Trade Liberalization and the Extensive Margin of
Differentiated Goods: Evidence from China*
Johannes Van Biesebroeck§
University of Leuven & CEPR
Yingting Yi
University of Leuven
October 2012
Abstract
We exploit tariff reductions associated with China’s entry into the WTO to
evaluate whether the response of trade flows at the external margin depends on
the degree of product differentiation. We adopt a triple difference-indifferences approach to identify the effects as cleanly as possible by comparing
each WTO member’s market entry into China with its entry into India and
Indonesia. The absolute magnitudes of the average responses are estimated to
be relatively large and we provide both new evidence and justification for the
heterogeneity of the effects across products and countries. Specifically, tariff
reductions are more likely to induce export market entry for differentiated
goods sectors (Rauch, 1999), for goods subject to a lower Chinese import
elasticity (Broda et al., 2006), and for exports from OECD countries.
Key words: WTO, import tariff, trade policy, exports
JEL Classification: F10, F13, F14
*
We would like to thank Matthew Cole, Joep Konings, Frank Verboven, and seminar participants at
ETSG, the Belgium Trade Workshop, and the University of Leuven for useful comments. Financial
support from an ERC grant of the E.U. and Program Financing of the University of Leuven is gratefully
acknowledged.
§
Corresponding author: Department of Economics, University of Leuven, Naamsestraat 69, 3000
Leuven, Belgium; E-mail: [email protected].
1
1. Introduction
The controversial finding of Rose (2004)—a lack of evidence that the WTO has increased world
trade 1 —started a lively debate to assess various dimensions of the WTO’s effectiveness to
promote international trade. Focusing on sector-level bilateral trade volumes in the gravity
framework, Subramanian and Wei (2007) find that the effect on trade growth is strong and robust,
but highly uneven across countries and sectors. Distinguishing between the number of products
traded (extensive margin) and the average trade volume per product (intensive margin), Dutt,
Mihov, and Van Zandt (2011) find that the impact is almost exclusively on the extensive margin.
Exploiting variation in tariff rates to identify the effects, rather than relying on a WTO
membership dummy, Debaere and Mostashari (2010) find only weak effects of tariff reductions
on the number of imported products in the United States. Using export data for French firms,
Buono and Lalanne (2012) find that the extensive margin effects at the product level are the
result of existing exporters adding products or export destination.
We contribute to this line of inquiry in two ways. First, we develop a novel and extremely
flexible estimation strategy and second, we identify an effect of tariff reductions on the extensive
margin of trade that is particularly strong for differentiated goods. We focus specifically on
import tariff reductions of China which entered the WTO at the end of 2001. We study Chinese
imports rather than exports, because many important trading partners of China already applied
MFN rates to its exports before 2001. China’s entry into the WTO made this practice irrevocable
and is likely to have had an impact on exports as well, but identifying this effect in a precise way
would be very difficult.2 We use product rather than firm-level data to avoid relying on crosscountry differences for identification, as in most applications that use the gravity framework.
Buono and Lalanne (2012) highlight that the estimated importance of the extensive margin
response to trade liberalization depends crucially on the nature of the identification strategy.
We start from the triple-difference estimator of Frazer and Van Biesebroeck (2010). The
relationship between tariff reductions and changes in export propensity is measured relative to a
baseline probability at the exporter-product level that captures comparative advantage in a very
disaggregate way. Product-year and country-year dummies control flexibly for variations in
product-level import demand and country-level export supply over time. The effect of a tariff
reduction is thus identified solely from differential changes over time. Information at the same
level of detail on exports to a control country allows an additional normalization of this `within’
effect. It guarantees that even changes at the exporter-product level that apply across destinations
1
This literature evaluates the impact of both the General Agreement on Tariffs and Trade (GATT) and its
successor, the World Trade Organization (WTO). A related literature investigates whether free trade
agreements boost their members’ trade, see for example Baier and Bergstrand (2007). Haveman et al.
(2003) broaden the topic further and include the effects of non-tariff barriers and trade preferences.
2
Information on import tariffs faced by China prior to its WTO entry in countries that did not charge it
the MFN tariff rate is not recorded in the TRAINS database. Shi (2010) studies the effect on Chinese
exports, but she does not distinguish between changes in tariffs or shipping costs.
2
are not mistaken for tariff effects. We can incorporate heterogeneous tariff elasticities by
interacting the tariff variable with country or product characteristics.
Our second contribution is to highlight that the response at the extensive product-margin of
trade with respect to a tariff cut varies across sectors. In particular, we show that the effect is
more pronounced for differentiated goods or for goods subject to a lower elasticity of
substitution. This finding can be explicitly related to the theory, taking the generalization of the
Melitz-model to many destinations by Chaney (2008) as the starting point. While the model
predicts that the export volume of a firm that enters the export market following a reduction in
variable cost is negatively related to the substitution elasticity, it also predicts that the effect on
the number of exporters should be independent of the substitution elasticity. However, Cole
(2012) points out that this independence does not hold for changes in tariffs if these are applied
to the price including the markup, rather than to the marginal cost. Even though reductions in
tariff rates or variable costs have the same effect on a firm’s optimal price and quantity, their
impact on the absolute level of profit differs and this is what matters for the export market entry
decision. A stronger effect of a tariff reduction on the probability of exporting or on the number
of exported products for low-elasticity products is consistent with the heterogeneous firm model.
Several studies have found that proxies for the fixed cost of exporting have their primary
effect on trade through the extensive margin, as measured by the number of products exported.
Examples are trade facilitation (Persson, 2008), familiarity with the export destination
(Andersson, 2007), or immigration networks (Peri and Requena-Silvente, 2010). In each of
these three cases, the evidence is consistent with the prediction of Chaney (2008) that the
elasticity of the minimum productivity threshold for exporters with respect to the fixed cost of
exporting is higher for goods with lower substitution elasticity.3 Lawless (2010) finds that even
variables that are included in the gravity equation primarily as proxies for variable trade costs,
such as distance or the total cost of import restrictions, also have their primary influence on trade
through the extensive margin. The evidence using our novel identification method indicates that
the effects of tariff reductions on the extensive margin are also highly significant and
quantitatively important for differentiated, low-elasticity goods.
Given these constraints, we measure the extensive margin of trade by the existence of
positive trade flows in a given product category and year from each exporter country. In a
heterogeneous firm model with a fixed cost of exporting, such as Chaney (2008), this indicator
will be one if the productivity of the best firm exceeds the threshold for export market entry. A
tariff reduction has an impact on firm-level export behavior through the change in the
productivity threshold. The impact on the product extensive margin (at least one firm exports in
a given sector to a destination market) and on the number of firms that exports (often used as
3
In each of these applications, the degree of substitution is proxied with a dummy for product
differentiation taken from Rauch (1999).
3
firm extensive margin) are necessarily of the same sign. The relative magnitude of the two
effects is determined exogenously by the shape of the firm-level productivity distribution.
We use the classification in Rauch (1999) to divide products into three ordered categories:
homogeneous, reference-priced and differentiated goods. The degree of product differentiation
is, by definition, increasing in this order and the elasticity of substitution is assumed to be lower
on average. As a robustness check, we also use the estimates for the (Chinese) import demand
elasticities from Broda, Greenfield, and Weinstein (2006) to define an alternative partitioning of
products into a low, medium, and high elasticity group.
