Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

Private Saving and Terms of Trade Shocks: Evidence from Developing Countries

Author(s): Jonathan D. Ostry and Carmen M. Reinhart


Source: Staff Papers (International Monetary Fund), Vol. 39, No. 3 (Sep., 1992), pp. 495-517
Published by: Palgrave Macmillan Journals on behalf of the International Monetary Fund
Stable URL: http://www.jstor.org/stable/3867471
Accessed: 24-11-2015 07:29 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

International Monetary Fund and Palgrave Macmillan Journals are collaborating with JSTOR to digitize, preserve and
extend access to Staff Papers (International Monetary Fund).

http://www.jstor.org

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
IMF StaffPapers
Vol. 39,No. 3 (September
1992)
C 1992International Fund
Monetary

PrivateSavingand Terms
of Trade Shocks
EvidencefromDevelopingCountries

JONATHAN D. OSTRY andCARMEN M. REINHART*

Therelationshipbetween temporary terms oftradeshocksandhousehold


savingin developing countriesis examined. shownthat,froma
It isfirst
theoretical
standpoint, this is
relationship ambiguous: privatesavingmay
riseorfallinresponse toa transitory
terms oftradeshock,depending on
thevaluesof theintertemporal elasticityof substitutionand theintra-
temporal ofsubstitution
elasticity between tradedand nontraded goods.
Empirical estimatesof thesetwoparameters areobtained usingdatafrom
a sampleof13 developing and thenusedtodrawimplications
countries,
for the to
responseofprivatesaving transitory termsof tradeshocks.
[JELE21, F32, F41, 010, 053,054,055]

T HETERMS beenoneofthemostimportant
OFTRADEhavehistorically
exogenousdeterminantsof the externalpositionsof developing
Overthepasttwodecades,sharpfluctuations
countries. inworldmarket
pricesforprimary andtwooil shocks,whichsubstantially
commodities
increasedthepriceofimported
energyproductsfornon-oildeveloping

*Jonathan D. Ostry,an Economistin the ResearchDepartment, holdsa


doctoratefromtheUniversity ofChicago,as wellas degreesfromtheLondon
Schoolof Economicsand PoliticalScience,OxfordUniversity, and Queen's
University.
CarmenM. Reinhart intheResearchDepartment.
is an Economist Sheholds
a Ph.D. fromColumbiaUniversity.
The authorswishto thankMikeGavin,MohsinS. Khan,Leo Leiderman,
EnriqueMendoza,PeterMontiel,AssafRazin,and PeterWickham foruseful
comments.
495

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
496 D. OSTRYandCARMENM. REINHART
JONATHAN

countries, wereassociated withincreased variabilityinthesaving, invest-


ment,and current accountbehaviorofthesecountries.
The theoretical literature on therelationship betweenthetermsof
tradeand the current accounthas focusedalmostexclusively on how
termsof tradechangesaffectprivatesaving,ignoring anyadditional
effectson investment andpublicsaving.'The traditional explanation-
associatedwiththenamesofHarberger (1950)andLaursenandMetzler
(1950)-suggeststhatan improvement in the termsof traderaisesa
country's real income level, measured as thepurchasing powerof its
exports inworld and
markets, hence, on the assumption themarginal
that
propensity to consume is lessthan unity, raisesprivatesaving.Thus,the
Harberger-Laursen-Metzler (HLM) effect, it has becomeknown,
as
that
hypothesized improvements in a country's termsoftradewouldbe
associated withincreases inprivate and
saving, conversely, adverseterms
oftradeshockswouldreducesaving.
Thisviewwentlargely unchallenged fornearlythreedecades,andwas
generally supported bytheavailableempirical evidence.(See, forexam-
ple,KhanandKnight (1983).)Intheearly1980s,however, severalstudies
re-examined thetheoretical underpinnings oftheHLM effect, a crucial
building blockofwhichwastheKeynesian (static)relationshipbetween
consumption (orsaving)andincome.Thesestudies, including,forexam-
ple,thosebySachs(1981,1982)andSvenssonandRazin(1983),argued
thathouseholdsavingdecisionsshouldbe derivedfromsolutionsto a
dynamic optimization problemofchoosing consumption levelsat differ-
entpointsin time.As faras theHLM effect was concerned, thekey
insightprovided by these models was that the relationshipbetween the
termsoftradeand savingdependedcrucially on theexpectedduration
of the termsof tradeshock.For example,ifhouseholdsexpectedan
improvement in thetermsof tradeto be permanent, thentheywould
reviseupwardtheirestimate ofpermanent incomeinproportion to the
increased purchasing power oftheir income today. Under the hypothesis
thatthemarginal propensity toconsume(save)outofpermanent income
isunity(zero),a permanent change in the terms oftrade would therefore
haveno effect on saving,contrary to theHLM view.2By contrast, in a
situationin whichtheimprovement in thetermsoftradewasexpected
tobe onlytemporary, theincreaseinpermanent incomewouldbe smaller
thantheincreasein current and
income, savingwouldaccordingly rise.
1Fora discussion
ofinvestment
effectsofterms oftradechangesina somewhat
different
context,see Corden(1988).
2
Theviewthatpermanent termsoftradeshockshaveno effect onthecurrent
accounthasbeendisputed byObstfeld (1982),Ostry(1988),and,morerecently,
byGavin(1990).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 497

Therefore, theHLM hypothesis wassatisfied fortransitoryterms oftrade


but
disturbances, apparently not for permanent ones.
Atthesametime,theviewthattransitory changesintheterms oftrade
haveunambiguous effectsonprivate is
saving misleading fortwo reasons.
Whena country experiences a temporary adversetermsoftradeshock
thatraisesthepriceofcurrent imports relativeto futureimports, con-
sumershavean incentive to postponetheirpurchases-that is, to save
more.So, whileconsumption-smoothing considerations-the basisfor
theHLM effect-imply thatprivatesavingshoulddeclineinresponseto
the temporary real incomedecline,the so-calledconsumption-tilting
motives imply thatprivate savingshouldincreaseas agentsreducecurrent
consumption in line with the increasein its relativeprice.3On these
grounds alone, therefore, whathappensto savingis theoretically am-
biguous and depends on the relativemagnitudes of the consumption-
smoothing and tiltingmotives.The parametergoverning thislatter
motiveis the intertemporal elasticityof substitution.Relatively large
valuesofthisparameter implythat,in responseto a given(transitory)
movement inthetermsoftradeand,hence,intheintertemporal relative
price(consumption rateof interest),consumers increasetheir savingby
a relatively largeamount;itfollowsthatthelargeris thiselasticity, the
greateris the increase (the smaller the in in
fall) privatesaving response
to a transitory adverseshockto thetermsoftrade.
In addition,however,whenthereare nontraded goods,an adverse
terms of trade shock will lead consumers to substituteawayfromrela-
tivelyexpensive imports in favor ofhome goods,thereby bidding uptheir
relativeprice.If thetermsof tradeshockis temporary, theresulting
temporary realappreciation willcontribute to a furtherincreasein the
consumption interestrateand,hence,a further increaseinsaving.4 The
parameter governing theswitch from imports tohomegoodsand,hence,
themagnitude of thetemporary real appreciation and increasein the
consumption rateofinterest istheintratemporal ofsubstitution
elasticity
betweentradablesand nontradables. A relativelylargevalue of this
parameter impliesa largeincreaseintheconsumption rateofinterest and
a commensurately largerisein saving.It maybe concluded, therefore,

