Journal Issue
Share
Article

Devaluation, Relative Prices, and International Trade Evidence from Developing Countries

Author(s):
International Monetary Fund. Research Dept.
Published Date:
January 1995
Share
  • ShareShare
Show Summary Details

DEVALUATIONS HAVE often been used by developing countries to reduce large external imbalances, correct perceived perceived “overvaluations” of the real exchange rate, increase international competitiveness, and promote export growth. The 50 percent devaluation in early 1994 by the CFA franc zone countries stands out as a recent example of such a policy (see Ostry (1994)).1 However, a devaluation can only accomplish these tasks if, in the first place, it translates into a real devaluation and, second, if trade flows respond to relative prices in a significant and predictable manner.

With regard to the response of the real exchange rate to a nominal devaluation, the empirical literature appears to agree that in most devaluation episodes the real exchange rate does indeed respond significantly to a nominal exchange rate shock, at least in the short run.2 Examining the behavior of the real exchange rate in the aftermath of 29 devaluation episodes. Edwards (1989) finds that, in most instances (the real effects in chronic high‐inflation countries appear to be much less), there are significant real effects one year after the devaluation; the effects, however, appear to erode completely beyond the third year. Kiguel and Ghei (1993) further conclude that, in low‐inflation economies with a tradition of a fixed exchange rate, the real effects of the devaluation may be even longer‐lived than was suggested by the earlier work of Kamin (1988) and Edwards (1989).

The next question is whether trade flows systematically respond to the change in relative prices produced by the devaluation and, if so, what is the order of magnitude of the response. The earlier literature that modeled trade in developing countries (see, for instance, Khan (1974), Rittenberg (1986), Bond (1987), and Marquez and McNeilly (1988)) commonly found evidence that relative prices play a significant role in the determination of trade flows, buttressing policies of devaluation as a way to correct trade imbalances. Their evidence often came in the form of significant t–statistics on the relative price variable in static or “long‐run” specifications of import demand or export supplies and, hence, calls to mind the work on the inference problems inherent with variables that have unit roots.

More recent empirical work (see Rose (1990) and (1991); and Ostry and Rose (1992)), however, has suggested that, once the time‐series properties of the variables are properly taken into account in the estimation, there is little evidence that relative prices have a significant and predictable impact on trade.3 While Rose (1990) does not model imports and exports separately (as is done in some of the earlier literature), using data for 30 developing countries he finds that changes in the real exchange rate do not have a significant effect on changes in the balance of trade.

The latter conclusion would, of course, imply that a devaluation is likely to be ineffective in its “expenditure‐switching” role and, therefore, in achieving its main goals of reducing trade imbalances and stimulating export growth.

In light of such conflicting evidence and policy implications, this paper re‐examines the relationship between relative prices and the imports and exports for a sample of developing countries along the lines of the earlier literature on this subject (see Khan (1974), among others). The analysis, however, is conducted in light of the inference problems that arise when the variables used to estimate behavioral relationships are nonstationary. The paper connects the growing literature on estimating relationships among variables that are nonstationary, including the contributions of Engle and Granger (1987); Johansen (1988) and (1991); and Stock and Watson (1989) to a practical policy problem endemic to developing countries. The cointegration approach to estimating preference parameters employed here is found elsewhere in the recent empirical literature (see, for instance Ogaki (1992) and, for an application very similar to ours, Clarida (1994)). This approach provides reliable estimates of the long‐run price and income elasticities of developing countries’ import demand as well as industrial countries’ demand for developing country exports.

By analyzing whether and to what extent the imports and exports of developing countries respond to relative price changes, conclusions can be drawn as to the effectiveness of the often‐used devaluation policies. In addition, the empirical results presented in this paper can also be used to evaluate the efficacy of various commercial policies aimed at altering the relative price of traded and nontraded goods.4

Section I discusses the theoretical underpinnings of import and export determination in the context of an intertemporal optimizing version of the simple imperfect substitutes model that has dominated this literature (see, for instance, Goldstein and Khan (1985)). In the developing country, utility‐maximizing consumers choose between a nontraded domestic good and an imported good. Similarly, in industrial countries, households choose among the domestically produced good and the export of the developing country. Section II first establishes the time‐series properties of the variables used in the analysis and then applies the cointegration tests of Johansen (1988) and (1991) to determine if the specifications suggested by theory adequately define the steady‐state behavior of imports and exports. With these relationships in hand, Section III presents an estimator (see Stock and Watson (1990)) that is nuisance‐parameter free to tackle the issue of whether or not relative prices affect trade flows in a significant and systematic way and to obtain reliable parameter estimates. Section IV pools into regional blocks the time‐series data of the various countries in order to highlight some of the stylized facts that characterize trade flows among developing and industrial countries and the distinct patterns that prevail among geographical regions. The final section summarizes the key results and reviews some of the policy implications.

I. A Simple Model of Developing Country Foreign Trade

The modeling of foreign trade relationships has a long history, as illustrated by Goldstein and Khan (1985). In effect, there is a remarkable degree of consensus in the profession on the empirical form of the demand for imports and exports. The standard approach to specifying and estimating trade equations, the model most prevalent in the empirical trade literature, is the “imperfect substitutes model.” The central assumption of that model is that neither imports nor exports are perfect substitutes in consumption for domestic nontraded goods. The assumption of imperfect substitutability has found broad empirical support. For instance, Ostry and Reinhart (1992), who estimate the intratemporal elasticity of substitution between traded and nontraded goods for a broad panel of developing countries, find that this parameter is in the 1.0–1.5 range in all the regions considered in the study, implying gross substitutability.5 Similar results were found for individual developing countries (see Ogaki, Ostry, and Reinhart (1995)). This section describes such an economy.

