trade eﬀects of the europe agreements

This article was downloaded by: [INASP - Pakistan (PERI)]On: 27 March 2014, At: 05:41Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH,UK

The Journal of InternationalTrade & Economic Development:An International andComparative ReviewPublication details, including instructions for authorsand subscription information:http://www.tandfonline.com/loi/rjte20

Trade effects of the Europeagreements: A theory-basedgravity approachJulia Spies a & Helena Marques ba Institute for Applied Economic Research (IAW) ,72074, Tübingen, Germanyb Manchester Business School, The University ofManchester , Booth Street West, Manchester, M15 6PB,UKPublished online: 07 Apr 2009.

To cite this article: Julia Spies & Helena Marques (2009) Trade effects of the Europeagreements: A theory-based gravity approach, The Journal of International Trade &Economic Development: An International and Comparative Review, 18:1, 11-35, DOI:10.1080/09638190902757368

To link to this article: http://dx.doi.org/10.1080/09638190902757368

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all theinformation (the “Content”) contained in the publications on our platform.However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, orsuitability for any purpose of the Content. Any opinions and views expressedin this publication are the opinions and views of the authors, and are not theviews of or endorsed by Taylor & Francis. The accuracy of the Content shouldnot be relied upon and should be independently verified with primary sources

http://www.tandfonline.com/loi/rjte20

http://www.tandfonline.com/action/showCitFormats?doi=10.1080/09638190902757368

http://dx.doi.org/10.1080/09638190902757368

of information. Taylor and Francis shall not be liable for any losses, actions,claims, proceedings, demands, costs, expenses, damages, and other liabilitieswhatsoever or howsoever caused arising directly or indirectly in connectionwith, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes.Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expresslyforbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

Trade effects of the Europe agreements: A theory-based gravity

approach

Julia Spiesa* and Helena Marquesb

aInstitute for Applied Economic Research (IAW), 72074, Tubingen, Germany;bManchester Business School, The University of Manchester, Booth Street West,

Manchester, M15 6PB, UK

(Received 24 July 2007; final version received 19 December 2008)

In this paper, we develop a new version of a theory-based gravityequation to properly account for the relative price indices initiallyproposed by Anderson and van Wincoop (2003). The partially time-varying character of our multilateral resistance variables overcomes thebias present in earlier studies that solely rely on country or country pairfixed effects. Applying the augmented gravity equation to the process ofEuropean Union (EU) integration during the 1990s, we find robustevidence that the Free Trade Agreements (FTAs) with the Central andEastern European Countries (CEECs) have substantially increasedintra-group trade, in the case of the Czech and Slovak Republic andSlovenia at the expense of the Rest of the World (ROW). Sincedecreasing multilateral trade resistance negatively influences a country’sbilateral imports but may be positively correlated with a bilateral FTA,earlier East–West studies, which ignored the relative price term’s time-varying character, tend to be downward biased. Indeed, our resultsindicate that once we correct for the omitted variable bias, the FTAswith the CEECs created 7 to 20% more new trade compared with thescenario where only time-invariant country pair effects were included.

Keywords: free trade agreements; gravity equation; Central and EasternEurope; panel data

1. Introduction

Since 1989, Europe has been the stage of an ongoing process of regionalintegration involving 15 European Union (EU) member states and tenCentral and Eastern European Countries (CEECs).1 The EU admission ofeight CEECs on 1 May 2004 represented a temporary peak in theintegration process, but it was not the end of it. Bulgaria and Romaniaalso joined the EU in January 2007 after almost 15 years of preferential

*Corresponding author. Email: [email protected]

The Journal of International Trade & Economic DevelopmentVol. 18, No. 1, March 2009, 11–35

ISSN 0963-8199 print/ISSN 1469-9559 online

� 2009 Taylor & Francis

DOI: 10.1080/09638190902757368

http://www.informaworld.com

Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

trade relations guided by Free Trade Agreements (FTAs) that becameknown to the public as the Europe Agreements (EAs). Since the EAs had togo through a long process of ratification by each individual member state,the European Community (EC) gave provisions on trade and trade-relatedmeasures effective by means of Interim Agreements (IAs) at an earlierstage.2 Subsequent to the gradual reduction of trade barriers, one couldobserve a relative increase in the EU’s total imports from the CEECs ascompared to its imports from the Rest of the World (ROW).3 This relativeboost raised questions about the extent to which the geographicalrestructuring of trade flows has taken place and how much of it canactually be attributed to the FTAs signed and implemented in the course ofthe 1990s.

After the fall of the Iron Curtain, economists started to examine the‘natural’ trade patterns of the CEECs (see for example Bussiere, Fidrmuc andSchnatz 2005 for an overview, and Egger, Pfaffermayr and Schmidt 2007 fora recent contribution to this stream of literature) without explicitlyquantifying the impact of FTAs. Early cross-sectional studies thatspecifically point to the geographical restructuring of trade flows arisingfrom the implementation of the EAs report insignificant or even negativecoefficients for East-West integration (Laaser and Schrader 2002; Paas 2003).Studies relying on panel data, however, find significant positive estimates inthe range of 11 to 130% (Martın and Turrion 2001; Adam, Kosma andMcHugh 2003; De Benedictis, De Santis and Vicarelli 2005; Herderscheeand Qiao 2007). With the exception of the study by De Benedictis, De Santisand Vicarelli (2005), a major shortcoming of all of these studies is that theysuffer from an omitted variable bias. Anderson and van Wincoop (2003)stress that bilateral trade does not only depend on bilateral trade costs butalso on a trading pair’s resistance to trade with the ROW.

In this paper, we will employ a new version of a theory-based gravityequation to correct for biases stemming from the omission of the Andersonand van Wincoop (2003) relative price terms. We find that earlier East–Weststudies that solely rely on country pair fixed effects underestimate the trade-promoting effect of the EU’s FTAs. Describing multilateral trade resistancethrough all factors that also influence the bilateral resistance to trade,our results indicate that the FTAs with the CEECs have boosted EUimports from the CEECs by 72% (up to 80% for the new members Bulgariaand Romania), whilst not decreasing imports from the ROW by morethan 14%.

In Section 2, we briefly lay out the concepts of trade creation and tradediversion and relate them to the empirical literature on gravity modelling.Section 3 develops the theoretical model, which builds the basis for theestimated equation. Section 4 discusses our approach to the measurement ofmultilateral trade resistance. Section 5 deals with econometric and dataissues. We present the estimation results in Section 6. Section 7 concludes.

12 J. Spies and H. Marques

Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

2. Trade creation and trade diversion in East–West integration

Theoretical insights on allocation effects of FTAs were first given by Viner(1950) and Bye (1950). These authors argue that a fractional reduction oftrade barriers leads only to a shift, but not to an elimination of thediscrimination of different sources of supply. Viner named the resultingeffects trade creation and trade diversion. Trade creation is then associatedwith the portion of the new trade between member countries that is whollynew, resulting in an improvement in the international resource allocation. Itoccurs when, subsequent to the formation of a customs union, domesticproduction at high costs is replaced by lower-cost sources from the newpartner country. Trade diversion refers to the part of the new trade betweenmember countries that is only a substitute for trade with third countries. Itdescribes a situation in which the preferential trade liberalisationcauses higher-cost production from the new partner country to replaceimports from low-cost sources in the ROW. In this case, the resourceallocation is worsened. The concepts of trade diversion and trade creationin their original version refer only to producers and consumers inside theFTA area. Trade diversion can, however, seriously harm excludedcountries.

