Population ageing and economic growth in China

Population Ageing and Economic Growth in China

Emiel Mijnen

Master Thesis

Stud. Nr: 10000208

Email: emiel_m@hotmail.com

Field: International Economics and Globalization. Thesis supervisor: Drs. N. J. Leefmans.

2

Statement of Originality

This thesis belongs to:

Emiel Mijnen

Please do not print/copy without prior permission. All information is a product to be used to demonstrate

my work experience and related skills. This statement certifies that the thesis is based upon original research undertaken by the author and that the thesis

was conceived and written by the author alone and has not been published elsewhere. All information and ideas from

3

Table of Contents

1. Introduction ... 4

2. Literature Review ... 7

2.1 General views ... 7

2.2 China ... 8

2.3 Relevant variables ... 9

2.3.1 Economic growth ... 9

2.3.2 Population ageing ... 11

2.4 Brief Conclusion ... 11

3. Variables and Data ... 13

4. Methodology ... 20

5. Empirical Research Results and Interpretation ... 23

5.1 Hausman Test ... 23

5.2 Regression results using fixed-effects ... 24

5.3 Estimating future effects of the ageing problem in China... 27

6. Conclusion ... 29

References ... 31

Appendices ... 34

Appendix 1: Descriptive Statistics ... 34

Appendix 2: Projections ... 34

Appendix 3: DemProj projections ... 35

Appendix 4: Linear Polarisation ... 35

4

1. Introduction

China has experienced an increase in population growth during the 1950s. The so-called ‘baby boom’ period after World War II was stimulated by the Ministry of Health through introducing abortion restrictions (Yao et al., 2014). In 1953, the total population of China was approximately at 582 million, but the population continued growing rapidly (see Graph 1) until the Chinese authorities decided to intervene. In 1979, when the population was at 975 million, the ‘one-child’ policy was implemented and led to a continuous decline in population growth. Despite this, the total population reached 1.340 billion in 2010. Recent developments show that the policy has become more flexible in November 2013. People were only allowed to have a second child if both parents were a single child. This has been relaxed to parents having a second child if one of the parents was a single child. The United Nations predict that the growth will continue until 2030 when the population is expected to be at its peak level, 1.460 billion. After this, the population will start to reduce (Yao et al., 2014). Graph 1: Total Population China 1960-2013 (in billions).

Data source: Graph constructed by author based on data from The World Bank (2015).

Besides changes in the total population level, the age structure of the population changed as well. The United Nations (2015) stated that the population of ages 0-14 has entered a long-term decline (see Graph 2). From 39,7% in 1960, it is expected to hit 18.8% in 2020 and 15.2% in 2050. They also believe that the working-age population (currently about 73%) will begin to shrink in 2020 and will be around 60% of total population in 2050. The 65+ population is expected to increase from 7% in 2000 to 15.3% in 2030 and eventually 23.3% in 2050 (Yao et al., 2014). Scholars and policy makers have become more concerned about the rapid ageing of the population of China. They predict that it could potentially have adverse effects on sustainable economic growth (Peng, 2008). China hasn’t constructed a social protection system to support the elderly yet. The changing age structure is not favorable to continued economic growth. Potential growth rates are therefore

0,0 0,2 0,4 0,6 0,8 1,0 1,2 1,4 1,6 19 60 19 62 19 64 19 66 19 68 19 70 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 20 12 20 14 B ill io n s

5

decreasing, which make it even more difficult to deal with the consequences of population ageing (Du & Yang, 2014).

Graph 2: Age-structure population China 1960-2013.

Data source: Graph constructed by author based on data from The World Bank (2015).

It is therefore in the interest of China to know to what extent the population ageing problem would have an effect on economic growth. Doing a quantitative research could potentially contribute to the current literature concerning the population ageing problem. The aim of this thesis is to estimate the future effect of population ageing on economic growth for the next 30 years. I chose to use 30 years because it is useful to know the upcoming 15 years as the United Nations predicts that China will hit the peak level of their population in 2030. After that it might be interesting to see what will happen in the next 15 years to the economic growth of China, when the population level is decreasing. Increasing the prediction range to 50 years or more might produce less accurate estimates since the estimates of the UN become less accurate as time increases. This is confirmed by a study of Prskawetz et al. (2007) which found that the ageing structure of a country can be forecasted accurately for at most two to three decades. The research question will therefore be: ‘To what extent will population ageing in China affect its economic growth for the next 30 years?’.

The methodology used for this thesis is based on other relevant empirical studies concerning population ageing and economic growth. First, a cross-country regression will be done to estimate the effect of population ageing on economic growth using panel data. According to Bloom, Canning & Fink (2010), the use of panel data is essential in cross-country regression as it accounts for differences in timing which may also mitigate the negative impacts of ageing. For instance, as rich countries age faster than poor countries, a rich country can use immigrants from a poor country to compensate for the ageing problem. This could theoretically slow the ageing process in the developed country, but increase social pressure and unrest in a country. Therefore, to account for these effects, it is relevant for this research to use panel data. The methodology in this thesis is based on the research of Lee et al. (2013) and Bloom & Finlay (2009), that both perform a cross-country

0 10 20 30 40 50 60 70 80 19 60 19 63 19 66 19 69 19 72 19 75 19 78 19 81 19 84 19 87 19 90 19 93 19 96 19 99 20 02 20 05 20 08 20 11

Population ages 0-14 (% of total) Population ages 15-64 (% of total)

Population ages 65 and above (% of total)

6

regression to study the impact of population ageing on economic growth. However, it will include more economic growth factors in the regression. The dependent variable will be economic growth and the explanatory variables will be the level of initial income (measured in real GDP per capita), political instability, population growth, youth & old dependency ratio’s, human capital (life expectancy & primary school enrollment), working-age population growth rate, fertility rate and government consumption. The dependency ratio’s, fertility rate and government consumption are neither included in the research of Bloom & Finlay (2009), nor in the regression of Lee et al. (2013). They also used data from 80 countries for every 5 years, while this research will use 30 countries and annual data. This country selection consists of 13 countries which are located in East Asia, Southeast Asia and South Asia from the selection of Bloom & Finlay (2009), since this is our area of interest (China). However, these countries have experienced similar demographic changes over time and therefore another 17 countries were added to the dataset, to increase variation in the data.

Next, the estimated coefficient of the old dependency variable will be used to produce a forecast for the next 30 years for China for the effect of population ageing on economic growth in China. This methodology allows us to only examine the effect of the old dependency variable on economic growth. The U.N. forecasts for the old dependency ratio for China until 2045 will be inserted into the dependency ratio variable to give an estimate of the effect of population ageing on economic growth in China for the next 30 years. This method is based on the study of Bloom, Canning and Fink (2010) and Bloom & Finlay (2009), where they also made projections. However, the study of Bloom, Canning and Fink (2010) used data until 2009 (and Bloom & Finlay (2009) until 2005), while I use more recent data (until 2013). Bloom, Canning and Fink (2010) also didn’t do a cross-country regression first to determine the effect. They focused mainly on the OECD-countries in general while this research focuses on China specifically. Bloom & Finlay (2009) included China in their projections but focused more on Asia in general. At last, both studies didn’t account for the effect of the relaxed one-child policy in their projections (this happened in 2013), while this study uses DemProj to forecast different scenarios to account for this.

