Japan & The World Economy 51 (2019) 100963

Contents lists available at ScienceDirect

Japan & The World Economy

journal homepage: www.elsevier.com/locate/jwe

Nonlinear effects of population aging on economic growth⋆ T

a b,⁎
Hyun-Hoon Lee , Kwanho Shin
a
Kangwon National University, Department of International Trade and Business, Chuncheon, 200-701, Republic of Korea
b
Korea University, Republic of Korea

ARTICLE INFO ABSTRACT

Keywords: Using panel data for 142 countries for the period from 1960 to 2014, we assess the effects of population aging on
Population aging economic growth. We find that population aging proxied by old-age population share (or old-age dependency
Non-linear effects ratio) negatively affects economic growth only when it reaches a certain high level and its negative effects grow
Economic growth stronger as population aging deepens. The nonlinear relationship between population aging and economic
growth is associated with the historical nonlinear relationship between the shares of old and working-age po-
pulation. In the early stages of a demographic shift in most countries, as the old-age population share increases,
the working-age population share also tends to increase. Only when the share of old-age population is suffi-
ciently high, the increase in the share of old-age population has coincided with the decline in the share of
working age population, thereby having a negative relationship with economic growth. These results can clarify
why some previous papers failed to uncover a negative relationship between aging population and economic
growth. We also find that population aging has hampered economic growth during more recent years, especially
in more aged countries which are mostly developed countries.

1. Introduction and Niimi, 2017). Thus, if a number of people in the working-age po-
pulation are replaced by an equivalent number among the old-age po-
During the past few decades, most countries have experienced rapid pulation, other things being equal, then population aging in a country
changes in the age structure of their populations. For example, many will hamper its economic growth, thus imposing a significant demo-
East Asian countries have experienced a rapid shift in their age struc- graphic burden.1
ture from a high youth-age population share to a high working-age However, previous empirical studies examining the overall impact
population share and then to a high old-age population share. A rapid of population aging on economic growth have often yielded mixed re-
increase in working-age population share occurred in these countries sults, as reviewed in Nagarajan et al. (2013). For example, using a panel
during the 1970s and 1980s and this transition contributed sub- dataset for the period 1960–2005, Bloom et al., 2008aBloom et al.,
stantially to East Asia’s so-called economic miracle (Bloom and (2008a, 2008b) find that the effect of old age on growth is negative in
Williamson, 1998). Such a period of demographic dividend did not last the short run but insignificant in the long run. Similarly, using the
long, however. These countries in Asia are now experiencing popula- partial adjustment model in a panel framework and a dataset for 80
tion aging at a rate higher than in any other regions. countries for the period 1960–2005, Lee et al. (2013) find that popu-
As compared to medium-age workers (more generally, the working- lation aging does not appear to hold back economic growth. More re-
age population), the elderly participate less actively in the labor force, cently, Acemoglu and Restrepo (2017) argue that countries experien-
their productivities are lower (Skirbekk, 2003; Aiyar et al., 2016; Liu cing more rapid aging have grown more rapidly because of the more
and Westelius, 2016), and they save less (Park and Shin, 2012; Horioka rapid adoption of automation technologies in these countries.

⋆
Presented at the 14th Annual Conference of Asia-Pacific Economic Association, University of Sothern California, USA, August 3–4 and at the 15th Korea and the
World Economy Conference, Jeju, Republic of Korea, June 25–26, 2018. Also presented at ADB Workshop on Technology and Aging Workforce Maximize the Gains
from Longevity and Long Working Life, Asiatic Research Institute, Korea University, Republic of Korea, 17–18 May 2018. We thank Yongok Choi, HyungGun Kim,
Kiseok Hong and anonymous referees for providing valuable comments. We also thank Dung Ly-My for her research assistance. We acknowledge the support by the
National Research Foundation of Korea Grant funded by the Korean Government (NRF-2017S1A5A2A03069146).
⁎
Corresponding author at: Department of Economics, Korea University, 5-1 Anam-Dong, Sungbuk-Ku, Seoul, 136-701, Republic of Korea.
E-mail addresses: [email protected] (H.-H. Lee), [email protected] (K. Shin).
1
The negative burden of aging can be mitigated by behavioral responses such as higher savings for retirement, higher investment in education, and greater labor
force participation. This line of argument can be found in Bloom et al. (2008a, 2008b), and Bloom et al. (2011).

https://doi.org/10.1016/j.japwor.2019.100963
Received 28 December 2018; Received in revised form 17 June 2019; Accepted 20 June 2019
Available online 29 June 2019
0922-1425/ © 2019 Elsevier B.V. All rights reserved.
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Fig. 1. Trend of population structure in Japan and Republic of Korea.(1960–2016).

Notes: Young-age share: population ages 0–14 (% of total), Working-age share: population ages 15–64 (% of total), Old-age share: population ages 65 and above (% of
total); Young dependency ratio: 100*young/work; and Old dependency ratio: 100*old/work
Source: Authors’ calculations based on World Bank’s World Development Indicators

