This section reports empirical results from several different approaches to modeling the determination of the current account for a set of 21 industrial countries.1 The ultimate objective is to develop a measure for “normal” or “equilibrium” current account balances from a medium-term perspective. Because the current account must equal the difference between domestic saving and investment (i.e., the saving-investment balance), current account developments are examined from the perspective of the fundamental medium- to long-run determinants of saving and investment behavior.

This section reports empirical results from several different approaches to modeling the determination of the current account for a set of 21 industrial countries.1 The ultimate objective is to develop a measure for “normal” or “equilibrium” current account balances from a medium-term perspective. Because the current account must equal the difference between domestic saving and investment (i.e., the saving-investment balance), current account developments are examined from the perspective of the fundamental medium- to long-run determinants of saving and investment behavior.

Using cross-sectional and panel data, the empirical analysis examines to what extent a common set of underlying determinants has been relevant historically in explaining current account balances across countries and over time.2 Using various economic theories of saving and investment as a guide, the analysis investigates the role of the stage of development, demographics, macroeconomic policies, and other considerations in underpinning longer-term saving-investment balances. Using a systems approach, the current account model is further examined in the context of joint estimation of individual equations for saving, investment, and the saving-investment balance.

The findings can be summarized as follows. In the cross section, we find a significant effect of the stage of development on the current account. The general finding is that relatively more advanced economies are more likely to run a smaller deficit or a larger surplus, other things equal. We also find an effect of demographics on the long-run current account position. A country that has an above-average dependency ratio (ratio of dependents to the working age population) tends to have a larger current account deficit. In the cross section, we find little impact of the budget position in explaining current account balances across countries. When turning to a time-series perspective, we find that the relative fiscal position has a significant effect on the current account and net foreign asset position in a given country over time, in addition to the effects of the stage of development and demographics. In the dynamic model, we also identify short-run effects of the exchange rate, the terms of trade, and the cyclical position on the current account.

The next subsection provides a brief review of standard theories of current account determination, largely from the perspective of the saving-investment balance, to motivate the choice of explanatory variables for the empirical analysis. This is followed by discussions of data and estimation issues and a broad set of results from cross-sectional and panel regressions. The final subsection presents some more specific empirical results for the current account and explains how they may be used to construct medium-term saving-investment norms for the sample of industrial countries.

Theories of Current Account Determination

As discussed in Section V, one commonly used framework for examining developments in the external account is the standard trade model, based on the elasticities approach to the balance of payments.3 This approach has the benefit of straightforward empirical predictions, which are often found to be helpful in examining the short-run implications of exchange rate changes on the trade balance. Due to its partial-equilibrium nature, however, the elasticities approach is inherently limited in its ability to explain normal or equilibrium current account positions. In particular, while the exchange rate may play a key role as an equilibrating mechanism in the adjustment process, it is not generally regarded as a primary driving force behind longer-term developments in the levels of saving, investment, and the current account. Hence, the methodology described in this study combines the elasticities approach with a second model that focuses on the fundamental determinants of the longer-term saving-investment balance.

Much of the recent theoretical analysis on the current account, which can be viewed as a modern absorption approach, has sought to explain current account developments through closer examination of intertemporal consumption, saving, and investment decisions. Indeed, the basic insight of the intertemporal approach to the balance of payments is that the current account can act as a shock absorber that enables a country to smooth consumption and maximize welfare in the wake of temporary changes in national cash flow or net output (i.e., output less investment and government spending).4

While the basic permanent income model has been fruitful in explaining current account movements at business-cycle frequencies, the consumption smoothing perspective has generally had less to say on sustained current account imbalances and trend developments.5 Nevertheless, the model can be used to analyze longer-term variation in current account balances, as illustrated by the relation between the current account, investment, and the stage of economic development in the permanent income model.

In particular, a small open economy that is initially capital (and income) poor, provided it has access to international capital markets, will run current account (and trade) deficits for a sustained period of time to build its capital stock while maintaining its long-run rate of consumption. During the adjustment, a relatively high marginal product of capital will attract capital inflows and raise external indebtedness. Eventually, as output grows toward its longrun level and the return on capital converges to its value abroad, the current account will improve to-ward (zero) balance as net exports move sufficiently into surplus to pay the interest obligations on the accumulated external debt.

Long-run growth somewhat complicates the analysis by allowing for the possibility of non-zero current account balances in steady state. Assuming that the stock of net foreign assets does not outpace growth in the overall economy indefinitely, the level of the current account (as a share of GDP) required to stabilize net external indebtedness can be deter-mined. In particular, given that the current account, CA, equals the change in net foreign assets, NFA, a stable ratio of NFA to GDP (denoted Y) implies that in steady state:


where g = ΔY/Y.6 Hence, structural determinants of the current account could be viewed in terms of the factors that underpin the desired net foreign asset position in the long run. Equivalently, one could view this stock-flow equilibrium relationship in terms of the underlying determinants of saving and investment behavior.

Among the broader explanations for sustained current account imbalances, many of the insights into longer-term saving and investment behavior and the current account have relied on life-cycle extensions of the basic intertemporal (representative-agent) approach that obtain from an overlapping generations framework.7 With heterogeneity across age groups, factors such as demographic trends become relevant as a source of long-run variation in the current account.8 According to the life-cycle model, consumption and saving behavior are directly tied to the stage in the life cycle. Hence, systematic changes in the age structure of the population will affect national saving behavior. To the extent that capital-labor ratios are also affected (via the number of available workers), changes in demographics may affect investment as well.

Similarly, the life-cycle framework can also be used to examine the real effects of fiscal policy on the current account.9 In the absence of Ricardian equivalence, for example, tax policies will have implications (through net wealth effects) for national saving. In particular, changes in public saving and debt (i.e., the timing of taxes) will not be fully offset by counterpart changes in private saving, leading to changes in the current account balance.10 Government spending will have a further impact on the current account, even in the permanent income model, through its direct effect on absorption given income. Consequently, the stance of fiscal policy may have important long-run implications for net foreign assets and the saving-investment balance.11

Extending the permanent income model to include the impact of uncertainty and risk aversion, one could also examine the effects of variability in national income and precautionary saving on the level of the current account. Without certainty (or certainty equivalence), economies facing variable income streams due to, say, terms of trade volatility may find it desirable (without full insurance) to have additional saving as a buffer. Consequently, systematic changes in variability and uncertainty in relevant income measures could possibly affect the current account balance.12

Finally, in the absence of freely mobile capital, one could approach current account determination by focusing more explicitly on developments in its counterpart: the capital account. From this flow-of-funds perspective, the impact of capital controls on the international flow of saving and the current account becomes more apparent. Countries that maintain a relatively closed capital account through barriers and controls, or countries with limited access to foreign borrowing due to country risk, are likely to have smaller current account imbalances than otherwise.13 Correspondingly, financial liberalization and changes in capital mobility may have important long-run implications for overall current account positions.14

Data and Estimation Issues

The above discussion suggests a number of factors that might be important in determining the current account through their implications for saving and investment: fiscal policy, the stage of development, demographics, capital controls, and the terms of trade. The methodology adopted in estimating the determinants of current account deficits is outlined below, together with a discussion of the different data samples used.

The sample we use to examine the determinants of the current account deficit consists of data over the period 1971–93 for 21 industrial countries. Some of the data are drawn from the saving study by Masson, Bayoumi, and Samiei (1995), while some additional variables are from the databases of the IMF and Organization for Economic Cooperation and Development (OECD) and the Summers and Heston (1991) database. The list of countries used and the data documentation are given in Appendix I.

The dependent variable in most of the regressions is the ratio of the current account to GDP. In the panel estimation, the change in the NFA-to-GDP ratio is also used in an error-correction specification. The stance of fiscal policy is captured in various forms: the general government budget surplus (including interest payments on government debt), government current expenditure, and government investment expenditure, all expressed as ratios to GDP.

Stage-of-development effects were measured by real GDP per capita, calculated relative to that in the United States.15 As a proxy for the marginal productivity of capital, we use two measures: the ratio of the capital stock to GDP and the ratio of the capital stock to labor.16 These ratios will also capture stageof-development effects to some extent, which may give rise to a problem of multicolinearity in the estimation. Both linear and quadratic terms were included in the regression, so as to capture any potential nonlinearities in the effect of the stage of development, reflecting the tendency of countries to borrow from abroad at relatively early stages of development and repay capital at later stages. We also estimate a specification that allows for an exponential relation between the capital stock and the current account, which accords more closely with the theory discussed above.