The structure of this thesis is as follows. First there will be a literature review in Chapter 2 to search for relevant variables and to analyze different views on the effect of population ageing on economic growth. This should provide a sufficient background of information which can be used for the rest of the paper. In Chapter 3 the variables used for the regression and the data will be explained. After this, the methodology of the regression will be explained in Chapter 4, whereas the results will be presented and analyzed in Chapter 5. Based on the literature review and the empirical research, the main conclusions will be presented in Chapter 6 along with some possibilities for further research.

7

2. Literature Review

In this chapter the existing literature concerning the effect of population ageing on economic growth in China will be discussed and analyzed. First, the results of the effects in general (not China specific) will be presented. Their findings will be linked and compared to each other. The studies related to China will be viewed and interpreted next. Subsequently, the methodologies used will be described. Different research methods will be analyzed first to see what variables are useful for determining the effect of population ageing on economic growth. These variables will be split into two groups: ‘economic growth variables’ and ‘population ageing variables’. At last, a brief conclusion will be drawn based on the findings of the literature review.

2.1 General views

Recent studies have investigated what effect population ageing has on economic growth in general and have found similar results, namely that ageing has a negative effect on economic growth. Most studies used a cross-country regression (Sala-I-Martin (1997); Barro (1991); Barro & Lee (1994); etc.), one used a theoretical approach (Prettner, 2013) whereas only the study of Yao, Kinugasa & Hamori (2014) did a time-series regression for China using an ARDL model. The empirical studies of Bloom, Canning and Fink (2010) and Bloom et al. (2007) also used a cross-country regression with panel data and showed that the age structure might be influential on economic growth (measured by income per capita growth). They state that large youth and elderly cohorts slow the pace of economic growth, while a large share of working-age workers speeds it. This is in line with the research of Lee et al. (2013), which found that a larger share of workforce within a population increases economic growth. The latter statement is based on the studies of Park and Rhee (2007) and Bloom, Canning and Fink (2008) which state that labor supply and savings are higher among working-age people (15-65 years). Lee et al (2013) also found that an increase in the young population (0-14) relative to the elderly population (65+) inhibits economic growth in the short- and long run. However, the theoretical research of Prettner (2013) shows a controversial result. He finds that increasing population ageing does not slow down the technological progress of a country and therefore economic growth. One thing to mention here is that he assumes that technological progress is the main driver of economic growth. The benefits of private investments into knowledge creation that individuals make increases with ageing as their created knowledge might pay off more when they have a longer time horizon. This is also in line with the study of Prskawetz et al. (2007), where they found a significant positive effect of life expectancy on economic growth. They also found a significant negative effect of the youth dependency ratio (which is total population aged 0-14

8

divided by workforce), which confirms the results of Bloom, Canning and Fink (2010) and Lee et al. (2013).

There were also some studies that focused more on the population ageing in Asia (Bloom & Williamson (1998); Bloom, Canning & Malaney (2000); Bloom & Finlay (2009)). The study of Bloom & Williamson (1998) found a significant negative effect for the growth rate of the young aged population (0-14), a significant positive effect of the working-age population level growth (same as Bloom, Canning & Malaney (2000); Bloom & Finlay (2009)) and a positive (but insignificant) effect of the elderly population growth (65+). The latter result contradicts earlier results but their explanation is that elderly people take care of younger people, save more, sometimes work in part-time and have a smaller negative impact when compared to younger people on the economic growth. This is due to the labor participation and savings-rate for younger people which are very low. Combining all these effects might explain the positive coefficient.

2.2 China

Several studies focused on the population ageing problem specifically in China. The empirical study of Yao, Kinugasa and Hamori (2014) shows that the share of workforce (age 15-65) in total population has a positive impact on economic growth in China (same result has been found by the research of Hsu, Liao & Zhao (2011)). This means that an increase in the share of potential workers would increase the economic growth. The research of Peng (2008) shows that the decline in the growth rate of the workforce level negatively affects the growth rate of the economy, which will reduce the demand for new investments. Similar results have been found by an empirical study of Du & Yang (2014). They found that when the productive population (workforce) increases relatively to the total population, the growth rate of income per capita will increase in China. This is mainly because economic growth is affected by production and consumption lifecycles. People aged between 24 and 60 are more productive in comparison with their consumption level and in China it can be seen that a large group of the total population is in this age-region. They also state that population ageing leads to rapid accumulation of capital in China. The ageing process can stimulate saving rates to support investments. When this occurs, the capital-intensity will raise the productivity of the labor force (output per worker) which will also stimulate economic growth. They also stated that the age structure is correlated with the consumption pattern. The study of Du & Wang (2011) shows that people aged 60+ tend to reduce their work-related costs but increase their overall expenditure on health care. The latter is proven not to be favorable for economic growth, since the health expenditure could crowd out household consumption on other products.

9

2.3 Relevant variables

To determine what factors might be influential on economic growth, several articles will be analyzed in paragraph 2.3.1 and linked to each other. The resulting factors will be used as control variables in the regression. More articles will be analyzed in paragraph 2.3.2 to determine what variables can be used to proxy for population ageing.

2.3.1 Economic growth

Several studies tried to determine what factors are key to use for determining economic growth. One of the most commonly used explanatory variables was the level of initial income. This factor was found by a research of Sala-I-Martin (1997) where he did two million cross-sectional regressions to determine what factors are useful in determining economic growth. One way to measure the initial income is by taking the logarithm of the real GDP per capita in the initial year (Barro (1991); Barro & Lee (1994); Bloom & Sachs (1998); Sachs (1997)). The research of Barro & Lee (1994) explains that the use of an earlier value of Log GDP as an instrument diminishes overestimating the convergence effect because of measurement error in GDP.

Another variable that was extensively used was human capital (Barro (1991); Barro & Lee (1994); Bloom & Sachs (1998)). The research of Romer (1989) showed that an increase in human capital would increase the innovation rate concerning new products due to technological progress. In this way countries with a higher amount of human capital tend to innovate faster and experience more economic growth. An earlier research of Nelson & Phelps (1966) showed that the ‘technological spillovers’ is another channel through which human capital affects economic growth. If the human capital increases, the chance of absorbing ideas that where discovered elsewhere increases as well. In this way a follower country catches up faster with the technological leader if its amount of human capital is larger. To proxy for human capital, the research of Barro (1991) used primary and secondary school enrollment. Bloom & Sachs (1998) used the natural logarithm of the years of secondary schooling to account for this effect. They also used public health in their regression to measure human capital. This is in line with the study of Barro & Lee (1994) which showed that human capital can be measured through education and health. To account for health, Barro & Lee (1994) and Sachs and Warner (1997) used the life expectancy which they obtained from the U.N. database. The life expectancy variable and the primary school enrollment variable can both be found in the significant variable list in the research of Sala-I-Martin (1997).

