Current Accounts in Debtor and Creditor Countries Author(s): Aart Kraay and Jaume Ventura Source: The Quarterly Journal of Economics, Vol. 115, No. 4 (Nov., 2000), pp. 1137-1166 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/2586921 Accessed: 19/08/2010 23:24 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=mitpress. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. 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 [email protected].

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Quarterly Journal of Economics.

http://www.jstor.org

CURRENT ACCOUNTS IN DEBTOR AND CREDITOR COUNTRIES* AART KRAAY AND JAUME VENTURA What is the current account response to transitoryincome shocks such as temporarychanges in the terms oftrade, transfersfromabroad, or fluctuationsin production?We propose this new rule: the current account response equals the saving generated by the shock multipliedby the country'sshare offoreignassets in total assets. This rule implies that favorable shocks lead to deficits(surpluses) in debtor(creditor)countries.This rule is a natural implicationofthe intertemporal approach to the currentaccount ifinvestmentrisk is high and diminishingreturns are weak. Evidence fromindustrial countriesbroadlysupports this rule.

What is the currentaccount response to transitoryincome shocks such as temporarychanges in the terms of trade, transfersfrom

in production? We proposethis new rule: abroad,or fluctuations

the currentaccount response is equal to the saving generated by the shock multiplied by the country'sshare of foreignassets in total assets. This rule implies that favorableincome shocks lead to current account deficitsin debtor countries and currentaccount surpluses in creditorcountries. This rule is a natural implication of the intertemporalapproach to the current account if investment risk is important and diminishing returns are weak. Evidence from thirteen industrial countries broadly supports its empirical validity. A simple thought experiment reveals how natural the new rule is. Consider a small country that receives a favorable transitoryincome shock and saves a part of it. To the extent that this shock does not affectthe expected return to future investments at home and abroad, a reasonable guess is that investors allocate the marginal unit of wealth (the income shock) among assets in the same proportions as the average unit of wealth. Consider firsta countrywith negative foreignassets, that is, with a foreigndebt that exceeds its holdings offoreignequity.Since by definitionthe share of this country'swealth invested in domestic capital exceeds one, an increase in wealth (saving) results in a greater increase in domestic capital (investment), leading to a deficiton the current account (saving minus investment). Con* We are gratefulto Rudiger Dornbusch fordiscussing these ideas withus. We also thank Daron Acemoglu, Ricardo Caballero, Xavier Gabaix, Jakob Svensson, and an anonymousrefereefortheircomments.The views expressed herein are the authors',and do not necessarily reflectthose ofthe WorldBank.

and FellowsofHarvardCollegeand theMassachusettsInstituteof e 2000 bythePresident Technology. TheQuarterly JournalofEconomics, November 2000

1137

1138

QUARTERLYJOURNALOF ECONOMICS

versely, in a country with positive foreign assets, the saving generated by the shock exceeds investmentat home, as a portion ofit is invested abroad. This produces a currentaccount surplus. The sharp result that comes out of this simple example followsfromour assumptions on how countriessave and invest an income shock. First, we assumed that the income shock is partly saved. This is a basic result of forward-lookingmodels of saving, and applies whenever consumption-smoothingoperates as a saving motive. As is customary in intertemporalmodels of the current account, we assume throughout that this is the case. Second, we assume that the countryinvests the marginal unit of wealth as the average one. Here we depart from traditional practice. In existingintertemporalmodels ofthe currentaccount, countriesinvest the marginal unit ofwealth in foreignassets. As a result, these models predict that favorable transitory income shocks generate currentaccount responses that are equal to the saving generated by the shock. We refer to this idea as the traditionalrule, and note that it implies that all countriesrespond to transitoryincomeshockswithsurpluses in the currentaccount.1 Why should countries invest income shocks in the same proportionsas their current portfoliosrather than invest them only in foreignassets? In other words, why should we preferthe new rule over the traditional rule to predict current account behavior? The main theme of this paper is that the effectsof transitoryincome shocks on investment depend on the relative importanceofinvestmentrisk and diminishingreturns.If investment risk is high,investorshave a strongdesire fordiversification that makes them reluctant to shift their portfoliostoward any single asset. If diminishing returns are strong, an increase in domestic capital generates a substantial reductionin its expected return that encourages investors to shifttheir portfoliostoward foreignassets. The traditional rule applies if investmentrisk is low and diminishing returns are strong, since investors are willing to change their portfoliosand income shocks offerthem strongincentivesto do so. The new rule applies ifinvestmentrisk is high and diminishing returns are weak, since investors are reluctant to change their portfoliosand income shocks provide them with weak incentivesto do so. Thus, one can interpretthese 1. The intertemporalapproach to the current account has a distinguished tradition that includes Sachs [1981, 1982], Obstfeld [1982], Dornbusch [1983], Svensson and Razin [19831,Persson and Svensson [19851,and Matsuyama [19871, among others.Obstfeldand Rogoff[19951surveythis research.

CURRENT ACCOUNTS

1139

two rules as alternative benchmarks (or limiting cases) from which to think about currentaccount issues. Which one of these benchmark rules, if either, is more consistentwith the data? To the extentthat income shocks are the main source of variability in saving, the traditional rule states that changes in saving lead to equal changes in the current account. The top panel of Figure I plots the current account against saving using a panel of thirteenindustrial countries for the period 1973-1995. The traditional rule predictsthat the slope coefficientof a regression of the current account on saving is positive and equal to one. Although the current account and saving exhibit a positive correlation of around 40 percent, the estimated slope coefficientis only 0.24. The null hypothesisthat this coefficient is one is easily rejected at conventionalsignificance levels. Of course,there are sources ofsaving variabilityotherthan income shocks. If these additional shocks also affectthe current account, the slope coefficientwill be biased. However, we show later that the findingof a low slope coefficientis robust to the inclusion of a set of control variables designed to capture this possibility.This negative evidence should not be surprising.Since the currentaccount is saving minus investment,the top panel of Figure I is just the well-knownfindingof Feldstein and Horioka [19801that saving and investmentmove almost one-to-oneacross countriesand over time.2 To the extent that income shocks are the main source of saving variability,the new rule states that changes in saving lead to changes in the current account that are proportional to the share offoreignassets in total assets. The bottompanel ofFigure I plots the currentaccount against an interactionterm:the share of foreignassets in total assets times the saving rate. The new rule predicts that the slope coefficientof a regression of the current account on this variable should be one. Not only does the current account exhibit a correlationof 61 percent with this interaction variable, but also the estimated slope coefficientis 0.95. The null 2. Although their initial study focused on cross-sectional comparisons of savings and investment,later studies confirmedthe same results fortime-series comparisons.Feldstein and Bachetta [19911,Tesar [19911,and Obstfeldand Rogoff [1995] review the facts and survey alternative explanations. Feldstein and Horioka [19801 interpretedthe slope coefficientof a regression of investmenton savings as a measure of international capital mobility.The findingthat this is large (in our sample is 0.76) led them to conclude that the evidence is coefficient ..) quite incompatible with the assumption of complete arbitrage in world capital markets." We do not think these coefficientscan be interpretedin this manner and do not share theirconclusion.

The TraditionalRule 0.15

y =0.2362x - 0.0597 R2= 0.1582

0.1 CL

-0.05

***t** * ~~~~~~~~*

0 01

0.4

0.3

-0.05

-0.15 Gross National Saving/GNP

The New Rule y=0.9547x- 0.0131 2= 0.3694

0.08 0.06 0.04

Z

*

I?