In contrast, using a sample of 142 countries for the period share also tends to increase. This is illustrated in Fig. 1. As can be seen
1960–2014, Lee et al. (2017) find that population aging has a negative in the figure, Japan’s old-age population (65 and above) share has been
impact on economic growth in both the short run and the long run. increasing, from 5.6% in 1960 to 26.6% in 2016. During this period,
Eggertson et al. (2017) show that in a more recent period (2008–2015), Japan’s old-age dependency ratio also increased from 8.8% to 43.9%.
as compared to the period 1990–2008, the correlation between aging However, the working-age population share increased from 64.5% in
and GDP growth has become negative. Maestas et al. (2016) find that 1960 to 69.8% in 1992 and since then began to decrease. The figures in
using the U.S. states data for the period 1980–2010, population aging the right panel reveal that this is also true for Republic of Korea: both
decreases the growth rate of GDP per capita. the old-age share and the old-age dependency ratio have increased
Most, if not all, previous studies used the old-age population share while the working-age population share has increased, from 53.4% in
or the old-age dependency ratio as a proxy for aging and presume that 1960 to 73.4% in 2013. Only from 2014 did the working-age share
the effect of population aging on economic growth is linear, irrespective begin to decrease.
of the level of population aging. In theory, however, as explained in Fig. 2(A) redraws such a remarkable relationship between the old-
Section (2), while the relationship between economic growth and age population share and the working-age population share. In the case
working-age population share is linear, that between economic growth of Japan, until the old-age population share reached 12.3% in the year
and the old-age or young population share is nonlinear. Hence popu- 1991, the annual increases in the old-age population share were ac-
lation aging may impact economic growth negatively only when it companied by increases in the working-age population share. Since
reaches a certain high level and its impact gets stronger as population then, the working-age population share has decreased, while the old-
aging deepens. age population share has increased. In the case of Republic of Korea,
One of the major reasons why aging exerts nonlinear effects on until the old-age share reached 11.5% in 2013, population aging pro-
economic growth might be the changing structure of the entire popu- gressed, while the working-age population increased. As seen in
lation in history. One may expect that as the old-age population share Fig. 2(B), a very similar pattern is also observed if the old-age de-
increases, there will be relatively fewer people included in the working- pendency ratio replaces the old-age share.
age population. However, as the old-age population share increases, the As long as the demographic burden due to the increasing share of
sum of the youth-age and working-age population shares should de- old-age population is smaller than (or similar to) the demographic di-
crease. Therefore, an increasing old-age population share may coincide vidend due to the increasing share of working-age population is greater,
with an increasing working-age population share if the youth-age po- the “overall” effect of population aging on economic growth may not
pulation share decreases at a higher rate. necessarily be significantly negative. In this study, therefore, we sys-
Indeed, in the early stages of a demographic shift in most countries, tematically show that the nonlinear relationship between population
as the old-age population share increases, the working-age population aging and economic growth is associated with the historical nonlinear

2
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Fig. 2. Relationship between population aging and working-age population share in Japan and Republic of Korea.(1960–2016).
Notes: young_sh: population ages 0–14 (% of total), work_sh: population ages 15–64 (% of total), old_ sh: population ages 65 and above (% of total); young_dep:
100*young/work; and old_dep: 100*old/work
Source: Authors’ calculations based on World Bank’s World Development Indicators

relationship between the shares of old and working-age population. As 2. Empirical specification
this historical nonlinear relationship is combined with the positive
linear relationship between the share of working-age population and It is assumed that the aggregate output is determined by a three-
economic growth, a nonlinear relationship arises between the share of factor Cobb-Douglas production function3 :
old-age population and economic growth. When the share of old-age (1)
Y = AK aH L1 a
population is sufficiently high, the increase in the share of old-age
population has coincided with the decline in the share of working age where Y is gross domestic product (GDP), K is physical capital, H is
population, thereby having a negative relationship with economic human capital, L is labor force, and A is the productivity level. We
growth. These results can clarify why some previous papers failed to normalize Eq. (1) by dividing both sides by population, P, and then take
uncover a negative relationship between aging population and eco- the natural logarithm of both sides; we thus obtain:
nomic growth.2 ln y = ln A + ln k + ln H + (1 )ln(L / P ) ln P (2)
The remainder of this paper is organized as follows. Section 2 ex-
plains the empirical framework. Section 3 reports and discusses the where y is GDP per capita and k is physical capital per capita. By taking
main results. Lastly, provided in Section 4 are a summary and con- the time difference, Eq. (2) becomes:
cluding remarks. L
lny= lnA + lnk + lnH + (1 ) ln lnP
P (3)

3
Below we follow the exposition of Lee et al. (2013) and Lee et al. (2017) in
2
We thank one of the reviewers for clearly bring out this point. deriving the estimation equation.

3
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 1
Non-linear Effects of population aging on economic growth.
Source: Drawn from . Sources: Columns 1–3 are from Lee et al. (2017). Other columns are from authors’ calculations.

Notes: 1 Panel estimation with country-specific and period-specific effects. 2. Robust standard errors are in par-
entheses. 3. ***, **, and * indicate the significance levels of 1, 5, and 10 percent, respectively.

4
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 2
Average marginal effects of population aging on economic growth at different levels of population aging.
Source: Authors’ calculations.
(A) Old-age population share (Old_Sh)

Delta-method

dy/dx Std. Err. z P>z [95% Conf. Interval]

