Population Aging and Comparative Advantage
Jie Cai and Andrey Stoyanov
December 21, 2015
Abstract
In this paper we show that demographic di¤erences between countries are a source
of comparative advantage in international trade. Since many skills are age-dependent,
population aging decreases the relative supply and increases the relative price of skills
which depreciate with age. Thus, industries relying on skills in which younger workers
are relatively more e¢ cient will be more productive in countries with younger labor force
and less productive in countries with older populations. Building upon the neuroscience
and economics literature, we construct industry-level measures of intensities in various agedependent skills and show that population aging leads to specialization in industries which
use age-appreciating skills intensively and erodes comparative advantage in industries for
which age-depreciating skills are more important.
JEL Classi…cation codes: F14, F16, J11, J24
Key words: trade patterns; comparative advantage, population aging, cognitive skills
Cai: Shanghai University of Finance and Economics, School of Economics; email: [email protected]. Stoyanov: York University, Department of Economics, Faculty of Liberal Arts and Professional Studies; e-mail: [email protected]. We are grateful for valuable suggestions from seminar participants at the Canadian Economics
Association 2015 meetings, Goethe University Frankfurt, International Monetary Fund, Midwest Economic Theory and Trade 2014 conference, University of New South Wales, University of Toronto, and York University.
We also thank Matilde Bombardini, Ben Li, Daniel Tre‡er, Benjamin Sand, and Alan Woodland for helpful
discussions.
1
1
Introduction
Many countries observe major changes in demographics of their populations as aging leads to
a shift in the age structure of the labour force towards older workers. These changes are likely
to have a profound in‡uence on the structure of economic activity within countries and on the
pattern of trade between them through a change in the relative supply of age-dependent skills.
Recent research on aging suggests that there is a negative relationship between age and some
cognitive abilities, with a number of studies showing that cognitive decline begins as early as at
the age of 25. This implies that aging societies experience a more rapid decline in the quality and
the stock of certain cognitive skills and may thus lose comparative advantage in industries which
use those skills intensively. In this paper, we study how di¤erences in age structure of populations
between countries a¤ect global trade ‡ows. Speci…cally, we …nd a novel and empirically sizeable
source of comparative advantage: the relative supply and quality of age-dependent skills that
vary across countries with di¤erent demographic compositions.
This article links economics, with its focus on skills and productivity, and psychology, where
the idea of di¤erent skills changing di¤erently with age …rst originated. A series of studies on
cognitive abilities and aging consistently report that speech and language abilities improve with
age, while memory, multitasking, and the speed of information processing decline with age. Furthermore, a decline in physical strength with age is well documented in the medical literature.
Knowing the importance of those skills for various occupations and the composition of occupations across industries, we are able to pin down industry-level demand for each age-dependent
skill. For instance, among the occupations which rely heavily on age-depreciating cognitive skills
are various types of machine operators, where coordination, divided attention, and perceptual
speed are very important. As a result, in industries where most workers enter as machine setters
and operators, such as yarn mills or wood product manufacturing, age-depreciating skills are used
intensively. Other industries, such as printing or beverages and tobacco, employ many workers in
occupations which require good written and/or oral communications skills (e.g. technical writers
or sales representatives) and use age-appreciating cognitive skills intensively.
One such mechanism operates through the e¤ect of population aging on the stocks of age2
dependent skills. If individuals are endowed with certain stocks of age-dependent skills which
they supply to the market, then a country with older labor force, all else being equal, will have
lower endowments and higher relative prices of age-depreciating skills and thus specialize in the
production of goods which utilize age-appreciating skills intensively. The second mechanism
through which demographics operate is the e¤ect on labor productivity. If aging a¤ects the
quality rather than the quantity of age-dependent skills, then older workers become less productive in tasks which require age-depreciating skills, and the age composition of the labor force
would determine relative productivity of an industry. In the presence of labor market frictions
which prevent perfect sorting of workers to industries based on their age, industries would inherit national age distribution, in which case population aging will shift the age distribution of
employees in all sectors and increase (decrease) labor productivity in industries which rely on
age-appreciating (age-depreciating) skills.
Both the Heckscher-Ohlin and the Ricardian mechanisms imply that countries with older
(younger) populations will specialize in industries which use age-appreciating (age-depreciating)
skills intensively. Figures 1 to 3 illustrate this prediction and plot the relationship between export
shares and skill intensities across industries for countries with the median age below 20 (young)
and above 35 (old) in the year 2000. As we expect, countries with older populations export more
in industries which use age-appreciating cognitive skills intensively, and less in industries with
a greater dependence on physical and age-depreciating cognitive skills. The opposite pattern is
observed among countries with younger populations, which specialize more on commodities that
rely on age-depreciating skills.
Further, if demographics a¤ect productivity and return on age-dependent skills, industries
which use age-depreciating skills intensively should have disproportionately smaller share of older
workers. As workers are getting older, one would expect them to leave industries which are
intensive in age-depreciating skills, as workers’productivity in those industries decline and return
to their skills fall. Figure 4 plots the relationship between the median age of the labor force in
eighty-three four-digit NAICS industries and intensity in three age-dependent skills for the US in
2000.1 As we expect, the relationship between median worker’s age and skill intensity is positive
1
The data on age composition of di¤erent industries is obtained from the U.S. Census Public Use MicroSamples.
3
for age-appreciating cognitive skills and negative for age-depreciating cognitive and physical
skills.
The relationship between skill development and aging allows us to analyze the e¤ect of unobservable endowments of cognitive skills on trade ‡ows by using observable cross-country variation
in demographic composition as a proxy. Surveying behavioral and neuroscience literature, we
identify cognitive skills which are known to change over the course of an individual’s life. To
measure sectorial intensities in those skills, we retrieve the indicators of their importance for
di¤erent worker occupations from the O*NET database. We then use occupational composition
for each 4-digit NAICS industry, obtained from the US Bureau of Labor Statistics, to construct
the weighed-average measure of importance of each age-dependent skill for every 4-digit NAICS
industry. Since many of these skill importance variables are highly correlated and their impacts
cannot be separately identi…ed with our data, we group all age-dependent skills in three main
categories using the principal component analysis: physical abilities, age-appreciating cognitive
skills, and age-depreciating cognitive skills.
We con…rm the main theoretical prediction about the e¤ect of aging on comparative advantage
using rich bilateral trade data which include 86 industries and cover the time period between
1962 and 2010. We …nd that countries with older (younger) populations capture larger shares of
world trade in commodities that more intensively use age-appreciating (age-depreciating) skills.
In the baseline regressions that include 136 exporting countries, we show that the interaction
of a country’s median age and the industry’s intensity in age-dependent skills is an important
determinant of bilateral trade ‡ows, both statistically and economically.2 This …nding is robust
to the inclusion of the standard determinants of comparative advantage such as endowments of
physical and human capitals, as well as institutional factors such as …nancial development and
the quality of the legal system. Moreover, age-appreciating and age-depreciating cognitive skills
often explain more variation in trade ‡ows than physical and human capitals. Furthermore, the
magnitude of the impact of age di¤erences on trade ‡ows, estimated in this study, is comparable
to institutional determinants of comparative advantage identi…ed in recent literatures, such as
product market institutions (Nunn, 2007; Costinot, 2009), …nancial market institutions (Manova,
2
Similar results are obtained when a country’s endowment of age-dependent skills is proxied by the share of
workers below a certain age threshold in the labor force.
4
2008), and labor market institutions (Cuñat and Melitz, 2012). The main …nding of this paper
is robust to di¤erent de…nitions of human capital and holds for di¤erent time periods.
Availability of historical demographic data allows us to test the dynamic predictions of the
model. Speci…cally, we would expect fast(slow)-aging countries to observe a decrease in the
relative price of age-appreciating (age-depreciating) skills and an increase in relative productivity
of industries which use those skills intensively, thus making them more competitive in the global
market. Analyzing the e¤ect of population aging on changes in comparative advantage allows
to address many omitted variable concerns in the estimation, and in particular for the e¤ect of
institutional factors which do not vary over time. We also pay close attention to other sources
of endogeneity in the empirical model and explore robustness of our results using instrumental
variable approach and alternative measures of countries’e¤ective endowment in age-dependent
skills.
We …nd substantial support for the dynamic predictions of the model in the data. We
demonstrate that an increase in the median age between 1962 and 2000 is associated with a
shift in a country’s production and export structure towards commodities that more intensively
use age-appreciating skills and away from commodities that rely more on age-depreciating skills.
This result implies that although population aging leads to a reduction in premia for skills that
are inherent to older workers, the problem can be alleviated by an increase in demand for those
skills through expansion in the production and exports of goods which use them intensively.
This paper contributes to the fast-growing literature that formally tests the relationship
between factor proportions and trade ‡ows. Speci…cally, it is related to the classical works
documenting the important role of physical and human capital endowments in comparative advantage.3 More recent developments in this literature emphasize non-traditional sources of comparative advantage, such as the cross-country variations in contract enforcement (Levchenko,
2007; Nunn, 2007), the quality of …nancial systems (Beck, 2003; Manova, 2008), the extent of
labor market frictions (Helpman and Itskhoki, 2010; Cuñat and Melitz, 2012; Tang, 2012), skill
dispersion (Bombardini, Gallipoli and Pupato, 2012), and water resources (Debeare, 2014). Our
3
See Tre‡er (1993), Harrigan (1997), Davis and Weinstein (2001), and Debeare (2003) for the empirical evidence
based on the factor content of trade analysis. Romalis (2004) and Blum (2010) analyze the role of capital and
skilled labor in specialization and comparative advantage using commodity trade and output data, respectively.
5
study contributes to this literature by proposing a novel factor of comparative advantage stemming from the di¤erences in endowments of cognitive and physical skills between countries. Using
the variation in demographic composition across countries and the variation in age-dependence
of di¤erent cognitive skills, we are able to construct a reliable proxy for a country’s e¤ective endowment in unobservable cognitive and physical skills. We thus demonstrate that demographics
can a¤ect cross-country di¤erences in e¤ective endowments of cognitive and physical skills and
is an equally important determinant of comparative advantage as are the di¤erences in physical
and human capitals.
The only study on the role of cognitive skills in trade we are aware of is by Wol¤ (2003), who
estimates the content of cognitive and physical skills in the US exports. Wol¤ demonstrates that
the US has comparative advantage in cognitive and interactive skills and disadvantage in motor
skills. However, without reliable data on factor endowments in other countries, these results are
hard to interpret as an empirical test of the role of cognitive skills in the Heckscher-Ohlin model.
Our model, in contrast, provides a theoretical underpinning for the Wol¤’s …nding on interactive
and motor skills. Since interactive skills improve with age while motor skills deteriorate, a country
with relatively old population, such as the US, must be abundant in interactive and scarce in
physical skills. Therefore, the US demographic composition can explain why this country has a
positive factor content of trade for the communication and negative for the physical skills. We
also show, using more comprehensive data, that failure to distinguish various cognitive skills by
their age-dependence may confound the e¤ect of demographics on comparative advantage.
Our research also relates to the literature on macroeconomic impacts of population aging.
Papers studying the role of aging in international capital and immigration ‡ows using an overlapping generations framework4 conclude that more rapid population aging in Northern countries
increases the rate of capital accumulation, stimulating capital ‡ow to Southern countries where
the return on capital is higher.5 Higgins (1997) and Narciso (2013) empirically investigate this
4
Holzmann (2002), Boersch-Supan, Ludwig, and Winter (2001), Boersch-Supan, Ludwig, and Winter (2006),
Domeij and Floden (2006), Attanasio, Kitao, and Violante (2007), Ludwig, Krueger, and Boersch-Supan (2007),
among others.
5
Ludwig and Vogel (2010) and Ludwig, Schelkle, and Vogel (2012) point out that intensi…ed accumulation
of human capital in the North can mitigate the decline in the marginal return to capital and slow down capital
out‡ow.
6
prediction and con…rm that demographic structure has signi…cant e¤ect on capital ‡ows. In the
context of international trade, Helliwell (2004) shows with a theoretical model that demographic
changes are associated with intensi…ed outsourcing of labor-intensive processes to countries with
younger populations and thus a¤ect a country’s comparative advantage. Our theoretical framework is also based on the e¤ect of aging on trade through changes in relative factor prices but
goes beyond the two factor model and introduces multiple age-dependent factors of production.
The paper is organized as follows. Section 2 discusses the empirical strategy for testing the
main predictions of the theoretical model. Section 3 describes the data and Sections 4 presents the
baseline empirical results, along with extensions, robustness checks, and solutions to endogeneity
problems. Section 5 shows the e¤ect of changes in age composition of a country’s laborforce
on comparative advantage. Section 6 concludes. We leave the detailed discussion of theoretical
models to Appendix A.
2
Theoretical background and empirical methodology
In the presence of age-dependent skills, demographic composition can a¤ect a country’s comparative advantage through two di¤erent channels. First, there is a Heckscher-Ohlin channel whereby
population aging can a¤ect the stocks and relative supply of age-dependent skills. When certain skills change over the course of an individual’s life, the stocks of those skills will vary
across individuals and hence across countries with the age structure of their populations. The
Heckscher-Ohlin model then implies that with cross-industry variation in skill intensities, the
age structure of a country becomes a source of comparative advantage. For example, a country with young population will have a comparative advantage in products which intensively use
age-depreciating skills. In this case, comparative advantage stems from cross-country variation
in skill premia.
Second, demographics can a¤ect comparative advantage through the Ricardian channel.
Workers of di¤erent ages may not be equally productive in tasks that require age-dependent
skills, which may a¤ect relative labor productivities of industries with di¤erent composition of
7
tasks. As long as industry-speci…c skills or other labor market frictions prevent workers of di¤erent ages from moving freely from one industry to another, the workers’age distribution in every
industry will resemble the country’s distribution. In this case, population aging will shift the age
distribution of employees in all sectors and increase (decrease) labor productivity in industries
which rely on age-appreciating (age-depreciating) skills. Therefore, countries with older populations will have higher Ricardian productivities and stronger comparative advantage in industries
which are intensive in age-appreciating skills.
In Appendix A we introduce age-dependent skills into the theoretical model of Ricardian
and Heckscher-Ohlin comparative advantage by Chor (2010). In that extension, population
aging a¤ects country’s comparative advantage both through a reduction in relative supply and
increase in relative price of age-depreciating skills (Heckscher-Ohlin channel) and through a direct
e¤ect on labor productivity in tasks that rely on age-dependent skills (Ricardian channel). The
empirical speci…cation implied by the model is similar to Chor (2010) and Bombardini, Gallipoli,
and Pupato (2012):
ln Xcpi =
X
k
k Ii
k2K
Agec +
X
f
f Ii
Fcf +
0
cp
+
c
+
pi
+ "cpi ;
(1)
f 2F
where Xcpi is exports from country c to country p in industry i, K is the set of age-dependent
skills and F is the set of other factors of production, Iik is the intensity of industry i in factor of
production k, Agec is the demographic structure in country c assumed to be increasing as population is getting older, Fcf is endowment of country c in factor f , and
Coe¢ cients
f
cp
is bilateral trade costs.
re‡ect the importance of conventional determinants of comparative advantage,
such as human and physical capital endowments.
In (1), the coe¢ cients
k
on interactions Iik
Agec capture both the Ricardian and the
Heckscher-Ohlin channels. Separating the two e¤ects is neither critical nor feasible for this study.
First, the estimates of model (1) are valid for the purpose of evaluating the e¤ect of population
aging on trade ‡ows and comparative advantage. Whether this e¤ect operates through changes
in skill premia or labor productivity is irrelevant for the main …nding of this paper that aging
countries experience structural changes in their production and trade patters away from industries
8
which rely on age-depreciating skills. Second, isolating one e¤ect from the other would require
di¤erent data and empirical strategy, which is outside the scope of this paper, and we thus leave
this questions for future research.6
The interaction Iik
Agec is the main variable of interest and the sign of
k
allows us to test
the key theoretical prediction that younger countries have a comparative advantage in industries
which intensively use age-depreciating skills. The theoretical model predicts that
skills which worsen with age and
k
k
< 0 for
> 0 for skills that improve with age. Furthermore, equation
(1) implies that for any pair of countries c1 and c2 exporting goods i and j to a third country p
the following holds
E ln
Xc1 pi
Xc2 pi
Xc1 pj
Xc2 pj
ln
=
X
k
Iik
Ijk
(Agec1
Agec2 )
(2)
k2K
If country c1 has younger population than country c2 , (Agec1
is more intensive in skill k than industry j, Iik
Agec2 ) < 0, and industry i
Ijk > 0, then we would expect country c1 to
export relatively more (less) of good i than j if skill k depreciates (appreciates) with age, which
would be the case when
k
<0(
k
> 0).
In our baseline speci…cations we control for two standard Heckscher-Ohlin factors of comparative advantage – the cross-country di¤erences in physical capital and skilled labor. Given
that countries export more in industries which use their abundant factors intensively, we expect
f
> 0 for all standard factors of production. The vector
cp
in equation (1) captures bilateral
trade frictions between countries c and p. Exporter …xed e¤ects
c
control for exporter’s ag-
gregate productivity level, size, remoteness from other countries, and other characteristics that
do not vary across industries. Importer-industry …xed e¤ects
pi
control for product prices in
the importing country and all other demand shifters, including those which may be driven by
cross-country demographic di¤erences.
6
A speci…c feature of the setup with multiple age-dependent skills is that the workers arrive with a bundle
of skills. Ohnsorge and Tre‡er (2007) demonstrate that in a setting with multiple skills embedded in workers, a
country’s comparative advantage is determined not only by its relative skill endowments but also by the second
moments of skill distributions. In Appendix A2 we use the Ohnsorge and Tre‡er (2007) framework to show that in
the presence of multiple skills the pattern of comparative advantage remains consistent with the Heckscher-Ohlin
theorem for the plausible range of the parameters on the second moments of the multivariate distribution of skill
endowments, and hence empirical methodology employed in the current study remains valid.
