1990a Bpea Cutler Poterba Sheiner Summers Akerlof

DAVID M.
CUTLER
MassachusettsInstituteof Technology
JAMES M. POTERBA
MassachusettsInstituteof Technology
LOUISE M. SHEINER
HarvardUniversity
LAWRENCE H. SUMMERS
HarvardUniversity
An Aging Society: Opportunity

or Challenge?
AN AMERICANwomanreachingchildbearingage in 1960wouldexpect 3.6

children;an identical woman in 1990 would expect only 1.9 children.
That dramaticdemographicchange makes it almost inevitablethat the
Americanpopulationwill age rapidlyover the next 50 years. By 2025,
the shareof the Americanpopulationthat is 65 or older will exceed the
share of Florida'spopulationthat is of retirementage today. The ratio
of retirees to workers will have risen by nearly two-thirds.Even more
dramaticdemographicchanges are occurringabroad.The share of the
Japanesepopulationthat is 65 or over will rise from 11 percent to 19
percent over the next two decades. If currentfertility levels in West
Germanyare maintaineduntil2050, the populationwill not only age but
shrinkmorethanone-third.
Thesedemographicchangeshave arousedconsiderableanxietyin the
United States. Economic concerns have focused on the burdenthat a
We are gratefulto George Akerlof, Robert Barro, Greg Duffee, Rachel Friedberg,
Ken Judd, Larry Katz, Laurence Kotlikoff, Julio Rotemberg, Robert Solow, and
membersof the Brookings Panel for helpful comments and to the National Science
Foundationand AlfredP. Sloan Foundationfor researchsupport.A data appendixfor
this projectis on file at the Inter-universityConsortiumfor Politicaland Social Research
in Ann Arbor,MI.
1
2 Brookings Papers on Economic Activity, 1:1990
growingelderlypopulationwill place on the economy in generalandthe

federaltreasuryin particular,as well as on a possible loss of dynamism
as populationgrowth slows. Those concerns have led to a potentially
radicalchange in Americanfiscal policy. To ensure that social security
taxes will be sufficientto fund benefits over the next 75 years, and to
help the nation save in anticipationof increaseddemographicburdens,
the social security legislation enacted in 1983 calls for social security
taxes to exceed benefits over the next 30 years. This surplus will be
accumulatedin a trust fund, which will peak at 29 percent of GNP in
2020and then be drawndown as the populationages.
Thispapersteps backfromthe currentpoliticaldebateover the social
security trust fund and examines the more general question of how
serious a macroeconomic problem aging is and how policy should
respondto it. We focus primarilyon issues relatingto savingand capital
accumulation.We do not consider the broaderquestion of whetherthe
currentU.S. nationalsavingrate is too highor too low, but focus on the
effect of demographicchanges on the optimallevel of nationalsaving.
In addition,we considerthe effects of demographicchange on produc-
tivity growthandthe optimaltimingof tax collections.
Our general conclusion is that demographicchanges will improve
Americanstandardsof living in the nearfuture,but lower them slightly
over the very long term. Other things being equal, the optimal policy
response to recent and anticipated demographic changes is almost
certainlya reductionratherthan an increase in the nationalsavingrate.
Slowing populationgrowth will reduce the investment that must be
devoted to equippingnew workers and housing new families, while
makingit easierfor the United States to attractforeigncapital.Although
therearemanyreasonsfor arguingthatthe United Statescurrentlysaves
too little, anticipateddemographicchangeis not one of them.
Our analysis proceeds in five steps. First, we assess the coming
dependencyburden.Whileit is truethatthe shareof the populationaged
65 or over will increase sharply,it is also true that the shareof children
in the populationwill graduallydecline, andthatthe fractionof the labor
force that is nearpeak productivitywill increase. Using informationon
projectedfertility, mortality,and laborforce participationrates as well
as dataon healthcare costs andthe spendingof differentage groups, we
assess past and future dependency trends. We find that demographic
changes unaccompaniedby changes in capital intensity would reduce
D. M. Cutler, J. M. Poterba, L. M. Sheiner, and L. H. Summers 3
per capita incomes by between 7 percent and 12 percent over the next
60 years, but would actuallyincreaseincomes over the next 20 years. In
only one of the next six decades will demographicchanges affect living
standardsas much as the "peace dividend" is likely to affect them in
this decade. The decline in living standardscaused by the increased
dependencewould be fully reversedby a 0.15 percenta year increasein
productivitygrowth.
Second, we consider the consequences of the slower labor force
growththat presages the increase in the retiredshareof the population.
Between 2010 and 2060, the laborforce is expected to decline slightly,
comparedwith an averageincreaseof 1.5percentannuallybetween 1950
and 1990.The projecteddecline in the laborforce growthratewill permit
a 3-4 percent reductionin the share of net investment in total income
withoutreducingcapitalintensity. Since reducedlaborforce growthwill
occur before dependency burdens increase, projected demographic
changes raise the short-termconsumptionpath even if the steady-state
consumptionlevel declines. We show that in a standardgrowth model
with plausibleparametervalues, optimalconsumptiontypically rises in
response to a demographicshock like that experienced in the United
States over the past three decades.
Third,we considerthe implicationsof integratedworldcapitalmarkets
for ouranalysis.The degreeand speed of populationagingin othermajor
industrializedcountries,particularlyWest GermanyandJapan,is more
dramaticthan that in the United States. The increase in dependency
abroadwill coincide with a deceleration in labor force growth rates.
Alongan optimalpath, therefore,the rest of the worldwill exportcapital
to the United States-thus increasingU.S. consumptionand reducing
savingin the shortrun.
Fourth, we go beyond the standardgrowth theoretic approachand
ask whether the coming demographicchanges are likely to affect the
rate of technicalchange. With slow laborforce growth, laboris scarce;
this scarcitymay induce more rapidtechnical change. Such a develop-
mentwould sharpenour conclusion that diminishedfertilityrepresents
an opportunityratherthana problem.Using internationalcross-section
time series data for 1960-85, we find some evidence that nations with
slowerlaborforcegrowthdo experiencemorerapidproductivitygrowth.
The estimatessuggestthatthe reductionin laborforce growthprojected
for the next 40 years may raise productivitygrowth enough to offset
fully the consequencesof increaseddependence.This finding,however,

is uncertain.A more definitivefindingis the absence of any empirical
support for the pessimistic view that aging societies suffer reduced
productivitygrowth.
Fifth, we consider the implicationsof our findingsfor fiscal policy.
Because demographicchanges over the next decades are not likely to
be associated with reducedprivatesaving, they constituteno argument
for reducingthe budgetdeficit. Thereremainsthe questionof efficiency
in tax collection. Maintainingcurrentservice levels for the elderly will
requirean increasein governmentspendingfromabout 32 percentto 37
percentof GNP. Since the deadweightloss from taxationrises with the
squareof the tax rate, financingthese expenditureson a pay-as-you-go
basis will involve higherdeadweightlosses than maintaininga constant
tax rate. We find, however, that these effects are likely to be small,
amountingto at most several tenths of a percentof annualGNP.
We conclude by discussing the implicationsof our results for social
security, for intergenerationalredistributionmore generally, and for
populationandimmigrationpolicy. Ourfindingssuggestthatpopulation
aging does not constitute a strong argumentfor accumulatinga large
social security trust fund, althoughif national saving is deemed to be
inadequatefor otherreasons, the trustfundmay be a convenientway to
increaseit.
The Burden of Increased Dependency

Theeconomicconsequencesof populationagingdependon the nature
of the underlyingdemographicchangeas well as the relationshipbetween
the resourceneeds of individualsat differentages and theircapacityfor
self-support.This sectionpresentsourestimatesof the economicburden
of increaseddependency, notingthe uncertaintiesassociated with each
step in the calculation.
Changing Demographic Structure

Figure 1plots the Social SecurityAdministration'sprojectionsof the
elderly dependency ratio, the numberof people aged 65 and over as a
fractionof the populationaged 20-64, and the total dependency ratio,
Figure 1. Actualand ProjectedDependencyRatios, UnitedStates, 1960-2065a

Ratio
Sources: Board of Trustees of the Federal Old-Age and Survivors Insurance and the Disability Insurance Trust
Funds (1988, table Al, p. 93) and unpublished data from the Social Security Administration underlying the published
projections.
a. Elderly dependency ratio is the population aged 65 and over divided by the population aged 20-64. Total
dependency ratio is population aged 65 and over plus the population under 20 divided by the population aged
20-64.
the number of children plus elderly as a fraction of the working-age

population,between 1960and2065.The figureshows the Social Security
Administration'sintermediateprojections (alternative II) as well as
outlyingprojectionsmakingmore extreme assumptionsabout fertility
and mortality changes. The projections agree in suggesting that the
fractionof the population65 and over will increase, and the fractionof
the populationunder20 will decrease, over the next 50 years. There is
very little change, however, over the next decade.
Decliningfertilityis the principalsourceof the changingdemographic
patterns.' In stable or decliningpopulations,young cohorts account for
a smaller share of the total populationthan they do in rapidlygrowing
populations.In the years followingWorldWarII, the total fertilityrate
in the United States rose from 2.4 in 1945 to a peak of 3.7 in 1957.
Fertility declined sharplyduringthe late 1960sand early 1970s, falling
to 1.7-well below replacementlevels-by 1976.Since then, fertilityhas
increased slightly, averaging1.8 in the mid-1980s. Preliminarydata for
1989suggest continuedincrease, to 2.0. These changes have important
implicationsfor the demographicstructureof the populationover the
next half-century.
The demographiceffects of fallingfertility have been reinforcedby
improvementsin old-age mortality. In 1960, life expectancy for a 65-
year-old man was 12.9 years, comparedwith 15.0 years in 1990. The
mortalityimprovementfor women has been even more pronounced,
with life expectancy at age 65 increasingfrom 15.9 to 18.9 years during
thepastthreedecades. Currentprojectionscallforfurtherimprovements
in life expectancies at age 65, to 18.0 years for men, and 22.1 years for
women, in 2060.2
Long-termdemographicprojectionslike those in figure1areuncertain
for severalreasons. First, fertilityforecasts are subjectto largestandard
errorsand are notoriouslyinaccurate,as is illustratedby figure2, which
displays historical total fertility rates and the various Social Security
Administrationprojections for the next half-century. The range of
historical experience dwarfs the range between the Social Security
1. The relativeimportanceof fertilitydeclines, mortalityimprovement,and interna-

tionalmigrationis discussedin OECD(1988).
2. These dataaredrawnfromBoardof Trusteesof the FederalOld-AgeandSurvivors
Insuranceandthe FederalDisabilityInsuranceTrustFunds(1990,table 11).Moredetailed
informationon mortalityimprovementscan be foundin PoterbaandSummers(1987).
Figure2. Actualand AlternativeProjectedTotal FertilityRates, 1920-2080

Totalfertilityrate
Source: Projected data are from Board of Trustees of the Federal Old-Age and Survivors Insurance and Disability
Insurance Trust Funds (1988, table 11, p. 37). Historical data are from Social Security Area Popuilation Projectionls.
1989 (1989, pp. 3-4).
Administration'soptimisticandpessimisticprojections.Even the factor

of two difference between the predicted share of the population65 or
over in 2050in the optimisticandpessimisticprojectionsshown in figure
1 probably understates the true degree of demographicuncertainty.3
Postwarfertilityprojectionsin the United States anticipatedneitherthe
beginning,nor the end, of the baby boom.
A second importantsource of demographicuncertaintyis the future
course of immigration.The Social Security Administration'sinterme-
diate forecasts assume net immigrationof 600,000 people a year until
2065-roughly the annuallevel of net legal and illegalimmigrationin the
late 1980s. Assuming a constant immigrantflow for the next 75 years
ignores potential changes in either immigrationpolicy or the level of
3. The pessimisticcase assumes an ultimatefertilityrate of 1.6, highfor examplein
contrast to West Germany'scurrentrate of 1.3. On the other hand, figure2 may be
deceptive, in that uncertaintyregardingthe averagefertilityrate over a 75-yearperiod
maybe muchless thanthe uncertaintyregardingfertilityratesat any pointin time.
illegalmigration.The age structureof the populationis sensitive to the

level of immigrationbecause immigrantson average are younger than
nonimmigrants.George Boras reports that only 3.1 percent of those
who immigratedto the United States between 1975and 1979were 65 or
over in 1980, comparedwith 10.6 percent of the nonimmigrantpopula-
tion.4 Higher immigrationduring the next half-centurywould reduce
dependencyburdens.
Uncertaintyaboutfuturemortalitygainsis a third,butless important,
source of randomnessin demographicprojections.Most of the forecast
rise in the numberof Americansaged 65 or over is the result of large
birthcohorts in the 1950sand 1960s.Even doublingthe projectedgains
in life expectancy at age 65 between 1990and 2060 would increase the
numberof elderly in 2060by less than 20 percent, and change the ratio
of the elderly to the working-agepopulationby less than 8 percentage
points.
Althoughthere is much uncertaintyregardingthe futureage compo-
sition of the U.S. population,the broad trend toward a rising average
age, a greaternumberof dependentelderly,andfewerdependentchildren
is indisputable.Moreover, uncertaintyabout long-termdemographic
change should not cloud the relativelycertainshort-termdemographic
outlook. Labor force growth in the next two decades, for example, is
largelyforecastablegiventhefertilityexperienceof the pasttwo decades.
Along many dimensions, the near-termeffects of demographicchange
operatein differentdirectionsfromthe long-termchanges. To illustrate
this we now explore alternativeways to calibratethe shiftingburdenof
demographicchange.
The Support Ratio

Demographicshifts affect the economy's consumptionopportunities
because they change the relative sizes of the self-supportingand de-
pendentpopulations.We summarizethese changes in the supportratio,
denoted a, which we defineas the effective laborforce, LF, dividedby
the effective numberof consumers, CON:
(1) o = LFICON.
The share of the populationaged 65 and over is one, but not the only,
determinantof this ratio. The support ratio is also influencedby the
4. Borjas (1990, pp. 41, 46).
relative consumptionneeds of people of differentages, as well as by

changes in the retirementage, labor force participationrates, and the
earningpower of those who are working. Because there are several
approachesto measuringandprojectingeach of these factors,we present
severaldifferentmeasuresof the supportratio.
The first issue in measuringthe supportratio concerns the relative
consumptionneeds of people at differentages. One assumption,which
we label CONI, defines effective consumption as if all people have
identicalresourceneeds:
99
(2) CONI = Ni,
where Ni is the numberof people of age i. This measure of needs is

implicitin the commonlycited total dependencyratio shown in figure1.
An alternativeapproachinvolves differentiatingthe resource needs
of people at different ages. We develop this approach in a second
measureof effective consumptionneeds, CON2, which has three parts:
privatenonmedicalexpenses, public education expenses, and medical
care. For private nonmedicaloutlays, we follow Edward Lazear and
RobertMichael in assumingthat all people 20 and older have identical
needs, while those under20 (18 in theirwork) have needs equal to one-
half those of adults.' For public education expenses, we assume per
capita outlays of $2,553 (1989 dollars) per person under 20, $309 per
personaged20-64, and$84perpersonaged65 andover. These estimates
are explainedin more detail below. For medical care, we assume that
needs are proportionalto total spendingby age: $1,262 per person per
yearforthoseunder64and$5,360forthose 65 andover.6Addingtogether
5. Lazear and Michael (1980, p. 102) estimate that a child raises equivalent scale
consumptionfor a husband-wifefamilyby 22.2 percent,or by 44.4 percentas muchas the
averageconsumptionof eitherparent.Thereis some evidencethatnonmedicalconsump-
tion needs of the elderly may be lower thanthose for youngerpeople. For example, the
U.S. Departmentof Agriculturepovertylineassumesthatfood expendituresby the elderly
are90 percentof those for prime-agedindividuals.The ongoingtrendtowardmoreelderly
livingin singlehouseholds,however, suggeststhat the relativeexpenditureneeds of the
elderlymayrise in the future.
6. These relativemedicalcosts are based on the currentage structureof the elderly
population.As the averageage of those 65 or over rises, the relativecost of medicalcare
for the elderlywill increase.In 1987totalannualpercapitahealthexpendituresfor people
aged65-69 were $3,728,comparedwith $9,178for those aged 85 and over. Holdingage-
specificexpenditurepatternsconstant at their 1987level, average spendingper person
aged 65 and over would be approximately10 percent higherwith the age composition
expectedin 2060ratherthanthatin 1990.See Waldoandothers(1989).
these three components, we construct a needs-weightedconsumption

measure, CON2, as 0.72 times the numberof people under20, plus the
numberaged 20-64, plus 1.27times the number65 and over.7
Therelativeneedsof elderlyandnonelderlyconsumerscanbe affected
by demographic factors such as mortality improvements. Edward
Schneiderand Jack Guralnikobserve that only 3 percent of men and 6
percentof women 65 andover reside in nursinghomes, while 15percent
of men and 25 percent of women 85 and over are in such homes. The
high cost of nursinghome care ($23,600per resident per year in 1985)
makes it an importantcontributorto the total cost of caringfor the aged
population.8
The appropriateweightingof young and old dependentsmay depend
on more than their consumption demands. Many of the transfers to
childrentake place withinthe family, while those to elderlydependents
are largely mediated by the government. A Scandinavianproverb,
broughtto our attentionby George Akerlof, suggests that "one mother
can care for ten children,but ten childrencannot care for one mother."
Individualsmayderivemorepleasurefromcaringfor childrenthanfrom
caring for elderly dependents, makingthe burdens of an increasingly
elderlypopulationmore onerous than the burdensof caringfor a young
population.9
We also considertwo differentmeasuresof the effective laborforce.
The first,LF1, assumes thatall people aged 20-64 are in the laborforce,
while individuals19and underor 65 and over are not:
64
(3) LF1 = E Ni.
i=20
Oursecond measure,LF2, recognizesthatboth humancapitalandlabor

force participationrates vary by age. We use data on the average 1989
earnings (w) of people of each age (measured in five-year intervals),
7. The needs-weightedconsumptionmeasure,CON2,is definedas
99
CON2 = E Si Ni,
i =,I
whereSi is the respectiveweightfor an individualat age i.
8. SchneiderandGuralnik(1990).
9. Providedthe "warmglow" of caregivingdoes not affect the marginalutility of
consuminggoods, it shouldnot affect our needs weightingof different-agedhouseholds.
It will affectthe totalutilityof households.
D. M. Cutler,J. M. Poterba, L. M. Sheiner, and L. H. Summers 11
alongwith Social SecurityAdministrationforecasts of age-specificlabor

force participationrates, PR, to estimateLF2:10
80
(4) LF2 = E wiPRiNi.
i= 15
Thisrecognizesthatthe earningcapacityof a society witha highfraction

of people in middle age is higherthan that of a society with many new
entrantsto the laborforce.11
Support Ratio Projections, 1990-2060