*--m--*

-

* Q2 0.04

0.06

0 C

-0.08 -0.06 -0.04 -O.0~W* *0C.,~

0

*4S

0.08

~ ~~ ~ ~ ~ ~~!-t0.08I

(Gross National Saving/GNP) x (Foreign Assets/Total Assets) FIGURE I

Accountin ThirteenOECD Economies,1973-1995 Savingand theCurrent See Appendix2 forvariabledefinitions and data sources.

1141

CURRENTACCOUNTS 0.25 0.2 0.15

O

0.1

U,

!

.......

o.o.

-010

i,~~~~~~~~~~~~~~~~~~~~~~~~ .. ....... ......_ 0)

(

0

Foreign Asset Positions in Thirteen OECD Economies, 1973-1995 See Appendix 2 forvariable definitionsand data sources.

hypothesis that this coefficientis one cannot be rejected at conventional significance levels. We shall show later that this findingis robust to a varietyofalternative specificationsand also to the inclusion of a number of controlvariables. This evidence seems to be consistentwith the new rule. Going beyond statistical performance,there is an additional reason to prefer the new rule over the traditional rule as a benchmark fromwhich to think about current account issues. Figure II shows another well-knownfact: there is a stronghome bias in the portfoliosof OECD countries. In fact, some countries are long in domestic capital and short in foreignassets.3 Many believe that this bias reflectsa variety of costs associated with internationalfinancialtransactions.The traditionalrule suggests that despite these costs, at the margin countries use only foreign assets to smoothincome shocks. By insistingon the validityofthis rule, one is placed in the uncomfortableposition of having to explain why countriesbehave on average as ifthe costs ofholding foreignassets were high,while at the margin countriesbehave as 3. The home bias in countryportfolioshas been documented by French and Poterba [1991] and Tesar and Werner [1992]. Lewis [1999] surveys alternative explanations.

1142

QUARTERLYJOURNALOF ECONOMICS

if these costs were low. The new rule provides a way out of this intellectual pirouette. Since countries smooth income shocks using a combination of domestic and foreign investment that resembles their existing portfolios,there is no discrepancy between average and marginal behavior. If one adopts this view,the Feldstein-Horioka findingthat saving and investment move almost one-to-oneturns out to be nothingbut the flowversion ofthe home bias in countryportfolios.4 The paper is organized as follows. Section I develops a stylized intertemporalmodel ofthe currentaccount that emphasizes the interplay of investment risk and diminishing returns. Section II uses this model to study how saving, investment,and the currentaccount respond to transitoryincome shocks. Section III discusses furtherchannels throughwhich income shocks can affectthe current account. Sections IV and V interpretthe data fromthe perspectivesofthe traditional and new rules. I. A STYLIZED INTERTEMPORALMODEL OF THE CURRENTACCOUNT

Consider a small country populated by a continuum of identical consumers.5There is a single good that can be used for consumption and investment. There are three assets: foreign loans, foreign capital, and domestic capital. Foreign loans are denominatedin termsofthe single good and pay an instantaneous risk-freeinterestrate p. To produce one unit offoreignor domestic capital, one unit ofthe finalgood is required. Capital is reversible. As a result, both types ofcapital have a constant price ofone and deliver a return that is equal to the flow of production net of depreciation.Foreign capital generates an instantaneous net flow 4. Feldstein and Bachetta [1991, p. 203] were clearly aware of this explanationwhen theywrotethat a mean-variance "(... .) investorwho has a high degree of risk aversion or who attributes a large subjective variance to long-terminvestments in foreignassets may want to invest a large share ofhis portfolioin domestic assets (depending on asset yield covariances) even when a substantial expected yield differenceexists in favor of the foreign assets. Since the mean-variance investor's optimal proportionalallocation ofthe assets is independent ofthe total value, an increase in saving that raises the total pool of funds will be invested primarilyin the domesticeconomy."Withhindsight,it is somewhat surprisingthat Feldstein and Bachetta did not pursue this idea. Instead, they argued that the evidence in the 1980s supports the conclusions of the original Feldstein-Horioka paper. 5. The use of the small country assumption implies that the shocks we or idiosyncraticsince theydo not analyze have to be interpretedas country-specific affectthe variables that describe the rest of the world. For the same reason, comparative statics exercises apply only to changes in the country'sappropriate parameter,holdingthe rest ofthe world constant.

1143

CURRENTACCOUNTS

of production that is normally distributed with mean u* and variance *2, where -n* and v* are nonnegativeconstants. Domestic capital generates an instantaneous net flowofproductionthat is also normallydistributedwith mean uTand variance c2, where a is a nonnegative constant and wTis a continuous, twice differentiable and nonincreasingfunctionofthe country'sstock ofcapital. The correlationbetween returnsto domesticand foreigncapital is *dt, where q E (- 1,1) is constant. We motivate diminishing returns to domestic capital bluntly as the result of congestion effectsor negative externalities. Since the representative consumer is infinitesimal,he/she understands that his/heractions have no influenceon the aggregate stock ofcapital.6 Each period, the representativeconsumer decides how much to save and consume, and how to distribute the stock of wealth among alternative assets. Let c be consumption. Consumption sequences are valued as follows:

E

(1)

In l(c) *e -"I* dt

(8 > ).

Let a, k, and k* be the representative consumer's stock of wealth and holdings ofdomesticand foreigncapital. Then, his/her budget constraintis (2)

da=

[Trek+ Trnk* + p(a-k-

k*)-c].

+ k*

* do

dt

+ k* *

* do*,

where X and a* are Wiener processes with increments that are normally distributed with E [dw] = E [dw*] 0, E [dW2] = E[dw*2] = dt, and E[dw *dw*] = q *dt. This budget constraint illustrates the standard risk-returntrade-offunderlyinginvestment decisions. Throughout,we impose the usual transversality conditionto rule out equilibria with bubbles. We also assume that the country'sholdings ofboth typesofcapital are nonnegative.We interpret this assumption as a restrictionon the shape of the functioniT and the set ofpermissiblevalues forwT*, &*, a, and a. The problem of the representative consumer was solved by 6. At the cost of furthernotation, we could generate this dependence by assuming that there is a factorofproductionthat is not priced or that the country faces a downward sloping demand curve forits exports. Since this is well-known, we dispense with the formalities.Consistent with the small countryassumption, we assume that the returns to foreigncapital are unaffectedby the country's investmentpolicy.

1144

QUARTERLYJOURNALOF ECONOMICS

Merton [1971], who showed that the first-orderconditions imply these equations (see Appendix 1):

c = 8 *a

(3)

'r - p =a 2 *ka

(4)

+ a *v**r * kla

la* - p U*2 k*la + a *v**. .*

(5)

k/a.