L.old_sh
_at
1 0.002103 0.014869 0.14 0.888 −0.02704 0.031246
2 0.000138 0.013834 0.01 0.992 −0.02698 0.027252
3 −0.00183 0.012816 −0.14 0.887 −0.02695 0.023291
4 −0.00379 0.011819 −0.32 0.748 −0.02696 0.019372
5 −0.00576 0.01085 −0.53 0.596 −0.02703 0.015508
6 −0.00772 0.009917 −0.78 0.436 −0.02716 0.011713
7 −0.00969 0.009031 −1.07 0.283 −0.02739 0.00801
8 −0.01166 0.008206 −1.42 0.155 −0.02774 0.004428
9 −0.01362 0.007463 −1.83 0.068 −0.02825 0.001007
10 −0.01559 0.00683 −2.28 0.022 −0.02897 −0.0022
11 −0.01755 0.006339 −2.77 0.006 −0.02998 −0.00513
12 −0.01952 0.006025 −3.24 0.001 −0.03133 −0.00771
13 −0.02148 0.005916 −3.63 0 −0.03308 −0.00989
14 −0.02345 0.006024 −3.89 0 −0.03525 −0.01164
15 −0.02541 0.006336 −4.01 0 −0.03783 −0.013
16 −0.02738 0.006826 −4.01 0 −0.04076 −0.014
17 −0.02935 0.007459 −3.93 0 −0.04396 −0.01473
18 −0.03131 0.0082 −3.82 0 −0.04738 −0.01524
19 −0.03328 0.009025 −3.69 0 −0.05096 −0.01559
20 −0.03524 0.009911 −3.56 0 −0.05467 −0.01582
21 −0.03721 0.010844 −3.43 0.001 −0.05846 −0.01595
22 −0.03917 0.011813 −3.32 0.001 −0.06232 −0.01602
23 −0.04114 0.012809 −3.21 0.001 −0.06624 −0.01603
24 −0.0431 0.013827 −3.12 0.002 −0.0702 −0.016
25 −0.04507 0.014862 −3.03 0.002 −0.0742 −0.01594
26 −0.04704 0.015911 −2.96 0.003 −0.07822 −0.01585
27 −0.049 0.016971 −2.89 0.004 −0.08226 −0.01574
28 −0.05097 0.018041 −2.83 0.005 −0.08633 −0.01561
29 −0.05293 0.019118 −2.77 0.006 −0.0904 −0.01546
30 −0.0549 0.020202 −2.72 0.007 −0.09449 −0.0153

(B) Old-age dependency ratio (Old_Dep)

Delta-method
dy/dx Std. Err. z P>z [95% Conf. Interval]
L.old_dep
_at
1 0.006451 0.008957 0.72 0.471 −0.0111 0.024006
2 0.005531 0.008515 0.65 0.516 −0.01116 0.022221
3 0.004611 0.008079 0.57 0.568 −0.01122 0.020445
4 0.003691 0.007647 0.48 0.629 −0.0113 0.018679
5 0.002771 0.007222 0.38 0.701 −0.01138 0.016927
6 0.001851 0.006805 0.27 0.786 −0.01149 0.015189
7 0.000931 0.006397 0.15 0.884 −0.01161 0.013469
8 1.07E-05 0.006 0 0.999 −0.01175 0.011771
9 −0.00091 0.005617 −0.16 0.871 −0.01192 0.010099
10 −0.00183 0.00525 −0.35 0.727 −0.01212 0.00846
11 −0.00275 0.004902 −0.56 0.575 −0.01236 0.006859
12 −0.00367 0.00458 −0.8 0.423 −0.01265 0.005307
13 −0.00459 0.004287 −1.07 0.284 −0.01299 0.003813
14 −0.00551 0.004031 −1.37 0.172 −0.01341 0.002392
15 −0.00643 0.00382 −1.68 0.092 −0.01392 0.001057
16 −0.00735 0.00366 −2.01 0.045 −0.01452 −0.00018
17 −0.00827 0.003558 −2.32 0.02 −0.01524 −0.0013
18 −0.00919 0.003521 −2.61 0.009 −0.01609 −0.00229
19 −0.01011 0.003549 −2.85 0.004 −0.01707 −0.00315
20 −0.01103 0.003641 −3.03 0.002 −0.01817 −0.00389
21 −0.01195 0.003793 −3.15 0.002 −0.01938 −0.00452
22 −0.01287 0.003998 −3.22 0.001 −0.02071 −0.00503
(continued on next page)

5
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 2 (continued)

(B) Old-age dependency ratio (Old_Dep)

23 −0.01379 0.004248 −3.25 0.001 −0.02212 −0.00546

24 −0.01471 0.004536 −3.24 0.001 −0.0236 −0.00582
25 −0.01563 0.004854 −3.22 0.001 −0.02514 −0.00612
26 −0.01655 0.005198 −3.18 0.001 −0.02674 −0.00636
27 −0.01747 0.005563 −3.14 0.002 −0.02837 −0.00657
28 −0.01839 0.005944 −3.09 0.002 −0.03004 −0.00674
29 −0.01931 0.006339 −3.05 0.002 −0.03174 −0.00689
30 −0.02023 0.006746 −3 0.003 −0.03345 −0.00701
31 −0.02115 0.007162 −2.95 0.003 −0.03519 −0.00711
32 −0.02207 0.007586 −2.91 0.004 −0.03694 −0.0072
33 −0.02299 0.008016 −2.87 0.004 −0.0387 −0.00728
34 −0.02391 0.008452 −2.83 0.005 −0.04048 −0.00734
35 −0.02483 0.008893 −2.79 0.005 −0.04226 −0.0074
36 −0.02575 0.009338 −2.76 0.006 −0.04405 −0.00745
37 −0.02667 0.009787 −2.73 0.006 −0.04585 −0.00749
38 −0.02759 0.010238 −2.69 0.007 −0.04766 −0.00752
39 −0.02851 0.010693 −2.67 0.008 −0.04947 −0.00755
40 −0.02943 0.011149 −2.64 0.008 −0.05128 −0.00758

Note: Calculated for Old-age population share (Old_Sh) in Column (3) of Table 1.
Note: Calculated for Old-age dependency ratio (Old_Dep) in Column (4) of Table 1.

where represents the time difference. Eq. (3) clearly shows that the
growth rate of per capita GDP depends positively on the growth rate of
a country’s labor force in the total population.
It is assumed that L is proxied by the working age population, so
that

P C O
lny= lnA + lnk + lnH + (1 ) ln
P
lnP (4)

where C is the youth population and O is the old population. Thus, Eq.
(4) also suggests that a country’s growth rate of per capita GDP depends
negatively on aged population as well as youth population. It should be
noted, however, that Eq. (4) suggests that the relationship between
population aging, measured by increase in the old population share
( ln( P )), and economic growth is not linear.
O

