WPS5590
Policy Research Working Paper 5590
Over the Hedge
Exchange Rate Volatility, Commodity Price Correlations,
and the Structure of Trade
Claudio Raddatz
The World Bank
Development Research Group
Macroeconomics and Growth Team
March 2011
Policy Research Working Paper 5590
Abstract
A long empirical literature has examined the idea that, in those whose international price is negatively correlated
the absence of hedging mechanisms, currency risk should with the nominal exchange rate of the country where they
have an adverse effect on the export volumes of risk operate--should be relatively benefited in environments
averse exporters. But there are no clear conclusions from of high exchange rate volatility, and capture a larger share
this literature, and the current consensus seems to be that of the country's export basket. This hypothesis is tested
there is at most a weak negative effect of exchange rate using detailed data on the composition of trade of 132
volatility on aggregate trade flows. However, most of this countries at 4-digit SITC level. The results show that the
literature examines the impact of exchange rate volatility commodities that offer natural hedge capture a larger
on aggregate trade flows, implicitly assuming a uniform share of a country's export basket when the exchange
impact of this volatility on exporters across sectors. This rate is volatile, but there is only weak evidence that the
paper explots the fact that, if exchange rate volatility is availability of financial derivatives to hedge currency risk
detrimental for trade, firms exporting goods that offer reduces the importance of a sector's natural hedge.
a natural hedge against exchange rate fluctuations--i.e.
This paper is a product of the Macroeconomics and Growth Team, Development Research Group. It is part of a larger
effort by the World Bank to provide open access to its research and make a contribution to development policy discussions
around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author
may be contacted at craddatz@worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
Over the Hedge: Exchange Rate Volatility, Commodity Price
Correlations, and the Structure of Trade
Claudio Raddatz
craddatz@worldbank.org. I am grateful to Kevin Cowan, Luis Serven, Andrei Levchenko, and Pierre Olivier
Gourinchas for helpful discussions. I also thank Yuki Masujima and Alfonso Astudillo for excellent research assistance,
and to the Japanese Trust Fund, the Knowledge for Change Program, and the Oce of the Chief Economist of Latin
America and the Caribbean for Financial Support. The oppinions expressed in this paper are the author's only and
do not necessarily represent those of the World Bank, its Executive Directors, or the countries they represent.
1 Introduction
A long literature has studied the impact of exchange rate volatility and exchange rate regimes on
trade ows.
1 The initial motivation behind this literature was the idea that, in absence of access to
hedging mechanisms, risk averse exporters would be adversely aected by currency risk and exports
would be reduced (Clark (1973)). This simple mechanism rests on a series of assumptions and has
been challenged both theoretically and empirically. Overall, the current consensus seems to be that
there is at most a weak negative eect of exchange rate volatility on aggregate trade ows.
A shortcoming with most of the literature is the focus on the impact of exchange rate volatility
on the aggregate volume of trade. From a theoretical perspective, what really matters for a trading
rm is the volatility of its prots. How this volatility relates to the volatility of exchange rates
depends on how this rate correlates with a rm's price and costs. In other words, what really
matters for a rm is the volatility of its individual real exchange rate. For instance, Clark (1973)
highlights the importance of the use of imported inputs as a natural hedge against uctuations
on the exchange rate. More recent papers also discuss the role of currency derivatives in reducing
exposure to exchange rate risk (Wei (1999)).
Unless the prices and costs of all rms in a country are equally correlated with the nominal
exchange rate, exchange rate uctuations may have consequences for the composition of trade,
even without having a large impact on the aggregate trade volume. This is consistent with recent
empirical evidence showing that the impact of exchange rate volatility varies across broad groups
of sectors. For instance, Broda and Romalis (2003),Clark et al. (2004), and Byrne et al. (2008)
show that rms producing homogeneous goods tend to be relatively less aected by exchange rate
volatility than rms producing dierentiated products.
This paper studies the impact of exchange rate volatility on the structure of trade using detailed
sectoral trade data. It starts from the simple premise that nominal exchange rate volatility should
aect relatively less the trade of those rms and industries whose prots do not uctuate much as
a result of exchange rate movements. In particular, industries exporting goods whose international
price co-moves negatively with nominal exchange rate uctuations, which I refer to as industries with
a natural hedge against exchange rate volatility, should be relatively less aected. Firms exporting
goods with this characteristic are naturally protected against exchange rate risk. Therefore, if this
risk matters they should be relatively less aected by exchange rate volatility.
The possibility that this natural hedge oers some protection to rms is supported by existing
evidence that rms whose income is positively correlated with the exchange rate, such as those in
1
McKenzie (1999) surveys the traditional literature that used time-series methods to estimate the impact of a
country's exchange rate volatility on its total trade and exports. The most recent literature has exploited detailed
panel data on bilateral trade ows and used gravity equations to estimate the impact of exchange rate volatility,
exchange rate regimes, and currency unions on bilateral or multilateral trade. See, for instance, Frankel and Wei
(1993), Frankel and Rose (2002), Rose (2000), Rose and Van Wincoop (2001), Klein and Shambaugh (2006), and
Tenreyro (2007), among others.Clark et al. (2004) provides a comprehensive recent survey.
1
more tradable sectors, have a higher fraction of foreign-currency denominated liabilities (Bleakley
and Cowan (2008)). If this mechanism is empirically relevant, the structure of trade should naturally
shift towards industries producing goods that oer a natural hedge against exchange rate uctuations
in countries with high exchange rate volatility, or as a result of sustained changes in exchange rate
volatility such as those resulting from exchange rate liberalization. This is the hypothesis that is
formally tested in this paper using data on the composition of exports of 129 countries across 749
commodities during 1984-2000, and the correlation of these commodities' global-unit-values and
each of these countries' nominal exchange rates.
The idea that the price of some goods may be correlated with uctuations in the nominal
exchange rate has been present in the recent literature on commodity currencies (Cashin et al.
(2004), Chen and Rogo (2003)), which are currencies whose value uctuates with the average price
of the commodities exported by the country. Typical examples of commodity currencies are New
Zealand and Australia. Because of this correlation, the price in local currency of these commodities
and of any other sector whose price is correlated with them will be stable relative to other products.
Thus, if uctuations in the local currency price of goods matter for trade and resource allocation,
rms in those sectors should become relatively more important after a sustained increase in exchange
rate risk.
The importance of a commodity's natural hedge emphasized in this paper as a determinant
of trade patterns may depend on the availability of other forms of hedging. For this reason, the
paper controls for the availability of currency derivatives to assess the quantitative importance of
the mechanism we emphasize. Moreover, in as much income uncertainty will impact nancially
constrained rms the most, one would expect that un-hedged exchange rate uncertainty will have
a larger eect in countries with underdeveloped nancial systems.
The results indicate that exchange rate volatility matters relatively more for those sectors that
lack a natural hedge and are more exposed to it. The natural hedge against exchange rate volatility
provided by a negative correlation between a commodity's international price and a country's nom-
inal exchange rate aects its export patterns, even after controlling for other standard determinants
of export composition, such as the factor content of trade and the export patterns of countries with
similar income levels. An increase in exchange rate volatility from the 10th to the 90th percentile,
corresponding to 57 percentage points, raises by 37 percent the ratio of the export share of a good
at the 10th percentile of correlation (correlation of -0.55) relative to a good at the 90th percentile
(positive correlation of 0.28). This dierence is about 15 percent of the typical within-country dif-
ference in export shares across commodities. Larger dierences in correlations are associated with
proportionally larger changes in relative shares resulting from the same increase in exchange rate
volatility. Without being a rst order determinant of the structure of trade, this mechanism can ex-
plain a non-trivial portion of the observed dierences in relative shares. Furthermore, instrumental
variables specications controlling for endogeneity suggest that the actual coecient is likely one
2
order of magnitude larger, which would result in more meaningful quantitative impacts.
Similar results are obtained for the bilateral trade patterns across countries. For a given exporter-
importer pair, an increase in bilateral exchange rate volatility raises the relative export share of a
commodity with a large negative correlation. The magnitude of the eects is, however, quantitatively
smaller, reaching only 10 percent of the typical dierences in relative shares within a given country
pair, which is not surprising considering the larger variability of bilateral trade shares.
There is also a positive relation between the natural hedge provided by a commodity's price
movements and the growth rate of its total and bilateral export shares during the sample period.
This relation can also explain about 10 percent of the typical dierences in export shares growth.
More importantly, when using the specication in growth rates to compute the long run impact of
changes in natural hedge, the estimated coecient is twice as large as when directly estimating the
specication in levels, and accounts for a similarly larger share of dierences in relative shares. In
addition, similar results are obtained when comparing the role of natural hedging across exchange
rate regimes instead of using an ex-post measure of exchange rate volatility. A commodity's natural
hedge is related to its importance in a country's commodity basket mainly during exible exchange
rate regime periods. A series of robustness tests show that these results are not crucially driven by
specic measures, countries, or commodities.
Somewhat surprisingly, the results show that the importance of a natural hedge against exchange-
rate uctuations is not clearly related to the availability of formal hedging instruments. Only when
looking at the growth of export shares I nd that a well developed market for foreign exchange
rate derivatives weakens the relation between a commodity price's correlation with the country's
exchange rate and its importance in the country's export basket in high exchange rate volatility
environments. When looking directly at the composition of trade (in levels), the relation between
the development of foreign exchange rate derivatives and the importance of natural share reverses.
This is consistent with evidence by Wei (1999) that the availability of hedging instruments does
not aect the relation between exchange rate volatility and trade. Broader measures of the de-
velopment of nancial markets do not seem to have such an eect on the importance of a natural
hedge for export composition, suggesting that the relevant dimension of nancial development for
this mechanism, if any, is the widespread availability of exchange rate derivatives.
This paper relates to several strands of literature. First and most directly, it relates to the
literature on the real eects of exchange rate volatility and exchange rate regimes (see, Frankel
and Wei (1993), Frankel and Rose (2002), Rose (2000), Rose and Van Wincoop (2001), Clark et
al. (2004),Klein and Shambaugh (2006), and Tenreyro (2007), among others; Clark et al. (2004)
and McKenzie (1999) also provide comprehensive surveys of the pre- and post- gravity equation
literature, respectively). It contributes to this literature by emphasizing the heterogeneity of the
impact of exchange rate risk across industries and by focusing on its consequences for the composition
3
of trade.
2 Second, it also contributes to the literature on commodity currencies (Cashin et al. (2004),
Chen and Rogo (2003)) by emphasizing the implications that the comovement of nominal exchange
rates with the price of goods has in terms of hedging of currency risk. Finally, this paper is also
related to the empirical literature on the structure of trade that has explored the role of technology
dierences and factor endowments on the composition of trade (Leamer (1980), Romalis (2004),
Fitzgerald and Hallak (2004), among others) by showing that factors besides these dierences can
inuence the structure of exports.
From a policy perspective, the ndings of this paper remind us of the endogeneity of the compo-
sition of exports, and show that it is aected by factors beyond standard relative factor abundance
and that are related to the ability of rms to cope with the risks associated with exporting. If the
composition of exports matters, as conjectured by Hausmann et al. (2007), addressing some of these
nancial market imperfections may be a way to move towards a rst best export composition with-
out engaging in industrial policy. However, the main caveat is that the results do not clearly show
that the development of exchange rate derivatives weakens the relation between a sector's natural
hedge and export composition. The results also highlight a potential channel between exchange
rate liberalization and macroeconomic volatility. It has been shown that oating exchange rate
regimes insulate output from terms of trade uctuations (Broda (2004), Edwards and Levy Yeyati
(2005)). However, since the results show that exchange rate exibility increases export concentra-
tion by shifting activity toward sectors with a natural hedge, the stabilizing benets of oating the
exchange rate may be reduced because of a simple diversication argument.
The rest of the paper is structured as follows. Section 2 describes the methodological approach
used to test the hypotheses that the natural hedge provided by some goods against exchange rate
uctuations increases the exports of sectors producing those goods vis-a-vis other sectors in countries
with high exchange rate volatility, and that the extent of this advantage depends on the availability
of formal sources of hedge for exchange rate risk. Section 3 describes the data sources and variables
used. Section 4 presents the results of the paper and their robustness. Section 5 concludes.
2 Methodology
2.1 The composition of trade and exchange rate volatility
The empirical equations estimated in this paper are derived from a simplied version of Chaney
(2008)'s version of Melitz (2003) model of trade with multiple goods and heterogeneous rms. This
2
Within this literature, this paper is closely related to Byrne et al. (2008) who nd a negative impact of exchange
rate volatility in trade using data for 22 industries and six countries trading with the US. However, Byrne et al.
(2008)does not consider relative sectoral prices to construct measures of sectoral exchange rate volatility because of
the lack of high frequency data in the price deators considered. As it will be discussed later, this paper makes
assumptions about the covariances of prices at dierent frequencies to construct measures of the components of
sectoral level exchange rate volatility.
4
model predicts that total exports from country i to country j of commodity
h
h, Ti,j , are given by
-1 -1
h Lj h j
Ti,j = h h
3 × × Li × h
L ij
where Li and Lj L is total world size, j is a measure of the
represent the size of each country,
h captures bilateral trade resistance between the two countries
importing country's remoteness, ij
and is commodity specic (this parameter will be later related to the volatility of exchange rate),
and 3 , h , and are technological parameters capturing the dispersion of productivity in a sector
and the elasticity of substitution between goods in a sector, among other characteristics.
This expression can be integrated across several dimensions to determine a country's exports of
a given commodity to the world, the world exports of a given commodity, and the total exports of a
commodity to a given country. These aggregates will be used to characterize a country's composition
of trade. For instance, a country's total exports of commodity
h
h, Ti,W are
h
Ti,W = h h × Li × ih ,
3
where ih is a weighted average of bilateral trade resistance parameters
h
ij . Similarly, country i's
total exports are
Ti,W = Li × i ,
with i again a weighted sum of the commodity specic trade resistance parameters ih . This implies
that the share of commodity h h
in a country's total exports, Xi,W , will be
h
h
Ti,W h h × ih
3
Xi,W = = .