Our first key finding is that the probability of export market entry is highly responsive to
tariff declines. This is an important finding for policy, as the effects at the extensive margin
determine whether the impact of trade liberalization will be magnified over time (Bustos, 2011;
Eaton et al., 2008). Our estimates are in contrast with the small effects found for U.S. imports in
Debaere and Mostashari (2010) and French exports in Buono and Lalanne (2012). The smaller
reductions in average tariff rates, 1.4 and 6.4 percentage points respectively on U.S. imports and
French exports over a ten year period, could be one explanation. Another difference is that we
work with a much more detailed product classification, especially compared to Buono and
Lalanne (2012), and the extensive margin tends to be more prominent with more disaggregate
data.4 The different identification strategy could be another explanation: we measure changes
over time relative to a flexible product-country benchmark while controlling for product and
export-specific trends and the evolution of exports to two similar destinations. A final
explanation for the large effects is the unique attractiveness of the large and fast-growing
Chinese market for any potential exporter.
Secondly, we show that the effect is strongest for differentiated, low-elasticity goods. Cole
(2012) illustrates theoretically how this can be rectified with the heterogeneous firm model of
Chaney (2008) by assuming that tariffs are assessed at the final good price while other variable
trade costs raise marginal costs. As the CES demand specification makes prices and marginal
costs proportional, this distinction does not matter for any infra-marginal decision—tariffs could
either be assessed at the final consumer price or foreign firms could engage in intra-firm trade
and only pay tariffs on the marginal cost. However, for market entry decisions and thus for
extensive margin responses, the absolute profit levels do matter.
The ordering of the magnitude of the effects across sectors that we find is consistent with
results in Feenstra, Markusen, and Rose (2001) and Egger, Egger, and Greenaway (2008). The
first authors argue from the difference in gravity parameters that export patterns for
differentiated goods are consistent with the theoretical predictions of a reciprocal-dumping
4
Buono and Lalanne (2012) do find reasonably strong responses at the product-level extensive margin if
they use cross-country differences in tariff rates to identify the effects, but not for their “within” estimator
that exploits changes over time similarly to us. They further show that the product-level effects are
primarily driven by existing exporters selling more products abroad, rather than export market entry by
new firms.
4
model with free entry, while for homogeneous goods entry barriers are important. The second
authors show that regional trade agreements boost the intra-industry trade share of differentiated
goods more than trade in homogeneous goods. Our results contrast with Shi (2011) who finds
for Chinese exports that the extensive margin response with respect to a decline in variable cost
is increasing in the substitution elasticity. Two caveats are that the measure for variable cost she
uses combines freight, which has a fixed cost element to it, with tariffs and that most of the
identifying power comes from variation across country-product pairs. Crozet and Koenig (2010)
take an opposite approach. They impose the model of Chaney (2008) on the data and estimate
the structural parameters using export information of French firms. They thus recover the
demand elasticity for each sector that makes the observed firm behavior consistent with the
model. Their estimates tend to be smaller than those of Broda and Weinstein (2006), but the
ordering of goods is very similar (a Spearman rank-correlation of 0.53).
Thirdly, we find that the estimated effects are also highly heterogeneous across country
groups. The few countries still outside the WTO in 2001 do not show any response to China’s
tariff declines, which supports the effectiveness of our methodology. More important is that
developed OECD countries show a much larger response than the other WTO members. Perhaps
the average demand elasticity for exports from developed economies is on average higher even
within each product category. The re-interpretation of the Melitz-model in terms of quality
differentiation by Baldwin and Harrigan (2011) provides one justification for such an assumption.
Reverse causality, where WTO members negotiate especially high tariff reductions for products
where they have strong export potential, could be a partial explanation. Note however, that this
cannot explain the heterogeneity across product categories as the rankings of the responses are
similar for the non-OECD WTO members which played a minor role at the negotiating table.
The remainder of the paper is organized as follows. Section 2 provides some theoretical
background to help with the interpretation. Section 3 contains the empirical model and Section 4
the data description. Empirical results are in section 5, followed by a discussion of potential
endogeneity problems and implications in Section 6. Section 7 concludes.
2. Theoretical background
The seminal contribution by Melitz (2003) introducing firm heterogeneity in an equilibrium
model of international trade features two key elements. Productivity is heterogeneous across
firms and firms have to pay a fixed cost when they export. In equilibrium, only the sub-set of
firms with a productivity level that exceeds a minimum threshold will be able to sell profitably
abroad and decide to enter the export market. Chaney (2008) generalizes the model to a world
with many asymmetric countries and derives predictions for the structure of bilateral trade flows.
The total adjustment of exports to a changes in fixed or variable trading cost is now the
combined effect of a change in the volume of trade for existing exporters (the intensive margin)
and the total exports by firms that enter or leave the export market (the extensive margin).
Similar to the model in Krugman (1980) with identical firms, the adjustment at the intensive
5
margin is increasing in a sector’s elasticity of substitution. However, the adjustment at the
extensive margin is decreasing in the substitution elasticity and with a Pareto distribution of
firm-level productivity the latter effect dominates.
In Chaney (2008), the extensive margin response to a decline in trade costs is defined as the
volume of trade by new exporters. In most of the empirical literature, however, it is defined as
the number of firms that export to a particular export destination.5 Working with product-level
bilateral trade data, the dependent variable we use is the probability a positive export flow is
observed, indicating that the most productive firm in a sector has entered a particular export
market. These last two definitions differ from the extensive margin term in the decomposition of
Chaney (2008). They only measure the sensitivity of the productivity threshold to changes in
trade barriers and ignore how much each new exporter sells.
Chaney (2008) derives the following threshold for the productivity level of exporters:
⁄
̅
( )
(
⁄
)
Firms in the exporting country c and a sector with substitution elasticity σ are predicted to export
to a destination country d if their productivity level exceeds this threshold. The fixed cost of
trading between the two countries is denoted by fcd and the variable cost, such as iceberg
transport costs or an ad valorem tariff, by τcd. The other elements are all constant for all sectors:
Yd /Y is the share of country d in world income, γ is the shape parameter of the Pareto distribution,
wc the wage in country c, θd the index of remoteness for country d, and λ is a function of several
parameters in the model.
The number of firms that export will then vary with fixed or variable trading costs according
to the following elasticities of the productivity threshold:
̅
̅
The impact of a change in the fixed cost on the number of firms predicted to export is negatively
related to a sector’s substitution elasticity, consistent with the findings in Andersson (2007),
Persson (2008), and Peri and Requena-Silvente (2010). The response to a changes in variable
cost should not depend on the substitution elasticity and be the same in all sectors or product
categories. However, when researchers use a variable like distance to proxy trade costs, see for
example Crozet and Koenig (2010), it is not obvious whether the effects should follow equation
5
This is the case for all papers mentioned in the introduction. Eaton et al. (2008) is one example of a
paper that empirically looks at the importance of the extensive margin defined as in Chaney (2008).
6
(2) or (3). One can clearly make the argument that distance raises the iceberg transport costs, but
overcoming a larger distance has a fixed cost aspect as well.6
An ad valorem tariff, on the other hand, is clearly a variable trade cost. Equation (3) predicts
that a reduction in the import tariff of country d should increase the number of firms exporting
from country c by the same amount in each sector or product category. The extensive margin
effect that drives the main result in Chaney (2008) is the larger volume of exports for each new
entrant in sectors with a low substitution elasticity. New exporters are by construction the least
productive firms on the export market and when product differentiation is low (the elasticity is
high), they will have great difficulties competing with more productive competitors and only sell
small volumes. This could lead to a higher increase in the number of exporters in less elastic
sectors if trade flows are only measured when they pass a minimum threshold. Each country has
a minimum threshold below which trade-transactions do not need to be reported and exports by
new entrants in highly elastic product categories are more likely to fall below any given
threshold.