Demographic effects were measured by the dependency ratio—the ratio of the nonworking age population to the working age population—more precisely, the ratio of the population aged 19 and under and 65 and over to the population aged between 20 and 64. We also split the dependency ratio into its two components: the ratio of the old (over 65) to the working age population, and the ratio of the young (under 19) to the working age population.

The dependency ratio variable for each country is expressed as the deviation from the average dependency ratio for all the countries in the sample, rather than the level of the dependency ratio. Thus, as the populations of all the industrial countries age in the coming decades, the effect on a country’s current account position is determined by whether its population ages relatively faster or slower than average, rather than by its absolute demographic position.17

The annual change in the terms of trade was used to capture the effects of export and import price movements on the current account. In the cross section, the average annual change should capture the effect of persistent movements or trends in the terms of trade. The mean annual change in the terms of trade was not significantly different from zero for any of the countries in the sample, suggesting that terms of trade movements should generally be viewed as temporary rather than permanent.

To look for possible effects of uncertainty on the current account via a precautionary savings motive, the variability of the terms of trade and inflation were also included. Price variability was measured by the variability of the annual (consumer price) inflation rate over the sample period, and similarly for the variability of the terms of trade. Other variables used include exports and imports of oil (as a percentage of GDP), the short-term ex post real interest rate, and the rate of inflation.

To measure the effects of financial liberalization we used an index of the level of capital controls developed by Milesi-Ferretti (1995). The measure varies between zero and six depending on whether various restrictions on capital or current account transactions were present in a particular country in a given year. An increase in the index suggests a more closed capital account.

Table 6.1 decomposes the variation in the dependent and the explanatory variables. It shows the proportion of the total variation in a given variable that is explained by the variation across countries in the time average of that variable. The table shows that variation in the current account is about evenly split between cross-country and within-country (or time) variation in the industrial country sample, suggesting that there is information to be gained in analyzing its determinants along both dimensions.

Table 6.1.

Variance Decomposition: Time Series Versus Cross Section (Percentage of total variance explained by cross-country variation)

article image
Note: The cross-country variance is based on the sample of the average values for each of the 21 countries. The total variance is based on the sample of 462 observations, reflecting 22 years of data for each of the 21 countries.

The regressions in the first part of the next sub-section seek to explain the cross-country variation in current account positions (averaged over time). In estimating the cross-sectional regressions, we assume that time averaging allows us to capture equilibrium relationships between the current account and its long-run determinants. Time averaging should net out the effects of transitory fluctuations in the explanatory and dependent variables and identify their longer-term relationships. The panel regressions seek to combine an explanation of the cross-country variation in the current account with an explanation of its time-series variation, assuming that the explanatory variables have the same impact on the current account across countries.

In the panel regressions, the choice of model in part depends on an assumption about the stationarity of the ratio of the current account to GDP, the ratio of net foreign assets to GDP, and the explanatory variables. In steady state, the ratio of the current account to GDP is linked to the NFA-to-GDP ratio by condition (6.1) above. Conceptually, this implies that whether the current account (as a share of GDP) is stationary (mean-reverting) depends on the long-run impact of shocks on the equilibrium net foreign asset position. If changes to the underlying determinants of saving and investment have only level effects on the stock of NFA, but not on the ratio of NFA to GDP, the effects of shocks on the current account to GDP ratio will tend to die out over time. If, however, certain shocks alter the entire path for NFA, the ratio of NFA and the current account to GDP would be permanently affected (absence of mean reversion).

Dickey-Fuller tests country by country are generally unable to reject the null hypothesis of difference stationarity for the ratio of net foreign assets to GDP and the ratio of the current account to GDP.18 It is well known that these tests have low power to reject in favor of stationary alternatives and that trend stationary and difference stationary representations may be observationally equivalent in finite samples.19 Similar considerations apply to the budget surplus and the ratio of government debt to GDP. Consequently, we estimate dynamic specifications that allow for the current account being either stationary or nonstationary. In particular, we estimate a partial adjustment model for the ratio of the current account to GDP as well as an error-correction model for the NFA-to-GDP ratio with possibly (co-)integrated variables. These models are described below.

But first, a number of general specification issues that arise in panel data estimation need to be addressed.20 For example, Ordinary Least Squares estimates that ignore the potential for country-specific effects will provide biased estimates if these considerations matter. Two estimation approaches that address this problem are the fixed effects and random effects models. The following general model of the panel specification will help illustrate these issues:


where y is the dependent variable and X is a matrix of explanatory variable. In equation (6.2), i is the country index where N is the number of countries (i.e., panel units), and t is the time index where T is the number of time periods. The error term ɛ can be decomposed as:


where μi represents country-specific effects (fixed or random) and uit is a residual error term.

The random effects estimates assume that the country-specific effects μi are distributed randomly across countries. Thus, it makes the assumption, as in OLS, that the country effects that are random are uncorrelated with the included exogenous variables, in which case both estimates will be consistent, although direct OLS estimation of equation (6.2) will be inefficient for not taking into account the variation in μi. The assumption regarding μi may not be an appropriate assumption in our model, however, especially when the lagged dependent variable is included in the model.21

Assume for the moment that γ = 0, so we leave aside for now the issue of the lagged dependent variable. Applying OLS to equation (6.2) will still result in biased estimates if the error term has the form of equation (6.3), and if the country-specific component μ is correlated with regressors in X, which is likely to be true in our model. For example, μ may capture the influence of initial conditions. That is, countries that have a large current account deficit at the beginning of the sample period may be more likely to continue to run large current account deficits over the sample. Some of the influence of the initial current account position, however, should be captured by the other right-hand-side variables.

The fixed effects estimator assumes that differences across countries are not systematic but fixed and can be captured by allowing for different constant terms, by including country-specific dummy variables. That is, fixed effects estimation will estimate a country-specific constant μi, which is equivalent to demeaning the data.22 The model can also be estimated in first-differences (i.e., changes).23

In a number of the specifications below, we also include a lagged dependent variable with coefficient γ in the model. In this case, by definition, μi is a fixed effect, and the OLS and random effects model will result in biased estimates. Fixed effect estimation will also result in inconsistent estimates because the error term will be correlated with the lagged dependent variable by construction. Islam (1995) shows, however, that while fixed effects with a lagged dependent variable is inconsistent as N→∞ given T (the number of time periods), as T→∞ given N (the number of countries) fixed effects is consistent and asymptotically equivalent to Maximum Likelihood Estimation. In our sample, with a fixed N and a large T, fixed effects should produce consistent estimates.

To address these various econometric concerns, we report fixed effect, OLS, and first-differenced estimates of the parameters. We also often report the random effect estimates, although in nearly every specification they will be inconsistent. As in the cross section, the constant appearing in the equations will in part capture the current account discrepancy. We assume that movements in the discrepancy across time in a particular country are not correlated with the right-hand-side variables. The random effects estimation could be interpreted as capturing the discrepancy more directly, by allowing for a country-specific component of the error term.

Empirical Results

We report three types of regression findings: cross-section results (based on data averaged over time), panel estimates of a partial adjustment model of the current account, and estimates of an error-correction model of the ratio of net foreign assets to GDP.

Cross-Section Results

Table 6.2 reports the results from a number of different specifications of the cross-sectional estimates using the sample of 21 industrial countries. Column 1 shows a basic specification that captures fiscal, demographic, and stage of development (relative income and capital stock) effects in a linear fashion. Columns 2 and 3 present nonlinear specifications for the stage of development variables. Columns 4 and 5 re-estimate the specifications in columns 1 and 2 including the initial (1971) ratio of net foreign assets to GDP.

Table 6.2.

Cross-Sectional Regression (Dependent variable, CA/GDP)

article image

Notes: denotes significance at the 10 (5) percent level; number of observations: 21; mean of dependent variable (current account):-1.42; standard error of dependent variable: 3.18.

The demographic variable (the ratio of the non-working age population to the working age population) enters with the expected negative sign in all cases and is significant in the nonlinear specifications (columns 2, 3, and 5): countries that have an above average number of dependents tended to have larger current account deficits. The coefficient suggests that a country that has a dependency ratio that is 5 percentage points above the average will run a current account deficit that is around 1 percentage point of GDP larger than otherwise. Over the sample period, dependency ratios varied from 64.3 percent in Japan to 99.7 percent in Ireland, while the average ratio was 75.1 percent. Hence, demographics imply a larger surplus for Japan and a larger deficit for Ireland.

