INTEREST RATES, EXCHANGE RATES AND … RATES, EXCHANGE RATES AND PRESENT VALUE MODELS OF THE ... Just as individuals may adjust consumption and saving behaviour in ... Following Dornbusch

INTEREST RATES, EXCHANGE RATES AND PRESENTVALUE MODELS OF THE CURRENT ACCOUNT�

Paul R. Bergin and Steven M. Sheffrin

This paper develops a testable intertemporal model of the current account that allows forvariable interest rates and exchange rates. These additional variables are channels throughwhich external shocks may in¯uence the domestic current account. The restrictions from thetheoretical model are subjected to present value tests using quarterly data from three smallopen economies. The paper ®nds that including the interest rate and exchange rate improvesthe ®t of the intertemporal model over what was found in previous studies. The modelpredictions better replicate the volatility of current account data and better explain historicalepisodes of current account imbalance.

In theoretical research dealing with the current account, it has becomestandard practice to use intertemporal models. The intertemporal approach tothe current account, in its simplest form, focuses on the optimal savingdecision of a representative household as it smooths consumption. Forexample, considering a small open economy experiencing a temporary fall inoutput, the country would be expected to smooth consumption by borrowingin world capital markets and thereby run a current account de®cit. This basicintertemporal model has been extended in many directions in the theoreticalliterature, to include investment, variable interest rates, nontraded goods, andeven monetary policy.1

Empirical work on the intertemporal approach to the current account haslagged behind the theoretical literature. Simple intertemporal models focus-ing on consumption smoothing have been tested empirically in Sheffrin andWoo (1990a,b), Otto (1992), Milbourne and Otto (1992), Otto and Voss(1995), and Ghosh (1995). Most of these studies adapt present value testsdeveloped by Campbell (1987) and Campbell and Shiller (1987), originallydeveloped to test consumption theory.

Present value tests are an approach that makes full use of the model'sstructure to derive testable hypotheses. The simple intertemporal modelimplies that a country's current account surplus should be equal to the presentvalue of expected future declines in output, net of investment and governmentpurchases. A vector autoregression involving the current account and outputcan be used to compute a forecast of this present value, conditional onhouseholds' information. According to the theory, the VAR forecast of thispresent value should be equal to the current account. This implication can beevaluated formally using a Wald statistic, or informally by comparing thehistorical movements of the current account with those of the prediction from

The Economic Journal, 110 (April), 535±558. # Royal Economic Society 2000. Published by BlackwellPublishers, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA.

[ 535 ]

� We would like to thank our colleagues Kevin Hoover and Wing Woo, as well as three anonymousreferees and seminar participants at U. C. Berkeley, U. C. Davis, UCLA, and U. C. Santa Cruz forvaluable comments.

1 See Obstfeld and Rogoff (1996) for a useful summary of this extensive literature.

the VAR. Similar tests have been useful in studies of consumption behaviourand government de®cits.2

To date, the results of such tests applied to the current account are mixed atbest. While the simple intertemporal current account model has often beenfound to work fairly well for large countries, it ironically tends to fail for manysmall open economies. This is surprising, inasmuch as one would expect theassumptions of the theory to be most appropriate in these cases. Small openeconomies can borrow from the rest of the world without inducing offsettingchanges in other variables such as the equilibrium world real interest rate.

A likely explanation for this failure is that small economies may be affectedstrongly by external shocks, a factor not considered in the simple version ofthe intertemporal model tested previously. To explain the current accountbehaviour of small open economies, it may be important not only to modelshocks to domestic output, but also shocks arising in the country's largerneighbours or the world in general. These external shocks will generally affectthe small open economy via movements in the interest rate or exchange rate.Just as individuals may adjust consumption and saving behaviour in responseto movements in real interest rates, countries may also adjust their currentaccount in response to movements of the real interest rate in world capitalmarkets.

Furthermore, Dornbusch (1983) has demonstrated that an anticipated risein the relative price of internationally traded goods can raise the cost ofborrowing from the rest of the world, when interest is paid in units of thesegoods. As a result, changes in the real exchange rate can induce substitution inconsumption between periods, and it thus can have intertemporal effects on acountry's current account similar to those of changes in the interest rate. Inaddition to these intertemporal effects, exchange rate changes of course canalso have more standard intratemporal effects, by inducing substitution be-tween internationally-traded goods and nontraded goods at a point in time.

This paper expands earlier present value tests of the current account toallow for variations in the interest rate and exchange rate. The paper derives atestable implication of an intertemporal model that allows for time-varyinginterest rates as well as a distinction between tradable and nontradable goods.This testable implication is then subjected to present value tests, usingquarterly data from three small open economies that have proved problematicin past studies: Australia, Canada, and the United Kingdom. In two of thethree countries it is found that including the interest rate and exchange ratesigni®cantly improves the ®t of the model over a benchmark model whichexcludes them. These extensions allow the model prediction to match thevolatility of current account data better, and they improve the model's abilityto explain historical episodes of current account imbalance. However, theresults suggest that the intratemporal elements of the theory, rather than theintertemporal elements, are primarily responsible for improving the ®t.

The next section of the paper outlines a theory of the current account for

2 For example, see Huang and Lin (1993).

536 [ A P R I LT H E E C O N O M I C J O U R N A L

# Royal Economic Society 2000

variable interest rates and exchange rates and develops the econometricframework. Section 2 discusses the data and parameter values. Section 3presents results from the present value tests. Our conclusion highlights someadditional issues in interpreting and extending intertemporal models of thecurrent account.

1. Theory and Econometric Methods

Our extensions are based on log-linearisation of the intertemporal budgetconstraint, following the lead of Campbell and Mankiw (1989) in their workon consumption and Huang and Lin (1993) in their work on ®scal de®cits.Our log-linear intertemporal budget constraint for the open economy is com-bined with the appropriate Euler equation to derive a fundamental testableimplication involving a transformation of the current account. This currentaccount condition can be subjected to present value tests.

