EXPORT INSTABILITY—uncertainty about the export earnings accruing to a country (which empirically arises mainly from price or terms of trade uncertainty rather than uncertainty about export volumes)—is an important source of macroeconomic uncertainty in developing countries. These countries have taken a number of different approaches to reducing their exposures to instability in export revenues, including diversification of the export base and the use of hedging instruments linked to commodity prices. These strategies have by no means eliminated the problem of export instability for these countries, however. Diversification of the export base is really only a long‐term solution toward the goal of self‐ insurance and may not even be desirable if such diversification runs counter to the country’s comparative advantage. As to available hedging instruments, these markets remain limited to a selected range of the commodities in which developing countries have a significant exposure. The risk associated with export instability in commodity‐producing developing countries must therefore largely be regarded as nondiversifiable.

The relevance of this issue has increased recently as policymakers in a number of developing countries have turned their attention to the problems associated with the sharp decline in commodity prices since the mid‐1980s. In addition to the negative trend in prices that has been debated in the literature at least since the contributions of Prebisch (1950) and Singer (1950), the steady increase in the volatility of commodity prices over the past two decades is an issue that has recently come to the fore.¹ Greater volatility in commodity prices translates into increased variability in the export earnings of a number of commodity‐exporting countries. Without domestic policies designed to eliminate or reduce the underlying volatility in world prices, an optimal response by domestic agents would be to build up precautionary balances to enable them to protect consumption (thereby increasing welfare) when particularly large real income shocks occur.² Controlling for the other determinants of saving behavior, an implication of the precautionary saving hypothesis is that greater uncertainty in export earnings should he correlated with increased saving. Moreover, in an open economy, this saving should be reflected at a macroeconomic level in the external current account balance, which simply measures a country’s national saving net of investment.

This paper considers the effect of export instability on the external balance in a dynamic optimizing model with incomplete markets for risky assets. As argued previously, the assumption of incomplete markets is more realistic than the alternative, since international capital markets have done little yet to diversify the risks facing developing countries. The assumption of a dynamic economy is made to capture the effects of export instability on the trade balance, which is intertemporal in nature. The main channel emphasized in this paper is the effect of export instability on saving behavior, through the precautionary saving motive. This channel has recently received renewed attention in the literature, which argues that labor income uncertainty is an important factor in explaining household saving behavior in the United States (see, for example, Caballero (1990)). We extend this idea to the aggregate level and consider a small commodity‐producing developing country that faces a nondiversifiable risk in its export earnings. In this case, export revenues play the role of household labor income in the existing literature, and the effect of export instability on saving behavior should show up in the trade balance or the current account, which measures the economy’s saving net of investment.

This paper draws on two strands in the literature: the first is the literature dealing with current account determination in an intertemporal context (intertemporal current account (ICA) models for short)³; the second, alluded to previously, is the recent (closed‐economy macroeconomic) literature on precautionary saving.⁴ The modern ICA literature, which mostly assumes full certainty, shows that the effect on the current account of a terms of trade shock depends on the persistence of that shock. If a favorable shock is expected to be short‐lived, its effects on saving behavior, and hence on the current account, will be much larger than if the shock is expected to last a long time. In fact, what really matters for the current account is the present discounted value of expected changes in real export earnings. If export earnings today are high relative to their average expected value in the future, the trade balance will show a surplus (smaller deficit) as agents accumulate foreign assets to keep consumption at desired levels when exports return to trend levels.

The ICA literature has generally ignored uncertainty. The incorporation of uncertainty into otherwise standard intertemporal models of the current account creates an additional channel through which export instability (induced either by relative price or volume effects) affects macroeconomic behavior. Specifically, if export earnings are uncertain, their volatility (in addition to their expected value, discussed above) will also influence the trade balance. This is the analogue to the effect of uncertainty in household labor income on savings behavior emphasized in the (closed‐economy) precautionary saving literature.

In an open economy, the current account measures the economy’s saving net of investment. If the economy’s export earnings are more uncertain, the economy will tend to increase its saving (and hence its current account surplus) to insure itself against the increased risk in its income stream. This additional saving (like the asset accumulations of an individual household) is made because it helps to reduce the volatility of consumption and utility when export earnings are stochastic. International borrowing and lending in this context thus has a double payoff to the country, enabling it to smooth consumption relative to both expected changes in revenues over time and to the volatility of the underlying revenue process.⁵

Our theoretical results thus emphasize two main channels through which export instability affects the external balance. The first is the standard consumption‐smoothing channel that has been emphasized in recent contributions to the current account literature, where certainty equivalence is assumed. The second is the volatility of export earnings, which directly affects the economy’s demand for precautionary saving, and thereby alters the external current balance. The higher this volatility, the higher will he the current account surplus (the smaller the deficit) that the country will wish to maintain to insure itself against future shocks. This channel has been ignored in previous work on the current account (that assumed certainty equivalence) and that is tested for in this paper.

