Goh 2020

Journal of Asian Economics 68 (2020) 101198
Contents lists available at ScienceDirect
Journal of Asian Economics
Full length article
Macroeconomic implications of population aging: Evidence

from Japan$
Soo Khoon Goha,* , Robert McNownb , Koi Nyen Wongc
a
Centre for Policy Research & International Studies, Universiti Sains Malaysia, Malaysia
b
Department of Economics, University of Colorado Boulder, USA
c
Department of Economics and Finance, Sunway University, 46150 Selangor, Malaysia
A R T I C L E I N F O A B S T R A C T
Article history: Population aging is an important feature of Japan’s economy, which since 2006 has become
Received 28 November 2019 a super-aged society. Changes in the age distribution of the population have important
Received in revised form 14 March 2020 macroeconomic implications. Using annual data for 1960–2015, this study tests whether
Accepted 13 April 2020
population age shares have long run influences on domestic saving, domestic investment,
Available online 1 May 2020
real GDP, inflation, the fiscal balance, and the current account balance. Cointegration is
found between each macroeconomic variable and the demographic variables, which is a
JEL classifications:
key finding of the analysis. The main empirical findings from the long-run cointegrating
J11
C32
equations are that the effects of demographic change on the macroeconomic variables are
statistically significant and quite strong. Alternative variants of the United Nation’s
Keywords: population projections provide further evidence of the importance of the demographic
Population aging changes for Japan’s macroeconomic future. This study finds that future trends of key
Augmented ARDL macroeconomic variables are not monotonic, but rather that long swings in the
Japan demographic factors produce a mixture of moderate growth periods and episodes of
GDP stagnation.
© 2020 Elsevier Inc. All rights reserved.
1. Introduction
Among the advanced economies, Japan stands out as having the most rapidly aging population and being the oldest
country by age (see The Economist, 2014). According to 2016 estimates, 26.7 % of the Japanese population are above the age of
65 (Ministry of Internal Affairs & Communication, Statistics Bureau, 2018). It is expected that this age group will reach 40 % of
the total population by 2050, the highest proportion in the world (Wan, Daniel, & Paul, 2016). As populations age, the
demographic dividend dissipates. Demographic changes through population aging could have substantial macroeconomic
implications (Bloom et al., 2015). For instance, when the proportion of the working-age population declines as the
population ages, fewer workers are available to support an increasing number of old-age dependents. Transfers from
workers to the elderly add to the tax burden on the working age population, discouraging investment in human capital and
lowering productivity growth. In addition, the aging population might result in a drawdown in savings, reducing the loanable
funds needed to finance real investment in plant, equipment, and research & development, again impairing productivity
$
This work was supported by research grant from Sumitomo Foundation, Japan and the Malaysia Higher Education, FRGS,203/PSOSIAL/6711781
* Corresponding author.
E-mail address: sookhoongoh@gmail.com (S.K. Goh).
http://dx.doi.org/10.1016/j.asieco.2020.101198
1049-0078/© 2020 Elsevier Inc. All rights reserved.
2 S.K. Goh et al. / Journal of Asian Economics 68 (2020) 101198
growth. The shrinking workforce could increase or decrease domestic investment, depending on whether capital and labour
are substitutes or complements.
The changes in saving and investment resulting from these demographic changes have direct consequences for a
country’s current account balance. If population aging leads to a decline in national saving relative to domestic investment,
then the current account balance will decline. Moreover, an increase in the dependency rate (i.e. the young and elderly
populations relative to the working-age population) will tend to increase the transfer of income from the working age
population to the non-working age population through government spending on pensions, medical care, welfare, and
educational expenditures, while tax revenues from the shrinking working age population are declining. Hence, population
aging might also worsen the fiscal balance while simultaneously increasing the current account deficit, resulting in a long
run twin deficits problems. In this scenario demographic changes work through political and societal channels to affect the
macroeconomy.
The existing literature includes panel data, general equilibrium models or survey data studies to examine the impacts of
population age structure on saving, domestic investment, and current account balances (see Weil, 1994; Bosworth &
Chodorow-Reich, 2007; Masson, Bayoumi, & Samiei, 1998; Edwards, 1995 and Higgins, 1998; Kim & Lee, 2007, 2008 and
additional articles cited in the literature review). There are several time-series studies using cointegration techniques to
examining the effects of aging on saving in China and Taiwan (Athukorala & Tsay, 2003; Modigliani & Cao, 2004). The present
paper presents a more comprehensive time series (or cointegration) analysis to examine the effects of an aging population on
multiple macroeconomic variables: saving, domestic investment, current account balance, fiscal balance, and hence the twin
deficits phenomenon.
The objective of this study is to examine the non-stationary characteristics of key demographic ratios and
macroeconomic variables, and to test for cointegration between demographics and saving, investment, real GDP, inflation,
and the fiscal balance of Japan. Demographic changes are inherently slow moving processes, resulting in long run
macroeconomic responses suitable for cointegration modelling. Cointegration tests are applied within a set of dynamic
autoregressive distributed lag equations that embody both long run equilibrium relations and short run dynamic
adjustments. This set of cointegrated equations is then solved to shed light on whether the historical path of current account
and fiscal account balances and other key macroeconomic variables can be explained by demographic changes. In addition,
based on age-specific population projections for Japan from the United Nations (UN), the model is solved to trace the future
paths of these key macroeconomic variables under alternative demographic scenarios. We are not aware of other studies that
have applied this methodology to examine the impact of demographic changes on saving, investment, real GDP, inflation, the
current account balance and the fiscal balance in the long run.
In the light of this development, the important innovations and contributions of this study to the existing literature are as
follows:
1 We examine the effects of aging on specific macroeconomic variables that characterize important aspects of the long run
health of Japan’s economy: the saving rate, the investment rate, the fiscal balance, and the current account balance. We
study the channels by which aging affects these key macroeconomic variables, which are not revealed when a single
macroeconomic variable, such as real GDP growth, is the object of analysis.
2 We examine the effects of different measures of aging on the macro variables, rather than representing aging by a single
measure such as the old-age dependency ratio. Our analysis employs three separate measures of aging – young, working
age, and old, each relative to the total population. The effect of changes in each ratio is expected to be targeted on different
macroeconomic factors. For example, the growth in the elderly population ratio is expected to most strongly affect the
government’s fiscal balance, while changes in the working-age population ratio affects more directly the level of real GDP
through productivity effects.
3 Before estimating the equations that relate the macroeconomic variables to the demographic ratios, it is necessary to
assess the stochastic properties of the individual variables. In particular, it is vital to determine the non-stationarity
characteristics of the variables – do they need to be differenced one or two times to become stationary? No sensible
functional relation can exist between variables with different orders of integration; for example, it is difficult to conceive
of a functional relation between one variable that persistently trends away from its initial values and another that
fluctuates around a constant mean. Pre-testing the variables for unit roots establishes that the demographic ratios are all
integrated of order two, so that their first differences are integrated of order one, matching the order of integration of the
macroeconomic variables. The requirement of identical orders of integration among the variables means that the long run
relations must contain the demographic ratios in first differenced form, while the macro variables are included without
differencing. Exisiting studies of these relations do not recognize these differing orders of integration. Therefore they
present relations that are unbalanced and consequently nonsensical because the variables will diverge by ever increasing
amounts (Banerjee, Dolado, Galbraith, & Hendry, 1993, p. 167).
4 Tests for cointegration establish the existence of long run equilibrium relations between each macroeconomic variable
and underlying demographic determinants. In cointegration analysis the emphasis is on long run relations, which is
appropriate for studying the impacts of slowly evolving demographic changes. The finding of cointegration is strong
confirmation of the theory that underpins the specification of each equation.
5 The estimated equations provide evidence that the effects of demographic changes on macroeconomic variables are
statistically significant and powerful. Furthermore, variations in the temporal patterns of the demographic factors will
S.K. Goh et al. / Journal of Asian Economics 68 (2020) 101198 3
produce macroeconomic outcomes that alternate between periods of economic weakness and episodes of more favorable
macroeconomic conditions. Population aging does not lead to the often cited dismal monotonic projections for the key
macroeconomic variables.
6 The empirical model is used to simulate future paths of the macroeconomy, given assumed trajectories of the
demographic variables. Since there is some uncertainty over Japan’s demographic future, we consider three alternative
fertility scenarios. A comparison across these scenarios indicates the sensitivity of the macroeconomic factors to
alternative fertility assumptions.
To preview the main findings, the three age shares of the total population are found to be I(2) variables, implying their
first differences are integrated of order one. Since the macroeconomic variables are found to be I(1), regression balance
requires the macro variables in their undifferenced form be related to the first differences of the demographic ratios.
Cointegration is found between each macroeconomic variable and theoretically appropriate measures of demographic
change; the existence of these long run cointegrating relations is a key finding of the analysis. The demographic variables
enter into the long run relations with statistically significant coefficients that also imply substantial responses of the
macroeconomic variables to representative changes in the growth of population age shares. Based on the system of
estimated equations, the baseline demographic and macroeconomic futures projected for Japan do not follow simple
monotonic trends. Rather, we observe long waves in macroeconomic variables that are traceable to the changing
demographics, as the historical decline in fertility rates has varying effects over time on the future age distribution of the
population. Furthermore, simulations of alternative future demographic scenarios show substantially differential effects for
the key macroeconomic variables, especially the fiscal and current account balances. For example, the simulations indicate
that the twin deficits phenomenon is most likely under the demographic scenario defined by the low fertility variant.
Since Japan is unique in several ways, the findings from this study may not be generalisable to other countries. This study
is part of a larger research agenda that applies the same time series methodology to other Asian countries, which will
indicate how demographic change plays out differently in other economies. In addition, because demographic variables may
respond to expectations of future macroeconomic events, the estimated relations should be cautiously interpreted as causal.
This paper is organised as follows. Section 2 reviews the existing literature in order to set the present study in context.
