A. The Baseline Model

Equilibrium

We briefly list the model’s equilibrium conditions before turning to the main results. First, domestic demand and supply of nontradables must be equal at all points in time:

$\begin{array}{ccc}{C}^n(t)=Q^n(t),\forall t.& & \mathbf{(}\mathbf{14}\mathbf{)}\end{array}$

On the demand side, there is a fixed relationship between spending on the two consumption goods at any moment:

$\begin{array}{ccc}[\alpha /(1-\alpha )]C^m(t)=p^n(t)C^n(t).& & \mathbf{(}\mathbf{15}\mathbf{)}\end{array}$

We assume that

$\begin{array}{ccc}r=\rho ,& & \mathbf{(}\mathbf{16}\mathbf{)}\end{array}$

so that the model possesses a stationary state. Next, we note that since this is a dynamic multi-sectoral model, both the inter-temporal and the intra-temporal elasticities of substitution play a role. In general, movements in the consumption of the two goods when out of stationary state are negatively (positively) correlated if the intertemporal elasticity of substitution is smaller (greater) than unity [cf. Brock (1988)]. Intuitively, given a unitary intra-temporal elasticity of substitution, if the intertemporal elasticity of substitution is smaller than unity, then agents prefer to rearrange their consumption bundle rather than shift total consumption over time. Hence, consumption of importables and of nontradables is negatively correlated during any period of transition. For simplicity, we now set

$\begin{array}{ccc}\sigma =1.& & \mathbf{(}\mathbf{17}\mathbf{)}\end{array}$

Consumption of importables is then constant along any perfect-foresight path, although it may still respond to unanticipated shocks:

$\begin{array}{ccc}{C}^m(t)=C^m,\forall t.& & \mathbf{(}\mathbf{18}\mathbf{)}\end{array}$

Using the intertemporal solvency condition, consumption of importables then equals the annuity on net national tradable wealth, W(t), defined as the present value of tradable output net of investment spending, minus the inherited foreign debt, all measured in terms of importables:

$\begin{array}{ccc}{C}^m=rW(t),W(t)\equiv {\displaystyle {\int }_t^{\infty }{{e}}^{-r(s-t)}[Q^m(s)-I(s)]ds+R(t)/r-D(t).}& & \mathbf{(}\mathbf{19}\mathbf{)}\end{array}$

On the supply side, labor market equilibrium requires that the value marginal product of labor be equated across sectors:

$\begin{array}{ccc}{{{\displaystyle {Q}}}^{{m}}}_2(K^m,L^m)=p^n{{{\displaystyle {Q}}}^{{n}}}_2(K^n,1-L^m).& & \mathbf{(}\mathbf{20}\mathbf{)}\end{array}$

Further, investment in each sector is chosen so that the cost of installing an extra unit of capital, given by its price (=1) and the marginal adjustment cost, equals the shadow value of installed capital, qⁱ:

$\begin{array}{lll}1+{{A}}^{{i}}\prime [j^i(t)-\delta K^i(t)]\ge q^i(t),K^i\ge 0,\text{complementary slack},i=m,n,& & \mathbf{(}\mathbf{21}\mathbf{)}\end{array}$

This establishes that net investment is increasing in qⁱ. In turn, qⁱ evolves over time according to

$\begin{array}{ccc}(r+\delta )q^i(t)-{\dot{q}}^i=p^i{{{\displaystyle {Q}}}^{{i}}}_1[K^i(t),L^i(t)]+\delta {{A}}^{{i}\prime }[J^i(t)-\delta K^i(t)]& & \mathbf{(}\mathbf{22}\mathbf{)}\end{array}$

$\begin{array}{cccc}\Rightarrow & r{q}^i(t)-{\dot{q}}^i\le p^i{{{\displaystyle {Q}}}^{{i}}}_1[K^i(t),L^i(t)]-\delta ,K^i\ge 0,\text{complementary slack},i=m,n,\text{by}(21).& & \mathbf{(}\mathbf{23}\mathbf{)}\end{array}$

(22) states that the “user cost” of installed capital must equal its value marginal product. The user cost is the sum of interest and depreciation charges, net of capital gains. The value marginal product of installed capital is the sum of its value marginal product in production, and of its marginal contribution to reducing the cost of installing a given flow of investment goods.

