Intensive and Extensive Margins of Exports and Real ...

Intensive and Extensive Margins of Exports and Real
Exchange Rates∗

Mariana Colacelli
Barnard College, Columbia University

Economics Department
[email protected]

April, 2010

Abstract

Current research in international trade literature centers on the importance of both
the intensive margin (i.e. the volume exported by an exporter) and extensive mar-
gin (i.e. new exporters) to model export behavior. In this paper I empirically study
the behavior of the intensive and the extensive margin to address the elasticity puzzle
in international economics. (The puzzle arises because the elasticity of substitution
between domestic and foreign goods has no consensus estimate and is thought to be
low in international macro models and high in trade models.) In particular, I study if
mechanisms related to sectoral differences in market structure or development status of
the traders have potential to address the elasticity puzzle. Lastly, I further study the
extensive margin of trade and its connection to credit constraints. A bilateral trade
sample of 136 countries is used for the period 1981-1997 and bilateral export responses
to bilateral real exchange rate fluctuations are decomposed into intensive and exten-
sive margin responses. I find that the extensive margin of trade has a significant role
in overall yearly export responses to real exchange rate fluctuations. Moreover the
extensive margin shows a larger role among less substitutable exports (as predicted
by Chaney, 2008). Taken together, these findings suggest that sectoral differences in
market structure may account for higher estimates of the substitution elasticity for dif-
ferentiated sectors. Alternatively, the development status of the exporter is found to
affect export responses suggesting that elasticity estimates would be higher for more
developed exporters. Lastly, credit constraints, measured as external finance depen-
dence of the sector, are found to play a significant role in extensive margin behavior.
Given the importance of the extensive margin in overall export responses, the evidence
on credit constraints may provide another explanation for heterogeneity in estimates of
the substitution elasticity.

∗Jae Bin Ahn provided excellent research assistance. I thank David Blackburn, Ariel Burstein, Don Davis,
Amit Khandelwal, Philippe Martin, Marc Melitz, Ken Rogoff, Francesco Caselli, LACEA 2009 conference,
CSWEP-AEA 2010 Workshop, MEA 2010 conference, Bowdoin College, Columbia Macro Lunch and Notre
Dame seminar participants for their comments. I thank David Weinstein for kindly providing the 6-digit
trade data. The Institute of Latin American Studies at Columbia University provided funding for part of
this project. All remaining errors are my own.

1

1 Introduction

The elasticity of substitution between domestic and foreign goods is a key parameter for

models in international economics. Different values for this elasticity importantly affect

the predictions of models that have an international channel. Calibrations of international

macroeconomic models that study real business cycles suggest that values of this elasticity

between 1 and 2 are appropriate. Alternatively, models that study trade patterns and the

impact of tariffs and trade liberalizations suggest that values of the elasticity of substitution
between 10 and 15 are appropriate.1 This lack of consensus on the elasticity of substitution

in the theoretical/calibration literature is also found in the vast empirical literature on

this parameter. Solutions to this puzzle have been offered concerning aggregation issues,
frictions in quantities traded, and the source of variation used in the elasticity estimation.2

Current research in the international trade literature centers on the importance of both

the intensive margin (i.e. the volume exported by an exporter) and extensive margin (i.e.

new exporters) to model export behavior. In this paper I empirically study the behavior

of the intensive and extensive margins in order to address the described elasticity puzzle.

First I study if margin behavior is able to solve the elasticity puzzle as proposed by Ruhl

(2008). As I uncover patterns that do not agree with Ruhl’s theory, I further examine

the data in search of alternative solutions to the elasticity puzzle. In particular I focus on

1 I use the stated values for the elasticity puzzle that Ruhl (2008) offers.
2 Imbs and Mejean (2009) propose that the level of aggregation used in the elasticity estimation can
explain the low aggregate estimates and large micro estimates. Drozd and Nosal (2008) build a model
where frictions in the adjustment of quantities traded explain low short run elasticities and higher long run
elasticities. Ruhl (2008) notes that elasticities estimated with high frequency data (time series) are small and
those estimated with cross-country data (or before-after trade liberalizations) are large. Ruhl proposes that
the different sources of variation used in the estimation can explain the puzzle given that temporary shocks
are predicted to deliver smaller elasticities than permanent changes, building a model with such predictions
where the decision to become an exporter (affecting the extensive margin of trade) is central for the results.

2

how the model of Chaney (2008) can shed light on the puzzle based on the type of sector
traded. In a model that directly incorporates extensive margin effects, Chaney (2008) adds
market structure to a gravity theory of trade with firm heterogeneity and predicts that
the extensive (intensive) margin response of a sector to changes in variable trade barriers is
diminished (amplified) by the elasticity of substitution of the sector. Moreover, I investigate
if the development status of exporters plays a role in explaining the elasticity puzzle. Last,
I further study the extensive margin of trade and its connection to credit constraints.

This paper uses yearly fluctuations in bilateral real exchange rates as measures of shocks
or changes in variable trade costs. Depending on the model that we have in mind, export
responses to real exchange rate fluctuations will deliver an estimate of the elasticity of
substitution between domestic and foreign goods. For example, simple models of trade
with monopolistic competition and representative firms predict that exports respond to
variable trade costs as a linear function of the substitution elasticity. These models imply
that the estimation strategy in this paper allows us to recover the elasticity of substitution.
Moreover, the estimated export responses to real exchange rate fluctuations are further
decomposed into extensive and intensive margin responses.

Following Hummels and Klenow (2005), I define the bilateral extensive margin of exports
as a weighted count of the exporter country’s varieties exported to the importer and I
define the bilateral intensive margin as the exporter country’s relative volume of exports.
I define a variety as each of the 440 4-digit sectors from the Standard International Trade
Classification (SITC) Revision 2 and use a bilateral trade sample of 136 countries for the
period 1981-1997.3

In a related exercise Kehoe and Ruhl (2009) build an alternative measure of bilateral
extensive margin using a country-pair specific cutoff to define if a 4-digit sector is traded or
not. Their study offers evidence consistent with Ruhl (2008) as their measure of extensive
margin responds to permanent but not transitory shocks in their sample of around twenty
country pairs. Note that their exercise is different from the present paper not only in their

3 Hummels and Klenow’s data include exports in 1995 from 121 countries to 59 importers in over 5,000
6-digit product categories. Section 3.4 performs a robustness check with 6-digit trade.

3

measure of extensive margin but also in that they focus on a selected group of countries and
episodes. I find that Kehoe and Ruhl’s (2009) results about the unimportance of extensive
margin response to transitory shocks may critically depend on their sample.

My results indicate that the extensive margin of trade plays a significant role in export
adjustments at the yearly frequency. In particular, I find that the extensive margin explains
on average 70% of overall export responses to real exchange rate fluctuations among the four
country samples with significant export responses. To the extent that the analysis in this
paper mostly relies on transitory as opposed to permanent shocks by using real exchange
rate fluctuations, results are inconsistent with Ruhl (2008) as I find that the extensive
margin plays an important role in the yearly changes of exports. This evidence suggests
that Ruhl’s theory of transitory versus permanent shocks and the extensive margin may
not be key in solving the elasticity puzzle.

Chaney (2008) predicts that the degree of substitutability of goods magnifies the inten-
sive margin response and diminishes the extensive margin response to changes in variable
trade costs. I interpret my results as supportive of Chaney’s predictions. This evidence
points to an alternative solution to the elasticity puzzle: as the evidence mostly indicates
that the extensive margin of trade is more responsive in less substitutable sectors, and
given that the extensive margin seems to dominate the export response, we may explain the
puzzle if the high estimated elasticities in the literature are associated with differentiated
sectors trade (and low estimates are linked to homogeneous sectors trade). It may be that
the higher estimated elasticities correspond to sectors with lower elasticity of substitution
where the extensive margin is relatively larger. Further study of existing estimates in the
vast empirical literature is needed in determining the validity of this proposed hypothesis.