Section 3 discusses the main theoretical arguments behind the equations relating demographic forces to the key
macroeconomic variables. Section 4 outlines the augmented autoregressive distributed lag (ARDL) framework for
cointegration analysis, and the data used in this study. Section 5 reports the results of unit root tests, augmented ARDL
estimation results, the in-sample simulations used to validate the system of dynamic equations, and the out-of-sample
projections under three alternative population projections. A discussion of implications and concluding comments appear in
Section 6.
2. Literature review
Most existing studies used panel data, general equilibrium models, or survey data to examine the impacts of population
age structure on saving, domestic investment and the current account. The bulk of the studies that employed
macroeconomic panel data found a strong negative link between dependency rates and saving rates and domestic
investment, as well as current account balances (Bosworth & Chodorow-Reich, 2007; Edwards, 1995; Masson et al., 1998;
Weil, 1994 and Higgins, 1998). For example, utilizing panel data of more than 100 countries, Higgins (1998) found that the
young and the elderly populations had strong negative effects on saving, domestic investment, and the current account
balance. The demographic effects on the macroeconomic variables were particularly strong for open economies.
Chinn and Prasad (2003) found that a younger population had a significant negative effect on the current account, and
Luehrmann (2003) showed that countries with high young and old population proportions tend to import more capital, and
hence, have a higher deficit in their current accounts as well. Using a panel VAR model, Kim and Lee (2007) found that an
increase in both youth and old populations significantly lowered saving rates and diminished the current account balance in
G7 countries. Using a similar method, Kim and Lee (2008) found an increase in the elderly dependency rate could
significantly lower saving rates, and in turn, lower current account balances in East Asian countries. Based on this empirical
evidence, the authors conjectured that aging populations would cause East Asia to have lower saving rates, leading to a
reduction in their current account surpluses, which would thereby reduce global current account imbalances. Since Asian
countries are all expected to exhibit increasing dependency ratios in the next 10 years, these four studies seem to contradict
Higgins’ analysis which projects an increase in current account balances in the face of rising dependency rates. One goal of
the current study is to resolve this apparent contradiction.
Studies based on household survey data did not generally find a negative link between age and saving rates, contradicting
the life-cycle hypothesis (LCH). For example, Mirer (1979) examined the saving behaviour of aged people in the U.S. and
found that the wealth of the elderly population rarely declined. In a similar study, Danziger, Van Der Gaag, Smolensky, and
Taussig (1982) concluded that elderly people had the lowest average propensity to consume, and hence, spent less than the
non-elderly at the same level of income. Carrol (1997) offered several explanations for this observed puzzling saving
behaviour of elderly people. First, the uncertainty in the timing of retirement and death could exert a major influence on the
pace of dissaving among the elderly. Second, the desire to leave bequests to children would make the elderly continue to save
rather than dissave to finance their own consumption. Carroll’s explanation was supported by Bernheim, Skinner, and
Weinberg (2001) and Banks, Richard, and Sarah (1998). Both studies found that retirees were prone to decrease consumption
expenditures substantially, which supports the precautionary or bequest motives for saving.
Fougere and M’erette (1998) used an open economy overlapping generations (OLG) model to test the LCH for selected
OECD countries. Their findings did not corroborate the implication of the LCH. However, the authors argued that
countries with aging populations are likely to have excess saving because the decline in domestic investment is
greater than the reduction in national savings. This behaviour results in a current account surplus, so that aging
countries export excess savings to countries with relatively young populations. However, other researchers such as
Jappelli and Modigliani (1998) were sceptical about the use of household survey data, which disregarded active
workers’ contributions to their pension funds, and also inappropriately included annuity income of retired people.
They argued that the contributions to pension funds should be considered as saving, while the annuity benefit should
not be counted as retirement income. Moreover, household survey data are biased especially since they covered
mostly richer elderly who did not live with their family members. Therefore, the elderly saving rate could be biased
upwards as their income and saving were directly correlated. The evidence from survey data studies leads to no clear
conclusion on the effects of an aging polulation on saving rates, and our estimated saving equation will inform this
debate using a different methodology.
Based on our literature search, we found no studies using cointegration analysis to determine the effects of an aging
population on multiple key macroeconomic variables such as saving, domestic investment, current account balance, fiscal
balance, and hence the twin deficits phenomenon. Nevertheless, there were several time-series studies with partial analysis
examining the effects of aging on saving. For example, based on a simple cointegration test, Modigliani and Cao (2004) found
that the major systematic determinants of the private saving rate in China were income growth and the demographic
structure of society. Using an ARDL cointegration test, Athukorala and Tsay (2003) found both old and young dependency
ratios have negative impacts on the saving rate in Taiwan. Horioka (1992a) used a simple time series regression (not
cointegration analysis) to examine the determinants of saving in Japan for the period 1956 – 1987. The study found that age
structure (i.e. the ratio of old age to the working-age population) is an important determinant of saving in Japan, in addition
to the level and rate of growth of income, the unemployment rate, and the inflation rate. Horioka (1992b) further projected
the future trends in Japan’s saving rate and its impact on future current account surplus. The projection of future saving was
based on a survey of several studies on the trends in saving over time. As a result of the rapid aging of the population, it was
predicted that saving would decline sharply in the following years, and Japan’s current account surplus would also decline,
albeit less than the rate of saving.
Besides time series analysis, there were some studies using general equilibrium analysis to examine the impact of
population aging on saving in selected countries. For example, Curtis, Lugauer, and Mark (2017) found that increases in the
saving rates in China and India were affected by changes in the child dependency ratio. On the other hand, the study also
found that the decrease in Japan’s saving rate since the mid-1970s was partially attributable to the rising proportion of the
retirement-aged population. Similar findings for Japan can be found in Braun, Ikeda, and Joines (2009) using a computable
dynamic OLG model. Based on projections, both studies suggested that the Japanese saving rate would be much lower in
the future with aging population, even under optimistic assumptions about total factor productivity growth and
demographics.
There is a growing body of empirical studies examining the effect of population age structure on inflation using panel data
analysis (see Lindh & Malmberg, 2000; Gajewski, 2014; Juselius & Takáts, 2015; Yoon, Kim, & Lee, 2014), general equilibrium
modeling (see Konishi & Ueda, 2013; Anderson et al., 2014), and overlapping generations models (see Konishi & Ueda, 2013).
In addition, Bobeica, Lis, Nickel, and Sun (2017) employed cointegration analysis to study the relationship between
demographic change and inflation for the Euro Area from 19752016. In general, the empirical evidence of the impact of
population aging on inflation is mixed. Some empirical studies found that aging is deflationary (see Konishi & Ueda, 2013;
Anderson et al., 2014; Yoon et al., 2014; Gajewski, 2014). In contrast, empirical studies by Juselius and Takáts (2015) showed
that aging could lead to higher inflation. By the same token, the study by Bobeica et al. (2017) supported the proposition that
there exists a positive relationship between Euro Area core inflation and the growth rate of the working-age population.
Given this mixed evidence from the literature on the effects of demographic change on inflation, this is a question that is
addressed empirically in the present study.
?A3B2 twb=.35w?>The relation between population aging and economic growth has also been addressed in a large
number of studies, but again with ambiguous results. The conventional view is that aging is accompanied by lower
labour force participation rates, reduced rates of saving and investment, and dampened productivity growth that could
impede economic growth (Bloom, Canning, & Fink, 2010). Empirical studies that support this proposition can be found
in Bloom et al. (2010) and Maestas, Mullen, and Powell (2016). However, Curtis and Lugauer (2019) argued that there
could exist a positive association between population aging and output per person, if production is sufficiently capital
intensive, which their cross-country regression evidence confirms. Other empirical studies that also support this
proposition are Futagami and Nakajima (2001), Lee, Mason, and Park (2011), and Prettner (2012). Since this current
study does not focus solely on the relation between population aging and economic growth, we do not enter into this
debate. Rather, the goal of this study is to consider the implications of various measures of aging on several key
macroeconomic variables, which in their turn will have futher implications for the long run economic health of the
Japanese economy – of which GDP growth is only one aspect.
3. Theoretical framework
This section presents the theoretical and empirical bases of how changes in demographic variables could affect selected
macroeconomic variables, i.e., saving, investment, fiscal balance, inflation, and GDP growth. Other factors may be included as
control variables in each macroeconomic equation. The empirical specifications begin with a dynamic ARDL representation
of the underlying theoretical equations as presented here. These ARDL equations are simplified through the general-to-
specific modelling approach, leading to much simpler empirical equations than suggested by the theory.
The relationship between household saving and the demographic variables is explained by the Life Cycle Hypothesis
(LCH). Individuals who are in their working age save to accumulate wealth to support their consumption expenditures in
retirement, while those who are retired and under the age of 15 decrease their saving or even dis-save because of low or
possibly zero income as children or after retirement. Hence, aggregate saving rates are expected to be negatively related to
the young age population ratio (YOUNG) and old age population ratio (OLD), both expressed relative to the total population
(Kwack & Lee, 2005).
However, as observed by Bosworth and Chodorow-Reich (2007), the retired tend to increase saving, which is contrary to
the LCH. Using a 1983 survey conducted by the Federal Reserve Board, Carrol (1997) highlighted that a higher propensity to
save is due to the precautionary saving motive to accommodate income shocks, uncertainty about life expectancy,
emergencies, and inheritance for children. From this perspective, saving may alternatively be a positive function of OLD. The
empirical analysis that follows provides evidence one way or the other.
þ= þ þ= þ= þ=
Saving ¼ f ðYOUNG; OLD; RGDP ; R ; INF; UNÞ ð1Þ
1 2
where YOUNG is the ratio of the youth population to the total population, and OLD is the ratio of the elderly population to
the total population. We also include a set of control variables including real income (RGDP), rate of interest (R),
unemployment (UN), and inflation (INF). RGDP is believed to influence saving based on the Keynesian absolute income
hypothesis. The interest rate effect depends on the relative strength of income and substitution effects. For example, an
increase in the interest rate (R) would tend to encourage people to substitute saving for consumption due to higher rewards