Following Dornbusch (1983), we consider a small country producing tradedand nontraded goods. The country can borrow and lend with the rest of theworld at a time-varying real interest rate. The representative household solvesan intertemporal maximisation problem, choosing a path of consumption anddebt that maximises discounted lifetime utility:

max E0P1t�0

â t U (CTt , CNt) (1)

s:t Yt ÿ (CTt � PtCNt)ÿ I t ÿ Gt � rtB tÿ1 � Bt ÿ Btÿ1, (2)

where U (CTt , CNt) � 1

1ÿ ó(C a

Tt C1ÿaNt )1ÿó

ó . 0, ó 6� 1, 0 , a , 1:

Consumption of the traded good is denoted CTt , and consumption of thenontraded good is CNt . Yt denotes the value of current output, I t is investmentexpenditure, and Gt is government spending on goods and services, all meas-ured in terms of traded goods. The relative price of home nontraded goods interms of traded goods is denoted Pt . The stock of external assets at thebeginning of the period is denoted Bt . Finally, rt is the net world real interestrate in terms of traded goods, which may vary exogenously over time. The left-hand side of this budget constraint may be interpreted as the current account.We may express total consumption expenditure in terms of traded goods asCt � CTt � PtCNt .

Appendix A derives the ®rst-order conditions for this problem and usesthem to derive the following optimal consumption pro®le:

1 � E t âã(1� r t�1)ãCt

C t�1

� �Pt

P t�1

� �(ãÿ1)(1ÿa)" #

: (3)

In this condition ã � 1=ó is the intertemporal elasticity of substitution. Thisderivation generally follows the well-known methods in Dornbusch (1983) and

2000] 537M O D E L S O F T H E C U R R E N T A C C O U N T


Obstfeld and Rogoff (1996). This involves de®ning a Cobb-Douglas consump-tion index corresponding to the utility function, and then ®nding a relatedprice index. These are used to transform the optimisation problem into a forminvolving a single composite good. The resulting intertemporal Euler equationthen can be written in terms of this composite good and price index, oralternatively as condition (3) above, in terms of total consumption expenditureand the relative price of nontraded goods.

Assuming joint log normality and constant variances and covariances, condi-tion (3) may be written in logs:

E tÄc t�1 � ãE t r�t�1, (4)

where r� is a consumption-based real interest rate de®ned by:

r�t � rt � 1ÿ ã

ã(1ÿ a)

� �Ä pt � constant: (5)

We de®ne Äc t�1 � log C t�1 ÿ log Ct and Äpt�1 � log P t�1 ÿ log Pt . For theworld real interest rate (de®ned in terms of traded goods) we use the approxi-mation: log(1� rt) � rt .3 The constant term at the end of the expression willdrop out of the empirical model when we later demean the consumption-based real interest rate using (5).

This condition characterises how the optimal consumption pro®le is in¯u-enced by the consumption-based real interest rate, r�, which re¯ects both theinterest rate r and the change in the relative price of nontraded goods, p.Previous empirical studies of the intertemporal approach to the currentaccount have not allowed for these variables.4 Such models imply a consump-tion pro®le where the expected change in consumption is zero; householdsalways try to smooth consumption over time by borrowing and lending withthe rest of the world. In contrast, the representative consumer here may beinduced to alter the consumption pro®le and ùnsmooth' consumption, in theface of changes in the terms of such borrowing and lending. First consider theinterest rate. An increase in the conventional real interest rate, r , makescurrent consumption more expensive in terms of future consumption fore-gone, and induces substitution toward future consumption with elasticity ã.

A similar intertemporal effect can result from a change in the relative priceof nontraded goods. If the price of traded goods is temporarily low andexpected to rise, then the future repayment of a loan in traded goods has ahigher cost in terms of the consumption bundle than in terms of traded goodsalone. Thus the consumption-based interest rate r� rises above the conven-tional interest rate r , and lowers the current total consumption expenditure byelasticity ã(1ÿ a).

3 Campbell (1998) discusses evidence that the variance and covariance terms in the constant term of(5) may be time-varying. Existing evidence suggests this is mainly a problem for frequencies higher thanthat used in this study. Campbell also suggests it may be important to consider the case of time-varyingrisk aversion. While such extensions are potentially useful, they are beyond the scope of the presentpaper.

4 At the time of writing this paper, Fahrion (1997) concurrently has developed a present valuemodel, which allows for variable interest rates; it does not consider nontraded goods.



In addition to this intertemporal substitution, a change in the relative priceof nontraded goods also induces intratemporal substitution. Again if the priceof traded goods is temporarily low relative to nontraded goods, households willsubstitute toward traded goods by the intratemporal elasticity, which is unityunder a Cobb-Douglas speci®cation. This raises total current consumptionexpenditure by elasticity (1ÿ a). This intratemporal effect will be dominatedby the intertemporal effect if the intertemporal elasticity, ã, is greater thanunity.

The representative agent optimisation problem above entails an intertem-poral budget constraint. De®ne Rs as the market discount factor for date sconsumption, so that

Rs � 1Ys

j�1

(1� r j)

:

Using the budget constraint of the optimisation problem (2), the currentaccount (CA) may be written:

CAt � Yt ÿ (CTt � PtCNt)ÿ I t ÿ Gt � rtB tÿ1 (6)

or as

CAt � NOt ÿ Ct � rtB tÿ1, (7)

where we de®ne net output as follows: NOt � Yt ÿ I t ÿ Gt . Summing over allperiods of the in®nite horizon, and imposing the following transversalitycondition:

limt!1E0(RtBt) � 0, (8)

we may write an intertemporal budget constraint:P1t�0

E0(RtCt) �P1t�0

E0(RtNOt)� B0, (9)

where B0 is initial net foreign assets. We log linearise this intertemporal budgetconstraint, following Campbell and Mankiw (1989) and Huang and Lin(1993). We show in Appendix B that equation (9) may be log linearised asfollows:

ÿP1t�1

â t Änot ÿ Äct

Ùÿ 1ÿ 1

Ù

� �rt

� �� no0 ÿ c0

Ù� 1ÿ 1

Ù

� �b0, (10)

where lower case letters represent the logs of upper case counterparts, exceptin the case of the world real interest rate, where we again use the approxima-tion that log (1� rt) � rt . Here Ù is a constant slightly less than one,Ù � 1ÿ B=

P1t�0 RtCt , where B is the steady state value of net foreign assets.

Next, take expectations of (10) above and combine it with the Euler (4) towrite:



ÿE t

P1i�1

âi Äno t�i ÿ ã

Ùr�t�i ÿ 1ÿ 1

Ù

� �rt

� �� not ÿ ct

Ù� 1ÿ 1

Ù

� �bt : (11)

The right side of this equation is similar to the de®nition of the currentaccount in (6), except that its components are in log terms. We label thistransformed representation of the current account as CA�. We will follow theconvention of choosing the steady state around which we linearise to be theone in which net foreign assets are zero. In this case, Ù � 1 and the conditionabove may be written:5

CA�t � ÿE t

P1i�1

âi(Äno t�i ÿ ãr�t�i), (12)

where CA�t � not ÿ ct : (13)

This condition says that if net output is expected to fall, the current accountwill rise as the representative household smooths consumption. But thecondition also says that aside from any change in domestic output, a rise in theconsumption-based interest rate will raise the current account by inducing therepresentative household to lower consumption below its smoothed level. Forcomparison, we also test a simpler version of the intertemporal model, wherethe consumption-based interest rate is assumed to be constant, and conse-quently only the ®rst of the two effects described above will occur. Thisamounts to testing a condition similar to (12) above, where the second term inthe brackets is not present.