Because the level of the support ratio is less informativethan its
changesfrom year to year, we focus on &'t,the percentagechangein the
supportratiobetween 1990and year t:
(5) (Xt = (LFtICONt)I(LF1990ICON190) - 1.
We reportsupportratioscorrespondingto each combinationof effective
laborforce and effective consumptionmeasures.
Table 1 shows the historicaland projectedchanges in LF and CON
and demonstratesthat regardless of measurementmethod, growth in
both the laborforce and consumptionrequirementsdeclines duringthe
next half-century.For example, the earnings-weightedlaborforce grew
at a 1.7 percent annualrate duringthe 1980s, but will shrinkin four of
the five decades between 2010and 2060. In the nearerterm, laborforce
growthalso slows. By the first decade of the next century, labor force
growthis only one-fourthits rateduringthe 1970s.Totalneeds-weighted
consumption,which grew at a 1.1 percentannualrate duringthe 1980s,
rises by less thanone tenth of 1 percenta year between 2040and 2060.
Table2 andfigure3 show the percentagechangein the fouralternative
measuresof the supportratio.Fourconclusionsstandout. First,because
both our measuresof the laborforce grow more slowly thanpopulation
10. These data are from the Bureauof Labor Statistics, Usual WeeklyEarningsof
Full-Time Wage and Salary Workers and Usual Weekly Earnings of Employed & Part-
Time Wage and Salary Workers.We adjust part-timeworkers to full-timeequivalent
employees.
11. This laborforce concept includes only marketactivity, neglectingthe value of
labordevotedto householdproduction.It may thereforeoverstatethe historicalchanges
in the effectivelaborforce thatwere partlydue to risingmarketlaborforce participation
by women.
Table 1. Actual and Projected Average Annual Growth in Labor Force
and Consumption, United States, 1950-2060
Percent
Laborforce
Earnings- Consumption
Population weighted Needs-
20-64 population Unweighted weighted
Period (LF1) (LF2) (CON1) (CON2)
1950-1960 0.74 1.18 1.77 1.66
1960-1970 1.25 1.19 1.23 1.28
1970-1980 1.73 2.05 0.91 1.13
1980-1990 1.29 1.69 0.95 1.08
1990-2000 0.83 1.07 0.70 0.75
2000-2010 0.80 0.48 0.57 0.65
2010-2020 0.06 -0.03 0.48 0.60
2020-2030 -0.26 -0.10 0.29 0.42
2030-2040 0.11 0.07 0.14 0.17
2040-2050 0.00 -0.03 0.04 0.05
2050-2060 -0.06 -0.02 0.03 0.05
Source: Boardof Trusteesof the FederalOld-Ageand SurvivorsInsuranceand DisabilityInsuranceTrustFunds
(1988).Projecteddatafor 1990-2060use alternativeIlb. Data show geometricaverageannualchangesin laborforce
and consumptionneeds underIIb.
duringthe next 70 years, there is a long-rundecline in the supportratio.

The size of the decline is more sensitive to our assumptions about
consumptionthan to our measure of the effective labor force.'2 When
consumptionneeds are assumed to be equal for people of all ages, the
supportratioforLFI (LF2)declinesby 7.4 percent(7.8 percent)between
1990and 2060. Whenwe adjustconsumptionusing our needs-weighted
measure, the decline in the support ratio is more pronounced: 11.5
percentand 11.8percentfor LF1 andLF2, respectively.
It is difficultto know whether these estimates representa large or a
smallburdenspreadover 70 years. They correspondto between a 0.10
percent and 0.15 percent reduction in the annual productivitygrowth
rate, which is small relative to the uncertaintyin secular productivity
growth. They representthree to four times as large a cost as the peace
dividendthat the United States is likely to enjoy over the next decade.
12. Forthe period1950-90,the supportratiosare sensitiveto ourchoice of laborforce
concept, primarilybecauseof significantchangesin laborforce participationrates, most
notablyamongwomen.
Table 2. Changes in Support Ratio Relative to 1990, United States, 1950-2060
Percent
Unweighted Earnings-
Unweighted Earnings- population weighted
population weighted aged 20-641 populationl
aged 20-641 populationl needs- needs-
unweighted unweighted weighted weighted
consumption consumption consumption consumption
Year (LFIICON1) (LF2/CON1) (LF1/CON2) (LF2/CON2)
1950 -1.4 - 11.5 1.4 -9.0
1960 - 10.9 - 16.5 - 7.4 - 13.2
1970 -10.8 - 16.9 -7.7 - 14.0
1980 -3.3 -7.0 -2.0 -5.8
1990 0.0 0.0 0.0 0.0
2000 1.3 3.7 0.8 3.2
2010 3.8 2.8 2.3 1.4
2020 -0.5 -2.3 -3.1 -4.8
2030 -5.9 -6.0 -9.5 -9.6
2040 -6.2 -6.6 - 10.0 - 10.5
2050 -6.5 -7.3 - 10.4 - 11.2
2060 -7.4 - 7.8 - 11.5 - 11.8
Source: Same as table 1. The earnings-weighted laborforce measureuses contemporaneousand projectedlabor
force participationratesand the 1987age-earningsprofilesfor men and womento formeffectivelaborforces.
In yet anothermetric, a three- to four-yearincrease in the average age

at retirement,or a 19 percentagepoint increase in female labor force
participation,would be needed to offset the increase in dependency.
Second, in the next two decades there is a decline in economic
dependency(a rise in the supportratio)because the decliningnumberof
dependentchildren more than offsets the rising numberof dependent
elderly. Between 1990and 2010, when the baby boom generationis part
of the laborforce andrelativelysmallbirthcohortsare retiring,the labor
force grows more rapidlythan the dependentpopulation.This leads to
an improvementin the supportratioby 2010.13
Figure4 provides furtherdetail on the differentialburdensof young
and aged dependents.It plots the contributionsof both childrenand the
elderly to the supportratio definedusing LF2 and CON2. In this case
13. Measuresthat define effective consumptionwith less weight on childrenshow

smallergainsin the supportratioduringthe next two decades. If the consumptionweight
based on needs is set equal to zero for children,the supportratio actuallydeclines by
between 1percentand2 percentduring1990-2010.
Figure 3. Actual and Projected Changes in Support Ratio (Relative to 1990),

Four Alternative Measures, United States, 1960-2065
Ratio
1.04
Unweightedpopulation *;
j Unweightedpopulation aged
1.02 aged 20-641needs- ./ .'z ' 20-641unweightedconsumption
1 -weighted '- (LFIICONl)
consumption ,/:
0.98 (LF1/CON 2)1 4'
-
0.96 , . Earnings-weightedpopulation!
" ,' .- ~~~~~~~~~needs-weighted
consumption
0.84 _'-' .< Eatnings-weightedEpopulation! (LF21CON2)
unweighted consumption
0.82 . . (LF2/ CON1)
0.8 l
1970 1980 1990 2000 2010 2020 2030 2040 2050 2060
Year
Source: Table 2.
= P/(C + P + E), where P is the numberof prime-agedadults, C the

numberof effective children, and E the number of effective elderly.
Then the percentagechange in the supportratiocan be writtenin terms
of the percentagechangein its components:
(6) = (P - C) [C(C + P + E)] + (P -E) [EI(C + P + E)].

The first term is due to differentialgrowth rates of the prime-agedand
dependentchildrenpopulations, the second to the differentialgrowth
between the prime-agedand elderly groups. Figure 4 plots these two
terms, showingthat virtuallyall the improvementin the supportratioin
the near term is from a shrinkingshare of children in the population.
Mostof the long-rundecline is a resultof risingnumbersof elderlyduring
2010-2035.
Third, the changes in the supportratio between 1990 and 2060 are
usually no larger than, and in some cases significantlysmaller than,
Figure 4. Contributions of Young and Elderly Dependents to Percentage Changes

in the Support Ratio (Relative to 1990), 1955-2065
Change in consumptionrelative to 1990 (percent)
4% ~~" ..-------a------- ---

.. ......... --- ... . ..... .. .... .
~~~...
2 -
A X Tota change
a~~~~~~~~~~
-8 t\ e ~ ofeldet-iyA,
~~~~Sha-e \
-10lo. - Shat-eof childi-en'
-12 _ - ...s
- 16 X l l l l l l l l l l l l l l l l l l l
1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 2060
Year
Source: Authors' calculations using equation 6. The earnings-weighted labor force measure (LF2) and the needs-
weighted consumption measure (CON2) are used. See text for details.
those between 1960 and 1990. With our preferredmeasures, LF2 and
CON2, the supportratio was 14.0 percent lower in 1970than 1990. By
2060, it is projectedonce again to be below the 1990level, this time by
11.8percent. Oursupportratiopeaks around2010. One reason why the
slow growthof real wages in the U.S. economy since 1973has been less
burdensomethan it might have otherwise been is that the labor force
participationrate has risen. The figures show clearly that the gains in
sustainable consumption from demographic developments are now
nearlyexhausted.
Finally, while the decline in the supportratio by the middle of the
next century is large, there is still substantialuncertaintyabout the
ultimateburden.Figure5 presents supportratios using LF2 and CON2
underthe three Social SecurityAdministrationdemographicforecasts.
Thereare substantialdifferencesin the scenarios, particularlybetween
the more pessimistic alternative III and alternative II, which is our
Figure 5. Percentage Change in Support Ratio (Relative to 1990), Alternative

Demographic Assumptions, 1950-2065
Index relative to 1990
4
-4 \- lesntv
-18 a 0 | l | l | i i*
-16 AlternativeIII iv
1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 2060
Year
Source: Authors' calculations using Board of Trustees of the Federal Old-Age and Survivors Insurance and
Disability Insurance Trust Funds (1988) and the needs-weighted consumption (CON2)and earnings-weighted labor
force (LF2) measures.
standardcase. The decline in the supportratio is almost twice as large

in the pessimistic scenario as in our benchmark. Even in the optimistic
alternative I, the support ratio still declines by almost 8 percent between
1990 and 2060.
Capital Accumulationand Shifting Dependency Burdens
This section explores how the demographic shifts described above

affect the economy's sustainable level of consumption, and how society
should plan for these changes. We find that sustainable consumption
increases for the next several decades and that an economy with
otherwise optimal national saving would reduce its saving in response
to the coming demographic changes.
Steady-State Consumption Opportunities

Demographicchange has two effects on consumptionopportunities.
First, anincreasein dependencylowersoutputperperson,thusreducing
consumptionper capita. Second, slower labor force growth reduces
investmentrequirements,thus reducingthe need for savingandincreas-
ing consumptionper capita.
To examine the importanceof these two changes for consumption
opportunities,we assume that output per worker,f(k), where k is the
capital-laborratio, is divided between consumption and investment.
Maintainingconstantcapitalintensityrequiresinvestmentof nk, where
n is the laborforcegrowthrate.14 Forexpositionalease, we have assumed
away depreciationand technicalchange.'5 Whenthe laborforce and the
populationare not the same, consumptionper capita is only a fraction
of outputnet of investmentper worker. This fractionis the ratio of the
numberof workers to the size of the population,precisely the support
ratio (o) definedabove. The resultingequationfor per capita consump-
tion is:
(7) c = ot(k) - nk].
This expression can be rewrittento find the change in steady-state

consumptionfor changesin a and n:
(8) Aclc = Ao/ot - [o(klc) An + Aot(klc) An],
with c, k, and a evaluated at the initial steady state.16Equation 8
illustratesthe two steady-stateeffects of demographicchange.A decline
in the labor force-populationratio (o) reduces the level of per capita
consumptionthat is feasible given the economy's capital stock. At the
same time, a decline in the growth rate of the labor force (n) permits
more consumptionfor a given capital-outputratio. Society receives a
"consumptiondividend"when it is able to invest less and still maintain
14. A substantialpartof the U.S. capitalstockis residentialcapital.Thenaturalsteady-
state conditionfor housingrequiresinvestmentat the rate of populationgrowth,not the
rateof laborforce growth.In steadystate, these two growthrateswill coincide.
15. We incorporatebothin our numericalsimulationsbelow.
16. We have arbitrarilyassigned the second-orderterm to the second effect in our
decomposition.We have also assumedthat the capital-laborratio, and thus the capital-
consumptionratio,do not changewithdemographicchange.The modelwe presentbelow
justifiesthis assumption.
Table 3. Shifting Steady-State Per Capita Consumption from Demographic

Shocks, 1960-2065a
Percent
Unweightedconsumption(CON1) Needs-weightedconsumption(CON2)
Total Total
change in change in
Effect of per capita Effect of per capita
Effect of laborforce consump- Effect of laborforce consump-
Year dependency growth tion dependency growth tion
Population20-64 as effectivelaborforce (LF1)
1960 - 10.9 0.4 -10.6 -7.4 0.4 -7.1
1970 - 10.8 - 1.8 -12.6 -7.7 - 1.8 -9.5
1980 -3.3 -2.4 -5.7 -2.0 -2.4 -4.5
1990 0.0 0.0 0.0 0.0 0.0 0.0
2000 1.3 0.0 1.3 0.8 0.0 0.8
2010 3.8 1.0 4.8 2.3 1.0 3.4
2020 -0.5 3.0 2.5 -3.1 2.9 -0.2
2030 -5.9 2.5 -3.4 -9.5 2.4 -7.1
2040 -6.2 1.8 -4.3 -10.0 1.8 -8.3
2050 -6.5 2.5 -4.0 -10.4 2.4 -8.0
2060 -7.4 2.1 - 5.3 -11.5 2.0 -9.4
2065 -7.4 2.2 -5.2 - 11.5 2.1 -9.4
Earnzings-weighted
laborforce (LF2)
1960 - 16.5 1.1 - 15.4 - 13.2 1.2 - 12.1
1970 - 16.9 -0.4 - 17.3 - 14.0 -0.4 - 14.4
1980 - 7.0 -1.9 -9.0 -5.8 -2.0 - 7.8
1990 0.0 0.0 0.0 0.0 0.0 0.0
2000 3.7 2.2 5.9 3.2 2.2 5.4
2010 2.8 3.5 6.3 1.4 3.5 4.9
2020 -2.3 4.4 2.1 -4.8 4.2 -0.6
2030 -6.0 3.7 -2.3 -9.6 3.6 -6.0
2040 -6.6 3.6 - 3.0 - 10.5 3.5 -7.0
2050 -7.3 3.9 -3.5 - 11.2 3.7 -7.5
2060 -7.8 3.6 -4.1 - 11.8 3.5 -8.3
2065 -7.8 3.7 -4.2 - 12.0 3.5 -8.4
Source:Authors'calculations.
a. The table shows the steady-statechangein consumptionrelativeto the 1990base if demographicchangewere
to reacha steady state at the level of the indicatedyear.
a given level of per capitaoutput.This "Solow effect" offsets the long-

rundependencyeffect on per capitaconsumption.
Table 3 reportsthe size of these two effects. For each year, we show
the steady-stateconsumptionchangeassociated with changesin a (first
column), n (second column), and the combined effect (thirdcolumn).
The consumptionchanges due to the dependencyincrease are the same
as the changes in the supportratio shown in table 2; the other columns

show the extent to which changinginvestmentneeds offset this effect.
Two resultsemergefromtable3. First, the consumptionbenefitsfrom
reducedinvestmentrequirementsare substantial.Duringthe next two
decades, the benefits of slower labor force growth will be about a 1
percentto 3.5 percentincreasein per capitaconsumption,usingthe 1990
base. Since the labor force was growingmore rapidlyin the 1970s and
1980s than in 1990, the effect of reduced investment requirementsis
even largerrelativeto earlieryears. By the middleof the next century,
the benefitsof slowerlaborforce growthwill be between 2.1 percentand
3.7 percentof per capitaconsumption.This is between one-quarterand
one-halfof the adverse dependency effects of the changingpopulation
mix.
Second, while the investment effect offsets a substantialpart of the
long-termdependencyincrease, it magnifiesthe short-runeffect of rising
supportratios. Reduced dependency and slowing labor force growth
both increase consumptionpossibilities so that by 2010, society will be
between 3.4 percentand 6.3 percentricher,dependingon the combina-
tion of laborforce andneeds measures.Onlyafter2020does the increase
in dependency outweigh the decline in investment needs and reduce
consumptionbelow its 1990level.
The steady-state consumption decline between 1990 and 2060 is
estimatedat between 4.2 percent(with effective consumersset equalto
total populationand the earnings-weightedlaborforce) and 9.4 percent
(witheffective consumerscomputedusing our needs-weightedmeasure
andthe unweightedlaborforce). As with the supportratios, this finding
is more sensitive to our definitionof consumption needs than to our
definitionof the effective labor force. For almost all cases, however,
society is richerin the new steady state thanin 1970or 1980.
Demographic Change and Optimal Capital Accumulation

Theresultspresentedso farsuggestthatin the shortrun,demographic
changeswill raise the level of consumptionthat can be sustainedwhile
maintainingthe level of capitalintensity.In the long run,they will reduce
the sustainablelevel of consumption.The question then becomes how
society shouldadjustits savingpolicy to these developments.To study
this question,we use the standardRamsey optimalgrowthmodel.
We assume that a social plannerseeks to maximize