In general, the evolutions of the two sectors’ capital stocks are linked. Changes in one sector’s capital stock will induce a shift of labor between sectors; in turn, this may also change the price of the other sector’s output relative to capital goods. Both factors will alter the other sector’s value marginal product of capital, and hence its investment incentives. Nevertheless, given certain assumptions discussed in Appendix I, this intersectoral linkage vanishes. The model may then be efficiently summarized by combining the dynamic equations for Kⁱ and for qⁱ to yield two phase diagrams, one for each sector, as shown in Figure 1. For each sector i, the horizontal axis measures Kⁱ, while the vertical axis measures qⁱ. The ${\dot{K}}^i=0$ locus is a horizontal line at the price of capital (unity). Also, a higher Kⁱ is associated with a lower sectoral value marginal product of installed capital. In equilibrium, it must therefore be associated with a lower user cost of installed capital, that is, either a lower qⁱ or a higher ${\dot{q}}^i$. Hence, the ${\dot{q}}^i=0$ locus is a downward-sloping line; to its right, qⁱ grows over time, and conversely to its left.

View Full Size

Dynamics of investment and capital in sector i.

Citation: IMF Working Papers 1999, 134; 10.5089/9781451855586.001.A001

Short dashes and arrows represent the saddle-path.

• Download Figure
• Download figure as PowerPoint slide

Natural-resource booms

We now analyze the impact of natural-resource booms. Assume that the economy is initially in stationary state. Let there be a discrete increase in R, that is, a previously unanticipated (current or future) natural-resource boom. The boom raises wealth, and hence demand for nontradables. This acts to increase the price of nontradables, which in turn draws labor away from the importable and towards the nontradable sector. Nontradable investment is stimulated both by the increase in nontradable prices, and by the shift in labor towards the nontradable sector. The latter, however, discourages importable output and investment; this is the Dutch Disease.

Figures 2 and 3 illustrate the dynamics. When the shock occurs, neither the ${\dot{K}}^n=0$ nor the ${\dot{K}}^m=0$ locus change. The ${\dot{q}}^n=0$ locus, and hence the nontradable saddle-path, shift up, whereas the ${\dot{q}}^m=0$ locus, and hence the importable saddle-path, shift down, qⁿ, nontradable investment, and nontradable output increase discretely upon impact, and at all future dates are higher than in the absence of the shock. q^m, importable investment, and importable output decrease discretely upon impact, and at all future dates are lower than in the absence of the shock.

View Full Size

Response of investment and capital in the nontradable sector to a natural-resource boom or to a shift in tastes towards nontradables.

Citation: IMF Working Papers 1999, 134; 10.5089/9781451855586.001.A001

Long dashes represent the pre-shock ${\dot{q}}^n=0$ locus. Short dashes and arrows represent the response of nontradables to the shock. Only the post-shock arrows of motion are shown.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Response of investment and capital in the importable sector to a natural-resource boom or to a shift in tastes towards nontradables.

Citation: IMF Working Papers 1999, 134; 10.5089/9781451855586.001.A001

Long dashes represent the pre-shock ${\dot{q}}^m=0$ locus. Short dashes and arrows represents the response of importables to the shock. Only the post-shock arrows of motion are shown.

• Download Figure
• Download figure as PowerPoint slide

After the initial impact, and during the transition to the new stationary state, the nontradable capital stock and output increase monotonically, while net nontradable investment decreases monotonically. The importable capital stock and output decrease monotonically, while net importable investment increases monotonically. The relative price of nontradables decreases monotonically. Hence, nontradables prices initially overshoot their new stationary-state value. Here, real exchange rate overshooting arises not because of price stickiness, but simply because it is optimal to spread out changes in the capital stock over a discrete period of time.

