____________________________________________________________________________________________________________________
What has been the effect of
demographic changes on economic
growth in Latin American countries?
_________________________________________________________________________________________________________
Lychelle de Lannoy
10522271
Amsterdam School of Economics:
M.sc Economics
International Economics and Globalisation
Supervisor: Ms. Naomi Leefmans
Second reader: Dr. D.J.M. Veestraeten
Date: 15 December 2017
Statement of Originality
This document is written by Lychelle de Lannoy who declares to take full responsibility for the contents of this document.
I declare the text and the work presented in this document are original and that no source other than those mentioned in the text and its references have been used in creating it.
Table of content
1. Introduction ... 4
2. Literature review ... 9
2.1 The transition model ... 9
2.2 Demographic dynamics in Latin America ... 11
2.3 Demographic changes and economic growth ... 14
2.4 Trade policy and institutional quality ... 16
3. Methodology and data ... 19
3.1 Theoretical model ... 19
3.2 Empirical model ... 22
3.3 Data description ... 24
4. Results ... 30
4.1Main results analysis ... 30
4.2 Demographic dividend conditional on trade and political stability ... 34
5. Conclusion ... 37
Reference ... 39
1. Introduction
In the past half-century we have seen a lot of demographic changes in the world. The world population has increased from 2.5 billion people around the 1960s to 7.5 billion people nowadays, which means the world population increased threefold since the 1960s (figure 1). During this period, the population of Latin America rose from 1.6 million to 6.3 million people. Such an increase was partially caused by the policy changes that improved the health system, thereby reducing death rates and increasing population growth in these countries. The global economy also experienced a growth spurt in the last half-century (figure 1). Therefore, economists such as David Bloom started executing research on the relation between demographic changes and population growth on the one hand and economic growth of countries and groups of countries on the other hand. His research was primarily focused on Asian countries, where he partially tried to explain the East-Asian growth miracle by a change in demographic variables and he indeed found a significant and substantial effect of the demographic dividend on economic growth of the countries (see for instance Bloom et al., 2010).
Figure 1: increase of population over the years (above) / increase of global real GDP over the years (below)
1
source: data from World Bank obtained on 15-08-2017
1 World real GDP is converted into US dollars using PPP-method. 3 4 5 6 7 8 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014
Wold population (billion)
population
0 50 100
1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 2011 2014
World real GDP (trillion)
1Besides David Bloom, other researchers with similar interest focused on Asian countries. The reason for their interest was obvious, Asian countries were among the first developing countries where the population grew, the demographic structure changed, and economic growth simultaneously increased (Cotlear, 2011). For this reason, investigating whether demographic changes contributed to the economic growth of those countries was very interesting at that time. But now that more countries in different regions have had the opportunity to develop, it is of crucial importance to also investigate those countries. Therefore, this thesis will concentrate on the Latin American region, which has sparsely been investigated with respect to this topic so far. The lack of evidence regarding the effect of demographic changes on economic growth for Latin America provides the motivation to investigate this relation for Latin America in this thesis.
East Asia and Latin America have a rather similar demographic history, which means that their demographic variables were relatively comparable to each other from the 1950s onwards (Bloom & Canning, 2004). Figure 2 for instance shows that the share of the working-age population in the total population has increased for both regions since around 1975.
Figure 2: comparison of working age population over time for Latin America and East Asia
While the demographic history of the two regions has been rather comparable, their economic performance however differed greatly (see figure 3). This was in favor of Asia, which experienced the “East Asian miracle” with growth rates up to almost 7% for the period 1975-1995. In the same period, a mixture of a rigid labor market, almost closed economies and weak governance caused Latin American countries to fall behind, with an average growth rate of approximately 1% (Bloom, Canning & Sevilla, 2003). Over time, a lot of things have changed in Latin American countries with respect to the degree of openness, policies and the political situation, which ultimately led to an increase in economic growth since 1970s (Saad, 1998). It is possible that while the increased share of the working-age population in the total population, as illustrated in figure 2, has led to a demographic dividend for East Asia, it did not have a similar contribution to economic growth in Latin America in the period 1975-1995 due to the above-mentioned policies in Latin America, notably the fact that the countries were rather closed with respect to trade and the poor institutional quality.
Figure 3: comparison of economic growth over time for Latin America and East Asia
source: data obtained from the World Bank database
With respect to trade policy, Ross (2004) touches upon the fact that if a country has a favorable openness policy it can benefit more from the advantage of the demographic dividend and increase its GDP growth. His argument states that such an open trade policy will have an effect on the labor market, where more people in the group of working-age population will be able to work and contribute
to the economy. This argument implies that the demographic dividend may be conditional on open trade policies, which is what will be tested in this thesis. This has not been tested before using regression analysis. Bloom, Canning & Sevilla (2001) state that between 1965 and 1985 Latin America was relatively closed to the world economy, but by 1995 about 70% of Latin America opened up to the world economy. This development occurred when Latin American countries established the MERCOSUR agreement in 1991 and signed the NAFTA agreement in 19942.
Furthermore, with respect to institutional quality, Bloom, Canning & Fink (2007) conclude that the demographic dividend does not go without a certain degree of institutional quality3. The majority of Latin American countries have a
reputation of a lack of good institutional quality, which has slowly improved over the years. Because these were major changes for the Latin American countries, we will pay special attention to these institutional variables and investigate if they have affected the relation between the demographic changes and economic growth of these countries. In other words, in this thesis we will also investigate whether the impact of demographic changes on economic growth has been conditional on institutional quality. This has been investigated before, but not particularly for Latin America.
The contribution of this thesis is the investigation of the relation between the demographic dividend and economic growth for a more recent period compared to previous studies, where the most recent period was 1965-2005 which was researched by Bloom & Finlay (2009). And as mentioned before, we will investigate the relation only for Latin American countries, which has not been done extensively up to this point. At last, we will also investigate whether it is of crucial importance to take both trade-policy and institutional changes into account when trying to benefit from the demographic dividend. For these reasons I find it interesting to investigate whether demographic changes also contributed to the economic growth of Latin American countries. The main research question this thesis will be analyzing is formulated as follows: Did changes in the demographic structure of Latin American
2 MERCOSUR = free trade agreement for South American countries.
NAFTA = free trade agreement between North America, Mexico and Canada
countries contribute to an increase in economic growth over the years of 1990 to 2015? This question will be answered based on data of 18 Latin American countries
which can be found in appendix 2. After we have complemented previous research by analyzing more recent years, we will also investigate if trade-policy and institutional quality were of crucial importance for the relation between demographic changes and the economic growth of Latin American countries. This will lead to the following additional question: is the effect of demographic changes on
economic growth in Latin America in the period 1990-2015 conditional on the trade openness and the institutional quality of a country?