A simple calculation allows for a first insight into the relative change inthe aggregate imports of EU15 countries from the CEECs and from theROW during the EU integration process of the candidate countries. Torender the sizes of the two geographical regions comparable, the yearlyimport values have been normalised with respect to the base year (1991).Taking the quotient allows to assess relative changes. To be precise, therelative imports from the CEECs (MCEECs) with respect to the ROW(MROW) in each sample year4 measured with 1991 as the base year has beencalculated as follows:

MCEECst=MCEECs91

MROWt=MROW91ð1Þ

Looking at Figure 1 it can readily be seen that the growth of EU15imports from the CEECs has been over three times higher than that fromthe ROW. Moreover, the relative boost seems to have taken placesteadily and continuously since the fall of the iron curtain until the eve ofthe CEECs’ EU membership. These stylised facts match our a prioriexpectations well and also fit into the findings of the empirical literatureon this subject.

While early studies on East–West integration detect highly unex-hausted trade potentials of the CEECs’ trade with the EU (see e.g.Hamilton and Winters 1992; Baldwin 1994), more recent studies statethat the CEECs have meanwhile returned to their ‘natural’ trade patterns

The Journal of International Trade & Economic Development 13

Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

(see e.g. Egger 2002; Fidrmuc and Fidrmuc 2003 and Bussiere, Fidrmucand Schnatz 2005). The political distortions in place at the start of theliberalisation process make it difficult to analyse the geographicalrestructuring of trade flows after the fall of the Iron Curtain in theframework of the Vinerian terms of trade creation and trade diversion.Nevertheless, most studies formally assessing the impact of any kind ofintegration arrangement implement different sets of dummy variables intoa gravity equation to measure the effects of preferential liberalisation onintra- and extra-group trade (see for example Bayoumi and Eichengreen1995; Frankel and Wei 1998; Soloaga and Winters 2001 or for a morerecent study Carrere 2006).5

Martın and Turrion (2001) provide the first study that directlycalculates the impact of integration efforts on East–West trade byestimating a gravity equation via a two-step fixed effects panel estimator.Although the authors conclude that the EU has strengthened its traderelations with the CEECs more than with other OECD countries, theresult does not seem to hold for these regions’ exports to the EU(coefficients of 0.83 and 1.99, respectively). With a similar estimator but aslightly different model specification and time coverage, Adam, Kosmaand McHugh (2003) calculate a trade effect of the EAs of 32%. In themost recent paper on this issue, Herderschee and Qiao (2007) assessvarious preferential trading agreements of the EU and find a significantpositive effect of the EAs of 26% and 55% in their static and dynamicspecification, respectively. In the field of the East–West integrationliterature, only De Benedictis, De Santis and Vicarelli (2005) account formultilateral trade resistance by including a country pair specific timetrend. The authors calculate an impact of the EU-CEEC FTAs of 11%,although this estimate excludes Bulgaria and Romania.

Figure 1. Relative changes in EU imports.Source: Own calculations, data from OECD ITCS.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

3. Theoretical foundation of the gravity equation

Even though the gravity equation’s initial success stemmed from its goodempirical properties, it possesses nowadays ‘more theoretical foundationsthan any other trade model’ (Baldwin 2006). The repeated ignorance ofwhich has, however, produced a number of commonly-accepted mistakes ingravity model estimation, so that we attach importance to laying out brieflythe derivation of the equation we are going to test.

Assuming identical, homothetic Constant Elasticity of Substitution(CES) preferences and ‘iceberg’ type transport costs, country i’s aggregatetotal value of imports from country j can be expressed as

Mij ¼ NjYipijPi

� �1�sð2Þ

with Nj representing the variety of products sold by country j and Yi beingcountry i’s nominal expenditure. Pij/Pi is the relative price determining theshare of country i’s expenditure spent on country j’s goods with Pi beingcountry i’s price index for all import-competing goods and pij standing forthe ‘landed’ price. s is the above-unity elasticity of substitution betweengoods originating from country i and country j.6 Since prices on individualgoods are hardly available, we define the landed price

pij ¼ tijPjeij ð3Þ

as a function of bilateral trade costs tij, country j’s producer price index Pj

and the nominal exchange rate eij (in price quotation).7 Substitutingequation (3) into equation (2) yields

Mij ¼ NjYiðtijreijÞ1�s with reij ¼eijPj

Pið4Þ

as the real exchange rate. Equation (4) already looks close to commonlyestimated gravity equations. However, as stated by Anderson and vanWincoop (2003), bilateral trade does not solely depend on bilateral tradecosts, but also on the average resistance to trade with the ROW. Only byconsidering these multilateral terms can it be explained why a certain regionis pushed towards trade with a given partner when barriers towards all tradepartners increase. Employing general equilibrium conditions has theconvenient side-effect of eliminating the number of varieties Nj, for whichdata are not on-hand.8 Producer prices in country j must then adjust, suchthat the market clearing condition is satisfied:

Yj ¼XIi¼1

Mij ð5Þ


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

Substituting the import demand equation (4) into the market clearingcondition (5), we can solve for Nj as follows:

Nj ¼YjPI

i¼1 reijtij� �1�s

Yi

ð6Þ

Plugging equation (6) into equation (4), we obtain our testable gravityequation

Mij ¼YiYj tijreij

� �1�sPI

i¼1 Yi tijreij� �1�s ð7Þ

If we further define world income as YW ¼PI

i¼1 Yi and the share of countryi’s income in world income as si ¼ Yi

YW, equation (7) can be rewritten as

Mij ¼YiYj

YW

tijreij� �1�s

PIi¼1 si tijreij

� �1�s ð8Þ

where country i’s total imports from country j are not only dependent on therelative incomes of the two countries and on their bilateral exchange rateand trade costs, but they also depend on the importers’ share of worldincome and on their average trade costs and exchange rate with respect to allexporters.

4. Measuring multilateral trade resistance in gravity equations

The existence of unobservable factors that simultaneously influence importsand the explanatory variables on the RHS introduces an unobservedheterogeneity bias (Cheng and Wall 2004). Baier and Bergstrand (2007a)specify that the FTA coefficient tends to be underestimated in cross-sectionstudies if there are, for example, welfare-reducing domestic policy regulationsthat induce countries to select into FTAs. Anderson and van Wincoop (2003)drew attention to the biases resulting from ignoring multilateral traderesistance and modelled it using price terms. The empirical measurement ofprice effects is tricky and has so far been addressed in four ways. First, Baierand Bergstrand (2001) approximate the relative price terms through GDPdeflators. Since published price indexes do not reflect most factors thatinfluence a country’s multilateral trade resistance, the coefficient estimate isclose to zero and thus economically not important in explaining the growth inbilateral trade. Second, Anderson and van Wincoop (2003) themselvesproposed to measure the terms directly with a nonlinear least-squaresestimator. Since this method proved to be computationally costly, Feenstra(2002) proposed a third approach to capture price effects via importer and


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

exporter dummies that has becomemost popular. Introducing an individual orbilateral fixed effect allows to control for time-invariant variables thatsimultaneously affect trade flows and RHS variables. Baldwin (2006) formallydemonstrates that only including country (pair) fixed effects in a panel setting,is, however, an insufficient solution to the omitted variable bias, since the time-varying part of the multilateral trade resistance is still ignored. This time-varying residual may well be correlated with other time-varying trade costmeasures causing a bias that becomes more important with the length of thesample. In which direction does the bias go? Ex-ante predictions are difficult: ifthe omitted term is positively correlated with the probability to select into anFTA (e.g. a high multilateral trade resistance increases a country’s incentive toform an FTA) and positively correlated with the dependent variable (highmultilateral resistance pushes a country towards trading with a specificpartner), one would expect an upwardly biased FTA coefficient in conven-tional estimations. Since every implemented FTA alters not only the bilateralbut also the multilateral trade resistances (by lowering the trade costs and therelative price terms) and lessens thereby the positive effect on the bilateral tradevolume, it is also possible that omitting these effects downward biases theestimated trade impact over time. Recently, various authors suggested a fourthapproach, which accounts for the overall trade resistance through time-varying exporter and importer dummies (e.g. Baltagi, Egger and Pfaffermayr2003; and Broto, Ruiz and Vilarrubia 2006). Whilst this method allowscapturing all the unobserved characteristics in a very intuitive way, the amountof dummies needed makes the estimation for the full sample computationallyunfeasible. Furthermore, the time–country interactions absorb the GDP effect,such that one cannot estimate the trade impact of these traditional gravityvariables. A similar difficulty occurs with Bun and Klaassen’s (2007)suggestion to include country pair specific time trends: the dummies absorbexplanatory power of other bilateral time-varying variables.

