Asian tigers, African lions : African demographics and economic growth

Asian Tigers, African Lions

African Demographics and Economic Growth

Author:

Supervisor:

Saskia Brink

Naomi Leefmans

11377194 Second Reader:

Dr. D.J.M. Veestraeten

Thesis submitted in partial fulfillment of the requirments for the degree of

Master of Science in Economics

14 December 2017 Word Count: 12154

1

Statement of Originality

This document is written by Saskia Brink, who declares to take full responsibility for the contents of this document.

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

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

2

Contents

3

LIST OF ABBREVIATIONS

SSA Sub Saharan Africa WAS Working Age Share

HIV/AIDS Human Immunodeficiency Virus/Acquired Immune Deficiency Syndrome ROW Rest of World

GDP Gross Domestic Product UN United Nations

4

1. INTRODUCTION

As global population growth declines, the world is set to undergo significant demographic changes. Yet global statistics mask regional heterogeneity. As shown in Figure 1, whilst most regions will see their population numbers level off or even decline, Sub-Saharan Africa 1 is set to experience an unprecedented population surge in the next 30 years, expanding from 1 billion today to 2 billion in 2050, steadily increasing towards 4 billion at the end of the century.2

Figure 1: Global Population Trends

Data Source: UN Population Prospects (2017 Revision)

Significant age-structural changes will accompany the above population trends. As the majority of the developed world will be faced with ageing populations, Sub-Saharan Africa (SSA) will in contrast be home to the proportionally youngest populations in the world (Drummond, Thakoor, & Yu, 2014). A variable commonly used to examine demographic dynamics is the dependency ratio. It measures the portion of a country’s working-age population to its dependents, where the dependents consist of both the population under the age of 15 and aged 65+. The higher the ratio, the larger the proportion of dependents per working-age person, which represents a drag on the economy as a whole. In contrast, a lower dependency ratio represents a proportionately larger working age population that is able to support the dependent population. As illustrated in Figure 2, the dependency ratio of SSA diverges strongly from the rest of the world. Excluding SSA, the rest of the world steadily moved towards lower dependency ratios roughly since the 1960s until now, with some regions, such as Europe,

1

See Appendix for Map of Sub Saharan African countries 2

5

already experiencing a renewed increase in dependency ratios as their populations start ageing. In contrast, SSA has exhibited a consistently high dependency ratio, which only started to decrease towards the end of the century, at a markedly slower pace. As of 2020 until 2050, it will be the only region still experiencing declining dependency ratios.

Figure 2: Dependency Ratios

Why are these age structural changes important? Lifecycle theory tells us that agents’ economic behaviour is not constant over a lifetime, and thus people’s economic behaviour and productive capacities will vary across different age groups, which can lead to different economic outcomes (Bloom, Canning, & Fink, 2008, p. 4). Yet, for all the possible implications these substantial age-structural changes may have, neoclassical economics has dedicated relatively little focus on the interaction between demographic changes and economic performance. Although many empirical economic growth models do add population variables, they are usually limited to population size and growth, not structure. This omits important features of population dynamics.

Only recently have demographic factors been included in economic growth regressions. A seminal paper by David Bloom and Jeffery Williamson (1998) coined the so-called “demographic dividend”3, a window of economic opportunity that can arise when a country

3

They actually defined it as the “demographic gift” and the term “demographic dividend” only appeared in subsequent literature

6

transitions from high fertility and mortality rates, to low fertility and mortality rates, as the age structure of the population then changes. By including populations’ evolving age structures in a cross-country growth regression, they estimated that demographic changes, i.e. declining dependency ratios, contributed to approximately a third of East Asian economies’ high economic growth in the period between 1960-1990 (henceforth referred to as Asian Tiger economies).

Whilst most of the world has already passed through its demographic transition from high fertility and mortality rates to low fertility and mortality rates, SSA has only just begun this transition. As shown in Figure 3, fertility rates in SSA were at rather similar levels in 1950 compared to other developing regions. However, whilst other developing regions saw rapid decreases in their fertility rates, SSA’s has remained persistently high until 1980, with a delayed and slower rate of decrease ever since. Figure 4 illustrates a further dynamic of SSA’s demographic lag. Whilst the rest of the world has already experienced its peak change in working age population, SSA will only experience this towards the middle of the 21st _century.

Figure 4: Change in Working Age Population

The decline of the dependency ratio in SSA since the mid 1990s has coincided with an economic growth revival in the region. After exhibiting dismal economic performance throughout the 70s and 80s, the late 1990s ushered in a period of economic prosperity for many SSA economies. As this growth has largely been maintained throughout the 2000s, there is optimism that SSA has experienced a permanent change in its growth trajectory. Several parallels have been drawn between the Asian Growth Miracle and SSAs recent economic performance. Accordingly, the term “African Lion” economies has been coined as an African equivalent of Asian Tiger economies, identifying high performing economies in the respective regions (The Economist (2011), McKinsey Global Institute (2010)).

7

Although there is a defined categorization for Asian Tiger economies (see Figure 5), there is no current consensus or classification as to which countries should be included in African Lion economies. Due to the sheer variety of economic growth levels, size of economies and regional factors, it is difficult to define a single set of conditional parameters.

This thesis broadly follows Bhorat and Tharp’s (2016) identification of “fast growing and/or economically dominant” African economies: Ethiopia, Ghana, Kenya, Mozambique, Nigeria, and South Africa. Yet one important substitution is made: Instead of Mozambique, this thesis will include Tanzania. The overall country choice was made not only on the basis of economic performance and dominance, but also on future population projection, where Tanzania far outweighs Mozambique. More importantly, Mozambique has not exhibited the recent demographic changes that others in this subgroup have. Therefore these countries are not only chosen on the grounds of recent economic performance, but also due to future demographic dominance. This the reason why high performing countries such as Botswana and Rwanda were not included.

With a combined population of 477 million people, these 6 countries represent approximately 50% of SSA’s current population and have shown a somewhat consistent economic growth turnaround since the turn of the century, increasing from an annual average of 1% to a current annual average of approximately 4%, as illustrated in the figures below.

