- Email: [email protected]
Demographic change and economic growth: An inverted-U shape relationship
Economics Letters 92 (2006) 447 – 454 www.elsevier.com/locate/econbase
Demographic change and economic growth: An inverted-U shape relationship Chong-Bum An a,*, Seung-Hoon Jeon b a
Department of Economics, Sungkyunkwan University, 3-53, Myongryun-dong, Chongro-gu, Seoul, 110-745, South Korea b National Assembly Budget Office (NABO), Seoul, South Korea Received 5 May 2005; received in revised form 17 March 2006; accepted 24 March 2006 Available online 24 July 2006
Abstract The cross-country regression and non-parametric kernel estimation using the panel data from OECD countries over the 1960–2000 periods show the inverted-U shape relationship between demographic changes and economic growth; growth rates initially increase and then decrease with population aging. D 2006 Elsevier B.V. All rights reserved. Keywords: Demographic change; Inverted-U shape relationship; Non-parametric estimation JEL classification: J11; O40; O50
1. Introduction The demographic trends over the past 50 years – which can be summarized as showing a steady decline in fertility rates and increase in life expectancy – is thought to have a powerful impact on the rate of economic growth. However, most empirical studies on the economic consequence of demographic change including Cutler et al. (1990), Bloom et al. (2000), Jones (2002), etc. find little cross-country evidence. Instead, there is a continuing debate over the demographic effects on economic growth. As discussed in Bloom et al. (2003), the debate involves three positions, such that demographic change (or * Corresponding author. Tel.: +82 2 760 0435; fax: +82 2 765 8138. E-mail address: [email protected] (C.-B. An). 0165-1765/$ - see front matter D 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2006.03.030
448
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
population growth) restricts, promotes or is independent of economic growth, representing the bPessimistic,Q bOptimistic,Q and bNeutralistQ theories respectively. The debate motivates us to attempt two tasks in this paper. First, we attempt to use more adequate demographic variables for studying its effect on economic performance. Fertility rate or life expectancy alone cannot be an accurate demographic indicator, since each captures only one part of population structure. As emphasized in Bloom and Canning (2004), the age structure of the population is an important measure of demographic change. So the index – the share of population over age 65 or the dependency ratios – provides richer information to determine economic performance than fertility rate or life expectancy alone. Since the age structure changes due to the combined effect of the fertility rate and life expectancy, both old age and young age dependency ratios can capture overall shape of demographic change in more appropriate way.1 So we use the share of the old (young) and the old age dependency ratio (young age dependency ratio) as the proxies of the demographic transition. For several decades we have observed a demographic transition from high fertility and high mortality to low fertility and low mortality.2 Since fertility and mortality rates do not decline at the same time, we can divide the demographic transition into three stages: high fertility/high mortality, high fertility/low mortality, and low fertility/low mortality. In the first stage, the share of the old (young) and the old (young) age dependency ratio are very low (high). Due to the longer life expectancy, the old share increases in the second stage. However, due to the increase in the young population caused by high fertility, the share of the old increases not rapidly but slowly. On the other hand, due to the increase in the old population, the share of the young decreases slowly. In the third stage, the share of the old increases rapidly and the share of the young decreases rapidly. Hence, increasing trends of both the old share and old age dependency ratio and decreasing trends of both the young share and young age dependency ratio can be used as the proxies for the demographic transition. Second, pursuing a more rigorous investigation, we attempt to test the relationship of growth to the demographic change in terms of parametric and non-parametric estimates. In fact, the entire body of previous empirical research on demographic influence on economic performance has never paid attention to the non-monotonic relationship between economic growth and demographic changes. However, the demographic impact on economic growth is not as monotonic as perceived in recent works.3 It may fluctuate to reflect a non-monotonic or non-linear relationship. As such, the nonlinearity can be tested in terms of its specifications including square and (or) cubic terms of demographic variables in the economic growth equation. In addition to this parametric test, we also attempt a nonparametric test, estimating the functional form itself in terms of the kernel regression. This paper is organized as follows. The first section introduces the importance of demographic change and explains the motivation to analyze its impact on economic growth. In Section 2, we explain the
1 Using dependency ratio rather than using both old age dependency ratio and young age dependency ratio together seems to be more appropriate, since dependency ratio reflects changes in both shares of the old and the young. However, it is very difficult to distinguish the high dependency ratio caused by high young age dependency ratio from that caused by high old age dependency ratio. The former appears in the earlier stage of demographic transition and the latter does in the later stage. These two show different effect on economic growth. Thus, we do not use the dependency ratio as the variable representing age structure. 2 See Lee (2003). bThe classic demographic transition starts with a mortality decline, followed after a time by reduced fertility, leading to an internal of first increased and then decreased population growth and, finally, population aging.Q 3 Zhang et al. (2003) theoretically analyzed the non-monotonic relationship of economic growth to demographic changes.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
449
econometric method and data. In Section 3, we report the estimation results. We state conclusions in Section 4.
