Explaining and forecasting the impact of demographic changes on economic growth in the BRICS countries

(1)

Explaining and Forecasting the Impact of Demographic Changes on Economic Growth in the BRICS countries

Author: Supervisor:

Stijn J. Allema Drs. Naomi J. Leefmans

6119808 Second Reader:

stijnallema@outlook.com Dr. Kostas Mavromatis

(2)

Statement of Originality

This document is written by Stijn Allema who declares to take full responsibility for the contents of this document.

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

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

(3)

1. Introduction

The world is in the midst of a major demographic transition, global population growth is slowing down and the age structure of countries (that is, the way in which the population is distributed across different age groups) is changing. Although population growth worldwide is declining, growth rates in developing countries are still relatively high. Besides that, the global share of the young is falling and the elderly dependency ratio is increasing. During recent years there has been an increasing awareness of a direct influence of the age structure of the population on the macro economy (Prskawetz et al. 2007). Since the economic needs and contributions of people vary at different stages of their life, the question arises what effects changes in a country’s age structure can have on economic growth.

This thesis will focus on five major developing economies, also known as BRICS: Brazil, Russia, India, China and South Africa. Together, these countries account for about 42% of the world population and 22% of the world’s gross domestic product (GDP)1. The acronym BRIC was coined in 2001 by Jim O’Neill from investment bank Goldman Sachs in a paper entitled ‘Building Better Global Economic BRICs’ to emphasize the economic growth potential of the countries Brazil, Russia, India and China in the coming decades. After the BRIC countries formed a political organization among themselves, they later expanded to include South Africa in 2010, becoming the BRICS.

Across three continents these newly industrialized countries have experienced faster economic growth compared with the advanced economies over the last two decades (see Table 1). Many factors have been cited as contributing to this growth, including increased fuel

Table 1. Historical and projected GDP growth rates for the BRICS countries

Data source: IMF Staff calculations, World Economic Outlook update January 2016

1

Source: The World Bank, based on total population and GDP at current US$ from the World Development

(5)

and manufacturing exports, stronger educational systems, improved institutional quality, increased FDI inflows and the large amount of land area (Goel and Korhonen, 2011). One explanation that has been insufficiently examined centres on the link between demographic change and economic growth. It is argued that the changes in the demographic structures of these countries have undoubtedly played an important role in their recent success. However, the projections in Table 1 show that for Brazil, Russia and South Africa economic growth is slowing down and forecasted to be lower than the growth in the advanced economies. For 2016 the projections for Brazil and Russia are even negative. Also China’s economic growth is forecasted to be markedly lower, but still rather high. Only for India the projected growth rates are higher than the average growth rate of the past two decades. Part of the explanation for the expected fall of growth rates might be changes in the demographic structure, such as a declining labour force and an ageing population.

A commonly used indicator in the analysis of the demographic structure of a country is the total dependency ratio, which is computed as the ratio between the dependent population (aged under 15 or above 64) and the active population (aged between 15 and 64). The lower the dependency ratio, the less is the dependency burden to an economy. As shown in Figure 1, the BRICS countries have all moved from high to lower ratios in the last decades

Figure 1. Dependency ratios for the BRICS countries

Data source: United Nations, Department of Economic and Social Affairs, Population Division (2015). World Population Prospects: The 2015 Revision.

(6)

in which they have experienced high economic growth. In the past, the large share of the population under age 15 drove the high dependency ratio. As they entered the labour market after reaching the working-age, this boosted labour supply, moderated wage pressures and contributed to rapid economic growth. This contribution to economic growth is usually referred to as the demographic dividend. In Figure 2 it is shown that between 1970 and 2015 all five countries experienced growth in the share of their working-age population (growth rates shown are five-year growth rates). In contrast, in the next decades, the rise of the dependency ratio will be entirely dominated by a large increase in the share of the population in older ages (aged above 64). Figure 2 shows that from 2005 onwards the growth rates of the share of the working-age population are decreasing and will be negative for all five countries by 2050. This research will investigate the impact these demographic changes have had on economic growth in the past and will have in the future.

Figure 2. Growth rates of the share of the working-age population for the BRICS countries

Recent empirical studies that concentrate on the inclusion of demographic variables into economic growth regressions find that age distribution matters (Bloom and Williamson, 1998; Bloom, Canning and Malaney, 2000; Kelley and Schmidt, 2005; Young, 1994,1995). This is a useful addition to the existing economic literature since earlier research focused on the pure impact of population growth on economic growth (Boserup, 1981; Coale and

(7)

Hoover, 1958; Ehrlich, 1968; Kuznets, 1967; Simon, 1981). Young (1994, 1995) argues in a detailed study of national accounts that most of the economic growth in the East Asian NICs2 was due to the accumulation of factor inputs, particularly labour, rather than pure productivity gains. He finds that demographic changes account for up to 2.6 percentage points of growth of output per capita in the East Asian NICs in the post-war period (1970-1993). Bloom, Canning and Malaney (2000) claim the demographic dividend accounts for nearly a third of the extraordinary growth in East Asia during the period 1965-1990. In line with these results Bloom and Williamson (1998) find that 1.4 to 1.9 percentage points of the ‘East Asian Miracle’3 can be attributed to the fact that the working-age population was growing faster than the dependent population. They conclude that economic growth will be less rapid when the growth rate of the working-age population falls short of that of the population as a whole, and economic growth will be more rapid when the growth rate of the working-age population exceeds that of the entire population. Bloom and Canning (2003) show that a dramatic rise in the share of the working-age population had the effect of accelerating Ireland’s economic growth, although these results are not as stunning as in East Asia.

In this thesis we review the links between demographic changes and economic growth and explore the extent to which it can help us understand the economic performance of the BRICS countries over the last decades. To conduct this study the following research question is formulated: “To what extent have changes in the age structure been responsible for the growth rates of income per capita in the BRICS countries in the period 1972-2007 and what will be the consequences of future demographic changes for income per capita growth rates in these countries?” First, we estimate the determinants of real GDP per capita growth for the period 1972-2007 in a cross-country growth model for a panel of 74 countries including the BRICS. Our empirical study is based on the theoretical convergence framework that will be outlined in subsection 3.1. In our regressions demographic variables like the growth rate of the working-age population, growth rate of the labour force participation rate and several other control variables will be included as explanatory variables for economic growth. As estimation methods we use ordinary least squares (OLS) and the method of instrumental variables (IV) to control for the potential endogeneity of the demographic variables. We expect that changes in the demographic structure will have a significant impact on economic growth. Next we compare the fitted growth rates from our econometric model with the actual

2_{East Asian New Industrialized Countries: Hong Kong, Singapore, South Korea, Taiwan} 3

The so-called East Asian Miracle refers to the East Asian NICs growth rates in the period 1965-1990. On average these countries experienced GDP per capita growth rates of 6 to 7% per annum in this period.

(8)

growth rates from the BRICS countries. Subsequently, we can determine the impact demographic changes have had on economic growth in the BRICS countries, after decomposing the fitted growth rates into three components, one of which is the estimated effect of changing age structure. Finally, we will forecast the future impact of demography on economic growth in these countries using the obtained regression coefficients from our econometric model and population prospects from the United Nations4.