The budget surplus was insignificant in all cases and had the wrong sign in the first three columns, suggesting that changes in governments’ fiscal policies have no impact on the current account in this cross section of countries.24 The lack of any fiscal effects in the cross section can be compared with the results from the saving regression in Masson, Bayoumi, and Samiei (1995). Their results identify a partial offset of private saving to an increase in government saving (an increase in the fiscal surplus). Our results may reflect the fact that investment may also respond positively to an increase in the fiscal surplus, so that the net effect on the current account is closer to zero.25

When the initial stock of net foreign assets is included, fiscal consolidation has a positive but still in-significant effect on the current account. The initial stock of NFA itself had a positive effect on the average current account position—countries that had large net foreign asset positions in 1971 tended to run larger current account surpluses on average over the period 1971–93.

In regressions not reported here, we also investigated whether the effect of fiscal policy on the current account varied in high-debt versus low-debt countries, distinguished by public debt above or below 50 percent of GDP. When the initial stock of NFA is included in the specification, we found that in low public debt countries, the effect of the fiscal surplus on the current account was significantly positive. The effect in high public debt countries was close to zero. This is consistent with the notion that high-debt countries are “more Ricardian” than low-debt countries because there is an expectation of imminent fiscal adjustment.26

The results in Table 6.2 show that a quadratic specification captures the effect of the stage of development reasonably well, although the coefficient on relative income is not significant in the specifications that include quadratic terms. This latter result may be caused by the multi-collinearity between the relative income variable and the capital output ratio. The positive coefficient on the squared relative income term in column 2 supports a U-shaped relation between the current account and the stage of development. The turning point in this quadratic specification occurs at an economy with per capita GDP that is around 66 percent of the level of the United States (slightly less than that of New Zealand in the data set). Prior to this level, increases in relative income are associated with a deterioration in the current account, while after this point they are associated with improvements.

The U shape supports the hypothesis that as countries develop they initially import capital in increasing amounts at lesser stages of development, but then at higher stages of development they become increasingly large capital exporters. The smaller current account deficits for countries that are relatively very poor may reflect liquidity constraints.

The quadratic specification is also significant in capturing the effect of the capital-stock-to-GDP ratio. The negative coefficient on the squared capital stock term suggests that as the ratio of the capital stock to GDP increases from a low level, the current account deficit improves until the capital-output ratio reaches around 2, after which time the current account worsens (an inverted U shape). The worsening of the current account when the capital stock reaches higher levels does not directly accord with the theory described earlier.27 The result appears to be driven by the Scandinavian and Australasian countries (Finland in particular), which have relatively large capital-output ratios but have also tended to run current account deficits over the sample period. These countries are in general relatively well endowed with raw materials, the extraction of which is relatively capital intensive. Consequently, the results for the capital stock may reflect the fact that we have not controlled for this aspect of these economies. The United States and Japan on the other hand have capital-output ratios around 1.75.

When using the capital-labor ratio as a proxy for the marginal product of capital, the signs on the quadratic terms were reversed from those above. The relationship between the capital-labor ratio and the current account is a positive quadratic. The turning point in this relation is around the capital-labor ratio of Australia, which is slightly above the average for the industrial country sample. Beyond this point, as the capital-labor ratio increases, the current account deteriorates.28

We tested the effect of a number of other variables that may also influence the current account, beyond those included in the specification in column 3 of Table 6.2, such as the real interest rate, inflation, productivity growth differentials, variability measures, and the terms of trade.29 Table 6.3 reports the marginal significance value of these variables when they were added individually to the quadratic specification reported in column 3.30

Table 6.3.

Marginal Significance Levels of Additional Variables

article image
Note: Marginal significance (p-values) of variables when included in the specification in column 3 of Table 6.2.

To summarize, in the cross section of industrial countries, we find evidence that the stage of economic development—proxied by GDP per capita, or capital-output ratios, or capital-labor ratios—has an effect on the current account. We find some evidence of an effect of demographics on the current account but no effect of fiscal policy in this cross section, unless we control for the initial stock of net foreign assets and the level of public debt.

Panel Results

Thus far, we have only exploited the cross-sectional information from our sample. In this subsection, we use the time series information as well. We estimate two types of models here, depending on the assumption about the stationarity of the NFA-to-GDP ratio and that of the current account: a partial adjustment model of the current account deficit using the panel data, and an error-correction model of the NFA-to-GDP ratio.

Table 6.4 shows the results of the panel estimation of the determinants of the current account deficit.31 The underlying assumption is that the current account and its determinants are stationary variables. The lagged dependent variable captures the partial adjustment effect. The specification captures fiscal effects (i.e., from the budget surplus), business cycle effects (i.e., from the domestic output gap),32 relative price effects (i.e., from changes in the real exchange rate and terms of trade), and the long-run influences of demographics and relative income. The fourth column reports the fixed effects estimates when the terms of trade is excluded from the specification. The final column excludes relative income and demographics.33

Table 6.4.

Panel Regression: Partial Adjustment Model (Dependent variable, CA/GDP)

article image

Notes: denotes significance at the 10 (5) percent level; number of countries (N) = 21; number of time periods (T) = 20; mean of dependent variable: -1.04; standard error of dependent variable: 3.72.

The Hausman test statistic of whether fixed effects or random effects is appropriate suggests the hypothesis of random effects is rejected at the 1 percent level. This confirms our prior judgment about the nature of our panel data, implying that fixed effects is a preferable specification. The model estimated using first differences with instrumental variables is very similar to that using fixed effects in levels, suggesting that problems associated with including the lagged dependent variable in the fixed effects model are not severe. The adequate number of observations in the time dimension suggests that the fixed effects estimator should be consistent.

Based on F-tests, the dummy variables are jointly significant in every fixed effect specification in Table 6.434 and are significantly different from each other, confirming the relevance of country-specific fixed effects in the current account. These dummy variables capture, in part, the country means that the previous cross-sectional regressions sought to explain. Unfortunately, the hypothesis that all the slope coefficients in the equations in Table 6.4 are the same as imposed by the panel specification is also rejected.35 Nevertheless, the regression can be regarded as capturing the effect of the explanatory variables on an average country’s current account position. Meanwhile, individual countries may have significantly different coefficients from the panel point estimates on at least some of the variables.

Overall, the panel results identify, in contrast to the cross-sectional results, a large impact of fiscal policy on the current account. The coefficient suggests that an increase in the budget deficit of 1 percentage point of GDP results in an increase in the current account of around ⅙ of 1 percentage point of GDP in the short run and around ½ of 1 percentage point in the long run.36 The OLS results find a much weaker effect of fiscal policy on the current account, which may explain why we found no effect in the cross section. That is, it is necessary to control for the country-specific effects to identify the influence of fiscal policy.

As in the cross-sectional regression, we also investigated whether the impact of fiscal policy on the current account differed across high government debt and low government debt countries (defined as before). We again found that the effect of fiscal policy on the current account was significantly greater in low-debt countries. The estimated coefficient on the budget surplus was 0.26 for low-debt and 0.16 for high-debt countries, consistent with the idea that high-debt countries exhibit more Ricardian behavior.

The pattern of coefficients on the change in the real exchange rate variables in column 4 of Table 6.4 (when terms of trade are excluded) suggests a J-curve effect. A depreciation (fall) in the real exchange rate has a positive (but insignificant) effect on the current account contemporaneously but this effect subsequently becomes negative. When the terms of trade are included, an increase (appreciation) in the real exchange rate always exerts a negative influence on the current account (consistent with the Marshall-Lerner condition).37 Increases in the terms of trade meanwhile have a positive effect in the short run. This supports the consumption-smoothing theory that temporary increases in the terms of trade are reflected in saving rather than consumption, leading to improvements in the current account in the short run.38

The coefficient on the contemporaneous output gap has a negative sign, perhaps reflecting the dominance of the accelerator effect of the output gap on investment over the positive effect of the output gap on saving suggested by the permanent income model.39 While the first lag had a significant positive effect when it was included in the specification (not reported), the overall effect of the output gap was still negative.

The results also show that the dependency ratio had the expected negative sign but was generally in-significant (with the exception of the OLS specification), in contrast to the results from the cross-sectional regression. This suggests that demographics matter more for a country’s time-averaged level of the current account than for its time variation, as one might expect. The relative income variable entered significantly with the expected positive sign. The results were unchanged when the capital-output ratio replaced the relative income variable; when both variables were included, the capital-output ratio was not significant.