This restriction in (12) is tested using the approach of Sheffrin and Woo(1990b), augmented to consider the additional variable, r�. To test therestriction that the current account depends on expected future values of netoutput and interest rate, we ®rst must have proxies for these two sets ofexpected values. Under the null hypothesis of (12), the current account itselfshould incorporate all of the consumers' information on future values of thelinear combination of the interest rate and net output changes speci®ed inthat equation. This leads us to estimate a VAR to represent consumers'forecasts:

ÄnoCA�

r�

24 35t

�a11 a12 a13

a21 a22 a23

a31 a32 a33

24 35 ÄnoCA�

r�

24 35tÿ1

�u1 t

u2 t

u3 t

24 35: (14)

Or written more compactly: z t � Az tÿ1 � u t , where E(z t�i) � A iz t . This mayeasily be generalised for higher orders of VAR by writing a pth order VAR in®rst order form. A test of the simpler model that holds interest rates constantwould involve a VAR that omits the third equation and the third variable, r�.

Using (14), the restrictions on the current account in (12) can be expressedas:

5 Otto and Voss (1995) demonstrate it can be valuable to consider the net foreign asset position.



hz t � ÿP1i�1

âi(g1 ÿ ãg2)A iz t , (15)

where g1 � [1 0 0], g2 � [0 0 1], and h � [0 1 0]. (Again this can be general-ised for a larger number of lags.) For a given z t , the right-hand side of (15) canbe expressed as: cCA�t � kz t , (16)

where

k � ÿ(g91 ÿ ãg92)âA(Iÿ âA)ÿ1:

This expression gives a model prediction of the current account variableconsistent with the VAR and the restrictions of the intertemporal theory. Thiscan be compared graphically with the actual data as an indication of how wellthe restrictions of the theory are satis®ed. Note that kz t is not a forecast of thecurrent account in the conventional sense, but rather is a representation ofthe model's restrictions.

In addition, if the restrictions of the theory were consistent with the data,such that cCA�t � CA�t , then the vector k should equal [0 1 0]. This implies themodel may then be tested statistically by using the delta method to calculate a÷2 statistic for the hypothesis that k � [0 1 0]. Let ~k be the difference betweenthe actual k and the hypothesised value. Then ~k9((@k=@A)V(@k=@A)9)ÿ1~k willbe distributed chi-squared with three degrees of freedom, where V is thevariance-covariance matrix of the underlying parameters in the VAR, and(@k=@A) is the matrix of derivatives of the k vector with respect to theseunderlying parameters, which can be computed numerically.

2. Data and Parameter Values

We test (12) using quarterly data from three countries: Canada, Australia, andthe United Kingdom. Small open economies are of special interest, since thesehave been the most problematic in past studies. This is surprising, since thetheory should be most applicable to a small open economy that can borrowwithout affecting international capital markets. Canada and the United King-dom are studied because they were shown to be especially problematic for theintertemporal theory in the previous work of Sheffrin and Woo (1990b).6

Australia was found to be problematic in work by Milbourne and Otto (1992).All three countries have long quarterly data for the required series. Tests usingonly annual data often are unable to reject the restrictions of the model, eventhough the estimates clearly do not coincide with the theory, simply becausethere is so much uncertainty around these estimates that almost no value canbe rejected. All data are from International Financial Statistics (IFS), seasonallyadjusted at annual rates.

We compute a measure of the world real interest rate, rt , following the

6 Belgium and Denmark were also considered in this previous study, but the quarterly data were notavailable for the ®rst of these countries and only available for the recent decade for the second.



method of Barro and Sala i Martin (1990). We collected short-term nominalinterest rates, T-bill rates or the equivalent, on the G-7 economies. We useshort-term interest rates because we wish to adjust for in¯ation expectations,which are much more reliably forecast over a short-time horizon. In¯ation ineach country is measured using that country's consumer price index, andexpected in¯ation is forecast using a six-quarter autoregression. The nominalinterest rate in each country then is adjusted by in¯ation expectations tocompute an ex-ante real interest rate. An average real interest rate then iscomputed, using time-varying weights for each country based on its share ofreal GDP in the G-7 total. This same series is used for each of the three subjectcountries in the study.

The net output series, NOt , was constructed for each of the three subjectcountries by subtracting investment and government purchases from GDP,adjusting by the 1990 GDP de¯ator and by national population. Equation (12)uses this in logged and differenced form, Änot . The series for the currentaccount variable, CA�t , was constructed for each country by subtracting the logof consumption, adjusted for population and the 1990 GDP de¯ator, from thelog of net output.

We follow Rogoff (1992) in using as a proxy for Pt a measure of the realexchange rate derived from IFS. For Canada and the United Kingdom, the IFScomputes an effective nominal exchange rate index based on unit labour costs.This is not available for Australia, so we instead use a market exchange rateindex. These nominal indexes are converted to real terms using a consump-tion price index for industrial countries provided by IFS and a national priceindex for each subject country, respectively.7 An ex-ante expected exchangerate appreciation is computed, E tÄ pt�1, using a six-quarter autoregression,logging and differencing. Finally, the consumption-based real interest rate, r�t ,is computed for each country using the common world real interest rate andthe country-speci®c exchange rate series derived above, as speci®ed by (5).Because we are interested in the dynamic implications of the intertemporalmodel, the three series, Äno, CA� and r� all are demeaned.

Tests of condition (12) are contingent on values for the parameters â, a,and ã. Previous studies that tested the simpler version of the intertemporalmodel needed to deal only with the ®rst of these parameters, but we followtheir strategy in considering a range of values for unknown parameters.

The present model has an advantage in assigning a value to â, the discountfactor. Denoting as r the sample mean for the real interest rate in our data set,the model implies we may compute â � 1=(1� r), which here equals approxi-mately 0.94.