Until now little can be found in the economic literature about the impact of demographic changes on economic growth in the BRICS countries. In their study on the contribution of population health and demographic change to economic growth in China and India Bloom et al. (2010) find that the major drivers of the economic growth take-offs in these two countries are improved health, increased openness to trade, and a rising labour force-to-population ratio due to fertility decline. This rising labour force-to-force-to-population ratio is associated with an increasing working-age population. The only study found for Brazil focuses on regional economic growth in Brazilian states and the results from this study suggest that demographic change matters for regional income convergence in Brazil (Porsse et al., 2012). Demographic changes in South Africa are highlighted in some studies on Sub-Saharan countries (Bloom, Canning et al., 2007), but a specific study for this country cannot be found. The author did not come across any economic study based on demographic changes for Russia.

In India and South Africa the share of the working-age population will continue to expand until the middle of the current century. The Chinese and Russian share of the working-age population is declining as from 2010 onwards, while for Brazil this decline will start around 2020. These labour supply dynamics seem favourable for India and South Africa and therefore in our forecasts we expect positive demographic contributions to economic growth for these countries, while for China, Russia and Brazil these contributions could become negative.

If the BRICS economies can maintain their economic growth path of the last 25 years, they could within decades jointly become larger than the G65 (measured in total GDP). Growth patterns in the Japanese economy over the past 40 years, however, offer an alternative prospect. The annual average growth rate in Japan was 5.3% between 1960 and 1990 but only

4_{United Nations Population Division, World Population Prospects the 2015 Revision.}

5_{The Group of Six or G6 is}_{a government political forum of the world's (former) major industrialized countries}

(9)

1.5% during the 1990s. Some researchers (Hewitt, 2003) attribute the sustained stagnation of the Japanese economy to its rapidly ageing population and inadequate pension system. The dependency ratio in Japan first declined and then rose again, suggesting that, as the population aged, Japan lost an additional source of economic growth (Cai and Wang, 2005). Since the dependency ratios of the BRICS countries show a similar pattern (see Figure 1), forecasting the impact of demographic changes on GDP per capita growth in the BRICS countries will be an interesting contribution to the existing economic literature.

This thesis is structured as follows, section 2 consists of a literature review. The literature review aims to clarify the effects of demographic changes on economic growth and gives an overview of the empirical studies performed on this topic. Section 3 will present the model and methodology used in this thesis. Section 4 contains the empirical results and their analysis. Finally section 5 will provide a conclusion.

2. Literature Review

The purpose of this section is to provide an overview of the literature dealing with the topic of demographic changes and their impact on economic growth. The first subsection will focus on the theoretical background and aims to clarify how demographic changes are related to economic growth for a specific country. Subsequently, the second subsection will explain the different phases of demographic transition the BRICS countries are currently in. Finally, the third subsection will provide insights on the results and the methodologies used in the existing empirical literature in order to form the base for the model and methodology used in this thesis.

2.1 Theory: The Demographic Transition Process

The debate on the impact of population growth on economic growth is long and profound. Pessimists believe that rapid population growth is immiserizing because it will tend to overwhelm any induced response by technological progress and capital accumulation (Coale and Hoover, 1958; Ehrlich, 1968). Optimists believe that rapid population growth allows economies of scale to be captured and promotes technological and institutional innovation (Boserup, 1981; Kuznets, 1967; Simon, 1981). Recent research however shows that in numerous studies using cross-country data a statistically significant association between the growth rates of population and of per capita output cannot be found (Barro, 1997; Bloom, Canning and Malaney, 2000; Bloom, Canning and Sevilla, 2003a). This neutralist view that

(10)

population growth neither systematically impedes nor promotes economic growth has been the dominant academic belief ever since.

One of the shortcomings in these conventional economic theories of growth is that they only focus on the magnitude and growth of population, while neglecting changes in the age structure which may affect economic growth (Williamson, 1998; Bloom and Canning, 2003). Yet these changes are arguably as important as population growth. Therefore, it is worthwhile to decompose the population growth rate into birth and death rates and investigate the effects of changes in these rates. The birth (death) rate is the total number of births (deaths) per 1000 of a population in a year.

The process from high birth and death rates to low birth and death rates as a country develops from a pre-industrial to an industrialized economic system is known as the demographic transition. Figure 3 presents a stylized view of this process. Declines in death rates, especially in infant and child mortality, mark the beginning of almost all demographic transitions. After a lag, birth rates also begin to decline indicating the next phase of a demographic transition. The population growth rate is implicitly shown as the difference between the birth and death rates and as shown increases temporarily (Bloom and Williamson 1998).

The result of this temporary increase in population growth, shown in Figure 4, is a ‘bulge’ in the age structure, a ‘demographic wave’ that works its way through the population. First the youth dependency increases, subsequently the working-age population increases and finally the elderly dependency increases. A demographic transition could affect economic growth through these changes in the age distribution. A high child and elderly dependency increases the burden on society of dependents, therefore negatively affecting economic growth. Similarly, when the working population is relatively larger, the population structure is more productive because labour supply and savings rates are larger (Bloom and Canning, 2005; Cai and Wang, 2005). The demographic transition also has significant effects on investments in human capital; effects that are the least tangible, but may be the most significant and far-reaching (Bloom, Canning and Sevilla, 2003a). Contrary to the neutralist view, the emerging evidence indicates that population does matter to economic growth, with age structure playing a central role. As the share of working-age to non-working-age people increases, opportunities for economic growth tend to rise, creating what is now referred to as a ‘demographic dividend’ (Bloom and Canning, 2003).

However, it should be mentioned that this so called demographic dividend in the middle phase of the transition provides economic growth potential whose realization depends

(11)

Figure 3. Demographic Transition Process

Data source: Bloom & Williamson (1998)

Figure 4. Population Growth and Changes in the Age Structure

Data source: Bloom & Williamson (1998)

on the social, economic and political environment. A focus on the policy environment will maximize the chance of capturing a demographic dividend in a country. Factors and policies that may help bring about, or increase the size of, the demographic dividend include quality of governmental institutions, labour legislation, macroeconomic management, trade policy and education policy (Bloom, Canning and Rosenberg, 2011). To capture this effect in our growth regression an interaction term between the growth rate of the share of the working-age population and a quality of institutions variable will be introduced.

(12)

2.2 Demographic Transition in the BRICS countries

The BRICS countries are currently in different stages of their demographic transition. During the investigated period, all five countries have experienced periods of growth of their working-age population share, although this growth is less pronounced in Russia than in the other four countries. In the 1970s, declines in mortality in China, which included large declines in infant and child mortality, led to large cohorts of young people. Subsequent declines in fertility, mainly due to family planning policies, produced a bulge generation (Bloom et al., 2010). These declines in mortality and fertility coincide with the declining birth and death rates shown in Figure 5. The rapid fall of the birth rate triggered a subsequent sharp rise in the ratio of working-age to non-working-age people, corresponding with the middle phase of the demographic transition process. The share of the working-age population in China has been growing vastly, but peaks around 2010 at 74% and declines from there on as shown in Figure 5. In the nearby future a decreasing working-age population associated with an increasing elderly dependency ratio could possibly have a significant negative effect on economic growth in China.

