I. Introduction

The current international environment is characterized by larger current account imbalances among major countries than at any point since the early post-war period. At present, the United States is running a current account deficit equal to about 3 percent of its Gross National Product, and Japan and the Federal Republic of Germany have current account surpluses of the same order, relative to their output levels. Moreover, current account positions of this size have persisted from 1982-83 to the present, leading to massive changes in net foreign assets and liabilities of these three countries. The United States, which was the largest net creditor country during most of the post-war period up until the early 1980s, is now estimated to be the world’s largest net debtor, having net foreign liabilities of about $600 billion--larger than the net liability position of the 10 most heavily indebted developing countries combined. Over this period, Japan and Germany accumulated sizeable net foreign claims; they are now estimated to be net creditors to the extent of some $350 and $200 billion, respectively (IMF [1988], Table 32, p. 89).

Relative to their output levels such claims positions are not unprecedented--for instance, the net foreign liability position of the United States is about 7 percent of GNP, while Canada’s has long exceeded 20 percent of GNP--but because of the size of the countries concerned, their importance for international credit flows, and the key role of their currencies for the international monetary system, the payments imbalances are potentially cause for concern. An important issue is whether forces exist that bring about smooth adjustment of net foreign assets to equilibrium levels, or whether instead tendencies toward increasing asset and liability positions will continue, and will be associated with instability in international credit and foreign exchange markets.

The causes of the present current account imbalances are not the subject of this paper; the extent to which they are related to the relative stance of fiscal policies in the United States, Japan and Germany since the early 1980s has been discussed extensively elsewhere. ^2/ Instead, the analysis will consider stabilizing mechanisms that result both from private sector behavior and from fiscal policy. The adjustment mechanisms may include induced changes in aggregate supply or domestic absorption, either the direct result of net foreign asset accumulation, or prompted indirectly by changes in real exchange rates and interest rates.

It is useful in considering the question to distinguish between two conceptually different but interrelated issues: first, whether a long-run relationship exists between net foreign assets and other variables, and second, what short-run or medium-run adjustment mechanisms exist to bring about the long-run relationship. We specify a general model that allows for feedbacks consistent with various transmission mechanisms. In order to assess whether a long-run relationship exists, we both test whether deviations from alternative long-run relationships seem to reverse themselves, and examine whether some of the key short-run feedback mechanisms can be identified within a dynamic model framework. The existence of a stable long-run equilibrium and of short-run equilibrating feedback can in fact be viewed as facets of the same phenomenon (Hendry [1986], Engle and Granger [1987], and Section III below).

The model we consider is related to a number of strands in the literature. First, since Feldstein and Horioka [1980], a number of articles have considered why national saving and investment tend to be so highly correlated, and hence why current account imbalances tend to be small (though the data cited above for the three largest industrial countries modifies this judgement to some extent). We consider the stock counterpart of the above question; we are interested not only in the level of the current flows, but also in their persistence, ^3/ and hence the extent they cause asset stock accumulation. Even if current accounts are small, they may still bring about large accumulations of net claims or indebtedness; we are interested in forces that would prevent this happening. ^4/ The stabilizing feedbacks of asset stocks on goods and asset prices, and thence on saving and investment, may help explain the empirical regularity identified by Feldstein and Horioka.

This paper is also related to the literature on the transfer problem (see, among others, Johnson [1956] and Frenkel and Razin [1987]). That literature considers whether transfers of wealth between countries have effects on expenditure that feed back onto the current account balance, allowing the transfer to be effected without terms of trade changes. Since accumulations of net foreign claims correspond to transfers of wealth, depending on spending propensities, stabilization of foreign claims may or may not require real exchange rate changes.

Finally, in the absence of Ricardian Equivalence (Barro [1974]), the boundedness of net foreign assets as a ratio to GNP is related to whether government bonds as a ratio to GNP are bounded. Net foreign liabilities play the same role for the economy’s intertemporal budget constraint as bonds for the government’s budget constraint. There have been several recent studies examining the empirical implications of the government’s intertemporal budget constraint for the conduct of budgetary policy (Hamilton and Flavin [1986], Smith and Zin [1988], and Kremers [1989]). As in that literature, the proper question is not whether the economy ultimately satisfies its intertemporal budget constraint, but rather whether there are feedback mechanisms that operate smoothly to ensure that it is satisfied.

The plan of the paper is as follows. A theoretical model explaining net foreign asset positions is presented in Section II. This is followed by a description of the econometric methodology in Section III. The empirical results are in Section IV, and Section V concludes.

II. The Model

We begin by considering the pattern of net foreign claims in a model with optimizing consumers. In such a model, it is possible to relate the equilibrium values of variables to assumptions concerning preferences of households. We choose a framework that is general enough to allow the government’s financing decisions to have real effects. In particular, we use a two-country version of the Blanchard-Yaari model (see Blanchard [1985], Buiter [1984], and Frenkel and Razin [1986]). It is shown that the equilibrium for net foreign assets depends in general on levels of output, government spending and government debt in the two countries. The discussion at the end of this section considers alternative models for the net foreign asset position that also take into account the following features: (1) endogenous capital accumulation; (2) life cycle effects on saving; (3) endogenous government policies; and (4) a risk premium that depends on the net claims position. Such models suggest that in general there are many factors that may affect the behavior of net foreign assets.

We assume that as in Blanchard [1985] consumers maximize the expected discounted present value of utility, subject to a fixed probability of death that is independent of age; this probability p is assumed to apply to the representative consumer in each country. Moreover, the birth rate is also equal to p, so that the population is stationary. ^5/ Each individual’s discount rate is equal to θ, both in the home country and in the foreign country. ^6/ Countries are specialized in the production of one good; production is exogenous. Utility in each country is assumed to be Cobb-Douglas in the home good and the foreign good, implying constant expenditure shares. For the home country those shares are 1-β on the home good and β on the foreign good; for the foreign consumers, corresponding expenditure shares are 1-β* and β*. ^7/ It is assumed that consumption is biased toward home goods, so that β and β* are each less than one-half, and hence 1 - β - β* > 0. As in Blanchard [1985], insurance is assumed to exist, such that the insurance company inherits all the consumer’s wealth upon death, in exchange for an actuarially fair insurance payment; there are therefore no bequests from parents to children. Financial wealth in the home country is held in the form of government bonds (B for the aggregate economy) and net claims on foreigners (F, assumed to be denominated in foreign currency). It is assumed for the time being that domestic and foreign bonds are perfect substitutes, so that interest parity holds between real interest rates r and r*:

r = r* + ė

The real exchange rate between the goods produced in the two countries, e, is endogenous.