Cole (2012) highlights a different reason why the response in the number of firms should
depend on the substitution elasticity. He uses the framework of Chaney (2008), but argues that
changes in foreign import tariffs (denoted by tcd) affect the level of profits differently from
changes in variable trade costs (denoted by τcd). In the CES demand model, the price-cost
markup is constant and tcd and τcd will change the price and quantity of all exported units in the
same way. A difference, however, is that the tariff is applied to the final price, including the
markup, and it will take away more of the revenue from the differentiated goods sector than a
variable trade cost that is applied to the marginal cost. In particular, it will also take away some
of the profit markup, which is decreasing in the substitution elasticity, and in equilibrium this
lower the number of exported varieties through an income effect.
Taking this into account, Cole (2012) obtains the following productivity threshold for
exporters:
⁄
̅
( )
(
⁄
)
̅
The elasticity of the productivity threshold with respect to changes in fixed costs or variable
trade costs is the same as in Chaney (2008), but the tariff elasticity is higher. Actually, the tariff
elasticity (5) equals the sum of the fixed cost and variable cost elasticities (3) and (4). As a result,
the effect of a reduction in tariff rates on the number of firms that export is now also decreasing
in the substitution elasticity. The intuition is that the amount of income that was taxed away by
6
The evidence in Hummels and Skiba (2004) that shipping costs have an important per-unit dimension
goes in the same direction.
7
government with a tariff and assumed to be spent on the numeraire good, i.e. not on the
differentiated goods subject to tariffs, is magnified for low-elasticity goods as the tariff is also
applied to the profit markup. The intensive margin effect of a tariff reduction, the consumer
price falls and quantity rises, is augmented by an extensive margin effect as more firms start
exporting, especially in low-elasticity sectors.
3. The empirical model
To identify the tariff elasticity of exports as cleanly as possible, we adopt the triple difference-indifferences approach of Frazer and Van Biesebroeck (2010). The dependent variable is an
indicator which equals one if we observe positive exports of product p from country c to country
d in year t and equals zero otherwise. The key explanatory variable is the (logarithm of the)
import tariff that applies to this trade flow and its coefficient captures the extensive margin
response to a tariff decline.
We estimate the equation with least squares, which implies a linear probability model, as it
allows the inclusion of extremely flexible controls for demand and supply effects. 7 In
expectation, the dependent variable captures the probability that the productivity of the most
efficient firm exceeds the threshold ̅ in equation (4). The productivity dispersion,
substitution elasticity, and tariff rates certainly differ by product, making the productivity
threshold product specific. In addition, the fixed and variable trade costs as well as the wage rate
in country c are likely to be product specific too, but also to vary over time. To control flexibly
for all these factors, the probability of positive export flows is modeled as a function of
interaction fixed effects for each possible pairing of the three dimensions: product, time, and
exporter-importer pairs.
For the baseline results, we estimate a separate regression for each product category k:
[
]
(
)
A unit of observation is a country pair cd, measuring exports from 124 exporter countries c to
each of 3 destination countries d, in a particular product category p and year t. China is one of
the three destination countries and we use India and Indonesia as control destinations. Almost
every product imported in China has experienced a tariff reduction over the 2000-2006 period.
The import status of each product in China from 124 countries is the treated group of
observations and the import status of the same products in India or Indonesia is the control group
as tariffs to those destinations were constant over time. The tariff elasticity is then identified by
the strength of the association between product-level tariff reductions and increased import
penetration in China relative to changes in import penetration in the control destinations.
7
As a robustness check we also estimate a probit model that incorporates the discrete nature of the
dependent variable, but does not allow as flexible controls.
8
The fixed effects included in the regression are crucial. Product-year effects soak up changes
in demand, quality, and production technology at the product level that are global in nature. We
have chosen India and Indonesia as control destinations as we expect the evolution of their
import demand for different products to be similar to China, because they are also WTO
members, important manufacturing bases in Asia, large developing countries, and located at a
similar distance from important exporters. Country-pair/year effects soak up not only the level
of import demand and export supply of each country, but also changes in the national business
cycle, level of development, and overall trade expansion due to greater integration in the world
economy. Country-pair/product effects capture the baseline probability of bilateral trade for
every product. We do not aim to explain the level of import penetration, only to what extent
relative changes can be explained by tariff reductions.8
The parameter of interest on the logarithm of import tariffs measures the elasticity of the
probability a product is exported to China with respect to the import tariff. In this specification,
we implicitly assume constant export elasticities, but we also estimate equation (6) using the
level of tariffs which assumes a constant effect of each percentage point reduction in tariffs, i.e. a
constant semi-elasticity. k indexes product categories which are expected to differ in their
elasticity of demand. The different βk parameters are estimated in separate regressions, which
would be the same as a single estimation with the γcdt interaction effects varying by product
category. According to the model in Cole (2012), we expect the tariff elasticity to be the larger
in absolute value if the substitution elasticity for the product category is lower.
The intuition for equation (6) can be illustrated by explicitly writing out the triple-difference,
over each dimension: time, product, and country-destination:
(
)
(
)
(
)
(7)
For two products (1 and 2) and two export destinations (C and I) we calculate the change over
time in the probability that a particular country has positive exports. The relative increase in
import penetration for product 1 versus 2 in China is normalized by the relative increase in
import penetration in India or Indonesia. Given that treatment here is continuous and not a zeroone decision, we normalize the double-difference in growth rates by the relative change in
Chinese import tariffs for product 1 versus 2 to obtain the elasticity estimate . 9 The
corresponding difference for India drops out as its tariffs were constant.
The advantage over a usual difference-in-differences (DID) estimation at the country or
product level is the robustness to policy endogeneity (Besley and Case, 2000) or misattribution
of changes to the policy variable due to omitted variables. An aggregate country-level DID
estimation is that anything unusual about the evolution of China’s trade will be differenced out in
8
We use India and Indonesia as control destinations, but do not need their actual tariff levels as these are
almost perfectly constant over the sample period and absorbed by the country-pair/product effects.
9
The right-hand side in the equation approximates (Δ%t)C,1 – (Δ%t)C,2.
9
our case using the product dimension. The advantage over a product-level DID estimation using
only a single destination is that strong performance of product 1 exports to China for other
reasons than tariff reductions are normalized by the relative export performance to
India/Indonesia. Observations in the data sets used by Debaere and Mostashari (2010) and
Buono and Lalanne (2012) also vary across all three dimensions—country, product, and time—
but these authors only use additive fixed effects. As a result, much of their identifying power
still comes from variation in absolute growth rates of import penetration at the country-product
level. Only one specification in Buono and Lalanne (2012) uses the same triple-difference set-up
and it completely wipes out the contribution of the extensive margin effect in their case, which is
a key finding of their analysis.
Following China’s entry in the WTO, its tariffs declined for almost all products, but to
varying extents.
By identifying separate elasticities for different product categories
(homogeneous, referenced-priced, and differentiated goods), the export responses are allowed to
differ across categories and are only identified by assuming a constant elasticity within each
category. We first estimate equation (6) by pooling across all exporters, again assuming constant
tariff elasticities across countries, but we relax this subsequently and estimate the model
separately for OECD countries, countries that are members of the WTO but not the OECD, and
countries belonging to neither group. Following Subramanian and Wei (2007), who argue that
developed countries participate more actively in trade negotiations to obtain tariff cuts most
likely to benefit their firms, we expect the largest elasticities for the first group of countries.