This paper is set up as follows: In section 2 the previous literature will be discussed to clarify how demographic changes can influence economic growth. Next, the demographic dynamics and important aspects of Latin American countries will be analyzed and discussed. In section 3 the theoretical model and the empirical model will be explained extensively followed by a description of the data used for the analysis. Section 4 will analyze and discuss the results intensively and finally we complete this paper in section 5 with a conclusion based on the analysis.
2. Literature review
In this section we will explain the concept of the demographic transition of a country and the effect of such a transition on economic growth. Furthermore the demographic dynamics of Latin American countries will be discussed followed by a review of recent studies of demographic changes in Latin America. Also literature about the effect of institutional quality on the relation between demographic changes and growth will be discussed following by previous literate about trade policies.
2.1 The transition model
As discussed in the introduction, the main part of the population growth process that can contribute to the economic development of a country is the change in the age structure of the population in a country, which is influenced by the fertility rate and the mortality rate (Cotlear, 2011).
Since the 1950s, the mortality rate and the fertility rate began to decrease for developing countries, as part of the development progress of those countries. The process from a high fertility and mortality rate to a lower fertility and mortality rate is called the demographic transition. This transition is closely correlated with the economic development progress of a country (Szreter, 1993). Bloom and Williamson (1998) constructed a model for this process that consists of four stages (see figure 4).
Figure 4 - the demographic transition model
In the first stage, the fertility and mortality rate are both high and population growth rate is relatively constant. In the second stage, the mortality rate begins to fall and the fertility rate remains the same. This accelerates the growth of the population and leads to an increase in the amount of young people in the population, because it is mostly child mortality rates that decline. The dependency ratio4 will increase in this stage, due to this increased amount of young people in the
population. In the third stage of the demographic transition, the fertility rate will also decline and the population growth rate will slowly decrease. In this stage, the share of the working age population will increase and the dependency ratio will decrease, as all the children that were born in stage 2 become a part of the working-age population. This third stworking-age is where the demographic dividend advantworking-age comes in. After a while, in the fourth stage, the fertility rate and mortality rate will both settle at a lower value than in stage 1. This might eventually lead to an increase of the dependency ratio, because people age and move from the working-age category to the elderly category. Eventually population growth will also stabilize at a constant (long run) rate in this stage.
The initial decline of the mortality rate starting from the second stage is associated with the epidemiological transition of a country. This means that there is a decline in the incidents of infectious and contagious diseases, especially affecting the children of age five and younger. Vaccinations against diseases, improved hygiene and better access to clean water contributed to this process (Bloom and Williamson, 1998). Infant mortality is one of the main reasons why mortality rates are high in low-income countries. Those countries are therefore still in the first stage of the demographic transition (World Bank, 2015b). For the mortality rate to decrease, infectious diseases have to be cured. The supply of public health services that the government offers plays a vital role in this case. In less-developed countries the supply of public health was at a minimum in the 1960s. According to the Pan American Health Organization (PAHO), the supply of public health in Latin American countries has improved over the years, which contributed to the decline of the
4 Dependency ratio = the amount of people aged 15 and below plus the elderly (ages 65 and above), divided by the working
mortality rate (Atun, 2015).
With respect to the fertility rate, Doepke (2004) argues that its decline over the years is influenced by education, income and health. Women who have a higher education, tend to have fewer children. The reason for that is that for them not to participate in the labor market incurs a high opportunity cost. Moreover, if the income per capita increases, parents face a dilemma of having more children or investing in human capital of their children. The conclusion of empirical studies is that as per capita income increases, people prefer investing in their children (quality) rather than having more children (quantity). The last factor that influences the fertility rate is the health system. This is one of the minor parts that influence the fertility rate, but its effect on the fertility rate has been proven. Women, who have easy access to the health system, tend to use more contraceptive methods. In low-income countries the health system is not optimally regulated, therefore the access to optimal health care is challenging. Over time, better access has contributed to the decline of the fertility rate as shown in figure 5.
2.2 Demographic dynamics in Latin America
Paolo M. Saad (2011) clarifies the demographic trends of Latin American countries from 1950 to 2050, which are influenced by the mortality rate and fertility rate of those countries. He analyses the different variables that might affect the demographic dividend, but does not test for any hypothesis. He argues that the mortality rate started to decrease from the 1950s onwards in Latin American countries, mainly due to the fact that child mortality decreased substantially (figure 5). As a consequence, life expectancy increased during the years and is now on average 75 years5 in 2016 compared to 55 in the 1950s. The average life expectancy
of Latin American countries increased at such a rate, that it currently is eigth years higher than the averages of other developing regions and only 1.2 years below Europe’s average. This drastic increase in life expectancy was partially due to the improvement of the health system in Latin American countries (Atun et al., 2015).
The second factor that Saad (2011) discusses is fertility. The fertility rate dropped during the last 50 years, from almost six children per woman to two
5 data from World Bank Database
children per woman in Latin American countries, as shown in figure5 (Cotlear, 2011)6. These trends are different across countries, but it is assumed that in the
long run all countries will converge to a certain level. Studies have shown that the fertility rate also varies within a country. People living in urban areas of a country, are most affected by the outside world, which contributed to an earlier decrease of their fertility rate compared to the rural population (Cotlear & Tornarolli, 2011).
Figure 5 – Fertility and mortality rate in Latin America over the years.