The next section presents a stylized model of a small commodity‐ exporting country that allows the current account to be solved for analytically in terms of the two channels described above. Then we test this model against data for a large sample of developing countries. Because of the relative paucity of data in the time‐series dimension, we estimate the model for a panel consisting of about 60 developing countries, covering Africa, Asia, the Middle East, and Latin America, over a 25‐year period. The empirical results, taken up more fully below, reveal that export instability has, in general, exerted a systematic effect on saving behavior and on the external balance of developing countries. An increase in export instability, other things being equal, raises the precautionary demand for saving and the external balance of developing countries. Moreover, the effect is not insignificant—with precautionary saving accounting for perhaps 3½ percent of average imports for the nonfuel primary commodity exporters and up to 14 percent of average imports for the fuel exporters. The results therefore suggest that agents in these countries may have been more successful in using financial markets to insure themselves against shocks to their export earnings than had previously been believed. Finally, the main conclusions of the paper are summarized in Section III.

I. The Model

The model developed in this section is highly stylized, and the structure is kept as simple as possible while still permitting an analysis of the question of how export instability affects saving behavior and the external balance in developing countries. We consider a small developing country that consumes a single composite commodity, which we assume is not produced domestically. In addition, the country produces a good whose domestic consumption is assumed to be negligible. Since all domestic production is therefore shipped abroad, this good is referred to as exports. The model ignores the additional considerations arising from the incorporation of nontraded goods, since they have been emphasized elsewhere. They would only complicate the theoretical analysis (by causing the relative price of nontradables to be endogenous), and thereby make an empirical investigation of the issue of concern here—the effect of export instability on the external balance—intractable.⁶

Agents are assumed to maximize the expected value of the discounted sum of current and future utilities:

where β is the subjective discount factor, $u(\cdot )$ is the instantaneous utility function, and m_t denotes consumption of the single good, which, as noted previously, is imported from abroad. In addition to the transversality condition that rules out Ponzi‐type schemes, consumers’ decisions must satisfy their dynamic budget constraints, which hold that in any period t:

where b_t denotes the economy’s bond holdings at time t, p_t denotes the (stochastic) terms of trade at time t, x_t denotes the volume of exports and m_t the volume of imports at time t.⁷

As is known from the formal theory of consumption under uncertainty, not all utility functions are consistent with the existence of a precautionary demand for saving, even when agents are risk averse. Since the purpose of this paper is to test for the existence of such a demand, we need to adopt a class of utility function that allows for precautionary saving. Following the work of Leland (1968), Sandmo (1970), Drèze and Modigliani (1972), and Miller (1976), who showed that precautionary saving is consistent with utility functions with a positive third derivative, we assume here that the instantaneous utility function has the constant‐absolute‐risk‐aversion (CARA) form:

where $\alpha >0$ denotes the Arrow‐Pratt measure of (absolute) risk aversion.

Under the simplifying assumption that the exogenous foreign interest rate is equal to the rate of time preference, the first‐order necessary condition is given by

This condition states that the marginal utility cost of giving up one unit of good m at time t should be equated to the expected utility gain from consuming one more unit of mat t + 1. Alternatively, dividing the left‐hand side of equation (4) by the right‐hand side, the condition states that the intertemporal marginal rate of substitution should equal the ratio of the prices of present and future consumption, which here is unity.⁸

Since the issue at hand involves how saving responds to changes in the variance of export earnings over time, some assumption must be made about how agents perceive the future behavior of this variance. One way to proceed would be to consider a comparative‐statics exercise in which the current account responds to exogenous changes in the variance of (the lifetime innovation to) export earnings.⁹ The problem with such an approach is that it tries to identify agents’ systematic response to changes in uncertainty even though, within the model, these agents are assumed to have assigned a zero probability to such changes. Therefore, we choose to abandon this type of comparative‐statics exercise and assume instead that agents make their consumption decisions taking explicit account of the stochastic properties of the variance process.¹⁰

Here, we assume that the variance follows an AR(1) process with parameter ρ.¹¹ To solve for the consumption function, we use a “guess and verify” method. Our guess for the consumption process is

where ξ_t is the lifetime innovation in export earnings,

${\mathrm{\Lambda }}_{t-1}$ is the stochastic slope of the consumption path between periods t − 1 and t, which depends on the variance of ${\xi }_{t-1}$ denoted ${\sigma }_{{\xi }_{t-1}}^2$; and s_t is the innovation to Λ_t.¹² Intuitively, under certainty equivalence, the first difference of consumption would just be equal to ξ_t. In our case, however, there are two additional terms, which reflect precautionary saving behavior. A high value of the variance last period raises ${\mathrm{\Lambda }}_{t-1}$, which increases the growth rate of consumption (lowers m_t−1), in line with the precautionary saving hypothesis. A positive innovation to the variance today—which implies a positive drawing for the shock s_t—lowers consumption today m_t thereby reducing the growth rate of consumption. If ρ = 1, so that the innovation in today’s variance is permanent, agents revise upward their estimate of the future variance by the full amount of the shock, and, therefore, the effect on consumption growth is equal to the annuity value of the innovation s_t/r. If $\rho <1$ the shock is reversed in the future, and the effect of the innovation on consumption is accordingly smaller.