In the new stationary state, the relative price of nontradables and the manufacturing capital-labor ratio will exceed their pre-shock value. Holding constant the price of foreign nontradables, this yields two corollaries. First, the short-run deviations from purchasing-power parity (PPP) exceed the long-run deviations. Second, even in the long-run, PPP does not hold.

The response of aggregate investment and of the aggregate capital stock is in general ambiguous. However, they will increase if labor reallocation only has a small impact on the marginal product of capital. To see this, let the marginal product of manufacturing capital be independent of the sectoral supply of labor. Then, upon impact, the price of nontradables and their share in total employment still increase discretely, but manufacturing investment is unaffected.²⁷ Hence, both nontradable and aggregate investment increase discretely, and must at all future dates be higher than in the absence of the shock. In the general case where labor reallocation does affect the marginal product of capital, the resource boom will raise the marginal product of capital in nontraded and lower it in the traded sector; the former provides a further boost to aggregate investment, while the latter reduces it.

A further effect is related to the capital intensities of the sectors. To see this, let the production function be linearly homogenous in capital and labor alone. Then, for a given stationary-state user cost of capital, the stationary-state capital-labor ratios in both sectors are uniquely determined. Hence, if nontradables are relatively capital-intensive (labor-intensive), the aggregate stationary-state capital stock increases (decreases). In turn, this implies an increase (decrease) both in “average” investment during the transition, and in stationary-state investment.

At the time of the boom, the instantaneous change in savings is

$\begin{array}{ccc}{\Delta }S(t)={\Delta }[p^r(t)Q^r(t)-R(t)]+{\Delta }[Q^m(t)-r{\displaystyle {\int }_t^{\infty }{{e}}^{-r(s-t)}{Q}^m(s)ds}]+{\Delta }r{\displaystyle {\int }_t^{\infty }{{e}}^{-r(s-t)}I(s)ds}.& & \mathbf{(}\mathbf{24}\mathbf{)}\end{array}$

The first term on the R.H.S. depends on the change in the anticipated time path of natural-resource export earnings. It is positive for a transitory increase in natural-resource prices. It may be positive even for a permanent price increase, so long as natural resources output is expected to decline over time, e.g., because oil reserves will be quickly exhausted. It must be negative for an anticipated future increase in natural-resource prices or output. The second term is positive, since importable output is expected to decline over time. The last term corresponds to the change in “permanent” aggregate investment. As discussed above, it is in general ambiguous. The intuition behind (24) is that agents want to smooth consumption over time.

The instantaneous change in the trade balance and current account is

$\begin{array}{ccc}{\Delta }[S(t)-I(t)]={\Delta }[p^r(t)Q^r(t)-R(t)]+{\Delta }[Q^m(t)-r{\displaystyle {\int }_t^{\infty }{{e}}^{-r(s-t)}{Q}^m(s)ds}]+{\Delta }[r{\displaystyle {\int }_t^{\infty }{{e}}^{-r(s-t)}I(s)ds-I(t)]}.& & \mathbf{(}\mathbf{25}\mathbf{)}\end{array}$

The first and second term on the R.H.S. were discussed above. The last term is negative if and only if aggregate investment increases discretely upon impact.²⁸

B. Extensions

C. Summary

The intuition behind the model, and its formal predictions, is as follows. A natural-resource boom increases national wealth. This raises the consumption of importables. In addition, it increases the demand for nontradables, which acts to increase the price of nontradables relative to importables. In turn, this has two effects. First, for a given sectoral allocation of labor, so long as some fraction of the capital goods is importable, investment in nontradables increases; and so long as some fraction of the capital goods is nontradable, investment in importables decreases. Second, assuming that the natural-resource sector constitutes an enclave, labor is drawn out of the importable and into the nontradable sector. Hence, output and investment decrease in the importable sector (the Dutch Disease), but again increase in the nontradable sector. Since nontradable investment increases, the supply of nontradables gradually rises over time. This acts to reduce the price of nontradables, after its initial increase. In other words, the real exchange rate initially overshoots its new long-run equilibrium.