(9) V = f e-PIPt U(ct)dt,
where Pt denotes the numberof individualsalive in period t, ct is per

capitaconsumptionin period t, and p is the social time preferencerate.
We denote the currentperiodas time zero. This social welfarefunction
weights the utility, denoted as U, of a representativeindividualin each
generationby the generation'ssize.'7 Using our earliernotation, Pt =
Ntlot, where Nt is the laborforce in periodt and ot is the supportratio.
Ouranalysis abstractsfrom the overlappinggenerationsstructureof
the actualpopulation.CalvoandObstfeldformallyjustify this procedure
by demonstratingthat if age-specifictransferprogramslike social secu-
rity are available, and if individual utility functions are additively
separable,then "the Cass-Koopmans-Ramseyframeworkcan be used
to evaluate paths of aggregate consumption even in models where
different generations co-exist. . . . the planning problem facing the
government can be decomposed into two subproblems, a standard
problem of optimal aggregatecapital accumulationand a problem of
distributingconsumptionoptimallyon each date amongthe generations
alive then."18 The social planner maximizes equation 9 subject to a
capitalaccumulationconstraintanalogousto equation7:19
(10) kt =f(kt) - ctIot - ntIkt.
If at = 1, equation 10 reduces to the standardresource constraint in

neoclassical growth models. The consumptionprofile that solves this
problemsatisfies:
(11) &tIct= u[ f'(kt) - P]
where u = [- U'(ct)Ict] [U"(ct)], the elasticity of substitution in con-
sumption.
17. Some mightarguefor usingan alternativeobjectivefunctionthatdoes not weight
the averageutilityof differentgenerationsby the numberof peoplein the generation.This
will lead the social plannerto raise averageconsumptionin smallcohortsrelativeto that
in largercohorts,because the aggregateresourcecost of raisingthe averageconsumption
of peoplein smallcohortsis less thanthatfor largecohorts. We see no compellingethical
argumentfor weightingpeoplein differentsized cohortsdifferently.
18. CalvoandObstfeld(1988,p. 163).
19. The optimalplanmustalso satisfytransversalityconditionsnotedfor exampleby
BlanchardandFischer(1989).
Figure6. Steady-StateConsumptionResponseto an Increasein Dependency
(Declinein a)
Percapita =O
consumption(c)
Initial c* -----------
New c*
k* ~~~~~~~~~~Capital-labor
ratio (k)
In steady state with no technical progress, per capita consumption

and the capital-laborratio must be constant. From the Euler equation
11, we findthat constantconsumptionrequires
(12) k* =f'-lI(p).
This locus, a verticalline in (c, k)space, is drawnin figure6. Constancy

of the capital-laborratio given in equation 7 yields the second locus
depictedin figure6 as the solid line k = 0.
Permanentreductionsin a, the supportratio, scale back the feasible
level of per capita consumptionfor each k, shiftingthe k = 0 locus as
shownby the dashedline in figure6. The steady-statecapital-laborratio
is unaffected by this change, so the only effect of this shock is an
immediateandpermanentdeclinein consumptionpercapita.Reductions
in n, the laborforce growthrate, wouldhave the opposite effect, shifting
the k = 0 frontierout. The steady-stateconsumptioneffect of a demo-
graphic shift such as a fertility decline, which reduces both a and n,
dependson which of these effects may be larger.Reductionsin n would
unambiguouslyreduce the optimal steady-state saving rate while in-

creases in a would have no effect on steady-statesavings.20
The actual demographicprojectionsfor the United States are more
complex thanan immediateshift in either a or n, however. For the next
several decades, the net effect of demographicchange is an outward
shift in the k = 0 locus, followed by a period of inward shift that
terminateswith the locus below its currentlevel. Whenconsumershave
perfectforesightand recognize the complex natureof the demographic
transition,the initialconsumptionresponse to news of the demographic
transitionis theoreticallyambiguous.
This ambiguitysuggests the need for explicit numericalsimulations
to addressthe optimalconsumptionresponse. We assumethatthe utility
functionin equation9 has the form
(13) U(c,) = (c,- 1/u- 1)/(1 - 1/U),
where u is the elasticity of substitutionin consumption.We also assume
a constantelasticity of substitutionproductionfunction:
(14) f(kt) = [akt111-+ b]1-P.
The elasticity of substitutionin productionis P. To find the transition
path between one steady state and another, we discretize differential
equations10and 11andemploy a grid-searchalgorithmto findthe initial
consumptionlevel that will lead the economy to the new steady state.21
Oursimulationsalso allow for labor-augmentingtechnicalchange(g)
and depreciation(s), which are introducedinto equations 10 and 11 in
the standardway.22Althoughconsumptiongrows over time when there
is technical progress, the consumptionnumberswe reportare relative
to the consumptionthatwould have been possible withoutdemographic
change. We assume that technicalchangeis equal to 1.4 percenta year,
the Social Security Administration'ssteady-stateprojection.23The de-
preciationrate is set equalto 4.1 percent, the U.S. averageduring1952-
20. This is easily seen fromthe Harrod-Domarconditionk/fl k) = s/n, where s is the
saving rate out of nationalincome, and the observationthat neitherchanges in (x nor
changesin n affectoptimalsteady-statecapitalintensityin the Ramseymodel.
21. Because the Social Security Administrationforecasts populationin every fifth
year, we interpolateannualobservationsusinga smoothinterpolator.The resultsare not
sensitiveto the frequencyof the data.
22. FollowingBlanchardandFischer(1989),we expresscapitalper "effectiveworker,"
whereeffectiveworkersgrowat n + g. Consumptionis expressedper "effectiveperson."
In equation11,the discountfactorbecomes (p + 8 + g/l).
23. Ourresultsareinsensitiveto the choice of g.
87.24Finally,we use datafor this periodon paymentsto laborandcapital

to estimatecapital'ssharein gross output-33.2 percent. Overthis same
period,the capital-outputratioaveraged2.3.25These two numbersimply
a steady-statemarginalproductof capitalof 14.4percent.Fromequation
11, this impliesan effective discountrate (p + glu) of 10.3percent:that
is, the steady-statemarginalproductof capitalless depreciation.
We present results using two values of u, a benchmarkcase of unit
elasticity (u = 1) and an alternativeelasticity of substitutionof one-
tenth (u = 0.1). We also choose two values for the elasticity of
substitutionin production,a benchmarkof unitelasticity (,B= 1)and an
alternative elasticity of one-half (I = 0.5). When the elasticity of
substitutionin consumptionis low, consumptiontoday is not a good
substitutefor consumptiontomorrow,andwe expect moreconsumption
smoothing. When the elasticity of substitution in production is low,
savingdoes notget a highreturnsincethe extracapitaldoes not substitute
well for the smaller labor force, and we expect less consumption
smoothing.
Demographicchange has occurredgraduallyover the past 25 years,
as the baby boom has given way to the baby bust. It is not obvious how
best to model these changes as a single shock. Initially,we assume the
economy is in steady state with values of otandn correspondingto those
prevailingin 1990,and ask how consumptionand saving should evolve
henceforth.Because some of the consequences of demographicchange
were already known by 1990, we go on to examine how consumption
and saving should have responded in 1970and 1980if news of demo-
graphicchangehad suddenlyarrived.
For all our simulations,we use the trajectoriesof oxand n impliedby
the Social Security Administration'salternativeIlb forecasts, and fur-
ther assume that the predictedvalues for 2065persist as the economy's
final steady state.26The resulting consumption changes are thus the
24. Ourdepreciationrate is estimatedas capitalconsumptionallowancesdividedby
the aggregatecapital stock. We define aggregatecapital stock as nationalassets minus
consumerdurablesminusone half of the value of land. Consumerdurablesare excluded
since they are not includedin output. One half of land is includedin capitalto allow for
naturalresourcevaluesto change.
25. Capital'sshare in output is total output less wages and salaries, two-thirdsof
proprietors'income (the estimated labor compensation),and indirect business taxes,
dividedby outputless indirectbusinesstaxes.
26. AlternativeIIb projectionsembody the alternativeII forecasts regardingdemo-
graphicchange,as well as an intermediateset of economicforecasts.
Table 4. Optimal Consumption Response to Demographic Shocks,

United States, 1990-2060a
Perfectforesight with alternativeelasticitiesb

Static B =1 [ =1 [B = 0.5 B = 0.5
Item expectations a =1 u = 0.1 =I
=1 = 0.1
Case 1. Laborforce population20-64 and unweightedconsumption
Initial steady state 100.0 100.0 100.0 100.0 100.0
Initial adjustment 100.0 100.6 101.1 100.4 101.0
Time path
2000 101.3 101.4 101.3 101.4 101.3
2010 104.8 103.3 101.7 103.8 102.1
2020 102.5 102.3 101.4 102.5 101.7
2030 96.6 98.3 100.3 97.7 99.8
2040 95.7 96.2 99.0 95.9 98.1
2050 96.0 95.9 98.1 95.9 97.1
2060 94.7 95.1 97.3 95.0 96.2
New steady state 94.8 94.8 94.8 94.8 94.8
Case 2. Earnings-weightedlaborforce and needs-weightedconsumption

Time path
2000 105.4 104.1 103.0 104.5 103.3
2010 104.9 104.1 102.8 104.5 103.0
2020 99.4 100.4 101.5 100.2 101.1
2030 94.0 95.7 99.6 95.0 98.3
2040 93.0 93.5 97.8 93.2 96.0
2050 92.5 92.7 96.3 92.6 94.5
2060 91.7 92.0 95.1 91.8 93.4
New steady state 91.6 91.6 91.6 91.6 91.6
Source: Authors'calculations.
a. Each column is the simulatedpath of consumptionin responseto a demographicshock like that which the
UnitedStates will experiencebetween 1990and 2060.The static expectationscolumnis the changein consumption
if agentsin each periodassumethat the currentlevel of otand it will persistforever.The perfectforesightcolumns
assumecurrentknowledgeof the entirepathof demographicchange.The initialsteady state is the 1990value of at
and n.
b. ,Bis the substitutionelasticityin production;a is the substitutionelasticityin consumption.
optimalresponse to the demographictransitionthat the United States

will undergoover the next seven decades, assumingthese changeswere
unforeseenas of 1990.
The results of these simulationsare shown in table 4. The level of per
capita consumptionin the 1990steady state is normalizedto 100. The
firstcolumn,the staticexpectationsresponse, is the changein consump-
tionif consumershave no foresightaboutdemographicchangebutrather

assume at each date that currentconditionswill persist forever. It thus
correspondsto the consumptionpathin table 3. The otherfour columns
assume that consumersin 1990have perfect foresight regardingfuture
demographicchanges.
For all the parametervalues, consumptionrises initiallyin response
to the demographictransition,by up to 2.8 percentrelativeto the steady
state implied by 1990 demography. This result is insensitive to the
parameterchoices we present.27Consumptionremains above its 1990
level until 2020 or later. Thus, demographicshifts duringthe next half-
centuryoptimallyraise present consumption.The initialeffect is more
pronouncedwhen consumptionis less substitutableover time(a is small)
and less pronouncedwhen productionis less substitutableover time (
is small).
Figure 7 shows the movements of consumptionand capital for the
simulations using unit elasticities of substitution in production and
consumption. The corresponding saving rate is shown in figure 8.
Consumptioninitiallyrises by 2.3 percent. This is followed by a period
when capitalper effective workerdeclines, duringwhich consumption
continuesto increase. The shiftingopportunitylocus due to the decline
in laborforce growthultimatelycauses an increasein savingand thus in
capital intensity, even at the higher level of consumption. After the
period of capital deepening, consumption begins to decline. Finally,
when the increase in dependencyovertakes the favorableeffects of the
slowinglaborforce growth,both consumptionand capitaldecline to the
new steady state, and savingremainslow.
As figure8 demonstrates,the saving rate falls almost 2 percentage
points duringthe 1990 initial adjustment.It then increases for a few
years, thoughit never attainsits initialsteady-statevalue. This increase
is dueto the increasein the supportratio,whichallowsbothconsumption
perpersonandthe savingrateto increase. Finally,the savingratebegins
to fall toward its new long-runlevel, equal to the amount of saving
necessaryto equipthe more slowly growinglaborforce.
Wealso ranthe simulationsusingthe Social SecurityAdministration's
alternativeI and alternativeIII, with no substantivechanges in results.
27. Inadditionto theparametervaluesreported,we haveexperimentedwithelasticities
in substitutionandproductionup to 10. For none of these cases is therean initialincrease
in savings.
Figure 7. Optimal Consumption and Capital-Labor Trajectory Following Demographic

Shock, 1990-2060
Consumption per effective person (index, 1990 = 100)
104 200 2005
1995 2015
102 1990
100 - Initial steady state 2020
98-
2025
96
2030
94 2 35
2040
2050
92 2060
Nev steady state
270 274 278 282 286 290
Capitalper effective worker
Source: Authors' calculations. Elasticities used are , = 1, a= 1. Labor force measure is earnings-weighted (LF2);
consumption measure is needs-weighted (CON2). Consumption at the initial steady state is normalized to 100. The
resulting index yields capital per effective worker equal to 274 in the initial steady state.
Consumptionrises less with the alternativeI assumptionsthanwith our

benchmarkalternativeII assumptions,becausethe numberof dependent
childrendoes not decrease as quickly, and more with the alternativeIII
assumptions,wherethereis an even largershort-runbenefit.In all three
cases the responseto the demographicnews is a decrease in saving. We
have also experimentedwith changingcapital's share or the assumed
initiallevel of capitalintensityin orderto vary the discountrate'8 Even
witha purediscountrateas low as zero, ourconclusionthatconsumption
rises followinga demographicshock remainsvalid.
Finally,we exploredhow consumptionwould changeif we beganthe
simulations in 1970 or 1980. As table 2 demonstrated, consumption
28. Because the discountfactormust equal the marginalproductof capitalin steady
state, and the marginalproductof capitalis the ratio of capital's sharein outputto the
capital-outputratio,changesin the discountfactorhave to be accompaniedby changesin
eithercapital'sshareor the capital-outputratio.
Figure 8. Optimal Saving Rates in Response to Demographic Shock, 1990-2070

Saving rate (percent)
6.5 -
5.5
4.5-
4-
3.5-
3 -
2.5-
2 l l l l l l l
2000 2010 2020 2030 2040 2050 2060 2070
Year
Source: Authors' calculations. Elasticities used are , = 1, a= 1. The consumption measure is needs-weighted
(CON2); the labor force measure is earnings-weighted (LF2).
possibilitiesare higherin 1990thanin any of the threepreviousdecades.

Figure9 shows the deviation of the saving rate from its initial steady-
state level afterthe demographicnews. In all of the simulations,saving
falls immediatelyfollowingthe demographicnews, andis always falling
by 2000.Even in the cases where savingbeginsto increasein the 1990s-
when we begin the simulationin 1980or 1990-the saving rate is lower
throughoutthe 1990sthanthe originalsteady state, andit begins a period
of prolongeddecline by 2000.
Whilethese figureshelp to develop perspective on the recent decline
in U.S. savingand investmentrates, the actual decline in U.S. national
savingfrom an averageof 7.1 percentin the 1970sto about 2 percent in
the late 1980sis considerablymore than our demographicanalysis can
justify.
Theanalysisinthis sectionreachesa clearconclusion.Foraneconomy
choosingits consumptionpathin accordwith a standardoptimalgrowth
model, the right response to the upcoming U.S. demographicchange
Figure 9. Optimal Saving Responses to Demographic Change: Sensitivity to Date

of Demographic Discoverya
Differencein saving rate from steady state (percent)
Simulation in 1970
stat-ting
-2
X _ Simulation
star-ting 19809*.0. in ----
Simulations
st*.igin180. ... .............
-5 -
I I i I. I Il
1970 1.980 1990 2000 2010 2020 2030 2040 2050 2060 2070
Year
Source: Authors'calculationsusing Board of Trustees of the Federal Old-Ageand SurvivorsInsuranceand
Disability Insurance Trust Funds (1988).
a. This figurepresentsthe percentagepoint differencein saving rates from the initialsteady state followinga
demographic Thethreelinesrepresent
shock. saving
paths arerunstarting
whenthesimulations and
in1970,1980,
1990. These saving plcuations u sinBardof Tusg needs-weighted
n consumption(CON2) and the earnings-
weightedlaborforce (LF2),and unitelasticitiesof substitutionin both productionand consumption(, = 1, a= 1).
wouldbe an increasein consumptionanda reductionin nationalsaving.