The simple continuous‐time model of a representative utility‐maximizing household described below is meant to be illustrative, as it yields representations of the demand for imports that are quite common in the trade literature (see, for example, Khan and Ostry (1992)). The model outlined below describes a small open exchange economy populated with identical agents that possess perfect foresight. These agents have inherited an outstanding stock of internationally traded debt; since there is perfect capital mobility, the residents in this economy take the world interest rate as given. As the framework outlined below does not include monetary considerations (i.e., it is a real model), it takes as given, as previously discussed, that a nominal devaluation is capable of altering the relative price of traded and nontraded goods. Hence, the households are faced with a “real” shock.

Developing Countries’ Import Demand

The representative infinitely‐lived household consumes a nontraded and an imported good, denoted as ht and mtrespectively; they have endowments of the home good, qt and the export good, xt which is not domestically consumed.6 Thus, the total endowment expressed in terms of the home good is defined as yt=qt+xt(px/p)t. This endowment, in turn, is used to service the interest payment on the fixed stock of outstanding debt, which is denominated in terms of the export, with the remaining income used to finance consumption or accumulate the asset.7 The representative consumer maximizes the lifetime utility function

where β > 0 is the subjective rate of time preference. We assume for simplicity that the household derives utility in each period according to a Cobb‐Douglas utility function.8

Equation (2) defines the household’s flow budget constraint. Taking the home good as the numeraire, (Px/P)t defines the price of the export relative to the home good, and (Pm/P)t is the relative price of imports. The world interest rate is rt* and A, represents the stock of debt.9

Combining u(.) with the budget constraint, and introducing the costate variable, μt leads to a Hamiltonian of the form

The first order conditions yield the familiar relationships between con‐ sumption of the home and imported goods that hold at each point in time:

Equation (4) equates the intratemporal marginal rate of substitution between importables and nontradables to the relevant relative price. Dynamics place consumption of the importable along the optimal path given by the Euler equation:

Equation (5) is analogous to equation (4), as it relates the marginal rate of substitution between current and future consumption to the relevant intertemporal price, the world real interest rate.

Industrial Countries’ Demand for Developing Country Exports

The optimization problem faced by consumers in the industrial countries directly parallels the foregoing simple framework. The representative infinitely‐lived household consumes a nontraded and an imported good, denoted as ht and xt respectively. The imported good is the export of the developing countries. There are endowments of the home good qt*, and the export good, mt (imported by developing countries), which is not domestically consumed. Households in the industrial countries are assumed to be net lenders, who receive interest income; they can consume or accumulate the asset. The representative consumer problem and solution are summarized by equations (6)—(9),

where the preference parameters are assumed to be the same as those of households in developing countries.10 As before, the home good servesas the numeraire, (Px/P*)t defines the price of the imported good (which is exported by the developing country) to the home good. The total endowment in terms of the home good is defined as yt*=qt*+mt(Pm/P*)t. The first order conditions yield relationships between consumption of the home and imported goods that hold at each point in time:

while dynamics are given by the Euler equation:

The Steady State

The dynamics of imports in developing and developed countries are given by the Euler equations (equations (5) and (9), respectively). However, our primary interest in the analysis that follows is to employ cointegration analysis to examine the “long‐run” steady‐state relationships that describe import demand.11 Market clearing conditions for the home goods markets (ht=qtandht*=qt*) determine the relative prices of the nontraded goods. A steady‐state solution requires that the subjective rate of time preference equal the world rate of interest (β=r*); the latter ensures that there is no saving (dissaving) in the steady state. We solve the budget constraints (equations (2) and (9)) to obtain an expression that links imports to their price relative to the home good and to permanent income, which in our nonstochastic framework is defined as the endowment of the exportable plus or minus interest incomes. Its log‐linear versions for developing and developed countries are given by equations (10) and (11):

This is a nonstochastic version of the “long‐run” relationship that describes the behavior of imports, and what will be termed a cointegrating equation that is estimated in the following section. The model can be made to accommodate a stochastic element by either reconsidering the maximization problem under uncertainty, or by assuming that some or all the underlying concepts are measured with error.

Thus, a simple theoretical framework provides a number of testable implications. First, it suggests that a scale variable (such as permanent income or wealth) and relative prices are both necessary and sufficient to define the long‐run behavior of imports. This would argue against the inclusion of any other variables in an ad hoc manner. Second, it assigns a predictable and well‐defined role for relative prices in affecting trade flows. Third, the simple Cobb‐Douglas utility function employed predicts that the income and price elasticities are one and minus one, respectively. The sections that follow will test all these hypotheses for a variety of developing countries and for an aggregate “industrial country” bloc.

II. Empirical Analysis

The structural model outlined in the previous section links the steady‐ state consumption of the imported good to real permanent income and relative import prices. In this simple two‐good setting, the relevant deflator for import prices is the price of nontraded goods. However, because of data limitations, the analysis that follows uses imports as a proxy for consumption of importables, and consumer prices and real GDP to proxy the price of nontraded goods and permanent income, respectively. The extent to which these variables imperfectly proxy the underlying concepts introduces a measurement error that is likely to vary across countries and across time. The only assumption that underlies the estimation is that such measurement errors are stationary processes with well‐defined variances. Industrial countries’ consumption of developing countries’ exports similarly depends on permanent income and the relative price of the exportable. The data used are annual and cover the period 1968–92. Details of the data and sources are presented in the Appendix; the countries included in the sample are listed in the tables that follow.

Time–Series Preliminaries

We establish the time‐series properties of the relevant variables through the standard unit root tests: the Dickey‐Fuller (DF) and Augmented Dickey‐Fuller (ADF) tests (Dickey and Fuller (1981)). These tests deal with a variable, zt that admits a simple autoregressive representation,

where et is a random error term drawn from an unchanging and independent distribution. The two tests examine the null hypothesis that the series zt is I(1), that is, if the absolute value of p equals or exceeds 1, versus the alternative hypothesis that the absolute value of ρ<1.. Because shocks do not die out, a nonstationary series has no well‐defined asymptotic variance. It is simple to show that the unconditional variance of z depends on the variance of e and the coefficient p, as in σz2=σe2/(1ρ). So, if ρ=1, the variance of z is unbounded.