Regarding the share of traded goods in private ®nal consumption, a, ourdata set is of little value for making any inference. So we turn to outsideempirical studies. Stockman and Tesar (1995) estimate the share of tradablesin two ways. First, using services as a proxy, they estimate the average share of

7 Note that the real exchange rate measured here incorporates both the relative price of nontradedgoods and the terms of trade, whereas the model is speci®ed in terms of just the relative price ofnontraded goods.



nontradables over seven countries to be approximately one-half. A secondmethod breaks down expenditure by categories used by Kravis et al. (1982),and uses expenditure on the following categories as a proxy for nontradables:rent, fuel, transportation and communication. This method produces anaverage estimate of the traded share close to two-thirds. The present value testsbelow consider both values for the share parameter, though results are similarin both cases.

The intertemporal elasticity, ã, is the most problematic of the three para-meters, in that outside empirical estimates range widely. Hall (1988) estimatesthe intertemporal elasticity to be small, concluding that it is unlikely to bemuch above 0.1 and it may well be zero. This is based on the observation thatconsumption tends to respond weakly to the real interest rate. On the otherhand, the reciprocal of the intertemporal elasticity, ó , may be interpreted asthe coef®cient of relative risk aversion in the model. Mehra and Prescott(1985) have suggested that sensible values for ó should be less than 2.0,implying a value of ã greater than 0.5. These two concepts, though distinct, arelinked as reciprocals because of the form of the utility function. Because thefocus of this paper is on the response of consumption to interest rates, ratherthan household attitudes toward risk, we are sympathetic to Hall's estimates.

Our tests consider a range of values for the intertemporal elasticity, but wefocus attention on the case in which we estimate the elasticity in the context ofour model. This estimation uses a method developed in an entirely analogouscase by Campbell and Shiller (1989), in which the restrictions of the model areused to identify the parameter. Mechanically, we search for the value of ã thatminimises the ÷2 statistic of the present value test. Campbell and Shiller (1989)demonstrate this can be interpreted as a method of moments estimation.However, when we eventually use the minimised ÷2 statistic to evaluate a test ofthe overidentifying restrictions, a penalty must be imposed by reducing thedegrees of freedom for the distribution by one.

As an alternative method of choosing parameter values, we experimentedwith conventional GMM estimation of (3). However, we found that thesemethods gave imprecise estimates of the three parameters with large standarderrors, and estimates of the tradeable goods share were outside the rangepermitted by the theory.8

Before we test the model, we must check the assumption that the variablesin the VAR, CA�, r� and Äno, are stationary. We run a standard procedure byregressing

ÄCA�t �Pni�1

biÄCA�tÿi � cCA�tÿ1 � ç t (17)

and testing whether the coef®cient c is negative and signi®cantly differentfrom zero using the appropriate Dickey-Fuller statistics.9 We perform this test

8 Detailed GMM results are available from the authors.9 We do not include a time trend or constant in the above regression, because the three series have

been demeaned, and they ¯uctuate around a level of zero without apparent trend throughout thesample period.



for a range of lags, n, on the differenced term. In addition, we test fornonstationarity by the Phillips-Perron test, which controls for higher-orderserial correlation by making a correction to the t-statistic for the coef®cient c,rather than adding a set of lagged difference terms as in the Dickey-Fuller tests.We use the Newey-West heteroskedasticity autocorrelation consistent estimateof the adjustment, and we consider a range of values for the number of periodsof serial correlation. We perform these stationarity tests also for the two othervariables used in our present value tests, the consumption-based real interestrate, r�, and the change in net output, Änot . In computing r� for thestationarity tests, we use a value for the parameter a of one-half, and for ã weuse the values estimated using the method of Campbell and Shiller (1989), asdiscussed above. The results are reported in Table 1. For each of the threevariables in each country, both the Dickey-Fuller test and the Phillips-Perron

Table 1Unit Root Tests

no. of lags 1 3 5

AustraliaCurrent account (CA�):ADF ÿ2.599�� ÿ3.126�� ÿ3.160��PP ÿ3.212�� ÿ3.970�� ÿ3.441��Interest rate (r�):ADF ÿ7.421�� ÿ5.515�� ÿ4.327��PP ÿ10.660�� ÿ10.667�� ÿ10.673��Change in net output (Äno):ADF ÿ8.181�� ÿ5.576�� ÿ3.854��PP ÿ14.153�� ÿ14.242�� ÿ14.377��Share of traded goods � 0.5, intertemporal elasticity � 0.087, range: 1961-Q2 to 1996-Q2.

CanadaCurrent account (CA�):ADF ÿ2.768� ÿ2.661�� ÿ2.388��PP ÿ2.975�� ÿ3.020�� ÿ2.966��Interest rate (r�):ADF ÿ6.697�� ÿ5.119�� ÿ4.435��PP ÿ10.144�� ÿ10.274�� ÿ10.367��Change in net output (Äno):ADF ÿ10.959�� ÿ6.505�� ÿ4.999��PP ÿ15.255�� ÿ15.734�� ÿ15.911��Share of traded goods � 0.5, intertemporal elasticity � 0.039, range: 1960-Q1 to 1996-Q2.

United KindomCurrent account (CA�):ADF ÿ2.233� ÿ2.438� ÿ2.293�PP ÿ2.396� ÿ2.502� ÿ2.553�Interest rate (r�):ADF ÿ8.327�� ÿ5.974�� ÿ4.526��PP ÿ11.873�� ÿ11.876�� ÿ11.872��Change in net output (Äno):ADF ÿ9.191�� ÿ4.995�� ÿ4.587��PP ÿ14.048�� ÿ14.062�� ÿ14.036��Share of traded goods � 0.5, intertemporal elasticity � 0.085, range: 1960-Q1 to 1996-Q2.

Notes: ADF indicates the augmented Dickey-Fuller test; PP indicates Phillips-Perron.`�' indicates the test statistic is signi®cant at the 5% signi®cance level: `��' indicates the 1% signi®cancelevel. Regressions do not include a constant or time trend.



test reject the presence of a unit root at least at the 5% signi®cance level for allnumbers of lags considered.

3. Results

The results from present value tests are summarised in Tables 2±4. All tableshave the same format, where each column represents an alternative speci®ca-tion of the model. The ®rst and second columns are most important. The ®rstcolumn shows a benchmark model, which ignores changes in the interest rateand exchange rate. The second column shows a model augmented with thesetwo variables, in which the intertemporal elasticity is estimated in the contextof the model. Each column reports the estimated k-vector, as well as theassociated ÷2 statistic and its p-value. Finally the volatility of the predictedcurrent account is reported as a ratio to that for the actual data. Note thatsince the degrees of freedom vary by case, the ÷2 statistic is not comparableacross cases, but the p-value is useful for such comparisons. To illustratefurther how well the restrictions of the model are satis®ed, Figs 1±6 plot themodel prediction for the current account variable, derived using (16), andcompare this to the data.10

To preview brie¯y the results discussed below in detail, the statistical test inall three countries rejects the benchmark model, which ignores changes in theinterest rate and exchange rate. But for two of the countries, the modelaugmented with these variables is not rejected, and even in the third case,there appears to be improvement relative to the benchmark model.

