Economic Analysis and Policy 44 (2014) 266–278

Contents lists available at ScienceDirect

Economic Analysis and Policy

journal homepage: www.elsevier.com/locate/eap

Full length article

Competitive devaluations and the trade balance in less

developed countries: An empirical study of Latin American
countries
Rodrigo Costamagna ∗
IESE, Universidad de Navarra, Camino del Cerro del Águila, 3, 28023 Madrid, Spain

article info abstract

Article history: Policymakers in less developed countries (LDCs) usually support competitive devaluations
Received 12 May 2014 to improve the trade balance (TB). In this paper, we estimate the effects of appreciated
Received in revised form 15 August 2014 and depreciated real exchange rates (RER) in the TB of Argentina and Brazil during the pe-
Accepted 15 August 2014
riod, 1990–2010 using co-integration tests for non-stationary data and vector error correc-
Available online 26 August 2014
tion models (VECM). The estimations confirmed the existence of a long-term relationship
among the TB and the RER and foreign and domestic incomes for the countries during op-
Keywords:
Competitive devaluation
posite RER policies. Based on our estimations, the Marshall–Lerner condition held during
Trade balance periods of more flexible and depreciated RER, and the estimation of the general impulse re-
Marshall–Lerner condition sponse functions demonstrated that devaluations in both countries did not follow a J-curve
J-curve pattern in the short-term.
Co-integration test © 2014 Economic Society of Australia, Queensland. Published by Elsevier B.V. All rights
reserved.

1. Introduction

In this paper, we attempt to provide a new empirical contribution to understanding the relationship between exchange
rate depreciation (appreciation) and the trade balance (TB) in a sample of Latin American countries, namely, Argentina and
Brazil. Since the beginning of the floating era, conventional wisdom supports devaluations (competitive devaluations) to im-
prove the TB of countries. On the one hand, competitive devaluations are expected to make a country’s exports cheaper in
terms of foreign currency, and thus lead to increasing exports. On the other hand, a decline in a country’s imports is expected
due to increasing prices in the domestic currency. Based on this simplistic equation, policymakers in less developed countries
(LDCs) often support competitive devaluations to increase exports and to plan import substitution programs. However, de-
spite the vast empirical evidence that devaluations can cause significant and sustained improvement in the TB, this causality
remains mixed. Hence, the elasticity-price approach seems to contribute partially to a nation’s competitiveness policy.
It is not uncommon that countries facing a macroeconomics crisis resort to currency devaluation as a key factor in
stabilization programs when they face external imbalances. The expected impact of devaluation (appreciation) is to decrease
(increase) the relative cost of domestic labor (priced in domestic currency) and materials, and hence, to increase exports
and decrease imports. In other words, if a country’s currency is devalued, a resulting decrease in prices should increase the
quantity of exports and decrease the quantity of imports, but the trade balance can only improve if the export or import
quantities respond sufficiently to offset the deterioration in price.
However, let us consider the following example. Britain’s trade deficit has been growing over the last 20 years, but
has worsened since 2008. Between 2008 and 2009, the sterling depreciated by 25% against trading partners’ currencies to

∗ Tel.: +34 91 211 30 00.

E-mail address: rcostamagna@iese.edu.

http://dx.doi.org/10.1016/j.eap.2014.08.002
0313-5926/© 2014 Economic Society of Australia, Queensland. Published by Elsevier B.V. All rights reserved.
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 267

BRA Period 1990 - 2010 ARG Period 1990 - 2010

5 0.4 3 0.8
0.3 2.5 0.6
4
0.2 2 0.4

Log TB
Log TB

RER
3 0.2
RER

0.1 1.5
2 0 1 0

1 0.5 -0.2
-0.1
-0.2 0 -0.4
0

1990
1991
1992
1993
1995
1996
1997
1998
2000
2001
2002
2003
2005
2006
2007
2008
2010
1990
1991
1992
1993
1995
1996
1997
1998
2000
2001
2002
2003
2005
2006
2007
2008
2010
RER Log TB RER Log TB

Fig. 1. Correlation between RER and TB of goods during the period 1990–2010.
Source: International financial statistics.

mitigate the country’s macroeconomic imbalances. A competitive devaluation was expected to support domestic producers,
more precisely, to produce a direct impact on labor costs (wages) to lower export prices. However, exports deteriorated
and, consequently, the current account continued to fall by more than £100 billion per year since 2011. Britain’s external
performance after the sterling depreciation policy suggests that the price-elasticity analysis reflects a partial view of the
factors influencing the TB behavior. Certainly, several factors are associated with the improvement of the TB, such us high
levels of productivity, high technological content exports, and rising commodity prices.
The contribution of this paper is based on the specific historical context of the country samples under analysis: Argentina
and Brazil. Both countries have commodity-based economies; however, they are highly committed to the industrialization
of their economies. During the period under analysis (1990–2010), the countries’ exchange rate policies have changed in
response to economic crisis, mainly with the expectation of improving external indicators. However, during periods of high
depreciation of the domestic currency, the external performance showed unexpected results. Fig. 1 shows that the TB does
not follow the expected theoretical behavior through time. For instance, the TB in Brazil showed the highest performance
during the period of domestic currency appreciation (2000s) rather than the period of strong depreciation (1990s).
The main theoretical framework used to explain the effects of currency depreciation in a country’s TB is formally studied
through the use of the ‘J-curve’ effect for the short-term, and the Marshall–Lerner (ML) condition for the long-term. The
J-curve effect indicates that currency depreciation improves the TB in the long term but worsens it in the short-term. The
initial deterioration in the TB occurs because an initial depreciation increases spending on imports over any initial increase in
export revenue. In other words, the TB is likely to decline subsequent to a depreciation episode because it is expected that the
export and import volumes (quantities) adjust slowly to movements in relative prices, but import prices respond quickly
to exchange rate changes. Implicit in this discussion is the assumption that in the short-term elasticities are sufficiently
low and in the long-term elasticities are sufficiently high, or in the long-term the ML condition holds. According to the ML
condition, currency devaluation improves the trade balance in the long-term only if the sum of the absolute values of import
and export demand elasticities exceeds unit prices.
To verify the ML condition and the J-curve effect in this paper, we conducted a non-structural technique that models the
TB as a function of the real exchange rates (RER), and domestic and foreign expenditures. The econometric techniques are
based on the estimation of the ‘partial reduced form’ equation for the TB of goods. We first tested for unit-roots to examine
whether the stationary properties of the data are non-stationary in the series. Next, we applied the procedure developed by
Johansen and Juselius (1990) to search for co-integrating relationships from a stable vector autoregressive (VAR) modeling
specification. Based on the vector error-correction model (VECM) formulation and generalized impulse response function
(GIR), we analyzed the long and short-term trade balance dynamic for a sample of Latin American countries.
Finally, results in the estimations predicted an improvement of the TB in the long term under depreciated RER policy for
both countries. In addition, the GIR function revealed that the TB did not follow a J-curve pattern for both countries during
the periods. The controversial results shown in the estimates suggest strong policy implications for policymakers regarding
economic growth and development strategies for LDCs based on competitive devaluations.
In Section 2, the literature review is developed. Section 3 contains the theoretical framework and suggested econometric
methodology. In Section 4, the dataset is described and the empirical test is applied. Finally, in Section 5, conclusions and
policy implications are discussed.

