A Study on Economic Integration Agreements and Margins of Trade

This paper investigates the impact of Economic Integration Agreements (EIAs) on international trade flows, specifically focusing on the intensive and extensive margins of trade. Using gravity equations and panel data from 1962-2000, the study provides evidence that EIAs affect both margins, with different types of EIAs having varying effects. The research also reveals a differential timing of the two margins' effects, with the intensive margin reacting sooner than the extensive margin. The findings are robust to potential biases, including sample selection, firm heterogeneity, and reverse causality. This study contributes to the literature by offering empirical insights into the relationship between trade liberalization and trade margins, which is crucial for estimating welfare gains from trade.

Economic Integration Agreements
and the Margins of International Trade∗
Scott L. Baier†
, Jeffrey H. Bergstrand‡
, Michael Feng§
April 20, 2013
Abstract
One of the main policy sources of trade-cost changes is the formation of an economic
integration agreement (EIA),which potentially affects an importing country’s welfare.
This paper:(i) provides the first evidence using gravity equations of both intensive and
extensive (goods) margins being affected by EIAs employing a panel data set with a large
number of country pairs, product categories, and EIAs from 1962-2000; (ii) provides the
first evidence of the differential (partial) effects of various “types” of EIAs on these inten-
sive and extensive margins of trade; and (iii) finds a novel differential “timing” of the two
margins’(partial) effects with intensive-margin effects occurring sooner than extensive-
margin effects,consistent with recent theoreticalpredictions.The results are robust to
correcting for potential sample-selection, firm-heterogeneity, and reverse causality biases.
Key words: Free Trade Agreements; International Trade; Extensive Margins; Inten-
sive Margins
JEL classification: F1; F15
∗Acknowledgements:The authors are gratefulto co-editor Dan Trefler and two anonymous referees for
excellent comments that have improved this paper substantively, and to Matthew Clance for excellent research
assistance.We are also gratefulfor comments from presentations at the 2011 Aarhus University conference
on “Globalization:Strategies and Effects” in Koldingfjord (Denmark),ETH University in Zurich, Florida
International University, Drexel University, Georgia Tech, and the University of Notre Dame.We thank Adam
Ayers,R. Ethan Braden,Francisco Briseno,Emma Buckley,Brandon Caruthers,Tom Foote,Mitch Gainer,
Pedro Gimenez, Nick Goode, Drew Hill, Christine Hsieh, Tahir Imtiaz, Humberto Kravetz, Cherrica Li, Lindsey
Lim, Ron Mariutto, Ben O’Neill, Mo Sabet, James Schappler, Andrew Weiler, Alex Wheeler, and Chris Wittman
for excellent research assistance on construction of the EIA data base.
† Affiliation:John E. Walker Department of Economics, Clemson University, Clemson, SC 29634 and Federal
Reserve Bank of Atlanta, Atlanta, GA, USA. E-mail:[e-mail].
‡ Affiliation: Department ofFinance,Department ofEconomics,and Kellogg Institute for International
Studies,University ofNotre Dame,Notre Dame,IN 46556 USA and CESifo, Munich, Germany. E-mail:
[e-mail].
§ Affiliation: John E. Walker Department ofEconomics,Clemson University,Clemson,SC 29634 USA.
E-mail: [e-mail].

1 Introduction
The gravity equation has long dominated the international trade literature as the main econo
metric approach toward estimating ex postthe “partial” (or direct) effects ofeconomic inte-
gration agreements and other natural and policy-based bilateral trade costs on aggregate bi
eraltrade flows.1 Economic integration agreements (EIAs) refer broadly to preferentialtrade
agreements,free trade agreements,customs unions,common markets,and economic unions.2
Recently, Baier and Bergstrand (2007) demonstrated that estimation (ex post ) of the (partia
effects ofEIAs suffered from endogeneity bias,mainly due to self-selection ofcountry-pairs’
governments into agreements.They showed that – after accounting for such bias using panel
techniques – EIAs had much larger effects on trade flows than revealed in the earlier gravity
equation literature and these estimates were more precise.Anderson and Yotov (2011) con-
firmed these findings using panel data also.Such results followed in the footsteps of empirical
trade studies such as Trefler (1993) and Lee and Swagel(1997) that showed that previous
estimates oftrade-policy liberalizations on imports were underestimated considerably due to
endogeneity bias.