for saving. However, higher R might affect saving adversely owing to a strong income effect; i.e. savers would recognize that
an increase in interest income payments allows them to increase future consumption while saving less currently. Empirical
evidence of a negative relationship between the saving ratio and the rate of interest can be found in Ouliaris (1981). In
addition, Athukorala and Tsay (2003) argued that both UN and INF could have a positive as well as a negative impact on
saving. For instance, when UN (or INF) increases, job security (or future asset prices) would become vulnerable (or uncertain),
leading to an increase in precautionary saving for a rainy day (or higher saving in order to maintain future consumtion levels).
On the other hand, an increase in UN (or INF) could result in temporary dissaving for consumption smoothing by the
unemployed (or could cause savers to invest savings in a higher-return financial asset in order preserve the purchasing
power of savings).
Apart from saving, the population’s age structure could influence real investment demand. For instance, a country with a
high working age population3 would encourage domestic investment if capital and labour are complements in production. In
this case, an aging population, accompanied by a relative decline in the workforce, would lead to a decrease in the demand
for new investment. On the other hand, capital and labour may be substitutes in production, so that a shrinking labour supply
is compensated for by increased investment. The demand for domestic investment is therefore either a positive or a negative
function of working age population as shown in Eq. (2).
!
þ= þ þ
Investment ¼ f WORKING; R ; RGDP; K K1 ð2Þ
where WORKING is the ratio of working-age population to total population. Similarly, R, RGDP, and the accelerator effect (AE)
are the control variables that could potentially influence investment. For instance, an increase in R could discourage
investment spending. In the case of an increase in RGDP, firms would require more capital to produce new goods and services,
which is expected to lead to an increase in investment. Moreover, the level of investment is dependent on the AE, which is
contingent on the gap between the last period’s capital stock (K-1) and the desired capital (K*). The firm’s rate of investment
increases if the gap K* - K-1 increases.
As the process of population aging continues, governments tend to spend more on health and public pensions, putting
more pressure on public budgets. Likewise, an increase in YOUNG is expected to lead to an increase in fiscal spending on
education. Hence, the fiscal account balance is a negative function of YOUNG and OLD.
þ þ þ=
Fiscal Balance ¼ f ðYOUNG; OLD; RGDP; TBR; INF Þ ð3Þ
1
The youth population is defined as those people aged less than15 (see https://data.oecd.org/pop/young-population.htm#indicator-chart).
2
The elderly population is defined as people aged 65 and over (see https://data.oecd.org/pop/elderly-population.htm#indicator-chart).
3
According to the OECD Labour Force Statistics, the working age population is defined as those aged between 15 and 64 (see https://data.oecd.org/pop/
working-age-population.htm).
where RGDP, treasury bill rate (TBR), and inflation (INF) are some other variables that could affect the fiscal balance. For
instance, an increase in RGDP is expected to generate a higher level of income tax revenue that could result in an increase in
the fiscal balance, ceteris paribus. On the other hand, an increase in TBR could lead to higher interest payments on previously
borrowed funds, thereby increasing government expenditure and reducing the fiscal balance. The variable INF could have
price effects on government expenditures (e.g. social security payments and wages) as well as progressive income taxes (see
Tujula & Wolswijk, 2004). To be more precise, social security recipients, civil servants, and income taxpayers would,
respectively, suffer lower real social benefits, lower real wages, and higher income tax brackets during times of inflation
without price indexation of social benefits. Moreover, high inflation also erodes the real value of nominal government debt,
lowering the cost of financing the debt and increasing the fiscal balance. Thus, the overall effect of inflation on the fiscal
balance is not certain.
Given the relations for saving, investment, and fiscal balance, the current account balance (CAB) is determined by the
identity:
CAB = Saving – Investment + Fiscal Balance (4)
The demographic factors and other macroeconomic variables affect the CAB indirectly, working through the relations for
saving, investment, and fiscal balance. Therefore, the model is a quasi-structural model, rather than a simple reduced form of
CAB. This quasi-structural approach is much more informative than a reduced form, single equation model of the CAB since it
shows the channels by which demographic change affects the CAB, both historically and into the future.
In addition to the equations presented above, the projections for saving, domestic investment, fiscal balance, and CAB into
the future require the formulation of dynamic equations for macroeconomic variables such as real GDP and inflation that
appear as explanatory variables in the saving, investment, and fiscal balance equations. These additional macroeconomic
variables are also influenced by the same demographic variables previously defined. For example, theoretically an increase in
the working age population has a positive impact on the growth of real GDP given
þ þ
Y ¼ F L; K ð5Þ
where L and K are represented by WORKING and the acquisition of productive physical capital, respectively. An increase in
WORKING or physical capital could lead to higher RGDP.
With regard to inflation, an increase in the share of dependants is expected to put upward pressure on prices because
more people are consuming goods and services than are producing them (see Lindh & Malmberg, 2000; Bobeica et al., 2017).
We postsulate both the young and the elderly shares are postively related to inflation. Nevertheless, there is a growing body
of theoretical and empirical studies arguing that aging causes deflationary outcomes. For instance, based on constrained
competitive equilibrium, the aging population, who are less productive and prefer a higher return on savings, tend to
contribute to lower inflation (Bullard, Garriga, & Waller, 2012). Konishi and Ueda (2013) pointed out that aging could be
associated with deflation when there is an unanticipated increase in life expectancy. With these diverse theoretical and
empirical findings, the effect of an increase in OLD on inflation is uncertain. Hence, the equation for Inflation is written as
!
þ þ= þ þ
Inf lation ¼ f YOUNG; OLD; RGDP; R ð6Þ
where we include as well RGDP and R. RGDP and R are expected to influence inflation positively and negatively, respectively,
because the former represents a higher level of aggregate demand, while the latter is likely to discourage aggregate demand
through reduced investment activity.
4. Methodology and data
The empirical specification for the system of equations described in section 3 is grounded in the cointegration framework,
emphasizing the long run nature of the relations between demographic changes and macroeconomic responses. In
particular, the model is specified as a system of autoregressive distributed lag (ARDL) equations that simultaneously provide
tests for cointegration and estimates of the short run and long run dynamic relations among the variables. The empirical
equations are dynamic representations of the theoretical equations in ARDL form illustrated in Eq. (7) below. The final
empirical equations are the result of general-to-specific specification searches, eliminating variables or lagged terms with
statistically insignificant coefficients.4 The final specifications are accepted into the empirical model only after passing
standard diagnostic tests.
The specific cointegration testing methodology is the augmented ARDL bounds test, which is an extension by McNown,
Sam, and Goh (2018) and Sam, McNown, and Goh (2018) of the framework presented by Pesaran, Shin, and Smith (2001).
Pesaran et al. (2001) introduced two tests in their ARDL bounds testing framework: the overall F-test on all the coefficients
4
The general-to-specific approach is meant to result in parsimonious specification. The use of this method explains why some of the estimated equations
include only a small number of explanatory variables since some variables have been eliminated through the search procedure, as supported by the data.
on the lagged level variables and a t-test on the coefficient on the lagged level of the dependent variable. However, these tests
require the assumption of an I(1) dependent variable to rule out the possibilities of degenerate cases and to draw a valid
conclusion. Degenerate cases imply non-cointegration. Instead of assuming the dependent variable to be I(1), McNown et al.
(2018) introduced an additional test on the significance of the coefficients on the lagged levels of independent variable(s).
The advantage of the additional test is to overcome the reliance on the assumption of an I(1) dependent variable to rule out
the degenerate case. This reduces the risk of relying on standard unit root tests that have low power.
The augmented ARDL equation with an unrestricted error correction mechanism for three series is specified as follows:
X
p1 0 X
q1 0 X
r1 X
q
^y ^ Dy þ d^ Dx
0 0
Dyt ¼ ^c þ f ^ x1;t1 þ ’
t1 þ g ^ x2;t1 þ li ti j 1;tj þ ^ k Dx2:tk þ
p vl Dt;l þ êt ð7Þ
i¼1 j¼1 k¼1 l¼1
where i, j, k, and l are indices of lags i ¼ 1; 2; :::; p, j ¼ 0; 1; 2; :::; q, k ¼ 0; 1; 2; :::; r, s ¼ 1; 2; :::; l; t denotes the time periods
t ¼ 1; 2; :::; T; yt is the dependent variable; x1t and x2t are independent variables; Dt;k is a dummy variable to deal with
possible structural breaks in the equation. The dummies are specified as one in the identified year and zero otherwise. In an
ARDL equation this type of dummy variable allows a permanent shift in the relation.
The null hypotheses for all three tests are defined as follows:
^ =g
(i) F-test on the coefficients on the lagged level of all variables: H0 : f ^ =’ ^, g
^ = 0, against H1 : f ^, ’
^ 6¼ 0
^ ¼ 0 against H : f
(ii) t-test on the coefficient on the lagged level of the dependent variable: H0 : f ^ <0
1
(iii) F-test on lthe coefficients on the lagged levels of the independent variables: H0 : g^¼’ ^ ¼ 0 against H1 : g^; ’
^ 6¼ 0
McNown et al. (2018) proposed the third test, the F-test on the lagged level of the independent variables in addition to the
first and second tests proposed by Pesaran et al. By combining this new test with the other two, we gain a complete picture of
the cointegration status of the system.
^ =g^ =’ ^ ¼ 0 are rejected but hypothesis (iii) H : g
^ = 0 and (ii) H0 : f
If the first two null hypotheses (i) H0 : f 0 ^ ¼ ’
^ ¼ 0 is not,
a case of degenerate lagged independent variables arises. This is because the deletion of the lagged level of the independent
variables reduces the equation to a generalized Dickey-Fuller equation. The statistical significance of the overall F-statistic is
due solely to the significance of the coefficient on the lagged level of the dependent variable, indicating that the dependent
variable is I(0). On the other hand, if the null hypotheses from (i) and (iii) are rejected but (ii) is not, then this falls into the
degenerate lagged dependent variable case. Only if all the three null hypotheses are rejected, can one conclude there is
cointegration.
Pesaran et al. (2001) provided bounds on critical values for the overall F-test and the t-test for the dependent variable
coefficient. In introducing the test on the coefficients on the lagged levels of the independent variable(s), McNown et al.
(2018) adopted a bootstrap procedure to generate the critical values for this additional test. Since the bootstrap
implementation involves programming and computation that researchers may wish to avoid, Sam et al. (2018) derived the
limiting distributions for this test statistic, and generated critical value bounds for both asymptotic and small sample cases.
In this study the first two test statistics are compared with the critical value bounds for the overall F-test and the t-test for the
dependent variable coefficient from Pesaran et al. (2001, p. 300 & 303, Table CI (iii) & Table CII(iii)), while the critical value
bounds for the test on the coefficients on the lagged levels of all independent variables are obtained from Sam et al. (2018), p.
9, Table 3.
If cointegration is established, the long-run effects can be extracted from the unrestricted ECM. Refering to Eq. (7), if
Dyt = 0, then