9
There are two potential endogeneity concerns with Iik
Agec variables in equation (1). The
…rst one relates to the demographic composition of a country’s population. A country’s median
age is predetermined relative to industry-level trade ‡ows, and it is di¢ cult to think of other
reasons for why the median age could a¤ect the export structure other than through the e¤ect on
either supply or demand.7 Yet it may be related to other unobservable countries’determinants
of comparative advantage which may have di¤erential impact on productivities in industries with
di¤erent skill intensities. In Section 4.3 we discuss how we address potential endogeneity issues
with demographic composition using instrumental variable approach and alternative measures
of e¤ective endowment of age-dependent skills. Second, the skill intensity measures constructed
from the occupational structures in the US industries are plausibly exogenous, as long as the
US employment composition is una¤ected by bilateral trade ‡ows between other countries. In
Section 4.2 we provide some evidence in support of this assumption. In particular, if there
is a feedback from trade ‡ows to employment structure, the simultaneity would especially be
a problem for the US trade ‡ows. However, removing the US from the set of importing and
exporting countries does not a¤ect our results. Moreover, skill intensities, constructed with 2010
data, predict trade ‡ows in 1970 just as well as in 2010, suggesting that our results are unlikely
to be subject to the reverse causality.
It is also important to emphasize that while the focus of this study is on the e¤ect of population
aging on supply, the demographic composition, in principle, can also a¤ect trade through the
e¤ect on demand if consumer preferences change with age. We explore this possibility in Section
4.2 and …nd that although consumption behavior does change with age, there is little evidence
that age-related changes in preferences are systematically related to production technology or
skill intensities in manufacturing industries.
To test the prediction regarding the e¤ect of demographic transformations on trade ‡ows, we
7
Galor and Mountford (2008) …nd that trade openness, measured by the ratio of total trade ‡ow over GDP,
may have di¤erential e¤ect on the fertility rate and investment in human capital in developed and developing
countries. However, this potential feedback from trade openness to demographic composition is not a concern for
us because our focus is not at the level of openness but at the share of trade across industries for a given level of
openness.
10
estimate equation (1) in di¤erences:
ln Xcpi =
X
k2K
where
k
k Ii
Agec +
X
f
f Ii
Fcf +
0
cp
+
c
+
pi
+ "cpi
(3)
f 2F
is the time-di¤erence operator. Equation (3) thus assumes that while trade structure,
age composition, and factor stocks can change within a country over time, industries’factor intensities are constant. Thus, rapidly aging countries should lose comparative advantage in industries
which rely on age-depreciating skills and specialize in industries which use age-appreciating skills
intensively. Therefore,
k
are expected to be positive for age-appreciating and negative for age-
depreciating skills. For physical capital and skilled labour the Rybczynski prediction implies
f
> 0.
The advantage of adding a panel dimension to our data structure is that it allows to address
additional omitted variable concerns in the estimation. In particular, time di¤erencing controls
for time-invariant exporter-industry characteristics, such as the e¤ect of some institutions on
industry-level productivity. Many institutional factors which previous literature identi…ed as
important determinants of comparative advantage vary little over time for a vast majority of
countries in our sample and cannot be studies in a panel data analysis. Time di¤erencing
transformation will also eliminate the e¤ect of any geographic characteristics on productivity of
di¤erent industries, such as proximity to main markets or to natural resources.
3
Data
Estimation of the main model (1) requires four sets of data: industry-level data on bilateral trade
‡ows; determinants of bilateral trade costs; industries’ intensities in age-dependent skills and
other factors of production; and country-level measures of abundance in those factors. Equation
(1) is estimated with bilateral trade data for the year 2000. To estimate equation (3), we employ
the change in trade structure, age composition, and factor endowments between 1962 and 2000.
In what follows we describe the data sources for this study and discuss the issues with construction
of the key variables.
11
Trade data. The data on industry-level bilateral trade ‡ows for estimation of equation (1)
are obtained from the UN-TRAINS database at 6-digit Harmonized System classi…cation and
aggregated into 4-digit North American Industry Classi…cation System (NAICS) using concordance from Feenstra, Romalis, and Schott (2002). The resulting data is an unbalanced panel
of 235 exporters, 159 importers, and 85 industries for the year 2000. Changes in bilateral trade
‡ows between 1962 and 2000, which we use to estimate the Rybczynski e¤ect, are obtained
from NBER-UN International Trade Database with the NBER concordance tables employed to
convert 4-digit ISIC data into 4-digit NAICS. Equation (3) is estimated for 80 exporters, 135
importers, and 76 industries.
Bilateral trade costs are controlled for with the standard set of geographical and institutional
variables used in the gravity model literature. The vector
cp
in equations (1) and (3) includes the
log of distance, de…ned as the distance between the major cities of the two countries, common
land border indicator, common o¢ cial language binary variable, colonial ties binary variable
(separately for before and after 1945), and a binary variable taking the value of one if importer and
exporter were ever part of one country.8 We also use two binary variables, which we constructed
from the WTO database on Regional Trade Agreements, for the presence of a free trade agreement
or a customs union between a pair of countries.
Intensities in cognitive skills and physical abilities. Estimating the e¤ect of age-dependent
skills on trade ‡ows is the main focus of this paper, and in what follows we provide a detailed discussion of how the industry-level measures of intensities in age-dependent skills were constructed.
Appendix B describes two categories of age-dependent skills –cognitive and physical –and
reviews the literature that analyses the evolution of these skills over the course of an individual’s
life. For cognitive skills, studies in neuropsychology literature provide convincing evidence that
speech and language abilities improve with age, while memory, divided attention,9 and the speed
of information processing deteriorate with age. The negative impact of aging on nearly all
aspects of physical and psychomotor skills – such as muscular strength, stamina, coordination,
8
All of these variables were obtained from the Centre d’Etudes Prospectives et d’Informations Internationales
(CEPII).
9
Divided attention is the ability to process information from two or more sources at the same time or to switch
from one task to another.
12
and dexterity –is well documented in the medical literature.
To construct industry-level measures of intensity in cognitive skills and physical abilities,10 we
use information on occupational composition for every industry, obtained from the US Bureau of
Labor Statistics. Occupational employment shares were matched though a common occupational
classi…cation (7-digit Standard Occupational Classi…cation) to the information on the importance
of di¤erent skills and abilities across occupations, retrieved from the Occupational Information
Network (O*NET) database.11 Using occupational employment shares as weights, we construct
an industry-level measure of intensity in a particular skill as a weighted average of the importance
of that skill across occupations within an industry. Therefore, for a given skill, the variation in
intensity of its use across industries comes from the di¤erences in occupational composition
between industries. At the same time, the within-industry variations in intensities of di¤erent
skills comes from the variation in the importance of those skills between occupations.
To quantify the importance of cognitive skills and physical abilities for di¤erent occupations,
we use the O*NET database. O*NET ranks all occupations in several dimensions which are
closely related to the age-depreciating skills described in the Appendix B. The importance of
speech and language abilities between occupations is captured by the following four skill indicators from the O*NET database: oral comprehension; oral expression; written comprehension;
and written expression. The intensities in memory and divided attention are constructed from
the O*NET indicators of importance of memorization and time sharing, respectively. The speed
of information processing is captured by the indicators on perceptual speed and speed of closure.
Averaging these indicators across occupations within industries, we obtain eight measures of
industry-level intensities in cognitive skills. Given the high degree of correlation between cognitive skills indicators (see Tables 1A and 2A in the Appendix), we group four language indicators
into a single indicator for age-appreciating cognitive skills, cog_appi , using the principle component analysis (PCA)12 . Similarly, the four indicators associated with age-declining cognitive
skills were also grouped into one measure for age-depreciating cognitive skills, cog_depi . Panels
A and B of Table A3 report the results of the PCA. The third column shows the total variance
10
Appendix C describes the construction of skill intensity variables in more details.
Several recent studies analyze the role of various skill measures from O*NET in international context, see for
example Oldenski (2012) and Bombardini, Gallipoli, and Pupato (2012).
12
See Jolli¤e (2002).
11
13
accounted for by each factor and the last column reports factor loadings.
For physical abilities, we construct nine industry-level measures of skill intensities based
on the O*NET questions capturing the importance of dynamic ‡exibility, dynamic strength,
explosive strength, extent ‡exibility, gross body coordination, gross body equilibrium, stamina,
static strength, and trunk strength. As with the cognitive skills, all nine indicators were combined
into a single indicator for physical abilities, physicali , through the PCA.
The industry-level intensity in age-dependent skills is constructed relative to a reference
cognitive skill which choice is determined by two considerations. First, it should not be a¤ected
by aging, otherwise it could confound the e¤ect of the age-dependent variables. Second, because
cog_appi and cog_depi are highly correlated, most likely due to high degree of complementarity
between the two groups of cognitive skills, the reference skill should also be positively correlated
with age-dependent skills in order to break multicollinearity between cog_appi and cog_depi . We
choose inductive reasoning as the benchmark reference factor, and in Section 4.2 we demonstrate
the robustness of our results to alternative reference skills that satisfy the above criteria.
Table 1 lists the ten most and the ten least intensive occupations for physical and both types
of cognitive skills. Many of the occupations that are the most intensive in age-appreciating skills
are related to sales where oral and written communication skills are critical. The top of the
list for age-depreciating skills is dominated by various machine setters and operators, for whom
coordination, divided attention, and perceptual speed are the most important. It is important to
note that many of the least age-depreciation intensive occupations are high-skill occupations that
are not intensive in physical skills. This results in strong positive correlation between cog_depi
and physicali measures and negative correlation with human capital measures (see Table 2).
Therefore, multicollinearity between factor intensities remains a problem and will be a concern
in the empirical analysis.
Other data. Our primary measure of a country’s age structure is the median age, obtained
from the United Nations.13 As an alternative, we also use the share of young workers in the labour
force, constructed as a fraction of 20-40 year-olds in the 20-to-65 age group. The information
13
See Table 5A in the Appendix for information on median age in 2000 and change in median age between 1962
and 2000 for all exporting countries in our sample.
14
on the age structure of population comes from the World Development Indicators database,
maintained by the World Bank.
Industry-level measures of intensities in skilled labor and physical capital are derived from the
US Census of Manufacturers for 1998.14 Capital-intensity is constructed as the ratio of capital
stock over total employment, and skill intensity as the share of non-production workers in total
employment.
Data on country’s stock of physical capital, measured in 2005 prices, is retrieved from the
Penn World Table. Human capital stock is obtained from Barro and Lee (2013) and is measured
as a share of population with secondary and tertiary education.15 The full sample includes 136
exporting countries, 155 importing countries, and 83 industries.
4
4.1
Results
Baseline results
Table 3 reports estimation results for equation (1) with the country’s median age used as a proxy
for the stocks of age-depreciating skills. For comparability, all tables report standardized coe¢ cients. The …rst column con…rms the main prediction of the Heckscher-Ohlin model for capital
and skilled labor: countries that are abundant in capital and skilled labor export more in industries which use those factors intensively. Adding Iik
Agec interactions to the main speci…cation
in columns (2) to (5), we …nd that all coe¢ cients are consistent with the theoretical model and
are statistically signi…cant,16 thus supporting the hypothesis that age di¤erences across countries
are the source of comparative advantage in international trade. The estimates in columns (2)-(5)
reveal that older countries export more in industries which use age-appreciating cognitive skills
intensively and less in industries which are intensive in physical and age-depreciating cognitive
14
Under the assumption of no factor intensity reversals, the ranking of factor intensities across industries does
not vary by country.
15
Our three alternative measures are the average years of schooling attained, the share of workers with at least
primary education, and the share of workers having completed tertiary schooling.
16
The estimates remain statistically signi…cant if standard errors are clustered by exporter.
15
skills.
Columns (6) of Table 3 reports results for a complete speci…cation with all skill measures
included in the regression. This extension does not substantially a¤ect the coe¢ cient estimates
for the two types of cognitive skills, however, the coe¢ cients on physical skills become insignificant. Furthermore, industry-level intensities in physical skills and age-depreciating cognitive
skills are highly correlated (see Table 2), and the interaction (cog_depi
Agec ) picks up the
e¤ect of physical skills.
Because the reported coe¢ cients in Table 3 are standardized, we can directly compare the
magnitudes of the e¤ect of di¤erent factors of production on trade using the estimates for equation
(2). Suppose industry i has one standard deviation higher intensity in all factors of production.
Then, focusing on the most complete speci…cation in column (6), a country which has one
standard deviation higher median age than another one will export 11% more in industries which
are intensive in age-appreciating skills, 15:2% less in industries which use age-depreciating skills
intensively, and 6:2% and 5:2% more in capital and labor intensive industries, respectively.1718
In Table 4 we report the estimates of equation (1) with the age composition being measured
by the share of young workers in the labour force, which we de…ne as the fraction of 20-40 yearolds in the 20-to-65 age group. The advantage of this measure over the median age is that it
represents the age structure of the working-age population only. At the same time, since we
do not know the exact onset of the age-related cognitive decline, the age threshold of 40 in
the de…nition of young workers is somewhat ad hoc.19 The estimates in Table 4 are similar in
magnitude to those obtained with the median age (note that since the share of young workers
is inversely related to the population’s median age, the coe¢ cient estimates in Tables 3 and
17
With the standard deviation of log exports being equal to 3:36, the di¤erence in exports of k-factor intensive
industry between the two countries is exp (3:36 k ). The standard deviation of the median age is equal to 7:4 in
our sample.
18
It is not unusual in the literature to …nd the e¤ect of various institutional factors of trade to be more important
than the e¤ect of capital and skilled labor (see for example Bombardini et al, 2012, and Nunn, 2007). One possible
explanation is the measurement problem inherent to both variables: the de…nitions of physical capital vary a lot
across countries and the number of years of schooling does not re‡ect cross-country di¤erences in the quality of
education systems.
19
This threshold level is motivated by studies on aging and cognition, surveyed in Appendix B, which show
that cognitive decline begins at around 30 and after 40 reaches the level of 20 year olds. Changing the threshold
to 35 or 45 does not alter our main estimation results.
16
4 are of opposite signs). In column (6) we include the interaction of industries’ skill intensity
and countries’abundance in skilled young workers (the share of young workers with secondary
and tertiary education) to control for the di¤erences in human capital of young workers across
countries. The coe¢ cients of interest remain broadly the same as before.
4.2
Robustness tests
In Tables 5 and 6 we present several extensions of the main speci…cation and explore the robustness of our results to changes in econometric speci…cation and de…nitions of the key variables.
Alternative measures of human capital. In columns (1) to (3) of Table 5 we use three alternative de…nitions of human capital to measure endowments in high and low skilled workers. In
column (1) we construct a country’s abundance in skilled labor as the average number of years
of schooling, in column (2) as the share of population with completed tertiary education, and in
column (3) the endowment is measured with the human capital index based obtained from the
Penn World tables. These modi…cations leave the coe¢ cients of interest nearly unchanged.
Alternative reference skills. In Section 3 we emphasize the need to normalize age-dependents
skill intensities by a reference age-neutral skill. In the benchmark speci…cation “inductive reasoning”was used as a reference skill. In columns (4) to (6) of Table 5 we demonstrate robustness
of the main results to using alternative reference skills and report regression estimates when all
age-dependent skill intensities are normalized by O*Net measures of “deductive reasoning”, “‡uency of ideas”, and “information ordering”. The main coe¢ cients of interest preserve expected
signs and remain statistically signi…cant at least at 90% con…dence level.
Alternative occupational weights. We use occupational employment shares as weights to construct industry-level measure of intensity in a particular skill as a weighted average of importance
of that skill across occupations within an industry. For the benchmark speci…cation, the data on
occupational composition come from the US Bureau of Labor Statistics for the year 2012. A potential concern is that these occupational shares could re‡ect industries’adjustments to changes
in relative supply of age-dependent skills caused by accelerated population aging in the US dur-
17
ing the last two decades. This would result in endogeneity if industries di¤er in their ability
to adjust to demographic changes. We address this concern by reconstructing factor intensities
using occupational shares from 1980 and 1990 US Census years.
A major issue with using older census data is having consistent occupational categories over
time so that the share of employment of each category in an industry could be matched to the
O*Net measure of importance of age-dependent skills. We use occupational coding scheme by
Firpo, Fortin, and Lemieux (2011), who crosswalk O*Net occupation classi…cation into 1980 and
1990 Census occupation classi…cation (COC). However, the limitation of this approach is that
COC data cannot be precisely mapped to O*Net classi…cation, introducing measurement error
in factor intensities measures.20 Industry Classi…cation codes used in earlier Census data were
converted into 1987 SIC and then into 1997 NAICS using concordance tables provided by the
US Census Bureau.21
Columns (7) and (8) present the estimation results for equation (1) when industry-level intensity in age-dependent skills are constructed with occupational composition for 1980 and 1990,
respectively. For comparison, column (9) reports the estimates for the benchmark speci…cation using the same sample as in columns (7) and (8). Consistently with previous …ndings,
the coe¢ cient on age-depreciating cognitive skills is negative and statistically signi…cant, and
the magnitude is comparable to the estimates from the benchmark speci…cation. Furthermore,
the coe¢ cient estimate on physical skills also becomes negative and signi…cant. By contrast,
the coe¢ cient on the interaction of median age with age-appreciating cognitive skills become insigni…cant. However, it is hard to say whether the di¤erence in the results is driven by di¤erences
in occupational compositions across census years or by the measurement error.