We use the same dataset as was utilized by Lee et al. (2017). GDP
per capita growth rates at five-year intervals are calculated for 142
countries for the period 1960–2014, using the Penn World Table ver-
sion 9.0′s national-accounts real GDP (RGDPNA).4 As is common in the
literature, we minimize the influence of business cycle fluctuations by
using five-year averages of variables for eleven sub-periods: (Period 1:
1960–1964), (Period 2: 1965–1969), (Period 3: 1970–1974), (Period 4:
1975–1979), (Period 5: 1980–1984), (Period 6: 1985–1989), (Period 7:
1990–1994), (Period 8: 1995–1999), (Period 9: 2000–2004), (Period
10: 2005–2009), and (Period 11: 2010–2014). We then calculate
average growth rates of GDP per capita for each 5-year interval.
Following Mankiw et al. (1992), we assume that y approaches its
steady state with partial adjustment and that the steady state is de-
termined by Eq. (2). As illustrated in Lee et al. (2013), if we further
assume that the change in the productivity of the economy, Δ ln(A) , is
influenced by the initial levels of physical capital, human capital and
trade openness, Eq. (5) can be converted to the following empirically
Fig. 3. Average marginal effects of population aging on economic growth at testable equation:
different levels of population aging.
Source: Drawn from Table 2

4
Among the five types of real GDP in the Penn World Table 9.0 reports, real
GDP using national-accounts (RGDPNA) is recommended by PWT for studies on
cross-country growth rates (Feenstra et al., 2015, 2016). In PWT version 9.0,
2014 is the latest year for which data is available.

6
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 3
Nonlinear specifications for old age population share.
Source: Authors’ calculations.

Notes: 1. The dependent variable is five-year growth rate of per capita GDP. 2. Panel
estimation with country-specific and period-specific effects. 3. Robust standard errors are
in parentheses. 4. ***, **, and * indicate the significance levels of 1, 5, and 10 percent,
respectively.
7
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 4
Nonlinear specifications for working-age population share.
Source: Authors’ calculations

Notes: 1. The dependent variable is five-year growth rate of per capita GDP. 2. Panel estimation with country-
specific and period-specific effects. 3. Robust standard errors are in parentheses. 4. ***, **, and * indicate the
significance levels of 1, 5, and 10 percent, respectively.

(ln rgpdnait ln rgpdnait 1) = ln rgdpoit + 9.0]

1 1 ln Popit 1
hc = Human capital index [PWT 9.0]6
+ 2 Youngit + 3 Oldit 1
1 ln Pop = Log of total population [World Bank’s WDI]
+ 4 ln Capit 1 + 5 hcit 1 + 6 ln Trdit 1 Young = YoungSh (Population aged below 15 as % of total) or
+ 7 (ln Popit ln Popit 1 ) + 8 (Youngit Youngit 1) YoungDep (100*population aged below 15/population aged between 15
and 65) [World Bank’s WDI]
+ 9 (Oldit Oldit 1)
Old = OldSh (Population aged above 65 as % of total) or OldDep
+ 10 (lnCapit ln Capit 1 ) + 11 (hcit hcit + 12 (ln Trdit ln Trdit
1) 1)
(100*population aged above 65/population aged between 15 and 65)
+ µi + µt + t [World Bank’s WDI]
(5) ln Trd = Log of trade as % of GDP (World Bank’s WDI).
Note that is the rate of adjustment (or speed of conversion), which
where
is bounded by zero and one. We estimate panel regressions with country
(ln rgdpnait ln rgdpnait 1) = Difference of log of real GDP per ca-
fixed effects.7 We also include period dummies as explanatory variables
pita between t
and t-1 [in 2011 national prices; PWT 9.0]
ln rgdpoit 1 = Log of initial level of output-side real GDP per capita 6
For details of human capital index, see http://www.rug.nl/ggdc/docs/
[in 2011 human_capital_in_pwt_90.pdf.
US$; PWT 9.0]5 7
Some researchers rely on the system GMM to estimate dynamic panel
ln Cap = Log of capital stock at PPPs [in million 2011 US$; PWT equations that take a similar form to Eq. (5). Note that, since we use output-side
real GDP per capita (RGDPO) instead of national-accounts real GDP per capita
(RGDPNA) as the initial level of real GDP per capita, our specification is not for
5
Instead of real GDP per capita in local currency, real GDP per capita in US$ dynamic panel equations. While RGDPNA is recommended to calculate real
is used for the initial level of GDP per capita, for the sake of comparison across growth rates, for the comparison of per capita GDP across countries, RGDPO is
countries. preferred by PWT.

8
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Fig. 5. Relationship between changes in demographic variables (1960–2014).

Fig. 4. Relationship between demographic variables.(1960–2014).
Notes: Young_age_share_c: Five-year changes in population ages 0–14 (% of
Notes: Young_age_share: population ages 0–14 (% of total), Working_age_share:
total), Working_age_share_c: Five-year changes in population ages 15–64 (% of
population ages 15–64 (% of total), and Old_age_share: population ages 65 and
total), and Old_age_share_c: Five-year changes in population ages 65 and above
above (% of total).
(% of total).
Source: Authors’ calculations based on World Bank’s World Development
Source: Authors’ calculations based on World Bank’s World Development
Indicators
Indicators

to control time-fixed global factors such as global capital market shocks. 3. Empirical results
As explained in Lee et al. (2013), we can interpret the coefficients
on level variable as the long-run effects, whereas the coefficients on the 3.1. Benchmark results
first-difference variables as the short-run adjustments to con-
temporaneous changes in the determinants of ln y . Columns (2) and (3) of Table 1 report the estimated results of Eq.
Lastly, as seen from Eq. (4), the relationship between population (5), whereas Column (1) reports the result when the working-age po-
aging ( ln( P )) and economic growth is not linear. Indeed, this point
O
pulation share is included in place of the old-age and young-age vari-
was briefly illustrated in Section 1 and will be further illustrated in ables. These results are identical to those reported in Tables 1 and 2 in
Section 3. Therefore, we need to add a quadratic term of old age po- Lee et al., 2017. The coefficients both on the initial value of old-age
pulation share in Eq. (5).8 Further, we will utilize young-and old-age population share and on its change are negative and significant.
population shares as percentages of total population as well as young- Therefore, population aging may have a negative impact on economic
and old-age dependency ratios (i.e. young and old-age population re- growth not only in the short run but also in the long run. More speci-
lative to working age population). fically, a one percentage point increase in the old-age population share
In contrast, as seen from Eqs (2) and (3), the relationship between decreases the five-year economic growth rate by 2.0 percentage points