For the baseline results, we estimate equation (6) using observations for year 2001 and 2006,
Rauch (1999)’s conservative classification to group products into three categories, and both India
and Indonesia as control destinations. We perform robustness checks on different time periods
and including all intervening years, using Rauch (1999)’s liberal classification or the Broda et al.
(2006)’s estimated import demand elasticities to categorize goods, using only one of the two
control destinations or neither, and estimating a probit model.
4. Data
Chinese import tariffs
The first piece of information we need is the import tariff that firms face when they export to
China. We obtained Chinese import tariffs at the 8-digit level of the Harmonized System (HS)
classification of goods from the Chinese National Bureau of Statistics. Because we use bilateral
trade flows between different country-pairs, e.g. US-China and Germany-India, we need to work
at the 6-digit HS level, because more detailed definitions of product categories are not uniform
across countries.
Ideally, tariff rates should be aggregated using as weights trade flows unimpeded by
protectionist barriers. In practice this is not possible. One possibility would be to use
information from the last observed period—when tariff rates and non-tariff barriers are lowest—
10
to construct an import-weighted average with constant weights over the entire period. A
problem we face is that the HS classification at the 8-digit level did undergo several changes
between the 1996 and 2002 versions. To avoid using current import weights in early years
which would bias the average tariff downward and to avoid introducing tariff changes that are
due to reclassifications, we use a simple unweighted average. Given that many 6-digit categories
only contain a single 8-digit sub-category and even if there are several sub-categories their tariff
rates tend to be rather similar, the exact choice of weight is not that important.
The evolution over time of the average Chinese import tariffs in the three product categories
defined by Rauch (1999) is depicted in Figure 1. The average declines significantly in each
category between 2000 and 2006. In fact, Chinese tariffs had already declined substantially even
before 2000, but we do not rely on those tariff cuts as they coincide with tariff cuts in other
countries that were negotiated in the Uruguay Round WTO agreement. Import tariffs for India
and Indonesia are almost perfectly constant over the sample period and all identifying power
comes from the product-level variation in the extent of Chinese tariff reductions.
[Insert Figure 1 approximately here]
Bilateral export flows
The dependent variable in the analysis is an indicator of positive import flows in China, India,
and Indonesia between 2000 and 2006, also at the 6-digit HS level. Chinese trade data is from
China Customs and we aggregate them up from the firm-level. It allows us to include only
imports that enter the country under the ordinary trade regime and are thus subject to import
tariffs. Imports into India and Indonesia are taken from the BACI database of CEPII. 10 We
limit the sample to manufactured products, omitting agricultural and mining products. Exporting
countries are included in the sample if they show positive exports to at least one of the three
destinations in the first or last year of the sample. The list of 135 countries is provided in Table
A.1 in the Appendix and they are classified in three groups depending on membership in the
WTO and/or OECD.
To classify goods in categories that differ in substitution elasticity, we rely on the three-way
classification in Rauch (1999) which is at the 4-digit SITC level. A concordance table between
the HS and SITC classifications is used to link this information to the trade data.11 From the
5,040 products in the BACI data set (at the 6-digit HS level), we are left with 3,378
manufactured goods across the three categories. Rauch (1999) developed a conservative and
10
BACI reconciles the declarations of exporters and importers in the United Nations’ COMTRADE
database. It extend considerably the number of country-pairs for which trade data are available. Details
on the harmonization procedure are provided in Gaulier and Zignago (2010) and the data set itself is
available online at http://www.cepii.fr/anglaisgraph/bdd/baci.html.
11
Both the Rauch classification and concordance table are available at:
http://www.macalester.edu/research/economics/page/haveman/Trade.Resources/
11
liberal classification and we show for both cases the breakdown of products in Table 1. Most
products fall in the category of differentiated goods, but note that this is meant to indicate further
differentiation within each product. Homogenous products are fairly rare, within our sample of
manufactured goods, representing between 4 and 5.7 percent of the sample.
[Insert Table 1 approximately here]
As an alternative way to classify products, we perform a robustness check using the
estimated import demand elasticities for China from Broda et al. (2006). They are provided at
the 3-digit HS level (170 categories) and we use them to classify the 6-digit products evenly into
low, medium, and high elasticity groups. We also use the elasticity estimates directly, which
have a median of 3.4, mean of 6.2, and standard deviation of 11.9.
Once the set of countries and products is determined, zero export observations are added to
the sample by balancing the panel for each country along the product and time dimensions. The
number of observations in each regression is then simply obtained by multiplying the number of
years by the number of exporting countries, import destinations, and products. In the baseline
analysis, we use 2 years (2001 and 2006), exports from 124 WTO members to 3 destinations
(China, India and Indonesia), and 3,378 products for a total of 2,513,232 observations: 1,737,984
for differentiated goods, 675,552 for referenced goods, and 99,696 for homogeneous goods.
In Table 2, we list the number of products for which we observe positive imports in 2000 or
in 2006 by category and by export destination. The initial import penetration varies between 7.0
and 8.4 percent of the exporter-product pairs, leaving a lot of scope for growth at the extensive
margin of trade. In the following six year, the import penetration increases everywhere, but most
strongly for China and for differentiated goods.
[Insert Table 2 approximately here]
5. Empirical Results
Baseline estimates of equation (6) are reported in Table 3. As mentioned, this uses data for 2001
(the year before China entered the WTO) and 2006, 124 exporters (all WTO members), 3
destinations (China, Indonesia, India), and all manufactured products in the conservative
classification of Rauch (1999). We estimate the tariff semi-elasticities first using the form
suggested by the theory, interacting the product category dummies with ln(1+t) and report the
results in the first column. The estimates using the tariff rates directly, in the second column,
exhibit the same pattern.
A first thing to note is that all estimates are negative and significant at the 1 or 5 percent level.
This is in contrast with the previous literature that only finds small effects of tariff reductions at
the extensive margin. Second and most important, the elasticity is much higher in absolute value
for differentiated goods which are assumed to have the lowest demand elasticities, followed by
12
reference-priced goods, and then the homogeneous goods. Both findings are consistent with the
predictions in Cole (2012). Note that the differentiation is taking place within each product
category and our dependent variable measures the first export instance of any product from an
exporter. The possible existence of many more varieties within a particular HS 6-digit product
that is classified in the differentiated goods category does not influence our estimates.
[Insert Table 3 approximately here]
Results in Table 4 illustrate that the estimates are highly robust when several elements of the
benchmark set-up are changed. In columns (1) and (2) each of the control destinations is used
separately. In column (3), we use the liberal classification of Rauch (1999) identifies fewer
products as differentiated. In column (4), the sample period already starts in 2000, two years
before the negotiated tariff declines start to take effect. Tariff declines in the pre-WTO period
were voluntarily and they could be concentrated in sectors where the scope for increased import
penetration was low.12 In column (5) we do not limit the sample to the observations in the
beginning and end periods, but use all the intervening years. In the last two columns we change
the estimation strategy. First, we omit the control destinations and conduct a product-level
difference-in-differences analysis. Next, we estimate a probit model with random effects for
each product, as in Baldwin and Harrigan (2011).
[Insert Table 4 approximately here]
A few patterns are worth highlighting. First of all, the qualitative pattern is robust in each
variation. The estimate for differentiated products is always the largest and highly significant in
all but one instance. The only case where it becomes insignificant is when all intervening years
(all 6 years from 2001-2006) are included. The inclusion of many instances where tariffs do not
change lead to much higher standard errors in the within estimator.13 The results in column (6)
confirm the findings in Buono and Lalanne (2012) that the estimated responses are a lot higher
when cross-sectional variation contributes to the identification. Results for the other two product
categories are always lower in absolute value and sometime become insignificant. In six of the
seven columns, the elasticity for referenced-priced goods is more negative than for homogenous
goods.