Controlling for bilateral trade costs. In column (1) of Table 6 we estimate equation (1)
with importer-product and exporter-importer …xed e¤ects. The latter is used to account for
20
The signi…cance of the measurement error caused by di¤erent occupational classi…cations becomes apparent
when factor intensities constructed with employment shares from di¤erent census years are compared with each
other. The coe¢ cient of correlation in skills intensities across industries in 1980 and 1990, which are based on
a very similar COC classi…cations that can be easily mapped one-to-one, is 0.96-0.97. However, the correlation
with the measure based on O*Net occupational classi…cation in 2012 is 0.61 for age-appreciating and 0.74 for
age-depreciating cognitive skills.
21
Because Census industry classi…cation is more aggregated than 4-digit NAICS, we exclude industries that
cannot be precisely mapped to the broader NAICS classi…cation.
18
unobserved country-pair heterogeneity that does not vary across industries and is not captured
by distance and other controls for bilateral trade costs. While this extension of the model
substantially improves the …t to the data, it does not materially a¤ect the coe¢ cients of interest.
Zero trade ‡ows. A potential problem with estimating equation (1) by the OLS is that
it discards observations with zero trade ‡ows, which constitute about two-thirds of the data.
Excluding those observations from the sample can result in systematically biased OLS coe¢ cients
(Santos Silva and Tenreyro, 2006; Helpman et al, 2008). We address this problem by using a twostep procedure to correct for sample selection as in Helpman, Melitz, and Rubinstein (2008).22
The second stage results, reported in Column (2) of Table 6, do not suggest that sample selection
a¤ects the OLS estimates as all coe¢ cients remain unchanged.
Results for other time periods. In columns (3) and (4) we demonstrate that the results are not
con…ned to a particular time period by estimating equation (1) using trade and median countries’
age data for the years 2010 and 1970, respectively. The coe¢ cients on cognitive skills remain
qualitatively unchanged. An di¤erence with the benchmark results is the insigni…cant coe¢ cient
on physical capital in both time periods.23
Results without the USA. The result that skill-intensities, constructed with the US data for
2000, can equally well predict bilateral trade ‡ows in 1970 and in 2010 suggests that the reverse
causality from trade to occupational composition across industries is unlikely to be a problem.
Furthermore, if reverse causality were present, it would be stronger for the US trade data.
However, excluding the US from the sample (column 5 of Table 6) produces virtually identical
results to those in the benchmark speci…cation, which corroborates the conjecture that our main
results are not subject to simultaneity bias.
Other determinants of comparative advantage. Recent studies on comparative advantage have
identi…ed a number of institutional determinants of trade ‡ows. In what follows we consider
22
As in Helpman, Melitz, and Rubinstein (2008), we use the interactions of market entry regulation costs in
importing and exporting countries to identify variation in the extensive export margin at the …rst stage of the
estimation procedure.
23
The cross-section regressions of equation (1) for the years 1962, 1980, and 1990, not presented in the paper
but available upon request, produce similar results: the coe¢ cients on (cog_appi Agec ) and (cog_depi Agec )
variables are always positive and negative, respectively, and are statistically signi…cant.
19
four such factors and show that our results are robust to inclusion of these factors. The …rst
one is the level of a country’s …nancial development interacted with industry’s dependence on
external …nance. Manova (2008) proposed that more …nancially developed countries have a
comparative advantage in industries that rely heavily on external …nancing. The second control
is the interaction of a country’s ability to enforce contracts and industry-level measure of contract
intensity (Nunn, 2007) to capture the holdup problem that may undermine productivity in
industries in which relationship-speci…c investments are important. The third measure is the
strength of a country’s legal system interacted against a measure of industry’s task complexity.
Costinot (2009) argue that countries where …rms are better able to monitor workers will gain
more from task specialization and thus specialize in industries which require many tasks. Lastly,
following Cunat and Melitz (2012), we include the interaction of the ‡exibility of a country’s labor
markets and industry’s output volatility. The data on all four interactions is from Chor (2010)24
and is recorder at 2-digit SIC-87. We use US Census Bureau concordance to convert our measures
of skill-intensities into that format, which results in a reduction in the number of industries from
85 to 20.25 The estimated coe¢ cients on all four institutional factors are positive and statistically
signi…cant, in line with previous studies (Column 6 of Table 6). More importantly, the signs and
statistical signi…cance of cognitive age-dependent skills remain unchanged.
4.3
Addressing endogeneity of the median age
The OLS estimates of (1) provide consistent estimates only if age composition is independent of
other country characteristics which may have a di¤erential e¤ect on productivity in industries
depending on their intensities in age-dependent skills. While this may be the case, in this section
we explore two additional strategies to identify the e¤ect of age composition on comparative advantage using the instrumental variable approach and alternative measure of a country’s e¤ective
endowment in age-dependent skills.
24
We thank Davin Chor for sharing the data with us.
We exclude physical skills from this speci…cation because with only twenty SIC industries the measure becomes
highly collinear with the intensity in age-depreciating skills.
25
20
4.3.1
Instrumenting demographic composition
Although it is hard to …nd a purely exogenous source of variation for demographic composition,
we employ several instruments which explain di¤erent sources of variation in the median age and
rely on weaker identi…cation assumptions than the OLS. The …rst instrument is the birth rate
per capita in 1960, which is highly correlated with the median age in 2000 but is independent
of many socioeconomic in‡uences that took place between 1960 and 2000, such as changes in
industrial policies or reduction in gender disparities in education system.26 The data for this
instrument is obtained from the World Development Indicators database.
Our second instrument for a country’s demographics is the religious composition. It is a
well established result that religious families have higher fertility rates than non-believers. Many
demographic studies also demonstrate that religious a¢ liation is an important determinant of
demographic behavior. In particular, countries with large Muslim and Catholic populations tend
to experience higher fertility rates. Our instrument for median age is based on six variables measuring the share of population in …ve main religious groups –Buddhism, Christianity, Hinduism,
Islam and Judaism –plus the share una¢ liated with any religion.27 Other religion groups serve
as an omitted category. We instrument the interactions of country’ median age and industry’
skill intensities in model (1) with Iik
d c , where Age
d c is obtained as …tted values from the
Age
cross-country regression of median age on religious group shares.28
The choice of our third instrument is motivated by Spolaore and Wacziarg (2014) who argue that the variation in fertility decline among European countries is related to their cultural
distance to France where the demographic revolution had begun. The authors con…rm that
the genetic distance from France, which captures ancestral and cultural di¤erences, is a strong
predictor of the onset of demographic transition to lower fertility levels. We use the data from
Spolaore and Wacziarg (2009) to construct genetic distance of each country in our sample to
France and use it as a source of exogenous variation in demographic composition to predict the
26
We continue to assume that industry factor intensities are exogenous and instrument Iik Agec variables with
interactions of Iik and the instruments for median age.
27
The data is retrieved from the Pew Research Center.
28
The R-squared from this regression is 0.34 and the p-value of the F-test for joint signi…cance of religious
group share variables is 0.00.
21
median age of populations in di¤erent countries. That instrument performs well in the …rst stage.
As expected, the coe¢ cient on the genetic distance in the regression for median is negative and
the t-statistics of the test of signi…cance equals to
9:31. It is important to emphasize that
genetic distance and religious composition capture di¤erent sources of variation in median age
as the coe¢ cient of correlation between the median age predicted from these two models is only
0:16.
We present the results with the birth rate in 1960 in column (1) of Table 7. The instrument
perform well in the …rst stage: the p-value for the Angrist-Pischke weak instrument test is
always equal to 0.0 and the F-statistics from the …rst stage regression are in the range from 93
to 112.29 The IV estimations results are consistent with our previous …ndings: the coe¢ cient
on age-appreciating cognitive skills is positive and highly signi…cant, and the coe¢ cients on agedepreciating cognitive and learning skills are negative. These results indicate that countries with
high fertility rates and with the large share of rural population in the past tend to specialize in
industries which depend on age-depreciating skills.
Results with religious composition and genetic distance as instruments are presented in
columns (2) and (3) of Table 7. The corresponding Kleibergen-Paap F-statistics for the endogenous variables is always greater than 20, which is well above the critical values tabulated by
Stock and Yogo (2005) and suggest that a bias from weak instruments is unlikely to be a concern.
As can be seen, the estimates with the religious composition are very similar to OLS and are
consistent with our earlier …ndings about the e¤ect of demographics on comparative advantage.
When the median age is instrumented with genetic distance, the e¤ect of age-appreciating is
also in line with our expectations, however, the coe¢ cient on age-depreciating cognitive skills
becomes small and insigni…cant. This result should not be particularly surprising. Although
the genetic distance from France is a statistically signi…cant determinant of the median age, it
is not capturing many other sources of exogenous variation in demographic composition across
countries which may a¤ect comparative advantage. Results with all three instruments together
are reporter in column (4) of Table 7 and again provide little evidence for the endogeneity of
median age: the coe¢ cient estimates are not much di¤erent from the OLS estimates and the
29
The full set of the …rst stage results is presented in Table 6A in the Appendix.
22
overidenti…cation test cannot reject the hypothesis of exogeneity of instrument.
Overall, while the exclusion assumption underlying some of our instruments can be argued,
the three instruments that we employ isolate di¤erent sources of potentially endogenous variation
in the median age and yet the results paint a similar picture to that documented previously.
4.3.2
An alternative measure of the stock of age-depreciating cognitive skills
Our second strategy to con…rm that the main result of this paper is not driven by correlation
between age composition and other unobservable determinants of a country’s comparative advantage is to use an alternative measure for a country’s relative productivity in age-dependent
skills. To this point, we have used a country’s age structure to proxy for its e¤ective endowment
of age-dependent skills in all regression speci…cations. In what follows we use an alternative
proxy variable for a country’s e¤ective endowment of one of the age-depreciating skills –memory
–and show that trade data is still consistent with predictions of our theoretical model.
The proposed measure is based on a standardized memory test conducted in 27 di¤erent
countries among seniors.30 The test consists of verbal registration and recall of a list of ten
words one minute after an interviewer read them to respondents. The test score is the fraction
of the number of words that were recalled correctly. This test provides an alternative measure of
cross-country di¤erences in e¤ective memory stocks and in aggregate productivities in memoryintensive tasks. Therefore, all else being equal, countries with higher memory tests scores, both
unconditional and conditional on observable characteristics, are expected to have comparative
advantage in memory-intensive industries, just like countries with younger labor force. But the
main advantage of the memory test is that, for a given age cohort, it provides a cross-country
variation in the quantity and quality of skill endowment which is largely independent of the
variation in population aging and demographic composition, and is thus not susceptible to the
possible omitted variable bias which may be inherent to the median age variable.
In Column (1) of Table 8 we present the results for the benchmark speci…cation (1) using
30
A detailed test description can be found in Appendix D. Although the sample is overrepresented by countries
with high median age, there are four countries from the …rst and second quartiles of the median age distribution.
23
median age as a proxy for skill endowment but with memory as a single cognitive skill. As
with the composite age-depreciating skill variable, the coe¢ cient on memory intensity interacted
with a country’s median age is negative and statistically signi…cant. Thus, consistently with our
previous …ndings, younger countries capture larger market shares in industries where employees
are required to have good memory. In column (2) we use the unconditional word recall test
score as a measure of skill endowment, averaged among individuals of 50-55 years of age within
each country in order to control for di¤erent age composition of respondents across countries.
The e¤ect of the interaction remains large and statistically signi…cant at 5% con…dence level.31
Column (3) shows that results of column (1) remain robust when estimated for the sample of
countries for which the memory test scores are available. It is also important to note that the
coe¢ cient estimates with memory test scores are remarkably similar to the estimates with the
median age. Since the two variables rely on di¤erent sources of variation in memory endowment
to identify country’s comparative advantage in memory-intensive industries, we believe that the
similarity of the estimates is reassuring that neither set of results is driven by the omitted variable
bias.
To make our results with memory test more comparable to those for the median age, we
use memory test scores to identify the contribution of demographic structure on comparative
advantage. All test participants complete a survey questionnaire, which includes information on
individuals’s age and gender. The survey also collects standard data on the highest completed
education and the number of years of schooling, which makes it comparable across countries.
In order to isolate the e¤ect of age and education from other sources of variation in cognitive
skills, we …rst regress the memory test scores on age and years of schooling.32 Using the estimated
elasticities of the test score with respect to age and the number of years of schooling from the …rst
stage regression, which are
0:005 and 0:011, respectively (column 2 of Table 7A in Appendix),
and assuming that the latter elasticity is age-independent, we construct the predicted value of
memory endowment for every country in our sample. In this way, we construct a measure of
memory endowment that captures the variation in memory test scores across countries which
31
Note that since the test score measures skill endowment, while the median age measures the depreciation of
skill endowment, the expected sign of the interaction is positive with the former measure and negative with the
latter.
32
In the …rst stage regression we also control for country …xed e¤ects. The main result is unchanged when age
enters non-linearly. Column (4) of Table 7A in the Appendix reports the results for this …rst-stage regression.
24
is due to demographics and educational attainment only. Results, presented in column (4), are
very similar to the benchmark. Next, we use the elasticities estimated in column (4) to construct
the out-of-sample predicted value of the test score for all countries in our trade data sample.
Interacting this measure of skill endowment with memory intensity in column (5) leaves the
main result unchanged. Finally, in column (6) we construct a measure of skill endowment for
senior workers only. At the …rst stage, we estimate the e¤ect of age and education level for every
5-year cohort group among people aged 50-65, and construct a country-level measure of e¤ective
memory endowment using the data on educational attainment reported in Barro and Lee (2000)
for every country and every 5-year age group. The results are robust.
Overall, Table 8 shows that no matter which measure of skill endowment we use, the coe¢ cient
on its interaction with sectorial memory intensity is always of the expected sign and statistically
signi…cant, with the magnitude of the coe¢ cient being remarkably stable across speci…cation.
4.4
4.4.1
Extensions
The role of education and health care in age-related cognitive development
The use of a country’s demographic composition as a proxy for unobserved endowment of agedependent skills in equation (11) relies on the assumption that the stock of each age-dependent
skill is a linear function of the median age only. However, age-related decrements in cognition
may not be entirely due to biological aging process. The presence of other factors that a¤ect
cognitive functioning at di¤erent ages may result in biased estimates of
k
coe¢ cients if the
cross-country variation in those factors is correlated with demographics. In this subsection we
consider two additional factors of cognitive development and demonstrate that our main …ndings
remain quantitatively and statistically robust to alternative de…nitions of the proxy variables for
the age-dependent skills.
The …rst such factor is the quality of the health care system. Results of numerous medical
studies reveal a strong e¤ect of psychological and systemic diseases on cognitive functioning for
25
individuals of all ages,33 but the e¤ect is particularly pertinent to older individuals who are
subject to increased incidence and prevalence of such diseases. Thus, insofar as the age-related
cognitive and physical decline is driven by deteriorating health conditions, it may, to some extent,
be reversible with appropriate medical treatment. In this way, an e¢ cient health care system
could remediate the e¤ect of population aging on the e¤ective stock of cognitive skills and physical
abilities.
The second factor which may a¤ect the relationship between aging and cognitive functioning
is education and cognitive training. A large body of literature documents a positive relationship
between education and old age cognitive functioning, and several recent studies have been able
to identify a causal e¤ect of childhood schooling on cognitive abilities (primarily on memory)
at older ages by exploiting exogenous variation in education policies (e.g. Glymour et al, 2008,
Banks and Mazzonna, 2012). Moreover, it has also been established that education and training
can moderate the course of intellectual decline as individuals get older (Schaie, 1986, and Schaie,
2005). Thus, we may expect that increasing rates of educational attainment, especially among
older workers, can increase the e¤ective stock of age-dependent cognitive skills.
Based on the above evidence, we extend equation (11) by allowing for country c’s endowment
of age-dependent skill k to be a function of the median age (Agec ), the quality of the health care
system (Healthc ), and education level (Educc )
ln Fck =
where we expect
and
k
3
k
1
k
0
+
k
1 Agec
+
k
2 Healthc
< 0 for age-depreciating skills,
k
1
+
k
3 Educc ;
(4)
> 0 for age-appreciating skills, and
k
2
>0
> 0 for any k. Then equation (1) becomes
ln Xcpi =
X
k k
1 Ii
k k
2 Ii
Agec +
Healthc +
k k
3 Ii
Educc +
(5)
k2K
+
X
n
n Ii
Fcn +
0
cp
+
c
+
pi
+ "cpi ;
n
where
i
j
=
j
+
j
j
i
. Estimation results for equation (5) are presented in Table 9. Education
33
See Stern and Carstensen (2000) for a survey of the literature. In subsection 4.3.2 we also provide some
evidence on the relationship between health and cognition.
26
is measured with the share of population with secondary and post-secondary education, although
results do not change when we use alternative measures of education or when the measures of
educational attainment are constructed only for senior workers. In column (1), the e¢ ciency of
the health care system is measured with the share of total health expenditure in GDP, obtained
from the World Bank for the year 2000. All interactions of education variables and skill intensities
in column (1) are insigni…cant, suggesting that education does not a¤ect accumulation of agedependent skills. This result, however, may be driven by poor quality of educational data,
which does not take into account di¤erences in the quality of education across countries. As
for the e¤ect of the share of health expenditure in GDP, only the interaction with the intensity
in age-appreciating skills is statistically signi…cant and positive, as expected. Yet insigni…cant
coe¢ cients for age-depreciating skills provide no evidence that increase in health expenditure at
the national level can remediate the e¤ect of cognitive decline in aging population on comparative
advantage. The estimated coe¢ cients on the interactions of median age with intensities in agedependent skills remain statistically signi…cant and similar in magnitudes to the benchmark
values.
4.4.2
Population aging and changes in preferences
Our theoretical model focuses on the e¤ect of population aging on trade ‡ows through the
e¤ect on supply and thus suggests that demographics can be an important factor of a country’s
comparative advantage. However, the demographic composition can also a¤ect trade through
the e¤ect on demand. In particular, it is possible that preferences and demand for di¤erent
manufacturing products may change with age. In this case, our Iik
Agec variables may capture
changes in preferences rather than in specialization if aging is associated with a reduction in
demand for products which use age-appreciating skills intensively and/or an increase in demand
for products which are intensive in age-depreciating skills.