Aggregate investment is likely to increase under the following circumstances. First, if capital goods have a large tradable content. Second, if Dutch Disease effects are weak, for instance because the aggregate labor supply is very elastic, or else because labor is substantially immobile across sectors. Third, if the expanding nontradable sector is more capital-intensive than the contracting importable sector. Fourth, if natural-resource investment responds strongly to the shock.

The response of the current account is more likely to be negative if aggregate investment increases, if the natural-resource boom is perceived as largely permanent, and if importable output is not expected to decline through Dutch Disease effects.

A shift in private preferences towards nontradables, or an increase in government spending that falls mainly on nontradables, also raises demand for nontradables. Hence, the implications for importable and nontradable output and investment are analogous to those of a natural-resource boom. However, consumption of importables decreases rather than increases.

D. The Link Between Natural-Resource Wealth and the Terms of Trade

So far, we have conducted the analysis in terms of R, the annuity on natural-resource wealth, measured in terms of importables. We now clarify the relationship between changes in R and terms-of-trade shocks.

In the model, so long as imports are strictly positive, the terms of trade are well-defined and equal to p^r. Even in this simple case, the relationship between the response of the terms of trade and of R to a shock generally depends both on the source and nature of the shock, and on the economy’s ability to influence world natural-resource prices.

In an intertemporal setting, a shock affects wealth as soon as it is anticipated, rather than when it actually occurs. Hence, if the shock is foreseen sufficiently far in advance, the economy’s response will be largely complete by the time we observe any change in the terms of trade. Again, if a shock is expected to be largely transitory, its effect on wealth is minimal, whatever the change in the terms of trade. In what follows, we therefore assume the shocks are completely unanticipated, and are expected to display some persistence.

Assume that p^r is exogenous to the economy. First, let there be an increase in p^r. Then R must increase. The greater the anticipated persistence of the shock, and the more natural-resource output is free to respond through optimal changes in investment, the greater the increase in R. Second, let there be an increase in natural-resource output and exports, for instance, because of a discovery of natural-resource reserves. Theni R increases without any corresponding change in the terms of trade.

Now let the economy, either by itself or as part of a cartel, be large in the world market for natural resources. First, let there be an increase in foreign demand for domestic natural resources. Then, both p^r and R increase. Second, let there be an increase in natural-resource output and exports. If the increase occurs because of a cartel breakdown, and so long as the country had initially joined the cartel to promote its economic interests, then both p^r and R decrease. If instead the increase reflects greater productivity in the natural-resource sector, say because of a discovery of natural-resource reserves, then p^r decreases but, assuming that optimal export taxes are in place, R increases.

In summary, an increase in the terms of trade must imply an increase in R, if two conditions are met. First, the change in the terms of trade must be unanticipated and must be expected to display some persistence. Second, either the terms of trade must be exogenous, or else the terms-of-trade shock must reflect the strengthening and weakening of a cartel, and/or a change in foreign demand for domestic exports. In contrast, productivity shocks to the natural-resource sector of a large natural-resource exporter may cause the terms of trade and R to diverge.

Even if the above conditions are met, an increase in R need not be associated with an increase in the terms of trade. Specifically, if the volume of natural-resource exports or reserves changes over time, then R will change without any corresponding change in the terms of trade. Again, if countries differ in the size of the exportable sector, or in their perceptions about the persistence of a shock, then the impact of a given terms-of-trade shock on R must also differ across countries. Hence, changes in the terms of trade are at best an imperfect measure of changes in natural-resource wealth. This caveat applies a fortiori to changes in total wealth.

A. Interpretation of the Terms of Trade

Throughout, we interpret changes in the terms of trade as a measure of changes in wealth. Our justification is as follows. First, the major changes in the price of oil were largely unanticipated, and were expected to display some persistence. Specifically, the main oil price increases were triggered by essentially unpredictable historical events, that is, the 1973-74 Arab-Israeli war, the 1978-79 Iranian revolution, and the start of the Iran-Iraq war in 1980. In addition, surveys of informed participants in the oil market during the 1980s suggest four conclusions.³² First, the long price decline starting in 1981 was not anticipated. Second, it quickly led to a wave of successive reductions in both short- and long-term forecasts of oil prices. Third, at any given point in time, the economically relevant agents generally expect slow, constant growth in oil prices. Fourth, market expectations are predominantly adaptive and extrapolative.