Productivity growth is assumed to proceed at rate η in each country; as already mentioned there is no population growth, and hence outputs Y and Y* (assumed exogenous) also expand at n (all variables are measured in real terms). Letting C, C* equal aggregate consumption, and T, T* equal taxes in the home and foreign countries, then under the above assumptions consumption in each country is a constant fraction of total wealth:

where human wealth levels H, H* are given by

Goods market equilibrium requires that output is equal to the sum of domestic and foreign consumption and government spending; each government is assumed for convenience only to buy domestic goods, in amounts G, G*, respectively. ^8/ So goods market clearing requires

The change in the net foreign asset position is equal to the current account balance, that is, to foreign purchases of home goods, minus home purchases of foreign goods, plus investment income receipts from abroad (or payments, if F is negative). Recall that these claims or liabilities are assumed to be denominated in the foreign country’s currency and hence to pay rate r*, equal to the rate paid on the debt of the foreign government. Hence the change in net foreign assets is given by

Government budget constraints relate the change in bond stocks to government spending, taxes, and real interest rates:

As noted above, we consider the case where outputs in the two countries grow at the same rate n; we assume that the interest rates r, r* exceed that growth rate in equilibrium. ^9/ We further assume that, in equilibrium, government spending grows at the same rate as output, as do taxes and the public debt. We then examine the long-run equilibrium for net foreign assets, assuming first that a steady state exists with constant interest rates and real exchange rate, and where asset stocks grow at the same rate as output. The stability of such an equilibrium is subsequently discussed.

The existence of a steady state presupposes a constant real exchange rate; interest parity therefore implies that r = r* in steady state. With constant interest rates and constant growth rates for income and taxes, human wealth is given by the following:

therefore, if Ḟ = nF, we can solve for F using (1) and (4). This yields

The equilibrium level for r can be obtained by summing the goods market equilibrium conditions (3), with the second equation multiplied by the equilibrium exchange rate e. Writing the resulting world variables Ȳ - Y + eY*, etc., yields

Assuming that Ḃ = nB, we can use the government budget constraint, summed for the two countries, to solve for $\overline{\mathrm{T}}$, namely,

Substituting (9) into (8), we can solve for r:

Thus, when either world government debt $\overline{\mathrm{B}}$ or the probability of death p is zero, the real interest rate equals n + θ; when debt and p are positive, the real interest rate is higher, by an amount that is increasing in the ratio of debt to world output net of government spending. ^10/

Using the expression for the equilibrium interest rate, we can derive expressions for the steady state exchange rate and net foreign asset position. First define $\delta ={\left[1+\mathrm{p}\overline{\mathrm{B}}/\left(\overline{\mathrm{Y}}-\overline{\mathrm{G}}\right)\right]}^{-1}$. Substituting the expressions for consumption, (1), and equilibrium human wealth, (6), and the government budget constraint, (5), into one of the pair of equations (3), say the first one, yields

Equation (11) corresponds to goods market equilibrium. It gives the combinations of e and F such that home-country output is equal to demand; an increase in F, on the assumption that consumption is biased towards domestic goods so 1-β-β*>0, raises the relative demand for home goods. Therefore, it is associated with a real appreciation (a decrease in e).

Next, equation (7) gives an expression for asset market equilibrium:

Equation (12) corresponds to equality between the current and capital accounts. Together, equations (11) and (12) determine equilibrium net foreign assets and the real exchange rate.

After considerable manipulation, it can be shown that the home country’s net foreign claims as a ratio to its output in steady state are given by

The denominator of the RHS of (13) is positive, for reasonable levels for the variables. ^11/ Therefore, the sign of F is the same as the sign of the numerator of (13), and higher domestic government spending and debt are associated with lower net claims on foreigners (keeping world totals unchanged) and conversely for foreign country variables.

The stability of the model will not be analyzed here; it has been treated by Buiter [1984]. He shows that a more complicated version of the model that includes capital accumulation is saddlepoint stable, for plausible parameter values and assuming adjustment of taxes to satisfy the government’s intertemporal budget constraint. Given that the model is stable, it can be expressed as adjustment equations toward a steady state growth path which, in the case of F, H, and H*, involves growth at rate n. Scaling eF and H by Y, and H* by Y*, gives variables that are stationary in steady state. If we call the resulting ratios f, h and h*, and their equilibrium levels $\hat{\mathrm{f}}$, ĥ and ĥ*, respectively, and if the equilibrium level of the exchange rate is denoted by e, then the model can be linearized around those equilibrium values to give the following fourth-order system:

The matrix A has three negative and one positive characteristic roots, in keeping with saddlepoint stability. The error-correction models discussed in Section IV estimate discrete-time versions of a subset of (14).^12/

There are several theoretical extensions of the above model that may be of empirical relevance. First, production and investment can be made endogenous. Current account surpluses and deficits correspond not just to consumption smoothing but also result from investment flowing to its most productive uses. Long-run net foreign asset positions then result from cumulative investment flows in excess of domestic saving.

Second, equilibrium net foreign asset positions have also been considered in a framework that takes account of the age structure of the population. In a model with two overlapping generations, it can be shown that the country whose residents have a higher value of consumption when young will run a steady-state current account deficit, and thus will be a net debtor in equilibrium (Buiter [1981]). Persson [1985] shows in a two-country, one-good version of that model that net foreign assets will be affected by government debt in substantially the same way as described above. More generally, life-cycle models (see Ando and Modigliani [1965]) posit that saving varies over a person’s life for such reasons as accumulating down-payments on a house and saving for children’s education or retirement. Such a model would suggest that aggregate saving, and perhaps also the current account and net foreign asset positions could depend on the age structure of a country’s population. Lee and Lapkoff [1988] consider the effects of increased fertility--implying an age structure with proportionately more young people--on intergenerational transfers and consumption. They conclude that higher fertility would lower life cycle consumption, though this is the net result of several effects with different signs, all of which subject to a “considerable range of uncertainty.”

Third, the possibility that government policies are modified in order to influence the net foreign asset position may have to be taken into account. For instance, Summers [1988] argues that the apparent immobility of capital may be due to policies of governments that aim to restrict international investment positions, and that a reason they may want to do so is that the social return to domestic investment exceeds that of foreign investment. Governments may choose to impose capital controls or taxes; it may also be the case that budgetary positions are influenced by targets for net national saving.

Fourth, interest parity may be violated because domestic and foreign bonds are not perfect substitutes, as a result of currency risk or country risk. There may therefore be a systematic portfolio balance relationship between interest differentials and net foreign asset positions in the long run. As an example, Dooley and Isard [1986] have recently argued that fear of taxation helps to explain why large net claims positions are not allowed to develop, despite favorable investment opportunities.

To sum up, the model suggests that there should be a long-run relationship between ratios to GNP of net foreign assets, government spending, and government debt. In the empirical part of the paper, we first examine the net foreign assets ratio to see whether it exhibits stability over time. We then attempt to relate its secular movement to the government variables mentioned above, and to demographic variables.

The model also suggests that the dynamic feedbacks allowing net foreign assets to maintain a stable pattern may be complex. Possible linkages include the following:

• (1) An increase in F changes home and foreign wealth, changing expenditure patterns; given the bias of consumption towards domestic goods, this leads to an increase in the relative demand for the home goods.

• (2) Changes in F will be associated with changes in the value of the real exchange rate needed to maintain goods market equilibrium, leading to expenditure switching.

• (3) If a portfolio balance relationship exists, then changes in F will affect interest differentials directly, which will also feed back onto exchange rates and expenditure.

These feedbacks may operate with a lag, especially in a model where the accumulation of physical capital is taken into account. Therefore, the dynamics of adjustment may be quite complex. In the empirical part of the paper, Section IV, we attempt to identify some of these feedbacks by estimating dynamic adjustment equations for net foreign assets and for domestic absorption. In each case, we test whether the deviations from the long-run equilibrium net foreign asset ratio (discussed above) have significant feedback effects.