Forallplausiblecombinationsof parametervalues, the effects of reduced
laborforce growthand reductionsin the numberof childrenexceed the
effects of increasesin long-rundependency.
Open-EconomyAspects of the Demographic Shift
Ouranalysis thus far has focused on the demographicchange in the

United States. When capital markets are integrated, however, the
demographicshift in the United States must be measurednot only in
absoluteterms but relativeto the coincident shifts in our majortrading
partners. This section compares the degree of population aging in
differentnations and extends our earlier simulationmodel to consider

the United States in relationto the other countries in the Organization
for Economic Cooperation and Development (OECD). Our earlier
findingthat, other things being equal, demographicchanges justify a
reductionin optimalsavingis reinforcedwhen we allow for international
capital flows, because demographicchange is less pronouncedin the
United States thanelsewhere in the OECD.
Relative Rates of Population Aging

To compare rates of population aging, we use projections by the
OECD.29These projectionsdifferin two importantways fromthe Social
Security Administrationprojections for the United States. First, the
OECD treats the 15-19 age group as workers ratherthan dependents.
Second, and more important,the OECD assumes that fertilityrates in
all countries will converge to the replacement level of 2.1 by 2050.
Because U.S. fertilityrates are currentlywell above those in most of the
OECD, this understatesthe likely contrastbetween the futureU.S. and
foreigndemographicexperiences.
Figure10shows the historicalandprojectedelderlydependencyratio
for the United States, Japan, and the European Community.30The
elderly dependency ratio increases substantiallyin all countries, with
the most rapidincrease in Japan. By 2050, even with a 19 percentage
point increasein the elderly dependencyratiofrom 1950,the U.S. ratio
will be roughly 5 percentage points lower than those of the other
countries.
Figure 11 shows the path of supportratios correspondingto the LF1
and CON1 assumptions. The broad outlines for all three regions are
similar.All have highersupportratios in 1990than in 1960,and all will
have much lower supportratios by the middleof the next century than
they do today. The ultimatelevel of U.S. dependencywill be lower than
thatabroad.
Two differences in these indexes are notable, however. First, the
United States will be better off for the next two decades than it is now,
while the other countries experience declines in the support ratio
29. OECD(1988).
30. Themulticountryindexis a GDP-weightedaverageof the indexesforthe individual
countries.
Figure 10. Actual and Projected Elderly Dependency Ratios, United States, Japan,
and European Community, 1950-2050
Percent
42
40 -
38 -
36-
34-
32 Japat -
30 - .
24~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~J
22 t EulopeanCommruni/y
........./. t
14 - -J
18 _ UniteStates
122/
10
I I I I I I I II
1950 1970 1990 2010 2030 2050
Year
Source: OECD (1988). Working-age population includes all those aged 15-64.
beginning in 1990. Second, the U.S. and EC dependency ratios are

driven principallyby fertility changes, while the Japanesechanges are
drivento a muchlargerextent by reductionsin mortality.31The decline
in the supportratio in the 1950sin the United States and in the 1960sin
the EC is due to increased numbersof children;the rise in the support
ratio throughoutthe postwar period in Japan, in contrast, is caused by
reduced mortalityat middle and older ages. Because the labor force
grows faster when fertilityis higher,the reductionin laborforce growth
over the next several decades, and thus the consumptiondividendfrom
reducedinvestmentrequirements,will be largerin the United States and
the EuropeanCommunitythanin Japan.
To evaluate the size of the demographictransitionabroad, table 5
reports the optimal consumption and saving responses to projected
31. OECD(1988)presentsevidenceon theimportanceoffertilityandmortalitydeclines
for the differentcountries.
Figure 11. Actual and Projected Support Ratios, United States, Japan, and European
Community, 1950-2050
Supportratio (1990 = 1)
United States
1.00- J n
0.95
_
***;/..*....
/ "v^^ s~~~ ~~~~~Eur-opean
0.90 *. ......... Community
_~~~~~ s
_ .
0.95 -l
I S
0l9S0 19019 01 0025
demorapic cangs intheUnitd Sape,Jaa,ahnurpanCm
United tates, onsumpton rise only sightly nitiall, contiues in

2010 and thendeclinesto the ne steady sate. Thi
creasing ntil
consumptio increase s accompaied by an ncreased sving rate
however, sncethe reative incrase in theworking-ag populatio
1950in
increases outpu 2010,
1970199
per 2050clnsttenwstaysat.Ti
personby ore than consumtion 2030per person
For Japan,thecomin demographicchange reduce optimal con
Suc:OD(18)Tconsumption measuree totalppopulation
is (CncrTelaboefrc rthe,
measuren i
deogevrapi changesitheUnatveitcedsi
Stthes Japain,-theEuopelantCom-
aesnd thmoetotalnED3 consumptionpeprsn
muity,ethesnon-pu.S OECD hs
usming
patsore simuan,thed previusesection. Fornth
themodelphofthanes
32. The table uses the case of unit elasticities of substitutionand production.We
assumethatdepreciationratesandratesof labor-augmenting technicalprogressare equal
in all countriesand are the same as the Social SecurityAdministrationforecasts for the
United States. The assumptionof equal productivitygrowth is obviously wrong but
probablydoes not havea largeimpacton estimatesof the changein savingdue to changes
in demographicstructure.
Table 5. Autarky Response of Consumption and Saving to Demographic Shocks,

United States and Various Nations, 1990-2050a
Country
United European Non-U.S. Total
Item States Japan Community OECD OECD
Consumnption
response
Time path
2000 100.6 97.2 99.7 99.3 99.8
2010 101.5 92.2 98.8 97.8 99.2
2020 99.1 89.0 97.1 95.5 97.0
2030 94.4 88.5 92.8 92.1 93.0
2040 92.0 86.3 89.1 88.8 89.9
2050 92.1 84.8 88.2 87.9 89.1
New steady state 92.3 84.4 87.9 87.8 89.0
Saving rate response
Initial adjustment -0.1 0.7 -0.1 0.0 0.0
Time path
2000 0.1 - 1.4 -0.9 - 1.0 - 0.6
2010 0.5 - 3.0 -0.7 - 1.1 -0.5
2020 - 1.4 - 2.6 -1.5 -1.8 - 1.6
2030 -2.4 - 1.3 -2.8 -2.5 -2.5
2040 - 1.5 -3.1 -2.9 -2.9 -2.3
2050 - 1.4 - 1.5 -0.8 -1.2 - 1.3
New steady state - 1.5 - 1.3 -0.7 - 1.1 - 1.3
Source: OECD (1988) and authors' calculations.
a. The values in the table are the optimal consumption and saving paths for each country without international
capital flows. We use needs-weighted consumption (CON2) and the unweighted labor force (LF1). Consumption is
relative to the initial steady state, which is normalized to 100. Saving paths are defined as the percentage point
difference between the saving rate along the path and the initial steady state. The elasticities of substitution in
production and consumption both equal unity.
sumptioninitiallyby just under 1 percent, and consumptioncontinues

to decline throughoutthe next 60 years, even as the savingratedeclines.
For the European Community,there is also a slight increase in con-
sumption,but by 2000 consumptionis lower, and continues to decline
throughoutthe next half-century.This patternof decliningconsumption
aftera smallincreasein initialconsumptioncarriesover to the non-U.S.
OECDand total OECD simulations.
The initial decrease in the saving rate in the United States and the
increasein the non-U.S. OECDimply that in an open economy, capital
wouldinitiallyflowfromthe non-U.S. OECDto the United States. After
the initialchangein saving, however, capitalflows are more difficultto
predict.In additionto the changein savingrates in the autarkycase, the

countriesalso have differentchanges in labor force growth rates, and
thus in investment requirements.Because the desired capital inflow
dependson the differencebetween savingandinvestmentrequirements,
lookingat savingrates or consumptionalone does not indicatewhether
each countrywould borrowor lend. To addressthis issue, we turnnext
to simulationsthat allow for capitalmobility.
A Two-CountrySimulationModel
Ouropen-economysimulationsaggregatethe EuropeanCommunity,
Japan,and the other countriesof the OECD to form a non-U.S. OECD
index. Figures 12 and 13 show the supportratio and laborforce growth
rates for the United States and this aggregate.The supportratios are
consistentwith those in figure11. The United States has a 5 percentage
point highersupportratioin 2050 than the non-U.S. OECD, and unlike
the restof the OECDhas a risingsupportratioover the next two decades.
By 2050the laborforce in both areas is projectedto stabilize, not grow.
Between now and then they fluctuate,but with U.S. laborforce growth
always higher.
To assess the optimalresponse of U.S. saving in an open-economy
context, we extend the modelof the previoussection to allow for capital
mobility. We distinguishasset ownershipfromasset locationby denoting
periodt asset ownershipper person in country 1 by a1 t. Asset accumu-
lationis given by
(15) a1,t = wt + a1,t(rt - ni,t) -(cl,t1o,),
where the wage, wt, and the interest rate, rt, are equalized across
countries.The laborforce growthrate, the supportratio, and the level
of percapitaconsumptioncan differacross countriesandthereforehave
both time and countrysubscripts.
Thecommoncapital-laborratiois a weightedaverageof asset holdings
in the two nations:
(16) kt = 01t a1,t+ (1 - 01,t)a2 t,
where 01,tis country I's share of world population. From equation 16 we
derivethe capitalaccumulationconstraintfor the two-countrymodel:
(17) k,= 61,t (a1,t - a2,t) + 01,t bl,t + (I 01,t) a2,t
Figure 12. Actual and Projected Support Ratios, United States and Non-U.S.
OECD, 1960-2050
Supportratio (1990 = 1)
UnitedStates
1.00
0.95 - / Non-U.S. OECD "\
0.90 _ %%s
*.85 1970 1980 1990 2000 2010 2020 2030 2040 2050
Year
Source: Authors' calculations based on data in OECD (1988). Non-U.S. OECD includes all OECD countries
except the United States, weighted by GDP. The consumption measure is needs-weighted (CON2), with the same
weights as for the United States. The labor force is the population 15-64 (LF1).
This constraint replaces equation 10 in the one-country model. The

optimalconsumptionprofile(equation 11) and the steady-state saving-
investmentrelation(equation12)areidenticalto those in the one-country
case.
We calibrate the two-country model assuming that both countries
have Cobb-Douglasproductionfunctions and logarithmicutility func-
tions. We assume that one nation is the United States and the other is
the non-U.S. OECD,and set the relativelaborforce in the United States
at four-tenthsof the two-countrytotal, roughlythe value of the produc-
tivity-weightedU.S. labor force share for 1990. In addition, we begin
the simulationsassumingno net foreigninvestmentposition.33We also
assume equalratesof technologicalprogressand equaldiscountrates in
the two countries.
33. This correspondsto the averageU.S. net foreignasset positionduringthe 1980s,
butunderstatesforeignholdingsof U.S. assets at the beginningof the 1990s.
Figure 13. Actual and Projected Growth Rate of Labor Force, United States
and Non-U.S. OECD, 1960-2050
Annualgrowthrate (percentagepoints)
1.8
1.6-
1.4 UnitedStates
1.2 A
1 "____\
0.8
0.6-
-
0.4 Non-U.S. OECD\
0.2 -%
-0.2 -
-0.4
-0.6 -
I I l l l
1960 1970 1980 1990 2000 2010 2020 2030 2040 2050
Year
Source: Sameas figure12.
Table 6 presents the two-country simulationresults. We normalize

consumptionto be 100initiallyin both countries. Whilethe shape of the
consumptionresponseis similarin the open- andclosed-economycases,
the size of the responses is different.For the United States, the closed-
economy analysis suggests a 0.1 percent consumptionincrease relative
to the 1990steady state. Withcapitalflows between the relativelyslowly
aging United States and the more rapidly aging rest of the OECD,
however, the U.S. consumption increase is 1 percent of the 1990
benchmark.Consumptionin the United States increases more in the
open-economycase because highsavingelsewherein the worldreduces
the rate of returnto capital, inducinga positive shock to the value of
humanwealth.34
34. Althoughwe assume that utilityis logarithmic,the interestelasticityof saving is
positive. When interest rates increase, holding wealth constant, saving is unaffected.
However, with interest rates higher, the present discounted value of labor income
decreases,and hence consumptionfalls. Wagesin the United States also increasein the
shortrun.
36 Brookings Papers on Economic Activity, 1.1990
Table 6. Effect of Demographic Shock on Consumption and Foreign Capital Ownership,
United States and Non-U.S. OECD, 1990-2050a
UnitedStates Non-U.S. OECD
Foreign Foreign
capital capital
Consump- ownership Consump- ownership
Item tion (percent) tion (percent)
Initial steady state 100.0 0.0 100.0 0.0
Initial adjustment 101.0 0.0 99.4 0.0
Time path
2000 100.8 - 5.5 99.2 3.7
2010 100.2 - 6.4 98.7 4.7
2020 97.9 - 3.5 96.4 2.6
2030 93.9 -4.7 92.4 3.7
2040 91.0 -7.9 89.6 6.7
2050 90.5 - 8.7 89.1 7.7
New steady state 90.5 -8.7 89.1 7.7

Source: OECD(1988)and authors'calculations.
a. The table shows the results from the open-economydemographicsimulation.We use the needs-weighted
measurefor consumption(CON2)and the unweightedlaborforce (LFI). Consumptionis normalizedto 100in the
initialsteady state. Foreigncapitalownershipis the percentageof assets in each countryownedby foreigners.It is
initiallyzero in bothcountries.The elasticitiesof substitutionin productionand consumptionboth equalunity.
To financethe additionalconsumptionindicatedin the simulations,

the United States runs a currentaccount deficit. Figure 14 shows the
path of net national saving and net investment, and, as residual, the
current account. For about 15 years, the United States runs current
account deficits, so that more than 6 percent of U.S. assets are owned
by foreignersin 2010. High savingfor the subsequent15years results in
currentaccount surplusesand reduces foreign capitalownershipto 3.5
percent. Past 2020, however, with the rapidincrease in the numberof
elderly, the United States againrunscurrentaccount deficits, so that in
the steady state almost9 percentof U.S. assets areowned by foreigners.
For the non-U.S. OECD, consumption declines 0.6 percent when
tradewith the United States is permitted.The availabilityof investment
projects in the United States means that higher saving in the short run
will not depress rates of returnby as much as in the closed-economy
case.
The open- and closed-economy cases yield different consumption
levels in both the short run and the steady state. In the open-economy
Figure 14. Net Saving and Investment Rates, United States, from Two-Country
Simulations, 1990-2050
Percentof GNP
5.5
Investmentrate
5 J1/
4.5
4 %
%
Saving l-ate
3.5-
-
3
2.5-
2-
1990 2000 2010 2020 2030 2040 2050

Year
Source: Authors' calculations and OECD (1988).
case, U.S. consumption is higher in the early stage of the transition

because of the availabilityof foreign capital. The resultingdecrease in
asset accumulationtranslatesinto a 1.8 percentagepoint reduction in
steady-stateconsumptionfromthe closed-economy simulation.For the
non-U.S. OECD, the effect is reversed: greater capital accumulation
along the transitionpath leads to steady-stateconsumption1.3 percen-
tage points higherthanin the autarkysteady state.
These results suggest two conclusions. First, the pattern of demo-
graphicchangein otherdeveloped nationscan have a largeeffect on the
optimal consumptionresponse to demographicchange in the United
States. The importanceof these effects depends criticallyon the degree
of capitalmarketintegration.Second, because the United States is aging
more slowly than other OECD nations, the optimal consumption re-
sponse in the open economy entails higherinitial consumptionthan in
the autarkycase and thus a currentaccount deficit.
Demographic Change and Productivity Growth
The foregoingcalculationsassume that demographicchanges affect

productivityonly by causing changes in capital intensity. The rate of
technicalchange, or equivalentlyof totalfactorproductivity,is assumed
to be independentof demographicdevelopments. But if demographic
factors affect technical change, the implicationscould be quite signifi-
cant, both for living standardsandfor optimalcapitalaccumulation.
Thereare severalpotentiallinksbetween demographicdevelopments
and the rate of technologicalchange. One argument,stressed recently
by JulianSimonand Ben Wattenberg,holdsthat slow populationgrowth
reduces the rate of technicalprogress.35The argumenthas two strands.
First, a rapidlygrowingpopulationenlargesthe marketfor capitalgoods
(the Solow effect noted above), makinginnovationmore profitableby
permittinggreaterspreadingof fixed costs. As populationgrowthslows,
innovationbecomes less profitable.Second, as the shareof the popula-
tion that is young and innovates declines, the aging society loses some
of its "dynamism" and experiences slower technical change. As de-
scribedby FrenchdemographerAlfredSauvy, such a futurewould hold
"a society of old people, living in old houses, ruminatingabout old
ideas."36
A more optimistic argument,advanced by H. J. Habakkuk,is that
incentives to innovate are strongest when labor is scarce.37Habakkuk
arguedthatindustrializationproceededfasterinAmericathanin England
because attractiveagriculturalopportunitiesraisedthe price of laborin
the United States relativeto that in England,where laborwas abundant
and less expensive. Paul Romerhas formalizedthis argumentand used
it to explainthe apparenttendencyfor abnormallyrapidU.S. productiv-
ity growthin periodsof relatively slow laborforce growth.38
The relative importanceof these mechanismscan only be assessed
empirically.Unfortunately,there are no ideal experimentsfor consid-
eringthe effects of demographicchangeon productivitygrowth. Below
we draw on the differingdemographicexperiences of relatively high-
35. Simon(1981)andWattenberg(1987).
36. Wattenberg(1987,p. 65).
37. Habakkuk(1962).
38. Romer(1990).
income countries to assess the likelihood that an agingpopulationwill

lead to economic stagnation.
Evidence on Productivityand Demographic Composition

Our empiricalwork uses the 1960-85 internationalcomparisondata
of Robert Summers and Alan Heston.39Unfortunately,data on total
factor productivityare not available for a wide sample of countries.
Instead, we study the relation between labor force growth and labor
productivitygrowth.40
We selected countrieswith 1960laborproductivityat least 30 percent
of U.S. productivity(we use income per worker as our productivity
measure)and excluded the OPEC countries, thus generatinga sample
of 29 countries. Selecting on initialincome avoids the bias of including
only countries that have experienced large productivity growth, as
BradfordDeLong highlights.41We omit countries with very low initial
productivitybecause the role of laborforce growthmaybe very different
in pre-industrialsocieties. Japanis omittedbecause its productivitywas
only 25 percentof U.S. productivityin 1960.
Figure 15 plots annual productivitygrowth and annual labor force
growth during 1960-85. The data show a strong negative correlation.
Slower-growingcountries, including most European nations, exhibit
above-averageproductivitygrowth, while more rapidlygrowingcoun-
tries such as Canadaand Australiahave lower productivitygrowth.
To controlfor additionalfactors affectinggrowth,we estimatecross-
section regressionsof the form:
(18) ln (yj1/y0j)1T = o0 + al ln (LFj,j/LF0,j)/T
+ (X2 ln (yo,i)+ (X3 (I Y)i + Ei,
whereyli andyo, are, respectively, finalandinitialoutputperlaborforce

member;LF1,jand LF0,jare the finaland initiallaborforce; (I/Y)iis the
averageinvestmentrateduringthe sampleperiod;i denotes the country;
and T is the lengthof the sampleperiod.The investmentrate is included
to control for changes in capital that affect labor productivitybut not
39. SummersandHeston(1990).
40. We presentlimitedevidence below suggestingthat the differencebetween labor
productivityandtotalfactorproductivitydoes not have a largeeffect on our results.
41. DeLong(1988).
Figure 15. Productivity Growth and Labor Force Growth, Selected Countries, 1960-85a
Productivitygrowth (percentannually)
6
S OJapan
4-
*Italy
3 00 France
Germanya
2_ Canad
Kingdom _
1 * * ~~~~~~~~~~UnitedSt
O I I I I I I I I I I * I I I I
0.4 0.8 1.2 1.6 2 2.4 2.8 3.2
Laborforce growth (percentannually)
Source: Authors' calculations based on data in Summers and Heston (1990). The G-7 countries are labeled on the
figure.
a. Japan is included in the figure but not in the regression line.
total factor productivity. Initial income is included to capture the