To explore this possibility, we use the Canadian Survey of Household Spending for the year
2000 kept by Statistics Canada. These data are representative of an open market economy, a
context which is applicable to other developed countries that account for around half of exporterindustry observations in our sample. The survey includes complete information on household
27
expenditure during the whole calendar year for over 14,000 households in Canada. Figure 5 shows
the variation in the share of di¤erent consumer goods in total household expenditure across four
cohorts.34 The …gure reveals that, while for most products there are no substantial di¤erences
in expenditure shares among di¤erent cohorts, older individuals tend to spend more on food and
medical equipment and less on sports, audio, and video equipment. Hence, not controlling for
di¤erences in consumption patterns between older and younger individuals can result in omitted
variable bias in
k
coe¢ cients in equation (1) if those di¤erences are systematically related to
variation in skill intensities across industries. Yet, in the Canadian data this relationship is
weak: the Spearman rank correlation between skill intensities and the gap in expenditure shares
of 55-65 and 25-35 age cohorts is only around 0.1, while the same correlations with physical and
human capital intensities are 0.23 and -0.34, respectively.35
As an additional robustness test, we estimate equation (1) on a subset of industries which
produce consumer goods. If our results are mainly driven by the relationship between age and
consumption behavior, we would expect the e¤ect to be more pronounced for …nal goods. Results
in column (2) of Table 9 show that the magnitudes of
k
coe¢ cients on the sample of consumption
goods are similar to our baseline estimates for the entire sample, suggesting that age-dependent
preferences are unlikely to play a major role in our results.
5
The e¤ect of population aging on comparative advantage
We have shown in the previous section that a country’s age structure is a source of comparative
advantage. In Section 2 we also argue that population aging at a rate faster than in other
countries should alter a country’s export structure via a decrease in relative productivity and
increase in relative costs in industries which use age-depreciating skills intensively. To test this
prediction of the model, we estimate equation (3) and relate changes in exports to the interaction
of industry factor intensities and changes in population age structure between 1962 and 2000.
34
35
We keep only households with either one person or married couples from the same cohort.
These results are presented in Table 8A in the Appendix.
28
Estimation results are reported in Table 10. The results are based on the sample of 82
exporters, 135 importers, and 76 industries. Insigni…cant coe¢ cients on capital and skilled labor
in the …rst column reveal that, contrary to our expectations, accumulation of physical and human
capitals is not associated with a shift in the export structure towards capital- or skill-intensive
industries.36 Both results are in contrast with Romalis (2004), who shows that changes in capital
and, in some speci…cation, skilled labour stocks imply changes in countries’structure of exports
to the US. We …nd that the di¤erence between our results and those by Romalis is primarily
driven by the choice of the US as a single importing country. When we restrict our sample of
importers to the US only, we obtain large and positive coe¢ cients on both capital and skilled
labor, with the latter also being statistically signi…cant.37
Turning to the estimates with age-dependent skills in columns (2)-(4), we see that the coe¢ cients on all skills are signi…cant and have expected signs. These results imply that rapid
population aging, on one hand, increases a country’s specialization in industries which use ageappreciating cognitive skills intensively, and on the other hand, it erodes competitive advantage
in industries which rely on age-depreciating cognitive skills and physical abilities. When both
age-dependent cognitive skills are estimated in one regression in column (5), the magnitudes become smaller but the coe¢ cients remain signi…cant. We do not include physical skills in column
(5) because it is highly collinear to age-depreciating cognitive skills and the two coe¢ cients cannot be jointly identi…ed. For this reason, in speci…cations that follow we do not include physical
ability in the list of covariates.
To illustrate the magnitude of the implied estimates, consider two countries, US and Argentina, and two industries, motor vehicles and meat products. Between 1968 and 2000 the
median age in the US increased by one standard deviation relative to Argentina, or by 4.7 years.
Also, the intensity of motor vehicles in age-appreciating cognitive skills is approximately one
36
Using data on factor inputs and output for 28 manufacturing industries and 27 countries, Blum (2010) also
…nds that changes in capital and skilled labor endowment between 1973 and 1990 did not a¤ect a country’s output
mix, but did a¤ect its relative factor prices and factor intensities at the industry level.
37
It is also important to note that lack of evidence for the Rybczynski e¤ect for capital may be caused by
changes in capital intensities over time. As column (3) of Table 6 shows, interactions of capital intensities,
constructed with 2000 data, are insigni…cant determinants of trade ‡ows in 1970. Poor measurement of e¤ective
capital stock and human capital can also be a problem as it is notoriously di¢ cult to measure both types of
capital in a consistent way across countries.
29
standard deviation higher than in the meat products. Given that the standard deviation of
change in log exports is 2:4, the estimates in column (5) imply that the more rapid population aging in the US and the associate increase in the stock of age-appreciating cognitive skills
induced exports of motor vehicles relative to meat products to increase by e2:4
0:027
= 6:7%
more in the US than in Argentina. Similarly, Argentinian exports of textile products relative to
household appliances, the two industries with one standard deviation di¤erence in the intensity
in age-depreciating cognitive skills, increased by e2:4
0:029
= 6:7% more than in the US.
In Table 11 we report several robustness tests for the e¤ect of demographic changes on comparative advantage. Column (1) presents the results with the share of young workers (aged 20-40)
in the labour force as a measure of abundance in age-dependent skills. The results are fully consistent with the previous …ndings that population aging and the reduction in the share of young
workers are associated with an increase in exports of products that use age-appreciating cognitive skills intensively, and decrease in exports of products which are intensive in age-depreciating
cognitive skills.
In columns (2), (3), and (4) we report results using the base year of 1970, 1980, and 1990,
respectively, to construct the di¤erences in trade ‡ows and factor endowments. The shorter
span for time-di¤erencing increases the number of exporting countries and observations, since
many countries, especially less developed ones with younger populations, do not report trade
data for 1960s and 70s. With the more representative sample of exporting countries in column
(2), the e¤ect of population aging becomes even more pronounced for age-dependent skills –
the coe¢ cients on both skills have expected signs and are statistically signi…cant. However,
with di¤erencing over progressively shorter periods, the e¤ect becomes weaker in column (3) and
disappears entirely in column (4), where 1990 is used as the base date. That the results change
substantially with shorter-span time di¤erencing may indicate that the general-equilibrium e¤ects
due to resource relocation between industries are less visible in higher-frequency data if it takes
more than ten years for the economy to adjust to changes in the supply of factor inputs. Yet
weak results in column (4) may also be due to low variation in population aging over shorter
periods of time, which would make the identi…cation of the e¤ect of our interest more di¢ cult.
Our baseline speci…cation (3) implies a linear relationship between population aging and trade
30
‡ows for given factor intensities. However, it may be the case that increases and decreases in the
median age of population may have di¤erential impacts on the structure of trade. For example,
we know that countries with rapidly aging populations experience an increase in the relative
supply and a decrease in the relative price of age-appreciating skills. Therefore, if relative goods’
prices do not change much over time, aging countries may observe more substantial changes
in the production structure as compared to countries with stable demographics and constant
relative supply of age-dependent skills. In column (5) of Table 10 we present the results for
speci…cation (3) where the e¤ect of Iik
Agec interaction is estimated separately for slowly and
rapidly aging countries:
ln Xcpi =
X X
j j k
k Dc Ii
k2K j2Y;O
Agec +
X
f
f Ii
Fcf +
0
cp
+
c
+
pi
+ "cpi ;
(6)
f 2F
where DcY and DcO are dummy variables which take the value of one if country c is below and
above the median of
O
k
Age distribution, respectively. The results reveal that only coe¢ cients
are statistically signi…cant, implying the e¤ect of age structure on trade is mainly driven by
countries with fast population aging.
The …ndings of this section provide new insights for evaluation of economic consequences
of population aging and carry important public policy implications. The results imply that
demographic changes a¤ect country’s comparative advantage and the structure of output across
industries. In particular, rapidly aging countries lose comparative advantage in industries which
are intensive in age-depreciating cognitive and physical skills but gain comparative advantage in
industries that utilize age-appreciating cognitive skills. As a result, population aging changes the
relative supply of age-dependent skills, and international trade acts as the adjustment mechanism
that works through increase in relative demand for more abundant factors by employing them in
export-oriented sectors. Therefore, as long as the rates of population aging di¤er across countries,
the e¤ect of demographic changes on skill premia for age-dependent skills and relative income
levels of senior and junior workers will be muted by changes in output and exports composition.
31
6
Conclusions
Variations in relative productivities and factor endowments across countries are the sources
of comparative advantage. This paper contributes to the comparative advantage literature by
analyzing the e¤ect of cross-country di¤erences in demographic composition and endowments
of age-dependent skills on the structure of commodity trade. We incorporate age-dependent
skills into the extension of the Eaton-Kortum model by Chor (2010) to show that cross-country
di¤erences in demographic structure determine the pattern of trade. The model also predicts
that population aging results in a reduction in productivity and increase in unit output costs in
industries that utilize age-depreciating skills intensively.
We apply the two main predictions of the model to bilateral trade data for a large panel of
countries and 86 industries over the time period from 1962 to 2010, and con…rm that population
age structure is an important factor of a country’s comparative advantage. First, we show that
countries with younger labour force tend to specialize in industries which are intensive in agedepreciating skills, and the result is remarkably robust to the inclusion of controls for alternative
sources of comparative advantage and is not con…ned to a particular time period. Furthermore,
the e¤ect of a country’s age structure on trade is economically sizable and explains more of the
variation in trade ‡ows than physical and human capital endowments combined.
Second, we establish that population aging results in a shift in a country’s export structure
towards industries which intensively use age-appreciating skills and away from industries which
rely heavily on age-depreciating skills. Population aging could therefore play an important role
in the structure of economic activity within and between countries. Yet, our …ndings point to an
optimistic perspective for the long-run impact of population aging on relative wages of older and
younger workers. The results demonstrate that rapidly aging countries can adjust to demographic
changes by utilizing the growing share of older workers in industries that intensively utilize ageappreciating skills. Thus, free trade and imports of products that embody age-depreciating skills
can alleviate the e¤ect of a reduction in supply of skills of younger workers in fast-aging countries
on relative income levels.
32
References
Attanasio, O., S. Kitao, and G. L. Violante (2007): “Global Demographic Trends and
Social Security Reform,”Journal of Monetary Economics, 54(1), 144–198.
Banks, J., and F. Mazzonna (2012): “The E¤ect of Education on Old Age Cognitive Abilities:
Evidence from a Regression Discontinuity Design,”Economic Journal, 122(560), 418–448.
Barro, R. J., and J. W. Lee (2013): “A New Data Set of Educational Attainment in the
World, 1950-2010,”Journal of Development Economics, 104(C), 184–198.
Beck, T. (2003): “Financial Dependence and International Trade,” Review of International
Economics, 11(2), 296–316.
Blum, B. S. (2010): “Endowments, Output, and the Bias of Directed Innovation,” Review of
Economic Studies, 77(2), 534–559.
Boersch-Supan, A., A. Ludwig, and J. Winter (2001): “Aging and International Capital
Flows,”Working Paper 8553, National Bureau of Economic Research.
(2006): “Ageing, Pension Reform and Capital Flows: A Multi-Country Simulation
Model,”Economica, 73(292), 625–658.
Bombardini, M., G. Gallipoli, and G. Pupato (2012): “Skill Dispersion and Trade Flows,”
American Economic Review, 102(5), 2327–48.
Chor, D. (2010): “Unpacking Sources of Comparative Advantage: A Quantitative Approach,”
Journal of International Economics, 82(2), 152–167.
Costinot, A. (2009): “On the origins of comparative advantage,” Journal of International
Economics, 77(2), 255–264.
Cunat, A., and M. J. Melitz (2012): “Volatility, Labor Market Flexibility, And The Pattern
Of Comparative Advantage,”Journal of the European Economic Association, 10(2), 225–254.
Davis, D. R., and D. E. Weinstein (2001): “An Account of Global Factor Trade,”American
Economic Review, 91(5), 1423–1453.
33
de Zwart, B., and M. Frings-Dresen (1995): “Physical Workload and the Ageing Worker: A
Review of the Literature,”International Archives of Occupational and Environmental Health,
68, 1–12.
Debaere, P. (2014): “The Global Economics of Water: Is Water A Source of Comparative
Advantage?,”American Economic Journal: Applied Economics, 6(2), 32–48.
Debeare, P. (2003): “Relative Factor Abundance and Trade,” Journal of Political Economy,
111(3), 589–610.
Domeij, D., and M. Floden (2006): “Population Ageing and International Capital Flows,”
Discussion paper.
Eaton, J., and S. Kortum (2002): “Technology, Geography, and Trade,”Econometrica, 70(5),
1741–1779.
Feenstra, R., J. Romalis, and P. Schott (2002): “U.S. Imports, Exports and Tari¤ Data,
1989-2001,”Discussion paper, NBER Working Paper 9387.
Firpo, S., N. M. Fortin, and T. Lemieux (2011): “Occupational Tasks and Changes in the
Wage Structure,”Iza discussion papers, Institute for the Study of Labor (IZA).
Galor, O., and A. Mountford (2008): “Trading Population for Productivity: Theory and
Evidence,”Review of Economic Studies, 75(4), 1143–1179.
Glymour, Maria; Kawachi, I. J. C. B. L. (2008): “Does Childhood Schooling A¤ect Old
Age Memory or Mental Status? Using State Schooling Laws as Natural Experiments,”Journal
of Epidemiology and Community Health, 62(6), 532–7.
Harrigan, J. (1997): “Technology, Factor Supplies, and International Specialization: Estimating the Neoclassical Model,”American Economic Review, 87(4), 475–94.
Hedge, Jerry; Borman, W. L. S. (2005): The Aging Workforce: Realities, Myths, And
Implications For Organizations. American Psychological Association.
Helliwell, J. F. (2004): “Demographic Changes and International Factor Mobility,” NBER
Working Papers 10945, National Bureau of Economic Research, Inc.
34
Helpman, E., and O. Itskhoki (2010): “Labour Market Rigidities, Trade and Unemployment,”Review of Economic Studies, 77(3), 1100–1137.
Helpman, E., M. Melitz, and Y. Rubinstein (2008): “Estimating Trade Flows: Trading
Partners and Trading Volumes,”The Quarterly Journal of Economics, 123(2), 441–487.
Higgins, M. (1997): “Demography, National Savings and International Capital Flows,”Discussion paper.
Holzmann, R. (2002): “Can Investments in Emerging Markets Help to Solve the Aging Problem?,”Journal of Emerging Market Finance, 1(2), 215–241.
Jolliffe, I. (2002): Principal Component Analysis. Springer, Berlin.
Levchenko, A. A. (2007): “Institutional Quality and International Trade,” Review of Economic Studies, 74(3), 791–819.
Ludwig, A., D. Krueger, and A. H. Boersch-Supan (2007): “Demographic Change, Relative Factor Prices, International Capital Flows, and Their Di¤erential E¤ects on the Welfare
of Generations,”Working Paper 13185, National Bureau of Economic Research.
Ludwig, A., T. Schelkle, and E. Vogel (2012): “Demographic Change, Human Capital
and Welfare,”Review of Economic Dynamics, 15(1), 94–107.
Ludwig, A., and E. Vogel (2010): “Mortality, Fertility, Education and Capital Accumulation
in a Simple OLG economy,”Journal of Population Economics, 23(2), 703–735.
Manova, K. (2008): “Credit Constraints, Equity Market Liberalizations and International
Trade,”Journal of International Economics, 76(1), 33–47.
McDowd, J., and F. Craik (1988): “E¤ects of Aging and Task Di¢ culty on Divided Attention
Performance,”Journal of Experimental Psychology: Human Perception and Performance, 14,
267–280.
McDowd, J., and R. Shaw (2000): Attention and Aging: A Functional Perspective. Erlbaum;
Mahwah, NJ.
35
Narciso, A. (2013): “The Impact of an Ageing Population on International Capital Flows,”
MPRA Working Papers 26457, Copenhagen Business School.
Nunn, N. (2007): “Relationship-Speci…city, Incomplete Contracts, and the Pattern of Trade,”
The Quarterly Journal of Economics, 122(2), 569–600.
Ohnsorge, F., and D. Trefler (2007): “Sorting It Out: International Trade with Heterogeneous Workers,”Journal of Political Economy, 115(5), 868–892.
Oldenski, L. (2012): “Export Versus FDI and the Communication of Complex Information,”
Journal of International Economics, 87(2), 312–322.
Romalis, J. (2004): “Factor Proportions and the Structure of Commodity Trade,” American
Economic Review, 94(1), 67–97.
Salthouse, T. (1998): “Independence of Age-Related In‡uences on Cognitive Abilities Across
the Life Span,”Developmental Psychology, 34, 851–864.
(2009): “When Does Age-related Cognitive Decline Begin?,”Neurobiology of Aging, 30,
507–514.
(2010): “In‡uence of Age on Practice E¤ects in Longitudinal Neurocognitive Change,”
Neuropsychology, 24, 563–572.
Schaie, W. (1986): Adult Development and Aging. Boston: Little, Brown & Co.
(1994): “The Course of Adult Intellectual Development,” Americal Psychologist, 49,
304–313.
(2000): “The Impact of Longitudinal Studies on Understanding Development from
Young Adulthood to Old Age,” International Journal of Behavioral Development, 24, 257–
266.
(2005): Developmental In‡uences on Adult Intelligence: The Seattle Longitudinal Study.
New York: Oxford University Press.
36
Silva, J. M. C. S., and S. Tenreyro (2006): “The Log of Gravity,”The Review of Economics
and Statistics, 88(4), 641–658.