2. Data, variables and estimation methods Most previous empirical works have relied on findings from one country’s experience or crosscountry data within a limited period of time. As the findings can neither be generalized to show evidence of the demographic effect on economic growth, nor can they be applied to the countries which have not experienced a significant population aging. Such the data limitation motivates us to use a richer panel data from 25 OECD countries over a 41 year period (1960–2000)4 to generalize the empirical findings on the relationship of economic growth to demographic change. Among the OECD countries, Luxembourg, Hungary, Poland, Slovak Republic, and Czech Republic were excluded as they do not have continuous annual series for most variables used in this paper. The growth account variables were extracted from Penn World Table version 6.1, while the demographic change variables and other control variables except average schooling years came from the world development indicator 2003 published by World Bank. Average schooling years came from Barro and Lee (2000). Economic growth is measured by log GDP per capita growth rate. The first estimation method is the simple cross-country regression using the pooled data covering 25 countries over the period 1960–2000. The basic specification for the cross-country regression is: PGDPGR ¼ C þ a1 LPGDPINI þ a2 INVR þ a3 OPEN þ a4 EDU þ a5 AGESTR where the dependent variable of PGDPGR is log GDP per capita growth rate and the explanatory variables LPGDPINI, INVR, OPEN, EDU, and AGESTR indicate the logarithm of initial GDP per capita, the total investment per GDP, import and export per GDP, average schooling years of the population aged 15 and over, and the variables representing the age structure respectively. Here, LPGDPINI captures conditional convergence and INVR is a measure of physical capital accumulation.5 Other control variables are meant to detect cross-country differences in the level and rate of growth of technology. In our formulation, we explore the shape of the relationship between demographic change and economic growth in three different functional forms: linear, quadratic and cubic. A key estimation method in this paper is the non-parametric kernel regression. When we explore the relationship between two variables, we generally rely on specific functional forms. For example, when we test the Kuznets’ hypothesis or age–income profile, we specify the regression function as a quadratic form and conclude that the inverted-U shape does exist. However, different functional forms can also generate the inverted-U shape. For example, Y = ax + b(1 / x) as well as Y = ax + bx 2 can yield the invertedU shape, depending on (a, b). Hence, specifying the regression function as a quadratic form and trying to draw conclusions from it may be misleading. The best way to resolve this can be using the non-
4
The single exception is Germany, which covers a shorter period, 1970–2000. Capturing conditional convergence reduces the possibility that the estimated contribution of aging to growth reflects that of initial per capita income. Kelly and Schmidt (1995), and Barro and Sala-I-Martin (2004) also used initial GDP per capita as an explanatory variable in the growth equation. 5
450
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
Table 1 Results of the cross-country regression 1 (1960–2000) Intercept LPGDPINI INVR OPEN EDU P65R
Spec.1
Spec.2
Spec.3
Spec.4
Spec.5
Spec.6
23.5858*** (5.1253) 2.7516*** (0.6554) 0.0881** (0.0368) 0.0013 (0.0070) 0.2001* (0.1103) 0.0963 (0.0700)
26.9573*** (4.5429) 3.5298*** (0.6246) 0.0816** (0.0315) 0.0033 (0.0060) 0.2361** (0.0952) 0.8810*** (0.2842) 0.0364** (0.0129)
27.8905*** (5.7108) 3.5240*** (0.6415) 0.0814** (0.0324) 0.0036 (0.0063) 0.2387** (0.0982) 0.4999 (1.3775) 0.0053 (0.1481) 0.0014 (0.0049)
23.0359*** (5.0466) 2.6954*** (0.6530) 0.0928** (0.0363) 0.0016 (0.0071) 0.1962* (0.1110)
27.1309*** (4.7517) 3.6418*** (0.6897) 0.0893** (0.0321) 0.0036 (0.0064) 0.2647** (0.1019)
28.4847*** (5.7163) 3.5656*** (0.7256) 0.0880** (0.0330) 0.0041 (0.0066) 0.2631** (0.1043)
0.0602 (0.0474)
0.6053** (0.2212) 0.0159** (0.0063)
0.5361
0.6372
0.1604 (1.0142) 0.0137 (0.0659) 0.0006 (0.0014) 0.6204
P65R2 P65R3 OAGDEP OAGDEP2 OAGDEP3 Adj. R 2
0.5423
0.6653
0.6472
Note: P65R2 (OAGDEP2) and P65R3 (OAGDEP3) represent quadratic and cubic term of P65R (OAGDEP) respectively. Standard errors in ( ). *: 10%, **: 5%, ***: 1% significance level.
parametric kernel regression, which does not depend on any functional form but estimates the functional form itself.