2. Literature review

In the theoretical field, analysis of the effects of devaluations in the TB is mainly based on the elasticity approach (e.g., see
Robinson, 1947; Metzler, 1948), which describes the sufficient conditions for improvement in the TB in terms of elasticities
of demand and supply: if the demand’s elasticity is sufficiently large and the supply sufficient small, devaluations should
improve the TB. Khan (1974) and Rittemberg (1986), among others, have argued that relative prices play an important role
in the determination of trade flows in LDCs, thus, improvements in the TB after devaluations are naturally expected. In
this regard, early studies have provided mixed empirical evidence from developing and developed countries. For instance,
268 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

in Rose (1990, 1991) and Ostry and Rose (1992), among others, we found empirical evidence that relative prices have not
provided a significant and predictable impact in the TB, supporting the notion that devaluations are likely to be ineffective
in reducing trade imbalances and stimulating export growth. In addition, Reinhardt (1995) found evidence that trade flows
were significantly responsive to relative prices in developing countries, supporting the argument that a nominal devaluation
is likely to improve the TB if it is transferred into a real devaluation. For instance, in the presence of inflationary processes, the
more indexed an economy, the less the likelihood of devaluations to produce a real effect in the TB (Cooper, 1971). Regarding
the effectiveness of devaluations as a tool to improve the external accounts, Edwards (1989) found that after the third year
of a devaluation episode, currencies appreciated by the effects of inflation. However, the current account improved with
time, which is contrary to what was expected.
Recent empirical studies have focused on the examination of the long-term and short-term effects of devaluations in the
TB. Magee’s (1973) pioneer contribution supported the idea that while exchange rates adjust instantaneously, there exists
a lag in the time by which consumers and producers experience changes in relative prices.1
Thus, a favorable effect of the exchange rate depreciation in the long-term is argued. The long-term effect of devaluation
in the TB is captured by analyzing the ML condition and the J-curve effects in the short-term. According to the latter,
devaluations improve the TB in the long-term, but worsen it in the short-term. The initial deterioration occurs if the
immediate effect of depreciation is to increase spending on imports (measured in local currency) by more than an
initial increase in export revenues. Early studies covering a wide range of econometric models and estimation techniques
found evidence of the J-curve pattern and verified the ML condition in developing countries after devaluation episodes
(see Krugman and Baldwin, 1987; Bahmani-Oskooee and Malixi, 1992; Wilson, 1993; Bahmani-Oskooee and Alse, 1994;
Demirden and Pastine, 1995; Bahmani-Oskooee and Niroomand, 1998; Boyd et al., 2001). However, some authors found
mixed results (Haynes and Stone, 1982; Bahmani-Oskooee, 1985; Marwah and Klein, 1996; Bahmani-Oskooee and Ratha,
2004; Prawoto, 2007). Finally, Rose and Yellen (1989), Rose (1991), Upadhyaya and Dhakal (1996), and Shirvani and
Wilbratte (1997), among others, have not found empirical evidence of the J-curve employing co-integration techniques.
Kalyoncu et al. (2009), found mixed evidence of the effects of devaluations in the short- and the long-term using
Johansen–Juselius’ (1991) co-integration test and impulse response function in a group of Latin American countries. For
instance, only the TB of Argentina and Peru followed the J-curve pattern and, furthermore, the ML condition was held. In
contrast, Bustamante and Morales (2009), employing the co-integrated VAR model in Peru during the period, 1991–2008,
found no evidence of a J-curve pattern, but the ML condition was held. Matesanz and Fugerolas (2009) found controversial
results using VAR-based co-integration tests and the impulse response function to assess the long- and short-term effects
of RER in the TB of Argentina during the period, 1962–2005. They showed that after devaluation episodes, the TB did not
follow the J-curve pattern and, paradoxically, the ML condition holds during the period of a fixed and over-valued exchange
rate. Mahmud et al. (2004) provided evidence that the ML condition holds under fixed exchange rate policies in developed
countries. In a more recent and detailed work, Cao-Alvira and Ronderos-Torres (2014) confirmed empirically that the M–L
condition holds for the cases of Colombian non-commodity trade between the US and Venezuela. The Colombian commodity
trade was unresponsive to changes in its terms of trade. Hsing (2010) examined the functional form of the ML condition of the
bilateral trade between eight Asian countries and the US. This author found an interesting fact that adds more complexity
toward the interpretation of this type of analysis: ML condition results depend upon which relative price index is used.
Concluding, a revealing study by Bahmani-Oskooee et al. (2013) applied the recently developed ARDL co-integration method
to estimate 29 countries’ long-term import and export elasticities, and test the M–L condition using recent data. Their
findings suggested that support for the M–L condition is much weaker than commonly thought. Nonetheless, in light of
the mixed conclusions, caution is advisable for policy implications regarding the effects of devaluation in the long-term, and
the predictive power of the ML condition methodology.
Concluding, the findings reported for Latin American countries regarding the short- and long-term effects, using the J-
curve phenomenon and ML condition, are inconclusive. In addition, the findings suggest that the real exchange rate might
be insufficient to cope with the high current account deficit. The aim of this paper is to add new evidence to the existing
literature with new empirical evidence in a sample of Latin American countries. Hence, in the next sections we conduct
econometric checks to verify the short- and long-term effects of competitive devaluations in the TB.