Ti,W i
This expression can be used to obtain the following equation for the log total export shares
h
log Xi,W = i + h + log(ih ). (1)
Proceeding in a similar fashion, one can derive an equation for the (log) share of commodity h
in total exports from country i to j (bilateral export shares),
h
Tij
h h
log(Xi,j ) = log = jh + ij + log(ij ). (2)
Tij
Transforming the expressions above into estimable equations requires specifying the form of the
trade resistance terms at dierent levels of aggregation. This paper follows Tenreyro (2007) and
5
assumes that it is a parametric function of the volatility of the sector-specic real exchange rate:
h h
log(ij ) = ij + h + V REXij ,
with
h
V REXi,j a measure of the volatility of product h's real exchange rate between countries i and
j. This is a simple, reduced-form, manner of capturing the possibility that a high degree of bilateral
exchange rate volatility will aect trade between two countries.
3 Analogously, I will assume that
at an aggregate level, the multilateral trade resistance is parametrically related to product h's real
exchange rate relative to the world international prices
log(ih ) = i + h + V REXih .
Under these assumptions, expressions (1) to (2) can be transformed into the following estimable
equations
h
log Xi,W = i + h + V REXih + h
ij (3)
h h
log Xi,j = jh + ij + V REXij + h
ij (4)
2.2 Derivation of the volatility of the product level real exchange rate
Let
h
Ri be the sector-specic real exchange rate of product h from country i vis-à-vis the dollar,
computed as
h ei,U S ph
Ri = (5)
pi
where ei,U S is country i's nominal exchange rate against the US dollar, ph is the price of good h (in
US dollars), and pi is the country's price level (in local currency). The volatility of the growth or
log-deviations from steady state of this sector-specic exchange rate is, therefore,
V REXih = 2 (^i,U S + ph - pi ),
e ^ ^
= 2 (^i,U S ) - 2Cov(^i,U S , pi ) + 2 (^h - pi ) + 2Cov(^i,U S , ph ),
e e ^ p ^ e ^
= 2 (^i,U S ) - 2Cov(^i,U S , pi ) + 2 (^h - pi ) + 2(^i,U S , ph )(^i,U S )(^h ),
e e ^ p ^ e ^ e p
where the hats indicate growth rates or log deviations of a variable from its steady state value.
The expression above clearly shows that, at the sectoral level, real exchange rate volatility not
only depends on the volatility of the nominal exchange rate but also on the volatility of relative
3
In the model, ij measures the fraction of the exported units that is lost during transportation. From the
viewpoint of the exporter, uncertainty on the exchange rate may be seen as a transportation cost. An appreciation
during transportation will reduce the amount of local goods that the exporter will be able to buy. Of course, the
reverse argument also applies. Which one dominates ex-ante will depend on the degree of risk-aversion of the exporter.
The specication above leaves the data speak about the sign and signicance of .
6
prices and on the comovement between these two. A good whose international price is highly volatile
relative to domestic prices will face volatility in its international competitiveness even if there is no
exchange rate volatility. Similarly, a good whose international price comoves negatively with the
exchange rate will face less volatility in its real exchange rate than a good whose international price
comoves positively with the nominal exchange rate.
It is the last covariance term in the expression above that I associate with the natural hedge of
a good against exchange rate uctuations, although in what follows I will also loosely use this term
to refer to the correlation between the international price of a good and exchange rate uctuations
e ^
((^i,U S , ph )), since this correlation determines whether the covariance term hedges or increases the
exposure to exchange rate volatility. However, it is important to remember that what really matters
is the whole covariance term. For a given correlation and exchange rate volatility, the intensity of
hedging also depends on the variability of a the international price of a good. For instance, the
price of a good may have a large negative correlation with the nominal exchange rate, but if the
price uctuates little it will provide little hedging possibilities, even when the volatility of exchange
rates is large.
Most existing papers on exchange rate volatility try to measure within-year volatility, because
this higher frequency volatility is more likely to matter for ongoing production decisions.
4 Measuring
the within-year monthly volatility of the sectoral real exchange rates requires monthly data on
sectoral prices, which is typically unavailable. To make use of the monthly data available for
the nominal exchange rate I will assume that the correlation between nominal exchange rates and
sectoral relative prices at monthly frequency is proportional to the correlation at annual frequency. I
will also assume that the variance of the relative sectoral price at monthly frequency is proportional
the variance at annual frequency. These assumptions conjecture that there is not enough mean
reversion to reverse the ranking of monthly volatility when moving at annual frequency. In the
next section I will show these are reasonable assumptions using data from an alternative, more
aggregated, data source. Under these assumptions, the within year variance of the sectoral real
exchange rate can be written as
2
V REXih = 2 (^i,U S ) - 2Cov(^i,U S , pi ) + 0 A (^h - pi ) + 1 A (^i,U S , ph )(^i,U S )A (^h )
e e ^ p ^ e ^ e p (6)
where the subscript A denotes that the moment is computed using annual data, and the parameters
are the proportionality coecients for the relation between monthly and annual moments.
Substituting into equation (3) we obtain the following estimable expression that describes the
4
This does not mean that lower frequency volatility does not matter, but that is likely to be more important for
entry-exit decisions (see Caballero and Lorenzoni (2007)). The dierential role of volatility at dierent frequencies
has not been thoroughly studied and is an interesting direction for future research.
7
composition of the trade of country i with the rest of the world.
h
log Xi,W = i + h + V ARPh,i + CORRi,h × SDXRi × SDPh + Zi,h + i,h , (7)
where the variables V ARPi,h , CORRi,h , SDXRi , and SDPh are empirical measures of
2 p
A (^h - pi ),
^
e ^ e
A (^i,U S , ph ), (^i,U S ), and p
A (^h ), respectively, and the rst two terms in equation (6) are absent
because they are absorbed by the country xed-eect i . The variables in Zi,h include other potential
determinants of a country's trade structure that may be included as controls. As it follows from
the discussion in the previous paragraph, the triple interaction term CORRi,h × SDXRi × SDPh
measures good h's natural hedge against exchange rate uctuations, and captures the importance
of natural hedge. However, in the discussion below I will on occasion, and when not confusing,
use the term natural hedge to refer only to the correlation component (CORRi,h ), understanding
that for given volatilities of the exchange rate and international prices, it is the correlation that
determines the level of hedging.
Similarly, the estimable equation for the composition of the bilateral trade with country j is
given by
h
log Xi,j = i j + h + V ARPi,j,h + CORRi,j,h × SDXRi × SDPj,h + Zi,j,h + h .
i,j (8)
where some of the variables have an extra sub-index that shows that the measure also varies with
the importing country. For instance, while in the aggregate specications ph is the global price
of good h, in the bilateral specication it is the price of good h in country j (which is the relevant
price for countries exporting commodity h to country j ). The main deviation from following the
exact same steps as in the derivation of equation (7) is that the bilateral sectoral real exchange rates
were computed using the US dollar as reference currency. The reason is that in the data prices in
all importing countries are quoted in US dollars (see the detailed description of the data below), so
instead of transforming the prices into local currency and transforming the pairs of nominal exchange
rates against the dollars into a bilateral exchange rate, I directly used the dollar based data. Thus,
in equation (8), SDXRi is also the standard deviation of the exporter's country nominal exchange
against the US dollar, and SDPj,h is the standard deviation of the average price of good h in country
j in US dollars, and CORRi,j,h is the correlation between these two variables. A caveat with this
approach is that the triple interaction term in equation (8) is only part of the covariance between
a country-pair's bilateral exchange rate and the price of a commodity in the importing country in
local currency. The other part corresponds to the covariance between the exchange rate against the
dollar of each of the countries in the pair, which is non-parametrically absorbed by the country-pair
xed eect i,j .5
5
Notice also that while equation (4) has a non-parametric importing country-sector xed eect (jh ), this speci-
cation has only a sector xed eect (h ). The justication of this substitution is reducing computing time. In the
8
2.3 Measurement and other issues
The variable CORRi,h is the time-series correlation between good h's international (or bilateral)
price (in US dollars), ^
ph,t , and country i's nominal exchange rate (local currency per US dollar),
^
ei,t , computed as
1
Ti,c p
t (^h,t - ph )(^i,t - ei )
¯ e ¯
CORRi,h = ,
p e
(^h,t )(^i,t )
where the notation follows from above. In what follows, the term correlation and the variable CORR
will be used to refer to this particular correlation measure, unless explicitly stated otherwise. The
bilateral correlation CORRi,j,h is computed analogously, replacing commodity h's global price by
the price in importing country j.
The volatility of nominal exchange rates is captured by their average within-year standard
deviation, SDXR,
12
1
SDXRi,t = (^i,t,m - ec,t )2 ,
e ¯
12
m=1
1
SDXRi = SDREXi,t .
T t
The variance of relative prices and the standard deviation of a commodity price are computed using
annual data, since monthly data is unavailable at high levels of disaggregation:
T
1 2
V ARPi,h = (^h,t - pi,t ) - (^h - pi )
p ^ p ^
T
t=1
T
1 2
SDPh = ph,t - ph
^ ^
T
t=1
and the term CORRi,h × SDXRi × SDPh in equations (7) and (8) denotes the interaction of these
three variables and is the broad measure of the natural hedge provided by a god in a given country (or
country-pair). According to the hypothesis that goods with natural hedging are relatively beneted
in more volatile environments, the coecient on this interaction, , should be signicantly negative.
The specication includes good and country xed eects, denoted by h and i , respectively, and
in the bilateral specication it includes country-pair xed eects i,j that control for the fact that
estimations below, the country-pair xed eects (ij ) are not estimated but properly removed from the estimation
by properly de-meaning the data. A set of importing country-sector xed eects would not be orthogonal to the
country-pair xed eects and would have to be directly estimating, making the estimation extremely slow. I checked
the results with the full set of xed eects described in (4) for the baseline specication, obtaining estimates of
that are quantitatively and statistically similar to those obtained in this simplied version of the specication. The
results are available upon request.
9
some goods are more heavily traded than others and that some countries (or pairs of countries)
trade relatively more and in more goods than the average.
The vector Zi,c includes all potential determinants of a country's trade structure that need to
be controlled for, in particular, the role of countries' factor intensity in shaping its trade patterns.
Standard trade models predict that countries should export relatively more those goods that are
intensive in its abundant production factors (Heckscher et al. (1991)). Although this prediction has
received mixed empirical support (Leamer (1980), Romalis (2004), Fitzgerald and Hallak (2004),
among others), it is important to control for it by adding the interaction of measures of the factor
intensity of a good and a measure of the factor abundance of a country, to avoid wrongly attributing
to exchange rate volatility the role of dierences in factor abundance. Therefore, the empirical
specication in equation (7) could be considered as an extension of the quasi-Heckscher-Ohlin model
estimated by Romalis (2004), that adds the possibility that natural hedge may distort trade patterns
beyond those explained by factor content. In this sense, it is also similar to the reduced form relation
between sectoral output and endowments capturing the intuition of the Heckscher-Ohlin model
estimated by Fitzgerald and Hallak (2004) to explain output composition among OECD countries.
Since there is no data on factor intensities for the broad set of goods considered in this paper,
I construct a rough estimate of the factor intensity of a good by using actual trade patterns to
compute the trade weighted average of the factor abundance of the countries that export it.
6 Under
the assumption that trade patterns reect dierences in factor abundance, this measure would be
akin to a revealed measure of factor intensity. If countries that are relatively abundant in a factor
tend to concentrate a larger fraction of their exports in goods that are relatively intensive in the
use of that factor, a revealed measure of factor intensity of commodity h for factor F , h (F ), can
be built from the within-country composition of trade, as
h
i Xi,W i (F )
h (F ) = h
,
i Xi,W
where i (F ) is country i's relative abundance of factor F, and
h
Xi,W are the total export shares
described above. Three measures of factor intensity are built using this procedure, capturing skilled
labor intensity, capital intensity, and resource intensity. Following Romalis (2004), I also use GDP
per capita as a measure of overall factor abundance. Also following Romalis (2004), the country
level factor abundances i (F ) are taken from Hall and Jones (1999) and from the World Bank
(2008) but instead of expressing factor intensities relative to the US, they are expressed relative to
the world average. The reason is that while Romalis (2004) tries to explain bilateral exports to the
US, this paper focuses on a country's trade with the rest of the world. Notice that, when applied
to the output per capita of dierent countries instead of their factor endowments, this measure
6
Romalis (2004) measures factor intensity directly from US Census of Manufactures (1992), but this would seriously
constrain the set of goods that can be considered.
10
corresponds to the P RODY measure introduced by Hausmann et al. (2007).
Using these revealed factor intensity measures departs from the structure imposed by standard
trade models on the coecients of the Heckscher-Ohlin equations, but it has the advantage of
providing a broader, more agnostic description of trade patterns. It recognizes that, for whatever
reason, some goods are typically more intensively traded by countries that are, for instance, more
abundant in skilled labor, without taking a stance on whether this is due to factor endowments only
or to a combination of endowments with technological dierences. Since this paper does not aim to
document the validity of dierent models of trade, but to test for the role of natural hedge in the
most general way, this agnostic approach is sensible.
7
The main concern with the specication described in equation (7) is the possibility of reverse
causality. The argument for reverse causality is that a country's nominal exchange rate may en-
dogenously appreciate with an increase in the price of goods that represent a large fraction of a
country's exports, inducing a negative correlation between a country's nominal exchange rate and
the price goods it exports the most. This reverse mechanism would only aect the correlation of
a country's exchange rate with the price of dierent goods and not the volatility of the exchange
rate, but it is more likely to be strong when a country has a exible exchange rate regime where the
volatility of the exchange rate is likely to be higher. Therefore, the argument cannot be dismissed
on these grounds. Nevertheless, there are two aspects of the specication that should ease these
concerns. First, the mechanism relates the price each country gets from its most exported goods
to its nominal exchange rate, but the estimated international price used in this paper is a world
average of individual prices. As it will be described below, these individual prices suer important
variation across countries, so the mechanic relation with the international price is less likely. Sec-
ond, this argument is much weaker for bilateral trade patterns. An increase in the price paid for a
commodity by a single trading partner is unlikely to aect the bilateral exchange rate between the
exported and importer, which is determined in a global equilibrium.
