JOURNAL OF

Development
Journal of Development Economics ECONOMICS
ELSEVIER Vol. 54 (1997) 77-100

Equilibrium real exchange rate, volatility, and

stabilization
Gustavo M. Gonzaga *, Maria Cristina T. Terra
PUC-Rio, Rua Marques de Sao Vicente 225, CEP 22453-040, Rio de Janeiro, Brazil

Abstract

This paper is composed of two parts. The theoretical part studies the effect of real
exchange rate (RER) volatility on trade using a general equilibrium framework. The
volatility of the RER is derived endogenously, and is caused by a demand shock, which
may be influenced by inflation volatility. The model shows that RER volatility affects
positively the equilibrium RER. In the empirical part, the behavior of several RER volatility
indexes over the last fifteen years for Brazil is examined, identifying the influence of
stabilization plans and inflation volatility. In fact, inflation volatility explains most of the
variation in RER volatility in Brazil. In addition, export supply equations that include RER
volatility as one of the explanatory variables are estimated for Brazil. For most specifica-
tions the RER volatility coefficient is negative, although not significantly different from
zero. The implied elasticity for the most significant RER volatility coefficient is -0.05.
© 1997 Elsevier Science B.V.

Keywords: Equilibrium real exchange rate; Volatility; Export supply

I. Introduction

There is a large theoretical and empirical literature on the effects of real

exchange rate (RER) volatility on international trade (see Crt6, 1994 for a recent
survey). Traditional models consider risk averse exporters for whom the RER is
the source o f uncertainty. Two assumptions are crucial for the volatility of the
RER to affect the exporting decision. One is that there is no perfect h e d g i n g - -

* Corresponding author.

0304-3878/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved.

PH S0304-3878(97)00030-8
78 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100

access to exchange rate forward market would reduce the effect. The other is that
exporters have to be very risk averse. As Caballero and Corbo (1989) point out,
profit is a convex function of prices; hence, increased variability of prices
increases profits. To capture the behavior of a risk averse agent facing risk,
exporters are assumed to maximize a concave function of profits, because a
concave function has the property of decreasing with the variability of its
argument. Hence, exporters maximize a concave function (a utility function) of a
convex function (the profit function) of prices. They will ultimately be maximizing
a concave function of prices if the utility function is sufficiently concave, that is, if
they are sufficiently risk averse.
There is an alternative line of research, in which the international market for the
country's export sector is non-competitive. There are costs associated with enter-
ing a n d / o r exiting the market; therefore, an increased volatility of the RER would
make the option of entering a n d / o r exiting the market more valuable. More
volatility of the RER would then make exports less responsive to variations in the
RER level, as shown by Dixit (1989).
Traditional models, however, are better suited for the purposes of this paper
than the 'option' framework. This paper intends to study situations related to
developing economies, for which the export sector is composed mostly of products
that present high degree of competitiveness in the foreign markets. Therefore, the
behavior of the sector is better captured by the assumption of risk averse exporters
in a competitive environment.
One common feature of all models that relate the volatility of the RER to trade
is that they use a partial-equilibrium approach. They usually focus on the export
sector, and study the effect of an exogenously given volatility of the RER on the
quantity of exports. This paper makes a step into using a general equilibrium
framework. Substitution across sectors is considered, and the model allows the
study of the effect of volatility on the equilibrium real exchange rate, that is, on
the value of the RER that yields equilibrium in all markets of the economy. The
volatility of the RER is derived endogenously. Its original source in the model is a
demand shock.
The theoretical part of the paper shows that the value of the equilibrium real
exchange rate is affected by its own volatility. Risk averse exporters, that make
their exporting decision before observing the realization of the RER, choose to
export less the more volatile is the RER. Therefore, the RER that ensures
equilibrium in all markets will be more devalued.
Another point this paper intends to make is that price stabilization plans may
affect the variability of the RER. The effect on volatility is clear when the price
stabilization plan embodies a change in the exchange rate regime. If the exchange
rate was flexible before the plan, and is fixed after the plan, for instance, then, a
lower volatility of the RER should be expected. However, even if the exchange
rate regime remains unchanged, as in our model, price stabilization may affect the
variance of the inflation rate. Price stabilization means that the inflation rate
 G.M. Gonzaga, M. C.T. Terra~Journal of Development Economics 54 (1997) 77-100 79

moves to a lower level, and that may affect its variability. 1 The lower inflation
volatility would then probably affect the variability of the RER.
In sum, the message this paper wants to convey is that the variability of the
RER may affect its equilibrium level, and price stabilization may affect that
variability. If our theoretical results are correct, empirical studies should include
the RER volatility as one of the explanatory variables of the RER itself.
The empirical section consists of two parts. In the first part, we investigate an
alternative source of RER volatility: the effects of stabilization plans in high-infla-
tion countries. It contains an extensive description of the behavior of several
measures of the RER volatility for Brazil, using monthly, weekly and daily data
over the last fifteen years. Interesting patterns of volatility could be associated to
the nature of the several stabilization plans adopted, and to changes in the
exchange rate regimes. Simple OLS regression shows that RER volatility was
largely explained by movements in inflation volatility.
The effect of RER volatility on its equilibrium level suggested by the theoreti-
cal model depends crucially on whether exports decisions are affected by the
variability of the RER. The second part of the empirical section performs the
estimation of export supply equations for Brazil that include RER volatility as one
of the explanatory variables. For most specifications, the RER volatility coefficient
was negative, although not significantly different from zero.
This paper is organized as follows. Section 2 presents the model. Section 3
contains the empirical results. Section 4 concludes and points directions for future
research.

2. The model

In a simple general equilibrium framework, the model presented here tries to

capture the effect of the volatility of the real exchange rate on its equilibrium
level. The effect arises from the assumption that firms are risk averse and the
decision on how much to export is made before the RER is observed. When the
export activity is riskier relative to the others, less resources will be allocated to it.
To maintain external balance, the RER has to depreciate. Therefore, there is a
negative relation between the equilibrium RER and its volatility.