3A transitory adverseshockto thetermsof traderaisesthecostof current


consumption relativeto future consumption (theconsumption rateofinterest)
becauseittemporarily raisestherelativepriceofimports,
whichentersintothe
consumer priceindex.The latter,however, returns
to itstrendleveloncethe
termsoftradereturn totheirtrendlevel.Forfurther
detailson theconsumption
rateofinterest, see Dornbusch(1983).
4 See Ostry(1988).The reasonis thesameas givenin thepreviousfootnote.
Thetransitory riseintherelativepriceofnontradables
raisestheconsumer price
indextemporarily, making currentgoodsmoreexpensive relativetofuturegoods.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
498 JONATHAND. OSTRY and CARMEN M. REINHART

thatthelargerare eithertheintertemporal or intratemporal elasticities


the
ofsubstitution, greater willbe the increase (the smaller the decrease)
in privatesavingin responseto a temporary adversemovement in the
termsoftrade.The outcomein anycase is an empirical matter thatcan
onlybe addressedthrough estimation ofthesetwocritical parameters.
The approachtakeninthispaperinvolves estimating the"structural"
parameters ofa representative household's utility function. Thebasisfor
suchan approach,inpreference tothealternative ofestimating reduced-
formconsumption or savingfunctions, is relatedto theLucas critique.
As is wellrecognized, theLucas critiqueimpliesthattheremaynotbe
anything thatcouldproperly be calleda consumption orsavingfunction,
inthesenseofa stablefunctional relationship that is independent ofthe
widermacroeconomic context.5 In contrast to previousstudies,we em-
ploya disaggregated commodity structure accordingto whichagents
consumebothtradedandnontraded goods.Disaggregation permits esti-
mationofthetwoparameters ofinterest: theintertemporal elasticity of
substitution andtheintratemporal elasticityof substitution between trad-
ablesandnontradables. The datasetemployed is also suitableforcom-
our
paring findings tothose ofprevious studies that employed a one-good
structure. In contrast to manysuchstudies,we findevidencethatthe
intertemporal elasticityofsubstitution issignificantly different fromzero
and lies in the0.3 to 0.8 range,depending on theregionconsidered.
Intratemporal substitution elasticitiesare estimated to be significantly
higher, and indicate that this parameter-which to our knowledge has
beenentirely in
ignored previous Euler for
equationestimations develop-
ingcountries-playsa critical roleindetermining thesignandmagnitude
oftheHLM effect in thesecountries.
Finally,althoughtheempiricalresultsof thispapercan be used to
analyzea varietyof otherissues-including the effects of permanent
termsoftradeshocksandtheimpactoftradereforms (whichalterthe
internaltermsoftradeofthecountry thatundertakes them)-wefocus
in whatfollowson temporary termsof tradeshocks,mainlybecause
recentempirical evidencerelating to thedeveloping countries suggests
thatthetransitory component ofsuchshocksisquantitatively important.6
Forinstance, Cuddington and Urzua(1989)foundthatfully60 percent
of all shocksto commodity priceswereof a temporary nature,and
Mendoza(1992)reported a similarresultrelating to thetermsoftrade
ofdeveloping countries.
5See Hall ofthisview.
(1988,p. 340),foran elegantrestatement
6 On theusefulness ofourestimates to theissueofpermanent termsoftrade
shocks,see, forexample,Ostry(1988),Gavin(1990),andEdwardsand Ostry
(1992); on theirapplicability
to tradereform issues,see Calvo (1987),Ostry
1990,1991,1992),Edwardsand Ostry(1990),and Ostryand Rose (1992).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATESAVINGAND TERMSOF TRADE SHOCKS 499

Theremainder ofthispaperisorganized as follows.


SectionI illustrates
theroleofpreference parameters in theHLM effectin thecontextof
a simpletwo-periodmodelthatadmitsclosed-form solutions.For the
purposesofempiricalimplementation, however, SectionII considersthe
infinite-horizon
stochastic, versionofthismodelandpresents theopti-
conditions
mality foranintertemporal equilibrium modelinwhichhouse-
holdsconsumebothtradedand nontraded goods.SectionIII describes
theapproachto estimationandpresents theempirical results.
The main
are containedin SectionIV.
conclusions

I. A SimpleModeloftheHLM Effect
Considera smallopen exchangeeconomywheretherepresentative
C, in each periodaccording
householdderivesutility, to thefollowing
constant of
elasticity substitution
(CES) function:7
-
C = (am - 1/+ nl 1/e) 1, a, > O, (1)

wherem(n) denotesconsumptionofimportables
(nontradables).Agents
are assumedto livefortwoperiods.8
Intertemporal
consumption deci-
sionsmaximizethefollowing function
CES utility subjectto constraints
below:
specified

U = (C - + pC -l/) 1 , a > 0, p < 1, (2)

wherethesubscripts 1 and 2 denoteperiods1 and 2, respectively,and


where 3 denotesthe subjectivediscountfactor.In equation(1) the
parameter E denotestheintratemporal ofsubstitution
elasticity between
tradables(importables) andnontradables. Largervalues ofthisparame-
terimplygreaterresponsiveness to relativeprice(real exchangerate)
changes.A valueof unitycorresponds to thelogarithmic case,
utility
whilevaluesabove (below)unityimplygrosssubstitutability (comple-
mentarity).In equation(2) theparameter crdenotestheintertemporal
ofsubstitution.
elasticity Largervaluesofthisparameter implygreater
7 The model ofthissectionis a versionoftheone developedin Ostry
simplified
(1988). A stripped-downversion is presentedhere only for the purposes of
illustrating the role of preferenceparametersin the HLM effect.The model to
be estimatedempiricallyis presentedin Section II.
8 As is well known,thetwo-periodassumptionis notrestrictive here,sincethe
secondperiodmayrepresenttheaggregationofa large(possiblyinfinite)number
of futureperiods.The motivationforthetwo-periodstructure is thatit allowsus
to obtainclosed-form solutionsfortheresponseofprivatesavingto termsoftrade
shocks,somethingthatis precludedin the infinite-horizon versionof the model
developed later.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
500 JONATHAND. OSTRY and CARMEN M. REINHART

tomovements
responsiveness inintertemporalrelativeprices(consump-
tionratesof interest).Equation(2) collapsesto a logarithmicutility
functionwhen oC= 1.
Perfectcapitalmobility is assumed,andtherefore thecountry facesa
given(in terms of the numeraire) worldinterestrate.9All debts are
requiredtobe repaidby the end ofthesecond period.These assumptions
implythattherepresentative householdmaximizes equation(2) (given
equation(1)), subjectto theconstraint thatthepresent valueofexpen-
dituresnotexceedthepresentvalueofresources. The latter,whichwe
refer
tobelowas lifetime wealth,is assumedtotaketheformofa stream
ofendowments oftradable(importable andexportable) andnontradable
goods.The solution to thisoptimization problemyieldsdemandsofthe
form
mi = m [pi, q, Pi C,(R, W)] (3a)
ni = n[pl, ql,P1C(R, W)] (3b)
m2 = m2[P2, q2,P2C2(R, W)] (3c)
n2= n2[P2, q2,P2C2(R, W)], (3d)
wherepi andqidenote,respectively,
therelativepriceofimportablesand
and
nontradables, Pi denotesthe consumer priceindexin periodi; is
R
theconsumption discount which
factor, is givenby R = 1/(1+ r),where
ristheconsumption rateofinterest;
andWisrealwealth.1' Theconsump-
tiondiscountfactor,R, is relatedto the worlddiscountfactor,R*,
accordingto
R = R*P2/P,. (4)
Thus,theconsumption discount factortakesintoaccountthattherele-
vantinterestrateforintertemporal consumption decisionsdependson
the evolutionof the relativepricestructurethrough time.Since the
consumer priceindexin anyperioddependson therelativepricesof

9Without lossofgenerality,thenumeraireis takentobe theexportable good.