Data source: World bank Database
Figure 6 – Dependency ratio in Latin America
Data source: World bank Database
In figure 5, we can observe that during the years 1960-1963 the mortality rate started to decline before the fertility rate. Due to the lack of data availablility we could only show this from 1960s onward in this figure, however Saad (2011) argues that such a decline in mortality rate started from the 1950s onward, whereas the fertility rate only started around the 1960s. Hence according to the transition model we explained in the previous part, population growth has taken place and therefore change in the population structure. In figure 6, we see that the dependency ratio also decreased over the years. This changing population structure corresponds with the fact that more people have been entering the working-age population over the years (as was shown in figure 2 in the introduction) and thus the dependent population has been declining. Based on our observations, we can conclude that Latin American countries are in the third stage of the demographic transition. Some might argue that they already entered the fourth stage, but the fertility rate is still declining. This means that it has not reached its stable (long run) rate yet. On the other hand, the mortility rate has been stable for about 15 years already (figure 5), thus we can conclude that it has reached its (long run) stable rate. For a country to enter the fourth stage, both of the variables have to have reached there (long run) stable rate.
2.3 Demographic changes and economic growth
In the last few years, the amount of research regarding the effect of demographic changes on economic growth has increased.
Bloom and Williamson (1998) investigated the question: does the change in
the population structure caused by the demographic transition have an effect on the economic growth in East Asian countries? By answering this question they tried to
explain the “economic miracle” in East Asia. They used a cross-country OLS and IV analysis of 78 Asian and non-Asian countries covering the period 1965-1990 and controlled for different variables. The results show that there was a significant negative effect between population growth and economic growth, but there is a significant positive effect of the growth rate of the working-age population on economic growth. This resulted in the conclusion that East Asia has benefited from the demographic dividend. The authors indicated that this has especially been the case because the region had social, economic and institutional policies that granted them the possibility to benefit from the demographic transition.
Bloom & Finlay (2009) analyze the same question as Bloom and Williamson (1998) and therefore re-estimate their results, but for a longer and more recent period. They extended this research by also asking themselves the question: what is
the demographic effect on economic growth of a country in the future? Based on the
same methods used in Bloom and Williamson (1998), the cross-country OLS and IV regressions are analyzed for the period 1965-2005. In this paper they use 5 year averages for each variable during the entire period. For example, the average GDP growth rate is calculated for the period 1990-1995 and for all other five-year periods in 1965-2005. The results are similar as the ones of Bloom and Williamson (1998) where there is a significantly positive effect for the share of working–age population on growth and a significantly negative effect on population growth. The other question they tried to answer is to forecast the future demographic effect by using the coefficients of the demographic variables obtained from the previous question, and combining these with available projections for the demographic variables.
The overall conclusion of this paper is that East Asia did benefit from the demographic dividend, but by expanding the period and forecasting future periods they realized that this effect is decreasing over the years.
Another article from Bloom, Canning & Sevilla (2001) evaluates the relation between the demographic changes and economic growth of five different parts of the world, namely East Asia, Latin America, Sub-Saharan Africa, the Middle East & North Africa, and Japan. He tries to answer the question: how does the age structure
in different countries affect the economic performance of those countries. Each group
of countries is situated in a different stage of the demographic transition. For example, the Middle East and North Africa were still in the first stage of the transition process in 2001, but Japan already had an aging population (meaning it was and still is in its last stage of the transition). Although this article does not econometrically test any empirical hypothesis, it does explain all the different aspects of a particular region with regard to demographic changes and economic performance. By comparing the different aspects, the authors conclude that East Asia experienced the most success with their demographic dividend, whereas Latin America had a less positive transition. The overall conclusion of this article is that such a demographic transition can be beneficial for an economy, only if the country has the necessary policy measures, such as a certain degree of openness to trade and other macroeconomic variables. As indicated, this is however not tested econometrically.
In contrast to the studies discussed above, Beaudry and Collard (2003) focused their research on industrialized countries7. They tried to show that
differences in growth rates of working age population are the key to explaining the difference in economic performance of a country. This question is analyzed by using OLS regression analyzing for two time-periods, 1960-1974 and 1975 – 1997 including 18 different industrialized countries. The results obtained show that working-age population growth had a positive effect on economic performance during the period 1960-1974, but a negative effect during the period 1975-1997. Beaudry and Collard (2003) further investigate if this effect had a relation with technology transition. The conclusion is that a country with a lower growth rate of the working-age population will invest in more capital-intensive technology,
whereas a country with a higher growth of working-age population will depend on human capital.
All the above papers used the share of working age population as the demographic independent variable in their analysis. But James Feyrer (2007) argued that one should not focus on the age structure of the entire population, but rather focus on the age structure within the workforce. His argument was that this way one could observe the productivity of people in different age groups and see who contributes the most to the economic growth of a country. Therefore in his paper, he tries to answer the question: what is the effect of workforce demographics
on productivity of the working age population and in return the economic performance of a country? This question is answered by analyzing an OLS and IV
regression for 2 samples of countries for the period 1960-1990, where the first sample consist of 87 non-oil countries from all over the world and the second sample consist of 19 OECD countries. The independent variables are the different 10-year proportion groups of the working-age population divided by the total working age population. For example, such a proportion group might cover the group of people aged 40-49. The results indicate that there is a positive relation between the different age groups and productivity, however there is a difference in magnitude between the two samples. The conclusion of this article is that the productivity gap between rich and poor countries can be explained through the different age structures within the working force.
2.4 Trade policy and institutional quality
As discussed in the introduction, the demographic dividend may be conditional on trade policy and institutional quality. With respect to the latter, institutional quality has already been found to be of crucial importance in order to benefit from the demographic dividend. Bloom, Canning, Fink & Finlay (2007) tried to answer the question: are the effects of demographic changes on growth different
for Africa compared to the rest of the world with specific interest in institutional quality? To analyze this effect they collected 5-year period interval data for the
period 1960-2000 and included 85 countries where 19 of them where Sub-Saharan African countries. Additional to the standard regression, where they regress
economic growth on demographic variables, they expanded their research by adding an interaction term with institutional quality. This variable enters with a significant positive coefficient into the regression. The conclusion based on these results is that Africa differs significantly from the rest of the world. The lack of good institutional quality will overshadow the positive effect from the demographic dividend.