3. Theoretical framework

Our interest in this paper relates to the dynamic response of the TB to devaluation episodes in a sample of Latin American
countries. Some influential works are based on the use of structural demand and supply equations for exports and imports
to estimate the effects of RER on the TB (see e.g., Bahmani-Oskooee, 1985; Krugman and Baldwin, 1987). In this paper we
follow the straightforward modeling introduced by Rose and Yellen (1989) and Rose (1990), called the partial ‘reduced
form equation’ for the TB examination. This analysis does not distinguish between price and volume effects. As Rose and

1 Two possible explanations for a deterioration of the trade balance by the devaluation effect are advanced. On the one hand, there is a short-term
deterioration in the trade balance by contract rigidities that take time to wear off; and second, there is a pass-through effect of currency depreciation on
domestic prices which may not take place until some time has passed after devaluation. Thus, a favorable effect of the exchange rate depreciation in the
long-term is argued.
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 269

Yellen noted, to determine the nature of the effects of RER in the TB in the short-term, the ‘partial reduced form’ equation is
preferable than the detailed structural approach. Thus, the TB dynamic is testable and directly avoids the identification and
estimation of structural parameters.
Following Rose and Yellen (1989), the ‘partial reduced form’ equation for the TB takes the following form:

TB = B(E , Y , Y ∗ ). (1)
Then, we begin defining the standard model specification for long-term demand function for exports (X ) and imports (M ):
ω  ∗ µ
Xt = P / P ∗ · E

t
· Yt (2)
ϕ
Mt = P · E /P t · (Yt )δ
 ∗
(3)
where X and M are the volume of exports and imports, respectively. E is the nominal exchange rate; p, p∗ denote the
domestic and foreign price levels and, Y , Y ∗ domestic and foreign incomes. ω and ϕ are the RER elasticities for exports
and imports, and µ and δ are the income elasticities for imports and exports.
Then, computing Eqs. (1) and (2) in logarithm form, takes the following form:

ln Xt = ω ln Pt − ln Pt∗ + ln Et + µ ln Yt∗
 
(4)

ln Xt = ϕ ln Pt − ln Pt + ln Et + δ ln Yt
∗
 
(5)
 
where, ln et = ln Pt − ln Pt∗ + ln Et is the natural logarithm of the RER and, the TB is defined as the ratio between exports
and imports:

ln TBt = µ ln Yt + δ ln Yi∗ + ψ ln et (6)

where, ψ = − (ω + ϕ). In the model, the coefficient on ln et indicates the long-term effects. For instance, if µ and δ are
positive and ω and ϕ negative, the ML condition holds whenever ψ is positive, thereby indicating that a higher RER (real
depreciation), or negative, indicating that a lower RER (appreciation) appears to improve the TB in the long-term.
Finally, we considered the Rose and Yellen (1989) approach in this study due to restrictions imposed by data availability.

3.1. Econometric procedure

The methodological approach introduced in this paper entails the direct estimation of the non-structural equation in (6).
We introduced the estimation techniques for the dynamics between the TB and the RER for the cases of Argentina and Brazil.
The dynamic specification of the TB was assumed to be related to the lagged values of RER and, additionally, for the domestic
and foreign income variables. Furthermore, because the aim of this paper was to examine the short- and long-term effects
of RER behavior on the TB, the error correction model was estimated.
The TB (TB) was defined as the ratio between exports of goods (X ) to imports of goods (M ), depending on the RER (RER)
and the real domestic and foreign incomes (Y , Y ∗ ). The RER was computed as the ratio of the US consumer price index
(CPIusa) to domestic consumer price index (CPIn) multiplied by the nominal exchange rate of the domestic currency with
US dollars. The national income (Y ) and foreign income (Y ∗ ) was defined as the gross domestic product (GDP) volume in
national currency and the US GDP was taken as a proxy of foreign output. The world income variable was included in the
model to take account of the potential effects. All the variables were analyzed in logarithmic form.
To determine a long-term equilibrium relationship among the variables under analysis, a co-integration analysis was per-
formed. Taking into account the ‘partial reduced equation’ form in (6) to model the TB dynamic, the long-term co-integrating
relationship can be written in the following log-linear form:

ln TBt = β0 + β1 ln Yt + β2 ln Yt∗ + β3 ln RERt + ut (7)

where ut is the random error term and β0 is a constant. Any positive value in the estimator β3 (β3 > 0) verifies the ML
condition. In the estimation of β1 , it was expected to be positive (β1 > 0) if an increase in the countries’ incomes induced
higher demand for imports and a rise in foreign incomes induced an increase in a countries’ demand for exports; thus, a
negative estimate of β2 was expected (β2 < 0). The RER will improve or deteriorate the TB if the coefficient of β3 was posi-
tive or negative. It was assumed that a higher RER would improve the TB and, in contrast, a lower RER would deteriorate it,
as is commonly expected.
The variables were expected to be integrated in order 1 to determine a stable long-term relationship in the series and
the linear combination was expected to be stationary in order 0. In other words, the linear combination of non-stationary
data series in Eq. (7) is stationary.
With this purpose, we first tested the stability of the VAR model running the univariate Augmented Dickey–Fuller (ADF)
unit root test to examine the behavior of each variable over time, determining whether the linear combination of the non-
stationary data series in Eq. (7) was stationary, thus describing a non-spurious regression. Following this methodology,
a necessary condition for the examination of the long-term relationship between the variables was to assess the order
of integration of the time series. In this regard, the next stage was to implement the co-integration test in Johansen and
270 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