2.1. Production

A small open economy is considered, which produces three goods: a non-trada-

ble good (QN), an importable good (QM) and an exported good (Qx). The
importable good is a perfect substitute of the country's imports. The exported good
is that good produced exclusively for export, and is not consumed locally. It can

i For an empiricalstudy on the relation between inflation level and its variabilityfor Brazilian data,
see lssler (1991).
80 G.M.Gonzaga,M.C.T.Terra~Journalof DevelopmentEconomics54 (1997)77--100

be thought as a good that is produced to attend foreign specifications. It is

assumed that at the beginning of the period, before observing the realization of the
RER, finns have to make a binding contract specifying the amount to be produced
of the exported good. Assuming that the producer is risk averse, she will maximize
the expected value of a concave function, which will be called utility function, of
her profit. The problem of a representative finn at the beginning of the period is
represented by:
MaxE[U(~xQx(Lx)+tMQM(LM)+QN(Z-Lx-LM))]. (1)
For simplicity, only one mobile factor of production exists, L,_ presenting
decreasing returns, and the firm's endowment of this factor is L. Qx(Lx),
QM(LM), and Q N ( L - L x -LM) represent the production functions of the ex-
ported, importable and non-tradable goods, respectively. The non-tradable good is
the numeraire, and the relative prices of the exported and importable goods, Px
and tM, are uncertain. The prices of the tradable goods are equal to the
international price, pj*, exogenous and assumed constant, multiplied by the
nominal exchange rate relative to the price of non-tradable goods, g, as repre-
sented in Eq. (2)
tj=~pj*,forj= X,M. (2)
By solving the maximization problem above, the producer chooses how much
labor to allocate for the production of the exported good: ~
Qx = qx( E( pxU' ) , E(tMU'), E( U') ), (3)
where U' is the derivative of the utility function with respect to profits. The
production of the exported good is positively related to E[ t x U']--which can be
interpreted as the marginal utility for the producer of producing one extra unit of
the exported good--and is negatively related to E(U')--which can be interpreted
as the marginal utility for the producer of producing one extra unit of the
non-tradable good.
After the realization of the random variable, the firm decides how much to
produce of the importable and non-tradable goods with the labor net of the amount
used in the production of the exported good. The offer functions of the two goods
are represented as the functions:
QM = qM(PM, E ( t x U ' ) , E(pMUt), E(U')) (4.a)
and
QN = qN(PM, E ( t x U ' ) , E ( t i U ' ) , E(U')), (4.b)
where PM is the realization of the random variable tiM" The derivatives of the two
functions have the following signs: (OqM/OpM) > O, (OqM/OE [tMU']) > 0,

2 The solution also yields the amount planned to be allocated to the importable and to the
non-tradable goods. However,the decisionof how much labor is allocatedbetweenthem is made after
the realizationof the real exchangerate.
 G.M. Gonzaga,M.C.T. Terra~Journalof DevelopmentEconomics54 (1997) 77-100 81

(OqM/OE[U']) < O, (OqN/OpM) < O, (OqN/OE [/3MU' ]) < 0, (OqN/OE[U']) > O,

and (aqJaE[ ~xU']) < 0 for j = M, N.
2.2. Consumption
Now that the production side of the economy is defined, let's turn to the
consumption decisions. The consumer in this model economy consumes two types
of goods: importables and non-tradables. She maximizes her utility from consump-
tion, subject to her budget constraint. A cash-in-advance constraint is introduced,
so that the demand for good depends positively on the amount of money the
consumer holds. The role of introducing money in this model is to create a
demand shock. Hence, money should be viewed here as a source of demand
shocks. 3 The demand for each type of good may be represented by:
CM = CM( PM, m) (5.a)
and
CN -----CN( PM, m), (5.b)
where m is the real amount of money in terms of the price of the non-tradable
good, and (3CM/OPM) < O, (3CN/OPM) > 0, and (Ocj/Om) > 0 for j = M, N.
2.3. Equilibrium conditions
There are two equilibrium conditions in this economy. The first one is that the
total production of the non-tradable good must equal its total consumption.
Ex-post relative prices of importables and real money supply must satisfy the
equilibrium condition in the non-tradables market, represented in Eq. (6):
CN( PM, m) = qN( PM, E ( / 3 x U ' ) , E(/3MU' ) , E(U')). (6)
The second equilibrium condition is the one that determines money supply. The
money supply is endogenous in this model economy where the exchange rate is
fixed, and it equals the money stock (m 0) plus the variation in money supply due
to reserve variations (AR). It is represented in Eq, (7) in real terms:
= m 0 + ~AR. (7)
Reserves variation, on the other hand, matches the result of the balance of
payments. If the capital account is assumed exogenous, the current account will
determine reserve variations necessary to maintain the fixed exchange rate.
Without loss of generality, the capital account is assumed to equal zero, so that
reserves variation (AR) equals the current account result:
AR = Px qx( PM, E ( / 3 x U ' ) , E(/3MU' ) , E(U')) - p ~ { C M ( PM, m, ]3 )
--qM(PM, E ( / 3 x U ' ) , E(/3MU' ) , E ( U ' ) ) } . (8)
Uncertainty is introduced in the model through the money stock. All economic

3 An alternative way to accomplish this could be made, for instance, by introducing government
expenditures that would have a positiveeffect on the demand for both goods.
82 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100

pM M~I;