Unless otherwise noted, it was found that the null hypothesis of a unit root could not be rejected for the level of the variable but was rejected for the first difference of the variable. In other words, the results suggest (subject to the usual caveats about the low power of the unit root tests) that the variables in question are I(1) processes. In most instances, real imports, real GDP, the ratio of real exports to real GDP, and relative prices are integrated of the same order.

The next task is to determine across variables if these shocks coincide in a way predicted by economic theory. For instance, is a permanent increase in income associated with a permanent increase in imports, as predicted by the simple theoretical model? The strategy is to determine if one or more linear combinations of these variables is drawn from a stationary distribution. If that holds, then the individually integrated variables are said to be cointegrated. If so, then it can be concluded that these variables define steady‐state trade relationships and our simple theoretical model finds support in the data.

Cointegration

In much of the earlier literature, estimates of the preference parameters (i.e., the price and income elasticities) were frequently obtained by applying ordinary least squares (OLS) to a specification that was often very similar to the import equation, (10), and to the export equation, (11). These specifications of the demand for imports and exports usually yielded parameters in accordance with the model’s priors, the scale variable entering positively while relative prices entered negatively. Most often, the estimates were statistically significant. However, as Granger and Newbold (1974) first showed, two nonstationary variables may appear to have a relationship only because they have similar time‐series properties. Indeed, this could be the case here since, as discussed, all the variables of interest are nonstationary. What we set out to do in the next two subsections is to ensure that inference regarding relative prices is not clouded by a spurious element.

A large body of econometric literature (see Banerjee and others (1986)) tells us that even if cointegration obtains, inference problems still remain in OLS estimation. OLS provides consistent, but inefficient, estimates of the true parameters; biases arising from serially correlated errors and simultaneity problems are present and could be quite large for samples as small as those considered here (see Campbell and Perron (1991)). Under such circumstances, standard errors and t–statistics do not provide an adequate measure of statistical significance.

Lack of cointegration is even more problematic, since the OLS estimates are no longer consistent. Further, failure to obtain cointegration may reflect a fundamental misspecification in the model, possibly arising from the omission of one or more variables. Again, no valid inferences can be drawn. In either case, such problems raise questions about the findings of the earlier literature.

To re‐examine the role of relative prices in light of these developments and assess if the implied theoretical model is capable of describing the data, we proceed in two steps. First, we test for cointegration; this tells us whether the long‐run behavior of import demand is adequately specified. Second, in the next subsection we employ an estimator that is free from nuisance parameters and, hence, provides reliable estimates of the price and income elasticities and allows us to test whether relative prices significantly affect trade. 12

The cointegration test most commonly employed in the literature is that suggested by Engle and Granger (1987). However, a more powerful test that allows for the detection and estimation of the number of cointegrating vectors was developed by Johansen (1988) and (1991) in the context of a vector autoregression model (VAR),13 This is the test employed here.

In the Johansen (1988) and (1991) procedure. maximum likelihood is applied to an autoregressive representation of the form given by equation (12).

where F(L) is a 3 x 3 matrix of polynomials in the lag operator, which shifts a series back in time, that is, Lyt=yt1.

The intuition is as follows: a stationary variable, such as Δmt, cannot depend on a set of variables that are individually blowing up (such as mt1yt1, etc.). Statistically, this implies that the coefficients on the lagged variables appearing on the right‐hand side of equation (12) should be insignificantly different from zero unless that set of variables is cointegrated.14 That is to say, a linear combination of these I(1) variables produces a stationary process. The lagged first differences of the dependent variables included in the right‐hand side ensure that any serial correlation in the residuals is corrected.

These tests thus focus on the properties of the matrix of coefficients, Π In the absence of cointegration Π is a singular matrix (its rank, r = 0). Hence, in our case, the rank of Π could be anywhere between zero, if no cointegrating vector exists, and three, the number of variables in the system. The λ‐Max tests the null hypothesis of rcointegrating vectors versus the alternative hypothesis of r +1 cointegrating vectors.15 If the largest eigenvalue of Π (λ–Max) exceeds the critical value tabulated under the null hypothesis, we can reject the null hypothesis in favor of the alternative. The trace test has the same null hypothesis as the λ‐Max test; however, the alternative hypothesis is the rank of Π is n–r, where n represents the number of variables in the system. If the trace of Π exceeds the critical value the null hypothesis is rejected.

Table 1 and 2 present the results of these two tests and their attendant critical values. The null hypothesis tested is that there is no cointegrating vectorr=0. 16 In the case of developing countries’ import demand (Table 1), we can reject the hypothesis of no cointegration (using either or both tests) for 10 of the 12 countries in our sample. The results for industrial countries’ demand for developing countries’ exports (summarized in Table 2) are somewhat less conclusive. The rank tests detect cointegration in 8 of the 12 countries. No cointegrating vector was found for Brazil, Congo, Costa Rica, and Indonesia. In the case of industrial country demand for developing country exports, the lower incidence of cointegration may simply reflect the fact that for some developing countries the demand for their exports is increasingly coming from other developing countries (see Muscatelli, Stevenson, and Montagna (1994) and Milesi‐Ferretti (1994)). This would imply equation (11) is misspecified. In most instances, however, the simple relationships suggested by the theoretical framework seem to find fairly broad support in the data.

Table 1.Testing for Cointegration: Developing Countries’ Import Demand, 1970‐92
Maximum likelihood rank tests

(null hypothesis r = 0)
Countryk‐MaxTrace
Africa
Congo30.13441.063
Kenya23.16335.527
Morocco27.59238.553
Asia
Hong Kong29.73240.552
Indonesia20.17829.354
Pakistan50.41458.986
Sri Lanka41.95662.842
Latin America
Argentina25.13237.457
Brazil24.03635.678
Colombia35.38947.920
Costa Rica22.21434.787
Mexico24.78338.985
Critical values for p – r = 3
90 percent22.3236.13
95 percent24.8439.41
Notes: We test for no cointegrating vector: hence, p, the number of variables, is 3, and r, the number of cointegrating vectors, is 0. The critical values are taken from Osterwald‐Lenum (1992) for Case 1* and adjusted using the small sample correction factor suggested by Cheung and Lai (1993).
Notes: We test for no cointegrating vector: hence, p, the number of variables, is 3, and r, the number of cointegrating vectors, is 0. The critical values are taken from Osterwald‐Lenum (1992) for Case 1* and adjusted using the small sample correction factor suggested by Cheung and Lai (1993).