3.1. Australia

In the case of Australia, the Akaike information criterion suggests that two lagsbe used.11 Fig. 1 shows the current account variable computed from the dataand the prediction generated by the version of the intertemporal model thatexcludes interest rates and exchange rates, over the range 1961-Q4 to 1996-Q2.This simple model does fairly well in predicting the general direction ofcurrent account ¯uctuations, such as the run of sizeable de®cits in the early1980s and another in the middle of the decade. Indeed, a cursory examinationof output data for Australia suggests transitory dips in output roughly corre-sponding to these periods.

However, the statistical test, presented in column 1 of Table 2, soundlyrejects the model. The intertemporal theory suggests that with two lags andtwo variables the k-vector should be [0 0 1 0]. The k-vector coef®cient on thecurrent account at date t is 0.406, and while it is signi®cantly different fromzero it also is signi®cantly different from the value of unity suggested by thetheory. Further, the values on net output and lagged current account are

10 Note that the Figs. do not offer a way to control for the number of variables or free parameters inthe model. Therefore in comparing alternative models, we will rely primarily on the statistical tests, anduse the Figs. mainly for illustration.

11 Similarly, Ghosh (1995) uses between 1 and 3 lags for his VARS on quarterly data.



signi®cantly different from their theoretical values of zero. Overall, the ÷2 teststrongly rejects the model, with a p-value of zero. This result is typical of mostpast tests in this area: while a simple graphical analysis suggests the simpleintertemporal model can explain much, the model rarely satis®es statisticaltests. The table shows that part of the problem is that the model prediction isonly about two-thirds as volatile as the actual data. The graph con®rms this;while the model captures the direction of most current account ¯uctuations, itrepeatedly underpredicts the magnitudes.

Next consider an intertemporal model which includes a time-varying con-sumption based real interest rate. We believe this partly explains someepisodes of current account de®cit in Australia, such as the middle 1980s, asthe world real interest rate series computed here is unusually low during thisperiod. We focus on the model in which we use the method of Campbell andShiller (1989) to estimate the intertemporal elasticity. The resulting estimatefor ã is 0.087, which is low but not out of line with estimates by Hall (1988)discussed in the previous section. Fig. 2 shows that the model prediction isimproved over that of the simpler model, mainly in that it better captures themagnitude of ¯uctuations. This improvement is con®rmed in column 2 ofTable 2. The intertemporal theory suggests that with two lags and threevariables, the k-vector should be [0 0 1 0 0 0]. The coef®cient on lagged cur-rent account is 0.934, signi®cantly different from zero and not signi®cantly

Fig. 1. Australia Current Account Variable Excluding Interest Rate and Exchange Rate



Table 2Australia Present Value Tests

Primary models Alternative models

Cases:

(1)Benchmark

model(r� constant)

(2)Optimal model

(ã chosen tominimise ÷2)

(3)ã chosen to

match variance(4)

ã � 0.5(5)

ã � 1.0

(6)just interest rate(exchange rate

excluded)(7)

a � 2=3

ã ± 0.087 0.022 0.500 1.000 0.078 0.097k-vector

not 0.308 0.048 0.048 ÿ0.140 0.401 0.328 0.120(0.052) (0.120) (0.121) (0.234) (0.401) (0.069) (0.094)

notÿ1 0.059 0.124 0.124 0.288 0.212 0.084 0.108(0.030) (0.071) (0.071) (0.137) (0.236) (0.041) (0.055)

CA�t 0.406 0.934 0.988 1.008 1.063 0.523 0.749(0.105) (0.231) (0.234) (0.413) (0.772) (0.134) (0.180)

CA�tÿ1 0.206 ÿ0.047 ÿ0.034 ÿ0.402 0.479 0.252 0.010(0.060) (0.151) (0.151) (0.304) (0.448) (0.078) (0.120)

r�t ± 0.009 0.001 1.169 12.945 1.841 0.021(0.013) (0.003) (0.357) (3.280) (0.568) (0.018)

r�tÿ1 ± ÿ0.004 ÿ0.001 0.389 ÿ5.680 ÿ0.805 0.000(0.009) (0.002) (0.187) (1.836) (0.318) (0.012)

÷2-statistic 51.288 5.825 5.942 12.137 16.586 37.645 9.342p-value 0.000 0.324{ 0.312{ 0.059 0.011 0.000{ 0.096{ó bCA�/óCA� 0.662 0.944 1.000 1.483 5.472 1.000 0.823

Notes: Standard errors in parentheses. Regressions are for 1961-Q4 to 1996-Q2.Share of tradables in consumption, a, is 0.5, unless otherwise stated. â � 0.94.{ indicates degrees of freedom equal to 5 in this case instead of 6 because of extra estimated parameter.

2000]547

MO

DE

LS

OF

TH

EC

UR

RE

NT

AC

CO

UN

T

#R

oyal

Eco

no

mic

Society

2000

different from the theoretically prescribed value of unity. All other coef®cientsare very small and insigni®cantly different from zero. The present value test isfar from rejected, with a p-value of 0.324. We compute this p-value for adistribution of ®ve rather than six degrees of freedom, to take into considera-tion the penalty for using one restriction to identify the elasticity we estimated.Con®rming the impression from the ®gure, the table notes that the volatilityof the current account forecast has risen; the standard error is now 94.4% ofthat for the actual data.

As an alternative, we consider a value for the intertemporal elasticity thatenables the model to match the volatility of the current account data. Theestimate for the elasticity again is small, 0.022. As shown in column 3 of Table2, the k-vector coef®cient on current account at date t is now 0.988, very closeto the theoretically predicted value of unity. The statistical test again does notreject the model.