Scheme 1,

agents know the distribution for the possible realizations of money stock, and the
government sets a fixed nominal exchange rate without observing its realization. If
it were not for the uncertainty on m 0, the total money supply could be perfectly
anticipated by the economic agents.
In several inflationary economies with fixed exchange rates, nominal exchange
rate is set by the government targeting the equilibrium real exchange rate.
However, prices are collected with a lag. Therefore, in an inflationary environ-
ment, prices may be different from expected, and the real exchange rate may result
misaligned. The setting modeled here represents such a situation. Hence, the real
exchange rate will be different from its equilibrium value whenever inflation, or,
in the context of our static model, whenever money supply is different from
expected. This means that the variability of the real exchange rate increases with
inflation variability, or with money supply variability in our model.
The equilibrium is represented in Scheme 1. The vertical axis depicts the
relative price of importables, and the horizontal axis represents the real money
supply. Ex-post relative price of importables and the real money supply must
satisfy the equilibrium condition for the non-tradable good market, Eq. (6), which
is represented by the NT schedule, and the equation for the real money supply,
Eqs. (7) and (8), represented by the MS schedule. 4
Money stock affects the position of the curves in the graph. A higher value of
m o, for instance, shifts the MS schedule to the right, yielding a lower equilibrium
value for the relative price of importables.
2.4. The equilibrium real exchange rate and its volatility
The real exchange rate is the ratio of the price of tradables and the price of
non-tradables. In terms of the model presented here, the RER is:
RER = H ( / 3 M, fix), (9)

4 The slope of the MS schedule can be either positive or negative, depending on the interaction of
price and quantity effects of a change in the price of importables on the trade balance. The slope will
be positive as drawn when the quantity effect is larger than the price effect. Either way, the slope of
this schedule does not alter the results, as long as the slope of the MS schedule, when negative, is lower
than the slope of the NT schedule, which is a reasonable assumption.
 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100 83

pM S' MS

m
Scheme 2.
where the derivative of the function //(.) is positive with respect to both
arguments. The volatility of the RER will then depend positively on the volatility
of the relative prices of importable and exported goods, which, in turn, are related
to the variability of the relative nominal exchange rate, as shown in Eq. (2).
From Eq. (2), it is clear that the variability of the relative nominal exchange
rate is positively related to the variability of the prices of exported and importable
goods. This means that the volatility of the relative price of importables and
exportables are positively related, and, as stated in the previous paragraph, both
volatilities have a positive effect on the volatility of the RER.
Furthermore, from the solution of the model it is straightforward to see that the
higher the money supply volatility, the higher will be the volatility of the price of
importables, and, therefore, the higher will also be the volatility of the price of
exported goods. Hence, one of the results of this model is that the volatility of
inflation and the volatility of the RER are positively related.
To assess the effect of changes in the variability of RER on its equilibrium
level, one has to determine the effect of that variability on the equilibrium
conditions, more specifically, on Eqs. (6) and (7). To exemplify, the effect of an
increase in the volatility of the RER will be analyzed. From Eqs. (6) and (7), it is
clear that the effect of the increased volatility of the RER will depend on its
effects on E[/3xU'], E[pMU'] and E[U'], and this effect, in turn, will depend on
the concavity of the functions fix U', PM U', and U'. The intuition is that for a risk
averse individual, the volatility of the price affects the marginal utility from
producing one more unit of the good, even if the expected value of the price
remains unchanged. The effect of volatility will then ultimately depend on the
concavity properties of the utility function. An increase in volatility will decrease
the marginal utility of production if this function is concave in prices. Assuming
that the functions /3xU' and /3MU' are concave in prices, and the function U' is
not concave in prices, 5 an increase in the variability of the RER will decrease
E[/3xU'] and E[~MU'], and will not decrease E[U']. This is a sufficient

5 The function px U' can be concave and U' convex in Px at the same time, because
(32(pxU')/&p 2) = U" px(Qx )2 +2U"Qx, and (,92 U/,gPx)
t 2
= U,it (Qx) 2 . The utility function U(P)
= (pl- r)/(1 _ Y) for T :g 1, for instance, satisfiesthis condition.
84 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100

condition for the increase in the variability of the RER to increase the amount
produced o f the non-tradable good (in Eq. (6)), and to decrease the amount
produced of the exported and importable goods. 6
Scheme 2 represents the changes in the equilibrium conditions caused by an
increase in the variability of the RER. The non-tradables market equilibrium
condition will shift upwards: the increased variability in the tradables sector will
cause an increase in the production o f non-tradable goods, so that, for any given
price of importables, the real money supply has to increase to clear the market.
As for the money supply equilibrium, the increased volatility depresses trade
balance, which lowers the money supply for each relative price of importables.
Hence, the money supply curve shifts left.
The new point that satisfies both equilibrium conditions (point ~ in Scheme 2)
has a higher relative price o f importable goods.
Summing up, the model shows that the volatility of the money supply has a
positive effect on the equilibrium real exchange rate, through its effect on RER
volatility.
In a dynamic context, inflation volatility takes the role of money supply
v o l a t i l i t y - - h i g h e r inflation volatility would affect positively the equilibrium RER.
Furthermore, as Issler (1991) shows, inflation rate and inflation volatility are
positively related. Hence, lower inflation rates would be associated with less RER
volatility, for a given exchange rate regime. Stabilization plans could, thus, cause
an appreciation of the equilibrium RER.