9
()
change in working-age population share ( ln P ) and economic growth
L
in the long run, leading to a lower steady state income per capita. If the
change in the old-age share increases by one percentage point over a 5-
( lny) is linear.
year period, the five-year economic growth rate will decrease by 3.4

8
When we add a cubic term of old-age population share, its estimated coef-
ficient is statistically insignificant. (footnote continued)
9
A decrease in the working-age share will correspond to an increase in the share can be used as a proxy for the old-age share only when the young-age
sum of the shares of the young-aged and old-aged. Therefore, the working-age share is kept constant. We thank an anonymous referee for this point.

9
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

percentage points.
Column (3) of the table reports the results when old and young
population dependency ratios (YoungDep and OldDep) are used in place
of the old and young population shares (YoungSh and OldSh). These
results also reveal that population aging has a negative impact on
economic growth both in the long run and in the short run. In terms of
the size of the estimated coefficients, old-age population, as compared
to youth population, has roughly double the negative impact on eco-
nomic growth.10
As seen in Section 2, a linear regression may not reveal the asso-
ciation between population aging and economic growth because the
true relationship is nonlinear. Column (4), therefore, reports the results
with the addition of a quadratic term of old-age population share
(OldSh).11 As expected, the quadratic term of old-age population share
enters with a negative and significant coefficient, while OldSh is no
longer statistically significant. A similar result is obtained when old-age
dependency ratio (OldDep) is used instead of OldSh.
In Tables 2(A) and (B) and in Figs. 3(A) and (B), we present how the
marginal effect of population aging on economic growth changes as
population aging deepens. They show clearly that, as either old-age
population share (Table 2(A) and Fig. 3(A)) or old-age dependency
ratio (Table 2(B) and Fig. 3(B)) increases, its marginal effect on eco-
nomic growth becomes greater. Standard errors and 95% confidence
levels are calculated by the Delta-method. In Table 2(A), the marginal
effect of OldSh becomes statistically significant at the five percent level
only when it reaches 10 percent of total population. As population
aging deepens, its negative marginal effect on economic growth con-
tinues to become greater economically and more significant statisti-
cally. More specifically, at the 10/ 15/ 20 percent level of old-age po-
pulation, a one-percentage point increase in old-age population
decreases the five-year economic growth rate by 1.6/ 2.5/ 3.5 per-
centage points. Table 2(B) also indicates that at the 15/ 20/ 25 percent
level of old-age dependency ratio, a one-percentage point increase in
old-age dependency ratio decreases the five-year economic growth rate
by 0.6/1.1/1.6, respectively.12
In order to confirm the robustness of our finding that there exists a
nonlinear relationship between old-age population share and economic
growth, we explore various specifications in Table 3. In Table 3A, we
report estimation results when a quadratic term of young-age popula-
tion share is also included.13 In column (1)-(3), we ignore short-term
adjustment terms by assuming that the adjustment speed is infinite. In
columns (1) and (4), we do not include human capital and trade
openness as regressors. In columns (2) and (5), we exclude just trade
openness. Interestingly we find that the coefficient of a quadratic term
of young-age population share is not statistically significant in any
specification. In contrast, the coefficient of a quadratic term of old-age
population share is statistically significant at the 5 percent level in
columns (2), (3) and (5) and at the 10 percent level in column (6). In
Table 3B, we report regression results when we exclude young-age
population share from regressors. In this case, the coefficient of a

10
Among the control variables, the coefficient on the initial value of the log
of population is negative and significant. Countries with a larger amount of
Fig. 6. Relationship between change in old-age share vs. change in working-age physical capital are found to enjoy higher rates of economic growth both in the
share.(1960–2014). short run and in the long run. Countries that are more open to trade are also
Notes: Working_age_share_c: Five-year changes in population ages 15–64 (% of found to grow faster.
11
total), and Old_age_share_c: Five-year changes in population ages 65 and above Since the coefficient of a cubic term is not statistically significant, we do
(% of total). In our sample, the average level of old-age population share is not explore higher order terms.
12
6.32. As a reference, old-age population share and old-age dependency ratio of
Source:Authors’ calculations based on World Bank’s World Development the Republic of Korea in 2010 were 10.3 and 14.6, respectively, while those of
Indicators Japan in 2010 were 22.5 and 35.1, respectively.
13
In fact, a non-linear relationship that arises when we take the second-order
Taylor expansion of the log-term in Eq. (4). This non-linear relationship would
apply to both the old-age share and the young-age share. However, as explained
in the text, the evidence for the nonlinear relationship for the young-age share
is weak.

10
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 5
Nonlinear relationship between working-age population share and old/young population share.
Source: Authors’ calculations.