Figure 5. Demographic Transition and Population Distribution by Broad Age Group - China

Brazil has experienced significant changes in their demographic structure over the last decades. The substantial decrease in birth and death rates has contributed to boosting the demographic transition, although it is argued that Brazil has struggled to fully exploit their demographic dividend. As shown in Figure 6, the increase in the share of the working-age population accelerates from 1970 and is projected to continue until around 2020. This

(13)

acceleration is associated with a sharp decline in the share of the child population. The ageing of the Brazilian population tends to become more prominent from 2020 onwards and will govern the dependency ratio from there on (Porsse et al. 2012).

Figure 6. Demographic Transition and Population Distribution by Broad Age Group – Brazil

Data source: United Nations, Department of Economic and Social Affairs, Population Division (2015). World Population Prospects: The 2015 Revision

The demographic transition process in Russia is a different story. Similar to China, the share of the working-age population peaks around 2010 and starts declining from there on, as shown in Figure 7. However, this peak has not been preceded by a sharp rise, as was the case in China. While in most BRICS countries life expectancy has been increasing during our investigated period, an exception has been the mortality pattern in Russia after the dissolution of the Soviet Union. Since the early 1990s Russia saw a precipitous drop in life expectancy and death rates became higher than birth rates, as shown in Figure 7. This decline in life expectancy corresponded to roughly 1.5 million premature deaths, which affected

working-age men more than any other demographic group (Bloom, 2011). These demographic changes

can be linked to the downfall in economic growth for this similar period (1993-1997). By

2009, life expectancy had recovered to its initial level. However, with a decreasing working-age population the demographic future for Russia does not look promising.

As mentioned above China, Brazil and Russia are currently approaching a more advanced stage of the demographic transition process. Therefore, most of their economic growth potential due to demographic changes lies behind. This is not the case for India and South Africa for which most of their demographic dividend lies ahead. Although infant mortality in India has been declining, it is currently still twice the size of that of China and Brazil. As shown in Figure 8, birth rates began declining in India roughly at the same time as

(14)

Figure 7. Demographic Transition and Population Distribution by Broad Age Group – Russia

in China and Brazil, but the decline has been much slower. The fact that the ratio of working-age to non-working-working-age people is currently the same as it was in Brazil in the late 1990s and in China in 1985 illustrates this slower change. The share of the working-age population continues to grow until the middle of the current century. Since the demographic transition takes place more slowly, the greater portion of the potential demographic dividend in India still lies ahead (Bloom, Canning and Rosenberg, 2011).

Figure 8. Demographic Transition and Population Distribution by Broad Age Group – India

Similar to India, South African birth and death rates have been declining since the 1970s, but birth rates remain relatively high compared with the rest of the world. After a steady decline, South Africa saw its death rate increase again after the abolition of apartheid in the 1990s, as shown in Figure 9. In their growth outlook for Sub-Saharan countries Bloom,

(15)

Canning et al. (2007) are very positive about South Africa as current regional leader in terms of their institutional quality, even though their prospects for profiting from a demographic dividend until 2025 are relatively small since the share of the working-age population is expected to grow only very moderately. Though similar to India, their share of the working-age population will continue to grow until around 2050.

Figure 9. Demographic Transition and Population Distribution by Broad Age Group – India and South Africa

2.3 Impact of Demographic Changes on Economic Growth

Numerous studies have been conducted on the effect of demographic changes on economic growth. The most widely used theoretical starting point for analysing these effects is the convergence or technology-gap framework, which is rooted in the neoclassical growth theory of Solow (1956). As Kelley and Schmidt (2001,2005) point out, this model is highlighted as the empirical framework under which the demographic effect on economic growth can be fully demonstrated.

Various demographic variables have been included in empirical cross-country growth studies. Amongst them, demographic structure, represented by the dependency ratio or the proportion of the working-age population to the total population, is generally regarded as the key variable for exploring economic implications of demographic change on growth (Wei and Hao, 2010). According to Bloom and Canning (2001), this variable can measure demographic effects on the labour market more directly than crude birth and death rates.

Since most empirical studies focus on the demographic transition that occurred in Asia, this literature review starts with an overview of these studies. Next, the existing empirical

(16)

literature on the BRICS countries will be reviewed. Finally, an interesting study on sources of income growth in Ireland will be presented.

2.3.1. Empirical Literature – Asia

One of the first studies that estimated the impact of demographic changes on economic growth is an empirical growth accounting study of Young (1995). In this study he tries to explain the extraordinary post-war growth of the East Asian NICs. His results suggest that although these countries experienced fast economic growth in this investigated period (around 6 to 7% per annum), productivity growth has not been extraordinarily high in the period 1970-1993. Instead, demographic changes like rising participation rates mainly due to declining birth rates accounted on average for 1.5 percentage points of the growth rates per annum.

Unlike the research of Young (1995), most econometric analyses that investigate the demographic transitions that occurred in the Asian economies are based on the earlier mentioned theoretical convergence framework. Bloom and Williamson (1998) base their cross-country growth regression on 78 Asian and non-Asian countries. The period analysed runs from 1965 to 1990. Bloom, Canning and Malaney (2000) investigate the demographic changes that took place in East Asia using panel data for the five consecutive five-year periods spanned by the years 1965-1990. In our analysis we will use five-year averages instead to avoid short-term disturbances, business cycle fluctuations and serial correlations (Islam, 1995).

Bloom and Williamson (1998) use the growth rate of real GDP per capita in purchasing power parity as dependent variable and by using ordinary least squares (OLS) they confirm the neutralist finding that there is no significant relationship between the level of population growth and GDP per capita growth. They illustrate that this conclusion fails to pay any attention to the sources of the population growth, and to the stages of the demographic transition. Therefore the growth rate of the working-age population enters the regression and the coefficient of this demographic structure variable is positive, statistically significant and big: a one percent increase in the growth rate of the working-age population is associated with a 1.46 percent increase in the growth rate of GDP per capita. In contrast, the effect of the total population growth is insignificant and negative. Bloom, Canning and Malaney (2000) find that the demographic dividend accounts for a third to a half of the sustained high economic growth in East Asia and that changes in life expectancy have had a significant impact on economic growth rates in Asia. Bloom and Finlay (2009) reconfirm these results with a similar approach, but for a longer time period 1965-2010.

(17)

Next to demographic variables several other determinants of economic growth are included in these studies, like initial income per capita, life expectancy, schooling, openness to international trade and institutional quality. Also geographic factors are added such as the percentage of land area located in the tropics and whether or not a country is landlocked (see for example, Barro 1991; Barro and Sala-i-Martin 1995). Life expectancy is used as a proxy for the health of the workforce and as Bloom, Canning and Malaney (2000) point out it is an important determinant of economic growth. It is quite common that schooling variables lack significance in cross-country growth regressions. This may be due to measurement error in the schooling variables, as discussed by Krueger and Lindahl (2001) or may reflect a real lack of impact, as suggested by Pritchett (2001). The government variable used in most studies is based on the index created by Knack and Keefer (1995), which gives an average indicator of the quality of institutions. Bloom, Canning and Malaney (2000) deliberately omit savings and investment rates as explanatory variables. As Bloom and Williamson (1998) argue, investment might be endogenous and thereby its effects on growth might have already been explained by demographic changes especially through the mechanism of savings. Wei and Hao (2010) do include the investment ratio in their empirical growth equation in their study on economic growth in China but find the estimated coefficient to be statistically insignificant.