Additionally, Bloom et al. (2010) also investigate the relation between the demographic dividend and the economic performance, but this time for China and India. The question they are trying to answer is: what is the effect of the demographic
dividend on the economic performance in India and China? They obtained data for the
period 1960-2000. During their analysis, they also focused on the institutional quality of the countries. Therefore they include an interaction term between growth of the share of working age population and their institutional variable. They find a significant positive result for this interaction term. Hence they conclude that an increase in the share of working age population benefits a country which has a favorable institutional environment. Although this paper does not pay a lot of attention to institutional quality, that particular part where they do analyze it is of high importance for this thesis. The overall conclusion of the paper is that China and India’s economic growth spurt is primarily due to the increase in life expectancy, the degree of openness of the country and the growth of the share of working age population.
This confirms that another important variable that goes hand in hand with an increase in the share of working-age population is the trade openness policy, as indicated by Ross (2004). Ross (2004) based his conclusion on the paper of Bloom, Canning & Sevilla (2001), which we discussed in the previous chapter. The conclusion is drawn by comparing East Asia with Latin America. The obvious difference was indeed that East Asia had become more open compared to Latin America, which resulted in a higher economic growth rate. Although there is no study that specifically analyzes the effect of trade policies on the demographic dividend and in return the economic performance, in all the other research discussed in the previous part, they do take trade openness into their regression as a control variable and argue that such a policy is needed to absorb the bulk of working age population (Nayad, 2007; Bloom et al., 2010, Bloom & Williamson,
1998; Bloom & Finlay, 2009). They however do not include trade openness as an interaction variable with the demographic variables, which is what will be done in this research.
3. Methodology and data
In section 3.1 we will introduce the theoretical model and show how this model will be used to do an empirical analysis in section 3.2. Furthermore, we will also introduce an interaction term to see if the effect of the demographic dividend is conditional on trade policy and institutional quality. Section 3.3, finally provides a description of the data that will be used in our analysis.
3.1 Theoretical model
We start out with a traditional neoclassical theory where the long run growth rate is determined by three factors, labor, capital and technology (Mankiw, Romer & Weil, 1992). We assume that the production function has a standard Cobb Douglas form:
𝑌 = 𝐴𝐾𝑎𝐿1−𝑎 (1)
where, Y stands for total output, A represents the total factor productivity, K represents capital, L represents labor and is the elasticity of output with respect to capital.
We choose to follow the approach of endogenous growth models, where we assume that both variables, capital per worker and total factor productivity, are determined endogenously. In such a model all variables converge to a steady state (Mankiw, Romer & Weil, 1992). Because we are interested in output per worker, we divide both sides of equation 1 by L to get the steady state per worker:
𝑌
𝐿 = 𝐴 ( 𝐾 𝐿 )
α (2)
where, we get an equation that links output per worker with labor productivity and capital per worker.
There are different factors that influence the steady state of the variables. These variables are denoted by X, which means that the steady state of output per worker is:
where z*8 is the steady state of log of output per worker, X is a vector of different
factors that influence the steady state of total factor productivity and the capital accumulation, which are the two factors that influence the steady state of output per worker.
If an economy is in a steady state, there will be no economic growth (Mankiw, Romer & Weil, 1992). By definition we say that an economy can only reach its steady state in the long run. On the path towards the long run steady state, all variables can change. For this reason the growth rate of output per worker is given by:
𝑔𝑧= 𝜆(𝑧 ∗ −𝑧0) (4)
where gz is the growth rate of output per worker, z* is the steady state level of the log of output per worker, z0 is the initial level of the log of output per worker and is the convergence rate towards the steady state. As can be seen in equation 4, the distance between the initial level and the steady state level thus determines the growth rate of output per worker. Because all countries have different levels of the factors X that determine their steady state and their own structural characteristics, we can therefore conclude that all countries have different growth rates depending on their own steady state level. This concept is called “the conditional convergence hypothesis”.
Rewriting equation 4, given 𝑧∗= 𝑋𝛽 and including a random error term, we get:
𝑔𝑧 = 𝜆(𝑋𝛽 − 𝑧0) + 𝜀 (5)
Over the years there has been some discussion on how to measure the economic growth of a country. Up until now most economists agree that output per capita is the most acceptable form to measure such growth. There are some limitations to this approach, for example: it does not take into account the welfare effects of the population. In this thesis, this limitation does not affect our analysis. Therefore we settle for this measure.
8 *indicates that variables are in a steady state
Bloom and Williamson (1998) derived an accounting identity that links the working age population with economic growth. This identity is as follows:
𝑌 𝑃
=
𝑌 𝐿 𝐿 𝑊 𝑊 𝑃(6)
where Y represents aggregate income, P stands for total population, L stands for the labor force and W represents the working age population9. According to this
identity, income per capita is equal to income per worker multiplied by labor participation rate and the ratio of working age share to total population.
Taking logs of both side of equation 6, we get:
log𝑌 𝑃 = log 𝑌 𝐿+ log 𝐿 𝑊+ log 𝑊 𝑃 (7)
which can also be written as 10
𝑦 = 𝑧 + 𝑝 + 𝑤 (8)
by differentiating with respect to time, we can get the growth rate of the variables:
𝑔𝑦 = 𝑔𝑧+ 𝑔𝑝+ 𝑔𝑤 (9)
This equation states that the growth rate of output per capita is equal to the growth rate of output per worker plus the growth rate of the participation rate and the share of working-age population.
By combining equation 5, 8 and 9 and adding an error term, we get:
𝑔𝑦 = 𝜆(𝑋𝛽 − 𝑦0+ 𝑝0+ 𝑤0) + 𝑔𝑝+ 𝑔𝑤 + ε (10)
9Based on the world bank, that states that people between 15 and 64 years are able to work 10Defining 𝑦 = log𝑌 𝑃 , 𝑧 = log 𝑌 𝐿, 𝑝 = log 𝐿 𝑊, 𝑤 = log 𝑊 𝑃
This final equation -which is no longer an identity, because equation (5) which is included in equation (10) is no identity- will also be used as the foundation for our empirical model. The equation states that the growth of output per capita is determined by the variables that affect the steady state of output per worker (X), the log of the initial level of output per capita (𝑦0), the log of the initial level of the participation rate (𝑝0) and the log of the initial level of the share of working age
population (𝑤0) and the growth of the participation rate (gp) and the share of working age population (gw). What we can also observe is that the growth variables
have a direct effect on the economic growth, whereas the initial variables have an indirect effect through the steady state level of initial income per capita, y0.
