-GMM analysis (although insignificant). They argue that within the modest range of educational attainment levels in low-income countries, it might be impossible to identify a positive relationship between human capital and growth, even though such a relationship may exist for developing countries as a whole. Unfortunately, research focused upon developing countries, such as SSA, has to deal with a lot of missing data. For this study specifically, the data for the education variable contained a lot of gaps which may also have influenced the counterintuitive result8. When looking at regression (1.2), where the two management variables are included, the property rights index has a significantly negative effect on economic growth as well. This would imply that when a country ranks very high on the property rights index and thus has an efficient public sector in terms of law-enforcement, the lower the GDP per capita growth rate would be. This strange finding could be due to two things, the first of which could once again be the presence of gaps in the data. The second of which concerns the potential inability of the property rights index as a variable to accurately measure institutional quality. The latter of the two seems the most likely, especially since inflation does have the expected significantly negative impact on economic growth. When looking at the population growth rate, it has a positive impact on economic growth, which is contrary to the findings of Clements et al. 8 Estimating the model without the education variable did not change the signs of the other variables and barely influenced the significance levels, which is why it was decided to include the variable despite the gaps in data.
(2003) and Pattillo et al. (2002). Recall that theoretically, the population growth rate could take any sign. In the case of Sub-Saharan Africa, the population growth rate has a positive effect (although not always significant), being in favour of the idea of Berry (2014) where “growing economies need growing populations”. Interestingly enough, the two trade variables OPENit and BTOTit do not seem to have a significant impact on
economic growth in regression (1.2). However, they do have a significantly positive effect on growth in regressions (1.5) and (1.6), so only when debt is included and when it is measured specifically as a share of exports instead of as a share of GNI. It is difficult to conclude anything definitive about the impact of the “metals” hypothesis in on economic growth in context of debt relief, when the results are so dependent on the way the debt variable is measured. Of course, the focus of this study lies upon the relationship between debt relief and economic growth. The variables of interest are therefore the face value of total external debt stock as a share of GNI (FVGNIit) and the face value of total external debt stock as a share of exports (FVEXPit). The first of which can be found in regressions (1.3) and (1.4), where the debt variable did not seem to have a significant impact on economic growth. Only when including the management variables, do the signs of the coeffecients correspond with the existing literature. As in Pattillo et al. (2002) and Clements et al. (2003), when including a quadratic term for debt, the coefficient for the quadratic term tends to be negative, while that of the linear term tends to be positive. The negative quadratic term implies increasing returns, indicating that the impact of additional debt (or debt relief) would be bigger where it is already present. So, the marginal effect of debt becomes negative at a certain turning point, while at very low levels of debt, the effect of external debt on economic growth appears to be positive. As Pattillo et al. (2002) explained, it is very difficult to estimate the exact turning point because of the limited variation in the data between the growth experiences of HIPCs and those with low-to-moderate indebtedness. The combination of the positive effect of FVGNIit and the negative effect of FVGNI2it thus provides mild support for the debt
overhang effect, as discussed in chapter 2.1, although the effect is not significant. When looking at FVEXPit in regression (1.5) and (1.6), there is no significant effect to be
distinguished on economic growth either. However, in the regressions (1.3), (1.4) and (1.6), the public debt service variable (DEBTSit) remains significant, sometimes even at
payments as a share of exports go down by 1%, the annual growth rate of GDP per capita would go up by for example 0.233% in regression (1.4). This finding is quite remarkable, since the previously discussed literature failed to find a significant effect of the public debt service/exports ratio on economic growth. As can be recalled from chapter 4, the public debt service ratio was included in the model to capture the third channel of the effect of debt on economic growth, as distinguished by Elbadawi et al. (1997). This channel captured the direct effect of external debt on economic growth through the crowding out of public funds, where (part of) the available funds are transferred to foreign creditors in the form of debt service payments instead of invested in pro-growth strategies. This result thus emphasizes the importance of public investment for economic growth in Sub-Saharan African countries. When looking at the test results in table 2, it can be observed that the p-values of the Arellano-Bond test (AR(2)) lay above the 1% level in all regressions. This value is even stronger in regression (1.1) and (1.2), where it lays above the 10% level. This implies that the null hypothesis of a serially uncorrelated error term cannot be rejected at the 1% level, confirming that the second lags of the endogenous variables are appropriate as instruments. The second test which has been discussed in chapter 5.2 is the Sargan-Hansen test, where the null hypothesis states that the over-identifying restrictions are valid. Recall that this tested the crucial assumption within system-GMM that the instruments together are exogenous and thus uncorrelated with the error term. The p-value of the Hansen statistic lies above the 1% level in all regressions (except for regression (1.4)) confirming the exogeneity of the instruments. In conclusion, the results show three noteworthy findings, the first of which concerns the effect of “metals” on economic growth in Sub-Saharan Africa. Even though Beny and Cook (2009) found strong support for the effect of increasing export prices on economic growth in Africa, the obtained results in this study seem less clear. Only in some of the regressions do the trade variables have a significant positive effect on growth and even then, the effect is often quite small. Beny and Cook (2009) did not include debt variables in their growth analysis, suggesting that the presence of debt diminishes the effect of “metals” on economic growth. The second noteworthy finding concerns the “management” variables, which have a significant impact on economic growth in regression (1.2). Inflation even remains significant across most of the considered regressions. This provides sufficient evidence for the effect of institutional