In Table 5, we allow the elasticities not only to differ by product category, but also by
country grouping. The ordering of the elasticities, highest for differentiated and lowest for
homogenous goods, applies to OECD countries as well as to non-OECD countries. Within each
The need to observe “ordinary trade” flows separately from “duty-free” processing trade flows prevent
us from going further back in time.
13
The same is true comparing the results in Rose (2004) and Subramanian and Wei (2007) for the WTO
effects – including intervening years lowers the significance of the estimates notably. We also obtained
results that confirm the delayed effect of WTO entry in Chang and Lee (2011). Varying the end year, the
results are only estimated significantly if we extend the sample period at least to the year 2005.
12
13
product category, exports from OECD countries are always more elastic. The difference is
especially large for differentiated products: a point estimate of –0.221 versus –0.045. One
explanation is that exporters from OECD countries face a lower average substitution elasticity
within each category, possibly because they sell more quality-differentiated goods. With this
interpretation, it is natural that the difference in estimates is a lot smaller for homogenous goods,
–0.043 versus –0.020, point estimates which are not significantly different.14
[Insert Table 5 approximately here]
In the last column of Table 5 we report results for exports by non-WTO members. The three
estimates are now all insignificant, indicating that firms from these countries did not increase
their import penetration following China’s entry into the WTO. This can be considered a
placebo test for our identification strategy. The tariffs that China applies to these countries did
not have to change in the same way as its MFN tariff rate and it is thus not surprising that they
have no effect on their extensive margins. The estimates in Table 5 confirm that the WTO
promotes trade strongly but unevenly across countries (Subramanian and Wei, 2007).
For the next set of results, we use an alternative to the Rauch (1999) classification to measure
the heterogeneity of different products. We rely on the import demand elasticities estimated for
China from Broda et al. (2006) to proxy each product’s substation elasticity.15 For the results in
Table 6 we add an interaction between the continuous elasticity estimates and the tariff variable
in the regression equation. In each cases, the coefficient estimate on the interaction term
strongly indicates that the absolute value of the extensive margin response is decreasing in the
substitution elasticity. These estimates directly confirm the comparative statics prediction in
Cole (2012).
[Insert Table 6 approximately here]
We have also run the regressions separately for the category of differentiated products. The
reduction in the estimated tariff elasticity for products that have higher demand elasticities is a
lot more pronounced on this sample. The positive coefficients on the interaction term are
approximately twice as high as on the full sample and they are always statistically significant at
14
Another distinction is that firms in non-OECD countries had to confront fiercer competition in their
home markets as Chinese products could now be imported at lower MFN-tariffs. In the U.S. and the E.U.
Chinese imports already benefitted from these low rates prior to China’s entry in WTO. With heightened
competition at home, expanding abroad might not be the first concern for these firms.
15
We use these estimates to show robustness rather than for the benchmark results because one might be
concerned about endogeneity as the elasticities are themselves estimated from Chinese import flows. It is
only a minor issue as import growth at the extensive margin accounts for only a small fraction of total
import growth and the sample periods barely overlap. Moreover, to the extent that endogeneity is an
issue, it will bias the estimates towards larger responses in high elasticity sectors, while the key insight
from Table 6 is the negative correlation between the extensive margin effect and the demand elasticity.
14
the 1 percent level. To compare the responses between OECD and non-OECD countries, we
need to take both coefficients into account. While the reduction in the extensive margin effect
for more elastic products is more pronounced for OECD countries, no reasonable value for the
substitution elasticity can overturn the large difference on the uninteracted tariff coefficient. On
both samples, the tariff elasticity is still estimated much larger for OECD countries.
Alternatively, we can use the import demand elasticities from Broda et al. (2006) to divide
products evenly into low, medium, and high elasticity groups and estimate a different response
for each group. These results, reported in Table A.2 in the Appendix, confirm that the response
is strongest for low elasticity products. Even the point estimates, at –0.123 on the sample of all
countries, –0.243 and –0.079 for OECD and non-OECD countries, are similar to those for
differentiated products in Tables 3 and 5, which were –0.092, –0.221 and –0.045 respectively.
and for the latter group the ordering of the three estimates coincides entirely with the inverse
ordering of the elasticities.16 In most cases, the ordering by magnitude of the effect is the inverse
of the elasticity ordering.
In Figure 2, we plot the predicted effects together with the initial import penetration to
illustrate the economic magnitude of the estimates. The average effects of the tariff cuts can be
calculated directly from equation (7) and the estimated elasticities. We report results separately
for OECD and non-OECD countries because the average import penetration is very different—
note the different scaling of the two vertical axis.
Tariff declines clearly have the greatest impact for differentiated product with a low demand
elasticity. The average decline in import tariffs is 0.062 (in log-points) and the estimated effect
of –0.352 for OECD countries implies an increase in the probability of positive imports into
China by 2.2 percentage points. This accounts for almost one third of the total increase in import
penetration of 7.1 percent over the same time period (up from 31.1 in 2001 to 38.2 percent in
2006). As the estimated effect is a lot lower for non-OECD countries, –0.113, the comparable
decline in tariff rates is now predicted to have raised the import penetration by only 0.7
percentage points. However, this is from a much lower initial level and out of a much lower total
increase. The tariff cut is now estimated to account for almost half of the total increase in import
penetration. The effects are more modest when averaging over all differentiated goods, with the
tariff cuts accounting for respectively 16.3 and 15.2 percent of the increased import penetration
for OECD and non-OECD countries. Nevertheless, these fractions are substantially higher than
the 5 and 12 percent shares obtained by Debaere and Mostashari (2010) over the 1989-1999 and
1996-2006 periods.
[Insert Figure 2 approximately here]
16
For OECD countries, the estimate is higher in absolute value for the high elasticity than for medium
elasticity group. This could be due to erroneous elasticity estimates for some products which in a few
cases even exceed one hundred.
15
6. Discussion
We have interpreted our empirical findings using the heterogeneous firm models with
horizontally differentiated goods from Chaney (2008) and Cole (2012). The larger tariff
elasticity at the extensive margin for goods facing a less elastic demand is driven by the higher
profit mark-ups applied on these goods. When tariffs decline, the equilibrium price adjusts
similarly everywhere, but a greater absolute amount of profit is released for low elasticity goods.
This makes it easier to cover the fixed cost of exporting and enticing even more firms into the
export market. There are of course other forces that operate on firms which are omitted from the
model. We explore several of them in this section.
A first worry is that less market entry might be possible for homogenous goods, simply
because many countries do not produce them. The above results are for a balanced panel, which
implicitly assumes that all countries have the potential to export all 3,378 manufacturing
products to China. Agricultural and mining products were already omitted from the sample as
they are not produced in each country, simply for comparative advantage reasons. A high
substitution elasticity will lead to very competitive markets, low markups, and few surviving
producers. Even for manufactured goods we expect fewer potential entrants in such sectors. In
the theory, there is a full productivity distribution of firms in each country for each sector, but
that is clearly a strong assumption.
To take into account that some countries may never export certain products, even if Chinese
import tariffs fall to zero, we restrict the sample for each country to those products with positive
exports to the rest of the world at any point in the 2000-2006 period. Estimation on such an
unbalanced panel is a lot more time-consuming, because we now need to perform the withintransformation also on the interaction fixed effects (see Van Biesebroeck and Frazer, 2010).