possibility that lagging countries grow more rapidly as they converge
toward leading ones. Productivitygrowth and labor force growth are
expressed at annualrates.
The upperpanel of table 7 reportsordinaryleast squaresestimates of
equation 18. The coefficients in the bivariateregressions, analogousto
figure 15, imply that a 1 percentagepoint decrease in the annuallabor
force growthrate raises productivitygrowthby 0.62 percentagepoint a
year. Controllingfor the initiallevel of productivityandinvestmentrates
has little effect on the laborforce growthcoefficient, with the estimates
still negative (-0.64) and large. The data also suggest that more rapid
investment leads to faster productivity growth, although there is no
evidence of productivityconvergencefor this sample.
We estimated equation 18 with other samples of countries, with
similarresults. If we includethe six OPECcountrieswith 1960produc-
Table 7. Demographic Change and Productivity Growth, 1960-85, Various Periodsa
No controls Withcontrols
Logarithm
Laborforce _ Laborforce of initial Investment
Period growth R growth produictivity rate R
Ordinaryleast squaresestimates
1960-85 -0.617 0.281 -0.637 -0.346 0.063 0.421
(0.179) (0.161) (0.434) (0.022)
1960-73 - 1.061 0.389 -1.044 -0.730 0.064 0.460
(0.245) (0.232) (0.647) (0.028)
1973-85 - 0.258 0.025 -0.295 0.154 0.075 0.175
(0.198) (0.195) (0.546) (0.031)
Fixed effects -0.903 0.085 -0.446 -6.290 0.032 0.600
(0.477) (0.355) (1.177) (0.057)
Instrumentalvariablesestimates
1960-85 -0.711 ... -0.742 -0.337 0.064
(0.216) (0.189) (0.438) (0.022)
1960-73 - 0.977 . .. - 0.956 - 0.763 0.064
(0.297) (0.272) (0.651) (0.028)
1973-85 - 0.436 ... -0.610 -0.150 0.085 ...
(0.272) (0.296) (0.610) (0.033)
Fixed effects - 0.840 ... 0.332 -7.273 -0.017
(1.151) (1.440) (2.174) (0.107)
Source: Authors'calculationsbasedon data in Summersand Heston (1990).
a. The dependentvariableis the annualproductivitygrowthrate.The laborforcegrowthrateand investmentrate
are both annualrates. The sampleconsists of the 29 non-OPECcountrieswith 1960income per workerabove 30
percentof the U.S. level. The upperpanelreportsordinaryleast squaresestimates.The lowerpanelinstrumentsfor
the growthrate of the laborforce with the growthrate of the population.Standarderrorsare in parentheses;R2s
are adjustedfor degreesof freedom.
tivityabove30percentof the U. S. level, the coefficientinthe multivariate

regressionrises to -0.517 (0.144). If we limit the sample to countries
with 1960productivityat least 50 percent of that in the United States,
the coefficient becomes -0.263 (0.192). If we consider the current
OECDcountries,the coefficientis - 0.372 (0.161). Finally, if we include
all 114countriesin the Summersand Heston data with at least 20 years
of data, the coefficient becomes - 0.507 (0.159).
Dividingthe period 1960-85 into two shorterintervals, 1960-73 and
1973-85,shown in the second and thirdrows of the upperpanel, allows
us to examinethe importanceof the productivityslowdown in the mid-
1970s.The resultsfromthese regressionsare consistent with those from
the full sample,althoughthe evidence is strongerin the 1960-73period.
In the earlierperiod, the coefficient (- 1.044) is much larger and still
statisticallysignificantlydifferentfrom zero. In the post-1973 period,
the coefficientfalls to - 0.295 and is no longer statisticallysignificant.
The fourth row of the table presents the results of treatingthe two
sampleperiodsas a panelandestimatinga fixed-effectsregression.This
specificationcontrolsfor otherfactorsthatcan explainpersistentdiffer-
ences in growth rates across countriesbut that are not includedin our
set of explanatoryvariables.The resultsarequalitativelysimilarto those
without the fixed effects. The coefficient in the multivariateregression
(- 0.446)is withinthe rangeof the estimatesfor the two sampleperiods,
althoughthe coefficient is not statistically significantlydifferentfrom
zero when we controlfor initialincome and the investmentrate.
The lower panel of the table reportsinstrumentalvariablesestimates
of the sameequations,usingthe populationgrowthrateas an instrument
for labor force growth. If rapidproductivitygrowth leads to less rapid
increases in labor force participation,the ordinaryleast squares esti-
mates will be biased, but the instrumentalvariablesregressionswill not.
The instrumentalvariablesestimatesstronglyconfirmthe ordinaryleast
squares estimates. In the 1960-85 regression, the coefficient on labor
force growth becomes more negative in the instrumentalvariables
regression( - 0.742) and is still statisticallysignificant.The coefficients
on the othervariables,in contrast,changelittle.
As the second and third rows of the bottom panel suggest, this is
principallydue to a more pronouncednegative relationbetween labor
force growthandproductivitygrowthduring1973-85.This is consistent
with RichardFreeman'sclaim thatthe decline in productivitygrowthin
post-1973 Europe discouragedlabor force participation,leading to a
positive bias in the coefficienton laborforce growthrates.42
The final row presents the results for the instrumentalvariables
regressionwith the fixed-effects specification.While the coefficient in
the bivariateregressionis similarto the ordinaryleast squaresestimate,
the coefficient in the multivariateregressionis positive. In both cases,
the coefficientson laborforce growthare not statisticallydifferentfrom
zero.
Because reductions in the labor force growth rate tend to increase
capitalintensity,one would expect themto be associatedwith increases
in labor productivitygrowth even if they had no impact on technical
change.43We doubtthatthe equationsin table 7 areprimarilypickingup
42. Freeman(1988).
43. Mankiw,Romer,andWeil(1990)explorethis possibilitywith particularattention
to the role of humancapitalaccumulation.
this effect for two reasons. First, its theoretical magnitudeis much
smallerthanthe effects impliedby the cross-countryequations. Over a
25-year period, a 1 percentage point reduction in labor force growth
holding the saving rate constant would raise labor productivityby at
most 0.17 percentagepoint assuminga Cobb-Douglasproductionfunc-
tionwith a 67 percentlaborshare." Second, for a smallsampleof OECD
countrieswith available data (18 countries) we estimated productivity
growthequationsusing both laborproductivityand total factor produc-
tivity andfound only negligibledifferencesin the results.45
These regressions imply substantivelylarge effects of demographic
changeon futuregrowth. Because the annuallaborforce growthrate is
predicted to fall by about 1 percentage point between 1990 and 2050,
with most of the changeoccurringbetween 1990and2010, ourestimates
imply an increase of about 0.6 percentagepoint in annualproductivity
growth. Such effects are large enough to offset the decline in living
standardsthatwe presentedabove. Even a 0.2 percentagepointincrease
in annualproductivitygrowth between 1990and 2040 would offset the
roughly 10 percent decrease in per capita consumptionas a result of
rising dependency burdens over that period. Thus, even if the effects
aremuchsmallerthanthose fromourregressions,they are likely to have
a largeimpacton futureliving standards.
Theregressionsthusfarpresentlittleevidenceforthemorepessimistic
view of demographicchange. It may be, however, thatpartof the effect
of demographicchange occurs throughthe investment rate. If slower
laborforce growthreduces the rate of innovationbecause of decreased
demand for capital goods, that will show up as a positive effect of
investmentrateson productivitygrowth,ratherthanas an effect of labor
force growth.
To considerthis hypothesis, we reestimatedthe equationsin table 7
withoutcontrollingfor the rate of investment. The results change little
fromthose reported.For the fulltimeperiod,for example,the coefficient
44. The predictedeffect is this largeonly if the base year for our observations(1960)
is the firstyearof the new laborforce growthrates. If the countrieswerealreadyin steady
state with differentlabor force growth rates, there would be no predicted effect on
productivityfromthis explanation.
45. Withoutcontrolsfor initialproductivityand investmentrates, the coefficienton
the growthrateof the laborforce is - 0.788 (0.207)in the equationfor laborproductivity
and - 0.696 (0.257) in the equation for total factor productivity.In the multivariate
regression,the coefficientsare - 0.305 (0.216)and - 0.259 (0.324)in the two equations.
on laborforce growthfalls only slightly,from - 0.637 (0.161)to - 0.617

(0.182). In no case does the coefficient on labor force growth fall
substantially,and in manycases it becomes more important.
It is also possible that our measureof demographicchange is not the
best measurefor examiningthe productivityconsequences of changes
in the populationstructure.The argumentthatolderworkforces are less
innovativethantheiryoungercounterpartssuggests that a variablelike
the averageageof the workforceis a moredirectmeasureof demographic
conditions. Our measure of labor force growth rates is only partly
correlatedwith this type of demographicvariable.
We explored this possibility by addingthe average age of the labor
force to the equationsin table 7.46In the basic specificationin the first
row of the table,whenthe averageage variableis included,the coefficient
on the labor force growth rate declines to - 0.483 (0.225), and that on
the average labor force age is 0.135 (0.138). Neither the coefficient on
initialproductivitynorthaton the investmentratechangessubstantially.
Similarconclusionsemergefor the other specifications.47
To the extent thatthe laborscarcityhypothesisis correct,it reinforces
our conclusion that the maturingof the labor force expands society's
opportunities.Fasterproductivitygrowthhas a theoreticallyambiguous
effect on the level of currentconsumption,however. It tends to increase
consumptiontoday because of the income effect of increased output,
but this effect can be offset by a substitutioneffect from the increased
investmentreturnas the effective supplyof laborgrows more quickly.
To evaluate these effects for current consumption decisions, we
calculated the optimal consumption path when productivity growth
changes over time. We assumedthat each percentagepoint decrease in
labor force growth increases productivity growth by 0.5 percentage
point, a numberin the range of those in table 7. Figure 16 shows the
resultingconsumptionpath, as well as the consumptionpathwithoutthe
46. To accountfor changesin the averageage of the laborforce over the time period
of our productivitygrowthmeasurements,we definedthe averagelaborforce age over
any periodas the meanof the averageage at the endpointsof the period.
47. We intendto explorethese issues furtherin subsequentwork. Preliminaryresults
suggestthatthe evidencefor beneficialeffects of slow laborforce growthis muchweaker
for the 1870-1960periodthanfor the post-1960period.This may be a consequenceof the
simultaneitycausedby muchlargerimmigrationflows in the earlyperiod.At this point,it
seems fairto concludethatthereis no internationalevidenceforthe dynamismhypothesis
that morerapidpopulationgrowthor a youngerpopulationraisesproductivity,and some
evidencefor the contrarylaborscarcityhypothesis.
Figure 16. Consumption Response with Induced Productivity, United States, 1990-2065
Per capita consumption (index, 1990 = 100)
420
380 -
340 -
300 IIncreasingproductivity
220-
100~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~1
180 - --- -Constant pr-oductivity
140-
100 I
2000 2010 2020 2030 2040 2050 2060
Year
Source: Authors' calculations. For increasing productivity, it is assumed that productivity grows 0.5 percent for
each I percent decline in annual labor force growth. The consumption measure is needs-weighted (CON2), and the
labor force is earnings-weighted (LF2). The simulations assume unit elasticities of substitution in production and
consumption.
productivityeffects, followinga demographicshock like those examined

in the previous sections.48The effect of increased productivitygrowth
is to increase current consumption even more, by an additional 0.2
percent. Further,because most of the productivitybenefitsoccur in the
next several decades, when the laborforce grows slowly, consumption
remainsabove its initiallevel throughoutthe transitionpath to the new
steady state.
DemographicChange and Fiscal Policy
The preceding section suggests that, other things being equal, the
optimal response to recent and projected demographicchanges is a
laborforce andneeds-
48. Weuse the supportratiodefinedwiththe earnings-weighted
weighted consumptionmeasure. We also assume unit elasticities of substitutionin
productionandconsumption.
46 Brookings Papers on Economic Activity, 1.1990
decline in the national saving rate. The implicationsfor fiscal policy

depend upon how the private saving rate responds to demographic
changes. Tax-smoothingconsiderationsmay also imply particularpat-
terns of optimalfiscal policy. This section examines these issues.
The effect of the population'saging on private saving has been the
subjectof a numberof analyses, but no firmconclusionhas yet emerged.
From the standpointof the life-cycle hypothesis, slowing population
growthand an aging society shouldbe associated with reductionsin the
private saving rate. As the aged share of the populationincreases, the
ratio of dissavers to savers rises and so the private saving rate falls.
David Weil has recently pointed out that this effect may be reinforced
by anincreasein expectedbequestspermemberof the adultpopulation.49
On the other hand, many analysts have arguedthat the maturingof the
baby boom generationwill raisepersonalsavingbecause people borrow
when youngand save as they approachmiddleage. Increasesin personal
saving may also resultfrompeople havingfewer children.
Summersand Carrollexplore the impactof demographicchanges on
savingbehaviorby assumingconstantage-specificsavingrates.50Figure
17uses the age-specificsavingratesfromtheiranalysis, as well as Social
SecurityAdministrationpopulationforecasts, to projectpersonalsaving
rates over the next 30 years. The results suggestthatthe maturingof the
populationwill be associatedwith a smallincreasein savingratesduring
the next three decades. Calculationsby Alan Auerbachand Laurence
Kotlikoffreach similarconclusions.51
A near-termincreasein privatesavingprovides a furtherreason why
an economy with an initially optimal saving rate should loosen fiscal
policy in response to changingdemographicconditions.
There is, however, a different argumentfor a tight fiscal policy.
Projecteddemographicchangesimplysignificantfluctuationsin the level
of government spendingover the next century, since transfersto the
49. Weil (1989).
50. SummersandCarroll(1987).
51. Auerbachand Kotlikoff(1989). Both sets of calculationsare flawed in ignoring
pension saving, which may changeas the age structureof the populationchanges. They
also take no account of changes in the numberof childrenor in the numberof people
supportingagedparents,althoughthese factorsmayaffectage-specificsavingrates.David
Weil (1990)uses aggregatedata on OECD countriesto study saving, recognizingthese
effects. His resultssuggestthatprivatesavingin the UnitedStatesmayriseabout1percent
in the next decadeas a resultof demographicfactors.
Figure 17. Projected Private Saving Rate, United States, 1990-2065a

Saving rate (percent)
5.0 -
native H
k ~~~~~~~~~~~~~~~~~~~Alt
4.5 - Altei-native I
AlternativeIII
4.0_ .
1 1, 1 1 1 l l I I l I l l I
1990 2000 2010 2020 2030 2040 2050 2060
Year
Source: Summers and Carroll (1987); Board of Trustees of the Federal Old-Age and Survivors Insurance and
Disability Insurance Trust Funds (1988).
a. The calculations assume that the saving rate is held constant by age.
elderly are much larger than those to any other group. Efficiency
considerationsargue for higher current taxes to fund foreseeable in-
creases in governmentoutlays. Because the deadweightloss of taxation
increaseswith the squareof the tax rate, financingthe anticipatedrise
in governmentoutlays on a pay-as-you-gobasis, with lower tax rates
duringthe next few decades and higherones in the middle of the next
century, entails a larger deadweight burden than a constant tax rate
policy.52 This argumentparallels the traditionaljustificationfor using
debtto financewarsandothertransitoryshocksto governmentspending.
To evaluate the empirical significanceof tax-smoothingconsidera-
tions, we begin by describingthe age-specific pattern of government
outlays.Wethenpresenta simpleframeworkforevaluatingtheefficiency
52. Barro(1979)describesthe "tax-smoothing"view of optimalgovernmentfinancial

policy.
Table 8. Per Capita Government Spending, by Age, United States, 1989

1989dollars
Social
security
and Health
Age group disability care Education Total
0-4 132 872 674 1,678
5-14 132 690 3,353 4,175
15-19 132 298 2,930 3,360
20-24 16 298 1,112 1,426
25-44 83 298 233 614
45-64 811 218 84 1,113
65 and over 6,138 3,526 84 9,748
Total 925 824 873 2,622
Sources: Social Security Administration (1987); Waldo and others (1989); Board of Trustees of the Federal
Supplementary Medical Insurance Trust Fund (1988). For details regarding construction of the data, see notes to
table 9.
gains from tax smoothing and report suggestive calculations. These

findingsimply relatively small efficiency improvements-on the order
of 1 percent of one year's GNP-from stabilizingtax rates throughout
the next half-century.
Age-Specific Patterns of Government Spending

Governments spend different amounts on individuals of different
ages. Outlays on education, for example, benefit primarilychildren,
while the elderly are the principalbeneficiaries of most government
spending on health care and social security. Even without changes in
the structureof governmentprograms,demographicshiftscan therefore
affect the level of governmentspending.
Table 8 presents age-specific governmentexpenditurepatterns for
the United States, focusing on the three largest social expenditures:
social security, health care, and education. The first column shows
spendingon Old-Ageand Survivorsand DisabilityInsurance.Virtually
all such expendituresare directed to individualsaged 65 or over, with
averageoutlays in 1986of $6,138per person. The second column shows
analogousage-specific spendingpatternsfor health care, with average
expendituresper person aged 65 and over ($3,526)morethanfour times
largerthan outlays for any other age group. The third column reports
Table 9. Projected Government Expenditures, United States, 1990-2060