In fact, equation (8) introduces a novel way of modelling multilateral traderesistance in gravity equations that presents several advantages with respect tothe existing literature. Hence, we propose a fifth approach where, in contrast tothe work by Anderson and van Wincoop (2003), multilateral resistance is notdescribed through relative price terms, but through all variables that alsoinfluence the bilateral resistance to trade. Their partially time-varyingcharacter overcomes the bias present in earlier estimations that solely relyon country (pair) fixed effects to proxy for the multilateral resistance terms. Atthe same time we are able to apply standard panel data estimation techniqueson the full sample. Note, however, that to the extent that there are unobservedor unobservable sources of time-varying multilateral trade resistance, theapplied procedure does not fully correct the omitted variable bias. The loss is,however, believed to be minor since the most common trade cost variables areexplicitly captured. The multilateral terms in the denominator of equation (8)are defined as averages over all partner countries to account for the fact that


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

when estimating the equation there will be a number of importers (i countries).Hence, the joint share of the importers’ income in world income is defined asPI

i¼1Yi

YW and the real exchange rate of each importer’s currency against theaverage of all exporters’ currencies takes the form 1

J

PJj¼1 reij 8i. Furthermore,

as proposed by Baier and Bergstrand (2007b), the trade cost variables aredefined as multilateral and world resistances,

MWRij ¼1

I

XIi¼1

tij þ1

J

XJj¼1

tij �1

IJ

XIi¼1

XJj¼1

tij 8i; j ð9Þ

The first two terms on the right-hand side (RHS) represent the multilateraltrade resistances of the respective trading partners. Holding bilateral tradecosts constant, a rise in these terms implies a lower ratio of bilateral tomultilateral trade costs and thus a boost of bilateral trade. The last term,however, resembles the world resistance to trade and, as such, lowers thetrade value between every pair of countries.9 The opposite interpretation ofthe multilateral and world resistance terms holds true, of course, for tradestimulating factors such as cultural proximity or trade arrangements.

In line with the basic idea behind gravity models, that the intensity withwhich a pair of countries trades is subject to pull and push factors, we adopta broad interpretation of the bilateral and multilateral trade resistance termsand assume the unobservable trade cost variable tij to be a log-linearfunction of a set of observable variables that influence trade costs,10

tij ¼ ðDijÞd1 ½ed2LLij�d3Bij�d4CLij�d5DEPij�d6FTAijþd7FTAi � ð10Þ

with Dij as the great-circle distance between the importing and the exportingcountry, LLij as a dummy variable equalling 1 if 1 and 2 of both countries in atrading pair are landlocked and 0 otherwise, and Bij as a dummy variablecontrolling for the length of the common border between countries i and j.Supposing that cultural proximity beats down the landed price throughtransaction cost savings, the dummy variable CLij equals 1 when the importerand the exporter have the same official language and 0 otherwise. Finally,DEPij is a dummy taking the value of 1 whenever country j is a non-independent entity being legally associated with an independent state and 0otherwise.11 To separate the ex-post effects of the FTAs with the individualCEECs, a set of stepwise dummy variables has to be included into thetheoretically derived gravity equation. FTAij equals 1 for the contractingparties for the years following the entry into force of the IAs and 2 for the yearsfollowing the entry into force of the EAs (intra-bloc bias) to capture the impactof the FTAs on intra-group trade. FTAi equals 1 for non-contracting partiesfor the years following the entry into force of the IAs and 2 for the yearsfollowing the entry into force of the EAs (extra-bloc openness) to capture theimpact of the FTAs on trade of group members with non-members.12


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

Following this specification, we will be able to examine whether theFTAs signed in the 1990s between the EU15 and the CEECs increased tradebetween the EU and the associated countries at the expense of lower tradewith third countries while overcoming the biases present in earlier East–West studies. Taking into account the modifications of the theoreticallyderived equation discussed above, the log-linearised13 reduced-form gravityequation is given by

lnMijt ¼ aþ b1 lnYit þ b2 lnYjt þ b3 ln reijt þ b4ðlnÞtijðtÞ

þ b5 lnXIi¼1

sit þ b6 ln1

J

XJj¼1

reijt þ b7ðlnÞMWRijðtÞ þ eijt ð11Þ14

where 1/YW is absorbed into the constant term a,15 common to all years andall country pairs, eijt is the i.i.d. error term and the expected coefficient signsare

b1 > 0;b2 > 0;b3 < 0;b4 ¼XTd¼1ð1� sÞd1 < 0; ð1� sÞd2 < 0; ð1� sÞ

d3 > 0; ð1� sÞd4 > 0; ð1� sÞd5 > 0;

ð1� sÞd6 > 0; ð1� sÞd7 < 0;

b5 > 0b6 > 0b7 ¼XTr¼1ð1� sÞr1 > 0; ð1� sÞr2 > 0; ð1� sÞ

r3 < 0; ð1� sÞr4 < 0; ð1� sÞr5 < 0; ð1� sÞ;r6 < 0

5. Econometric issues and data

In accordance to the findings of Egger (2002), panel data methodology isapplied. First, and in contrast to cross-section analysis, panels enable us tocapture relevant relationships between variables over time. Second, theyallow monitoring unobservable country pair individual effects. As outlinedabove, not controlling for country heterogeneity yields biased estimates.Since the random effects model only yields consistent estimates when theunobservable bilateral effects are not correlated with the error term,a Hausman test is conducted to test for that correlation. As the test clearlyrejects the null-hypothesis of no correlation, the country pair effects will betreated as fixed. The relevant fixed effects (FE) regression thus givesunbiased estimates of the time-varying variables (reported in columns 1 and2 of Table 1). Nevertheless, to provide comparability, we also present theestimated parameters of the random effects (RE) in columns 3 and 4 ofTable 1. The use of fixed effects still presents another problem as it does notallow for the estimation of time-invariant variables. To overcome this


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

problem, we use the fixed effects vector decomposition (FEVD) regressionmethod (reported in columns 5 and 6 of Table 1). The latter has beendeveloped by Plumper and Troeger (2007) and equals a stepwise fixed effectsestimation technique, rendering the estimation of the time-invariantvariables possible (econometric details are provided in Appendix A4). Anadditional advantage of the FEVD estimator as compared to the basic fixedeffects estimator comes from the estimation of hardly time-varyingvariables, such as, for example, the FTA dummies. Plumper and Troeger(2007) argue that although the within-groups transformation does noteliminate these variables, their estimation becomes inefficient in the fixedeffects approach. This does not only lead to higher standard errors, but mayalso cause unreliable coefficients.