Further, results indicate that export responses from less developed countries are smaller
than those from high-income countries (0.04 versus 0.12). This evidence points to an alter-
native solution to the elasticity puzzle. If most high estimates from the literature correspond
to samples of high-income exporters (and low estimates correspond to less developed ex-
porters) we may propose that the puzzle relates to a development status story. It may
be that higher estimated elasticities mostly correspond to more developed exporters. Again,

4

further study of existing estimates in the vast empirical literature is needed in determining
the validity of this proposed hypothesis.

Last, I find evidence in support of the importance of sectoral credit constraints (mea-
sured as external finance dependence) on the extensive margin response, with strongest
evidence in the case of entry among homogeneous sectors after a large depreciation. Such
evidence is consistent with recent findings in Manova, Wei and Zhang (2009) where desti-
nation markets and product scope of Chinese firms are affected by credit constraints. This
suggests that in solving the elasticity puzzle credit constraints may also play a role.

Overall the evidence presented in this paper suggests that the elasticity puzzle may well
be solved by considering the sectoral composition of trade and the development status of
the traders as these dimensions are found to deliver heterogeneous export responses to real
exchange rate fluctuations.

1.1 Trade Theories and Empirical Implications

Several models are used in framing the empirical analysis from Section 3. A brief description
of such models and the studied empirical implications follow. Note that the focus is on
implications related to variable trade costs as those arguably better parallel the studied
real exchange rate fluctuations.

The model from Krugman (1980) where all firms are identical includes only a variable
cost of trade. This structure, without fixed trade costs, delivers no extensive margin of
trade as every firm exports to every country. Given that consumers have a preference for
variety, identical countries trade the differentiated goods (by country of origin) even though
there are costs of trade. The implied gravity equation from this model is:

 =  ∗  ∗  (1)
( )

where  represents the exporter country,  represents the importer country and  is the

elasticity of substitution between domestic and foreign goods. Empirically, Krugman’s

model implies that the substitution elasticity can be estimated with the partial derivative

5

of bilateral exports with respect to variable trade barriers:

− ln() =  (2)
 ln()

Within this model a higher  implies a higher impact of trade barriers on bilateral exports.

Chaney (2008) proposes a model of heterogeneous firms (through a random productiv-

ity shock based on a Pareto distribution for productivity) with variable and fixed costs of

exporting. The more productive firms select into exporting delivering an extensive margin

of trade. Chaney derives a bilateral trade gravity model using multiple asymmetric coun-

tries and trade barriers where the function () affects bilateral export responses to trade

barriers:

 =  ∗  ∗  (3)
( )()

Contrary to Krugman, Chaney’s model predicts that the export response to variable trade

barriers does not depend on :

− ln() =  (4)
 ln()

where  is the shape parameter of the Pareto distribution.4 However, this model implies

that margin responses to changes in variable trade barriers depend on . In particular

the response of the extensive margin of trade to variable trade costs is predicted to be

amplified for lower , and the response of the intensive margin of trade is predicted to

be diminished for lower . The intuition of this result relates to the fact that in less

substitutable sectors (low ) firms capture a more even market share (as market shares

are less responsive to productivity differences for less substitutable sectors). When variable

trade barriers decrease, new entrants to the export market in less substitutable sectors will

increase the export volume relatively more than in more substitutable sectors, implying a

larger extensive margin response.

4 Chaney’s model predicts that the export response to changes in fixed trade barriers increases for lower
.

6

Lastly, Ruhl (2008) builds a model where firms have heterogeneous productivity, firms
are exposed to permanent and temporary productivity shocks and it includes variable and
fixed costs of exporting. As in Chaney (2008), the more productive firms select into export-
ing delivering a model with extensive and intensive margins. Export status is affected by
permanent changes in tariffs and/or (persistent but) transitory productivity shocks. From
model calibration, Ruhl determines that a transitory productivity shock has a small impact
on the extensive margin of trade (and implies a small ), where a permanent change in
tariffs (of the same magnitude of the transitory productivity shock) has a large effect on the
extensive margin (and implies a large ). Again, to the extent that yearly fluctuations in
real exchange rates mostly capture transitory as opposed to permanent shocks,5 we inter-
pret Ruhl’s prediction as implying a small extensive margin response to yearly fluctuations
in real exchange rates.6

2 Extensive and Intensive Margins of Trade

Recent literature has studied the relative importance of extensive and intensive margins
of trade in exports. Descriptive studies such as Bernard et al (2009) and Eaton et al
(2007) find that yearly changes in exports are mostly driven by the intensive margin. They
find that the contribution from the extensive margin rises for longer time horizons and
for the cross-section.7 Using margin measures to study cross-country patterns of exports,
Hummels and Klenow (2005) determine that their measure of the extensive margin at the
exporter level allows them to account for more than 60% of the greater exports of larger
economies. Lastly, in linking the extensive margin behavior with the business cycle, Broda
and Weinstein (2007) find that product creation (at the barcode level) is pro-cyclical at

5 Section 3.3 further supports this claim.
6 A related model is Bilbiie at al (2007) which uses the extensive margin of trade as a propagation
mechanism in an RBC framework. Assuming sunk entry costs and a time-to-build lag, when aggregate
productivity increases permanently for all firms, the model delivers that GDP first increases because of the
intensive margin and later entry activates the extensive margin which further increases GDP.
7 Bernard et al use US firms’ trade data (10-digit) for 1993-2004 and define the extensive margin at the
firm-product level. Eaton et al use Colombian firms’ exports for 1996-2005 and define the extensive margin
as the number of exporting firms.

7

quarterly business cycle frequency.8
In the spirit of Hummels and Klenow (2005), who build on Feenstra (1994), I define

the bilateral extensive margin as the weighted count of the 4-digit sectors in which the
exporter exports to the importer in a given year. Weights correspond to the relevance of
the exported sectors in total exports from the world to the importer (excluding exports
from the exporter) in a given year. More precisely , the extensive margin between
exporter  and importer  in year , is:

X


 = X (5)




where  is the "rest of the world" by including all exporters to  except for .  is the
set of sectors  in which exporter  exports to  in year .  is the set of all 440 sectors
included. The numerator of  measures exports from the rest of the world to the
importer  in those sectors  in which  exports to  in year . The denominator includes
all exports from the rest of the world to  in all sectors  in year . If all sectors  have
equal importance for  during , then  is the fraction of sectors in which  exports to
 during . In general weights on sectors reflect their relevance in ’s exports to . 

is between 0 and 1.

The bilateral intensive margin measures exports from the exporter to the importer rela-

tive to total exports to the importer (excluding exporter’s exports) in those sectors in which

the exporter exports to the importer in a given year. In particular:

X


   = X (6)




8 Broda and Weinstein derive this finding with ACNielsen Homescan data which include all goods with
barcodes (~40% of all expenditure on goods in CPI) purchased by a sample of approximately 55,000 US
households across 23 cities between 1999 and 2003 measured at the quarterly level.

8

 measures the intensive margin between  and  in  by comparing exports from  to 

in  with exports from the rest of the world to  in  in the sectors  in which  exports to

 in .  is positive and can be below or above 1.

As defined by Hummels and Klenow (2005), the product of both margins delivers overall

bilateral exports from  to  in , , as the ratio of exports from  to  in  over exports

from the rest of the world to  in  (numerator of  is equal to denominator of ):

X


 =  ∗    = X (7)




Note that Hummels and Klenow (2005) work on decomposing the measure of  into

extensive margin and intensive margin. This work will study the relative importance of

the trade margins on the absolute level of bilateral trade flows, , as opposed to the

relative trade measure . Such strategy will allow us to tie our findings with previous

work on the response of  to fluctuations in the bilateral real exchange rate and on the

substitution elasticity. In order to do such decomposition of the variation of bilateral trade

flows  we state the relationship between  and . Note that  is the numerator

of the defined  in equation (6) (and is also the numerator in equation (7)). Therefore,
X

defining  as , we can express  as:



X (8)
 =  ∗  =  ∗  ∗ 



Given the definitions of margins used, bilateral trade flows  can be decomposed into

extensive margin, intensive margin and . Therefore variation in bilateral trade flows

 will be explained by variation in the margins of trade (as defined by Hummels and

Klenow (2005)) and also variation in , where  measures exports from the rest of

the world to  in all sectors in year .