Juselius (1990) to apply the maximum likelihood procedure to a VAR model. A co-integration test looks for stable long-term
equilibrium relationships.
Testing the existence of long-term relationships requires a pth-order structural and dynamic VAR model on the variables
which, related to the Granger representation theorem, can be written as an unrestricted VECM up to p lags (Granger
representation theorem):
p

1xt = µ + α zt −1 + Γi 1x∗t −i + µt
i=1

in our case,
p
 p
 p
 p

ln TBt = α0 + ϖj ∆ ln TBt −j + λj ∆ ln Yt −j + ϑj ∆ ln Yt∗−j + νj ∆ ln RERt −j
j =1 j =1 j=1 j =1
 
+ η ln TBt −1 − β̂0 − β̂1 ln RERt −1 − β̂2 ln Yt −1 − β̂3 ln Yt∗−1 + εt (8)

where ∆ is the first difference operator, and η provides information on the speed-of-adjustment coefficient to long-term
equilibrium and, εt is a disturbance assumed to be white noise; j is the lag order and p is the maximum number of the lag
length.
The error correction modeling denoted in Eq. (7) gives the short-term dynamic behavior model of the TB using the
estimate of past disequilibrium. In this equation, it is looking for the pattern of dynamic adjustment that occurs in the short
term to establish these long-term relationships in response to shocks in the system. In this fashion, it requires determination
of the appropriate lag length of each variable by which it is used in the Akaike Information Criterion (AIC) tests.
Next, for the unit root test, we tested for co-integration by using the maximum likelihood procedure of Johansen and
Juselius (1990) and Johansen (1989) that derives from two statistics tests to determine the number and estimation of co-
integration vectors:
(i) Trace statistic
n

λtrace (r0 |k) = −T ln(1 − λ̂i ). (9)
i=r0 +1

(ii) Maximal-eigenvalue statistic

λmax (n − 1) = −T ln(1 − λ̂i ). (10)

Once the existence of co-integration between the variables is determined, we proceeded to identify the co-integration
coefficients.
In the final step of the empirical analysis of the study, we estimated the VECM for the transitional dynamics of the system
to for the long-term equilibrium. We then applied the generalized impulse response function (GIR) to examine if shocks to
RER induced the TB to follow a J-curve for each country. Following the J-curve theoretical assumptions, under competitive
devaluations an initial deterioration in the TB is expected, followed by an improvement in the long-term.

4. Data and empirical results

Quarterly data from the International Financial Statistics database compiled by the International Monetary Fund (IMF)2
are used for the analysis in this paper. They include two Latin American countries that are the focus of the present study,
Argentina and Brazil, for the period, 1990–2010. For Argentina, the RER was calculated based on additional data from the
Foundation for Latin American Economic Research (FIEL),3 more precisely, the domestic CPIarg. TB was constructed using
goods, exports and imports: F.O.B. end of period, in nominal USD terms. The national CPI indices were based on those for
2005. National and foreign income referred to real gross domestic product (GDP) in national currency.

4.1. Test for unit-roots

A necessary condition to test the long-term relationship with time series variables is to first test the unit-roots in the
variables included in reduced equation form. In this paper, we use the Dickey–Fuller test in Eq. (6). We first tested the
augmented ADF test that involves estimation of the following regression for each individual variable:
m

1Xt = α + βt + ρ Xt −1 + φ i1Xt −1 + εt (11)
i=t

2 http://www.imf.org/.
3 http://www.fiel.org/.
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 271

Table 1
Average industry exports on total exports for Argentina and Brazil.
Source: Data obtained from World Trade Organization (http://stat.wto.org/). Calculations made
by the author.
Exports Argentina Brazil
1990–2000 2000–2010 1990–2000 2000–2010

Agricultural products 53.70% 51.75% 32.74% 32.81%

Fuels and mining products 14.09% 16.51% 12.20% 20.01%
Manufactures 32.21% 31.74% 55.07% 47.18%

Table 2
ADF-testing for unit roots for Argentina and Brazil.
Argentina Brazil
RER RER

−1.565 0.131
1990–2001 1990–1999
(lag = 3) (lag 3)
−2.175 0.141
2002–2010 2000–2010
(lag = 5) (lag 5)
TB TB
−1.998 0.107
1990–2001 1990–1999
(lag = 1) (lag 4)
−1.953 0.204
2002–2010 2000–2010
lag(2) (lag 5)
GDP GDP
−1.892 0.196
1990–2001 1990–1999
(lags = 2) (lag 5)
1.213 0.075
2002–2010 2000–2010
(lags = 4) (lag 4)
GDP GDP
0.073 0.169
1990–2001 1990–1999
(lag = 4) (lag 5)
−1.680 0.0628
2002–2010 2000–2010
(lag = 3) (lag 5)
ADF: critical value at 5% = −2.904.

where εt is the error term and m (i.e., the number of lagged first-differenced terms) is determined such that εt is
approximately white noise. The null hypothesis that 1Xt is a non-stationary time series translates into Ho: β0 = 0. The
null hypothesis is rejected if β0 is significantly negative. The sample periods for Argentina were 1990 through 2001 and
2002 through 2010; in the case of Brazil, the periods were 1990 through 1999 and 2000 through 2010. Critical values of
rejection are also reported for all tests.
Table 2 summarizes the results in both periods under analysis and indicates the presence of unit roots of each variable
for Brazil and Argentina. The ADF test is consistent with the hypothesis that a non-stationary unit root characterizes each of
the variables. For instance, ln Y , ln Y ∗ and ln RER contain unit-root I (1) in their levels form but not in their first differences
form. The null hypothesis of a unit roots in the univariate representation cannot be rejected for any of the variables at a
reasonable significance level.
Once it was established that series are I (1), we proceeded to test for a long-term, co-integrating relationship between
ln TB, ln RER, ln Y and ln Y ∗ using Johansen’s (1989) co-integration test procedures.