Another estimator that has recently gained popularity in the context ofgravity modelling is the Hausman–Taylor (HT) estimator (Hausman andTaylor 1981). Since the estimator instruments possibly endogenousvariables, it accounts for the potential problem of reverse causality betweenFTAs and trade flows. A drawback, however, is the selection of validinstruments. Since the instruments simultaneously have to fulfil theconditions of correlation with the endogenous variables and independencefrom the unit effects, their quality is sometimes poor. This results inefficiency losses. With little success, we have looked for high qualityinstruments to perform an HT estimation. The most likely problem in ourcase of models is that the variables determining FTAs and trade flows are toa large extent the same (see also Baier and Bergstrand 2007a). Whenspecifically looking at the process of European integration of the CEECs,however, an economic explanation seems plausible as well: Whilst the EU15has been the biggest trading partner for the CEECs, their relevance for theEU15 was only minor at the start of the transformation process. Hence, hightrade intensity was possibly not the basis for decision-making when signingthe IAs.

A valuable extension to the static approach is a dynamic specification ofthe gravity equation. By including lagged imports as an additionalexplanatory variable, we account for the potential inertia of trade flowsrelated to sunk costs. The Arellano-Bond difference General Methods ofMoments (GMM) estimator is an instrumental variables estimator wherelagged levels of the dependent variable and first differences of the strictlyexogenous variables are used as instruments. One potential drawback of theestimator is that lagged levels are poor instruments in case the dependentvariable follows a random walk. Applying an augmented Dickey–Fuller unitroot test, we are, however, able to reject the null–hypothesis of non-stationarity of imports. Hence, it is appropriate to add the dynamic GMMestimator as a robustness check to our static analysis.

Heteroscedasticity and serial correlation of the error terms is correctedfor in all regressions.16 Finally, we controlled for a possible selection bias


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

by including three variables that approximate the Heckman correctionterm: HC1 is a variable containing the number of years of a tradingpair in the sample; HC2 and HC3 are dummies, taking the value of 1 ifthe trading pair is observed over the entire period 1991 to 2003 and if thetrading pair is present in the sample in t71, respectively (and 0otherwise).17

As for the data, we consider EU15 countries’ imports from a worldwidesample of 204 countries18 over the period 1991–2003, forming anunbalanced panel data set with roughly 32,245 observations. The datasources and definitions of all variables entering the tested gravity equationare listed in Table A3 in the appendix.

6. Results

This section starts with the results of estimating equation (11), laying aspecial focus on the effect of including multilateral resistance terms. Thebasic regression and the robustness checks provide evidence that the FTAelasticities move up, once the time-varying part of the relative price terms isappropriately accounted for. This finding is finally put into the context ofearlier studies in the field of East–West integration.

6.1. Regression results

The results of the regressions with and without the multilateral and worldresistance terms are presented in Table 1. For each regression methodexplained in Section 5, the first column shows the regression results omittingthe multilateral terms. Except for some FTA dummies, all parameterestimates of the FE model show the expected signs and are highlysignificant. As for the traditional gravity variables, the positive parameterestimates for GDP indicate that the import value increases with theimporter’s GDP, owing to a higher import demand, and with the exporter’sGDP, owing to a higher export supply. The coefficients are, however,somewhat away from the theoretically predicted unitary elasticity.19 Notethat the theoretically justified inclusion of the real exchange rate exhibitsempirical importance as well. A 10% depreciation20 of the importingcountry’s currency against its trading partner’s currency reduces the importvalue from the latter by 2.8%. Moving to the FEVD regression, we find thatour distance coefficient of –1.30 lies within the usual range.21 When at leastone of the countries in a trading pair is landlocked, bilateral importsdecrease by 74%.22 Being legally dependant on the importing country andsharing a common language significantly boost the propensity to trade. EUmembership does not have a significant influence on imports in this setting.Since there is little time variation in the EU dummy (only for Austria,Finland and Sweden, the variable jumps from 0 to 1 in 1995), most


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

Table 1. Estimation results.

FE RE FEVD

(1) (2) (3) (4) (5) (6)

w/o MT with MT w/o MT with MT w/o MT with MT

ln Yit 0.22**(0.10)

0.44***(0.13)

1.02***(0.03)

1.09***(0.03)

0.22***(0.01)

0.44***(0.01)

ln Yjt 0.67***(0.08)

0.71***(0.08)

1.14***(0.01)

1.15***(0.01)

0.67***(0.00)

0.71***(0.01)

ln reijt 70.28***(0.07)

70.38***(0.07)

70.08***(0.01)

70.07***(0.01)

70.28***(0.00)

70.38***(0.00)

ln Dij 70.85***(0.05)

71.81***(0.20)

71.30***(0.01)

72.30***(0.03)

Bij 0.00(0.00)

70.00**(0.00)

0.00***(0.00)

70.00***(0.00)

LLij 70.56***(0.07)

72.04(3.58)

71.33***(0.02)

70.70(0.49)

DEPij 0.75*(0.45)

1.69***(0.57)

0.98***(0.12)

1.22***(0.14)

CLij 1.39***(0.14)

0.77***(0.15)

1.07***(0.03)

0.94***(0.03)

EUijt 70.01(0.04)

70.00(0.06)

0.26***(0.07)

70.15**(0.06)

70.01(0.05)

70.00(0.03)

EUit 70.30***(0.08)

70.32***(0.08)

70.24***(0.06)

70.29***(0.06)

70.30***(0.05)

70.32***(0.05)

FTAirobut 0.47***(0.08)

0.59***(0.09)

0.27***(0.06)

0.13**(0.07)

0.47***(0.07)

0.59***(0.08)

FTAihupot 0.26**(0.10)

0.31***(0.11)

0.24***(0.08)

0.26***(0.09)

0.26***(0.08)

0.31***(0.09)

FTAiczslt 0.39***(0.10)

0.47***(0.11)

0.16**(0.08)

0.11(0.08)

0.39***(0.08)

0.47***(0.09)

FTAisvt 0.04(0.04)

0.10(0.06)

0.05(0.05)

70.16***(0.06)

0.04(0.03)

0.10*(0.05)

FTAibalticst 0.49***(0.08)

0.58***(0.09)

0.56***(0.07)

0.67***(0.07)

0.49***(0.06)

0.58***(0.08)

FTAit

(Ro, Bu)0.11*(0.06)

0.12*(0.06)

0.04(0.05)

70.06(0.05)

0.11(0.07)

0.12*(0.07)

FTAit

(Hu, Po)70.08(0.06)

70.10(0.06)

70.02(0.06)

0.00(0.06)

70.08(0.07)

70.10(0.07)

FTAit

(Cz, Sl)70.14**(0.06)

70.15**(0.06)

70.27***(0.06)

70.31***(0.06)

70.14*(0.08)

70.15*(0.07)

FTAit

(Sv)70.03(0.03)

70.08**(0.03)

70.07***(0.02)

70.22***(0.03)

70.03(0.03)

70.08**(0.03)

FTAit

(Baltics)0.14***(0.05)

0.13***(0.05)

0.12***(0.04)

0.26***(0.04)

0.14**(0.06)

0.13**(0.06)PI

i¼1 sit 0.02(0.23)

71.62***(0.12)

0.02(0.17)

ln 1J

PJj¼1 reit 0.99***

(0.23)70.30***(0.05)

0.99***(0.02)

(continued)


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

explanatory power is already absorbed in the country pair fixed effects.Although the standard errors are smaller in the FEVD estimation, theefficiency gain does not render the coefficient significant.

Looking at the results of the regressions, including the multilateralterms, we find that the relative size of the importing EU15 countries is notsignificantly different from zero. The average exchange rate variable shows apositive sign and a nearly unit coefficient, indicating that imports from acertain trading partner increase almost proportionally to a depreciation ofthe importing country’s currency against all currencies. A 10% rise incountry i’s geographical distance from all trading partners (remoteness)pushes it to trade 16% more with country j. Dependency and beinglandlocked does not seem to matter on a multilateral basis and border effectsalso play quantitatively a minor role in this sample. The coefficient on themultilateral language variable does not show the expected opposite signs oftheir bilateral counterparts. This may be due to the last term on the RHS of

Table 1. (Continued).