Singh-Manoux, Archana; Kivimaki, M. G. M. E. A. (2012): “Timing of Onset of Cognitive
Decline: Results from Whitehall II Prospective Cohort Study,” British Medical Journal, 344,
18–26.
Spolaore, E., and R. Wacziarg (2009): “The Di¤usion of Development,” The Quarterly
Journal of Economics, 124(2), 469–529.
(2014): “Fertility and Modernity,”Discussion paper.
Stern, P. C., and L. L. Carstensen (2000): The Aging Mind: Opportunities in Cognitive
Research. The National Academies Press.
Stock, J. H., and M. Yogo (2005): “Testing for Weak Instruments in Linear IV Regression.
In Stock, J. H. and Andrews, D. W., editors,”Discussion paper, Identi
cation and Inference for Econometric Models: Essays in Honor of Thomas J. Rothenberg.
Cambridge University Press.
Tang, H. (2012): “Labor Market Institutions, Firm-Speci…c Skills, and Trade Patterns,”Journal
of International Economics, 87(2), 337–351.
Trefler, D. (1993): “International Factor Price Di¤erences: Leontief Was Right!,”Journal of
Political Economy, 101(6), 961–87.
Tsang, P. (1998): “Age, Attention, Expertise, and Time-Sharing Performance,” Psychology
and Aging, 13, 323–347.
Verhaeghen, P., and J. Cerella (2002): “Aging, Executive Control, and Attention: A
Review of Meta-Analyses,”Neuroscience Biobehavioral Review, 26, 849–57.
Wolff, E. N. (2003): “Skills and Changing Comparative Advantage,”The Review of Economics
and Statistics, 85(1), 77–93.
37
-2
-4
-6
-8
Coef Young = 0.48 [t-stat = 2.47]
Coef Old = -0.56 [t-stat = -4.59]
-10
Log of industry's share in total exports
Figure 1. Correlation between intensity in age-depreciating cognitive skills and exports
for old and young countries
-4
-2
0
2
4
Intensity in age-depreciating cognitive skills
Industry share in total exports, young countries
Linear fit, young countries
Industry share in total exports, old countries
Linear fit, old countries
Note: Young and old countries are the ones with the median age less than 20 and greater than 35, respectively
-8
-6
-4
-2
Coef Young = -0.57 [t-stat = -3.19]
Coef Old = 0.18 [t-stat = 1.79]
-10
Log of industry's share in total exports
Figure 2. Correlation between intensity in age-appreciating cognitive skills and exports
for old and young countries
-2
-1
0
1
2
Intensity in age-appreciating cognitive skills
Industry share in total exports, young countries
Linear fit, young countries
Industry share in total exports, old countries
Linear fit, old countries
Note: Young and old countries are the ones with the median age less than 20 and greater than 35, respectively
-2
-4
-6
-8
Coef Young = 0.39 [t-stat = 2.45]
Coef Old = -0.51 [t-stat = -5.20]
-10
Log of industry's share in total exports
Figure 3. Correlation between intensity in physical abilities and exports for old and
young countries
-4
-2
0
2
Intensity in physical abilities
Industry share in total exports, young countries
Linear fit, young countries
Industry share in total exports, old countries
Linear fit, old countries
Note: Young and old countries are the ones with the median age less than 20 and greater than 35, respectively
Figure 4. Relationship between skill intensities and median age of labor force across
industries
46
Age-appreciating skills
44
42
40
38
36
Median age of the labor force
slope = 0.068
[std.er. = 0.005]
80
90
100
110
Intensity in age-appreciating cognitive skills
38
40
42
44
slope = -0.071
[std.er. = 0.046]
36
Median age of the labor force
46
Age-depreciating skills
-30
-20
-10
0
Intensity in age-depreciating cognitive skills
46
Physical skills
44
42
40
38
36
Median age of the labor force
slope = -0.016
[std.er. = 0.016]
20
40
60
Intensity in physical skills
80
100
0
0
Cloth
Source: Canadian Survey of Household Spending, 2000
.1
5
Product rank
10
Household textile
Computer equip.
Reading materials
Furniture
Audio and video equip.
Alcohol
Sport equipment
Household equip. and appliances
Tobacco
Medical equip.
Automobiles
.05
Share in total household expenditure
Food
.15
Figure 5. Variation in consumer goods expenditure shares among different age cohorts
Age 25-35
Age 35-45
Age 45-55
Age 55-65
15
Table 1. Occupations with extreme skill intensities
Rank
10 Most skill intensive occupations
SOC
1
273042
3
434151
2
4
5
6
7
8
9
10
113111
414012
414011
519083
113061
113121
439031
433061
1
514194
3
519041
2
4
5
6
7
8
9
10
516064
519021
537051
519111
434151
537072
514021
519121
1
475051
3
499044
2
4
5
6
7
8
9
10
453011
537062
472211
472111
499096
519197
512011
519012
10 Least skill intensive occupations
Occupation
Rank
SOC
Age-appreciating cognitive skills
Occupation
Technical Writers
Compensation and Benefits
Managers
Order Clerks
Sales Representatives, Except
Technical and Scientific Products
Sales Representatives, Technical
and Scientific Products
Ophthalmic Laboratory
Technicians
1
516051
Sewers, Hand
3
517021
Furniture Finishers
Maintenance and Repair Workers,
General
Human Resources Managers
8
Purchasing Managers
Desktop Publishers
2
4
5
6
7
9
Procurement Clerks
10
191022
499071
514052
519031
517041
519198
519123
518093
Age-depreciating cognitive skills
Tool Grinders, Filers, and
Sharpeners
Textile Winding, Twisting, and
Drawing Out Machine Setters,
Operators, and Tenders
Extruding, Forming, and Pressing
Machine Setters, and Operators
Crushing, Grinding, and Polishing
Machine Setters, and Operators
Industrial Truck and Tractor
Operators
Packaging and Filling Machine
Operators
Order Clerks
Pump Operators, Except Wellhead
Pumpers
Extruding and Drawing Machine
Setters, Operators, and Tenders,
Metal and Plastic
Coating, Painting, and Spraying
Machine Setters and Operators
Microbiologists
Pourers and Casters, Metal
Cutters and Trimmers, Hand
Sawing Machine Setters and
Operators
Helpers--Production Workers
Painting, Coating, and Decorating
Workers
Petroleum Pump System Operators,
Refinery Operators, and Gaugers
1
191022
Microbiologists
3
452041
Graders and Sorters, Agricultural
Products
2
4
5
6
7
8
9
10
Physical abilities
172031
191021
172131
111011
172071
192031
172041
172112
Biomedical Engineers
Biochemists and Biophysicists
Materials Engineers
Chief Executives
Electrical Engineers
Chemists
Chemical Engineers
Industrial Engineers
Rock Splitters, Quarry
Fishers and Related Fishing
Workers
Millwrights
Laborers and Freight, Stock, and
Material Movers, Hand
Sheet Metal Workers
1
131081
Logisticians
3
172131
Materials Engineers
Tire Builders
Aircraft Structure, Surfaces,
Rigging, and Systems Assemblers
Separating, Filtering, and Still
Machine Setters, and Operators
8
Electricians
Riggers
2
4
5
6
7
9
10
271024
112021
271021
173013
172141
172031
173012
273042
Graphic Designers
Marketing Managers
Commercial and Industrial Designers
Mechanical Drafters
Mechanical Engineers
Biomedical Engineers
Electrical and Electronics Drafters
Technical Writers
Note: Only occupations with at least 10% employment in manufacturing sector are included in the rankings.
Table 2. Correlation between industry-level intensities in factor inputs
(1)
(2)
(3)
(4)
(5)
1
(1) cog_app
-0.23
1
(2) cog_dep
-0.40
0.93
1
(3) physical
0.39
-0.91
-0.24
1
(4) capital intensity
0.63
-0.70
-0.83
0.35
1
(5) skill intensity
Note: correlation coefficients are calculated over 86 3-digit manufacturing NAICS
industries.
Table 3. Baseline specification with median age
Cog_appi ×
(Median age)c
(1)
Cog_depi ×
(Median age)c
(2)
0.040**
(8.17)
Physicali ×
(Median age)c
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
ln(Distance)
Customs Union dummy
FTA dummy
Common border
Common language
Common ethnisity
Ever in colonial
relationship
Common colonizer
Current colony
Ever same country
Importer-industry FE
Exporter FE
Trade costs controls
R-squared
N
(3)
-0.058**
(-10.59)
(4)
-0.063**
(-10.62)
-0.029*
(-2.10)
-0.036*
(-2.46)
(6)
0.031**
(5.73)
-0.042**
(-2.83)
-0.013
(-0.83)
0.030**
(6.23)
0.018**
(3.60)
0.029**
(5.83)
-0.358**
(-107.13)
-0.359**
(-106.94)
-0.358**
(-106.92)
-0.358**
(-107.15)
-0.353**
(-110.99)
-0.358**
(-107.20)
0.110**
(16.44)
0.109**
(16.15)
0.108**
(16.09)
0.108**
(16.06)
0.105**
(16.14)
0.108**
(16.11)
0.049**
(10.31)
0.189**
(21.20)
0.218**
(32.43)
0.095**
(13.44)
0.036**
(5.27)
0.195**
(22.74)
0.242**
(26.01)
0.234**
(5.90)
0.047**
(4.47)
YES
YES
YES
0.551
414,918
0.038**
(7.59)
0.192**
(21.61)
0.215**
(31.66)
0.097**
(13.68)
0.035**
(5.17)
0.196**
(22.77)
0.243**
(25.83)
0.231**
(5.80)
0.050**
(4.69)
YES
YES
YES
0.553
413,466
0.023**
(4.16)
0.190**
(21.44)
0.214**
(31.63)
0.097**
(13.59)
0.036**
(5.25)
0.196**
(22.90)
0.241**
(25.76)
0.230**
(5.78)
0.049**
(4.68)
YES
YES
YES
0.554
413,466
0.024**
(4.66)
(5)
0.030**
(6.10)
0.018**
(3.17)
0.191**
(21.56)
0.214**
(31.60)
0.097**
(13.60)
0.035**
(5.22)
0.197**
(22.92)
0.241**
(25.76)
0.230**
(5.79)
0.049**
(4.65)
YES
YES
YES
0.554
413,466
0.188**
(22.05)
0.213**
(33.92)
0.066**
(9.73)
0.069**
(10.68)
0.175**
(20.87)
0.242**
(28.01)
0.143**
(4.06)
0.053**
(5.28)
YES
YES
YES
0.557
462,045
0.018**
(3.46)
0.015**
(2.64)
0.192**
(21.65)
0.214**
(31.53)
0.097**
(13.66)
0.035**
(5.18)
0.197**
(22.91)
0.242**
(25.84)
0.230**
(5.77)
0.049**
(4.66)
YES
YES
YES
0.555
413,466
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in industry i in year
2000. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust standard errors are clustered by
exporter-industry. * significant at 5%, ** significant at 1%.
Table 4. Baseline specification with the share of young workers
Cog_appi ×
(share of young workers)c
Cog_depi ×
(share of young workers)c
(1)
-0.038**
(-8.03)
Physicali ×
(share of young workers)c
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
(Skill int.)i ×
(Skill abund. young)c
Importer-industry FE
Exporter FE
Trade costs controls
R-squared
N
0.020**
(4.20)
0.040**
(8.03)
(2)
0.054**
(10.58)
0.029**
(5.99)
0.026**
(4.91)
(3)
0.058**
(10.54)
0.025**
(5.10)
0.022**
(4.02)
(4)
-0.029**
(-5.66)
-0.028**
(-5.38)
0.005
(0.34)
-0.000
(-0.02)
0.045**
(3.12)
0.021**
(4.30)
0.019**
(3.54)
YES
YES
YES
YES
0.553
0.554
0.555
0.555
YES
YES
411,362
YES
YES
411,362
YES
YES
411,362
(5)
YES
YES
411,362
0.046**
(3.19)
0.021**
(4.41)
0.015**
(2.94)
-0.022**
(-4.63)
YES
YES
YES
0.555
411,362
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in
industry i in year 2000. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust
standard errors are clustered by exporter-industry. * significant at 5%, ** significant at 1%.
Table 5. Robustness tests.
Cog_appi ×
(Median age)c
Cog_depi ×
(Median age)c
Physicali ×
(Median age)c
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
R-squared
N
(1)
0.028**
(5.29)
(2)
0.029**
(5.50)
(3)
0.029**
(5.37)
0.030**
(6.14)
-0.004
(-0.27)
-0.022
(-1.28)
-0.043**
(-2.92)
-0.043**
(-2.90)
-0.044**
(-2.93)
0.017**
(3.33)
0.017**
(3.24)
0.555
0.556
-0.004
(-0.23)
0.027**
(4.31)
413,466
-0.007
(-0.44)
0.035**
(6.70)
413,466
(4)
(5)
0.015+
(1.75)
(6)
0.011+
(1.88)
-0.040*
(-2.43)
-0.050**
(-4.81)
-0.089**
(-4.60)
0.021**
(3.95)
0.017**
(3.33)
0.026**
(5.06)
0.027**
(5.19)
0.560
0.555
0.555
0.555
0.025**
(3.90)
401,633
0.015**
(2.64)
413,466
-0.013
(-0.93)
0.018**
(3.23)
413,466
(7)
0.001
(0.09)
(8)
0.007
(0.98)
0.037**
(5.51)
-0.023+
(-1.82)
-0.002
(-0.11)
-0.027*
(-2.21)
-0.045**
(-3.99)
0.014*
(2.31)
0.018**
(2.82)
0.551
0.551
0.016
(0.83)
-0.042**
(-2.87)
0.017**
(2.91)
0.020**
(3.33)
413,466
299,639
(9)
-0.038+
(-1.94)
0.016*
(2.56)
0.020**
(3.42)
0.018**
(2.84)
299,639
299,639
0.550
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in industry i in year 2000. All specifications include
exporter and importer-industry fixed effects. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust standard errors are clustered by
exporter-industry. + significant at 10%, * significant at 5%, ** significant at 1%. In columns (1), (2) and (3) human capital endowment (Skill abund.)c is measured,
respectively, with the average years of schooling, the share of population with tertiary education, and the human capital index from Barro and Lee (2013). In
columns (4), (5), and (6) industry intensities in all age dependent skills are normalized, respectively, by alternative age-neutral O*Net measures of "deductive
reasoning", "fluency of ideas", and "information ordering". In columns (7) and (8) industry-level intensities in age-dependent skills were constructed using
occupational employment shares from 1980 and 1990 US Census, respectively. Column (9) represents estimates of the benchmark specification using the same
sample as in columns (7) and (8).
Table 6. Robustness tests.
Cog_appi ×
(Median age)c
Cog_depi ×
(Median age)c
Physicali ×
(Median age)c
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
(External fin. Depend.)i ×
(Fin. development)c
(1)
0.034**
(6.06)
(2)
0.037**
(6.12)
(3)
0.040**
(7.24)
0.034**
(3.74)
-0.005
(-0.28)
-0.038+
(-1.65)
-0.044**
(-2.88)
-0.051**
(-3.12)
-0.052**
(-3.41)
0.019**
(3.61)
0.018**
(2.76)
0.010+
(1.90)
-0.011
(-0.64)
0.018**
(2.91)
-0.017
(-0.94)
0.024**
(3.86)
(4)
0.025**
(4.58)
(5)
0.031**
(5.80)
-0.040**
(-2.73)
-0.121**
(-8.72)
-0.090
(1.40)
0.019**
(3.58)
0.158**
(8.88)
0.024**
(2.43)
-0.016
(-0.95)
0.008
(1.42)
Importer-exporter FE
Trade costs controls
R-squared
N
0.202**
(16.77)
0.170**
(8.09)
(Sales volatility)i ×
(Flexible lab. markets)c
Exporter FE
0.043+
(1.78)
0.219**
(14.25)
(Job complexity)i ×
(Judicial quality)c
Importer-industry FE
0.051**
(3.38)
-0.041*
(-2.05)
(Contract intensity)i ×
(Judicial quality)c
Sample:
(6)
benchmark
benchmark
2010
1970
no USA
NO
YES
YES
YES
YES
YES
YES
NO
419,401
YES
NO
YES
0.548
297,485
YES
NO
YES
0.574
453,390
YES
NO
YES
0.529
160,718
YES
NO
YES
0.525
395,666
0.062**
(5.05)
Chor
(2010)
YES
YES
NO
YES
0.596
42,332
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in industry i in
year 2000. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust standard errors are
clustered by exporter-industry. + significant at 10%, * significant at 5%, ** significant at 1%. Specification (2) includes
additional unreported probability of exports obtained from the first stage. Column (7) uses data and empirical specification
from Chor (2010)
Table 7. Instrumenting demographic composition.
(1)
Cog_appi ×
(Median age)c
Physicali ×
(Median age)c
Hansen J-statistics (p-value)
IV-GMM
-0.040**
(-2.62)
-0.057*
(-1.97)
-0.004
(-0.11)
-0.045*
(-2.54)
0.013+
(1.92)
0.017*
(1.96)
0.038**
(3.98)
0.016*
(2.50)
0.001
(0.05)
Birth rate,
1960
Religious
composition
409,910
409,545
N
0.036*
(2.32)
-0.026
(-0.82)
0.016**
(3.00)
(Skill int.)i ×
(Skill abund.)c
(4)
IV-GMM
-0.013
(-0.58)
(Capital int.)i ×
(Capital abund.)c
(3)
IV-GMM
0.037**
(5.85)
Cog_depi ×
(Median age)c
Instrument for median age
(2)
IV-GMM
0.031**
(4.95)
-0.004
(-0.09)
-0.010
(-0.55)
0.016*
(2.55)
0.031*
(2.27)
0.007
(1.05)
Genetic
distance to
France
All three
0.48
309,736
309,736
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in
industry i in year 2000. Standardized beta coefficients are reported, with t-statistics in parentheses. + significant
at 10%, * significant at 5%, ** significant at 1%. All standard errors are obtained by bootstrap with 100
replications and are clustered by exporter-industry. All specifications include exporter fixed effects, importerindustry fixed effects, and controls for bilateral trade costs. The null hypothesis for the Hansen overidentification
test is that instruments are exogenous.