4.2. Co-integrating VAR analysis

Before undertaking co-integration tests, the relevant order of lags (ρ ) of the vector autoregression model (VAR) was
specified. The appropriate number of lags of the VAR model was selected on the basis of the Akaike (AIC) criteria in all
samples. The results obtained from the Johansen and Juselius methods are shown in Table 3.
The results showed that the best lag order in Argentina during the first period (1990–2001) was three years and one year
for the second period (2002–2010). In the case of Brazil, the best lag order was one year for the first period (1990–1999)
and two years for the second period (2000–2010). Next, we applied the Johansen and Juselius procedure to test the number
and estimation of the co-integrating relationship. Let r be the number of co-integration equations from zero to k − 1, where
k = 4 is the number of endogenous variables in the model. The trace statistic tests the hypothesis of the existence of r co-
integrating vector (H0 : r = a against the alternative HA : r ≥ a + 1), while the eigenvalue tests the hypothesis of existence
of the r co-integrating vector against the alternative that r + 1 exists (H0 : r = a against the alternative HA : r ≥ a + 1).
In Table 4, the results suggest that the null hypothesis of no co-integration (r = 0) for the variables in Eq. (6) is rejected by
272 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

Table 3
Selection order criteria for Argentina and Brazil.
LL LR df ρ EPE AIC HQIC SBIC

Argentina, 1990–2000
0 124.3 2.5 −6.2 −6.1 −5.99
1 190.5 132.3 16 0.000 1.9 −8.7∗ −8.43∗ −7.89∗
2 214.5 47.93 16 0.000 1.3∗ −9.2 −8.6 −7.61
3 227.5 26.1 16 0.05 1.6 −9 8.2 −6.78
4 242.5 30.02∗ 16 0.02 2 −8.9 7.9 −6.04
Endogenous variables: TB, GDParg, GDPusa, RER.
Argentina, 2002–2010
0 297.35 7.2 −14.3 −14.24 −14.14
1 418.44 242.19 16 0.000 4.3 −19.43 −19.13 −18.60
2 448.77 60.65 16 0.000 2.2 −20.13 −19.58 −18.63
3 508.01 118.48 16 0.000 2.8∗ −22.24∗ −21.43∗ −20.07∗
4 523.66 31.31∗ 16 0.012 3.2 −22.22 −21.19 −19.38
Endogenous variables: TB, GDParg, GDPusa, RER.
Brazil, 1990–1999
0 147.94 1.7 −6.54 −6.48 −6.38
1 289.77 283.6 16 0.000 5.6 −12.26∗ −11.96∗ −11.45∗
2 310.73 41.9 16 0.000 4.5 −12.48 −11.94 −11.02
3 343.06 64.6 16 0.000 2.3 −13.23 −12.44 −11.12
4 362.99 39.8∗ 16 0.001 2.1 −13.40 −12.38 −10.65
Endogenous variables: TB, GDParg, GDPusa, RER.
Brazil, 2000–2010
0 110.99 3.1 −5.94 −5.88 −5.76
1 278.16 334.3 16 0.000 7.0 −14.34 −14.03 −13.46
2 306.06 55.7 16 0.002 3.7∗ −15.00∗ −14.45∗ −13.42∗
3 317.82 23.5 16 0.101 5.2 −14.76 −13.96 −12.48
4 331.83 28.0∗ 16 0.031 7.0 −14.65 −13.61 −11.66
Endogenous variables: TB, GDParg, GDPusa, RER.

Table 4
Johansen co-integration test results. Co-integration equation for Argentina and
Brazil.
Rank Parms LL Eigenvalue Trace stat. 5% Critical value
Argentina, 1990–2001

0 36 450.57 – 106.54 47.21

1 43 493.62 0.85 106.54 29.68
2 48 500.93 0.27 20.44∗ 15.41
3 51 503.21 0.09 5.81 3.76
4 52 503.84 0.02 1.26
Brazil, 1990–1999
0 20 274 – 73.13 47.21
1 27 296.6 0.69 27.84∗ 29.68
2 32 304.5 0.34 11.96 15.41
3 35 310.2 0.25 0.56 3.76
4 36 310.5 0.01 0

the trace statistic test that confirms one co-integrated vector at the 5% level of significance among the variables for the case
of Argentina and Brazil in all periods.
Long-term elasticity estimates from the VECM and their standard error are reported in Table 5 for Y , Y ∗ and RER. Setting
the estimated coefficient ln TB at −1 we normalized the income variables and RER coefficients and, thus, the estimates of
β1 –β3 indicate the long-term TB elasticities. Each co-integrating parameter (β1 , β2 and β3 ) measures the TB elasticity with
respect to Argentina and Brazilian income, US income and RER.
During periods of appreciated RER in Argentina (1990–2001), the TB is negatively associated with domestic output (GDP)
as predicted in the model. During periods of flexible and depreciated exchange rate policy (2002–2010), the TB showed
unexpected effects: the TB was positively associated with both, domestic and international output (GDP), meaning that
exports and imports increased simultaneously. To elucidate the point above, in Fig. 2 we introduced new data that added
complexity to the interpretation of the estimates. During the period of high depreciation, manufacturing exports increased
at levels superior to the period of a fixed and appreciated exchange rate; however, trade deficits of manufacturers have
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 273

Table 5
Johansen co-integration test results. Co-integration equation for Argentina and Brazil.
Beta Coef. Std. error z P > IzI (95% conf. interval)
Argentina, 1990–2001

TB 1 – – – – –
GDParg −2.19 0.60 −3.62 0.000 −3.38 −1.005
GDPusa 4.86 0.55 8.75 0.000 3.77 5.95
RER −8.22 0.50 −16.1 0.000 −9.22 −7.2
Cons −53.0 – – – – –
Notes: The vectors are normalized for TB: b1, b2 and b3 denote the ARG
GDP, USA GDP, and TCR elasticities of trade balance, respectively. Results
carried out by STATA.
Brazil, 1990–1999
TB 1 – – – – –
GDPbra 0.042 0.009 4.68 0.000 −3.38 −1.005
GDPusa −0.289 0.457 −0.6 0.527 3.77 5.95
RER −0.275 0.154 −1.8 0.076 −9.22 −7.2
Cons 4.24 – – – – –
Notes: The vectors are normalized for TB: b1, b2 and b3 denote the BRA GDP, USA GDP, and
TCR elasticities of trade balance, respectively. Results carried out by STATA.