Figure 5: Classification Asian Tigers

8

Figure 6: GDP Growth and Population

World Development Indicators and UN Population Division (2017)

Whilst several studies exist that examine the prospects of the future size of the demographic dividend, none were found that consider the impact of current age-structural changes on SSA’s recent economic growth turnaround. A commonly held belief is that improved political and macroeconomic policy, as well as an increase in commodity prices have been the key drivers behind Africa’s recent growth revival (Beny & Cook, 2009). Although these factors are by no means trivial, little emphasis is placed on how the demographic structures of these countries have influenced their past and will influence future economic growth paths. Furthermore, there seems to be little consensus on the determinants of the recent growth spurt. For instance, Rodrik (2011) in contrast finds little evidence for the impact of changes to macroeconomic management, trade or institutions. This is result is also found in Easterly (2005). Finally, most studies treat SSA as one homogenous group in their analysis, which may overlook important differentiating factors of the 48 nations. This lack of consensus and the variety of results in the literature motivates further research into the determinants of this recent growth spurt in a selected group of countries, and whether age-structural changes have contributed to it. This motivates the following research question:

To what extent have demographic changes contributed to the recent economic growth turnaround in African Lion economies in the period between 1990-2014, and how will future changes influence future economic growth prospects?

A cross country convergence model will be used to estimate the determinants of growth for a panel over the period from 1970-2014. Although the research question specifically defines the estimation period as between 1990 and 2014, which was chosen due to data constraints of the Lion economies, the cross country convergence model will include data from 1970 onwards. Whilst Lion economies are the countries of interest, in order to estimate the coefficient for the

9

broad impact of demographic changes on economic growth, a worldwide sample is used. This coefficient can then be used to estimate the impact of demographic changes in the Lion economies between 1990 – 2014. These estimates will also be used to calculate the predictions for the future demographic changes on economic growth in the African Lion economies, as defined above.

I limit my analysis to countries for which explanatory variables are available, which limits the analysis to 84 countries across the world in all regions (see Appendix for complete list). This is based on commonly applied empirical studies, such as Bloom et al. (2004, 2009, 2010). Empirical estimation techniques are applied in the form of Ordinary Least Squares (OLS) and Instrumental Variables (IV) regression analysis.

The expectation of this research is that age-structural changes will have a positive effect on GDP per capita growth. Furthermore, I expect that gains from demographic changes will be contingent on institutional factors, and will test this by adding an interaction term between institutional quality and demographic changes, as found in Bloom and Canning (2003), Bloom et al. (2007) and Lee, Lee and Mason (2006). This is especially relevant in the Sub-Saharan African context, where institutional quality is still lagging behind the rest of the world, although promising progress is being made.

This research could offer further insights as to whether Africa’s recent growth turnaround was a fleeting economic boom or rather the start of a period of higher economic growth. Additionally, this thesis will focus on the effect of demographic changes for future performance in a subset of African economies. This research will also further contribute to the understanding of how age structures affect economic growth in SSA: How to affectively design policy that can cater for the unprecedented population influx depends on how much emphasis governments and policy makers place on future demographic changes today: investment in reproductive health, education and long-term employment strategies will be essential in determining whether such a demographic change will spell opportunity or disaster in the African context.

10

2. LITERATURE REVIEW

The literature review is organized into 3 parts. The first part introduces the demographic dividend theory and discusses the current literature of demography in conjunction with economic performance. The second part introduces stylised facts on demographic changes and economic growth in African Lion economies and reviews literature examining economic growth determinants in SSA.

2.1 INTRODUCTION ON DEMOGRAPHY AND ECONOMIC GROWTH

“The power of population is indefinitely greater than the power in the earth to produce subsistence for man.” (Malthus, 1798)

The inclusion of demographic factors in the analysis of economic growth has historically been skewed towards population growth and size. As early as 1789, Thomas Malthus hypothesized that due to humans’ innate desire to procreate, rapid population growth would continuously outstrip resources, such as land, capital and knowledge, leading to a rather doomsday type scenario. In the Malthusian view, population growth is perceived as a burden to society and the only way population can be kept in check is either through moral restraint, war, disease or famine. Therefore, in the absence of technological improvements, population sizes are self-equilibrating. Any increases in resources would eventually be outstripped by the subsequent population growth through capital dilution (Galor & Weil, 2000). This view grew in dominance in the post war periods and influenced policies that aimed to slow population growth (Coale & Hoover, 1958).

Besides this “pessimistic” Malthusian theory, two other schools have emerged within the literature: The population Optimists and the population Neutralists. The “optimistic” theory holds that there is a positive relationship between population growth and economic performance, so that an increase in population promotes economic growth by increasing the labour size as well as the market for goods and the ability of both labours and markets to innovate (Kuznets, 1967) (Simon, 1996). Research that emphasized no significant association between population growth and economic growth emerged in the 1980s and was coined the “neutralist” theory. According to this view, changes in population size have no effect on economic performance (Kelley & Schmidt, 1995). This view garnered significant backing from the empirical literature, which is also reflected by the waning development policy emphasis on

11

population and reproductive health during this period (Bloom, Canning, & Sevilla, 2003, pp. 4-5).

Yet none of these 3 theories has emerged as the long term dominant theory with reference to how population size and growth affects economic performance. A reason that is commonly cited for the inconclusiveness and lack of consensus of these studies has been the simplification of demographic processes (Kelley & Schmidt, 2005). The majority of these early studies focused primarily on the relationship between population growth and size on economic growth, but did not include the composition or age structural dynamics of other population variables. This is often reflected by the omission of demographic dynamics in neoclassical growth models, which usually only include a constant population growth variable. Whilst this may be an appropriate assumption for economies that exhibit relatively stable population trends, it overlooks important dynamics in countries that are experiencing important age-structural changes. In the case of some developed countries, this means managing increasingly older populations, whereas in the case of some developing regions, like SSA, this means facing a young population boom.

Adding demographic variables to economic growth models only gained momentum in the 1990s. A forerunner was Robert Barro (1991), who incorporated demographic variables into the classical convergence (technology-gap) framework (Kelley & Schmidt, 2007). Notably, he distinguished between fertility and mortality rates, instead of using population growth per se, with fertility being negatively correlated with economic growth and mortality rates showing no significant effect. Kelley and Schmidt (1994) added population density and size to this framework. Additionally, Mankiw, Romer and Weil (1992) used the Solow convergence framework and augmented it with a human capital component. By adding the human capital component, the estimated negative magnitude of the effect of population growth on economic growth was significantly reduced.

Barro’s framework evolved further in the late 1990s, primarily pioneered by David Bloom. By building on Barro’s framework, Bloom included transitional demographic variables to model periods of population instability, such as when a country transitions from high fertility and high mortality rates to low fertility and low mortality rates. This transition is discussed in the following section.

12

2.2 THEORY ON DEMOGRAPHIC TRANSITION AND DIVIDEND

As countries transform from poor to rich societies, they simultaneously move from high birth and high mortality rates to low birth and low death rates4 _{(Bloom & Williamson , 1998). This}

is commonly referred to as the demographic transition. Figure 7A illustrates the declining mortality and fertility rates and their implication for total population size. Figure 7B illustrates the population pyramids of the corresponding 5 different stages of the demographic transition.