Table 8. Alternative measure of the stock of cognitive skills
Memory_Ii ×
(Median age)c
Memory_Ii ×
Scorec
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
Importer-industry FE
Exporter FE
Trade costs controls
R-squared
N
(1)
-0.022**
(-4.55)
(2)
0.020*
(2.45)
(3)
-0.019*
(-2.08)
(4)
(5)
(6)
0.020*
(2.49)
0.027**
(5.90)
0.019**
(3.69)
0.052**
(6.72)
0.048**
(9.96)
0.048**
(10.15)
0.038**
(7.46)
0.057**
(4.75)
0.054**
(4.45)
0.049**
(4.01)
YES
YES
YES
YES
0.043**
(8.81)
YES
YES
0.552
413,466
0.032**
(4.14)
YES
YES
0.669
160,945
0.033**
(4.36)
YES
YES
0.640
160,945
YES
YES
0.666
150,178
0.036**
(7.10)
YES
YES
YES
0.552
413,466
0.036**
(7.06)
YES
YES
YES
0.552
413,466
Notes: The dependent variable is the normalized natural logarithm of exports from country c to country p in
industry i in year 2000. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust
standard errors are clustered by exporter-industry. + significant at 10%, * significant at 5%, ** significant at 1%.
Table 9. Extensions.
Cog_appi ×
(Median age)c
Cog_depi ×
(Median age)c
Physicali ×
(Median age)c
Cog_appi × Healthc
Cog_depi × Healthc
Physicali × Healthc
Cog_appi × Educc
Cog_depi × Educc
Physicali × Educc
(Capital int.)i ×
(Capital abund.)c
(Skill int.)i ×
(Skill abund.)c
R-squared
N
(1)
0.021**
(3.48)
-0.045**
(-6.77)
-0.016
(-0.75)
0.031**
(5.02)
(2)
0.030**
(4.80)
-0.040**
(-5.19)
-0.022
(-1.14)
-0.034
(-1.93)
0.033
(1.73)
0.003
(0.48)
0.016
(0.84)
-0.055*
(-2.45)
0.023**
(4.67)
0.026**
(3.32)
0.559
0.549
219,657
-0.006
(-0.65)
401,851
0.019*
(2.38)
Notes: The dependent variable is the normalized natural logarithm of exports
from country c to country p in industry i in year 2000. Standardized beta
coefficients are reported, with t-statistics in parentheses. Robust standard
errors are clustered by exporter-industry. + significant at 10%, * significant at
5%, ** significant at 1%. All specifications include exporter fixed effects,
importer-industry fixed effects, and controls for bilateral trade costs. Column
(2) is estimated on a subsample of industries that produce consumer goods.
Table 10. Estimates for the effect of population aging on comparative
advantage
Cog_appi ×
(ΔMedian age)c
(1)
Cog_depi ×
(ΔMedian age)c
(2)
0.035**
(3.18)
Physicali ×
(ΔMedian age)c
(Capital int.)i ×
(ΔCapital abund.)c
(Skill int.)i ×
(ΔSkill abund.)c
Importer-industry FE
Exporter FE
Trade costs controls
R-squared
N
0.010
(0.83)
-0.002
(-0.23)
YES
YES
YES
0.476
62,488
0.012
(0.90)
-0.004
(-0.39)
YES
YES
YES
0.477
61,835
(3)
-0.037**
(-3.19)
0.019
(1.39)
-0.004
(-0.40)
YES
YES
YES
0.477
61,835
(4)
-0.039**
(-3.29)
0.014
(1.03)
-0.004
(-0.45)
YES
YES
YES
0.477
61,835
(5)
0.027*
(2.34)
-0.029*
(-2.47)
0.012
(0.91)
-0.005
(-0.47)
YES
YES
YES
0.478
61,835
Notes: The dependent variable is the change in the normalized natural logarithm of exports from country
c to country p in industry i between years 2000 and 1960. Standardized beta coefficients are reported,
with t-statistics in parentheses. Robust standard errors are clustered by exporter-industry. + significant
at 10%, * significant at 5%, ** significant at 1%.
Table 11. Robustness and extensions for the effect of population aging on comparative advantage
Cog_appi ×
(ΔAge)c
Cog_depi ×
(ΔAge)c
(Capital int.)i ×
(ΔAge)c
(Skill int.)i ×
(ΔAge)c
Dependent variable:
Measure for Agec
End year
Start year
Importers
Importer-industry FE
Exporter FE
Trade costs controls
R-squared
N
(1)
-0.027*
(-2.05)
(2)
0.037**
(3.84)
(3)
0.034**
(3.71)
(4)
0.002
(0.25)
0.027*
(2.05)
-0.039**
(-4.00)
-0.030**
(-2.76)
-0.008
(-0.90)
-0.003
(-0.32)
-0.010
(-1.38)
-0.002
(-0.31)
0.007
(1.39)
0.003
(0.21)
ln(Expcpi)
0.005
(0.55)
ln(Expcpi)
0.017*
(2.03)
ln(Expcpi)
0.018*
(2.47)
ln(Expcpi)
βO
0.059**
(4.98)
(5)
-0.038**
(-3.07)
βY
-0.023
(-0.92)
-0.025
(-0.92)
0.004
(0.46)
-0.008
(-1.14)
ln(Expcpi)
(6)
0.120*
(2.00)
0.036
(0.52)
0.139+
(1.90)
0.126**
(2.85)
(Import
share)pci
young
worker share
median age
median age
median age
median age
median age
all
1970
all
1980
1990
1970
1970
2000
1962
YES
YES
YES
0.477
62,488
2000
YES
YES
YES
0.444
83,976
2000
all
YES
YES
YES
0.425
101,977
2000
all
YES
YES
YES
0.324
93,352
2000
all
YES
YES
YES
0.445
83,976
2000
USA
NO
NO
YES
0.113
2,691
Notes: The dependent variable is the change in the normalized natural logarithm of exports from country c to country p in industry i between years 2000 and
1960. Standardized beta coefficients are reported, with t-statistics in parentheses. Robust standard errors are clustered by exporter-industry. + significant at
10%, * significant at 5%, ** significant at 1%. Column (6) includes exporter and industry fixed effects.
Appendix A. Theoretical framework
6.1
A1. Heckscher-Ohlin and Ricardian models with age-dependent
skills
We illustrate the role of population age structure on a country’s comparative advantage through
the two main models of comparative advantage –the Heckscher-Ohlin and the Ricardian –using
the extension of the Eaton and Kortum (2002) model by Chor (2010). The model accommodates
both productivity and factor endowment di¤erences in a setting with multiple countries, industries, and factors of production. In this setup, for any pair of countries c1 and c2, their relative
exports of product i to country p is given by
'ic1 =mcic1 dic1p
Xc1pi
=
;
Xc2pi
'ic2 =mcic2 dic2p
(7)
where Xc1pi is the value of exports of good i from c1 to country p, dicp is the iceberg trade cost for
shipping one unit of i from c to p, and
is the inverse of productivity shock variance. The term
mcic is the unit production costs of country c in industry i which captures the Heckscher-Ohlin
forces and 'ic is the Ricardian productivity of country c in industry i. Following Chor (2010), we
parametrize the productivity term and the unit cost function as follows, distinguishing coe¢ cients
that relate to demographics with stars:
ln 'ic =
c
+
i
+
X
k
k Ii
Agec +
k2K
mcic =
Y
and
i
n
nm Li
Mcm
(8)
fn;mg
(wck )ski
k2K
c
X
Y
(wcf )sf i
f 2F
are country and industry productivity parameters, Lni and Mcm are country and
industry characteristics, such as institutional factors, which determine country’s productivity
edge in that industry, and coe¢ cients
nm
re‡ect the strength of the e¤ect of interactions Lni Mcm
on productivity. If senior workers become less productive in tasks that require age-depreciating
38
skills, productivity will also depend on the interaction of industry’s intensity in age-dependent
skill k, Iik , and a measure of a country’s demographic composition, Agec , which we assume is
increasing as a country’s population becomes older. To the extent that industries inherit age
distribution of a country, older population would imply productivity advantage for industries
which require age-appreciating skills (
depreciating skills intensively (
k
> 0) and disadvantage for industries which use age-
k
< 0).
The unit cost function is a Cobb-Douglas aggregator of factor prices in country c, where K
is a set of age-dependent skills, F is a set of other factors of production, such as human and
physical capital, and sji is the share of factor j 2 fK; F g in total costs of industry i. If the
Heckscher-Ohlin channel plays a role, then, as in , relative factor prices are inversely related to
relative factor endowments, and the log unit costs becomes
X
ln mcic =
k ski
ln Fck
X
f sf i
ln Fcf
(9)
f 2F
k2K
where Fcj is the endowment of factor j 2 fK; F g in country c measured relative to some reference
factor, and
k
1
> 0,
ln
f
> 0. Substituting (8) and (9) into (7) we obtain
Xc1pi
Xc2pi
=
X
k
k Ii
(Agec1
Agec2 ) +
k2K
X
c1
k ski
k
Fc1
k
Fc2
ln
k2K
n
nm Li
fn;mg
(
X
c2 )
nm
(Mc1
dic1p
nm
Mc2
)+
dic2p
X
f 2F
f sf i
ln
f
Fc1
f
Fc2
(10)
+
!
+
The relative exports are determined by combination of six factors: Ricardian forces, as captured by the di¤erential e¤ect of age composition and institutional factors on productivity (the
…rst and the third terms); the Heckscher-Ohlin forces, operating through the di¤erence in factor
endowments (the second and the fourth terms); productivity shifters (…fth term) and trade costs
(sixth term). On one hand, if there are no Ricardian forces in the model and population aging affects only the stock of age-dependent skills but not the quality, then the …rst and the third terms
disappear from equation (10), which becomes similar to the prediction of the Heckscher-Ohlin
39
model by Romalis (2004). On the other hand, if population aging does not a¤ect relative premia
of di¤erent age-dependent skills, then the second term vanishes and demographic composition
would a¤ect trade only through variation in labor productivity across industries.
Speci…cation (10) also allows to analyze the e¤ect of changes in demographic composition
over time on comparative advantage. It is easy to see that di¤erent rates of population aging in
countries c1 and c2 will a¤ect their relative exports through two complementary channels. First,
more rapidly aging countries should observe a decrease in endowments of age-depreciating skills
and increase in their premia. This, in turn, will shift a country’s export structure away from
industries which use age-depreciating skills intensively through the Rybczynski e¤ect. Second,
di¤erent aging rates a¤ect relative exports of two countries directly through the e¤ect on age
composition of workers and labor productivity across industries (Ricardian e¤ect).
Equation (10) demonstrates both channels, Heckscher-Ohlin and Ricardian, through which
age composition can a¤ect comparative advantage. However, separating one channel from the
other empirically would require data on either the endowments or relative prices of age-dependent
skills, which are not available. In the absence of such data, we proxy the stock of age-dependent
skills in equation (10) with the country’s median age Agec :
ln Fck =
k
0
+
k
1 Agec
(11)
so that for age-appreciating skills the stock of skills increases with population age (
for age-depreciating skills the stock decreases with age (
k
1
k
1
> 0) and
< 0 ). This transformation results in
an empirical speci…cation which is similar to Chor (2010) and Bombardini, Gallipoli, and Pupato
(2012):38
ln Xcpi =
X
k
k Ii
Agec +
k2K
X
f
f Ii
Fcf +
0
cp
+
c
+
pi
+ "cpi
(12)
f 2F
k
=
k
+
38
k
k 1
Note that both Iik and ski measure intensity of industry i in skill k and we assume that Iik = ski . In this
speci…cation we also do not consider the role of institutional factors of production but introduce some of them in
extensions.
40
In (12),
k
combines both
k
and
k,
and the interactions Iik
Agec capture both the Ricardian
and the Heckscher-Ohlin channels. The model predicts that
with age and
k
k
< 0 for skills which worsen
> 0 for skills that improve with age. This follows from the fact that for age-
depreciating skills
k
< 0 and
For age-appreciating skills both
k
1
< 0, which implies that
k
and
k
1
k
< 0 since
are positive, and so is
k
is positive for all k.
k.
A2. The role of worker heterogeneity in age-dependent skills
An important feature of the above setup is that both age-appreciating and age-depreciating skills
are bundled in workers. Although workers are perfectly mobile across industries, how skills are
bundled across workers may also matter for aggregate productivity. Ohnsorge and Tre‡er (2007)
demonstrate that in a setting with multiple skills embedded in workers, international comparative
advantage is determined not only by the relative skill endowments across countries but also by
the second moments of skill distributions. In what follows, we use the Ohnsorge and Tre‡er
(2007) framework to analyze how the presence of multiple skills a¤ects theoretical predictions
and empirical methodology of the current study.
Each worker is endowed with two skills –age-appreciating skill L and age-depreciating skill H.
Type (H; L) worker in industry i produces task T (H; L; i) ; and the aggregate industry output
is the sum of tasks of all workers it employes. Workers are paid the value of their marginal
product W (H; L; i) = P (i) T (H; L; i) where P (i) is the price of output produced by industry
i. Assuming that the task function is constant return to scale in H and L, the log of worker’s
earnings can be expressed as w (H; L; i) = p (i) + t (s; i) + l; where lower-case letters are in logs,
l = ln(L), and s = ln(H=L) is the log of worker’s relative endowment of s. Ohnsorge and
Tre‡er (2007) show that when workers choose industries where their earnings are the largest,
s determines worker’s comparative advantage across industries so that high-s workers will sort
into s-intensive industries. However, the international comparative advantage of an industry
depends not only on s but also on the amount of other factor l that workers bring in. Under the
assumption that the distribution of skills across workers is bivariate normal
41
2
4
s
l
the log output of industry i is
ln Y (i) =
si
3
5
+
l
02
s
N @4
1
2
2
l
0
32
54
2
1
2
s
s l
s l
2
l
31
5A
1
ln 2
2
+ t (si ; i)
1
2
2
s
2
si
(13)
s
and country’s comparative advantage is completely characterized by the parameters of the distribution function ; ;
s;
and
l.
For the purpose of this study, it is important to note that
is decreasing with population
aging since s re‡ects relative endowment of age-depreciating skills of a worker. Furthermore,
parameter , which measures the correlation between worker’s absolute advantage l and comparative advantage s, must also be smaller in older countries because population aging increases l
and decreases s.
In order to analyze the e¤ects of skill endowments and the correlation in worker’s absolute
and comparative advantage on output, di¤erentiate equation (13) with respect to
@ ln (Yi )
=
@
@ ln (Yi )
=
@
l
+
si
2
s
s
l
(si
)
s
1
=
2
l
2
s
=
(si
l
s l
(si
s l
and :
(14)
)
(15)
)
s
One can see that the two derivatives have the same sign on the entire domain of industry’s
factor intensity si . When both
and decrease due to population aging, Yi decreases in industries
that are intensive in age-depreciating skills (when si >
that are intensive in age-appreciating skills (when si <
39
The relative magnitude of these two e¤ects depends on
@ ln(Yi )
@
@ ln(Yi )
@
When the joint skills distribution is more dispersed (
to changes in production and trade pattern than .
s l
+
+
s l
s l
) and increases in industries
).39 Hence, for given
and the relative change in
and
s
and
l,
:
=
s l
=
42
is greater), changes in
contribute relatively more
the standard Heckscher-Ohlin prediction regarding the e¤ect of population aging on comparative
advantage continue to hold in the presence of multiple age-dependent skills per worker. In
such case, changes in both the means of age-dependent skill endowments and their covariances,
caused by demographic transformations, have the same e¤ect on comparative advantage. As a
result, a country’s median age used in the empirical analysis will pick up both channels through
which demographic changes a¤ect comparative advantage in industries with di¤erent intensities
in age-dependent skills.
We next turn to the analysis of the e¤ect of variance in skill distributions on comparative
advantage. Di¤erentiating (13) with respect to
s
@ ln (Yi )
=
@ l
@ ln (Yi )
=
@ s
It is easy to see that
in
l
@ ln(Yi )
@ l
l
(
l
si )
2
s
and
l,
1
2
1
+
we obtain:
(16)
(
si )2
(17)
3
s
s
is always positive and independent of si , hence, any possible change
does not a¤ect industrial comparative advantage.
In contrast,
@ ln(Yi )
@ s
is non-monotone in si , so that if
s
is not constant the exact e¤ect of
population aging on output depends on the e¤ect of aging on
s.
The variance of relative
endowment of age-depreciating skills can be expressed as follows:
2
s
= var (H) (E (1=L))2 + var (1=L) (E (H))2 + var (H) var (1=L)
We know that both E (H) and E (1=L) decrease as population is getting older, however, we
have no priors about the e¤ect of demographic changes on variances of factor endowments and
without micro-data on joint skills distribution we can not tell the magnitude of changes in var (H)
2
s
and var (1=L). Assume that var (H) and var (1=L) are constants over age cohorts, then
is
decreasing in the median age of a country’s population.40
When
s
< 0;
@ ln(Yi )
@ s
s
is an inverse U-shape curve, and the combined impact of
40
;
If the relationship between
s and population aging is positive, the Heckscher-Ohlin prediction continues
to hold for all parameter values of the model as long as s is not too big.