3.2 Empirical model
By translating our theoretical model into an empirical model, in the form of a panel data regression we take each year from 1990 to 2015 as an observation for each of the 18 Latin American countries.
Empirical models that analyze economic growth struggle with the issue of having endogeneity problems, which results in biased results. One of those endogeneity problems is simultaneous causality. This implies that there are variables that influence the growth of the country but simultaneously the economic growth also influences these variables. For example, an increase in the growth of the share of working age population might affect the economic performance of a country, but at the same time it might also be the case that economic performance might affect the share of working age population. Therefore we will base our empirical model on the two stage least squares model that is used in Bloom, Canning and Malaney (1999). As discussed before, our variable of interest, the growth of the working share population, might suffer from such an endogeneity issue. The growth of the participation rate might also be exposed to such a problem. Therefore we use their lagged values as instruments in the first stage of the two stage least squares model. These lagged values eliminate the potential simultaneous causality problem, because current economic growth cannot affect variables of previous years, but variables of previous years can still affect the current economic growth.
For the initial variables, such as the initial income per worker, we use the values of the base years 1990.
Ultimately, we include time dummies for each year11 to control for worldwide
shocks (e.g crises). This leads to the following regression equation:
Growth_yi,t = +1 B_yi,0 + 2 B_wi,0 + 3 B_pi,0 + 4 Growth_ pi,0 +
5 Growth_ wi,0 + 1 Educationi,t + 2 Expectancyi,t + 3 Populationdensityi,t + (11) 4 Opennessi,t + 5 QualityIndexi + 6 Landlocki + 7 Tropicsi + 8 CPI + t + i,t
In this equation, the Growth_yi,t is the real income per capita growth rate
which corresponds to 𝑔𝑦 from equation 10. The initial demographic variables B_wi,0,
B_pi,0 , B_yi,0 that are determined by the base year 1990 value represent the natural
logarithm of the share of working-age population, the participation rate and the income per capita respectively. The growth rate of the participation rate (gp) and the working age to population ratio (gw) are represented by Growth_ pi,0 , Growth_ wi,0.
In a recent study, Bloom, Canning & Fink (2010) concluded that the participation rate is insignificant and therefore excluded it from the model in most of his studies. The reason was that there was only reliable data available from 1990 onwards. As the time period of this thesis is from 1990 onwards we are going to include it in our model, and test to see if we find significant results. If not, we will eliminate the initial participation rate and the growth rate of the participation rate from our model, as was done by the above-mentioned authors.
Ultimately, the number of years of secondary education, life expectancy, population density, openness of a country, quality index, landlock , tropics and the inflation rate are the control variables that are indicated as the vector X in equation 10. Each of these variables will be discussed separately in section 3.3. These variables are likely to determine the income per worker in each country. For example, a person that has a higher education is more likely to get a higher income. The time dummies are represented by t and the error term is represented by i,t.
For the second part of this thesis, we want to investigate whether the effect of the demographic dividend is conditional on trade policy and institutional quality. For this reason we include an interaction term between the growth of the share of working-age population and the openness variable and between the growth of the share of working-age population and the institutional quality index. This leads to the following regression equations, which will also be estimated after equation (11):
Growth_yi,t = +1 B_yi,0 + 2 B_wi,0 + 3 B_pi,0 + 4 Growth_ pi,0 +
5 Growth_ wi,0 + 6 (Growth_ wi,0 x Openness) + 1 Educationi,t + (12) 2 Expectancyi,t + 3 Populationdensityi,t + 4 Opennessi,t +
5 QualityIndexi + 6 Landlocki + 7 Tropicsi + 8 CPI + t + i,t
Growth_yi,t = +1 B_yi,0 + 2 B_wi,0 + 3 B_pi,0 + 4 Growth_ pi,0 +
5 Growth_ wi,0 + 6 (Growth_ wi,0 x Quality) + 1 Educationi,t + (13) 2 Expectancyi,t + 3 Populationdensityi,t + 4 Opennessi,t +
5 QualityIndexi + 6 Landlocki + 7 Tropicsi + 8 CPI + t + i,t
3.3 Data description
In this thesis we focus on 18 countries over a period of 25 years (1990-2015)12. As
discussed in the previous section, we use 13 variables in our regression, where our main focus is on the growth of the share of working population.
Because we have 18 countries in this analysis it would be too much to include a graph for all variables for each country in the main text. Therefore, we included the graphs in the appendix. The remaining of this paragraph will be as follows. We will first describe how each variable is constructed followed by the source of the data. We will also mention what the expected effect is of the particular variable on economic growth of the Latin American countries and describe any notable trends in the variables.
All the initial variables, i.e the initial income per capita, initial participation rate and the initial share of working-age share population are obtained from the World Banks’ database. The income per capita is indicated by the real GDP over the
total population in constant 2011 international dollars, where it is converted using the purchasing power parity (PPP) approach. The participation rate is the ratio between the entire economically active population, those who supply labor for the production of goods and services relatively to the working-age population. The World Bank gathered the data from the International Labour Office (ILO)13.
The share of the working-age population variable is based on the population aged between 15 and 64 years over the total population, which counts for all residents regardless of legal status or citizenship. The growth rate variables are based on the same construction as the initial variables.
We expect the initial participation rate (p0) and the initial share of working age
population (w0) to have a positive effect on economic growth of a country. This
comes down to the fact that if there are more people working or are legally able to work, the economy would be better off because they can produce more goods and services.
Control variables
As most cross-country economic growth papers, we included control variables. The reason for this is that there are a lot of other factors next to our variable of interest that affect the economy of a country. Therefore for us to get reliable results, we included eight control variables, namely education, life expectancy, population density, trade, institutional quality, a landlocked dummy, a tropics dummy, and the inflation rate.