3. Real exchange rate volatility

There is a vast empirical literature on the effects of (nominal and real)
exchange rate variability on trade flows and trade prices in several countries. 7
However, the results of this literature are still inconclusive. While most papers
found a negative significant exchange rate volatility effect, the magnitudes of the
coefficients are in general relatively small. Some recent work has used developing
countries data, but the results are also mixed. 8

6A decrease in E[/SxU'] will depress exports production, but the decrease in E[/3MU']will
increase it. Hence, here we have to further assume that the direct effect of the decrease of the marginal
utility of producing exported goods (E[ fix U']) is greater than the indirect effect of the decrease of the
marginal utility of producing importable goods (E[/3MU']), so that there will be a net decrease of
exported goods production when the variability of the RER increases. The same is true for the
importable goods production.
7 For a good survey on the most recent empirical literature on exchange rate volatility and trade, see
CrtE (1994). Among other important contributions, see Gotur (1985), Kenen and Rodrik (1986),
Caballero and Corbo (1989), and Gagnon (1993).
8 For instance, Coes (1981), Paredes (1989), Caballero and Corbo (1989) and Grobar (1993) found
evidence of a negative relationship between real exchange rate (RER) volatility and trade for some
developing countries, while Paredes (1989), Caballero and Corbo (1989) and Steiner and Wullner
(1994) did not find any significant RER volatility effect on trade for some other developing countries.
 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100 85

On the other hand, the empirical literature usually considered the end of the
Bretton Woods system and the introduction of economic integration areas (Nafta,
the European Union, and Mercosul) as the main sources of change in exchange
rate volatility. In fact, most authors tend to find an increase in exchange rate
volatility after the collapse of the Bretton Woods regime, most notably in
developing countries (see, for example, Edwards, 1989), and that regional eco-
nomic integration reduces the effect of exchange rate variability on trade (see, for
example, Frankel and Wei, 1993).
In this section, we perform two empirical exercises. First, we investigate an
alternative source of changes in RER volatility: the effects of stabilization plans in
high-inflation countries. As our model suggests, for a given exchange rate
indexation regime, inflation variability and RER variability are positively corre-
lated. Changes in the indexation regime or changes from a fixed to a floating
exchange rate system are also other potential sources of RER variability. By
examining the behavior of several RER volatility indexes over the last fifteen
years for Brazil, we try to identify the influence of stabilization plans and inflation
volatility on these variability measures.
The second experiment is the estimation of export supply equations for Brazil
that include RER volatility as one of the explanatory variables. Coes (1981) and
Paredes (1989) performed the same task with Brazilian data, obtaining mixed
results. In the second part of this section, we update their results using more recent
data and new measures of RER volatility.
3.1. Real exchange rate volatility: measures and sources

In order to measure RER volatility, we compute (unconditional) standard

deviations of RER changes within pre-determined periods. This procedure is the
most traditional way of measuring volatility (see, for example, Kenen and Rodrik,
1986; Grobar, 1993; Caballero and Corbo, 1989; among others). The first volatil-
ity measure examined is the standard deviation of 12-month (moving) consecutive
observations of monthly RER changes. It is centered in the middle of each
12-month period, although this does not affect much the regression results of this
section.
This RER volatility measure is computed for Brazil, Argentina and Mexico.
Table l shows RER volatilities and inflation means for the three countries. Mexico
experienced the lowest average level of inflation (3.5% a month) and the lowest
RER volatility measure, in absolute terms. Argentina, on the other hand, experi-
Table 1
Real exchange rate volatility and inflation: means (%)--monthly data, 1980-1995
RER volatility Inflation
Brazil 3.02 15.39
Argentina 9.27 12.69
Mexico 1.72 3.45
86 G.M. Gonzaga, M. C.T. Terra/Journal of Development Economics 54 (1997) 77-100

enced the highest RER volatility measure, both in absolute terms and relative to
the inflation rate.
Fig. 1 displays the evolution of the monthly RER volatility measure for the
three countries. We ignored the large 1989-1991 numbers for Argentina which
were above 60% a month, so as to provide a better visual comparison between the
three countries. Mexico experienced the lowest RER volatility among the three
countries, specially after the events of late 1982. It remained below 2% a month
throughout most of the period. Argentina's RER volatility is the highest of the
group. After the convertibility plan in 1991, however, RER volatility remained
below 0.5% a month, the lowest level reached by any of the three countries in the
period.
The second measure of RER volatility used in this paper is identical to the first,
with the difference that multilateral monthly real exchange rates are used, which
are based on export weights of the seven largest Brazilian trade partners between
1985 and 1995.9 It will be examined more closely later.
By construction, however, these two traditional measures of RER volatility,
based on monthly RER changes, ignore the effects of any within-month RER
movements. Since we believe these, within-month variations are significant in
countries with high (and volatile) inflation, two alternative measures of RER
volatility using Brazilian data are proposed.
The first one, a daily measure, makes use of the daily bilateral nominal
exchange rates. 10 By assuming a constant exponential growth for prices within
each month, we were able to compute daily RERs for a period starting in January
1st of 1980. iJ
The second measure is also based on monthly RER changes. However, we
make use of a weekly consumer price index for the state of S~o Paulo (FIPE-CPI),
which measures monthly price averages ending in each week of each month since
1986 (4 indexes for each month). By computing the number of weekdays for each
4-week period since 1986, we were able to obtain 4-week nominal exchange rate
averages. These series were then deflated by the FIPE-CPI index and inflated by
weekly-interpolated WPI *, generating a weekly series of monthly (4-week) RER
changes. The weekly volatility measure is the standard deviation of these monthly
RER changes within a (moving) 3-month period.

9 We thank Paulo Levy, from the Grupo de Acompanhamento Conjuntural, IPEA/RJ, for providing
this data.
10 Daily nominal exchange rates (e) are published by the Central Bank: sell quotation, domestic
currency/dollar. We thank Dionfsio Dias Cameiro for providing the complete series. As usual in small
economy models, real exchange rates are proxied by RER = (e WPI * )/CPI, where WPI * is the U.S.
wholesale price index and CPI is the Brazilian consumer price index (INPC, also from IBGE). See
Edwards (1989) for a discussion on alternative measures of real exchange rates.
11 The assumption of a constant exponential price growth within each month does not seem damaging
to our results, since ex-ante RER expectations should be based on a hypothesis like this, as other
distributions are unknown.
 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100 87

i i
"o
L

0
~0

,ca

0 I ........................

iN

0
°~
~a

°~
>
a

oO :.J
¢n

f-
t-
O
n~
UJ
r~
I
f-
"5
f- >
0

i
i m
g
i m
,ca

n
¢0
0
>

i
\
o m u~ o-,"
 Real E x c h a n g e Rate - Brazil D a i l y - R e a i s of 6 / 1 5 / 9 5 ~o
2.2.