43
and
s
and
can be non-monotone for some values of si . At the right tail of si distribution,
s
;
work together to reduce comparative advantage in these industries. However, at
the left tail of si , the impact of
s
o¤sets the e¤ect of
and
and even dominates when
si is small enough. Figure 3A illustrates the pattern of comparative advantage for the following
parameter values:
= 1,
=
0:5;
s
= 1,
l
= 1 and
+
= 1. For this parametrization
of the model, the total impact of population aging can be summarized as
@ ln (Yi )
@
= (si
ln (Yi ) =
Figure 3A shows how
si )2
s
ln (Yi ) varies across industries for di¤erent values of parameter
=
2 [0; 1], which measures the relative importance of monotonic and non-monotonic forces.
s
+
If
@ ln (Yi )
@ ln (Yi )
+
s
@
@ s
)(
+
)+
si 1 + (
+
s
is small relative to
+
( < 0:25) and
s
is not extremelly large or low, the main
prediction of the HO model is preserved even if population aging a¤ects all three parameters
of skill distribution –an ageing country gains comparative advantage at the lower tail of factor
intensity distribution and loses at the upper tail. However, in the unlikely event when
s
is relatively large and dominates the e¤ect of population aging on the mean of relative skill
endowments, or when
> 0:25, the prediction of the HO model does not hold any more for lower
values of si distribution: an ageing country gains comparative advantage only in the middle
range of factor intensity and loses at the both ends of si distribution.
Appendix B. Age-dependent skills
Age and cognitive skills
Of all studies on aging and cognition we primarily focus on those which utilize both cross-sectional
and longitudinal analysis. Studies that derive their results from a pure cross-sectional comparison
of individuals of di¤erent age, may not accurately distinguish the e¤ect of chronological age on
individual’s performance from the e¤ect of fundamental di¤erences between age cohorts which
44
may arise from changes in social or cultural environment. At the same time, longitudinal studies
that test the same individuals multiple times arguably underestimate the e¤ect of aging on
cognitive skills due to selection and the positive e¤ect of re-testing on test performance, even if
the two tests are many years apart.
One of the …rst comprehensive research programs on cognitive abilities and aging has been
conducted by Schaie (1986, 1994, 2000, 2005) in the Seattle Longitudinal Study (SLS). The SLS
has followed several cohorts of adults aged 25 to 88, with more than 5,000 subjects in total.
Participants were …rst tested in 1956 and then re-tested at every 7-year interval along with the
new cohort added to the study in every cycle. The test battery included the measures of verbal
comprehension, verbal memory, spatial orientation, inductive reasoning, numerical ability, and
perceptual speed. The results of this program demonstrate that not all cognitive skills change
uniformly through adulthood. Cross-sectional comparison of score means of individuals from different age groups shows that inductive reasoning and spatial orientation decrease monotonically
with age starting at 25, whereas verbal and numerical abilities peak at mid-life and decline slowly
afterwards. The longitudinal analysis reveals that all scores, except for perceptual speed, initially
increase, and then level o¤ for verbal ability and decrease for reasoning, spatial orientation, and
numerical ability after 60-65.
More recent works support the main message of the earlier studies that not all cognitive skills
decline uniformly with age. Salthouse (1998, 2009, 2010) reports results of another laboratory
testing with over 2,000 participants tested twice in 1-7 years intervals. He con…rmed that in the
cross sectional comparison nearly all cognitive abilities are found to be monotonically declining
with age, including memory, reasoning, and space orientation. Only vocabulary was found to be
improving throughout the adulthood. Results from the longitudinal analysis, adjusted for prior
testing experience, paint a di¤erent picture: only memory demonstrates considerable decline
with age, while reasoning and vocabulary remain stable throughout lifecycle. In another study
by Singh-Manoux (2012), 7,400 participants aged 45-70 were tested three times every …ve years
beginning in 1997. The authors found, both in cross-sectional and longitudinal analysis, a signi…cant decline in memory and reasoning between 45 and 55 years, and slight but statistically
signi…cant improvement in vocabulary.
45
Overall, the research on age and cognitive development seems to reach a consensus on a
number cognitive abilities. First, nearly all studies that examine the e¤ect of aging on speech
and language abilities …nd that they improve with age. Older individuals usually have more
extensive vocabularies and are more skilled conversationalists than younger adults, and older
people out-perform younger people on tasks that involve language and writing skills. Second,
there is a general consensus in the literature that memory declines signi…cantly with age. Older
adults exhibit di¢ culties with activities that involve manipulations of the content of their working
memory. Third, while older adults experience no di¢ culties in maintaining selective attention on
a particular activity without being distracted by others (McDowd and Shaw, 2000; Verhaeghen
and Cerella, 2002), divided attention is associated with signi…cant age-related decline. When
individuals are required to process information from two or more sources at the same time
or to switch from one task to another, older adults are more a¤ected by the cost of dividing
attention into multiple tasks (McDowd and Craik, 1988; Tsang, 1998; McDowd and Shaw, 2000;
Verhaeghen and Cerella, 2002).41
Finally, there is little disagreement in the literature that the speed of information processing
is decreasing with age. Schaie (1994, 2005) and Salthouse (2010) identify perceptual speed as
one of the abilities associated with the most rapid age-related decline which begins between ages
25 and 32. In fact, many researchers argue that much of the evidence for age-related decline in
some cognitive abilities can be explained by the slowing of information processing. Salthouse
has demonstrated in numerous works that an age-related decline in some cognitive tasks, such as
reasoning, can be accounted for by decreasing speed of information processing. Moreover, some
studies emphasize that older individuals may show poorer results on the measures of reasoning
because the tests on those measures are administered with tight time limits and may thus pick
up the speed of information processing e¤ect. However, Salthouse suggests that the e¤ect of
aging on memory is independent of processing speed.
Therefore, in our empirical analysis we primarily focus on memory, language skills, divided
attention, and processing speed as the four cognitive skills that are known to change substantially
with age.
41
McDowd and Shaw (2000) have demonstrated that devided attention impairment is a signi…cant factor of
higher car accidents among adults.
46
Aging and physical abilities
A decline in physical strength with age is well documented in the medical literature (see de Zwart
and Frings-Dresen (1995) and Hedge (2005) for a survey). The loss of strength is a natural process
caused by changes in body composition with a reduction in muscle mass being the main factor.
Beginning in the late 20s, the lean muscle mass start to decrease and the fat levels start to
increase. Moreover, the quality of muscle mass deteriorates: the decline in strength is more
rapid than the reduction in muscle mass and the maintaining of muscle mass does not prevent
age-declining muscle strength.
47
Appendix C. Construction of industry-level measures of
skill intensities
Industry-level measures of intensity in cognitive skills are constructed using two sources of data.
First, from the O*NET database we retrieve measures of the importance of di¤erent cognitive
skills for all occupational types, recorded at 7-digit Standard Occupational Classi…cation (SOC).
O*NET surveys workers and experts in di¤erent occupations and collects their responses to
questions about the importance of various skills and abilities for successful performance. The
responses, ranked on a scale from 1 to 5, are averaged across respondents in each occupation.
Second, for every 4-digit NAICS industry, we obtain occupational composition using Occupational Employment Statistics (OES) Survey, maintained by the US Bureau of Labor Statistics.
The OES database includes information on the number of workers in all 7-digit SOC occupational categories for the year 2012. This allows us to construct the share of each occupation in
total employment in every industry, which we match to O*NET scores by SOC. Finally, using
occupational employment shares as weights, we generate industry-level measures of skill intensity as the weighted average of importance of that skill across occupations within an industry.
Therefore, the cross-industry variation in skill intensity is driven by the variation in occupational
composition across industries.
We rely on the O*NET database to construct measures of intensity in four age-dependent
cognitive skills described in Appendix B. Below we explain which O*NET skill indicators were
used to construct those measures.
Communication skills ranking across occupation is based on the following four O*NET indicators: oral comprehension - the ability to listen to and understand information and ideas
presented through spoken words and sentences; oral expression - the ability to communicate
information and ideas in speaking so others will understand; written comprehension - the ability
to read and understand information and ideas presented in writing; and written expression - the
ability to communicate information and ideas in writing so others will understand.
Memory: memorization - the ability to remember information such as words, numbers, pic48
tures, and procedures.
Divided attention: time sharing - the ability to shift back and forth between two or more
activities or sources of information such as speech, sounds, touch, or other sources.
Speed of information processing: perceptual speed - the ability to quickly and accurately
compare similarities and di¤erences among sets of letters, numbers, objects, pictures, or patterns;
and speed of closure - the ability to quickly make sense of, combine, and organize information
into meaningful patterns.
Industry-level measures of intensity in physical skills are constructed in the same way as
cognitive skills. The following questions about physical abilities were used from the O*NET to
generate nine measures of physical skill intensities: dynamic ‡exibility — the ability to quickly
and repeatedly bend, stretch, twist, or reach out with your body, arms, and/or legs; dynamic
strength — the ability to exert muscle force repeatedly or continuously over time (this involves
muscular endurance and resistance to muscle fatigue); explosive strength — the ability to use
short bursts of muscle force to propel oneself (as in jumping or sprinting), or to throw an object;
extent ‡exibility — the ability to bend, stretch, twist, or reach with your body, arms, and/or
legs; gross body coordination — the ability to coordinate the movement of your arms, legs, and
torso together when the whole body is in motion; gross body equilibrium — the ability to keep
or regain your body balance or stay upright when in an unstable position; stamina — the ability
to exert yourself physically over long periods of time without getting winded or out of breath;
static strength — the ability to exert maximum muscle force to lift, push, pull, or carry objects;
trunk strength — the ability to use your abdominal and lower back muscles to support part of
the body repeatedly or continuously over time without ‘giving out’or fatiguing.
A measure of intensity in age-neutral cognitive skills was constructed based on the following
O*NET indicators: category ‡exibility — the ability to generate or use di¤erent sets of rules for
combining or grouping things in di¤erent ways; deductive reasoning — the ability to apply general
rules to speci…c problems to produce answers that make sense; ‡exibility of closure — the ability
to identify or detect a known pattern (a …gure, object, word, or sound) that is hidden in other
distracting material; ‡uency of ideas — the ability to come up with a number of ideas about a
49
topic (the number of ideas is important, not their quality, correctness, or creativity); information
ordering — the ability to arrange things or actions in a certain order or pattern according to
a speci…c rule or set of rules (e.g., patterns of numbers, letters, words, pictures, mathematical
operations); mathematical reasoning — the ability to choose the right mathematical methods or
formulas to solve a problem; number facility — the ability to add, subtract, multiply, or divide
quickly and correctly; originality — the ability to come up with unusual or clever ideas about a
given topic or situation, or to develop creative ways to solve a problem; problem sensitivity —
the ability to tell when something is wrong or is likely to go wrong. It does not involve solving
the problem, only recognizing there is a problem; spatial orientation — the ability to know your
location in relation to the environment or to know where other objects are in relation to you;
visualization — the ability to imagine how something will look after it is moved around or when
its parts are moved or rearranged.
50
Appendix D. Measuring country-level endowment in memory using standardized memory tests
This appendix explains the details of construction of a proxy measure for a country’s stock of
cognitive skills which does not depend on the age structure. The measure that we propose relies
on the variation in cognitive functioning across countries and is available from the international
standardized tests of cognitive abilities. These tests are based on comparable cross-national
household surveys conducted in di¤erent parts of the world and include Survey of Health, Aging
and Retirement in Europe (SHARE), World Health Organization study on global aging and adult
health (SAGE), English Longitudinal Study of Aging (ELSA), and Health and Retirement Study
(HRS). These surveys cover 27 di¤erent countries: Austria, Belgium, Czech Republic, Denmark,
Estonia, France, Germany, Greece, Hungary, Ireland, Israel, Italy, Netherlands, Poland, Portugal,
Slovenia, Spain, Sweden, and Switzerland in the SHARE survey; China, Ghana, India, Mexico,
Russia, and South Africa in the SAGE survey; UK in the ELSA survey; and USA in the HRS
survey. Although the sample is overrepresented by countries from the top quantile of the median
age distribution, four countries from the …rst and second quantiles (China, Ghana, India, and
Mexico) are also present. In this study we use data from the last wave of surveys conducted in
2010-11.
The target population includes individual aged 50 years and over and their spouses. Participants of all four surveys take a standardized short-term memory test which consists of verbal
registration and immediate recall of a list of ten simple words (e.g. sky, ocean, home, butter,
child, etc.). After one minute of asking other survey questions the respondents are asked to recall
the nouns previously presented as part of the immediate recall task. The test score is the count
of the number of words that were recalled correctly and ranges from zero to ten. Tests of other
cognitive skills, such as numeracy or long-term memory test, are not covered in all surveys and
are thus not included in our study. We standardized the memory test score to 0-1 interval.
Along with the tests, survey participants also …ll out a questionnaire which provides detailed
information on age, gender, household composition, health, social status, income and assets,
employment, etc. Unfortunately, most of these questions are either not included in every survey
51
or not measured in a comparable way. As a measure of individual’s health we use a self-perceived
health condition assessment, which is the only health-related indicator measured consistently
across surveys. The respondents were asked to rank their health as excellent, very good, good,
fair, or poor. For education, we use two outcome variables. The …rst one is the number of
years of formal education. The second measure is based on the highest education level, which
we categorized into four groups: less than primary, primary school completed, secondary school
completed, post-secondary.
Figure 1A illustrates the relationship between the word recall test scores and age for selected
countries. As expected, there is a strong age-related decline in short-term memory in all countries as measured by the proportion of words recalled correctly within one minute. Furthermore,
this negative relationship is preserved when we control for the e¤ect of gender, education, and
self-reported health status on test outcomes in Figure 2A. Figure 2A also reveals substantial
di¤erences in the levels of test scores across countries even after controlling for cross-country differences in educational attainment, which suggests the presence of unobservable country-speci…c
determinants of cognitive performance.
52
Appendix Tables and Figures
5.5
5
4
4.5
Memory test score
6
6.5
Figure 1A. The relationship between unconditional word recall test score and age.
50
70
65
60
55
Age
China
Germany
Russia
India
Mexico
USA
Note: Curves are smoothed by quadratic trends
5
4.5
4
Memory test score
5.5
Figure 2A. The relationship between conditional word recall test score and age.
50
55
60
65
70
Age
China
Germany
Russia
India
Mexico
USA
Note: Curves are smoothed by linear trends
Figure 3A. Comparative advantage and second moments of skill distribution
Table 1A. Correlation between industry-level intensities in cognitive
skills
(1)
(2)
(3)
(4)
1
(1) Oral comprehension
0.98
1
(2) Oral expression
0.93 0.91
1
(3) Written comprehension
0.96 0.94 0.97
1
(4) Written expression
0.91 0.92 0.87 0.91
(5) Memorization
0.55 0.57 0.61 0.61
(6) Time sharing
0.43 0.49 0.44 0.43
(7) Perceptual speed
0.69 0.74 0.73 0.73
(8) Speed of closure
(5)
(6)
(7)
(8)
1
0.66
0.57
0.84
1
0.80
0.83
1
0.84
1
(6)
(7)
(8)
(9)
1
0.91
0.88
0.84
1
0.98
0.98
1
0.98
1
Note: correlation coefficients are calculated over 86 3-digit manufacturing NAICS industries.
Table 2A. Correlation between industry-level intensities in physical abilities
(1)
(2)
(3)
(4)
(5)
1
(1) Dynamic flexibility
0.73
1
(2) Dynamic strength
0.82 0.73
1
(3) Explosive strength
0.72 0.95 0.65
1
(4) Extent flexibility
0.72 0.97 0.68 0.95
1
(5) Gross body coordination
0.68
0.89
0.64
0.87
0.91
(6) Gross body equilibrium
0.74 0.97 0.70 0.96 0.98
(7) Stamina
0.76 0.98 0.74 0.95 0.96
(8) Static strength
0.74 0.97 0.73 0.95 0.96
(9) Trunk strength
Note: correlation coefficients are calculated over 86 3-digit manufacturing NAICS industries.