9

2.1 Descriptive Statistics for Margins

As in Colacelli (2009), I build a country-pair level sample including 136 countries for the
period between 1981 and 1997 where the bilateral trade flows are those compiled by Feenstra
(2000).9 The sample includes 13,860 country pairs and 140,013 bilateral-level observations.
The country-pair level data on margins are built with approximately 8 million observations
from 4-digit SITC Rev. 2 bilateral sector trade data. Table 1 provides summary statistics
for the described measures of bilateral extensive margin, bilateral intensive margin, bilateral
exports , and .

Statistics are reported for five different samples of countries, depending on the develop-
ment status of the exporter and importer. The World sample includes all 136 exporters and
importers. Following the World Bank 2006 classification of countries based on 2004 GNI
per capita, I classify 34 countries as high income and 102 countries as developing countries,
as in Colacelli (2009). The “HI” sample includes exporter and importer countries from the
high-income group, the “HI&MIX” sample includes exporters from the high-income group
but importers from the high-income and developing country groups, and the “DC” and
“DC&MIX” samples are built similarly for developing countries.

Statistics in Table 1 are reported by type of product exported for the five country
samples. As in Colacelli (2009), following Rauch (1999), the focus is on homogeneous and
differentiated bilateral trade. Rauch classifies export goods by the availability of information
on their price. Differentiated products are defined as those without a reference price or
“branded” — i.e. their price can be quoted once mentioning the manufacturer. Homogeneous
products are those traded on organized exchanges where reference prices are quoted (for
example in the London Metal Exchange). For simplicity this paper ignores the middle
category of sectors, reference-price products, which are not "branded" and have prices
listed only in trade publications and may have specialized traders who centralize price
information.

As described above,  is the weighted fraction of sectors in which  exports to 

9 The raw bilateral trade data is in thousands of US dollars. I obtain the 1995 dollar measure using the
US GDP deflator.

10

during . The mean exporter in the World sample exports to its typical destination in
33% of all the weighted 4-digit sectors imported by this destination during an average year.
The mean extensive margin increases to 59% when we restrict the sample to high-income
exporters and importers, and it is only 22% for the DC sample. When focusing on exports
in homogeneous sectors only, the bilateral extensive margin is reduced approximately to
half of that for all exports. For differentiated sectors the mean bilateral extensive margin
is similar to that for all exports in high-income country samples but is reduced to around
15% for samples with developing countries.

 was defined as the ratio of exports from  to  in  over exports from the rest of the
world to  in  in the sectors  in which  exports to  in . The mean exporter in the World
sample during an average year exports to its typical destination 54% of the volume exported
by the rest of the world to this destination in those sectors. When focusing on the HI sample
this statistic is just 6% which reflects the smaller concentration in exports among high-
income countries. When looking at the DC sample the average bilateral intensive margin
increases to 90%. Focusing on homogeneous sectors only, the bilateral intensive margin
is higher than 100% for all country samples (152-426%) reflecting high bilateral export
concentration in such sectors. On the other hand, differentiated sectors show bilateral
intensive margins between 6% and 27% for all samples.

3 Contribution of Extensive and Intensive Margins

In order to study the composition of the margin response to real exchange rate fluctuations
I use a set of bilateral gravity equations. This approach is a combination of standard tech-
niques first developed by Tinbergen (1962) and the work by Hummels and Klenow (2005).
Bilateral trade as well as bilateral margins of trade are explained, in separate equations, by
the scale of the exporter and importer and measures of trade resistance between trading
partners including the measured bilateral real exchange rate. The estimating equation for

11

bilateral trade flows  is:

ln() = 1 ln() + 2 ln() + 3 ln() +   ln() (9)
4

+  ln(d) +  +   + 


where  is the GDP of the importer at time ,  is the GDP of the exporter at time ,

 represents country-pair specific measures of trade resistance that affect bilateral trade,

  represents a time specific effect on bilateral trade, and  represents country-pair-year


specific error. As additional measures of time-varying exporter and importer activity we add

GDP per capita of the exporter and the importer represented as  and . As mentioned,
this paper interprets the measured real exchange rate between a pair of countries, d,
as a measure of variable trade cost between them.10

In order to decompose the bilateral export response  we simply estimate the following

parallel models for the extensive and intensive margin defined. The parallel estimating

equation for the extensive margin is:

ln() =   ln() + 2 ln() + 3 ln() + 4  ln() (10)
1

+  ln(d) +  +   + 


And similarly, the estimating equation for the intensive margin is:

ln(   ) =   ln() + 2 ln() +   ln() +   ln() (11)
1 3 4

+  ln(d) +  +    + 


Given that  is defined as  ∗  ∗ , it must be that:

10 Colacelli (2009) and Bayoumi (1999) follow a similar procedure to obtain the real exchange rate elasticity
of exports. Colacelli (2009) (and also Section 3.3 of this paper) includes a time varying measure of trade
regulations (presence of regional trade agreements) without changing the estimated patterns of the elasticity
 . Feenstra (1989) finds supporting evidence for the symmetric pass-through of tariffs and exchange rates
on US import prices of Japanese cars, trucks and motorcycles.

12

 =  +  +  (12)

because of linearity of OLS.  represents the coefficient of ln(d) in a parallel es-
timating equation where the dependent variable is ln(). Property (12) of the defined
margins allows for a simple decomposition of their contribution to bilateral trade responses
to real exchange rate fluctuations.

The estimated  from equation (9) is a measure of  under the Krugman (1980) model
discussed in Section 1.1. Broda and Weinstein (2006) pursue an alternative identification
strategy to uncover  using prices and quantities for US imports between 1972 and 2001.
The authors estimate a supply and demand system for US imports identifying  with cross-
country variation in prices of 10-digit sector flows. They assume that each exporter of a
given 10-digit sector sells a different variety of that good.

The chosen identification strategy in this paper relies on bilateral real exchange rates
as opposed to direct measures of prices or trade costs. The reason for this choice is that to
increase country coverage we use more aggregated trade data (4- or 6-digit). Prices for such
level of aggregation are less precise (as prices are estimated with volume and units traded
which are more heterogeneous when more aggregated). Measures of real exchange rate are
available and represent variation in prices or trade costs bilaterally. Moreover, if we were
to use 4-digit price data we would still need to rely on an identification strategy similar to
Broda-Weinstein’s to get around the simultaneity problem present when export volumes and
prices are jointly determined.11 The key identifying assumption used is that the variation in
the bilateral real exchange rate (beyond exporter-importer and time fixed-effects) is not de-
termined by variation in bilateral exports or bilateral extensive and intensive margins. Note
that this identification strategy is pursued at two levels. The first identifying assumption,
used at the aggregate level, requires that the variation in the bilateral real exchange rate
(beyond exporter-importer and time fixed-effects) is not determined by variation in bilateral
exports or bilateral extensive and intensive margins. The second identifying assumption,

11 Such a strategy would still face the issue that inferred prices are mechanically determined from trade
value and unit information.

13

used at the sectoral level, requires that the variation in the bilateral real exchange rate (be-
yond exporter-importer and time fixed-effects) is not determined by variation in bilateral
sectoral exports or bilateral sectoral extensive and intensive margins.

To perform this analysis, on top of the described margins variables and data, I use
measures of exchange rate, income and GDP deflators obtained from the World Development
Indicators (2001). Real exchange rate, d, is measured by the nominal exchange rate
for each country in the pair and GDP deflators such that an increase in d represents a
real depreciation of the exporter country  with respect to the importer country . Colacelli
(2009) details measurement issues of true real exchange rate, , and how the inclusion
of country-pair fixed effects in the estimation solves it. Descriptive statistics for d for
the defined samples are detailed in Colacelli (2009).