Despite these departures from a mechanical reverse relation between export shares and corre-
lations, it may still be the case that the correlation between global and country prices, or between
global and bilateral shares are particularly strong for homogeneous goods. This concern implies
that the estimated relation should be signicantly stronger among homogeneous goods, so it can
be addressed by comparing the coecients obtained for homogeneous and dierentiated goods.
Furthermore, the importance of reverse causality can be further addressed by restricting the esti-
mation to only those goods that do not represent a large share of a country's exports (e.g. below
10 percent). Finally, it is also possibly to tackle this concern directly through instrumental variable
estimation. A potential instrument for the correlation between the international price of a good
7
I also considered an alternative measure that weighted country level factor abundances by the share of a country's
in a commodity global trade instead of the within country shares. The results described below are robust to this
alternative choice, but these revealed measures exhibited less variability across goods because larger countries tend
to grab a relatively large world share of the exports of most products.
11
and the exchange rate of a country vis-a-vis the dollar is the correlation between the price of that
good and the value of the dollar in international markets vis-a-vis other major currencies.
2.4 Stationary versus transition equations
The empirical specications in equations (7) to (8) describe the long run equilibrium of a country's
trade pattern, conditional on other characteristics. Since the force that leads sectors with natural
hedge to become more prevalent in more volatile environments probably operate over long periods,
this is an accurate representation of the long run steady state composition of exports. However,
since the available trade data covers at most 17 years, during which many countries engaged in
structural reforms that might have aected the steady state composition of exports, such as trade
liberalization, from an empirical perspective it may be also appropriate to estimate the transitional
version of the equation that relates natural hedge to the growth of the export share of a commodity
over the sample period, given by,
T -1 log Xi,W
h h
= i +h + log Xi,W +V ARPh,i +CORRi,h ×SDXRi ×SDPh + Zi,h +i,h , ,
t,t+T t
(9)
where the notation is as above, except that T
-1 log h
Xi,W is the average growth rate of
t,t+T
the global export share during the sample period of length T, and
h
log Xi,W is the (log) initial
t
value of this share for each country-good pair. The long run, stationary equilibrium of this equation
corresponds to equation (7), where the long run coecients correspond to the short run ones divided
by - . The long run coecient associated with the interaction would, therefore, be (-/). A
similar transition equation applies to the bilateral export shares model of equation (8).
In addition to allow for non-stationarity, this specication is also less likely to be aected by the
reverse causality problem discussed above, since it controls for initial size dierences across sectors
and convergence eects. Finding similar evidence on growth rates would strongly support the view
that it is the natural hedge oered by the correlation that favors certain sectors.
3 Data
Data on export composition come from Feenstra et al. (2005), NBER-UN World Trade Flows
database, which reports bilateral trade ows (imports and exports) for a large number of countries at
detailed 4-digit SITC (Revision 2) level during the period 1962-2000. This database has been broadly
used in the empirical trade literature, specially in the estimation of gravity models of disaggregated
trade (Chaney (2008), Colacelli (2008), Do and Levchenko (2007)). Because of restrictions to the
availability of other variables, the analysis focuses on the 1984-2000 period. Our nal sample includes
12
129 countries and 749 goods, after dropping some regions and countries with limited coverage or
no mapping with exchange rate data and special SITC codes used by Feenstra et al. (2005) to
make 4-digit ows compatible with 3-digit values. In particular, all countries exporting less than
10 4-digit SITC commodities and all commodities exported by less than 10 countries are excluded,
as well as the US, which is the country of the anchor currency. Country-commodity pairs where
the correlation between international prices and exchange rates was constructed using less than 8
annual observations or those with a correlation larger than 0.9 (in absolute value) were also dropped.
Using this dataset the export shares of a country in dierent goods was constructed as described in
the previous section. The countries included in our main sample as well as their average exports at
SITC level, and number of goods exported are reported in the appendix.
The international price of a good Pi,t is measured by its global unit value, using data on total
values and quantities exported across the world, also from the Feenstra et al. (2005) database,
Vi,t
Pi,t = ,
Qi,t
where Vi,t is the total value of global gross exports of good i, and Qi,t is the total global quantity
exported (in tons). The price index built in this manner corresponds to a quantity weighted average
of country-level unit values.
Despite is simplicity and wide availability there are several shortcomings in using unit values as
measures of international prices, most importantly that they do not control for changes in quality.
Thus, an increase in the quality of goods could result in an increase in unit value that should
not be captured by an ideal price index that controls for the characteristics of a good. However,
since this paper focuses on global unit values, this would not be a problem if the average quality
of traded goods across the world does not change signicantly. Although there is no evidence on
this regard, the average quality of exported goods is more likely to be stable than the quality of
individual country exports, as during the life-cycle of a good, production of the cheapest varieties
shifts to poorer countries. The concerns about quality dierences are also reduced by using a highly
disaggregated classication, which is the motivation for working with the lowest level of aggregation
reported by Feenstra et al. (2005). Also, problems with the level of unit values should not aect
this measure because this paper focuses on the correlation between the growth rate of these global
unit value indexes and the growth of the nominal exchange rates of dierent countries. Dierences
between the growth of this index and that of the actual price of goods exported by a country within
an SITC class may arise from dierences in the growth of the quality of a country's exports relative
to global exports. However, these dierences would only introduce measurement error that would
make it less likely to nd any signicant result.
A more pragmatic reason to use uctuations in global unit values as proxies for price uctuations
13
is that there are no better detailed comprehensive choices available.
8 The Bureau of Labor Statistics
(BLS) reports data on import and export price indexes for the US, but these indexes are available
only at a fairly aggregate level: 129 commodities overall, and only 72 commodities at 3-digit SITC
(Rev 3). Furthermore, the BLS indexes are reported using the third revision of the SITC, while the
trade data described above use the second revision. Imputing BLS prices to trade data requires a
concordance between these two classications. Since the concordance is not one to one (a 4-digit
sector from SITC Rev 2 may belong to more than one 3-digit sector in Rev 3), the number of well
mapped sectors would be even smaller. Working at this level of aggregation would miss the point of
the paper that exposure to exchange rate risk varies across sectors. Nevertheless, to ease concerns
and provide some further validation to the use of global unit values as proxies for prices, I computed
the global unit values for 70 commodities that have BLS price data and where there was a clear map
to SITC Rev 2 commodities and checked the within-commodity across-time correlation between the
two price series by running a xed eect regression between the growth rates of prices from the
two sources. Although the time span is short because BLS data are available only since 1993, the
regression shows a positive and highly signicant relation between the two series. The results are
summarized in the two plots reported in Figure 1 that show the partial regression between the two
price growth series for the whole sample, and dropping the extremes. Each scatterplot also reports
the estimated coecient in each subsample and shows that both relations are highly signicant.
This evidence suggests that, despite their shortcomings, uctuations in global unit values are good
proxies for uctuations in international prices.
The BLS data also shed some light on the validity of the assumption that the correlation be-
tween nominal exchange rate growth and the growth of the sectoral prices at monthly frequency is
proportional to the correlation at annual frequency. Using the monthly BLS data it is possible to
compute and compare the correlations at monthly (within year) and annual frequency, as well as
the variance of relative prices (ph
^ - pi )
^ ^
and the standard deviation of sectoral prices (ph ) at these
two dierent frequencies. The results, summarized in Table 2, show that, within a country, the
correlations, variances, and standard deviations at these two dierent frequencies are positively and
signicantly related. Not surprisingly, the weakest relation is for the correlation measures, where
the relation between the within and across year measures is signicant only at the 2 percent level. It
is also the case that all the estimated coecients are much smaller than 1, which suggests that the
coecients of the specication described in equation (7) estimated using annual frequency proxies
will be smaller than the true coecients associated with the within year measures of correlations,
variances, and standard deviations. Furthermore, it is clear from the gure that the relationship
between annual and monthly correlations, while signicant, is imprecise and annual correlations
2
explain only a fraction of monthly correlations (R is only 0.05). This means that the annual proxies
8
I attempted using US import price indexes from BLS as an alternative, but doing so signicantly reduces the
number of goods covered because of a higher level of aggregation and to an imperfect match between BLS sectors
(based on SITC rev 3) and Feenstra sectors (based on SITC rev 2).
14
may have important measurement error, leading to attenuation bias in the estimated coecients.
This should be considered for the quantitative interpretation of the results below.
Data on nominal and real eective exchange rate at monthly and annual frequencies come from
the International Monetary Fund 2008, and data on the development of foreign exchange derivatives
in dierent countries, measured by the ratio of average daily turnover to total GDP and total exports
come from the Bank of International Settlements (1998, 2001, 2004, and 2007) Triennial Survey of
Central Banks. Since this survey is conducted only every three years, the average value across
available years is considered as representative country value. Data on a country's degree of nancial
development come from Beck et al. (2000) (2006 revision).
The correlation between the annual growth of dierent international good prices and the growth
of the nominal exchange rates of the countries in our sample was computed as described in section
2, using the data detailed above. This procedure yields a matrix of estimated correlations where
each row corresponds to a 4-digit SITC good, each column corresponds to a country, and each entry
is the estimated time-series correlation between the growth rate of the price of the good described
in the row and the country described in the column. The distribution of the estimated correlations
is shown in Figure 2. It shows that while the correlations cover a broad set of values, the mass
of the distribution is shifted to the negative side. There are 57484 correlations in the database
across 129 countries and 749 commodities with available data. Since our sample covers at most
17 observations, most of the estimated correlations are not statistically signicant, in fact only 22
and 32 percent of the estimated correlations are signicantly dierent from zero at the 10 and 20
percent level, respectively. This introduces an additional source of noise that should make even
more dicult nding any signicant result.
The 20 country-good pairs with the most positive and negative correlations are reported in
Table 1. The commodities are distributed across many countries. For instance, the 20 commodities-
countries with the most negative correlations are distributed across 12 countries and 18 commodi-
ties, and the 20 with the largest positive correlation are distributed across 15 countries and 17
commodities. In each group, the commodities cover a wide variety of industries. Figure 3 shows
that this variation is a robust pattern of the data. The average correlation across commodities
within countries (Panel A) uctuates between -0.4 and 0.4. As expected from the histogram, the
average correlation is negative in most countries, but there is a sizable number of countries where it
is positive and there is no individual country that stands as an outlier. The same applies to the dis-
tribution of average correlations across countries within commodities (Panel B). This suggests, as it
is indeed the case, that the results reported below are robust to dropping any particular commodity
and country.
15
4 Results
This section presents and discusses the results of the estimation of the empirical specications
described in equations (7), (8), and (9) that test for the qualitative and quantitative importance
of a sector's natural hedge for a country's export structure. It rst presents the baseline results in
levels and growth rates, addresses several concerns about these specications through a battery of
robustness tests, and explores the role of the availability of formal nancial hedging instruments for
exchange rate risk.
4.1 Baseline results
Countries with higher nominal exchange rate volatility tend to export relatively more goods whose
international price is negatively correlated with their exchange rate uctuations, especially when
this price is also volatile. This is shown in the dierent columns of Table 3. Panel A shows the
estimated coecients for three versions of equation (7) that include only the interaction of CORR,
average exchange rate volatility, and average price volatility CORR × SDXR × SDP , and that
sequentially adds controls for factor content and product dierentiation. In all regressions, the
coecient on the interaction, , is negative and statistically signicant.
The economic magnitude implied by the coecient is small but meaningful. Since the coecient
on natural hedge is akin to a di-in-di estimator, a useful way to gauge its quantitative importance
is to conduct a counter-factual experiment of what would be the change in relative export shares of
dierent commodities resulting from an increase in volatility. Let s10 and s90 be the export shares
of the good at the 10th and 90th percentile of correlation, respectively. An increase from the 10th
to the 90th percentile in the combined volatility of nominal exchange rate and international prices
(SDXR × SDP ) would raise s10 /s90 , the ratio of the export share of a good at the 10th percentile
level of correlation (correlation of -0.55) relative to the share of a good at the 90th percentile of
correlation (correlation of 0.28), by 45 percent. This is about 20 percent of the standard deviation
of relative shares (s
10 /s90 ).
Alternatively, to gauge only the importance of exchange rate volatility, one can keep the volatility
of international prices at its average level and repeat the counter-factual. In this case, an increase
in exchange rate volatility from the 10th to the 90th percentile raises by 37 percent the ratio
of the export share of a good at the 10th percentile level of correlation relative to a good at
the 90th percentile, which corresponds to 15 percent of the standard deviation of (log) export
shares. Of course, the relative dierences are larger when comparing commodities with larger
dierences in natural hedge. Similar increases in combined volatility and in exchange rate volatility
result in relative increases in export shares of 32 and 26 percent of the standard deviation of
relative shares when comparing commodities with correlations of -0.7 and 0.7. Therefore, according
to the estimates, changes in natural hedge could induce changes in relative shares of about one-
16
fth of the within-country dispersion of export shares. While not a rst order determinant of
export composition, the mechanism is quantitatively meaningful. Furthermore, as discussed above,
measurement error in the use of annual correlations and variances to proxy for their within-year
counterparts, as well as the evidence from BLS price data that the within-year measures have less
dispersion than their annual counterparts, suggest that the true coecient for the correlation in
equation (7) may be larger than the ones reported in Table 3. Therefore, it is likely that these
quantitative impacts are a conservative estimate of the true economic relevance of this mechanism.
Bilateral regressions show similar results. The coecient is also signicantly negative across
specications. In this case, an increase in the volatility of a country's exchange rate against the
US dollar from the 10th to the 90th percentile (75 percentage points increase) raises by 20 percent
the ratio of the export share of a good at the 10th percentile level of correlation (-0.65) relative
to a good at the 90th percentile (0.29) (measured at the average level of price volatility). This
corresponds to 11 percent of the average within country-pair standard deviation of log export
shares. Comparing hypothetical commodities with correlations of -0.7 and 0.7 yields a 30 percent
increase in the relative export share of the former, corresponding to 16 percent of the within country-
pair variation of relative shares. This smaller explanatory power is not surprising since the term
included in the regression is only part of the actual covariance between bilateral exchange rates and
importer's prices, the other term being the covariance between the importer and exporter exchange
rates against the dollar, which is absorbed non-parametrically.
