Content uploaded by Konstantin Kholodilin
Author content
All content in this area was uploaded by Konstantin Kholodilin on May 02, 2023
Content may be subject to copyright.
Does aging influence structural change? Evidence from
panel data
Boriss Siliverstovs
a
, Konstantin A. Kholodilin
b
, Ulrich Thiessen
b,
*
a
ETH Zurich, KOF Swiss Economic Institute,Weinbergstrasse 35, 8092 Zurich, Switzerland
b
DIW Berlin, Mohrenstrasse 58, 10117 Berlin, Germany
1. Introduction
The process of aging plays an important role in shaping modern society. It is no longer limited to a
handful of wealthy industrial countries but increasingly extends to the developing world. Moreover,
during the recent past, the aging process intensified. Demographic projections for the next decades
depict a society dominated by the gray-heads.
Acknowledging the ongoing process of population aging, a still limited but growing literature
investigating its effects on various economic processes emerged. One branch of this literature
addresses the question whether aging increases the size of the welfare state (e.g. Shelton, 2008; Razin
Economic Systems 35 (2011) 244–260
ARTICLE INFO
Article history:
Received 30 March 2009
Received in revised form 4 May 2010
Accepted 12 May 2010
Available online 14 October 2010
JEL classification:
J11
O57
C33
Keywords:
Structural change
Aging
Employment shares
Dynamic panel data
ABSTRACT
Our study represents a first attempt to single out the effects of aging
on the entire structure of the economy that is approximated by
employment shares in different sectors. We find that even after
controlling for the effects of other relevant factors – e.g., income per
capita, share of trade in GDP, government consumption share in
GDP, population size – aging does have a statistically significant
differentiated impact on the employment shares. In particular, we
find that an increase in aging exerts a statistically significant
adverse effect on the employment shares in agriculture, manu-
facturing, construction, and mining and quarrying industries. At the
same time, an increasing share of the elderly (decreasing share of
the youth) in society positively affects employment shares in
community, social, and personal services as well as in the financial
sector.
ß2010 Elsevier B.V. All rights reserved.
* Corresponding author. Tel.: +49 30 89 789 346.
E-mail addresses: boriss.siliverstovs@kof.ethz.ch (B. Siliverstovs), kkholodilin@diw.de (K.A. Kholodilin), uthiessen@diw.de
(U. Thiessen).
Contents lists available at ScienceDirect
Economic Systems
journal homepage: www.elsevier.com/locate/ecosys
0939-3625/$ – see front matter ß2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.ecosys.2010.05.004
et al., 2002). Another strand concentrates not only on the effects of aging on aggregate economic
performance (e.g. Oliveira Martins et al., 2005; Bo
¨rsch-Supan, 2003b; McMorrow and Roeger, 2003,
1999; Bo
¨s and von Weizsa
¨cker, 1988; Denton and Spencer, 1998) but also on the channels through
which aging affects the economy, including savings and capital flows (e.g. Bo
¨rsch-Supan et al., 2006;
Bo
¨rsch-Supan, 2001), productivity (e.g. Bo
¨rsch-Supan et al., 2005; Prskawetz et al., 2007), labor
demand and supply (e.g. Sapozhnikov and Triest, 2007; Prskawetz et al., 2007; Bo
¨rsch-Supan, 2003a),
and income inequality (e.g. Fehr et al., 2008). There is also a branch that investigates changes in
consumption patterns induced by a growing share of elderly people in society (e.g. Buslei et al., 2007;
Lu
¨hrmann, 2005; Oliveira Martins et al., 2005; Bo
¨rsch-Supan, 2003a,b; Serow and Sly, 1988).
The latter branch of research is of most relevance to our paper as it attempts to indirectly infer the
extent to which aging affects the structure of the economy through assessing changes in private
consumption caused by aging in the first place. But this is obviously a limited approach and hence
there is a need to estimate the effects of aging on the structure of the economy directly. The re-
allocation of resources among different sectors – brought about by rapid senescing of modern
societies, among other things – is likely to result in considerable economic and social costs. Therefore,
it is important already now to provide estimates of the potential scale and directions of this process.
To the best of our knowledge, a direct assessment of the potential effects of aging on the whole
economic structure and in particular on sectoral employment shares has been given rather little, if
any, attention in the literature. Hence, the main motivation of our paper is to fill this gap by analyzing
the impact of aging on the employment shares in different economic industries. In addition, by
identifying aging of societies as an additional factor which influences the structure of an economy, we
also contribute to a rather extensive literature that studies patterns and determinants of structural
change in modern economies.
In this paper, we intend to single out the effects of aging on the employment shares from those
caused by other factors reviewed above. To this end, we employ dynamic panel data analysis based on
51 countries and covering a time period from 1970 till 2004. We utilize three proxies of the aging
process in our analysis. The first one is a conventional demographic measure based on the ratio of
elderly to total population, whereas the second measure based on the ratio of elderly to working age
population is related to the mounting burden on working tax-payers in rapidly senescing societies.
The third measure representing the ratio of youth in the total population is used for the purposes of the
robustness check of the results obtained for the previous two variables. Our results suggest that the
aging variables do exert a statistically significant differentiated impact on the employment shares
when controlling for other relevant factors, i.e., income per capita, share of trade in GDP, government
consumption share in GDP, population size, and resource endowment. In particular, we find that an
increase in the aging proxies exerts a statistically significant adverse effect on the employment shares
in agriculture, manufacturing, construction, and mining and quarrying industries.
At the same time, an increasing share of elderly people in society positively affects employment
shares in community, social, and personal services as well as in the financial sector. The remainder of
the paper is structured as follows: Section 2summarizes the main features of the aging process in
selected OECD countries in the past, present and future, given the demographic projections. Section 3
briefly outlines the possible channels through which aging may affect employment shares and Section
4describes the data set employed in this study. Section 5discusses the empirical model, the
econometric estimation method and our results. Finally, Section 6concludes.
2. The aging process over the last 150 years
It should be noted that in most industrialized countries the increase in the ‘‘dependency ratio’’ (as
measured by the ratio of persons not involved in the working and income-generating process relative
to those that are involved) has a far longer history than is usually assumed. In many industrial
countries it already started several centuries ago. It is difficult, however, to illustrate this on a
comparative scale using the conventional dependency ratios, i.e., the ratios of young and/or old
persons to those 15–64 years of age. The reason is that in such ratios the only criterion used to
distinguish between those involved and not involved in the income-generating process is age.
However, in former times, a larger share of the persons younger than 20 years were working, i.e., had a
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
245
gainful occupation, than today. Moreover, in many countries the share of young people enrolled in
higher education has been steadily growing, as has the average age at which gainful employment
starts. In addition, there may be persons in the age group of the commonly defined labor force (people
of 20–65 years) that are, in fact, not participating. Finally, detailed population statistics only started in
the second half of the 19th century. Therefore, taking all these factors into account, one may conclude
that in many industrialized countries the conventional youth dependency ratio underestimates the
scale of the long-term phenomenon of increasing youth dependency. This also implies that the rising
long-term trend of the total dependency ratio may be somewhat underestimated. Hence, the
statistical indicators of dependency, at least those covering the late 19th and early 20th century as
well as the projected
1
dependency ratios, have to be interpreted with great care.
