A. The Intertemporal Approach to the Current Account

129. The intertemporal approach to the external accounts investigates the extent to which the current account balance reflects an “optimal” path of domestic savings and consumption arising from a model of optimal borrowing and lending. This class of model predicts that the amount of foreign savings absorbed by the economy responds primarily to agents’ desire to smooth consumption, in view of expected changes in disposable income, and subject to the external solvency of the country. This approach can entail at least two broad sets of policy implications. First, to the extent that the actual current account deviates from its “optimal” level, one may inquire what distortions or constraints are limiting the ability of agents to smooth consumption through borrowing and lending. Second, as the approach builds on partial equilibrium models, wide fluctuations of the “optimal” current account prompt questions about underlying imbalances conditioning agents’ decisions. In answering these questions, exogenous external shocks and government policies would emerge as natural candidates for scrutiny. In particular, a large “optimal” deficit may not imply a sustainable situation, but an extreme (although optimal from the point of view of economic agents) response to a serious macroeconomic imbalance. For instance, if the fiscal policy stance is perceived as unsustainably lax, but imported durable goods are more sheltered from taxes than domestic financial assets, the optimal reaction of residents in anticipation of a reversal of such a stance could lead to a marked deterioration of the current account.

130. The consumption smoothing approach to analyzing the current account is based on the permanent income theory of consumption and saving. Essentially, this theory postulates that forward-looking agents choose their consumption level taking into account their permanent income—broadly, the constant income yielded by their net wealth. They will, therefore, decrease their saving whenever new information leads them to expect their income to increase. The model will, therefore, predict a very different impact on the current account of a change in income, depending on whether the change is perceived as temporary or permanent.⁵⁸ This reasoning would thus imply that a decline in national income which is expected to be temporary will be accompanied by a temporary deterioration in the current account, as agents would not alter their consumption schedule. Instead, they would avail themselves of foreign savings, counting on a rebound of the economy. By contrast, after a permanent shock (i.e., a shock that reduces permanent income as it will not be reversed in the future), the current account should not change, as agents should instead fully adjust their consumption schedule to the new situation.

131. The “optimal” current account balance thus depends fundamentally on agents’ expectations about future income. This, of course, complicates the calculation of such a balance on the basis of the observation of macroeconomic variables, as expectations are not observable. While past behavior of income can provide useful information about its future path, it hardly reflects the expectations that economic agents can formulate based on all the information they have available at each period. In particular, such projections may not capture expected policy changes that may have a bearing on future income, or within-period external shocks.

132. Although the information set used by agents could be approximated by expanding the number of variables used to estimate future income, Campbell (1987) and Campbell and Shiller (1987) suggest another approach. They note that, under some not overly restrictive assumptions (notably that income follows a random walk), the permanent income theory predicts a particular joint behavior of saving and income, and makes saving itself the best instrument to forecast changes in income. This permits the econometrician to extract the “optimal” current account from the actual current account and observed changes in income. Campbell’s insight has been extensively used to analyze savings decisions and, in particular, the current account (Ghosh, 1995; Ghosh and Ostry, 1994; Sheffrin and Woo, 1990).⁵⁹ It permits to test the fit of the consumption smoothing model statistically, and to compare “optimal” and actual current account balances over time (it also permits to use the observed current account to forecast changes in national income). The model is described in greater detail in the next section. The hurried (or less technically inclined) reader can forego this section and pass on to the subsequent sections, where the empirical results are presented and analyzed.

B. A Present-Value Model of the Current Account

133. The application of the permanent income theory to the analysis of the current account starts with a representative agent maximizing a time-separable utility function $U={\mathrm{\Sigma }}_t{a}^{t}u\left(c_t\right),$ where a is the discount rate and c_t is consumption at time t, under a sequence of budget constraints:

where y_t is the disposable income, and b_t is the holdings of an indexed bond. Under certain conditions (such as a quadratic utility function, and a constant real interest rate r), it can be shown that at each period the optimal consumption (c^*) will amount to a fixed fraction of the present value of the expected stream of future income plus the current yield of the bond b (see Campbell, 1987). Optimal saving (s^*) is, in turn, equal to the expected present value of future changes in income:

Which implies that

cannot be forecast on the basis of information available at time t−1, as it reflects the revision from t−1 to t in the expected value of future income.

134. Simple manipulation of (1), (2), and (4) will show that changes in optimal consumption cannot be forecast either. Moreover, under the assumption that changes in income are stationary, it is easy to see that saving (as the present value of future changes in income) will also be stationary, while consumption will be a random walk. This will imply that the vector w comprising total income (i.e., including the yield of the bond) and consumption is cointegrated (saving being the difference between the two terms). Therefore, the vector w has an error-correction representation, in which changes in income are predicted using lagged changes in income and consumption, together with lagged deviations of consumption from income (i.e., roughly, lagged saving).