III. Econometric Methodology

The purpose of the empirical analysis is first to identify the existence of possible stabilizing mechanisms that ensure a long-run equilibrium of the ratio of net foreign assets to GNP. As we have seen above, this equilibrium could be a function of other variables, including the ratio of government debt to GNP. Given that some of these variables may appear to be non-stationary, that is, not to settle down to a constant value, particularly in the small post-war sample, ^13/ the concept of cointegration offers a suitable framework for the empirical analysis.

Cointegration allows one to study the behavior of variables moving apart in the short run, but brought together again to a stationary equilibrium by government policy or market forces, or both, in the long run. A variable is defined to be integrated of order one [denoted by I(1)] if it must be differenced once to induce stationarity [denoted by I(0)]. ^14/ A set of variables, each I(1), is said to be cointegrated if a linear combination of them is I(0) (i.e., if they “move together in the long run”). That linear combination is characterized by a cointegrating vector.

The empirical analysis in the next section starts by testing whether the ratio of net foreign assets to GNP is stationary; such stationarity would indicate the existence of a long-run equilibrium as a function of stationary variables only. ^15/ In the event that non-stationarity of this ratio cannot be rejected, it will subsequently be assessed whether the ratio is cointegrated with other non-stationary variables suggested by the theoretical considerations presented in the previous section. Such cointegration would suggest that the arguments of the long-run equilibrium also included these other non-stationary variables.

Given the small sample, it is evidently important to follow an approach to testing that has high power to reject non-stationarity. The remainder of this section explains the testing strategy adopted below. Two alternative approaches to testing for cointegration have been proposed by Engle and Granger [1987]. One strategy consists of applying the usual univariate tests for (unit root) non-stationarity to deviations of the data from a long-run cointegrating combination. The other involves testing for the presence of equilibrating feedback in the context of a multivariate dynamic specification. It has been established that the latter approach tends to be more powerful (Banerjee et al. [1986], Engle and Granger [1987]). Therefore, graphical evidence will be examined and univariate tests will be applied, but only as a first step of the empirical analysis; the results of this first step will subsequently be scrutinized within a dynamic framework.

IV. Empirical Results

1. The long run

The post-war pattern of the ratio of net foreign assets to GNP for the United States, Japan and Germany is shown in Chart 1. ^18/ The graphical evidence suggests no sustained tendency on the part of any of these ratios to return to a stationary long-run equilibrium. The net foreign asset positions of Japan and Germany, relative to GNP, have fluctuated along an upward trend, while, in contrast, that of the United States has fallen dramatically in the 1980s after following a more stable pattern in the previous decades.

Chart 1.

Net Foreign Assets (Percent of GNP), 1952 to 1987¹

View Expanded

¹ Source: See Appendix; Data for Japan and Germany begin in 1954 and 1953 respectively, data for Japan end in 1986.

The graphical impression is confirmed by the fact that, for this sample, non-stationarity of the F/Y ratio cannot be rejected for any of the three countries (Table 1). However, the non-stationarity of the first difference of this variable can be rejected, indicating that F/Y is I(1). ^19/ Hence, if a stationary long-run equilibrium exists, then it must be a function of at least one other non-stationary variable.

Table 1.

Tests for Order of Integration ^1/

Source: Appendix I.

^1/This table reports tests for non-stationarity of the variables included in the cointegration tests. The tests are the Sargan-Bhargava (SB) statistic, with critical values in Sargan and Bhargava [1983, Table I] and Bhargava [1986, Table I] and the Augmented Dickey-Fuller (ADF) statistic, with critical values in Fuller [1976, Table 8.5.2]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation.

^2/The sample for each of the ADF tests equals that for the corresponding SB test, minus the observations at the beginning of the sample needed to obtain the number of lags necessary for an ADF test regression without residual autocorrelation.

Table 1.

Tests for Order of Integration ^1/

Variable		Sargan-Bhargava		Augmented Dickey-Fuller 2/	Inference
		Result	Sample
United States
F/Y		0.20	1950-86	-0.03	F/Y at least I(1).
Δ(F/Y)		0.42	1951-86	-0.50	?
	F/Y	0.28	1950-81	-0.03	F/Y at least I(1).
	Δ(F/Y)	1.28*	1951-81	-3.56*	F/Y is I(1).
B/Y		0.05	1950-87	-1.50	B/Y at least I(1).
Δ(B/Y)		0.74*	1951-87	-5.74*	B/Y is I(1).
RDEM1		0.06	1956-86	-2.55	RDEM1 at least I(1).
Δ(RDEM1)		0.20	1957-86	-1.78	RDEM1 at least I(2).
ΔΔ(RDEM1)		2.17*	1958-86	-5.62*	RDEM1 is I(2).
RDEM2		0.02	1956-86	-1.64	RDEM2 at least I(1).
Δ(RDEM2)		0.39	1957-86	-1.96	RDEM2 at least I(2).
ΔΔ(RDEM2)		2.02*	1958-86	-5.18*	RDEM2 is I(2).
Japan
F/Y		0.06	1954-86	-0.76	F/Y at least I(1).
Δ(F/Y)		1.26*	1955-86	-3.82*	F/Y is I(1).
B/Y		0.03	1954-86	-0.68	B/Y at least I(1).
Δ (B/Y)		0.51	1955-86	-2.03	B/Y is I(1)?
ΔΔ(B/Y)		2.56*	1956-86	-7.07*	B/Y is I(1)?
DEM1		0.06	1956-86	-4.28*	?
Δ(RDEM1)		0.21	1957-86	-1.42	RDEM1 at least I(2).
ΔΔ(RDEM1)		2.17*	1958-86	-5.56*	RDEM1 is I(2).
RDEM2		0.02	1956-86	-2.78	RDEM2 at least I(1).
Δ(RDEM2)		0.34	1957-86	-2.77	RDEM2 at last I(2).
ΔΔ(RDEM2)		2.88*	1958-86	-8.17*	RDEM2 is I(2).
Germany
F/Y		0.18	1955-86	-2.11	F/Y at least I(1).
Δ(F/Y)		1.04*	1955-86	-3.08*	F/Y is I(1).
B/Y		0.04	1955-86	0.13	B/Y at least I(1).
Δ(B/Y)		0.89*	1955-86	-3.19*	B/Y is I(1).
RDEM1		0.04	1956-86	-3.84*	?
Δ(RDEM1)		0.12	1957-86	-1.83	RDEM1 at least I(2).
ΔΔ(RDEM1)		1.13*	1958-86	-2.95*	RDEM1 is I(2).
RDEM2		0.04	1956-86	-2.47	RDEM2 at least I(1).
Δ(RDEM2)		0.19	1957-86	-1.11	RDEM2 at least I(2).
ΔΔ(RDEM2)		1.76*	1958-86	-4.60*	RDEM2 is I(2).

Source: Appendix I.

^1/This table reports tests for non-stationarity of the variables included in the cointegration tests. The tests are the Sargan-Bhargava (SB) statistic, with critical values in Sargan and Bhargava [1983, Table I] and Bhargava [1986, Table I] and the Augmented Dickey-Fuller (ADF) statistic, with critical values in Fuller [1976, Table 8.5.2]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation.

^2/The sample for each of the ADF tests equals that for the corresponding SB test, minus the observations at the beginning of the sample needed to obtain the number of lags necessary for an ADF test regression without residual autocorrelation.