When deciding the consumption profile,the representative consumer acts as a permanent-incomeconsumer a la Friedman. Equation (3) states that consumptionis a fixedfractionofwealth and is independent of the expected return and volatility of available assets. This is the well-knownresult that income and substitutioneffectsof changes in asset characteristics cancel for logarithmicconsumers. When deciding how to invest his/herwealth, the representative consumer acts as a mean-variance investor a la MarkowitzTobin. Equations (4) and (5) state that the expected excess returns to domesticand foreigncapital must equal the appropriate risk or ofrisk equity premium.Withlogarithmicinvestors,the coefficient aversion is one, and the risk premium is nothingbut the covariance between the returnto the appropriate capital and the return to wealth. The larger is the share of domestic (foreign)capital in consumers'wealth, the larger is this covariance and the larger is the risk premium that is required to hold the marginal unit of domestic (foreign) capital. The sign of q determines whether domestic and foreigncapital are substitutes or complements.For instance, if q is positive, the risk premium of domestic (foreign) capital increases with the holdings offoreign(domestic) capital. Equations (2)-(5) and the initial conditionforwealth provide a complete descriptionofhow the countryevolves over time. This countryis a stochasticversionofthe convexgrowthmodel ofJones and Manuelli [1990]. If 8 is low enough, the stock ofwealth drifts toward infinity.Otherwise, the stock of wealth has a tendencyto reverttoward a finitevalue. In any case, the countryis continuously subject to shocks that move it away fromits expected path. The linear model arises in two important special cases: (i) if diminishingreturns are weak, arrlak 0; and (ii) if investment risk is low,a 0. What happens to the share ofdomestic capital in the country portfolioas wealth increases? Using equation (5) to eliminate k* -

1145

CURRENTACCOUNTS

fromequation (4), we findthis equilibrium condition:

(6)

wp-

= (T* j p)*

Ace a. *2) +

~~~k

2

*_

a

and, applyingthe implicitfunctiontheorem,we findthat (7)_=

ak ~ aa

(1 -qI2)

*

2

(1 2) (1-- q 2)* U2 _(arrak)

*a

k a

That is, k is weakly increasing in wealth but the share of capital in the countryportfoliois weakly decreasing in wealth. For a given capital stock, an increase in wealth lowers the share of domestic capital in wealth and reduces the risk premium on domesticcapital. This leads investorsto increase theirholdings of it. The strongerare diminishingreturns (as measured by awrlak), the weaker is the response ofthe domesticcapital stock to a given increase in wealth. The lower panel shows that the higher is investmentrisk (as measured by c2), the larger is the response of the domestic capital stock to a given increase in wealth. If investmentrisk is negligible relative to the strengthof diminishing returns,c2I(a wrlak) 0, the marginal unit ofwealth is invested in foreignloans, aklaa = 0, and increases in wealth do not affect the country'scapital stock. If diminishingreturns are negligible relative to investmentrisk, c2I(arrlak) ) oc,the marginal unit of wealth is invested as the average one, aklaa = k/a,and increases in wealth do not affectthe compositionof the country'sportfolio. These results provide a rigorous theoretical underpinningto the investmenthypothesesdiscussed in the introduction. -

II. SAVING, INVESTMENT, AND THE CURRENT ACCOUNT DURING BOOMS

Next we examine the joint behavior of saving, investment, and the currentaccount during a temporaryeconomic boom. Let T1 and T2 be two dates with T2> T1. We consider the following 7. In keeping with the small countryassumption,we have implicitlyassumed that foreignholdings ofdomestic capital are constant. Increases in inward foreign investmentwould shiftthe excess-returnsfunction(the left-hand-sideofequation (6)) as diminishingreturnsset in. If inward foreigninvestmentresponds systematically to the same type of incentives as domestic investment,the excess-returns functionwould be flatter,since inward foreigninvestmentwould decline whenever diminishingreturnsset in and thereforeact as a moderatingforce.

1146

QUARTERLYJOURNALOF ECONOMICS

path ofshocks:

(8)

dw =

{0 E -*

dt

t E [OT1) U [T2,co) t E [T1,T2).

Equation (8) describes a sample path in which the country receives a sequence ofunexpected shocks that are E dt times the capital stock during the period [T1,T2),and zero afterward.We referto the period [T1,T2)as an economicboom. Figure III plots the paths of per capita saving (S = da), investment (I = dk), and the current account (CA = da - dk) before,during, and after the economic boom under alternative assumptions. In all cases, the permanent-incomeconsumers who populate this countrysave the income shocks in order to smooth their consumption over time. This is true regardless of our assumptions on investment risk and diminishing returns, and applies equally to debtorand creditorcountries.Having decided to save the shock, consumers must then decide how to allocate the additional savings between domestic capital and foreignassets. We depart from previous intertemporal models of the current account in how we model the investmentdecision. The top panel of Figure III shows the case in which investment risk is not very important and diminishing returns are strong;i.e., c2I(a-rlak) O 0. This limitingcase delivers the traditional rule. Despite the increase in saving that results fromthe boom, investment is not affected. Strong diminishing returns make new investment unattractive and encourage investors to shifttheirportfoliostoward foreignassets. Since investmentrisk is low, consumers have a weak desire to diversifyand easily accommodate a change in portfoliocomposition.In the limit, all the savings generated by the shock are allocated to foreignassets, and as a result, the current account goes into a surplus in both debtorand creditorcountries.8 The middle panel of Figure III depicts the opposite case in which investmentrisk is importantand diminishingreturns are co.This limitingcase delivers the new rule. weak; i.e., c2I(rrlak) As before,the saving rate jumps up during the boom and falls -

0, the 8. While this result depends only on the assumption that o-2/04rr/8k)effectsof an economic boom on expected returns and the risk premium depend crucially on whether a- is "small" or alr/akis "large." In the firstcase, expected returns and the risk premium remain roughlyconstant throughoutthe boom. In the second case, these two variables fall dramaticallyduringthe boom.

1147

CURRENTACCOUNTS Debtor Countries

CreditorCountries

Returns The TraditionalRule: No Investment Risk,Diminishing

-

.-

_

_

.-CA]

__

_..

__

_...

__

Returns The New Rule: Investment Risk,No Diminishing

IS

Returns Riskand Diminishing The GeneralCase: Investment

.-CAI =

IL;-CAII

O I

~~~~~~~~~~I

. ..

t

........._.._

FIGURE III

Saving, Investmentand the CurrentAccount duringBooms These figures are generated under the followingassumptions: (i) no foreign investment,k = 0; (ii) i(k) = a - a *k,with a = 0.04 and a = 0 (I = 0.001) for the case of no diminishing (diminishing returns); (iii) aT = 0.10 (a= 0.15) for debtor(creditor)countries; (iv) p = 8 = 0.02; (v) initial wealth ao = 1; and (vi) the shockE = 0.02.

back to its originallevel afterward.Weak diminishingreturns ensure that new investmentremains as attractiveas existing and so thereis no incentiveto changethe portfolio investment, In addition,highinvestmentrisk makes investors composition. reluctantto change the compositionof theirportfolios.In the

1148

QUARTERLYJOURNALOF ECONOMICS

limit, the shock is invested so as to keep the share of domestic capital in the consumers' portfoliosconstant. This leads to an increase in domestic investment that is more (less) than the increase in saving if these portfoliosare short (long) in foreign assets. This implies that the currentaccount exhibits a deficitin debtorcountriesand a surplus in creditorcountries. Finally,the bottompanel ofFigure III shows an intermediate case; i.e., 0 < u2I(arrlak) < oc.As in the top panel, both saving and investmentjump up during the boom and fall afterward.In this case, however, saving and investment decline during the boom and are lower after the boom than before it. This is due to a reduction in the expected rate of return that lowers saving directlyand lowers investmentboth because saving is lower and also because foreignassets are relativelymore attractive. In this case, the effectsof the boom on investment and the current account reflectthe trade-offbetween two forces.On the one hand, diminishingreturnsencourage consumers to change the composition of their portfoliostoward foreignassets. On the other hand, their desire to diversifyrisk discourages consumers fromchanging the compositionoftheirportfoliostoo much. The result is that the fractionof the marginal unit of wealth that is invested in domestic capital is positive but smaller than the average one. In creditor countries this necessarily implies that the current account goes into a surplus. In debtor countries, however, it is possible that an increase in the share of foreignassets can be achieved by simplyrunninga small currentaccount deficit. As this last example shows, one should not expect strong general results linking income shocks to current account responses. Even in such a stylized model as the one presented here, this response can take many forms depending on a variety of factors. Moreover,we shall next show that simple and realistic extensions ofthe theorylead to an even wider set ofpossibilities. III.