Next, consider the source of terms-of-trade shocks. For non-OPEC oil exporters, the terms of trade are best seen as exogenous. Even for OPEC members, terms-of-trade shocks mainly reflected two factors. First, the strengthening and weakening of the OPEC cartel. Second, changes in foreign demand for their oil, stemming from the emergence of new oil producers, the development and exploitation of alternative energy sources, and business cycles abroad. For most countries, productivity shocks to the domestic oil sector, that is, unanticipated discoveries or depletions of oil reserves, have had a minor impact on oil prices.

As argued in Section II. D, increases in the terms of trade then imply an increase in wealth.³³ The converse, however, need not hold. Given that oil production and reserves change over time and differ across countries, changes in the terms of trade are at best an imperfect measure of changes in wealth. We control for this by interacting the terms of trade with the trade ratio, defined as [(exports + imports)/2 GDP]. We use the average trade ratio over the sample period, for the following reason. If some countries can respond to an increase in the terms of trade by shifting resources over time towards their export sector, then using the initial-year trade ratio yields a downward-biased estimate of the increase in their wealth. Conversely, using the post-shock trade ratio neglects the opportunity and adjustment costs associated with the resource shift, and yields an upward-biased estimate of the wealth increase. Again, to the extent that the movements in resources are anticipated, using the current-year trade ratio yields a misleading picture of the timing of the wealth increase. Using the average trade ratio hopefully minimizes such problems. In any event, the results are not very sensitive to the precise measure used.

Finally, measurement of the terms of trade may be subject to index-number problems for two main reasons. First, even for a given producer, oil is not completely homogenous: some oil exporters have successfully diversified into oil refining. Second, the share of non-oil tradables in exports changes over time. However, given that the terms of trade are heavily correlated both across countries, and with the price of oil relative to U.S. manufactures, these problems appear to be minor.

B. Estimation

Throughout, we use a panel rather than a time-series methodology since, if the underlying assumptions are satisfied, this allows for more precise estimates and renders omitted-variable bias a less serious problem (see Section IV. C). We have no strong priors and theory provides little guidance on the speed at which our variables adjust to terms-of-trade shocks. Hence, we take a flexible, data-determined approach to dynamics. Specifically, we estimate the regression equation

The i subscript refers to the country, and the t subscript to the year, y denotes each of the dependent variables,³⁴ TOT denotes the terms of trade, and $\overline{{\mathit{\text{TR}}}_i}$ denotes a country’s average trade ratio over the sample period./?(L) is a polynomial in the lag operator, whose order is chosen using the Schwartz Information Criterion, and whose elements, multiplied by the relevant country’s average trade ratio, denote the elasticity of the dependent variable with respect to the terms of trade at various lags. CONTROL denotes the control variables, which include a debt crisis dummy, the world real interest rate, and a linear time trend. We control for the debt crisis because it is a conceptually distinct phenomenon from a terms-of-trade shock; our results are strengthened to the extent that the crisis was itself caused by the downturn in the price of oil. We use the world real interest rate because it raises fewer endogeneity problems than domestic real interest rates, and because most of the countries in our sample were relatively well integrated into the world capital markets, at least until the onset of the debt crisis. We use a time trend to capture the effects of technological progress and other underlying sources of growth.

In (26), the intercept α varies across countries. These country-specific effects reflect the cross-country heterogeneity in the absolute size of the dependent variable, stemming from differences both in the underlying economic structure and in the units of measurement. We do not allow for year-specific effects. Since the countries in our sample, by construction, rely heavily on oil for their exports, the shocks to their terms of trade are heavily correlated. Time effects would essentially soak up this cross-country common component of terms-of-trade shocks, leaving only the very small cross-country differences in the shocks. This would inevitably make any estimated response insignificant.