Substituting equation (5) into equation (4) yields¹³

If the innovations to export earnings have a normal distribution (with mean zero), then ξ will also. If, moreover, the innovations to the variance process follow a normal distribution, we can evaluate the expectations in equation (6) to yield

where ${\sigma }_s^2$ is the (known and constant) variance of s. Clearly, with Λ_t, as defined in equation (8). the guess for the consumption process in equation (5) satisfies the first‐order condition in equation (4).

Once Λ_t, has been obtained, we can guess a final form of the consumption function:

Thus, according to equation (9), consumption in any period is equal to permanent export earnings minus a term in the variance of export earnings. To check our guess, we need to show that equation (9) satisfies equation (5). Note from equation (9):

But from the budget constraint (2):

Substituting the process for Λ_t gives

which is equation (5), as was to be verified.

By definition, the current account is equal to the change in foreign assets. Using the budget constraint equation (2) together with the solution for the consumption function given in equation (9) gives a simple expression for the current account as the present value of expected declines in export earnings plus a term in the variance of the innovations to export earnings:

where Δ is the (backward) difference operator, $\mathrm{\Delta }{x}_t=x_t-x_{t-1},$, and where, from equation (8), the constant depends on the (known) variance of the shocks to the Λ process. Equation (13) clearly illustrates the implications of precautionary saving for the current account, revealing that both risk aversion (α) and the persistence of the shocks to the variance process (ρ) magnify the effect of the precautionary saving motive on the current account. Specifically, innovations to the variance process that die out quickly (low value of ρ) will have little effect on precautionary saving, and hence on the current account. In contrast, shocks to export earnings (embodied in the first term in equation (13)) that display low persistence will have a large effect on the current account as they will lead to a correspondingly large expected change in export earnings (since these will quickly return to their trend levels in the case of a transitory shock).

Equation (13) also carries a number of policy implications. It shows, for example, that the desirability of building up precautionary balances (either explicitly through a government‐sponsored stabilization fund or simply through the voluntary saving behavior of the various agents in the economy) depends on the persistence of the shocks to the variance process and also on how risk averse agents are. In addition, the feasibility of maintaining these balances is diminished if the country finds itself pushed below its subsistence level by a large permanent negative shock, or is denied access to international capital markets. The model may therefore be expected to perform best in countries in which achieving subsistence consumption levels is not difficult, and in which access to the international capital markets for consumption smoothing is possible. Also, the model may not find much evidence of precautionary saving behavior if the underlying fundamentals are relatively stable, which would likely be true for countries with a highly diversified export base.¹⁴

To implement the model empirically, we need an estimate of the variance of ξ¹⁵ The most obvious strategy would be to estimate the lifetime innovation to export earnings by fitting a univariate process for export earnings. As shown by Campbell (1987), however, such a procedure is likely to overestimate the extent of uncertainty as long as agents have more information about the future course of export earnings than is contained in its past values. When agents do possess such information, the current account ought to be a useful predictor of (Granger‐cause) subsequent movements in export earnings.¹⁶ Therefore, the procedure followed here is to estimate the following hivariate, first‐order VAR in $\left[\mathrm{\Delta }\right(p_t{x}_t),{ca}_t]$

In equation (14), the data on the current account are detrended before estimating the VAR.

The problem with equation (14) from the point of view of retrieving the innovation ξ is that ξ depends on innovations in the level of export earnings, whereas the VAR in equation (14) is in terms of the first difference of export earnings. Using the definition of ξ, and projecting equation (14) forward, however, one can show that:

Given ξ, its instantaneous variance, period by period, may be calculated by simply squaring the expression in equation (15), yielding a time‐series.${\sigma }_{\xi t}^2$¹⁷ In our empirical work, in addition to using just the current realization ξ_t we also experiment with taking an average of the current realization of ξ and one, two, and five leads and lags of ξ_t when calculating ${\sigma }_{\xi t}^2$. We refer to these various estimates of ${\sigma }_{\xi t}^2$ as the one‐(no leads or lags), three‐(one lead and one lag), five‐(two leads and two lags), and eleven‐ (five leads and five lags) period measures, respectively.