OTHER EFFECTS

OF INCOME SHOCKS

Income shocks are changes in the wealth of a country.The new rule applies if these changes in wealth do not affectthe compositionof the countryportfolio.In the simple model developed above, changes in wealth can only affectthe compositionof the countryportfoliothrough changes in expected returns. But there are reasons to believe that changes in wealth can affectthe compositionofcountryportfolioseven in the absence ofchanges in

1149

CURRENT ACCOUNTS

expected returns. This could be due to psychological reasons. Many believe, for instance, that risk aversion declines with the level of wealth and, ceteris paribus, increases in wealth should lead investors to pursue more aggressive investment strategies. This could also reflectthe investor's optimal response to changes in the compositionofhis/hertotal wealth (human plus financial). If labor income is less risky than financial income, increases in financial wealth raise the overall risk faced by investors and, ceteris paribus, should lead them to adopt more prudent investment strategies. The risk premiumshould take all these considerations into account.9 To explore these issues, we extend the model in two directions. First, we generalize the utilityfunctionas follows:

Ef

(9)

In (c - y) - e-8t *dt.

Consumers now exhibit decreasing (increasing) relative risk aversion in wealth if y > 0 (y < 0). Second, we assume that there is an additional technologythat uses labor to create a flow of A dt.10Appendix 1 shows that these assumpproductionequal toX tions lead to this generalization ofthe equilibrium condition(6): -p = (r* _ p)*

(10)

Ace

+ a' * (l-2)

~~~~k

+ a +(X

-y)Ip'

and, applyingthe implicitfunctiontheorem,we findthat

ak

aa

(1 -

q2)

(1- I2) 2 * U2 _-awlak*(a

k + yA-)/p) a + (A - y)/p

Note firstthat the simple model of Section I obtains if X = y. Unlike the previous model it is now possible that the share of 9. There exists a rather sophisticated literature that analyzes how optimal investment strategies depend on attitudes toward risk, the size and stochastic properties of labor income, and the correlation between asset returns and some aspects ofthe consumer's environment.See Merton [1995] foran overviewofthis research, and Bodie, Merton, and Samuelson [1992] for an example with risky labor income. An importantresult that we do not explore here is that, if actual returns are correlated with changes in expected returns, there is a hedging component in asset demands. This hedging component is positive or negative depending ofthe degree ofrisk aversion and, in the magical case oflog preferences, turns out to be zero. 10. The assumption ofan additivelyseparable aggregate technologybetween labor and capital is less restrictivethan it might seem at firstglance. It arises naturally in models where some formof factor-price-equalizationtheorem holds. See Ventura [1997]. This theorem also justifies the assumption that diminishing returnsare weak at the countrylevel, even ifdiminishingreturnsare strongat the industrylevel.

1150

QUARTERLYJOURNALOF ECONOMICS

domestic capital in the countryportfolioincreases with wealth. That is, the fractionofthe marginal unit ofwealth that is invested in domestic capital could be larger than the average one; i.e., aklaa > k/a. Thus, positive shocks could lead to current account deficitseven in creditorcountries. Not surprisingly,variable risk aversion and labor income do not affectthe conditionsunder which the traditional rule applies. If c2IG9-rrlak)0, the marginal unit ofwealth is invested in foreign loans, and aklaa = 0. Since the additional effectsof wealth on investment work through the risk premium (i.e., how investors manage risk), they do not operate if investmentrisk is negligible relative to diminishing returns. As a result, investor strategies continue to be the same as before,namely to allocate all of their wealth to the asset that pays the highest return. Variable risk aversion and labor income, however,affectthe conditionsunder which the new rule applies. In particular,it is no longer necessary or sufficientto assume that c2I(arrlak) o to obtain the new rule, since this conditioneliminates onlyone effect of changes in wealth on the country'sportfolio:the effectthrough changes in expected returns.Now there are two additional effects of changes in wealth. First, there is a risk-aversion effectmeasured by y. Ceteris paribus, in the realistic case where y > 0, increases in wealth reduce investors'risk aversion,inducingthem to increase the share of domestic capital in the countryportfolio. Second, there is a wealth-composition effectmeasured by X. Ceteris paribus, increases in wealth raise the share of financial wealth in investors' total wealth, inducing them to reduce the share of domestic capital in the countryportfolio.The new rule applies whenever these three effects cancel each other. For instance, in the simple model of Sections II and III, we assumed implicitlythat X = y = 0 so that the risk-aversion and wealthcompositioneffectsare nil, and our sophisticated investors act as simple mean-variance ones. Consequently,the new rule applies if and only if we shut down the effectthat income shocks have on countryportfoliosthrough changes in expected returns. This, of course, requires that c2I(a wrlak) , oc. Should we be discouraged by this myriad of possible current account responses to a simple income shock? We think that this should not be the case. First, this is not an "almost anythinggoes" type ofresult. The theorytightlylinks currentaccount responses to measurable parameters, such as the volatility of production, the curvature of the aggregate productionfunction,the degree of -

CURRENT ACCOUNTS

1151

risk aversion, and the share of labor in income. This should eventually permit other researchers to calibrate models and perform quantitative analyses of current account movements generated by specificevents such as a temporaryimprovementin the terms oftrade or a transitorydrop in production.Second, and perhaps more important,the generalization of the intertemporal approach to the current account developed here moves us away fromwhat we think is an impasse in currentresearch. We make this point next. IV. TRADITIONAL INTERPRETATIONS OF THE DATA

The traditional rule states that countries use only foreign assets to smooth income shocks. According to this rule, these shocks generate equal changes in saving and the currentaccount. The natural way to test this implication is to estimate the followingregression: (12)

CAct=

a

+

Sct + acts

where c and t index countriesand years; CA and S are the current account and saving, both expressed as a share ofGNP; and u is an errortermthat captures other sources ofvariation in the current account that are not considered by the theoryand are assumed to be orthogonalto saving. Under the null that the traditionalrule is true, we should findthat 3 is one. We estimate equation (12) by ordinaryleast squares (OLS), using data on currentaccounts and saving for an unbalanced panel of thirteen industrial countries during the 1973-1995 period.11Since the saving variable is a residual, i.e., the sum of direct measures of the currentaccount and domestic investment, it is likely to contain substantial measurement error. Therefore, we also present instrumental variables (IV) estimates that mitigatethe attenuation bias in the OLS estimates.12 11. Although data on current accounts and saving are available for many more countriesand years, we restrictthe sample to those countriesforwhich data on stocks offoreignassets are also available, in orderto ensure that our tests ofthe traditional rule and the new rule are comparable. Appendix 2 provides a detailed descriptionofour data sources. 12. To correctthe attenuation bias due to measurement error,we use the rank of the dependent variable as the instrumental variable, as suggested by Greene [1993, Chapter 91.This variable satisfiesthe requirementthat it be correlatedwith saving. In the first-stageregressions this variable was highly significant.Moreover, if measurement errors are small relative to the size of saving, they are unlikelyto scramble the rankingofsaving, and, as a result,the rankingshould not be correlatedwiththese errors.

1152

QUARTERLYJOURNALOF ECONOMICS

Table I presents the results. The firsttwo columns show the OLS and IV estimates of I in a regressionthat pools all country/ year observations. The point estimates are 0.236 and 0.229, and we can comfortablyreject that 3 is equal to one. The estimate obtained fromthe pooled regression uses all the available variation in saving and current accounts. To determine whether this estimate is driven by persistent (between-countryvariation) or transitory(within-country variation) differencesin saving and the current account, we estimate a cross-sectional regression using time-averagesofall variables, and a fixed-effects panel regression to obtain two additional estimates of P. The estimates in columns (3) and (4) use only the between-countryvariation, while the estimates in columns (5) and (6) use only the within-country variation. Although the estimates vary across specificationsand range from0.182 to 0.269, we always overwhelminglyreject the null that 3 is equal to one. Overall, these results are quite negative for the view that the traditional rule provides a good descriptionofthe data. This should come as no surprise,since we have simply confirmedthat the Feldstein-Horioka findingalso applies in our sample. The traditional rule followsfroma view ofthe world in which there are no unexploited trading opportunities and financial markets play an important role in eliminating them. These assumptions are formallyembedded in equation (10), which, in the absence of investmentrisk, states that returns are equalized across countries; i.e., wr= p. Provided that this equation holds, income shocks cannot affect the domestic capital stock and thereforeinvestment.A firstset of explanations ofthe FeldsteinHorioka findingare based on the notion that equation (10) is a poor descriptionofinternationalfinancialmarkets,and we should drop it fromour models. A second set ofexplanations tries instead to modifythis equation so as to reconcile the theory with the data.13

Why would equation (10) fail? Perhaps financial markets are not well integrated in the sense that there are unexploited arbitrage opportunities.For example, due to asymmetricinformation problems or the existence of sovereign risk, debtor countries mightface binding constraintson how much they can borrow,as creditor countries might find it in their interest to restrain 13. Once again, see Feldstein and Bachetta [1991], Tesar [1991], and Obstfeld and Rogoff[19951 who survey proposed explanations for the Feldstein-Horioka finding.

1153

CURRENT ACCOUNTS 0c

oc

?

to o -

C( V]

c

0 8

80

0 u? to.0SU cq

av

c t

0

o

t-

o

O

'Rd (M

0

ooe>o

0~~~~~

o0

oc

'I eto6- -6666

o

co; r;~~~~~~~~~c

av

_=

o

a

ce

Z l|

Q

Ev

j

CI,~~~~~~~~~~~~~~~~~oC c; | as~~0c; c | _ s to04RoLC)U"o

666b.66 Ccq ? ~~~~~~~~~~~C)

CQ

g So _ e;

v0

o

;2E -

*>0c a>;

CO

w)

L

0~~~

c; c; c

$;

pq

0~~~~

0~~~~~~0O

~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~1 X

~

C~~~~0

c

00

0w

0~~~~~~0(

O|

6

LO

, w

C

i'

2

@ tcS:c

X

rn

i?

0~~~~~~~~~~~~~~~~~~~~~~Q

-

0 C1 0c;

c;

00

C