Although in the above-mentioned studies time dummy variables are used to account for worldwide shocks, country fixed effects are not included, because the necessary presence of the initial income level would then result in inconsistent (biased) coefficient estimates. While this problem can be overcome (see Islam 1995), the techniques required involve instrumenting all the explanatory variables, with a consequent large loss of precision in the estimates.

In economic growth studies causality is a serious issue. Because all explanatory variables are determined to some extent by income level, there is the potential for a strong feedback from growth to the right-hand side variables. To account for the possibility of reverse causality between (working-age) population growth and economic growth, instrumental variables (IV) are used as an extension to the OLS analyses. Bloom and Williamson (1998) use the average growth of the population from 1950-1960, the percentage of the urbanized population in 1965 and population policy variables as instruments. The test statistics of their Hausman specification tests (Hausman, 1978) suggest that the null hypothesis that the IV and OLS estimates are statistically equivalent cannot be rejected. Bloom and Williamson (1998) interpret this result as meaning that there is no endogeneity

(18)

problem. Bloom, Canning and Malaney (2000) and Bloom and Finlay (2009) use the growth rates of the working-age and total population during 1960-65 and the 1965 fertility and infant mortality rates as instruments.

The key idea in the method of determining the direction of causality is that variables from before each 5-year growth period can be treated as predetermined and exogenous (Bloom et al., 2010). However, the growth in the share of the working-age population is contemporaneous with the economic growth being explained and as such possibly endogenous. To try to control for this potential endogeneity we will use the lagged growth rates (from the previous five-year period) and the beginning-of-period fertility rate as instruments.

2.3.2 Empirical Literature – BRICS Countries

In a study on the contribution of population health and demographic change to economic growth in China and India, Bloom et al. (2010) find that the main reasons behind the rapid growth spurt of these two countries are a rise in life expectancy, a rise in trade or openness of the economy, and an increasing share of the working-age population. Together these factors accounted for 3.3 percentage points of the GDP growth rate in China and respectively for 2.7 percentage points in India. The model of Bloom et al. (2010) predicts slower economic growth in China after 2010, based on projections of modest further increases in life expectancy and a rising dependency rate as the population ages. By contrast, they expect to see somewhat higher growth rates in India over the next 30 years as the effects of the declining birth rate and of the ‘bulge’ population cohort creates a rise in the working-age share of the total population.

Wei and Hao (2010) find that the substantial decline in the dependency ratio, particularly the youth dependency ratio resulting from fertility falls, has accounted for about one-sixth of the provincial growth rate of GDP per capita in China from 1989 to 2004. Cai and Wang (2005, 2006) examine the demographic effect on economic growth in the convergence model using pooled data for 28 Chinese provinces from 1982 to 2000. They find that a 1% increase in the total dependency ratio lowers 0.115% of China’s GDP per capita growth rate. In the period 1982–2000, the total dependency ratio has declined by 20.1%, which thereby corresponds with one-quarter of China’s economic growth.

Barely any empirical research has focused on the impact of demographic changes on economic growth in Brazil, Russia and South Africa. Porsse et al. (2012) find that the child

(19)

dependency ratio in Brazil is negatively correlated with economic growth and that the elderly dependency ratio is positively correlated with economic growth (the latter outcome is not in line with the theory described in subsection 2.1). Bloom and Canning (2004) argue that Latin American countries are unable to realize their demographic dividend due to their unstable political and economic environment. It will be interesting to see whether our empirical results confirm this argument or not.

Previous research for Russia cannot be found on this topic. Relative to other parts of the world, Russia has had a high working age share since at least 1950, but it is argued that they have failed to benefit economically from it (Bloom, Canning and Sevilla, 2003a). In the next 50 years, Russia will see a growth in its elderly population and shrinkage in its working-age and youth populations. Our results in section 4 will show the impact these demographic changes have had and will have on economic growth in Russia.

High fertility has been the major component of Africa’s sluggish demographic transition and a major cause of its rapid population growth. Since the late 1990s many African countries, including South Africa, have improved their institutional quality and fertility rates have begun to fall. Bloom, Canning et al. (2007) argue that the Sub-Saharan African countries have the potential to exploit their demographic dividend in the future as long as the policy and institutional context is compatible and conducive. The results in section 4 will indicate whether or not South Africa has already benefitted from the start of their demographic transition.

2.3.3. Sources of Income per Capita Growth – Case Study Ireland

Bloom and Canning (2003) perform a case study of economic growth in Ireland and suggest that the legalization of contraception in 1980 resulted in a sharp decline in fertility and a sizeable increase in the relative share of the working-age population. This rise, although not quite as stunning as that in East Asia, had the effect of accelerating Ireland’s economic growth. Similar to the earlier studies mentioned, they use OLS and IV regressions to explain income per capita growth in the five-year period by a fairly similar set of explanatory, demographic and time dummy variables, for a panel of countries over the period 1965-1995.

An interesting analysis of this study is the comparison of their fitted growth rates from their estimated econometric model with the actual growth rates of income per capita in Ireland. The model tracks the actual growth rates extremely well in terms of growth rates changes. It also tracks the data reasonably well in terms of growth rate levels. Another

(20)

appealing aspect of this study is that the fitted growth rates are decomposed into three components: the catch-up effect, the world-income growth effect and the estimated effect of changes in the age structure, see Figure 10.

Figure 10. Sources of Income per Capita Growth in Ireland

Data source: Bloom and Canning (2003)

The catch-up effect is the change in income per capita that corresponds to the partial adjustment of a country’s actual income to its steady-state income. The world-income growth component corresponds to the time dummy variables that are included in the regression. Interestingly, the component of the estimated effect of changing age structure was zero for Ireland through the 1970s and grew steadily to nearly 3 percentage points by the mid-1990s. By that time this estimated demographic change effect was the largest component of economic growth. In subsection 4.2 we will compare the fitted growth rates from our estimated econometric model with the actual growth rates of income per capita in the BRICS countries. Also a similar decomposition of the fitted growth rates will be performed to observe whether or not the age structure component has dominated income per capita growth in the BRICS countries.

Bloom and Canning (2003) also investigate the effect of labour force participation on economic growth in Ireland and find the estimated effect to be negative, in contrast to the expected positive effect. Bloom et al. (2010) argue that this appears to be attributable to the poor quality of the participation rate data as an indicator of labour supply. Based on this

(21)

argument the labour participation rate is assumed constant and excluded from empirical growth equations in several studies (Bloom and Canning, 2005; Bloom, Canning et al., 2007; Bloom et al., 2010). In this study we will perform a robustness check with the labour participation rate for the period 1992-2007 to see whether we will obtain similar results as Bloom and Canning (2003) or not.

3. Model and Methodology

This section provides the regression model and methodology used in this thesis to empirically test the effect of demographic changes on economic growth in the BRICS countries. The first subsection explains the theoretical model that will form the base of the empirical model. The latter is presented in the second subsection. The third subsection will describe the data used in this thesis. Finally, in subsection 3.4 we explain how we will use the estimated coefficients from our regression in combination with population forecasts to gauge the magnitude of the demographic dividend in the BRICS countries over the coming decades.