The third important factor that I came across was government consumption (Barro (1991); Barro & Lee (1994); Bloom & Sachs; Sachs & Warner (1997)). The study of Barro (1991) showed that the government consumption had no direct effect on the private productivity or private property

10

rights. However, the distorting effect of taxation and government expenditures affects the economic growth in a negative way. To proxy for government consumption, they used the ratio of real government consumption expenditure to GDP. Later on, the research of Barro & Lee (1994) showed that the government consumption variable should be lagged because if the real GDP goes up, the real government consumption expenditure to GDP goes down. The lag in the variable would deal with this problem. The study of Sachs & Warner (1997) used central government saving as a variable (calculated as current government expenditures minus current government revenues) and Bloom & Sachs (1998) used the government deficit as a variable to proxy for government consumption.

Fourth, many studies used population growth in their regressions (Barro & Lee (1994); Bloom & Sachs (1998); Sachs & Warner (1997)). The research of Barro & Lee (1994) showed that population growth would negatively affect economic growth. However, including fertility rates made the coefficient positive but insignificant. They explain that for a given fertility, the higher population growth reflects net immigration or lower mortality (and these elements are plausible to be positively affecting economic growth). Including all variables might give biased results since fertility and population growth might be influenced by life expectancy (Barro & Lee, 1994). The population growth variable is not included in the list of significant explanatory variables for economic growth by Sala-I-Martin (1997).

The fifth factor that seems important is the fertility rate (Barro (1991); Barro & Lee (1994)). The study of Barro & Lee (1994) shows that an increase in fertility rates involves an increase in the value of a parent’s time. In this way, raising a child would cost more. Any change that tends to increase the cost of raising a child would trigger the adult to save more and results in lower fertility. This means that people shift from saving in the form of children to saving in the form of physical capital and human capital. The increase in savings reduces the per capita growth rate. The fertility rate is however not included in the list of significant explanatory variables for economic growth by Sala-I-Martin (1997).

Another channel through which economic growth is affected is through political instability (Sala-I-Martin (1997); Barro (1991); Barro & Lee (1994)). One explanation for this is given by Barro & Lee (1994), namely that an instable government has an adverse effect on property rights and thereby negatively influences investment and growth. Barro (1991) measured the political instability using ‘Number of political assassinations per year per million of population’ and ‘Number of revolutions and coups per year’. The latter was also used in the study of Barro & Lee (1994). The research of Sala-I-Martin (1997) showed some other variables which positively affect economic growth, like: ‘Rule of Law’, ‘Political Rights’ and ‘Civil Liberties’. He also states that the amount of revolutions and a war dummy show negative impact on economic growth.

11

The last channels which were commonly used were the youth dependency ratio (Barro & Lee (1994); Bloom & Sachs (1998)) and the growth rate of the working-age population level (Bloom & Sachs (1998); Sachs & Warner (1997)). Barro & Lee (1994) explains that the youth dependency ratio (which is the absolute number of people below 15 years divided by the amount of total population) affects economic growth in two ways. First, the share of non-workers increases which tends to lower per capita GDP growth. Second, if the youth dependency ratio increases, the work effort of adults will focus more on raising children instead of working. This will lower GDP growth as well.

2.3.2 Population ageing

To account for population ageing, many studies seem to use the growth rate of the working-age population (Bloom, Canning & Malaney (2000); Bloom & Finlay (2009); Yao, Kinugasa & Hamori (2014); Bloom, Canning & Fink (2010); Przkawets et al. (2007)). The study of Przkawets et al. (2007) found a significant positive coefficient for the working-age population growth rate. Studies also used the workforce share of total population (Bloom, Canning & Malaney (2000); Bloom & Finlay (2009); Hsu, Liao & Zhao (2011); Peng (2008)). The study of Bloom & Finlay (2009) states that a decline in the working-age share will tend to depress economic performance in the region. In fact, all studies seem to find a significant positive coefficient on the share of workforce of total population when regressed on economic growth.

Another commonly used variable is the share of younger- and older people in the total population, called the young- and old-dependency ratios respectively (Du & Yang, 2014; Lee et al. (2013); Barro & Lee (1994); Przkawets et al. (2007)). Bloom & Williamson (1998) used the growth of these shares to measure population ageing. The dependency ratios are calculated by dividing the amount of younger people aged 0-14 (or elderly people, 65+) by the workforce population ( aged 15-65). The study of Lee et al. (2013) shows that the coefficient for these variables is negative, thus an increase in the share of younger- or older people would reduce economic growth. However, the share of workforce might be highly correlated with the dependency ratio variables since these variables can be linearly predicted from the share of workforce with a substantial degree of accuracy. In this way we might encounter a multicollinearity bias and this variable will therefore be left out of the equations.

2.4 Brief Conclusion

Most studies examining the impact of ageing on economic growth use cross-country regressions. There are four most commonly used variables to proxy for population ageing (assuming workforce is 15-65 years): young- and old-dependency ratios, the share of workforce in total

12

population and the level of workforce growth. For the young dependency ratio, all the studies find a significant negative coefficient. However, for the old-dependency ratio, the research of Bloom & Williamson (1998) found a positive (but insignificant) effect which contradicts the results of other studies (who found a negative coefficient). Moreover, all studies find a significant positive effect of the share- and the growth rate of the level of workforce within a population on economic growth. For the cross-country regression, across the studies I found 10 variables of interest. These are as follows (same as in the introduction/section 2.3): Real initial GDP per capita, population growth, youth & old dependency ratio’s, political instability, life expectancy, primary school enrollment, growth rate of the working-age population level, fertility rate and government consumption. To proxy for population ageing, the dependency ratios and the working-age population growth rate are used.

13

3. Variables and Data

Based on the literature review, I chose to use 10 variables which might affect economic growth (listed on the right). The most interesting variable for this research is the old dependency ratio, whose coefficient will be used to project future effects of the population ageing problem in China with respect to economic growth. As mentioned earlier, there are 30 countries of interest which I used in the dataset. However, to present a

graph in this chapter for every variable for all countries would require a lot of space and might make things cluttered. Therefore, the data summary description can be found in Appendix 1. In this chapter, each paragraph describes one variable that will be used for the regression equations. The paragraph will consist of a brief description, followed by the expected effect that it will have on economic growth. At last, the source of the data will be given and some interesting features in the data will be described shortly.