Our first control variable is the education level. We obtained this variable from Barro and Lee (2013), who measure education as a ratio of total amount of people enrolled for secondary education over the total population aged 15 and older. We chose to only focus on secondary schooling, because Barro and Lee (2013) concluded that secondary schooling is the most influential part of education. This indicates that if people are more educated, it will lead to an improvement of the labor productivity and in return increase the income per capita level in countries. Therefore our expectation for this variable is a positive coefficient. Unfortunately, in the dataset of Barro & Lee (2013) we could only find the first year of each 5-year
13We collected the data from the World Bank database and tried to compare it with the data from ILO-website. But could not find the data, therefore continued with the World Bank’s data.
period. For example, data is given for only 1990 in the period 1990-1994. We used this data to interpolate results for the years for which we do not have any data. Economic performance of a country might have an effect on education, which causes simultaneous causality issues in our regression. To control for this potential problem, we include the lagged value in our regression.
By analyzing this variable for the different countries in our sample (appendix 4), we can conclude that the overall trend for the enrollment for secondary education is upward. However, an exception is the enrollment rate in Bolivia, which has been decreasing over the years. Another slightly odd observation is that the enrollment rate in Costa Rica increased up until 2005 and decreased after. Moreover, in Chile the enrolment rate stayed constant up to 2000 and after started to increase. Another strange observation is the enrollment rate in Paraguay. It slightly decreases, than catches up with the rest of the countries during the period of 1995 to 2005, but eventually decreases again. For this variable we could not find any data on the enrollment rate in Suriname.
Our next control variable is ”life expectancy”. This variable refers to the expected years of life for a newborn infant, if the circumstances remain the same during its life. This variable is a proxy for the health of the population. Bloom, Canning & Malaney (1999) argue that recent studies have concluded that there is a positive effect between economic growth and health of the population. This benefit is due to more savings, higher return to investment in human capital and higher productivity. Life expectancy might suffer from the simultaneous causality issue. Because a population who is healthy might add more value to the economy, however some might also argue that because of an increase in economic performance, the health system of a county improved. To adjust for this potential problem, we enter this variable in our regression as a lagged value.
In appendix 5 we can see that for all the countries in our sample the life expectancy has increased during the period 1990-2015, which indicates that the life circumstances in Latin American countries have been improving. In Bolivia the life expectancy increased the most, whereas in Guyana’s life expectancy increased the least.
Our third control variable is the population density variable. This variable is constructed by the total population divided by land area in square kilometers and is
obtained from the World Bank’s database. The effect of population density on economic growth is dubious. Bloom, Canning & Mallaney (1999) argue that if natural resources are fixed, an increase in population density might affect the economy of a country negatively. But it can also benefit from the economies of scale14 due to increased population density, which will positively influence income
per capita. Our expectation is that we will get a negative coefficient. Because Latin American countries are developing countries, we expect that the negative effect will be stronger in this case. Based on the graphs in appendix 6, we observe that for about half of the countries in our sample, the population density has been relatively stable. However, for the other half of the countries such as Costa Rica, Ecuador, Guatemala, Mexico, Nicaragua, Panama and El Salvador there has been an upward trend. A particularly interesting observation is that Guatemala started off with the highest degree of population density and is still increasing.
To control for openness of a country we include the trade variable. Sachs and Warner (1995) argue that a country that is more open to trade, experiences a higher income growth. Bloom et al. (2010) also included this variable and used the proxy that Sachs and Warner (1995) constructed. This proxy can take the value of 1 or 0. The value 1 stands for a country that is completely open to trade and 0 otherwise. In this paper we will take another approach to proxy openness, because the Sachs and Warner proxy is not available for our time period and it does not capture the change of openness during the years. Therefore in this thesis we will be using the trade openness variable that is constructed as follows: total amount of import plus export in current US dollars divided by total GDP. The data is obtained from the World Bank Database. This standard measure is adequate for our analysis, because it captures the change immediately after trade policies have been implemented. We expect that there is a positive relation between trade openness and economic growth of a country. For most of the countries this variable has been relatively stable, as can be seen in appendix 7. Some exceptions are: Guyana’s trade ratio was high in the early 1990s and has been decreasing ever since whereas for Panama, the ratio was somewhat stable until 2010, then it started to decrease rapidly. Besides those striking observations, we can also see that for Honduras, Nicaragua, Paraguay,
Suriname and Uruguay had a fluctuating trade variable during the period of 1990-2015.
To measure the political stability in a country, we included the quality of institutions variable. Institutional quality is a broad concept that includes law, individual rights, government regulation and much more. We constructed our own proxy by using data from the political risk group (PRG) database15. Our political risk
proxy consist of government stability index, investment profile index, corruption index, law and order index and ethnic tension index. These variables all had different maximum values, therefore we gathered all the information and normalized all the values so that all the variables obtained a maximum of 12 points. After that we summed up all the values of the different indices per country, which eventually turned into proxy for institutional quality. A higher value for this variable means that there is less political risk in that country. The data we obtained only covered 1990, therefore we took this value and assumed that it would be the same for the entire period 1990-2015, because the institutional quality of a country does not change rapidly. We expected that this variable would have a positive effect on the growth of the countries. Economic performance might also have an effect on the institutional quality, which implicates that it might suffer from simultaneous causality. But because we assume that the value for 1990 holds for the entire period, we do not have to adjust for the potential endogeneity issue.
By looking at appendix 8 we can divide the 18 countries into 3 groups based on this index. Group 1 consists of the countries with relatively low values, which are Bolivia and Guyana. The second group consists of countries with better values of this proxy compared to the first group, however there is stillroom for improvement. The countries in this group are: Argentina, Brazil, Colombia, Costa Rica, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Peru, Paraguay, El Salvador. Our last group consists of the countries with relatively no political risk, which are Chile and Uruguay.
The next control variable is the landlocked variable. This variable is a dummy variable that indicates if a country is surrounded by other countries (land) and therefore has no direct access to the ocean. Bloom et al. (2010) state that countries without a coastline, will suffer from higher transportation costs, which translates
into less opportunities in trade. This will lead to a lower income per capita. Based on this argument we expect that the coefficient will be negative. In our sample of countries, there are only two countries that are landlocked, namely Bolivia and Paraguay.