Percent of GNP
Social
security
and Health
Year disability care Education Other Total
1990 4.7 4.1 4.7 18.0 31.8
2000 4.5 5.3 4.9 18.0 32.9
2010 4.6 5.9 4.9 18.0 33.4
2020 5.6 6.5 4.8 18.0 35.0
2030 6.5 7.4 4.9 18.0 36.7
2040 6.5 7.8 4.9 18.0 37.1
2050 6.5 7.8 4.9 18.0 37.1
2060 6.5 7.8 4.9 18.0 37.0
Sources: Social security and disability spending are predicted from projected population growth rates. For the
1989 distribution of spending, we projected the year-end 1986 distribution from the Social Security Administration
(1987) to 1989, using the GNP deflator. Spending at below retirement ages is the sum of OASDI payments to the
disabled, payments to early retirees, and payments to surviving children and spouses. Spending on all persons below
20 years of age was treated as applying uniformly to the members of this group.
For health care spending, we combined four types of spending. We obtained 1987 estimates of Medicare and
Medicaid per capita spending on the elderly for hospital care, physicians' services, nursing home care, and other
personal health care from Waldo and others (1989). For the nonelderly population, we calculated government spending
on each of these categories as the difference between the Division of National Cost Estimates, Office of Actuary,
Health Care Financing Administration (1987) estimate of 1987 total government spending for that category and the
implied spending on the elderly. This estimate includes both Medicaid spending for the nonelderly and medical care
spending for government employees. We distributed this spending by age on the basis of Medicaid spending, as
presented in Public Health Service (1989). All of the estimates were converted to 1989 dollars using category-specific
projections of 1987-1990 inflation in Division of National Cost Estimates, Office of Actuary, Health Care Financing
Administration (1987).
We forecast spending using estimates of inflation rates for the four categories of spending and projections of the
age distribution of the population. Hospital care estimates are from the Board of Trustees of the Federal Supplementary
Medical Insurance Trust Fund (1988). They imply a steady-state inflation rate above general inflation but below the
growth rate of output. Inflation rates for the other three categories were projected to 2000 using the Division of
National Cost Estimates, Office of Actuary, Health Care Financing Administration (1987) estimates, and were
assumed to grow at the rate of hospital price inflation after that.
Finally, for age-specific spending on education, we obtained 1986 age-specific enrollment rates in school as well
as the aggregate amounts spent on primary and secondary education, and higher education. We assumed that all
persons under 17 who were enrolled in school were in primary and secondary schools, and all persons 18 and over
who were enrolled were in higher education. Spending per person was then the weighted average of the population
in each age group and the share of each age group in the two types of education. Our projections assume that
education spending would grow at the rate of GNP growth, so that changes in the share of GNP devoted to education
change only with changing numbers of young people.
the age profileof educationspending.Percapitaexpenditureson schools

for the younger cohorts are substantial, reaching $3,353 a year for
childrenbetween the ages of 5 and 14. For the threeprogramscombined,
spendingon the elderlyis more thandouble that of any other group.
Demographicshifts can significantlyaltergovernmentoutlays. Table
9 reports projections of total government outlays as a share of GNP
underthe assumptionthat age-specificexpenditurepatternsremain at
1989 levels for the next 60 years. Primary government spending is
assumed to equal a constant fraction of GNP. In these projections,
governmentspendingrises from 31.8 percent of GNP in 1990 to 37.0

percent of GNP in 2060, with nearly all the increase due to changes in
medical expenditures and transfer programsto the elderly. Our tax-
smoothingcalculationassesses the efficiency gains from smoothingthe
timepathof revenuesneededto collect this variableexpenditurestream.
The Efficiency Gains from Tax Smoothing

We evaluate the efficiencygainfromtax smoothingby assumingthat
the deadweightburdenof raisingTpercent of nationaloutputin taxes is
given by
(19) DWL, = ET2Yt/2.
The parameterE depends on the elasticities of aggregate supply and
demandand Y,is nationalincome.53The marginaldeadweightloss per
dollar of revenue raised is ET We calibrate E by setting the marginal
deadweightloss from raising one dollar equal to 30 cents, the upper-
bound estimate in CharlesBallard,John Shoven, and John Whalley's
generalequilibriumanalysis of the U.S. tax system.54Theircalculation
employs 1973data, when federal and state-localtaxationin the United
States was 31 percentof GNP, and thereforeimpliesE - 1.0.
We assume that a governmentplannerseeks to minimizethe present
discountedvalue of the deadweightlosses fromtaxationover a Tperiod
horizon:
T [t
(20) V f7 (1 + r)- ET2Y,/2,

t=l il
where ri denotes the one-periodnominalinterest rate in period i. This

minimizationis subject to an intertemporalbudget constraint linking
taxes and spending as a share of GNP (T, and ht, respectively) with
governmentdebt as a shareof GNP (di). For each period, this constraint
is:
(21) dt = dt 1 [(1 + r1)I(1+ yt)] + ht - Tt,
53. If governmentsset taxes to minimizedeadweightloss, the marginaldeadweight

burdenper revenuedollarshouldbe equalacross tax instruments.The aggregatetax-to-
GNP ratiois then a simpleproxyfor the level of tax burdens.This convenientassumption
neglects the voluminouspublic finance literaturesuggestingthat marginaldeadweight
losses varyacrosstax instruments.
54. Ballard,Shoven, andWhalley(1985).
where r, is the interest rate and y, is the rate of output growth.

Summingthis forwardyields a budgetconstraintof the form:
T T
(22) E
t=l
Ttt =
t=1
ht8t + do - d7ZT,
where 8t = H=1 (1 + ys)/(l + rs). Minimizing equation 20 subject to

equation22 yields first-orderconditionsof the form
(23) ETt= A,
so the optimalpolicy calls for equal tax rates in each period.

In the case where r, = -yt,the benefits of tax smoothing take a
particularlysimpleform. The budgetconstraintis
T T
(24) >LTt= lht+do-dT-

t=l t=1
If, further,dT = do, then with a pay-as-you-gopolicy, taxes just cover

government spending: T, = h. Under the constant period-by-period
debt-to-GNPpolicy, the deadweightloss is
(25) DWL1 = (Y0 e/2)>ht2.
The constanttax rate satisfyingthe governmentbudgetconstraintis just
the averagevalue of governmentspending,so that
(26) DWL2 = YoE/2*T (>ht/T)2.
Thus,the relativedeadweightloss fromthe optimaltax-smoothingpolicy
is:
(27) DWL2/DWLI = (Eht1J)21(Eht211).
For the expenditurepath in table 9, the deadweightloss reduction in

equation27 is 0.3 percent.
The incrementaldeadweightloss fromtime-varyingtax ratesdepends
on the precise time path of taxes, hence on the government'schoice of
debtpolicy. We considertwo such policies. The firstassumes a constant
debt-to-GNPratio in every year, and the second assumes a constant
primarysurplus(equalto its value in 1989of 0.5 percentof GNP) in each
year.
Table 10 presents our estimates of the efficiency gains from tax
smoothing.The upperpanel presents results assuminga constant debt-
52 Brookings Papers oni Economic Activity, 1 :1990
Table 10. Efficiency Gains from Smoothing Taxes, United States, 1990-2050a
Percent
Average
deadweight
loss as
Year petcent of
avetrage
Tax rate 1990 2010 2030 2050 GNP
Constant debt-GNP ratiob
Variable rate 32.6 33.5 36.8 37.1 6.23
Constant rate 35.3 35.3 35.3 35.3 6.22
Constant pritnaty slurplusc
Variable rate 32.3 33.9 37.2 37.6 6.52
Constant rate 35.7 35.7 35.7 35.7 6.51
Source: Authors' calculations based on sources described in table 9.
a. For each spending category, total government expenditures are projected to 2060, as described in table 9. The
two cases are described in more detail in the text.
b. Constant debt-GNP ratio is fixed at 1989 level of 50.2 percent.
c. Constant primary surplus assumes that a federal and state-local surplus equals its 1989 share of GNP throughout
1990-2000.
to-GNP ratio, fixed at its 1989level of 50.2 percent. In this case the pay-
as-you-gotax raterises from32.6 percentof GNP in 1990to 37.1 percent
by 2050.The averagedeadweightloss fromthis policy, shown in the last
column, is 6.23 percent of the average value of GNP. The constant tax
rate that achieves the same debt-to-GNPratioof 50.2 percentin 2050is
35.3 percent. Under this plan, taxes would rise by 3 percent of GNP-
roughly $150 billion-in 1990. Despite this large change in the debt
trajectory,however, the change in excess burdenis small. The average
value of deadweightloss is 6.22 percentof averageGNP when tax rates
are smoothed. The improvementin deadweight loss averages 0.017
percent of GNP annually,or less than $1 billion a year in 1990dollars.
The changein the present value of deadweightlosses between 1990and
2060equals 1.1 percentof 1990GNP, or approximately$55 billion.
The lower panel in table 10 shows parallelcalculationsassumingthe
combinedfederal and state-local primarysurplusequals its 1989 share
of GNP value throughoutthe 1990to 2050 period. The results indicate
that the averageexcess-burden-to-GNPratiounderthis scenariois 6.52
percent, comparedwith 6.51 percent if the tax rate is smoothed. The
difference between these two efficiency costs is similarto that in our
firstcase, 0.017 percentof GNP. Plausiblevariationsin ourassumptions
aboutthe debt-to-GNPtrajectorythereforedo not appearto have large

effects on the efficiencygains from tax smoothing.The generalconclu-
sion of these calculationsis that there is only a weak tax-efficiencycase
for prepayingthe costs of the futuredependencyburden.
Conclusions
Demographicchanges currentlyin progress do not appearlikely to

worsen economic performancein the United States, at least duringthe
next several decades. While increased dependency will reduce living
standardsby 5-10 percentin the long run, demographicchanges will be
beneficialover the next 20 years. In the shortrun, demographicchange
will have two importanteffects. First, slowing populationgrowth will
permita smallershareof nationaloutputto be devoted to investmentin
plant, equipment,andhousing. Second, the shareof the populationthat
is workingwill rise, largely as a result of the fallingrelative population
of children. These positive effects of demographic change may be
reinforcedby increasedforeigncapitalinflowsandacceleratingtechnical
changeas firmsrespondto an increasingscarcityof labor.
Recent and prospective demographic changes do not appear to
warrantincreasing the national saving rate. These changes increase
wealth in the short run, reduce the rate of returnto saving, and attract
foreign capital. Holding all else equal, their net effect would be a
reduction in the optimal national saving rate. Nor do tax-smoothing
factorsrepresentan importantargumentfor tryingto prepaythe govern-
ment's prospectiveliabilityto supporta dependentpopulation.Thereis
little efficiencyloss in followinga pay-as-you-gopolicy with variabletax
rates.
Ourconclusiondepartsfrom analyses, such as that of Henry Aaron,
BarryBosworth,andGaryBurtless, thatrecommendaccumulationof a
large social security trust fund to bolster U.S. nationalsaving.55These
positions are not necessarily inconsistent, however. A first line of
reconciliationwould hold that apartfrom demographicconsiderations,
Americannationalsavingis much too low rightnow and that the social
securitytrustfundprovidesa politicallyconvenientway of reducingthe
55. Aaron,Bosworth,andBurtless(1989).
54 Brookings Paipers on Economic Activity, 1:1990
federalgovernment'sabsorptionof private saving.56The decline in the

private saving rate from an average of 7.1 percent duringthe 1970s to
about3 percentduring1986-89is greaterthanwhatouranalysissuggests
can be justified by demographicfactors. There are even some reasons
for advocating an increase in the U.S. national saving rate to levels
above those observed historically,particularlyin light of the emerging
need for capitalin EasternEuropeand the signs that savingis declining
outside of the United States.
A second potential reconciliationof these views involves questions
of optimal intergenerationalredistribution.Some argue for using the
social security trust fund to raise the national saving rate in order to
avoid unfairlyburdeningour children.The primarythrustof this argu-
ment-that we need to preparefor the anticipatedburdenof increased
dependency-is exactly what our supportratiocalculationsreject. This
is because the dependencyburdenis remote, and because slower labor
force growth means more rapidly diminishing returns to additional
saving. Admittedly, our approachfocuses on the economy's year-by-
year consumptionlevel, ratherthanthe welfareof individualcohorts. It
is thereforepoorly suited to addressingargumentsthat certain cohorts
will be greatly disadvantagedwithout additionalcapital accumulation.
However, we do not find compellingthe claim that our childrenwill be
unfairlyburdenedunless we increasecapitalaccumulationtoday.
Two argumentsmilitateagainstthe intergenerationalequity case for
trust fund accumulation.First, if the aforementionedfears of inequity
were correct, the appropriateresponse should be an adjustmentin the
level of prospective intergenerationaltransfers,not a change in capital
accumulationpolicy. Just as concerns about the income distributionat
a particulartime are better addressed through transfer policies than
throughchangesin the mix of productsproduced,transfersare the right
way to respondto concerns aboutintergenerationalequity.57
Second, other considerationsoperateto make the baby boom gener-
ationless well off thanits successors. The baby boomers systematically
lose because of theirlargecohort size. Duringtheirworkingyears, wage
growthis slow becauseof low capital-laborratios.Duringtheirretirement
years, the numberof potential purchasersof capital will fall, thereby
56. See, for example,Hatsopoulos,Krugman,and Summers(1988).

57. Thisis the centralpointmadeby CalvoandObstfeld(1988).
reducingthe rate of returnon saving.58Moreover, given productivity

growth,the next generationwill be considerablymore affluentthan the
current one. If slower population growth or foreign capital inflows
acceleratethis tendency, the case for intergenerationalredistributionis
reduced. Even with our estimates of the path of optimalconsumption,
alongwhich a decliningsupportratioreduces consumption,the lifetime
utility of a person who lives for eighty years rises for those born from
1990until 2020. Only after 2020 does the lifetime utility of new cohorts
fall below that of those who lived before the demographicchange.
Our aggregateanalysis cannot resolve policy debates about raising
the birthrate or increasingimmigration,since these debates often focus
on microeconomiceffects and distributionalconsequences. Moreover,
there is a fundamentalpoliticaldifficultyof deciding "who is us." How
should the welfare of immigrantsbe treatedin decidingwhether or not
to accept more of them?How should the utility of an otherwise unborn
childbe treated?Policy recommendationsareimpossiblewithouta clear
philosophicalresolutionof these questions.
Our analysis does, however, cast some doubt on the view that in
narroweconomic terms, higherfertilityis helpfulin reducingthe burden
of dependencyin old age. Dependency at the beginningof the life cycle
is between 50 percent and 100 percent as costly as dependency at the
end of the life cycle. It also comes 60 years earlier. Furthermore,the
weak availableevidence suggeststhat slowerpopulationgrowthis more
likely to raise thanto reduce productivitygrowth.
For the set of issues capturedby our analysis, there is a strongercase
for increased immigrationas a way of reducing dependency. Most
immigrantsarriveas young adults and so begin workingwithout being
dependentsfirst. To the extent that they immediatelystartpayingtaxes
for the supportof the elderly, they may increase economic welfare of
the preexistingpopulation,even if they areultimatelyeligiblefor transfer
paymentsin old age.
We have only scratchedthe surfacein assessing the macroeconomic
implicationsof demographicchange. Among the main priorities for
future research, we would include the following. First, any effects of
58. MankiwandWeil(1989)predictthatrealhousepriceswill fallby almost50 percent
over the next 20 yearsbecauseof demographicchanges.Whiletheirresultsmayoverstate
the comingdecline,even smallreductionsin house priceswouldtransferlargeamountsof
wealthto peoplewho arevery youngtoday.
demographyon the rate of technicalchange are likely to dwarfits other

consequences.It wouldbe valuableto refineourestimatesby considering
data spanning longer periods and by experimentingwith alternative
control variables. Second, how demographicchanges affect private
savingremainsuncertain.Investigatingthe internationalexperience on
this question seems worthwhile, particularlyif long-termdata can be
located. Third,our calculationshave assumedthat the nonmedicalcare
needs of the elderly are equal to those for the nonelderly.Whetherthis
assumptionis correct, and whether it will remain correct as the aged
populationages, needs to be investigated.Fourth, it would be useful to
analyze more systematicallythe impactof demographicchanges on the
welfare of differentcohorts. This would requirea life-cycle analysis of
the questionswe addresswith an infinitehorizonsetting.59
It would also be useful to explore the microeconomicimplicationsof
changingdemography.For example,ouraggregationof capitalmaywell
be inappropriateif demographicchange alters the relative demandsfor
housing and nonhousingcapital. Similarly,demographicchanges may
have importantimplicationsforthelabormarketpositionof agedworkers
andfor the relativedemandsfor workersin differentoccupations.
Furtherresearch on these and other related topics is likely to refine
the conclusionsaboutdemographicchangereachedhere. We doubtthat
it will alter our primaryconclusion that demographicchange provides
opportunitiesas well as challenges.
59. Auerbachand Kotlikoff(1987)and Auerbachand others (1989)use a life-cycle
modelto considerdemographicissues, but they assumecounterfactuallythat consumers
actuallyvary their savingrates as the model would predictand do not use the modelfor
normativeanalysis.
Comments
and Discussion
George A. Akerlof: The authorsdocumentthe dramaticdemographic

changesthat have been takingplace in the United States since 1940:the
steep increase in fertility in the 1940sand 1950sfollowed by the sharp
decline in fertility in the 1960s.1Although the ratio of elderly to the
workingpopulationis projected to decline over the next two decades
while the babyboomersremainactive in the work force, largeincreases
in this ratio are projected after 2010 when the baby boomers become
retirees. The authorsexplore whether these demographicchanges are
cause for concern, focusing especially on the adjustmentsthat should
occur in nationalsavingrates.
Theirpapershows, convincinglyin my opinion,thatdecreasedfertility
rates are not a reason for increased nationalsaving even though lower
fertilityeventually causes an increase in the ratio of the retiredto the
workingpopulation.In the simpleSolow modelof economicgrowth,the
change in an economy's capital-laborratio at a point in time is the
differencebetween saving per workerand "capital-widening"require-
ments per worker;the capital-wideningrequirementper worker is the
investmentper workerneeded to equip new entrantsto the work force
with the same capital as existing workers. A decline in fertility, other
thingsequal, lowers the economy's capital-wideningrequirementsand
increases the growth rate of the capital-laborratio. Intuitively, the
desirabilityof addingto society's stock of capital depends on the gap
between the marginalproductivityof capital and society's rate of time
preference.With the saving rate fixed, a decline in fertility promotes
"capitaldeepening"(it raises the growthrate of the capital-laborratio)
that, in turn, lowers the marginalproductivityof capitalrelative to the
1. Thesecommentswere preparedjointly withJanetYellen.
57
rate of time preference;as a result, the optimalsaving rate is reduced.