Forrecent evidence theviewthatdeveloping
supporting ingeneral,
countries, can
be o
characterized as financially
openeconomies, see Haque andMontiel(1991).
Although it is straightforward
to obtainexplicitsolutionsforthedemand
functionsin thiscase, thereis no particular
interestin doingso. In equations
(3a)-(3d),wehavemadeuseofthefactthattheoptimization problem as specified
satisfies
theassumptions fortwo-stage
necessary budgeting (GoldmanandUzawa
(1964)).Accordingly, demandsina givenperioddependonlyon relative prices
inthatperiodandaggregate spendinginthatperiod.Therealvalueofaggregate
spending, in turn,dependsonlyon lifetime wealthand on theintertemporal
price,R (theconsumption
relative discountfactor,whichisequalto 1 over1 plus
theconsumption rateofinterest).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATESAVINGAND TERMSOF TRADE SHOCKS 501

importablesandnontradables,theconsumption andworlddiscount fac-


from
torswilldiffer oneanother whenevertheterms oftrade(therelative
priceof importables)or thereal exchangerate(thereciprocalof the
relative of is not
price nontradables) expected toremain constantthrough
conditions
time.To close the model,market-clearing fornontradable
goodsare specified:
n [pl, ql, PI C1(R, W)] = nf (5a)
n2[p2, q2,P2C2(R, W)] = n2, (5b)
wherenifrepresents theendowment goodsin periodi.
of nontradable
we
Finally, can the
define of
ratio to
saving
private GDP (grossdomestic
product)(s) as follows:"
x1 - l (ml - i) (6)
X1 + pi m + qi nl
wherewe haveusedthemarket-clearing forhomegoodsand
conditions
whereYx,mi represent theendowments oftheexportable andimportable
goods,respectively, in thefirstperiod.
Considernowtheeffect of a transitory in thetermsof
deterioration
trade-thatis, a riseinpl, with p2 constant.To simplifytheanalysis,it
to consideran initially
is convenient stationaryequilibrium in whichall
pricesandquantities areconstant overtime.Differentiatingequation(6)
aroundan initialequilibrium withs = 0, andusing(5a) and(5b) tosolve
fortheeffectson ql and q2gives
s = b(1
ds )E
b (1 - k)E¢T -
- b(1 k)X, (7)
d log p- bE + (1 - b)r
whereb is theinitialexpenditure shareon importables (a positivefrac-
tion),k is the ratioof currentto lifetimespendingorwealth, andXis the
ratioofexports to production The first
oftradables.12 termon theright-
handsideofequation(7) represents theintertemporalsubstitutioneffect,
whichis equal to (minus)theproductoftheelasticity ofcurrent spend-
ingwithrespectto theconsumption discountfactor,(1 - k)cr,andthe
changein the discountfactor,be/(be+ (1 - b)a). This expression is
increasing in both e and c, which shows thatsavingrises bymore, the

"Under theassumption ofno historicaldebtcommitments, thisratiois also


equal to theratioof thecurrent accountbalanceto GDP, sincethereis no
investment savingin themodel.
or government
12Clearly,bothkandXarepositive Ifthehorizon
fractions. ofhouseholds were
a goodproxy
infinite, rate.Itshouldalsobe noted
forkwouldbe therealinterest
thatifthereis no domestic
production (or endowment)ofimport substitutes,X
is equal to unity.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
502 JONATHAND. OSTRY and CARMEN M. REINHART

largerare eithertheintratemporal or intertemporal elasticitiesof substi-


tution.For a givenrisein theconsumptioninterestrate,largervalues of
o( implylargerincreasesin saving.For a givenelasticityof saving,larger
valuesofe implylargerincreasesin theconsumptionrateofinterestand,
hence,largerincreasesin saving.The secondtermon theright-hand side
of equation (7) representsthe consumption-smoothing effect,which
dependson theinitialvolumeof exports.Withreal incomefallingbelow
itstrendlevel,theconsumption-smoothing effecttendsto reduceprivate
saving.Equation (7) summarizesthe mainresultof thissection,namely
thatprivatesavingwill increaseby more (fall by less) in responseto a
temporarydeteriorationin the termsof trade, the larger are either
intertemporal or intratemporal elasticitiesof substitution.

II. The Stochastic Euler Equations

The model of SectionI was presentedin orderto illustratethe role of


preferenceparametersin theHLM effect.Witha viewtowardempirical
implementation, however,we need to generalizethatmodel by allow-
ing for uncertaintyand more than two periods. Accordingly,consider
an economy with an infinitelylived representativehousehold whose
objectiveis to choose a consumptionstreamthatmaximizes
- 1/ 1/
[(r/(a - 1)] Eo E p(aml + nl -le)
t=O

a, 3, E, ( > , p < 1, (8)


subject to the series of budgetconstraints
ptmt + qtnt = ptmt + qt it + Xt
+ At - (l/R*-1)At- 1, t > 0, (9)
and the transversality
condition13
t
lim (1/R*)At = 0, (10)
t-- 00i = o

whereEo is the expectationsoperatorconditionalon information avail-


able at time 0; At denotes the real level of debt carriedfromperiod t
to period t + 1;14 (1/R*) - 1 = r* is the real interestrate (in termsof
13 oftheappropriate
See Chamberlain andWilson(1984)fora fuller
discussion
noponzi-game constraintinaninfinite-horizon
consumption modelunderuncer-
tainty.
14We assumethattheinherited levelof debt,A_1,is givenand,forconve-
nience,setequal to zero.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATESAVINGAND TERMSOF TRADE SHOCKS 503

thenumeraire) on thedebt;15
and remaining notationis as specified
in
SectionI.
The problemoftheconsumer, then,is tochoosean optimalsequence
equation(8),subjecttoequations(9) and(10).
(m,,nt,At)thatmaximizes
The first-order conditions
necessary foran optimum are
I -m +l- Ct - 1(-
aml - lI + n' - /e_ l
RPt m, (. 3
Et
P,-+ + +l a(C-1) t+1 -1 1
R*pt+-am1-+e + n'1-,e ,
nMt

a (n,/m)' =ptIqt. (13)