While such positive estimates for EIA dummy variables were interpreted in the context of
either Armington or Krugman models as EIAs increasing trade volumes ofexisting homoge-
neous firms (i.e., the “intensive margin”), consideration of zeros in bilateral trade, fixed expo
costs, and firm heterogeneity have led researchers more recently to examine various “exten
margins” of trade.Such extensive margins fall under three general categories:country, goods
(or products), and firm.The existence of zeros in aggregate bilateral trade flows among many
country pairs has led some researchers to explore the probability that a pair of countries tra
at all; to the extent that an EIA affects this probability,this changes the country extensive
margin of trade and potentially economic welfare.
A second margin is known as the “goods” margin of trade.Hummels and Klenow (2005), or
HK, introduced this notion by examining zeros in bilateral trade flows at highly disaggregate
product-category levels.The motivation for HK was to explore in a cross section ofa large
number of products and among a large number of U.S. trading partners a fundamental quest
1Partial (or direct) effects refer to the absence of general-equilibrium (or indirect) effects; see Anderson and
van Wincoop (2003) and Baier and Bergstrand (2009) on partialversus generalequilibrium trade effects of
trade-cost changes.“Aggregate” refers to all “goods” (or industries or product categories).
2In this study, we use the term “preferential trade agreement” to denote one with only partial liberalization
(not free trade).
2

Do large economies export more because they export larger quantities ofa given good (i.e.,
intensive goods margin) or a wider set ofgoods (extensive goods margin)?3 They found in
their cross section that about 60 percent of larger exports of large economies was attributab
to the extensive goods margin;specifically,as the exporter country’s economic size grew,it
exported a larger number of product categories (or “goods”) to more markets.However,HK
did not investigate the relationship between trade liberalizations and the intensive and exten
goods margins of trade.The purpose of this paper is to address this shortcoming.
In this paper,we explore the impact ofEIAs on aggregate trade flows,intensive (goods)
margins, and extensive (goods) margins for a large number of goods, country pairs, and yea4
This is important for at least three reasons.First, the relative impacts on intensive versus
extensive margins oftrade liberalizations may matter for estimating the welfare gains from
trade.Traditionally, the welfare gains from trade liberalizations in models such as Armington
and Krugman arise due to terms-of-trade changes; this is summarized succinctly in Arkolakis
Costinot and Rodriguez-Clare (2012).In Eaton and Kortum (2002),trade liberalizations in-
crease welfare due to an increase in economic efficiency a la the Dornbusch-Fisher-Samuels
model.In the Melitz (2003) model, trade liberalizations lead to gains due to firm heterogene-
ity and resulting increases in aggregate productivity.Second,while Arkolakis,Costinot,and
Rodriguez-Clare (2012) recently argued that the welfare gains are iso-morphic across many
modern “quantitative trade models,” they note that the gains can vary across models allowi
heterogenous firms depending upon the type of Melitz model;hence,the distinction between
intensive margin effects and extensive margin effects is important for ultimately quantifying
with more precision the “gains from trade.”5 Third, the HK analysis limited itself to a cross
section.In a panel, however, intensive margin and extensive margin effects of EIAs may have
differential “timings.” For instance,Arkolakis,Eaton and Kortum (2012) recently introduced
staggered “Calvo pricing” into their Ricardian modelof trade and showed that the intensive
margin likely reacts sooner to trade liberalizations than does the extensive margin.Moreover,
since the two margins have different “trade elasticities,” the quantitative path of the welfare
3Each “good” was a 6-digit SITC category.They also explored the effects ofcountry size and per capita
GDP on the quality of goods exported, as well as the two margins.
4Because firm-leveldata is not available for a large number ofcountry-pairs for a large number ofyears,
we are constrained to investigating EIAs impacts on products defined at the 4-digit SITC category level,as
in Hillberry and McDaniel(2002),Kehoe and Ruhl(2009),and Foster,Poeschl,and Stehrer (2011) discussed
below.