2.0

o~

1.8

1.6

1.4

~3
1.2

1.0

",4
L
0,8 t i i i i i i L i i i i i i i i i i
De¢-79 C~t~O Aug.1 Jun~2 Ap¢-83 J=n-84 Nov-84 Sep-85 Jul-86 Apt-87 Fete88 Oec~8 Oct-Sg Jul-go May-g1 Mar-~2 J=n-g3 Oct-l~3 Aug-94 Jun*g5

Fig. 2. Real exchange rate--Brazil; daily--reais of 6/15/95.

 Real E x c h a n g e Rate - Brazil Daily Variation (%)
20

C~

15

10

0 -~'v- . 4~

~q
'~~ ,,~-,~,~~~~IT!~1 I

-5
Dec.79 Oct-80 Aug.1 Jun~82 Apt-83 Jin~4 Nc~'-84 Sep85 Jul~ Apr.7 Fet:-88 De¢-88 Oct.9 Jul-90 May-91 Mar-g2 Jim-g3 Oct-iT3 Aug-g4 Jun-95

Fig. 3. Real exchange rate--Brazil: daily variation (%).

90 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100

In order to illustrate the gains in using the first of these two alternative series,
Figs. 2 and 3 display daily Brazilian RER levels and changes, respectively, from
the beginning of 1980 to the end of June 1995. A very volatile RER picture
emerges from these figures. Large one-day swings were observed in the 30%
maxi-devaluation episode of February 1983, and in mid-devaluations that preceded
most stabilization plans of the 1980s. Months of steep inflation acceleration/decel-
eration not only brought changes in the RER level but also caused an increase in
(within-month) volatility, which is not perceived in monthly data.
Fig. 4 shows the behavior of our daily RER volatility measure. Note that most
peaks in the volatility measure are related to large devaluations, that were made
either to deal with the external debt crisis, as in February of 1983, or preceding
stabilization plans. 12 The remaining peaks reflect large appreciations that fol-
lowed both the Collor and the Real plans.
If we do not consider those peaks, at least six volatility patterns can be
identified.
The first one, observed in the period running from 1980 to February 1983, is
characterized by a relatively high volatility (around 0.7% a day), which resulted
from a loose crawling peg regime without a fixed-period indexation rule.
The second period, observed from February 1983 to mid-1985, was one of a
daily exchange rate indexation. However, inflation acceleration (from 100% a year
to 200% a year) compensated the indexation effect, resulting in a RER volatility of
around 0.7% a month, similar to that observed in the first period.
The third period, from mid-1985 to the end of 1988, was characterized by a low
daily RER volatility (around 0.2% a day), which resulted from two price (and
exchange rate) freezing attempts and a policy orientation of keeping the RER
unchanged, despite the inflation acceleration at the end of the period.
The fourth period runs from 1990 to the middle of 1991, period in which the
Central Bank did not follow any indexation (parity) rule, letting the exchange rate
float under unspecified thresholds, which characterized what is usually called a
'dirty floating' regime. The result was a very high RER volatility (above 1% a
day).
A fifth period, from mid-1991 to mid-1994, illustrates the return to a daily
indexation regime and was characterized by some attempts to point to a more
devalued currency. Continuous inflation acceleration in that period, together with a
lag on current price observation, resulted in a continuous increase in the daily RER
volatility measure. The magnitudes, however, were not large, compared to other
periods, averaging around 0.3% a day.
The sixth pattern was the one introduced by the floating exchange regime that
followed the Real plan. After a brief period of relatively high volatility, which was

lz We stress the point that these peaks should be kept as part of the volatilityindex, since in all these
periods, forward-looking agents were uncertain about future levels of the RER, anticipating the
possibility of large devaluationsor price freezing.
 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100 91

&

t_

e"

t_

0
!

t- x_.____
O
°m
I
°~

.y,
c~

(1)

N r~
/

|
==

e- ,¢

e'- Y

iii
;>
.4
m
, m
¢=

°. i

mm

O
>
i
I
O ~
 92 G.M. Gonzaga, M.C.T. Terra/ Journal of Development Economics 54 (1997) 77-100

~=._.
J~
-~

(~'.~

ILl - ~

rr

z ~
m

>0
 '(%) +([qluouJ--UO!II~!A~p"pls :l!zl~all--s~u'ett3 ~]~J ~uuq3xo leoJ :£1![!11~[oA "9 "~!d
.~B4,eFI HPJelN ¢6-uet" ~6-AON 06",des 6g-lnr gg-,~alPl Zg-JelPl ~l~-IJel" t~'^ON £g-cles ~:e-lnc &g-XelN 0e-Je~l

I
P~

L~

0~

peJa;ueO 'q;uolN ~ I. - Xlqiuolq 'leJelellg peJelueo 'q)uof,q E !. -/q q ~uolP,I 'leJelel!;Inl/~ 1

(%)/~lq~UOlN - UOp, elAe(] P~S

I!ZeJe - se6ueqo ele~l e6ueqox3 lee~l :~!lP, elOA
94 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100