Notes: 1. The dependent variable is working-age population share (panel A) and its five-year difference (panel B).
2. Columns (1) and (4) are pooling OLS estimation, columns (2) and (5) are panel estimation with random effects
and columns (3) and (6) are panel estimation with country-specific and period-specific effects. 3. Robust standard
errors are in parentheses. 4. ***, **, and * indicate the significance levels of 1, 5, and 10 percent, respectively.

quadratic term of old-age population share is even more statistically share exceeds the 10 percent level, higher old-age population shares
significant. It is statistically significant at the 1 percent level in columns appear to be associated with smaller working-age population shares.15
(2), (3), (5) ad (6) and at the 5 percent level in the rest of the columns. Fig. 5 illustrates the same relationship in changes. Fig. 5(A) reveals
Overall our findings strongly suggest that there exists a nonlinear re- that countries with greater increases (five-year) in youth-age popula-
lationship between population aging proxied by old-age population tion share are those with smaller rates of change in working-age po-
share and economic growth. pulation share. However, Fig. 5(B) reveals that the relationship be-
However, as seen from Eq. (3), the theoretical relationship between tween changes in old-age population share and changes in working-age
working-age population share and economic growth is linear. In population share is not clearly identifiable.
Table 4, we report regression results when we use working-age popu- In Table 5, to show the above evidence more systematically, we
lation share as a regressor, instead of old-age population share (and report regression results. In Table 5A, we report estimation results for
young-age population share). In Table 4, the coefficient of a quadratic the nonlinear specifications in levels between working-age population
term of working-age population share is not statistically significant share and old-age population share (or young-age population share). In
except in column (6) where it is marginally significant at the 10 percent all columns the dependent variable is working-age population share.
level. The regressors are both linear and quadratic terms of old-age popula-
As briefly discussed in Section 1, the crucial point is that higher old- tion share (columns (1)-(3)) and young-age population share (columns
age population share does not necessarily coincide with lower working- (4)–(6)). Columns (1) and (4) are pooling OLS estimation, columns (2)
age population share. Using the panel data for our empirical analysis, and (5) are panel estimation with random effects and columns (3) and
Figs. 4–6 support this view. Specifically, Fig. 4(A) reveals that countries (6) are panel estimation with country-specific and period-specific ef-
with greater youth-age population shares are those with smaller fects. In all columns, the coefficient of a quadratic term, irrespective of
working-age shares. Fig. 4(B), however, reveals that the relationship whether it is of old-age population share or of young-age population
between old-age share and working-age share is nonlinear. For the
countries with old-age population share of about 10% or smaller, a
greater share of old-age population appears to coincide with a greater
(footnote continued)
share of working-age population.14 However, if old-age population countries in the sample may affect our key results but we still find similar re-
sults.
15
We also find similar results when we examine the correlation between
14
ARE (United Arab Emirates), QAT (Qatar), and BHR (Bahrain) appear to be working-age share and old-age dependency ratio (defined as the ratio of old-age
outliers and hence in the regression analysis we check if dropping these three to working age population).

11
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 6
Stronger effects of population aging on economic growth in recent years.
Source: Authors' calculations.

Notes: 1. Panel estimation with country-specific and period-specific effects. 2. Robust standard errors are in par-
entheses. 3. ***, **, and * indicate the significance levels of 1, 5, and 10 percent, respectively.

12
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 7
Average marginal effects of population aging on economic growth at different periods of time.
source: Authors' calculations.
(A) Old-age population share (Old_Sh)

Delta-method

dy/dx Std. Err. z P>z [95% Conf. Interval]

L.old_sh
_at
1 0.000846 0.008956 0.09 0.925 −0.01671 0.018399
2 −0.00156 0.008289 −0.19 0.851 −0.01781 0.014684
3 −0.00397 0.007662 −0.52 0.604 −0.01899 0.011046
4 −0.00638 0.007085 −0.9 0.368 −0.02026 0.007508
5 −0.00879 0.006572 −1.34 0.181 −0.02167 0.004094
6 −0.01119 0.006138 −1.82 0.068 −0.02322 0.000837
7 −0.0136 0.005802 −2.34 0.019 −0.02497 −0.00223
8 −0.01601 0.005582 −2.87 0.004 −0.02695 −0.00507
9 −0.01842 0.00549 −3.35 0.001 −0.02918 −0.00766
10 −0.02083 0.005534 −3.76 0 −0.03167 −0.00998
11 −0.02323 0.005711 −4.07 0 −0.03443 −0.01204

(B) Old-age dependency ratio (Old_Dep)

Delta-method

dy/dx Std. Err. z P>z [95% Conf. Interval]

L.old_dep
_at
1 0.00425 0.004841 0.88 0.38 −0.00524 0.013738
2 0.002465 0.004452 0.55 0.58 −0.00626 0.011191
3 0.000681 0.004099 0.17 0.868 −0.00735 0.008716
4 −0.0011 0.003794 −0.29 0.771 −0.00854 0.006332
5 −0.00289 0.003548 −0.81 0.416 −0.00984 0.004065
6 −0.00467 0.003374 −1.39 0.166 −0.01129 0.00194
7 −0.00646 0.003283 −1.97 0.049 −0.01289 −2.2E-05
8 −0.00824 0.003283 −2.51 0.012 −0.01468 −0.00181
9 −0.01003 0.003374 −2.97 0.003 −0.01664 −0.00341
10 −0.01181 0.003548 −3.33 0.001 −0.01876 −0.00486
11 −0.0136 0.003794 −3.58 0 −0.02103 −0.00616

Note: Calculated for Old-age population share (Old_Sh) in Column (5) of Table 6.
Note: Calculated for Old-age population share (Old_Sh) in Column (6) of Table 6.