FE RE FEVD

(1) (2) (3) (4) (5) (6)

w/o MT with MT w/o MT with MT w/o MT with MT

MWDij 1.16***(0.21)

1.46***(0.04)

MWLLij 1.47(3.61)

70.56(0.50)

MWBij 0.00***(0.00)

0.03***(0.00)

MWCLij 2.06***(0.29)

1.88***(0.08)

MWDEPij 712.78***(3.49)

0.23(1.26)

MWFTAijt 70.24***(0.09)

0.17***(0.06)

70.24***(0.08)

MWEUijt 70.04(0.08)

0.41***(0.11)

70.04(0.06)

HC1 0.47***(0.04)

0.43***(0.04)

0.61***(0.01)

0.52***(0.01)

HC2 71.85***(0.20)

71.49***(0.20)

71.82***(0.05)

71.74***(0.05)

HC3 0.11(0.11)

0.12(0.10)

0.14(0.10)

0.11(0.10)

0.11(0.11)

0.12(0.11)

Observations 32245 32245 32245 32245 32245 32245R-squared 0.91 0.91 0.84 0.85 0.91 0.91

Robust standard errors in parentheses. *Significant at 10%; **significant at 5%; ***significantat 1%.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

equation 9, the world resistance term, dominating the multilateral terms.Thus, in the world as a whole, there are many common languagesfacilitating trade between every pair of countries and outweighing possiblenegative consequences for bilateral trade of the multilateral languagevariables.

Turning to the interpretation of the FTA coefficients, the results displaythe meaningfulness of the agreements for the CEECs’ integration into theEU. In the FE regression without the multilateral terms, four out of fivedummy variables argue for a significant boost of the EU15 countries’imports brought about by the agreements. Most trade has been created bythe FTAs signed with the Baltic countries (63% above the normal level),closely followed by the Romanian and Bulgarian FTAs (60% above thenormal level). These arrangements are also the ones featuring extra-blocopenness. They increased EU imports from the ROW by 12% and 15%,respectively. The result for the Czech and Slovak Republic agreement issomewhat mixed. While it led to 48% more imports than what would havebeen predicted by the baseline-scenario gravity model, it has reducedimports from third countries by 13%. For the agreements with Hungary,Poland and Slovenia effects on third countries could not be detected.

In general, the results keep holding true when the multilateral resistanceterms are included. Nevertheless, it has to be noted that the coefficients of theFTA dummies (as well as the GDP and distance measures) move up in thesecond set of regressions. As laid out by Baldwin and Taglioni (2006),estimates of currency union dummies are likely to be biased if the relativeprice terms are omitted. Since we rely on a fixed effects model, even in ourfirst regression, only the time-variant part of the Anderson and van Wincoop(2003) terms is ignored. Not accounting for their time-varying charactermeans, however, that the omitted variables enter into the error term. Sincethe ignored multilateral variables are correlated with their bilateral counter-parts they bias the estimates of the latter. The results presented in Table 1confirm this hypothesis. The two time-varying multilateral FTA terms arenegatively correlated with bilateral trade, but may indeed exhibit a positiveimpact on the decision to form an FTA. The average exchange rate, in turn,has a positive effect on the dependant variable. Moving to the estimationresults of the true model in columns 2, 4 and 6 of Table 1, we find that theagreements with the CEECs actually have created between 7 and 20percentage points more trade than suggested by the regression, ignoring thetime-varying component of the relative price terms. In the case of the Czechand the Slovak Republic as well as Slovenia, we also find stronger evidencefor reduced EU15 imports from the ROW. For the FTAs with Romania andBulgaria and with the Baltic States, there is no evidence that the large tradecreation with the EU-15 was at the expense of third countries. To thecontrary, in both specifications, the ROW also gained import shares in thecourse of the implementation of the two arrangements.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

6.2. Robustness checks23

Although the country groupings used in Table 1 follow the historicalcriterion of signing and entry into force of the IAs and EAs (see Table A1),it could be argued that an economic criterion, based on structural similarity,might be more appropriate. Such a criterion would point towards the threecountry groupings commonly used in the East–West European tradeliterature: Visegrad (Czech and Slovak Republics, Hungary and Poland),Balkans (Bulgaria, Romania and Slovenia) and Baltics (Estonia, Latvia andLithuania). Table 2 shows the trade effects obtained from the regressionresults for these alternative country groupings, allowing thereby for a bettercomparison to previous studies. The parameter estimates of the explanatoryvariables underline the robustness of the previous estimation results shownin Table 1. The FTA coefficients on an aggregate level confirm the resultsobtained on an individual basis, although they seem to translate thedifferences in country size, that is the results for the Visegrad countriesmimic Poland and Hungary (which are larger than the Czech Republic andSlovakia) and the results for the Balkan countries mimic Bulgaria andRomania (which are larger than Slovenia). For all CEECs taken together,the IAs and EAs boosted EU imports from that region 72% above theotherwise predicted level without negatively affecting third countries.

In addition to the static approach, we also estimated a dynamicspecification of equation (11), where the lagged dependent variable isincluded among the regressors to control for persistence of import flows.The results are not directly comparable since differencing our stepwise [0;1;2]

Table 2. Robustness to alternative country groupings.

FE FEVD

(1) (2) (3) (4)

FTAiCEECst 0.54*** (0.06) 0.54*** (0.04)FTAit (CEECs)* 0.06*** (0.02) 0.06*** (0.02)FTAivisegradt 0.53*** (0.08) 0.53*** (0.06)FTAibalkant 0.43*** (0.07) 0.43*** (0.05)FTAibalticst 0.48*** (0.08) 0.48*** (0.05)FTAit (Visegrad) 0.00 (0.05) 0.00 (0.05)FTAit (Balkan) 0.07** (0.03) 0.07** (0.03)FTAit (Baltics) 0.01 (0.03) 0.01 (0.03)Observations 32245 32245 32245 32245R-squared 0.91 0.91 0.91 0.91

Robust standard errors in parentheses. *Significant at 10%; **significant at 5%; ***significantat 1%. The other coefficients are omitted but the full list is as in Table 1.

Note: *Since the dates of entry into force of the agreements differ for the countries in theaggregate, we took the ‘mean’ years in assigning the values of 1 and 2 to the dummies measuringthe extra-bloc openness.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

FTA dummy variable wipes out the trade liberalising effects except for theyears the IAs and the EAs entered into force. The estimated coefficient of0.46 (58%) for the FTAijt and 0.07 (7%) for the FTAit variables representtherefore only the initial impact their implementation had on EU imports,but to a large extent cannot capture the gradual nature of tariff reductionsunder the agreements. Nevertheless, the results reveal that the currentimports depend on their previous levels and therefore provide a rationale fora dynamic approach. The remaining coefficients generally behave similarlyin both the static and dynamic specifications.

In order to compare our approach to recent studies that account for theAnderson and van Wincoop (2003) price terms via time-varying dummyvariables, we repeated the regression including country pair specific timetrends as suggested by Bun and Klaassen (2007) instead of the multilateraltrade cost variables. In order to make the regression computationallypossible, we had to restrict the sample to OECD countries and limit thedummies to take triannual steps. Since this transformation dropped thenumber of observations considerably, a comparison can only be made to afixed effects regression with MWR terms on an identical sample. With a timetrend, the coefficient of the FTAijt variable is reduced to 0.15 (0.25 for thefixed effects estimation with MWR variables). Since both the FTA variablesand the country pair dummies are defined bilaterally and time-variant, thedummy might absorb a large part of the explanatory power of the FTAvariables. Baldwin and Taglioni (2006) mention in this context that time-varying pair dummies will make it impossible to estimate factors that affectbilateral trade costs even if they are time-varying as well.