Error-Correction Model for NFA

In estimating an error-correction specification, we allow for the possibility that the NFA-to-GDP ratio and the current-account-to-GDP ratio are nonstationary variables.40 We first estimate the long-run or levels regression where the dependent variable is the stock of net foreign assets as a share of GDP.41 In the case where the dependent and explanatory variables are difference-stationary (i.e., stationary in first differences but not levels), the equation has the interpretation of the long-run cointegrating relationship, wherein changes in the determining variables have permanent effects on the level of net foreign assets.42 The results from this (first-stage) regression are then subsequently used to estimate a short-run error-correction specification, in which changes in net foreign assets are related—through an error-correction mechanism (ECM) consistent with the long-run cointegrating relationship—to additional factors that have transitory effects on NFA. When all the variables are stationary, the error-correction framework would have the following interpretation. Stock equilibrium is attained only gradually over time, but in the short run, flows reflect both equilibrium adjustment (i.e., error-correction) and disequilibrium fluctuations.

The explanatory variables in the levels specification are basically those from the cross-sectional regression estimated above, namely, the dependency ratio, the capital stock and relative income variables, and fiscal policy.43 Corresponding to our use of the stock of net foreign assets as the dependent variable, we use the stock of government debt to represent fiscal effects. The results from estimating this levels regression are shown in Table 6.5.

Table 6.5.

Panel Regression: Level Equation (Dependent variable, NFA/GDP)

article image

Notes: denotes significance at the 10 (5) percent level; the standard errors are indicative only; number of countries (N) = 21; number of time periods (T) = 23; mean of dependent variable (NFA/GDP): -7.27; standard error of dependent variable: 32.48.

The first column shows the estimates when country specific dummies (or fixed effects) are included. The second column shows the result of an OLS regression with a constant that is common across countries. The third column is the random effects estimate.44 The slope coefficients are almost identical across the random and fixed effects specifications. The hypothesis that all the constant terms are the same is, however, rejected (critical value 60.1). The model estimated using the first difference estimator (shown in column 5) suggests that the fixed effect specification is still an appropriate specification.

The results again show a much larger impact of fiscal policy in this specification compared with the cross-sectional regression earlier. An increase in government debt of 1 percentage point of GDP is associated with a reduction in the NFA-to-GDP ratio of about 0.6 of a percentage point,45 in line with the long-run impact of fiscal policy in the partial adjustment equation. The relationship between the level of government debt and the net foreign asset position is stronger possibly because the government debt variable may be better capturing the long-run impact of fiscal policy.

Demographics again have a negative although not significant impact on the NFA ratio in this specification.46 Relative income has a positive influence on the current account, while the capital-output ratio has a negative influence in the fixed effects regressions, leading us to prefer column 1 over column 4. Again, the latter results may be caused either by the Scandinavian countries or by the collinearity between the capital-output ratio and relative income.

The panel unit root test rejects the null of nonstationary residuals (i.e., no cointegration) in the OLS specification but narrowly fails to reject the null in our preferred fixed effects specification. The country dummies are jointly significant in the fixed effects specification. The hypothesis that the coefficients on the explanatory variables are the same across countries is again rejected, however (the F statistic is 28.4). Nevertheless, as stated above, the regression can be regarded as capturing the effect on an average country’s net foreign asset position of changes in the explanatory variables.

The estimation results of the error-correction model are shown in Table 6.6. The short-run, error-correction specification includes the residuals (i.e., error-correction mechanism terms) from the levels regression estimated with fixed effects in column 1 of Table 6.5. The specification also includes other variables that may have shorter-term effects on the current account discussed earlier. The dependent variable in the short-run model is the change in the net foreign assets ratio.47 The presence of the lagged dependent variable in the regression suggests that fixed effects is the appropriate specification; moreover, the Hausman test rejects the random effects assumption at the 5 percent level.

Table 6.6.

Panel Regression: Error-Correction Model (Dependent variable, Δ NFA/GDP)

article image

Notes: denotes significance at the 10 (5) percent level; number of countries (N) = 21; number of time periods (T) = 20; mean of dependent variable: -0.98; standard error of dependent variable: 5.70.

The error-correction term was significant and had the expected negative sign. The coefficient of -0.1 in column 1 suggests that deviations from the equilibrium level of NFA to GDP have a half life of six or seven years. The other short-run variables generally have the same impact as in the current account estimation in Table 6.4. The exchange rate again has a J-curve type impact on the change in NFA when terms of trade are excluded in column 4, with the initial effect being positive before turning negative in the second year. This may in part reflect valuation effects on the stock of NFA. Fiscal policy, meanwhile, has a large influence on the change in NFA in the short run. A change in the stock of government debt of 1 percent of GDP will improve the NFA position in the short run by around 0.5 percent of GDP. The domestic output gap term enters significantly with the expected negative sign.

The error-correction model for NFA thus unites, in a general sense, the findings from the cross-sectional regressions (Table 6.2) and the panel estimates of the partial adjustment model (Table 6.4). In the long run, the stage of development, government debt, and demographics affect the current account, associated with longer-term stock equilibrium, but in the short run, the flow current account is affected by movements in relative prices, the stage of the business cycle, and the fiscal budget balance.

Partial Adjustment Model Extensions and the Saving-Investment Norm

The previous discussion has shown that, for our panel of 21 industrial countries, both a partial adjustment model of the current account and an error-correction model of net foreign assets can provide satisfactory explanations of the longer-term behavior of the current account in terms of explanatory variables that can be thought of as operating on the external balance through their effects on underlying saving and investment behavior. This section further examines—in the context of the partial adjustment model—the role of relative fiscal, demographic, and stage of development variables in determining the saving-investment balance; it does so to construct measures of “normal” current account positions (i.e., saving-investment norms) for our sample of indus-trial countries. The decision to base the saving-investment norms on the partial adjustment model, rather than the error-correction model of NFA, reflected an initial desire to have separate norms for saving and investment, along with the fact that adequate data were not available on cumulative stocks of saving and investment. In extending the partial adjustment model for constructing norms, we examine issues such as country-specific constants and coefficients, omitted variables, and the implications of the saving-investment model for the global discrepancy.

Following the reduced-form framework described in Section IV, we can express the current account as part of a larger system involving separate equations for saving, investment, and the saving-investment balance (with the appropriate cross-equation restrictions). With common coefficients across countries, and expressing the set of explanatory variables in relative terms (as deviations from global averages), this framework has the desirable property of preserving the level of the global current account (i.e., global discrepancy) in the face of changes in own-country variables or their global averages—which can be thought of as reflecting changes in the world real interest rate. We will return to this system of equations shortly.

Partial Adjustment Model: Single Equation Estimates

Before turning to the system estimates, it is useful to describe many of the issues relevant for constructing the saving-investment norms in the context of a single equation for the saving-investment balance. Single-equation estimation of the partial adjustment model for our panel of 21 industrial countries over the sample period from 1972 to 1993 yields the following result:


In this equation, note that the current account, CA, is expressed as a ratio to GDP, but the fiscal surplus (as a share of GDP), SUR, and the dependency ratio, DEM, are now standardized, expressed as deviations from their GDP-weighted averages for the industrial countries; YPCAP is income per capita relative to the United States, and GAP is the domestic output gap.48 The fitted equation also contains country-specific constants and a measure for German unification (1990–93).49

The common coefficients imposed across countries in equation (6.4) are significant at around the 5 percent level or higher with the exception of the demographic variable (t-stat = -1.2, p-value = 0.2). The common partial adjustment coefficient of 0.66 suggests a significant degree of inertia in the current account. Recall from the earlier discussion, however, that the restriction that all countries have common coefficients is rejected by the data. The issue of introducing country-specific coefficients is pursued in Appendix II, which also examines the in-sample forecasting performance of this model.

Overall, the empirical results are very similar to the earlier fixed effects estimates of the partial adjustment equation reported in Table 6.4. Note, however, that changes in the real effective exchange rate and the terms of trade, which entered (as short-run factors) in the earlier specification, have been dropped from equation (6.4). Omitting these relative price measures greatly simplifies subsequent calculations of equilibrium exchange rates, which reconcile the underlying current account with the saving-investment norm, when the latter measure is taken to be independent of the exchange rate.50 But since it is the effects of relative price changes that are dropped, the omission is perhaps not a particular concern when considering the medium-term level of the current account, which is more likely to be affected by the level variables (e.g., fiscal policy stance) that enter the equation.