View Table

Table 1.

Tests for Order of Integration ^1/

Variable		Sargan-Bhargava		Augmented Dickey-Fuller 2/	Inference
		Result	Sample
United States
F/Y		0.20	1950-86	-0.03	F/Y at least I(1).
Δ(F/Y)		0.42	1951-86	-0.50	?
	F/Y	0.28	1950-81	-0.03	F/Y at least I(1).
	Δ(F/Y)	1.28*	1951-81	-3.56*	F/Y is I(1).
B/Y		0.05	1950-87	-1.50	B/Y at least I(1).
Δ(B/Y)		0.74*	1951-87	-5.74*	B/Y is I(1).
RDEM1		0.06	1956-86	-2.55	RDEM1 at least I(1).
Δ(RDEM1)		0.20	1957-86	-1.78	RDEM1 at least I(2).
ΔΔ(RDEM1)		2.17*	1958-86	-5.62*	RDEM1 is I(2).
RDEM2		0.02	1956-86	-1.64	RDEM2 at least I(1).
Δ(RDEM2)		0.39	1957-86	-1.96	RDEM2 at least I(2).
ΔΔ(RDEM2)		2.02*	1958-86	-5.18*	RDEM2 is I(2).
Japan
F/Y		0.06	1954-86	-0.76	F/Y at least I(1).
Δ(F/Y)		1.26*	1955-86	-3.82*	F/Y is I(1).
B/Y		0.03	1954-86	-0.68	B/Y at least I(1).
Δ (B/Y)		0.51	1955-86	-2.03	B/Y is I(1)?
ΔΔ(B/Y)		2.56*	1956-86	-7.07*	B/Y is I(1)?
DEM1		0.06	1956-86	-4.28*	?
Δ(RDEM1)		0.21	1957-86	-1.42	RDEM1 at least I(2).
ΔΔ(RDEM1)		2.17*	1958-86	-5.56*	RDEM1 is I(2).
RDEM2		0.02	1956-86	-2.78	RDEM2 at least I(1).
Δ(RDEM2)		0.34	1957-86	-2.77	RDEM2 at last I(2).
ΔΔ(RDEM2)		2.88*	1958-86	-8.17*	RDEM2 is I(2).
Germany
F/Y		0.18	1955-86	-2.11	F/Y at least I(1).
Δ(F/Y)		1.04*	1955-86	-3.08*	F/Y is I(1).
B/Y		0.04	1955-86	0.13	B/Y at least I(1).
Δ(B/Y)		0.89*	1955-86	-3.19*	B/Y is I(1).
RDEM1		0.04	1956-86	-3.84*	?
Δ(RDEM1)		0.12	1957-86	-1.83	RDEM1 at least I(2).
ΔΔ(RDEM1)		1.13*	1958-86	-2.95*	RDEM1 is I(2).
RDEM2		0.04	1956-86	-2.47	RDEM2 at least I(1).
Δ(RDEM2)		0.19	1957-86	-1.11	RDEM2 at least I(2).
ΔΔ(RDEM2)		1.76*	1958-86	-4.60*	RDEM2 is I(2).

Source: Appendix I.

^1/This table reports tests for non-stationarity of the variables included in the cointegration tests. The tests are the Sargan-Bhargava (SB) statistic, with critical values in Sargan and Bhargava [1983, Table I] and Bhargava [1986, Table I] and the Augmented Dickey-Fuller (ADF) statistic, with critical values in Fuller [1976, Table 8.5.2]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation.

^2/The sample for each of the ADF tests equals that for the corresponding SB test, minus the observations at the beginning of the sample needed to obtain the number of lags necessary for an ADF test regression without residual autocorrelation.

In order to explore this possibility, we examine the cointegration of F/Y with government debt stocks and demographic variables. The theoretical analysis of Section II suggests that, in equilibrium, higher net foreign assets (scaled by GNP) would be associated with lower government debt at home and higher government debt abroad (equation (13)). Given that comparable data on government debt is not available for a reasonably large number of industrial countries over the time period of interest to us, we focus on the possible cointegration of F/Y with the domestic debt-to-GNP ratio. ^20/ The two demographic variables included in the analysis are the foreign ratio of the population aged under 15 years to that from 15 to 64 years, divided by the corresponding ratio for the home country (RDEM1), and an analogous ratio for the population aged 65 years and over (RDEM2).

The first step is to ascertain that all these variables are indeed I(1). The non-stationarity of B/Y cannot be rejected but that of Δ(B/Y) can be (Table 1), confirming that B/Y is I(1). ^21/ But, for the sample under investigation, each of the demographic variables appears to be I(2). Still, the finding that, for each of the three countries, a linear combination of the pair of relative demographic variables seems to be I(1), justifies including them jointly in a cointegrating combination with other I(1) variables (F/Y and B/Y). ^22/

The tests for cointegration of the set of variables comprising F/Y, B/Y, RDEM1, and RDEM2 are presented in Table 2. In the cases of Japan and Germany, the ADF tests reject the null hypothesis of no cointegration at a significance level of 5 percent (for Germany, the regression includes a trend term as well); for the United States the significance level is somewhat higher. In all three cases the SB test statistics lie in the 5 percent inconclusive region, but, given the relatively large number of variables included, this region is very wide. Taken together, this evidence suggests that, for the three countries under investigation, a stationary long-run equilibrium for the F/Y ratio exists, and that it is a function of B/Y and demographic factors. Of course, as mentioned in Section III, the equilibrium may depend on one or more I(0) variables as well--this will be examined in the dynamic analysis.

Table 2.

Cointegration Tests ^1/

Source: Appendix I.

^1/This table reports static OLS regressions of the ratio of net foreign assets to GNP (F/Y) on: a constant; the ratio of government debt to GNP (B/Y); the foreign ratio of the population aged under 15 years to that from 15 to 64 years relative to the corresponding ratio in the home country (RDEM1); the foreign ratio of the population aged 65 years and over to that aged from 15 to 64 years relative to the corresponding ratio for the home country (RDEM2); and, for Germany, a trend term. The data are annual for the period 1956-86. The tests for non-stationarity of the residuals from these regressions are given by the Sargan-Bhargava test statistic, with critical values in Sargan and Bhargava [1983, Table I] and the Augmented Dickey-Fuller statistic, with critical values in Engle and Yoo [1987, Table 3]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation. For further explanations, see Table 1.

^2/Test result lies in the 5 percent inconclusive region of Sargan and Bhargava [1983, Table I]. Given the large number of regressors, these regions are very wide.

Table 2.

Cointegration Tests ^1/

Variables included in regression					Sargan-Bhargava	Augmented Dickey-Fuller
Constant	B/Y	RDEM1	RDEM2	TREND
United States
0.62	-0.44	0.46	-0.87		0.62 2/	-3.51
(0.06)	(0.05)	(0.08)	(0.09)
Japan
0.43	-0.21	0.18	-0.39		0.94 2/	-4.22*
(0.06)	(0.06)	(0.02)	(0.03)
Germany
-0.076	-0.19	-0.22	0.49	0.004	0.67 2/	-4.63*
(0.086)	(0.15)	(0.14)	(0.31)	(0.001)

Source: Appendix I.