As an example of the aging process in a typical industrialized country the case of Germany is
presented in Fig. 1. It shows the ratios of the old (65 years and older) and young (between 0 and 14
years) to the whole population as well as the ratio of the sum of young and old to the total population
over the period 1871–2050. It can be seen that the whole period was characterized by two
demographic changes going in opposite directions. Firstly, the ratio of young to the total population
has been steadily decreasing from about 44% in the end of 19th century to 20% in 2005. For 2050, a
further decrease up to 15% is projected.
Secondly, the ratio of the elderly to the total population has been almost uninterruptedly
increasing from around 9% in the late 19th century to 19% in 2005. According to demographic
projections, in 2050 it should reach 33%. As a result of these changes, from the beginning of the sample
period and up to the early 1930s, the ratio of the sum of young and old to the total population
experienced a decrease from 48% to about 38%, at which level it stabilized during the next 80 years,
interrupted for a short period between 1964 and 1979 as a consequence of the World War II human
losses and the baby boom of the 1960s. However, it is projected to increase further in the forecast
period and attain 48% by 2050.
In recent decades, the increase of the elderly ratio has been a characteristic feature of the majority
of the industrialized countries. Fig. 2 illustrates this for the selected OECD countries and an average of
EU15 members. To facilitate readability of the graph it was split in two panels: one showing the
countries where the ratio of elderly is projected to stabilize in the period after 2030, and another one
displaying countries where this ratio is projected to increase further. Notice that the Asian countries,
such as Japan and South Korea, are experiencing the largest increase in the ratio of elderly to total
[()TD$FIG]
Fig. 1. Population structure in Germany 1871–2050.
1
Population forecasts should be treated carefully because they can only extrapolate the current state. Neither wars, nor deep
changes in fertility ratios (such as the so-called German baby boom) nor technological progress (for instance, the invention of
the anti-baby pill) are predictable.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
246
population among all OECD countries. In 2050, the elderly ratio in Japan and South Korea should be
respectively 8 and 12 times bigger than it was in 1950. The smallest increase is projected in France,
Norway, Sweden, Ireland, and in the USA, where the elderly ratio will be 2.3–2.5 times larger than in
1950. This is slightly less than the projected average of EU15 of about 3 times. In 2050, the highest
elderly ratio among the selected countries is expected to be achieved in Japan, South Korea, and Spain
(39.6%, 38.2%, and 35.7%, correspondingly), whereas the lowest elderly ratio is expected for Turkey,
USA, and Mexico (17.0%, 20.6%, 21.1%, correspondingly).
3. Mechanisms how aging may affect employment shares
Before presenting our empirical analysis we briefly summarize potential mechanisms how aging
may affect employment shares. To structure this overview we use six major potential macroeconomic
channels: the labor supply, consumption patterns, the supply of capital, total factor productivity,
financial markets (i.e., interest rates, exchange rates, capital movements), and public debt. It becomes
clear that each of these effects of aging depends on several behavioral characteristics and parameters,
[()TD$FIG]
Fig. 2. Ratio of population aged 65 and above to total population for selected OECD countries 1950–2050.Source: World
Development Indicators.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
247
which may even change in the process of aging. Hence, each of these effects is theoretically uncertain,
as is their overall effect on the employment structure. The discussion is thus admittedly highly
tentative and only emphasizes the importance of empirical analysis in detecting channels through
which aging may influence structural change in general and sectoral employment shares in particular.
3.1. Labor supply/labor markets
Firstly, ceteris paribus, aging tends to dampen the labor supply and thus also economic growth (e.g.
Oliveira Martins et al., 2005). However, for economic analysis the effective labor supply is relevant, i.e.,
the multiplication of four elements, namely the number of working age persons, the participationrate of
male and female persons, the number of workinghours, and not least the efficiencyor quality of labor. So
even if the first element would decline under aging, the effective labor supply may not shrink due to
increases in one ormore of the other elements. To the extent that the labor supply and with it economic
growth would decline, which is the assumption of most studies of macroeconomic effects of aging,
implicationsfor employment sharesare unclear. On the one hand,a growth decline would be expectedto
dampen relative growth particularly of employment in services and dampen the relative decline of
employmentin industry (e.g., due to lower increasesin living standards, less rapid specialization of labor
and less demand for and outsourcing of services in industrial enterprises). On the other hand, there are
some exceptions because of changes in consumption patterns due to a relatively higher demand for
particularservices especially wantedby the elderly (e.g., health andtraveling services). And itis not clear
whether economic growth is in fact negatively affected even if the effective labor supply declines as this
depends especially on the evolution of the supply of capital and productivity.
3.2. Consumption patterns
Secondly, consumption patterns change due to aging (e.g. Lu
¨hrmann, 2005; Buslei et al., 2007). A
comparison of the structure of goods and services demanded by the different age groups reveals marked
differencessuch that the elderly consumerelatively less clothing,furniture, and education,transport and
leisure services,and relatively more energy, housing, and health and traveling services. Forecasts under
different scenarios regarding population aging, economic growth and social security reform all broadly
confirm a non-negligible impact of aging on the production structure: Production declines in relative
terms regarding food, clothing, furniture, transport, education, leisure, and increases in relative terms
regardingenergy, health and travelingservices. However, additionalinfluences may be considered: First,
social securityreforms would be required to maintain the financial stability of the pay as yougo pension
system in many countries and would, in essence, somewhat lower the growth of financial resources
during retirement. But private savingstend to increase and become moreimportant for financing old age.
And, of course, if young and middle aged people increase their savings, this reduces their consumption
growth to some extent. Overall, these effects appear to reinforce the production changes mentioned
above. Second,technological progress couldinfluence consumption patternsof the elderly: their average
health may improve and their average consumption structure may deviate less from younger
households, for instance, to the extent that technological progress facilitates consumption of particular
goods and services such as education and training through electronic means. Thus, the expected relative
decline of such services would be less pronounced.
3.3. Supply of capital
Thirdly, the effects of aging on capital supply appear to be even more difficult to ascertain. The
extent to which aging influences the private and national saving rate is crucial because savings
determine both the investment rate and capital-to-labor ratio (the capital intensity of production). If
the capital-to-labor ratio and thus capital intensity would fall due to declining savings, GDP per capita
would tend to fall and the real interest rate would tend to rise. The employment structure would
change such that growth of services at the expense of the industry would slow markedly, because with
falling living standards demand for most services would tend to fall, despite possibly increased
demand for those services related to aging. On the other hand, if savings would decline only mildly or
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
248
even increase so that the capital-to-labor ratio would continue to rise, there would be no rise in the
real interest rate and the negative effect on GDP per capita coming from the reduced labor force could
even be more than offset, providing for continued growth of GDP per capita. Employment shares could
continue their long-run pattern of change as identified in the literature on structural change cited
above (such that employment shares of industry and agriculture decline in relative terms and labor
productivity in these sectors continues to increase). Labor shedding in these sectors is associated with
a trend increase in employment in the services sectors as described above. This is reinforced because
productivity growth in the services sectors is subject to natural limits.
2
Looking at this in more detail, aging may reduce the supply of capital, i.e., reduce the national
saving rate, if the aggregate saving propensity of households declines. It has been shown since long
that private saving rates across countries and time are systematically linked to the age structure such
that a higher share of elderly (people above the age of 65) is associated with lower savings rates (e.g.