Phase 1

During the first phase, birth and death rates are both high and rather similar, leading to relatively constant population numbers and in turn stable population age-structures. The corresponding population pyramid in Figure 7B has a strong triangular shape and a wide base.

Phase 2

In phase 2, mortality has decreased, due to various advances in public health services, eradication of communicable diseases, improvements in sanitation and improvements in primary healthcare, such as vaccinations (Bloom & Canning, 2004, p. 4). Yet, as mortality drops due to these advances, birth rates remain high, as there is a lag between the decrease in mortality rates and a decrease in birth rates. Because birth rates are higher than death rates, the lag creates an increase in the population, i.e. an increase in children since it is mostly infant mortality rates that decline. This results in “boom generations” in younger population cohorts, that progressively move upward through the population pyramid (Bloom & Canning, 2004, p. 6). The population pyramid for this stage therefore widens further towards the top than in phase 1.

Phase 3

With decreased rates of infant mortality and higher childhood survival rates, along with increased economic prosperity, the need to have multiple children diminishes and birth rates decline, evolving into phase 3. The decrease in fertility can also be due to educational development and improved access to family planning resources (Bloom & Canning, 2004, p. 7). Although fertility rates are still higher than mortality rates, population growth is steadily declining, eventually leading to an end of the generational boom. The population pyramid for

4

Following Bloom and Williamson (1998) external migration is ignored in this paper, who assume that migration in current times, though relatively larger than before, is not significant enough to matter for demographics

13

this phase is still triangular in shape, but the sides are less steep and the share of the population towards the top of the pyramid have increased.

Phase 4

Once countries have reached stage 4, both crude birth and death rates are at stable and similar levels again, resulting in a return of stable population numbers as shown in figure 7A. The corresponding population pyramid has relatively straight sides and a rounded top, as the number of elderly increases.

Phase 5

Stage 5 is characterized by low mortality rates and even lower birth rates, meaning that the per capita replacement level is too low to sustain stable population numbers, leading to contracting populations (Magrath, et al., 2013). This is also emphasised by the bottom of the corresponding population pyramid being inverted.

Figure 7: The Demographic Transition

From Magrath et al. 2013

Eastwood and Lipton (2012) adapt this transition model by defining three different phases of a population’s dependency ratio. The first phase sees mortality rates decreasing, especially in younger age cohorts.This leads to an increase in the dependency ratio, due to the increase in

14

the share of young dependents within the population (Eastwood & Lipton, 2012, p. 27). In phase 2, “this population bulge” is now of working age, automatically leading to a decrease in the dependency ratio. In the 3rd _{phase, the population bulge is at retirement age, leading to an}

increase in the dependency ratio. Reviewing Figure 2, one can clearly see how SSA lags behind others in dependency ratio trends by approximately 20 years. It also shows how parts of Europe are already experiencing increases in dependency ratios as more people are moving towards retirement age.

Combining these two approaches, in phase 2, a population “boom” occurs as the mortality rate drops but the birth rate remains high, indicating an increase in the dependency ratio, as the number of children increases. This young cohort initially represents a drain on the economy: they require higher investments in education and healthcare and represent little or no economic output. Yet, as this large young cohort moves from the dependent population towards working age, it presents a window of economic opportunity: the demographic dividend. The demographic dividend is the potentially favourable relationship between increases in the working age share in population (so decreases in the dependency ratio) and economic growth. As subsequent young population cohorts are smaller relative to the “boom’ cohort due to decreases in fertility, they demand relatively low public spending and investment, meaning that fewer private and public funds need to be allocated to the dependent population, potentially freeing up funds for increased savings, investments, and consumer spending, all boosting economic growth (Mason, 2005). Modigliani’s Life-cycle theory describes how different age cohorts exhibit different consumption and production behaviours over the course of a lifetime. Therefore the dependent portions of the population represent net consumers and the working-age share represents net producers, provided that they are economically active.

The temporary boost in the share of working age adults could affect economic growth through various channels, the main ones being the relative size of the labour force, savings and human capital (Bloom, Kuhn, & Prettner, 2016, p. 2). The relative size of the labour force indicates an increase in the labour supply, potentially boosting economic output. The increase in aggregate savings is due to this larger share in working age population, which in turn can be used for capital and productive investments, further boosting economic output. The potential increase in human capital is due to fewer (young) dependents, meaning that, potentially, more resources are devoted to education per child. The drop in fertility can also indicate a potential increase in female labour force participation leading to a further economic boost (Bloom, Kuhn, & Prettner, 2016, p. 3).

15

An important caveat exists with the demographic dividend: it is not automatic. It merely represents a potential window of opportunity, rather than a guarantee, of increased economic growth. Reaping the dividend is contingent on various factors such as the quality of institutional and political environment, macroeconomic management, trade regulation and educational policies (Bloom & Canning, 2004, p. 12). Rigid labour markets, inward facing economies and bad governance are often cited as a reason for why Latin America was not able to efficiently harness its demographic dividend, although it underwent similar age-structural changes as East-Asia did in the 1970s, but GDP per capita between 1975 -1995 only grew by 0.7% annually, which is significantly lower than the 6.8% annual growth of Asian Tiger economies (Bloom D. , et al., 2010). Lee, Lee and Mason (2006) confirm that the potential of the demographic transition yielding a demographic dividend is highly contingent on quality institutions. The increase in the workforce will only yield economic improvements if these additional workers are able to find employment. Without adequate policy planning, the increase in the WAS may lead to increasing unemployment and exacerbate economic and social progress (Drummond, Thakoor, & Yu, 2014, p. 6).

2.2 EMPIRICAL FINDINGS ON THE DEMOGRAPHIC DIVIDEND

In this section, empirical findings on the demographic dividend are presented. Since early studies were concentrated mainly on Asia, this section will start with a review of the empirical literature on the demographic dividend with reference to the Asian growth miracle. This is then followed by a contextualization of SSA and Lion demographics, after which existing literature of the demographic dividend in SSA is discussed.

As discussed above, Bloom and Williamson (1998) are often identified as the pioneers of the demographic dividend literature. 5 They showed that population dynamics neither affected economic performance through population growth or size, nor through fertility or mortality rates, but rather through age-structural changes. Their initial findings indicated that up to one third of the Asian growth miracle could be attributed to changes in age structure in these countries (Bloom & Williamson , 1998). These results indicate that demographic dynamics were in fact the largest contributing factor to the Asian growth performance. This is significant

16

as it suggests that other variables, such as macroeconomic policies and institutional quality may have been overstated in previous studies.