The following control variable is the tropics variable. This variable stands for the percentage of land that is situated in the tropics and is obtained from the World Bank Database. Bloom et al. (2010) argue that if a higher share of land is situated in the tropics, this will lead to a lower steady state level of income per capita. This inter
alia has to do with the fact that there are more diseases in the tropics. We also
expect this variable to have a negative coefficient. Analyzing the tropics variable in appendix 9, we see that Suriname and Guyana have the highest percentage of land situated in the tropics. On the contrary, Argentina, El Salvador and Uruguay have a relatively low percentage situated in the topics. In general there has been a decreasing trend for the Latin American countries, except for Costa Rica and Uruguay. Costa Rica has a strange pattern. Its tropics variable decreases slightly over the period 1990-2000 and increases after.
Our last control variable is the inflation rate. Inflation is an important indicator to see how the economy is doing in macroeconomic terms. Therefore we also include it in our regression. This variable is constructed on the basis of change in the Consumer Price Index (CPI). The CPI is a variable that is constructed by taking the average price of all the consumption products that the average inhabitant consumes. The data for this variable is also obtained from the World Bank database. The Producer Price Index (PPI) was also a possible estimator for inflation, but there is a fundamental difference between CPI and PPI. CPI includes all the products the average inhabitant consumes, while the PPI includes the products that are produced in a country. Because the produced products can also be sold on the international market, this is not the ideal measure. Therefore we choose to use CPI. In appendix 10 we can see that in inflation has been relatively stable for all the countries from 2000 onwards. Argentina Brazil, Nicaragua, Peru and Suriname experienced a higher inflation rate before the year 2000, where Brazil had the highest of them all. Ecuador has a relatively stable inflation rate, with a small increase between 1998-2001.
4. Results
This section analyses the results obtained from the empirical model. It starts of in section 4.1 by first focusing on the demographic variable of interest, the growth rate of the share of the working age population, and then also looks at the control variables included. The second part of this section, 4.2, looks at whether the demographic effect on economic growth in Latin American countries is conditional on either trade or institutional quality.
4.1Main results analysis
In this section the regression results of table 1 will be discussed. First we start with a regression where we include all the variables except the growth of the share of working age population. This is done, because we first wanted to determine if the participation rate is of interest to this study. As discussed before, in the paper of Bloom, Canning & Fink (2010) the participation rate variables are excluded due to the fact that there was not enough (reliable) data available before 1990. However, during the time period this thesis covers, 1990 – 2015, there is sufficient data, therefore we included it in our regression and test if they are significant. In column 1 of table 1, the initial participation rate is significant at a 1% level. However, the growth of the participation rate is insignificant. In column 2 we try to correct for potential endogeneity, and include the lagged value of the growth of the participation rate as an instrument. This gives us very strange results for the growth of participation rate, but the initial participation rate stays highly significant.
Therefore we only include the initial participation rate in the following regressions and exclude the growth rate of the participation rate.
Table 1- Coefficients and p-values of six regressions with dependent variable: growth of output per
capita
1 2 3 4 5 6
OLS IV OLS IV OLS IV
Constant -22.740* -11.800 -28.290* -26.910** 41.020*** 43.560*** (0.064) (0.631) (0.074) (0.048) (0.002) (0.000) Initial output per capita -0.080* -0.102* -0.095** -0.093** -0.090* -0.096**
(0.073) (0.096) (0.049) (0.023) (0.051) (0.019) Initial share of working-age population 23.990*** 16.900 27.900** 26.950**
(0.007) (0.246) (0.040) (0.023)
Initial participation rate -8.419*** -7.960*** -8.818*** -8.730*** -8.596*** -8.788*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Growth of participation rate 5.897 -26.190
(0.186) (0.732) Growth of share of working age
population 0.730 0.597
(0.209) (0.209)
initial dependency ratio -4.799** -5.200*** (0.017) (0.003)
Growth of dependency ratio -0.118 -0.243
(0.616) (0.227) Life expectancy 0.429 0.514 0.557 0.537 0.463 0.497 (0.246) (0.154) (0.138) (0.101) (0.196) (0.126) Trade 1.345** 1.267** 1.498*** 1.468*** 1.436*** 1.519*** (0.015) (0.011) (0.005) (0.001) (0.009) (0.001) Population density -0.004** -0.003** -0.004*** -0.004*** -0.004** -0.004*** (0.017) (0.036) (0.009) (0.001) (0.016) (0.002) Education -0.057** -0.039 -0.062** -0.061** -0.057** -0.060*** (0.039) (0.357) (0.028) (0.011) (0.036) (0.009) CPI -0.688*** -0.670*** -0.732*** -0.724*** -0.702*** -0.717*** (0.001) (0.000) (0.001) (0.000) (0.001) (0.000) Institutional quality 0.036 0.068 0.033 0.035 0.031 0.027 (0.279) (0.305) (0.377) (0.299) (0.426) (0.459) Tropics -0.008 -0.002 -0.012 -0.011 -0.009 -0.012 (0.449) (0.885) (0.344) (0.323) (0.462) (0.324) Landlock 0.217 0.041 0.192 0.190 0.173 0.174 (0.459) (0.933) (0.565) (0.520) (0.565) (0.544)
Time Fixed effects yes yes yes yes yes yes
N 433 433 433 433 433 433
R-squared 0.373 0.298 0.374 0.374 0.371 0.37
In regression 3, we include the growth rate of the share of working age population, which is our variable of interest. Unfortunately this variable is insignificant, but it does have a positive relation with the economy of a country. For this regression, we also control for potential endogeneity issues, by including the lagged value of the share of working age population. This fails to change the results.
The 4th column deviates somewhat from the accounting identity that we discussed
before, but we wanted to test if another variable that also measures the change in the share of the working age of population would have different results. Additionally, this is also a robustness check. Therefore we substitute the share of working age population with the dependency ratio16. The initial dependency
variable has the same outcome as before, where it is negative and significant. It is negative because if the share of working age population increases, the dependency decreases. The variables are somewhat different structured, but eventually lead to the same conclusion. Thus the sign of the coefficient is in line with our hypothesis. Also in this regression, the growth of the dependency ratio is not significant. In column 6, we correct again for the possibility of endogeneity issues, by using lagged instead of current values. This is in contrast to the finding method of Bloom et al. (1999). Based on the findings in column 4 we can conclude that for the Latin American countries that we used in our dataset, the demographic dividend is not a major determinant of economic growth, which it was in the case of the East-Asian miracle. An argument for this might be that Latin America did not have the optimal combination of policy changes, certain degree of openness and other macroeconomic variables to get the maximum advantage of a higher share of working age population.