Kohl and O’Brien, 1998). Sectors with relatively high capital intensity are manufacturing, energy,
agriculture, other natural resource extraction, and construction. Hence, if the supply of capital and
therefore growth of capital intensity would decline due to aging this would be expected, ceteris
paribus, to dampen productivity growth particularly in those sectors with relatively high capital
intensity. This, in turn, would dampen the long-run trend decline of employment particularly in
manufacturing, agriculture, and energy. Simultaneously, growth of employment in the services
sectors would also be dampened. On the other hand, the fear of insufficient pensions from the public
pension system and tax incentives introduced by governments may cause younger households to save
more for retirement. In addition, the elderly may save relatively more compared to past patterns so as
to support their younger family members. (This, however, may cause the younger families to expect
higher bequests and dampen their increased savings efforts). These three factors would work against a
declining supply of capital. Hence, in our view, there is no clear theoretical expectation as to the
overall effect on both the supply of capital and the evolution of capital intensity under aging. And thus
a priori it is not clear whether aging has a dampening effect on the long-run employment decline in
industry and agriculture and on employment growth in services.
3.4. Total factor productivity
Fourthly, another complex issue is that of the evolution of labor efficiency under aging. In the
literature it is often assumed that aging tends to lower overall productivity growth because older
people tend to be less productive and because the capital–labor ratio may decline. However, there is
neither theoretical nor empirical evidence convincingly linking aging to productivity trends (e.g.
Prskawetz et al., 2007; Feyrer, 2007).
3
Two simple facts may be noted: all generations of elderly faced
better preconditions for old age and the need for life long learning tends to be increasingly accepted in
all societies. Even if productivity declines with age for people with paid work, there are several other
important productive contributions of aging, namely volunteer work, support within family and
private networks which could have a positive influence on GDP growth and labor productivity. And as
discussed above it is not clear whether the capital–labor ratio should decline. Hence, it appears
reasonable to adopt a neutral position with regard to the potential influence of aging on productivity,
i.e., productivity evolution may be unaffected.
3.5. Financial markets (i.e., interest rates, exchange rates, capital movements)
If in one country aging would reduce the labor force and national savings would decline more than
investment, its current account position would, ceteris paribus, tend to deteriorate and the exchange
2
For instance, in education and health services, the student/patient to teacher/doctor ratio cannot be increased above certain
limits but rather would be wanted to be reduced.
3
Feyrer (2007) shows that low productivity levels in poor countries are associated with workforces that are very young and
that demographics help explain cross-country productivity differences. This suggests that aging could even promote
productivity. But he is careful to emphasize that his evidence is not sufficient to establish a causal link between demographic
changes and productivity growth.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
249
rate would tend to depreciate. Assuming for simplicity that developed countries are similarly affected
by aging, whereas aging is less pronounced in developing countries, the joint current account position
of developed countries vis-a
`-vis the group of developing countries would tend, ceteris paribus, to
deteriorate. Assuming rising investment needs in developing countries on account of a long-run
catching up process there could be upward pressure on the average world real interest rate level
slowing down worldwide economic growth. However, since the time pattern and the strength of aging
differs from country to country and since national savings do not necessarily have to be negatively
affected by aging, the effects on current account positions, interest rates, exchange rates and capital
movements are highly uncertain and theoretically ambiguous.
3.6. Public debt
Aging is often assumed to increase pressure on public finances and contribute to increasing
public debt to GDP ratios in the period of rising dependency ratios, i.e., during the next several
decades. This effect would, ceteris paribus, tend to cause rising real interest rates and dampen both
investment and economic growth. Slower growth would influence the employment structure as
discussed before, i.e., less growth of employment in services and less decline of industrial
employment. But the number of measures already taken by governments to reduce the growth of
expenditures in public pension and health systems improve the efficiency of public administrations
and public enterprises and provide incentives for households to raise their retirement savings,
suggesting that this effect is uncertain. And even if the public-debt-to-GDP ratio would tend to rise
with aging, one needs to consider that today the public-debt-to-GDP ratio and even the
‘‘generational gap’’ are well publicized concepts and discussed in the media, putting pressure on
households to increase savings. Thus, chances for ‘‘Ricardian equivalence’’ rise, meaning that
changes in public indebtedness tend to be offset by changes in private sector savings which would
work against higher real interest rates.
4
Overall, then, there appears to be only one channel with an a priori theoretical expectation of the
direction of the influence of aging on employment shares, namely changes in the consumption
structure. But again this effect is, like all others, subject to feedback effects from aging on economic
growth, changing behavioral characteristics, and the influence of technological progress.
4. Data
Our data set includes a maximum of 51 selected developing and developed economies.
5
We use the
employment data provided by the International Labor Organization (ILO) for nine sectors, namely total
employment by economic activities as defined in the classification ISIC-Rev.3. All other data – young,
working-age, and old population – were taken from the World Development Indicators data base of
the World Bank. The collected panel data set is unbalanced. The number of observations per country
varies: the shortest time period available for an individual country is seven years, and the longest time
period covers 1970–2004.
Since this paper concentrates on analyzing the effects of aging on the employment structure, we
use sectoral employment shares as the dependent variable in our regressions. Specifically, we use
employment shares in the following nine sectors: agriculture (lfa), manufacturing (lfm), mining and
quarrying (lfmq), electricity, gas and water (lfe), construction (lfc), wholesale, retail trade, restaurants
and hotels (lfw), transport, storage, communication (lft), financial services, real estate and related
4
In its pure form Ricardian equivalence is not confirmed by empirical research but many studies found Ricardian effects in
saving behavior. See Briotti (2005) for a literature overview.
5
Only market economies were included that do not have unusual characteristics, such as, for instance, a very small
population (less than one million) or an extremely large share of GDP derived from extraction of natural resources. The chosen
countries were: Argentina, Australia, Austria, Belgium, Bolivia, Canada, Chile, Colombia, Costa Rica, Cyprus, Denmark,
Dominican Republic, Ecuador, Egypt, El Salvador, Finland, France, Germany, Greece, Honduras, India, Indonesia, Ireland, Israel,
Italy, Jamaica, Japan, Korea, Malaysia, Mauritius, Mexico, Netherlands, New Zealand, Nicaragua, Norway, Pakistan, Panama,
Peru, Philippines, Portugal, Spain, Sri Lanka, Sweden, Switzerland, Thailand, Trinidad and Tobago, Turkey, United Kingdom, USA,
Uruguay, and Venezuela.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
250
services (lff), and community, social, and personal services (lfcs).
6
Over the whole period, about 80% of
total employment were concentrated in four sectors: agriculture (lfa), manufacturing (lfm), wholesale,
retail trade, restaurants and hotels (lfw), and community, social, and personal services (lfcs), whereas
the remaining five sectors together comprised about 20% of total employment. The shares of
agriculture (lfa) and manufacturing (lfm) declined from 26% and 21% to 13% and 15%, respectively,
whereas the shares of wholesale, retail trade, restaurants and hotels (lfw) and community, social, and
personal services (lfcs) went up from 15% and 21% to 19% and 27%, correspondingly. However, when
comparing changes in the employment structure one has to bear in mind that the sample size steadily
increases from 7 countries in 1970 to 46 countries in 2004. The structural change can be illustrated by
taking Germany as a particular example. For Germany we have employment data available from 1991
until 2004. In this relatively short time period, one can observe rather substantial changes in
employment composition across various sectors. Thus, the share of agriculture declined from 4% to 2%,
that of manufacturing went down from 28% to 23%, the share of wholesale, retail trade, restaurants
and hotels increased from 16% to 17%, whilst that of community, social, and personal services went up
from 27% to 31%. Taken together, these four largest sectors account for about 75% of total employment
in Germany, which is very close to the average computed for the whole panel.