Table 3A. Principle component analysis
Variance
Panel A. Age-appreciating cognitive skills
2.26
Factor1
1.05
Factor2
Factor3
Factor4
0.49
0.19
Panel B. Age-depreciating cognitive skills
3.04
Factor1
0.59
Factor2
0.25
Factor3
0.13
Factor4
Panel C. Physical abilities
Factor1
Factor2
Factor3
Factor4
Factor5
Factor6
Factor7
Factor8
Factor9
7.06
1.46
0.24
0.12
0.05
0.03
0.02
0.01
0.01
Proportion of
explained
variance
0.57
0.26
Factor 1
components
0.482
0.538
0.05
Oral comprehension
Oral expression
Written
comprehension
Written expression
0.76
0.15
0.06
0.03
Memorization
Time sharing
Perceptual speed
Speed of closure
0.416
0.525
0.512
0.538
0.12
0.78
0.16
0.03
0.01
0.01
0.00
0.00
0.00
0.00
Dynamic flexibility
Dynamic strength
Explosive strength
Extent flexibility
Gross body
coordination
Gross body
equilibrium
Stamina
Static strength
Trunk strength
0.464
0.513
0.205
0.366
0.194
0.363
0.367
0.337
0.370
0.370
0.365
Table 4A. Intensity in age-dependent skills by 4-digit NAICS industries
4-digit
AgeAgeIndustry Name
NAICS
appreciating
depreciating
code
cognitive skills
cognitive skills
Value
Rank
Value
Rank
Animal food
Grain and oilseed milling
Sugar and confectionery product
Fruit and vegetable preserving and
specialty food manufacturing
Dairy product
Meat product
Seafood product preparation and
packaging
Bakeries and tortilla
Other food
Beverage
Tobacco
Fiber, yarn and thread mills
Fabric mills
Textile and fabric finishing and fabric
coating
Textile furnishings mills
Other textile product mills
Clothing knitting mills
Cut and sew clothing
Clothing accessories and other clothing
Leather and hide tanning and finishing
Footwear
Other leather and allied product
Sawmills and wood preservation
Veneer, plywood and engineered wood
product
Other wood product
Pulp, paper and paperboard mills
Converted paper product
Printing and related support activities
Petroleum and coal product
Basic chemical
Resin, synthetic rubber, and artificial and
synthetic fibers and filaments
Pesticide, fertilizer and other agricultural
chemical
Pharmaceutical and medicine
Paint, coating and adhesive
Soap, cleaning compound and toilet
preparation
Other chemical product
Plastic product
Rubber product
3111
3112
3113
Rank
8
1.05
22
32
14
6
1.23
0.31
3
40
0.88
1.63
12
2
23
7
5
45
1
25
0.68
0.87
0.62
-0.14
1.72
0.81
21
13
25
57
1
15
7
50
38
1.09
0.72
0.82
4
14
9
0.64
-0.99
23
70
-0.34
3117
0.11
3115
3116
Value
1.46
0.00
0.20
3114
57
41
Physical ability
0.85
0.32
39
0.66
0.44
0.84
1.25
3
3118
3119
3121
3122
3131
3132
0.82
0.74
2.30
1.35
-1.48
-0.74
15
22
1
8
78
66
0.53
0.90
1.07
0.19
3.12
0.53
3141
3149
3151
3152
3159
3161
3162
3169
3211
-1.38
-1.24
-1.85
-1.88
-1.84
-1.61
-1.30
-1.52
-2.18
76
74
84
85
83
81
75
80
86
0.77
10
0.92
0.39
1.23
0.13
0.39
1.02
0.37
0.42
1.24
10
35
5
48
36
7
39
34
4
-0.69
-1.76
-0.92
1.72
-0.97
0.61
68
6
19
43
44
32
28
70
46
21
3219
3221
3222
3231
3241
3251
-0.95
1.02
0.60
0.28
0.26
0.40
0.51
-0.49
0.19
0.57
65
82
67
4
69
25
0.74
0.34
0.11
-0.18
-0.75
-0.40
11
38
49
60
74
65
0.91
0.38
0.30
-0.33
-1.11
-1.05
11
38
42
62
78
74
3253
1.30
10
3254
3255
81
52
-1.87
-0.39
66
3
-2.09
0.06
-0.43
1.73
12
2
37
3256
0.87
2.23
0.34
81
65
3259
3261
3262
1.65
0.35
-0.35
5
33
58
0.40
0.73
0.38
33
13
34
-0.11
0.62
0.72
55
26
18
3133
3212
3252
-0.38
0.36
60
32
0.48
-0.10
0.31
30
58
41
0.37
0.93
-0.62
-0.02
40
9
70
53
Clay product and refractory
Glass and glass product
Cement and concrete product
Lime and gypsum product
Other non-metallic mineral product
Iron and steel mills and ferro-alloy
Steel product manufacturing from
purchased steel
Alumina and aluminum production and
processing
Non-ferrous metal (except aluminum)
production and processing
Foundries
Forging and stamping
Cutlery and hand tool manufacturing
Architectural and structural metals
Boiler, tank and shipping container
Hardware
Spring and wire product
Machine shops, turned product, and
screw, nut and bolt
Coating, engraving, heat treating and
allied activities
Other fabricated metal product
manufacturing
Agricultural, construction and mining
machinery
Industrial machinery
Commercial and service industry
machinery
Ventilation, heating, air-conditioning and
commercial refrigeration equipment
Metalworking machinery
Engine, turbine and power transmission
equipment
Other general-purpose machinery
Computer and peripheral equipment
Communications equipment
Audio and video equipment
Semiconductor and other electronic
component
Navigational, measuring, medical and
control instruments
Manufacturing and reproducing magnetic
and optical media
Electric lighting equipment
Household appliance
Electrical equipment
Other electrical equipment and
component
Motor vehicle
Motor vehicle body and trailer
Motor vehicle parts
3271
3272
3273
3274
3279
3311
0.23
0.03
0.77
0.11
0.09
-1.52
37
48
19
40
42
79
0.28
0.36
1.46
0.74
0.52
0.61
42
35
2
12
27
17
0.46
0.69
0.95
0.79
0.38
0.60
31
20
8
17
37
28
3313
-1.13
71
0.61
18
0.63
24
3315
3321
3322
3323
3324
3325
3326
-1.42
-0.44
0.39
-0.14
-0.41
0.81
0.05
77
64
31
51
61
18
46
0.54
-0.02
0.12
-0.04
0.08
-0.05
0.16
3328
-0.36
59
0.52
3312
3314
3327
3329
3331
0.07
-0.32
-0.28
0.56
0.33
56
55
0.69
0.53
0.44
27
34
-0.18
15
24
0.42
0.33
22
54
48
55
50
56
47
0.69
0.03
-0.16
0.24
0.18
-0.22
0.16
26
0.56
31
0.00
33
41
19
50
58
43
45
59
47
52
30
-0.43
61
-1.19
66
-0.31
61
78
-0.37
-1.44
80
64
3332
3333
1.49
14
6
-1.17
77
-1.43
79
3334
3335
0.29
-0.19
62
-0.07
54
3336
0.04
36
-0.15
53
-0.70
73
-0.83
73
3344
0.15
3339
3341
3342
3343
0.82
44
0.75
0.63
0.76
0.32
-0.29
63
71
86
85
79
-0.82
-3.69
-2.91
-1.11
72
86
85
77
39
-2.49
83
-2.30
83
-2.81
84
-2.79
84
9
-0.48
68
-0.57
68
75
-0.59
69
1.34
3351
3352
3353
0.53
0.57
-0.42
28
26
62
-0.79
0.07
-1.45
76
51
80
3361
3362
3363
0.09
0.01
-0.15
43
49
52
0.64
0.35
-0.15
16
36
59
0.06
67
-0.67
-3.60
-2.93
-1.29
17
3359
-0.50
21
24
20
35
3345
3346
0.81
47
45
-0.78
-0.66
0.06
-1.08
0.64
0.61
0.00
71
49
76
23
27
51
Aerospace product and parts
Railroad rolling stock
Ship and boat building
Other transportation equipment
Household and institutional furniture and
kitchen cabinet
Office furniture (including fixtures)
Other furniture-related product
Medical equipment and supplies
Other miscellaneous
3364
3365
3366
3369
-0.19
0.40
-1.15
0.86
54
30
72
13
-2.27
-0.33
-0.46
-0.48
82
64
67
69
-2.21
-0.13
0.17
-0.31
82
56
46
60
3372
3379
3391
3399
-0.44
0.51
0.96
0.82
63
29
11
16
0.04
0.59
-0.68
-0.10
53
20
72
57
0.23
0.58
-1.05
-0.36
44
29
75
63
3371
-1.17
73
0.49
29
0.80
16
Notes: For each skill, the first column reports the value of the standardized skill-intensity; the second column reports industry's rank in intensity of that skill.
Ten most and ten least skill intensive industries are market with red and blue colors, respectively.
Table 5A. Median age in 1962 and 2000 for Countries
Median Median Change
Country Name
age,
age,
19622000
1962
2000
41.3
25.5
15.81
1
Japan
2
Italy
40.2
31.6
8.64
5
Finland
39.4
28.4
11.00
32.7
5.90
3
4
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
Bulgaria
Sweden
Belgium
Croatia
Switzerland
Hungary
Denmark
Greece
Austria
Slovenia
Estonia
Latvia
Portugal
United Kingdom
Spain
Ukraine
Czech Republic
France
Netherlands
Luxembourg
Norway
Canada
Russian Federation
Hong Kong
Malta
Lithuania
Australia
Poland
USA
Romania
New Zealand
Singapore
Slovakia
Barbados
Iceland
Cuba
Ireland
Korea, Republic of
39.7
39.4
39.1
39.0
38.6
38.6
38.4
38.3
38.2
38.0
37.9
37.9
37.8
37.6
37.6
37.6
37.6
37.6
37.3
37.3
36.9
36.8
36.5
36.5
35.9
35.9
35.4
35.3
35.3
34.4
34.3
34.1
33.9
33.5
32.8
32.8
32.5
32.1
30.4
36.0
35.0
29.2
32.2
33.0
29.1
35.6
29.3
32.2
32.3
27.9
35.5
29.4
28.9
33.1
33.0
28.7
35.2
34.3
9.27
3.37
4.09
9.79
6.41
5.40
9.22
2.59
8.68
5.73
5.67
9.82
2.16
8.24
8.73
4.46
4.54
8.60
2.14
2.53
26.5
10.33
23.1
13.39
29.6
5.77
27.2
21.5
28.5
26.6
29.6
28.4
27.4
9.34
14.44
7.39
8.75
5.72
5.93
6.89
18.8
15.33
25.4
7.37
27.7
22.4
22.9
30.1
19.8
6.27
11.17
9.89
2.41
12.30
Country Name
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
Malaysia
Venezuela
Guyana
South Africa
India
Mexico
Peru
Ecuador
Kyrgyzstan
Morocco
Libya
Fiji
Mongolia
Egypt
Algeria
Saudi Arabia
Bangladesh
Iran
El Salvador
Philippines
Paraguay
Bolivia
Tonga
Botswana
Papua New
Guinea
Nepal
Jordan
Gabon
Namibia
Belize
Haiti
Pakistan
Ghana
Syria
Nicaragua
Congo
Lesotho
Côte d'Ivoire
Maldives
Central African
Rep.
Cambodia
Median
age,
2000
23.8
Median
age,
1962
17.6
Change
19622000
6.25
23.2
19.7
3.49
23.3
23.2
23.0
23.0
22.9
22.9
22.5
22.3
22.2
22.1
22.0
22.0
21.6
21.1
21.0
20.9
20.7
20.5
20.4
20.0
19.9
19.8
19.6
19.6
19.5
19.5
19.5
19.3
19.1
18.9
18.9
18.8
18.8
18.8
18.7
18.7
18.5
18.5
18.4
17.2
17.0
20.3
17.1
18.4
18.6
6.04
6.23
2.76
5.87
4.45
4.33
24.0
-1.51
15.8
6.31
18.1
19.5
4.20
2.72
23.0
-1.04
18.5
2.54
19.8
17.6
18.6
19.5
17.5
16.5
16.1
18.7
17.3
17.2
18.8
2.15
4.01
2.38
1.45
3.23
3.95
4.36
1.27
2.63
2.59
0.80
20.2
-0.62
19.3
0.15
18.0
27.1
17.8
1.51
-7.53
1.52
19.8
-0.68
17.1
1.74
19.8
17.7
16.3
-0.91
1.25
2.51
19.5
-0.71
20.5
-1.97
18.3
19.2
21.4
17.2
0.38
-0.52
-2.90
1.19
42
43
44
45
46
47
48
49
50
Cyprus
31.8
23.0
8.76
110
Qatar
30.3
19.4
10.92
113
16.7
12.31
Uruguay
Armenia
Thailand
China
Mauritius
Chile
51
Kuwait
United Arab
Emirates
54
Trinidad and Tobago
52
53
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Israel
Argentina
Kazakhstan
Sri Lanka
Bahrain
Albania
Brunei Darussalam
Brazil
Costa Rica
Turkey
Tunisia
Jamaica
Indonesia
Panama
Vietnam
Colombia
31.6
30.3
30.1
29.6
29.0
28.7
28.6
28.2
28.0
27.9
27.8
27.7
27.5
26.5
26.2
25.6
25.3
24.8
24.7
24.6
24.5
24.5
24.4
24.2
23.8
28.9
22.5
18.7
21.3
20.6
24.0
18.5
24.1
26.8
18.7
22.8
19.2
19.6
19.5
19.4
18.6
18.0
19.1
18.3
19.6
20.2
18.0
21.8
16.9
2.69
7.87
11.36
8.27
8.12
4.57
9.71
3.90
1.01
9.09
4.84
8.33
6.84
6.67
6.22
6.75
6.76
5.66
6.28
4.87
4.25
6.39
2.36
6.95
111
112
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
Honduras
18.4
17.1
Tajikistan
18.2
21.9
Mauritania
Iraq
Sierra Leone
Liberia
Lao PDR
Mozambique
Sudan
Guatemala
Togo
Kenya
Tanzania
Swaziland
Benin
Senegal
Cameroon
Malawi
Zambia
Gambia
Mali
Rwanda
Niger
Yemen
Burundi
Uganda
Afghanistan
18.4
18.3
18.1
18.1
18.0
17.9
17.9
17.6
17.5
17.4
17.4
17.2
17.2
17.2
17.1
17.0
16.9
16.9
16.8
16.4
16.1
15.6
15.4
15.3
15.3
17.7
1.31
0.69
19.6
-1.33
19.0
-0.92
20.9
19.1
18.8
17.6
17.2
-3.75
-2.79
-1.05
-0.91
0.36
0.48
18.5
-0.96
17.4
-0.21
20.1
-2.94
17.2
17.0
20.9
18.4
17.0
17.4
20.8
19.9
16.3
16.2
19.4
18.4
17.1
18.0
0.26
0.36
-3.74
-1.27
0.08
-0.50
-3.92
-3.19
0.15
-0.13
-3.82
-2.94
-1.71
-2.77
Table 6A. First stage results for IV regressions in Table 7
Instrumented variable:
Cog_appi ×
(1960 birth rate)c
Cog_depi ×
(1960 birth rate)c
Physicali ×
(1960 birth rate)c
Cog_appi ×
Cog_appi × (Median age)c
(1)
-0.843
(-0.064)***
(2)
0.011
(0.213)
-0.040
(-0.232)
Cog_depi ×
Physicali ×
0.114
(0.185)
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
Cog_appi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Cog_depi ×
Physicali ×
Cog_appi ×
(1960 birth rate)c
Cog_depi ×
(1960 birth rate)c
Physicali ×
(1960 birth rate)c
Cog_appi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
Cog_depi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
Physicali ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
Cog_appi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Cog_depi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Physicali ×
0.093
(0.095)
-0.001
(-0.281)
0.446
(0.160)**
-0.016
(-0.214)
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
F-stat
Instrumented variable:
-0.065
(-0.356)
-0.033
(-0.171)
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
112.52
(1)
-0.024
(-0.073)
28.68
0.045
(0.106)
0.000
(0.291)
-0.004
(-0.309)
(3)
(4)
-0.008
(-0.082)
24.77
Cog_depi × (Median age)c
(2)
0.012
(0.300)
0.111
(0.236)
-0.854
(-0.145)***
0.023
(0.201)
(4)
0.743
(0.104)***
0.042
(0.336)
0.543
(0.069)***
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
(3)
116.52
-0.798
(-0.181)***
0.029
(0.236)
0.055
(0.071)
-0.005
(-0.063)
-0.077
(-0.191)
-0.060
(-0.197)
0.564
(0.175)***
0.138
(0.186)
0.071
(0.105)
0.515
(0.272)*
-0.085
0.010
(0.070)
0.079
(0.192)
0.006
F-stat
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Instrumented variable:
Cog_appi ×
(1960 birth rate)c
Cog_depi ×
(1960 birth rate)c
Physicali ×
(1960 birth rate)c
Cog_appi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
93.35
26.49
(3)
(4)
-0.014
(-0.080)
26.98
(2)
0.009
(0.181)
-0.840
(-0.166)***
Physicali ×
-0.720
(-0.191)***
0.060
(0.070)
0.454
(0.188)**
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
Cog_appi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Cog_depi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
Physicali ×
98.37
109.58
-0.043
(-0.216)
0.006
(0.061)
0.020
(0.173)
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
F-stat
(0.203)
Physicali × (Median age)c
(1)
-0.028
(-0.071)
Cog_depi ×
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 )
(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
(-0.287)
28.9
-0.025
(-0.181)
0.074
(0.076)
0.105
(0.192)
0.014
(0.068)
0.040
(0.197)
-0.005
(-0.187)
24.82
104.79
0.371
(0.201)*
0.090
(0.199)
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 ) is obtained as the fitted values from regression of the median age in country c
Notes: (𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
1
�𝑎𝑎𝑎𝑎𝑎𝑎𝑐𝑐 ) is obtained as the fitted values from regression of the
on religious dummy variables. (𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀
2
median age in country c on its genetic distance from France.
Table 7A. The effect of age, education and health on memory test scores
Age
Male
Education:
Years of schooling
(1)
-0.006**
(-16.96)
-0.028**
(-3.59)
Primary
(2)
-0.005**
(-14.88)
-0.038**
(-6.36)
0.011**
(10.86)
Secondary
Post-Secondary
Self-assessed health:
Excellent
(3)
-0.005**
(-14.48)
-0.035**
(-5.11)
0.042*
(2.77)
0.101**
(8.60)
0.171**
(16.01)
Very good
Good
Fair
Country FE
R-squared
N
YES
0.101
80,673
YES
0.161
65,296
YES
0.144
80,673
(4)
-0.004**
(-12.53)
-0.040**
(-6.66)
0.006**
(5.78)
0.020
(1.41)
0.055**
(3.68)
0.094**
(5.32)
0.104**
(17.57)
0.101**
(20.70)
0.075**
(17.82)
0.046**
(11.44)
YES
0.187
65,296
Notes: The dependent variable is the memory test score, measured as the fraction of correctly recalled
words by an individual. The sample is restricted to individuals of 50-70 years of age. The excluded
education category is "less than primary" and the excluded health category is "poor". * significant at 5%,
** significant at 1%. Robust standard errors are clustered by country.
Table 8A. Correlation between factor intensities and the gap in expenditure shares of 55-65 and
25-35 age cohorts
(1)
(2)
(3)
(4)
(5)
(6)
1
(1) cog_app
-0.28
1
(2) cog_dep
-0.18
0.93
1
(3) physical
0.64
-0.02
0.11
1
(4) capital intensity
0.59
-0.49
-0.47
0.34
1
(5) skill intensity
difference in consumption shares
(6)
0.13
0.13
0.12
0.23
-0.34
1
between 55-65 and 25-35 cohorts
Note: Spearman rank correlation coefficients are calculated over 14 consumption product categories.