Overall, the results for the baseline specication indicate that dierences in natural hedge across
commodities have a small but non-negligible impact of natural hedge on trade patterns. This is
encouraging considering the typically poor t of Heckscher-Ohlin regressions.
The results presented in columns (2), (3), (5), and (6) show that the measures of revealed
factor intensities matter for trade patterns. Column (2) introduces the interaction of the baseline
measures of factor intensity with a country's measure of factor abundance. The regression in Column
(3) replaces these interactions by the interaction of the measure of the overall factor intensity of a
commodity and a country's relative output per worker, which is a measure of a country's overall
factor abundance (see Romalis (2004)). This measure is also the P RODY measure introduced by
Hausmann et al. (2007) and controls in a general way for the possibility that richer countries tend to
export similar products, beyond what can be explained by relative factor endowments. The results
of these two columns show that, in fact, countries tend to concentrate a relatively larger fraction of
their exports in goods that are intensive in their more abundant factors. All estimated coecients
are positive and statistically signicant, except for the one for capital intensity in Column (2), but
the sign and signicance of the main coecient remain unaected. There is, however, a decline
in the magnitude of the main coecient in the regressions for global trade shares, which is about
half the size of the baseline when controlling for overall factor intensity in column (3). Part of this
decline is due to the smaller sample of countries for which there are data on the factor endowments.
17
There is no such decline in the regressions explaining bilateral trade patterns.
Adding these other controls also permit comparing the economic magnitude of the natural hedge
coecient with that of other determinants of trade patterns. I use the same approach described
above of comparing the impact of an increase in a country level variable on goods at the 10th and
90th percentile of the distribution of a variable to quantify the impact of the measures of factor
intensities. The impact of natural hedge on trade composition is 16 percent of that of skill intensity,
80 percent of that of capital intensity, 61 percent of that of resource intensity, and 10 percent of
that of overall factor intensity. Therefore, the economic size of the natural hedge mechanism is
smaller, but in most cases of the same order of magnitude of other signicant determinants of trade
patterns.
Estimating the growth version of equation (7) presented in equation (9) allows dealing with
non-stationarity, which may be appropriate considering the length and time period of the data, and
also reduces concerns about reverse causality between the export share of a good on a country and
the correlation of that country's exchange rate with the price of the good (more on this below).
Table 4 reports the results of estimating the same regressions presented in both panels of Table 3,
but in growth form instead of levels. The evidence is clear and the results are stronger than those
in levels, easing the concerns that the bulk of the previous ndings could be due to reverse causality
(more on this below). The estimated coecients for are negative and statistically signicant
regardless of whether the share of exports represented by a good is measured within or across
countries. According to the estimated coecients, an increase from the 10th to the 90th percentile
in the volatility of exchange rate, would increase in three percentage points the growth rate of the
export share of the sector at the 10th percentile of correlations relative to that of the sector at the
90th percentile. This is about 10 percent of the typical dierences in growth rates of export shares
across sectors.
In the growth equation, the magnitude of the natural hedge coecient is also large compared
with the economic eect implied by the coecients associated with other determinants of trade
composition, which are in most cases statistically insignicant. For instance, an increase from the
10th to the 90th percentile in either the country level of skills or capital per worker would increase
the dierence in growth rates of sectors at the 10th and 90th percentile of revealed skill intensity
in about three percentage points; about the same impact of natural hedge computed at the mean
level of commodity price volatility. Only the coecient associated with overall factor abundance has
a larger economic impact three times larger than that of natural hedge, although not statistically
signicant. Thus, the natural hedge mechanism is not only statistically signicant, but also has a
similar economic impact as that of more standard determinants of trade.
It is also possible to use the estimated coecients to gauge the long run relation between
natural hedge, exchange rate volatility, and trade composition. As mentioned in section 2, the long
run coecient associated with the main interaction is (-/). The value of this coecient, and
18
its standard error are reported at the bottom of the table. The coecient is about twice as large
as that obtained in the levels equation, and it implies that an increase in exchange rate volatility
from the 10th to the 90th percentile would result in a relative share increase of about 82 percent,
which is about 34 percent of the typical variation in relative shares across industries. Of course, the
relative comparison with other determinants of trade is the same as in growth rates.
4.2 Reverse causality issues
The concern about reverse causality is addressed in the regressions reported in Table 5. As a
reminder, the concern is that changes in the international price of goods that capture a high share
of exports could put pressure on the nominal exchange rate. This is what the commodity currency
argument has in mind, and could induce a negative correlation between the price of a country's main
exports and its nominal exchange rate. As argued above, the use of an average international price
index for each good instead of a country specic price should ease concerns. However, the concern
remains if there is little variation in the price paid to dierent countries for a given good. This is
most likely to be the case for homogeneous goods that have a well dened international price. Thus,
the regressions in columns (1) and (2) separately estimate the baseline specication for dierentiated
and homogeneous goods, respectively (dened according to Rauch (1999) classication).
The results show that the ndings are not driven by homogeneous goods, and in fact, the re-
sults are stronger among dierentiated goods. Considering total export shares, the natural hedge
coecient is even positive among homogeneous goods, although very imprecisely estimated. Con-
sidering bilateral export shares the coecient for homogeneous goods is slightly smaller than for
dierentiated goods, but also insignicant. Since international prices of dierentiated goods are
not mechanically related to an individual country's terms of trade, this should ease concerns about
the role that an increase in the prices of heavily exported goods has on the nominal exchange rate.
These results are also consistent with those of Broda and Romalis (2003), who nd that exchange
rate volatility mattered mainly for dierentiated goods.
Another way of tackling the reverse causality issue is to exclude from the estimation those goods
that represent a sizable share of a country's exports. The mechanism behind the reverse causality
issue is roughly that an increase in the price of an exported good increases the local supply of dollars,
putting pressure for a nominal appreciation. While this is plausible in countries that have large and
very concentrated exports (e.g. commodity exporters), this macroeconomic eect is unlikely to occur
for goods that represent a small fraction of a country's exports. To check for this, the regressions
in columns (3) to (5) sequentially drop from the sample those goods with average export shares
above 10, 5, and 1 percent. The main coecient remains largely unchanged. Finally, the regression
in Column (6) reproduces the baseline specication using a country's share of global commodity
exports in each commodity. That is, instead of measuring the importance of a commodity relative
to a country's total exports, this regression measures the importance of a country's exports of a
19
commodity relative to the global exports of that commodity. This specication can be easily derived
in the same way that equation (7) was derived in section 2. Since a commodity may be important
for a country without necessarily representing a high share of global trade in that commodity, this
specication should be less aected by the reverse causality mechanism. The results again show a
signicantly negative relation between natural hedge and global export share of a commodity.
9
Finally, the best way of addressing any concern of reverse causality is to use instruments. Of
course, the challenge is to nd relevant exogenous instruments. The regression in Column (7)
show the results obtained instrumenting the correlation between a goods' international price a
country's nominal exchange rate by the correlation observed for that good in the UK, France, and
Japan. These instruments should capture global movements in the value of the dollar vis-a-vis
major currencies, and, therefore, are not aected by the relation between individual country's main
export price and its exchange rate. The triple interaction associated with the natural hedge is then
instrumented by the interaction of the correlations in each of these three countries and the product
of a country's exchange rate volatility and a commodity's price volatility. These IV regressions were
only conducted for the global trade share specication (Eq. [7]) because the instruments capture
the correlations resulting from global uctuations of major currencies against the dollar and cannot
explain the bilateral dimension of the correlations. In any case, as discussed above, the reverse
causality argument does not clearly apply to bilateral trade shares.
The results yield supportive evidence for the presence of a causal link between natural hedge and
export composition. The coecient for the interaction of correlation and exchange rate volatility
is negative, statistically signicant, and an order of magnitude larger than those reported in Table
3. The instruments are valid according to the Hansen J-statistic (see the p-value of this test at
the bottom of the table) and they are not weak: the F statistic of the rst stage regression is 55,
and the hypothesis that the actual size of the nominal 5 percent test is larger than 5 percent is
rejected based on the Kleibergen and Paap (2006) LM statistic and the Stock and Yogo (2002)
empirical distribution for the Cragg-Donald statistic. Similar results are obtained for the regression
including the factor endowments, although the coecients for the factor endowment are similar to
those obtained by OLS. Interestingly, the coecient for the variance of real commodity price is now
negative and signicant.
These results, and the discussion above on the role of measurement error and the use of annual
moments instead of within-year ones, indicate that the true relevance of natural hedge for the com-
position of exports may be signicantly higher than that obtained in the baseline OLS estimation,
and that the OLS coecient could be considered a lower bound on the true importance of natural
hedge. This reinforces the view that the mechanism is quantitatively relevant.
Overall, these results indicate that concerns for reverse causality are not behind the qualitative
ndings of the paper. Together with the evidence on growth rates that is not mechanically aected
9
Of course, the global shares concept cannot be replicated at the bilateral level.
20
by this problem, this strongly support the view that dierences in natural hedge matter for the
composition of a country's exports.
4.3 Robustness
The baseline results were obtained measuring the correlation of the international prices and nominal
exchange rates, as well as the volatility of the exchange rate using the growth rates of each series
to remove trends. Under the assumption that the series are dierence stationary (in logs), this
procedure correctly removes trends, but it may still lead to spurious correlations if the growth rates
are trending, which could happen over short periods of time. An alternative procedure would be
to de-trend the series by ltering their long run trends using a Hodrick-Prescott lter (HP lter).
This more complex detrending procedure allows for changes in the low frequency trends followed by
the series.
10 To check for the importance of the choice of detrending procedure I also estimated the
regressions for the global share specication (Eq. [7]) using standard deviations and correlations
computed from HP ltered series.
11 The results, reported in Table 6, are somewhat smaller but
very much in line with those obtained in the baseline regressions, both in sign and magnitude.
A potential concern with the results is that extreme and potentially unreliable observations for
the correlation may be behind the ndings. The baseline sample already dropped country-industries
with correlations larger than 0.9 in absolute value, with correlations computed using too few annual
observations (less than 8), as well as countries that exported too few goods and industries exported
by few countries. Still, to check the role of extreme observations, the regressions in Table 7 impose
further constraints on those variables. Column (1) drops country-industries with correlations above
0.8 (in absolute value), Column (2) drops those computed with less than 12 annual observations,
and Column (3) drops sectors exported by less than 20 countries and countries that export less than
20 goods. As before, the results are largely unaected.
The ndings are not crucially driven by individual countries or goods. This is shown in Figure
4 that present histograms of the values for the main coecient obtained after dropping one
country and one good at a time from the sample, respectively. The gure presents ve histograms.
Panels A and B show the histogram of the coecient for the regression with global export shares as
dependent variable after dropping one exporting country (Panel A) and one industry (Panel B) at
a time. Panels C to E show the histogram for the coecient of the bilateral export share regression
obtained by sequentially dropping one exporting country (Panel C), importing country (Panel D),
and industry (Panel E) at a time. Summary statistics of the distribution of coecients depicted
in each histogram are reported at the bottom of each gure. It is apparent that the estimated
10
The annual series were ltered using a smoothing parameter of 100. The monthly series were ltered with a
parameter of 14400.
11
I did not compute the HP ltered series for the bilateral specications because the length of importer based unit
values are shorter and less continuous than those for global unit values, which dicults the use of the HP ltering
algorithm.
21
coecients are tightly distributed around the baseline level. For instance, the mean coecient in
Panel A is -3.18 and has a standard deviation of 0.08. The situation is similar for all other panels.
Therefore, dropping individual countries or industries has little impact on the estimated coecient,
which shows that the results are not crucially driven by specic dimensions of the data, but respond
to a more robust pattern of variation.
4.4 The role of exchange rate regimes
An alternative way of testing for the relevance of natural hedge is to test whether it matters relatively
more for the composition of exports during periods of exible exchange rates than during various
types of pegs. By construction the exchange rate can uctuate more in the former than in the latter
type of regime. To test this hypothesis, the regressions reported in Table 8 estimate the parameters
of the baseline specication (Eq. [7]) across dierent exchange rate regimes, classied according to
Reinhart and Rogo (2004) in ve categories comprising the spectrum between full pegs and fully
exible regimes. In these estimations, the triple interaction term CORRi,h × SDXRi × SDPh is
replaced by the double interaction CORRi,h × SDPh because the goal is to test for dierences in
the importance of natural hedge across regimes, and the dierences in exchange rate regimes are
proxying for dierences in the volatility of exchange rate. The specication for the bilateral export
shares were not estimated because of the diculty of dening a country's exchange rate regime
against each of its trading partners.
The results for dierent regimes are presented in columns (1) to (4) by degree of exibility, from
the harder pegs to free oating.
12 They show that the main coecient is clearly more negative
in fully exible regimes (Column (4)) than in the other ones. In fact, the regression reported in
Column (5) shows that the natural hedge coecient is signicantly more negative than in other
regimes. The negative coecients during hard pegs is a bit surprising, but if the peg currency is
not the US dollar there may still be important uctuations in the dollar exchange rate (remember
that prices are denominated in US dollars). The export share of a sector at the 10th percentile of
correlation is 41 percent larger than that of a sector at the 90th percentile level in a oating regime
than during hard and soft pegs. This unambiguously shows that natural hedge is signicantly more
important for export patterns in exible exchange rate regimes than during pegs.
4.5 Natural hedge and country characteristics
The natural hedge against exchange rate uctuations provided by some goods should be relatively
more important in open countries where trade represents a larger fraction of GDP and is determined
by market forces. One would also expect it to be less important when there are formal hedging
instruments available to protect exporting rms against exchange rate risk or when nancially sound
12
I use the Reinhart and Rogo (2004) coarse annual classication and exclude from the sample those regimes
classied as free falling and those with dual pegs with missing data (categories 5 and 6).