In allowing for country-specific effects, we compute both the fixed-effects (FE) and the random-effects (RE) estimators. These estimators differ in both their efficiency and their consistency properties. The FE estimator is asymptotically efficient if the country effects are treated as fixed regressors, or if the model is to be analyzed conditional on the effects present in the observed sample, as is the case for in-sample predictions. The RE estimator is asymptotically efficient if the country effects are treated as random and if the model is to be analyzed unconditionally, as is the case for out-of-sample forecasts. In addition, the FE but not the RE estimator remains consistent if the country effects are correlated with the explanatory variables. Both for this last reason, and because the differences in the estimates are minor, we focus on the results from FE estimation.

Analysis of the residuals from a preliminary regression suggests the presence of first-order autocorrelation, cross-country heteroskedasticity, and cross-country correlation. We assume that the disturbances in all countries share a common AR(1) process and a common crosscountry correlation. We then estimate the autoregressive coefficient, use the Prais-Winsten transformation to remove the autocorrelation, estimate the country-specific variances, and the cross-country error correlation, compute the FGLS estimator, and iterate to convergence. The results are reported in Appendix III, Table 1, and are discussed in Section V.

Next, we carry out a series of specification tests. First, we test the hypothesis that there are country-specific effects, that is, α varies across countries. Under FE estimation, this involves a standard F test. Under RE estimation, we employ the Breusch and Pagan (1980) Lagrange Multiplier (LM) test. In all regressions, both the F and the LM test statistic are significant at the 1 percent level. We therefore reject the null of no country-specific effects. Since the LM test statistic has a block-diagonal information matrix, its use in a pretest does not affect the standard errors of the other estimators we compute.

Second, we test the hypothesis that the long-run elasticity with respect to the terms of trade is constant over time. To do this, we allow the estimated coefficients on the terms of trade to differ across the sub-periods 1965-80 and 1981-89. Since these roughly correspond to the periods of rising and of falling terms of trade, this also allows us to examine whether the responses to positive and to negative terms-of-trade shocks differ. The results of the F-test are reported in Appendix III, Table 2, and are discussed in Section V.

Third, we test the hypothesis that the long-run elasticity with respect to the terms of trade is constant across countries. The results of the F-test are reported in Appendix IV, available upon request. The null hypothesis of stability across countries is often rejected. The rejection is hardly surprising, given that we are dealing with such a broad sample of countries, but it does imply that we must determine how representative the panel estimates are. Accordingly,^ Appendix IV also presents each country’s estimated response to changes in the terms of trade.

C. Omitted-Variable Bias

Omitted-variable bias needs to be treated differently in a panel context. As shown in Appendix II, a necessary and sufficient condition for consistency of the country-by-country OLS estimator of the coefficients on the terms of trade is that, asymptotically, the terms of trade be orthogonal to any omitted variables. In contrast, a necessary and sufficient condition for consistency of the panel estimator is that, asymptotically, a weighted average of the biases in the country-by-country OLS estimates be zero.³⁵

Hence, the panel estimator can be consistent even if the terms of trade are not asymptotically orthogonal to the omitted variables in any one country. What matters is that the country-specific asymptotic biases be negatively correlated across countries, so that they average to zero. This last assumption is more credible to the extent that the relationship between the omitted variables and the terms of trade is idiosyncratic to each country and not positively correlated across countries. To the extent that the omitted variables are world variables which affect all countries, then the country-specific asymptotic biases will have the same signs. We therefore control for the debt crisis and the world interest rate. It might also be desirable to control for world output. However, since periods of rising oil prices coincided with changes in OECD activity that would have depressed growth in our sample, and vice-versa, this should strengthen the links we find between terms-of-trade changes and activity variables.

D. Non-Stationarity

There are two ways to interpret our econometric model. First, we may view the terms of trade, which are driven mainly by the world price of oil, as a fixed regressor. (26) is then the mechanism generating the rest of the data. The presence or absence of stationarity is not an issue here.

Alternatively, we may treat the terms of trade as generated by some time-series statistical model. Here, the order of integration of our data is an issue. We take a flexible approach to the spurious regression problem, adopting the Prais-Winsten adjustment for first-order serial correlation in the residuals. Blough (1992) demonstrates that, if the variables in the regression are 1(1) and not cointegrated, this procedure is asymptotically equivalent to differencing the data before estimating the model. In addition, the procedure avoids the misspecification problems associated with differencing when the regression variables are cointegrated.