These results have subsequently been reproduced and confirmed by multiple articles, using different estimation methods and data. Whilst Bloom and Williamson (1998) adopt cross sectional data for the period of 1965-1990, later studies have frequently used panel data (Kelley and Schmidt 2005, Bloom et al. 2007 & 2015)). A precursory study was conducted by Islam (1995) who also used panel data estimation methods and used the initial level of technology as a country fixed effect.

Although the organization of data has differed between studies, the empirical estimation techniques have remained relatively similar over time. Bloom and Williamson (1998) use ordinary least squares (OLS) in combination with instrumental variables (IV) to obtain their results and to correct for possible endogeneity due to reverse causality. Kelley & Schmidt (2005) use a similar setup but add country and time fixed effects.

Cuaresma, Lutz and Sanderson (2014), in their analysis of demographic changes and educational characteristics, use a Blundell-Bond “system” GMM estimator as their empirical estimation strategy (Blundell & Bond, 1999), concluding that the demographic dividend is highly conditional on educational improvement. Besides these standard regression techniques, other models have also been used to obtain similar results, albeit not as frequently.

The next sections will discuss characteristics of Lion economies with reference to demographic changes.

There are several key demographic differences between Lion economies and the rest of the world (ROW). As shown in Figure 3, whilst the rest of the world saw steep declines in fertility from the middle of the 20th century, decreases in fertility only occurred with a 20 year lag in Lion economies and at a markedly slower rate. This has caused not only the demographic transition to be delayed, it will also occur at a more sluggish pace. Dependency ratios peaked in the mid-80s, lagging behind Asia by 20 years.

The persistence of high fertility rates and large family sizes even with improved infant and child survival rates in the region seems puzzling compared to other regions. However, one reason may be that in many SSA countries, children are still regarded as a valuable source of labour and as a de facto pension. The reliance on children as future pension resources is partially due to low savings rates, caused by weak financial infrastructure (Bloom, Canning, &

17

Fink, 2007, p. 6). Whilst the current fertility projections may not be as prominent for Lion economies as compared to the past experiences of other regions, it is also important to note that there is still room for policies to induce a steeper fertility decline.

Figure 8

In terms of the evolution of the WAS of the population, Lion economies do not only differ from the rest of the world, but there is also some heterogeneity between Lion economies. Whilst South Africa already experienced increases in its WAS from the middle of the century, others at the time were still experiencing declining WAS shares. Ghana was the second to undergo the WAS-share turnaround, namely in the middle of the 1970s, followed by Kenya, Tanzania and Nigeria in the 80s. Ethiopia only began its turnaround in the middle of the 1990s.

The impact of HIV/AIDS has also affected the working age population in these regions, leading to lower proportions of working age population, as HIV/AIDS has largely affected healthy, working age adults, with the majority of deaths occurring between the ages of 20 and 59 (Bloom & Canning, 2004, p. 10) (Bloom, Canning, & Fink, 2007).

War and civil unrest has also had a significant impact on many African economies. War does not only kill soldiers and civilians, directly affecting demographics; it also debilitates infrastructure and functioning institutions, which are integral in reaping any benefits from the demographic transition (Bloom, Canning, & Fink, 2007, p. 6). Whilst none all of the lion economies defined above have experienced long standing wars, the spill over effects, such as displaced persons and regional instability could have contributed to impaired economic performance. 45,0 50,0 55,0 60,0 65,0 70,0

Working Age Share

Percentage of Total Population

Ethiopia Kenya Tanzania South africa Nigeria Ghana

18

The rest of this section is dedicated to the discussion of empirical literature on the demographic dividend of SSA.

In their seminal paper, Bloom and Sachs (1998) attribute nearly 19% of economic growth differences between SSA and other regions for 1965-1990 to differences in demographics and confirm these results with cross country and panel data analysis. In economic growth research, regional dummy variables are often added for Africa and East Asia, with the coefficient for the Africa dummy usually appearing negatively and for the East Asia dummy appearing positively in the results. However, once age-structural dynamics are added to these regressions, the regional dummies are no longer significant (Bloom & Canning, 2004, p. 12) . This means that a large part of Africa’s lack of growth from the 1960s to the 1990s can potentially be explained due to demographic variables, possibly refuting many articles of the 1990s that point towards other explanatory variables for its poor growth performance. This does not mean that other variables are irrelevant, but merely that the omission of age structures may have been an important omission.

Eastwood and Lipton (2011) claim that future fertility declines are overestimated and thus the demographic dividend in SSA will be smaller than it was in Asia, unless there is an increased policy emphasis in fertility decline.

Bongaarts and Bulatao (1999) also find that due to factors such as high fertility rates and the HIV/AIDS pandemic, SSA will not be able to reap the same demographic dividend as the rest of the world. This is however countered by Bloom et al. (2007). In Realizing the Demographic

Dividend: is Africa any Different, they assess whether SSA’s growth is determined by the same

variables as the rest of the world, particularly with respect to demographic changes. This builds on Sachs & Warner’s (1997) research on the determinants of economic growth in SSA, noting that there have been several improvements in fertility and institutional quality since then. They use the standard IV estimation (as discussed earlier) and confirm that the determinants of African growth are the same as for the rest of the world. Importantly, they find that the growth effects of the demographic dividend are highly conditional on institutional quality by interacting the growth in WAS with institutional quality (Bloom D. , Canning, Fink, & Finlay, 2007, p. 13). This confirms Lee and Mason (2006) and Bloom et al. (2003) who also emphasise institutional quality as a contingent factor for the demographic dividend.

Drummond et al. (2014) estimate the potential magnitude of the demographic dividend for SSA, using a worldwide panel of 172 countries for the period 1960-2010. They construct a

19

simple general equilibrium overlapping generations model to predict the potential growth effects from theory. They deviate from previous econometric methods by adding country and fixed effects (FE) based on Hausman tests and apply a system GMM estimator (Drummond, Thakoor, & Yu, 2014, p. 4). They conclude that the results from system GMM, although similar to those of FE, could potentially lead to too many instruments if too many variables are included, opting for the FE as baseline estimates instead (Drummond, Thakoor, & Yu, 2014, p. 13). They also do not include institutional quality, arguing that it unnecessarily curtails sample size. Yet this directly contradicts Bloom et al (2007). Instead, they conclude that the demographic dividend is strongly contingent on human capital investments. This is along the lines of Ashraf et al. (2013) and Mason et al. (2016) , who use an accounting simulation approach to show that investments in physical and human capital are key drivers for the demographic dividend and fertility decline. Drummond et al. (2014) also estimate the potential demographic dividend for SSA between 2010-2100. They define the demographic dividend as the difference between GDP per capita excluding the demographic transition and GDP per capita including the demographic transition. They consider 2 scenarios: one where the WAS remains constant and one where the WAS rises as per the UN median predictions. They use the coefficients of previous regression results (holding all other terms constant) and apply this to country data for growth of WAS for the forecast. This method will also be followed in this thesis.