Transforming the partial adjustment equation, we can write the “steady-state” version of the model (i.e., using CA = CA-1) as the following long-run equation:


Beyond the constant terms and unification effects, the equilibrium current account or saving-investment norm can be expressed in terms of relative fiscal, demographic, and stage-of-development variables, which presumably are evaluated at their respective longer-run values.51 The longer-term value for the output gap is taken to be zero (i.e., internal balance), as likely would be the case of relative price changes had they been included. The longrun coefficients β¯i in equation (6.5) build from the short-run coefficients βi or impact effects seen in equation (6.4), after all the partial adjustment or transitional dynamics have taken place:


Note that while steady state is often associated with equilibrium over the very long run, equation (6.5) is a steady-state condition in flows (rather than stocks), which may reflect a more appropriate medium-term equilibrium benchmark for the saving-investment balance. Meanwhile, stock variables (e.g., net foreign asset ratios) may still be adjusting, depending on initial stocks and growth rates, and need not have reached the steady-state values associated with full stock-flow equilibrium.

Finally, note that an important component used in deriving the long-run current account—but not shown in equation (6.5)—is the country specific fixed effect. This measure takes the form of a constant and captures factors not reflected in the explanatory variables.52 Econometrically, we require that these factors have a fixed (rather than systematic) effect on the current account; otherwise, the equation would suffer from misspecification bias from omitted variables. Standardizing the stage of development variable in equation (6.5), Table 6.7 shows normalized fixed effects for each country. These are derived by reestimating the partial adjustment model, using a standardized variable for the stage of development, to obtain the following equation for the longer-term saving-investment balance:

Table 6.7.

Normalized Fixed Effects in Equation 6.5

article image

Note: An indicates significance at the 5 (1) percent level.


This is essentially identical to the estimates reported earlier in equation (6.2), except that the income per capita variable has been normalized differently. In this equation, the fiscal surplus SUR, dependency ratio DEM, and relative income per capita (versus the United States) YPCAP, are all expressed as deviations from their (weighted) averages for the industrial countries. The corresponding (normalized) fixed effects μi from this equation are no longer negatively valued systematically across countries.

With the normalization, the country-specific constants or fixed effects in the table measure longer-term current account balances not explained by the regressors—that is, the current account that would obtain if the values of the explanatory variables were zero. In other words, if a particular country were average in terms of its fiscal position, dependency ratio, and relative income per capita, the normalized fixed effect would identify the longer-run level of the current account balance (as a share of GDP) attributable to factors outside the model but specific to that country. Hence, a large fixed effect (in absolute terms) would tend to suggest that the model has some difficulty in explaining the observed level of the saving-investment balance for a particular country over the sample period.

From the table, we see that among the major industrial countries, fixed effects are statistically significant for the United States, Germany, Italy, and Canada. For the United States, the explanatory variables alone would predict, despite the presence of fiscal deficits, a larger current account surplus—primarily due to its advanced stage of development— than what has been observed over the estimation period. The resulting negative fixed effect for the United States may reflect (among other things) the relatively low rate of private saving, to the extent that this behavior reflects considerations (e.g., preferences) not fully captured by the regressors. For Germany, the actual current account balance (on average) has been higher than what relative fiscal, demographic, and stage-of-development factors would predict according to the model. The estimate of the positive fixed effect in this case, however, is affected by the inclusion of unification effects, which act to lower the saving-investment balance.53 Italy is an example where country-specific factors are found to be highly significant. Large public sector deficits over the sample period lead to an overprediction (based on the common fiscal coefficient) of the size of the saving-investment deficit in Italy.54 Consequently, a large positive fixed effect is required as an offsetting factor.

Saving, Investment, and the Current Account: System Estimates

The saving-investment balance can also be examined within the broader context of a system of separate equations for the current account, saving, and investment. This approach has the advantage of identifying the separate saving and investment channels through which relative fiscal, demographic, and stage-of-development factors may operate in determining the medium-term, saving-investment balance or norm. The system is also capable of generating fitted or predicted values of medium-term saving and investment levels separately, in addition to the saving-investment balance.

Table 6.8 presents estimates of the system of equations for saving, investment, and the current account, all as ratios to GDP. In the estimation, several cross-equation restrictions are imposed. Specifically, we impose the same partial adjustment coefficient in each equation; second, for the relative fiscal, demographic, and stage-of-development variables, we impose the restriction that the coefficient in the current account equation equal the difference between the respective coefficients in the saving and investment equations;55 however, country constants (not reported) entering each equation are estimated unconstrained.56 Finally, following the discussion in Section IV, aggregate variables—based on the average values for the fiscal, demographic, and output gap variables—were also included (where significant) in the saving and investment equations, with equal coefficients and a zero coefficient (i.e., exclusion restriction) in the current account equation.57

Table 6.8.

Panel Estimates: Saving, Investment, and Current Account Equations

(Dependent variables, SIGDP, I/GDP, and CA/GDP)

article image
Source: World Economic Outlook database, except as indicated in table notes.

Note: An indicates significance at the 5(1) percent level.

General government balance as percent of GDP minus GDP-weighted average ratio of general government balance to GDP for all countries.

Ratio of population of 65 and older or 19 and younger to population aged 20–64, minus GDP-weighted average ratio for all countries (United Nations, World Population Prospects).

Relative per capita GDP, PPP-adjusted (Organization for Economic Cooperation and Development, Main Economic Indicators).

Actual output minus potential output (logarithmic difference).

GDP-weighted average ratio of general government balance to GDP for all countries.

This latter type of variable is introduced to incorporate factors that affect saving and investment levels but not their difference. For example, in the case of a global fiscal consolidation, saving and investment levels would rise across all countries, as indicated by the coefficients on SUR in Table 6.8, but the global current account balance would remain unchanged. Implicitly, as discussed in Section IV, the world real interest rate would adjust so that the changes in aggregate saving and investment were equalized.

From Table 6.8, note that changes in relative fiscal or demographic positions affect the saving-investment balance primarily through their effect on domestic saving; relative income per capita operates roughly equally through both saving and investment; while the effect of the output gap on the current account largely works through its effects on investment.

Solving for the long-run version of the model as before, the medium-term saving-investment balance based on the estimates in Table 6.8 can be written as follows:


It is interesting to note that the long-run equation based on the single partial adjustment equation for the current account—recall equation (6.5)—yields a somewhat lower fiscal multiplier than that obtained from the system of equations for saving, investment, and current account. The long-run fiscal coefficient based on the single-equation model is 0.45 versus 0.65 from the system when the cross-equation (adding-up) restriction is imposed across the corresponding saving coefficient (0.80) and investment coefficient (0.15).

The estimates in Table 6.8 can also be used to generate medium-term fitted or predicted values for saving and investment levels individually. To ensure adding-up between estimates of the medium-term saving, investment, and current account positions would further require that the fixed effects satisfy an adding-up constraint across these equations. Because saving is often considered to be the most poorly measured of the three variables, one implementation of the model would set the country constants in the saving equation equal to the sum of the estimated fixed effects in the other two equations to ensure that the accounting identity (CA = S - I) holds exactly.

Calculation of the Saving-Investment Norms

Equation (6.8)—with country-specific, fixed-effects included—serves as the basis for computing saving-investment norms for each country in our sample. The medium-term fiscal position that is used in the calculation is based on structural (rather than actual) fiscal balances, which are also smoothed to abstract from shorter-run changes in fiscal policy; similarly, the stage of development measure used in the calculation is given by the cyclically adjusted level of income per capita relative to that of the United States.

In constructing the norms, we adjust Germany’s country-specific constant to broadly reflect the persistent effects of unification. Specifically, based on a smoothing of the estimated coefficients on a series of dummy variables for 1990–93 (as discussed in the context of equation (6.4) above), the estimated country-specific constant term for Germany in the partial adjustment equation (not shown) from which equation (6.8) is derived is lowered by 1½ in 1991, which amounts to a 4½ percent of GDP decline in Germany’s medium-term saving-investment norm in that year.58 In addition, counterpart or offsetting adjustments are made in the saving-investment balances of other countries (distributed in proportion to their shares in Germany’s imports), to ensure that the global current account is unaffected (in dollar terms). We also impose a gradual unwinding of this unification effect over time at a 10 percent annual rate of decay.