Fig. 2. Trade balance and exchange rate.

Source: Data obtained from World Trade Organization. http://stat.wto.org/.

prevailed under the depreciation policy. In a sense, evidence in Fig. 2 suggests that exports of commodity goods compensate
the trade deficit of manufacturers, which would explain the long-term improvement of the TB in the estimates. For instance,
the fact that international and domestic outputs (GDP) had increased during the period is likely to be associated with
increasing imports in the former case, and with the increasing prices of commodity exports. Certainly, the commodity boom
mainly characterized the 2000s.
During the period 1990–2000 Brazil encountered several macroeconomic crises with strong implications for monetary
policy; however, an appreciated exchange rate policy prevailed. During this period, the TB was negatively associated with
international output (GDP), as expected. In contrast, during the period of floating and appreciated rates (2000–2010), the
TB was also negatively associated with domestic output (GDP), but positively associated with international output (GDP).
Data in Fig. 3 shows that the manufacturing balance deteriorates to the extent that the exchange rate policy showed a trend
toward appreciation. Therefore, in the mid 2000–2010 period, the TB of manufacturers became a deficit, which potentially
explains the decline in domestic GDP and the increase in international output (GDP).
The TB and the RER were positively co-integrated, and the elasticity coefficient in β3 was positive and statistically
significant at the 95% coefficient level during periods of more depreciated RER in both countries. The estimates allow us
to conclude that the ML condition is verified under depreciated exchange rate periods, that is, for Argentina during the
2002–2010 period, contrasting with a more appreciated exchange rate in Brazil during the 2000–2010 period. For the
Brazilian TB behavior, our findings oppose those of Gomes and Paz (2005), who found that the ML condition did not hold
during the period of more appreciated exchange rate policy (1990–1999), suggesting that the TB deteriorated under a fixed
exchange rate regime. Nevertheless, Figs. 2 and 3 showed that the improving trend in the TB in the long-term should
be interpreted with caution. For instance, while exports of manufactured goods increased during the period of floating
and depreciated exchange rates, the manufacturing sector was in deficit in both countries. Furthermore, the data showed
274 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

Fig. 3. Trade balance and exchange rate.

Source: Data obtained from World Trade Organization. http://stat.wto.org/.

vec_br, arg_rer, arg_tb

.006

.004

.002

0
0 10 20 30
step
Graphs by irfname, impulse variable, and response variable

Fig. 4. Argentina 1990–2001. Response of ln TB. One S.D. ln RER innovation.

that rising commodity exports compensated the trade deficit in manufactured goods in both countries. Thus, competitive
devaluation policies causing the TB of manufactured goods to improve in the long-term need to be verified in detail in our
country samples. In this sense, the appreciation of the currency in Brazil may explain the increasing levels of manufacturing
imports, showing low price-elasticity in the manufacturing sectors.
In conclusion, estimates showed that a flexible and depreciated RER policy improved the TB for Argentina in the long-
term. In addition, an appreciation policy improved the TB in the long term for Brazil. However, based on the new data we
added to the analysis, results showed that countries found difficulties in reducing the inflow of imports of manufactures
under competitive devaluation policies. Hence, this potentially explains the TB deterioration in the manufacturing sectors.

4.3. Generalized impulse response analysis (GIR)

To examine the J-curve pattern, the generalized impulse response function (GIR) of ln TB was calculated for a one
standard-deviation RER innovation. For this purpose, the information from Johansen’s correction model and a one-time
shock to the RER was traced. The results are reported in Figs. 4 and 5 for Argentina, and in Figs. 6 and 7 for Brazil. Through
depreciation, an initial worsening in the TB by an improvement would give rise to J-curve effects. The effect arises because
in the short term, price effects prevail, and quantity adjustments dominate in the long-term.
Fig. 4 shows no evidence of the J-curve pattern in Argentina for the period 1990–2001. The result shows an initial positive
response to a shock in the exchange rate (improvement of the TB in the short-term). Likewise, an improvement in the long-
term prevails. Surprisingly, Fig. 5 shows a strong positive response to a shock in the exchange rate (improvement of the TB
in the short-term) until the second quarter, with a prompt fall of the TB in the long-term. This response indicates that the ML
condition was not verified under a policy of depreciation. Supporting these findings, data in Fig. 1 shows that the inflation
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 275

vec_br, arg_rer, arg_tb

.01

.005

-.005

-.01
0 10 20 30
step
Graphs by irfname, impulse variable, and response variable

Fig. 5. Argentina 2002–2010. Response of ln TB. One S.D. ln RER innovation.

vec_br, bra_rer, bra_tb

.02

.01

-.01
0 10 20 30
step
Graphs by irfname, impulse variable, and response variable

Fig. 6. Brazil 1990–1999. Response of ln TB. One S.D. ln RER innovation.

vec_br, bra_rer, bra_tb

.01

.005

0
0 10 20 30
step
Graphs by irfname, impulse variable, and response variable

Fig. 7. Brazil 2000–2010. Response of ln TB. One S.D. ln RER innovation.

rate increased during the period and, thus, produced real appreciation of the RER. Therefore, the fact that the economy
reacted swiftly to depreciation suggests that manufacturing exports were tied to competitive devaluations.
276 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

In Fig. 6, evidence of the J-curve pattern was not observed in Brazil during the period 1990–1999. An initial strong positive
response to a shock in the exchange rate (improvement of the TB in the short-term) is shown and a deterioration of the TB
prevails in the long-term. Fig. 7 shows that a J-curve pattern was not found during the period 2000–2010, but was followed
by a continuous improvement in the long term.
In conclusion, the J-curve pattern was not verified in Argentina and Brazil for any of the periods under analysis using
quarterly data. These effects showed a common pattern: exports were sensitive to competitive devaluations; however,
countries were not able to sustain a depreciated exchange rate over time.