It is well documented in the literature that a rise in per-capita income is typically accompanied by
the following changes of relative importance of different industries and hence sectoral employment
shares. First, one observes a continuous decline in the primary and secondary sectors, here
represented by agriculture, manufacturing, mining and quarrying, energy, and construction. Second,
employment shares rise continuously with increasing per-capita income in the tertiary sector
represented by wholesale and retail trade, restaurants and hotels, transport, storage, communication,
financial services, real estate and related services, and community, social, and personal services.
7
The earlier literature, represented by such seminal studies as Bean (1946),Clark (1957), and
Kuznets (1956), viewed rising income levels and associated changes in the composition of demand as a
primary factor explaining the rising share of manufacturing at the expense of more traditional
endowment-related industries. However, their results were based on the data for the first half of the
20th century, when industrialization was still going on. Chenery (1960) argued that in addition to
demand-related factors, changes in supply conditions like the capital stock per worker, education and
skill levels should be taken into account when looking at the determinants of structural change. Also,
the established fact that patterns of trade change systematically with income levels (Hilgerdt, 1945)is
accommodated by his theoretical model. In a more recent paper by Rowthorn and Ramaswamy
(1999), the close connection between growth, trade, and (de-)industrialization is also emphasized.
Hence the empirical analysis begun by Chenery (1960) on the determinants of growth in
manufacturing was initially based only on two variables, namely income per capita and population
size. Chenery and Taylor (1968) extended this model by directly incorporating factors related to
resource endowment and the degree of openness of the economies. In addition, they experimented
with different functional forms of the estimated equations. For instance, by including squared log-
income as an additional explanatory variable, they allowed for non-constant income elasticity in
sectoral equations. This enabled them to consider the empirical observation of a slowing down or even
reverting of the growth of manufacturing for the benefit of services (e.g., see Fuchs, 1968; Rowthorn
and Wells, 1987; Sachs and Shatz, 1994; Rowthorn and Ramaswamy, 1997, 1999, for theoretical
explanations of a declining share of [employment in] manufacturing). As argued in Chenery and Taylor
(1968), by considering different types of variables and functional forms it is possible to accommodate
different theories of development patterns.
Inspired by the literature on the determinants of structural change reviewed above our
parsimonious specification of the empirical models includes the explanatory variables per capita
income (gdppc), measured in purchasing power parities, the size of the economies proxied by
6
The employment data have a residual due to unexplained employment (‘‘activities not adequately defined’’), i.e., the shares
do not always add up to one and hence our dataset does not fulfill this aggregation constraint. At the same time, using this
dataset, we are able to track changes for most of employment in the economy as the unexplained portion is relatively small
(below 5%).
7
The latter sector includes government.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
251
population (pop), a proxy for ‘‘openness’’ (tr), i.e., the sum of exports and imports as a ratio to GDP, and
a variable to capture the effect of government policies, namely the government consumption
expenditure share (gcogdp). In order to reflect the argument of Chenery and Taylor (1968) on the
importance of flexible functional forms we included squared log-income as an additional explanatory
variable. Regarding the agricultural sector, a proxy for the endowment with agricultural resources was
also included (the share of exports of agricultural products in exports of merchandise goods, xaxme).
However, our variable of primary interest is aging, represented by the three proxies already described,
namely the ratio of elderly either to the total population (age) or to the labor force (odep), or the ratio of
youth to the total population (young). In the following analysis, we use the logarithmic transformation
of per capita income and of total population. All other variables, including the dependent variables, are
left intact since they are already expressed as percentage shares.
5. Empirical findings
5.1. Econometric model
In the following, we estimate the parameters of the following dynamic panel data model
em p
it
¼dem p
i;t1
þb
0
X
it
þlaging
it
þu
0
D
it
þu
it
(1)
where emp
it
is the employment share in either of the nine industries in question, X
it
is a K1 vector of
the explanatory variables which were discussed in Section 4above. The variable aging
it
represents our
variable of interest and the corresponding coefficient measures the impact of a changing proportion of
elderly people in society on the employment shares in different industries. As usual, a vector of the
cross-sectional fixed effects, represented by D
it
, can be added to the regression model (1). Note that in
order to account for temporal correlation of the dependent variable, we include its first lag in the
model. Since we use yearly data, such lag length proved to be appropriate for our empirical modeling.
It is well-known that in the presence of the fixed effects, estimation of the parameters of the
dynamic panel data model is subject to estimation bias, which can be quite severe if the panel time
dimension is rather small (Nickell, 1981; Kiviet, 1995). As the solution to the estimation bias, a number
of panel data estimators have been proposed, including the instrumental estimator of Anderson and
Hsiao (1982) that uses the first-differences of the data in order to eliminate the fixed effects.
Expanding on the Anderson and Hsiao (1982) estimator, Arellano and Bond (1991) show that there are
many more instruments available within the GMM framework than used by conventional
instrumental variable estimation. The GMM estimator of Arellano and Bond (1991) is the two-
step estimator. In the first step, the parameters are estimated using the identity matrix for weighting
the moment conditions. In the second step, an asymptotically more efficient estimation is conducted
by optimal weighting of the moment condition using the first-step estimation results.
As a consequence, the efficiency of GMM-based estimators is greatly enhanced when compared to
that of the Anderson and Hsiao (1982) estimator. However, as Baltagi et al. (2000) point out, the
estimators of Anderson and Hsiao (1982) and Arellano and Bond (1991) may eliminate the estimation
bias, but with a large loss of information. The application of the first-difference transformation
destroys the economic structure formed between the levels of the variables across the time series
dimension. Fortunately, the two-step GMM estimator developed by Arellano and Bover (1995)
addresses this issue by employing both the first differences as well as the levels of the variables by
specifying the appropriate sets of instruments for both types of equations. Below, we estimate the
parameters of Eq. (1) by means of the two-step GMM estimator suggested in Arellano and Bover
(1995). Our inference on the significance of the estimated parameters is based on the small sample
correction of Windmeijer (2000) applied to the estimated covariance matrix in the second step of the
corresponding GMM procedure.
As noted in Phillips and Sul (2007), the performance of the GMM estimators can be unsatisfactory
in situations where time series exhibit high persistence. Hence, in order to obtain sensible results, we
need to check how persistent the dependent variable is. In order to check this, we have estimated the
first-order panel autoregressive model for each of the nine industrial sectors using the Arellano and
Bover (1995) GMM estimator. The results of our exercise provide very strong evidence that our
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
252
dependent variable is highly persistent: the estimated value of the autoregressive parameter is within
the range of 0.971–0.995 for employment shares in eight sectors, and regarding the ‘financial service’
sector its value is 1.005, indicating non-stationarity of our data. A similar pattern of high temporal
persistence is observed for the explanatory variables collected in X
it
.