3. Empirical findings In Tables 1 and 2, we report the estimation results from the cross-country analysis. We use four different variables capturing the age structure in various specifications; the ratio of the old population aged 65 and over (P65R), old age dependency rate (OAGDEP), the ratio of young population aged between 0 and 14 (P014R), young age dependency rate (YDEP). In Table 1, we use the share of the old as variables that represent age structure in Spec.1, Spec.2, and Spec.3, and the old age dependency ratio in Spec.4, Spec.5 and Spec.6. In Table 2, we use the share of the young in Spec.7, Spec.8, and Spec.9, and the young age dependency ratio in Spec.10, Spec.11 and Spec.12. Here, we do not include both OAGDEP and YDEP (both P65R and P014R) together, since these are so highly correlated that entering both is unnecessary.6
6
The correlation coefficients are 0.754 and 0.881, respectively.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
451
Table 2 Results of the cross-country regression 2 (1960–2000) Intercept LPGDP60 INVR OPEN EDU P014r
Spec.7
Spec.8
Spec.9
Spec.10
Spec.11
Spec.12
28.9931*** (5.7202) 2.9338*** (0.5691) 0.0603 (0.0378) 0.0004 (0.0062) 0.1941* (0.0983) 0.0814** (0.0360)
21.8857*** (6.1244) 2.8098*** (0.5216) 0.0672* (0.0345) 0.0008 (0.0057) 0.1415 (0.0927) 0.3726* (0.2080) 0.0076** (0.0034)
8.0829 (14.3557) 2.9917*** (0.5472) 0.0617* (0.0348) 0.0008 (0.0057) 0.1419 (0.0923) 2.0948 (1.6346) 0.0674 (0.0565) 0.0007 (0.0006)
28.6040*** (5.2262) 2.9143*** (0.5360) 0.0556 (0.0371) 0.0009 (0.0061) 0.1809* (0.0948)
23.9284*** (5.9254) 2.7078*** (0.5362) 0.0671* (0.0367) 0.0003 (0.0059) 0.1411 (0.0954)
17.7562** (7.1823) 3.0349*** (0.5685) 0.0589 (0.0362) 0.0002 (0.0057) 0.1503 (0.0930)
0.0410** (0.0164)
0.0809 (0.0820) 0.0011 (0.0007)
0.6211
0.6452
0.6962 (0.4354) 0.0134 (0.0086) 0.0001 (0.0001) 0.6650
P014r2 P014r3 YDEP YDEP2 YDEP3 Adj. R 2
0.6032
0.6706
0.6729
Note: P014R2 (YDEP2) and P014R3 (YDEP3) represent quadratic and cubic terms of P014R (YDEP) respectively. Standard errors in ( ). *: 10%, **: 5%, ***: 1% significance level.
In Table 1, all the variables have reasonable signs. The sign of the LPGDPINI is negative, implying the existence of the conditional convergence. The signs of the INVR and EDU are positive, as expected. As Table 1 shows, the linear specifications in Spec.1 and Spec.4 and the cubic specifications in Spec.3 and Spec.6 are not significant. But the quadratic specifications in Spec.2 and Spec.5 show a significant inverted-U shape relationship, implying that the per capita GDP growth rate initially increases then decreases as either the ratio of the population aged 65 and over or old age dependency ratio rises. The findings of the inverted-U relationship in Table 1 highlight the effects of the demographic transition on economic growth. As mentioned above, we can divide the demographic transition into three stages: high fertility/high mortality, high fertility/low mortality, and low fertility/low mortality. This kind of demographic transition may affect economic growth with an inverted Ushaped relationship via its effects on labor supply. In the first and second stages of the demographic transition, the labor supply continuously increases and a large mass of the age pyramid is reflected in a young labor force. Thus the demographic effect on economic growth may appear to be positive. But in the third stage, a large population mass keeps moving upward to the older generation as the labor supply decreases. In summary, the demographic transition through three stages shows that the positive effect of demographic structure on growth becomes weaker, with the possibility of an eventual negative effect. The effect of the demographic
452
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
12
12
10
10
8
8
Growth Rate(%)
Growth Rate(%)
transition on the savings may also affect economic growth. According to the dependency-rate hypothesis proposed by Leff (1969), as the dependency rate increases, the working generation has a heavier family consumption burden, which then decreases the family saving rates and physical capital accumulation. In Table 2, the linear specifications in Spec.7 and Spec.10 are significant and the effects of age structure are shown to be negative. Since we observed the decreasing trend in the share of young and young age dependency ratio since 1960s, the negative sign indicates that growth rates increase as the share of the young and the young age dependency ratio decreases. The quadratic specification in Spec.8 shows a significant inverted-U shape relationship, implying that the growth rate initially increases then decreases as the share of the young keeps decreasing. The quadratic specification in Spec.11 and the cubic specifications in Spec.9 and Spec.12 are not significant. The implications of these results in Table 2 are analogous to those in Table 1. Decrease in the share of the young results in increase in both the share of working age group and total labor supply in the economy. And it also induces a smaller burden in family consumption for the working age group, which in turn increases the aggregate saving rate, the physical capital accumulation and economic growth. However, a continuous fall in the share of the young begins to produce less working population, which decreases total labor supply and the growth rate. Hence, the share of the young may have both the