In a final robustness test, we included multiplicative terms between thetrade dummies and some of the gravity variables (income, distance andcommon border). The results prove to be robust to the introduction of theseinteractions, indicating that a rise in exporter GDP and geographicaldistance lowers the trade gain of an FTA. This is a sensible result, as smalleconomies typically gain the most from liberalizing trade, and higherdistances among integrating partners increase trade costs, thus loweringtrade gains.

6.3. Comparison with earlier studies

Evaluating our results in the context of other East–West trade studies, wefind that our FTA coefficient for all CEECs of 0.54 (thus, indicating a tradecreation elasticity of 72%) lies at the upper bound of previous parameterestimates (Table 3).

The differences in the trade creation elasticities seen in Table 3 stem fromdifferent specifications of the gravity equation, varying estimation techni-ques, country samples and time spans. Paas (2003) and Laaser and Schrader(2002) use cross-sectional data and find that integration among the members


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

of the Baltic sea region is more intense than between the CEECs and the EU.Ignoring the time dimension results, however, in a substantial loss ofinformation. This is especially true for FTAs that imply a gradual ratherthan an all-at-once liberalisation process. In a panel data study, Martın andTurrion (2001) estimate a coefficient of 0.83 for intra-group and 1.99 onextra-group export shares. De Benedictis, De Santis and Vicarelli (2005)appropriately account for multilateral trade resistance by including acountry pair specific time trend. Part of their lower estimate of 11% may beexplained by the small country sample (compare the respective robustnesscheck in Section 6.2.) and the exclusion of Romania and Bulgaria. TheFTAs with the two countries lead to a 60% and 80% increase in EU importsin our regressions without and with MWR, respectively (see Table 1). Evenmore important should be the estimation properties. First, the author’s useof a bilateral time trend may absorb some of the explanatory power of thetime-varying bilateral FTA variables. Second, the study applies a systemGMM estimator, reasoning that orthogonal deviations transform help tocircumvent the problem of first differencing (in the difference GMMestimator) taking out most of their time-varying FTA dummy variable. Thisconclusion is, however, incorrect. By subtracting the average of all futureobservations, the effect on the FTA variable is essentially the same: it cancelsout at all but one point in time. Hence, the estimate that De Benedictis, DeSantis and Vicarelli (2005) provide can only capture the trade effect in 2003,the last year of their sample. Close to our procedures appear the approachesof Adam, Kosma and McHugh (2003) and Herderschee and Qiao (2007).The studies rely on time-invariant country (pair) specific fixed effects toaccount for the multilateral resistance terms. Since part of the resistance,namely the multilateral FTA variable, is time-varying and negativelycorrelated with the dependent variable, the results are likely to be downwardbiased. Herderschee and Qiao (2007) also provide evidence of tradedevelopments on an individual country level. In line with our estimated

Table 3. Trade creation elasticities in previous studies.

TCelasticity Estimation technique

This paper 72% FE and FEVDDe Benedictis, De Santis and Vicarelli (2005) 11% System GMMHerderschee and Qiao (2007) 35% FE and system GMMAdam, Kosma and McHugh (2003) 32% Two-step FEMartin and Turrion (2001) 130% Two-step FEPaas (2003) 730% Cross-sectionLaaser and Schrader (2002) 286%*/

14%**Cross-section

Note: *Baltic states’ exports to members of the Baltic sea region, **Baltic states’ exports to therest-EU (estimate not significant).


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

impact of the individual FTAs, the authors find Poland, Romania andEstonia experience the biggest trade gains, whereas the EA with Slovenia ismeasured to have had the lowest impact on the country’s exports to the EU.

7. Conclusions

This paper has paid particular importance to theoretically deriving a newversion of a correctly specified gravity equation to avoid biases present inprevious studies. The specification proposed gives additional importantinsights. First, we are able to show that the exchange rate variablesfrequently empirically employed do stand on a sound theoretical ground andexhibit econometric importance. Second, new measures for multilateraltrade resistance are introduced that mostly show the expected coefficientsigns in the empirical estimation. Third, looking at the agreements on anindividual country basis we conclude that the FTAs have supported andaccelerated the CEECs’ integration into the EU. The process has not beenfree of charge, however, as we also find evidence that although each FTAcreated new trade within the trade bloc, the increase has in the case of theCzech Republic, Slovakia and Slovenia been at the expense of imports fromthe ROW. Finally, as for the aggregate trade effects of the IAs and EAs, ourresult is in line with previous estimates. However, we believe that earlierstudies relying on fixed effects underestimate the agreements’ effect sincethey only partly eliminate the omitted variable biases.

In a modern application of Viner’s theory, other issues arise that gobeyond the scope of this paper. For example, it has to be questioned to whatextent the classic concepts of trade creation and trade diversion, which havebeen conceived in a world without capital mobility, do make sense in ascenario in which the fastest growing component of trade is the onestemming from capital mobility, i.e. from foreign affiliates of MultinationalEnterprises (MNEs). The EU Commission (2006) now recognizes that tradere-orientation is in large part led by the international fragmentation ofproduction of EU multinationals, which have delocalized labour-intensiveparts of their production processes in the East, and then re-importedintermediates through special custom regimes (OPT). These issues (verticalFDI, trade in intermediates, OPT), are points still to be empirically analysedwhen talking about trade effects of regional integration.

Acknowledgements

Thanks are due to two anonymous referees and to Dennis Novy and otherparticipants in the European Economic and Finance Society Conference (Sofia, June2007), in the European Trade Study Group Annual Conference (Vienna, September2006), in the workshop ‘Internationale Wirtschaftsbeziehungen’ (Gottingen, March2007) and in a seminar at the University of Murcia (June 2007) for their valuablecomments and suggestions that contributed to improving this paper. Any remainingerrors are the authors’ full responsibility.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

Notes

1. In this paper, the CEECs are the group formed by the Baltic States (Estonia,Latvia and Lithuania), Bulgaria, the Czech Republic, Hungary, Poland,Romania, Slovakia, and Slovenia.

2. As being subject to Article 133 of the EU treaty (Common Trade Policy),the IAs fell under the Community’s Competence. Details on the exactdates of entry into force of the agreements are provided in Table A1 in theappendix.

3. Compare also Figure 1.4. Our sample period is 1991–2003.5. For a survey on other methods to measure the EU’s trade effects, see for

example Marques (2008).6. Usual estimates of s range from 5 to 8. Consequently a rise in the relative prices

by 1% would cause the total import value to fall by 4 to 7%.7. An exchange rate variable has first been formally introduced into the gravity

equation by Bergstrand (1985).8. Appendix A2 describes the case for a restricted country sample.9. To give an example, for the distance variable this means that a higher distance

of the trading partners i and j towards all other countries in the sampleincreases country i’s imports from j, whilst a high world distance (everyone isfar away from everyone) lowers trade between every country pair. In the case ofdummy variables, the relevant economic interpretation is given by theproportion of ‘1’ versus the proportion of ‘0’ values of the dummies for thevarious country pairs.

10. Compare Melitz (2007) for a similar interpretation of the bilateral trade costvariable.

11. This includes French Polynesia and New Caledonia for France, Aruba and theNetherlands’ Antilles for the Netherlands, and Bermuda and the CaymanIslands for the United Kingdom.

12. The countries are grouped by dates of entry into force of the IAs and EAs. SeeTable A1 for details.

13. The brackets after b4 and b7 indicate that the dummy variables included in tijand MWRij will not be log-linearised whereas distance of course, will.

14. Please find the complete list of variables in Table A3 in the appendix.15. Since YW is constant we implicitly assume no world growth, although countries

i and j may grow. As a consequence, we assume that the positive growth ofsome countries is cancelled by the negative growth of others so that the world asa whole does not grow.