There is also someliterature using other models and estimation techniques with reference to the future effects of demographic changes on economic growth.

Ahmed et al (2016) use a LINKAGE, global computable general equilibrium (CGE) model that links economic data with global trade flow data and a microsimulation model that combines cross-country household survey data. Their results estimate that the demographic dividend has the potential to account for between 11-15% of GDP growth (absolute) by 2030, contingent on institutional quality. They estimate that this could be even large if larger gains are made with regards to educational outcomes (Ahmed, Cruz, Go, Maliszewska, & Osorio-Rodarte, 2016, p. 762).

20

3. THEORETICAL FRAMEWORK, MODEL AND METHODOLOGY

Section 3.1 presents the theoretical framework on which the empirical work is based, while section 3.2 presents the actual empirical model. The chapter ends with a description of the data that is used for the estimation that follows.

3.1 THE THEORETICAL FRAMEWORK

The focus of most economic growth models is on growth in income per capita. However, conventional neoclassical models, such as the Solow-Swan model, do not include the dimension of population dynamics, which is crucial if we study income in per capita terms. Following Bloom and Finlay (2009) and Drummond et al (2010) this section derives an augmented Solow model that does include demographic dynamics.

A useful starting point is the Cobb-Douglas production function, which relates output to factor inputs and productivity:

𝑌 = 𝐴𝐾&_𝐿()& ₍₁₎

Where Y is aggregate income, A represents technology, K is capital and L is the labour force, proxied by the working-age population (Bloom & Finlay, 2009). Alpha (𝛼) represents the input share of capital, whereas 1- α represents the input share of labour.

Dividing this equation by labour (L) leads to the following output/income per worker equation:

.

/ = 𝐴 01/231

/ (2)

𝑧 = 𝐴𝑘& ₍₃₎

Where and 𝑧 =._/ and 𝑘 =0_/

Equation 3 expresses output/income per worker as a function of capital per worker and worker productivity. In the next step of the model derivation, income per capita is related to income per worker (z), eventually referring back to this initial equation.

Following Bloom et al. (2010) output/income per capita, can be related to output/income per worker in the following way, by starting with an accounting identity that links output/income per capita Y/N, to

21

output per worker Y/WA, where WA represents the working age population (15-64 years) and N is the total population: 𝑌 𝑁= 𝑌 𝑊𝐴 𝑊𝐴 𝑁

This identity states that output/income per capita is equal to output/income per worker times the ratio of the working age population to the whole population (working age share).

Taking logs of the above produces the following equation: log 𝑌 𝑁 = log 𝑌 𝑊𝐴 log 𝑊𝐴 𝑁 One can further define y, z and w as follows

𝑦 = log 𝑌 𝑁 𝑧 = log 𝑌 𝑊𝐴 𝑤 = log 𝑊𝐴 𝑁 Which results in 𝑦 = 𝑧𝑤 Totally differentiating with respect to time leads to:

ẏ = ż + ẇ (4)

The growth rate of income per capita, ẏ, can be decomposed into the growth rate of income per worker (ż) and the growth rate of the working share of the population (ẇ), assuming a constant labour participation growth rate. The assumption of constant labour participation rate is important: some studies have included labour participation rates, such as Bloom and Canning (2003), yet the estimated effect is negative, contrary to expectations. Bloom et al. (2010) argue that due to the poor quality and lack of availability of labour participation data, the empirical results are generally weak and one should omit it. This is especially true for SSA countries,

22

where the data availability is even more limited. One simplifying assumption one could make is that increases in the labour participation will be captured by the growth in WAS.

Returning to the model, the focus will be on the growth rate if z (= ż). Neoclassical growth theory measures growth in terms of a convergence framework: economic growth is measured as the speed of convergence towards a so-called “steady-state” level of output/income per worker, defined as z*. This implies that the further an economy is from its steady state during the initial period, the faster the country will grow, or “converge” towards its steady state. As the growth rate of income per worker depends on the initial income per worker ( z0 ), one can

further define:

ż = 𝜆(𝑧∗_{− 𝑧}

E) (5)

where z* is the steady-state level of the log of output per worker and z0 is the initial level of

the log of output per worker. λ represents the speed of convergence towards the steady state. The growth rate, ż , is therefore defined by how far z is away from its steady state value. Variables that determine the steady-state level z* are captured by the vector X, leading to the following equation:

𝑧 = 𝜆 𝑋𝛽 − 𝑧E (6)

Where 𝑋𝛽=z*

Relating the initial income per capita to the initial income per worker:

𝑦E = 𝑧E+ 𝑤E → 𝑧E = 𝑦E− 𝑤E (7)

Combining this with (6) leads to:

𝑧 = 𝜆 𝑋𝛽 − 𝑦_E− 𝑤_E Substituting 𝑧 into (4) leads to the following equation:

23

Equation 8 relates the growth of income per capita (𝑦) to the growth of the working-age share of the population (𝑤), as well as to the initial level of WAS, 𝑤_E, the initial level of GDP per capita, 𝑦_E , and the set of variables that determine the steady-state level of output per worker, X. The set of variables included in X for the purpose of our regression equation, will be elaborated in the next section, complemented with the remainder of the variables in equation 8. Equation 8 motivates my empirical model and the regression equation that will be used in this study. The regression equation will be presented in the next section.

3.2 EMPIRICAL MODEL

This section will use the theoretical model that was derived above to motivate the empirical model.

One key question that growth literature concerns itself with is which variables to include in growth regression analysis. On the one side of the debate, scholars such as Sala-i-Martin (2004) and Islam (2008) propose a move away from monocausal economic growth theory, as they find multiple variables that have an effect on economic growth and therefore should be included. On the other hand, Durlauf et al (2008), criticises the “kitchen sink” approach, by which multiple regressors are included, without actually having to reconcile them with a specified empirical growth model or theory.

As much as an economic model that integrates all of the proposed variables in the following regressions would further add legitimacy to the results, the development of such a model is beyond the scope of this paper. Therefore I will substantiate the inclusion of variables by referring to general cross-country regression literature, which have yielded relatively robust results.