III. The Role of Relative Prices: Empirical Evidence

Given the prominence of devaluation in adjustment programs, and particularly since a recent strand in the empirical trade literature has called into question whether relative prices have any effect on trade balances (for instance, see Rose (1990)), our next goal is to assess whether and, if so, to what extent trade flows respond to relative prices. Hence, this section focuses on obtaining estimates of the price and income elasticities and tests hypotheses about these parameters.

Table 2.Testing for Cointegration: Industrial Countries’ Demand for Developing Countries’ Exports, 1970–92
Maximum likelihood rank tests

(null hypothesis r = 0)
Countryλ‐MaxTrace
Africa
Congo22.14334.273
Kenya25.98136.024
Morocco25.13337.641
Asia
Hong Kong24.98536.434
Indonesia19.12527.541
Pakistan23.44536.974
Sri Lanka25.96538.397
Latin America
Argentina27.32538.916
Brazil17.74323.981
Colombia29.32939.912
Costa Rica18.63125.983
Mexico29.18041.821
Critical values for p – r= 3
90 percent22.3236.13
95 percent24.8439.41
Notes: We test for no cointegrating vector; hence, p, the number of variables, is 3, and r, the number of cointegrating vectors, is 0. The critical values are taken from Osterwald‐Lenum (1992) for Case 1* and adjusted using the small sample correction factor suggested by Cheung and Lai (1993).
Notes: We test for no cointegrating vector; hence, p, the number of variables, is 3, and r, the number of cointegrating vectors, is 0. The critical values are taken from Osterwald‐Lenum (1992) for Case 1* and adjusted using the small sample correction factor suggested by Cheung and Lai (1993).

To that end, we adopt Stock and Watson’s (1989) specification, which deals with the biases introduced in the cointegrating regressions by simultaneity and serial correlation in the errors. By eliminating these nuisance parameters, we can obtain reliable estimates of the long‐run relationship among these variables. The nonlinear specification reproduced below in equation (13) was estimated for imports and exports, respectively, and the results are summarized in Tables 3 and 4:

In 11 of the 12 countries, relative import prices were significant with the anticipated sign; the exception was Morocco. The price elasticities range from –0.156 to –1.363. Estimates of industrial countries’ demand for developing country exports show that prices are significant in seven of the nine countries where cointegration obtains (Kenya and Mexico are exceptions). To examine the robustness of these results, we also tested the significance of relative prices in the context of a VAR, Johansen framework. We compared the unrestricted system given by equation (12) with the restricted version reproduced below in equation (14).

Table 3.Stock and Watson Estimates of Developing Countries’ import Demand. 1970–91
CountryConstantPm/PyR2
Africa
Congo–12.42–0.1561.3590.859
(1.675)(0.025)(0.143)
Kenya1960–0.6500.0950.675
(0.809)(0.340)(0.391)
Morocco–4.7160.2751.2040.940
(2.368)(0.279)(0.650)
Asia
Hong Kong–1.247–1.2801.4020.985
(0.623)(0.362)(0.049)
Indonesiaa–9.704–0.9271.6200.950
(1.036)(0.170)(0.106)
Pakistan–4.046–0.3981.1500.941
(0.418)(0.147)(0.083)
Sri Lanka–6.668–0.3041.9760.852
(0.793)(0.158)(0.249)
 Latin America
Argentina–1.377–0.4671.0920.404
(1.380)(0.147)(0.583)
Brazil13.791–0.5532.7590.850
(1.364)(0.147)(0.320)
Colombia1.184–1.3631.1380.901
(2.520)(0.537)(0.121)
Costa Ricaa–0.381–0.7470.9750.519
(0.279)(0.263)(0.257)
Mexico–3.360–0.3930.8930.884
(3.128)(0.143)(0.388)
Notes: Standard errors are in parentheses. Description of the data and their sources are in Table Al.

These results are reported but are not reliable, as cointegration did not obtain.

Notes: Standard errors are in parentheses. Description of the data and their sources are in Table Al.

These results are reported but are not reliable, as cointegration did not obtain.

Table 4.Stock and Watson Estimates of Industrial Countries’ Demand for Developing Country Exports, 1970–91
CountryConstantPx/P*y*R2
Africa
Congoa–13.189–0.3202.0560.909
(1.062)(0.088)(0.209)
Kenya–5.8680.1881.3520.503
(8.301)(0.179)(1.793)
Morocco–8.963–0.3571.8520.957
(0.767)(0.103)(0.164)
Asia
Hong Kong–19.360–0.5444.4100.994
(0.960)(0.165)(0.222)
Indonesiaa–7.201–0.0152.0220.944
(0.652)(0.052)(0.129)
Pakistan–7.172–0.9701.4540.935
(0.897)(0.244)(0.209)
Sri Lanka–4.937–0.6070.8899.971
(0.369)(0.057)(0.079)
Latin America
Argentina–5.280–0.4151.3590.901
(1.052)(0.099)(0.222)
Brazila–9.527–0.1482.4470.940
(1.011)(0.157)(0.221)
Colombia–7.254–0.5221.6260.914
(0.993)(0.111)(0.210)
Costa Ricaa–5.189–0.4861.0710.936
(0.357)(0.136)(0.078)
Mexico–13.7040.3123.3790.949
(1.036)(0.173)(0.206)
Notes: Standard errors are in parentheses. Description of the data and their sources are in Table Al.
aThese results are reported but are not reliable, as cointegration did not obtain.
Notes: Standard errors are in parentheses. Description of the data and their sources are in Table Al.
aThese results are reported but are not reliable, as cointegration did not obtain.