Equation (6.8), adjusted for German unification, generates plausible estimates of saving-investment norms for most of the industrial countries. Among the major industrial countries, the exception is Italy, where as we move further back into the historical period, the model increasingly overpredicts the size of the current account deficit, while as we move into the future, the model predicts large and increasing surpluses over the World Economic Outlook horizon. The very steep trend in the predicted saving-investment balance, compared with that in the observed current account for Italy, is a direct consequence of the significant (pooled) fiscal coefficient and the magnitude of Italy’s fiscal adjustment over time. The effects of fiscal adjustment may be somewhat different, however, in a country where fiscal sustainability issues have been pronounced, as suggested by the empirical results discussed earlier for high-debt versus low-debt countries. The size of the fixed effect for Italy shown in Table 6.7 also suggests that the model has difficulty explaining the observed current account over the sample period. Consequently, the saving-investment norm for Italy is based on panel estimates of the coefficients on the demographic and stage-of-development variables and country-specific estimates of the fiscal coefficient and the constant term. Specifically, the norm for Italy is based on the panel estimates described in Appendix II, but with a fiscal coefficient that is essentially zero (i.e., the fiscal coefficient in the equation estimated for Italy alone) and a consistently estimated country-specific constant.

The saving-investment norms for the major industrial countries, calculated from the World Economic Outlook database and projections as of August 1997, are plotted in Figure 2.3 of Section II.

Appendix I: Data Sources

Most of the data was obtained from the IMF World Economic Outlook (WEO) and International Financial Statistics (IFS) databases. Additional data was obtained from the OECD database, the Summers and Heston Penn World Tables (PWT).

article image

The real exchange rate (trade-weighted, CPI-based): calculated in the Research Department of the IMF, in logs, 1982 = 1.

The following countries were included in the industrial country sample; numbers refer to the IFS country code:

article image

Appendix II: Further Results for the Partial Adjustment Model

This appendix reexamines the partial adjustment model for the current account, focusing on the issues of in-sample forecasting performance and country-specific coefficients within the model. For illustration, we consider, for the most part, the single-equation model for the current account alone—the estimates of which are reported in equation (6.4)—rather than the (stacked) system containing the current account, saving, and investment. As the stacked system is treated as an undated (i.e., cross-sectional) data set with no explicit time dimension, it is difficult to consider issues of dynamic forecast performance in that context; moreover, when considering country-specific estimates, it is more convenient to work with the single-equation model.

In-Sample Fit

As the equilibrium current account is an unobservable variable, we compare the short-run implications of the model to the actual current account series to get a sense of the “goodness of fit” of the estimated equation. In terms of regression diagnostics, one could examine the in-sample forecasting performance of the model. To do this, we estimate the model over the initial part of the sample from 1972 to 1979 and then update these estimates recursively, by extending the sample period by one year at each iteration, over the remainder of sample from 1980 to 1993. Forecasting issues aside, the recursive estimates provide some information regarding issues of parameter stability. In general, the recursive point estimates (not shown) remain stable as the time span of the data is increased. Not all coefficients are significant at the 5 percent level. In particular, the 95 percent confidence bands around the point estimate for demographics are centered about zero throughout the estimation period, and similarly for the stage of development variable except at the very end of the sample. The point estimates for the other coefficients are more strongly significant (bounded away from zero) statistically.

Regarding in-sample forecasts, we consider the one-step-ahead forecast errors implied by the model. Specifically, based on the recursive esti-mates of the parameters β at time t and information regarding the right-hand-side variables, Z at t + 1, we form a forecast for the current account at t + 1. The difference between the forecast and the realization for the current account defines the (one-step) forecast error u:


The forecast standard errors for each country and the standard error of estimate for the panel (over the entire sample) are shown in Table 6.9.

Table 6.9.

Standard Error of Estimate and Forecast Errors: Partial Adjustment Model for the Current Account

article image

Country-Specific Estimates

In terms of the parameter restrictions, one could of course relax the imposition of common coefficients in the panel estimates. This approach is implemented here. In particular, we allow the major industrial countries to have different coefficients from each other and from the smaller industrial countries, whose coefficients are constrained to be identical. The point estimates are presented in Table 6.10, where the estimates for the major industrial countries are expressed as deviations from the smaller industrial country coefficients. As seen in the table, the (pooled) estimates for the smaller industrial countries are broadly similar to the earlier panel estimates, with a somewhat larger coefficient on the stage of development and demographic variables. The latter coefficient is now also closer to the 5 percent level of significance (t-stat = -1.8, p- value 0.08).

Table 6.10.

Panel Versus Country-Specific Parameter Estimates: Partial Adjustment Model for the Current Account

article image

Note: An indicates significance at the 5(1) percent level.

As for the major industrial country estimates, they are generally found to be insignificantly different from the pooled estimates. Also, country-specific estimates for the major industrial countries, assuming a common lag structure (AR coefficient), are shown in Table 6.11. Various alternative specifications were also conducted (but not reported), testing for example whether fiscal multipliers differed among the major industrial countries when other coefficients were imposed to be the same. These tests further confirm that, in general, we cannot reject the condition that the major industrial country coefficients, taken individually, are the same as the pooled estimates for the smaller industrials based on this pairwise comparison. The joint hypothesis that all the coefficients, collectively, are the same for each variable across all countries is, however, rejected (as discussed earlier).

Table 6.11.

Panel Versus Country-Specific Parameter Estimates with Common Lag Structure: Partial Adjustment Model for the Current Account

article image

Note: An indicates significance at the 5 (1) percent level.


Much of the analysis was originally presented in Debelle and Faruqee (1996).


Because of data limitations and to ensure greater cross-country comparability, the sample of countries is restricted to a panel of 21 industrial countries, which presumably have broadly similar economic characteristics, including reliable access to international capital markets.


See Goldstein and Khan (1985) for a general review of this approach.


See Sachs (1981) for a discussion of the intertemporal optimizing approach and current account developments following the oil shocks in the 1970s. See Obstfeld and Rogoff (1995) and Razin (1995) for recent surveys of the theory and evidence for the modern intertemporal approach to the current account.


In the permanent income model, longer-term developments are generally limited to consumption-tilting effects resulting from changes in the rate of time preference (which are difficult to measure). Consequently, tests of the (present-value) model have examined detrended current account series. See, for example, Ghosh and Ostry (1995).


If there are real exchange rate trends, the proportional factor g would also take account of the long-run rate of appreciation to account for differing valuation effects on NFA and Y.


Extending the Blanchard (1985) model to include age-dependent income (e.g., Faruqee, Laxton, and Symansky (1997)) or labor supply (e.g., St. Paul (1992)) would generate life-cycle-(OLG-) type implications. The seminal paper on OLG analysis in an open economy is Diamond (1965).


Masson, Bayoumi, and Samiei (1995) find evidence of demo-graphic factors (and fiscal variables) explaining the cross-country variation in saving rates.


See Frenkel and Razin (1992) for a comprehensive analysis of the role of fiscal policies in an open economy.


In a large open economy, changes in the real interest rate (cost of capital) and its effects on investment should also be taken into account.


See Masson, Kremers, and Home (1994) for evidence of a long-run impact of public debt and demographics on net foreign assets in the United States, Japan, and Germany.


See Ghosh and Ostry (1994) for some empirical evidence on the effects of income uncertainty on precautionary saving and the current account.


In a well-known paper, Feldstein and Horioka (1980) implicated low capital mobility as the explanation for the high correlation between saving and investment across countries. See Montiel (1994) for a recent review of measurement issues surrounding capital mobility, and also Obstfeld and Rogoff (1995) for a critical review of the Feldstein-Horioka results and its implications for the intertemporal approach.


A related issue (not addressed here) is the composition of the capital account (i.e., the nature of financing in connection with a given current account deficit). Whether capital inflows predominantly reflect short-term borrowing and portfolio flows or longer-term foreign direct investment has also received significant attention recently in the context of current account sustainability.


Data on real GDP per capita are expressed in the common benchmark of 1990 international prices to allow for cross-country comparisons. See Summers and Heston (1991) for further discussion.


With a Cobb-Douglas production function, the marginal productivity of capital is proportional to either of these two ratios.


Note that specifying variables in this fashion will not alter the coefficient of the variable, but only the size of the constant term in the cross-sectional regression. See Glick and Rogoff (1995) for further discussion on global versus country specific shocks.


On the other hand, panel unit root tests (following Im, Pesaran, and Shin (1995)) suggest that in the case of the current account, the null of difference stationarity may possibly be rejected (the p-value is 0.12).


See Hsiao (1986) for a complete discussion of these issues. Cashin and Loayza (1995) and Islam (1995) discuss similar issues in the context of cross-country growth regressions using panel data. Keane and Runkle (1992) discuss the pitfalls of fixed effects estimates.