We are now in a position to test the model of the current account postulated in equation (13) and to examine whether export instability plays an important role in determining the current account in developing countries. It is useful to separate that part of the current account that does not depend on export instability and that from equation (13) is simply equal to the expected present value of future declines in real export earnings, denoted pdv:¹⁸

Then, from equation (13) the actual (optimal) current account should equal pdv plus a term in the variance of ξ. Thus, our final regression is of the form:

where, under the null, a₁=1 and $a_2/\left\{2\right[r+(1-\rho )\left]\right\}$. Since, by construction, pdv will be correlated with the error term u_t in equation (17), this regression cannot be estimated via ordinary least‐squares. To overcome this problem, an instrumental variables procedure is used to estimate equation (17), in which the instruments consist of the first difference of export earnings and two lags of the current account.

II. Empirical Results

Before turning to the empirical results, it bears emphasizing that the quantitative importance of precautionary saving will depend on both the magnitude of the coefficient a_t defined previously, as well as on how large is the variance to the lifetime innovation in export earnings. Although the empirical results reported below will shed light on this issue, it is perhaps worth trying to obtain some rough idea of the orders of magnitude involved. To this end, if a typical value of the risk aversion parameter is taken, say, about 2.0, and the autoregressive coefficient on the variance assumed is about 0.25,¹⁹ this would yield a value for a₂ of about one‐third for any plausible value of the real interest rate.²⁰ Higher values of the risk aversion or autoregressive parameters could easily result in a value for a₂ of about 0.5.²¹

Table 1.

Volatility of Lifetime Innovation in Export Earning^a

^aThe volatility measures are rendered unitless by dividing the (one period measure of the) variance of the lifetime innovation by the square of average imports, taking the square root, and multiplying by 100 percent. Average imports, rather than the innovation, are used to deflate the variance, because the innovation has a mean of zero.

Table 1.

Volatility of Lifetime Innovation in Export Earning^a

Country grouping	Volatility (in percent)
By region
Asia	22.88
Africa	37.97
Middle East	72.37
Western Hemisphere	18.72
By predominant export
Fuel	64.75
Manufactures	17.25
Primary products	31.24
Services	7.60
Diversified	15.96

^aThe volatility measures are rendered unitless by dividing the (one period measure of the) variance of the lifetime innovation by the square of average imports, taking the square root, and multiplying by 100 percent. Average imports, rather than the innovation, are used to deflate the variance, because the innovation has a mean of zero.

View Table

Table 1.

Volatility of Lifetime Innovation in Export Earning^a

Country grouping	Volatility (in percent)
By region
Asia	22.88
Africa	37.97
Middle East	72.37
Western Hemisphere	18.72
By predominant export
Fuel	64.75
Manufactures	17.25
Primary products	31.24
Services	7.60
Diversified	15.96

^aThe volatility measures are rendered unitless by dividing the (one period measure of the) variance of the lifetime innovation by the square of average imports, taking the square root, and multiplying by 100 percent. Average imports, rather than the innovation, are used to deflate the variance, because the innovation has a mean of zero.

As to the variance, Table 1 provides information on the volatility of the lifetime innovation in export earnings. The underlying volatilities vary considerably across different regions and by type of predominant export.²² For example, the coefficient of variation varies from about 19 percent for the Western Hemisphere countries to 72 percent for the Middle Eastern countries, reflecting the variability in oil prices over the two oil shocks that our sample spans. Regarding type of predominant export, actual volatilities for the diversified exporters and for the World Economic Outlook’s (WEO) classification of exporters of services and recipients of private transfers are 16 percent and 8 percent, respectively, Much higher volatilities are recorded by exporters of primary products (31 percent) and fuel exporters (65 percent). The volatility of exporters of manufactures, at just over 17 percent, lies in the middle range.

How large is precautionary saving relative to, say, average imports in our sample? Using the entries in Table 1 and a conservative value for the coefficient a₂ of one‐third gives some indication. Since, as explained below, the regressor in the empirical work is the variance divided by average imports,²³ the entries in Table 1 divided by 100 percent and squared give the value of the regressor as a fraction of average imports. In percentage terms, one can see that the numbers range from 0.6 percent to 41.9 percent by type of export, and from 3.5 percent to 52.4 percent by region. Multiplying these numbers by one‐third thus gives some indication of the relative importance of precautionary saving. For the African region, for example, precautionary saving would account for about 5 percent of average imports over the sample, whereas for the commodity‐exporting group, they would account for about 3½ percent, and for the fuel exporters. they would make up as much as 14 percent.