6 4 2
6 4 2
0
0
-2
-2
-4
-4 2
4
6
8
10
12
14
16
18
4
12
12
10
10
8
8
6 4 2
12
16
20
24
28
6 4 2
0
0
-2
-2
-4
8
Old Age Dependency Ratio
Growth Rate(%)
Growth Rate(%)
Share of the Old
-4 12 16 20 24 28 32 36 40 44 48
Share of the Young
20
30
40
50
60
70
80
90 100
Young Age Dependency Ratio
Fig. 1. Result of the non-parametric kernel estimation.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
453
negative and positive effects on economic growth. This can explain the inverted-U shaped relationship shown in Spec.9. Fig. 1 reports the results of the non-parametric kernel regression. Its advantage is that we can investigate the functional form itself. The upper left side of the Fig. 1 shows the inverted-U shaped relationship between growth rate and the ratio of the population aged 65 and over. Economic growth rate increases during the early stage of demographic change, then decreases as the ratio of the population aged 65 and over further rises. The upper right side of the Fig. 1 also shows the inverted-U shaped relationship between growth rate and old age dependency ratio. As appeared in the lower two diagrams in Fig. 1, the inverted-U shaped relationship is also found in the relationships between the shares of the young (or young age dependency ratio) and economic growth. Based on these findings from the nonparametric kernel regression, we can also conclude that there exists the inverted-U shaped relationship between demographic changes and economic growth. Another interesting finding from the kernel fit is that economic growth rates decrease as the ratio of the population aged 65 and over (old age dependency ratio) rises over a certain level, roughly 7% (12%). This implies that the negative effect of demographic change on economy begins when the country begins to experience baging society.Q
4. Conclusion and further studies The empirical findings from two estimation methods attempted in this paper show that demographic changes appear to first increase and then decrease economic growth. This can be named as the Demographic U Hypothesis (Curve) — an inverted U-shape relationship between demographic change and economic growth. Based on these empirical findings, three subjects can be addressed for the further study. First, these findings will enable us to make a more accurate projection of the effect of aging on growth for the countries which recently began to experience the effects of an aging population. Second, while the impact of demographic change itself plays a significant role in economic performance, also very important is how quickly the change occurs. Korea is now confronting the fastest population aging in the world, while Japan has already suffered from a long-term economic recession due to a fast aging rate. Hence, as demographic change occurs more rapidly, we can expect the bigger impact on the economic growth. In addition to studying the demographic impact on economic growth, it is also meaningful to address the speed of the demographic changes which vary across countries depending on the timing of baby booms and busts, and the extent to which fertility rate and mortality rate decline in a given period. Third, we need a theoretical approach to the demographic U hypothesis, which might support the empirical finding of the demographic U curve. The theoretical investigation should also be applied to the channels of the demographic impact on the economic growth.
Acknowledgement We are grateful to Ronald Lee, Charles Jones and anonymous referee for their helpful comments and suggestions. This work is supported by the Brain Korea21 Project.
454
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
References Barro, Robert J., Lee, Jong-Wha, 2000. International Data on Educational Attainment: Updates and Implications. CID Working Paper, vol. 42. Barro, R.J., Sala-I-Martin, X., 2004. Economic Growth, Second edition Mcgraw-Hill, New York. Ch. 12. Bloom, D., Canning, D., 2004. Global Demographic Change: Dimensions and Economic Significance. NBER Working Paper, vol. 10817. Sept. 2004. Bloom, D.E., Canning, David, Malaney, Pia N., 2000. Demographic change and economic growth in Asia. Population and Development Review 26, 257 – 290 (Suppl.). Bloom, D., Canning, D., Sevilla, J., 2003. The Demographic Dividend: A New Perspective on the Economic Consequences of Population Change. RAND, Santa Monica, California. MR-1274. Cutler, David M., Poterba, James M., Sheiner, Louise M., Summers, Lawrence H., 1990. An Aging Society: Opportunity or Challenge? Brooking Papers on Economic Activity, vol. 1. Jones, C., 2002. Sources of U.S. economic growth in a world of ideas. American Economic Review 92 (1), 220 – 239. Kelly, A., Schmidt, R., 1995. Aggregate population and economic growth correlations: the role of the components of demographic change. Demography 32, 543 – 555. Lee, R., 2003. The demographic transition: three centuries of fundamental change. Journal of Economic Perspectives 17 (4), 167 – 190 (Fall). Leff, N.H., 1969. Dependency rates and saving rates. American Economic Review 59, 886 – 895. Zhang, J., Zhang, J., Lee, R., 2003. Rising longevity, education, savings, and growth. Journal of Development Economics 70, 83 – 101.