A. The Importance of Investment

Overall, we find strong evidence of a positive relationship between terms-of-trade shocks and investment. Panel estimates suggest the 95 percent confidence interval (0.067, 0.382) for the mean long-run elasticity of aggregate investment with respect to the terms of trade. This investment response helps explain why the trade balance was strongly negatively related to terms-of-trade shocks, even though the export-price effect in isolation should cause a higher trade surplus.

Investment responds more strongly to a decline than to an increase in the terms of trade, even after controlling for the debt crisis. This suggests that adjustment costs are substantially larger for increases than for decreases in the capital stock. Specifically, the positive response to a favorable terms-of-trade shock may be choked off by physical bottlenecks arising, for instance, from inadequate transportation infrastructure and port facilities, which limit the ability to import capital goods. We now examine what the available evidence says about the sectoral breakdown and the causes of this investment response.

It is likely that much of this investment occurred in the nontradable sector of the economy. Although data on investment by sector are simply not available for most of these countries,³⁶ so that this claim cannot be supported directly, much of the other data are supportive. First, we do have evidence that, in the long run, output expanded in nontradable sectors such as construction and services. It is reasonable to view this increase as partly due to a higher capital stock in these sectors.

Second, our model concludes that, in any given sector, private investment is determined by the ratio of marginal q (the present value of the value marginal product of capital) to the relevant price deflator for investment goods. We know that the relative price of nontradables to importables is positively associated with terms-of-trade shocks. Also, Appendix III, Table 3, shows that the countries in our sample import many capital goods. Further, the world prices of capital goods relative to most noncommodity tradables did not change dramatically over the period. Hence, the price of nontradables relative to physical capital, and therefore the price incentive for investment in the nontradable sector, probably rose in the 1970s and declined in the 1980s.

The model also suggests that the relative price changes should stimulate investment in the oil sector. Although this would probably hold for countries experiencing exogenous changes in their export prices, OPEC raised oil prices in the 1970s partly through a deliberate policy of output restriction. Hence, for much of our period the oil sector was engineering price changes, rather than passively responding to them. In a setting where countries exercise their power in the world oil market, it is a priori unclear which way the incentives for investment in the oil sector work. Oil production quotas may be seen as a permanent ceiling on oil production, leaving little reason to invest in the oil sector’s productive capacity. On the other hand, an increase in such capacity may strengthen one’s hand in the periodic quota negotiations; in addition, there is still an incentive for cost-reducing investment.

We do know that real oil output per worker fell between 1973 and 1981. Perhaps the countries invested in modernizing the oil sector, rather than in expanding its capacity. Perhaps they did invest in expanding capacity, but just did not use this capacity fully over the period. We simply do not have the data to pin this down very precisely.

B. Consumption, Savings, and the Trade Balance

As expected, after a permanent, positive terms-of-trade shock, consumption increases. Panel estimates suggest the 95 percent confidence interval (0.107, 0.194) for the mean long-run elasticity of consumption with respect to the terms of trade. As with investment, consumption responds more strongly to a decline than to an increase in the terms of trade, even after controlling for the debt crisis. Again, this suggests that adjustment costs are substantially larger for increases than for decreases in consumption. For instance, the positive response to a favorable terms-of-trade shock may be choked off by physical bottlenecks in supply. Alternatively, it may prove psychologically difficult to increase consumption levels rapidly, whereas the capital markets may quickly enforce any required downward adjustment.

Government consumption responds more strongly than private sector consumption, particularly when we allow for a structural break over time. Three interpretations are possible. First, oil rents generally accrue to the government, which may find it politically advantageous to spend directly, rather than rebate the revenues to the private sector through lower taxes or higher transfer payments. Second, governments may systematically view shocks as more permanent than do private agents. Third, private demands for publicly-provided goods and services may be characterized by relatively high income elasticity.