The other columns of Table 2 are included for sensitivity analysis. Columns(4) and (5) explore larger values for the intertemporal elasticity. As theintertemporal elasticity rises, the volatility of the current account rises in excessof the volatility of the actual data, and the ®t of the model worsens. Next,column (6) considers a model in which the exchange rate is not permitted tovary, but the world real interest rate is. This is intended to distinguish theseparate effects of the two components of our composite variable, r�. The

Fig. 2. Australia Current Account Variable Including Interest Rate and Exchange Rate



value of ã used here is that which matches the current account volatility,because the value which minimises the value of the ÷2 statistic is negative. Themodel is rejected, suggesting that the exchange rate, rather than the interestrate, is primarily responsible for the improvement in the model's ®t. Indeed, arelatively appreciated Australian real exchange rate at the beginning of the1980s partly explains the current account de®cit during this period. Column(7) tests sensitivity to the assumption of the share of nontraded goods. Theexperiment of column two is replicated under the assumption that a � 2=3rather than 1=2: The model performs somewhat less well, but the qualitativeconclusions are unchanged.

Clearly, including the consumption-based interest rate improves the ®t ofthe model to the data. In particular, it offers a way to increase the volatility ofthe model prediction for the current account to better match the data. Oneinterpretation of this result is that the new variable helps to capture importantexternal shocks. These are transmitted to the home country through changesin the real interest rate and exchange rate, which then induce a response inconsumption and hence ¯uctuations in the current account.

Small intertemporal elasticities appear to work best in the present model.This is consistent with outside estimates discussed earlier, which suggest thatconsumption responds only moderately to interest rate changes. The theorydeveloped in this paper implies that as the intertemporal elasticity grows small,so do the intertemporal effects of both the world real interest rate and theexpected change in the relative price of nontraded goods. However, the theoryimplies that the relative price of nontraded goods also has an intratemporaleffect, and this effect grows larger as the elasticity approaches zero. Ourestimation of a low intertemporal elasticity suggests that it is mainly thisintratemporal effect that is improving the ®t of the model, more so than theadditional intertemporal effects. Yet both sets of effects are consistent with thetheoretical model. Note also that a small intertemporal elasticity means house-holds aggressively smooth their consumption. So expected changes in netoutput, a central feature of the basic intertemporal current account theory,also play a large role in our results.

3.2. Canada

The Akaike information criterion suggests that two lags be used in the case ofCanada. Fig. 3 shows the actual data for CA�, and the forecast based on thesimple model that ignores changes in the interest rate and exchange rate. Themodel prediction is much less volatile than the actual data. While it showssome variability from quarter to quarter, the prediction misses the larger,medium-term swings of the current account away from balance. In particular,it misses the large surpluses in the early to mid 1980s and again in the middleof the next decade, toward the end of our sample. This poor prediction isre¯ected in the statistical test reported in column (1) of Table 3. The pointestimate of the k-vector coef®cient on the current account is 0.083, far fromthe value of unity prescribed by the theory. The ÷2 test rejects the hypothesis



that the k-vector as a whole is [0 0 1 0], with a p-value of 0.003. The model alsofails in that the forecast of the current account is only a ®fth as volatile as thedata.

Fig. 4 suggests that the model prediction for the current account improveswhen the interest rate and exchange rate are included, and the statistical testin column (2) of the table con®rms this. The intertemporal elasticity isestimated at 0.039, which minimises the ÷2 statistic. Now the longer-run devia-tions of the current account from balance depicted in the ®gure are bettercaptured by the model prediction, including the improvement in the early tomid 1980s and again in the middle of the 1990s. A plausible explanation maybe that this pattern is due to shocks originating in the United States, whosecurrent account follows a basically inverted pattern to that seen in Canada inthose years. These external shocks are re¯ected in the effective Canadianexchange rate, which weights the U.S. dollar heavily. The Canadian exchangerate depreciates in the early and mid 1980s and again in the mid 1990s. Thetheory implies that the current account surpluses could be generated by theintratemporal effect, in which it becomes more expensive to purchase tradablegoods from abroad. In the table, the element of the k-vector corresponding tothe current account is 0.636, an improvement over the benchmark model incolumn (1), but not as close to the theoretical value of unity as was thecorresponding value for Australia. However, the other elements of the k-vector

Fig. 3. Canada Current Account Variable Excluding Interest Rate and Exchange Rate



are close to zero, and the ÷2 test does not reject the hypothesis that the k-vectoris [0 0 1 0 0 0] . The volatility of the current account prediction also isimproved relative to the benchmark in column (1).

Column (3) shows the less successful result if the intertemporal elasticity isestimated to match the second moment of the current account data. Theelement of the k-vector corresponding to the current account is much im-proved, 0.95, but other elements of the k-vector are now signi®cantly differentfrom zero. The statistical test rejects, once the penalty is imposed for the factwe estimate the intertemporal elasticity. Columns 4 to 7 reiterate the conclu-sions from their counterparts for Australia in Table 2. Larger intertemporalelasticities generate excessive volatility. And consideration of a variable ex-change rate is an essential component of the model's success.

3.3. United Kingdom

Results for the United Kingdom are less successful than in the previous twocountries. The Akaike information criterion suggests only one lag be used. Fig.5 offers an especially dramatic example of how the benchmark modelproduces predictions that are much too ¯at. The model completely fails topredict the large swings in the current account, such as the large de®cit in the

Fig. 4. Canada Current Account Variable Including Interest Rate and Exchange Rate



Table 3Canada Present Value Tests


Cases:

(1)Benchmark

model(r� constant)

(2)Optimal model


(3)ã chosen to

match variance(4)

ã � 0.5(5)

ã � 1.0


excluded)(7)

a � 2=3

ã ± 0.039 0.177 0.500 1.000 3.740 0.397k-vector

not 0.192 ÿ0.196 ÿ0.336 ÿ0.304 0.813 2.105 ÿ0.196(0.099) (0.192) (0.233) (0.435) (0.643) (2.286) (0.362)

notÿ1 0.068 0.011 ÿ0.018 0.113 ÿ0.052 ÿ0.436 0.111(0.059) (0.103) (0.123) (0.242) (0.399) (1.418) (0.201)

CA�t 0.083 0.636 0.945 0.733 0.020 0.645 0.558(0.232) (0.420) (0.502) (0.979) (1.510) (5.369) (0.815)

CA�tÿ1 0.036 ÿ0.208 ÿ0.261 ÿ0.528 0.128 0.325 ÿ0.425(0.103) (0.195) (0.235) (0.485) (0.678) (2.411) (0.403)

r�t ± 0.021 0.209 2.969 8.587 49.050 2.459(0.010) (0.070) (0.853) (2.070) (12.469) (0.708)

r�tÿ1 ± 0.004 0.057 1.000 ÿ2.066 ÿ22.100 0.835(0.007) (0.046) (0.442) (0.889) (6.842) (0.367)



552[

AP

RIL

TH

EE

CO

NO

MIC

JOU

RN

AL

#R

oyal

Eco

no

mic

Society

2000

start of the 1990s. The statistical tests in Table 4 con®rm this: the test ®rmlyrejects the intertemporal model restriction that k � [0 1].