^1/This table reports static OLS regressions of the ratio of net foreign assets to GNP (F/Y) on: a constant; the ratio of government debt to GNP (B/Y); the foreign ratio of the population aged under 15 years to that from 15 to 64 years relative to the corresponding ratio in the home country (RDEM1); the foreign ratio of the population aged 65 years and over to that aged from 15 to 64 years relative to the corresponding ratio for the home country (RDEM2); and, for Germany, a trend term. The data are annual for the period 1956-86. The tests for non-stationarity of the residuals from these regressions are given by the Sargan-Bhargava test statistic, with critical values in Sargan and Bhargava [1983, Table I] and the Augmented Dickey-Fuller statistic, with critical values in Engle and Yoo [1987, Table 3]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation. For further explanations, see Table 1.

^2/Test result lies in the 5 percent inconclusive region of Sargan and Bhargava [1983, Table I]. Given the large number of regressors, these regions are very wide.

View Table

Table 2.

Cointegration Tests ^1/

Variables included in regression					Sargan-Bhargava	Augmented Dickey-Fuller
Constant	B/Y	RDEM1	RDEM2	TREND
United States
0.62	-0.44	0.46	-0.87		0.62 2/	-3.51
(0.06)	(0.05)	(0.08)	(0.09)
Japan
0.43	-0.21	0.18	-0.39		0.94 2/	-4.22*
(0.06)	(0.06)	(0.02)	(0.03)
Germany
-0.076	-0.19	-0.22	0.49	0.004	0.67 2/	-4.63*
(0.086)	(0.15)	(0.14)	(0.31)	(0.001)

Source: Appendix I.

^1/This table reports static OLS regressions of the ratio of net foreign assets to GNP (F/Y) on: a constant; the ratio of government debt to GNP (B/Y); the foreign ratio of the population aged under 15 years to that from 15 to 64 years relative to the corresponding ratio in the home country (RDEM1); the foreign ratio of the population aged 65 years and over to that aged from 15 to 64 years relative to the corresponding ratio for the home country (RDEM2); and, for Germany, a trend term. The data are annual for the period 1956-86. The tests for non-stationarity of the residuals from these regressions are given by the Sargan-Bhargava test statistic, with critical values in Sargan and Bhargava [1983, Table I] and the Augmented Dickey-Fuller statistic, with critical values in Engle and Yoo [1987, Table 3]. Statistics suggesting rejection of non-stationarity at the 5 percent significance level are marked by an asterisk (*). The ADF auxiliary test regressions (not reported) were checked for the absence of autocorrelation. For further explanations, see Table 1.

^2/Test result lies in the 5 percent inconclusive region of Sargan and Bhargava [1983, Table I]. Given the large number of regressors, these regions are very wide.

Even though the coefficient estimates obtained from the static regressions in Table 2 are biased (given the autocorrelation induced by the absence of dynamics and other explanatory variables), they are consistent ^23/ and it is therefore noteworthy that the coefficients on B/Y are of the hypothesized sign. In addition, they are of very similar magnitude for Japan and Germany, while the coefficient for the United States is twice as large in magnitude. As regards the influence of demographic variables, it appears that, for the United States and Japan, relatively high average age of the population corresponds to high net foreign assets; this result seems consistent with Lee and Lapkoff [1988], cited above. However, for Germany the relation appears to be the reverse.

In order to help understanding of the implications of the estimated equation for long-run equilibrium of the ratio of U.S. net foreign assets to GNP, we present actual and fitted values, given in Chart 2. It appears that this equilibrium, which is conditional on the paths of B/Y and the demographic variables, and which assumes that any possible I(0) influences are zero, sharply declined in the 1980s. According to the estimated coefficients, this primarily reflected the change in stance of U.S. fiscal policy, as evidenced by Chart 3 which decomposes the hypothetical long-run equilibrium into the relative contributions of the debt and demographic variables. During the 1950s and the 1960s, a fairly stable F/Y ratio was the result of offsetting forces: a declining debt-to-GNP ratio and a more rapidly aging population than in other countries. Both the B/Y ratio and the combined effect of demographic factors stabilized in the 1970s, but in the 1980s large Federal deficits had the effect of rapidly increasing the government’s indebtedness. In the absence of counterbalancing movements in demographic variables, the implied equilibrium for net foreign assets became strongly negative. Since the equation overpredicts at the end of the period, the actual net foreign asset position of the United States in 1986 was even more negative than this conditional long-run equilibrium would imply.

Chart 2

UNITED STATES: Actual and equilibrium values of net foreign assets, 1956-86 ^1/

View Expanded

Source: Appendix I and Table 2. ^1/ Conditional on demographics and the ratio of government debt to GNP. 2/ Coefficients are in Table 2; determinants are in Chart 3. Chart 3

UNITED STATES: The determinants of net foreign asset equilibrium, 1956-86 ^1/

View Expanded

Source: Table 2. ^1/ This chart shows the contribution of demographics and government debt to changes since 1956 in the conditional net foreign asset equilibrium depicted in Chart 2. 2/ B/Y multiplied by its coefficient from Table 2, shifted so that 1956=0. 3/ Linear combination of RDEM1 and RDEM2 using the coefficients from Table 2, shifted so that 1956=0.

In contrast, the long-run equilibrium of the Japanese net foreign asset position has shown an upward movement throughout most of the postwar period (Chart 4), reflecting the aging of the population in combination, after the mid-1970s, with fiscal consolidation (Chart 5). In 1986, its net foreign asset position lay above this hypothetical equilibrium.

Chart 4

JAPAN: Actual and equilibrium values of net foreign assets, 1956-86 ^1/

View Expanded

Source: Appendix I and Table 2. ^1/ Conditional on demographics and the ratio of government debt to GNP. 2/ Coefficients are in Table 2; determinants are in Chart 5. Chart 5

JAPAN: The determinants of net foreign asset equilibrium, 1956-86 ^1/

View Expanded

Source: Table 2. ^1/ This chart shows the contribution of demographics and government debt to changes since 1956 in the conditional net foreign asset equilibrium depicted in Chart 4. 2/ Linear combination of RDEM1 and RDEM2 using the coefficients from Table 2, shifted so that 1956=0. 3/ B/Y multiplied by its coefficient from Table 2, shifted so that 1956=0.

In Germany, before the mid-1970s the debt and demographic variables made virtually no contribution to the rising net foreign asset equilibrium (Charts 6 and 7), the rise reflecting other influences which are proxied by the trend term. ^24/ From the mid-1970s onward, however, the net foreign asset equilibrium showed a downward movement as the government budget turned into deficit and the average age of the German population rose slightly relative to other countries. More recently, the government has stabilized B/Y and the net foreign asset equilibrium has resumed its upward path.

Chart 6

GERMANY: Actual and equilibrium values of net foreign assets, 1956-86 ^1/

View Expanded

Source: Appendix I and Table 2. ^1/ Conditional on demographics and the ratio of government debt to GNP. 2/ Coefficients are in Table 2; determinants are in Chart 7. Chart 7

GERMANY: The determinants of net foreign asset equilibrium, 1956-86 ^1/

View Expanded

Source: Table 2. ^1/ This chart shows the contribution of demographics and government debt to changes since 1956 in the conditional net foreign asset equilibrium depicted in Chart 6. 2/ Trend term multiplied by its coefficient in Table 2, shifted so that 1956=0. 3/ Linear combination of RDEM1 and RDEM2 using the coefficients from Table 2, shifted so that 1956=0. 4/ B/Y multiplied by its coefficient from Table 2, shifted so that 1956=0.