The empirical analysis was undertaken using annual data for 60 developing countries over 1965–91. The source for all data was the World Bank’s World Tables. Developing countries from all regions were included in the study, which covered 26 countries from Africa: Algeria, Benin, Burkina Faso, Cameroon, the Central African Republic, Congo, Côte d’Ivoire, Ethiopia, Gabon, The Gambia, Ghana, Guinea Bissau, Kenya, Liberia, Morocco, Madagascar, Mali, Mauritius, Malawi, Niger, Nigeria, Senegal, Tunisia. Uganda, Zaire, and Zimbabwe; 11 countries from Asia: Bangladesh, China, Indonesia, India, Korea, Malaysia, Pakistan, the Philippines, Papua New Guinea, Singapore, and Thailand; 7 countries from the Middle East: Egypt, Israel, Jordan, Kuwait, Lebanon, Libya, and Saudi Arabia; and 17 countries from the Western Hemisphere: Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Ecuador, El Salvador, Guatemala, Honduras, Jamaica, Mexico, Panama, Paraguay, Peru, Uruguay, and Venezuela.

Equation (17) was estimated using an instrumental variables procedure over the entire sample as well as over various subgroupings to he discussed below. Country‐specific constants (“fixed effects”) were allowed for in the regressions. In addition, because the coefficient on the variance in equation (17) depends on the degree of absolute risk aversion (α), and because the latter is unlikely to be independent of the level of consumption or imports. we deflate each country’s variance measure by average imports. This implies that the coefficient on the deflated variance measure should now depend on relative risk aversion, which is more likely to be constant across countries with very different income levels, as in our sample. Finally, because the net interest income component of the current account is a predetermined variable (because it depends on the last period’s asset stock), precautionary saving will be reflected mainly in the trade balance, which is therefore used as the dependent variable in the regressions reported below.

Table 2 reports the estimation results for the full sample, using the one‐, three‐, five‐, and eleven‐period measures of the variance. Overall, the model works well, in terms of the overall fit of the regressions, the sign of the point estimates, and, for the coefficient on pdv, the magnitude of the parameter estimate. As mentioned previously, pdv represents the certainty‐equivalent portion of the model, and, theoretically, the parameter estimate should be unity. Table 2 shows that in most cases the null hypothesis that this parameter is in fact unity cannot be rejected.²⁴ In all cases, the variance measure exerts a positive influence on the trade balance, so that as uncertainty about real export earnings increases, agents increase their precautionary saving to insure themselves against increased risk in their income streams. In all cases—independent of the number of time periods used in calculating the variance—the coefficient on the variance is statistically different from zero.

Several factors will influence the model’s overall performance. First, turning to the certainty‐equivalent portion of the model, it is unlikely that all countries in the sample will he equally able to smooth away the cyclical component of shocks to export earnings by using the international capital market. Some countries may be completely excluded from the international capital market (either because private agents are subject to capital controls and governments do not behave optimally or because foreign banks perceive the countries to be insolvent and thus unable to repay loans), whereas others may find it easy to borrow abroad. It would thus be inappropriate from an econometric standpoint to constrain the coefficient on pdv—which measures the country’s ability to use the international capital market to smooth away the cyclical component of export earnings—to be equal across all countries in the sample.

Second, an important factor in being able to detect a systematic effect of instability about the terms of trade or export receipts of a country is that the instability actually gets transmitted to the agents in the economy that are going to do the saving. If institutional arrangements result in an artificial smoothing of prices in some countries even though underlying world prices are unstable, and if the government, say, does not undertake the saving that would otherwise have been undertaken by the private sector, then the effect of export instability on the external balance will not be detectable. Since the extent to which such institutional arrange-meats are important is likely to differ across countries, this factor too argues in favor of looking at the results in a more disaggregated fashion.

Table 2.

Effect of Export Instability on Trade Balance: Full Sample ^a

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

Table 2.

Effect of Export Instability on Trade Balance: Full Sample ^a

		Coefficient estimate	White‐standard error
One‐period variance measure
pdv		0.86*	0.14
Variance		0.13*	0.04
	R2 = 0.76
Three‐period variance measure
pdv		0.78*	0.11
Variance		0.25*	0.08
	R2 = 0.81
Five‐period variance measure
pdv		0.63*	0.23
Variance		0.44*	0.17
	R2 = 0.82
Eleven‐period variance measure
pdv		0.86*	0.14
Variance		0.13*	0.04
	R2 = 0.76

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

View Table

Table 2.