These results are presented in Table 5. In the case of imports, the x2 tests comparing the restricted system (true under the null hypothesis) with the unrestricted system, which includes relative prices, also indicate that for imports the restriction excluding prices was rejected by the data in 8 of the 10 countries where cointegration obtained. For exports, the restricted system was rejected in 7 of the 8 countries where cointegration obtained. The cumulative evidence from the test results for these countries appears to indicate that relative prices play a significant role in the determination of imports, exports, or both.

Hence, in the majority of cases, income and relative prices are sufficient to define a steady state, that is, these variables are cointegrated with imports in a way predicted by theory. Second, when the model is compared with a restricted model that excludes relative prices, the data reject this restriction in the majority of cases. Not surprisingly, where the model fares the worst is in its predictions of the income and price elasticities; the joint hypothesis of (1, –1) income and price elasticities was rejected in the overwhelming majority of cases.17 This result could be due be due to the presence of measurement error in the scale and relative price variables, as these imperfectly proxy the underlying concepts. Just as likely, however, the data could be rejecting the Cobb‐Douglas specification in favor of a more general specification. For instance, in the case of a CES utility function, the relative price elasticity will depend on the intratemporal elasticity of substitution, and is not restricted to equal minus unity (see Ostry and Reinhart (1992) who find evidence favoring a CES specification).

Table 5.Can Relative Prices Be Excluded? 1970–92
CountryX2(1)

Imports
X2(1)

Exports
Africa
Congo8.57012.620
(0.00)(0.00)
Kenya5.9104.340
(0.02)(0.04)
Morocco1.5601.840
(0.21)(0.08)
Asia
Hong Kong2.9900.230
(0.08)(0.64)
Indonesia10.1401.520
(0.00)(0.22)
Pakistan36.5902.720
(0.00)(0.10)
Sri Lanka2.22012.220
(0.04)(0.00)
Latin America
Argentina0.50014.340
(0.48)(0.00)
Brazil8.1403.290
(0.00)(0.07)
Colombia20.1502.230
(0.00)(0.04)
Costa Rica8.1203.360
(0.00)(0.07)
Mexico6.70012.610
(0.01)(0.00)
Note: Probability values are in parentheses.
Note: Probability values are in parentheses.

In general, the countries in the sample appear to meet the static Marshall‐Lerner condition for stability, as changes in relative prices do produce long‐run reallocation of trade flows. However, Backus, Kehoe, and Kydland (1992) show in the context of an intertemporal model of international trade that the sign of the relation between the terms of trade and the trade balance will depend on the elasticity of substitution between the imported and home goods rather than in the fulfillment of the static Marshall‐Lerner condition. In these models what remains essential is that consumption respond to price changes, a condition for which we find ample empirical evidence.

IV. Regional and Aggregate Evidence

For most of the countries in the sample, only annual data for the variables of interest are available. As such, sample sizes are limited to 25 observations or fewer. The usual small sample handicaps can, however, be circumvented by pooling together countries within a geographical region. Grouping together countries within each region not only increases the efficiency of the estimates of the parameters of interest, but also helps highlight the broader stylized facts that may be obscured in the country‐specific analysis. The following section explores these regional trade patterns in greater detail.

The model sketched in Section I assumed, for the sake of simplicity, that preferences were identical across countries, industrial and developing alike. Yet country‐specific parameter estimates, as Tables 3 and 4 attest, tend to vary over fairly broad ranges. However, the individual country estimates make it difficult to discern if differences in responses to prices and income follow any broader regional pattern, or if the assumption that developing and industrial countries are alike is met by the data.

To attempt to address these issues we pooled countries within regions and across regions. Several caveats are in order. First, the individual country cointegration tests revealed that, in some of the countries (albeit a minority) that are pooled to make up the regional estimates, no co‐integration was established. To date, little is known about unit root testing and cointegration tests with panel data. The recent work of Levin and Lim (1993) suggests that there are gains from pooling, as the power of these tests increases. With these caveats in mind, we present the panel estimates in Table 6. These estimates were obtained using the fixed effects estimator, which allows the intercept to vary across countries while imposing the restriction that the slope coefficients are the same and by correcting, along the lines of Stock and Watson (1989), for any potential simultaneity and serial correlation that may be present.

Table 6.Regional and Aggregate Estimates: Fixed Effects Specification, 1970–91
CountryRelative

price
yR2
Developing countries’ import demand
Latin America–0.3570.9640.569
(0.070)(0.079)
Asia–0.4031.3860.904
(0.073)(0.045)
Africa–1.3631.1380.901
(0.537)(0.121)
All Countries–0.5311.2190.737
(0.052)(0.041)
Industrial countries’ demand for developing country exports
Latin America–0.1922.0690.780
(0.084)(0.115)
Asia–0.3982.4940.777
(0.090)(0.140)
Africa–0.2661.2530.472
(0.099)(0.171)
All Countries–0.3242.0520.718
(0.053)(0.078)
Notes: Standard errors are in parentheses. The Stock and Watson (1989) estimator was used.
Notes: Standard errors are in parentheses. The Stock and Watson (1989) estimator was used.

When all developing countries are pooled, and similarly, when the demands for developing country exports are grouped into a single panel. the Houthakker and Magee (1969) results re‐emerge. The income elasticity of industrial countries’ demand for imports is 2.05, compared with an income elasticity of 1.22 for developing countries’ demand for imports.18 Hence, if developed and developing countries grow at the same rate, industrial countries’ trade balance would deteriorate over time. However, this “behavioral” discrepancy, which tends to favor developing countries, does not apply uniformly to all regions. While the HouthakkerMagee result characterizes Asian and Latin American trade patterns quite well, it does not apply to the case of Africa. Industrial countries’ demand for African exports has an income elasticity of 1.25, about one half of what it is for Asia and well below the 2.07 for Latin America. Indeed, it is not significantly different from the income elasticity of African import demand. As discussed in Baban and Greene (1992) and Reinhart and Wickham (1994), the high primary commodity content of African exports is a probable explanation for this result.19 The income elasticity estimates for industrial countries closely resemble those obtained by Clarida (1994), who employs a similar estimation strategy for the United States, and are somewhat lower than those found by Marquez (1989).20

Not surprisingly, the panel estimates confirm what the country‐specific results showed, namely, that relative prices play a significant role in affecting trade flows. Both industrial and developing countries’ demand for imports (irrespective of the region considered) respond to relative prices as predicted by theory. However, with the exception of African import demand, relative price elasticities are well below unity, suggesting that large relative price swings are necessary to produce an appreciable reallocation of trade flows. These estimates are smaller (in absolute value) than Clarida (1994) estimates for the United States and about in line with the Marquez (1989) estimates.