5For instance,welfare estimates could be sensitive to the presence or absence of intermediates or multiple
sectors.See also Melitz and Redding (2013) and Feenstra and Weinstein (2013).
3

gains is time sensitive.
Our paper extends the literature by offering three potential empirical contributions.First,
we extend the Baier and Bergstrand (2007) paneleconometric methodology for the (partial)
effects of EIAs on aggregate trade flows using a gravity equation to examine in a setting with
a large number of country pairs the effects of virtually all EIAs on the extensive and intensiv
goods margins,using the HK trade-margin-decomposition methodology.In the context of an
econometric analysis,we are the first to find economically and statistically significant EIA
effects on both the intensive and extensive (goods) margins in the context of a large numbe
of country pairs, EIAs, and years.
Second, we examine the effects of various types of EIAs – one-way preferential trade agre
ments (OWPTAs),two-way preferentialtrade agreements (TWPTAs),free trade agreements
(FTAs), and a variable for customs unions, common markets and economic unions (CUCME-
CUs) – on trade flows,extensive margins,and intensive margins.6 While two recent studies
have adapted the Baier-Bergstrand methodology for estimating the effect of differing “types
of EIAs on bilateralaggregate trade flows,no econometric study has examined the effect of
various types ofEIAs on the (goods) extensive and intensive margins oftrade using a large
number of country pairs and EIAs.7 Neither Helpman, Melitz, and Rubinstein (2008) nor Eg-
ger, Larch, Staub, and Winkelmann (2011) distinguished among various types of EIAs in thei
analyses of country intensive and extensive margins.We find not only that deeper EIAs have
larger trade effects than FTAs,and the latter have larger effects than (partial) two-way and
one-way PTAs,but we distinguish between these various trade effects at the extensive and
intensive margins using a panel of (disaggregate) bilateral trade flows from 1962-2000 cove
98 percent of world exports.
Third, Bernard, Jensen, Redding and Schott (2009) is likely the only empirical study to dat
to explore the “timing” ofextensive and intensive margin responses to shocks.Using cross-
sectionalvariation to examine long-run aspects,Bernard,Jensen,Redding and Schott (2009)
find that variation in trade flows across country pairs is explained largely by the extensive
margin, using firm-level data (the “firm” margin); this result is consistent with HK using their
6The HK methodology is based on Feenstra (1994).Due to few observations on common markets and
economic unions,we combine these two types of“deeper” EIAs with customs unions to form the variable
CUCMECU, representing “deep” EIAs.
7The two studies that extended the Baier-Bergstrand framework to differing types of EIAs are Magee (2008)
and Roy (2010); both found that customs unions had larger aggregate trade flow effects than FTAs.However,
neither study examined extensive versus intensive margin issues.
4

“goods” margin.But using time-series variation, Bernard, Jensen, Redding and Schott (2009)
find that a larger proportion oftrade variation can be explained by the intensive margin at
short (five-year) time intervals.They show that,following the Asian financialcrisis of 1997,
virtually all of the variation in trade flows within 2-3 years could be explained by the intensiv
margin. This finding is consistent with two recent theoreticalstudies arguing that the low
trade-cost elasticity found in macroeconomic analyses of business cycles should be associat
with the intensive margin oftrade compared with the relatively higher trade-cost elasticity
found in international trade, which reflects the intensive and extensive margin effects.8 In this
paper,we allow for differential“timing” ofEIA effects using paneldata. We find the first
comprehensive empirical evidence that the shorter-term effects of EIAs on trade flows are m
at the (goods) intensive margin and longer-term effects are more at the extensive margin (th
latter entailing either fixed export costs or staggered “Calvo pricing” by consumers), consist
with intuition and results in Bernard, Jensen, Redding and Schott (2009).Moreover, our results
shed empirical light on theoretical conjectures for the relative quantitative effects on intensi
and extensive margins of variable trade cost changes in a Melitz-type model.Finally, we show
our results are robust to potentialcountry-selection,firm-heterogeneity,and reverse causality
biases.