Given our findings, one may consider testing for the presence of unit roots in the variables of
interest and subsequently testing for the presence of cointegration. However, given the very
unbalanced nature of our data and the significant share of countries where the time dimension is
either smaller or of a comparable magnitude to the number of the explanatory variables, testing for
cointegration in our data set would involve the omission of a substantial number of countries (i.e.,
observations) and hence it may bias our results. Instead, we have chosen to transform the variable in
question by taking first differences. Hence the estimated model takes the following form
Dem p
it
¼dDem p
i;t1
þb
0
DX
it
þlDaging
it
þy
it
(2)
where
D
is the first-difference operator. Note that such transformation implies that the application of
the GMM estimator of Arellano and Bover (1995) involves the moment conditions among the first-
differences of the levels of the original variables as well as among their second-differences. An
additional advantage of estimating the model in first differences is that we avoid the spurious
regression problem caused by nonstationarity of the modeled variables. As usual, the validity of the
moment conditions and of the model assumptions is checked with the Sargan over-identification test
as well as with the Arellano–Bond first- and second-order residual autocorrelation tests.
5.2. Estimation results
Before reporting estimation results it is instructive to look at the correlations between the three
measures of the aging process – ratio of old (>65 years) people to working or middle aged persons
(odep), ratio of old people to the population (age), ratio of young (<15 years) persons to the population
(young) – that are used in our study. As expected, there is a very high positive correlation between the
age and odep measures (0.996), explaining the very similar results obtained for these measures.
Unsurprisingly, we also observe a negative correlation between the pairs of age and young and odep
and young (0.888 and 0.921, respectively), implying that an opposite sign on the variable young to
that observed for either age or odep is expected in the regressions.
Tables 1–3 contain the estimation results of our panel data regressions using the share of elderly in
total population, the share of elderly in the labor force, and the share of youth (younger than 15 years)
in the total population. First, one observes that the first two sets of regressions yield very similar
results, i.e., the coefficient values and their significance for all explanatory variables, except our
proxies for the aging variable, match each other quite closely. The difference in the values of the
estimated coefficients of the first two aging proxies is due to different scaling, i.e., we scale the same
variable (number of elderly) either by total population or by labor force. In contrast, in the case of the
third aging proxy (young) the estimation results are somewhat different. Second, according to the
model specification tests, the estimated regressions do not deviate from model assumptions: we
cannot reject the null hypothesis of the validity of moment conditions according to the Sargan test
and, as expected, the Arellano–Bond autocorrelation tests indicate the presence of the first-order
autocorrelation in the panel regression residuals. At the same time, no second-order autocorrelation is
detected in the regression residuals. The own lag of the dependent variable is significant in five out of
nine sectors and takes negative values for all sectors. Thus, the absolute value of the autoregressive
parameter is much less than unity, indicating that by taking the first-difference transformation of our
data we have solved the problem of extremely high persistence and/or of non-stationarity. We also
find that our estimation results suggest that changes in employment shares in agriculture are
negatively correlated with changes in per capita income, whereas there is a positive association
between changes in employment shares in manufacturing, construction, and financial sector on the
one hand and changes in per capita income on the other hand. As the coefficient estimate of the square
of per capita income is insignificantly different from zero in all sectors but construction, we find no
statistical evidence for the presence of nonlinear effects between changes in per capita income and
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
253
Table 1
Estimation results: the aging proxy—the share of elderly in total population.
D
lfa
D
lfam
D
lfmq
D
lfe
D
lfc
D
lfw
D
lft
D
lff
D
lfcs
Own lag 0.146
***
0.010 0.025 0.176
***
0.022 0.090 0.158
**
0.250
**
0.128
0.055 0.037 0.049 0.052 0.054 0.050 0.073 0.118 0.104
D
lgdppcpc 8.024
***
3.087
*
0.209 0.102 3.951
***
0.780 0.382 0.934 1.413
1.790 1.610 0.150 0.127 0.830 1.126 0.472 0.602 2.083
(
D
lgdppcpc)2 3.774 28.693 1.588 0.257 32.434
*
11.024 4.013 7.355
13.590 36.680 1.960 3.392 17.310 26.580 10.660 22.830
D
tr 0.018
*
0.004 0.001 0.000 0.001 0.006 0.001 0.002 0.003
0.009 0.008 0.001 0.001 0.003 0.005 0.001 0.003 0.007
D
lpop 10.154
***
2.674 0.055 0.017 4.169
*
5.771 0.695 3.677
**
1.588
2.873 3.317 0.431 0.115 2.256 4.488 1.044 1.702 5.435
D
gcogdp 0.052 0.002 0.009 0.001 0.015 0.021 0.018 0.016 0.048
0.039 0.073 0.006 0.004 0.025 0.030 0.015 0.022 0.047
D
xaxme 0.046
0.037
D
age 0.688
**
0.805
**
0.063
***
0.024 0.239
**
0.321 0.032 0.831
***
1.077
***
0.314 0.336 0.020 0.025 0.115 0.253 0.066 0.174 0.399
s
1.332 1.020 0.139 0.198 0.460 0.693 0.290 0.498 1.050
N50 51 51 51 51 51 51 51 51
Tmax28282828 2828282728
Tmin3333 33333
Obs. 825 862 861 862 862 862 862 852 862
Sargan 46.43 [1.000] 43.21 [1.000] 41.15 [1.000] 46.58 [1.000] 42.65 [1.000] 44.33 [1.000] 48.39 [1.000] 46.32 [1.000] 46.61 [1.000]
AR(1) 2.38 [0.017]
*
2.15 [0.032]
*
2.67 [0.008]
**
3.90 [0.000]
**
4.00 [0.000]
**
3.54 [0.000]
**
3.58 [0.000]
**
1.97 [0.049]
*
2.26 [0.024]
*
AR(2) 0.13 [0.900] 1.18 [0.239] 0.43 [0.668] 0.50 [0.619] 1.62 [0.105] 1.01 [0.312] 0.61 [0.541] 1.31 [0.189] 1.02 [0.307]
The sectors are named as follows: agriculture (lfa), manufacturing (lfm), mining and quarrying (lfmq), electricity, gas and water (lfe), construction (lfc), wholesale, retail trade, restaurants
and hotels (lfw), transport, storage, communication (lft), financial services, real estate and related services (lff), and community, social, and personal services (lfcs). The explanatory variables
include per-capita income (gdppc), measured in purchasing power parities, population (pop), ‘‘openness to trade’’ (tr), and the government consumption expenditure share (gcogdp). Table
entries are estimated coefficients with the corresponding standard errors in parentheses. Bold font represent the coefficients of the variable of main interest – the aging proxy – that are
significant at the usual levels.
*
10% significance level.
**
5% significance level.
***
1% significance level.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
254
Table 2
Estimation results: the aging proxy—the share of elderly in the labor force.