Table 2
RER volatility measures--summary statistics
Volatility Obs Mean (%) Std. error (%)
Daily 187 0.653 0.612
Weekly 109 2.839 2.484
Monthly-Bilateral 170 3.161 1.974
Monthly-Multilateral 116 3.835 1.780

caused by the large appreciation of the Real, inflation stabilization at low levels
(below 2% a month), together with the introduction of exchange rate bands in
March 1995 helped to decrease the RER variability. However, RER volatility is
still larger than the one observed in the fourth period, period in which the average
inflation rate was around 25% a month.
Of the four RER volatility measures used in this paper, we believe that the
daily volatility measure gives the best picture in terms of picking up the effects of
inflation acceleration/deceleration and of changes in the indexation regime for the
last fifteen years, as described in the previous paragraphs, despite the arbitrary
assumptions that were made in its construction. 13 The question, then, is to
examine how all these volatility measures are related to each other.
Figs. 5 and 6 display the two pairs of RER volatility measures. Fig. 5 pictures
the monthly averages of the daily and weekly volatilities, both based on three-
month periods (centered in the middle of each quarter). 14 Note that, despite their
different methodologies, the two series follow similar paths. The main differences
are that they are measured in different scales and the weekly series is more
sensitive to the large price swings observed in the period 1990-199l.
Fig. 6 shows the bilateral and multilateral volatility indexes, both based on
12-month periods (centered in the middle of each period). They also follow similar
paths, with the most notable exception being the period from October 1991 to June
1994. Historically, the Brazilian government has been more concerned with
indexing the bilateral (domestic currency/US$) rate, letting the multilateral index
flow, which makes the multilateral index more volatile whenever the dollar
fluctuates relative to other currencies.
Table 2 shows the means of each RER volatility measure. The daily volatility
index averaged 0.65% a day between 1980:1 and 1995:6. Remember that these
averages are taken over weekdays only, excluding weekends, holidays, and bank
holidays. The magnitude of the daily volatility measure is, thus, relatively high, of
about the same size as the daily rate of inflation.
The multilateral real exchange rate volatility mean, on the other hand, is the

13 The weekly measure, although technically superior to the daily measure, begins in 1986, missing
the first two periods described above.
14 Monthly and quarterly averages of the RER volatilities are used in all monthly and quarterly
regressions of this section, respectively.
 G.M. Gonzaga, M.C.T. Terra / Journal of Development Economics 54 (1997) 77-100 95

Table 3
Correlation matrix: RER volatility measures. Monthly data: 1986:4-1995:2
Volatility Daily W e e k l y Monthly-Bilateral Monthly-Multilateral
Daily 1.000 0.742 0.416 0.398
Weekly 0.742 1.000 0.477 0.502
Monthly-Bilateral 0.416 0.477 1.000 0.671
Monthly-Multilateral 0.398 0.502 0.671 1.000

highest among the monthly volatility indexes, which is a symptom that the
government, in most of the period studied, looked much more closely at the
bilateral domestic currency/dollar rate. The monthly average of the weekly index
is the most volatile, which is in part due to the fact that it is based on 3-month
periods, contrary to the two other monthly measures.
All four RER volatility averages are much larger than those observed for
developed countries. Kenen and Rodrik (1986), for instance, reported (24-month)
monthly volatility means for eleven OECD countries from 1975 to 1984. Eight of
the eleven countries have experienced RER volatilities below 1.5%, while the
highest mean value, Sweden's, was 2.7%.
Table 3 below displays the correlations between monthly observations of our
four measures of volatility (daily, weekly, monthly-bilateral and monthly-multi-
lateral) over comparable periods (1986:4 to 1995:2). The results confirm that the
series are positively correlated. The daily and the weekly volatilities are the most
highly correlated (0.74).
One explanation for such high RER volatilities observed in Brazil, suggested by
our model and the experiences of Mexico and Argentina, is high inflation
volatility. We test this proposition by running OLS regressions, correcting for
first-order serial correlation using the iterative Cochrane-Orcutt technique, of our
RER volatility measures on compatible inflation volatility measures. 15 The results
are displayed in Table 4 below. They confirm that inflation volatility should be
considered as one of the main sources of RER volatility changes in Brazil. The
inflation volatility coefficients are statistically significant and positive in all
regressions. The large adjusted R2s indicate that inflation volatility explains most
of the variation in RER volatility in Brazil over the last fifteen years.

3.2. Real exchange rate uolatility and export f l o w s

In this sub-section, we estimate export supply equations for Brazil that include
RER volatility as one of the explanatory variables. The four measures of volatility

~5The inflation volatility measure in each regression was computed as follows: (i) monthly averages
of the 3-month standard deviation of daily inflation rates, used in the daily RER regression; (ii) the
standard deviation of 3-month changes in the FIPE-CPI index, used in the weekly volatility regression:
(iii) the 12-month standard deviation of monthly inflation in the monthly volatility regressions.
96 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100

Table 4
Regression results: RER volatility on inflation volatility a
RER volatility Constant Inflation volatility Adjusted R 2 Period
Daily 0.54 b (4.19) 0.62 b (3.51) 0.66 1980:4-1995:3
Weekly 2.46 b (3.00) 0.12 b (5.74) 0.95 1986:3-1995:3
Monthly-Bilateral 2.76 b (2.97) 0.06 b (2.45) 0.93 1980:6-1995:1
Monthly-Multilateral 3.31 b (3.01) 0.04 c (1.77) 0.93 1985:6-1995:1

at-statistics in parentheses.
bDenotes significance at the 5% level.
cDenotes significance at the 10% level.

described earlier are used in alternative specifications o f the export supply

function. W e test w h e t h e r their coefficients are n e g a t i v e and significant as our
m o d e l predicts.
All specifications used here take the f o r m o f a typical export supply equation:
X = f ( R E R , Y, V O L ) , w h e r e X denotes exports; R E R is the real e x c h a n g e rate
defined as ( e W P I * ) / C P I ; 16 y is a m e a s u r e o f d o m e s t i c activity ( G D P in the
quarterly regressions; industrial production, IP, in the m o n t h l y versions); and V O L
is the R E R volatility measure. W e take logs o f all variables with the e x c e p t i o n o f
the volatility measures. W e also include a constant, a trend and a set o f seasonal
d u m m i e s in all regressions.
W e run two sets o f regressions, with different m e a s u r e s o f exports: export
v o l u m e and e x p o r t s / G D P . T h e export v o l u m e index was taken f r o m Pinheiro
(1993). A l t h o u g h this series ends in June o f 1992, it is still the m o s t c o m p l e t e and
accurate m e a s u r e o f export v o l u m e s available for Brazil. 17 To m a k e use o f the
m o s t recent observations, w e use an alternative specification with exports o v e r
G D P (both m e a s u r e d in dollars) as our d e p e n d e n t variable, as in Grobar (1993). 18
T h e o r y predicts a n e g a t i v e c o e f f i c i e n t for both the d o m e s t i c activity and the
R E R volatility variables, and a positive coefficient for the R E R series. If the
country is small in the w o r l d market, O L S coefficients are consistent. 19 H o w e v e r ,
one can confidently reject absence o f serial correlation for all regression residuals,