V. Conclusions

An older empirical literature on trade commonly found evidence that relative prices play a significant role in the determination of trade flows. These results, in turn, lent support to policies of devaluation as a means of correcting trade imbalances and promoting export growth. However, some of the recent studies that have taken into account the time‐series properties of these variables have arrived at a very different conclusion, namely, that no systematic relationship between trade balances and relative prices is discernible from the data.

This paper has re‐examined the role of relative prices in affecting trade and therefore, implicitly, the effectiveness of devaluation policies in light of the recent time‐series literature that deals with variables that have unit roots and no well‐defined limiting distributions. Several empirical regularities emerge. First, the analysis suggests that, in accordance with standard microeconomic theory, income and relative prices are, more often than not, both necessary and sufficient to pin down steady‐state trade flows. However, the “traditional” specification appears to fare better when modeling developing‐country demand for imports than when applied to industrial‐country demand for developing‐country exports. The latter may suggest that a fruitful area to investigate is intradeveloping‐country trade. Second, it is found that, for the majority of cases, relative prices are a significant determinant of the demand for imports and exports. Third, while relative prices have a predictable and systematic impact on trade, price elasticities tend to be low, in most instances well below unity. The latter suggests that large relative price swings are required to have an appreciable impact on trade patterns. Finally, while industrial‐country income elasticities are well above their developing‐country Asian and Latin American counterparts, suggesting that in a scenario of balanced growth the developing country trade balance should improve, this is not the case for Africa. The high primary commodity content of African exports probably accounts for this result.

APPENDIX Description and Sources of Data

All data are annual and cover the period 1968–92. The source is the IMF’s World Economic Outlook.

Variable definitions:

mt = Nominal imports deflated by import unit values.

yt = Real gross domestic product (in the domestic currency).

(Pm/P)t = Import unit values (converted to domestic currency) deflated by consumer prices.

xt= Nominal exports deflated by export unit values.

yt*=Real gross domestic product of industrial countries (in U.S. dollars).

(Px/P*)t = Export unit values deflated by industrial countries’ consumer prices (in U.S. dollars).

REFERENCES

    BabanRoy C. and Joshua E.Greene“The Export Performance of Sub‐Saharan Africa. 1970–90: A Survey,”IMF Working Paper 92/55 (Washington: International Monetary FundJuly1992).

    • Search Google Scholar
    • Export Citation

    BackusDavid K.Patrick J.Kehoe and Finn E.Kydland“Relative Price Movements in Dynamic General Equilibrium Models of International Trade,”NBER Working Paper No. 4243 (Cambridge, Massachusetts: National Bureau of Economic Research. December1992).

    • Search Google Scholar
    • Export Citation

    BanerjeeAnindyaJuan J.DoladoDavid F.Hendry and Gregor W.Smith“Exploring Equilibrium Relationships in Econometrics through Static Models: Some Monte Carlo Evidence,”Oxford Bulletin of Economics and StatisticsVol. 48 (August1986) pp. 25377.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BondMarian E.“An Econometric Study of Primary Commodity Exports from Developing Country Regions to the World,”Staff Papers. International Monetary FundVol. 34 (June1987) pp. 191227.

    • Crossref
    • Search Google Scholar
    • Export Citation

    CampbellJohn Y. and PierrePerron“Pitfalls and Opportunities: What Macroeconomists Should Know About Unit Roots.” “NBER Macroeconomics Annual1991”ed byOlivier JeanBlanchard and StanleyFischer (Cambridge: MIT Press1991) pp. 141201.

    • Crossref
    • Search Google Scholar
    • Export Citation

    CeglowskiJanet“Intertemporal Substitution in Import Demand,”Journal of International Money and FinanceVol. 10 (March1991) pp. 11830.

    • Crossref
    • Search Google Scholar
    • Export Citation

    CheungYin‐Wong and Kon S.Lai“Finite‐Sample Sizes of Johansen’s Likelihood Ratio Tests for Cointegration,”Oxford Bulletin of Economics and StatisticsVol. 55 (August1993) pp. 31328.

    • Crossref
    • Search Google Scholar
    • Export Citation

    ClaridaRichard H.“Co–integration, Aggregate Consumption, and the Demand for Imports: A Structural Econometric Investigation,”American Economic ReviewVol. 84 (March1994) pp. 298308.

    • Search Google Scholar
    • Export Citation

    DickeyDavid A. and WayneA. Fuller“Likelihood Ratio Statistics for Auto‐ regressive Time Series with a Unit Root,”EconometricaVol. 49 (July1981) pp. 105772.

    • Crossref
    • Search Google Scholar
    • Export Citation

    EdwardsSebastian“Devaluations Real Exchange Rates and the External Sector” in his Rea/ Exchange Rates Devaluation and Adjustment: Exchange Rate Policy in Developing Countries (Cambridge: MIT Press1989) pp. 254309.

    • Search Google Scholar
    • Export Citation

    EngleRobert F. and C.W.J.Granger“Co‐integration and Error Correction: Representation, Estimation and Testing,”EconometricaVol. 55 (March1987) pp. 25176.

    • Crossref
    • Search Google Scholar
    • Export Citation

    GoldsteinMorris and Mohsin S.Khan“Income and Price Effects in Foreign Trade” in Handbook of International Economics Vol. IIed byRonald W.Jones and Peter B.Kenen (Amsterdam: North‐Holland. 1985) pp. 1041105.