D
lfa
D
lfam
D
lfmq
D
lfe
D
lfc
D
lfw
D
lft
D
lff
D
lfcs
Own lag 0.148
***
0.010 0.025 0.184
***
0.019 0.089
*
0.166
**
0.249
**
0.127
0.056 0.040 0.049 0.051 0.055 0.050 0.077 0.119 0.105
D
lgdppcpc 8.135
***
2.887
*
0.196 0.099 4.023
***
0.846 0.331 1.093
*
1.306
1.802 1.578 0.159 0.136 0.836 1.083 0.518 0.628 2.112
(
D
lgdppcpc)2 0.670 27.789 1.942 0.337 34.016
*
19.987 5.197 5.120
14.230 32.550 2.508 4.365 18.790 27.970 11.140 23.560
D
tr 0.018
*
0.003 0.001 0.000 0.002 0.007 0.001 0.002 0.003
0.010 0.009 0.001 0.001 0.003 0.006 0.001 0.003 0.007
D
lpop 10.301
***
3.608 0.015 0.082 3.800 5.613 0.867 4.013
**
0.089
3.026 3.734 0.355 0.200 2.770 4.019 1.121 1.797 6.620
D
gcogdp 0.063
*
0.008 0.008 0.003 0.016 0.024 0.020 0.018 0.055
0.036 0.071 0.007 0.004 0.024 0.030 0.016 0.023 0.046
D
xaxme 0.049
0.038
D
odep 0.394
**
0.451
**
0.042
***
0.017 0.168
**
0.146 0.007 0.533
***
0.560
**
0.181 0.190 0.014 0.013 0.074 0.135 0.037 0.109 0.247
s
1.332 1.021 0.139 0.198 0.460 0.693 0.290 0.499 1.053
N50 51 51 51 51 51 51 51 51
Tmax28282828 2828282728
Tmin3333 33333
Obs. 825 862 861 862 862 862 862 852 862
Sargan 46.9 [1.000] 44.39 [1.000] 43.64 [1.000] 46.78 [1.000] 43.35 [1.000] 42.68 [1.000] 49.18 [1.000] 46.43 [1.000] 46.92 [1.000]
AR(1) 2.374 [0.018]
**
2.138 [0.033]
**
2.696 [0.007]
***
3.903 [0.000]
***
3.971 [0.000]
***
3.583 [0.000]
***
3.483 [0.000]
***
1.969 [0.049]
**
2.251 [0.024]
**
AR(2) 0.1227 [0.902] 1.174 [0.240] 0.4341 [0.664] 0.6277 [0.530] 1.588 [0.112] 1.005 [0.315] 0.5281 [0.597] 1.312 [0.190] 0.9968 [0.319]
The sectors are named as follows: agriculture (lfa), manufacturing (lfm), mining and quarrying (lfmq), electricity, gas and water (lfe), construction (lfc), wholesale, retail trade, restaurants
and hotels (lfw), transport, storage, communication (lft), financial services, real estate and related services (lff), and community, social, and personal services (lfcs).The explanatory variables
include per-capita income (gdppc), measured in purchasing power parities, population (pop), ‘‘openness to trade’’ (tr), and the government consumption expenditure share (gcogdp). Table
entries are estimated coefficients with the corresponding standard errors in parentheses. Bold font represent the coefficients of the variable of main interest – the aging proxy – that are
significant at the usual levels.
*
10% significance level.
**
5% significance level.
***
1% significance level.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
255
Table 3
Estimation results: the aging proxy—the share of youth (<15 years) in total population.
D
lfa
D
lfam
D
lfmq
D
lfe
D
lfc
D
lfw
D
lft
D
lff
D
lfcs
Own lag 0.146
***
0.024 0.051 0.165
***
0.019 0.092
*
0.164
***
0.235
*
0.119
0.052 0.036 0.039 0.045 0.052 0.054 0.062 0.122 0.100
D
lgdppcpc 7.378
***
1.821 0.163 0.101 3.176 1.000 0.208 1.113 1.904
1.723 1.280 0.149 0.101 0.797 1.081 0.401 0.691 1.755
(
D
lgdppcpc)2 4.003
*
2.178
D
tr 0.020
**
0.001 0.001 0.001 0.004 0.005 0.001 0.003 0.000
0.009 0.008 0.001 0.002 0.003 0.005 0.001 0.003 0.007
D
lpop 7.208
***
2.188 0.138 0.365 3.778 5.248 0.411 3.782 4.712
2.248 2.574 0.639 0.300 1.497 3.736 1.405 2.429 5.251
D
gcogdp 0.063 0.000 0.010 0.002 0.001 0.021 0.016 0.021 0.058
0.038 0.068 0.008 0.004 0.026 0.029 0.014 0.023 0.043
D
x ax me 0.063
0.039
D
young 0.456
**
0.217 0.024 0.028
*
0.146 0.213
*
0.090
*
0.149
*
0.506
**
0.193 0.137 0.022 0.017 0.081 0.125 0.051 0.089 0.204
s
1.325 1.024 0.139 0.198 0.462 0.691 0.290 0.512 1.051
N50 51 51 51 51 51 51 51 51
Tmax 28 28 28 28 28 28 28 28 28
Tmin3333 3 3 333
Obs. 825 862 862 862 862 862 862 852 862
Sargan 47.01 [1.000] 44.35 [1.000] 45.85 [1.000] 41.89 [1.000] 42.96 [1.000] 46.89 [1.000] 46.07 [1.000] 43.08 [1.000] 44.66 [1.000]
AR(1) 2.407 [0.016]
**
2.122 [0.034]
**
2.75 [0.006]
***
4.201 [0.000]
***
3.992 [0.000]
***
3.486 [0.000]
***
3.68 [0.000]
***
1.972 [0.049]
**
2.366 [0.018]
**
AR(2) 0.1707 [0.864] 1.014 [0.311] 0.5175 [0.605] 0.2679 [0.789] 1.541 [0.123] 1.046 [0.295] 0.5817 [0.561] 1.206 [0.228] 0.9614 [0.336]
The sectors are named as follows: agriculture (lfa), manufacturing (lfm), mining and quarrying (lfmq), electricity, gas and water (lfe), construction (lfc), wholesale, retail trade, restaurants
and hotels (lfw), transport, storage, communication (lft), financial services, real estate and related services (lff), and community, social, and personal services (lfcs). The explanatory variables
include per-capita income (gdppc), measured in purchasing power parities, population (pop), ‘‘openness to trade’’ (tr), and the government consumption expenditure share (gcogdp). Table
entries are estimated coefficients with the corresponding standard errors in parentheses. Bold font represent the coefficients of the variable of main interest – the aging proxy – that are
significant at the usual levels.
*
10% significance level.
**
5% significance level.
***
1% significance level.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
256
employment shares for these eight sectors. It is also interesting to note that neither changes in trade
share nor in the government consumption share have significant effects on changes of the sectoral
employment shares.
Next, we assess the impact of our aging variables on the employment shares in the different sectors.
Since the estimation results for the first two aging proxies are very similar, we restrict our discussion
to the results presented in Table 1, i.e., for the share of elderly to the total population. First, we notice
that the corresponding variable is significant for six sectors out of nine. We find that the corresponding
parameter estimate is not significantly different from zero for electricity, gas and water (lfe), wholesale
and retail trade, restaurants and hotels (lfw) and transport, storage, and communication (lft) sectors.
At the same time, we find that the increase in the aging variable has a statistically significant negative
impact on employment shares in the sectors agriculture, manufacturing, mining and quarrying, and in
construction. The strongest effect occurs in agriculture and manufacturing. In two services sectors,
namely, financial and related services and community, social, and personal services, we find that
changes in the share of elderly positively affect employment shares. Thus, aging appears to have a
statistically significant differentiated effect on employment shares in various sectors. The overall
impact of aging is that it promotes ongoing long-run structural change by contributing further to
shrinking agricultural and manufacturing shares in the economy and, at the same time, to enhancing
employment in the financial industry and communal, social, and personal services. This result holds
for both proxies of aging we use, i.e., the share of elderly in total population as well as in the labor force.