Results in Table 7 illustrate that the basic pattern is maintained. In particular, the estimates for
OECD countries are almost unchanged from those obtained from the balanced panel. For nonOECD countries, the estimates become a lot more negative, as expected, but the change is
similar for all three product categories. For homogenous goods the point estimate rises from
–0.020 to –0.037, but the difference with heterogeneous goods remains similar as those estimates
increase by approximately the same amount (from –0.045 to –0.059). The lack of exports for
some goods is primarily an issue for poorer countries, but it is not concentrated for homogenous
products.
[Insert Table 7 approximately here]
A second concern is the potential endogeneity of tariff reductions. They were decided in
bilateral negotiations between China and existing WTO members and it is reasonable to believe
that countries would push harder for reductions in categories where they expect strong export
potential, for example because of comparative advantage (Subramanian and Wei, 2007). This is
especially a concern for OECD countries, as the United States and the European Union were the
16
most active in the negotiations and other OECD countries are likely to have similar comparative
advantage. Such endogeneity could contribute to the larger estimates for differentiated goods as
OECD exports are concentrated in that category and tariff rates also started out higher (Brandt et
al., 2012).
There are several reasons to believe that this concern is unlikely to explain the entire effect.
Evidence in Blonigen and Cole (2011) suggests that since 1993 the presence of FDI in a sector is
a better predictor for Chinese import tariffs than foreign export elasticities. Brandt et al. (2012)
illustrate that tariff cuts between 1995 and 2006 are almost perfectly proportional to the initial
tariff rate showing very little discretion in rate reductions. The averages in Figure 1 suggest that
goods in each of the three categories have enjoyed rather similar tariff declines. If anything, the
decline was largest for homogenous goods. Moreover, given that non-OECD countries are not as
involved in WTO negotiations and that their comparative advantage is likely to differ from the
U.S. and E.U., Chinese tariff reduction can be considered exogenous for them. The fact that the
absolute values of the estimates for the different product categories in Tables 3 and 5 are also
inversely related to the substitution elasticity for these countries is reassuring.
The next potentially important element that is absent from the model is quality differentiation.
To the extent that higher quality goods enjoy lower rates of substitution in demand, the
mechanism we focus on encompasses quality effects. However, Baldwin and Harrigan (2011)
augment the Melitz model and illustrate that even differentiation within a sector, holding
substitution elasticity constant, can be caused by quality differences. They predict that, all else
equal, the highest quality goods will be shipped the farthest. This mirrors the predictions and
empirical evidence in the paper by Hummels and Skiba (2004), who assume and find evidence
for per-unit, instead of iceberg, shipping costs.
Given that tariffs, like distance, raise the variable trading costs, the above suggests that
countries with higher tariffs will also be receiving, all else equal, imports of higher quality. At
the margin, lowering tariffs just enough to entice the first firm to start exporting is similar
(hypothetically) to lowering the distance just enough to make exports profitable. Both Hummels
and Skiba (2004) and Baldwin and Harrigan (2011) predict that the first exporter will be
producing a high quality product. The evidence in Schott (2004) demonstrates that richer
(OECD) countries are more likely to export high quality goods. Conditional on a sector’s
substitution elasticity, the extensive margin tariff elasticity should be higher for producers of
high quality goods, i.e. OECD countries, consistent with our findings.
Note that this reasoning does not reveal whether the response to further tariff cuts for
additional firms will always be larger for higher quality goods. We only argued that the first
good exported from exporter c to destination d will be high quality and this is the particular
dependent variable we focus on. Without an explicit model, we cannot say anything about the
responsiveness of the extensive margin for additional exporters, needed to use number of
exporters as dependent variable, or the level of sales for new exporters, to compare with the
17
predictions in Chaney (2008). Finally, the fact that that the ordering of tariff elasticities across
product categories also holds for non-OECD countries, which are unlikely to be the quality
leaders in all but a few examples, suggests that quality is unlikely to be the whole explanation for
the observed pattern.
Another element missing from the model is an endogenous mark-up. Marginal exporters
have lower productivity and thus higher marginal cost than infra-marginal exporters. In models
that do not assume the CES demand structure, marginal exporters are predicted to choose a lower
mark-up in equilibrium as they face a more elastic residual demand. This difference between
marginal and infra-marginal firms is especially pronounced in industries with on average lower
demand elasticity. It is an open question whether a differential impact on mark-ups for firms in
high or low elasticity sectors will reinforce or diminish the difference in extensive margin
response derived in the CES demand case. The analysis of different types of liberalization in
Melitz and Ottaviano (2008) highlights the important general equilibrium effects that influence
entry in all countries. Spearot (2012) uses a model with varying mark-ups to study the impact of
tariff declines, but his theoretical results are derived abstracting from the extensive margin
response.
7. Conclusion
We have used the Chinese experience following its entry into the WTO in 2001 to verify how the
response at the extensive import margin to a tariff decline varies across different types of
products. We use a triple difference-in-differences approach to control flexibly for demand and
supply shocks and the comparative advantage of trading partners. The responses are purely
identified from the “within” changes within a country-pair-product observation and the export
response to China is normalized by the export responses to India and Indonesia.
Three findings are worth highlighting. First, the extensive margin response to tariff cuts is
strong, precisely estimated, and robust to variations in the estimating assumptions. The absolute
magnitude of the response to the tariff decline represent a large fraction of the total extensive
margin response in some cases. Second, the response is always estimated higher for
differentiated goods (using Rauch, 1999) or for goods with a low import substitution elasticity
(using Broda et al., 2006). This pattern fits the theoretical predictions of Cole (2012) who has
shown that the tariff elasticity at the extensive margin should differ from a variable cost elasticity.
Third, the responses are a lot higher for OECD than for non-OECD countries, but reassuringly
they are completely absent for non-WTO members. We offer several explanations for the
stronger response for rich countries, but the consistency of the pattern across the product
categories also for the poorer countries suggests that the differences in import demand elasticities
are likely to be an important part of the explanations for the differential responses.
While the overall magnitude of the tariff response is always estimated higher for OECD
countries, tariff cuts often account for a larger fraction of the observed increase in import
18
penetration for non-OECD countries. This holds in most cases and especially in the three
product categories with strong responses as illustrated in Figure 2. It underscores that trade
liberalization can be an important source of long-run trade growth for developing countries. As
current levels of import penetration tend to be relatively low for them, extensive margin
responses are especially important.
19
References
Andersson, M., 2007. “Entry Costs and Adjustments on the Extensive Margin – An Analysis of
How Familiarity Breeds Exports.” Working Paper Series in Economics and Institutions of
Innovation No. 81, Royal Institute of Technology, CESIS.
Baier, S.L., Bergstrand, J.H., 2007. “Do Free Trade Agreements Actually Increase Members’
International Trade?” Journal of International Economics 71: 72-95.
Baldwin, R., Harrigan, J., 2011. “Zeros, Quality, and Space: Trade Theory and Trade Evidence.”
American Economic Journal: Microeconomics 3: 60-88.
Besley, T.J., Case, A., 2000. “Unnatural Experiments? Estimating the Incidence of Endogenous
Policies.” Economic Journal 110(467), 672-694.
Blonigen, B., Cole, M.T., 2011. “Optimal Tariffs with FDI: The Evidence.” University College
Dublin Working Paper No. 11/21.
Broda, C., Weinstein, D., 2006. “Globalization and the Gains from Variety.” Quarterly Journal
of Economics 121(2): 541-585.
Brandt, L., Van Biesebroeck, J., Wang, L., Zhang Y., 2012. “WTO Accession and Performance
of Chinese Manufacturing Firms.” CEPR Discussion Paper No. 9166.