share, is statistically significant at the 1 percent level. Hence we find a The nonlinear relationship between working-age population share
strong evidence of the nonlinear relationship. However, the size of the and old-age population share makes the relationship reversed as we
coefficient is much larger for old-age population share (-0.09, -.13, and group countries based on the degree of their aging level. Fig. 6 illus-
-.16) than for young-age population share (-.01, -.01 and -.01), in- trates this by grouping countries in three different levels of aging.
dicating that working-age population share has a much stronger non- Fig. 6(A) clearly reveals that among the non-aged countries (old-age
linear relationship with old-age population share than with young-age population share below 6.32, the average level in our sample), an in-
population share. Indeed, as shown in Fig. 4(A), the relationship be- crease in old-age population share generally coincides with an increase
tween young-age population share and working-age population share in working-age population share. However, Fig. 6(B) reveals that for the
can be approximated by a linear relationship. However, as presented in group of aged countries (old-age population share > 6.32), the re-
Fig. 4(B), the relationship between old-age population share and lationship appears to be negative, and Fig. 6(C) further reveals that
working-age population share can hardly be approximated by a linear such a negative relationship becomes stronger for the group of more
relationship. aged countries (old-age population share > 10.0).
In Table 5B, we also report the regression results for the relationship A very similar pattern is observed even when we replace old-age
in changes. The dependent variable is change in working-age popula- population share with old-age population dependency ratio, which is
tion share. The regressors are both linear and quadratic terms of change defined as the ratio of old-age to working-age population.16
in old-age population share (columns (1)-(3)) and of change in young- Thus, a linear regression may not reveal the true association be-
age population share (columns (4)-(6)). Here, again, the nonlinear re- tween population aging and economic growth when population aging is
lationship is much more pronounced for old population share. While still at a low level, because the positive impact of increasing working-
the coefficient of a quadratic term of old population share is statistically age share on economic growth is simultaneously at play.
significant at the 1 percent level in all columns (1)-(3), that for young-
age population share is not statistically significant in columns (5) and
(6).

16
Not shown for the sake of brevity but available upon request.

13
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Section 3.1. As seen in Columns (5) and (6), the interaction terms enter
with very significant negative coefficients. Table 7(A) and (B) as well as
Fig. 7(A) and (B) illustrate that as the data include more recent periods,
the marginal effect of population aging on economic growth becomes
greater.
This result is consistent with the findings of Eggertson et al. (2017),
who find a negative correlation between aging and GDP growth when
they use the data for the period 2008–2015 instead of 1990-2008. Thus,
as the world’s aging population continues to grow older dramatically
(United Nations, 2017), it is likely that the world may enter an age of
secular stagnation, as suggested by Summers (2013, 2015) and
Eggertson et al. (2017).

3.3. Effects of population aging in different groups of countries

Rising life expectancy and decreasing fertility rate emerged first in

more developed countries and hence population aging is more ad-
vanced in developed countries than in developing countries. If our
findings above are correct, the negative impact of population aging on
economic growth should be stronger in developed countries, where
population aging is more advanced. Table 8 reports the estimated re-
sults for 29 OECD countries17 as well as for other countries. In our
sample, the mean value of old-age population share for 29 OECD
countries is 12.4, while that for other countries is 4.9.
When population aging is expressed as old-age population share
(Columns 1 and 3), the absolute size of its coefficient is slightly smaller
in the OECD country group but its significance is far greater in the
OECD country group. When old-age dependency ratio is used, the po-
pulation aging variable no longer carries a significant coefficient for
non-OECD countries, while it continues to have a highly significant
negative impact on economic growth for the 29 OECD countries.
We also ran regressions separately for more aged country groups
and less aged country groups where more age country groups are
compirised of countries whose old-age population shares or old-age
dependency ratios are greater than the averages of the sample. As seen
Fig. 7. Average marginal effects of population aging on economic growth at
different perioods of time. in Table 9, population aging has significant negative impacts on eco-
nomic growth only in more aged country groups, not only in the long
run but also in the short run.

3.2. Effects of population aging in different periods of time 4. Summary and concluding remarks

As discussed in Section 1, earlier studies in the literature often fail to Earlier studies in the literature often fail to find a negative impact of
find a statistically significant negative impact of population aging on population aging on economic growth. This might be due to the fact
economic growth. This might be due to the fact that the data used by that in early stages of population aging, most countries experiencing
earlier studies did not include the recent period of high population population aging do not experience a decline of the proportion of their
aging and hence with a relatively lower level of population aging, the working age population (15–64 years) to total population.
negative impact of population aging was not captured. Using panel data for 142 countries for the period from 1950 to
In order to assess this possibility, we estimate Equation (6) while 2014, we have assessed the effects of population aging on economic
restricting the sample to the period until the year 1999. The estimated growth.
results are reported in Columns (3) and (4) of Table 6. The corre- We have found that population aging proxied by old-age population
sponding results when using the whole sample are also reported in share (or old-age dependency ratio) hampers economic growth to the
Columns (1) and (2) for the sake of comparison. As expected, the es- greatest extent in countries where population aging has already reached
timated coefficients of population aging variables (OldSh and OldDep) a high level. In contrast, there is a positive linear relationship between
are no longer statistically significant, even though they enter with ne- the share of working-age population and economic growth. Thus, the
gative signs. nonlinear relationship between population aging and economic growth
In order to assess whether the negative effects of population aging is due to the historical nonlinear relationship between the shares of old
on economic growth have been continuously growing stronger over and working-age population, i.e. at the early stage of aging, both shares
time, we re-estimated Eq. (5) with the addition of an interaction term of of old and working-age population increased together and only at the
the population aging variables (OldSh and OldDep) and a period vari-
able. Note that there are eleven periods in our sample, as explained in

17
Other OECD member countries which became members after 2010 are not
included. They are Chile, Estonia, Israel, Latvia, Slovak Republic, and Slovenia.
Even if these countries are also included in the OECD country group, the results
are similar.