3.1 Theoretical Model

Most empirical models of economic growth have focused on the growth of income per capita. It is a convenient summary of the standard of living and a useful measure of the level of economic development. However, economic theories typically address the level of output per worker (Bloom and Canning, 2005). Therefore, we start with an accounting identity that links income per capita (Y/N) to income per worker (Y/L):

𝑌𝑌 𝑁𝑁 = 𝑌𝑌 𝐿𝐿 𝐿𝐿 𝑊𝑊𝑊𝑊 𝑊𝑊𝑊𝑊 𝑁𝑁 .

In this identity Y is aggregate income, L is the labour force, WA represents the population of working age (aged 15-64) and N is the total population. The identity merely states that the level of income per capita equals the level of income per worker times the labour participation ratio (L/WA) times the ratio of working-age to total population (WA/N). Defining,

𝑦𝑦 = 𝑙𝑙𝑙𝑙𝑙𝑙_{𝑁𝑁 , 𝑧𝑧 = 𝑙𝑙𝑙𝑙𝑙𝑙}𝑌𝑌 𝑌𝑌_{𝐿𝐿 , 𝑝𝑝 = 𝑙𝑙𝑙𝑙𝑙𝑙}_{𝑊𝑊𝑊𝑊 , 𝑤𝑤 = 𝑙𝑙𝑙𝑙𝑙𝑙}𝐿𝐿 𝑊𝑊𝑊𝑊_𝑁𝑁

and totally differentiating the identity, we see that the growth rate of income per capita equals the growth rate of income per worker plus the growth rate of labour participation plus the

(22)

growth rate of the ratio of working-age to total population (Bloom et al. 2010). That is: 𝑙𝑙𝑦𝑦 = 𝑙𝑙𝑧𝑧+ 𝑙𝑙𝑝𝑝+ 𝑙𝑙𝑤𝑤 .

We now focus on the growth of income per worker (𝑙𝑙_𝑧𝑧). Most theoretical models of income growth rely on the growth model of Solow (1956). In this standard neoclassical model of economic growth it is assumed that the production function is the same for all countries and it is a constant return to scale production function (i.e. Cobb Douglas production function). Let us begin by assuming a Cobb Douglas production function of the form

𝑌𝑌 = 𝑊𝑊 𝐾𝐾𝛼𝛼_𝐿𝐿1−𝛼𝛼

where Y is total income, K represents capital, L represents labour, A represents total factor productivity, and α is the elasticity of income with respect to capital.Now dividing both sides by L gives: 𝑌𝑌 𝐿𝐿 = 𝑊𝑊 � 𝐾𝐾 𝐿𝐿� 𝛼𝛼 .

Assuming that the capital stock per worker (𝐾𝐾 𝐿𝐿⁄ ) and total factor productivity (𝑊𝑊) are determined endogenously and that the endogenous processes generating capital accumulation and total factor productivity converge to a steady state, we can write steady-state income per worker as 𝑧𝑧∗ _{= �}𝑌𝑌 𝐿𝐿� ∗ = 𝑊𝑊∗_�𝐾𝐾∗ 𝐿𝐿 � 𝛼𝛼

where * denotes a variable’s steady-state value (Bloom, Canning and Malaney, 2000). Standard neoclassical growth theory states that the growth rate of income of a country is proportional to its initial distance from its steady-state income level. First, this implies that the poorer a country is with respect to its steady state, the faster it can be expected to grow. Second, the higher a country’s ceiling level of income, the faster its expected rate of growth for a given level of initial income. This property of standard neoclassical growth models, in which each country’s income continually approaches a steady state that is determined by that country’s characteristics, is known as ‘conditional convergence’ (see Barro 1991; Barro and Sala-I-Martin, 1995; Kelley and Schmidt 2005 and Figure A in the Appendix). This means that the growth rate of productivity (labour in this case) depends on how far productivity falls

(23)

short of its steady-state value. Letting 𝑧𝑧₀ be the initial level of income per worker, we can write the growth rate of income per worker as

𝑙𝑙𝑧𝑧= 𝜆𝜆 (𝑧𝑧∗− 𝑧𝑧0)

where 𝑧𝑧∗ is the steady state level of income per worker and 𝜆𝜆 is the speed of convergence. The steady state level of income per worker depends on many factors that may affect labour productivity. If we now write the vector of variables that can affect steady state labour productivity as 𝑋𝑋, which gives us 𝑧𝑧∗ = 𝑋𝑋𝑋𝑋 and so

𝑙𝑙𝑧𝑧 = 𝜆𝜆 (𝑋𝑋𝑋𝑋 − 𝑧𝑧0).

The question arises what types of 𝑋𝑋 variables should be included as determinants of run output per worker. Kelley and Schmidt (2005) argue that factors that influence long-run physical and human capital stocks, technology, and natural resource stocks should be considered. Barro (1997) additionally notes that endogenous growth theories that include the discovery and diffusion of new technologies suggest that 𝑧𝑧∗ depends upon governmental actions such as taxation, maintenance of law and order, provision of infrastructure services, protection of intellectual property rights, and regulation of international trade, financial markets, and other aspects of the economy. Additionally, various authors have suggested access to ports, climate, education, health, and many other factors as possible influences.

Now returning to the initial equation of growth of income per capita and since 𝑦𝑦0 = 𝑧𝑧0+ 𝑝𝑝0+ 𝑤𝑤0 ,

we can write

𝑙𝑙𝑦𝑦 = 𝜆𝜆 (𝑋𝑋𝑋𝑋 + 𝑝𝑝0+ 𝑤𝑤0− 𝑦𝑦0) + 𝑙𝑙𝑝𝑝+ 𝑙𝑙𝑤𝑤.

This final equation is similar in form to the standard regressions run in economic growth analyses and we will use this equation as the base for our empirical model. It relates growth in output per capita to a range of variables, 𝑋𝑋, that influence the steady-state level of output per worker and the log of the initial level of income per capita, 𝑦𝑦₀. However, several other terms appear. The log of the initial participation and working-age to total population ratio (𝑝𝑝₀ and 𝑤𝑤₀) also affect the steady-state level of income per capita, while the growth terms (𝑙𝑙_𝑝𝑝 and 𝑙𝑙_𝑤𝑤) affect the growth of income per capita directly. Due to the identity used to derive this regression, the coefficients on these two terms are fixed (equal to 𝜆𝜆, or minus the coefficient on initial income per capita for the level term and equal to one for the growth

(24)

term) (Bloom and Canning, 2005). In other words, they represent what will happen if the extra labour supply per capita due to these age structure effects is fully employed and leads to no diminution in the level of inputs per worker given by 𝑋𝑋. Rather then impose coefficients on these demographic variables we prefer to estimate all the parameters of the above equation. This allows for the possibility that some of the enlarged labour force may not be gainfully employed, and that labour force growth may dilute non-labour inputs per worker (Bloom et al. 2010).