22
exporting rms can easily borrow to overcome temporary uctuations in real costs associated with
exchange rate movements. The most appropriate nancial instruments to hedge exchange rate risk
are nancial derivatives, such as futures and forwards of exchange rates that allow rms to x the
exchange rate at a future time, or exchange rate swaps that allow rms to trade exchange rate
risks. However, these derivatives are relatively sophisticated and are not widely available, especially
exchange rate futures that require some form of standardization and are traded in open exchanges.
In most cases where there is a market for exchange rate risk this takes place over the counter.
As mentioned in section 3, data on the depth of foreign exchange rate derivative markets comes
from the Bank of International Settlements. The survey conducted by the BIS covers a set of 52
countries where these markets have some meaningful degree of activity. In the rest of the world,
forex derivatives are largely nonexistent. However, even if there are no derivative markets that
allow a rm to hedge exchange rate risk ex-ante, a well functioning nancial market may provide
liquidity to exporters facing temporary diculties arising from exchange rate uctuations. Thus,
even in absence of direct hedging options the role of natural hedging may be stronger in nancially
underdeveloped countries.
To test the hypotheses that the degree of trade openness, and the development of forex deriva-
tives and nancial markets reduce the importance of natural hedge, I estimate the parameters of
equation (7) separately for dierent groups of countries formed according to the median level of
each these characteristics, allowing therefore the natural hedge parameter to vary with these
characteristics. The results of the estimation are reported in the two panels of Table 9. The rst
two columns separately estimate the coecient among closed and open countries (i.e. those with
total trade to GDP below and above the sample median, respectively). The results in both panels
show a larger coecient for natural hedge among open countries (Column (2)) than among closed
ones (Column (1)), but the dierence between these two groups is not signicantly dierent from
zero (see the Test of Equality of Interaction row at the bottom of the table that reports the p-value
for the test that the coecients in the two groups are identical).
The next two columns compare those countries where there is no forex derivatives market data
and those where there is data, regardless of the size of the market (Columns (3) and (4), respectively)
Surprisingly, the results clearly show that the coecient is more negative and signicant among
countries where there is forex market data (Column (4)); in fact, in Panel A the coecient is
one order of magnitude larger among these countries than among countries with no data on forex
markets. In Panel B that looks at bilateral trade patterns the dierence between the two groups
is smaller and statistically insignicant, although still the coecient is larger among countries with
forex data.
The regressions in columns (5) and (6) estimate the parameters of equations (7) and (8) among
those countries with data on forex derivatives. Even within this group, the coecient on natural
hedge is more negative in countries with a high foreign exchange derivatives turnover, although the
23
dierence with the coecient obtained in countries with low turnover is not signicant in either of
the two panels.
Even if foreign exchange derivative markets are not well developed, natural hedge may be rel-
atively unimportant if exporting rms can easily borrow to overcome temporary uctuations in
exchange rates (Caballero and Lorenzoni (2007)). To test for this possibility, the regressions in
columns (7) and (8) estimate the baseline specications among countries with nancial develop-
ment above and below the sample median (nancial development is captured by the average ratio of
private credit to GDP over the sample period, from Beck et al. (2000), revision 2006). Similarly to
the results on the role of the development of derivative markets, the results surprisingly show that
the coecient is more negative among countries with high nancial development, indicating that
natural hedge is more important among these countries. This evidence suggests that the availability
of formal hedging instruments or credit does not reduce the importance of the natural hedge oered
by the comovement between international prices and nominal exchange rates at the sector level.
To check if these results come from persistent dierences in trade patterns across countries, the
specication in growth rates (Eq. [9]) was also estimated for each of the groups of countries described
above. The results, reported in Table 10, are more favorable to the hypothesis that the availability of
formal hedging attenuates the relevance of natural hedge. The results for global trade shares (Panel
A) show that, while the coecient for natural hedge is still larger among nancially developed
countries, it is now larger among countries with no derivatives data, and, among those countries
with derivatives data, it is larger among countries with relatively lower development of derivatives
markets. However, as shown at the bottom of the panel, none of these dierences is statistically
signicant. The results for bilateral trade shares (Panel B) oer stronger support for the hypothesis.
In these regressions, the coecient for natural hedge is more negative among countries with relatively
lower development of derivative markets and among nancially underdeveloped countries (Columns
(5) and (7), respectively). The former dierence is also signicant at the 5 percent level.
Overall, the results provide only mixed support for the view that the availability of formal
hedging instruments reduces the trade impact of the natural hedge of a good against exchange
rate volatility. Industries with stronger natural hedge do not capture a relatively larger share of
exports in countries with low availability of hedge instruments. However, these industries' shares
of a country's exports have increased during the sample period. These results can be reconciled if
industries with stronger natural hedge had relatively larger initial shares in countries that developed
formal hedging instruments. This could happen if the development of these instruments was an ex-
post reaction to the role of exchange rate volatility. Since for data availability reasons the measures
on the development of forex derivative markets is only available on the latest years of the sample,
it could be that the results are picking on the endogenous response of forex at the aggregate level.
This would also be consistent with the growth of the sectors with natural hedge slowing down as a
result of this subsequent development. Of course, an alternative interpretation is that natural and
24
nancial hedges are complements rather than substitutes. If that is the case, sectors with natural
hedge would be the ones better able to use nancial instruments and would be stronger in countries
where these instruments are available. Disentangling these competing explanations is an interesting
topic for future research.
5 Conclusion
This paper has shown evidence of an indirect causal eect of exchange rate volatility on trade by
looking at the dierential eect of this volatility on the export share of sectors producing com-
modities whose price correlation with the home country's exchange rate oers dierent degrees of
natural hedge against exchange rate uctuations. The results indicate that sectors that oer natural
hedge (i.e. those whose international price comoves negatively with the country's nominal exchange
rate) are indeed larger relative to others, even after controlling for standard determinants of trade
composition, in countries with more volatile exchange rates. Thus, exchange rate volatility tilts
the structure of trade towards sectors that produce goods that oer a natural hedge against this
volatility. This dierential eect comes mainly from countries with a oating exchange rate regime.
Contrary to expectations, there is only weak evidence that the relation between natural hedge,
exchange rate volatility, and export composition is stronger in countries where rms have no access
to nancial instruments to hedge currency risk, such as exchange rate derivatives.
Overall, the results provide evidence that exchange rate volatility has real consequences for trade
and that the structure of trade responds not only to standard factors, such as endowments, but
also to other more institutionally driven constraints. If what you export matters, as suggested by
Hausmann et al. (2007), addressing these sources of ineciencies would be a better way of reaching
a trade pattern that is not driven by risk hedging than engaging in some sort of industrial policy.
The analysis of this paper can be extended along several dimensions. First, using more disaggre-
gated data may permit obtaining better proxies of international price uctuations and, therefore,
better pinpoint the relevance of the natural hedge mechanism. This can be done by using other new
sources of data, such as the Gaulier et al. (2007), BACI database, which has a shorter time span
but covers trade at 6-digits of the Harmonized System. Second, the hedge provided by an industry
depends not only on the commodity it produces, but also on how the price of their imported inputs
comoves with the nominal exchange rate. This paper has taken the stance that that source of hedg-
ing operates largely independently of the one on the product side, but of course this does not need
to be the case. Addressing this issue faces some serious problems of measuring that would require
further assumptions, but it is an interesting area of future research.
25
References
Beck, T., A. Demirguc-Kunt, and R. Levine, A New Database on the Structure and Devel-
opment of the Financial Sector, The World Bank Economic Review, 2000, 14 (3), 597605.
Bleakley, H. and K. Cowan, Corporate Dollar Debt and Depreciations Much Ado About Noth-
ing?, Review of Economics and Statistics, 08 2008, 90 (4), 612626.
Broda, C., Terms of trade and exchange rate regimes in developing countries, Journal of Inter-
national Economics, 2004, 63 (1), 3158.
and J. Romalis, Identifying the Relationship Between Trade and Exchange Rate Volatility,
Technical Report, Chicago GSB manuscript 2003.
Byrne, J.P., J. Darby, and R. MacDonald, US trade and exchange rate volatility: A real
sectoral bilateral analysis, Journal of Macroeconomics, 2008, 30 (1), 238259.
Caballero, R.J. and G. Lorenzoni, Persistent Appreciations and Overshooting: A Normative
Analysis, Technical Report, NBER Working Paper 2007.
Cashin, P., L.F. Céspedes, and R. Sahay, Commodity currencies and the real exchange rate,
Journal of Development Economics, 2004, 75 (1), 239268.
Chaney, Thomas, Distorted Gravity: The Intensive and Extensive Margins of International
Trade, American Economic Review, 2008, 98 (4), 17071721.
Chen, Y. and K. Rogo, Commodity currencies, Journal of International Economics, 2003, 60
(1), 133160.
Clark, P., N. Tamirisa, S.J. Wei, A. Sadikov, and L. Zeng, Exchange rate volatility and
trade ows-some new evidence, Technical Report, Washington, DC: International Monetary Fund
2004.
Clark, P.B., Uncertainty, Exchange Rate Risk, and the Level of International Trade, Economic
Inquiry, 1973, 11 (3), 302313.
Colacelli, M., Export Responses to Real Exchange Rate Fluctuations: Development Status and
Exported Good Eects, Technical Report, Columbia University 2008.
Do, Q.T. and A.A. Levchenko, Comparative advantage, demand for external nance, and
nancial development, Journal of Financial Economics, 2007, 86 (3), 796834.
Edwards, S. and E. Levy Yeyati, Flexible exchange rates as shock absorbers, European Eco-
nomic Review, 2005, 49 (8), 20792105.
26
Feenstra, R.C., R.E. Lipsey, H. Deng, A.C. Ma, and H. Mo, World Trade Flows: 1962-
2000, Technical Report, NBER Working Paper 2005.
Fitzgerald, D. and J.C. Hallak, Specialization, factor accumulation and development, Journal
of International Economics, 2004, 64 (2), 277302.
Frankel, J. and A. Rose, An Estimate of the Eect of Common Currencies on Trade and Income
1, Quarterly Journal of Economics, 2002, 117 (2), 437466.
Frankel, Jerey A. and Shang-Jin Wei, Emerging Currency Blocs, Working Paper 4335,
National Bureau of Economic Research April 1993.
Gaulier, G., R. Paillacar, and S. Zignago, BACI: A World Database of International Trade
at the Product-level 1995-2004 Version, Technical Report, CEPII World Paper 2007.
Hall, R.E. and C.I. Jones, Why Do Some Countries Produce So Much More Output Per Worker
Than Others?*, Quarterly Journal of Economics, 1999, 114 (1), 83116.
Hausmann, R., J. Hwang, and D. Rodrick, What you export matters, Journal of Economic
Growth, 2007, 12, 125.
Heckscher, E.F., B.G. Ohlin, and M.J. Flanders, Heckscher-Ohlin Trade Theory, The MIT
Press, 1991.
International Monetary Fund, International Financial Statistics, 2008.
Kleibergen, F. and R. Paap, Generalized reduced rank tests using the singular value decom-
position, Journal of Econometrics, 2006, 133 (1), 97126.
Klein, M.W. and J.C. Shambaugh, Fixed exchange rates and trade, Journal of International
Economics, 2006, 70 (2), 359383.
Leamer, E.E., The Leontief Paradox, Reconsidered, The Journal of Political Economy, 1980, 88
(3), 495.
McKenzie, M.D., The Impact of Exchange Rate Volatility on International Trade Flows, Journal
of Economic Surveys, 1999, 13 (1), 71106.
Melitz, M.J., The impact of trade on intra-industry reallocations and aggregate industry pro-
ductivity, Econometrica, 2003, 71 (6), 16951725.
Rauch, J.E., Networks versus markets in international trade, Journal of International Eco-
nomics, 1999, 48 (1), 735.
27
Reinhart, C.M. and K.S. Rogo, The Modern History of Exchange Rate Arrangements: A
Reinterpretation, Quarterly Journal of Economics, 2004, 119 (1), 148.
Romalis, J., Factor Proportions and the Structure of Commodity Trade, American Economic
Review, 2004, 94 (1), 6797.
Rose, A.K., One money, one market: the eect of common currencies on trade, Economic policy,
2000, 15 (30), 746.
and E. Van Wincoop, National money as a barrier to international trade: The real case for
currency union, American Economic Review, 2001, 91 (2), 386390.
Stock, J.H. and M. Yogo, Testing for Weak Instruments in Linear IV Regression, Technical
Report, NBER Working Paper 2002.
Tenreyro, Sylvana, On the trade impact of nominal exchange rate volatility, Journal of Devel-
opment Economics, 2007, 82 (2), 485508.
Wei, S.J., Currency hedging and goods trade, European Economic Review, 1999, 43 (7), 1371
1394.
World Bank, World Development Indicators, 2008.