Effect of Export Instability on Trade Balance: Full Sample ^a

		Coefficient estimate	White‐standard error
One‐period variance measure
pdv		0.86*	0.14
Variance		0.13*	0.04
	R2 = 0.76
Three‐period variance measure
pdv		0.78*	0.11
Variance		0.25*	0.08
	R2 = 0.81
Five‐period variance measure
pdv		0.63*	0.23
Variance		0.44*	0.17
	R2 = 0.82
Eleven‐period variance measure
pdv		0.86*	0.14
Variance		0.13*	0.04
	R2 = 0.76

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

Third, the underlying uncertainty of export earnings may differ according to how much the country’s export base is heavily concentrated in a particular commodity. Indeed, export diversification strategies are often recommended as a means of attaining self‐insurance, since export earnings are likely to be less volatile for countries with more diversified exports. If export diversification results in highly stable export earnings, precautionary saving behavior may be very difficult to identify empirically²⁵.

The first type of disaggregation that we turn to is the WEO classification of developing countries by predominant export. The WEO classifies developing countries into four broad categories by predominant export (usually meaning more than 50 percent of the total): fuel exporters; exporters of manufactured products; exporters of primary products; and exporters of services and recipients of private transfers. Countries whose export earnings are not dominated by any one of the above categories are classified as having a diversified export base. Thus, there are five categories, denoted below by fuel, manufactures, primary commodities, transfers, and diversified.

Table 3 reports the empirical results for the five country groupings, and the one‐, three‐, five‐, and eleven‐period variance measures, respectively. In the table, the constrained regression assumes that the effect of pdv is the same in all countries within a region, whereas the unconstrained regression allows this parameter to vary across countries. Since the various slope coefficients (one for each of the 60 countries) on the pdv are not of immediate interest, they are not reported in the table but are available from the authors on request.

As mentioned previously, our priors tell us that the results of the unconstrained regression are likely to provide more reliable information, since they do not assume that factors that affect the ability of countries to smooth away the cyclical component of export earnings are equal in all countries. Nevertheless, if the constraint that the slope coefficients on pdv is equal across countries is true for a particular country grouping, the efficiency of parameter estimates will increase, which could result in a higher coefficient of determination (which occurs in a few cases in Table 3).

The results show a consistent pattern. Turning first to the effect of the variance of export earnings, this parameter is statistically significant for the fuel exporters, exporters of manufactures, and exporters of primary commodities, in virtually all cases (except the five‐year variance measure), independent of which variance measure is used. This result contrasts sharply with those for the recipients of private transfers and countries with a diversified export base. The latter results may simply reflect the fact that for these countries, the underlying uncertainty is relatively small and therefore the data are unable to reveal much evidence of precautionary saving.²⁶ Second, the model’s overall performance seems to be best for the fuel exporters, as measured by the high coefficient of determination for this subgroup (usually above 90 percent). For the manufacturing exporters (unconstrained regressions), even this simple model explains between 25 percent and 30 percent of the variation in the trade balance, whereas for the commodity exporters. R²s vary between 5 percent and 20 percent for the unconstrained regressions (they are higher for the constrained regressions for some variance measures). For the remaining subgroupings, the model performs well overall, but the variance is not found to play a statistically significant role, with the slope coefficients on the pdv variable having much more explanatory power.²⁷

Table 3.

Effect of Export Instability by Type of Exporter: Selected Variance Measures^a

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

Table 3.

Effect of Export Instability by Type of Exporter: Selected Variance Measures^a

	Constrainedb				Unconstrainedc
	1‐period	3‐period	5‐period	11‐period	1‐period	3‐period	5‐period	11‐period
Fuel exporters
Coefficient on variance	0.09*	0.22*	0.43*	0.09*	0.08*	0.20*	0.47	0.08*
White‐standard error	0.03	0.08	0.19	0.03	0.03	0.10	0.33	0.03
R2	0.96	0.94	0.90	0.96	0.97	0.95	0.87	0.97
Exporters of manufactures
Coefficient on variance	0.54*	0.75	0.67	0.54*	0.46*	0.80*	0.54	0.46*
White‐standard error	0.19	0.41	0.93	0.19	0.16	0.35	0.78	0.16
R2	0.02	0.01	0.31	0.02	0.25	0.27	0.32	0.25
Exporters of primary commodities
Coefficient on variance	0.15*	0.23*	0.31*	0.15*	0.20*	0.25*	0.41*	0.20*
White‐standard error	0.04	0.04	0.03	0.04	0.06	0.05	0.07	0.06
R2	0.09	0.32	0.69	0.09	0.05	0.11	0.19	0.05
Exporters of services and recipients of private transfers
Coefficients on variance	5.78	–4.10	–16.07	5.78	–2.00	–1.73	‐0.30	–2.00
White‐standard error	4.67	7.52	9.62	4.67	2.47	2.64	2.07	2.47
R2	0.03	0.01	0.06	0.03	0.96	0.96	0.97	0.96
Diversified exporters
Coefficients on variance	0.02	–0.46	–0.87	0.02	–0.02	–0.57	–1.16	–0.02
White‐standard error	0.18	0.36	0.51	0.18	0.23	0.48	0.66	0.23
R2	0.49	0.42	0.43	0.49	0.27	0.22	0.27	0.27

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

View Table

Table 3.