Equation(11) is theintertemporal
Eulerequationassociatedwithim-
portables consumptionin two consecutiveperiods; it states that the
costofgiving
utility
marginal uponeunitofmattimetshouldbe equated
totheexpectedutility
gainfrom onemoreunitofm att + 1.
consuming
Equation (12) is the analogous conditionrelatingthe marginalrate of
substitution ofgoodn attandt + 1to therelevant
betweenconsumption
relativeprice. Finally,equation (13) is the nonstochastic
intertemporal
first-orderconditionequatingtheintratemporal marginal rateof sub-
stitutionbetweenimportables and nontradables to the corresponding
relativpe it
c ratio.It can be verified
thatequations(11)-(13) are not
independent. combining
Specifically, (13) wmoeither of the two re-
maining equationsyieldsthettha . or
te ivnat nonstochas-
ticfirst-orderconditionholds,equations(11) and (12) do notprovide
independent restrictions
on theevolution ofconsumption through time.
Itisperhapsworth emphasizing thedifferences between equations(11)
() and thecorreportables
and (1on2)tradabs nd
corresponding
model.'6In sucha model,the relativepriceratiothatis relevantfor
transforming presentintofuture consumption is therealinteresrate-
thatis,thenominalratedeflated bytherateofchangeoftheaggregate
priceindex.Whenrelativepricesare notconstant, however,as when
thereare termsoftradeshocks,theappropriate intertemporal relative
price ratio needs to take account of such changes. This is why,for
in
example, equation(11) thepriceratio that the
premultiplies marginal
rateofsubstitution insidetheexpectation signis therealrateofinterest

Clearly,then,R * is theassociatedworldrealdiscount
factor.
16Thisis particularly
relevant,sinceestimationofconsumption
Euler
equations
fordevelopingcountrieshas been confinedto environmentsin whichthereis a
singleconsumptiongood; see, forexample,Giovannini(1985) and Rossi (1988).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
504 JONATHAND. OSTRY and CARMEN M. REINHART

in termsofthenumeraire, 1 , adjustedfortherateofchangeofthe
1/R*,
termsoftradeovertime,Pt/Pt + 1(thatis,the"own"rateofinterest). If
therelative of is
price imports expected to decline through time,current
importablesconsumption isexpensive relative tofuture con-
importables
In
sumption. consequence,offsetting changes in the marginal of
rate
are requiredinprecisely
substitution thesamedirection as wouldoccur
iftheworldrateofinterest wereto rise(R * wereto fall).An analogous
interpretation carriesoverto equation(12), whereinthe appropriate
relativepriceforthepurposeof determining themarginal rateofsub-
stitutionbetweennontradables consumption in consecutive periods
involvestherealexchangerateratio,q,/q,+i.
Giventime-series dataon importables andnontradables consumption
and on interest ratesand import, export,and nontradables prices,itis
possibleto estimatethe systemconsisting of equations(11)-(13) and
recoverthemainparameters ofinterest. Since(13) mustholdidentically
(in theabsenceofmeasurement error),and since(11) and (12) are not
independent, giventhat(13) holds,it is sufficient in theestimation to
considerequation(11) alone.The restrictions on thejointbehaviorof
consumption ofimportables and nontradables, thetermsoftrade,and
therelevant rateofreturn impliedby the maximization oftheexpected
function
utility givenbyequation(8), subject to the constraints
givenin
(9) and (10), are summarized in equation(11). In addition,giventhe
assumption of rational
expectations, we can use equation(11) to define
thedisturbance
-
PPt=
amt;1 +
amt + 1n}^^ m/r(E Mt + 1 e +1 1 14)
ut -1 + -m
R t+ 1am n m,

withanyvariablethatisintheinformation
whereutmustbe uncorrelated
setof agentsat timet.

III. EmpiricalResults

Theparameters oftherepresentativehousehold's function


utility out-
linedintheprevioussectionswereestimated usingannualpooledtime-
data for 13 developingcountries.The countries
series,cross-section
examined intheanalysisincludefourAfricancountries-Egypt,Ghana,
Cote d'Ivoire,and Morocco;fiveAsian countries-SriLanka, India,
Korea, Pakistan,and the Philippines;and four Latin American
countries-Brazil,Colombia,CostaRica, and Mexico.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 505

Data Issues
Data coverageforeach country beginsin 1968and endsanywhere
between1983and1987;see theAppendixfora listofthesampleperiod
foreach country and thesourcesofthedata.
As equation(14) highlights, estimation oftheintertemporal and in-
tratemporal elasticitiesofsubstitution requiresdataon householdcon-
sumption oftradedandnontraded goodsandthetermsoftrade.While
timeserieson thetermsoftradeare readilyavailable(see Appendix),
consumption dataare generally notdisaggregated intotradedandnon-
tradedcomponents. Guidedbythetheoretical framework, theseseries
wereconstructed usingdatafroma variety ofsources.17
The timeseriesforconsumption of importables was constructed as
follows.The agricultural, mining, and industrialsectors produce traded
goods;GDP originating inthesesectors thusdefines domestic production
of tradedgoods. Privateand publicservicescomprisethe nontraded
goodssector.Domesticproduction ofimport substitutes is calculatedas
domestic production oftradedgoodslessexports, ontheassumption that
exportables are notconsumedat home.l8If markets clear,all domestic
production ofimport is consumedat home.Consumption
substitutes of
import substitutes plusconsumer goodsimports, whicharetotalimports
less imports of intermediateand capitalgoods,makeup theseriesof
interest-consumption ofimportables. Nontraded goodsconsumption is
residuallycalculatedas totalprivateconsumption less consumption of
importables.19
The relevant pricedeflatorsfortheconsumption oftradedand non-
tradedgoods are priceindicesforimportsand services,respectively.
Depositratesofinterest wereusedwhenavailable,and,intheirabsence,
a moneymarket rate.Allconsumption dataareconverted toa percapita
basisbydividing theaggregates by the existingpopulation.

Methodology
Weestimatetheparameter vector,x= [, E,or]byfitting
thefirst-order
conditiondefinedin equation(14) to the panel data usingHansen's
7 All seriesare availableuponrequest.
18
bea restrictive
Thismayadmittedly assumptioninthecaseofsomecountries,
thedatadonotpermit
butunfortunately, ustodisaggregate
consumptionfurther.
19Toensure alltheseriesusedtodisaggregate
consistency, into
consumption
itstradedandnontradedcomponents (GDP bysector, private
consumption,
exports, areona national
andimports) income accounts(NIA)basis.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
506 JONATHAN
D. OSTRYandCARMENM. REINHART

(1982) generalizedmethodof moments(GMM).20 The residualsin the


estimatedequation are partlyforecasterrors,which,by the assumption
ofrationalexpectations,are uncorrelatedwithanyvariablein theagent's
information set at timet; in technicalterms,thoseerrorsare orthogonal
to anychoseninstrument knownto agentsat timet. The assumptionthat
all available information is used in forecastingfutureconsumptionand
prices(thatis, m,+ 1,n,+ 1,qt+ 1,andp + 1) allowsus to use a largenumber
of instruments to estimatea smallernumberof parameters.That excess
of instruments over estimableparametersyieldsa testableset of over-
identifying restrictions.In reality,however,the errorterm may also
include measurementerror. Any systematicpart of that noise-say,
owing to serial correlation-should be allowed for in the estimation.
Simply,it may be more efficientto fitthe orthogonality conditionless
tightlyin those periods when it is knownmeasurmenterrorswells the
compositeresidual.
Understandingthe complex natureof the disturbances-thatis, the
ut-is criticalto the estimationstrategy.Serial correlationamongthe u,
mayarisefora varietyofreasons.First,as illustrated byHayashiand Sims
(1983), current values of consumption,m, and n,, may not be observed
before expectationsof futureconsumption(m,+1, n,+) are formed,
implying some laggedvalues ofu,are notin theinformation set; thismay
make today'sforecasterrorcorrelatedwithlastperiod'syetunobserved
error.21
Second, the natureof themeasurementerrormaymake the residuals
seriallycorrelated;time aggregationproblemsin annual consumption
data, as discussedin Hall (1988), introducea first-order movingaverage
process with a knownparameterin the errorterm. Since the moving