The estimation results using the proportion of young people in total population reported in Table 3
are somewhat different than those obtained for the first two aging proxies, age and odep. Therefore, we
describe them separately. Judging from reported regression standard error s
_
the regression fit
obtained with aging measure young is comparable to that reported for the first two aging proxies in
Tables 1 and 2. As expected, the estimated coefficient on the variable young has the opposite sign to
that observed for the variables age and odep in all but one case (for electricity, gas and water, lfe). In
general, however, we observe that the statistical strength of the estimated relationship between the
variable young and employment shares, measured by the corresponding p-values of the estimated
coefficients on the aging proxy, is somewhat weaker than that observed for the variables age and odep.
We have only two coefficient estimates that are significant at the 5% level (for agriculture, lfa, and for
community, social, and personal services, lfcs) and five that are significant at the 10% level.
It is instructive to consider the effect of one standard deviation change in the aging proxies on the
sectoral employment shares expressed in terms of the standard deviation of the dependent variable.
8
This calculation enables us to assess the relative importance of the effect of aging for the employment
shares. First, consider the effects on the employment shares exerted by the first two aging proxies. As
can be seen in the upper and middle panel of Table 4, the effect of aging is strongest regarding
financial, real estate and related services (lff): One standard deviation in the
D
age variable results in
an increase of about 0.163 (0.173) standard deviations of the employment share in financial, real
estate and related services, depending on the aging proxy. Such a strong impact can be explained by
the size effect. Indeed, from 1969 to 2004 the employment share of this sector more than tripled,
starting from a very low level of 3%. Thus, it was the most rapidly growing sector in terms of
employment.
The third aging proxy, young, has a somewhat different (in terms of magnitude) impact on the
employment shares of different sectors. The community, social, and personal services (lfcs) sector is
second with the corresponding impact amounting to about 0.113 (0.097) standard deviations. The
third strongest effect is measured for the employment share of manufacturing (lfm) – 0.090
(0.083). And regarding the three remaining sectors for which we found statistically significant
coefficient estimates (i.e., lfa, lfmq, and lfc), the impact of aging is smallest and relatively similar in
magnitude, in the range of 0.050 to 0.064, depending on the aging proxy and the industry.
For the two sectors with the lowest p-values (lfa and lfcs) we find that the impact of the aging proxy
young on the employment share in agriculture (in absolute value) appears to be about 1.5 times larger
8
Observe that due to the non-stationarity of the variables of interest in levels (recall that the sample variance of the
nonstationary variable increases without limit) such exercise requires a stationarity transformation of these variables by taking
first differences.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
257
compared to that observed either for age or odep, and the impact magnitude is about the same but with
the opposite sign for employment share in community, social, and personal services. In financial and
real-estate services (lff), however, the impact of age and odep is much more pronounced than that of
young. In addition, for lfc,lfw, and lft the measured impact of young also seems to be substantially
larger than that of age and odep.
The use of three different aging proxies based on the share of old people in the total population, age,
and in the labor force, odep, as well as on the share of young people in total population, young,
generally leads to a comparable impact on employment share. The only difference is that in the latter
case the aging proxy has an expectedly opposite sign to that obtained for the former two aging proxies.
For some sectors we observe a stronger, for others a somewhat smaller effect, but no definite pattern
emerges. All in all, our results appear to be robust with respect to the use of different aging proxies.
6. Conclusion
In this paper, we investigated the relationship between the changes in the demographic structure
of the population and the employment shares in the various industries based on an unbalanced panel
of 51 countries covering a maximum time period from 1970 till 2004.
Our results suggest that the aging variable – approximated by the ratio of elderly either to the total
population, or to the labor force, or of the youth to the total population – does have a statistically
significant differentiated impact on the employment shares when controlling for other relevant
factors, e.g., income per capita, share of trade in GDP, government consumption share in GDP,
population size, etc. In particular, we find that an increase in the aging proxies exerts a statistically
significant adverse effect on the employment shares in agriculture, manufacturing, construction, and
mining and quarrying industries. At the same time an increasing share of elderly people in the society
positively affects employment shares in community, social and personal services as well as in the
financial sector. Moreover, we also find that the effect of aging has mostly been pronounced in the
financial, real estate, and related services as well as in the community, social, and personal services,
followed by the manufacturing sector. Hence, we conclude that aging further accelerates the ongoing
long-run structural change usually associated with an ever-increasing role of the services sector at the
Table 4
Estimated effect of one standard deviation change in aging proxies (Row 5) in terms of standard deviations of the dependent
variable.
Row
D
lfa
D
lfm
D
lfmq
D
lfe
D
lfe
D
lfw
D
lft
D
lff
D
lfcs
1 Coef. age 0.688 0.805 0.063 0.024 0.239 0.321 0.032 0.831 1.077
2 st.dev. age 0.113 0.113 0.113 0.113 0.113 0.113 0.113 0.113 0.113
3 Product (row2 row3) 0.078 0.091 0.007 0.003 0.027 0.036 0.004 0.094 0.122
4 st.dev. of dep. var. 1.481 1.017 0.141 0.186 0.491 0.728 0.299 0.577 1.085
5 Ratio (row3/row4) 0.053 0.090 0.051 0.015 0.055 0.050 0.012 0.163 0.113
1 coef. odep 0.394 0.451 0.042 0.017 0.168 0.146 0.007 0.533 0.560
2 st.dev. odep 0.188 0.188 0.188 0.188 0.188 0.188 0.188 0.188 0.188
3 product (row2 row3) 0.074 0.085 0.008 0.003 0.032 0.027 0.001 0.100 0.105
4 st.dev. of dep. var. 1.481 1.017 0.141 0.186 0.491 0.728 0.299 0.577 1.085
5 Ratio (row3/row4) 0.050 0.083 0.056 0.017 0.064 0.038 0.004 0.173 0.097
1 coef. young 0.456 0.217 0.024 0.028 0.146 0.213 0.090 0.149 0.506
2 st.dev. young 0.253 0.253 0.253 0.253 0.253 0.253 0.253 0.253 0.253
3 product (row2 row3) 0.115 0.055 0.006 0.007 0.037 0.054 0.023 0.038 0.128
4 st.dev. of dep. var. 1.481 1.017 0.141 0.186 0.491 0.728 0.299 0.577 1.085
5 Ratio (row3/row4) 0.078 0.054 0.042 0.039 0.075 0.074 0.076 0.065 0.118
The upper and middle panel contains the estimated effect of one standard deviation change in the share of elderly in total
population and in the labor force on the sectoral employment shares as expressed in terms of standard deviation, see the row 5
in the respective panel. The lower panel reports estimated effect of one standard deviation change in the share of youth in total
population on the sectoral employment shares as expressed in terms of standard deviation. All aging proxies and sectoral
employment shares have been transformed to stationarity by taking first difference. For description of the sectors see notes in
Table 3.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
258
expense of traditional sectors like agriculture and manufacturing. Therefore, such a structural change
may imply sizable social and economic costs. In order to minimize these and make the transition
process as smooth as possible, appropriate structural policy measures by governments might be
needed.