~~~~~~~~~~0 U

U)0

1~~~4 Q)

00

0

~ ~

6

C12~~~~~cqC

>

0 CO

~~~~~ 66

6

0~~~~~~~~~~ 6~~~~~~~~~~~~~~~~~~~~~Q

C~~~~~~~~~~~~~~~~ 0 :U$ ~~~~~ CI~~~~~~~~~~~ ~5 3

1154

QUARTERLYJOURNALOF ECONOMICS

themselves from lending too much. To the extent that income shocks have little or no effectson these constraints, countries would invest these shocks at home despite the ensuing fall in expected returns. Equation (10) would also fail if financial markets were not needed to eliminate arbitrage opportunities. A central idea of the Heckscher-Ohlin theory is that commodity trade can equalize factor returns across countries even in the absence of capital flows. This is Samuelson's factor-priceequalization theorem. If commodity trade already eliminates returndifferentials,small costs ofinternationalfinancialtransactions would induce countriesto invest income shocks at home. While the notion that asymmetric information,sovereign risk, and commodity trade might play an important role in shaping existingpatterns ofcapital flowsis appealing, this line of research has not yet generated strongempirical predictionsthat can be broughtto the data. In any case, the findingthat changes in wealth lead to changes in the current account that are proportional to the share offoreignassets in total assets (see Figure IV and Section V) poses a new challenge to this set ofexplanations. It is not immediate to see why we should observe this empirical regularity in a world in which there are unexploited trading opportunitiesor, alternatively,they are eliminated without the help offinancialmarkets. How mightwe rescue equation (10)? Perhaps the theoryis not wrong,but instead we are just tryingto test an overlysimplistic version of it in which countries receive only idiosyncratic or country-specificincome shocks. Consider the possibility that countries receive common or global income shocks. Since the world is a closed economy,these shocks would affectsavings and investment in all countries. Consider also the possibility that countries receive persistent shocks to their rates of population and productivitygrowth. Standard growth models show how these shocks affectthe investmentrate that is required to keep the marginal product of capital constant. Modigliani's life-cycle theoryof saving predicts that these shocks also affectaggregate saving, as the savings ofyoungergenerations increase relative to the dissavings of the older ones. Since commonincome shocks or shocks to the rates ofpopulation and productivitygrowthsimultaneously affectsaving and investment, the error term uct in a regression such as (12) is negatively correlated with saving and

0.35 y =1.0378x + 0.001 1 R' 0.9775

03 0.25 0.2 0.15 i

45-Degree Line

0.1i 0.05 1 -0.15

0.05

O

0.15

0.25

*;,fp.5

0.35 X(t-1)

-0.15

0 35 X

0.3

y I

025.

11892x + 0 0033 R2 = 0 871

02: 0 15 |

O11

.

0 05

-0 15

1

*

Line

*/>~t45-Degree

+^

*

005

0 15

025

035 X(t-5)

I;

-01 -0 15

_

0.35 0 35 i

y = 1.4647x + 0.0062 R2

0 30

= 0.7636

0.251 0.2 0.15 45-Degree Line

0.1 0.05

-0.15

.

0.05 - 'J

0.15

025

0.35 X(t.10)

-0.15

FIGURE IV The Persistence ofCountryPortfolios This figureplots the share of foreignassets in total assets (X(t)) on the vertical axis, against the share offoreignassets in total assets lagged one year (X(t - 1)), fiveyears (X(t - 5)) and ten years (X(t - 10)). See Appendix 2 fordata definitions and sources.

1156

QUARTERLYJOURNALOF ECONOMICS

the estimate of 3 is biased toward zero. A low estimate of this coefficienttherefore does not warrant the conclusion that we should abandon the idea that unexploited trading opportunities are eliminated by financialmarkets. It just means that we should treat the traditional rule as a conditionalresult. This line of argument has been popular among economists because, besides being plausible, it generates a strong empirical prediction:if we controlforthese additional shocks in equation (12), we should find that 1 = 1. We test this prediction by reestimating 1 using time dummies and measures of population and productivitygrowthas controlvariables. Columns (7) to (12) of Table I show the results. Consistent with the results of other researchers,we findthat the estimates of 13increase, but they are still much lower than one. Perhaps there are other missing variables that are responsible forthe Feldstein-Horioka finding, but they have not been found yet. We agree with Feldstein and Bachetta [1991, p. 319] that the results obtained so farplace "( ...) on the defendersofthat hypothesisthe burden ofidentifyingsuch commoncausal factors."'14 V. A

NEW INTERPRETATION OF THE DATA

We propose an alternative empirical strategy that rescues equation (10) by placing the risk premium at center stage. If investmentrisk is not negligible,this equation no longer reduces to the simple statement that expected returns are equalized across countries, and a whole new range of theoretical possibilities arises. In particular, equation (10) shows that the current account response to a simple income shock can be positive or negative and depends on a number of country characteristics

14. There are also papers that assume the traditionalrule is correct,and then use the current account to indirectly test the permanent-income theory of consumption.Sheffrinand Woo [19901,Otto [19921,Ghosh [19951,and Ghosh and Ostry [19951 assume that investment follows an exogenously given process and compute "permanent"or net present values ofincome net of investment.They use these series to test whether the current account tends to be positive (negative) when income net of investmentis above its net present value using the technique developed by Campbell [1987] to test the permanent-incometheoryof consumption. An innovation in this line of research is Glick and Rogoff[1995]. This paper uses a model with adjustment costs to capital and persistentshocks to productivity to derivethe "permanent"or net presentvalue ofincomenetofinvestment,and also distinguishes between global and country-specificshocks. Remember that the theoryis onlyconcernedwith the latter (see footnote5).