The dependent variable for the regression equations is economic growth. This variable is measured as the GDP per capita percentage growth rate based on constant local currency, aggregates are based on constant 2005 U.S. dollars. It is calculated as the gross domestic product divided by the midyear population level. This data is obtained from The World Bank (2015). It is interesting to see what the economic development regarding growth was for the countries in the past years (from 1960 to 2013). The data show that the economic growth patterns of the sample countries are rather similar. It seems relatively constant with some exceptional peaks. One remarkable peak is the -26% growth of China in 1962. However, this can be explained as there was a war between China and India in 1962. Since wars cost money and can hurt economic growth, it is logical to assume that the war caused the sudden drop for obvious reasons. However, after the drop, China has experienced higher growth rates than any other country in the sample. From the data it can be seen that sometimes South Korea and Singapore seem to overrule this statement (one time even Japan) but overall it seems that China is the leader when it comes to economic growth.

The old dependency ratio will be the only explanatory variable of interest in the regression equations since the aim of this research is to isolate the effect of population ageing on economic growth. The definition of dependency ratio is as follows: ‘The dependency ratio measures the % of dependent people (not of working age) / number of people of working age (economically active)’. This variable is therefore the variable of interest since this variable shows the amount of elderly

Variables of interest

Old dependency ratio

Initial Real GDP per capita

Population growth

Young dependency ratio

Political instability

Life expectancy

Primary school enrollment

Working-age population growth

rate

Fertility rate

14

people as a percentage relative to the working-age population. The higher this value will be, the higher the ageing level within a population will be. One might expect a negative coefficient here since most studies find a negative relationship here due to the fact that an increase in the share of older people decreases the share of working-age people (except when the inflow of younger people is from the same magnitude) which negatively affects economic growth. Explanations are: there are more elderly to support (Bloom & Williamson, 1998), work-related consumption goes down (Du & Wang, 2011) and it will slow the economic growth (Bloom, Canning & Fink, 2010; Lee et al., 2013). However, we might also find a positive coefficient as elderly can teach the young, work part-time and save more (Bloom & Williamson, 1998). The data for this variable until 2013 was obtained from The World Bank (2015), while the data for the projections were obtained from the United Nations (2015) database. From the data we can see that most of the Asian countries have a similar pattern with respect to the old dependency ratio, remaining relatively constant and low over time compared to the rest. The other non-Asian countries seem to have a slight upward trend in the data. However, Japan is the only country which seems to deviate from the ‘trending’ pattern from these countries. With an increase of 31,6% of the old dependency ratio over 54 years (1960-2013) it clearly deviates from the rest of the Asian countries since the number two of Asia (South Korea) only had an increase of 9,9% over 54 years. Graph 4 in Appendix 2 presents the projections for the old dependency ratio for China from the U.N. until 2045. These values will be used to project the future effects of this variable on economic growth later on. As we can see from the graph, there is a continuous increase in the old dependency ratio in China for the next 30 years. This refers to the ‘population ageing problem’ in China. With a difference of 25,0% over 35 years it might be interesting to see to what extent this will affect China’s economic growth. From The World Bank (2015) data it can be seen that there was a slight increase of 5,1% over the 54 years in the period 1960-2013. However, the projected impact is almost 5 times higher in a shorter time period.

To account for effects of other factors on economic growth (which might influence our coefficient estimation) some control variables will be added to the regression equations. The initial real GDP per capita variable will be used as a control variable since this variable is directly correlated with the dependent variable economic growth. It is calculated as the gross domestic product divided by the midyear population level using 2005 as a base year assuming a constant dollar exchange rate. However, the natural logarithm of this variable will be used to take the growth of the real GDP per capita into account (Bloom & Sachs (1998); Barro & Lee (1994); Przkawets et al. (2007); Yao et al. (2014)). This variable will be lagged one period as according to Barro & Lee (1994) this is essential to deal with overestimated effects due to measurement errors in GDP (see literature review for more info). One might expect a negative coefficient here since this variable accounts for the ‘convergence effect’ (also known as the catch-up effect) according to Sala-I-Martin (1997). This effect states that

15

poorer countries in terms of GDP per capita tend to catch up with ‘richer’ countries since diminishing returns in countries with lower capital stocks are not as strong as in countries with higher capital stocks. This means that an increase in this variable should negatively affect economic growth. The data for this variable has been gathered from The World Bank (2015). This data seems to have an upward trend. However, the countries South Korea, Japan and Singapore, which were marked as East Asia by Bloom & Finlay (2009), have experienced a booming period compared to the countries which were marked as South-East Asia and South Asia (with an exception for China) as these countries seem to remain at a constant low level. This effect might be explained by the research of Nelson and Phelps (1966) suggesting that poor countries tend to catch up with rich countries if the poor countries have a higher amount of human capital per person. Norway is the clear leader in real GDP per capita since it experiences higher levels throughout the time-period of 54 years.

Another control variable will be population growth. This is measured as the growth rate in year t of the midyear population level from time t-1 to time t, given as a percentage. According to Bloom & Williamson (1998) the population growth should affect the economic growth through the channel that it affects the ratio of the working-age population to dependent population. If the population growth increases the elderly share of the population due to a higher life expectancy, it should have a negative direct effect as more elderly need more support. However, population growth due to a diminishing mortality rate would have no effect since the level of workforce-to-dependent-population ratio won’t be affected. It might also occur that population growth goes in line with an increasing fertility rate. Bloom & Williamson explain that then the effect would be negative on economic growth since there are more mouths to feed (this effect should also occur if there is a fall in infant mortality). However, this effect will have a delayed positive effect as the workforce population will be booming two decades later. Therefore we might expect a negative coefficient here, since the variable is lagged for one period. From a research of Bloom & Canning (2000) it can be shown that including the population growth will increase the explanatory power of the regression. One important remark from the research of Barro & Lee (1994) is that the population growth might be influenced by life expectancy. The data consists of the population growth numbers from 1960-2013, obtained from The World Bank (2015). From the data it shows that there is a general, slight decline in population growth for all countries (downward trend). China shows a steep decline in population growth in 1962, which indicates the war of China with India. Furthermore, Singapore is highly fluctuating whereas the rest is relatively comparable with each other.

Another effect that needs to be taken into account is the effect of young-aged people on economic growth, the younger dependency ratio. This ratio works the same as the old dependency ratio as it is calculated as the level of young-aged people divided by the workforce population (people aged 0-14 / people aged 15-65). One might expect that the coefficient of the variable will be

16

negative since studies of Bloom & Williamson (1998), Lee et al. (2013) and others show that it will negatively affect the economic growth. From the research of Barro & Lee (1994) it follows that an increase in this share will affect the economic growth negatively in two channels. First, it will reduce the per capita growth since the amount of non-working age people increases. Second, younger people demand effort of working-age people as they need care. In this way working-age people will have less time available to be productive to increase economic growth. The data is from the U.N. (2015) for all 30 countries and it includes projections until 2045 for China. From the data it can be seen that there seems to be a downward trend. One important remark here is that all the countries experienced this. In graph 3 (in Appendix 2) we can see the projections of the young dependency ratio made for China until 2045 using the U.N. database. The dependency ratio will hit its peak in 2020 whereas it will have its lowest point in 2035. These projections show that in the long-run (here 2045) the difference with the starting point is a decline of 1,3 over 35 years. I chose not to include the growth rate of the youth dependency ratio since several studies showed that the growth rates have no influence on the level of savings, which ultimately affects the growth of output per working age person (Bloom et al., (2003); Kelley & Schmidt, (2005); Kögel, (2003)).