The remainder of this paper is as follows.Section 2 discusses our methodology, based on the
HK linear trade-margins-decomposition method and the Baier and Bergstrand (2007) approa
for estimating partial effects of EIAs on trade flows in gravity frameworks.Section 3 discusses
data and measurement issues.Section 4 provides the main empirical results and findings from
three sensitivity analyses.Section 5 concludes.
2 Methodology
Only three empirical studies have explored the effects of trade liberalizations – and, in partic
ular, EIAs – on the intensive and extensive goods margins of trade using the HK methodology
The earliest study using the HK decomposition to explore this issue is Hillberry and McDanie
(2002),focusing solely on the North American Free Trade Agreement (NAFTA).Although
8Ruhl (2008) explains the delayed effect of the extensive-margin effects to fixed export costs on the supply
side,while Arkolakis,Eaton, and Kortum (2011) explain the delayed effect of the extensive-margin effects to
“Calvo pricing” by consumers on the demand side.
5

they do not attempt to establish causaleffects from NAFTA to trade increases,they provide
a decomposition of post-NAFTA trade among the three partners into goods intensive and ex-
tensive margins using 4-digit Standard InternationalTrade Classification (SITC) data.They
find evidence of both margins changing between 1993-2001.Kehoe and Ruhl (2009) examined
NAFTA, the earlier Canada-U.S.FTA trade liberalization,and some structuraltransforma-
tions using a modified version of the HK decomposition methodology and applied to a series
of cross sections.Similar to Hillberry and McDaniel (2002), they do not conduct an economet-
ric analysis trying to explain the effect of NAFTA (or the Canada-U.S.FTA) on trade flows
conditionalon other variables.They decompose actualgoods extensive and intensive margin
changes post-agreement also using 4-digit SITC data for goods categories from Feenstra et a
(2005).They find significant evidence ofboth extensive and intensive margin changes using
their modified HK decomposition methodology.Both studies’ evidence of goods intensive and
extensive margins of trade expanding following the signing of NAFTA suggests the need for a
comprehensive econometric analysis (conditional on other covariates) of the effects of EIAs
generalon the goods intensive and extensive margins oftrade,in the spirit ofHK’s original
analysis of the effect of country size and per capita GDP on the two goods’ margins.
The only study to our knowledge that like us uses a data set for a large number of country
pairs and years,a large number ofEIAs, and the HK methodology is Foster,Poeschl,and
Stehrer (2011).However, the partial effect they found of an EIA on the goods extensive margin
was an economically insignificant 10 percent, and they found virtually no effect of EIAs on th
intensive margin.The latter result is a puzzle because it is existing exporters and importers
that seek EIAs and the theoretical studies noted above suggest that the shorter-term effect o
EIAs should be on the intensive margin.Yet, there are severaldifferences between our study
and theirs.First, their estimation was based upon a traditional gravity-equation specification
ignoring recent theoreticaldevelopments that emphasize the importance ofrelative price or
“multilateralresistance” terms.Our paper is based upon state-of-the-art gravity-equation
specifications,such as discussed in Arkolakis,Costinot,and Rodriguez-Clare (2012).Second,
Foster,Poeschl,and Stehrer (2011) use a short three-year window on both sides ofthe EIA
formation, and consequently can only capture short-term EIA effects; this likely explains the
economically small partial effects but does not explain finding only an extensive margin effe
As shown in Baier and Bergstrand (2007), EIAs can take 10-15 years to have their full impac
6

aggregate bilateral trade flows.Moreover, by allowing longer lags, we can distinguish between
short-term vs.longer-term effects.Third, Foster,Poeschl,and Stehrer (2011) examine the
impact ofEIAs using a single dummy variable;we use multiple EIA variables to distinguish
the effects of one-way PTAs, two-way PTAs, FTAs, and deeper EIAs on aggregate trade flows,
extensive margins,and intensive margins.Even disregarding the potential endogeneity biases
introduced by their ignoring relative prices,their study did not distinguish between various
“types” of EIAs and did not distinguish between the “timing” of intensive and extensive marg
effects.Indeed, Foster, Poeschl, and Stehrer (2011) suggest in their concluding paragraph tha
examination of the shorter-run versus longer-run effects and accounting for the differing “de
and breadth” of EIAs would be useful extensions.