135. Although an error-correction model is not a VAR, it can be put into VAR form (see Campbell, 1987). In the case at hand, equation (3) allows the model to be expressed as a VAR relating current and lagged saving and changes in income. This is so because equation (3) includes both the variable being forecast (changes in income), and the optimal forecast, that is, the variable that best captures the information available about future changes in income (saving). The number of lags in the VAR will correspond to the number of lags in the error-correction model that would make the forecast error a white noise. Empirically, the number of lags will be chosen based on the fitting of the VAR.

136. The above results can be extended to an economy with production, investment, and government at the cost of a few additional assumptions. These boil down to postulating that investment is optimally chosen to maximize the country’s wealth and that the government relies on nondistortionary taxes (as Ricardian equivalence is implicit in the representative agent assumption, the solvency assumption extends to the government budget constraint). In this case, Fisherian separability implies that investment decisions may be treated as exogenous to the consumption/saving rule (Sheffrin and Woo, 1990).⁶⁰ The relevant income flow z to be considered comprises national income net of public consumption and total investment. In this case, the optimal external account can be written as:

137. Under the separability hypothesis discussed, the reasoning developed in the previous paragraphs is still valid, and the auto correction model can be transformed into a bivariate vector autoregression, with lagged saving (i.e., lagged current account balances) and lagged changes in income explaining current savings and changes in income:^{61 62}

where ε_t is a white noise. This system may be written more compactly as:

where ${x^{\prime }}_t=\left[\begin{array}{cc}\mathrm{\Delta }{z}_t& {ca}_t\end{array}\right],$, and the £-step ahead expectation of x, is given by:

In this case, (5) becomes a geometric sum and the current account can be expressed as follows:

Note that, owing to the restrictions mentioned above, the expression of the optimal current account turns out to be a combination of the contemporaneous actual current account and change in income.

C. Empirical Results

138. Two set of estimations were performed, reflecting the importance of international transfers in the case of Portugal. The first estimation starts from national disposable income (as measured in the recently released “long series” of the national accounts). In this case, the disposable cashflow corresponds to national disposable income, minus government consumption and total investment net of international capital transfers. The relevant saving concept is the current account balance minus capital transfers from the European Union. The second estimation uses less precise concepts which, however, permit extending the analysis to more recent years (the “long series” end in 1993). In the second estimation, the disposable cashflow corresponds to GDP net of government consumption and total investment, and the current account includes capital transfers. Results varied relatively little, and the discussion will thus concentrate on the second estimation.

139. As a first step, the saving variable (i.e., the current account at constant prices) was tested for stationarity, the hypothesis of a unit root being rejected at the 10 percent significance level. The rejection of the hypothesis of nonstationarity of the current account can be interpreted as evidence of long-run current account solvency.⁶³ The VAR model was estimated with the number of lags specification being determined based on the Akaike and Schwarz criteria. While the Akaike criterion was essentially indifferent between one and three lags, the Schwarz criterion unmistakably favored the one-lag specification.

140. The coefficient estimates for the VAR (table below) indicate that the information in the current account is relevant in estimating future changes in the disposable cashflow. Indeed, the coefficient (−0.18) is correctly signed and significant. Moreover, a Granger causality test indicates that the information on the current account makes a significant contribution to predictions of changes in disposable income (the hypothesis of block exogeneity, i.e., of non-Granger causality is rejected at 2.5 percent significance level). Therefore, increases in saving tend to anticipate a decline in income, or conversely, a deterioration of the current account reflects expectations of an improvement in the disposable cashflow in the period ahead.

VAR Parameters and Implications for the Γ vector

Note: * denotes significance of-the coefficient at 5 percent level. (Estimation period 1953-93).

VAR Parameters and Implications for the Γ vector

a. VAR Parameters
	dzt-1	Standard error	cat-1	Standard error
dz	0.58*	0.12	−0.18*	0.09
ca	0.00	0.15	0.72*	0.11
b. Implied Γ vector
	Γy	Standard error	Γca	Standard error
	−1.25*	0.64	1.15*	0.71

Note: * denotes significance of-the coefficient at 5 percent level. (Estimation period 1953-93).

View Table

VAR Parameters and Implications for the Γ vector

a. VAR Parameters
	dzt-1	Standard error	cat-1	Standard error
dz	0.58*	0.12	−0.18*	0.09
ca	0.00	0.15	0.72*	0.11
b. Implied Γ vector
	Γy	Standard error	Γca	Standard error
	−1.25*	0.64	1.15*	0.71

Note: * denotes significance of-the coefficient at 5 percent level. (Estimation period 1953-93).