One solution to this problem is to use a technique such as Chamberlain’s Π estimator. Cashin and Loayza (1995) and Islam (1995) provide examples of the application of this technique. Cashin and Loayza argue that this approach is especially appropriate if there are problems of measurement error.


Keane and Runkle (1992) point out that the fixed effects estimator may not be consistent unless there is a strong assumption of strict exogeneity—that is, that the error terms uit are not correlated with the regressors X at any time horizon. This assumption is unlikely to hold in our model, where shocks to the current account are likely to induce policy reactions or exchange rate adjustments in future periods.


Keane and Runkle (1992) suggest first differencing as a solution to the exogeneity problem described in the preceding footnote, in which case, the explanatory variables need only be pre-determined rather than strictly exogenous. When the first differencing procedure is adopted, an instrumental variables approach is applied. The instruments are the lagged levels of the explanatory variables; standard errors are also corrected for the first-order serial correlation that results from the first differencing.


Milesi-Ferretti and Razin (1996) also conclude that the fiscal balance is not a good indicator of current account sustainability.


When we estimated separate saving and investment equations similar to the partial adjustment model of the current account below, we did indeed find a positive impact of fiscal policy on investment.


A nonlinear relationship that more closely reflects the implications of the theory is an exponential one in which increases in the capital-stock-to-GDP ratio are associated with an improvement in the current account, which levels off. We estimated such a nonlinear specification with mixed results. The coefficient on the exponential term and the exponent itself had the expected negative sign, but the results were very sensitive to the choice of the starting values used in the nonlinear estimation.


The relationship between relative income and the current account is also inverted when the capital-labor ratio is used. The relationship suggests that as per capita income increases in this sample the current account tends to improve. Only Canada, the United States, and Switzerland lie to the right of the turning point where further increases in income lead to a deterioration in the current account. Thus, essentially there is a positive relationship between relative income (stage of development) and the current account for this sample.


Greece is an outlier in the sample, both in terms of its current account deficit and in terms of some of the explanatory variables. Consequently, we also re-estimated the specifications in Table 6.2 excluding Greece. The major difference is that the significance level of the capital stock variables declines, while the significance level of the income variables increases. The capital stock quadratic terms are now only jointly significant at the 19 percent level. When Greece and Switzerland are both excluded from the specifications, the dependency ratio is still significant but the stage of development variables are no longer significant.


The insignificance of the capital control variable (bottom row of Table 6.3) might reflect a problem of endogeneity: countries with larger current account deficits may be more likely to impose capital controls (see Milesi-Ferretti, 1995). Alternatively, the im-position of capital controls may force the current account toward zero. Accordingly, we interacted the capital control variable with the explanatory variables in the different specifications in Table 6.2. We found no evidence of a significant effect of capital controls in this sample of countries, although the effect was to bias the coefficients toward zero. There may also be a measurement problem with the capital control variable in that the number of control measures may not be a good proxy for the effectiveness of the controls.


Estimating the equations in Table 6.4 excluding the lagged dependent variable resulted in a large amount of first-order serial correlation. Including the lagged dependent variable mitigated the problem, although the residuals still exhibited some higher-order auto-correlation. Meanwhile, the standard errors are not corrected for serial correlation or heteroscedasticity. Corrections using the Newey-West procedure did not change the standard errors significantly. However, the standard errors in the first differenced specification are corrected for first-order serial correlation.


Foreign output gaps were also tried but were found to be in-significant. This finding may be caused by the correlation between domestic and foreign output gaps in this sample of industrial countries. The domestic output gap is defined as actual less potential output.


We tested the inclusion of a number of other variables in the specification in column 1. The inflation rate also had a negative but insignificant sign. The real interest rate was positive but insignificant.


The F-test statistic F(21,396) for the specification in column 1 is 3.56.


The F-test statistic F(159,236) for the specification in the first column in Table 6.4 is 1.78. We return later to the issue of country-specific coefficients.


The long-run effect is obtained by dividing the short-run co-efficient by one minus the coefficient on the lagged current account.


The change in the real exchange rate is in log differences. Thus a 1 percent increase in the real exchange rate reduces the current account in the first period by 0.04 in the first specification.


This finding suggests that the income effect outweighs the substitution effect following an innovation to the terms of trade. See the discussion in Ostry and Reinhart (1992).


This finding is confirmed when separate saving and investment equations are estimated later in this paper.


Nonstationarity implies that a variable does not revert to a fixed mean but instead exhibits a stochastic trend, reflecting the permanent effects of certain shocks.


Since the variability of innovations to the stochastic trend (relative to total variability) is much smaller for the current account than for NFA, estimation of the trend in NFA should be more robust.


As the time that series properties of the squared variables may be integrated are not obvious, we estimate the levels equation including the income and capital stock variables in linear form.


A Hausman test (with critical value 3.46) fails to reject the random effects model in favor of the fixed effects model.


Masson, Kremers, and Home (1994) find coefficients of 0.3, 0.5, and 0.7, respectively, in Germany, Japan, and the United States for the effect of government debt on NFA.


Under a cointegration interpretation, note that the t-statistics from the levels regression have nonstandard limiting distributions and, hence, are only indicative.


While the first difference of the ratio of net foreign assets to GDP used in the levels regression is not precisely the current-account-to-GDP ratio, as discussed earlier, the current-account-toGDP ratio (ca) maps into the change in the NFA-to-GDP ratio (Δnfa) through the growth rate of output (g) in steady state. Similarly, the budget surplus maps into changes in the level of government debt.


A standardized measure for the stage of development variable, YPCAP, expressed as a deviation from its average, is considered later. The discussion in Section IV also suggests including a global average for the output gap (to standardize this variable). A foreign output gap term—constructed on a trade-weighted (rather than GDP-weighted) basis—was included in the regression but not found significant.


The effects of unification are introduced through a series of time dummies for 1990–93 in the current account equation for Germany. The dummies have the expected negative sign, reflecting the significant decline in the saving-investment balance (as a share of GDP), reaching a peak decline in 1991 with a point estimate of -3 (percentage points) followed by -1 and -1½ (percentage points) in subsequent years. As described later, in constructing the saving-investment norm, we adjust Germany’s country-specific constant to broadly reflect the (persistent) effects of unification, with counterpart adjustments in the saving-investment balances of Germany’s partner countries (distributed according to trade shares). We further impose a gradual unwinding of these effects at a 10 percent annual rate of decay over time.


See Wren-Lewis (1992) for discussion of the implications of this (recursive) structure between fundamentals and the exchange rate for the Fundamental Equilibrium Exchange Rate (FEER) approach


When constructing saving-investment norms, we use filtered measures of the structural fiscal balance to represent its longerrun value. Note that because fiscal and demographic variables are standardized, a tighter, say, fiscal stance at home will not necessarily lead to an improvement in the current account unless it out-paces the (average) degree of fiscal consolidation abroad. Because only relative positions matter at the country level, changes in saving-investment balances sum to zero across all countries. But while changes in, say, fiscal policy leave the aggregate current account for the industrial countries unaffected, this aggregate balance need not equal zero, and in light of the global current account discrepancy in the data, no adding-up constraint has been imposed on individual country constants.


Also, because the stage of development measure is not a mean-zero (but positive-valued) variable, country constants are uniformly negative to adjust the fitted value of equation (6.5) to the average level of the current account over the sample period (i.e., to obtain mean-zero residuals for the equation).


Excluding unification dummies from the Germany equation, we obtain a fixed effect estimate of 3½ percentage points, which is no longer significant at the 5 percent level.


As shown in Appendix Table 6.10, the fiscal coefficient, when country-specific, is basically zero in the case of Italy.


The F-test that the cross-equation restrictions are valid is marginally rejected (4 percent significance level), suggesting that the restrictions are not grossly inconsistent with the data.


The cross-equation restrictions were imposed on the unification dummies (not reported) in the equations for Germany. The point estimates suggest that the decline in the saving-investment balance was attributable to both higher investment and lower saving. The pickup in investment and the drop-off in saving was responsible for roughly 60 percent and 40 percent, respectively, of the decline in the current account balance between 1991 and 1993.


It may be noted that the aggregate output gap that was tried was constructed with trade weights rather than GDP weights, and the domestic output gap is not measured relative to a world average, since the reported equations are superior to equations in which the domestic output gap was measured relative to trade-weighted averages of trading-partner output gaps. In further estimation work there would be scope for paying attention to these inconsistencies; but because output gaps are set to zero in constructing measures of equilibrium saving and investment positions, we have regarded this as a low-priority concern.


The observed current account in Germany fell by some 4 percentage points of GDP in 1991.