Demographic change and economic growth: An inverted-U shape relationship Chong-Bum An a,*, Seung-Hoon Jeon b a
Department of Economics, Sungkyunkwan University, 3-53, Myongryun-dong, Chongro-gu, Seoul, 110-745, South Korea b National Assembly Budget Office (NABO), Seoul, South Korea Received 5 May 2005; received in revised form 17 March 2006; accepted 24 March 2006 Available online 24 July 2006
Abstract The cross-country regression and non-parametric kernel estimation using the panel data from OECD countries over the 1960–2000 periods show the inverted-U shape relationship between demographic changes and economic growth; growth rates initially increase and then decrease with population aging. D 2006 Elsevier B.V. All rights reserved. Keywords: Demographic change; Inverted-U shape relationship; Non-parametric estimation JEL classification: J11; O40; O50
1. Introduction The demographic trends over the past 50 years – which can be summarized as showing a steady decline in fertility rates and increase in life expectancy – is thought to have a powerful impact on the rate of economic growth. However, most empirical studies on the economic consequence of demographic change including Cutler et al. (1990), Bloom et al. (2000), Jones (2002), etc. find little cross-country evidence. Instead, there is a continuing debate over the demographic effects on economic growth. As discussed in Bloom et al. (2003), the debate involves three positions, such that demographic change (or * Corresponding author. Tel.: +82 2 760 0435; fax: +82 2 765 8138. E-mail address: [email protected] (C.-B. An). 0165-1765/$ - see front matter D 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.econlet.2006.03.030
448
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
population growth) restricts, promotes or is independent of economic growth, representing the bPessimistic,Q bOptimistic,Q and bNeutralistQ theories respectively. The debate motivates us to attempt two tasks in this paper. First, we attempt to use more adequate demographic variables for studying its effect on economic performance. Fertility rate or life expectancy alone cannot be an accurate demographic indicator, since each captures only one part of population structure. As emphasized in Bloom and Canning (2004), the age structure of the population is an important measure of demographic change. So the index – the share of population over age 65 or the dependency ratios – provides richer information to determine economic performance than fertility rate or life expectancy alone. Since the age structure changes due to the combined effect of the fertility rate and life expectancy, both old age and young age dependency ratios can capture overall shape of demographic change in more appropriate way.1 So we use the share of the old (young) and the old age dependency ratio (young age dependency ratio) as the proxies of the demographic transition. For several decades we have observed a demographic transition from high fertility and high mortality to low fertility and low mortality.2 Since fertility and mortality rates do not decline at the same time, we can divide the demographic transition into three stages: high fertility/high mortality, high fertility/low mortality, and low fertility/low mortality. In the first stage, the share of the old (young) and the old (young) age dependency ratio are very low (high). Due to the longer life expectancy, the old share increases in the second stage. However, due to the increase in the young population caused by high fertility, the share of the old increases not rapidly but slowly. On the other hand, due to the increase in the old population, the share of the young decreases slowly. In the third stage, the share of the old increases rapidly and the share of the young decreases rapidly. Hence, increasing trends of both the old share and old age dependency ratio and decreasing trends of both the young share and young age dependency ratio can be used as the proxies for the demographic transition. Second, pursuing a more rigorous investigation, we attempt to test the relationship of growth to the demographic change in terms of parametric and non-parametric estimates. In fact, the entire body of previous empirical research on demographic influence on economic performance has never paid attention to the non-monotonic relationship between economic growth and demographic changes. However, the demographic impact on economic growth is not as monotonic as perceived in recent works.3 It may fluctuate to reflect a non-monotonic or non-linear relationship. As such, the nonlinearity can be tested in terms of its specifications including square and (or) cubic terms of demographic variables in the economic growth equation. In addition to this parametric test, we also attempt a nonparametric test, estimating the functional form itself in terms of the kernel regression. This paper is organized as follows. The first section introduces the importance of demographic change and explains the motivation to analyze its impact on economic growth. In Section 2, we explain the
1 Using dependency ratio rather than using both old age dependency ratio and young age dependency ratio together seems to be more appropriate, since dependency ratio reflects changes in both shares of the old and the young. However, it is very difficult to distinguish the high dependency ratio caused by high young age dependency ratio from that caused by high old age dependency ratio. The former appears in the earlier stage of demographic transition and the latter does in the later stage. These two show different effect on economic growth. Thus, we do not use the dependency ratio as the variable representing age structure. 2 See Lee (2003). bThe classic demographic transition starts with a mortality decline, followed after a time by reduced fertility, leading to an internal of first increased and then decreased population growth and, finally, population aging.Q 3 Zhang et al. (2003) theoretically analyzed the non-monotonic relationship of economic growth to demographic changes.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
449
econometric method and data. In Section 3, we report the estimation results. We state conclusions in Section 4.