28
Table 1. Highest 20 Country-Commodity Pairs of Positive and Negative Correlation
Country SITC code Commodity Correlation
Mexico 372 CRUSTACEANS AND MOLLUSCS,PREPARED OR PRESERVED 0.8448
Liberia 1124 SPIRITS;LIQUEURS, OTHER SPIRITUOUS BEVERAGES,N.E.S 0.8492
Slovenia 573 BANANAS,FRESH OR DRIED 0.8517
Indonesia 14 POULTRY, LIVE (I.E., FOWLS, DUCKS, GEESE, ETC.) 0.8517
Ethiopia 6551 KNITTED/CROCH.FAB.NOT ELAST.NOR RUBBERIZ.OF SYNT.F 0.8533
Yemen 6544 FABRICS,WOVEN,OF FLAX OR OF RAMIE 0.8537
Saudi arabia 6118 LEATHER,SPECIALLY DRESSED OR FINISED 0.8557
Ethiopia 6574 ELASTIC FABRICS AND TRIMMINGS 0.8568
Zambia 6344 WOOD-BASED PANELS,N.E.S. 0.8580
Haiti 7432 PARTS OF THE PUMPS & COMPRESSORS OF 743.1- 0.8595
Mozambique 7518 OFFICE MACHINES, N.E.S. 0.8631
El salvador 5913 WEED KILLERS (HERBICIDES)PACKED FOR SALE ETC. 0.8645
Slovenia 6575 TWINE,CORDAGE,ROPES & CABLES.& MANUFACTUR.THEREOF 0.8746
Mexico 5331 OTHER COLOURING MATTER 0.8759
Guatemala 5911 INSECTICIDES PACKED FOR SALE ETC. 0.8789
Peru 2877 MANGANESE ORES AND CONCENTRATES 0.8796
Israel 6118 LEATHER,SPECIALLY DRESSED OR FINISED 0.8843
Bolivia 5334 VARNISHES AND LACOUERS;DISTEMPERS,WATER PIGMENTS 0.8918
Israel 5334 VARNISHES AND LACOUERS;DISTEMPERS,WATER PIGMENTS 0.8955
Liberia 14 POULTRY, LIVE (I.E., FOWLS, DUCKS, GEESE, ETC.) 0.8994
Country SITC code Commodity Correlation
Seychelles 5323 SYNTH.ORG.TANNING SUBSTANCES,& INORG.TANNING SUBST -0.9000
Switzerland 2927 CUT FLOWERS AND FOLIAGE -0.8995
Kenya 2686 WASTE OF SHEEPS/LAMBS WOOL OR OF OTHER ANIM.HAIR -0.8990
Cyprus 7264 PRINTING PRESSES -0.8990
Seychelles 5332 PRINTING INK -0.8988
Spain 6912 STRUCTURES& PARTS OF STRUC.;ALUMINIUM;PLATES,RODS -0.8985
Cyprus 6584 BED LINEN,TABLE LINEN,TOILET & KITCHEN LINEN ETC. -0.8984
France 6794 CASTINGS OR IRON OR STEEL,IN THE ROUGH STATE -0.8979
Belgium 620 SUGAR CONFECTIONERY AND OTHER SUGAR PREPARATIONS -0.8978
Belgium 546 VEGETABLES,FROZEN OR IN TEMPORARY PRESERVATIVE -0.8978
Denmark 620 SUGAR CONFECTIONERY AND OTHER SUGAR PREPARATIONS -0.8972
Italy 7753 DISH WASHING MACHINES OF HOUSEHOLD TYPE -0.8972
Myanmar 484 BAKERY PRODUCTS (E.G.,BREAD,BISCUITS,CAKES) ETC. -0.8971
Spain 5323 SYNTH.ORG.TANNING SUBSTANCES,& INORG.TANNING SUBST -0.8970
Norway 3224 PEAT,WHETHER/NOT COMPRES.INTO BALES NOT AGGLOMERA -0.8969
Switzerland 2926 BULBS,TUBERS & RHIZOMES OF FLOWERING OR OF FOLIAGE -0.8966
Norway 583 JAMS,FRUIT JELLIES, MARMALADES,FRUIT PUREE,COOKED -0.8963
Malta 6973 DOM ESTIC-TYPE,NON-ELECTRIC HEATING,COOKING APPAR. -0.8960
Seychelles 8821 CHEMICAL PRODUCTS & FLASHLIGHT MATERIALS -0.8960
Norway 6635 SLAG WOOL.ROCK WOOL AND SIMILAR MINERAL WOOLS -0.8959
Table 2. Relation Between Measures of Correlations and Variances at Different
Frequencies
Correlation Variance relative Std. dev. Sectoral
exchange rate- sectoral price (BLS price (BLS) within
sectoral price and CPI) within years
(BLS) within years years
(1) (2) (3)
Correlation exchange 0.007**
rate-sectoral price (BLS) (0.003)
Variance relative 0.017***
sectoral price across (0.000)
Std. dev. Sectoral price 0.149***
across years (0.001)
Cons -0.026*** -0.001*** 0.004***
(0.001) (0.000) (0.000)
N 17599 19268 19642
R-sq 0.056 0.638 0.541
The table shows the estimated coefficients from regressions between various statistics
computed using within-year, monthly data on nominal exchange rate, CPI, and sectoral prices
from BLS, and the same statistics computed using across-years, annual frequency data, after
controlling for country fixed effects. In column (1) the dependent variable is the within-year
correlation between a country's monthly (growth of) nominal exchange rate and a sector's price
(from BLS). In column (2), the dependent variable is the within-year monthly variance of (the
growth of) a sector's relative price (the difference in the growth of the BLS index and a
country's CPI). In column (3), the dependent variable is the within-year monthly standard
deviation of a sector's price. In each column, the explanatory variable is the corresponding
statistic computed using annual data. . *, **, and *** denote significance at 10, 5, and 1
Table 3. Natural Hedge and the Composition of Trade. Baseline Results
A. Total Trade Shares B. Bilateral Trade Shares
(1) (2) (3) (4) (5) (6)
0.154 -0.355 -0.278 -0.290*** -0.267*** -0.270***
Variance real commodity price
(0.370) (0.493) (0.440) (0.037) (0.040) (0.040)
CorrelationX SD Exch. Rate X SD -3.179*** -2.393** -1.367* -0.839*** -0.800*** -0.840***
commodity price (0.909) (0.870) (0.774) (0.128) (0.140) (0.139)
Skill Intensity X Human Cap./ 4.419*** 0.582***
Worker (0.567) (0.146)
Capital Intensity X Capital / 1.164 0.594**
Worker (0.774) (0.248)
Resource Intensity X Land Area / 0.233*** 0.113***
Worker (0.037) (0.013)
Factor Intensity X Output / 1.219*** 0.244***
Worker (0.074) (0.019)
Observations 57484 49798 49948 157433 144396 144396
R-squared 0.361 0.390 0.411 0.198 0.205 0.210
Industry FE Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes -- -- --
Country-pair FE Yes Yes Yes
In Panel A the dependent variable is the log of a commodity's share of total global country exports, so the regressions explain a country's
total trade patterns. In Panel B, the dependent variable is the log of a commodity's share of total country exports to each trading partner.
These regressions therefore explain bilateral trade patterns. Variance real commodity price is the variance of the growth rate of the ratio of
a commodity's price to a country's aggregate price level. In Panel A the commodity price is the average world price and in Panel B is the
price in each trading partner. Correlation is the correlation between a country's nominal exchange rate growth and the growth the world
price of each commodity (across years from annual data). SD Exch Rate is the standard deviation of a country's nominal exchange rate
growth (within a year from monthly data) and SD commodity price is the standard deviation of the growth rate of a commodity's
international prices. In panel B, the growth of a commodity price is computed within each trading partner. The symbol X denotes the
interaction between variables. Skilled, Capital, Resource, and Factor Intensity are measures of a commodity's intensity of use of skilled
labor, capital, resources, and all factors in general, and are computed as the weighted average of factor aboundance of countries that export
a commodity. The weight is the share of a country's exports represented by a commodity. These commodity specific measures are interacted
with each country's aboundance of different factors (human capital, capital, resources, and resources). Heterskedasticity robust standard
errors reported in parenthesis. *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Table 4. Natural Hedge and the Composition of Trade. Specification in Growth Rates
A. Total Trade Shares B. Bilateral Trade Shares
(1) (2) (3) (4) (5) (6)
-0.031*** -0.030*** -0.030*** -0.043*** -0.042*** -0.042***
(log) Initial Export Share
(0.002) (0.001) (0.001) (0.001) (0.001) (0.001)
0.011 -0.020 -0.020 -0.023*** -0.021*** -0.021***
Variance real commodity price
(0.030) (0.039) (0.040) (0.004) (0.004) (0.004)
CorrelationX SD Exch. Rate X -0.222*** -0.150** -0.133** -0.035** -0.031* -0.031*
SD commodity price (0.057) (0.061) (0.060) (0.018) (0.019) (0.019)
Skill Intensity X Human Cap./ 0.072 -0.025**
Worker (0.049) (0.010)
Capital Intensity X Capital / 0.076 0.014
Worker (0.077) (0.016)
Resource Intensity X Land Area 0.008*** 0.005***
/ Worker (0.002) (0.001)
Factor Intensity X Output / 0.023 -0.002
Worker (0.005) (0.001)
Long run coefficient -7.1386*** -5.0342** -4.3957** -0.827** -0.748* -0.756*
Observations 43249 38784 38784 157357 144322 144322
R-squared 0.119 0.125 0.125 0.177 0.180 0.179
Industry FE Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes -- -- --
Country-pair FE -- -- -- Yes Yes Yes
In Panel A the dependent variable is the growth of a commodity's share of total global country exports. In Panel B, the dependent
variable is the growth of a commodity's share of total country exports to each trading partner. Initial Export Share is the share of a
commodity in a country's total (bilateral) trade at the beginning of the sample period (or the first value if the initial value is zero) in
Panel A (Panel B). Variance real commodity price is the variance of the growth rate of the ratio of a commodity's price to a country's
aggregate price level. In Panel A the commodity price is the average world price and in Panel B is the price in each trading partner.
Correlation is the correlation between a country's nominal exchange rate growth and the growth the world price of each commodity
(across years from annual data). SD Exch Rate is the standard deviation of a country's nominal exchange rate growth (within a year
from monthly data) and SD commodity price is the standard deviation of the growth rate of a commodity's international prices. In
panel B, the growth of a commodity price is computed within each trading partner. The symbol X denotes the interaction between
variables. Skilled, Capital, Resource, and Factor Intensity are measures of a commodity's intensity of use of skilled labor, capital,
resources, and all factors in general, and are computed as the weighted average of factor aboundance of countries that export a
commodity. The weights are the shares of a country's exports represented by a commodity. These commodity specific measures are
interacted with each country's aboundance of different factors (human capital, capital, resources, and resources). Heteroskedasticity
robust standard errors reported in parenthesis. *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Table 5. Controlling for Potential Feedback from Trade Shares to Correlations
Diff. Homo Share<.0 Share<.0 Global
goods goods Share< .1 5 1 Shares IV
(1) (2) (3) (4) (5) (4) (5)
A: Global Trade Shares
0.205 -0.879 0.198 0.175 0.149 0.1653 -1.659*
Variance real commodity price
(0.330) (1.542) (0.366) (0.357) (0.323) (0.3716) (1.006)
CorrelationX SD Exch. Rate -3.277*** 2.735 -3.121*** -3.082*** -2.834*** -3.1804*** -28.80**
X SD commodity price (0.828) (1.895) (0.887) (0.852) (0.806) (0.9212) (10.98)
Observations 47100 5653 57332 57158 55833 57484 55267
R-squared 0.382 0.363 0.363 0.366 0.367 0.624 --
Industry FE Yes Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes Yes Yes Yes Yes
(1) (2) (3) (4) (5) (4) (5)
B: Bilateral Trade Shares
-0.309*** -1.078*** -0.270*** -0.253*** -0.167*** -- --
Variance real commodity price
(0.038) (0.245) (0.037) (0.038) (0.038)
CorrelationX SD Exch. Rate -0.875*** -0.728 -0.704*** -0.649*** -0.469*** -- --
X SD commodity price (0.134) (0.946) (0.126) (0.125) (0.123)
Observations 140222 8283 152527 148694 127144 -- --
R-squared 0.179 0.293 0.183 0.177 0.156
Industry FE Yes Yes Yes Yes Yes -- --
Country-pair FE Yes Yes Yes Yes Yes -- --
In Panel A the dependent variable is the log of a commodity's share of total global country exports, except in column (6) where it
is the log share of a country's commodity's exports on global exports of that commodity. In Panel B, it is the log of a commodity's
share of total country exports to each trading partner. Variance real commodity price is the variance of the growth rate of the
ratio of a commodity's price to a country's aggregate price level. In Panel A the commodity price is the average world price and in
Panel B is the price in each trading partner. Correlation is the correlation between a country's nominal exchange rate growth and
the growth the world price of each commodity (across years from annual data). SD Exch Rate is the standard deviation of a
country's nominal exchange rate growth (within a year from monthly data) and SD commodity price is the standard deviation of
the growth rate of a commodity's international prices. In panel B, the growth of a commodity price is computed within each
trading partner. The symbol X denotes the interaction between variables. Columns (1) and (2) report regressions estimated in the
sample of differentiated and homogeneous goods respectively (according to Rauch (1995) classification). The regressions in
columns (3) to (5) sequeantially exclude from the sample all commodities where a country has a total (bilateral in Panel B)
export share larger than 10, 5, and 1 percent respectively. The regression in column (7) was estimated by two-stages leaste
squares using as instruments for the correlation in each country-industry the correlation computed for each industry in the UK,
France and Japan. Heterskedasticity robust standard errors reported in parenthesis, clustered at the country (country-pair) level
in Panel A (B). *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Table 6. Natural Hedge and the Composition of Trade. Moments Based on HP Filtered
Series
Total Trade Shares
(1) (2) (3)
-0.003 0.017 0.027
Variance real commodity price
(0.085) (0.080) (0.083)
CorrelationX SD Exch. Rate X SD commodity -1.825*** -1.455** -1.016**
price (0.494) (0.466) (0.403)
4.421***
Skill Intensity X Human Cap./ Worker (0.568)
1.175
Capital Intensity X Capital / Worker
(0.773)
0.233***
Resource Intensity X Land Area / Worker
(0.037)
1.220***
Factor Intensity X Output / Worker
(0.074)
Observations 57484 49798 49948
R-squared 0.360 0.390 0.411
Industry FE Yes Yes Yes
Country FE Yes Yes Yes
The dependent variable is the log of a commodity's share of total global country exports. Variance real
commodity price is the variance of the growth rate of the ratio of a commodity's price to a country's aggregate
price level. The commodity price is the average world price. Correlation is the correlation between a country's
nominal exchange rate growth and the growth the world price of each commodity (across years from annual
data). SD Exch Rate is the standard deviation of a country's nominal exchange rate growth (within a year
from monthly data) and SD commodity price is the standard deviation of the growth rate of a commodity's
international prices. The symbol X denotes the interaction between variables. Skilled, Capital, Resource, and
Factor Intensity are measures of a commodity's intensity of use of skilled labor, capital, resources, and all
factors in general, and are computed as the weighted average of factor aboundance of countries that export a
commodity. The weights are the share of a country's exports represented by a commodity. These commodity
specific measures are interacted with each country's aboundance of different factors (human capital, capital,
resources, and resources). Heterskedasticity robust standard errors reported in parenthesis. *, **, and ***
denote significance at 10, 5, and 1 percent, respectively.