CURRENT ACCOUNTS

1157

including the expected returns and volatilityof production('rr,u), the size of labor income (X), the attitudes toward risk (y), and the level of wealth (a). Recognizing that investors demand a risk premiumthereforeshows how special the traditionalrule is. More important, recognizing that investors demand a risk premium also suggests another special case that turns out to be more relevant empirically:the new rule. The new rule states that countries smooth income shocks with a combinationofdomestic and foreignassets that resembles their portfolio.According to this rule, these shocks generate changes in the currentaccount that are equal to saving times the share of foreignassets in total assets. To test this prediction,let Xctbe the share offoreignassets in total assets and consider this regression: (13)

CAct= (x +

XCt*Sct + uct

Under the null that the new rule is true, we should findthat the estimate of 3 is one. To estimate this equation, we use additional data on foreignasset positions of countries (see Appendix 2 for details on how we constructthis variable). Since stocks offoreign assets are measured with substantial error, there is now an additional reason to use an instrumental variable procedure to estimate 3. The results are presented in Table II. The differentspecifications correspondto those used in Table I forthe traditional rule. The OLS estimates are generally smaller than the IV estimates, suggesting that the formerare contaminated by attenuation bias due to measurement errorin foreignassets, and so we focuson the latter. The pooled regression generates an estimate of 13equal to 1.034. Columns (4) and (6) show that the between and within estimators are also very close to one, indicating that the pooled estimate is driven by both cross-countryand within-country variation simultaneously.Columns (8), (10), and (12) confirmthat these results hold after controllingfor year effects,population, and productivitygrowth.In none of the specifications(including both the OLS and IV estimates) can we reject the null that P is equal to one at the 5 percent significance level. Overall, this evidence supports the view that the new rule provides a good descriptionofthe data. This conclusion is reinforcedif we directlyexamine country

QUARTERLY JOURNAL OF ECONOMICS

1158

o oqUCf tC8>e)Cs N~~~~~~~~~~~~~~c

cq i>

C,

00000t

Lo

m

C

C

U-) to

t-mo

co

0

~

o

3

$

u

0

0 o- oq o oc'qt oq Cl0~~~'~~t~~e~ o~ oo o4C.C.C ;C 66,66ooo '-~~~~~~~~~~~~~~~~~~~~~~~~~0~~~~~~~~

oo

t- -C t? F)

(3

C

00 n coo

t0

I

6 00 O

o

00 CI

1CI2

0 O CIO 0oOc8> co C t0 0 0 I

CItO0

co

;d0

>

~~~~~~

ooO cO

O; ooi)

i

3

00

t U:

0o

.l 6

~~~~~~~~~~~~~~~L q CD CD (3 d3 "t'-'I Z~~~~~~~~~~~~~~~~~~~~~0 X

St--

m

o

cn

m Q Q

oNo ]U

m

C 6t

>0~~~~~~~~~~~~~c t

C0

~~~~~~ 4

-

1

co <

o

c

En

w

Oc

-0 1 g M o0 s: ?~~~>

t

O

00 o

Q
co

0O

(x) CO

C sO 0voD

r-

s,*

X~~~~~~~~~~~~~~~~~~co ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~~c ~~~Cl~~~~~~C ~~~~~00Cl ~~~~g

CIOt-

a0

t-

00

:I

U-

0 66 r-l I'lCOc11

~~~U2 -

,3v t~~co

CIO.

( CI coXQ

0

03

j

o

C; '-4 C;

0~~~~~~;

a

C; C3C 3 Cdcq C0 Cl 0 ~0

:'-

X

C"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 00~~~t CO

(3)

1~~~~~~~4 co C) 8

~ c;

cq m 0

~

~ -

~ 0

a)

0

~ CI O

CO-

~~~~ c~~~~~~~c I)

:.

t

~

~

w .-c-

C12+0

_

1~~4

..

co

cio

1

> co

CURRENT ACCOUNTS

1159

portfolios.Figure IV plots the currentportfolioshare against its one-year,five-year,and ten-yearlag. The firstpanel shows that the year-to-yearvariation in portfolioshares is negligible. Over longer periods, however, there is a tendency for the share of foreignassets to increase in absolute value over time,as shown by the gradual counterclockwiserotationofthe regressionline as the lag length increases. Figure IV also shows why the new rule is consistent with the Feldstein-Horioka finding:there is a strong home bias in countryportfolios.In our sample, the mean absolute value forthe share offoreignassets in total assets is 5.5 percent. Under the new rule, the fact that all countries have such low shares in absolute value implies that they invest most of their income shocks at home. Although the evidence seems to support the new rule, one should view the results in Table II with a healthy dose of skepticism.Thus far,we have imposed the restrictionthat 1 is the same across countries and over time. In Table III we relax this restrictionand present estimates of 1 foreach cross-section(21 years) and time-series (13 countries) in our sample. The crosssectional estimates average 1.145 with a relativelylow standard deviation of 0.444, and we do not reject the restrictionthat they are all equal. For only 2 (1,0) out of 21 cross-sectionalestimates, can we reject the null that 1 is one at the 10 percent (5 percent,1 percent) significance level. Although the time-series estimates average 1.087, they have a relativelylarge standard deviation of 1.348, and we do reject that they are equal. For 6 (5,4) out of 13 time-seriesestimates, we can also rejectthe null that 1 is equal to one at the 10 percent (5 percent, 1 percent) significance level. While the cross-sectional estimates are fairlystable and consistent with the new rule, the time-series estimates vary substantially in ways that are inconsistentwiththe new rule. To some extent,this is not surprising.The new rule has been developed in a model with a perfectlyelastic supply ofcapital and full informationabout the nature of shocks. With these assumptions, the model should be interpretedas a model ofthe long-run equilibrium trajectory.Since the time-seriesestimates reflectthe transitory variation in saving and the current account, the discrepancy between time-series and cross-sectional estimates might indicate short-run departures from the new rule. One would expect that the path countries follow to return to their