3.1 Estimation Results on Margins’ Contribution

Table 2 presents estimation results for equations (9), (10) and (11) in columns (1), (2) and
(3) respectively for the World sample. Column (4) presents the estimated  from equation
(12). Results for overall bilateral trade are reported in the top panel, for homogeneous
bilateral trade in the middle panel and for differentiated bilateral trade in the bottom
panel.

Tables 3, 4, 5 and 6 present parallel results for samples HI, HI&MIX, DC and DC&MIX
respectively. Four out of the five country samples studied (all except HI) show that the
extensive margin of trade plays a significant role in the overall export response to yearly real
exchange rate fluctuations. The extensive margin response varies from 37% (0.045/0.122)
of the export response in HI&MIX to 91% (0.039/0.043) in DC&MIX.

Table 2 shows that bilateral exports of differentiated goods have a larger significant
response than those of homogeneous goods (0.08 versus 0.03) for the World sample. This
table indicates that the intensive margin has no contribution to the trade response of either
homogeneous or differentiated trade (its contribution is significant but negative for differen-
tiated exports) in the World sample. The extensive margin is shown to explain 46% of the
differentiated trade response (0.037/0.080) and essentially none of the homogeneous trade

14

response in the World sample.
Tables 3 and 4 for samples HI and HI&MIX show an extensive margin contribution to

trade responses of differentiated goods of 37% for HI (0.161/0.433) and 13% of the trade
response for HI&MIX (0.023/0.178). The intensive margin has no contribution to the trade
response of differentiated sectors in either sample. The extensive margin explains 68% of
the trade response of homogeneous goods in the HI&MIX sample (no role of IM), and the
intensive margin explains over 100% of the trade response of homogeneous goods in the HI
sample (no role of the EM).

Tables 5 and 6 present results for samples DC and DC&MIX. Differentiated exports show
significant responses to real exchange rate fluctuations in both samples while homogeneous
exports have insignificant trade responses. The extensive margin contributes around 50%
of the trade response of differentiated exports in both samples and the intensive margin
shows no significant contribution.

A summary of the above results is presented in Table 7, where I include the estimated
coefficients  ,  and  that are significantly different from zero from estimating
equations (9), (10) and (11). The last column of the table computes the contribution of
the extensive margin to the export response as discussed above. Table 7 allows for a simple
inspection of the results across samples regarding the overall importance of the extensive
margin contribution in the export response. This evidence is not consistent with Ruhl’s
(2008) prediction of unimportant extensive margin response to temporary shocks such as
those measured by the yearly real exchange rate change. Therefore Ruhl’s proposed solution
to the elasticity puzzle seems less promising.

In search for an alternative solution to the elasticity puzzle we first analyze the evidence
on margins and sectors. In order to test Chaney’s (2008) prediction regarding the impact
of the elasticity of substitution on the margin responses we perform a non-linear test of the
estimated coefficients. Chaney predicts, as explained in Section 1.1, that the response of
the extensive (intensive) margin of trade to variable trade costs is amplified (diminished)
for lower . (As others in the literature we associate a higher  with homogeneous sectors
and a lower  with differentiated sectors.) Note that the intensive margin is almost never

15

significant in terms of contribution to export responses. The intensive margin response is
significantly different from zero only for homogeneous sectors in the HI sample, which we
can interpret as being in favor of Chaney’s prediction on the intensive margin (as the inten-
sive margin is more important for homogeneous than differentiated sectors in this sample).
However, given the evidence, we focus on the test of the contribution from the extensive
margin of trade between homogeneous and differentiated sectors. The null hypothesis for
this test is that the extensive margin contribution in homogeneous sectors is larger than in
differentiated sectors, for each of the five samples studied. The null in the corresponding no-
tation with the estimated coefficients reads as (  )   (  ) . Rejection
of the null hypothesis offers supportive evidence for Chaney’s extensive margin implication.
A Wald-type test of smooth nonlinear hypotheses about the estimated parameters is used
where the p-values are based on the delta method (an approximation appropriate in large
samples). Results of the tests, reported in Table 8, offer overall support for Chaney’s impli-
cation. Four out of the five samples studied (all except HI&MIX) show evidence in support
of a significantly larger extensive margin contribution in differentiated than homogeneous
sectors. The evidence in support of Chaney’s mentioned prediction offers a possible alter-
native solution to the elasticity puzzle where high elasticities could be due to the extensive
margin dominating in less substitutable sectors. Lower elasticities may be attributable to
more substitutable sectors where the role of the extensive margin is minor.

Lastly, as mentioned, another alternative solution to the elasticity puzzle involves the
development status of exporters. Note that export responses to real exchange rate fluctu-
ations from less developed countries to the rest of the world are smaller than those from
high-income countries (0.043 for DC&MIX sample versus 0.122 for HI&MIX sample). This
evidence would suggest a solution to the substitution elasticity puzzle where high elasticities
may correspond to high-income exporters and low elasticities may correspond to developing
countries.

16

3.2 Kehoe-Ruhl (2009) Sample

Kehoe and Ruhl (2009) conclude from their empirical study of the extensive margin of
trade that growth in the extensive margin is a consequence of structural changes or trade
liberalizations but does not respond much to business-cycle events. To the extent that the
study in this paper fits within their business-cycle category, our results contradict Kehoe
and Ruhl’s (KR) because we find that the extensive margin is an important component of
the yearly export response to real exchange rate fluctuations. In this section I argue that
results from both papers can be reconciled by expanding the coverage of KR’s sample. I
argue that if KR were to include all bilateral trade flows for the countries in their business-
cycle sample their results would likely change.

KR study a total of 9 countries over varying 10-year periods depending on the episode
they focus on. Trade liberalization and business-cycle episodes are studied for the period
1988/1989-1998/1999 where structural transformations are studied for the period 1975-1985
or 1995-2005. However, even among the 9 countries they study, KR limit their study to
a sub-sample of all bilateral trade relationships. On top of using a different sample, this
work is different from KR’s in the exact definition of the extensive margin. Where this
paper adopts the Hummels-Klenow (2005) definition detailed above, KR use a country-pair
specific cutoff to determine if a country-pair trades or not in a given sector. We both use
similar bilateral trade data with 4-digit SITC detail for trade flows.

To understand the differences between results, I first apply my methodology to their
sample of 4 countries (and 6 country-pairs) of business-cycle episodes in order to determine
if the different methods are responsible for the different results.12 I find, similar to KR, that
the extensive margin does not have a significant contribution to the yearly export response
to real exchange rate fluctuations within this restricted sample. Next I repeat the exercise
including all bilateral trade relationships (within my sample of 136 countries) where their
4 countries are present either as exporters or importers. This change increases the sample
from 6 to 1,068 country-pairs. In this expanded but still restricted sample, results show that

12 Country-pairs used to study business-cycle events by KR include: US-Germany, US-Japan, US-UK,
Germany-US, Japan-US and UK-US.

17

the extensive margin of trade plays an important role in the export response to yearly real
exchange rate fluctuations. Table 9 shows results for KR sample and for the expanded KR
sample. Expanded KR sample shows significant contributions from the extensive margin in
export responses for overall bilateral trade flows and for differentiated bilateral trade flows
between 88 and 90% .

This finding confirms the importance of the particular sample of country-pairs chosen
when studying the behavior of the extensive margin. By choosing a reduced sample of
bilateral relationships corresponding to their 4 chosen countries, Kehoe-Ruhl (2009) explore
a partial aspect of the extensive margin of trade during business-cycle events.

3.3 Transitory versus Permanent Shocks

While addressing Ruhl (2008) and Kehoe-Ruhl (2009), this paper has argued that real
exchange rate fluctuations represent transitory shocks to a country-pair along the lines of
business-cycle events, in contrast to permanent shocks to a bilateral trade relationship such
as structural transformations and trade liberalizations.