Durlauf et al (2008) considers 7 categories to group economic growth variables into: 1. Neoclassical Growth Theory (Mankiw et al. 1992; Solow, 1956)

2. Demography and Health (Shastry and Weil, 2003; Weil, 2005) 3. Macroeconomic policy(Barro, 1996),

4. Religion (Barro and McCleary, 2003; Durlauf, 2006), 5. Geography (Sachs and Warner, 1995, Sachs 2003)

6. Ethnic Fractionalisation (Alesina et al., 2003; Easterly and Levine, 1997) 7. Institutions (Acemoglu et al. 2001; Acemoglu and Johnson, 2005)

24

The variables that will be included in our empirical model are presented below by using the above-mentioned categorization.

Neoclassical Growth Theory

Neoclassical growth theory is at the heart of this model: The initial GDP per capita reflects the convergence hypothesis, as already discussed.

Education is also a factor often emphasised in standard growth theory. Average years of schooling will be used as a proxy for the quality and productivity of the labour force (Bloom D. , et al., 2010, p. 6). Whilst Bloom & Canning (2003) use average years of total schooling, Bloom et al (2010) use average years of secondary schooling. The change has not been justified and there seems to be no consensus on which variable is more appropriate, but I will follow Bloom’s latest paper and use average years of secondary schooling.

Demography and Health

The inclusion of demographic variables has been discussed. Some studies distinguish between fertility and mortality rates, but the growth of the working age share captures the effect more concisely, as it combines both variables. WAS will be used as the demographic variable in this analysis. Life expectancy will be used as a proxy for health of the labour force, as higher life expectancy is in general associated with overall better health and lower morbidity (Bloom & Canning, 2003; Murray and Chen, 1992; Murray and Lopez, 1997).

Macroeconomic Policy

In order to capture macroeconomic policy, I include trade openness and natural resource rents as explanatory variables.

A frequently used variable for openness is the Sachs and Warner index. However, their variable is a binary variable that does not adequately capture changes over time and that may not be sufficient. Even Bloom et al. (2006) suggest that it may be too crude to capture the effects of changes in trade policy that may be important. Therefore I use another measure for trade openness, namely the ratio of imports plus exports to GDP is also frequently used as a measure of openness (Bloom D. , et al., 2006). This is also not an ideal measure, as high performing economies tend to be more active in trade, and large countries tend to be more self-sufficient, yet it is arguably a more reliable variable than the Sachs and Warner index. Bloom and Canning (2003) also create an interaction term between trade openness and the WAS, indicating that this will be an important contributor to the growth regression.

25

I also add natural resource rents as a percentage of overall GDP. I expect that a low score will capture the diversification away from natural resources. This is especially important in the SSA context, as multiple countries still rely heavily on natural resource extraction and, as mentioned, resource exports have been suggested as a factor that has contributed to the recent growth turnaround.

Geography

To capture the differences in steady states across countries, I include a further set of control variables: Geographical variables are characterized by a climate variable and a geographic accessibility variable. The climate variable measures the percentage of a country’s land area that is located in tropical or subtropical climate zones, which is based on the Kӧppen Geiger classification of climate zones. The second geographical variable added is a dummy variable for whether a country is landlocked or not. Both of these variables are expected to enter with negative coefficients, negatively impacting economic performance, as in Rodrik et al. (2002) and Sachs (2003).

Fractionalization:

Bloom (2007, 2009) often includes ethno-linguistic fractionalization variables into his growth regressions, which will be followed in this thesis. This could be relevant in the African Lion countries, which are often linguistically and ethnically diverse. I add a fractionalization variable, which is a combination of ethnical and linguistic fractionalization indicators, that captures the degree of ethnic and linguistic diversity, as measured by Alesina et al. (2003). The expectation is that this will negatively impact economic performance.

Religion

I do not add a separate variable that captures religious differences, with the expectation that the ethno-linguistic fractionalization variable will partially pick up this effect.

Institutions

Institutional quality is a key variable in the analysis. I follow Bloom et al.’s (2007) definition for institutions, who define institutions as “a general term to include rule of law, efficiency of the bureaucracy, corruption, political freedom and expropriation risk, openness (political system, trade barriers, black market premium), freedom of political representation and freedom of speech”. I use a variable that is a slight adaptation of the Knack and Keefer index. I create

26

an index from the following components of the International Country Risk Guide (ICRG) published by the PRS group; Corruption, Law and Order, Quality of Bureaucracy and Government Stability. Institutional quality is an important variable in the subsequent analyses. I expect that the institutional quality index has a positive effect on economic growth, due to the evident impact of factors such as Law and Order and Government stability on economic growth.

As discussed, multiple studies have stressed that the demographic dividend is not an automatic phenomenon, but rather that functioning institutions needs to be in place to capture the dividend. This motivates the use of an interaction term between the institutional quality variable and the demographic variable, which is expected to be positive.

Combining all the above variables leads to the following regression equation:

yi,t = β0 + β1lnWASi,0 + β1∆WASi,t – β2lnyi,0 + β3TradeOpennessi,t + β4InsitutionalQualityi,t +

β5Schoolingi,t + β6Landlockedi + β7Tropicali + β8Resourcei,t + β9lnLifeExpectancyi,t + δt +εit

In this equation, yi,t represents the growth of GDP per capita, lnWASi,0 and lnyi,0 represent the

log of the initial values of GDP per capita and the working age share. The rest of the variables are self-explanatory. Finally, δt represents time dummies and εit represents the error term. Lags

and interaction terms are not presented this regression equation, but do feature in the estimation. The interaction term is constructed by interacting the institutional quality of a country with the growth of the working age share of the population. The interaction term is added in order to ascertain the effect of institutions on capturing the demographic dividend.

3.3 DATA DESCRIPTION

This section outlines the data used for the regression analysis.

A country level panel from 1970-2014, including 84 countries and using 5-year averages, is constructed. (A full list of countries can be found in the appendix). Although the region of interest is SSA and the main time period of interest is from 1990-2014, other countries, at various stages of their demographic transitions, and over longer time periods, are included as well. This is so that the dataset includes as much variation as possible and includes effects that may not be picked up in the SSA sample. From these results one can then evaluate the estimated

27

contribution of demographic changes on economic growth over the 1990-2014 period. Additionally, these estimates will be used to estimate out of sample country results. One main concern is missing data or complete lack of data availability for several countries. In the case of the institutional quality variable, data was only available until 2009. Data until 2014 was extrapolated, using the previous 5 year average as an estimation. I include time dummies in all model specifications in order to control for any time variant shocks across sample countries. However, country fixed effects are not added as they would exclude some dummy variables and are not ideal when dealing with other slow moving variables, as is the case in this study (Kohler & Kreuter, 2009, p. 245)

Table 1: Descriptive Statistics 1970-2014 for 84 countries

Variable Summary statistics across 756 observations Mean St. Dev. Min Max Growth of GDP, 5 year averages 1.765 3.273 -37.7566 19.0977 Growth of WAS, 5 year averages 0.26434 0.448 -.997 2.555 Average years of schooling

(secondary)

2.058 1.37 .03 6.84

Average years of schooling (total) 6.37 3.02 .29 13.18 Life Expectancy 65.97 11.2 32 83.5 Trade Openness .6833 .5176 .054 3.8914 Resource Rents 5.763 8.3571 0 64.42 Institutional Quality 16.58 5.06 3.132 27 Ethno-linguistic Fractionalization .8483 .5496 .00411 1.852 Landlocked .1547 .361 0 1 Tropics .510 .473 0 1

28

4. EMPIRICAL RESULTS & ANALYSIS

The first section of this chapter will discuss and analyse the empirical results, whilst section 4.2 presents the estimated impact for the Lion economies.