Dx1t =Dx2t= 0 when the system is in equilibrium, and the long-run coefficients for x1 and x2 are equal to
   g ^    and ’
^ =f ^ respectively. Since the standard errors in the unrestricted ECM cannot easily be inferred for the
^ =f
renormalized long-run relation, the restricted ECM is estimated instead.
X
p1 0 X
q1 0 X
r1 X
q
^ ðy ^ Dy þ d^j Dx1tj þ
0 0
Dyt ¼ ^c þ f ^ ^ x2t1 Þ þ
t1 þ s x1t1 þ k l^ı ti
^ k Dx2tk þ
p vl Dt;l þ êt
i¼1 j¼1 k¼1 l¼1
ð8Þ
X
p1 0 X
q1 0 X
r 1 X
q
^z ^ Dy þ d^j Dx1tj þ
0
vl Dt;l þ êt
0
¼ ^c þ f t1 þ li ti
^ k Dx2tk þ
p
i¼1 j¼1 k¼1 l¼1
^ x1t1 þ k
where zt-1 = ðyt1 þ s ^ x2t1 Þ.
Once each estimated equation is found to be cointegrated, the system of ARDL equations can be used for in-sample
simulation, conditional forecasting, and scenario analysis. The estimated ARDL equations are ideal for simulating long run
population and macroeconomic trends because “their values are linked over the long run, and imposing this information can
produce substantial improvements in forecasts over long horizons” (Stock, 1995, p. 1). Moreover, the ARDL system is used to
analyse future macroeconomic trends across alternative demographic scenarios based on the United Nations high, medium,
and low fertility assumptions for Japan. In particular, future long run patterns of private domestic saving, domestic
investment, the fiscal account balance, and the current account are studied under these three alternative demographic
scenarios.
Table 1
Augmented Dickey Fuller and Phillips-Perron Unit Root Tests.
ADF test PP test
Level 1st 2nd Level 1st 2nd