5. Conclusions and policy implications

This paper empirically examined the effects of a depreciating (appreciating) exchange rate on the trade balance in
the Latin American countries of Argentina and Brazil. In light of the recurrent debate relating to the uses of competitive
devaluations as a path to improving the competitive position of LDCs, the exchange rate policies in Argentina and Brazil
provide an opportunity to contribute new empirical insights for policymakers. Furthermore, because of the countries’
common features, the causal analysis allowed us to deepen the understanding of the effects of a competitive devaluation
policy. Both countries are rich and abundant in natural resources, push industrialization programs through competitive
devaluations and share the same geo-economics. However, the greatest similarity is that they both underwent similar
macroeconomic crises during the period under analysis and followed similar strategies to overcome them, using exchange
rate depreciation as the key variable of the countries’ stabilization programs and competitiveness strategies.
To empirically explore the effects of devaluation (appreciations) on the TB, we examined the long-term effects of deval-
uation through the Marshall–Lerner condition, and the short-term effects through the J-curve analysis. The ML condition
was verified during periods of strong depreciation policies. In addition, GIR analysis showed that Argentina and Brazil’s
TB has not followed the J-curve pattern of adjustment after a shock in the exchange rate. Following the J-curve analysis,
results were contradictory. In the case of Argentina, the TB deteriorated in the long-term under a policy of depreciation
and, paradoxically, improved under fixed and appreciated RER. In addition, Brazil’s TB deteriorated in the long-term under
depreciated and fixed RERs, and improved in the long-term under a floating and appreciation policy.
These mixed results deviated from the estimates motivate a round table of policy discussions and implications. For
instance, the additional data introduced in this article in Figs. 2 and 3 suggest that the external performance of countries is not
fully associated with competitive devaluations, but with a set of events that cannot be entirely captured in the model. First,
results showed that the rise in commodity prices during the 2000s largely contributed to the countries’ TB. However, prices
and exports were not subject to competitive devaluations. Therefore, the second period (2000–2010), which showed better
long-term prospects than the previous period (1990–2000), might intuitively be attributed to the considerable contribution
of the balance of commodities. Certainly, data in Figs. 2 and 3 showed that while exports of manufacturing grew, the increase
in manufactured imports caused a deficit in the balance of manufacturing. Second, while the elasticity analysis confirmed
that exports of manufactured goods were sensitive to competitive devaluations, the economies were not able to sustain real
depreciations through time. Therefore, in light of the deterioration of the TB of manufacturers and the real appreciation of
the domestic exchange rate, new devaluations should be expected to balance the trade deficit. Hence, the evidence from
this study’s results on the dynamics of the TB suggests that a competitive devaluation policy could cause economies to
fall into a vicious circle of continuous devaluation to reach a positive TB of manufacturers. In this sense, we suggest the
following policy implications. First, the trade deficit of manufacturers during periods of export growth in general reveals
that the import substitution policy is likely to be inefficient. Therefore, we suggest that a competitive devaluation was
not the more effective tool to gain international competitiveness, but the efficient exploitation of the extractive sectors.
Second, country data in Table 1 help us to discuss industrial-policy implications. Almost 70% of exports in Argentina have, in
recent times, been commodities, and manufacturing exports are mainly in the Maquiladora industry,4 such as the automotive
sector. Similarly, Brazil’s commodity exports have represented almost 50% of total exports, but have shown a higher degree
of industrialization; however, exports of manufacturers fell 10% on average from one decade to another. Therefore, how
should countries increase exports or reduce imports? The country data in this study show that increasing levels of imports
threaten the TB (see Tables 6 and 7).
Thus, it makes sense for countries to increase industrialization programs in sectors where they have competitive
advantages that is the commodity sectors. In fact, we suggest that future research could empirically test the effects of
depreciation (appreciation) on investment in commodity industrialization, more precisely, R&D investments (innovations).
Seminal work by Grossman and Helpman (1995) supports the idea that R&D is likely to produce a greater impact on
competitiveness and exports than investments on physical capital.
The theoretical model utilized in this paper has limitations that have not allowed us to capture some potential effects on
trade, which should be useful to explore in future studies. For instance, both countries are primary partners in the MERCOSUR
trade block. However, the model cannot evaluate the intra-block trade effects and the market-size effects, which are assumed
to have strong implications for both the TB and exchange rate policy. Thus, while these research questions have been widely

4 A Maquila industry is based on assembly processes and low-cost strategies (processes and labor), which require high levels of imports of intermediate
goods for production.
 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278 277

Table 6
Johansen and Juselius test for co-integration for Argentina and Brazil.
Rank Parms LL Eigenvalue Trace stat. 5% Critical value
Argentina, 2002–2010

0 20 176.7 – 114.3 47.21

1 27 221.8 0.91 24.07∗ 29.68
2 32 230.4 0.38 6.85 15.41
3 35 232.6 0.11 2.45 3.76
4 36 233.8 0.06 1.26
Brazil, 2000–2010
0 4 255.37 – 68.78 47.21
1 11 277.73 0.63 24.08∗ 29.68
2 16 285.44 0.29 8.65 15.41
3 19 288.64 0.13 2.25 3.76
4 20 289.77 0.05

Table 7
Johansen co-integration test results for Argentina and Brazil. Co-integration equation.
Beta Coef. Std. error Z P > IzI (95% conf. interval)

TB 1 – – – – –
GDParg 5.86 0.35 16.7 0.000 5.17 7
GDPusa 0.091 0.17 0.53 0.595 −0.24 0.43
RER 4.63 0.27 16.6 0.000 4.09 5.18
Cons −70.3 – – – – –
Notes: The vectors are normalized for TB: b1, b2 and b3 denote the ARG GDP,
USA GDP, and TCR elasticities of trade balance, respectively. Results carried
out by STATA.
Brazil, 2000–2010.
Beta Coef. Std. error z P > IzI (95% conf. interval)
TB 1 – – – – –
GDPbra −0.89 0.28 −3.1 0.002 −1.46 0
GDPusa 2.01 0.31 6.34 0 1.39 2.63
RER 0.23 0.4 0.59 0.557 −0.55 1.02
Cons −22.2 – – – – –
Notes: The vectors are normalized for TB: b1, b2 and b3 denote the BRA GDP, USA GDP, and TCR
elasticities of trade balance, respectively. Results carried out by STATA.

discussed in the literature, we suggest testing these effects in the MERCOSUR trade block as a basis for discussing trade-block
formation.
Finally, in our view, if the growth of manufacturing exports leads to manufacturing TB deficits, caution is advisable when
pushing competitive devaluations to improve the TB in general. The evidence in this paper suggests that these economies
could fall into the vicious circle of depreciation dependence.