Of course, given the relatively small sample upon which the static regressions are based, the implications for long-run equilibrium of net foreign asset positions should be interpreted with caution. Nevertheless, the dynamic analysis does provide strong supporting evidence.

2. The dynamics

The purpose of this sub-section is two-fold. First, dynamic error correction models are estimated in order to scrutinize in a more powerful fashion the conclusions regarding long-run foreign asset equilibrium reached above. Second, by estimating these models some light is shed on the dynamic forces underlying the long-run equilibrium, i.e., the adjustment mechanisms discussed in Section II.

The conceptual background for the dynamic analysis is equation (14) of the theoretical model, which expresses foreign assets and other variables as adjusting toward the long-run steady-state equilibrium. Rather than estimating a complete reduced-form multi-country model, however, we examine whether significant foreign asset disequilibrium feedback is discernible in selected structural equations of the model. In order to be able to distinguish between different channels of feedback, we transform (14) so as to arrive at a semi-reduced form equation for foreign assets conditional both on foreign asset disequilibrium and on several variables usually present in models of the current account, particularly domestic and foreign absorption, and the real exchange rate. The choice of this specification is motivated by the desire to identify “expenditure changing” channels (through the possible influence of changes in net wealth on relative absorption) and “expenditure-switching” channels (through the real exchange rate). Hence, the disequilibrium feedback term in this equation may be interpreted as capturing other channels that may be significant empirically (for example, supply side effects and non-price competitiveness factors such as non-tariff barriers, which are not modeled explicitly). ^25/

The common specification of the dynamic net foreign asset equation estimated for each of the three countries is:

$\begin{array}{cc}\mathrm{\Delta }\left(\mathrm{F}/\mathrm{Y}\right)={\mathrm{c}}_0-{\mathrm{c}}_1{\hat{\mathrm{u}}}_{-1}-{\mathrm{c}}_2{\left[ln\left(\mathrm{A}/\mathrm{Y}\right)-ln\left(\mathrm{AF}/\mathrm{YF}\right)\right]}_{-1}-{\mathrm{c}}_{3}ln{\left(\mathrm{RE}\right)}_{-1}& \left(15\right)\end{array}$

The variable û represents net foreign asset disequilibrium (the residuals from the cointegration regressions reported in Table 2); A and Y are domestic absorption and GNP, respectively; AF and YF are their foreign equivalents; and RE is the real effective exchange rate (a rise signifying appreciation). ^26/ The feedback channel through absorption is subsequently explored further by testing whether the latter is influenced by foreign asset disequilibrium. We do not attempt to model exchange rates, given the lack of success that has plagued other research in this area.

Within the confines of the analytical framework, the choice both of additional explanatory variables and between lag structures was guided mainly by the ability of alternative specifications to satisfy a set of diagnostic tests. In addition to the tests for residual autocorrelation, heteroscedasticity and non-normality which are reported below, each regression was scrutinized for the stability over time of each of its estimated coefficients and of its estimated equation standard error. ^27/ All regressions are OLS; regressions containing current-dated variables were checked for simultaneity. ^28/ The standard errors of the coefficient estimates (reported in parentheses) are heteroscedasticity-consistent (White [1980]).

The main finding of the dynamic analysis is that error correction feedback from foreign asset disequilibrium is highly significant for each of the three countries, both in the equations for net foreign assets and in those for absorption. Moreover, the estimated coefficients on the disequilibrium terms tend to be large, particularly for Japan and Germany. In none of the dynamic equations can the parameter restrictions incorporated in the error correction terms be rejected. The dynamic analysis thus provides strong support for the cointegration findings of the previous section.

a. United States

The estimation of (15) for the United States produced an equation for net foreign assets with statistically significant influences from each of the variables, but the equation failed some of the diagnostic tests. This appeared to be due to the omission of two other dynamic variables, namely the change of the ratio of public debt to GNP and the lagged differential between real interest rates at home and abroad. These variables may reflect mis-specification or mis-measurement of the absorption and real exchange rate channels. For instance, if trade is dominated by goods which are sensitive to interest rates, then the use of aggregate absorption may fail to capture properly the effects on net foreign assets that operate through this channel. The preferred equation was the following:

$\begin{array}{l}\begin{array}{cccc}\mathrm{\Delta }(\mathrm{F}/\mathrm{Y})& =& -0.0014-0.34{\hat{\mathrm{u}}}_{-1}-\underset{\left(0.05\right)}{0.43}[\ln (\mathrm{A}/\mathrm{Y})-\ln (\mathrm{AF}/\mathrm{YF}){]}_{-1}& \left(16\right)\\ & & \underset{\left(0.007\right)}{\begin{array}{c}-0.014\end{array}}\ln (\mathrm{RE}{)}_{-1}-\underset{\left(0.06\right)}{0.14}\mathrm{\Delta }(\mathrm{B}/\mathrm{Y})-\underset{\left(0.06\right)}{0.15}(\mathrm{R}-\mathrm{RF}{)}_{-1}& \end{array}\\ \begin{array}{ccc}\mathrm{T}=1957-86& {\mathrm{R}}^2=0.87& \hat{\sigma }=0.0044\end{array}\\ \begin{array}{cc}\mathrm{AR}{(3,21)}_1^3=1.57\left(3.07\right)& \mathrm{AR}{\left(1.23\right)}_1^1=0.04\left(4.28\right)\end{array}\\ \begin{array}{cc}\mathrm{ARCH}(2,20)=0.04\left(3.49\right)& \mathrm{HET}(10,13)=0.65\left(2.67\right)\end{array}\\ \mathrm{NORM}\left(2\right)=0.40\left(5.99\right)\end{array}$

The variable R represents the real interest rate at home, and RF is its foreign counterpart. This equation satisfies all the diagnostic tests, and recursive estimation reveals no parameter instability. Chart 8 shows the actual and fitted values. ^29/ The error correction variable is quite significant: its t-value is 4.8. An F-test of the parameter restriction incorporated in the error correction variable (testing the joint significance of (B/Y)_-1, RDEM1_-1, and RDEM2_-1) has the value 1.1, which, given the 5 percent critical value of 3.1, clearly validates the estimate of the cointegrating vector obtained in Table 2.

Chart 8

UNITED STATES: Actual and fitted values of the model for net foreign assets ^1/

View Expanded

Source: Equation (16). ^1/ Estimation period is 1957-86.