16. One possible source of the serial correlation may stem from the inclusion of thereal exchange rate. In a dynamic setting, a shock to the real exchange rate islikely to affect not only trade, but also the relative prices, with possible feed-back effects.

17. The empirical estimation also contains an EU dummy, controlling for theaccession of Austria, Sweden and Finland in 1995.

18. For the complete country list see Table A5 in the appendix.19. Anderson (1979) shows that the presence of non-tradable goods implies

coefficients lower than unity. The fact that the coefficient of the importer’s GDPis lower than that of the exporter’s GDP suggests a larger non-tradables sectorin the EU15 than in the ROW, on average. Given the large number ofdeveloping countries in the sample, this implication is realistic.

20. We use price quotation throughout the paper so a depreciation corresponds toa rise in the exchange rate.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

21. The elasticity of transport costs to distance is usually associated with anestimate in the range of 0.2 5 d1 5 0.4 (Limao and Venables 2001). Combinedwith an average estimate of s ¼ 7, a distance coefficient between –1.2 and –2.4would be suggested.

22. Since being landlocked is captured by a dummy variable, the import elasticity isgiven by exp(–1.33)–1 ¼ 74%.

23. In order to save space the full results tables for the robustness checks are notpresented here, but can be made available by the authors upon request.

References

Adam, A., D. Kosma, and J. McHugh. 2003. Trade-liberalization strategies: Whatcould Southeastern Europe learn from CEFTA and BFTA? IMF Working Paper239.

Anderson, J. 1979. A theoretical foundation for the gravity equation. AmericanEconomic Review 69, no. 1: 106–16.

Anderson, J., and E. van Wincoop. 2003. Gravity with gravitas: A solution to theborder puzzle. American Economic Review 93, no. 1: 170–92.

Baier, S., and J. Bergstrand. 2001. The growth of world trade: Tariffs, transportcosts, and income similarity. Journal of International Economics 53, no. 1: 1–27.

Baier, S., and J. Bergstrand. 2007a. Do free trade agreements actually increasemembers’ international trade? Journal of International Economics 71, no. 1:72–95.

Baier, S., and J. Bergstrand. 2007b. Bonus vetos OLS: A simple method forapproximating international trade-cost effects using the gravity equation.Available at http://www.eco.rug.nl/*brakman/GravityConference/BergstrandOct2007.pdf.

Baldwin, R. 1994. Towards an integrated Europe. CEPR: London.Baldwin, R. 2006. The Euro’s trade effects. ECB Working Paper 594.Baldwin, R., and D. Taglioni. 2006. Gravity for dummies and dummies for gravity

equations. NBER Working Paper 12516.Baltagi, B., P. Egger, and M. Pfaffermayr. 2003. A generalized design for bilateral

trade flow models. Economic Letters 80, no. 3: 391–7.Bayoumi, T., and B. Eichengreen. 1995. Is regionalism simply a diversion? Evidence

from the evolution of the EC and EFTA. NBER Working Paper 5283.Bergstrand, J. 1985. The gravity equation in international trade: Some micro-

economic foundations and empirical evidence. Review of Economics & Statistics67, no. 3: 474–81.

Broto, C., J. Ruiz, and J. Vilarrubia. 2006. Firm heterogeneity and selection bias:estimating trade potentials in the Euromed region. Available at http://www.eco.uc3m.es/jruiz/MENA_Trade_Potentials.pdf.

Bun, M., and F. Klaassen. 2007. The euro effect on trade is not as large as commonlythought. Oxford Bulletin of Economics and Statistics 69, no. 4: 473–96.

Bussiere, M., J. Fidrmuc, and B. Schnatz. 2005. Trade integration of central and easternEuropean countries: Lessons from a gravity model OENB Working Paper 105.

Bye, M. 1950. Unions Douanieres et Donnees Nationales. Economie applique 3:121–57.

Carrere, C. 2006. Revisiting the effects of regional trade agreements on trade flowswith proper specification of the gravity model. European Economic Review 50:223–47.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

http://www.eco.rug.nl/~brakman/GravityConference/BergstrandOct2007.pdf

http://www.eco.rug.nl/~brakman/GravityConference/BergstrandOct2007.pdf

http://www.eco.uc3m.es/jruiz/MENA_Trade_Potentials.pdf

http://www.eco.uc3m.es/jruiz/MENA_Trade_Potentials.pdf

Cheng, I., and H. Wall. 2004. Controlling for heterogeneity in gravity models oftrade and integration. Federal Reserve Bank of St. Louis Working Paper 1999–010E.

Council of the European Union. 2006. Search the agreements database. Available athttp://www.consilium.europa.eu/cms3_applications/Applications/accords/search.asp? lang¼EN&cmsid¼297.

De Benedictis, L., R. De Santis, and C. Vicarelli. 2005. Hub-and-spoke or else? Freetrade agreements in the enlarged EU. European Journal of ComparativeEconomics 2, no. 2: 245–60.

Egger, P. 2002. An econometric view on the estimation of gravity models and thecalculation of trade potentials. The World Economy 25: 297–312.

Egger, P., M. Pfaffermayr, and P. Schmidt. 2007. What happened to trade in Westernand Eastern Europe in the Aftermath of COMECON? Oxford Economic Papers59, no. 1: 102–26.

European Commission. 2006. EU competitiveness and industrial location. Bureauof European Policy Advisers. Available at http://ec.europa.eu/dgs/policy_advisers/publications/eu_comp_en.htm.

Feenstra, R. 2002. Border effects and the gravity equation: Consistent methods forestimation. Scottish Journal of Political Economy 49, no. 5: 491–506.

Fidrmuc, J., and J. Fidrmuc. 2003. Disintegration and trade. Review of InternationalEconomics 11, no. 5: 811–29.

Frankel, J., and S. Wei. 1998. Open regionalism in a world of continental trade blocs.IMF Working Paper 10.

Hamilton, C., and A. Winters. 1992. Opening up for international trade with EasternEurope. Economic Policy 7, no. 14: 78–115.

Hausman, J., and W. Taylor. 1981. Panel data and unobservable individual effects.Econometrica 49, no. 6: 1377–98.

Herderschee, J., and Z. Qiao. 2007. Impact of intra-European trade agreements,1990–2005: Policy implications for the western Balkans and Ukraine. IMFWorking Paper 126.

Laaser, C., and K. Schrader. 2002. European integration and changing tradepatterns: The case of the Baltic states. Kiel Working Paper 1088.

Limao, N., and A. Venables. 2001. Infrastructure, geographical disadvantage andtransport costs. World Bank Working Paper 2257.

Marques, H. 2008. Trade and factor flows in a diverse EU: What lessons for theeastern enlargement(s)? Journal of Economic Surveys 22, no. 2: 364–408.

Martın, C., and J. Turrion. 2001. The trade impact of the integration of the centraland Eastern European countries on the European Union. European EconomyGroup Working Paper 11.

Melitz, J. 2007. North, south and distance in the gravity model. European EconomicReview 51: 971–91.

Paas, T. 2003. Regional integration and international trade in the context of EUeastward enlargement. HWWA Discussion Paper 218.

Plumper, T., and V. Troeger. 2007. Efficient estimation of time-invariant and rarelychanging variables in finite sample panel analyses with unit fixed effects. PoliticalAnalysis 15, no. 2: 124–39.

Soloaga, I., and A. Winters. 2001. Regionalism in the nineties: What effect on trade?CEPR Discussion Paper 2183.