Broda, C., Greenfield, J., Weinstein, D., 2006. “From Groundnuts to Globalization: A Structural
Estimate of Trade and Growth.” NBER Working Paper No. 12512.
Bustos, P., 2011. “Trade Liberalization, Exports and Technology Upgrading: Evidence on the
Impact of MERCOSUR on Argentinian Firms.” American Economic Review 101(1): 304-340.
Buono, I., Lalanne G., 2012. “The Effect of the Uruguay Round on the Intensive and Extensive
Margins of Trade.” Journal of International Economics 86(2): 269-283.
Cole, M. T., 2012. “Distorted Trade Barriers: A Dissection of Trade Costs in a Gravity Model.”
Working Paper, University College Dublin.
Chaney, T., 2008. “Distorted Gravity: The Intensive and Extensive Margins of International
Trade.” American Economic Review 98(4): 1707-1721.
Chang, P.-L., Lee, M.-J., 2011. “The WTO Trade Effect.” Journal of International Economics
85(1): 53-71.
Crozet, M., Koenig, P., 2010. “Structural Gravity Equations with Intensive and Extensive
Margins.” Canadian Journal of Economics 43(1): 41-62.
Debaere, P., Mostashari, S., 2010. “Do Tariffs Matter for the Extensive Margin of International
Trade? An Empirical Analysis.” Journal of International Economics 81(2): 163-169.
Dutt, P., Mihov, I., Van Zandt, T., 2011. “Does WTO Matter for the Extensive and the Intensive
Margins of Trade?” Working Paper, INSEAD.
Eaton, J., Eslava, M., Kugler, M., Tybout, J., 2008. “The Margins of Entry into Export Markets:
Evidence from Colombia.” In Helpman, E., Marin, D., Verdier, T. (eds.) The Organization of
Firms in a Global Economy, Cambridge, MA: Harvard University Press.
20
Egger, H., Egger, P., Greenaway, D., 2008. “The Trade Structure Effects of Endogenous
Regional Trade Agreements.” Journal of International Economics 74(2): 278-298.
Feenstra, R.C., Markusen, J.R., Rose, A.K., 2001. “Using the Gravity Equation to Differentiate
among Alternative Theories of Trade.” Canadian Journal of Economics 34(2): 430-447.
Frazer, G., Van Biesebroeck, J., 2010. “Trade Growth Following the African Growth and
Opportunity Act.” Review of Economics and Statistics 92(1): 128-144.
Gaulier, G., Zignago, S., 2010. “BACI: International Trade Database at the Product-level. The
1994-2007 Version.” CEPII Working Paper No. 2010-23.
Haveman, J.D., Nair-Reichert, U., Thursby, J.G., 2003. “How Effective are Trade Barriers? An
Empirical Analysis of Trade Reduction, Diversion, and Compression.” Review of Economics
and Statistics 85(2): 480-485.
Hummels, D., Skiba A. 2004. “Shipping the Good Apples Out? An Empirical Confirmation of
the Alchian-Allen Conjecture.” Journal of Political Economy 112(6): 1384-1402.
Lawless, M., 2010. “Deconstructing Gravity: Trade costs and Extensive and Intensive Margins.”
Canadian Journal of Economics 43(4): 1149-1172.
Melitz, M.J., 2003. “The Impact of Trade on Intra-Industry Reallocations and Aggregate
Industry Productivity.” Econometrica 71(6): 1695-1725.
Melitz, M.J., Ottaviano, G.I.P., 2008. “Market Size, Trade, and Productivity.” Review of
Economic Studies 75(1): 295-316.
Peri, G., Requena-Silvente, F., 2010. “The Trade Creation Effect of Immigrants: Evidence from
the Remarkable Case of Spain.” Canadian Journal of Economics 43(4): 1433-1459.
Persson, M., 2012. “Trade Facilitation and the Extensive Margin.” Journal of International
Trade and Economic Development, forthcoming.
Rauch, J., 1999. “Networks versus Markets in International Trade.” Journal of International
Economics 48(1): 7-35.
Rose, A.K., 2004. “Do We Really Know That the WTO Increases Trade?” American Economic
Review 13(4): 682-698.
Schott, P.K., 2004. “Across-Product Versus Within-Product Specialization in International
Trade.” Quarterly Journal of Economics 119(2): 647-678.
Shi, H., 2011. “Gravity or Distorted Gravity? Evidence from Chinese Data.” Ohio State
University, Working Paper.
Spearot, A.C., 2012. “Variable Demand Elasticities and Tariff Liberalization.” Journal of
International Economics, forthcoming.
Subramanian, A., Wei, S., 2007. “The WTO Promotes Trade, Strongly but Unevenly.” Journal
of International Economics 72(1): 151-175.
21
Figure 1: Evolution of average Chinese import tariffs by products category
20%
19.4%
19.0%
15%
10.9%
12.4%
10%
9.9%
6.8%
5%
differentiated goods
homogeneous goods
reference-priced goods
0%
2000
2001
2002
2003
2004
2005
2006
Notes: Products are classified using Rauch’s (1999) conservative standard. Unweighted average of tariffs
at the HS 6-digit level.
22
Figure 2: Estimated impact of tariff reductions on average import penetration
40%
4%
30%
3%
20%
2%
10%
1%
0%
0%
diff.
goods
low
elasticity
diff. &
low elas
diff.
goods
OECD countries
low
elasticity
diff. &
low elas
non-OECD countries
Increased penetration for other reasons (by 2006)
Increased penetration due to tariff cuts (2001-06)
Import penetration in 2001
23
Table 1: Breakdown of the sample by product category
Conservative classification
Liberal classification
Number
Fraction
Number
Fraction
2,336
69.1%
2,133
64.8%
Referenced-priced goods
908
26.9%
972
29.5%
Homogeneous goods
134
4.0%
185
5.7%
Differentiated goods
Total
3,378
3,290
Table 2: Number of positive imports flows (3,378 products from 135 countries)
China
2000
Differentiated
Change
26,068 38,309 +47.0%
Reference-priced 7,265
Homogeneous
2006
India
549
9,227 +27.0%
831 +51.4%
2000
2006
Indonesia
Change
28,711 39,843 +38.8%
8,905 10,853 +21.9%
722
952 +31.9%
2000
2006
Change
23,506 27,025 +15.0%
7,625
7,620
–0.1%
642
637
–0.8%
Total
33,882 48,367 +42.8%
38,338 51,648 +34.7%
31,773 35,282 +11.0%
(% of potential)
(7.4%) (10.6%)
(8.4%) (11.3%)
(7.0%)
Note: Uses the conservative classification of Rauch (1999)
24
(7.7%)
Table 3: Baseline estimates of extensive margin responses (tariff semi-elasticities)
Dependent variable:
Dummy for positive exports
ln(1+tariff) *
product category
tariff *
product category
Differentiated product dummy
-0.092***
(0.010)
-0.079***
(0.006)
Reference-priced product dummy
-0.037***
(0.010)
-0.029***
(0.008)
Homogenous product dummy
-0.026**
(0.012)
-0.017**
(0.007)
Observations
2,513,232
2,513,232
Explanatory variable:
Notes: Sample period is 2001-2006, export destinations are China, India, and Indonesia, using the
conservative classification of products in the three categories. ***, **, and * indicate significance at the
1%, 5% and 10% level.