Despite predictions of purchasing power parity theory, it has been empirically established
that the real exchange rate fluctuates over time. These fluctuations, at a yearly frequency,
are exploited in this paper. The literature on this issue debates about the correct estimate
of the half-life of the real exchange rate.13 On one end, Rogoff (1996) offers a consensus
estimate of three to five years where Imbs et al (2005) propose an aggregation correction
that decreases the half-life of the RER to slightly over one year. For this paper, though, it
is not crucial to know the exact half-life of the RER. As long as fluctuations in the RER
from one year to the next are recognized by exporters and importers as temporary, the
interpretation of results holds.

For additional evidence on the validity of this categorization of shocks, I re-estimate
the model from equation (9), adding the presence of a regional trade agreement among the
countries in the country-pair as an explanatory variable. Regional trade agreements (RTA)
data are obtained from Rose (2004). As long as RTA represents policies of more permanent

13 Half-life is defined as the time necessary for half the effect of a given shock to disappear.

18

nature than changes in real exchange rates, we expect to find a larger impact from RTA
on bilateral exports. Indeed, estimation results, not shown for brevity, deliver estimates for
the impact of RTA that are larger than those for RER. The estimated impact of having a
RTA is equal to the impact of a real depreciation of 450%, or more, among the four samples
where both variables have significant effects on bilateral exports (World, HI&MIX, DC and
DC&MIX).14 The important difference between the impact of both variables is consistent
with the transitory/permanent distinction made in this paper.

3.4 6-digit Bilateral Trade Data

As a robustness check, the margins decomposition analysis is repeated with 6-digit bilateral
trade data instead of 4-digit bilateral trade data. Harmonized System (HS) 6-digit bilateral
trade data are available for approximately 200 countries for the period 1993-2003 from
the UN Comtrade database.15 For this robustness check I use a sub-sample including our
original 136 countries for the period between 1993 and 1997 to better compare results with
the main results of the paper. The higher disaggregation of these data increases the total
number of sectors studied from 440 SITC 4-digit sectors to 4,795 HS 6-digit sectors. Over
10 million observations are used to build the bilateral margins measures for 136 countries
over 1993-1997 following the same procedure described in Section 2 for 4-digit data.

The extensive margin contribution (measured at the 6-digit) shows to be important for
yearly responses of differentiated trade to RER fluctuations. Four out of the five samples
show a significant EM contribution in differentiated trade ranging from 46 to 64% with an
average contribution of 59% (HI&MIX sample shows no significant EM contribution). The
average significant EM contribution to differentiated trade measured with 4-digit trade in
Table 7 was found to be 40% among the five samples. In the case of homogeneous trade,
EM does not show a significant contribution to the yearly export response to RER with
6-digit margins measures (where only one of the samples showed a significant contribution

14 Only in 0.1% of the observations in the World sample, yearly fluctuations in RER represent real depre-
ciations above 450%.

15 The data were kindly provided by David Weinstein who compiled it for the period 1993-2003 when
country coverage is extensive.

19

with 4-digit margins).16 Lastly, the EM contribution to overall bilateral trade response
is significant for two of the five samples (HI and HI&MIX, average of 41%) where four of
the five samples showed significant results with 4-digit margins (average of 71%). Overall,
results with 6-digit margin measures indicate that the extensive margin of trade plays an
important role in yearly export responses to RER fluctuations. In this analysis EM is found
to be particularly important for differentiated trade as results are significant for those trade
flows in World, HI, DC and DC&MIX samples. This evidence is viewed as supportive of
earlier conclusions on the relevance of the extensive margin in the yearly export behavior.

4 Extensive Margin of Trade and Credit Constraints

Results from Section 3 suggest a salient role of the extensive margin in explaining bilateral
trade responses to real exchange rate fluctuations. In order to further learn about deter-
minants of the extensive margin response to real exchange rate fluctuations we focus on a
different sectoral characteristic. In particular we study if there is a role of credit constraints
on the extensive margin response to real exchange rate fluctuations.

Two measures of sectoral vulnerability to credit constraints are obtained from Manova
(2006). Asset tangibility (AT) and external finance dependence (EFD) are sector level mea-
sures based on US data of publicly traded firms for the period 1986-1995. AT is the share
of net property, plant and equipment in total assets for the median US firm in each sector.
EFD is the share of capital expenditures not financed by cash flow from operations for the
median US firm in each sector.17 Both measures are commonly used to study vulnerability
of sectors to credit constraints as they are argued to measure sectoral differences in tech-
nologies. For example, sectors vary in their share of tangible assets which affects their ability
to borrow capital (where lower AT indicates a higher vulnerability to credit constraints).
Moreover, there are sectoral differences in the size of set up costs and the speed at which
investments generate revenues creating different financing needs (therefore a higher EFD

16 Note that none of the five samples show significant export responses in homogeneous trade for the
period 1993-1997.

17 Rajan and Zingales (1998) first measured sectoral financial vulnerability with this share.

20

indicates a higher vulnerability to credit constraints).
The fact that both measures are built with US data, and not with data for each studied

country, is viewed as a strength in the literature. One reason is that firms’ financing
under the developed US financial market arguably reflects firms’ desired financing structure,
which is what the measures would ideally capture. Moreover, using the US based measures
avoids endogenous changes of the sectoral vulnerability measures to countries’ financial
development. In developing countries where financial development is worse a sector may
increase its AT and decrease its EFD with respect to the same sector in the US. Building
the measures with local data would erroneously suggest that this sector is less vulnerable to
credit constraints in the developing country. This bias may lead us to conclude that credit
constraints do not diminish the extensive margin.

Both sector measures are available for manufacturing industries at the 3-digit ISIC clas-
sification. A matching is performed between the 3-digit ISIC sectors with credit constraint
data and our 4-digit SITC data when a unique 3-digit ISIC corresponds to each 4-digit
SITC. Such matching leaves us with 74 out of 440 4-digit sectors with complete data where
35 sectors are identified by Rauch (1999) as differentiated and 14 sectors are identified as
homogeneous.

Moreover 4-digit sectors are classified by the amount of entry and exit in the sector
in order to determine which sectors may have a bigger role in the extensive margin. The
fraction of entry in the sector is defined as the number of country-pair-year that starts
trading in the sector the year of a "large depreciation" (of the exporter with respect to
the importer), divided over the number of country-pair-year that were not trading in that
sector the year prior to the "large depreciation". The fraction of exit in the sector is
defined as the number of country-pair-year that stops trading in the sector the year of
a "large appreciation" (of the exporter with respect to the importer), divided over the
number of country-pair-year that were trading in that sector the year prior to the "large
appreciation".18 For example, imagine that we observe zero meat exports from Argentina

18 We have limited the analysis to cases of "large depreciations" and "large appreciations" which are
defined as the observations in the top/bottom 30 percentile of the distribution of RER fluctuations among
country-pair-years in the sample. For the four samples, the 30 percentile cutoffs range between 9 and 15%.

21

to Brazil in 1990, from Argentina to Chile in 1990, and from Argentina to Bolivia in 1990.
Imagine that there is a "large depreciation" between Argentina and those three trading
partners between 1990 and 1991. And imagine that we observe positive meat exports from
Argentina to Brazil in 1991, and zero meat exports from Argentina to Chile and Bolivia
in 1991. Our measure of entry for the meat sector for the period will be 1/3, as there
is one new exporter-importer-year trading in meat (after a large depreciation) and there
are three exporter-importer-year not trading on meat before the depreciation. The entry
measure varies between 0 and 0.0013 for homogeneous sectors and between 0 and 0.0052
for differentiated sectors among the four samples studied. The exit measure varies between
0 and 0.2857 for homogeneous sectors and between 0 and 0.0984 for differentiated sectors
among the four samples studied.