141. The analysis of the vector Γ, also confirms the relatively good fit of the model.⁶⁴ If the actual current account agreed completely with the optimal current account predicted by the permanent income hypothesis, the first component of the vector would be zero, and the second would be one. While Γ_y lies somewhat away from its theoretical value, Γ_ca closely agrees with it, and overall the model cannot be rejected at a 10 percent significance level.⁶⁵

D. Main Findings

142. In view of the general acceptance of the statistical restrictions implied by the consumption smoothing hypothesis, it would be interesting to verify the predictions of the model. To get a grasp of these predictions, the actual current account should be compared to the “optimal” account. The two series move remarkably closely over the forty-year period (Figure 19), with significant divergences occurring only in a few instances. Most notable among these occasions are the years in which Portugal entered into stand-by arrangements with the Fund (1978 and 1983), suggesting that liquidity constraints were binding at the time. Another instance when the model’s predictions and the actual current account diverged was in 1989-90. While the “optimal” current account deteriorated, the actual current account balance improved. This improvement reflected in large part the sharp increase in taxes in 1989 (20 percent in real terms), following income tax reform and an accompanying change in the collection period, which entailed the collection of taxes levied on income received in 1988 and 1989 (income tax collection grew by 35 percent in real terms). The effect of such tax increases on households’ disposable income, coupled with a liquidity shortage due to credit restraint, resulted in a rapid deceleration of private consumption, notably of durable goods (Bank of Portugal, 1990). Credit restraint stemmed in some cases from government action (e.g., with respect to the financing of car purchases), but it would appear that, in the event, liquidity constraints in general impinged on consumers’ ability to offset the temporary reduction in disposable income entailed by the change in the timing of tax collection.

Portugal: Consumption Smoothing and the Current Account

(Actual and “optimal” current account as percentage of GDP)

Citation: IMF Staff Country Reports 1997, 121; 10.5089/9781451832105.002.A004

Source: Staff calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Portugal: Consumption Smoothing and the Current Account

(Actual and “optimal” current account as percentage of GDP)

Citation: IMF Staff Country Reports 1997, 121; 10.5089/9781451832105.002.A004

Source: Staff calculations.

• Download Figure
• Download figure as PowerPoint slide

Portugal: Consumption Smoothing and the Current Account

(Actual and “optimal” current account as percentage of GDP)

Citation: IMF Staff Country Reports 1997, 121; 10.5089/9781451832105.002.A004

Source: Staff calculations.

• Download Figure
• Download figure as PowerPoint slide

143. The predictions of the model for recent years appear quite intuitive. It is noteworthy that, as signs of a slowdown in economic activity in the rest of EU started to emerge in 1992, the optimal current account improved, as if in anticipation of a negative shock to income in Portugal. In the same vein, the balance showed a deterioration in 1993, followed by some improvement in 1994 as economic activity firmed.

144. The success of the model in tracking the actual current account over the last forty years is somewhat surprising, in view of the existence of capital controls as recently as the early 1990s. This behavior, which has been observed also in some other small European economies (e.g., Denmark and Sweden),⁶⁶ appears to contradict the notion popularized by Feldstein and Horioka (1980) that international capital flows react insufficiently to changes in desired saving. Indeed, the variance of the optimal current account of Portugal is close to that of the actual current account. This suggests that, except on occasions of a very large widening of the current account deficit, Portugal has not faced major borrowing constraints.

145. The fact that the “optimal” current account shows relatively large swings can, on the other hand, be seen as consistent with the stylized fact that the business cycle in Portugal is very pronounced, as noted in Chapter I. While external shocks appear to have had a strong impact on the economy, shocks to income can probably also be traced to domestic fiscal or monetary policy. For instance, while the current account deteriorated following the two oil shocks, policy actions may have also contributed to this outcome. A deterioration of the current account can be exacerbated if corrective policies perceived as required to redress some macroeconomic imbalance, while amply anticipated, are delayed.⁶⁷ In these cases, financial saving may be diverted toward durable goods, with a further impact on the current account (since a good proportion of these goods is imported). Fluctuations of emigrants’ remittances, which have at times contributed to the observed swings in the current account, could also be seen under the light of optimal saving behavior in anticipation of policy changes (Schmitt, 1981 conjectures that they react inter alia to anticipation of changes in the exchange rate and financial policies).⁶⁸ It would be of interest to test these conjectures using the recently released “long series,” but such work extends beyond the scope of the present paper.

146. Finally, the deterioration of the optimal balance since 1995 could be interpreted as agents’ reactions to increasingly positive expectations, which were most likely not previously anticipated. In particular, the decline in saving may reflect the firming of the prospects of an early participation in EMU, with the associated perception that participation will strengthen economic performance. The anticipated acceleration of growth (fueled also by factors such as the 1998 World Expo) would thus justify a temporary deterioration of the current account. If this is the case, and provided financial policies remain prudent, a narrowing of the current account deficit would be likely looking forward.

E. Conclusion

147. Portugal’s current account seems in good measure to reflect a desire on the part of residents to smooth consumption. A simple intertemporal model aimed at explaining fluctuations in the current account based on the permanent income hypothesis could not be statistically rejected. Moreover, the “optimal” current account implied by the model tracked the actual current account quite closely. The emergence of a deficit in the optimal current account in recent years appears to signal agents’ expectations of a further acceleration of economic activity in the period ahead. The model would predict, however, that the current account deficit should narrow as the expected positive events, largely of an exceptional nature, materialize.