Table 7. Robustness to Dropping Extreme Observations
Restrict Restrict obs. to Restrict
Correlation compute corr. number of
(|Corr|<0.8) (obs>12) sectors
(1) (2) (3)
A: Global Trade Shares
0.170 0.149 0.158
Variance real commodity price
(0.378) (0.448) (0.369)
CorrelationX SD Exch. Rate X -3.086** -3.333** -3.169***
SD commodity price (0.921) (1.005) (0.908)
Observations 56714 53645 57451
R-squared 0.360 0.363 0.360
Industry FE Yes Yes Yes
Country FE Yes Yes Yes
(1) (2) (3)
B: Bilateral Trade Shares
-0.288*** -0.478*** -0.219***
Variance real commodity price
(0.037) (0.051) (0.040)
CorrelationX SD Exch. Rate X -0.926*** -1.482*** -0.819***
SD commodity price (0.131) (0.218) (0.148)
Observations 152758 78380 136546
R-squared 0.198 0.246 0.194
Industry FE Yes Yes Yes
Country-pair FE Yes Yes Yes
In Panel A the dependent variable is the log of a commodity's share of total global country exports.
In Panel B, it is the log of a commodity's share of total country exports to each trading partner.
Variance real commodity price is the variance of the growth rate of the ratio of a commodity's price
to a country's aggregate price level. In Panel A the commodity price is the average world price and
in Panel B is the price in each trading partner. Correlation is the correlation between a country's
nominal exchange rate growth and the growth the world price of each commodity (across years from
annual data). SD Exch Rate is the standard deviation of a country's nominal exchange rate growth
(within a year from monthly data) and SD commodity price is the standard deviation of the growth
rate of a commodity's international prices. In panel B, the growth of a commodity price is
computed within each trading partner. The symbol X denotes the interaction between variables.
The regression in column (1) include only sectors where the correlation between nominal exchange
rate and international price fluctuations is smaller than 0.8 in absolute value. Column (2) indludes
only those country-industries where the correlation was computed with more than 12 annual
observations (the baseline is 8 observations). Finally, column (3) includes only those commodities
that are exported by more than 20 countries and countries that export more than 20 commodities.
Heterskedasticity robust standard errors reported in parenthesis, clustered at the country (country-
pair) level in Panel A (B). *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Table 8. Differences Across Exchange Rate Regimes
Regime 1 Regime 2 Regime 3 Regime 4 Free floating
(1) (2) (3) (4) (5)
0.048 -0.019 -1.930 0.115 -0.383
Variance real commodity price
(0.461) (1.010) (1.829) (0.777) (0.279)
CorrelationX SD commodity -2.271** -1.812** -0.770 -3.608** -1.596***
price (0.779) (0.657) (0.658) (1.115) (0.125)
Free Float X Variance real 0.498
commodity price (0.669)
Free Float X Correl. X SD -2.012***
commodity price (0.581)
Observations 19952 26745 20425 3606 70728
R-squared 0.396 0.340 0.380 0.513 0.359
Industry FE Yes Yes Yes Yes Yes
Country FE Yes Yes Yes Yes Yes
The dependent variable is the log of a commodity's share of total global country exports. Variance real
commodity price is the variance of the growth rate of the ratio of a commodity's price to a country's aggregate
price level. The commodity price is the average world. Correlation is the correlation between a country's
nominal exchange rate growth and the growth the world price of each commodity (across years from annual
data). SD Exch Rate is the standard deviation of a country's nominal exchange rate growth (within a year from
monthly data) and SD commodity price is the standard deviation of the growth rate of a commodity's
international prices. The symbol X denotes the interaction between variables. Columns (1) to (4) show results
restricted to periods where the exporting country was in each of the four main exchange rate regimes according
to the classification of Reinhart and Rogoff (2004). The degree of flexibility of the regime increases from 1 (hard
pegs) to (4) flexible regimes. Column (5) nest the 4 specification including a dummy for countries with a flexible
exchange rate regime and interacting it with the variables of the baseline specification. Heterskedasticity robust
standard errors reported in parenthesis, clustered at the country (country-pair) level in Panel A (B). *, **, and
*** denote significance at 10, 5, and 1 percent, respectively.
Table 9. Differences Across Groups of Countries
Closed Open No Forex Forex Data Low Forex High Forex Low Fin. High Fin.
Data Develop. Develop
(1) (2) (3) (4) (5) (6) (7) (8)
A: Global Trade Shares
-0.008 0.096 0.241 -0.032 0.634 -0.287 0.554 -0.450
Variance real commodity price
(0.572) (0.654) (0.477) (0.378) (0.743) (0.445) (0.559) (0.689)
CorrelationX SD Exch. Rate X -3.147** -4.015** -5.658*** -0.445 -4.090** -5.926** 0.618 -6.238***
SD commodity price (1.085) (1.582) (1.232) (0.787) (1.389) (2.829) (0.834) (1.433)
Observations 28595 27135 25767 31717 15961 15756 19174 36209
R-squared 0.379 0.380 0.374 0.477 0.485 0.536 0.350 0.433
Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Test of Equality of interaction coef. 0.651 0.000 0.560 0.000
(1) (2) (3) (4) (5) (6) (7) (8)
B: Bilateral Trade Shares
-0.342*** -0.214*** -0.548*** -0.284*** -0.308*** -0.308*** -0.185** -0.369***
Variance real commodity price
(0.055) (0.045) (0.156) (0.038) (0.053) (0.051) (0.064) (0.047)
CorrelationX SD Exch. Rate X -0.681*** -1.028*** -0.212 -0.844*** -0.856*** -0.739*** -0.500** -0.839***
SD commodity price (0.160) (0.206) (0.525) (0.129) (0.159) (0.207) (0.205) (0.161)
Observations 84539 72405 9386 148047 98680 58753 52564 96343
R-squared 0.224 0.215 0.375 0.198 0.212 0.269 0.226 0.235
Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
Country-pair FE Yes Yes Yes Yes Yes Yes Yes Yes
0.183 0.242 0.654 0.193
In Panel A the dependent variable is the log of a commodity's share of total global country exports. In Panel B, it is the log of a commodity's share of total
country exports to each trading partner. Variance real commodity price is the variance of the growth rate of the ratio of a commodity's price to a country's
aggregate price level. In Panel A the commodity price is the average world price and in Panel B is the price in each trading partner. Correlation is the
correlation between a country's nominal exchange rate growth and the growth the world price of each commodity (across years from annual data). SD Exch
Rate is the standard deviation of a country's nominal exchange rate growth (within a year from monthly data) and SD commodity price is the standard
deviation of the growth rate of a commodity's international prices. In panel B, the growth of a commodity price is computed within each trading partner.
The symbol X denotes the interaction between variables. Columns (1) and (2) report regressions estimated among exporting countries with average total
trade to GDP below and above the sample median (Closed and open respectively). Columns (3) and (4) compare the coefficients of the model between
countries with a ratio of private credit to GDP below and above the sample median (financially underdeveloped and financially developed), respectively.
Columns (5) and (6) do the same for countries with and without data on the development of foreign exchange derivative markets on the Bank for
International Settlements triennial survey of central banks. Finally, columns (5) and (6) compare, among countries with forex derivative market data, those
with forex turnover (as a fraction of GDP) below and above the sample median. Heterskedasticity robust standard errors reported in parenthesis, clustered
at the country (country-pair) level in Panel A (B). *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Table 10. Differences across Groups of Countries. Specification in Growth Rates
Closed Open No Forex Forex Low Forex High Forex Low Fin. High Fin.
Data Data Develop. Develop
(1) (2) (3) (4) (5) (6) (7) (8)
A: Global Trade Shares
-0.031*** -0.032*** -0.040*** -0.032*** -0.034*** -0.030 -0.039*** -0.029***
(log) Initial Export Share
(0.002) (0.003) (0.002) (0.002) (0.002) (0.004) (0.004) (0.001)
-0.013 0.021 -0.069 0.032 0.038 0.043 0.041 -0.042
Variance real commodity price
(0.039) (0.051) (0.055) (0.034) (0.047) (0.052) (0.055) (0.033)
CorrelationX SD Exch. Rate X -0.239** -0.189* -0.245** -0.150** -0.163** -0.021 -0.157* -0.251**
SD commodity price (0.070) (0.102) (0.104) (0.065) (0.078) (0.137) (0.094) (0.098)
Observations 22272 19986 14106 29143 14494 14649 11814 30087
R-squared 0.139 0.134 0.125 0.198 0.221 0.223 0.156 0.140
Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Test of Equality of interaction coef. 0.686 0.439 0.368 0.489
(1) (2) (3) (4) (5) (6) (7) (8)
B: Bilateral Trade Shares
-0.044*** -0.042*** -0.048*** -0.042*** -0.045*** -0.039*** -0.046*** -0.041***
(log) Initial Export Share
(0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001)
-0.028*** -0.014** -0.046* -0.022*** -0.033*** -0.008 -0.033*** -0.016***
Variance real commodity price
(0.005) (0.006) (0.024) (0.004) (0.005) (0.005) (0.007) (0.004)
CorrelationX SD Exch. Rate X -0.013 -0.070** -0.021 -0.034* -0.060** 0.020 -0.060** -0.019
SD commodity price (0.021) (0.032) (0.085) (0.018) (0.022) (0.031) (0.028) (0.023)
Observations 84488 72380 9377 147980 98624 58733 52540 96306
R-squared 0.192 0.180 0.251 0.179 0.200 0.167 0.193 0.178
Industry FE Yes Yes Yes Yes Yes Yes Yes Yes
Country-pair FE Yes Yes Yes Yes Yes Yes Yes Yes
0.136 0.881 0.035 0.258
In Panel A the dependent variable is the log of a commodity's share of total global country exports. In Panel B, it is the log of a commodity's share of total
country exports to each trading partner. Variance real commodity price is the variance of the growth rate of the ratio of a commodity's price to a country's
aggregate price level. In Panel A the commodity price is the average world price and in Panel B is the price in each trading partner. Correlation is the
correlation between a country's nominal exchange rate growth and the growth the world price of each commodity (across years from annual data). SD Exch
Rate is the standard deviation of a country's nominal exchange rate growth (within a year from monthly data) and SD commodity price is the standard
deviation of the growth rate of a commodity's international prices. In panel B, the growth of a commodity price is computed within each trading partner.
The symbol X denotes the interaction between variables. Columns (1) and (2) report regressions estimated among exporting countries with average total
trade to GDP below and above the sample median (Closed and open respectively). Columns (3) and (4) compare the coefficients of the model between
countries with a ratio of private credit to GDP below and above the sample median (financially underdeveloped and financially developed), respectively.
Columns (5) and (6) do the same for countries with and without data on the development of foreign exchange derivative markets on the Bank for
International Settlements triennial survey of central banks. Finally, columns (5) and (6) compare, among countries with forex derivative market data, those
with forex turnover (as a fraction of GDP) below and above the sample median. Heterskedasticity robust standard errors reported in parenthesis, clustered
at the country (country-pair) level in Panel A (B). *, **, and *** denote significance at 10, 5, and 1 percent, respectively.
Figure 1. Within Commodity Correlation Between Growth in Price Indexes from
BLS and Global Unit Values
A. Regression including all growth rates smaller than 1 in absolute value
1
.5
Growth Global Unit Value
0
-.5
-1
-.5 0 .5
Growth BLS price index
coef = .86079594, (robust) se = .11666881, t = 7.38
B. Regression including all growth rates smaller than 0.2 in absolute value
.2
Growth Global Unit Value
.1
0
-.1
-.2
-.2 -.1 0 .1 .2
Growth BLS price index
coef = .54809105, (robust) se = .12146552, t = 4.51
The figure shows the partial scatter plots of a regression between the growth rate of global unit values
and the growth rates of BLS import price indexes suring the period 1993-2000. Each regression included
commodity level fixed effects. The regression in Panel A (B) restricts the sample to include only
commodity-years where the growth rate of both price indicators was below 1 (0.2) in absolute value.
Each figure also shows the fitted regression line. The estimated coefficient for the partial regression, its
standard error, and the corresponding t-statistic are reported at the bottom of each scatter plot.
Figure 2. Average Correlations within Countries and Commodities
A. Average correlation within a country
B. Average correlation within a 3-digit SITC commodity
The two panels of the figure show the average correlation between a country's annual growth in its
nominal exchange rate against the dollar and the annual growth of a commodity's international
price within a country (Panel A) and within a 3-digit SITC industry.
Figure 3. Distribution of Correlation Values Across Countries and SITC
1.5
2
Density
1
.5
0
-1 -.5 0 .5 1
Correlation btw. growth in country nominal exchange rate and commodity price
Number of 25 75 Fraction of Fraction of Fraction of
Average Median
Observatio percentile perceitile 5% 10% 20%
57484 -0.1122 -0.0813 -0.3362 0.1066 0.1535 0.2176 0.3167
The figure shows the histogram of the correlation between a country's annual growth in its nominal
exchange rate against the dollar and the annual growth of a commodity's international price. The two
vertical lines at -0.9 and 0.9 show the truncation points used in the analysis (all correlations outside
this range were not included). The table at the bottom shows several summary statistics for the
distribution of these correlations: the total number of correlations included, as well as their average,
median, an 25th and 75th percentiles. It also shows the fraction of the correlations that are
statistically significant at the 5, 10, and 20 percent level.