Optimalsaving is lower because, with decreasedfertilityrates, there is
less need for investmentto equipa growingworkforce.
This paper seems to fly in the face of conventional wisdom, which
argues that demographictrends are one reason why U.S. saving is
inadequate.Accordingto a commonargument,the increases in fertility
duringthe 1940sand 1950snecessitate an increase in saving to provide
for the baby boomers' retirement.In contrast, the authors argue that
currentdemographictrendsdo not necessitate highersaving.
Appearancesto the contrary,these two approachesare not mutually
inconsistent.The two pointsof view areanalyzingthe samephenomenon
from different time perspectives. The baby boom-baby bust demo-
graphiccycle, which the United States has experienced,consisted of an
increase in fertilitystartingin the 1940s,followed by a decline in fertility
beginningin 1960.The demographicchangesthatbeganin 1940optimally
called for an increase in the savingrate (to appropriatelyequipthe baby
boomerswith capitalduringtheirworkingyears)followed by a decrease
in the saving rate when the baby boom ended and declining fertility
lowered capital-wideningrequirements.
According to the conventional view, saving did not increase as it
should have when fertility increased, perhapsbecause of institutional
factors such as social security and retirementplans whose benefits are
directlylinkedto social security. Because the savingrate apparentlydid
not rise in response to the fertility increase, it should rise now. In
contrast, the authors, with a different startingpoint, ask how saving
should respond to recent and anticipateddemographicchanges. Their
answeris that, otherthingsbeingequal, savingrates shoulddecline now
because decliningfertilityratesjustifya declinein savings.If the insights
of these two views were combined,one mightconcludethatif the saving
rate hadincreasedin responseto the increasein fertilityin the 1940sand
1950s as it should have, demographicfactors beginning in 1960 and
projectedinto the next centurywouldoptimallycall for decliningsaving.
But because the saving rate did not increase when it should have, in
responseto earlierdemographictrends,it maybe unwiseto lower saving
rates now in response to currentand projectedfuturetrends. Properly
interpreted,I therefore agree with the authors' assertion that "other
thingsbeingequal, the optimalpolicy responseto recent andanticipated
demographicchanges is almost certainly a reduction rather than an
increase in the nationalsaving rate." But I also agree with the popular
wisdom that the as yet inadequateresponse to the baby boom, which
began in the 1940s, partly accounts for the inadequacy of the current
capitalstock.
The authorstell us: do not worryaboutthe futuredependencyof the
baby boomers. If you had saved a great deal in response to the baby
boomers(as the plannerwould have in theirmodel), now is the time for
an increase in consumption. However, most peoples' worry is the
opposite: they worry that we have not saved in response to the baby
boom and, therefore,that saving should increase now while some time
remainsto preparefor the dependencybulge.
An analogyis apt. A travelerleaves Chicagoboundfor CedarRapids.
He notes that his plane is travelingeast. He tells the stewardess about
his anxiety, sayingthat CedarRapidsis west of Chicagoandthe planeis
headingin the wrong direction.The stewardess is reassuring.She tells
him that if he had begun his trip in Salt Lake City, where, on her
assumptions,he should have startedit, then he should be headed east,
as the planeis currentlygoing.
The correct moral to draw from the authors' paper is subtle. They
have rightlyreassuredus thatthe consequences of ourprofligacyarenot
as dire as imagined, since the baby busters will requirerelatively little
investmentto satisfy capital-wideningrequirements.
The authorsalso examine the implicationsof populationagingin the
context of an explicit open-economy model. As they show, the trade
partnersof the United States will experience increases in dependency
that are even more pronouncedthan those in the United States. More-
over, the near-termdecline in dependency projected for the United
States will not occur abroad.Accordingto a naive view, these develop-
ments abroadare cause for concern. The United States is now highly
dependenton foreign saving to finance its current spending. If saving
abroadfalls as dependencythere increases, foreignlendingmay dry up,
resultingin risinginterestrates and a decliningdollar.The authorspoint
out the flaw in this naive scenario. The demographicchanges that are
occurringalter not only optimal consumptionbut also optimal invest-
ment.Overthe next severaldecades, investmentneeds will declinemore
quicklyin other countries than in the United States. In the absence of
internationalcapital flows, the returnto investment should fall abroad
andriseinthe UnitedStates. Inresponseto this differential,international
capitalflowsintothe UnitedStates shouldincrease,not decrease. Taking

into accountthe possibilityof internationalborrowingandthe likelihood
thatothercountrieswillwishto exportmore,notless, capitalstrengthens
the authors'conclusion that, other things being equal, saving need not
rise now.
The authorsadvancetwo reasonsfortheirview of the relationbetween
the growth and age of the labor force and productivitygrowth. First,
workersof differentages may contributedifferentiallyto technological
innovation. This is an age-embodiedmodel of technological change.
Second, diminishedlaborforce growthmay resultin laborforce scarcity
thatraises wages andresultsin labor-savingtechnologicalprogress.The
authors produce evidence that rapid labor force growth is negatively
associated with rapid productivityincrease and positively associated
with the averageage of the laborforce in cross-sectionregressions.How
much supportthis gives to theirmodel of technologicalprogressis hard
to say. The authors' regressions omit wages. Furthermore,a cause
common to both may yield a relationbetween productivitygrowth and
labor force growth and population age but in no way imply a causal
relation between the demographicfactors and technological change.
FollowingMaxWeber,to give one example,Protestantismis responsible
not only for the capitalist ethic and rapid innovationbut also for low
fertility. Thus productivitychange and demographicchange may have
the same root cause without any implicationthat one causes the other.
The authors'fixed-effectsregressioncontrols for intercountrycultural
differencesandpotentiallyanswersthe objectionthatpersistentcultural
differencesacrosscountriesaccountfor the relationbetween productiv-
ity growthandfertility.The slope of this regressionshows thatas OECD
countries (not includingthe United States) switched from high to low
fertility,the rateof productivitygrowthrose. Butit is notfullyconvincing
thatthe relationshipis causalratherthancoincidental.Third,in a cleaner
test, outputper unit input, the Solow residual, would be chosen rather
thanproductivityas the dependentvariable.
In the final section of the paper the authors address whether taxes
should be raised now in anticipationof higherburdensof government
expenditureas the populationages. The paper compares the relative
merits of a tax-smoothingstrategy, in which taxes would be increased
now, before dependencyratios increase, and of a pay-as-you-gopolicy
of raisingtaxes later. As table9 shows, the requiredincreasein tax rates
undera pay-as-you-gopolicy will be substantial-amountingto approx-

imately 4.5 percent of GNP. Thus, in the absence of tax smoothing,
futuregenerationswill have to pay a significantlyhigherportionof their
income for social security and medicarebenefits. An advantageof tax
smoothingover pay-as-you-go is the lower deadweightburdens from
taxation.Accordingto Barro,an optimaltax policy would smoothtaxes
to minimize deadweight loss. Although the authors agree that tax
smoothingis more efficient than a pay-as-you-gopolicy, they reassure
us that little will be lost if the governmentwaits to raise taxes until the
dependency ratio actually increases. Their calculations show that the
efficiency loss from pay-as-you-go is small even with a 30 percent
deadweightloss associated with the marginaldollarof revenue.
AlthoughI agree with the analysis, as far as it goes, I find the tax-
smoothingissue one of misplacedemphasis.Thereis moreat stake than
aggregateefficiencyconsiderationsin the tax now-tax laterdebate. Tax
policy mattersboth to income distributionbetween generationsandalso
to aggregatecapitalaccumulation.In theirsimulations,with tax smooth-
ing, early generations would be roughly 2?/2percent poorer and later
generationswouldbe 21/2percentricher.A pay-as-you-gopolicy benefits
the currentgeneration,but inflictssignificant,avoidableburdenson our
children.This assumes, of course, that Ricardianequivalencedoes not
hold, for under Ricardianequivalence, the timing of tax collections is
irrelevantto intergenerationalincome distribution,and thus efficiency
is the sole appropriateconcern of policy. In a Ricardianworld, if taxes
are imposed later rather than sooner, we will simply bequeath our
childrencorrespondinglyadditionalwealth.
As the authors implicitly recognize, in the absence of Ricardian
equivalence,the tax policy that is selected to financea given programof
government expenditures to the aged affects the pattern of capital
accumulation.A tax-smoothingpolicy would lower consumptionand
promotegreatercapitalaccumulationnow in advance of the increase in
the dependencyratio; in contrast, a pay-as-you-gopolicy would result
in lower capitalformationnow and later.The authorsdeal with capital
accumulationand taxes separately, whereas, in fact, these two issues
are not separable.
The authorsalso arguethat optimalcapitalaccumulationand income
distributionamong individualsliving at the same time are separable
issues. The Ramseymodel, which is the basis for the recommendations
concerningcapitalaccumulation,is primarilyconcernedwiththe optimal

intertemporaldistributionof consumption among individualsalive at
different times. The solution depends on the chosen social rate of
discount, p, reflectingconcernfor those alive now relativeto those who
will be alive later. Whenconsideringtax and savingspolicy, the authors
disregardthe distributionof consumptionamong membersof different
generationsalive at a given date, because they argue that these issues
are separable. According to the authors, an optimal distribution of
consumptionbetween membersof differentgenerationswho are alive
at a given momentshouldbe handledby age-specifictransferprograms.
However, practically,this entails eitherexplicit or implicitadjustments
to social securitybenefits. It is an odd exercise to projectcurrentlevels
of social security and other age-specificgovernmentexpendituresand
to advocate a pay-as-you-go system of taxation to fund these benefits
while, at the same time, suggesting that if the income distributional
consequencesof these taxes areconsideredundesirable"the appropriate
response shouldbe an adjustmentin the level of prospectiveintergener-
ational transfers. . . ." Such an adjustment, in effect, would amount to
offeringretireeslower social securitypaymentsto avoidthe highertaxes
on workersthat would be levied underthe pay-as-you-gopolicy.
I would agree with James Tobin that "the overridinglong-runissue
about OASI is the balance between the tax contributionsof the young
and the benefits of the old."2 The concern here lies not just with
intergenerationalincome distribution,but also with the perceived fair-
ness of the social securitysystem. If the system is seen as unfair,younger
generations may renege on the promises implicitly made to the older
generations. And, perhaps more important,government will lose its
moralauthorityto redistributenationalincome.
In conclusion, let me say that this is a very interesting paper. In
particular,it reinforces the wisdom of development economists and
family plannersthat fertility declines are almost always economically
beneficial.In the case of the United States, althoughthe fertilitydecline
has posed unexpected problemsfor our pay-as-you-go social security
system, still, it has resultedinless, ratherthana greater,needfor national
saving.
2. Tobin(1988,p. 42).
Martin Neil Baily: As I was readingthis paper I was remindedof a

phone call I received a few years ago from someone at the Readers'
Digest. She was a fact checker and she was workingon an article by a
futurologist.He had claimedthat GNP would be twice as large 30 years
fromnow as it is today. WouldI please say if this was correct. I took out
my calculatorand figuredout that this representeda 2.5 percent a year
growthrate, so I said it seemed fine to me. When I got off the phone, I
reflectedthat forecastinglong-termgrowth rates was a risky, perhaps
even a foolish, business. The whole idea of fact-checkingpredictions
seemeda littlewacky. But as this papermakesclear, long-termforecasts
aredirectlyrelevantto personalandpolicy decisions today. The optimal
savingrate dependsheavily on the expected rate of productivitygrowth
over the long term. Such forecastingis necessary.
This paperis a tour de force in many ways. It provides a systematic
attemptto applythe principlesof optimalgrowthto a problemof empirical
importance.It is comprehensivein takinginto accountnot only changes
in the work force but also the relationbetween the work force and the
population,the differentsocial demandsof the young and the old, and
the consequencesof internationalcapitalflows. My commentswill cover
some points of detail and then look at a morebasic concern.
First, in assessing the dependency ratio, the authorspoint out cor-
rectly that the rise in the number of old people will be offset by the
decline in the numberof young people. They take some account of the
fact that old people use more resources than young, but perhaps not
enough. The effect of many of the advances in medicine has been to
allow old people to live longer. As Summersand Poterba know from
theirpriorwork, the biggestgainsin reductionsof death'rateshave been
for those over 80. If these trends in medical advance and longevity
continue, they will have an impact on the effective dependency ratio.
The study of long-termcare by Alice Rivlin and Joshua Wiener has
indicatedthat a tremendousincrease in medicaland nursinghome care
will probablybe needed.I
The authorsalso take account of the increase in the time people are
spendingin retirement,but there too I wonder if the effect could have
been understated.The resource cost of the growingcohort of elderly is
1. RivlinandWiener(1988).
likely to become very largeindeedin the next century.My intuitionsays

that we shouldpreparenow for the social problemsthat this will create,
and the authors'paperhas not overcome this intuition.
Second, the argumentis made that the gains from tax smoothingare
not very great.This does not seem to fitwithobservedpoliticalbehavior.
Substantialpolitical battles have been waged over tax rate changes
roughlythe same as the changesgiven here. The politicalconsequences
of a shift from taxes being 32 percentof income to 37 percentof income
are likely to be substantial, particularlyif the increase translates in
practiceinto a 10percentagepoint increasein the payrolltax. If overall
deadweightloss were the only criterionby which to judge tax changes,
then presumablywe shouldconsiderthe poll tax an exemplarytax. The
marchersin TrafalgarSquarewho were throwingchairs, however, did
not seem overly impressedby the small deadweightlosses implied by
MargaretThatcher's poll tax.
Of course, much of the politicaldebate over taxes turns on distribu-
tionalissues, but supportingthe elderlywill be a distributionalissue too,
with the burdenfallingon workersandthe benefitsfallingon retirees. If
the Moynihanview that there is no need to accumulatea surplusin the
social security trustfund prevails, today's debate on that issue is likely
to reverberatein Congress30 years fromnow: "Why shouldwe impose
these huge taxes on our workers today when the currentbeneficiaries
refused to pay their share when they were working?" Presumably,
cuttingbenefitsto the elderlywill be the other alternative.
Third, the open-economy simulationsassume too much capital mo-
bility. One can makea prettygood argumentthat there is perfectcapital
mobilityfor short-termgovernmentsecurities. But therecertainlyis not
perfectmobilityfor factoriesor business capitalgenerally.Flows of real
capital, like real trade flows, are very sticky. It is certainlypossible to
financeour budget deficit with capital inflows, as we have discovered.
And those inflows may also inflateproperty marketsin Honolulu and
Manhattan,but they do not do much for business capitalformationor
productivity.The portfoliochoice thatan economy makesis affectedby
the preferencesof the owners of the capitaland hence where the saving
is coming from.
Fourth, there are better ways of capturingthe impact of changing
demographicson productivitythan the regressionapproachused here.
Estimates of age-earningsprofiles used by Dale Jorgenson, Edward
Denison, and others in their accounting studies of past growth to

constructadjustedlaborinputs could be used to simulatethe effects on
productivityof the projected demographicchanges.2It would also be
better to start with the standardhypothesis that relative wages reflect
relativemarginalproductsand that the projectedagingwill raise future
productivityas more workersmove into theirhigh-productivityyears.
The attemptin the paper to estimate that demographicchanges will
have a big bang for the buck is not convincing. The hypothesis is not
tested by startingwith the standardalternatehypothesis, based on age-
earnings profiles, and then seeing if the data suggest a significantly
greaterimpact. Anecdotal evidence abounds that older workers find it
harderto adjustto changes in technology.
Fifth, the authorsdo not commenton one demographicoffset to the
projectedbenefitsof aging.JohnBishop has investigatedthe hypothesis
that general intellectualachievement is importantto productivityand
that boosting general intellectualachievement is what schooling does,
at least up to the end of high school.3Variousmeasuresof that achieve-
menthave showndeclinesbeginningin the late 1960s.In an earlierpaper,
I arguedthatthis decline in test scores could not have had a majoreffect
on productivityin the 1970sbecause of the small size of the cohort that
had entered the labor force.4 Bishop suggests that the effects are
becominglargerin the 1990s-the low-score rabbitis now moving into
the middleof the laborforce snake.
I turnnow to the moregeneralconcern. I findit hardto put the paper's
findingsin context. The authors are looking for the partial effect of
demographicchanges on the optimal saving rate. That is obviously an
interestingissue to look at, but it is hard to read the analysis without
worryingthatthe U.S. economy is probablya long way froman optimal
path of accumulationfor other reasons, notably because of the growth
slowdown,governmentdissaving, and life-cycle savingrequirements.
My readingof the optimal growth literatureof the 1960s is that it
implied that the U.S. economy was saving too little. The optimality
conditionis that the rate of profitshouldequal the rate of growthplus a
discountfactor. Plausiblerates of discount left the profitrate too high
andhence the capitalintensityof the economy too low. Taxes and risks
2. Denison (1985); Jorgenson, Gollop, and Fraumeni (1987).
3. Bishop (1989).
4. Baily (1981).
are reasons why private decisions can lead to nonoptimalsocial out-

comes.
These findingsfromthe 1960swere beingappliedto an economy with
relativelyrapidgrowth.The productivitygrowthslowdown shouldhave
led to a substantialincrease in the optimallevel of saving, unless one
believes that the slowdown is just temporary.The living standardsof
futuregenerationsare going to be muchlower thanwe thought.If saving
was already too low and has subsequently fallen sharply, and if the
optimalrate of saving has risen because of the productivityslowdown,
then we now have a huge gap between the actual rate of saving and the
optimal rate. Even if the authors are correct that the partialeffect of
demographicchange is to lower the optimalsaving rate, their findingis
cold comfort.
Turningspecificallyto the demographicissue, I did not understanda
key element in the modeling.Once we take into account the humanlife
cycle, the objectivefunctionof the optimalgrowthplannerchanges. The
plannershouldnot maximizethe integralof discountedper capitautility
times the numberof people. The integralof the lifetime utility of each
memberof the populationtimes the numberof people would be a better
concept. And I thinkthataffectsthe conclusions. For example,compare
two economies with different demographic compositions. The two
economies mighthave differentrates of growth of outputand different
intertemporalconsumption patterns, but the lifetime utility of any
individualcould be the same in the two economies. And presumablythe
reverse situationcould hold, where the aggregateslook the same in two
economies, but individualshave ratherdifferentlifetimeutilities.
This issue of lifetime utility affects the concluding section of the
paper, where the authorsask how theirfindingscan be reconciled with
the idea that the social security trust fund should build up a surplusto
pay for the retirementof the baby boom generation.The place to start
when deciding whether social security decisions are correct is to ask
whatrationalfar-sightedpeoplewoulddo if therewere no social security.
The life-cycle model says thatindividualsin theirhigh-incomeyears will
save for retirement. That will surely translate into higher aggregate
saving in periods like the current one when a large fraction of the
populationis in its high-incomeyears.
The pureretirementelement of the social security system is presum-
ably based on the assumption that many people either do not make
rationalfar-sighteddecisions or lack self-discipline,and so the program
forces them to save for their retirement. Then to mimic the rational
privatedecision, the trustfund shouldnow buildup a surplus.Whydoes
this not show up in the authors'results?
Withouttinkeringat length with the optimalgrowthconditions, I am
not sure, but I thinkit is relatedto one of the thingsthatCharlesSchultze
and I pointed to in a recent paper.5The steady-state rate of growth of
the U.S. economy has declined, and so the rate of profit is declining.
The projecteddecline in the growth of the labor force will drive down
the rate of profitfurther.The decline in the warrantedrate of growth is
drivingthe authors'conclusionthat the optimalsavingrate will decline.
This result is very much a steady-stateconclusion, however, and there
is no guaranteethat the same result should hold even over periods of
severaldecades.
It is not clear from theory that a decline in the rate of profit should
reduce savingby rationalindividuals,and it is not clear from empirical
evidence that actual people do reduce saving when the rate of return
declines. I would expect the life-cycle effect to dominateover the rate
of profiteffect so that the aggregatesavingrate of rationalsavers would
increase as the baby boomers hit their peak earningsyears. If so, the
social securitytrustfund shouldfollow theirlead.
Of course social security is not only a forced retirementprogram;it
is also redistributive.And therethe authorshave a good point. The baby
boomers have had it tough competing against their own large cohort.
They are entitledto expect latergenerationsto contributeto the support
of the low-incomeelderly in the next century. The build-upof the trust
fund shouldnot try to do it all.
This is a provocative and importantpaper that questions a conven-
tional wisdom about the appropriateway to respond to demographic
changes. The model has not settled some importantissues, but it is not
wrongeither. It raises legitimatequestions about the level of saving we
shouldexpect or seek.
General Discussion
The paper's implicationsfor national saving policy sparkeda lively