Effect of Export Instability by Type of Exporter: Selected Variance Measures^a

	Constrainedb				Unconstrainedc
	1‐period	3‐period	5‐period	11‐period	1‐period	3‐period	5‐period	11‐period
Fuel exporters
Coefficient on variance	0.09*	0.22*	0.43*	0.09*	0.08*	0.20*	0.47	0.08*
White‐standard error	0.03	0.08	0.19	0.03	0.03	0.10	0.33	0.03
R2	0.96	0.94	0.90	0.96	0.97	0.95	0.87	0.97
Exporters of manufactures
Coefficient on variance	0.54*	0.75	0.67	0.54*	0.46*	0.80*	0.54	0.46*
White‐standard error	0.19	0.41	0.93	0.19	0.16	0.35	0.78	0.16
R2	0.02	0.01	0.31	0.02	0.25	0.27	0.32	0.25
Exporters of primary commodities
Coefficient on variance	0.15*	0.23*	0.31*	0.15*	0.20*	0.25*	0.41*	0.20*
White‐standard error	0.04	0.04	0.03	0.04	0.06	0.05	0.07	0.06
R2	0.09	0.32	0.69	0.09	0.05	0.11	0.19	0.05
Exporters of services and recipients of private transfers
Coefficients on variance	5.78	–4.10	–16.07	5.78	–2.00	–1.73	‐0.30	–2.00
White‐standard error	4.67	7.52	9.62	4.67	2.47	2.64	2.07	2.47
R2	0.03	0.01	0.06	0.03	0.96	0.96	0.97	0.96
Diversified exporters
Coefficients on variance	0.02	–0.46	–0.87	0.02	–0.02	–0.57	–1.16	–0.02
White‐standard error	0.18	0.36	0.51	0.18	0.23	0.48	0.66	0.23
R2	0.49	0.42	0.43	0.49	0.27	0.22	0.27	0.27

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

The second type of disaggregation that we undertake is a regional one, which is reported in Table 4. The picture that emerges is that the model seems to work best for the Middle East region, where the variance is highly significant in virtually all cases, and the predictive capacity of the model is also high, implying that the coefficient(s) on the pdv slope terms (not reported) are also highly significant. These results are consistent with those reported in Table 3, since this region is dominated by the fuel exporters. The results suggest that these economies in particular seem to save for precautionary purposes in response to increases in the underlying volatility of their export earning streams.

Table 4 also reveals that the data fail to find any significant effect of export instability on saving behavior in the African region. Since subsistence consumption levels are not always assured in these countries, it may not be feasible to maintain precautionary balances despite the increased volatility in export earnings, and, more generally, liquidity constraints and lack of access to international capital markets may play a role.

The model also seems to work well for the Asian countries, with a majority of the regressions finding a statistically significant effect of export instability on precautionary saving. The model’s overall predictive ability varies depending on the variance measure used and on whether the constraints imposed on the pdv slope coefficients are binding; nevertheless, in some cases, R²s of nearly 50 percent are achieved. Finally, if one recalls that we have no priors about which of the particular variance measures should perform best,²⁸ the results for the Western Hemisphere region are reasonable, at least for the five‐year variance measure where slope coefficients on the pdv variable are not constrained. In that case, the variance is significant at the 10 percent level and the R² is above 70 percent.

Table 4.

Effect of Export Instability by Region: Selected Variance Measures^a

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

Table 4.

Effect of Export Instability by Region: Selected Variance Measures^a

	Constrained b				Unconstrained c
	1‐period	3‐period	5‐period	11‐period	1‐period	3‐period	5‐period	11‐period
Africa
Coefficient on variance	0.10	0.11	0.23	0.10	0.05	0.04	0.18	0.05
White‐standard error	0.18	0.16	0.28	0.18	0.16	0.20	0.30	0.16
R2	0.53	0.54	0.52	0.53	0.53	0.41	0.12	0.53
Asia
Coefficient on variance	0.54*	0.75*	0.60	0.53*	0.43*	0.54	0.05	0.43*
White‐standard error	0.18	0.37	0.82	0.18	0.12	0.36	1.01	0.12
R2	0.06	0.07	0.12	0.06	0.47	0.18	0.14	0.47
Middle East
Coefficient on variance	0.08*	0.20*	0.47	0.08*	0.08*	0.20*	0.48	0.08*
White‐standard error	0.03	0.10	0.32	0.03	0.03	0.10	0.34	0.03
R2	0.97	0.95	0.90	0.97	0.98	0.96	0.90	0.98
Western Hemisphere
Coefficient on variance	0.19	0.22	0.35	0.19	0.01	0.07	0.32	0.01
White‐standard error	0.14	0.23	0.31	0.14	0.10	0.14	0.18	0.10
R2	0.66	0.63	0.63	0.66	0.72	0.63	0.71	0.72

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

View Table

Table 4.