4.1 DISCUSSION AND ANALYSIS

Using the above-mentioned regression equation, the strategy of estimation was to start parsimoniously and add variables one by one to establish their explanatory value. We first apply OLS and then IV in our estimations. The results are presented in Table 1.

The first regression (column 1) excludes all but 1 demographic variable (initial WAS) and just captures the neoclassical variable effects. Column 1 includes only geographical features into the model. Being landlocked and being in a tropical location both negatively impact economic growth, as per expectation. However, the coefficient of landlockedness is insignificant, which is contrary to findings by Rodrik et al. (2002) and Sachs (2003).

Overall, the first regression results are broadly in line with expectations: the coefficient of initial GDP is negative and significant, exhibiting the convergence effect. This implies that the further away countries are from their steady state, the faster they grow economically. The initial value of the WAS is positive but insignificant.

In column 2, human capital indices and economic variables are added. In line with expectations, the coefficient of institutional quality enters positive and significantly, indicating that the quality of institutions positively impacts economic performance. As expected, the coefficient of trade openness also enters positively and significantly, indicating that higher trade volumes are associated with increases in GDP per capita growth. Furthermore, the dummy for landlockedness is now positive, but still insignificant, indicating that the trade openness factor could have absorbed some of this effect. One could argue that landlocked countries trade less, and therefore the trade openness variable includes this.

As per expectation, life expectancy enters positively and significantly, indicating that the proxy for a healthy workforce affects economic growth positively. The estimation result of the schooling variable is contrary to expectations. The coefficient of average years of secondary schooling appears negative and significant, opposite to what several other studies, such Cuaresma, Lutz and Sanderson (2014) have found6. I included various schooling variables yet

29

none were found to be in line with expectations or statistically significant in any estimation results. Although this may be counterintuitive, as education is expected to be an important variable for estimating economic growth, this finding is not uncommon in the literature. One explanation is the frequency of measurement errors in schooling (see Krueger and Lindahl (2001)).

The ethno-linguistic fractionalisation index does have a negative coefficient as per expectation but it is statistically insignificant. The coefficient for resource rents is positive, yet insignificant. In column 3 the growth of working age share is added. The results confirm the hypothesis that age structural changes impact economic growth: the coefficient appears positive and strongly significant. It also improves the explanatory power of the model, as shown by the increase in the R-squared level.

Simultaneity/Endogeneity is a concern with these types of models (Bloom & Canning 2014). One could expect that periods of high economic growth having feedback effects on demographic changes. To control for simultaneity between economic growth and the growth in WAS, the WAS should be instrumented. Due to no credible external IVs being found, the lag of the growth of WAS will be used as an instrument for the growth of the WAS. This is a reasonable assumption due to the slow changing nature of the variable, which means that the variable is predetermined in this model. This also means that the first 5 year period 1970-75 is consumed by the construction of the IV instruments in the IV estimations. The IV estimation results are presented in column 4. The IV estimate for WAS? is significantly larger than the OLS estimates.

Finally, in column 5 I test whether the positive effect of the WAS on economic growth is contingent on the institutional quality of a country, as proposed by Bloom and Canning (2003) and Bloom et al. (2007). I therefore interact the growth of WAS with the institutional quality measure. As the term includes the contemporaneous growth rate of WAS, there may be an underlying endogeneity problem, as discussed earlier. To control for this, I instrument the interaction term with its lagged value. The interaction term appears positive and strongly significant, whilst the growth of WAS is no longer significant. This indicates that institutional quality is imperative in capturing the demographic dividend. This will be important for the analysis in the next section, as SSA economies tend to have lower values for institutional quality in general.

30

GMM estimation was also utilised but did not yield more efficient or conclusive results. This has also been confirmed by Drummond et al. (2014) who use a similar dataset and estimation techniques.

31

Table 2: Estimation Results

ESTIMATION RESULTS : REAL GDP PER CAPITA GROWTH

(1) (2) (3) (4) (5)

OLS OLS OLS IV IV

VARIABLES Growth of WAS 1.150*** 2.621*** -0.665 (0.310) (0.519) (0.797) Institutional Index 0.158*** 0.147*** 0.133*** 0.127** (0.0392) (0.0396) (0.0416) (0.052) Trade Openness 0.489** 0.525** 0.527*** 0.160 (0.220) (0.221) (0.200) (0.1873) Resource Rents 0.000292 -0.00482 -0.0161 -0.0119 (0.0539) (0.0541) (0.0294) (0.031) Average years of Secondary Schooling -0.600*** -0.518*** -0.377** -0.509*** (0.115) (0.121) (0.173) (0.184) Fractionalisation Index -0.370 -0.344 -0.223 0.155 (0.276) (0.271) (0.298) (0.304)

Log of Initial Life Expectancy 8.516*** 6.910*** 4.774*** 8.131*** (1.669) (1.750) (1.605) (1.514) Landlocked Dummy -0.560 0.238 0.231 0.179 0.408 (0.348) (0.321) (0.321) (0.307) (0.385) Tropical Dummy -0.973*** 0.293 0.283 0.0539 0.336 (0.335) (0.352) (0.345) (0.300) (0.346) Log of initial GDP -0.533*** -1.257*** -1.266*** -1.316*** -1.227*** (0.168) (0.226) (0.225) (0.190) (0.229)

Log of initial WAS 2.874 3.116 6.098** 9.809*** 6.17**

(1.876) (2.098) (2.444) (2.025) (2.275)

Interaction Term: Institutional Quality and Growth of WAS

0.116*** (0.0264) Constant -2.079 -35.62*** -41.07*** -48.51*** -50.17*** (6.594) (10.70) (10.96) (8.146) (9.484) Observations 756 756 756 672 588 R-squared 0.089 0.229 0.245 0.253 0.2237 Number of countries 84 84 84 84 84

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

32

4.2 ESTIMATED GROWTH IMPACT: LION ECONOMIES

This section uses the coefficients from the regression results to estimate the contribution of WAS on economic growth in African Lion economies, contingent on institutional quality. More specifically, the interaction term of institutional quality and WAS from column 4 (0.116) and the growth of WAS (-0.665) are used to construct the estimated economic contribution of age-structural changes for the period of 1970-20507, by applying them to the WAS and institutional quality data of African Lion economies, which is available for all of them. More specifically, the coefficients will first be used to measure the estimated impact of the growth of WAS, contingent on institutions, for the period between 1990-2014. The coefficients will then be used to construct a forecast for the estimated impact of the growth of WAS for the period 2015-2050, holding all other variables constant.