3.2 Empirical Model

As mentioned earlier we will use five-year period averages to avoid short-term fluctuations. Causality is a serious issue in studies of economic growth. Considering the issue of causality we present an empirical model based on a similar approach like Bloom et al. (2010). We treat as exogenous the variables measured at the start of the growth period being explained. For example, we assume that the beginning-of-period value of life expectancy (health) in 1993 affects economic growth over the period 1993–1997. Although income does affect health in this assumption, it is income in 1993 and earlier that has an effect, and not future income growth. Our schooling variables are an exception due to data availability. Barro and Lee (2013) construct these variables in the middle of our five-year periods (1980, 1985 etc.), but we will treat them as being beginning-of-period values (1978, 1983 etc.).

When we do have variables that are measured during the growth period (for example, the growth of the share of the working-age population) we instrument them with lagged values and the beginning-of-period fertility rate. We use 1972 as base year for the initial values of income per capita and the share of the working-age population. Finally, we include period binaries for all but one five-year period to control for global trends in growth. This results in the following empirical regression (without instrumental variables):

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅_𝑝𝑝𝑝𝑝𝑝𝑝_𝑐𝑐𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ𝑖𝑖,𝑡𝑡 = 𝛼𝛼1+ 𝛼𝛼2𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑝𝑝𝑖𝑖,0+ 𝛼𝛼3𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑤𝑤𝑖𝑖,0+ 𝛼𝛼4𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑦𝑦𝑖𝑖,0

+ 𝛼𝛼5𝑅𝑅𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑃𝑃_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ𝑖𝑖,𝑡𝑡 + 𝛼𝛼6𝑊𝑊𝑙𝑙𝑝𝑝𝑊𝑊𝑐𝑐𝑃𝑃𝑙𝑙𝑊𝑊𝑙𝑙𝑝𝑝_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ𝑖𝑖,𝑡𝑡

+𝑋𝑋1𝑆𝑆𝑐𝑐ℎ𝑙𝑙𝑙𝑙𝑙𝑙𝑐𝑐𝑃𝑃𝑙𝑙𝑖𝑖,𝑡𝑡+ 𝑋𝑋2𝐿𝐿𝑐𝑐𝐿𝐿𝑝𝑝𝐿𝐿𝐿𝐿𝑝𝑝𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑃𝑃𝑐𝑐𝑦𝑦𝑖𝑖,𝑡𝑡+ 𝑋𝑋3𝐼𝐼𝑝𝑝𝑐𝑐𝑇𝑇𝑝𝑝𝑇𝑇𝑝𝑝𝑝𝑝𝑃𝑃𝑃𝑃𝑝𝑝𝑇𝑇𝑇𝑇𝑖𝑖,𝑡𝑡+ 𝑋𝑋4𝐼𝐼𝑃𝑃𝐼𝐼𝑝𝑝𝑇𝑇𝑐𝑐𝐼𝐼𝑝𝑝𝑃𝑃𝑐𝑐𝑅𝑅𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑖𝑖,𝑡𝑡

+ 𝑋𝑋5𝐿𝐿𝑐𝑐𝑃𝑃𝑇𝑇𝑙𝑙𝑙𝑙𝑐𝑐𝑊𝑊𝑝𝑝𝑇𝑇𝑖𝑖+ 𝑋𝑋6𝐼𝐼𝑝𝑝𝑙𝑙𝑝𝑝𝑐𝑐𝑐𝑐𝑇𝑇𝑖𝑖+ 𝑋𝑋7𝐿𝐿𝑐𝑐ℎ𝑃𝑃𝑙𝑙𝐿𝐿𝑐𝑐𝑃𝑃𝑙𝑙𝑛𝑛𝑐𝑐𝑇𝑇𝑐𝑐𝑐𝑐𝑐𝑐𝑛𝑛𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑃𝑃𝑐𝑐𝑙𝑙𝑐𝑐𝑧𝑧𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑃𝑃𝑖𝑖

(25)

Our dependent variable 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅_𝑝𝑝𝑝𝑝𝑝𝑝_𝑐𝑐𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ_{𝑖𝑖,𝑡𝑡} corresponds with the 𝑙𝑙_𝑦𝑦 from the theory section. The variables 𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑝𝑝_𝑖𝑖,0, 𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑤𝑤_𝑖𝑖,0 and 𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑦𝑦_𝑖𝑖,0 represent the log of the initial participation ratio, working-age to total population ratio and level of income per capita. The 𝑅𝑅𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑃𝑃_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ_{𝑖𝑖,𝑡𝑡} and 𝑊𝑊𝑙𝑙𝑝𝑝𝑊𝑊𝑐𝑐𝑃𝑃𝑙𝑙𝑊𝑊𝑙𝑙𝑝𝑝_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ_{𝑖𝑖,𝑡𝑡}represent the 𝑙𝑙_𝑝𝑝 and 𝑙𝑙_𝑤𝑤 from the theoretical model. In addition to these variables, we included a number of indicators that potentially explain labour productivity differences across countries (the variables in our vector X in the theory section above). The period dummy variables correspond with the 𝛿𝛿_𝑡𝑡 and 𝜀𝜀_{𝑖𝑖,𝑡𝑡}is the error term.

In their case study on economic growth in Ireland Bloom and Canning (2003) attempt to use the participation rate but are rather unsuccessful. After controlling for potential endogeneity they find an insignificant estimate for the labour participation growth rate. Since reliable labour participation data is only available from 1990 onwards, we will perform a robustness test for the period 1992-2007 to see whether we obtain similar results. If so, we will exclude the initial labour force participation rate (𝐼𝐼𝑁𝑁𝐼𝐼𝐼𝐼_𝑝𝑝_𝑖𝑖,0) and the growth rate of the labour force participation (𝑅𝑅𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑝𝑝𝑐𝑐𝑐𝑐𝑐𝑐𝑙𝑙𝑃𝑃_𝑙𝑙𝑝𝑝𝑙𝑙𝑤𝑤𝑐𝑐ℎ_{𝑖𝑖,𝑡𝑡}) from our regression like most empirical studies (Bloom et al. 2005, Bloom et al., 2007; Bloom et al., 2010).

3.3 Data Description

We construct a panel of countries observed every five years from 1972 to 2007. We limit our data set to those countries where all explanatory variables are available, which leaves us with a sample of 74 countries6 (Russia being an exception since data is only available from 1990 and thereafter). We deliberately use data until 2007 to exclude the period of the global financial crisis (2007-2010). Data on real GDP per capita and the ratio of investment to GDP are obtained from the Penn World Table Version 8.17. Figure 11 shows how real GDP per capita grew in the BRICS countries between 1972 and 2007. Although more pronounced for some, all five countries experienced an upward trend from the 1990s onwards.

The data on the share of the working-age over total population and the labour force participation ratio used in our regressions are drawn from the World Bank’s World Development Indicators (World Bank, 2016). The labour force participation ratio of a country

6

The countries used in this study are described in Appendix B.

7_{Penn World Table Version 8.1 is an update by Feenstra, Robert C., Robert Inklaar and Marcel P. Timmer}

(2015). As stated by Bloom et al. (2010) data on GDP per capita are more complete and reliable in the Penn World Table compared with the corresponding World Bank data, hence that data has been taken for our regressions.

(26)

Figure 11. Real GDP per capita - BRICS Countries

Data source: Penn World Table Version 8.1, updated by Feenstra, Robert C., Robert Inklaar and Marcel P. Timmer (2015).

is measured by the size of its economically active population divided by the size of its working-age population using data from the International Labour Office (ILO). Unfortunately this data from the ILO is only available from 1990 onwards. We follow the general cross-country comparisons literature and use geographical characteristics, human development indicators, and institutional measures as explanatory variables. See Table C in the Appendix for a complete description of variables and their data sources.