20The a is somepositivenumber thatdenotestheweight attached


parameter
to theimported goodin theperiodutility function.In theanalysisthatfollows,
a (whichis notofimmediate isnotjointly
interest), estimated withtheremaining
parameters. Instead,thefollowing valueswereused: 0.85 forAfrica,1.14for
Asia,and0.58forLatinAmerica.Thesevalueswereobtainedbyestimating the
nonstochastic first-ordercondition
(equation(13)) usingordinary leastsquares
(OLS). Sincewe testedforand foundcointegration amongrelativepricesand
consumption ofimportables andnontraded goods,we knowthatOLS provides
consistentparameter estimatesfora. Byimposing inthesubsequent estimation
thevaluesof a, we increasetheefficiency of theestimates of theremaining
parameters. Theestimates ofe obtainedbyapplying OLS to (13) werealsoused
as thestarting valuesin thesubsequent GMM estimation.
21Thisproblemis notlikelyto arisewithpricesandinterest rates,whichare
generally availablemonthly withlittleor no lag. However,fortheconsumer
making two-period forecastsofconsumption, itisnotunlikely
thatoverestimating
(underestimating) today'sconsumption levelleadsto a similarerrorin thesub-
sequentperiod,makingthetwocorrelated.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 507

averageparameter is known,theconstraint thatthedisturbances follow


a first-ordermoving average process is taken into account by quasi-
differencingtherelevantseries.22
In addition, becauseofthediversity of
countriesincludedinoursample,as wellas thefairly longperiodofcov-
erageconsidered, we allowforthepresenceof moregeneralformsof
in
heteroscedasticitythedisturbances.
The estimation proceedsundertheassumption thattheparameters
thatcharacterize householdpreferences are identicalacrosscountries
andregions.Although, as wewilllatershow,homogeneity oftastesmay
be a restrictive
assumption, iteconomizes on thenumber ofparameters
to be estimatedandallowsforthemaximum degreesoffreedom.23 Two
differentsetsof instruments are employed.Neitherinstrument set in-
cludesvariablesmeasuredat timet,sincethemoving averageprocessin
theerrorterm would in
result these variables beingcorrelated withthe
residual,ut.The selectionofinstruments is nottrivial,sincetheuse of
instrumentsthatarecorrelated withtheresidualwouldresultininconsis-
tentestimates.Themostrecentpermissible instrumentisonelaggedtwo
periods.The firstvectorofinstrumental variables
zl/ = [constant,mt-/mt-2,nt-1lnt-2,
Pt-2(R*-2pt-1), r- , nt-1],
usessixinstruments.Thisimpliesthatthereare sixorthogonality
con-
ditions;withthreeparametersto be estimated,thereare threeover-
The secondinstrument
restrictions.
identifying setreplacesthelevelsof
of
consumption importables and nontradedwith theirratio
z2/ = [constant,
mt- /m- 2, lt- 1lt- 2, Pt- 2 (R- 2pt- 1), mt - 1 nt- 1].
threefreeparameters
In thelattercase thereare fiveinstruments, (as
two
before),and,therefore, overidentifying restrictions.
Whilethevari-
ationintheinstrument setisslight, oftheestimates
comparison produced
by each set sheds lighton which are mostsensitive
parameters to the
choiceofinstruments or,in other which
words, parameterestimatesare
morerobust.24

22Fora discussionofhowthemoving
complete averageparameteriscalculated,
see Working (1960) and Hall (1988).
23Thisassumption willbe relaxedlaterwhenregional estimatesoftheprefer-
enceparameters are estimated.
24A common procedure intheexisting literature
on estimationofEulerequa-
tionsis toallowtheinstrument settovarybyintroducing morelags,considering
instrument setssuchas z3' = z lt,,zl - l]. Ifoneis workingwithtimeseriesfor
a singlecountry, theaddedlaginvolves thelossof1 degreeoffreedom. However,
inthepresent analysistheaddedlagwouldentailthelossof13degreesoffreedom
(1 foreachcountry). Forthisreason,we consideronlythemostparsimonious
instrument sets.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
508 JONATHAND. OSTRY and CARMEN M. REINHART

EstimationResults

Table1 reports theparameter estimates foreachinstrument setandthe


minimized valueoftheobjective function,J,whichHansenandSingleton
(1982)showedto be a teststatistic forthevalidity oftheoveridentifying
Theparameter
restrictions.25 estimates for3,e, ando aresimilar forboth
instrument setsand are economically meaningful. The discount factor,
P, in
falls the0.96-0.99 range. The intertemporal elasticityof substitu-
tion,ca,is in the 0.38-0.50rangebutis largerelativeto itsstandard
errors.26The intratemporal elasticityofsubstitution liesinthe1.22-1.27
that
range,indicating importables gross are substitutes fornontraded
goods.27 The are
J-statistics small relative to the of
degrees freedom (for
eitherinstrument set), indicating that the overidentifying restrictions
imposedbythemodelare notrejectedbythedata; thatis, thethree
parameters estimated do a good job of satisfying eitherthefiveor six
orthogonality that
conditions depend on the instrument set.The quasi-
differencing of the data and the correctionfor heteroscedasticity
producedregression residualsthatare whitenoise.
Of notableinterest is thefactthat,in contrast to previouswork-
including Giovannini (1985)and,to a lesserdegree,Rossi(1988)forthe
developing countries,and Hall (1988)fortheUnitedStates-theinter-
temporal ofsubstitution
elasticity isestimated tobe significantly different
fromzero.Thismeansthat,in responseto shifts in real(consumption-
based)ratesofinterest, households wouldbe expectedto alterthetime
profileof theirconsumption, increasing thegrowth rateofthelatterin
response to an increase in realrates of return.
A possibleexplanation ofourfinding ofa statistically significant de-
of
gree intertemporal substitution relates to the restrictiveassumptions
employed byprevious researchers.28 Specifically, forthemostpart,these
studieseitherassumedthe existenceof a singleconsumption good-
making nodistinction between consumption oftradableandnontradable
goods;ortheyassumedthatstandard consumption andpriceserieswere

25TheJ-statisticis distributed
as X2(n) underthenullhypothesis.
The degrees
offreedom, n, are equal to thenumber ofoveridentifying
restrictions.
theseestimates
26Interestingly, areconsistent withvaluesinthe2.5-3.0range
forthecoefficient ofrelativeriskaversion(thereciprocal oftheintertemporal
ofsubstitution)
elasticity usedincalibratingrealbusinesscyclemodels:see, for
example,Stockman andTesar(1990).
27Thisis slightlyhigherthantheestimates obtainedbyBackus,Kehoe,and
Kydland(1991)fortheUnitedStates.
28Ofrelatedinterest arerecentempirical papersthathaveincludedmoneyin
modeling theconsumer's choiceproblem:see, forexample,Arrau(1990)and
Ecksteinand Leiderman(1992).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 509