2.1 The Gravity Equation
As Arkolakis, Costinot, and Rodriguez-Clare (2012) and others have noted, there is a wide cla
of quantitative trade models (Armington, Krugman, Ricardian, Melitz) that yield iso-morphic
gravity equations.Following Arkolakis,Costinot,and Rodriguez-Clare (2012),using a Melitz
model one can generate a standard gravity equation:
X m
ijt = Nm
it Y m
jt
( (am
Lit )−γm
w−γm
it τ−γm
ijt f −[γm /(σ m −1)−1]
ijt
∑ K
k=1 N m
kt (am
Lkt )−γm w−γm
kt τ−γm
kjt f −[γm /(σ m −1)−1]
kjt
)
(1)
where Xm
ijt is the trade flow from i to j in year t in “good” m,N m
it is the number of firms in
i (exporting and non-exporting) that produce output in goodm, Y m
jt is the expenditure inj
on good m, am
Lit (defined as unit input requirements of labor) is the lower bound of the Pareto
distribution ofproductivities in m in i,γm is an index ofproductivity heterogeneity among
firms in good m, wit is the wage rate in i, τijt is variable trade costs of exporting i’s products into
j, f ijt is fixed export costs from i to j, and σm is the elasticity of substitution in consumption.9
Note that the relative price term in large parentheses is a standard representation of relative
prices in the gravity equation, but now also reflecting productivity heterogeneity (through am
Lit
and γm) and fixed exporting costs (fijt ), cf., Melitz (2003),Chaney (2008),Redding (2011),
and Arkolakis, Costinot, and Rodriguez-Clare (2012).
In the context of these models, variable trade costs, τijt , affect Xm
ijt via both the intensive and
9We refer here to equation (23) in Arkolokis,Costinot,and Rodriguez-Clare (2012),under the assumption
that fixed export costs are paid in the importing country (i.e., μ = 0).
7

extensive margins.As Chaney (2008) demonstrates in his Melitz-type model, γm = (σm − 1) +
[γm −(σm − 1)].σm − 1 represents the intensive margin elasticity of variable trade costs wherea
γm − (σm − 1) is the extensive margin elasticity ofvariable trade costs.For finite means in
the theory, γm/(σm − 1) must exceed 1.Empirically, Chaney (2008) notes empirical estimates
of γm/(σm − 1) range between 1.5 and 2.Hence,these models suggest that the intensive
margin variable-trade-cost elasticity should be larger than the extensive margin variable-tra
cost elasticity.For instance,if γ = 1.5(σ − 1),then the intensive margin elasticity is twice as
large as the extensive margin elasticity; if γ = 2(σ − 1), then the intensive margin elasticity
equal to the extensive margin elasticity.10
Interestingly, the theoretical result that the (variable-trade-cost) intensive margin elastici
should be at least as large as the extensive margin elasticity conflicts with the empirical res
for the EIA partial effects in Foster, Poeschl and Stehrer (2011) discussed earlier.We evaluate
empirically this implication later, a potential contribution of this paper.
2.2 Accounting for Endogenous EIAs: The Baier-Bergstrand Method-
ology
Baier and Bergstrand (2007), or BB, re-evaluated usage of the gravity equation econometric
for estimating partial effects of EIAs on pairs of countries’ trade flows.11 The first of two main
contributions was that self-selection of country-pairs into EIAs (cf., Baier and Bergstrand, 200
likely created a significant endogeneity bias in previous gravity-equation estimates of the (p
tial) effects of EIAs on trade flows.This is precisely the concern raised in Arkolakis, Costinot,
and Rodriguez-Clare (2012,section V) for gravity-equation estimates of trade elasticities;the
observed variable trade cost measure may be correlated with unobservable trade costs hidd
10However, fijt works entirely through the extensive margin, so that there is no clear theoretical hypothesis
for the relative sizes of intensive and extensive margin effects of a given EIA formation.Yet, in light of estimates
of γ/(σ − 1) between 1.5 and 2,this additionalfixed-trade-cost elasticity can range feasibly between only 0.5
and 1. Such an effect is dwarfed by the intensive margin elasticity (σ − 1) with likely values between 4 and 9
(if σ ranges between 5-10),under certain assumptions.The key assumption pertains to the effect of an EIA
on fixed costs versus variable costs.Suppose ln τijt = ρ0 − ρ1 ln DISTij − ρ2EIA ijt + μijt , as is conventional
(μijt denoting a random error term).Suppose also ln fijt = δ0 − δ1 ln DIST ij − δ2EIA ijt + ϕijt (ϕijt denoting
a random error term).If δ2 > ρ2 by a large amount, the effect of fijt changes on the extensive margin may be
sufficient to cause the extensive margin trade elasticity to exceed the intensive margin trade elasticity.To date,
to our knowledge, there are no firm estimates of δ2 because there is no data on ln fijt .