Our geographical variables chosen are the percentage of land area located in the tropics and a dummy variable if a country is landlocked or not. We use tropical area based on the expectation that a larger share of tropical area in a country is associated with lower steady-state per capita income and therefore lower growth in that country. The binary variable whether a country is landlocked or not relates to transport conditions; the expectation is that countries without a coastline have fewer opportunities for trade and hence fewer growth opportunities (Bloom et al., 2015). The data for these variables are taken from the Country Geography Data of the Portland State University.

Human development indicators included in our regressions are the average years of secondary schooling of the population aged 15 years and older and life expectancy at birth. We included the average years of secondary schooling since Barro and Sala-i-Martin (1995) argue that secondary schooling appears to be the most important component of education. The idea is that a more educated workforce is likely to be more productive and therefore more

(27)

likely to seize opportunities for economic growth. Educational data is taken from Barro and Lee (2013). Life expectancy data are from the World Bank (2016). Life expectancy is used as a proxy for the health of the workforce and is expected to lead to economic benefits through lower morbidity, higher returns on investment in human capital and increased savings. Schooling and health therefore can be thought of as indicators of the quality of labour. The graphs in Figure 12 show that life expectancy in Brazil, India and China has been steadily increasing since 1972. However, life expectancy in Russia and South Africa is lower at the end of the investigated period, 2007, compared with the start of the period, 1972. With life expectancy below 55 years in 2007, South Africa is way behind the other BRICS countries. An increasing life expectancy (or the lack of it) is expected to have a positive (negative) impact on the growth of the working-age population and thus on economic growth.

Figure 12. Life Expectancy - BRICS Countries

Data source: Penn World Table Version 8.1, updated by Feenstra, Robert C., Robert Inklaar and Marcel P. Timmer (2015)

Institutional reforms that have promoted trade may have contributed to the economic success of the BRICS countries. One measure of trade policy is the Sachs and Warner (1995) measure of openness, a dichotomous variable that takes the value 1 for open and 0 otherwise. Wacziarg and Welch (2003) update this measure to an annual series up to 1999, which means that if we would use this measure we would have to extrapolate until 2007. Though, the measure is problematic for our analysis of Russia, China and India. Both countries have openness measures of zero throughout the entire investigated period, suggesting that this

(28)

standard indicator is too crude to measure any changes in trade policy in these countries. The ratio of imports and exports to GDP is often used as an alternative measure of a country’s trade openness. We have constructed this trade openness measure with data from the Penn World Table and will use it in our regressions. However, where rich economies tend to participate more in international trade, countries with large populations are more self-supporting and require less international trade. Therefore, it could be argued that this measure is not the best alternative for our analysis of the BRICS countries.

In addition, several studies use the measure of ethno-linguistic fractionalization (ELF) from Easterly and Levine (1997) to check whether homogenous populations experience higher economic growth or not. The ELF index measures the probability that two randomly selected persons from a given country will not belong to the same ethno-linguistic group. The indices are computed as one minus the Herfindahl (1950) index of group shares. In line with these studies we will experiment with relative new measurements of ethnic, linguistic and religious fractionalization that intend to be more comprehensive than the fractionalization measurements previously used in economical literature. The dataset is taken from Alesina et al. (2003) and contains data for only one year for each country. The language and religion indices are based on data from 2001. Most of the data used to compute the ethnic fractionalisation index are from the 1990s, but for some countries older data are used (as far back as 1979).

The quality of institutions, as captured by the Knack and Keefer (1995) index, is another possible determinant of economic growth, with institutional quality being positively correlated with economic growth. Our governance variable is based on this index, which gives an average indicator of the quality of public institutions. We will use our own constructed indicator based on the sum of the government stability, investment profile, corruption, law and order and bureaucracy quality indices from the ICRG8. Our quality of institutions index is based on data for 2012 since this was the only year available. These variables our very slow moving and we take the value of 2012 as fixed back to 1972. Not having annual data on institutional quality during our investigated period is an unfortunate limitation of our analysis, which may be particularly problematic for some of the BRICS countries, where institutional reforms occurred at several points in time during the investigated period of 1972-2007.

8Data are from the Political Risk by Component File (Table3B, June 2012) of International Country Risk Guide (ICRG), available from the UvA library at https://uvalibraryfeb.wordpress.com/2013/05/16/prs-group-icrg/. The ICRG measures are subjective but based on a consistent methodology.

(29)

Table 2. Descriptive Statistics 1972-2007 for 74 countries and for the BRICS

As mentioned earlier we restricted our analysis to the set of 74 countries for which we had complete data (except Russia for which data start in 1990). See Table 2 for the descriptive statistics of the 74 countries we used in the sample and for the BRICS countries. Although less useful for some variables, the mean, standard deviation, minimum and maximum broadly indicate that the sub-sample of 74 countries retains the diversity of the global sample, in terms of low and high Real GDP per capita growth rates, working age population growth rates and low and high income. Table 2 also presents BRICS’s data for comparison with those of the global sample. The most interesting figures are BRICS’s higher average real GDP per capita and working-age population growth rates. Though, the mean of the level of real GDP capita is only half of that of the global mean. Also noticeable are the lower mean for life expectancy, average years of schooling and trade openness compared with the rest of the world. The mean of the quality of institutions is close to the global mean, but the standard deviation is remarkably low, with only 1.5 points between the minimum and maximum value. This indicates that the public institutions of the BRICS countries are comparable.

(30)

3.4 Forecasting Model

The regression results from our empirical model will be used to make predictions about demographic contributions to future real GDP per capita growth in the BRICS countries from 2010 to 2040. Remember that our empirical model is based on the following equation:

𝑙𝑙𝑦𝑦 = 𝑙𝑙𝑧𝑧+ 𝑙𝑙𝑝𝑝+ 𝑙𝑙𝑤𝑤 .

In our forecasting model we will not include the growth rate of the labour participation ratio (𝑝𝑝 is assumed constant) and focus on the future impact on economic growth due to changes in the working-age population. Future demographic data are drawn from the United Nations World Population Prospects (United Nations, 2015), which include projections up until 2100.

We first project steady-state real GDP per worker growth rates (𝑙𝑙�) for all five BRICS _𝑧𝑧 countries. In our forecasting model these rates will remain constant for the entire forecasting period. We construct these projected rates by taking the average fitted real GDP per capita growth rates between 1992 and 2007 and subtract the average age structure component for this similar period,

𝑙𝑙� = 𝑙𝑙𝑧𝑧 �� − 𝑙𝑙𝑦𝑦 �� . 𝑤𝑤

To compute potential added or reduced growth due to demographic changes, we will make projections for the growth rate of the share of the working-age population (𝑙𝑙�) with the _𝑤𝑤 data from the United Nations. To estimate the potential demographic dividend the BRICS countries could enjoy, we use the regression coefficients for the growth rate of the share of the working-age population and the interaction term from our empirical model. Combining the projected steady-state real GDP per worker growth rates with the projections for the age structure effect results in the following forecasting model:

𝑙𝑙� = 𝑙𝑙𝑦𝑦 � + 𝑙𝑙𝑧𝑧 � . 𝑤𝑤

4. Empirical Results and Analysis

This section presents and discusses the estimation results of the empirical model used in this thesis. Furthermore, this chapter tries to answer the question to what extent changes in the demographic structure have been responsible for the growth rates of real GDP per capita in the BRICS countries and what the consequences of future demographic changes will be for future real GDP per capita growth rates. In subsection 4.1 we present the results from our

(31)

empirical regression analyses. Subsequently, we compare the actual growth rates of real GDP per capita in the BRICS countries to the fitted growth rates from our estimated econometric model in subsection 4.2. Finally, in subsection 4.3 we investigate the potential demographic dividend that the BRICS countries could enjoy in the future and the associated implications for future real GDP per capita growth rates.