To account for the stability of a given government/country, a political instability variable is included. Of course, there is no data available that consists of ‘political instability’ since this is hard to measure. However, to estimate this effect I will differ from Barro (1991), who used ‘Number of political assassinations per year per million of population’ and ‘Number of revolutions and coups per year’ to proxy for this. Instead, I will introduce two dummy variables. The first variable indicates whether there was at least one minor armed conflict, which is defined as a conflict with between 25 and 999 battle-related deaths in a given year for each country. If the battle-related deaths in a given year were 1000 or higher, the second dummy variable (war dummy) will be at value one. All other cases will give a value of 0 for both dummies. The war dummy is on the Sala-I-Martin (1997) list, however the other dummy differs since it can separate another ‘middle’ value between a war and non-war called the ‘minor armed conflict’ value. The research of Sala-I-Martin (1997) shows that the war dummy has a negative effect on economic growth, therefore we might expect to find negative coefficients here. The data for this variable is obtained from Themnér and Wallensteen (2014). The data showed the battle-related events from 1946-2013 for all the countries in the world (with at least 25 deaths). I transformed the data in useful panel data, indicating for every country in my sample whether there was a minor armed conflict, a war, or nothing at a given time.

Another factor that seems important is the amount of years that people are expected to live in a given country, their life expectancy. This life expectancy variable is used to proxy for health as according to Barro & Lee (1994) the human capital of a country can be estimated by using health and education. The life expectancy variable is a combined variable for male & female, since females are

17

expected to live longer this might be relevant. The variable shows the amount of years that a newborn would live if the mortality patterns remain the same throughout its life. It reflects the overall mortality level of a population, and summarizes the mortality pattern that prevails across all age groups in a given year. It is calculated in a period life table which reflects a snapshot of a mortality pattern of a population at a given time. One might expect the life expectancy variable to have a positive coefficient since the research of Kulish et al. (2006) showed that a higher life expectancy will reduce the demand for health care and many might be able to work and contribute to the economy longer. However, according to Barro & Lee (1994) the life expectancy might be biased when including fertility and population growth as well. Greater life expectancy might increase the survival rate of mothers, which might stimulate families to have more children (which increases the fertility rate). However, it might also reduce the fertility rate as fewer births are needed to obtain a desired amount of children reaching adulthood. Higher life expectancy might influence population growth directly since people are expected to live longer, which increases total population (and thus growth). The data is obtained from The World Bank (2015). From the data it can be seen that there is an upward trend. All countries have experienced an increased life expectancy over time. However, there are some countries which experienced a drawback, for example Malaysia (1965) and Sri Lanka (1985). But after the drawback these countries also continued to increase their life expectancy further. In 1960 the difference between the lowest life expectancy (38,456) from Nepal and the highest life expectancy (73,550) from Norway is a little more than 35 years. This difference has been reduced in 2013 with India being the lowest (66,456) and Japan being the highest (83,332). The difference is now approximately 17 years (reduced by 18 years). One interesting booming period is from China in 1965, which increased its life expectancy with approx. 14 years in a time period of 8 years (1965-1973).

To account for the effects of education on economic growth, one might think of the primary school enrollment variable. This variable can be used to proxy for human capital based on the study of Barro & Lee (1994). The primary school enrollment variable is on the list of significant explanatory variables for economic growth by Sala-I-Martin (1997). Unfortunately, no complete dataset was available from 1960-2013. Using the data from the World Bank (2015) would reduce my observations from 1499 to 667, which is more than 50% of my data. Therefore I chose to use a different variable. The study of Collins and Bosworth (1996) used the average schooling years as a measure of human capital. This variable was calculated using seven rankings from no-schooling to postsecondary schooling in combination with the adult’s estimation of the average years of schooling. The study of Barro and Lee (1994) found a significant positive effect of secondary school enrollment on economic growth. However, the research of Collins and Bosworth (1996) couldn’t find a significant coefficient for the average years of schooling variable. For secondary school enrollment no significant coefficient

18

was found by Bloom & Sachs (1998). Therefore the expected effect of this variable is ambiguous. They used a dataset constructed by Barro & Lee (2010) which contained data from 1960-1990 which is now updated to a version of 1950-2010 data. This data contains the educational attainment of people aged 15 years or older, both male and female combined. However, this data is available for 5-year intervals (1950, 1955, 1960 etc.). To interpolate between the census 5-years, I used a linear approach to fill in the missing blanks. In Appendix 4 one can find a brief explanation how the linear interpolation was performed. From the data it can be seen that there is an upward trend in average years of schooling for the sample countries. Nepal experienced the lowest average years of schooling with 0,13 years in 1960 and 4,64 years in 2013, whereas the USA experienced the highest with 9,17 years in 1960 and 13,37 years in 2013. One remarkable country here is Canada, as it experienced the slowest growth in average years of schooling (increase of 0,58 years over 54 years). Another remarkable country was Germany, which experienced a reduction period of 15 years (from 1960 onwards) before it started increasing.

Another factor that should be accounted for is the growth rate of the level of the working-age population. However, no data was available for this variable but instead I obtained the share of workforce of each country from 1960-2013. This share was calculated by taking all people aged 15-65 and divide them by the total population. To obtain the level of workforce in year t, I multiplied the share of workforce in year t with the total population in year t for each country. However, to calculate the growth rates of these levels the following adjustment had been made:

𝑌_𝑡 =𝑋𝑡− 𝑋𝑡−1 𝑋𝑡−1

Where Yt is the growth rate at time t and Xr is the level of workforce at time t (so Xt-1 is the share at time t-1). In this way the difference between the current period and one period before is calculated and divided by the period before. This should give us the growth rate of the level of the working-age population. One might expect a positive coefficient here since the research of Bloom & Sachs (1998) showed that when the size of working-age people increases relatively to the total population, the per capita productive capacity of the economy expands. This expansion can either be a result of an increase in employment or an increase in savings, which creates more potential for rapid growth in an economy. The research of Bloom & Williamson (1998) also showed that when the working-age population growth rate is higher than the population growth, the economic growth of the country increases. The data for this variable was obtained from The World Bank (2015). The data shows that the growth rates have roughly been the same for every country in the sample for the period 1960-2013.