2. Data, variables and estimation methods Most previous empirical works have relied on findings from one country’s experience or crosscountry data within a limited period of time. As the findings can neither be generalized to show evidence of the demographic effect on economic growth, nor can they be applied to the countries which have not experienced a significant population aging. Such the data limitation motivates us to use a richer panel data from 25 OECD countries over a 41 year period (1960–2000)4 to generalize the empirical findings on the relationship of economic growth to demographic change. Among the OECD countries, Luxembourg, Hungary, Poland, Slovak Republic, and Czech Republic were excluded as they do not have continuous annual series for most variables used in this paper. The growth account variables were extracted from Penn World Table version 6.1, while the demographic change variables and other control variables except average schooling years came from the world development indicator 2003 published by World Bank. Average schooling years came from Barro and Lee (2000). Economic growth is measured by log GDP per capita growth rate. The first estimation method is the simple cross-country regression using the pooled data covering 25 countries over the period 1960–2000. The basic specification for the cross-country regression is: PGDPGR ¼ C þ a1 LPGDPINI þ a2 INVR þ a3 OPEN þ a4 EDU þ a5 AGESTR where the dependent variable of PGDPGR is log GDP per capita growth rate and the explanatory variables LPGDPINI, INVR, OPEN, EDU, and AGESTR indicate the logarithm of initial GDP per capita, the total investment per GDP, import and export per GDP, average schooling years of the population aged 15 and over, and the variables representing the age structure respectively. Here, LPGDPINI captures conditional convergence and INVR is a measure of physical capital accumulation.5 Other control variables are meant to detect cross-country differences in the level and rate of growth of technology. In our formulation, we explore the shape of the relationship between demographic change and economic growth in three different functional forms: linear, quadratic and cubic. A key estimation method in this paper is the non-parametric kernel regression. When we explore the relationship between two variables, we generally rely on specific functional forms. For example, when we test the Kuznets’ hypothesis or age–income profile, we specify the regression function as a quadratic form and conclude that the inverted-U shape does exist. However, different functional forms can also generate the inverted-U shape. For example, Y = ax + b(1 / x) as well as Y = ax + bx 2 can yield the invertedU shape, depending on (a, b). Hence, specifying the regression function as a quadratic form and trying to draw conclusions from it may be misleading. The best way to resolve this can be using the non-
4
The single exception is Germany, which covers a shorter period, 1970–2000. Capturing conditional convergence reduces the possibility that the estimated contribution of aging to growth reflects that of initial per capita income. Kelly and Schmidt (1995), and Barro and Sala-I-Martin (2004) also used initial GDP per capita as an explanatory variable in the growth equation. 5
450
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
Table 1 Results of the cross-country regression 1 (1960–2000) Intercept LPGDPINI INVR OPEN EDU P65R
Spec.1
Spec.2
Spec.3
Spec.4
Spec.5
Spec.6
23.5858*** (5.1253) 2.7516*** (0.6554) 0.0881** (0.0368) 0.0013 (0.0070) 0.2001* (0.1103) 0.0963 (0.0700)
26.9573*** (4.5429) 3.5298*** (0.6246) 0.0816** (0.0315) 0.0033 (0.0060) 0.2361** (0.0952) 0.8810*** (0.2842) 0.0364** (0.0129)
27.8905*** (5.7108) 3.5240*** (0.6415) 0.0814** (0.0324) 0.0036 (0.0063) 0.2387** (0.0982) 0.4999 (1.3775) 0.0053 (0.1481) 0.0014 (0.0049)
23.0359*** (5.0466) 2.6954*** (0.6530) 0.0928** (0.0363) 0.0016 (0.0071) 0.1962* (0.1110)
27.1309*** (4.7517) 3.6418*** (0.6897) 0.0893** (0.0321) 0.0036 (0.0064) 0.2647** (0.1019)
28.4847*** (5.7163) 3.5656*** (0.7256) 0.0880** (0.0330) 0.0041 (0.0066) 0.2631** (0.1043)
0.0602 (0.0474)
0.6053** (0.2212) 0.0159** (0.0063)
0.5361
0.6372
0.1604 (1.0142) 0.0137 (0.0659) 0.0006 (0.0014) 0.6204
P65R2 P65R3 OAGDEP OAGDEP2 OAGDEP3 Adj. R 2
0.5423
0.6653
0.6472
Note: P65R2 (OAGDEP2) and P65R3 (OAGDEP3) represent quadratic and cubic term of P65R (OAGDEP) respectively. Standard errors in ( ). *: 10%, **: 5%, ***: 1% significance level.
parametric kernel regression, which does not depend on any functional form but estimates the functional form itself.