11Partial (or direct) effects ignore general-equilibrium (or indirect) effects.While techniques exist for esti-
mating the indirect effects, such estimation is beyond this paper’s scope, cf., Anderson (2011), Bergstrand and
Egger (2011), and Egger, Larch, Staub, and Winkelmann (2011) on these issues.
8

in the gravity equation’s error term.The second main contribution ofBB was that – given
the slow-moving nature of EIAs’ determinations – gravity equation estimation could use pane
techniques and data to avoid endogeneity bias and also capture lagged influences, incorpora
ing either bilateralfixed effects (in a log-levels specification) or first-differencing to account
for time-invariant bilateral unobservable RHS variables, as well as incorporating exporter-tim
and importer-time effects to capture time-varying unobservable “multilateral price/resistanc
terms of the exporter and importer.BB showed that EIAs on average increased two members’
bilateral trade by approximately 100 percent after 10-15 years.Such a panel approach allows
estimates of the “timing” of EIAs’effects on trade flows between short run and long run,as
well as offers an alternative approach to instrumental variables using cross-sectional data (a
potentially avoids possible shortcomings of the latter approach).12
Given the problems associated with accounting for endogeneity of EIAs using instrumenta
variables and cross-section data,BB argued that a better approach to eliminate endogeneity
bias ofEIAs is to use paneltechniques.In the context ofthe theory and endogenous self-
selection of country pairs into EIAs, BB argued that one method to obtain consistent estimat
of the partial effect of EIAs is by fixed effects estimation of:
ln Xijt = β 0 + β1(EIA ijt ) + ηij + δit + ψjt + ϵijt (2)
where ηij is a country-pair fixed effect to capture alltime-invariant unobservable bilateral
factors influencing nominaltrade flows and δit and ψjt are exporter-time and importer-time
fixed effects,respectively,to capture time-varying exporter and importer GDPs as wellas all
other time-varying country-specific unobservables in i and j influencing trade,including the
exporter’s and importers’“multilateralprice/resistance” terms (cf.,Anderson and Wincoop,
2003).We refer to this as the fixed-effects (FE) specification.It is important to note that, in
most gravity-equation applications using a comprehensive set of RHS variables, the vast bul
“bilateral” trade-cost variables are time-invariant,such as bilateral distance,common border,
12As argued in BB,the problem with using cross-section data and consequently having to employ IV tech-
niques to account for EIA selection bias is the inability practically of satisfying the “exclusion restriction” with
confidence.Most variables that influence trade flows also explain selection into EIAs, and it is difficult to find
a variable that explains EIAs that does not also explain trade flows.Egger,Larch, Staub and Winkelmann
(2011) used the IV approach to account for the endogeneity of EIAs in their single cross-section, also allowing
for selection into zeros trade (i.e., a bivariate probit model).They found using an approach similar to Helpman,
Melitz, and Rubinstein (2008),except also allowing for endogenous EIAs,that EIAs predominantly affected
trade at the (country) intensive margin.
9

common language, etc.BB showed that the partial effect of the typical EIA on nominal trade
flows was about 0.76,implying that the typicalEIA increased bilateraltrade by about 114
percent after 10-15 years.