Table 1. Estimates
oftheModelPoolingAll Regions
forAlternative
Instrument
Sets
All Countries
Parameters InstrumentsetI InstrumentsetII
e 1.279 1.223
(0.154) (0.351)
a 0.383 0.504
(0.087) (0.228)
[3 0.955 0.991
(0.033) (0.041)
Memorandum items:
J-statistic 5.707 2.590
(0.127) (0.274)
Numberofobservations 208 208
Note: The data and sampleperiodscoveredare describedin detailin the
Appendix.Standarderrorsare shownin parentheses. Forinstrument setI the
valueoftheJ-statistic,
0.95 critical whichis distributed
as X2(3)underthenull
hypothesis,is7.815.Forinstrument
setII therelevant
critical
valueis 5.991.The
probabilityvaluesoftheJ-statistic
appearin parentheses.

reasonableproxiesforthe "true" utility-based indices.29


Eitherassump-
tionis likelyto provetoo restrictive
in the case of developingcountries,
whichare frequentlysubjectedto termsof tradeshiftsand whichcom-
monlyexperiencelarge movementsin real exchangeratesthatalterthe
relativeprice between importablesand home goods. The practice in
previousliteratureof computingthereal interestratethatis relevantfor
consumptiondecisionsas thenominalratedividedby(one plus) therate
of changeof a standardaggregatepricedeflator-thatis, a deflatorfor
whichthe correctutility-based weightshave not been used-may thus
potentiallyimply a seriousmisspecification,especiallywhen the profile
of relativeprices(terms of tradeand real exchangerates)is notconstant
throughtime.
Forthemostpart,researchersinthepasthaveused a linearizedversion
of theEuler equationsconsideredhere fortheparticularcase of a single
consumptiongood:30
Ac, = a + or, + e,, (15)
29
Correctaggregationwouldapplyutility-based tothevarioustypesof
weights
goodsconsumed. However,availableaggregatepriceindicesdo notemploy such
a methodology.
3Noticethattheassumption oflinearityitself
involves
a number ofadditional
restrictions on thejointdistribution
(particularly of consumption and ratesof
relativeto themodelestimated
return), in thispaper.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
510 JONATHAND. OSTRY and CARMEN M. REINHART

where c, is (the naturallogarithmof) aggregateconsumption,r,is the


(conventionally measured)real rateofinterest,and etis a randomdistur-
bance. The coefficient on the real interestrate, a, is the intertemporal
of
elasticity substitution. Giovannini(1985) foundno systematic relation-
ship between in
changes consumption and the real interest rate. Rossi
(1988), who allowed for liquidity constraints, also failed to detect a
in
relationship many ofthe regions considered.Using the countries in our
sample,we estimatedthe morerestrictive versionof the model givenin
equation (15). As in Giovannini(1985), theestimatesobtainedby apply-
inginstrumental variablestechniquesyieldedno systematicrelationship
betweenconsumptionchangesand thereal interestrate.This,of course,
highlightsthat our finding,summarizedin Table 1, of a statistically
significantintertemporal elasticityof substitutionis not a productof the
choice of countriesor period coveredin our sample. It rathersuggests
that,in estimatingthe parametersof consumerpreferences,it is impor-
tantto relax some of the assumptionsunderlying a specificationsuch as
(15). Specifically,in our case, it indicatestheimportanceof disaggregat-
ing between traded and nontradedgoods. A futureline of research,
particularlyrelevantfor developingcountries,would retainthe multi-
good settingemployedhere, but would also relax the assumptionof a
perfectcapitalmarketand allowfortheexistenceofliquidityconstraints,
as in Rossi (1988).

Table2. Estimates
oftheModelUsingPanelData
for13 Developing
Countries:
Instrument
SetI
Latin
Parameters Africa Asia America
e 1.279 0.655 0.760
(0.474) (0.105) (0.172)
a~(xT~ ~0.451 0.800 0.373
(0.159) (0.201) (0.111)
P 0.945 0.995 0.995
Memorandum items:
J-statistic 6.492 6.928 8.333
(0.165) (0.140) (0.080)
SSR 2.857 7.451 1.234
Numberofobservations 62 81 65
Note:See noteto Table1. Thevalueof(3chosenis thatwhichminimizes the
sumofsquaredresiduals (SSR); Jis thevalueofthecriterion quadratic function.
The 0.95 critical
valueoftheJ-statistic, whichis distributedas X2(4)underthe
nullhypothesis, is 9.488. The probability valuesof theJ-statistic appearin
parentheses.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATESAVINGAND TERMSOF TRADE SHOCKS 511

Thus far,we have imposedthe restriction thatpreferenceparameters


are identicalacrossthe threeregionsin our sample,a restriction thatwe
feelis unlikelyto be satisfiedin practice.We nowpresenta set of results
thatrelax thisassumptionby allowingforpossible regionalvariationin
taste parameters.To offsetthe loss of degrees of freedomwhen the
sample is brokenup, we economizeon the numberof parametersto be
estimated.Rather than estimatethe parametervector,L = [p, e, o], as
before,we confineour estimationinsteadto theparameterse and a. This
is in keepingwithour overall objective of sheddinglighton the HLM
effect,since the parametere will not play a criticalrole in thiscontext
(see Section I). Using the same estimationtechnique as before, we
estimatee and a overa rangeoffeasiblevaluesforp. Giventheestimates
of p presentedin Table 1, the search was conductedover the range
0.900-0.995 at intervalsof 0.005. The value of P presentedin Tables 2
and 3 is thatwhichminimizedthe sum of squared residuals(SSR). This
searchprocedurenot only allows us to pinpointp foreach region,but
byimposingitsvalue (as well as imposingtherelevantvalue fora) in the
estimationof e and a, it increasesthe efficiency of these estimates.
The resultsforinstrument sets I and II are summarizedin Tables 2
and 3, respectively.In general,theparametersare estimatedwithpreci-
sion in all regions,and the overidentifying restrictions
imposed by the
model are notrejectedbythedata. However,interesting regionaldiffer-

Table3. EstimatesoftheModelUsingPanelData
SetII
Instrument
for13 DevelopingCountries:
Latin
Parameters Africa Asia America
E 1.441 1.152 1.107
(0.771) (0.270) (0.383)
a 0.443 0.803 0.430
(0.178) (0.235) (0.135)
e3 0.940 0.990 0.995
Memorandum items:
J-statistic 5.019 3.679 3.731
(0.170) (0.298) (0.292)
SSR 3.506 1.661 1.658
Numberofobservations 62 81 65
Note:See notetoTable1. ThevalueofB1chosenis theonethatminimizes the
sumofsquaredresiduals (SSR); Jisthevalueofthecriterion quadratic function.
valueoftheJ-statistic,
The 0.95 critical whichis distributedas X2(3)underthe
nullhypothesis, is 7.815. The probabilityvaluesof theJ-statistic appearin
parentheses.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
512 JONATHAND. OSTRY and CARMEN M. REINHART

encesinpreferences emerge.Irrespective oftheinstrument setused,the


intertemporal elasticityofsubstitution is estimatedto be about0.80 for
Asia, andto be roughly halfas largeforAfricaandLatinAmerica.31 In
effect,estimates ofo do notappearto be verysensitive to thechoiceof
instruments in anyoftheregionsconsidered. The valueofP thatmini-
mizesthesumofsquaredresiduals is around0.94forAfricaandaround
0.995forAsia andLatinAmerica,indicating thatfutureconsumption is
discounted moreheavilyin theAfricancountries considered.
Tables2 and3 alsorevealthatimportables andnontradables arecloser
substitutesinAfricathaninAsiaorLatinAmerica.Partoftheseregional
differences maybe accountedforby regionaldifferences in thecom-
moditycomposition of tradables and nontradables.In particular, the
shareofdurablesinimportables is lowerinAfricathaninotherregions,
and sincenontradables are overwhelmingly nondurable(thatis, ser-
this
vices), may account for thehigher of
degree substitutability inAfrica.
For instrument set I (Table 2), we findgrosssubstitutability between
importables and nontradables forthe African countries
only, whereas for
thesecondinstrument set(Table3), grosssubstitutability is obtainedin
all threeregions.