Acknowledgements
We are grateful to U. Fritsche for his helpful remarks and to J.-O. Menz for his research assistance.
We also would like to thank two anonymous referees for their constructive comments. The paper has
benefited from comments made at the following conferences: the 23rd Annual Congress of European
Economic Association (EEA), Milan, Italy; the 7th Annual Meeting of the European Economics and
Finance Society (EEFS), Prague, Czech Republic; the 3rd Warsaw International Economic Meeting
(WIEM), Warsaw, Poland. All computations were performed using the DPD package for Ox, see
Doornik et al. (2006), and using the R language, see http://cran.r-project.org/.
References
Anderson, T.W., Hsiao, C., 1982. Estimation of dynamic panel models with error components. Journal of the American Statistical
Association 76, 74–82.
Arellano, M., Bond, S.R., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to
employment equations. Review of Economic Studies 58, 277–297.
Arellano, M., Bover, O., 1995. Another look at the instrumental-variable estimation of error-components models. Journal of
Econometrics 68, 29–52.
Baltagi, B.H., Griffin, J.M., Xiong, W., 2000. To pool or not to pool: homogeneous versus heterogeneous estimators applied to
cigarette demand. The Review of Economics and Statistics 82, 117–126.
Bean, L.H., 1946. International industrialisation and per capita income. Studies in Income and Wealth, 8. NBER, New York.
Bo
¨rsch-Supan, A., 2001. Aging and International Capital Flows. NBER Working Paper 8553, Cambridge, MA.
Bo
¨rsch-Supan, A., 2003a. Gesamtwirtschaftliche Folgen des demographischen Wandels. Institut fu
¨r Angewandte Wirtschafts-
forschung Report 31 (1), Tu
¨bingen.
Bo
¨rsch-Supan, A., 2003b. Labor market effects of population aging. Review of Labour Economics and Industrial Relations 17, 5–
44.
Bo
¨rsch-Supan, A., Du
¨zgu
¨n, I., Matthias, W., 2005. Altern und Produktivita
¨t: Zum Stand der Forschung. Mannheim Research
Institute for the Economics of Aging Working Paper 73, Mannheim.
Bo
¨rsch-Supan, A., Ludwig, A., Winter, J., 2006. Ageing, pension reform and capital flows: a multicountry simulation model.
Economica 73, 625–658.
Bo
¨s, D., von Weizsa
¨cker, R., 1988. Economic Consequences of an Aging Population. Discussion Paper Serie A 191. University of
Bonn, Germany.
Briotti, M.G., 2005. Economic reactions to public finance consolidation—A survey of the literature. European Central Bank
Occasional Paper Series 38, Frankfurt/Main.
Buslei, H., Schulz, E., Steiner, V., 2007. Effects of demographic change on private demand for goods and services in Germany until
2050. DIW Berlin Politikberatung kompakt 26, Berlin.
Chenery, H., 1960. Patterns of industrial growth. American Economic Review 50, 624–654.
Chenery, H.B., Taylor, L., 1968. Development patterns: among countries and over time. The Review of Economics and Statistics
50, 391–416.
Clark, C., 1957. The Conditions of Economic Progress, 3rd ed. Macmillan, London.
Denton, F.T., Spencer, B.G., 1998. Economic Costs of Population Aging. Independence and Economic Security of the Older
Population Research Papers 32. McMaster University, Hamilton, ON.
Fehr, H., Jokisch, S., Kotlikoff, L.J., 2008. Dynamic globalization and its potentially alarming prospects for low-wage workers.
NBER Working Paper 14527, Cambridge, MA.
Feyrer, J., 2007. Demographics and productivity. The Review of Economics and Statistics 89, 100–109.
Fuchs, V.R., 1968. The Serive Economy. NBER, Columbia University Press, New York.
Hilgerdt, F., 1945. Industrialisation and Foreign Trade. League of Nations, Geneva.
Kiviet, J.F., 1995. On bias, inconsistency and efficiency of some estimators in dynamic panel data models. Journal of
Econometrics 68, 53–78.
Kohl, R, O’Brien, P., 1998. The Macroeconomics of Ageing, Pensions and Savings: A Survey. OECD Economics Department
Working Papers 200, Paris.
Kuznets, S., 1956. Quantitative aspects of the economic growth of nations: I. Levels and variability of rates of growth. Economic
Development and Cultural Change 5, 1–94.
Lu
¨hrmann, M., 2005. Population aging and the demand for goods and services. MEA Discussion Paper Series 05095, Mannheim
Research Institute for the Economics of Aging (MEA), Mannheim.
McMorrow, K., Roeger, W., 1999. The economic consequences of ageing populations. European Economy – Economic Papers
138, Commission of the EC, Directorate-General for Economic and Financial Affairs (DG ECFIN), Brussels.
McMorrow, K., Roeger, W., 2003. Economic and financial market consequences of ageing populations. European Economy –
Economic Papers 182, Commission of the EC, Directorate-General for Economic and Financial Affairs (DG ECFIN), Brussels.
Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49, 1417–1426.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
259
Oliveira Martins, J., Gonand, F., Antolin, P., de la Maisonneuve, C., Yoo, K.-Y., 2005. The Impact of Ageing on Demand, Factor
Markets and Growth. OECD Economics Department Working Papers 420, Paris.
Phillips, P.C.B., Sul, D., 2007. Bias in dynamic panel estimation with fixed effects, incidental trends and cross section dependence.
Journal of Econometrics 127, 162–188.
Prskawetz, A, Fent, T., Guest, R., 2007. Workforce ageing and labour productivity. The role of supply and demand of labour in the
G7. In: Prskawetz, A., Bloom, D.E., Lutz, W. (Eds.), Population Aging, Human Capital Accumulation, and Productivity
Growth: A supplement to Population and Development Review. Population Council, New York, pp. 298–324.
Razin, A., Sadka, E., Swagel, P., 2002. The aging population and the size of the welfare state. Journal of Political Economy 110,
900–918.
Rowthorn, R., Ramaswamy, R., 1997. Deindustrialization—Its causes and implications. Staff Studies for the World Economic
Outlook. International Monetary Fund, Washington, DC, pp. 61–77.
Rowthorn, R., Ramaswamy, R., 1999. Growth, trade, and deindustrialization. IMF Staff Papers 46 (1), 18–41.
Rowthorn, R., Wells, J.R., 1987. De-industrialisation and Foreign Trade. Cambridge University Press, Cambridge.
Sachs, J.D., Shatz, H.J., 1994. Trade and jobs in U.S. manufacturing. Brookings Papers on Economic Activity 1, 1–84.
Sapozhnikov, M., Triest, R.K., 2007. Population aging, labor demand, and the structure of wages. Federal Reserve Bank of Boston
Working paper 07-8, Boston, MA.
Serow, W., Sly, D., 1988. Economic aspects of structural change in the older population of the United States: 1980–2020. Florida
State University Working Paper 46, Tallahassee, FL.
Shelton, C.A., 2008. The aging population and the size of the welfare state: is there a puzzle? Journal of Public Economics 92,
647–651.
Windmeijer, F., 2000. Moment conditions for fixed effects count data models with endogenous regressors. Economics Letters 68,
21–24.
B. Siliverstovs et al. / Economic Systems 35 (2011) 244–260
260