One might again expect the intratemporal effect of exchange rates to beimportant here. For example, the United Kingdom real exchange rate isgenerally thought to be rather appreciated in the beginning of the 1990s asthe United Kingdom fought to remain part of the Exchange Rate Mechanismin Europe. This may well have contributed to the low current account. Later,the devaluation in 1992 may have contributed to the return to current accountbalance. Fig. 6 shows the prediction for the model augmented with a variableexchange rate and interest rate. The ®gure suggests the model prediction isstrongly affected. It now begins to capture the medium-run swings frombalance that were utterly absent previously. The statistical test shows that the p-value indeed is improved over that of the simple model. However, the ®t still isquite poor, and the model still is rejected by the test.

4. Conclusion

This paper has examined the question of why simple intertemporal models ofthe current account have not fared well in tests using data from small open

Fig. 5. United Kingdom Current Account Variable Excluding Interest Rate and Exchange Rate



Table 4United Kingdom Present Value Tests


Cases:

(1)Benchmark

model(r � constant)

(2)Optimal model


(3)ã chosen to

match variance(4)

ã � 0.5(5)

ã � 1.0


excluded)(7)

a � 2=3

ã ± 0.085 0.458 0.500 1.000 0.049 0.429k-vector

not 0.206 0.165 0.133 0.134 0.140 0.210 0.177(0.049) (0.049) (0.050) (0.050) (0.053) (0.053) (0.049)

CA�t 0.072 0.546 0.634 0.625 ÿ0.128 0.001 0.396(0.202) (0.203) (0.205) (0.206) (0.217) (0.217) (0.203)

r�t ± 0.002 0.581 0.838 11.544 1.235 0.002(0.005) (0.061) (0.077) (0.504) (0.502) (0.005)



554[

AP

RIL

TH

EE

CO

NO

MIC

JOU

RN

AL

#R

oyal

Eco

no

mic

Society

2000

economies. This failure is surprising, inasmuch as the underlying assumptionsof the theory should apply especially well in these economies. We show thatallowing both for a variable interest rate and exchange rate can improve the ®tof the model. In some cases movements in the interest rate and exchange ratecan explain much of the medium-term movements of the current accountfrom balance that had been unexplained under the simpler intertemporaltheory used in earlier tests.

This paper has offered an explanation for the explanatory power of theseadditional variables. The current account of a small open economy is likely tobe affected not only by shocks to domestic output or government expenditure,but also by external shocks to the economies of large neighbours. Suchexternal shocks should be expected to affect the domestic economy viachanges in the world interest rate and the country's real exchange rate, bothof which set the terms by which the small open economy can trade intertempo-rally with the rest of the world. The intertemporal theory implies that suchchanges in the interest rate and exchange rate affect the intertemporal pro®leof saving and hence the current account. However, the theory also allows forintratemporal effects arising from the presence of exchange rates, and itappears these intratemporal effects are signi®cantly responsible for the mod-el's improved ®t.

Fig. 6. United Kingdom Current Account Variable Including Interest Rate and Exchange Rate



The intertemporal model, with its dynamic budget constraint and intertem-poral trade-offs, is a useful starting point for current account analysis. Thesimplest versions of the intertemporal current account model admittedly arelimited as positive descriptions of the data. But results here suggest the model'sdescriptive power can be improved with modest extensions, notably theinclusion of certain intratemporal trade-offs. Future empirical work shouldconsider further extensions of the basic model that have been considered intheoretical work, such as investment dynamics, a distinction between durableand nondurable goods, labour supply decisions, and nominal rigidities.

University of California, Davis

Date of receipt of ®rst submission: June 1998Date of receipt of ®nal typescript: August 1999

Appendix A. Deriving the Optimal Consumption Pro®le

We follow Dornbusch (1983) and Obstfeld and Rogoff (1996) in deriving the optimalconsumption pro®le. De®ne an index of total consumption, C�t � C a

Tt C1ÿaNt . De®ne

also a consumption-based price index, P�t , as the minimum amount of consumptionexpenditure Ct � CTt � PtCNt such that C�t � 1, given Pt . Traded goods are thenumeraire. The household problem (1) and (2) implies UNt � PtUTt and hence thefollowing allocation of expenditure between tradables and nontradables:

CTt � aCt and CNt � (1ÿ a)Ct

Pt: (18)

Substitute these into the de®nition of C�t

C�t � (aCt)a (1ÿ a)

Ct

Pt

� �1ÿa

(19)

and use the de®nition of P� to write

(aP�t )a (1ÿ a)P�tPt

� �1ÿa

� 1: (20)

Solve this for the consumption-based price index:

P�t � P 1ÿat [aÿa(1ÿ a)ÿ(1ÿa)]: (21)

This allows us to rewrite the budget constraint of the optimisation problem (2) as

Yt ÿ P�t C�t ÿ I t ÿ Gt � rtB tÿ1 � Bt ÿ Btÿ1 (22)

and the utility function as U (C�t ) � [1=(1ÿ ó)](C�t )1ÿó . This implies an intertempor-al Euler equation:

Et â(1� r t�1)P�t

P�t�1

!C�t

C�t�1

!ó24 35 � 1: (23)

To facilitate empirical implementation, we rewrite this condition in terms of consump-tion expenditure and the relative price of nontraded goods:



Et â(1� r t�1)Ct

C t�1

� �ó Pt

P t�1

� �(1ÿó )(1ÿa)" #

� 1: (24)

Assume joint log normality for the gross real world interest rate (1� r t�1), consump-tion growth rate (Äc t�1 � log C t�1 ÿ log Ct), and the percentage change in the relativeprice of nontraded goods (Ä pt�1 � log P t�1 ÿ log Pt). Assume also that the variancesand covariances between these variables are not time-varying. Then following conven-tional methods, the expression above may be written in log-linearised form:12

E tÄc t�1 � ãE t r t�1 � 1ÿ ã

ã(1ÿ a)Ä pt�1

� �

� 1

2[ó 2

c � ã2ó 2r � (1ÿ ã)2(1ÿ a)2ó 2

p � 2ãó c, r

� 2(1ÿ ã)(1ÿ a)ó c, p � 2ã(1ÿ ã)(1ÿ a)ó r , p], (25)

where the variances and covariances refer to the three variables de®ned above, andwhere ã � 1=ó . Use has been made of the approximation: log(1� r t�1) � r t�1.De®ning the ®rst bracketed set of terms on the right as a consumption based realinterest rate, r�t�1, and noting that under our assumptions the second bracketed set ofterms on the right is constant, this becomes the optimal consumption pro®le in thetext (4).