References

Bahmani-Oskooee, M., 1985. Devaluation and the J-curve: some evidence from LDCs. Rev. Econ. Stat. 67, 500–504.
Bahmani-Oskooee, M., Alse, J., 1994. Short-run versus long-run effects of devaluation: error correction modeling and cointegration. Eastern Econ. J. 20 (4),
453–464.
Bahmani-Oskooee, M., Harvey, H., Hegerty, S.W., 2013. Empirical tests of the Marshall–Lerner condition: a literature review. J. Econ. Stud. 40 (3), 411–443.
Bahmani-Oskooee, M., Malixi, M., 1992. More evidence on the J-curve from LDCs. J. Policy Model. 14, 641–653.
Bahmani-Oskooee, M., Niroomand, F., 1998. Long-run price elasticities and the Marshall–Lerner condition revisited. Econom. Lett. 61, 101–109.
Bahmani-Oskooee, M., Ratha, A., 2004. The J-curve: a literature review. Appl. Econ. 36, 1377–1398.
Boyd, D., Caporale, G.M., Smith, R., 2001. Real exchange rate effects on the balance of trade: cointegration and the Marshall–Lerner condition. Int. J. Finance
Econ. 6, 187–200.
Bustamante, R., Morales, F., 2009. Probando la condición de Marshall–Lerner y el efecto curva-J: Evidencia empírica para el caso peruano, Banco Central de
Reserva del Perú, Estudios Económicos no. 16, pp. 103–126.
Cao-Alvira, J.J, Ronderos-Torres, C., 2014. Real exchange rate volatility on the short- and long-run trade dynamics in Colombia. Int. Trade J. 28 (1), 45–64.
Cooper, R.N., 1971. Currency Devaluation in Developing Countries. In: Essays in International Finance, vol. 86. Princeton University, International Finance
Section, Princeton, NJ.
Demirden, T., Pastine, I., 1995. Flexible exchange rates and the J-curve: an alternative approach. Econom. Lett. 48, 373–377.
Edwards, S., 1989. Exchange controls, devaluations, and real exchange rates: the Latin American experience. Econ. Dev. Cult. Change 37 (3), 457–494.
Gomes, F.A.R., Paz, S., 2005. Can real exchange rate devaluation improve the trade balance? The 1990–1998 Brazilian case. Appl. Econ. Lett. 12, 525–528.
Grossman, G., Helpman, E., 1995. Technology and trade. In: Handbook of International Economics, vol. 3. North Holland.
Haynes, Stephen E., Stone, J.A., 1982. Impact of the terms of trade the US trade balance: a reexamination. Rev. Econ. Stat. 64, 702–706.
Hsing, Y., 2010. Test of the Marshall–Lerner condition for eight selected Asian countries and policy implications. Glob. Econ. Rev. 39 (1), 91–98.
Johansen, S., 1989. Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica 59, 1551–1580.
278 R. Costamagna / Economic Analysis and Policy 44 (2014) 266–278

Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inferences on cointegration with applications to the demand for money. Oxford Bull.
Econ. Stat. 52, 169–210.
Kalyoncu, H., Ozturk, I., Artan, S., 2009. Devaluation and trade balance in Latin American countries. J. Econ. Bus. 27 (1), 115–128.
Khan, M., 1974. Import and export demand in developing countries. IMF Staff Pap. 21, 678–693.
Krugman, P., Baldwin, R., 1987. The persistence of the US trade deficit. Brookings Pap. Econ. Act. 1, 1–55.
Magee, S.P., 1973. Currency contracts, pass through and devaluation. Brooking Pap. Econ. Act. 1, 303–325.
Mahmud, S.F., et al., 2004. Testing Marshall–Lerner condition: a non-parametric approach. Appl. Econ. Lett. 11, 231–236.
Marwah, K., Klein, L.R., 1996. Estimation of J-curve: United States and Canada. Canad. J. Econom. 29, 523–539.
Matesanz, D., Fugerolas, G., 2009. Exchange rate policy and trade balance: a cointegration analysis of the Argentine experience since 1962. Appl. Econ. 41,
2571–2582.
Metzler, L., 1948. The theory of international trade. In: Ellis, Howard S. (Ed.), Survey of Contemporary Economics, vol. 1. Blakiston, Philadelphia.
Ostry, J., Rose, A., 1992. An empirical evaluation of the macroeconomics effects of tariff. J. Int. Money Finance 11, 63–79.
Prawoto, R.B., 2007. Cointegration analysis on trading behavior in four selected ASEAN countries before monetary crisis. Gadjah Mada Int. J. Bus. 9 (2),
273–290.
Reinhardt, C., 1995. Devaluations, relative prices and international trade. IMF Staff Pap. 42, 290–312.
Rittemberg, L., 1986. Export growth performance of LDC’s. J. Dev. Econ. 24, 167–177.
Robinson, J., 1947. The pure theory of international trade. Rev. Econom. Stud. 14, 98–112.
Rose, A.K., 1990. Exchange rates and the trade balance: some evidence from developing countries. Econom. Lett. 34, 271–275.
Rose, A.K., 1991. The role of exchange rates in a popular model of international trade: does the ‘Marshall–Lerner’s condition hold? J. Int. Econ. 30, 301–316.
Rose, A.K., Yellen, J.L., 1989. In there a J-curve? J. Monetary Econ. 24, 53–68.
Shirvani, H., Wilbratte, B., 1997. The relation between the real exchange rate and the trade balance: an empirical reassessment. Int. Econ. J. 11, 39–49.
Upadhyaya, K., Dhakal, D., 1996. Devaluation and the trade balance: estimating the long run effect. Appl. Econ. Lett. 4, 343–345. 1997.
Wilson, W., 1993. J-curve effect and exchange rate pass-through: an empirical investigation of the United States. Int. Trade J. 7, 463–483.