Differenced Differenced Differenced Differenced
Total Young to total population (YOUNG) 1.97 1.95 (c,1) 4.16*** (c,2) 1.92 (t) 2.15 (c) 6.26 *** (c)
(t,4)
Total Working to total population (WORKING) 0.52 (c,4) 1.53 (c,2) 5.60*** (c,0) 0.17 (c) 1.75 (c) 5.95*** (c)
Total Old to total population (OLD) 0.39 0.50 (c,1) 3.20** (c,2) 0.65 (t) 0.52 (c) 6.31*** (c)
(t,5)
Gross Domestic Saving to Gross Domestic Product (GDS) 2.04 2.80* (c,1)
(t,0)
Gross Fixed Capital Formation to Gross Domestic Product 3.05 5.50*** (c,1) 3.05 (t) 5.47*** (c)
(GFC) (t,1)
Cash Surplus Deficit to Gross Domestic Product (CSD) 2.18 (c,1) 4.50*** (c,1) 2.18 (c) 6.09*** (c)
Real Gross Domestic Product (RGDP) 2.74 4.00*** (c,0) 2.79 (t) 3.89*** (c)
(t,0)
Inflation (INF) 1.71 8.16*** (c,0) 2.55 9.68*** (c)
(c,2) (c)
Real interest rate (R) 1.34 11.27*** 2.61 5.61*** (c)
(c,3) (c,0) (c)
Notes: *, **, *** indicate statistical significance at 10 %, 5% and 1% levels, respectively.
All of the time-series data are annual with the sample period from 1960–2015, excepting the real rate of interest that
begins in 1970. Most of data are taken from the World Bank Country Data.5 The data on the variables of interest are the real
interest rate6 (%) (R), real Gross Domestic Product, Gross Domestic Saving (% of GDP), Gross Fixed Capital Formation (% of
GDP), inflation (annual %), population aged 65 & above (% of total population), population aged 15–64 (% of total population),
population aged 0–14 (% of total population).7 The theoretical discussion emphasizes that alternative aspects of an aging
population should have different impacts the macroeconomic variables. For example, an increase in the proportion of the
population that is elderly has strong effects on a country’s fiscal balance, while real investment activity is most directly
affected by changes in the working age population proportion. The differential effects of these demographic changes cannot
be proxied by a single measure of aging, such as the old age dependency ratio that is commonly used in other studies.
To allow a longer time span of some series such as the fiscal balance, we also refer to Statistitics Department of Japan.
Population projection data is taken from the Population Division of the Department of Economic and Social Affairs of the
United Nations which publishes five-year-interval population projections for most countries, including Japan. These
projections were interpolated into single-year data. In addition, real GDP is expressed in logarithmic terms.
5. Results
5.1. Unit root tests
Before running the Augmented ARDL test, it is necessary to ensure all variables are integrated with order not greater than
one. Both the Augmented Dickey-Fuller (ADF) and Phillips Perron (PP) tests are employed to test and identify the degree of
integration, I(d), for all series. Table 1 presents the results for the two unit-root tests, which include intercept and trend
components (when there is a visible trend in the plot of the data) in the test equation for testing the on level of the series;
only the intercept is included when testing the first differenced series. The lag length determination for the ADF test is based
on the Modified Akaike Information Criterion, while the PP test uses the Bartlett kernel for its spectral estimation and the
Newey-West method for its bandwidth determination. Table 1 indicates that all macroeconomic series contain a single unit
root, I(1), while the demographic variables (i.e. YOUNG, WORKING, OLD), become stationary only after second differencing,
indicating they are I(2). Since I(2) variables cannot cointegrate with I(1) series, the demographic variables must be
differenced to match the I(1) macroeconomic variables.
These unit root tests provide the first substantive finding of this study. In the long run relations, macroeconomic variables
can only be related to changes in the population age shares. Since this formulation of the long run relations is unprecedented
in this literature, it warrants further consideration. The statistical basis is clear: variables of different orders of integration
5
The World Bank database is the primary souce of data for this study. This is because this study is part of a wider research project involving application of
the same methodology to other countries experiencing different degrees of demographic aging. Using the same source of data allows comparison of various
results from these different countries.
6
The real interest rate is the lending interest rate adjusted for inflation as measured by the GDP deflator.
7
These definitions of the demographic variables provide flexibility in examining the impact of demographic factors on different macro variables. Kim and
Lee (2007, 2008) also used these definitions for their demographic variables.
Fig. 1. Saving rate and elderly population age share, 1970 – 2016.
cannot form a long run equilibrium relation, and a graphical presentation of representative demographic and
macroeconomic variables is convincing on this point. Fig. 1 shows the steady upward trend in the percentage of the
population that is 65 years of age or older (OLD), as well as the overall tendency of a declining saving rate. A simple regression
of the saving rate on the elderly population share will certainly pick up a significant negative relation,8 but the smooth trend line
of OLD cannot capture the temporal variation in the saving rate over time.
The first difference of the elderly age share, however, not only accounts for the overall upward trend in the saving rate, it
also allows accelerations and decelerations in the growth of this age share (DOLD) to account for fluctuations in the saving
rate. The rise in DOLD in the 1970s is accompanied by a decline in the saving rate, followed by a leveling of these two series
during the first half of the 1980s. DOLD begins its most dramatic rise in the mid-1980s, and with a lag the saving rate resumes
its downward trajectory. This inverse relation holds until 2010, when there is a leveling of both DOLD and the saving rate. Of
course, the saving rate is not fully determined by changes in the elderly population age share, and the more complete relation
is represented by the ARDL equation for gross private domestic saving presented below.
The strong statistical evidence that the demographic ratios have unit roots overrides the theoretical argument that ratios
cannot have unit roots since they are bounded between zero and one. Since the unit root test statistics on the demographic
ratios are never close to their critical values, these series clearly behave like unit root processes. Any statistical modeling
involving these variables will result in estimators and test statistics with distributions that are better approximated under
the assumption that the series have unit roots, as opposed to the less realistic assumption of stationarity (see Banerjee et al.,
1993. pp. 95–98).
5.2. Augmented ARDL estimation results
Table 2 reports the unrestricted ECMs, the long run cointegration relations, and diagnostic tests of Eqs. (1) to (5) using the
augmented ARDL tests. Dummy variables are added to these equations to capture global economic shocks, such as oil price
shocks (DU73, DU74, and DU80), the Asian Financial Crisis (DU98), and the sub-prime loan crisis (DU09). These indicator
variables are assumed to capture exogenous shocks to Japan’s macroeconomy, shocks that originated globally. Each dummy
takes on the values of one (which indicates the presence of the event) or zero (indicating the absence of the event). In an
ARDL equation a one-off dummy of this form creates a permanent shift in the relation. The numbers of lagged differences of
each variable included in the ARDL equations are determined by a search procedure. To avoid over parameterization, a
maximum lag length of 4 is imposed, and the selection of optimum lag length is based on the Akaike’s Information Criterion
(AIC). We began with a complete specification of each equation as expressed in the theoretical section, and then pared them
down using the general-to-specific modeling approach. All the insignificant variables were deleted to ensure the model is
parsimonious. The resulting specifications must pass various diagnostic criteria: the correlogram Q-statistic up to lag 12 for
8
The least squares estimated coefficient is -0.9 with a t-statistic of -18.6, an R-squared of 0.89, and a Durbin-Watson statistic of 0.36. This appears to be a
spurious regression.
Table 2
The augmented ARDL tests.
Equations Test statistic & diagnostic