TABLE III CROSS-COUNTRY AND TIMES SERIES EVIDENCE FOR THE NEW RULE

Coefficient

Standard error

P-value forHo: A= 1

Number of observations

Cross-sectional estimates byyear 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995

2.051 1.519 0.866 0.557 0.662 1.193 2.073 1.640 1.457 1.057 1.364 1.167 0.990 0.954 1.285 0.922 1.049 1.265 0.970 0.667 0.338

Average Standard deviation

1.145 0.444

0.694 0.258 0.465 0.641 0.455 0.341 0.598 0.428 0.259 0.461 0.420 0.477 0.371 0.405 0.386 0.240 0.243 0.278 0.265 0.322 0.429

0.130 0.045 0.772 0.490 0.458 0.571 0.073 0.135 0.078 0.901 0.386 0.727 0.979 0.910 0.459 0.745 0.839 0.341 0.909 0.302 0.123

9 9 9 9 10 10 10 12 12 12 12 12 12 12 13 12 12 12 12 12 12

Time-series estimates by country Australia Austria Canada Germany Spain Finland France United Kingdom Italy Japan Netherlands Sweden United States Average Standard deviation

2.384 3.580 0.596 2.373 0.072 1.192 0.582 -1.442 1.038 1.543 0.204 2.289 -0.286

0.690 0.900 0.641 0.352 0.509 1.470 0.257 0.808 1.097 0.332 0.163 1.022 1.910

0.045 0.004 0.528 0.000 0.068 0.896 0.103 0.003 0.972 0.102 0.000 0.207 0.501

23 23 23 15 23 21 7 23 23 17 14 14 23

1.087 1.348

This table reportsthe results ofestimating the second equation in Table II. The upper panel reportsthe results of cross-countryestimates for each year, and the lower panel reports the results of time-series estimates foreach country.Standard errors are correctedforheteroskedasticity.See Appendix 2 forvariable definitionsand data sources.

1161

CURRENT ACCOUNTS

long-run trajectory after a shock depends on the nature of adjustment costs to investmentand the process bywhich information comes to the economy.We study the nature of this path in Kraay and Ventura [1999]. One should also keep in mind that our sample includes only thirteen industrial countries. These countries have the most advanced financial markets in the world, and the ability of financial markets to eliminate unexploited trading opportunities might be considerably less in other samples of countries. For instance, it seems reasonable to predict that a theoryof capital flowsbased on well-functioningfinancialmarkets mightbe a poor approximation to the reality of many developing countries that have onlyrudimentaryfinancialmarkets. Despite these caveats, we regard the rule that countries smooth income shocks through a combination of assets that resembles their portfoliosas an attractive new benchmark from which to think about current account issues. It has a solid theoreticalgroundingand providesa reasonablygood firstapproximation to the behavior of industrial countries. Moreover,since it reinterpretsthe Feldstein-Horioka findingas the flowversion of the home bias in countryportfolios,it also unifies two central problems in international finance. The next step, of course, is to solve them.

APPENDIX 1: SOLUTION DETAILS

This appendix solves the extended model of Section III with variable-risk aversion and labor income. The simpler model in Section I is just a special case in which X = y = 0. Consider the problemof a representativeconsumer who chooses c, k, and k* so as to maximize (9) subject to the budget constraint: da = [r *k + r* k* + p *(a - k - k*)?+ + k-

j

- c] *dt

- do + k*

.(J*

.

dw*,

and the law ofmotionof ra: drp=p

dt+X.

dw +x*

dw*.

In equilibrium,-a,,u,X,and X*mightbe functionsofthe aggregate stocks of domestic capital, but the representative consumer is infinitesimaland does not take into consideration how his/her

1162

QUARTERLY JOURNAL OF ECONOMICS

choices affectaggregates. The Bellman equation ofthe representative consumer's problemis V=

max Iln(c-y)+



?-

aV

k[+k r r*?k*

aa

a2V 1 aV + X - c] +-? 2 * - * [k2* (j2 ,+ a~r aa2 2

+ 2 a k u -k* (*.q] +

a2v aaair

*~[k

-u

+

?

+ k*2

-k -k*)

*2

a2v 1 2 . - . [x2 + X*2 + 2 XX XX* X q]

adr22

loq)+ k* a

(x + x*

p(a

(x a -q + X*))],

conditionsassociated with this Bellman equaand the first-order tion are

0= O

aV -

1

aV

_~c - w aa

a2V

(u* * Aq) 2*r *(k r + k* aa2

- (,i- p) +

aa

+

o

aV aa

-p)

?

a2V aa2

u* (k*(*

?

+

a2V aaair

k. a2v

aaa-r

*(u(X*

+ X* * )

q) *~(r*

* (X * to + X*).

It is straightforward to verifythat V = a-1 In {a + (X - y)/pl+ f(,) solves the Bellman equation. Using this value functionand the first-order conditions,it followsdirectlythat all the equations in the paper are special cases ofthis model. APPENDIX 2: VARIABLE DEFINITIONS

AND DATA SOURCES

This appendix describes the data used in this paper. The completedata set is available fromthe authors upon request. Data on stocks offoreignassets are drawn frominternational investment positions (IIPs) reported in the International Mone-

CURRENT ACCOUNTS

1163

tary Fund's Balance ofPayments Statistics Yearbook, Revisions 4 and 5. This source reports annual estimates of stocks of foreign assets and liabilities for most OECD countries, and a few nonOECD countries,in currentU. S. dollars. For some countriesand forsome items, these stocks incorporatevarious adjustments for changes in valuation and exchange rates, using methodologies that vary across sources. We measure a country's holdings of foreign capital (k*) as direct and portfolioequity investment abroad (Revision 5 lines 8505 and 8610), and its net lending abroad (a - k - k*) as the net HIP balance (line 8995) less net directinvestment(line 8505 - line 8555) less net portfolioequity investment (line 8610 - line 8660). Data on these variables are generally available since the early 1980s under the Revision 5 presentation. For some countries, data for earlier years are available under the Revision 4 presentation of the IIP. For these countrieswe extend the Revision 5 data backward using changes in the Revision 4 stocks to the earliest available year. In particular, we use the Revision 4 lines 3L.V4 and 6D1V4 to extend outward direct and portfolioequity investment, line ... V4 to extend the net IIP, and lines 3L.V4-3Y.V4 and 6D1.V4 (6V1V4 + 6S1V4) to extend net directand portfolioequity investment. Due to data revisions undertaken for the Revision 5 and some minorconceptual differencesbetween the Revision 5 financial account and the Revision 4 capital account, there are some small discrepancies between the Revision 4 and Revision 5 figures forsome countries in some years where the two sources overlap. We then restrictthe sample to the thirteen OECD countries for which the most complete data are available. To make the panel somewhat more balanced, we exclude the handful ofobservations available prior to 1973. The sample of countries and the time series coverage by source are indicated in Table IV. We measure a country'sholdings ofdomesticcapital (k) as the gross domestic capital stock, less inward direct and portfolio equity investment.We constructthe gross domestic capital stock in currentU. S. dollars by cumulating gross domestic investment in current U. S. dollars fromthe World Bank's Global DevelopmentIndicators(WBGDI) (NY.GDI.MKTP.CD), assuming a depreciation rate of 4 percent per year, and in each year revaluing the previous year's stock using the U. S. gross domestic investment deflator(WBGDI, NY.GDI.MKTP.CD/NY.GDP.MKTP.KD). We estimate the initial capital stock in 1965 using the average capitaloutput ratio over the period 1960-1965 reported in Nehru and