Figure 4. Histogram of Coefficients for Natural Hedge Obtained after Dropping Countries, Country-Pairs, and
Industries
Global Shares
A. Dropping one exporting country at a time B. Dropping one commodity at a time
50
8
40
6
30
Density
Density
4
20
2
10
0
0
-3.5 -3.4 -3.3 -3.2 -3.1 -3 -3.3 -3.25 -3.2 -3.15 -3.1
Coefficient on CORR X SDXR X SDP Coefficient on CORR X SDXR X SDP
Mean = -3.18, Std. Dev= 0.08 Mean = -3.18, Std. Dev= 0.02
Bilateral Shares
C. Dropping one exporting country at a time D. Dropping one importing country at a time
60
60
40
40
Density
Density
20
20
0
0
-.9 -.85 -.8 -.75 -.9 -.85 -.8 -.75
Coefficient on CORR X SDXR X SDP Coefficient on CORR X SDXR X SDP
Mean = -0.84, Std. Dev= 0.01 Mean = -0.84, Std. Dev= 0.01
E. Dropping one commodity at a time
150
100
Density
50
0
-.87 -.86 -.85 -.84 -.83 -.82
Coefficient on CORR X SDXR X SDP
Mean = -0.84, Std. Dev= 0.005
Each panel shows the empirical distribution of the coefficient of natural hedge for global export shares (Panels A and B) and bilateral export
shares (Panels C to E) after dropping one exporting country at a time (Panels A and C), one importing country at a time (Panel D), and one
commodity at a time (Panels B and E). The mean and standard deviation of the empirical distribution are reported at the bottom of each
panel.
Appendix. Summary Statistics
(1) (2) (3) (4) (5) (6) (7)
Average Average Std. Average
Average Number of Minimum Maximum
Country Global Export Dev. of Nominal correlation
Export Share Commodity correlation correlation
Share Exch. Rate
Albania 0.0022 0.0002 416 0.67 -0.11 -0.83 0.74
Algeria 0.0020 0.0011 461 0.30 0.01 -0.65 0.65
Argentina 0.0014 0.0081 697 0.84 0.17 -0.69 0.68
Australia 0.0012 0.0147 738 0.27 -0.14 -0.78 0.54
Austria 0.0013 0.0122 715 0.31 -0.39 -0.90 0.48
Bahamas, The 0.0031 0.0005 309 0.00 0.00 0.00 0.00
Bahrain 0.0025 0.0005 387 0.00 -0.16 -0.76 0.41
Bangladesh 0.0026 0.0042 373 0.07 -0.05 -0.63 0.59
Barbados 0.0041 0.0001 211 0.00 0.00 0.00 0.00
Belgium 0.0011 0.0339 729 0.31 -0.39 -0.90 0.56
Belize 0.0043 0.0001 233 0.00 0.00 0.00 0.00
Benin 0.0078 0.0012 124 0.43 -0.15 -0.61 0.55
Bolivia 0.0031 0.0010 285 0.89 0.04 -0.78 0.89
Brazil 0.0014 0.0167 704 0.66 0.16 -0.56 0.60
Burkina Faso 0.0061 0.0004 133 0.43 -0.17 -0.61 0.41
Burundi 0.0104 0.0003 77 0.29 -0.36 -0.80 0.32
Cambodia 0.0047 0.0002 195 0.72 0.02 -0.68 0.57
Cameroon 0.0035 0.0011 287 0.43 -0.17 -0.70 0.56
Canada 0.0013 0.0351 739 0.12 -0.11 -0.71 0.78
Central African Republic 0.0032 0.0002 127 0.43 -0.20 -0.69 0.41
Chad 0.0155 0.0006 64 0.43 -0.08 -0.70 0.56
Chile 0.0015 0.0050 659 0.20 0.06 -0.77 0.84
China 0.0013 0.0509 739 0.20 0.16 -0.73 0.70
Colombia 0.0015 0.0020 640 0.14 0.11 -0.62 0.72
Congo, Dem. Rep. 0.0040 0.0016 168 1.47 -0.02 -0.50 0.49
Congo, Rep. 0.0055 0.0008 156 0.43 -0.15 -0.70 0.56
Costa Rica 0.0021 0.0012 462 0.06 0.12 -0.65 0.69
Cote d'Ivoire 0.0024 0.0025 398 0.43 -0.18 -0.70 0.56
Croatia 0.0017 0.0009 569 0.48 -0.08 -0.88 0.84
Cyprus 0.0016 0.0004 560 0.25 -0.44 -0.90 0.48
Denmark 0.0012 0.0112 714 0.30 -0.39 -0.90 0.48
Djibouti 0.0074 0.0001 111 0.00 0.00 0.00 0.00
Dominican Republic 0.0023 0.0023 395 0.47 0.12 -0.59 0.81
Ecuador 0.0019 0.0020 508 0.58 -0.01 -0.79 0.79
Egypt, Arab Rep. 0.0016 0.0010 628 0.31 0.08 -0.64 0.73
El Salvador 0.0037 0.0006 268 0.18 0.22 -0.68 0.86
Equatorial Guinea 0.0183 0.0003 54 0.43 -0.08 -0.57 0.50
Ethiopia 0.0045 0.0008 216 0.23 -0.12 -0.78 0.38
Finland 0.0014 0.0084 707 0.30 -0.40 -0.89 0.52
France 0.0012 0.0597 729 0.30 -0.38 -0.90 0.60
Gabon 0.0047 0.0017 208 0.43 -0.17 -0.69 0.56
Gambia, The 0.0050 0.0002 104 0.36 0.08 -0.66 0.54
Germany 0.0012 0.0872 706 0.30 -0.39 -0.90 0.55
Ghana 0.0031 0.0012 267 0.42 -0.09 -0.71 0.63
Continues
(1) (2) (3) (4) (5) (6) (7)
Average Average Number of Average Std. Average Minimum Maximum
Country
Export Share Global Export Commodity Dev. of Nominal
correlation correlation correlation
Share Exch. Rate
Greece 0.0013 0.0040 713 0.28 -0.31 -0.83 0.53
Guatemala 0.0028 0.0013 356 0.39 0.23 -0.61 0.88
Guyana 0.0039 0.0005 139 0.63 0.13 -0.61 0.61
Haiti 0.0051 0.0005 191 0.32 -0.19 -0.74 0.32
Honduras 0.0033 0.0012 299 0.17 0.07 -0.44 0.72
Hong Kong, China 0.0012 0.0098 720 0.01 -0.19 -0.74 0.54
Hungary 0.0013 0.0041 706 0.24 -0.29 -0.82 0.67
Iceland 0.0021 0.0013 428 0.23 -0.13 -0.67 0.57
India 0.0011 0.0110 724 0.18 -0.07 -0.77 0.53
Indonesia 0.0014 0.0103 709 0.42 0.03 -0.86 0.85
Iran, Islamic Rep. 0.0016 0.0017 603 0.56 -0.14 -0.72 0.52
Ireland 0.0013 0.0078 710 0.30 -0.40 -0.89 0.45
Israel 0.0010 0.0034 675 0.26 0.05 -0.77 0.90
Italy 0.0013 0.0513 744 0.30 -0.34 -0.90 0.45
Jamaica 0.0015 0.0004 389 0.27 0.02 -0.53 0.68
Japan 0.0013 0.0653 739 0.34 -0.14 -0.75 0.65
Jordan 0.0021 0.0007 467 0.11 0.02 -0.61 0.68
Kenya 0.0023 0.0015 424 0.32 -0.20 -0.90 0.40
Korea, Rep. 0.0013 0.0187 724 0.19 -0.02 -0.82 0.81
Kuwait 0.0020 0.0010 501 0.09 -0.21 -0.71 0.54
Lao PDR 0.0056 0.0001 177 0.78 0.08 -0.56 0.57
Latvia 0.0021 0.0006 463 0.27 -0.26 -0.81 0.51
Lebanon 0.0015 0.0004 481 0.68 0.30 -0.35 0.76
Liberia 0.0037 0.0006 185 0.82 -0.03 -0.72 0.85
Libya 0.0035 0.0010 283 0.21 -0.08 -0.61 0.50
Lithuania 0.0019 0.0007 519 0.15 -0.08 -0.89 0.81
Macao, China 0.0022 0.0009 446 0.01 -0.15 -0.75 0.67
Madagascar 0.0045 0.0015 212 0.49 0.16 -0.41 0.63
Malawi 0.0073 0.0010 136 0.44 -0.04 -0.46 0.46
Malaysia 0.0014 0.0117 718 0.18 -0.05 -0.78 0.81
Mali 0.0045 0.0003 198 0.43 -0.16 -0.70 0.50
Malta 0.0020 0.0003 479 0.22 -0.42 -0.90 0.35
Mauritania 0.0064 0.0004 153 0.28 -0.25 -0.74 0.33
Mauritius 0.0029 0.0006 321 0.21 -0.26 -0.79 0.41
Mexico 0.0013 0.0108 724 0.27 0.35 -0.41 0.88
Mongolia 0.0039 0.0006 237 1.20 -0.03 -0.69 0.70
Morocco 0.0016 0.0031 596 0.22 -0.36 -0.89 0.47
Mozambique 0.0031 0.0004 319 0.61 0.18 -0.50 0.64
Myanmar 0.0026 0.0009 357 0.17 -0.33 -0.90 0.58
Nepal 0.0033 0.0006 295 0.18 0.04 -0.72 0.52
Netherlands 0.0012 0.0443 728 0.31 -0.38 -0.89 0.55
New Zealand 0.0014 0.0055 682 0.29 -0.26 -0.81 0.59
Niger 0.0009 0.0001 242 0.43 -0.20 -0.71 0.50
Nigeria 0.0025 0.0015 397 0.84 -0.02 -0.71 0.80
Norway 0.0014 0.0071 687 0.27 -0.41 -0.90 0.46
Pakistan 0.0017 0.0035 569 0.14 -0.29 -0.86 0.47
Panama 0.0016 0.0005 565 0.00 0.00 0.00 0.00
Papua New Guinea 0.0038 0.0037 223 0.29 -0.10 -0.70 0.45
Continues
(1) (2) (3) (4) (5) (6) (7)
Average Average Number of Average Std. Average Minimum Maximum
Country
Export Share Global Export Commodity Dev. of Nominal
correlation correlation correlation
Share Exch. Rate
Paraguay 0.0025 0.0010 376 0.48 0.07 -0.63 0.70
Peru 0.0016 0.0032 570 1.06 0.17 -0.65 0.68
Philippines 0.0014 0.0055 679 0.23 -0.08 -0.76 0.81
Poland 0.0013 0.0063 721 0.60 0.17 -0.65 0.78
Portugal 0.0014 0.0061 719 0.28 -0.36 -0.86 0.45
Qatar 0.0033 0.0005 302 0.00 0.00 0.00 0.00
Romania 0.0015 0.0025 639 0.84 -0.17 -0.72 0.35
Rwanda 0.0105 0.0005 86 0.33 -0.03 -0.70 0.53
Samoa 0.0125 0.0003 78 0.20 -0.26 -0.76 0.38
Saudi Arabia 0.0015 0.0038 642 0.01 0.19 -0.68 0.86
Senegal 0.0040 0.0017 231 0.43 -0.16 -0.70 0.55
Seychelles 0.0063 0.0001 150 0.17 -0.34 -0.89 0.45
Sierra Leone 0.0022 0.0002 204 0.92 0.28 -0.51 0.69
Singapore 0.0013 0.0088 736 0.13 -0.23 -0.71 0.69
Slovenia 0.0016 0.0022 618 0.36 0.02 -0.85 0.87
South Africa 0.0010 0.0071 730 0.34 -0.19 -0.83 0.52
Spain 0.0011 0.0183 738 0.30 -0.39 -0.90 0.34
Sri Lanka 0.0019 0.0025 490 0.11 -0.19 -0.81 0.54
St. Kitts and Nevis 0.0020 0.0002 461 0.00 0.00 0.00 0.00
Sudan 0.0037 0.0022 262 1.06 0.02 -0.52 0.62
Suriname 0.0017 0.0003 177 1.22 0.01 -0.52 0.57
Sweden 0.0013 0.0164 724 0.27 -0.35 -0.86 0.42
Switzerland 0.0010 0.0178 731 0.35 -0.37 -0.90 0.52
Syrian Arab Republic 0.0022 0.0008 459 0.21 0.11 -0.75 0.72
Tanzania 0.0038 0.0012 252 0.39 0.29 -0.34 0.81
Thailand 0.0013 0.0098 717 0.20 -0.10 -0.79 0.58
Togo 0.0064 0.0008 149 0.43 -0.14 -0.70 0.56
Trinidad and Tobago 0.0031 0.0006 314 0.18 0.09 -0.63 0.59
Tunisia 0.0016 0.0017 597 0.26 -0.37 -0.83 0.44
Turkey 0.0014 0.0078 716 0.29 -0.13 -0.67 0.53
Uganda 0.0073 0.0007 136 0.80 0.23 -0.40 0.65
United Kingdom 0.0012 0.0460 749 0.29 -0.33 -0.84 0.46
Uruguay 0.0015 0.0013 583 0.19 0.32 -0.46 0.81
Venezuela, RB 0.0016 0.0021 623 0.67 0.02 -0.62 0.85
Yemen, Rep. 0.0059 0.0004 168 0.75 0.20 -0.62 0.74
Zambia 0.0051 0.0009 187 1.08 0.02 -0.69 0.51
Zimbabwe 0.0021 0.0012 447 0.41 -0.22 -0.77 0.60
Average export share is the average across commodities (3-digit SITC classification) of the share of a country's
total exports represented by each of the commodities that a country exported between 1985 and 2000. Average
global export share is the average across commodities of the share of global commodity exports represented by
each of the commodities exported by a country between 1985 and 2000. Number of commodities is the number of
3 digit SITC sector exported by a country during the same period. Average standard devation of nominal
exchange rate is the average of the within year standard deviation of the monthly growth of a country's nominal
exchange rate against the US dollar. Average correlation is the average correlation between a country's nominal
exchange rate growth and the growth the world price of each commodity (across years from annual data) across
the commodities exported by a country. The minimum and maximum (within a country) of the same correlation
are reported in columns (6) and (7).