3. Empirical findings In Tables 1 and 2, we report the estimation results from the cross-country analysis. We use four different variables capturing the age structure in various specifications; the ratio of the old population aged 65 and over (P65R), old age dependency rate (OAGDEP), the ratio of young population aged between 0 and 14 (P014R), young age dependency rate (YDEP). In Table 1, we use the share of the old as variables that represent age structure in Spec.1, Spec.2, and Spec.3, and the old age dependency ratio in Spec.4, Spec.5 and Spec.6. In Table 2, we use the share of the young in Spec.7, Spec.8, and Spec.9, and the young age dependency ratio in Spec.10, Spec.11 and Spec.12. Here, we do not include both OAGDEP and YDEP (both P65R and P014R) together, since these are so highly correlated that entering both is unnecessary.6
6
The correlation coefficients are 0.754 and 0.881, respectively.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
451
Table 2 Results of the cross-country regression 2 (1960–2000) Intercept LPGDP60 INVR OPEN EDU P014r
Spec.7
Spec.8
Spec.9
Spec.10
Spec.11
Spec.12
28.9931*** (5.7202) 2.9338*** (0.5691) 0.0603 (0.0378) 0.0004 (0.0062) 0.1941* (0.0983) 0.0814** (0.0360)
21.8857*** (6.1244) 2.8098*** (0.5216) 0.0672* (0.0345) 0.0008 (0.0057) 0.1415 (0.0927) 0.3726* (0.2080) 0.0076** (0.0034)
8.0829 (14.3557) 2.9917*** (0.5472) 0.0617* (0.0348) 0.0008 (0.0057) 0.1419 (0.0923) 2.0948 (1.6346) 0.0674 (0.0565) 0.0007 (0.0006)
28.6040*** (5.2262) 2.9143*** (0.5360) 0.0556 (0.0371) 0.0009 (0.0061) 0.1809* (0.0948)
23.9284*** (5.9254) 2.7078*** (0.5362) 0.0671* (0.0367) 0.0003 (0.0059) 0.1411 (0.0954)
17.7562** (7.1823) 3.0349*** (0.5685) 0.0589 (0.0362) 0.0002 (0.0057) 0.1503 (0.0930)
0.0410** (0.0164)
0.0809 (0.0820) 0.0011 (0.0007)
0.6211
0.6452
0.6962 (0.4354) 0.0134 (0.0086) 0.0001 (0.0001) 0.6650
P014r2 P014r3 YDEP YDEP2 YDEP3 Adj. R 2
0.6032
0.6706
0.6729
Note: P014R2 (YDEP2) and P014R3 (YDEP3) represent quadratic and cubic terms of P014R (YDEP) respectively. Standard errors in ( ). *: 10%, **: 5%, ***: 1% significance level.
In Table 1, all the variables have reasonable signs. The sign of the LPGDPINI is negative, implying the existence of the conditional convergence. The signs of the INVR and EDU are positive, as expected. As Table 1 shows, the linear specifications in Spec.1 and Spec.4 and the cubic specifications in Spec.3 and Spec.6 are not significant. But the quadratic specifications in Spec.2 and Spec.5 show a significant inverted-U shape relationship, implying that the per capita GDP growth rate initially increases then decreases as either the ratio of the population aged 65 and over or old age dependency ratio rises. The findings of the inverted-U relationship in Table 1 highlight the effects of the demographic transition on economic growth. As mentioned above, we can divide the demographic transition into three stages: high fertility/high mortality, high fertility/low mortality, and low fertility/low mortality. This kind of demographic transition may affect economic growth with an inverted Ushaped relationship via its effects on labor supply. In the first and second stages of the demographic transition, the labor supply continuously increases and a large mass of the age pyramid is reflected in a young labor force. Thus the demographic effect on economic growth may appear to be positive. But in the third stage, a large population mass keeps moving upward to the older generation as the labor supply decreases. In summary, the demographic transition through three stages shows that the positive effect of demographic structure on growth becomes weaker, with the possibility of an eventual negative effect. The effect of the demographic
452
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
12
12
10
10
8
8
Growth Rate(%)
Growth Rate(%)
transition on the savings may also affect economic growth. According to the dependency-rate hypothesis proposed by Leff (1969), as the dependency rate increases, the working generation has a heavier family consumption burden, which then decreases the family saving rates and physical capital accumulation. In Table 2, the linear specifications in Spec.7 and Spec.10 are significant and the effects of age structure are shown to be negative. Since we observed the decreasing trend in the share of young and young age dependency ratio since 1960s, the negative sign indicates that growth rates increase as the share of the young and the young age dependency ratio decreases. The quadratic specification in Spec.8 shows a significant inverted-U shape relationship, implying that the growth rate initially increases then decreases as the share of the young keeps decreasing. The quadratic specification in Spec.11 and the cubic specifications in Spec.9 and Spec.12 are not significant. The implications of these results in Table 2 are analogous to those in Table 1. Decrease in the share of the young results in increase in both the share of working age group and total labor supply in the economy. And it also induces a smaller burden in family consumption for the working age group, which in turn increases the aggregate saving rate, the physical capital accumulation and economic growth. However, a continuous fall in the share of the young begins to produce less working population, which decreases total labor supply and the growth rate. Hence, the share of the young may have both the
6 4 2
6 4 2
0
0
-2
-2
-4
-4 2
4
6
8
10
12
14
16
18
4
12
12
10
10
8
8
6 4 2
12
16
20
24
28
6 4 2
0
0
-2
-2
-4
8
Old Age Dependency Ratio
Growth Rate(%)
Growth Rate(%)
Share of the Old
-4 12 16 20 24 28 32 36 40 44 48
Share of the Young
20
30
40
50
60
70
80
90 100
Young Age Dependency Ratio
Fig. 1. Result of the non-parametric kernel estimation.