One important caveat arises: for the forecasted contribution of WAS from 2015 onwards, contingent on institutional quality, I used an extrapolated value for institutional quality from 2012 up to 2050. Institutional quality, already a contentious variable (Bloom, Canning, & Fink, 2007, p. 17), is difficult to measure and even harder to forecast. The future institutional quality variable was constructed by using previous 5 year averages to create a projection. This means that the forecasted contribution of WAS is interacted with a static institutional quality variable. This indicates that these results are dependent on the evolution of institutions in these countries, a factor that my estimation does not include. However, one could argue that it is a relatively slow moving variable, and therefore, although not ideal, this variable is still applicable. For the sake of simplicity, both the current estimated contribution of the WAS on economic growth, as well as the future projection, will be presented in one graph for each Lion economy. As previously mentioned, for all estimates the UN Population Prospects medium variant values are applied.

33

South Africa

As discussed, South Africa’s demographic transition is already at a more advanced stage compared to other Lion economies. This is also represented by our estimation values. The contribution of WAS changes on economic growth were already apparent in the early 1970s, yet the combination of an increase in institutional quality and higher growth of the working age share lead to a peak in the estimated contribution (1,25%) in the late 80s and early 90s. This confirms the hypothesis that changes in age-structures can be considered a contributing factor to economic performance over this period. The decrease in early 2000s is partially due to a slight decrease in institutional quality as well as sustained decreases in the growth of WAS. Most of the demographic window of opportunity has already surpassed South Africa, although small gains could still be experienced until 2045 (contingent on institutions), after which the decreasing growth of the working age share will represent a drag on the economy.

Ethiopia

Ethiopia, in comparison to South Africa, is still in its early stages of the demographic transition, which is also characterized by the estimated contribution below. The combination of lower institutional quality and growing dependency ratios (and therefore negative WAS growth rates)

34

lead to a drag on the economy until the early 2000s. With an increase in institutional quality and positive WAS growth rates, the estimated contribution turned positive in the early 2000s, again confirming our hypothesis that age-structural changes contributed to the growth turnaround. However, this estimated contribution is smaller than South Africa’s, due to the overall lower institutional quality scores.

In terms of the estimated future contribution, Ethiopia will experience its potential demographic dividend peak, with a positive contribution persisting past 2050. Should Ethiopia make significant progress in terms of institutional quality, these gains could be considerably larger.

Tanzania

Tanzania’s estimated contribution is somewhat disjointed: whilst small gains were made in the early 1980s and 1990s, this decreased for much of the 2000s, with a recent sharp recovery. However, this seemingly haphazard trend broadly follows Tanzania’s working-age share pattern. This leads to an ambiguous result regarding the hypothesis.

Regarding future projections, with a peak contribution at only 0.2% in the 2020’s the overall estimated contribution seems underwhelming, although continued gains can be expected past 2050, well into the second half of the century.

35

Nigeria

Nigeria’s estimated contribution shows a somewhat irregular pattern, similar to that of Tanzania. Whilst the values up until the 1990s represent a drag on the economy, the 1990s seems to usher in a period of positive estimated contribution, yet this suddenly decreases in the middle of the 2000s, recovering again by 2014. Again, this broadly tracks the evolution of WAS growth in Nigeria over this period. The lacklustre estimated contribution to GDP per capita growth can be attributed to Nigeria’s persistently low institutional quality score. If no significant gains are made, it is unlikely that Nigeria will be able to reap much of its potential demographic dividend, even though the estimated contribution stays positive well after 2050.

Ghana

Ghana’s demographic transition, somewhat similar to South Africa’s, began earlier than other Lion economies. Whilst the early 1970s still saw a negative estimated contribution, the second half of the decade saw a positive and growing contribution. Ghana’s estimated contribution also saw strong growth over the 1990-2014 period, lending support to the hypothesis that some of the gains in economic performance can be explained by age-structural changes. Incidentally, Ghana also exhibited strong improvements in institutional quality over this period. The rather dramatic decrease in 2015 and the subsequent increase follows Ghana’s WAS growth trend during this time.

36

In terms of future prospects, the estimated contribution of WAS, contingent on institutional quality, will be small, but positive past 2050.

37

5. DISCUSSION & CONCLUSION

This thesis studied the estimated contribution of age structural changes on economic growth in African Lion economies. Various factors have been proposed as explanations for SSA’s growth turnaround in the 1990s and 2000s, yet no literature was found that examines demographic changes in conjunction with this recent change in the growth trajectory.

Even on a more global level, empirical studies emphasising the impact of demographic changes on economic growth are relatively recent in the economic growth discourse. Yet many of these studies confirm a significant impact of age-structural changes on economic performance. Whilst most of the world is set to experience aging populations over the next century, and therefore the focus will increasingly move towards the impact of aging populations, Sub Saharan Africa will instead be faced with some of the youngest populations in the world. In an attempt to move away from treating the 48 nations of SSA as a homogenous group, this thesis aimed to estimate the contribution of the early demographic transition that a sub-group of SSA nations are experiencing, and what possible gains may be contingent on. This was done in the form of a cross country growth analysis for a panel of 84 countries over the period from 1970-2014. OLS and IV were used as the regression estimation methods.

The findings of this thesis are broadly in line with other studies: the growth of the working ages share is an important contributing factor to economic performance in this period for the cross country growth regression results. Economic growth is further enhanced by good institutions, the degree of openness of trade, as well as the life expectancy (proxied for health) of the population. Importantly, this thesis found that the potential gains from the increase in the working age population are highly contingent on the institutional quality of a country, which means that it largely determines how large the contribution of WAS could be. The higher the institutional quality, the more a country is able to capitalize on the potential economic gains of the demographic transition. This is in line with the findings of other research (Bloom and Canning (2003) and Bloom et al. (2007)).

In the context of African Lion economies, this is a significant limitation in terms of the potential gains of the demographic changes that these countries are set to undergo. Institutional quality is far lower than the rest of the world, and even though improvements have been achieved, the