With respect to the remaining variables, we can report the following. In all the columns we can ascertain that the coefficient of the initial income per capita is negative and significant. This result is in line with the “conditional convergence hypothesis”, hence we can conclude that each Latin American country has its own steady state. The second variable that is included is the initial share of working age population, which is also included as a lagged value, therefore by assumption,
16 The dependency ratio also looks at the demographic changes in a country. We used it to see if our previous analyses was
becomes an exogenous variable. This variable has a positive effect on the growth of income per capita and is significant except in column 2.
The four significant control variables are trade, population density, CPI, education. Firstly, trade has a positive coefficient. This is indeed in line with the theory of Sachs and Warner (1995), which argues that a country that is open for trade or has different trade agreements with other countries is more likely to have a higher growth rate for GDP per capita. Hence we can conclude that for the countries we included in our dataset, trade is indeed a determinant for the growth of their economy. In the next section we will explore this some more.
Second, the population density variable appears to be negative and significant. The argument of the positive effect of population density, where economies of scale play a role, is overruled by the negative effect. This is in line with the findings of Bloom, Canning & Mallaney (1999) who argue that the income per capita will fall if the natural amount of resources (i.g land) available is fixed and the population density increases, due to shortage of resources.
Third, our CPI variable has a significant negative effect on the growth of a country. This variable states that inflation is an issue in Latin American countries. This is in line with the conclusion of de De Gregorio’s (1993) paper, where he states that inflation has been an important explanation for the low growth in Latin America. He extensively clarifies that inflation brings uncertainty to the financial market of a country, which in return influences the economy. This uncertainty reduces the incentive to invest and therefore the economy of a country is unable to grow.
The fourth significant variable is the education variable, which is negative. This comes as a surprise to us. As we analyzed the years of secondary schooling for each country in our dataset, we noticed that for some countries the years of schooling decreased. Hence this might be a reason that the coefficient is negative.
The other control variables included are not significant. However, we can analyze them on their magnitude. These variables are life expectancy, institutional quality, and the tropics and landlock dummies. First, the life expectancy variable is positive. This sign is in line with the findings of Canning and Malaney (1999) who argue that people with higher life expectancy, can contribute longer to the labor market, hence it will benefit the country.
The next insignificant variable, is the institutional quality variable. This variable enters positively in the regressions. As described in chapter 3.3, we constructed our own variable by including proxies of different aspects with respect to institutional quality. The higher the value of this variable, the less political risk there is. Therefore our result for the sign is in line with previous studies from Busse & Hefeker (2006) where they concluded that there is a negative relation between investment and political risk. Hence if there is less political risk, there will be more investment and therefore a higher growth for the country.
Our third and fourth insignificant variables are the tropics and landlock dummies. The tropics variable enters negatively, whereas the landlock variable is positive. We argued that we expected the tropics variable to be negative. This is indeed the case, which is in line with Bloom et al. (2010, june) where he argued that if a higher share of land is situated in the tropics, the health circumstances are not optimal. The landlock variable is insignificant and positive. This sign is not in line with the theory of Bloom et al. (2015).
4.2 Demographic dividend conditional on trade and political stability
In the second part of this paper, we investigate the potential effect of the demographic dividend on economic growth conditional on the openness of a country and on their institutional quality. The results can be seen in table 2.
Table 2- Coefficients and p-values of six regressions with dependent variable: growth of output per capita 1 2 OLS OLS Constant -20.340 -18.890 (0.147) (0.227)
Initial output per capita -0.116** -0.101**
(0.013) (0.030)
Initial share of working-age population 24.630** 21.330** (0.012) (0.050)
Initial participation rate -9.497*** -8.998***
(0.000) (0.000) Growth of share of working age population -1.014 4.076**
(0.289) (0.035) growth of working share population * trade 1.628**
(0.025) growth of working age population * institutional
quality -0.111* (0.097) Life expectancy 0.786** 0.733* (0.016) (0.051) Trade 0.972** 1.564*** (0.044) (0.002) Population density -0.004*** -0.004** (0.006) (0.019) Education -0.063** -0.060** (0.024) (0.039) CPI -0.681*** -0.713*** (0.002) (0.001) Institutional quality 0.038 0.096* (0.196) (0.072) Tropics -0.008 -0.006 (0.489) (0.601) Landlock 0.019 0.097 (0.951) (0.765)
Time Fixed effects yes yes
N 433 433
In column 1 we can see the regression for the conditional term with trade. The growth of the share of working age population alone is insignificant, which we already saw in our first part of our analysis. However, the interaction term of the growth of the share of working age population and the openness of a country is positive and significant at a 5% significance level. This indicates that Latin American countries can benefit from the demographic dividend when it also has a favorable change in the openness policy. To make this effect clearer, consider the following extreme example. A demographic change took place in a country and eventually the share of working age population increased. However, the country was completely closed. This would mean that there would be limited amount of work opportunities, which in return would lead to a higher unemployment rate, which would not benefit the economy of the country. Therefore, we conclude that these two effects should go hand in hand to be able to benefit from such changes in demographic variables and trade policy.
On the other hand, the quality of institution in a country might also have an effect on the relation between demographic dividend and the economy of a county. In column 2 of table 2 you can find the results of the regression we used, where we test this hypothesis. The interaction term between the growth of the share of working age population and the institutional variable enters as a negative sign in our regression and is significant at a 10% level. This leads to the conclusion that a country with a high institutional quality is less likely to benefit from the demographic dividend. As mentioned in chapter 3.3, an increase in the institutional quality variable indicates that the country has a better institutional quality, therefore the negative sign was not expected. Méon & Weill (2010) argue that corruption can help adjust the inefficiency of the bureaucracy, and positively affect the productivity of factors in the country, which were inefficient due to laws and weak institutional framework. Because Latin American countries have long been famous for their corruption environment, we can argue that such an increase in share of working age population might benefit from such a corruption.
What we also observe is that the R-squared is higher in regression 1 than in regression 2. Therefore we can conclude that the trade policy condition contributes much more to the growth model than the institutional quality condition.