25
Table 4: Robustness checks
Dependent variable: Dummy for positive exports
Only India as
control
destination
Only Indonesia
as control
destination
Liberal
product
classification
Extended
period:
2000-2006
Include all
intervening
years
DID estimation
w/o destination
control
Probit
estimation
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Differentiated
-0.081***
(0.008)
-0.103***
(0.010)
-0.088***
(0.009)
-0.110***
(0.007)
-0.026
(0.041)
-0.206***
(0.061)
-0.062***
(0.007)
Referenced-priced
-0.041***
(0.010)
-0.034**
(0.014)
-0.052***
(0.011)
-0.024***
(0.009)
-0.008
(0.018)
0.006
(0.038)
-0.015
(0.010)
Homogeneous
-0.040***
(0.010)
-0.012
(0.012)
-0.035**
(0.016)
-0.019
(0.015)
0.022
(0.026)
-0.024
(0.018)
-0.0003
(0.012)
Observations
1,675,488
1,675,488
2,447,760
2,513,232
7,340,304
959,352
1,256,616
ln(1+tariff)
Notes: In the benchmark estimation, the sample period is 2001 and 2006, export destinations are China, India, and Indonesia, and the conservative classification
of products is used. In each panel one of these dimensions is changed. ***, ** and, * indicate significance at the 1%, 5% and 10% level.
26
Table 5: Estimation by country groups
Dependent variable:
Positive export dummy
OECD countries
(WTO members)
Non-OECD countries
(WTO members)
Non-WTO members
(non-OECD)
Differentiated product
-0.221***
(0.050)
-0.045***
(0.008)
-0.007
(0.015)
Referenced-priced prod.
-0.063**
(0.025)
-0.028***
(0.005)
0.000
(0.000)
Homogeneous product
-0.043**
(0.020)
-0.020*
(0.011)
0.015
(0.023)
668,844
1,844,388
222,948
ln(1+tariff) *
Observations
Notes: Same estimation assumptions as in the benchmark case: sample period is 2001-2006, export
destinations are China, India, and Indonesia, and using the conservative classification of products. ***,
**, and * indicate significance at the 1%, 5% and 10% level.
27
Table 6: Estimates using interactions with Broda et al. (2006) import demand elasticities
Dependent variable:
Positive export dummy
All countries
OECD countries
Non-OECD countries
ln(1+tariff)
-0.078***
(0.004)
-0.155***
(0.024)
-0.050***
(0.007)
ln(1+tariff) * demand elast.
0.0014***
(0.0005)
0.0018*
(0.0011)
0.0013***
(0.0003)
Observations
2,473,800
658,350
1,815,450
ln(1+tariff)
-0.113***
(0.010)
-0.254***
(0.051)
-0.062***
(0.009)
ln(1+tariff) * demand elast.
0.0026***
(0.0005)
0.0039***
(0.015)
0.0022***
(0.0002)
Observations
1,698,552
452,034
1,246,518
(a) All products
(b) Differentiated products
Notes: The tariff variable is included uninteracted as well as interacted with China’s import demand
elasticity. Same estimation assumptions as in the benchmark case: sample period is 2001-2006, export
destinations are China, India, and Indonesia, and using the conservative classification of products. ***,
**, and * indicate significance at the 1%, 5% and 10% level.
28
Table 7: Estimates using only confirmed exported products for each country
Dependent variable:
Positive export dummy
All countries
OECD countries
Non-OECD countries
Differentiated product
-0.113***
(0.016)
-0.221***
(0.065)
-0.059***
(0.019)
Referenced-priced prod.
-0.051**
(0.023)
-0.062*
(0.036)
-0.044**
(0.018)
-0.040
(0.029)
-0.029
(0.035)
-0.037
(0.030)
880,740
324,891
555,849
ln(1+tariff) *
Homogeneous product
Observations
Notes: Sample is an unbalanced panel in the product dimension. For each exporter country, only products
that are observed as exports to any country over the sample period are included. Same estimation
assumptions as in the benchmark case: sample period is 2001-2006, export destinations are China, India,
and Indonesia, and using the conservative classification of products. ***, **, and * indicate significance
at the 1%, 5% and 10% level.
29
Table A.1: List of exporters (135 countries)
Group
Countries
OECD WTO
members
(33 countries)
Australia, Austria, Belgium/Luxemburg, Canada, Chile, Czech Rep.,
Denmark, Estonia, Finland, France, Germany, Greece, Hungary,
Iceland, Ireland, Israel, Italy, India, Mexico, Netherlands, New Zealand,
Norway, Poland, Portugal, Republic of Korea, Slovakia, Slovenia,
Spain, Sweden, Switzerland, Turkey, United Kingdom, United States
Non-OECD WTO
members
(91 countries)
Albania, Angola, Antigua and Barbuda, Argentina, Bahrain,
Bangladesh, Barbados, Belize, Benin, Bolivia, Botswana, Brazil, Brunei,
Bulgaria, Burkina Faso, Burundi, Cameroon, Chad, Colombia, Congo,
Costa Rica, Croatia, Cuba, Cyprus, Djibouti, Dominica, Dominican
Republic, Ecuador, Egypt, El Salvador, Fiji, Gabon, Georgia, Ghana,
Grenada, Guatemala, Guinea, Guyana, Haiti, Honduras, Indonesia,
Jamaica, Jordan, Kenya, Kuwait, Kyrgyzstan, Latvia, Lesotho,
Liechtenstein, Lithuania, Madagascar, Malawi, Malaysia, Maldives,
Mali, Malta, Mauritania, Mauritius, Central Africa, Moldova, Mongolia,
Morocco, Mozambique, Myanmar, Namibia, Nepal, Nicaragua, Niger,
Nigeria, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru,
Philippines, Qatar, Cote d'Ivoire, Romania, Rwanda, Saint Lucia, Saudi
Arabia, Senegal, Sierra Leone, Singapore, Solomon Islands, South
Africa, Sri Lanka, Surinam, Swaziland, Tanzania, Thailand, Togo,
Trinidad and Tobago, Tunisia, Uganda, United Arab Emirates, Uruguay,
Venezuela, Zambia, Zimbabwe
Non-WTO
members
(11 countries)
Algeria, Equatorial Guinea, Ethiopia, Greenland, Iraq, Liberia, Dem.
People's Rep. of Korea, Russian Federation, Sudan, Turkmenistan,
Uzbekistan
30
Table A.2: Estimates on product categories defined by Broda et al. (2006) demand elasticities
Dependent variable:
Positive export dummy
All countries
OECD countries
Non-OECD countries
Low elasticity
-0.123***
(0.006)
-0.243***
(0.031)
-0.079***
(0.007)
Medium elasticity
-0.028***
(0.007)
-0.037
(0.031)
-0.024***
(0.009)
High elasticity
-0.047***
(0.007)
-0.117***
(0.027)
-0.021**
(0.009)
2,473,800
658,350
1,815,450
Low elasticity
-0.177***
(0.011)
-0.352***
(0.037)
-0.113***
(0.011)
Medium elasticity
-0.046***
(0.017)
-0.120**
(0.057)
-0.020***
(0.005)
-0.019
(0.016)
-0.137*
(0.076)
0.024
(0.017)
1,698,552
452,034
1,246,518
(a) All products
ln(1+tariff) *
Observations
(b) Only differentiated products
ln(1+tariff)
High elasticity
Observations
Notes: Product category dummies that interact the tariff variable are now defined using the elasticity of
China’s import demand as estimated by Broda et al. (2006). Products are divided equally in low, medium,
and high elasticity groups. Same estimation assumptions as in the benchmark case: sample period is
2001-2006, export destinations are China, India, and Indonesia, and using the conservative classification
of products. ***, **, and * indicate significance at the 1%, 5% and 10% level.
31