4.1 Estimation Results

The results are presented in Table 3. Not reported are the time dummies to capture worldwide shocks to growth in each five-year period relative to the base period, 1973-1977. We started parsimoniously and then added sets of variables to see how each addition affected the explanatory power of the regression. In column (1) we included the initial working-age share, initial income per capita9, geographical and human development variables and added the growth rate of the working-age share of the population in column (2). We see that the latter variable enters highly significant into the regression equation and has a positive effect on income per capita growth as expected. Furthermore, a high initial working-age share and life expectancy significantly increase the real GDP per capita growth rate for a country. Column (2) indicates that after controlling for the growth rate of the working-age population, the coefficient of life expectancy remains significant but decreases in magnitude. This suggests that age-structure dynamics that lead to increases in life expectancy play a role in explaining cross-country differences in economic growth (Bloom et al., 2015). The coefficient on the log initial level of income per capita is negative, indicating the catch-up to a steady state defined by the other variables as set out in the theory section. The demographic variables of being landlocked, or being located in the tropics, have the expected negative sign but are not statistically significant. Contrary to Bloom and Canning (2003) and Bloom et al. (2010) our schooling variable has the expected positive sign, but is not statistically significant.

In order to control for the potential endogeneity of the growth rate of the share of the working-age population we instrument this variable with its lagged values and the beginning- of-period fertility rates. The results in column (3) show that the explanatory power of the regression increases compared with column (2) and that our secondary schooling variable now appears to be significant, albeit at the 10% level. This result fortifies the argument from Barro and Sala-i-Martin (1995) that secondary schooling appears to be the most important component of education, since initially we included the total average years of schooling, but

9

We also experimented with beginning-of-period values for the log initial income per capita and log initial working-age share variables contrary to initial values (i.e. 1972), but this approach lead to undesirable results.

(32)

Table 3. Estimates of the Determinants of the Growth Rate of Real GDP per Capita (all regressions include time fixed effects)

(33)

found the regression coefficient on this variable never to be robustly significant in our regressions. In most empirical economic growth studies the labour force participation ratio is assumed constant and therefore not included in cross-country growth regressions.

In column (4) and (5) we add the log of the initial labour participation ratio and the growth rate of the labour force participation ratio to check if these variables have some explanatory power. Our OLS estimates of these variables, presented in column (4), have the correct positive sign but are insignificant. These results are opposite to the negative coefficients Bloom and Canning (2003) found. However, the growth in the participation rate and the ratio of working-age to total population are contemporaneous with the economic growth being explained and as such possibly endogenous. In order to control for this potential endogeneity we instrument these two growth rates with their lagged values and beginning-of-period fertility rates. The results, shown in column (5), are extremely different compared to column (4). Similar to Bloom and Canning (2003), the coefficients of the participation variables are now negative, but still insignificant. The coefficient of the growth rate of the working-age share is no longer statistically significant. Furthermore, our regression in column (5) is only based on 148 observations and we decide to exclude the labour participation rate variables from our empirical growth regressions, similar to most studies performed on this topic.

In column (6) we include our trade openness, investment ratio, ethno-linguistic fractionalization and quality of institutions variables. Our measure of trade openness enters positively and significantly and increases the explanatory power of the regression. We also experimented with the Sachs and Warner (1995) measure of openness, but it underperformed compared with our self-constructed trade openness measure. The OLS estimate for our geographical tropics variable now appears significant. Similar to Wei and Hao (2010) we find the coefficient of the investment to GDP ratio not to be statistically significant. The ethno-linguistic fractionalization and quality of institutions appear to have the expected sign but are not statistically significant. We examined both the fractionalization indices from Alesina et al. (2003) and the most widely used ELF index (Easterly and Levine 1997; Hall and Jones, 1999; Mauro, 1995), but neither of them appeared to be significant. This outcome is partially explained by Posner (2004) who argues that there is a critical mismatch in most studies between the causal mechanism that is claimed to link ethnic diversity with slow economic growth and the measure of diversity that is used to test that mechanism. Contrary to the assumptions of most scholars who seek to test the effects of ethnic diversity on growth, there is no single ‘correct’ accounting of the ethnic groups in a country, and thus no single ‘correct’

(34)

ethnic fractionalization index value. In most empirical studies the coefficient on institutional quality is positive and significantly correlated with GDP per capita growth (Bloom and Canning, 2003; Bloom, Canning et al., 2007; Bloom et al., 2015). The fact that this is not the case in our regression is most certainly due to our data availability problem for this variable. We repeat the regression shown in column (6) with our instrumental variable approach. These results are shown in column (7). The results do not change much, except that the significance of the coefficient of life expectancy decreases.

A higher working-age population share appears to represent a supply-side opportunity for a potential output boom. However, the availability of extra workers will have little effect if they are not employed. Therefore, we test whether the positive growth effects of the demographic dividend are conditional on the institutional quality of a country as suggested by Bloom and Canning (2003). In column (8) we report estimates of the parameters of a regression model that is specified to include an interaction effect between the growth rate of the working-age population share and the quality of institutions. Since this interactive term involves a contemporaneous growth rate it will be instrumented with its lagged value. We find that the interaction term has a significant and positive coefficient, while the coefficient on the growth of the working-age population share is negative, though not statistically significant. This is an important result, as it implies that the better its institutional quality, the more a country will gain from growth of the working-age share of its population. There is a strong argument that with good governance, many inputs into the growth process, and not just age structure, become more effective. We focus on the interaction between age structure and institutional quality, although we note that others interactions are potentially important as well. For the BRICS countries, whose institutional quality varies between 22.0 and 23.5, the estimated overall effect of growth in the working-age share (combining the direct effect with the interactive effect) is always positive. Another interesting result from column (8) is that the coefficient of the average years of secondary schooling appears significant again, like in column (3).

4.2 Actual versus Fitted Growth Rates – BRICS

We now use the regression results shown in column (8) of Table 3 to see how well the model fits the experiences of the BRICS countries. Figure 13 shows the actual and fitted values for real GDP per capita growth in Brazil, India, China and South Africa from 1977 to 2007. Each observation gives the average annual growth rate over the previous 5 years (for example, the value for 2002 is the average annual growth rate between 1997 and 2002).

Explaining and forecasting the impact of demographic changes on economic growth in the BRICS countries

Table of Contents

1. Introduction

2. Literature Review

3. Model and Methodology

4. Empirical Results and Analysis