1164

QUARTERLYJOURNALOF ECONOMICS TABLE IV COVERAGE OF IIP DATA

Country

Availability

Australia Austria Canada Germany Spain Finland France United Kingdom Italy Japan Netherlands Sweden United States

1973-1997 1973-1996 1960-1997 1975-1989 1972-1997 1975-1997 1989-1996 1973-1997 1972-1997 1979-1997 1982-1996 1982-1996 1972-1997

Data from BOPS4 1973-1985 1973-1979 1975-1979 1972-1980 1973-1979 1979 1972-1979

Dareshwar [1993], multiplied by GDP in current U. S. dollars (WBGDI, NY.GDP.MKTP.CD). We use the flowmeasure of the currentaccount reportedin the Balance of Payments Statistics Yearbook in current U. S. dollars (line 4993), and we measure gross national saving residually as the sum of the current account plus gross domestic investmentin currentU. S. dollars fromthe WBGDI. Our results are qualitatively verysimilar ifwe instead use directmeasures of saving fromthe national income accounts.15 The controlvariables in Tables I and II are constructed as follows.Population growthis the growthin the midyear population (WBGDI, SP.POP.TOTL). The Solow residual is the annual growth rate of GDP in constant 1995 U. S. dollars (WBGDI, NT.GDP.MKTP.KD), less the share of wages in GDP times the growth rate of total civilian employment(OECD Labour Force Statistics, Table 6), less one minus this share times the growth rate of the gross domestic capital stock in constant U. S. dollars. The share of wages in GDP is measured as the average over 1960-1993 ofcompensation ofemployees divided by GDP (OECD National Accounts,MOCOM/MOGDPE). We constructthe constantprice capital stock by cumulating constant 1995 U. S. dollar gross 15. This is not true as an identity,since all the countries in our sample provide direct estimates of saving in the national income accounts (see SchmidtHebbel and Serven [1997, Table Al]).

CURRENT ACCOUNTS

1165

domestic investment flows as above, using the Nehru and Dareshwar [1993] estimate of the capital-output ratio times the GDP in constant 1995 U. S. dollars as a base in 1965. THE WORLD BANK MASSACHUSETTS INSTITUTE OF TECHNOLOGY,CEPR, AND NBER

REFERENCES Bodie, Zvi, RobertMerton,and William Samuelson, "Labor Supply Flexibilityand PortfolioChoice in a Life Cycle Model," Journal of Economic Dynamics and Control,XVI (1992),427-449. Campbell, John,"Does Savings Anticipate Declining Labor Income? An Alternative Test of the Permanent-Income Hypothesis," Econometrica, LV (1987), 1249-1273. Dornbusch, Rudiger, "Real Interest Rates, Home Goods, and Optimal External Borrowing,"Journal ofPolitical Economy,XCI (1983), 141-153. Feldstein, Martin, and Philippe Bachetta, "National Saving and International Investment,"in B. D. Bernheim and J. B. Shoven, eds., National Savings and Economic Performance(Chicago: UniversityofChicago Press, 1991). Feldstein, Martin, and Charles Horioka, "Domestic Savings and International Capital Flows,"Economic Journal, XC (1980), 314-329. French, Kenneth, and James Poterba, "InvestorDiversificationand International Equity Markets,"AmericanEconomic Review,LXXXI (1991), 222-226. Ghosh, Atish, "Capital MobilityAmongst the Major Industrialized Countries: Too Little or Too Much,"Economic Journal, CV (1995), 107-128. Ghosh,Atish,and Jonathan Ostry,"The CurrentAccountin Developing Countries: A Perspective from the Consumption-SmoothingApproach," World Bank Economic Review, IX (1995), 305-333. Glick, R., and K. Rogoff,"Global versus Country-specificProductivityShocks and the Current Account,"Journal of Monetary Economics, XXXV (1995), 159192. Greene, William, Econometric Analysis (Englewood Cliffs, NJ: Prentice-Hall, 1993). Jones, Larry, and Rodolfo Manuelli, "A Convex Model of Equilibrium Growth: Theoryand Policy Implications,"Journal ofPolitical Economy,XCVIII (1990), 1008-1038. Kraay, Aart, and Jaume Ventura, "Foreign Demand Shocks," manuscript, the WorldBank and the Massachusetts InstituteofTechnology,1999. Lewis, Karen, "Trying to Explain Home Bias in Equities and Consumption," Journal ofEconomic Literature,XXXVII (1999), 571-608. Matsuyama, Kiminori, "CurrentAccount Dynamics in a Finite Horizon Model," Journal ofInternationalEconomics, XXIII (1987), 299-313. Merton, Robert, "Optimum Consumption and PortfolioRules in a ContinuousTime Model,"Journal ofEconomic Theory,III (1971), 373-413. , Continuous-TimeFinance (Cambridge, MA: Basil Blackwell, 1995). Nehru, Vikram, and Ashok Dareshwar, "A New Database on Physical Capital Stock: Sources, Methodology and Results," Revista de Analisis Economico, VIII (1993), 37-59. Obstfeld, Maurice, "Aggregate Spending and the Terms of Trade: Is There a Laursen-Metzler Effect?"Quarterly Journal of Economics, XC (1982), 251270. Obstfeld, Maurice, and Kenneth Rogoff,"The Intertemporal Approach to the CurrentAccount,"in Gene Grossman and Kenneth Rogoff,eds., Handbook of InternationalEconomics (Amsterdam,The Netherlands: Elsevier, 1995). Otto, Glenn, "Testing a Present-Value Model of the Current Account: Evidence fromU. S. and Canadian Time Series," Journal of International Money and Finance, XI (1992),414-430.

1166

QUARTERLYJOURNALOF ECONOMICS

Persson, Torsten,and Lars Svensson, "CurrentAccount Dynamics and the Terms of Trade: Harberger-Laursen-Metzler Two Generations Later," Journal of Political Economy,XCIII (1985), 43-65. Sachs, Jeffrey,"The Current Account and Macroeconomic Adjustment in the 1970s,"BrookingsPapers on Economic Activity,1 (1981), 201-268. ., "The CurrentAccountin the MacroeconomicAdjustmentProcess,"Scandinavian Journal ofEconomics, LXXXIV (1982), 147-159. Schmidt-Hebbel,Klaus, and Luis Serven, "Saving across The World: Puzzles and Policies,"the WorldBank, Discussion Paper No. 354, 1997. Sheffrin,Steven, and Wing Thye Woo, "Present Value Tests of an Intertemporal Model of the Current Account,"Journal of International Economics, XXIX (1990), 237-253. Svensson, Lars, and Assaf Razin, "The Terms of Trade and the CurrentAccount: The Harberger-Laursen-MetzlerEffect,"Journal of Political Economy, XCI (1983), 97-125. Tesar, Linda, "Savings, Investment and International Capital Flows," Journal of InternationalEconomics, XXXI (1991), 55-78. Tesar, Linda, and Ingrid Werner,"Home Bias and the Globalization ofSecurities," National Bureau ofEconomic Research WorkingPaper No. 4218, 1992. Ventura,Jaume, "Growthand Interdependence,"QuarterlyJournal ofEconomics, CXII (1997), 57-84.