BB also employed an alternative specification using first-differencing:
∆ 5 ln Xijt = β 0 + β1(∆ 5EIA ijt ) + δ5,it + ψ5,jt + υ5,ijt (3)
where ∆5 refers to first-differencing over 5 years.We refer to this as the first-difference (FD)
specification.Note that the bilateralcountry-pair fixed effects are eliminated;however,the
exporter-time (δ5,it) and importer-time (ψ5,jt ) fixed effects are retained to capture changes in
the time-varying exporter and importer GDPs and multilateralprice terms over the same 5-
year period.The latter effects were ignored in Foster,Poeschl,and Stehrer (2011),creating
potential omitted variables bias.
First-differencing the panel data yields some potential advantages over fixed effects.13 First,
it is quite plausible that the unobserved factors influencing the likelihood of an EIA (say, trad
below its “natural” level) are likely slow moving and hence serially correlated.If the ϵijt are
highly serially correlated,the inefficiency of FE is exacerbated as T gets large.This suggests
that differencing the data willincrease estimation efficiency for our large-T panel.Second,
aggregate trade flow data and realGDP data are likely “close to” unit-root processes.Using
FE is equivalent to differencing data around the mean (in our sample,year 1980);this may
create a problem since T is large in our panel.As Wooldridge (2000, p.447) notes, if the data
follow unit-root processes and T is large, the “spurious regression problem” can arise in a pa
using FE. FD yields data that deviates from the previous period of our panel, and thus is clos
to a unit-root process.In the following,we focus on a first-difference approach;however,the
FE estimates are provided in the online appendix to this paper.14
In this paper,we introduce one further innovation relative to BB.15 While changes over
13As Wooldridge (2010, Ch.10) notes, when the number of time periods (T ) exceeds two, the FE estimator is
more efficient under the assumption of serially uncorrelated error terms ϵijt . The FD estimator is more efficient
(when T > 2) under the assumption that the error term ϵijt follows a random walk (i.e.,that the error term
υ5,ijt = ϵijt − ϵij,t−5 is white noise).When the number of time periods is exactly two (T =2), estimation with
FE and FD produce identical estimates and inferences; then, FD is easier to estimate.When T > 2, the choice
depends upon the assumption one wants to make about the distribution of the error term ϵijt .
14It turns out that, for FTAs and deeper EIAs, the results using FE and FD are quite similar.As a practical
matter, the choice is more important for TWPTAs and OWPTAs; we discuss this later.
15We thank a referee for motivating this innovation.
10

time in exporter-specific and importer-specific unobservable variables are captured with δ5 and
ψ5, changes over time in pair-specific unobservables,such as falling variable and fixed export
costs unrelated to EIAs, are not accounted for.Wooldridge (2000) suggests a “random growth”
(also called,random trend) first-difference model,henceforth,RGFD model. Unobservable
pair-specific changes over time can be partially accounted for by including pair-specific ij fix
effects in equation (3), suggesting specification:
∆ 5 ln Xijt = β 0 + β1(∆ 5EIA ijt ) + δ5,it + ψ5,jt + ηij + υ5,ijt (4)
Consequently,if unobservable declines in bilateralvariable and fixed trade costs (say,due to
technologicalimprovements) evolve smoothly over time,the ηij ’s in equation (4) willaccount
for these influences.