checking
F-statistic: 12.18***
Unrestricted ECM:
t-statistic (lagged DV):
DGDSt ¼ 8:04 0:22GDSt1 6:10DOLDt1 2:90DYOUNGt1 þ 0:08DRt1 þ 0:11DRt2 þ 4:15DDYOUNGt2
ð1:80Þð0:04Þð1:36Þð0:59Þð0:03Þð0:03Þð1:45Þ -4.93***
2:62DU09 2:31DU82 þ 2:79DU15 F-statistic (lagged IDV):
ð0:73Þð0:76Þð0:75Þ 15.23***
The long-run regression: N = 45, R2 = 0.69, Q-stat
GDSt ¼ 26:75DOLDt 12:74DYOUNGt (12) = 18.28 (0.107), LM
ð2:69Þð2:63Þ (2) = 0.525 (0.769),
JB = 1.49 (0.47)
Unrestricted ECM:
DGFCt ¼ 3:33 0:14GFCt1 þ 0:96DWORKINGt1 þ 0:39DGFCt1 0:47DGFC t4 þ 13:50DLRGDPt4
ð0:85Þð0:03Þð0:40Þð0:09Þð0:10Þð4:27Þ -4.56***
þ3:09DU73 1:34DU09 F-statistic (lagged IDV):
ð0:73Þð0:68Þ 5.68*
GFC t ¼ 6:76DWORKINGt1 (12) = 13.99 (0.30), LM
ð2:66Þ (2) = 4.71 (0.10),
JB = 1.96(0.37)
Unrestricted ECM:
DCSDt ¼ 0:37 0:14CSDt1 2:32DOLDt1 0:09INF t1 1:71DU98 þ 3:87DU05 þ 3:44DU06 3:67DU09
ð0:33Þð0:05Þð0:75Þð0:02Þð0:70Þð0:70Þð0:70Þð0:72Þ - 2.91*
The long-run regression: F-statistic (lagged IDV):
CSDt ¼ 15:88DOLDt 0:65INF t1 6.20*
ð5:35Þð0:30Þ N = 54, R2 = 0.735, Q-
stat(12) = 14.82 (0.252),
LM(2) = 3.53 (0.17),
JB = 0.72 (0.69)
Unrestricted ECM:
DINF ¼ 0:03 0:50INF t1 þ 2:65DYOUNGt1 0:13DINF t1 þ 13:52DDOAPRt4 2:71DDYOUNGt3 þ 16:65DLRGDPt1
ð0:22Þð0:03Þð0:69Þð0:04Þð4:52Þð1:36Þð4:27Þ -12.55***
F-statistic (lagged IDV):
þ12:58DLRGDPt2 þ 15:63DLRGDPt4 þ 5:49DU73 þ 15:89DU74 þ 3:53DU80
ð4:49Þð4:51Þð0:88Þð0:93Þð0:84Þ 9.33***
INF ¼ 5:21DYOUNGt1 (12) = 7.75 (0.804), LM
ð1:22Þ (2) = 1.44 (0.48),
JB = 1.90 (0.38)
Unrestricted ECM:
DLRGDP ¼ 1:63 0:05LRGDPt1 þ 0:029DWORKINGt1 0:22DLRGDPt2 þ 0:06DDWORKINGt4 0:05DU09
ð0:29Þð0:01Þð0:01Þð0:11Þð0:03Þð0:02Þ - 5.51***
The long-run regression: F-statistic (lagged IDV):
LRGDPt ¼ 0:54DWORKINGt1 5.80*
ð0:29Þ N = 50, R2 = 0.67, Q-stat
(12) = 4.77 (0.96), LM
(2) = 4.76 (0.10),
JB = 3.35 (0.18)
Notes: Standard errors of estimated coefficients are reported in parentheses; *, **, *** indicate statistical significance at 10 %, 5% and 1% levels, respectively. F-
^ = g
statistics are the statistics for testing H0 : f ^ = ’ ^, g
^ = 0, against H1 : f ^, ’
^ 6¼ 0; t-statistics (lagged DV) are for tests on the lagged level of the
^ ¼ 0 against H : f ^ < 0; F-statistic (lagged IDV) is the statistic for testing on the independent variable: H : g
dependent variable: H0 : f 1 0 ^ ¼ ’^¼0
against H1 : g ^; ’
^ 6¼ 0. N is the sample size. Q-stat (12) denotes Q statistic at lag 12, LM(2) denotes the Breusch-Godfrey Lagrange Multiplier test
statistic at lag 2, JB indicates Jarque-Bera statistic for normality test. The values in parentheses indicate p values. Dummy variables D## are
defined as one at the specific year ## and zero otherwise.
white noise residuals; the Breusch-Godfrey LM test for absence of serial correlation in the residuals, denoted by LM(2); the
Breusch-Pagan-Godfrey test for homoscedastic residuals; and the Jarque-Bera test statistic (JB) for normality of the residuals.
One concern in testing for cointegration is possible low power that arises with a small to moderate sample size, here
between 45 and 54 annual observations. More relevant than the number of observations in this context is the span of time
covered by the sample, which is at least 45 years in this application. The historical period covered by these data reflect vastly
different episodes of demographic change and economic activity, and this historical variation allows the long run relations
among the variables to emerge. In any case, since the primary concern of cointegration testing is that of low power, the fact
that non-cointegration is rejected for every equation in this system implies that low power is not an issue in this study.
The first important finding reported in Table 2, is the rejection of non-cointegration for Eqs. (1) to (5). The calculated F-
test statistic on the coefficients of the lagged levels of all variables, t-test statistic on the coefficient on the lagged level of the
dependent variable, and the F-test statistic on the coefficients on the lagged levels of the independent variables reject the
null hypothesis of no cointegration, implying there is a long-run relationship between the dependent variable and the
independent variables in each equation.
For example, in the equation for gross domestic saving/GDP (S), the calculated overall F-statistic is 12.18, which exceeds
the upper bound at the 1% level of significance of [5.15, 6.36].9 Moreover, the computed t-statistic on the coefficient of the
lagged level of the dependent variable is -4.93, which in absolute value exceeds the upper bound [-3.43, -4.10] at the 1% level
of significance.10 Finally, the F-statistic on the coefficients of the lagged levels of the independent variables is 15.23, which is
larger than the I(1) upper bound of [2.41, 4.43]11 at the 1% significance level. The statistical significance of all three test
statistics indicates that there is a long-run relationship between S and the associated independent variables at least at a 1%
significance level. Similarly, there exist long-run cointegrating relationships for GFC (investment/GDP), CSD (fiscal balance/
GDP), INF (inflation), and RGDP (real GDP) with respect to their respective explanatory variables.
Table 2 shows that the demographic variables have significant effects on the macroeconomic variables. For instance, the
coefficients on the change in the old-age to total population ratio (DOLD) as well as the change in youth to total population
ratio (DYOUNG) are significantly different from zero and have negative impacts on gross domestic saving (refer to the long
run relation in panel 1 of Table 2). This evidence corroborates the life-cycle hypothesis that saving is inversely related to
changes in the youth and old-age populations, expressed as shares of the total population.
On the other hand, an increase in the change of the working age population (DWORKING) has a positive and significant
impact on gross capital formation (GFC) in Japan (refer to the estimated long run equation in panel 2 of Table 2), indicating
that capital and labour are complements in aggregate production. By the same token, DWORKING is also an important
determinant of RGDP (panel 5 of Table 2), suggesting the changing age structure of the workforce matters to the supply-side
of the economy, and in turn, economic growth. Furthermore, as the process of population aging continues, government tend
to increase spending on health and public pensions at the expense of fiscal deficits, ceteris paribus (see the estimated long
run equation in panel 3 of Table 2 where fiscal balance (CSD) is negatively influenced by DOLD). In the relationship between
demographic variables and inflation, the long-run estimated regression for INF conforms to the argument that when there is
an increase in the cohort of young dependents, their increased expenditures are seen as inflationary, which cooroborates the
findings by Konishi and Ueda (2013) as well as Anderson et al. (2014).
To understand the magnitudes of the effects of demographic changes on macroeconomic variables implied by the
long run relations, it is useful to represent typical demographic shocks by their standard deviations.12 Using the
renormalized coefficients in the long-run regression, for example, the long-run effect of a one standard deviation
change in the growth of the old age dependency ratio on gross domestic saving/GDP is (-26.75*0.19) = -5.08, implying
a one standard deviation increase in the DOLD leads to a 5 percentage point decrease in the gross saving ratio.13 This
compares with Japan’s recent saving rate of 21 percent in 2015 reported by World Bank. Likewise, a one standard deviation
increase in the growth of the youth population ratio leads to a 3.43 percentage point reduction in the saving ratio [–
12.74 0.27 = -3.43].
Similarly, the long run macroeconomic effects of representative demographic changes can be computed from the long run
coefficients of the other ARDL equations. When the DOLD is increased by one standard deviation, the fiscal account balance
ratio would decline by 3 percentage points. A one standard deviation increase in DYOUNG, increases the inflation rate by 1.4
percentage points. Changes in the working age population enter into the long run relations for both investment and real GDP.
A one standard deviation increase in DWORKING results in an increase in the rate of investment equal of 2 percentage points
and in real GDP of 16 percent.14 Demographic changes clearly have powerful long run impacts on Japan’s macroeconomy.
The ARDL estimation results also show that the dummy variables capturing structural breaks have significant coefficients
in each equation. For example, the coefficient on the dummy variable DU09, which was used to capture the sub-prime loan
crisis of 2009, is negative and significantly different from zero in the equations for fiscal account balance, saving, investment,
and GDP. In addition, the oil crisis that took place in early 1970s (refer to the estimated dummy variables DU73 and DU74 in
row 4 of Table 2) had significant and direct impacts on inflation.
The R2 in the estimated equations ranges from 0.64 to 0.93, suggesting the estimated equations individually track the
data quite well. For the complete system of equations, a dynamic within-sample simulation captures the full
interdependence of the model and can reveal whether the model solution deviates persistently from the actual
observations. In Fig. 2 the in-sample simulated values of the endogenous variables track the historical observations closely,
providing validation of the complete system. The only exogenous variables in this simulation are the demographic ratios
and the dummy variables.
Given that the estimated system of ARDL equations passes the diagnostic tests, provides strong evidence for the
existence of cointegration, and has good within sample fit (Fig. 2), it is useful for out-of-sample conditional forecasting.
9
With k=2, no intercept, and no linear trend in the cointegrating equation, the lower and upper bounds for the F-test at 10%, 5%, and 1% levels of
significance can be found in Table CI (iii) on p. 300 of Pesaran et al. (2001). Relevant bounds for each test are reported in brackets.
10
The I(0) amd I(1) bounds for the t-statistic at the 10%, 5%, and 1% levels of significance are available in Table CII (iii) on p. 303 of Pesaran et al. (2001).
11
Sam et al. (2018) provide the I(0) and I(1) bounds for the F-statistic at the 10%, 5%, and 1% levels of significance (refer to Table 2 on P.8).
12
The standarad deviations for DOLD, DYOUNG, DWORKING are 0.19, 0.27 and 0.3 respectively
13
In the unrestricted ECM, the long run coefficient for DOLD is –(-6.10/-0.22)=-27.72. In the long-run regression, the long run coefficient for DOLD is -26.75.
The difference between the coefficients implied by the two equations is due to rounding errors produced by the truncation of the extra digits after the
decimal in the reported estimates.
14
Since the dependent variable in the RGDP equation is in logarithmic form, the coefficient on the independent variable is multiplied by 100 to obtain the
approximate percentage change: (0.54*100)*0.3=16.
Fig. 2. In-sample plots of the actual and the fitted value.
The demographics are assumed to be exogenous, so that the forecasts of macroeconomic conditions are conditional on
population projections. The demographic ratios are the only exogenous variables for which projected future values are
required. Furthermore, the exogeneity assumption means that the model does not allow for feedback from the
macroeconomic variables to underlying determinants of age-specific populations: fertility rates, mortality rates, or
international migration. This is a limitation of this study that suggests an important extension for future analysis.
5.3. Out-of-sample projections: baseline scenario
The out-of-sample conditional forecasts for Japan are driven by the Population Projection 2015 from the Population
Division of the United Nations (UN). The Population Division of the UN biennially presents population estimates and
projections for all countries of the world with three core variants, i.e. high, medium, and low. The main projection is the
medium variant, which is based on a country’s most recent fertility rates, life expectancy, and migration.15 The high variant