Appendix B. Deriving the Log-linearised Intertemporal BudgetConstraint

We can write the intertemporal budget constraint (9) as

Ö0 ÿØ0 � B0, (26)

where Ö0 � C0 �P1

t�1 RtCt , and Ø0 � NO0 �P1

t�1 RtNOt . Taking logs and followingthe linearisation of Huang and Lin (1993), we have:

ö0 ÿ ø0 � 1ÿ 1

Ù

� �(b0 ÿ ø0), (27)

where ö0 � logÖ0, ø0 � logØ0, b0 � log B0, and Ù � 1ÿ (B=Ö0), where B is steadystate net foreign assets. Now a further linearisation yields:

c0 ÿ ö0 �P1t�1r t(rt ÿ Äct), (28)

where c0 � log C0, Äct � log Ct ÿ log C tÿ1, and r � 1ÿ (c=ö0) where c is the steadystate value of the log of consumption. Similarly,

no0 ÿ ø0 �P1t�1r t(rt ÿ Änot), (29)

where no0 � log NO0, and Änot � log NOt ÿ log NOtÿ1. Substitute (28) and (29) intothe intertemporal budget constraint, (27):

12 See Campbell et al (1997) pp. 306±7.



no0 ÿ ø0 �P1t�1r t(rt ÿ Änot)�

P1t�1r t(rt ÿ Äct)ÿ c0

� 1ÿ 1

Ù

� �(b0 ÿ ø0)� 1ÿ 1

Ù

� � P1t�1r t(rt ÿ Äct)ÿ c0

� �, (30)

which may be rewritten as (10) in the text:

ÿP1t�1

â t Änot ÿ Äct

Ùÿ 1ÿ 1

Ù

� �rt

� �� no0 ÿ c0

Ù� 1ÿ 1

Ù

� �b0: (31)

ReferencesBarro, R. J. and Sala i Martin, X. (1990). `World real interest rates.' In (O. J. Blanchard and S. Fischer

eds.) NBER Macroeconomics Annual. Cambridge, MA: MIT Press, pp. 15±61.Campbell, J. Y. (1987). `Does saving anticipate declining labor income? An alternative test of the

permanent income hypothesis.' Econometrica, vol. 55, no. 6, (November), pp. 1249±74.Campbell, J. Y. (1998). Àsset prices, consumption, and the business cycle.' NBER Working Paper

No. 6485 (March).Campbell, J. Y., Lo, A. W. and MacKinlay, A. C. (1997). The Econometrics of Financial Markets. Princeton:

Princeton University Press.Campbell, J. Y. and Mankiw, N. G. (1989). `Consumption, income, and interest rates: reinterpreting the

time series evidence.' In (O. J. Blanchard and S. Fischer eds.) NBER Macroeconomics Annual.Cambridge, MA: MIT Press, pp. 185±244.

Campbell, J. Y. and Shiller, R. (1987). `Cointegration and tests of present value models.' Journal ofPolitical Economy, vol. 95, no. 5, (October), pp. 1062±88.

Campbell, J. Y. and Shiller, R. (1989). `The dividend-price ratio and expectations of future dividendsand discount factors.' The Review of Financial Studies, vol. 1, no. 3, (Fall), pp. 195±228.

Dornbusch, R. (1983). `Real interest rates, home goods and optimal external borrowing.' Journal ofPolitical Economy,vol. 91, no. 1, (November), pp. 141±53.

Fahrion, M. (1997). `The current account and variable interest rates: evidence from U.S., U.K. andCanadian time series.' Mimeo, Department of Economics, Columbia University.

Ghosh, A. R. (1995). Ìnternational capital mobility among the major industrialised countries: too littleor too much?' Economic Journal, vol. 105, no. 428, ( January), pp. 107±28.

Hall, R. E. (1988). Ìntertemporal substitution in consumption,' Journal of Political Economy, vol. 96,no. 2, (April), pp. 339±57.

Huang, C. and Lin, K. (1993). `De®cits, government expenditures, and tax smoothing in the UnitedStates: 1929±1988.' Journal of Monetary Economics, vol. 31, no. 3, ( June), pp. 317±39.

Kravis, I., Heston, A. and Summers, R. (1982). World Product and Income: International Comparisons andReal GDP. Baltimore, MD: Johns Hopkins University Press.

Mehra, R. and Prescott, E. (1985). `The equity premium: a puzzle.' Journal of Monetary Economics, vol. 15,no. 2, (March), pp. 145±61.

Milbourne, R. and Otto, G. (1992). `Consumption smoothing and the current account.' AustralianEconomic Papers, vol. 31, no. 59, (December), pp. 369±84.

Obstfeld, M. and Rogoff, K. (1996). Foundations of International Macroeconomics. Cambridge, MA: the MITPress.

Otto, G. (1992). `Testing a present value model of the current account: evidence from U.S. andCanadian time series.' Journal of International Money and Finance, vol. 11, no. 5, (October),pp. 414±30.

Otto, G. and Voss, G. M. (1995). `Consumption, external assets and the real interest rate.' Journal ofMacroeconomics, vol. 17, no. 3, (summer), pp. 471±94.

Rogoff, K. (1992). `Traded goods consumption smoothing and the random walk behavior of the realexchange rate.' Bank of Japan Monetary and Economic Studies, vol. 10, no. 2, (November), pp. 1±29.

Sheffrin, S. and Woo, W. T. (1990a). `Testing an optimizing model of the current account via theconsumption function.' Journal of International Money and Finance, vol. 9, no. 2, ( June), pp. 220±33.

Sheffrin, S. and Woo, W. T. (1990b). `Present value tests of an intertemporal model of the currentaccount.' Journal of International Economics,vol. 29, no. 3±4, (November), pp. 237±53.

Stockman, A. C. and Tesar, L. (1995). `Tastes and technology in a two-country model of the businesscycle: explaining international comovements.' American Economic Review, vol. 85, no. 1, (March),pp. 168±85.


558 [ A P R I L 2000]T H E E C O N O M I C J O U R N A L

Documents

INTEREST RATES, EXCHANGE RATES AND … RATES, EXCHANGE RATES AND PRESENT VALUE MODELS OF THE ... Just as individuals may adjust consumption and saving behaviour in ... Following Dornbusch