    • Search Google Scholar
    • Export Citation

    GrangerC.W.J. and P.Newbold“Spurious Regressions in Econometrics,”Journal of EconometricsVol. 2 (July1974) pp. 11120.

    HoffmaisterAlexander“The Cost of Export Subsidies: Evidence from Costa Rica,”Staff PapersVol. 39 (March1992) pp. 14874.

    HouthakkerH. S. and Stephen P.Magee“Income and Price Elasticities in World Trade,”Review of Economics and StatisticsVol. 51 (May1969) pp. 11125.

    • Crossref
    • Search Google Scholar
    • Export Citation

    JohansenSoren“Statistical Analysis of Cointegration Vectors,”Journal of Economic Dynamics and Control. Vol. 12 (June‐September1988) pp. 23154.

    • Crossref
    • Search Google Scholar
    • Export Citation

    JohansenSoren“Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,”EconometricaVol. 59 (November1991) pp. 155180.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KaminSteven B.“Devaluation External Balance and Macroeconomic Performance: A Look at the Numbers” Princeton University Studies in International Finance No. 62 (Princeton, New Jersey: Princeton University, Department of Economics, International Finance Section. 1988).

    • Search Google Scholar
    • Export Citation

    KiguelMiguel A. and N.Ghei“Devaluation in Low‐Inflation Economies.”World Bank Policy Research Working Paper 1224 (Washington: World BankNovember1993).

    • Search Google Scholar
    • Export Citation

    KhanMohsin S.“Import and Export Demand in Developing Countries,”Staff PapersInternational Monetary FundVol. 21 (November1974) pp. 67893.

    KhanMohsin S. and Jonathan D.Ostry“Response of the Equilibrium Real Exchange Rate to Real Disturbances in Developing Countries,”World DevelopmentVol. 20 (September1992) pp. 132534.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KremersJereon J. M.Neil REricsson and Juan J.Dolado“The Power of Cointegration Tests,”Oxford Bulletin of Economics and StatisticsVol. 54 (August1992) pp. 32548.

    • Crossref
    • Search Google Scholar
    • Export Citation

    LevinAndrew and ChienFu Lin“Unit Root Tests in Panel Data: Asymptotic and Finite Sample Properties” IMF Seminar Series (International) No. 1993–44 (Washington: International Monetary FundSeptember1993).

    • Search Google Scholar
    • Export Citation

    MarquezJaime“Cyclical and Secular Trade Elasticities: An Application to LDC Exports,”Journal of Economic Dynamics and ControlVol. 12 (March1988) pp. 7176.

    • Crossref
    • Search Google Scholar
    • Export Citation

    MarquezJaime“Income and Price Elasticities of Foreign Trade Flows: Econometric Estimation and Analysis of the U.S. Trade Deficit” in Economics in Theory and Practice: An Eclectic Approached. byLawrence R.Klein and JaimeMarquez (Dordrecht. The Netherlands: Kluwer Academic Publishers1989) pp. 12676.

    • Search Google Scholar
    • Export Citation

    MarquezJaime and CarylMcNeilly“Income and Price Elasticities for Exports of Developing Countries,”Review of Economics and StatisticsVol. 70 (May1988) pp. 30614.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Milesi–FerrettiGian Maria“The Determinants of Korean Imports and Exports: An Empirical Investigation.” (unpublished; Washington: International Monetary Fund1994).

    • Search Google Scholar
    • Export Citation

    Muscatelli V. A.A. A. Stevenson and C. Montagna“Intra‐NIE Competition in Exports of Manufactures,”Journal of International EconomicsVol. 37 (August1994) pp. 2947.

    • Crossref
    • Search Google Scholar
    • Export Citation

    MussaMichael“Nominal Exchange Rate Regimes and the Behavior of Real Exchange Rates: Evidence and Implications,”Carnegie–Rochester Series on Public PolicyVol. 25(Autumn1986) pp. 117214.

    • Crossref
    • Search Google Scholar
    • Export Citation

    OgakiMasao“Engel’s Law and Cointegration,”Journal of Political EconomyVol. 100 (October1992). pp. 102746.

    OgakiMasaoJonathan D.Ostry and Carmen M.Reinhart“Saving Behavior in Low‐and Middle‐Income Developing Countries: A Comparison.”IMF Working Paper 95/3 (Washington: International Monetary FundJanuary1995); forthcoming in Stuff Papers.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Osterwalcl‐LenumMichael“A Note with Quantiles of the Asymptotic Distribution of the Maximum Likelihood Cointegration Rank Test Statistics,”Oxford Bulletin of Economics and StatisticsVol. 54 (August1992) pp. 46171.

    • Crossref
    • Search Google Scholar
    • Export Citation

    OstryJonathan D.“The Balance of Trade, Terms of Trade, and Real Exchange Rate: An Intertemporal Optimizing Framework.”Staff PapersInternational Monetary FundVol. 35 (December1988) pp. 54173.

    • Crossref
    • Search Google Scholar
    • Export Citation

    OstryJonathan D.“Exchange Rate Misalignment Among CFA Franc Countries,” (unpublished; Washington: International Monetary Fund1994).

    • Search Google Scholar
    • Export Citation

    OstryJonathan D. and Carmen M.Reinhart“Private Saving and Terms of Trade Shocks:Evidence from Developing Countries,”Staff Papers International Monetary FundVol. 39 (September1992) pp. 495517.

    • Crossref
    • Search Google Scholar
    • Export Citation

    OstryJonathan D. and Andrew K.Rose“An Empirical Evaluation of the Macroeconomic Effects of Tariffs,”Journal of International Money and FinanceVol. 11 (February1992) pp. 6379.

    • Crossref
    • Search Google Scholar
    • Export Citation

    ReinhartCarmen M.“Fiscal Policy, the Real Exchange Rate, and Commodity Prices,”Staff PapersInternational Monetary FundVol. 38 (September1991) pp. 50624.