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
453
negative and positive effects on economic growth. This can explain the inverted-U shaped relationship shown in Spec.9. Fig. 1 reports the results of the non-parametric kernel regression. Its advantage is that we can investigate the functional form itself. The upper left side of the Fig. 1 shows the inverted-U shaped relationship between growth rate and the ratio of the population aged 65 and over. Economic growth rate increases during the early stage of demographic change, then decreases as the ratio of the population aged 65 and over further rises. The upper right side of the Fig. 1 also shows the inverted-U shaped relationship between growth rate and old age dependency ratio. As appeared in the lower two diagrams in Fig. 1, the inverted-U shaped relationship is also found in the relationships between the shares of the young (or young age dependency ratio) and economic growth. Based on these findings from the nonparametric kernel regression, we can also conclude that there exists the inverted-U shaped relationship between demographic changes and economic growth. Another interesting finding from the kernel fit is that economic growth rates decrease as the ratio of the population aged 65 and over (old age dependency ratio) rises over a certain level, roughly 7% (12%). This implies that the negative effect of demographic change on economy begins when the country begins to experience baging society.Q
4. Conclusion and further studies The empirical findings from two estimation methods attempted in this paper show that demographic changes appear to first increase and then decrease economic growth. This can be named as the Demographic U Hypothesis (Curve) — an inverted U-shape relationship between demographic change and economic growth. Based on these empirical findings, three subjects can be addressed for the further study. First, these findings will enable us to make a more accurate projection of the effect of aging on growth for the countries which recently began to experience the effects of an aging population. Second, while the impact of demographic change itself plays a significant role in economic performance, also very important is how quickly the change occurs. Korea is now confronting the fastest population aging in the world, while Japan has already suffered from a long-term economic recession due to a fast aging rate. Hence, as demographic change occurs more rapidly, we can expect the bigger impact on the economic growth. In addition to studying the demographic impact on economic growth, it is also meaningful to address the speed of the demographic changes which vary across countries depending on the timing of baby booms and busts, and the extent to which fertility rate and mortality rate decline in a given period. Third, we need a theoretical approach to the demographic U hypothesis, which might support the empirical finding of the demographic U curve. The theoretical investigation should also be applied to the channels of the demographic impact on the economic growth.
Acknowledgement We are grateful to Ronald Lee, Charles Jones and anonymous referee for their helpful comments and suggestions. This work is supported by the Brain Korea21 Project.
454
C.-B. An, S.-H. Jeon / Economics Letters 92 (2006) 447–454
References Barro, Robert J., Lee, Jong-Wha, 2000. International Data on Educational Attainment: Updates and Implications. CID Working Paper, vol. 42. Barro, R.J., Sala-I-Martin, X., 2004. Economic Growth, Second edition Mcgraw-Hill, New York. Ch. 12. Bloom, D., Canning, D., 2004. Global Demographic Change: Dimensions and Economic Significance. NBER Working Paper, vol. 10817. Sept. 2004. Bloom, D.E., Canning, David, Malaney, Pia N., 2000. Demographic change and economic growth in Asia. Population and Development Review 26, 257 – 290 (Suppl.). Bloom, D., Canning, D., Sevilla, J., 2003. The Demographic Dividend: A New Perspective on the Economic Consequences of Population Change. RAND, Santa Monica, California. MR-1274. Cutler, David M., Poterba, James M., Sheiner, Louise M., Summers, Lawrence H., 1990. An Aging Society: Opportunity or Challenge? Brooking Papers on Economic Activity, vol. 1. Jones, C., 2002. Sources of U.S. economic growth in a world of ideas. American Economic Review 92 (1), 220 – 239. Kelly, A., Schmidt, R., 1995. Aggregate population and economic growth correlations: the role of the components of demographic change. Demography 32, 543 – 555. Lee, R., 2003. The demographic transition: three centuries of fundamental change. Journal of Economic Perspectives 17 (4), 167 – 190 (Fall). Leff, N.H., 1969. Dependency rates and saving rates. American Economic Review 59, 886 – 895. Zhang, J., Zhang, J., Lee, R., 2003. Rising longevity, education, savings, and growth. Journal of Development Economics 70, 83 – 101.