14
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 8
Stronger Negative Effects of population aging in developed countries.
Source: Authors' calculations

Notes: 1. Panel estimation with country-specific and period-specific effects. 2. Robust standard errors are in
parentheses. 3. ***, **, and * indicate the significance levels of 1, 5, and 10 percent, respectively.

15
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Table 9
Negative Effects of population aging only in more aged countries.
Source: Authors' calculations

Notes: 1. Panel estimation with country-specific and period-specific effects. 2. Robust standard errors are in par-
entheses. 3. ***, **, and * indicate the significance levels of 1, 5, and 10 percent, respectively.

later stage of aging, the increase in the share of old-age population We have further found that the negative effect of population aging
coincided with the decline in the share of working age population. on economic growth has become greater in more recent years. This
These results explain why aging has a negative relationship with eco- finding is also related to the fact that more aged economies appear in
nomic growth only when the share of old-age population is sufficiently more recent years.
high.

16
H.-H. Lee and K. Shin Japan & The World Economy 51 (2019) 100963

Appendix A

Table A1.

Table A1
Sources/Definitions of Variables.
Variables Description and construction Data Source

Real GDP Per Capita Output-side real GDP at chained PPPs (in mil. 2011US$) Penn World Table 9.0
Population Total Population World Bank’s World Development Indicators
Working Age Population Share Population ages 15-64 (% of total population) World Bank’s World Development Indicators
Young Population Share Population ages 0-14 (% of total population) World Bank’s World Development Indicators
Old Population Share Population ages 65 and above (% of total population) World Bank’s World Development Indicators
Young Age Dependency Ratio Ratio of people younger than 15 to the working age population World Bank’s World Development Indicators
Old Age Dependency Ratio Ratio of people older than 64 to the working age population World Bank’s World Development Indicators
Capital Stock Capital Stock at current PPPs (in mil. 2011 US$) Penn World Table 9.0
Human Capital Human Capital Index Penn World Table 9.0
Trade Share % of GDP World Bank’s World Development Indicators

References Lee, H.-H., Huh, H.-s, Lee, Y.-Y., Lim, J.-Y., 2013. Effects of Population Aging on
Economic Growth: A Panel Analysis. Seoul J. Econ. 26 (4), 401–432.
Lee, H.-H., Shin, K., Park, D., 2017. Population aging and its impact on economic growth:
Acemoglu, D., Restrepo, P., 2017. Secular stagnation? The effects of aging on economic implications for Korea. Economic Analysis, ESRI 196, 159–178.
growth in the age of automation. American Economic Review 107 (5), 174–179. Liu, Y., Westelius, N., 2016. The Impact of Demographics on Productivity and Inflation in
Aiyar, Sh., Ebeke, C., Shao, X., 2016. The Impact of Workforce Aging on European Japan. IMF Working Paper WP/16/237, International Monetary Fund.
Productivity”, IMF Working Paper, WP/16/238. International Monetary Fund. Maestas, N., Mullen, K.J., Powell, D., 2016. The Effect of Population Aging on Economic
Bloom, D.E., Canning, D., Finlay, J., 2008a. ), “Population Aging and Economic Growth in Growth, Labor Force and Productivity. NBER Working Paper No. 22452.
Asia,” PGDA Working Paper No.40. Harvard School of Public Health. Mankiw, N.G., Romer, D., Weil, D., 1992. A Contribution to the Empirics of Economic
Bloom, D.E., Canning, D., Fink, G., 2008b. “Population Aging and Economic Growth,” Growth. Q. J. Econ. 152 (May), 407–437.
PGDA Working Paper No.31. Harvard School of Public Health. Nagarajan, R.A., Teixeira, A.C., Silva, S., 2013. “The Impact of an Ageing Population on
Bloom, D.E., Canning, D., Fink, G., 2011. “Implications of Population Aging for Economic Economic Growth: An Exploratory Review of the Main Mechanisms”. FEP Working
Growth,” PGDA Working Paper No.64. Harvard School of Public Health. Papers No. 504.
Bloom, D.E., Williamson, J.G., 1998. “Demographic transitions and economic miracles in Park, D., Shin, K., 2012. Impact of Population Aging on Asia’s Future Growth. In: Park, D.,
emerging asia”. World Bank Econ. Rev. 12 (30), 419–455. Lee, S.-H., Mason, A. (Eds.), Aging, Economic Growth, and Old-Age Security in Asia.
Eggertson, G., Lancastre, M., Summers, L.H., 2017. Aging, Output Per Capita and Secular Edward Elgar.
Stagnation. Institute for New Economic Thinking, mimeo. https://www. Skirbekk, V., 2003. “Age and Individual Productivity: A Literature Survey”, MPIDR
ineteconomics.org/research/research-papers/aging-output-per-capita-and-secular- Working Paper WP 2003-028. Max-Planck Institute for Demographic Research.
stagnation. Summers, L., 2013. Why Stagnation Might Prove to Be the New Normal. The Financial
Feenstra, R.C., Inklaar, R., Timmer, M.P., 2015. The Next Generation of the Penn World Times.
Table. Am. Econ. Rev. 105 (10), 3150–3182. Summers, L., 2015. Demand side secular stagnation. Am. Econ. Rev.: Pap. Proc. 105 (5),
Feenstra, R.C., Inklaar, R., Timmer, M.P., 2016. “What Is New in PWT 9.0″. http://www. 60–65.
rug.nl/ggdc/docs/what_is_new_in_pwt90.pdf. United Nations, 2017. World Population Prospects The 2017 Revision: Key Findings and
Horioka, C.Y., Niimi, Y., 2017. Saving Behavior of Japanese Middle-aged and the Elderly. Advance Tables. Department of Economic and Social Affairs, United Nations.
ESRI International Conference on Empirical Analysis on Population Decline and
Aging in the Japanese Economy, ESRI, Tokyo.

17