IV. Conclusions

The traditional
explanation oftherelationshipbetweenthetermsof
tradeand the externalcurrentaccountbalancehas, formanyyears,
restedon theHarberger-Laursen-Metzler hypothesis.Accordingto this
hypothesis,an improvement inthetermsoftraderaisesa country'sreal
incomelevel,andsincepartofthatincreaseinrealincomewillbe devoted
to saving,theimprovement in thetermsoftradeimproves thecurrent
account.
Thispaperhas presented a firstattemptto obtainquantitative
esti-
matesof themainparameters thatdetermine theresponseof private
savingtotransitory
termsoftradeshocksfora cross-section
ofdeveloping
countriesin thecontextof a fullyarticulated
intertemporaloptimizing
model.The mainresultsofourstudyare as follows.
First,theestimatedparameters thatdescribeconsumer behaviorin a
31Rossi(1988)
arguedthatestimatesoftheintertemporal ofsubstitu-
elasticity
tionarebiaseddownward ifliquidity
constraints
arenottakenintoaccount.The
regionaldifferences
in estimatesofo mayreflectthisomission,
sinceempirical
evidence(see Haque andMontiel(1989)) indicatesthattheAsiancountriesin
oursamplearelessliquidityconstrainedthantheirAfrican Unfor-
counterparts.
theHaque-Montiel
tunately, sampledoesnotincludeanyoftheLatinAmerican
countriescoveredbythisstudy.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 513

simplethree-good setting(the intertemporal and intratemporal elas-


ticitiesof substitution, and the discountfactor)are all economically
meaningful, irrespective of thechoiceofinstruments employedand/or
theregionconsidered. Disaggregation of the panel data allowedus to
detectinteresting regional differences. The overidentifying restrictions
imposedby the model are not rejectedby the data.
Second,theestimates oftheintertemporal ofsubstitution
elasticity are
significantlydifferent fromzeroinall theregions.Thisfinding contrasts
withprevious work,including Giovannini (1985)and,to a lesserdegree,
Rossi(1988)forthedeveloping countries, andHall (1988)fortheUnited
States.The implication ofourfinding is that,inresponsetoshifts inreal
ratesofinterest, households in developing countries willgenerally alter
thetimeprofileoftheirconsumption, increasing thegrowth rateofthe
latterin responseto an expectedincreasein realratesofreturn.
Third,our estimatesof the intratemporal elasticityof substitution
suggest thatsubstitution between tradables and nontradables isanimpor-
tantchannelthrough whichtermsof tradeshocksare transmitted to
privatesaving and the current account. In particular, our resultsare
consistent withtheviewthattermsoftradeshocksarelikelytogenerate
substantial fluctuations in realexchangerates,whichin turnaltercon-
sumption rates of interest, therebyaffecting savingbehaviorand the
allocationoftotalexpenditure betweentradedand nontraded goods.
Fourth, theestimates forall theregionsconsidered castdoubton the
viewthatconsumption smoothing is theonlyrelevant factorgoverning
the responseof householdsto transitory termsof tradeshocks.An
important implication ofourestimates is thattransitory termsoftrade
shocksshouldgive rise to intertemporal shiftsin consumption both
directly and through themovements in real exchangeratesthatthey
induce.Calibration ofa dynamic stochastic equilibrium modelusingthe
econometric estimates ofthispaper(see, forexample,Mendoza(1992))
shouldenableone to obtainreasonablequantitative estimatesof the
effectsoftransitory termsoftradeshockson private saving.Preliminary
evidenceinthisregardsuggests that,although privatesavingis likelyto
declineinresponsetotransitory adversetermsoftradeshocks,themag-
nitudeofthisdeclineis likelyto be muchsmallerthanwouldhavebeen
predicted onthebasisofprevious estimates oftheintertemporal elasticity
ofsubstitution. A policyimplication is thattheneedto "finance"transi-
toryadversemovements in the termsof trademaybe smallerthan
previously believed.Giventheestimated parameter values,thisconclu-
sionis likelyto be especially truefortheAsiancountries in oursample
and lessso fortheLatinAmericanand Africancountries.
Finally,whilethepaperhasfocusedexclusively ontheeffects oftransi-

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
514 JONATHAND. OSTRY and CARMEN M. REINHART

torytermsoftradeshocks,theparameterestimatesobtainedhereshould
prove usefulin a varietyof othercontexts,includingthe assessmentof
the effectsof trade reformsand fiscalpolicies.

APPENDIX

Descriptionand Sources of Data

ThisAppendixprovidesa description
ofthedataanalyzedin SectionIII and
liststhesourcesused.

Descriptionof Data
Numberof
Country SamplePeriod Observations
Africa 74
Egypt 1968-87 20
Ghana 1968-83 16
Cote d'Ivoire 1968-85 18
Morocco 1968-87 20
Asia 96
SriLanka 1968-86 19
India 1968-85 18
Korea 1968-87 20
Pakistan 1968-87 20
Philippines 1968-86 19
LatinAmerica 77
Brazil 1968-86 19
Colombia 1968-87 20
CostaRica 1968-85 18
Mexico 1968-87 20

Seriesand Sources
International
FinancialStatistics
(International
Monetary Fund)
Grossdomestic product(GDP)
Privateconsumption (nationalincomeaccounts(NIA))
Exports(NIA)
Imports(NIA)
Interestrate
Exchangerate
Population
WorldEconomicOutlook(International Monetary Fund)
Importunitvalues
Exportunitvalues

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 515

WorldTables,1988-89(WorldBank)
GDP bysectoroforigin
Deflatorforservices
TradeData System(UnitedNations)
Importsofconsumer goods
Providedbytheauthors
Domesticproductionoftradedgoods
Domesticproductionofimportsubstitutes
Consumption oftradedgoods(importables)
Consumption ofnontradedgoods

Descriptionof Series Constructedby theAuthors


Domesticproductionoftradedgoods(NIA basis)
= GDP originatingin theagricultural, and industrial
mining, sectors.
Domesticproductionofimportsubstitutes(NIA basis)
= domesticproductionoftradedgoods(NIA basis)- exports(NIA basis).
Consumption oftradedgoodson an NIA basis(m)
= importsofconsumergoods(convertedtoNIA basis)+ domestic production
assumesexports
(describedabove). This definition
of importsubstitutes
consumed
arenotdomestically andalldomestic importsubstitutes
produced
are consumed.Importpricesare usedas therelevant deflator.
Consumption ofnontradedgoodson an NIA basis(n)
= personalconsumptionexpenditures (NIA basis)- consumption
oftraded
goods.The for
deflator is
services usedas a proxyforthenontraded
goods
deflator.