quality on economic growth. The third noteworthy finding concerns the debt variables, where the total debt stock variables did not seem to have a significant effect on economic growth, suggesting that the debt overhang effect may not be that important in SSA. However, the public debt service ratio does have a significantly negative effect in almost all of the regressions. This result suggests that instead of the debt overhang effect, the crowding out effect of public funds is especially prominent in SSA. 6.2. Robustness The following sections will perform two tests to check the robustness of the results above. The first will still consider the whole sample, but will exclude any possible outliers which could have influenced the results. The second will estimate the model including time dummies. For these regressions, only equation (1.3) and (1.4) will be considered as a benchmark. As has been discussed in chapter 5.3, the large cross-country variation suggests the existence of outliers within the sample. These extreme values could influence the results and may result in an incorrect outcome. One way of testing this, is to draw a q-q plot of the residuals and check for potential outliers. Figure 2 compares the distribution of the residuals with that of the normal distribution. This method of detecting outliers has been strongly recommended by Miller (1997), whose recommendations come from practical experience. He argues that when a deviation cannot be spotted by eye in this type of plot, it is not worth worrying about. As can be observed in figure 2, there are seven observations in the tails of the graph which might be cause of concern. The three observations in the upper right corner seem to differentiate substantially from the normal distribution, as do the four observations in the lower left corner. These extremes belong to Mozambique 1996, Gabon 1999, Madagascar 2002, Zimbabwe 2003, Nigeria 2004, Sierra Leone 2013 and Sierra Leone 2015. Figure 3 shows that when these outliers are removed from the sample, the distribution becomes smoother.
Figure 2: Q-Q Plot of the residuals against the normal distribution in the complete sample of 21 years9 Figure 3: Q-Q Plot of the residuals against the normal distribution in the sample of 21 years excluding outliers. Equation (2.1) in table 3 displays the results of regression (1.3), but this time for the sample without outliers. As can be observed, the signs of the coefficients remain the 9 The debt to export ratios have been dropped before predicting the residuals, since equation (1.5) and (1.6) will no longer be used and this particular variable has some extreme values. Therefore, the wrong observations could be deleted when including FVEXPit, -2 0 -1 0 0 10 20 30 R e si d u a ls -10 0 10 Inverse Normal -2 0 -1 0 0 10 20 R e si d u a ls -10 -5 0 5 10 Inverse Normal
same and the same variables remain significant in their effect on economic growth. When looking at regression (2.2), where the management variables are included as well, it can be seen that the signs remain the same as well (as compared to (2.1) and (1.4)). When comparing (2.2) with (1.4), it can be observed that even the exact value of the coefficients of the variables of interest (debt and management) remain almost the same. These findings argue in favour of robustness of the obtained results in table 2. However, the effect of the lagged income variable is no longer significant in the sample without outliers. Interestingly enough, even when removing outliers from the sample, the public debt service ratio is still the only debt variable that seems to have a significantly negative effect on economic growth. This finding shows that even when controlling for the changes in institutional quality, the “crowding out” effect of public funds and its subsequent negative effect on economic growth is very prominent in Sub-Saharan Africa. For the second part of this section, the second robustness test will be performed. Following Pattillo et al. (2002), the model will also be estimated including time dummies. This will be conducted to ensure that the results are not driven by any time specific effects. Panel regressions which fail to control for time specific effects, could pick up the influence of aggregate trends which have no relation to causal relationships. Including time dummies allows the model to attribute some of the variation in the data to unobserved events that took place each year (e.g. war or natural disasters). Once again, regression (1.3) and (1.4) are used as a benchmark in this analysis. When looking at regression (2.3), it can be observed that only the lagged value of the GDP per capita growth rate remains significant, while none of the debt variables seem to have a significant effect on economic growth anymore. In regression (2.4), where the management variables are included, the debt service ratio and the inflation rate have a significantly negative effect on economic growth again. When including time dummies, the effect of these two variables is magnified, where the inflation rate increases from -0.003 to -0.006 and the impact of the public debt service ratio increases from -0.233 to -0.351. It is also quite interesting to see that some control variables become negative instead of positive. For example, in regression (2.4), the coefficients of the degree of openness and the population growth rate become negative, but they remain insignificant.