16 For consistency, the monthly multilateral RER series is used when testing the monthly-multilateral
volatility measure. For the other three regressions of each set, the monthly bilateral RER series is used.
17 Using a very disaggregated data on exports, Pinheiro (1993) constructs a Fischer index of export
prices and volumes. For details on the adequacy of their methodology, see Pinheiro and Motta (1991).
is We also tested using manufactured exports (volumes and proportional to GNP) as the dependent
variable without much change with respect to the results presented here.
19 There is some evidence that the small country hypothesis is not very suitable to the Brazilian case
(see, for example, Portugal, 1993, for a survey of the empirical literature on exports determination in
Brazil and for some estimates). Instrumental variables were used to deal with real exchange rate
endogeneity. The instruments chosen were: a constant, seasonal dummies, a trend, the log of relative
prices (U.S. wholesale price index divided by Brazil's CPI), the log of GNP, and the RER volatility
measure. Instrumental variables estimation, however, did not improve on the results presented here.
Table 5
Export supply equation--quarterly results; dependent variable: exports volume ~
RER volatility measure RER Real GDP RER volatility Trend Adjusted R 2 Period
Daily 0.29 b (2.02) -0.70 (- 1.32) -0.44 (-0.13) 0.02 b (4.02) 0.73 1980:2-1992:2
Weekly 0.69 (1.63) - 1.22 (- 1.39) - 1.56 ( - 1.01) - 0.66 1986:3-1992:2
Monthly-Bilateral 0.40 b (2.22) -0.94 (- 1.66) - 1.05 (-0.60) 0.02 b (3.63) 0.68 1981:4-1992:2
Monthly-Multilateral 0.48 (0.92) - 0.94 (- 0.92) 2.11 (0.50) - 0.44 1985:4-1992:2 N

at-statistics in parentheses.
b . .
Denotes slgmficance at the 5% level.

e~

Table 6
Export supply equation--quarterly results; dependent variable: e x p o r t s / G D P a
RER volatility measure RER Real GDP RER volatility Trend Adjusted R 2 Period
Daily 1.00 b (6.34) - 1.98 b ( - 3 . 9 0 ) 0.57 (0.179) 0.02 b (4.12) 0.78 1980:2-1995:1
Weekly 1.21 b (3.68) -2.40 b (-3.66) 1.69 ( - 1.31) 0.02 b (2.25) 0.72 1986:3-1995:1
Monthly-Bilateral 1.13 b (6.93) - 2.23 b ( - 4 . 7 0 ) -0.10 (-0.08) 0.02 b (4.83) 0.80 1981:2-1995:1
Monthly-Multilateral 0.69 c (1.88) -2.02 b (-2.37) - 1.22 (-0.44) - 0.52 1985:4-1994:3

at-statistics in parentheses. 4~

bDenotes significance at the 5% level.

CDenotes significance at the 10% level.
,,q
98 G.M. Gonzaga, M.C.T. Terra/Journal of Development Economics 54 (1997) 77-100

according to Ljung-Box statistics, not reported here. First-order serial correlation

was dealt with by using the Cochrane-Orcutt method, which removed any
significant serial correlation in the residuals. To save space, only the results after
serial correlation correction and after any eventual removal of trend and seasonal
variables that were not significant in pre-testing regressions are presented. 20
Monthly regressions do not improve over the results presented below, with the
disadvantage of showing a positive industrial production coefficient in most
cases. 21
Table 5 shows the results for the quarterly exports volume regressions. Export
supply price elasticities are positive as expected in all four regressions. However,
they were not large, being significantly different from zero (at 5%) in only two
cases. The domestic activity term, real GDP, also has the expected (negative) sign,
although its coefficient is not significant at 10% in any of the regressions. The
RER volatility coefficients have the expected (negative) sign in the three regres-
sions that use the bilateral RER volatilities. However, RER volatility coefficients
are not significant in any of the four regressions for reasonable statistical levels of
significance.
Although not statistically significant, the coefficients of RER volatility (in
absolute value) are positively correlated with the export supply price elasticities.
This result was also obtained by Paredes (1989), and is intuitively appealing: the
more important the RER level series becomes, the more important should be its
variability.
The results for the quarterly exports/GDP regressions are much more in
conformity with theory. Table 6 summarizes them. Positive and significant export
supply price elasticities were obtained in the four regressions. Negative and
significant GDP coefficients were also found for the four regressions. Although
negative RER volatility coefficients were found in three cases, they are not
statistically different from zero at conventional levels of significance.
The most significant (negative) RER volatility coefficient, obtained in the
weekly volatility measure, is significant at the 20% level. Its implied elasticity is
- 0 . 0 5 . Gagnon (1993), based on a similar model, simulates the decision rules of
exporters for different choices of parameters. In the case that maximizes the effect
of volatility on exports, he found that an increase of 300% in his volatility measure
(from 0.02 to 0.08) would cause a decrease of exports of 1.2%, which corresponds
to an elasticity of -0.004. Not only our implied elasticity is much larger, but also