    • Crossref
    • Search Google Scholar
    • Export Citation

    ReinhartCarmen M. and PeterWickham“Commodity Prices: Cyclical Weakness or Secular Decline?”Staff PapersInternational Monetary FundVol. 41 (June1994) pp. 175213.

    • Crossref
    • Search Google Scholar
    • Export Citation

    RittenbergLibby“Export Growth Performance of Less‐Developed Countries.”Journal of Development EconomicsVol. 24 (November1986) pp. 16777.

    • Crossref
    • Search Google Scholar
    • Export Citation

    RoseAndrew K.“Exchange Rates and the Trade Balance: Some Evidence from Developing Countries.”Economic LettersVol. 34 (November1990) pp. 27175.

    • Crossref
    • Search Google Scholar
    • Export Citation

    RoseAndrew K.“The Role of Exchange Rates in Popular Models of International Trade: Does the Marshall‐Lerner Condition Hold?”Journal of International EconomicsVol. 30 (May1991) pp. 30116.

    • Crossref
    • Search Google Scholar
    • Export Citation

    StockJames H. and Mark W.Watson“A Simple MLE of Cointegrating Vectors in Higher Order Integrated Systems,”NBER Technical Working Paper No. 83 (Cambridge, Massachusetts: National Bureau of Economic ResearchDecember1989).

    • Search Google Scholar
    • Export Citation

    World BankGlobal Economic Prospects (Washington: World Bank. 1994).

*Carmen M. Reinhart is an Economist in the Research Department. She holds a Ph.D. from Columbia University. The author would like to thank José De Gregorio, Karl Driessen. Alex Hoffmaister, Mohsin Khan, Leonardo Leiderman. Peter Montiel, Jonathan Ostry, Vincent Reinhart, Anthony Richards. and Peter Wickham for helpful comments and suggestions.
1The CFA franc zone countries are: Benin. Burkina Faso, Cameroon, Central African Republic, Chad, Congo, COte divoire, Equatorial Guinea, Gabon, Mali, Niger, Senegal, and Togo. The Comorian Franc was also devalued at that time by 33 percent.
2This is also the case for industrial countries; see Mussa (1986).
3In particular, Rose (1990) finds that there is little evidence of a systematic relationship between changes in the terms of trade and changes in the current account for various developing countries. Changes in the internal terms of trade induced by commercial policies are examined in Ostry and Rose (1992), who find negligible effects on the trade balance. for both developed and developing countries.
5Gross complementarily would imply an intratemporal elasticity of substitution below unity, while perfect substitutability would imply an infinite intratemporal elasticity of substitution. In addition, if imports and exports were perfect substitutes, one would not observe two‐way trade, at least in a static framework.
6The assumption that the exportable is not consumed is fairly standard in the trade literature; for a similar framework see Ostry (1988) and Ostry and Reinhart (1992).
7None of the results hinge on the household being a net debtor. As the next subsection shows, the analysis carries over to the case in which the household is a net lender. The assumption, which follows Reinhart (1991), simply facilitates the link to the developing countries in our sample. most of which are net international debtors.
8The empirical evidence, however, tends to favor the more general constant elasticity of substitution (CES) utility function (see Ostry and Reinhart (1992) and Ogaki, Ostry, and Reinhart (1995)), although for many countries the intratemporal elasticity of substitution is not significantly different from unity.
9Clarida (1994) and Ceglowski (1991) employ an addilog utility function to examine the behavior of imports.
10The assumption of identical preferences only serves to simplify the discussion. In the empirical analysis that follows, we provide country specific estimates of the parameters of interest. Only when the data are pooled by region do we restrict the preferences to be the same across countries within each region.
11Cointegration analysis has been employed by Ogaki (1992) and by Clarida (1994) to examine the first order condition linking the consumption of imports to consumption of the domestic good, equations (4) and (8), since this relationship must hold at each point in time. Estimation of that first order condition, if cointegration obtains, yields estimates of the intratemporal elasticity of substitution between importable and home goods, assumed to be minus unity for the Cobb‐Douglas case.
12A similar approach was taken by Hoffmaister (1992), who examines the behavior of exports and imports for Costa Rica, and by Milesi‐Ferretti (1994), who analyzes these issues for South Korea.
13The difference in the power of the rank tests when compared with the Engle‐Granger (1987) test is analyzed in Kremers, Ericsson, and Dolado (1992), who present both theoretical and Monte Carlo evidence in favor of the specifications employed in the rank tests.
14The relevant joint test is an F‐test.
15For a concise discussion of these tests, see Campbell and Perron (1991).
16The critical values are adjusted using the small sample correction suggested by Cheung and Lai (1993), which is equal to 0.1 + 0.917(T – tin, where T is the number of observations, n is the number of variables in the system, and j is the number of lags in the VAR.
17These results are available from the author upon request. However, the results presented in Tables 3 and Table 4 already provide estimates of the unrestricted system. In the case of developing countries’ import demand, the unit income elasticity could not be rejected for half of the countries. For industrial countries’ demand for developing country exports. the unit income elasticity holds in four of the eight cases in which cointegration obtains. interest income. Hence, when GDP is used to proxy permanent income, it introduces systematic biases in the income elasticities. Specifically, since GDP excludes factor payments abroad, it overstates developing country income and understates industrial country income. This systematic error, other things equal, would bias upward the industrial countries’ income elasticity, with the opposite being true for developing countries.
18The conditions sketched in the theoretical model have the developing countries as the net debtors and the industrial countries as the net lenders, who receive interest income. Hence, when GDP is used to proxy permanent income, it introduces systematic biases in the income elasticities. Specifically, since GDP excludes factor payments abroad, it overstates developing country income and understates industrial country income. This systematic error, other things equal, would bias upward the industrial countries' income elasticity, with the opposite being true for developing countries.
19For estimation of the demand for specific primary commodities, see World Bank (1994).
20Clarida (1994) estimatesplace the U.S. permanent income elasticity in the 2.12–2.21 range. Marquez (1989), who estimates these parameters for an aggregate of industrial countries, estimates it to be around 2.6.

Other Resources Citing This Publication