The fertility rate also seems to be affecting economic growth. The definition for this variable is: ‘Total fertility rate represents the number of children that would be born to a woman if she were

19

to live to the end of her childbearing years and bear children in accordance with current age-specific fertility rates’ (The World Bank, 2015). One might find a negative coefficient here since Bloom & Williamson (1998) state that in case of high fertility there are more children to feed and raise which costs work-time. The negative coefficient is also found in the research of Barro & Lee (1994). One important remark is from Bloom et al. (2010), namely that the projections of the U.N. for the fertility rates until 2050 show substantial declines which would stimulate the female labor-force participation. For this variable I found two databases, the U.N. (2015) database and The World Bank (2015) database. I chose to use The World Bank data instead of the U.N. data as the U.N. calculated the average of each 5 year period based on different scenarios, whereas The World Bank provided annual data interpolated from the United Nations data (which will not reduce my observations). From the data we can see a clear downward pattern. The highest fertility rate in 1960 was in the Philippines (7,148) which dropped over time, hitting 3,043 in 2013. The difference between countries seems somewhat bigger in 1960 than in 2014. In 1960 the biggest difference was between the Philippines and Japan (5,147) and in 2014 the biggest difference was between the Philippines and South Korea (1,856). There seems to be a more common trend towards having fewer children.

The last interesting control variable is the government consumption. However, this variable will not be included in the regression. The most complete data for this variable has been obtained from The World Bank (2015). This data only covered the timeline from 1990-2013. In this way, all the observations before 1990 would be lost (which is 30 years). This will reduce precision and therefore this variable will be left out of the regression equations.

20

4. Methodology

From the ‘variables and data’ part it is clear that all data for each variable is panel data and therefore a panel data analysis will be implemented. The old-dependency ratio is the most important variable since this will be our indicator for population ageing. To estimate the impact of this variable on economic growth, a cross-country regression will be done first. The effect of 10 variables on the economic growth of 30 countries will be estimated through six equations. The abbreviations used in this chapter are described in table 1 below.

Table 1: Short descriptions of the abbreviations used. Abbreviations

GROWTHit-1 The GDP per capita growth rate for country i at time t-1.

RGDPit-1 The Real GDP per capita for country i at time t-1.

POPit-1 The population growth for country i at time t-1.

YOUNGit-1 The young dependency ratio for country i at time t-1.

OLDit-1 The old dependency for country i ratio at time t-1.

ARMCit Amount of battle-related deaths between 25 and 999 for country i at time t.

WARit Amount of battle-related deaths at 1000+ for country i at time t.

LIFEit-1 The life expectancy for country i at time t-1.

AVGSCHOOLit-1 The average years of schooling for people aged 15+ for country i at time t-1.

WORKGRit-1 The working-age population growth rate for country i at time t-1.

FERTit-1 The fertility ratio for country i at time t-1.

The methodology that will be used is a similar regression method as the research of Bloom & Finlay (2009) where they used seven different regression equations. However, this research will only capture four of these models since three models include geographic variables. As we account for fixed-effects in this regression, the geographic variables aren’t included. Instead, I will add two more regressions to account for the problem of Barro & Lee (1994) where they showed that both fertility and population growth might be influenced by life expectancy. The complete model for this research looks like this:

𝐺𝑅𝑂𝑊𝑇𝐻

= 𝛼 + 𝛽

𝑅𝐺𝐷𝑃

+ β

𝑃𝑂𝑃

+ 𝛽

𝑌𝑂𝑈𝑁𝐺

+ 𝛽

𝑂𝐿𝐷

+ 𝛽

𝐴𝑅𝑀𝐶

+ 𝛽

𝑊𝐴𝑅

+ 𝛽

𝐿𝐼𝐹𝐸

+ 𝛽

𝐴𝑉𝐺𝑆𝐶𝐻𝑂𝑂𝐿

+ 𝛽

𝑊𝑂𝑅𝐾𝐺𝑅

+ 𝛽

𝐹𝐸𝑅𝑇

+ 𝜀

All equations will be regressed using country-fixed effects to counter possible omitted variable bias by taking the variables that are country-specific as constant over time (Wei and Hao, 2010). According to Wei and Hao (2010), the use of panel data for variables counters the small sample bias

21

since the within-variation can be multiplied using time series. Using panel data also makes it possible to use variables that vary mostly across countries and vary in time and provinces (government expenditure for example). At last, using lagged variables allows us to counter endogeneity problems since variables of one period ago are exogenous in the present.

The first model will consist of the real GDP per capita variable, war- and armed conflict dummies, average years of education and life expectancy. This model is similar to the first model of the research of Bloom & Finlay (2009) where they used a basic growth model. In the second model, the population growth will be added to the basic growth model (Bloom & Finlay, 2009). This is similar to a model proposed by The World Bank (2015). In this model The World Bank accounted for demographic changes by adding population growth. The full model will be used in the third equation including the old- and young dependency ratios, the fertility rate and the growth of the working-age population. Bloom & Finlay (2009) also accounted for other demographic variables in their third equation by including age differentials. The fourth equation will use the same model as the third equation but will be regressed using IV-regression. Bloom & Finlay (2009) included the IV to counter a possible reversed causality bias since the growth of the population and the working-age population might affect the growth of the real GDP per capita. These variables will be estimated by lagging the growth rates one period and including the fertility rate as instruments. To check whether the instruments perform statistically well, a Wald-test will be performed. To test whether the instruments are consistent, a Durbin-Wu-Hausman test will be used. After this regression, the complete model will be estimated in the fifth equation including two interaction variables between life expectancy and population growth/fertility rate. These will be added to see whether the latter two are interacting with life expectancy. At last, the complete model minus the life expectancy variable will be regressed using two-stage least-squares (Bloom & Finlay, 2009). The exact regression equation can be seen in Table 3.

After performing the above regressions, the coefficient of the old-dependency ratio will be used to make projections for the estimated effect of ageing on economic growth in China. The projections equation will look like this:

𝐸𝐹𝐹𝐸𝐶𝑇_𝑡 =< 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 >∗ 𝑂𝐿𝐷_𝑡

The coefficient obtained from the regression will be filled in the ‘<estimated coefficient>’ fragment. The data from the U.N. is forecasted in 5-year intervals. To obtain yearly estimates, I used the DemProj program from the Futures Group website (which can be found at Futures Group (2015) in the references) to estimate yearly projections based on the U.N. data. I chose to do this as the projected estimates of the U.N. are based on the assumption that the one-child policy applies. But as mentioned earlier, the one-child policy was relaxed in 2013. The DemProj program allows filling in and making different projections for the old dependency ratio to account for this. One important

22

development is that Chinese families prefer not to have another child, since it’s more costly. This situation is called the ‘low fertility’ situation, estimated by the DemProj program assuming that the fertility rate will be around 1.3 children per woman throughout the period. There might also be a high case scenario where more people choose to have children, assuming a fertility ratio of 1.8 children per woman. The last scenario will be the one where the relaxed one-child policy is eliminated, assuming a fertility ratio of 2.3 children per woman (these values are the three different scenario’s available at the DemProj program). The table of the data produced by the DemProj program can be found in Appendix 3. After that, the projections from the DemProj program for the old-dependency ratio until 2045 will be inserted in the above equation. In this way we only look at the effect of the old-dependency ratio on the economic growth for the next 30 years which will answer the main research question of this thesis.