Panel estimates suggest the 95 percent confidence interval (-.226, -.0618) for the mean long-run elasticity of savings with respect to the terms of trade. Given the negative impact on savings, and the strong response of investment, the trade balance should be expected to decrease. Indeed, this was the estimated response, which proved statistically significant both in the aggregate, and for 10 individual countries.³⁷

C. Nontradable Prices and Quantities

Increases in the terms of trade are very strongly associated with real exchange rate appreciations, that is, increases in the relative price of nontradables. Panel estimates suggest the 95 percent confidence interval (-.515, -.413) for the mean long-run elasticity of the real exchange rate with respect to the terms of trade. Thirteen countries out of 18 displayed a significant response.

The shocks are also associated with significant increase in value added within the nontradable sector, and in particular in construction and services. This confirms the importance of spending effects as a key mechanism in the transmission of terms-of-trade shocks to the rest of the economy: higher wealth leads to greater demand for nontradables, and hence to an increase in their relative prices and quantities. As with investment and consumption, nontradable output generally responds more strongly to a decline than to an increase in the terms of trade. Again, this presumably reflects physical bottlenecks.

D. Dutch Disease Effects

The literature strongly focuses on Dutch Disease effects. Here, the two non-oil tradables, that is, agriculture and manufacturing, are the closest approximation we have to a Dutch Disease sector. Yet we failed to detect clear contractions in either in response to an increase in the price of oil. Oil value-added did respond negatively, but this simply reflects the imposition of OPEC quotas: the oil sector is not best seen as facing exogenously given terms of trade, since part of OPEC’s strategy at the time clearly was to engineer price rises by restraining output.

The absence of a Dutch Disease may be partly explained by either the compression in oil output, or by its being an “enclave” sector which does not participate in domestic factor markets. Both factors act to reduce the pull of resources away from non-oil sectors. Still, as discussed above, the “spending effect” was operational, increasing relative prices and output in the nontradable sector. To the extent that nontradables compete with agriculture and manufacturing for scarce factors of production, or that nontradables are themselves intermediate inputs into other sectors, one might have expected a bout of the Dutch Disease.

E. Growth

Given that the relative prices between oil and other products changed so dramatically over the period, and given that the oil sector is so important in our sample, very serious index-number problems arise in even defining and measuring aggregate output. We do not feel that it is analytically useful to talk about the aggregate economy, and instead separate the oil from the non-oil sector. The former has already been mentioned. Within the non-oil aggregate, we do see an effect on GDP, reflecting mainly the expansion in nontradables. Panel estimates suggest the 95 percent confidence interval (0.0711, 0.162) for the mean long-run elasticity of non-oil value added with respect to the terms of trade. Given our evidence on investment, this response is best seen as driven by capital formation in the nontradable sector.

F. Some Problems with Extrapolation

When interpreting and above all extrapolating our results, several issues must be considered. First, the causes of the oil-price shock were different from the causes of the forecast future increase in primary commodity prices. In our sample, export prices rose largely because of oligopolistic coordination aimed at reducing output. Some responses might well be different for truly exogenous terms-of-trade shocks. In particular, both investment and output in the favored sector would probably increase. Since output was no longer constrained, we should also expect a bigger increase in wealth and hence consumption.

Second, our results need only be valid for a permanent change in the terms of trade. A temporary terms-of-trade shock may lead to some intertemporal substitution in consumption and investment, but is unlikely to cause a permanent change in output.

Third, an economy’s response to an increase in the price of primary commodities clearly hinges on how dependent it is on primary exports. Its investment response also depends on what share of its capital goods is importable.

Fourth, we might expect the availability of external finance to be a crucial factor interacting with changes in the terms of trade, and oil exporters in the 1970s probably enjoyed better access to the world capital markets than is true of the developing world today. On the other hand, this might well change if trends in commodity prices changed.

Finally, it is occasionally argued that oil exporters in the 1970s and 1980s wasted a large portion of their windfall revenues on prestige projects with small economic returns. If so, and if developing countries have by now learned the lesson, they may be able to put their luck to better use next time around.