We did not succeed in finding a model for U.S. absorption with stable parameters covering the entire sample period 1957-86, but a satisfactory model for the period until the early 1980s is:

$\begin{array}{l}\begin{array}{cc}\mathrm{\Delta }(\mathrm{A}/\mathrm{Y})=\underset{\left(0.15\right)}{0.66}+\underset{\left(0.11\right)}{0.36}{\hat{\mathrm{u}}}_{-1}-\underset{\left(0.15\right)}{0.66}(\mathrm{A}/\mathrm{Y}{)}_{-1}-\underset{\left(0.05\right)}{0.16}\mathrm{R}& \left(17\right)\end{array}\hfill \\ \begin{array}{ccc}\mathrm{T}=1957-81& {\mathrm{R}}^2=0.59& \hat{\sigma }=0.0036\end{array}\hfill \\ \begin{array}{cc}\mathrm{AR}{(3,18)}_1^3=0.48\left(3.16\right)& \mathrm{AR}{(1,20)}_1^1=0.25\left(4.35\right)\end{array}\hfill \\ \begin{array}{cc}\begin{array}{c}\mathrm{ARCH}(2,17)=0.14\left(3.59\right)\end{array}& \mathrm{HET}(6,14)=1.08\left(2.85\right)\end{array}\hfill \\ \mathrm{NORM}\left(2\right)=0.79\left(5.99\right)\end{array}$

Again, the t-value on û_-1 is large (3.2), and the cointegration restriction is easily satisfied (the F-test is 0.1 against the 5 percent critical value of 3.2). Chart 9 displays the actual and fitted values for 1957-81, as well as the conditional forecasts for 1982-86. Although it passes parameter stability tests prior to 1982, the model subsequently dramatically underpredicts U.S. absorption relative to GNP. The large Federal deficits during the Reagan administration apparently not only raised the public debt to GNP ratio and thus lowered the conditional net foreign asset equilibrium (Chart 2), but, during the same episode, the ratio of absorption to GNP failed to fall despite low (lagged) net foreign assets, a high (lagged) absorption-to-GNP ratio, and high real interest rates. As far as the public part of absorption is concerned, this apparent change of regime is consistent with the finding of Kremers [1989], that after 1981 high Federal debt interest costs failed to produce deficit reductions in line with the pattern during most of the inter- and postwar period.

Chart 9

UNITED STATES: Actual, fitted and forecast values of the model for absorption ^1/

View Expanded

Source: Equation (17). ^1/ Estimation period is 1957-81; conditional forecasts for 1982-86.

b. Japan

The estimated dynamic equation for net foreign assets of Japan is:

$\begin{array}{c}\begin{array}{ccc}\mathrm{\Delta }(\mathrm{F}/\mathrm{Y})=& -\underset{\left(0.0015\right)}{0.0018}-\underset{\left(0.15\right)}{0.71}{\hat{\mathrm{u}}}_{-1}-\underset{\left(0.16\right)}{0.48}[\ln (\mathrm{A}/\mathrm{Y})-\ln (\mathrm{AF}/\mathrm{YF}){]}_{-1}\hfill & \left(18\right)\\ & \underset{\left(0.006\right)}{\begin{array}{c}-0.039\end{array}}\ln (\mathrm{RE}{)}_{-1}-\underset{\left(0.012\right)}{0.034}\mathrm{\Delta \Delta ln}\left(\mathrm{RE}\right)\hfill & \\ & +\underset{\left(0.14\right)}{0.40}\mathrm{\Delta }[{(\mathrm{B}/\mathrm{Y})}_{-1}^{\mathrm{US}}+{(\mathrm{B}/\mathrm{Y})}_{-2}^{\mathrm{US}}]/2\hfill & \end{array}\hfill \\ \begin{array}{ccc}\mathrm{T}=1957-86& {\mathrm{R}}^2=0.68& \hat{\sigma }=0.0072\end{array}\hfill \\ \begin{array}{cc}\mathrm{AR}{(3,21)}_1^3-0.21\left(3.07\right)& \mathrm{AR}{(1,23)}_1^1-0.00\left(4.28\right)\end{array}\hfill \\ \begin{array}{cc}\mathrm{ARCH}(2,20)=0.91\left(3.49\right)& \mathrm{HET}(10,13)=0.23\left(2.67\right)\end{array}\hfill \\ \mathrm{NORM}\left(2\right)=0.11\left(5.99\right)\hfill \end{array}$

This equation satisfies all the diagnostic criteria and its parameters seem to be stable; actual and fitted values are in Chart 10. In common with the U.S. results, the error correction term is significantly negative (the t-value is 4.7), and the cointegration restriction can not be rejected (the F-test is 1.6 against a 5 percent critical value of 3.1). The influences of relative absorption and the real exchange rate are also similar to those found for the United States. ^30/ One difference is the presence, prompted by an examination of various alternatives, of a foreign (U.S.) rather than domestic public debt variable. A domestic debt variable is already included in u; as described in Section II, in a symmetric two-country model, foreign debt should also appear (with opposite sign), and the U.S. debt variable may be playing this role.

Chart 10

JAPAN: Actual and fitted values of the model for net foreign assets ^1/

View Expanded

Source: Equation (18). ^1/ Estimation period is 1957-86.

Feedback from foreign asset disequilibrium was found to affect absorption in Japan as well:

$\begin{array}{c}\begin{array}{ccc}\mathrm{\Delta }(\mathrm{A}/\mathrm{Y})=& -\underset{\left(0.0019\right)}{0.0033}+\underset{\left(0.15\right)}{0.56}{\hat{u}}_{-1}-\underset{\left(0.11\right)}{0.21}\mathrm{\Delta }(\mathrm{A}/\mathrm{Y}{)}_{-2}\hfill & \left(19\right)\\ & -\underset{\left(0.07\right)}{0.24}{\mathrm{\Delta R}}_{-1}+\underset{\left(0.11\right)}{0.36}\mathrm{\Delta \Delta }(\mathrm{B}/\mathrm{Y}{)}_{-1}+\underset{\left(0.07\right)}{0.19}\mathrm{\Delta }(\mathrm{B}/\mathrm{Y}{)}_{-1}\hfill & \end{array}\hfill \\ \begin{array}{ccc}\mathrm{T}=1957-87\hfill & {\mathrm{R}}^2=0.68\hfill & \hat{\sigma }=0.0082\hfill \end{array}\hfill \\ \begin{array}{cc}\mathrm{AR}{(3,22)}_1^3=0.91\left(3.05\right)& \mathrm{AR}{\left(1.24\right)}_1^1=2.98\left(4.26\right)\end{array}\hfill \\ \begin{array}{cc}\mathrm{ARCH}(2,21)=0.48\left(3.47\right)& \mathrm{HET}(10,14)=0.24\left(2.60\right)\end{array}\hfill \\ \mathrm{NORM}\left(2\right)=1.08\left(5.99\right)\hfill \end{array}$

The t-value on û_-1 is 3.8, and the F-test for the cointegration restriction is 0.8 (the 5 percent critical value is 3.1). This is the preferred equation for Japanese absorption conditional on the set of variables involved in this study. It passes the reported tests, the estimated equation standard error appears to be stable, and its tracking performance is satisfactory (Chart 11), but, in the earlier part of the sample, some of the coefficient estimates exhibit signs of instability. However, the apparent instability seems to affect neither the signs of any of the coefficient estimates nor the inference regarding feedback from foreign asset disequilibrium.