Findings from Section 3 on the importance of the extensive margin suggest that we
should observe entry (exit) in the average sector facing a depreciation (appreciation) of the
exporter with respect to the importer. We aim to determine weather the vulnerability to
credit constraints of the sectors plays a role in the sectoral response. To that end we correlate
sectoral entry after large depreciations with measures of vulnerability to credit constraints
for homogeneous and differentiated sectors. Moreover we correlate sectoral exit after large
appreciations with measures of vulnerability to credit constraints for homogeneous and
differentiated sectors. Two sets of scatter plots are presented for four samples (DC&MIX,
DC, HI&MIX and HI) in Figures 1 and 2 for the measure of external finance dependence.1920

Figure 1 shows the relationship between external finance dependence (in the x-axis)
and entry (in the y-axis). Each of the four panels within Figure 1 focus on a different
sample. For example the top left panel shows data for DC&MIX sample. Each data point
in this graph represents measures of EFD and entry of a sector. (Homogeneous sectors
are represented with circles and differentiated sectors are represented with triangles.) Each

(For example, for the HI sample, observations with yearly growth in RER above 9% are defined as having
large depreciations, and observations with yearly contractions in RER above 9% are defined as having large
appreciations.)

19 Results for AT are mixed and not shown for brevity.
20 Figures 1 and 2 exclude outliers in entry and exit measures in order to better present the data in the
figures.

22

data point is labeled by the 4-digit code of the sector. (Table 11 shows the description
labels of each of the homogeneous and differentiated sectors included in these figures.) A
line of best fit is shown for homogeneous and differentiated sectors. Overall Figure 1 shows
expected negative correlations between entry and external finance dependence for all the
samples studied among both sets of sectors. Figure 2 shows a parallel exercise for the data
on EFD and exit following a large appreciation. Except for differentiated sectors in the
HI sample, we observe the expected positive correlations among exit of homogeneous and
differentiated sectors and their EFD.

As a complement, Table 10 shows the correlation coefficients between external finance
dependence and entry (exit) after large depreciations (appreciations) for the four samples
studied. Regarding entry, the evidence indicates that among homogeneous sectors, entry
and EFD have a strong correlation of the expected sign ranging from -0.59 to -0.80 (signifi-
cantly different from zero in all samples). Among differentiated sectors the entry correlation
ranges from -0.07 to -0.10, with the expected sign in all cases but not significant. Regarding
exit, the evidence in Table 10 shows expected positive correlations between EFD and exit
after large appreciations in most of the samples. The exit correlations are smaller in magni-
tude than those for entry and are significant only for differentiated sectors in the DC&MIX
sample.

In order to rule out the possibility that other sectoral differences correlated with EFD
drive results we run a regression analysis with sectoral factor utilization controls. Measures
of physical capital, human capital and natural resource intensities for the 49 studied sectors
are obtained from Manova (2006).21 Entry (exit) is regressed on EFD for homogeneous and
differentiated sectors (two regressions for entry and two for exit) for each of the four samples
studied. Secondly we add factor utilization controls to these regressions. A total of thirty
two regressions are run where sixteen include controls and sixteen do not include controls.

21 Measures are obtained at the 3-digit ISIC level and matched to the 4-digit SITC data. Natural resource
intensity is a dummy variable equal to one for: wood products, except furniture; paper and products;
petroleum refineries; miscellaneous petroleum and coal products; other nonmetallic mineral products; iron
and steel; and nonferrous metals. Physical and human capital intensities are computed with UNIDO’s
dataset on US firms for the period 1986-1995. Physical capital intensity is the median of the ratio of gross
fixed capital formation to value added in each industry and human capital intensity is the median of the
industry’s mean wage over the mean wage of the US manufacturing sector.

23

Results, not shown for brevity, indicate that adding sectoral factor utilization controls does
not change results. In particular, the EFD coefficient keeps its sign and significance in
fourteen out of the sixteen regressions.

Overall this section provides evidence in support of the importance of vulnerability to
credit constraints (measured by external finance dependence) in the extensive margin of
trade. Within homogeneous and differentiated sectors there is a negative (positive) cor-
relation between external finance dependence and entry (exit) after a large depreciation
(appreciation) in the four country samples studied.22 Correlations for entry among homo-
geneous sectors following a large depreciation show the largest magnitude and significance
levels. Given the importance of the extensive margin of trade on export responses, the evi-
dence on credit constraints may provide another explanation for the substitution elasticity
puzzle.

5 Conclusion

The puzzle regarding the elasticity of substitution (between domestic and foreign goods)
that affects international economics is found likely to be worse than previously claimed
in the literature as my estimates (below one) are lower than those previously considered
low. Evidence presented in this paper suggests that the puzzle may possibly be solved by
considering the sectoral composition of trade and the development status of the exporters.
However, mechanisms proposed by Ruhl (2008) regarding extensive margin responses to
temporary/permanent shocks seem unlikely to solve the puzzle.

I define the bilateral extensive margin of exports as a weighted count of the exporter
country’s varieties exported to the importer and I define the bilateral intensive margin as
the exporter country’s relative volume of exports. I define a variety as each of the 440
4-digit sectors from the Standard International Trade Classification (SITC) Revision 2 and
use a bilateral trade sample of 136 countries for the period 1981-1997. Bilateral export
responses to bilateral real exchange rate fluctuations are decomposed into extensive and

22 The only exception is the negative exit correlation for the HI sample.

24

intensive margin responses.
My results indicate that the extensive margin of trade plays a significant role in yearly

export adjustments. In particular I find that the extensive margin explains on average
70% of overall export responses to real exchange rate fluctuations, among the four country
samples with statistically significant export responses, and approximately 80% in the overall
sample.

Consistent with Chaney (2008), evidence indicates that the extensive margin of trade is
more responsive in less substitutable sectors (except in the HI&MIX sample). Further study
of estimates in the literature may allow us to determine if such sectoral factors can solve
the puzzle. In particular, finding that high elasticities are mostly derived from estimates for
differentiated sectors (and low elasticities from estimates for homogeneous sectors) would
address the puzzle. Further results indicate that export responses from less developed
countries to the rest of the world are smaller than those from high-income countries. If most
high estimates from the literature were to correspond to samples of high-income exporters
(and low estimates were to correspond to less developed exporters), we may propose that
the puzzle relates to a development status story. Lastly, from the study of credit constraints
and the extensive margin, I find evidence in support of the importance of sectoral credit
constraints (measured as external finance dependence) on the extensive margin response.
This suggests that, in solving the elasticity puzzle, credit constraints may also play a role.

25

References

[1] Bayoumi, Tamim, "Estimating Trade Equations from Aggregate Bilateral Data," IMF
Working Paper No. 74, May, 1999.

[2] Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding and Peter K. Schott,
"The Margins of US Trade (Long Version)," NBER WP 14662, January, 2009.

[3] Bilbiie, Florin, Fabio Ghironi and Marc J. Melitz, "Endogenous Entry, Product Variety,
and Business Cycles," NBER WP 13646, November, 2007.

[4] Broda, Christian and David E. Weinstein, "Globalization and the Gains from Variety,"
Quarterly Journal of Economics, 2006.

[5] Broda, Christian and David E. Weinstein, "Product Creation and Destruction: Evi-
dence and Price Implications," NBER WP 13041, April, 2007.

[6] Chaney, Thomas, "Distorted Gravity: The Intensive and Extensive Margins of Inter-
national Trade," American Economic Review, September, 2008.

[7] Colacelli, Mariana, "Export Responses to Real Exchange Rate Fluctuations: Devel-
opment Status and Exported Good Effect," Barnard College, Columbia University,
mimeo, 2009.

[8] Drozd, Lukasz A. and Jaromir B. Nosal, "Understanding International Prices: Cus-
tomers as Capital," Staff Report 411, Federal Reserve Bank of Minneapolis, August,
2008.

[9] Eaton, Jonathan, Marcela Eslava, Maurice Kugler and James Tybout, "Export Dy-
namics in Colombia: Firm-Level Evidence," NBER WP 13531, October, 2007.

[10] Feenstra, Robert C., "Symmetric Pass-Through of Tariffs and Exchange Rates Un-
der Imperfect Competition: An Empirical Test," Journal of International Economics,
August, 1989.

26

[11] Feenstra, Robert C., "New Product Varieties and the Measurement of International
Prices," American Economic Review, March, 1994.