23

5. Empirical Research Results and Interpretation

In this chapter I will describe the regression results. For this research it is useful to know (since we’re using panel data) whether we should use country-fixed effects in our regression, since earlier studies on the determinants of economic growth also used country-fixed effects when they used panel data. Therefore a Hausman test will be performed first. After that, it is crucial to get an unbiased estimate of the old-dependency ratio coefficient. Therefore several regressions will be performed to find the most accurate estimate following the research of Bloom & Finlay (2009). After doing the regressions, the obtained coefficient will be used to estimate predictions using the U.N. data until 2045 of the old-dependency ratio.

5.1 Hausman Test

Before we start the regressions, it is useful to know whether we should take into account country fixed effects or not. The country fixed effects are useful to assume with panel data because in that case you assume that omitted variables that differ between countries, are taken as constants over time. However, it might be the case that some omitted variables are constant between countries but differ over time. When it is unclear what the case is, one might use ‘random-effects’. The latter one might be more efficient since it will give better P-values as they are a more efficient estimator. To determine what is best for our model, I did a Hausman test. The Hausman test compares a more efficient model (our case: random-effects) with a less efficient model, which provides more consistent results (fixed-effects). It will check whether the more efficient model also still produces consistent coefficients. The null hypothesis for this test is formulated as follows: the more efficient model produces coefficients which are just as consistent as the one produced by the less efficient model. This means that when we find a significant p-vale (below .05), the null hypothesis will be rejected. Hence, the more efficient model doesn’t produce coefficients with the same consistency. The results are given in table 2.

The variables which have a ‘1’ behind their name mean that they are lagged for 1 period (t-1). The bottom two variables are the interaction variables. POPLIFE is the variable which shows the population growth times the life expectancy, whereas FERTLIFE is the variable of fertility times the life expectancy. The ‘Prob>chi2’ shows the P-value for which the null-hypothesis is not rejected. Since this value is below the 0.05 border (significance of 5%), the null-hypothesis is rejected. Hence, one might assume that in our case the fixed-effects regression is preferred over a random-effects regression since this will produce more consistent estimates.

24

Variable: (b) fixed (B) random Difference (b-B) S.E.

LRGDP1 -1,187056 -0,158884 -1,028172 0,450333 POP1 -1,601160 -1,641031 0,039871 0,044370 YOUNG1 -0,069651 -0,044184 -0,025467 0,012940 OLD1 -0,092608 -0,130283 0,037674 0,029361 ARMC -0,767605 -0,973281 0,205676 0,065883 WAR -1,591849 -1,721259 0,129410 0,077511 LIFE1 -0,321054 -0,245319 -0,075735 0,059962 AVGSCHOOL1 0,148990 -0,037892 0,186882 0,097569 WORKGR1 116,609300 118,794000 -2,184717 4,024967 FERT1 -4,604977 -3,979959 -0,625017 0,503434 POPLIFE -0,010113 -0,009448 -0,000665 0,000394 FERTLIFE 0,083157 0,072508 0,010649 0,007534

b = consistent under Ho and Ha

B = inconsistent under Ha, efficient under Ho Test: Ho: difference in coefficients not systematic chi2(10) = 29,75

Prob>chi2 = 0.0009

5.2 Regression results using fixed-effects

In this section, the results of the regression will be discussed and analyzed. The earlier mentioned equations are regressed in the order in which they were presented. The first column in table 3 shows equation 1, the second column shows equation 2 etc. until in the last column equation 6 is presented. All regressions are done using country fixed-effects as the Hausman test showed that it was relevant to do so (see section 5.1).

The first column shows the basic growth model estimated using OLS. The real GDP per capita term was lagged one period to be an exogenous estimator for the base year growth and to deal with other biases from Barro & Lee (1994) mentioned earlier in Chapter 3. It shows a significant negative coefficient which confirms the ‘convergence effect’ which was mentioned in the paper of Sala-I-Martin (1997). Furthermore, the political variables also show a negative significant impact on economic growth (war at 5% significance, armed conflict at 10%) which also confirms the findings in the research of Sala-I-Martin (1997). The study of Barro & Lee (1994) explains that an instable government has an adverse effect on property rights and therefore negatively influences investment and economic growth. The life expectancy variable shows a positive significant effect on economic growth. This also confirms the findings of Kulish et al. (2006) which showed that health costs per person are lower when their life expectancy is larger and that people may contribute longer to the economy which stimulates economic growth. At last, the average school variable shows a negative

25

significant coefficient which contradicts the findings of Barro & Lee (1994) which found a positive coefficient at the secondary school enrollment. However, our variable is considered as the ‘average years of schooling’ which consists of both primary and secondary school enrollment. For that reason, comparing these results might not be sufficient enough to draw conclusions.

Table 3: Results using fixed-effect regressions, dependent variable: real GDP per capita growth.

VARIABLES\COLUMN 1 2 3 4 5 6

OLS OLS OLS IV OLS IV

LRGDP1 -1.813*** -2.113*** -2.091*** -0.189 -1.187** -0.0842 (0.391) (0.389) (0.481) (0.130) (0.499) (0.126) ARMC -0.517* -0.606** -0.688** -1.088*** -0.768*** -1.124*** (0.301) (0.297) (0.285) (0.254) (0.281) (0.254) WAR -1.033** -1.147** -1.422*** -1.787*** -1.592*** -1.800*** (0.487) (0.482) (0.465) (0.452) (0.460) (0.453) LIFE1 0.223*** 0.206*** 0.104** 0.0907*** -0.321*** (0.0389) (0.0386) (0.0517) (0.0352) (0.0871) AVGSCHOOL1 -0.343*** -0.390*** -0.00906 -0.0433 0.149 0.00148 (0.128) (0.127) (0.133) (0.0519) (0.134) (0.0490) POP1 -1.070*** -1.986*** -2.777*** -1.601*** -2.495*** (0.176) (0.345) (0.379) (0.378) (0.353) YOUNG1 -0.0430 -0.0275 -0.0697** -0.0169 (0.0278) (0.0219) (0.0276) (0.0215) OLD1 -0.222*** -0.239*** -0.0926* -0.208*** (0.0442) (0.0347) (0.0492) (0.0317) WORKGR1 109.7*** 141.8*** 116.6*** 140.8*** (28.37) (28.94) (28.03) (28.98) FERT1 0.198 0.534 -4.605*** 0.0842 (0.359) (0.328) (0.854) (0.270) POPLIFE -0.0101*** (0.00370) FERTLIFE 0.0832*** (0.0132) Constant 5.371** 10.85*** 19.31*** 3.323 38.08*** 8.447*** (2.638) (2.754) (4.486) (2.355) (5.398) (1.222) Observations 1,499 1,498 1,472 1,447 1,47 1,447 #Countries 30 30 30 30 30 30 R-squared 0.039 0.063 0.109 0.146 0.136 0.144

Wald test p-value 0.0000 0.0000

Wu-Hausman test p-value 0.4806 0.5280

Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1.