Effect of Export Instability by Region: Selected Variance Measures^a

	Constrained b				Unconstrained c
	1‐period	3‐period	5‐period	11‐period	1‐period	3‐period	5‐period	11‐period
Africa
Coefficient on variance	0.10	0.11	0.23	0.10	0.05	0.04	0.18	0.05
White‐standard error	0.18	0.16	0.28	0.18	0.16	0.20	0.30	0.16
R2	0.53	0.54	0.52	0.53	0.53	0.41	0.12	0.53
Asia
Coefficient on variance	0.54*	0.75*	0.60	0.53*	0.43*	0.54	0.05	0.43*
White‐standard error	0.18	0.37	0.82	0.18	0.12	0.36	1.01	0.12
R2	0.06	0.07	0.12	0.06	0.47	0.18	0.14	0.47
Middle East
Coefficient on variance	0.08*	0.20*	0.47	0.08*	0.08*	0.20*	0.48	0.08*
White‐standard error	0.03	0.10	0.32	0.03	0.03	0.10	0.34	0.03
R2	0.97	0.95	0.90	0.97	0.98	0.96	0.90	0.98
Western Hemisphere
Coefficient on variance	0.19	0.22	0.35	0.19	0.01	0.07	0.32	0.01
White‐standard error	0.14	0.23	0.31	0.14	0.10	0.14	0.18	0.10
R2	0.66	0.63	0.63	0.66	0.72	0.63	0.71	0.72

^aAn asterisk denotes significance of the coefficient at the 5 percent level.

^bThe constrained regression assumes that the slope coefficients on the pdv variable are equal across countries.

^cThe unconstrained regression allows the slope coefficients on the pdv variable to vary across countries.

III. Conclusion

Uncertainty about the terms of trade and/or export revenues is an important source of macroeconomic uncertainty in a number of developing countries. Theory dictates that smoother consumption streams and hence higher welfare can be achieved if agents increase their saving when they perceive that the variability of their export receipts will increase. This precautionary saving motive has been argued to underlie a significant portion of aggregate saving in a number of industrial countries, including the United States. In this paper, we sought to test empirically whether this precautionary saving motive was present in developing countries that have uncertain export earnings. Our finding was that in general the precautionary motive has significantly influenced saving behavior and the external balance of developing countries; quantitatively, its importance has varied across the sample, accounting for about 14 percent of average imports for fuel exporters and about 3½ percent of average imports for exporters of nonfuel primary commodities.

Of course, whether export instability increases precautionary saving depends on a number of factors. For example, some countries have instituted explicit stabilization funds whose purpose is to build up precautionary balances to insure consumption (public or private) in the face of commodity price shocks. In some cases, however, the fund’s purpose can equally well be achieved by accumulating (foreign) assets whose returns are uncorrelated with the export revenues accruing to the country. In either case, the resulting balances are part of domestic saving. and, therefore, in a financially open economy, changes in the level of such balances will be reflected in the country’s external current account balance, which after all is identically equal to national saving net of investment.

However, in a number of developing countries, government policies may inhibit the transmission of relative price shocks to the domestic economy, which would mitigate the private sector’s incentives to accumulate precautionary balances. If the government does not accumulate these balances in place of the private sector—that is, if it does not behave optimally—one will be unable to find much evidence of precautionary saving in the data. Alternatively, if the underlying source of uncertainty is reduced because the country has a highly diversified export base, again one will not find much evidence that precautionary saving is important for such countries. Finally, the inability to maintain consumption above a subsistence level, generalized liquidity constraints, and lack of access to international capital markets may also adversely affect the model’s performance in some of the poorer countries in the sample. Agents there will not find it feasible to maintain precautionary balances when they perceive that the uncertainty of their export earnings has increased.

The evidence presented in this paper suggests that, for the developing countries as a whole, the data do indeed support the view that these countries have attempted to build up precautionary balances when export instability increases. Moreover, a more disaggregated view of the results suggests that the precautionary saving effect is stronger for countries with an export base more heavily concentrated in a few commodities than it is for countries with a highly diversified export structure. Finally, there was also some evidence to suggest that the precautionary saving motive varied from region to region, which might reflect both the differing extents to which price stabilization measures are used across the differing regions and the differing abilities of the various agents in these countries to use capital markets to smooth consumption. An implication would be that, for a given level of underlying uncertainty, countries that use public policy to try to insulate themselves from relative price shocks may significantly benefit by relying instead on international capital markets to insure themselves against commodity price shocks.