One of the other potential contribution’s of BB’s panel methodology was to show that the
full impact of EIAs on trade flows took 10-15 years.One reason is that most EIAs are “phased-
in” over 5-10 years.The second reason is the lagged effect ofthe trade-cost changes (such
as terms-of-trade changes) on trade flows.As in BB, using a panelallows for differentiating
the shorter-term effects (5 years) from the longer-term effects (5-10 years).In the context of
the recent developments in the trade literature emphasizing intensive versus extensive mar
effects, our panel approach allows for differential timing of these effects.In reality, one would
expect that the intensive margin would be affected by a trade-cost change sooner than the
extensive margin,because intensive margin changes in volumes do not require any startup
costs.Such costs – criticalto the extensive margin – may delay the entry ofnew firms into
exporting, and thus we should expect the intensive margin to be influenced in the shorter te
and the extensive margin in the longer term,as the results in Bernard,Jensen,Redding,and
Schott (2009) show.Our panel data approach allows for evaluating this hypothesis.16
It will be usefulnow to rationalize the use of 5-year differencing of data as in BB,rather
16These differential timing effects were ignored in Foster,Poeschl,and Stehrer (2011).As discussed earlier,
two recent theoretical papers suggest a reason for the low trade-cost elasticity of trade flows in macroeconomi
analyses using time-series data and the relatively higher trade-cost elasticities of trade in cross-sectional trade
analyses.Ruhl (2008) explains this puzzle by noting that the macroeconomic time-series approach is estimat-
ing the intensive margin effect oftrade,whereas the trade literature’s cross-sectionalapproach is capturing
the intensive and extensive margin effects,due to export fixed costs for new producers delaying trade effects
and entry.In a complementary approach,Arkolakis,Eaton, and Kortum (2011) present a demand-oriented
staggered-adjustment “Calvo-pricing” approach to explain the lower time-series elasticity in terms of solely an
intensive margin effect,and the higher long-run cross-section trade-cost elasticity capturing the longer-term
extensive margin elasticity also.
11

than,say,annualdifferencing.Cheng and Wall(2005) and Wooldridge (2000) both argue in
favor of using data differenced over a longer period than annually.Cheng and Wall(2005,p.
8) note that “Fixed-effects estimations are sometimes criticized when applied to data pooled
over consecutive years on the grounds that dependent and independent variables cannot fu
adjust in a single year’s time.” Wooldridge (2000, p.423) confirms the reduction in standard
errors ofcoefficient estimates using changes over longer periods oftime than using “year-to-
year” changes.Based upon these considerations, we chose as in BB 5-year differences; similar
considerations led to the use of 4-year differences in Anderson and Yotov (2011).Nevertheless,
in a sensitivity analysis later, we will confirm our findings using annual data.
BB did not estimate differentialeffects ofvarious types ofEIAs (in terms of depth of
integration) on trade flows.Magee (2008) and Roy (2010) using the methodology of BB found
that trade flows were impacted by larger amounts for customs unions relative to FTAs.However,
no empirical study has examined the differential impact of FTAs relative to deeper EIAs on th
extensive versus intensive margins – much less the differential timing of such effects; these
goals of this paper.17 The next section discusses how we decompose data into the two margins.
2.3 The Hummels-Klenow Margin-Decomposition Methodology
Hummels and Klenow (2005),or HK, was the first paper to highlight a tractable method for
decomposing transparently the extensive and intensive goods margins of trade for a large s
countries’ bilateral trade flows using publicly available disaggregate trade data.18
Let Xijt denote the value of country i’s exports to country j in year t.Following HK, the
17It is usefulto note here a parallelliterature examining the effect of GATT and/or WTO membership on
trade flows.For brevity, we note that there now appears little convincing evidence of substantive GATT/WTO
effects on trade, once one accounts for EIA dummies, multilateral resistance, and unobserved country-pair fixed
effects (as we do here).This is the conclusion of Eicher and Henn (2011) (though they found a non-trivial WTO
“terms-of-trade” effect) and of Felbermayr and Kohler (2010) who examined possible extensive margin effects;
Eicher and Henn (2011) ignored extensive versus intensive margin effects.We also note an issue raised in Martin
and Ng (2004), which is the role of multilateral tariff reductions under the GATT/WTO. Most-Favored-Nation
(MFN) tariff cuts could also be affecting results.However, such MFN tariff cuts by country would be accounted
for by the exporter-time and importer-time fixed effects.
18Studies have also used country-specific data on individualplants (or firms) to study the extensive and
intensive firm margins oftrade liberalization,but such studies have necessarily been confined to particular
countries because such data is widely known to be much more costly to access and such data sets have not
been concorded for international comparisons, as noted in Helpman, Melitz, and Rubinstein (2008).See Eaton,
Kortum, and Kramarz (2008) for a study of French firms, Trefler (2004) for a study of Canada and the United
States,and Pavcnik (2002) for a study of Indian firms.Another relevant theoreticaland empiricalpiece with
similar overtones is Arkolakis, Demidov, Klenow, and Rodriguez-Clare (2008).
12