Chart 11

JAPAN: Actual and fitted values of the model for absorption ^1/

View Expanded

Source: Equation (19). ^1/ Estimation period is 1957-87.

c. Germany

The net foreign asset equation for Germany is essentially similar to those for the United States and Japan, except that two variables were added to capture some of the dynamic effects of large terms-of-trade changes that have resulted especially from oil price changes: ^31/

$\begin{array}{c}\begin{array}{ccc}\mathrm{\Delta }(\mathrm{F}/\mathrm{Y})=& \underset{\left(0.002\right)}{\begin{array}{c}-0.010\end{array}}-\underset{\left(0.07\right)}{0.60}{\hat{\mathrm{u}}}_{-2}-\underset{\left(0.06\right)}{0.59}[\ln (\mathrm{A}/\mathrm{Y})-\ln (\mathrm{AF}/\mathrm{YF}){]}_{-1}\hfill & \left(20\right)\\ & -\underset{\left(0.015\right)}{0.065}\ln (\mathrm{RE}{)}_{-1}-\underset{\left(0.03\right)}{0.15}\mathrm{\Delta \Delta ln}\left(\mathrm{A}\right)\hfill & \\ & +\underset{\left(0.026\right)}{0.052}\mathrm{\Delta \Delta ln}(\mathrm{EUV}{)}_{-1}+\underset{\left(0.004\right)}{0.014}\mathrm{\Delta ln}(\mathrm{EUV}/\mathrm{POIL}{)}_{-1}\hfill & \end{array}\\ \begin{array}{ccc}\mathrm{T}=1958-87& {\mathrm{R}}^2=0.90& \hat{\sigma }=0.0046\end{array}\hfill \\ \begin{array}{cc}\mathrm{AR}{(3,19)}_1^3=0.90\left(3.13\right)& \mathrm{AR}{(1,21)}_1^1=0.51\left(4.32\right)\end{array}\hfill \\ \begin{array}{cc}\mathrm{ARCH}(2,18)=0.77\left(3.55\right)& \mathrm{HET}(12,10)=0.67\left(2.91\right)\end{array}\hfill \\ \mathrm{NORM}\left(2\right)=0.84\left(5.99\right)\hfill \end{array}$

The variable EUV represents an index of German export unit values and POIL is the price of oil (both in deutsche mark). The diagnostic tests are satisfied and the parameter estimates appear stable. The t-value for net foreign asset disequilibrium feedback is very high (8.6), and, at a value of 0.7, an F-test cannot reject the cointegration restriction (the 5 percent critical value is 2.9). Actual and fitted values are in Chart 12.

Chart 12

GERMANY: Actual and fitted values of the model for net foreign assets ^1/

View Expanded

Source: Equation (20). ^1/ Estimation period is 1958-87. 2/ Dummy variable included for 1985.

Finally, feedback from foreign asset disequilibrium is also identified as a determinant of absorption in Germany:

$\begin{array}{c}\begin{array}{ccc}\mathrm{\Delta }(\mathrm{A}/\mathrm{Y})=& -\underset{\left(0.002\right)}{0.006}+\underset{\left(0.12\right)}{0.72}{\hat{\mathrm{u}}}_{-2}-\underset{\left(0.14\right)}{0.60}\mathrm{\Delta }(\mathrm{A}/\mathrm{Y}{)}_{-2}\hfill & \left(21\right)\\ & -\underset{\left(0.14\right)}{0.53}\mathrm{\Delta \Delta R}+\underset{\left(0.05\right)}{0.17}\mathrm{\Delta ln}(\mathrm{M}1{)}_{-1}\hfill & \end{array}\\ \begin{array}{ccc}\mathrm{T}=1963-87& {\mathrm{R}}^2=0.76& \hat{\sigma }=0.0068\end{array}\hfill \\ \begin{array}{cc}\mathrm{AR}{(3,17)}_1^3=0.48\left(3.20\right)& \mathrm{AR}{\left(1.19\right)}_1^1=0.86\left(4.38\right)\end{array}\hfill \\ \begin{array}{cc}\mathrm{ARCH}(2,16)=1.03\left(3.63\right)& \mathrm{HET}(8,11)=0.25\left(2.95\right)\end{array}\hfill \\ \mathrm{NORM}\left(2\right)=0.70\left(5.99\right)\hfill \end{array}$

The variable M1 represents the M1 money stock, divided by the GNP deflator. ^32/ In the case of Germany, it appears to be appropriate to express both A and Y in constant prices (they were divided by their respective price deflators). The t-value on û_-2 is high (6.2), and the F-test for the cointegration restriction is 0.7 against the 5 percent critical value of 3.0. The tracking performance of this equation is displayed in Chart 13.

Chart 13

GERMANY: Actual and fitted values of the model for absorption ^1/

View Expanded

Source: Equation (21). ^1/ Estimation period is 1963-87.

V. Conclusions

This study has sought to confirm the presence of stabilizing mechanisms for the current account positions of the three largest industrial countries in the post-World War II period. Persistent and large current account imbalances have led to quite dramatic changes in the net international positions of the United States, Japan, and Germany. However, various feedback effects from foreign asset stocks onto variables such as domestic absorption, real exchange rates and real interest rates may act as stabilizing mechanisms to prevent a continued increase of these net foreign asset or liability positions, and to ensure an eventual return to long-run equilibrium.

A theoretical model of optimizing consumers was specified in which long-run net foreign asset equilibrium, relative to GNP, was shown to be related, inter alia, to government debt. Cointegration tests were used to investigate empirically the nature of the long-run equilibrium relationship for the United States, Japan and Germany using postwar data. These tests suggest the existence of a long-run relationship between the net foreign asset/GNP ratio, the public debt/GNP ratio, and demographic factors.

Dynamic error correction models for net foreign assets and domestic absorption provide strong support for the inference based on the cointegration tests: in each of the three countries deviations from the long-run equilibrium net foreign asset ratio appear to exercise significant, stabilizing feedback on current account imbalances through domestic absorption and other channels. Dynamic equations for net foreign assets of the United States, Japan and Germany as well as equations for absorption in Japan and Germany satisfy a broad set of diagnostic criteria, including tests for parameter stability. However, a model for U.S. absorption based upon data covering most of the postwar period satisfies the requirement of parameter stability only prior to the 1980s. After 1981, it predicts values of domestic absorption significantly lower than those actually recorded.

Developments in the 1980s therefore seem to have differed between the United States on the one hand and Japan and Germany on the other hand in two respects. First, the conditional long-run net foreign asset equilibrium of the United States swung very sharply negative, reflecting an accumulation of public debt at a pace so rapid that it represented a break with the policies of previous decades. In contrast, the Japanese and German authorities implemented policies to consolidate the public finances, and the net foreign assets of these countries increased strongly. Second, the dynamic behavior of U.S. absorption that prevailed during most of the postwar period changed significantly in the 1980s, giving rise to an historically high level of absorption relative to income. In contrast, no such behavioral changes seem to have taken place in Japan or Germany. Nevertheless, the further evolution of developments in the United States may naturally have repercussions for other countries, including Japan and Germany, even if these countries have not as yet experienced any significant behavioral changes themselves.

Against that background, it will be of interest to extend this research by analyzing in greater detail the various foreign asset feedback mechanisms. ^33/ Specifically, a more profound understanding of the issues raised in the previous paragraph may result from a breakdown of absorption into public and private components.

The empirical findings also have some bearing on the issue of the correlation between saving and investment. The analysis of this paper may be viewed as the stock analog of recent contributions by Feldstein and Horioka [1980] and others on international saving-investment flows. Evidence of net foreign asset feedback may help explain why time series data show a high correlation between saving and investment. A shock to either saving or investment, by leading to an accumulation of foreign assets or liabilities, will bring about forces that tend to feed back onto absorption and output, in such a way that over time the effect of the shock is reversed. The faster these mechanisms operate, the less likely is the emergence of large current account surplus and deficit positions.