20 Seasonal variables were always significant. Trend was removed from the regression in three
occasions.
2t Dynamic specifications that included lagged endogenous variables as regressors, as used by
Caballero and Corbo (1989) and Gagnon (1993), were also examined. In general, they produced
statistically insignificant coefficients for the RER and domestic activity variables. Persistence coeffi-
cients were not large, which may indicate that costs of adjustment are relatively low in Brazil.
 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100 99

RER volatility observed for Brazil presented changes above 300% over the last
fifteen years, as shown in Figs. 5 and 6. 22

4. Conclusions and directions for future research

This paper is composed of two parts. The theoretical part studies the effect of
real exchange rate (RER) volatility on trade using a general equilibrium frame-
work. Substitution across sectors is considered, and the model studies the effect of
RER volatility on the equilibrium real exchange rate through its effect on tradables
production decisions. The volatility of the RER is derived endogenously, and is
caused originally by a demand shock. The model shows that inflation volatility has
a positive effect on RER volatility, which, in turn, affects positively the equilib-
rium RER.
The empirical part consists of two experiments. In the first one, we examine the
behavior of several RER volatility indexes over the last fifteen years for Brazil,
identifying the influence of stabilization plans and inflation volatility. We show
that, in fact, inflation volatility explains most of the variation in RER volatility in
Brazil over the last fifteen years.
The second experiment performs the estimation of export supply equations for
Brazil that include RER volatility as one of the explanatory variables. For most
specifications, we found that the RER volatility coefficient is negative, although
not significantly different from zero. The implied elasticity for the most significant
RER volatility coefficient is - 0 . 0 5 , which is well above the highest elasticity in
the simulation exercise performed by Gagnon (1993).
According to our theoretical model, RER volatility affects positively the
equilibrium real exchange rate level. In a future paper, we intend to test empiri-
cally this result by estimating the equilibrium RER. A traditional method of
estimating equilibrium RER has been applied to other developing countries' data
by Edwards (1994) and Elbadawi (1994), among others, usually using real
variables such as terms of trade and fiscal and monetary variables as the long-run
determinants of the RER. We intend to apply this method to Brazil, including RER
volatility as one of the variables that possibly cointegrate with the RER.

Acknowledgements

We are grateful to Carlos Winograd, Afonso Bevilaqua, Marcos Lisboa, Ruy

Ribeiro, Maurlcio C~rdenas, Carlos Felipe Jaramillo, participants at the 8th
Annual Inter-American Seminar on Economics in Bogotfi, Colombia, and at
seminars at PUC-Rio, University of Brasflia, EPGE-FGV, and at the XIV Latin

22The changes in RER volatilityfor Argentina were even larger, as shown in Fig. 1,
100 G.M. Gonzaga, M.C.T. Terra~Journal of Development Economics 54 (1997) 77-100

American Meeting of the Econometric Society for useful comments and sugges-
tions. We thank Dionlsio Dias Carneiro, Paulo Levy, Armando Castellar, Mfircia
Le6n and Leda Hahn for helping with data. We also thank Cristiana Vidigal Lopes
for research assistance. Any remaining errors are our responsibility.

References

Caballero, R., Corbo, V., 1989. The effect of real exchange rate uncertainty on exports: empirical
evidence. World Bank Economic Review 3, 263-278.
Coes, D.V., 1981. The crawling peg and exchange rate uncertainty. In: Williamson, J. (Ed.), Exchange
Rate Rules. St. Martins, New York.
C6t& A., 1994. Exchange rate volatility and trade: a survey. Working Paper 1994-1995, Bank of
Canada.
Dixit, A., 1989. Entry and exit decision under uncertainty. J. Political Economy 97, 620-638.
Edwards, S., 1989. Real Exchange Rates, Devaluation, and Adjustment: Exchange Rate Policy in
Developing Countries. MIT Press.
Edwards, S., 1994. Real and monetary determinants of real exchange rate behavior: theory and
evidence from developing countries. In: Williamson, J. (Ed.), Estimating Equilibrium Real Ex-
change Rates. Institute for International Economics, Washington, DC.
Elbadawi, I., 1994. Estimating long run equilibrium real exchange rate. In: Williamson, J. (Ed.),
Estimating Equilibrium Real Exchange Rates. Institute for International Economics, Washington,
DC.
Frankel, J., Wei, S.-J., 1993. Trade blocs and currency blocs. NBER Working Paper No. 4335, April.
Gagnon, J., 1993. Exchange rate variability and the level of international trade. J. Int. Economics 34,
269-340.
Gotur, P., 1985. Effects of exchange rate volatility on trade: some further evidence. Int. Monetary Fund
Staff Pap. 32, 475-512.
Grobar, L., 1993. The effect of real exchange rate uncertainty on LDC manufactured exports. J. Dev.
Economics 41,367-376.
Issler, J.V., 1991. Inflation level and uncertainty: evidence using Brazilian data. Revista Brasileira de
Economia 45, 472-482.
Kenen, P., Rodrik, D., 1986. Measuring and analyzing the effects of short-term volatility in real
exchange rates. Rev. Economics Statistics 68, 311-315.
Paredes, C., 1989. Exchange rate regimes, the real exchange rate and export performance in Latin
America. Brookings Discussion Papers in International Economics No. 77.
Pinheiro, A.C., 1993. Uma an~ilise desagregada do comdrcio exterior Brasileiro no perfodo 1974/1992.
Texto para DiscussSo, No. 306. IPEA, Rio de Janeiro, Brazil.
Pinheiro, A.C., Motta, R.S., 1991. Indices de exportaqao para o Brasil 1974/1988. Pesquisa e
Planejamento EconSmico 21, 253-285.
Portugal, M., 1993. Brazilian foreign trade: fixed and time varying parameter models. Ph.D. Disserta-
tion. University of Warwick, UK.
Steiner, R., Wullner, A., 1994. Efecto de la volatilidad de la tasa de cambio en las exportaciones no
tradicionales. Coyuntura Economica 24, 139-153.