In the second column the variable population growth was added. This variable shows a significant negative coefficient which means that an increase of population growth would reduce economic growth within a country. This confirms the findings of Bloom & Williamson (1998) which

26

stated that population growth would negatively affect the economic growth in the short run. There might be a delayed positive effect two decades later when the younger-aged people enter the working-age group. However, this variable is only lagged one period and therefore shows a negative coefficient. In the third column, all demographic variables concerning age groups were added to the equation. The variable of interest, the old-dependency ratio, shows a significant negative coefficient which means that an increase of this variable (so an increase in elderly people relative to working-age people) would negatively affect economic growth. This confirms earlier findings of Du & Wang (2011), Bloom, Canning & Fink (2010), Lee et al. (2013) and Bloom & Williamson (1998) which are all explained in chapter 3. However, the research of Bloom & Williamson (1998) stated that the coefficient for the old-dependency ratio might be positive. This is not shown in the regression for any scenario, as the coefficients are all negative and highly significant. Furthermore, the growth of the working-age population positively affects economic growth which would confirm earlier findings of Bloom & Sachs (1998) and Bloom & Williamson (1998). This also confirms the channel that an increase in the old-dependency ratio (thus the elderly relative to working-age population) negatively affects economic growth through the decrease of workforce compared to elderly, which is stated as the problem of population ageing. At last, the young dependency ratio and fertility rate don’t seem to have a significant impact on economic growth according to equation three. In the fourth column, the population growth and working-age population growth were estimated using their lags and the fertility rate as instruments through an IV-regression. This was done to eliminate a possible reversed causality of both growth variables on the real GDP per capita variable, since these variables might influence the growth of the real GDP per capita variable (Bloom & Finlay, 2009). The IV regression increases the R² value from 0.109 to 0.146 using the same equation, which means that a higher part of the variance in the data used for the dependent variable is explained by the IV model. By estimating the variables using instruments, both variables increased in impact and significance which might confirm the reversed causality problem discussed by Bloom & Finlay (2009). The Wald test shows that the coefficients from the variables used in this model statistically differ from zero, which means that they are relevant. Next, the Durbin-Wu-Hausman test evaluates whether and estimator (in our case the instruments) is consistent when compared to another estimator that is less efficient but known to be consistent. The p-value is not below the 10% significance border which means that the instruments which were used are consistent. One remark for equation three and four is that the average years of schooling variable now shows the same result as the research of Collins and Bosworth (1996), where they found a positive but insignificant coefficient for this variable. In the fifth equation interaction variables were added to see whether the life expectancy variable was interacting with fertility and population growth. Both variables significantly interact with life expectancy which confirms the problem of a biased life expectancy variable according to Barro & Lee

27

(1994) when including population growth, fertility rate and life expectancy in the same equation (explained earlier in chapter 3). Therefore in the last equation the life expectancy variable was excluded from the equation and again the IV-regression was done. The Wald test and the Wu-Hausman test show the same results when compared to equation four. From the last equation we can see that the old-dependency ratio is still significantly negative (-0.208) which shows that population ageing negatively affects economic growth. So this means that when the amount of elderly increases with 1% relatively to the working-age population, the economic growth would be reduced with a little more than 0.2%. This is interesting for our research since the old-dependency ratio keeps increasing for at least the next 30 years in China, which means that economic growth in China would significantly decrease in this period.

5.3 Estimating future effects of the ageing problem in China

From table 3, one can see the significant estimated effect of the old dependency ratio on economic growth. In the sixth equation, the old-dependency coefficient was -0.208. I used this equation since this is the most complete IV-regression without the life expectancy bias from Barro & Lee (1994). In this chapter I will use the coefficient of this variable to estimate what will happen to the economic growth in China for the next 30 years, by multiplying the coefficient with the projected old-dependency ratio for each year. This can be illustrated in this formula:

𝐸𝐹𝐹𝐸𝐶𝑇_𝑡 = −0.208 ∗ 𝑂𝐿𝐷_𝑡

In the formula, EFFECTt shows the effect that the old-dependency has on economic growth in that year, according to the old-dependency ratio in that year. So from year to year one can see the pattern illustrated in graph 5 which shows the estimated effect of population ageing on the economic growth in China for the next 30 years.

Graph 6: The effect of population ageing on economic growth (in %) in China for the next 30 days,

Data source: graph based on the regression and the projections of the DemProj program.

-9 -8 -7 -6 -5 -4 -3 -2 -1 0 20 14 20 16 20 18 20 20 20 22 20 24 20 26 20 28 20 30 20 32 20 34 20 36 20 38 20 40 20 42 20 44 Low fert. High fert. Elimination

28

In the graph, OLDt-1 starts at t=2014 such that the first estimate for EFFECTt will be at t=2014. One can interpret the percentages in the graph as ‘the amount of reduction of the economic growth percentage due to the population ageing effect’. The graph shows three different scenarios. From the graph it can be seen that the scenarios initially align with each other as differences are small until 2030 and then start to deviate. This might be explained as the total population is expected to grow until 2030 where it will hit its peak level and after that will start to decline (United Nations, 2015). The first scenario is the case when the relaxed one-child policy would be completely abolished (green line), assuming a fertility ratio of 2.3. This situation would result in an estimated reduction of 7.31% economic growth over 30 years for China. However, if they choose to keep the relaxed one-child policy, economic growth would reduce even more. In the ‘high fertility’ scenario (assuming a fertility rate of 1.8), economic growth is estimated to be reduced with 7.85%. In the current scenario where many people choose not to have another child as it might be more costly (the low fertility scenario, 1.3), the economic growth is estimated to be reduced with 8.49% over 30 years for China. So depending on what the government of China will do concerning the relaxed one-child policy, the economic growth is estimated to suffer between 7.31% and 8.49% over the next 30 years due to the population ageing problem. The numbers are given in table 4.

Table 4: Effect of population ageing on future GDP per capita growth in China,

YEAR 2020 2025 2030 2035 2040 2045

Elimination -3,55% -4,16% -4,90% -6,18% -7,59% -8,49%

High fertility -3,55% -4,16% -4.89% -6,04% -7,23% -7,85%

Low fertility -3,55% -4,16% -4,88% -5,91% -6,89% -7,31%