Viner, J. 1950. The customs unions issue. New York: Carnegie Endowment forInternational Peace.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

http://www.consilium.europa.eu/cms3_applications/Applications/accords/search.asp?lang=EN&cmsid=297




http://ec.europa.eu/dgs/policy_advisers/publications/eu_comp_en.htm

http://ec.europa.eu/dgs/policy_advisers/publications/eu_comp_en.htm

Appendix

A2. Adjusting the model to a limited number of importing countries

In this study, we have to adjust our theoretical framework to the case of EU15countries’ imports (country i) from a worldwide sample of countries (country j).Say that there exist r other importing countries

PRr¼1 countryr ¼

PJj¼1 countryj�PI

i¼1 countryi, whose import prices can be described analogously to country i as

prj ¼ trjPjerj ðA1Þ

Under general equilibrium conditions, output in country j must then equal theaggregate expenditure spent by countries i and r on varieties produced in j,

Yj ¼XIi¼1

Mij þXRr¼1

Mrj ðA2Þ

Making a few mathematical transformations, we can solve for Nj

Nj ¼YjPI

i¼1 ðreijtijÞ1�sYi þ

PRr¼1 ðrerjtrjÞ

1�sYr

ðA3Þ

Plugging equation (A3) into equation (4), country i’s imports arise as

Mij ¼YiYj

YW

tijreij� �1�s

PIi¼1 ðtijreijÞ

1�ssi þPR

r¼1 ðtrjrerjÞ1�ssr

ðA4Þ

For our empirical estimation this means that 1YW

1PR

r¼1srðtrj rerjÞ1�sPI

i¼1siðtij reijÞ1�s

þ1will be absorbed in

the constant. This is equivalent to assuming: (i) a co-movement of the sum of ‘r’countries’ GDPs, average exchange rates and trade costs against j and of the sum of‘i’ countries’ GDPs, average exchange rates and trade costs against j; (ii) a constantworld growth.

Table A1. Dates of entry into force of the Interim and the Europe Agreements.

Dummy CountryInterimAgreement

EuropeAgreement

FTACEECs FTAvisegrad FTAhupo Hungary March 1992 February 1994Poland March 1992 February 1994

FTAczsl Czech Republic March 1992 February 1995Slovakia March 1992 February 1995

FTAbalkan FTArobu Romania December 1993 February 1995Bulgaria May 1993 February 1995

FTAsv Slovenia July 1997 February 1999FTAbaltics FTAbaltics Estonia January 1995 February 1998

Lithuania January 1995 February 1998Latvia January 1995 February 1998

Source: Council of the European Union (2006).


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

A4. The fixed effects vector decomposition estimator

The FEVD procedure estimates in the first stage a standard FE model by conductinga within-groups transformation,

~Mijt ¼ d ~Xijt þ ~eijt ðA5Þ

Table A3. List of variables.

Variable Definition Source

Mijt Yearly imports of countryi from country j

OECD ITCS

Yi(j)t Importer and exporterGDP (in current US$)

UN NAMAD

reijt Bilateral real exchange rate UN NAMAD (nom.exchange rates), IMF IFS(price indices and GDPdeflators), owncalculations*

Dij Great circle distances betweenthe respective trading pairs

CIA World Factbook,own calculations basedon the harvesine formula

LLij Dummy ¼ 1 if one and ¼ 2if both countries in a tradingpair are landlocked

CIA World Factbook

Bij Dummy controlling for theborder length betweencountries i and j

CIA World Factbook

DEPij Dummy ¼ 1 if country j legallydepends on country i

CIA World Factbook

CLij Dummy ¼ 1 if the trading partnersshare a common official language

Wikipedia

EUijt Dummy ¼ 1 for EU member statesEUit Dummy ¼ 1 for non-EU member statesFTAijt Dummy ¼ 1 for contracting parties for

the years following the entry intoforce of the Interim and ¼ 2 forthe years following the entry intoforce of the Europe Agreements

Council of the EuropeanUnion

FTAit Dummy ¼ 1 for non-contractingparties for the years following theentry into force of the Interimand ¼ 2 for the years following theentry into force of the EuropeAgreements

Council of the EuropeanUnion

Sit Share of country i’s GDP in worldGDP (in current US$)

UN NAMAD

Note: *When available the producer price index has been used for the calculation of the realexchange rate. This has been the case for 65 out of 204 countries, among those the ten CEECs.In all other cases we reverted to the consumer price index or the GDP deflator.


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

which removes the bilateral effects mij and the time-invariant variables Tij. From this,one obtains the estimated unit effects mij, including all time-invariant variables, theoverall constant term and the mean effects of the time-varying variables. Inthe second stage, mij is decomposed into an explained part (by the observed time-invariant and rarely changing variables) and an unexplained part hij,

mij ¼ lTij þ hij ðA6Þ

In the last stage, the full model including the residual hij from stage two, but leavingout mij is re-estimated using POLS.

Mijt ¼ aþ dXijt þ lTij þ uhij þ eijt ðA7Þ

Hence, if the orthogonality assumption between the time-invariant variables and theunobserved bilateral effects is correct, the estimator is consistent.

Table A5. Country list.

Afghanistan Cayman Islands GermanyAlbania Central Afr. Rep. GhanaAlgeria Chad GreeceAmerican Samoa Chile GreenlandAndorra China GrenadaAngola Colombia GuamAntigua & Bar. Comoros GuatemalaArgentina Congo D. Rep. GuineaArmenia Congo Rep. Guinea-BissauAruba Costa Rica GuyanaAustralia Cote d’Ivoire HaitiAzerbaijan Croatia HondurasBahamas Cuba Hong KongBahrain Cyprus HungaryBangladesh Czech Republic IcelandBarbados Denmark IndiaBelarus Djibouti IndonesiaBelgium Dominica Iran Islamic Rep.Belize Dominican Republic IraqBenin Ecuador IrelandBermuda Egypt IsraelBhutan El Salvador ItalyBolivia Equatorial Guinea JamaicaBosnia-Herz. Eritrea JapanBotswana Estonia JordanBrazil Ethiopia KazakhstanBrunei Faeroe Islands KenyaBulgaria Fiji KiribatiBurkina Faso Finland Korea Dem. Rep.Burundi France Korea Rep.Cambodia French Polynesia KuwaitCameroon Gabon Kyrgyz RepublicCanada Gambia Lao PDRCape Verde Georgia Latvia

(continued)


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

Table A5. (Continued).

Lebanon Nigeria St. LuciaLesotho N. Mariana Isl. St. VincentLiberia Norway SudanLibya Oman SurinameLiechtenstein Pakistan SwazilandLithuania Palau SwedenLuxembourg Panama SwitzerlandMacao Papua New G. Syrian Arab Rep.Macedonia FYR Paraguay TajikistanMadagascar Peru TanzaniaMalawi Philippines ThailandMalaysia Poland Timor-LesteMaldives Portugal TogoMali Qatar TongaMalta Romania Trinidad & Tob.Marshall Islands Russian Fed. TunisiaMauritania Rwanda TurkeyMauritius Samoa TurkmenistanMayotte San Marino UgandaMexico S. Tome & Pr. UkraineMicronesia Saudi Arabia United Arab Em.Moldova Senegal United KingdomMongolia Serbia-Mont. United StatesMorocco Seychelles UruguayMozambique Sierra Leone UzbekistanMyanmar Singapore VanuatuNamibia Slovak Republic VenezuelaNepal Slovenia VietnamNetherlands Solomon Islands Virgin Isl. (U.S.)Netherlands Ant. Somalia W. Bank & GazaNew Caledonia South Africa Yemen Rep.New Zealand Spain ZambiaNicaragua Sri Lanka ZimbabweNiger St. Kitts & Nevis


Dow

nloa

ded

by [

INA

SP -

Pak

ista

n (P

ER

I)]

at 0

5:41

27

Mar

ch 2

014

trade eﬀects of the europe agreements

Documents