[12] Feenstra, Robert C., "World Trade Flows, 1980-1997,” Center for International Data,
UC Davis, March, 2000.

[13] Hummels, David and Peter J. Klenow, "The Variety and Quality of a Nation’s Ex-
ports," American Economic Review, June, 2005.

[14] Imbs, Jean and Isabelle Mejean, "Elasticity Optimism," CEPR, February, 2009.

[15] Imbs, Jean, Haroon Mumtaz, Morten O. Ravn and Hélène Rey, "PPP Strikes Back:
Aggregation and the Real Exchange Rate," Quarterly Journal of Economics, February,
2005.

[16] Kehoe, Timothy J. and Kim J. Ruhl, "How Important is the New Goods Margin in
International Trade?" Federal Reserve Bank of Minneapolis Research Department Staff
Report 324, May, 2009.

[17] Krugman, Paul, "Scale Economies, Product Differentiation, and the Pattern of Trade,"
American Economic Review, 1980.

[18] Manova, Kalina, "Credit Constraints, Heterogeneous Firms, and International Trade,"
Harvard University mimeo, 2006.

[19] Manova, Kalina, Shang-Jin Wei and Zhiwei Zhang, "Firm Exports and Multinational
Activity under Credit Constraints," Stanford University mimeo, 2009.

[20] Rajan, Raghuram G. and Luigi Zingales, "Financial Dependence and Growth," Amer-
ican Economic Review, 1998.

[21] Rauch, James, “Networks Versus Markets in International Trade,” Journal of Interna-
tional Economics, 1999.

[22] Rogoff, Kenneth, "The Purchasing Power Parity Puzzle," Journal of Economic Liter-
ature, June, 1996.

27

[23] Rose, Andrew K., "Does the WTO Make Trade More Stable?" NBER WP 10207,
January, 2004.

[24] Ruhl, Kim J., "The International Elasticity Puzzle," University of Texas at Austin
mimeo, March, 2008.

[25] The World Bank, "World Development Indicators," CD-ROM, 2001.
[26] Tinbergen, Jan, "Shaping the World Economy," The Twentieth Century Fund, New

York, 1962.

28

A Figures and Tables

 

Sectoral Entry and Credit Constraints Sectoral Entry and Credit Constraints

.0 008 DC&MIX Sample for Large Depreciations .0008 DC Sample for Large Depreciations

9 80 699 7 874 8 4 83 52839423894269 97 2 690

4 84 26952793 26 967097 8734 2 8742352 6 19 8 45619879391 87 48 541 9
759 1
288742135 2 644 218392 2548375443084922 84 51642 4 541 9 4 22 697545791320 8 4782978464 899 6
8989469 1 875999 61
698949655622821638128291354312204 60 2864 782127887 2 2 874165 2 68 124 50 2 687216284782724

0 81 214 40 262 218364289 0 0 6 99956 28213 2828131432214460 2 6281264328 90

0 . 2 . 4 .6 .8 1 0 .2 .4 .6 .8 1
External Finance De pendence Exte rnal Finance D ependence

Ho m. Entry Diff. Entry H om. Entry Diff. Entry
Ho mogeneous Di ffer enti ated H omogeneou s Di ffer enti ated

0 .0 01 .0 02 Sectoral Entry and Credit Constraints .0025Sectoral Entry and Credit Constraints

HI&MIX Sample for Large Depreciations HI Sample for Large Depreciations

845 9 4 84 4 84 25893432894269 97
65 73
2974292699 7 84 5619264737264251 9 2 8743252 4 83 2 46956109879391 87 48 541 9
2622898888797744293165525624824228613832122548283981543513408341222014 640 22826246 6 19 8
4 22 697545719302 8 4782798446 759 1
77 84 85794591 916 899 6
2 874165 2 68 124 50 2 687216284787242
8178287643223198298848699 1 1
0 0 6 99956 28213 2828131432214640 2 6212836428 90

0 . 2 . 4 .6 .8 0 .2 .4 .6 .8 1
External Finance De pendence Exte rnal Finance D ependence

Ho m. Entry Diff. Entry H om. Entry Diff. Entry
Ho mogeneous Di ffer enti ated H omogeneou s Di ffer enti ated

Figure 1: Sectoral Entry and Credit Constraints

29

 

Sectoral Exit and Credit Constraints Sectoral Exit and Credit Constraints

DC&MIX Sample for Large Apprec iations DC Sample for Large Appreciations

0 .0 2.0 4.0 6 2266 1904 0 .02 .06 .1 2 690
2864
28 71562 2 23 32 8246 7819 4868 2 871 24 50
87587221672869487429796401251
789599 61 56 22 2 6829247672 899 6
541 9 988974439655529 649824 2881862493308 759 1
628988794493565 529 4 2818639240812524863987545412307849923212699 7 9 874 8 62 872 2 332 828 46475872617168889484296145 19 87 48 541 9
294486 1 2 1225886897931451543248794219320124649097 2 616439278390

28 72 95 1204 40 26 1639 73

0 . 2 . 4 .6 .8 0 .2 .4 .6 .8 1
External Finance De pendence Exte rnal Finance D ependence

Ho m. Exit Diff. Exit H om. Exit Diff. Exit
Ho mogeneous Di ffer enti ated H omogeneou s Di ffer enti ated

0 .0 1 .0 25 Sectoral Exit and Credit Constraints 0 .005 .015 Sectoral Exit and Credit Constraints

HI&MIX Sample for Large Appreciations HI Sample for Large Appreciations

562 2 24 60 2 871 2 68 124 2 682
56 22 2 614
4 22 2266 1842 50 440
2 2 686
2288 46467951867721301799286487898364297601425
2628988977443129655 925 8 188861408393122528468293798155345142340387491922232021464909 7 8844 5721289 0 9 874 8 587495919 196 2 8988794493556925 8 13 2 332 97 19 87 48 875594919 196
29644 26 98177863609719628479886843279612451 6 4 84 252869879315454123487949213221069
1 946 8129230 1

0 . 2 . 4 .6 .8 0 .2 .4 .6 .8
External Finance De pendence Exte rnal Finance D ependence

Ho m. Exit Diff. Exit H om. Exit Diff. Exit
Ho mogeneous Di ffer enti ated H omogeneou s Di ffer enti ated

Figure 2: Sectoral Exit and Credit Constraints

30

Table 1: Bilateral Trade and Margins of Exports by Exported Good and by Country Group,

Summary Statistics. 34 High Income and 102 Developing Countries, 1981-1997.

All Exports Homogeneous Differentiated

Mean St Dev Obs. M ean St Dev Obs. Mean St Dev Obs.

(T and X measured in 1,000 of 1995 US dollars)

World

T 370,434 2,902,003 140,013 95,864 581,673 93,617 189,782 1,733,801 125,835

EM 0.33 0.27 140,013 0.15 0.20 93,617 0.29 0.32 125,835

IM 0.54 44.72 140,013 2.26 89.72 93,617 0.13 6.90 125,835

X 38,600,000 85,700,000 140,013 8,489,768 16,800,000 93,617 19,000,000 45,500,000 125,835

HI

T 2,040,783 7,542,183 17,252 308,215 1,229,359 15,025 1,030,652 4,392,642 16,953

EM 0.59 0.27 17,252 0.26 0.26 15,025 0.60 0.35 16,953

IM 0.06 0.53 17,252 1.52 75.89 15,025 0.06 0.21 16,953

X 71,500,000 111,000,000 17,252 13,600,000 20,300,000 15,025 33,700,000 57,200,000 16,953

HI & MIX 760,709 4,393,713 56,639 135,693 769,854 40,802 385,323 2,547,835 54,802
T 0.48 0.28 56,639 0.20 0.23 40,802 0.46 0.35 54,802
EM 0.45 56,639 1.85 40,802 0.10 0.72 54,802
IM 47.92 56,639 78.06 40,802 54,802
X 26,300,000 69,100,000 5,863,345 13,800,000 12,400,000 35,300,000

DC

T 32,412 157,139 46,921 23,359 93,932 25,283 10,346 64,866 38,847

EM 0.22 0.22 46,921 0.10 0.15 25,283 0.14 0.21 38,847

IM 0.90 48.20 46,921 4.26 91.38 25,283 0.27 12.38 38,847

X 10,600,000 20,700,000 46,921 1,947,627 2,995,232 25,283 4,710,254 9,646,196 38,847

DC & MIX 105,306 924,510 83,374 65,094 373,760 52,815 38,923 514,663 71,033
T 0.23 0.22 83,374 0.11 0.16 52,815 0.15 0.22 71,033
EM 0.61 83,374 2.58 52,815 0.16 9.17 71,033
IM 42.41 83,374 97.78 52,815 71,033
X 46,900,000 94,500,000 10,500,000 18,500,000 24,200,000 51,400,000

31

Table 2: Bilateral Margins of Export Responses to Real Exchange Rate Fluctuations. Over-

all, Homogeneous and Differentiated Exports. 136 Sample, 1981-1997.

All Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.056*** 0.045*** -0.048*** 0.060***
[0.008] [0.007] [0.008] [0.003]
Observations
# country-pairs 140,013 140,013 140,013 140,013
R-squared 13,860 13,860 13,860 13,860
0.09 0.06 0.09 0.70

Homogeneous Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.031*** 0.001 0.005 0.026***
[0.011] [0.009] [0.009] [0.003]
Observations 93,617 93,617 93,617 93,617
# country-pairs 10,543 10,543 10,543 10,543
R-squared 0.02 0.29
0.03 0.04

Differentiated Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.080*** 0.037*** -0.035*** 0.078***
[0.008] [0.006] [0.008] [0.004]
Observations
# country-pairs 125,835 125,835 125,835 125,835
R-squared 12,991 12,991 12,991 12,991
0.16 0.09 0.10 0.77

1. Robust standard errors in brackets
2. * significant at 10%, ** significant at 5%, *** significant at 1%
3. GDP and GDPpc for the exporter and the importer included. Fixed Effects for

country-pairs and years included

32

Table 3: Bilateral Margins of Export Responses to Real Exchange Rate Fluctuations. Over-

all, Homogeneous and Differentiated Exports. HI Sample, 1981-1997.

All Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) -0.042 0.225*** -0.585*** 0.317***
[0.103] [0.049] [0.091] [0.022]
Observations 17,252 17,252 17,252 17,252
# country-pairs 1,096 1,096 1,096 1,096
R-squared 0.11 0.15 0.85
0.17

Homogeneous Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) -0.234* 0.128 -0.387*** 0.024
[0.123] [0.089] [0.086] [0.020]
Observations 15,025 15,025 15,025 15,025
# country-pairs 1,038 1,038 1,038 1,038
R-squared 0.08
0.05 0.10 0.38

Differentiated Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.433*** 0.161*** -0.240*** 0.512***
[0.062] [0.042] [0.056] [0.022]
Observations 16,953 16,953 16,953 16,953
# country-pairs 1,087 1,087 1,087 1,087
R-squared 0.32 0.11 0.13 0.91

1. Robust standard errors in brackets
2. * significant at 10%, ** significant at 5%, *** significant at 1%
3. GDP and GDPpc for the exporter and the importer included. Fixed Effects for

country-pairs and years included

33

Table 4: Bilateral Margins of Export Responses to Real Exchange Rate Fluctuations. Over-

all, Homogeneous and Differentiated Exports. HI MIX Sample, 1981-1997.

All Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.122*** 0.045*** -0.045** 0.123***
[0.017] [0.013] [0.019] [0.007]
Observations 56,639 56,639 56,639 56,639
# country-pairs 4,206 4,206 4,206 4,206
R-squared 0.11 0.06 0.09 0.66

Homogeneous Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.145*** 0.098*** -0.019 0.065***
[0.022] [0.018] [0.020] [0.008]
Observations 40,802 40,802 40,802 40,802
# country-pairs 3,606 3,606 3,606 3,606
R-squared 0.02 0.05 0.24
0.05

Differentiated Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.178*** 0.023* -0.009 0.164***
[0.014] [0.012] [0.014] [0.006]
Observations 54,802 54,802 54,802 54,802
# country-pairs 4,137 4,137 4,137 4,137
R-squared 0.20 0.75
0.08 0.10

1. Robust standard errors in brackets
2. * significant at 10%, ** significant at 5%, *** significant at 1%
3. GDP and GDPpc for the exporter and the importer included. Fixed Effects for

country-pairs and years included

34

Table 5: Bilateral Margins of Export Responses to Real Exchange Rate Fluctuations. Over-

all, Homogeneous and Differentiated Exports. DC Sample, 1981-1997.

All Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.065*** 0.050*** -0.034*** 0.049***
[0.011] [0.010] [0.011] [0.003]
Observations 46,921 46,921 46,921 46,921
# country-pairs 6,542 6,542 6,542 6,542
R-squared 0.09 0.08 0.06 0.64

Homogeneous Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.022 -0.006 0.009 0.019***
[0.015] [0.012] [0.013] [0.004]
Observations 25,283 25,283 25,283 25,283
# country-pairs 4,270 4,270 4,270 4,270
R-squared 0.32
0.03 0.04 0.03

Differentiated Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.083*** 0.042*** -0.022* 0.064***
[0.011] [0.009] [0.012] [0.004]
Observations 38,847 38,847 38,847 38,847
# country-pairs 5,886 5,886 5,886 5,886
R-squared 0.17 0.11 0.72
0.06

1. Robust standard errors in brackets
2. * significant at 10%, ** significant at 5%, *** significant at 1%
3. GDP and GDPpc for the exporter and the importer included. Fixed Effects for

country-pairs and years included

35

Table 6: Bilateral Margins of Export Responses to Real Exchange Rate Fluctuations. Over-

all, Homogeneous and Differentiated Exports. DC MIX Sample, 1981-1997.

All Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.043*** 0.039*** -0.041*** 0.045***
[0.009] [0.008] [0.009] [0.003]
Observations 83,374 83,374 83,374 83,374
# country-pairs 9,654 9,654 9,654 9,654
R-squared 0.09 0.07 0.08 0.74

Homogeneous Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.007 -0.016 0.006 0.017***
[0.012] [0.010] [0.010] [0.003]
Observations 52,815 52,815 52,815 52,815
# country-pairs 6,937 6,937 6,937 6,937
R-squared 0.36
0.03 0.03 0.04

Differentiated Bilateral Exports

Dep. Var. is Log of Bilateral Trade and Margins Measures:

ln(T) ln(EM) ln(IM) ln(X)

Ln(RER) 0.058*** 0.032*** -0.031*** 0.057***
[0.009] [0.008] [0.010] [0.004]
Observations 71,033 71,033 71,033 71,033
# country-pairs 8,854 8,854 8,854 8,854
R-squared 0.15 0.10 0.10 0.79

1. Robust standard errors in brackets
2. * significant at 10%, ** significant at 5%, *** significant at 1%
3. GDP and GDPpc for the exporter and the importer included.

Fixed Effects for country-pairs and years included

36

Table 7: Summary of Results on Margin Contributions in Export Responses to Real
Exchange Rate Fluctuations. Overall, Homogeneous and Differentiated Exports. Five
samples, 1981-1997.

T coefficient EM coefficient IM coefficient EM % in

(significant coefficients included only) Trade

World All trade 0.056 0.045 80.4
Homogeneous 0.031 0.037 46.3
Differentiated 0.080

HI All trade -0.234 -0.387
Homogeneous 0.433
Differentiated 0.161 37.2

HI&MIX All trade 0.122 0.045 36.9
Homogeneous 0.145 0.098 67.6
Differentiated 0.178 0.023 12.9

DC All trade 0.065 0.050 76.9
Homogeneous 0.083 0.042 50.6
Differentiated

DC&MIX All trade 0.043 0.039 90.7
Homogeneous 0.058 0.032 55.2
Differentiated

1. EM & IM coefficients included when they have the same sign of T coefficient