REFERENCES

Arrau,P., "Intertemporal in a Monetary


Substitution Framework: Evidence
fromChileand Mexico,"PRE Working Paper#549(Washington: World
Bank,December1990).
Backus,D., P. KehoeandF. Kydland, "Dynamics oftheTradeBalanceandthe
TermsofTrade:TheJ-Curve Revisited" Minneapolis,
(unpublished; Minne-
sota:FederalReserveBankofMinneapolis, 1991).
Calvo,Guillermo A., "On theCostsofTemporary Policy,"JournalofDevelop-
mentEconomics,Vol. 27 (October1987),pp. 245-61.
Chamberlain, G. and C. Wilson,"OptimalIntertemporal Consumption and
Uncertainty,"UniversityofWisconsin Working Paper8422(unpublished;
Madison:University ofWisconsin,1984).
Corden,W. Max, "TradePolicyand Macroeconomic Balance in the World
Economy,"IMF Working Paper88/101(Washington: Mone-
International
taryFund,November1988).

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
516 JONATHAN M.REINHART
D. OSTRYandCARMEN

Cuddington, JohnT., and CarlosM. Urzua,"Trendsand Cyclesin theNet


BarterTermsofTrade:A NewApproach,"TheEconomicJournal, Vol. 99
(June1989),pp. 426-42.
Dornbusch,Rudiger,"Real InterestRates,Home Goods, and OptimalEx-
ternalBorrowing," Journal ofPoliticalEconomy, Vol. 91 (February1983),
pp. 141-53.
Eckstein, Z. andL. Leiderman, "Seigniorage andtheWelfare CostofInflation:
EvidencefromanIntertemporal ModelofMoneyandConsumption," Jour-
nal ofMonetary Economics,Vol. 29 (June1992).
Edwards, Sebastian,andJonathan D. Ostry,"TermsofTradeDisturbances, Real
Exchange Rates, and Welfare: The Role of Capital Controlsand Labor
MarketDistortions," OxfordEconomicPapers,Vol. 44 (January 1992),
pp. 20-34.
, "Anticipated ProtectionistPolicies,Real ExchangeRates,andtheCur-
rentAccount:The Case ofRigidWages,"Journal ofInternational
Money
and Finance,Vol. 9 (June1990),pp. 206-19.
Gavin,Michael,"Structural Adjustment to a TermsofTradeDisturbance: The
Role of RelativePrices,"Journalof International Economics,Vol. 28
(May 1990),pp. 217-43.
Giovannini, Alberto,"SavingandtheReal Interest RateinLDCs," Journal of
Development Economics,Vol. 18 (August1985),pp. 197-217.
Goldman,S. M., andH. Uzawa,"A NoteonSeparability inDemandAnalysis,"
Econometrica, Vol. 32, (July1964),pp. 387-98.
, "Intertemporal SubstitutioninConsumption," Journal ofPolitical
Econ-
omy,Vol. 96 (April1988),pp. 339-57.
Hall,RobertE., "Intertemporal SubstitutioninConsumption." Journal
ofPolit-
ical Economy, Vol. 96 (April1988),pp. 339-57.
Hansen,Lars Peter,"Large SampleProperties of GeneralizedMethodof
MomentsEstimators," Econometrica, Vol. 50 (July1982),pp. 1029-54.
, andK. J.Singleton, "Generalized Instrumental VariablesEstimationof
Nonlinear RationalExpectations Models,"Econometrica, Vol.50 (Septem-
ber1982),pp. 1269-86.
Haque,NadeemU., andPeterMontiel, "Consumption inDeveloping Countries:
TestsforLiquidity ConstraintsandFiniteHorizons,"ReviewofEconomics
and Statistics,
Vol. 71 (August1989),pp. 408-15.
, "CapitalMobilityin DevelopingCountries-SomeEmpiricalTests,"
WorldDevelopment, Vol. 19 (October1991),pp. 1391-98.
Harberger, ArnoldC., "Currency Depreciation, Income,and theBalanceof
Trade,"Journal ofPoliticalEconomy, Vol. 58 (February 1950),pp. 47-60.
Hayashi,Fumio, and Christopher Sims,"NearlyEfficient EstimationofTime
Series Models withPredetermined, But Not ExogenousInstruments,"
Econometrica, Vol. 51 (May 1983),pp. 783-98.
Khan,MohsinS., andMalcolmD. Knight, "Determinants ofCurrent Account
Balancesof Non-OilDevelopingCountriesin the 1970s:An Empirical
Analysis,"StaffPapers,International Monetary Fund,Vol. 32 (Decem-
ber1983),pp. 819-42.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions
PRIVATE SAVING AND TERMS OF TRADE SHOCKS 517

Laursen,Svend,and LloydA. Metzler,"FlexibleExchangeRates and the


Theoryof Employment," Reviewof Economicsand Statistics, Vol. 32
(November1950),pp. 281-99.
Mendoza,Enrique,"Macroeconomic Effectsof IncomeShocksin an Inter-
temporalEquilibrium Model of a SmallOpen, Endowment Economy"
(unpublished; Washington: InternationalMonetary Fund,1992).
Obstfeld, Maurice,"Aggregate Spendingand theTermsofTrade:Is Therea
Laursen-Metzler Effect?,"Quarterly Journalof Economics,Vol. 97
(May1982),pp. 251-70.
Ostry,Jonathan D., "TheBalanceofTrade,TermsofTrade,andRealExchange
Rate: An Intertemporal Optimizing Framework," StaffPapers,Interna-
tionalMonetary Fund,Vol. 35 (December1988),pp. 541-73.
"Tariffs and the CurrentAccount:The Role of InitialDistortions,"
CanadianJournalofEconomics,Vol. 23 (May 1990),pp. 348-56.
RealExchangeRates,andtheTradeBalanceina Two-Country
, "Tariffs,
World,"EuropeanEconomicReview,Vol. 35 (July1991),pp. 1127-42.
, "TradeRestrictions withImported Intermediate Inputs:A Comment,"
Journal of Development Economics,Vol. 38 (1992),pp. 403-405.
, andAndrewK. Rose,"An Empirical Evaluation oftheMacroeconomic
Effectsof Tariffs," Journalof InternationalMoneyand Finance,Vol. 11
(February 1992),pp. 63-79.
Rossi,Nicola,"Government Spending, theRealInterest Rate,andtheBehavior
of Liquidity-Constrained Consumersin DevelopingCountries,"Staff
Papers,International Monetary Fund,Vol. 35 (March1988),pp. 104-40.
Sachs,Jeffrey D., "The Current Account and Macroeconomic Adjustment in
the1970s,"Brookings Paperson EconomicActivity: 1, (Washington:The
Brookings 1981),pp. 201-68.
Institution,
, "The Current Accountin theMacroeconomic Adjustment Process,"
Scandinavian JournalofEconomics, Vol. 84, No. 2 (1982),pp. 147-64.
Stockman, A., andL. Tesar,"TastesandTechnology in a Two-Country Model
of theBusinessCycle:Explaining InternationalComovements," (unpub-
lished;Rochester, NewYork:University ofRochester, 1990).
Svensson, LarsE. 0., andAssafRazin,"The TermsofTradeandtheCurrent
Account:The Harberger-Laursen-Metzler Effect,"Journalof Political
Economy,Vol. 91 (February 1983),pp. 97-125.
Working, Holbrook,"Noteon theCorrelation ofFirstDifferences ofAverages
in a RandomChain,"Econometrica, Vol. 28 (October1960),pp. 916-18.

This content downloaded from 129.81.226.78 on Tue, 24 Nov 2015 07:29:55 UTC
All use subject to JSTOR Terms and Conditions

You might also like