  • Blanchard, O., 1985, “Debt, Deficits, and Finite Horizons,” Journal of Political Economy, Vol. 93 (April), pp. 22347.

  • Blanchard, O., and S. Fischer, 1989, Lectures on Macroeconomics (Cambridge, Massachusetts: MIT Press).

  • Campbell, J., and P. Perron, 1991, “Pitfalls and Opportuni-ties: What Macroeconomists Should Know About Unit Roots,” in NBER Macro Annual (Cambridge, Massachusetts: MIT Press), pp. 141201.

    • Search Google Scholar
    • Export Citation
  • Cashin, P., and N. Loayza, 1995, “Paradise Lost? Growth, Convergence, and Migration in the South Pacific,” Staff Papers, International Monetary Fund, Vol. 42 (September), pp. 60841.

    • Search Google Scholar
    • Export Citation
  • Daniel, B., 1995, “Precautionary Saving and Persistent Current Account Imbalance: An Analytical Approximation” (unpublished; Pittsburgh: Department of Economics, University of Pennsylvania).

    • Search Google Scholar
    • Export Citation
  • Debelle, Guy, and Hamid Faruqee, 1997, “What Determines the Current Account? A Cross-Sectional and Panel Approach,” IMF Working Paper 96/58 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
  • Diamond, P., 1965, “National Debt in a Neoclassical Growth Model,” American Economic Review, Vol. 55 (December), pp. 112650.

  • Engle, R., C. Granger, 1987, “Cointegration and Error Correction: Representation, Estimation and Testing,” Econometrica, Vol. 55 (March), pp. 25176.

    • Search Google Scholar
    • Export Citation
  • Faruqee, Hamid, Douglas Laxton, Steven Symansky, 1995, “Government Debt, Life-Cycle Income and Liquidity Constraints: Beyond Approximate Ricardian Equivalence,” Staff Papers, International Monetary Fund, Vol. 44 (September), pp. 37482.

    • Search Google Scholar
    • Export Citation
  • Feldstein, M., and C. Horioka, 1980, “Domestic Saving and International Capital Flows,” Economic Journal, Vol. 90 (June), pp. 31429.

    • Search Google Scholar
    • Export Citation
  • Frenkel, J., and A. Razin, 1992, Fiscal Policies and the World Economy (Cambridge, Massachusetts: MIT Press).

  • Ghosh, A., 1995, “International Capital Mobility Amongst the Major Industrial Countries: Too Little or Too Much,” Economic Journal, Vol. 105 (January), pp. 0728.

    • Search Google Scholar
    • Export Citation
  • Ghosh, A., Ostry, J. 1994, “Export Instability and the External Balance in Developing Counties,” Staff Papers, International Monetary Fund, Vol. 41 (June), pp. 21435.

    • Search Google Scholar
    • Export Citation
  • Ghosh, A., 1995, “The Current Account in Developing Countries: A Perspective from the Consumption-Smoothing Approach,” World Bank Economic Review, Vol. 9 (Washington: World Bank), pp. 30533.

    • Search Google Scholar
    • Export Citation
  • Glick, R., K. Rogoff, 1995, “Global Versus Country-Specific Productivity Shocks and the Current Account,” Journal of Monetary Economics (April), pp. 15992.

    • Search Google Scholar
    • Export Citation
  • Goldstein, Morris, and Mohsin S. Khan, 1985, “Income and Price Effects in Foreign Trade,” Chap. 20 in Handbook of International Economics, ed. by, Vol. 2 Ronald W. Jones and Peter B. Kenen, (New York: North Holland Press), pp. 1041105.

    • Search Google Scholar
    • Export Citation
  • Hsiao, C., 1986, Analysis of Panel Data, Econometric Society Monographs No. 11 (Melbourne, Australia: Cambridge University Press).

  • Im, K., H. Pesaran, and Y. Shin, 1995, “Testing for Unit Roots in Heterogeneous Panels,” DAE Working Paper No. 9526 (June).

  • Islam, N., 1995, “Growth Empirics: A Panel Data Approach,” Quarterly Journal of Economics, Vol. 110 (November), pp. 112770.

  • Keane, M., and D. Runkle 1992, “On the Estimation of Panel-Data Models With Serial Correlation When Instruments Are Not Strictly Exogenous,” Journal of Business and Economic Statistics, Vol. 10 (January), pp. 43341.

    • Search Google Scholar
    • Export Citation
  • Levin, A., and C. Lin 1993, “Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties” (unpublished; University of San Diego).

    • Search Google Scholar
    • Export Citation
  • Masson, Paul, Tamim Bayoumi, and H. Samiei, 1995, “International Evidence on the Determinants of Private Saving,” IMF Working Paper 95/51 (Washington: International Monetary Fund, May).

    • Search Google Scholar
    • Export Citation
  • Masson, Paul, J. Kremers, and J. Horne, 1994, “Net Foreign Assets and International Adjustment: The United States, Japan, and Germany,” Journal of International Money and Finance, Vol. 13 (February), pp. 2740.

    • Search Google Scholar
    • Export Citation
  • Meade, J., 1951, The Balance of Payments (London: Oxford University Press).

  • Milesi-Ferretti, G.M., 1995, “Why Capital Controls? Theory and Evidence” (unpublished; Washington: International Monetary Fund).

  • Milesi-Ferretti, G.M., Razin, A. 1996, “Persistent Current Account Deficits: A Warning Signal?” (unpublished; Washington: International Monetary Fund, March).

    • Search Google Scholar
    • Export Citation
  • Montiel P. 1994, “Capital Mobility in Developing countries: Some Measurement Issues and Empirical Estimates,” World Bank Economic Review, Vol. 8 (Washington: World Bank), pp. 31150.

    • Search Google Scholar
    • Export Citation
  • Obstfeld, M., and K. Rogoff, 1995, “The Intertemporal Approach to the Current Account,” in Handbook of International Economics, ed. by G. Grossman and K. Rogoff, (Amsterdam, Holland: Elsevier Science Publishing).

    • Search Google Scholar
    • Export Citation
  • Ostry, J., and C. Reinhart, 1992, “Private Saving and Terms of Trade Shocks,” Staff Papers, International Monetary Fund, Vol. 39 (September),pp 495517.

    • Search Google Scholar
    • Export Citation
  • Otto, G., 1992, “Testing a Present-Value Model of the Current Account: Evidence from U.S. and Canadian Time Series,” Journal of International Money and Finance, Vol. 11 (October), pp. 41430.

    • Search Google Scholar
    • Export Citation
  • Razin, A., 1995, “The Dynamic-Optimizing Approach to the Current Account: Theory and Evidence,” in Understanding Interdependence: The Macroeconomics of the Open Economy, ed. by P. Kenen (Princeton, New Jersey: Princeton University Press).

    • Search Google Scholar
    • Export Citation
  • Sachs, J., 1981, “The Current Account and Macroeconomic Adjustment in the 1970s,” Brookings Papers on Economic Activity, pp. 20168.

    • Search Google Scholar
    • Export Citation
  • Sheffrin, S., andW. Woo, 1990, “Present Value Tests of an Intertemporal Model of the Current Account,” Journal of International Economics, Vol. 84 (August), pp. 23753.

    • Search Google Scholar
    • Export Citation
  • St. Paul, G., 1992, “Fiscal Policy in an Endogenous Growth Model,” Quarterly Journal of Economics, Vol. 107 (November), pp. 124360.

    • Search Google Scholar
    • Export Citation
  • Summers, R., and A. Heston 1991, “The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950–88,” Quarterly Journal of Economics, Vol. 106 (May), pp. 32768.

    • Search Google Scholar
    • Export Citation
  • Sutherland, A., 1995, “Fiscal Crises and Aggregate Demand: Can High Public Debt Reverse the Effects of Fiscal Policy?CEPR Discussion Paper No. 1246 (September).

    • Search Google Scholar
    • Export Citation
  • Swan, T., 1963, “Longer-Run Problems of the Balance of Payments,” in The Australian Economy: A Volume of Readings, ed. by H. Arndt and W. Corden (Melbourne, Australia: Chesire Press), pp. 45564.

    • Search Google Scholar
    • Export Citation
  • Wren-Lewis, S., 1992, “On the Analytical Foundations of the Fundamental Equilibrium Exchange Rate,” in Macroeconomic Modeling of the Long Run, ed. by C.P. Hargreaves (Aldershot, England: Edward Elgar), pp. 32338.

    • Search Google Scholar
    • Export Citation

Cited By