projection assumes a fertility trajectory in which the total fertility rate is 0.5 births per woman higher than for the medium
variant (i.e. women have, on average, half-a-child more than in the medium variant). Likewise, the low variant projection
uses a total fertility trajectory with a total fertility rate that is 0.5 births per woman lower than for the medium variant. These
high and low variants are often considered to provide the upper and lower bounds of the UN population projectons,
indicating the margin of uncertainty in the projection (Sawyer & Bassarsky, 2016). The medium variant is employed for the
baseline out-of-sample forecasts in Fig. 3.
The first striking feature of the plots in Fig. 3 is that the baseline demographic and macroeconomic futures for Japan do
not follow simple monotonic trends. Rather, the model projects long waves in the macroeconomic variables that are
traceable to the changing demographics, as the historical decline in fertility has varying effects over time on the age
distribution of the population stretching far into the future.
Japan’s total fertility rate declined steadily from 2.14 births per woman in 1973 to 1.26 births in 2005, followed by a rise to
1.45 in 2015. The medium variant of the UN projections calls for further increases in Japan’s total fertility, but at a declining
rate, through the end of our projection horizon (2060), reaching 1.73 births per woman (see United Nations, 2017).16
Consequently, the growth in the proportion of youths (DYOUNG) increases steadily through 2045 and then levels off for the
rest of the projection period. The pattern for the growth in the working age population reflects, in part, the maturation of the
young population into adulthood, as well as the aging of working adults into their retirement years. For the first ten years of
the projections, the growth in the working age populations rises, while remaining negative, reflecting the rise in fertility that
began in 2005. By 2025, DWORKING turns down again with the aging of older adults into their retirement years. This latter
effect is apparent in the mirror images of the patterns in DWORKING and DOLD. The first of these rises until 2026, declines
from 2026 to 2036, levels off through 2040, and rises again through 2060; the growth in the elderly population proportion is
the exact opposite.
These patterns in the three demographic measures influence the long run macroeconomic trends whose turning
points are related to the demographic changes. Real GDP, for example, responds positively to the projected upward trends
of the working-age population, growing steadily into the early 2030s. Then real GDP remains flat for the decade of the 2030s,
15
The sources of data for the UN population projections for all countries is available at http://esa.un.org/unpd/wpp/DataSources/ (United Nations, 2017).
16
http://data.un.org/Data.aspx?d=PopDiv&f=variableID%3A54.
Fig. 3. Out-of-sample Forecast – Baseline.
reflecting the downturn in DWORKING, but with a lag. With the revival of the growth in the working age population ratio
after 2040, real GDP again resumes positive growth until the end of the projection period. A similar pattern holds for the
gross private investment ratio, which peaks in 2030, declines from 2030 into the early 2040s, and resumes its positive
growth path thereafter. The mixtures of positive and adverse macroeconomic episodes, reflecting the changing patterns of
demographic variables, is a surprising finding of this analysis. Long run macroeconomic pessimism arises incorrectly from
the simple extrapolation of population aging into the future. Fig. 3 demonstrates the alternative future in which fluctuations
in the growth of demographic ratios lead to similar patterns of growth and stagnation for the macroeconomy.
Gross domestic saving is modelled as a function of DOLD and DYOUNG in the long run relation, both with negative
influences on saving. Graphically, the pattern for the projected saving ratio is the mirror image of that for DOLD. The
projections for the saving ratio exceed those for investment, contributing to an expected improvement in the current
account balance, which is projected to maintain a surplus after 2025. However, deficits in fiscal account attenuate the
tendency towards surplus coming from the private sector, so that a deficit in the current account holds in the early period
prior to 2025. Subsequent fluctuations in the current account balance reflect changes in the fiscal account, attributable to less
government spending on education and healthcare (due to less pressure from young and old dependency ratios), and an
increase in tax revenues owing to stronger growth in the workforce.
5.4. Out-of-sample projections: alternative fertility scenarios
Recent increases in Japan’s total fertility rate raise the possibility that the UN’s medium variant fertility
assumptions are too low. The total fertility rate in Japan bottomed-out in 2005 at 1.260 births per woman, then rose
steadily to 1.390 in 2010 and 1.450 in 2015 (United Nations, 2017).17 The high variant projections assume a path for the
total fertility rate that ultimately becomes 0.5 births per woman higher than in the medium variant. These two
scenarios are compared in Fig. 4. One notable difference in the demographic trends is the more rapid, but uneven,
growth of the youth population ratio in the high variant. The growth in the elderly population ratio is slower in the
17
http://data.un.org/Data.aspx?d=PopDiv&f=variableID%3A54.
Fig. 4. Out-of-sample Forecast, Baseline vs High Variant.
high variant scenario, and this contributes to lower fiscal deficits throughout the projection period. The reduced fiscal
deficit is reflected in the higher current account surplus, with differences in the trends in private saving and
investment more-or-less offsetting. In addition, the higher fertility rates early in the projection period result in slower
growth in the working age population ratio through 2035, followed by higher growth thereafter. Both the investment
ratio (GFPC) and real GDP differences from the baseline projection follow the trend in DWORKING, showing lower levels of
investment until 2038 and of real output until 2040, and higher levels in the later years.
The UN’s low variant projections assume a total fertility rate that approaches 0.5 fewer average births per woman than
in the medium variant. Therefore the comparisons of the low variant and baseline projections in Fig. 5 are the mirror image
of those in Fig. 4. For instance, the lower fertility rate results in higher growth of the share of old-age population and lower
growth of the youth population ratio. In this low variant the growth in the working age share of the population is higher
than the baseline until 2033, and consequently the private investment ratio and real GDP are expected to outperform their
baseline levels through 2038 and 2040, respectively. However, in the post-2033 years, the growth in the share of the
working-age population is expected to decline relative to its baseline, leading, after some lags, to underperforming levels of
real GDP, saving, and investment. In addition, the higher growth in the elderly population ratio adversely affects the fiscal
balance, which in turn results in lower current account surpluses, and even deficits over the 2037–2047 period. Thus, the
twin deficit phenomenon could occur if fertility is lower than under the baseline projections.
The interesting and innovative discovery from these alternative scenarios is that a monotonic change in fertility does not
have one-sided consequences for the macroeconomic variables. A decline in the total fertility rate, for example, results in
macroeconomic outcomes that are more favourable in some periods, and less favourable in others, than the macroeconomy
under the baseline scenario.
Fig. 5. Out-of-sample Forecast, Baseline vs Low Variant.
6. Discussion and conclusion
The underlying perspective of this study is that demographic changes arise slowly and have long run effects on key
macroeconomic variables. Over the period of this study from 1970 to 2015, Japan has experienced long waves of
demographic change. In the early years into the late 1980s, Japan reaped the benefits of the demographic dividend, with a
growing proportion of the working-age population contributing to strong economic growth and high rates of private saving.
This was followed by a long period of population aging, resulting from a steady decline in fertility rates, and rates of saving,
investment, and GDP growth fell persistently. These dramatic changes in the age distribution of the population,
characterized in the early period by the demographic dividend, as well as the later episode of population aging, make the
Japanese experience ideal for the analysis of long run equilibrium relations between demographic and macroeconomic
variables.
This study emphsizes cointegration analysis, which is inherently long run and appropriate for analyzing effects of slow
moving demographic variables. In particular, the cointegration tests are based on individual augmented autoregressive
distributed lag (Augmented ARDL) equations that incorporate both long run and short run interactions among the variables.
This framework allows flexibility in the selection of explanatory variables appearing in each equation, as well as different lag
structures for each variable in each equation. This contrasts, for example, with systems based cointegration approaches, such
as presented by Johansen, that generally impose identical lag structures and include the same right-hand-side variables in
every equation. These ARDL equations are simplified through the general-to-specific modelling approach, leading to much
simpler empirical equations than suggested by the theories. An important finding of this study is the strong statistical
evidence for cointegration in each equation relating key macroeconomic variables to underlying demographic drivers.
This study has emphasized the rigorous application of time series methods to testing and estimation of a system of
dynamic equations suitable for conditional forecasting. An important early finding from this analysis is that the population
age shares are highly non-stationary, characterized by two unit roots. Since the macroeconomic variables are typically
integrated of order one, the age shares can enter into the long run relations with the macro variables only after first
differencing, and this is a fundamental departure from other time sries studies in this literature. Only in this first-differenced
form can the demographic factors account for the long waves exhibited by the macroeconomic variables.
The age distribution of the population is represented by three population age shares: youths, working age, and the elderly.
This allows old and young dependency rates to have different effects on the macro variables. In addition, different age shares
can serve as determinants of different macro variables. This contrasts with related studies that attempt to summarize the
entire age distribution with a small number of parameters, or others that choose only one segment of the age distribution,
such as the working age population. The key macroeconomic variables of interest are the ratios of private saving, investment,
and fiscal balance relative to GDP. Although the effects of the demographic factors on these three variables are important in
their own right, they also determine the current account balance through the fundamental accounting definition. Indirectly,
these macroeconomic ratios are important indicators of the long run health of Japan’s economy and prospects for economic
growth.
One limitation of this analysis is the assumption that demographic changes are exogenous and unaffected by
macroeconomic variables. This exogeneity assumption is supported by the observation that the demographic ratios evolve
slowly, so that feedback from macroeconomic conditions to the demographic ratios is expected to be weak. However, an
interesting extension of this study would be to allow macroeconomic conditions to influence the demographic variables
through changes in fertility and mortality rates. The implied population projections from this more complete model could