discussion. WilliamNordhauspraisedthe paperfor raisingthe quality
of the analysis about the effects of demographictrends on the national
5. BailyandSchultze(1990).
saving rate. A number of other panelists expressed concern that the

paperwould be misunderstoodand misusedin the currentpolicy debate
over nationalsaving. GaryBurtlessemphasizedthat if the savingrate is
too low to begin with, slower labor force growth should not be an
argumentfor decreasing it further. Lawrence Summersagreed that it
would be desirable to raise the national saving rate for a variety of
reasons, but said the point of the paper is that demographicchange is
not one of them. CharlesSchultze suggested simulatingdesired future
saving using the assumptionthat the initial stock of capital is less than
optimal.Althoughestimatesof the marginaleffects of populationmight
be quite similarto those in the paper, such simulationswould avoid the
impression that policy should aim to reduce saving from current,
suboptimallevels. RobertGordondissented from the widely held view
that current national saving is too low. He noted that net private
investmentis close to the level necessary to maintainthe capital-output
ratio. The projecteddecline in the populationgrowth rate and increase
in privatesavingdue to the changein the age distributionshouldmitigate
any concerns that investment is being financed by a current account
deficit.He also pointedout thatthe changesin percapitaincomeresulting
from differentcapital-outputratios are small comparedwith the uncer-
tainty about the future growth in productivity. The real problem for
policy, he contended, is the lack of public investment and the need to
redistributethe peace dividendin that direction.
Various panelists questioned whether the results were robust with
respect to certainchanges in assumptions. While noting that the theo-
reticalpossibilitythat decliningfertilitymightincrease welfare was not
new, Henry Aaron wondered whether the propositionwould be valid
for one countryin a worldeconomy. In particular,he did not regardthe
OECD as an adequaterepresentativefor the "rest of the world," and
arguedthatthe inclusionof non-OECDcountriesmightwell reverse the
authors' presumptionthat the net supply of savings from abroad will
increase.AlbertWojniloweragreedthatnon-OECDcountriesshouldbe
considered.LawrenceSummersnoted that in lightof the low returnson
investmentin the southernhemisphere,both historicaland anticipated,
that regionwas unlikelyto be a user of saving.
Burtless doubted that the social-welfarefunction used in the paper
capturedthe unpleasanttrade-offbetween the consumptionof different
generations. He reasoned that, if our grandchildrenwere expected to
enjoy real income twice as high as our own, it might be difficult to

persuadeus to make an additionalconsumptionsacrificetoday so that
theirincomescouldbe 2.1 times higherthanourown. Onthe otherhand,
if we thoughtthatourgrandchildrenwouldbe barelybetteroff-or even
worse off-than we are, then many of us might be willing to make a
sacrificeof currentconsumptionin orderto raise theirliving standards.
Burtlesssuggestedthata majorargumentfor highersavingtoday is that
we have become less optimistic about the income prospects of our
grandchildren,primarilybecause of a sharpdropin productivitygrowth.
The demographicshock of a higherdependencyratio, which is empha-
sized in the paper, is not the main reason for pessimism about future
living standards.Hence, even if demographicconcerns by themselves
do not push us to raise currentsaving rates, other concerns push us in
thatdirection.EdmundPhelpswas curioushow the optimalconsumption
time path would vary if it was calculated with successively lower
intertemporalsubstitutionelasticities. He speculatedthat, in the limit,
currentconsumptionmightgo down, not up. ThomasJusterwondered
how the conclusions would be changedif the authorshad used a utility
functionthatwas consistentwithpeople's apparentpreferencefor rising
consumptionover steady or declining consumption, even when faced
with low real rates of return. Such preferences, if true, suggest that a
pathof firstrisingand then fallingconsumptionis not optimal.
Several panelists criticized the paper's treatment of technological
change. Nordhausfound unconvincingthe cross-countrycomparisons
suggesting that lower population growth causes higher productivity
growth,because, he said, so many otherfactors can be at work. Phelps
doubtedthatincreasedlaborscarcitywouldleadto greatertechnological
progress,and wonderedwhy decreasedcapitalscarcitywould not have
the opposite effect. Summersdefended the use of cross-countrycom-
parisons, remarkingthat it would be hard to think of another natural
experimentwith which to searchfor these effects. Furthermore,to the
extent that movements in some unobservedfactor are responsible for
the apparentcorrelationof productivityand fertility, there is no reason
to thinkthe samefactoris not at workin the United States. At a minimum
he felt that the cross-country comparisons defused any fear that a
slowdownin dynamismwould accompanya reductionin the population
growth rate. Franco Modigliani,while findingit plausible that greater
age is associated with higher levels of productivity, saw no reason it
should lead to faster growth in productivity.Joseph Stiglitz observed

that returns-to-scaleeffects in learning-by-doingand endogenousR&D
models could give the opposite effect, with lower labor force growth
adverselyaffectingproductivity.
Nordhaussuggestedit wouldbe moreaccurateto weightthe nonmed-
ical consumptionof the elderly at about 80 percent of that of working
people. He also conjectured that if subsequent generations consume
fewer marketgoods but moreleisure, then it may not be optimalto save
less now. Summersnoted thatthe procedurein the paperaccommodates
the possibilityof increasedleisure underthe assumptionthat it requires
no capital and is separablefrom ordinaryconsumption in the utility
function. Nordhausalso thoughtit useful to distinguishthree sources of
slowing labor input: declines in fertility, reduced fraction of lifetime
spent working,and the rise in the disabilityrate. Each of these sources
would have a differenteffect on optimalsavingbehavior.
Several panelists discussed ways in which private sector saving and
investmentare likely to be affectedby changeddemographics.Modigli-
ani noted that in the typical life-cycle model, when an economy grows
less fast, it saves less. He conjecturedthat the demographicchanges
would automaticallyreduceprivatesaving, perhapseven more thanthe
authorsbelieved optimal.Gordonnotedthatalthoughslowerpopulation
growthis likely to lower futurehousingprices, most of the effect of that
reductionwould be in smallerbequests to the young ratherthan lower
consumptionby the elderly.Phelpsobservedthatthe predictedreduction
in investmentcouldresultinanemploymentproblembecausethe capital-
goods sector is morelaborintensivethanthe consumption-goodssector.
Wojnilowerobserved that because the elderly will have more political
power because of their largershare of the population,perhapsgovern-
mentpolicy itself will be directlyinfluencedby the changein demograph-
ics. Gordon suggested two policy actions that offered solutions to the
dependency problem. The first was raisingthe immigration-to-popula-
tion ratio to the 1913level. The second was allowingthe retirementage
to increasewith life expectancy.
References
Aaron, Henry J., BarryP. Bosworth, and Gary Burtless. 1989. Can America
Afford to Grow Old?Paying for Social Security. Washington:Brookings.
Auerbach,Alan J., and Laurence J. Kotlikoff. 1989. "Demographics,Fiscal
Policy, and U.S. Saving in the 1980s and Beyond." WorkingPaper 3150.
Cambridge,Mass.: National Bureauof Economic Research (October).
. 1987.DynamicFiscal Policy. New York: CambridgeUniversityPress.
Auerbach,Alan J., and others. 1989. "The Economic Dynamics of an Aging
Population:The Case of Four OECD Countries. OECDEconomic Studies
12: 97-130.
Baily, MartinN. 1981. "Productivityand the Services of Capitaland Labor."
BPEA, 1:1981, 1-50.
Baily, Martin Neil, and Charles L. Schultze. 1990. "The Productivity of
Capital in a Period of Slower Growth." BPEA: Microeconomics, 1990,
369-406.
Ballard, Charles L., John B. Shoven, and John Whalley. 1985. "General
EquilibriumComputationsof the MarginalWelfare Costs of Taxes in the
United States." AmericanEconomic Review 75: 128-38.
Bishop,John. 1989."Is the Test Score Decline Responsiblefor the Productivity
GrowthDecline?" AmericanEconomic Review 79: 178-97.
Blanchard,OlivierJ., andStanleyFischer. 1989.Lectureson Macroeconomics.
Cambridge,Mass.: MIT Press.
Barro, Robert J. 1979. "On the Determinationof the Public Debt." Journal
of Political Economy 87: 940-71.
Board of Trustees of the Federal Old-Age and Survivors Insurance and
DisabilityInsuranceTrust Funds. 1988. 1988Annual Report. Washington:
Social Security Administration.
. 1990.1990AnnualReport.Washington:Social SecurityAdministration.
Board of Trustees of the Federal SupplementaryMedical Insurance Trust
Fund. 1988.1988AnnualReport. Washington:Social SecurityAdministra-
tion.
Borjas, George J. 1990. Friends or Strangers:The Impact of Immigrationon
the U.S. Economy. New York: Basic Books.
Calvo, GuillermoA., and MauriceObstfeld. 1988. "OptimalTime-Consistent
FiscalPolicy with Finite Lifetimes:Analysis and Extensions." In Economic
Effects of the GovernmentBudget, edited by Elhanan Helpman, Assaf
Razin, and EfraimSadka. Cambridge,Mass.: MIT Press.
DeLong, J. Bradford.1988."ProductivityGrowth,Convergence,and Welfare:
Comment."AmericanEconomic Review 78: 1138-54.
Denison, EdwardF. 1985. Trendsin American Economic Growth, 1929-82.
Washington:Brookings.
Division of National Cost Estimates, Office of the Actuary, Health Care

Financing Administration. 1987. "National Health Expenditures, 1986-
2000." Health Care Financing Review 8(Summer): 1-36.
Freeman, Richard B. 1988. "Labour Market Institutions and Economic
Performance." Economic Policy 6: 63-80.
Habakkuk, H. J. 1962. American and British Technology in the Nineteenth
Century.Cambridge,England:CambridgeUniversity Press.
Hatsopoulos,GeorgeN., PaulR. Krugman,and LawrenceH. Summers.1988.
"U.S. Competitiveness:Beyond the TradeDeficit." Science, July 15: 299-
307.
Jorgenson,Dale, FrankM. Gollop, and BarbaraM. Fraumeni.1987.Produc-
tivity and U.S. Economic Growth. Cambridge, Mass.: Harvard University
Press.
Lazear, Edward P., and Robert T. Michael. 1980. "Family Size and the
Distributionof Real Per CapitaIncome." AmericanEconomic Review 70:
91-107.
Mankiw, N. Gregory, and David Weil. 1989. "The Baby Boom, the Baby
Bust, and the Housing Market." Regional Science and Urban Economics,
235-58.
Mankiw, N. Gregory,David Romer, and David Weil. 1990. "A Contribution
to the Empiricsof Economic Growth" (February).
Organizationfor Economic Cooperationand Development. 1988.Aging Pop-
ulations: The Social Policy Implications. Paris: OECD.
Poterba, James M., and Lawrence H. Summers. 1987. "Public Policy Impli-
cations of Declining Old-Age Mortality." In Work,Health, and Income
among the Elderly, edited by Gary Burtless. Washington:Brookings.
Public Health Service. 1989. Health United States 1988. Hyattsville, Md.:
Departmentof Health and HumanServices (March).
Rivlin, Alice M., and Joshua M. Wiener. 1988. Caring for the Disabled Elderly:
Who WillPay? Washington:Brookings.
Romer, Paul M. 1990. "Capital, Labor, and Productivity."BPEA: Microec-
onomics 1990, 337-67.
Schneider, Edward L., and Jack Guralnik. 1990. "The Aging of America:
Impact on Health Care Costs. " Journal of the American Medical Association
263 (May 2): 2335-40.
Simon,JulianL. 1981.The UltimateResource. Princeton:PrincetonUniversity
Press.
Social Security Administration. 1987. Social Security Bulletin, Annual Statis-
tical Supplement. Washington (December).
. 1989. Social Security Area Population Projections: 1989. Actuarial
Study 105. SSA Pub. No. 11-11552.Washington(June).
Summers, Lawrence H., and Chris Carroll. 1987. "Why Is U.S. Saving So
Low?" BPEA, 2:1987, 607-35.
Summers,Robert, and Alan Heston. 1990. "The Penn WorldTable (Mark5):
An Expanded Set of InternationalComparisons." QuarterlyJournal of

Economics (forthcoming).
Tobin, James. 1988. "The Future of Social Security: One Economist's
Assessment." In Social Security: Beyond the Rhetoric of Crisis, edited by
TheodoreR. MarmorandJerryL. Mashaw.Princeton:PrincetonUniversity
Press.
Waldo, Daniel R., and others. 1989. "Health Expendituresby Age Group,
1977 and 1987." Health Care Financing Review 10 (Summer): 111-20.
Wattenberg,Ben J. 1987. The BirthDearth. New York: Pharos Books.
Weil, David. 1989. "Age and Saving in Micro and Macro Data." Harvard
University.
1990. "Prospects for Saving in the OECD over the Next 35 Years."
HarvardUniversity.

1990a Bpea Cutler Poterba Sheiner Summers Akerlof

Uploaded by

Copyright:

Available Formats

1990a Bpea Cutler Poterba Sheiner Summers Akerlof

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1990a Bpea Cutler Poterba Sheiner Summers Akerlof

Uploaded by

Copyright:

Available Formats

DAVID M.

An Aging Society: Opportunity

AN AMERICANwomanreachingchildbearingage in 1960wouldexpect 3.6

growingelderlypopulationwill place on the economy in generalandthe

fully the consequencesof increaseddependence.This finding,however,

The Burden of Increased Dependency

Changing Demographic Structure

Figure 1. Actualand ProjectedDependencyRatios, UnitedStates, 1960-2065a

the number of children plus elderly as a fraction of the working-age

1. The relativeimportanceof fertilitydeclines, mortalityimprovement,and interna-

Figure2. Actualand AlternativeProjectedTotal FertilityRates, 1920-2080

Administration'soptimisticandpessimisticprojections.Even the factor

illegalmigration.The age structureof the populationis sensitive to the

The Support Ratio

relative consumptionneeds of people of differentages, as well as by

where Ni is the numberof people of age i. This measure of needs is

these three components, we construct a needs-weightedconsumption

Oursecond measure,LF2, recognizesthatboth humancapitalandlabor

alongwith Social SecurityAdministrationforecasts of age-specificlabor

Thisrecognizesthatthe earningcapacityof a society witha highfraction

Support Ratio Projections, 1990-2060

duringthe next 70 years, there is a long-rundecline in the supportratio.

In yet anothermetric, a three- to four-yearincrease in the average age

13. Measuresthat define effective consumptionwith less weight on childrenshow

Figure 3. Actual and Projected Changes in Support Ratio (Relative to 1990),

= P/(C + P + E), where P is the numberof prime-agedadults, C the

(6) = (P - C) [C(C + P + E)] + (P -E) [EI(C + P + E)].

Figure 4. Contributions of Young and Elderly Dependents to Percentage Changes

4% ~~" ..-------a------- ---

Figure 5. Percentage Change in Support Ratio (Relative to 1990), Alternative

standardcase. The decline in the supportratio is almost twice as large

Capital Accumulationand Shifting Dependency Burdens

This section explores how the demographic shifts described above

Steady-State Consumption Opportunities

This expression can be rewrittento find the change in steady-state

Table 3. Shifting Steady-State Per Capita Consumption from Demographic

a given level of per capitaoutput.This "Solow effect" offsets the long-

as the changes in the supportratio shown in table 2; the other columns

Demographic Change and Optimal Capital Accumulation

We assume that a social plannerseeks to maximize

where Pt denotes the numberof individualsalive in period t, ct is per

(10) kt =f(kt) - ctIot - ntIkt.

If at = 1, equation 10 reduces to the standardresource constraint in

In steady state with no technical progress, per capita consumption

This locus, a verticalline in (c, k)space, is drawnin figure6. Constancy

unambiguouslyreduce the optimal steady-state saving rate while in-

87.24Finally,we use datafor this periodon paymentsto laborandcapital

Table 4. Optimal Consumption Response to Demographic Shocks,

Perfectforesight with alternativeelasticitiesb

Case 2. Earnings-weightedlaborforce and needs-weightedconsumption

optimalresponse to the demographictransitionthat the United States

tionif consumershave no foresightaboutdemographicchangebutrather

Figure 7. Optimal Consumption and Capital-Labor Trajectory Following Demographic

104 200 2005

100 - Initial steady state 2020

Consumptionrises less with the alternativeI assumptionsthanwith our

Figure 8. Optimal Saving Rates in Response to Demographic Shock, 1990-2070

possibilitiesare higherin 1990thanin any of the threepreviousdecades.

Figure 9. Optimal Saving Responses to Demographic Change: Sensitivity to Date