then be compared with those from the United Nations.
The main empirical findings implied by the long run cointegrating equations are that the effects of demographic change
on the macroeconomic variables are statistically significant and powerful. For example, a one standard deviation increase in
the growth of the elderly share of the population leads to a five percentage point decrease in the gross private saving rate in
the long run. Similarly, a one standard deviation increase in the growth of the youth population ratio leads to a 3.4 percentage
point reduction in the saving ratio. As a further example, when the growth in the elderly population increases by one
standard deviation, the fiscal account balance relative to GDP is expected to decline by three percentage points.
Population projections from the United Nations provide further evidence of the importance of the demographic changes
for Japan’s macroeconomic future. Future trends of key macroeconomic variables – saving, investment, real GDP, fiscal
balance, and the current account balance - match the temporal patterns of the demographic drivers. The uneven pace of
population aging means that future trends in the macroeconomy are more favorable in the medium term than would be
expected from simple measures of population aging. Recent increases in total fertility rates in Japan, for example, imply
improved prospects for the Japanese economy during the period from 2025 to 2035, with reductions in the fiscal deficit as a
proportion of GDP, increased saving and investment ratios, and a move towards current account surpluses. This implication
of the empirical model presented here is in sharp contrast with the findings of Kim and Lee (2007) that an increase in the
elderly dependency rate could significantly lower the saving rate, which in turn, reduces the current account balances in East
Asian countries. In addition, the empirical results of this study do not conform with Higgins’ findings that Japan will
experience substantial surpluses in its current account owing to a more rapid decline in investment compared with that of
saving.
Another perspective on the importance of demographic change in Japan’s macroeconomic future is provided by
alternative variants of the UN’s population projections. High and low variants assume total fertility rates that are ultimately
0.5 births per woman higher or lower, respectively, than the medium variant. The high fertility scenario, for example,
generates slower long run growth in the eldery share of the population, with favourable consequences for saving, fiscal
balance, and the current account in the long run. However, according to this high variant, the growth in the share of the
working age population is lower than in the baseline scenario until 2033, resulting in reduced investment ratios and a lower
real GDP trajectory until 2033. This period is followed by a cross-over to higher growth in the working age population share,
leading to a higher investment ratio as well as higher levels of real GDP compared with the baseline projections. On the other
hand, if fertility is lower than in the medium variant, this could pose a threat of a twin deficits phenomenon in the long run
due to lower private saving and tax revenues from slower growth in the working age share and more rapid increases in the
share of the elderly population.
References
Derek Anderson, Dennis Botman, Benjamin Hunt, 2014. Is Japan's population aging deflationary? IMF Working Papers, 14/139, International Monetary Fund.
Athukorala, P., & Tsay, P. L. (2003). Determinants of Household saving in Taiwan: Growth demography and Public Policy. The Journal of Development Studies,
39(5), 65–88.
Banerjee, A., Dolado, J., Galbraith, J. W., & Hendry, D. F. (1993). Co-integration, error correction, and the econometric analysis of non-stationary data. Oxford:
Oxford University Press.
Banks, J., Richard, B., & Sarah, T. (1998). Is there retirement-savings puzzle? The American Economic Review, 91(4), 769–788.
Bernheim, D., Skinner, J., & Weinberg, S. (2001). What Accounts for the Variations in Retirement Wealth among U.S. Households? The American Economic
Review, 91, 832–857.
Bloom, D. E., Canning, D., & Fink, G. (2010). Implications of population ageing for economic growth. Oxford Review of Economic Policy, 26, 583–612.
Bloom, D. E., Somnath, C., Paul, K., Peter, L.-S., Martin, M., Bernd, R., et al. (2015). Macroeconomic implications of population ageing and selected policy
responses. Lancet, 385, 649–657.
Bobeica, E., Lis, E., Nickel, C., & Sun, Y. (2017). Demographics and inflation. European central bank working paper, #2006. .
Bosworth, B., & Chodorow-Reich, G. (2007). Saving and demographic change: The global dimension. Center for Retirement Program (CRP) Working Paper 2007-02.
Boston College.
Braun, A., Ikeda, D., & Joines, D. (2009). The saving rate in Japan: Why it has fallen and why it will remain low. International Economic Review, 50, 291–321.
Bullard, J., Garriga, C., & Waller, C. J. (2012). Demographics, redistribution, and optimal inflation. Federal Reserve Bank of St. Louis Review419–420.
Carrol, C. D. (1997). Buffer-stock saving and the life cycle/permanent income hypothesis. The Quarterly Journal of Economics, 112(1), 1–55.
Chinn, M. D., & Prasad, E. S. (2003). Medium-term determinants of current account in industrial and developing countries: An empirical exploration. Journal
of International Economics, 59(1), 47–76.
Curtis, C., & Lugauer, S. (2019). “Population aging, economic growth, and the importance of capital in production” mimeo. University of Richmond.
Curtis, C., Lugauer, S., & Mark, N. (2017). Demographics and Household Saving in Japan, China, and India. Journal of Macroeconomics, 51, 175–191.
Danziger, S., Van Der Gaag, J., Smolensky, E., & Taussig, M. K. (1982). The life-cycle hypothesis and the consumption behavior of the elderly. Journal of Post
Keynesian Economics, 5(2), 208–227.
Edwards, S. (1995). Why are saving rates so different across countries? An international comparative analysis. NBER working paper No. 507. Cambridge, MA:
National Bureau of Economic Research.
Fougere, M., & M'erette, M. (1998). Population aging and the current account in selected OECD countries. Working Paper 98-03. Canada: Department of Finance
Canada.
Futagami, K., & Nakajima, T. (2001). Population aging and economic growth. Journal of Macroeconomics, 23(1), 31–44.
Gajewski, P. (2014). Is ageing deflationary? Some evidence from OECD countries. Applied Economics Letters. http://dx.doi.org/10.1080/13504851.2014.987911.
Higgins, M. (1998). Demography, national savings, and international capital flows. International Economic Review, 39(2), 343–369.
Horioka, C. (1992a). The determinants of Japan’s saving rate: The impact of the age structure of the population and other factors. The Economic Studies
Quarterly, 42(3), 237–253.
Horioka, C. (1992b). Future trends in Japan’s saving rate and the implications for Japan’s external imbalance. Japan and the World Economy, 3, 307–330.
Jappelli, T., & Modigliani, F. (1998). The age-saving profile and the life-cycle hypothesis. Working paper. Fisciano (Italy). Centre for Studies in Economics and
Finance.
Juselius, M., & Takáts, E. (2015). Can demography affect inflation and monetary policy?”, BIS Working Paper 485/. .
Kim, S., & Lee, J.-W. (2007). Demographic changes, saving, and current account in East Asia. Asian Economic Papers, 6(2), 22–53.
Kim, S., & Lee, J.-W. (2008). Demographic changes, saving, and current account: an analysis based on a panel VAR. Japan and the World Economy, 20(2), 236–
256.
Konishi, H., & Ueda, K. (2013). Aging and deflation from a fiscal perspective. IMES Discussion Paper Series, No. 2013-E-13. Institute for Monetary and Economic
Studies, Bank of Japan.
Kwack, S. Y., & Lee, Y. S. (2005). What determines saving rates in Korea? The role of demography. Journal of Asian Economics, 16(2005), 861–873.
Lee, S. H., Mason, A., & Park, D. (2011). Why does population aging matter so much for Asia? Population aging, economic security and economic growth in
Asia. ERIA discussion paper series, ERIA-DP-2011-04. .
Lindh, T., & Malmberg, B. (2000). Can age structure forecast inflation trends? Journal of Economics and Business, 52, 31–49. http://dx.doi.org/10.1016/S0148-
6195(99)00026-0.
Luehrmann, M. (2003). Demographic change, foresight and international capital flows. MEA discussion paper series 03038. Munich Center for the Economics of
Aging (MEA) at the Max Planck Institute for Social Law and Social Policy.
Maestas, N., Mullen, K. J., & Powell, D. (2016). The effect of population aging on economic growth, the labor force and productivity. NBER working paper
#22452. .
Masson, P. R., Bayoumi, T., & Samiei, H. (1998). International Evidence on the Determinants of Private Saving. The World Bank Economic Review, 12, 483–501.
McNown, R., Sam, C. Y., & Goh, S. K. (2018). Bootstrapping the autoregressive distributed lag test for cointegration. Applied Economics, 50, 1509–1521.
Ministry of Internal Affairs and Communication, Statistics Bureau (2018). Chapter 2: Population and households. Japan Statiscal Yearbook. Retrieved 15
November 2018 https://www.stat.go.jp/english/data/nenkan/67nenkan/1431-02.html.
Mirer, T. W. (1979). The wealth-age relation among the aged. The American Economic Review, 69(3), 435–443.
Modigliani, F., & Cao, S. L. (2004). The chinese saving puzzle and the life-cycle hypothesis. Journal of Economic Literature, 42(1), 145–170.
Ouliaris, S. (1981). Household saving and the rate of interest. The Economic Record, 57(3), 205–214.
Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16, 289–326.
Prettner, K. (2012). Population aging and endogenous economic growth. Journal of Population Economics, 26, 811–834.
Sam, C. Y., McNown, R., & Goh, S. K. (2018). An augmented Autoregressive distributed Lag bounds test for cointegration. Economic Modeling, 80, 130–141.
http://dx.doi.org/10.1016/j.econmod.2018.11.001.
Sawyer, C., & Bassarsky, L. (2016). Projection outputs and variants. Regional workshop on the production of population projections Addis Ababa, 7-11 March, 2016. .
Stock, J. H. (1995). Point Forecasts and Prediction Intervals for long horizon forecasts, manuscipt. J.F.K. School of Government, Harvard Unviersity.
The Economist (2014). Japan’s demography: The incredible shrinking country, the economist. 25 March 2014. Retrieved 28 November 2018. .
Tujula, M., & Wolswijk, G. (2004). What determines fiscal balances? An empirical investigation in determinants of changes in OECD budget balances, working paper
series No. 422. Frankfurt am Main: Euopean Central Bank.
United Nations (UN), Department of Economic and Social Affairs, Population Division (2017). World population prospects: The 2017 revision, methodology of
the united nations population estimates and projections,Working paper No. ESA/P/WP.250. New York: United Nations.
Wan, H., Daniel, G., & Paul, K. (2016). An aging world: 2015. International population reports, P95/16-1. Washington, DC: U.S. Census Bureau.
Weil, D. N. (1994). The saving of the elderly in Micro and macro data. The Quarterly Journal of Economics, 109, 55–81.
Yoon, J.-W., Kim, J., & Lee, J. (2014). “Impact of demographic changes on inflation and the macroeconomy” IMF working paper 14/210. .

Goh 2020

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Goh 2020

Uploaded by

Copyright:

Available Formats

Journal of Asian Economics 68 (2020) 101198

Contents lists available at ScienceDirect

Journal of Asian Economics

Full length article

Macroeconomic implications of population aging: Evidence

4. Methodology and data

ADF test PP test

Level 1st 2nd Level 1st 2nd

Notes: , , indicate statistical signiﬁcance at 10 %, 5% and 1% levels, respectively.

5.1. Unit root tests

5.2. Augmented ARDL estimation results

Equations Test statistic & diagnostic

Fig. 2. In-sample plots of the actual and the ﬁtted value.

5.3. Out-of-sample projections: baseline scenario

Fig. 3. Out-of-sample Forecast – Baseline.

5.4. Out-of-sample projections: alternative fertility scenarios

Fig. 4. Out-of-sample Forecast, Baseline vs High Variant.

Fig. 5. Out-of-sample Forecast, Baseline vs Low Variant.

6. Discussion and conclusion

You might also like