The Uneven Roles of FTAs: Selection Effect or Learning Effect?

Transcription

1 The Unversty of Adelade School of Economcs Research aper No October 200 The Uneven Roles of FTAs: Selecton Effect or Learnng Effect? Faqn Ln

2 The Uneven Roles of FTAs: Selecton Effect or Learnng Effect? Faqn Ln Abstract: revous studes on the role of FTAs n promotng members nternatonal trade have usually focused on FTA premum, gnorng the dfference between selecton effects trade developments before the formaton of FTAs and learnng effects trade growth after the formaton of FTAs. Ths paper consders ths dfference, usng a large blateral trade panel comprsng data coverng more than 50 years from 78 countres. South South FTAs and North South FTAs are most related to the selecton effect whle North North FTAs have a sgnfcant learnng effect. Key Words: Free trade agreements, selecton effect, learnng effect J.E.L Classfcaton: F0, F3, F5. School of Economcs, Unversty of Adelade, South Australa, 5005, Australa, School of Internatonal Trade and Economcs, Unversty of Internatonal Busness and Economcs, Beng, 00029, Chna Emal: faqn.ln@adelade.edu.au, lnfaqn@yahoo.com.cn. I apprecate the helpful suggestons and comments of my supervsors rofessor Chrstopher Fndlay and rofessor Rchard omfret.

3 2. Introducton Snce ts ntroducton by Tnbergen (962), the gravty model has become a standard approach to estmate the ex post role of FTAs n ncreasng members nternatonal trade. Although trade economsts have found mxed results about specfc FTAs, the concluson that such agreements strongly mprove members nternatonal trade domnates the lterature. More recently, Baer and Bergstrand (2007, 2009) consdered a possble endogenety problem of the FTA varable n gravty equatons by usng dfferent approaches ncludng non-parametrc matchng econometrcs to fnd sgnfcant premum effects of FTAs. However, prevous studes usually focused on the FTA premum wthout consderng the causes of such a premum, whch mght be ether a selecton effect,.e. countres wth more blateral trade are more lkely to form FTAs, or a learnng effect,.e. trade growth after the formaton of FTAs. The two effects wll have dfferent sgnfcance from a polcy perspectve. In ths paper, I use a large blateral trade panel comprsng data coverng more than 50 years from 78 cpuntres to re-estmate the effects of FTAs. The data n ths study cover the maorty of global trade, wth lttle data selecton bas. Furthermore, I control for other varables that may have an mpact on members blateral trade snce omtted varable bas s the most mportant source of endogenety (Baer and Bergstrand, 2007). The results suggest that North South and South South FTAs are related manly to the selecton effect whle North North FTAs have a sgnfcant learnng effect; North North FTAs have a sgnfcant mpact on members trade growth whereas North South and South South FTAs do not. I conclude that after the formaton of FTAs, North North countres acheve deeper For example, Brada and Mendez (985) found sgnfcant effects of the European Communty on trade flows among members, whle Frankel, Sten and We (995) found the effects were nsgnfcant.

4 3 ntegraton but for North South and South South countres, FTAs do not appear to promote blateral trade. The rest of the paper s structured as follows. I brefly dscuss the gravty model, data and related control varables n secton 2. Secton 3 examnes the results of FTA premum wth cross-secton data and secton 4 dscusses the results wth the panel approach and frst dfferencng. Secton 5 addresses the selecton effect test and learnng effect test. Secton 6 concludes. 2. Gravty Model and Data 2. Gravty equaton The gravty equaton s typcally used to explan cross-sectonal varaton n country pars trade flows n terms of the countres ncomes, blateral dstance, and dummy varables for common languages, common land borders, the presence or absence of FTA/WTO, etc. Tnbergen (962) was the frst to estmate the ex post role of FTAs n ncreasng members nternatonal trade wth a gravty equaton. For nearly half a century the gravty equaton has been a standard approach for cross-country emprcal analyses of free trade agreements on nternatonal trade flows. Here we use extended gravty equaton to control for other possble factors to reconsder the uneven roles of FTAs. ln( X ) FTA 0 Landl 7 GS Colony 2 7 ln D Island 8 2 ComNat 3 YY 3 ln( YY ) 4 ln( CU Bothn op op ln( Area Area ) ComCol ) Lang Onen 6 5 CurCol Border 6 () where and denote tradng partners. The varables are defned n Table : Table : Defntons of varables

5 4 Varables Defntons X Average value of real blateral trade between and FTA Bnary varable whch s unty f and both belong to the same regonal trade agreement D Dstance between and Y Real GD op opulaton Lang Dummy varable whch s unty f and have a common language and zero otherwse Border Bnary varable whch s unty f and share a land border Landl The number of landlocked countres n the country par (0,, or 2) Island The number of sland natons n the par (0,, or 2) Area The area of the country (n square klometres) ComCol Bnary varable whch s unty f and were ever colones after 945 wth the same colonzer CurCol Bnary varable whch s unty f s a colony of at tme t or vce versa Colony Bnary varable whch s unty f ever colonzed or vce versa ComNat Bnary varable whch s unty f and remaned part of the same naton durng the sample CU Bnary varable whch s unty f and use the same currency at tme t Bothn Bnary varable whch s unty f both and are GATT/WTO members Onen Bnary varable whch s unty f ether or s a GATT/WTO member GS Bnary varable whch s unty f was a GS (generalzed system of preferences) benefcary of or vce versa Represents the omtted other nfluences on blateral trade, assumed to be whte nose process In the past 30 years, some trade economsts have sought formal theoretcal economc foundatons for the gravty equaton (Anderson, 979; Bergstrand, 985; Deardorff, 998; Baer and Bergstrand, 200; Eaton and Kortum, 2002; Anderson and van Wncoop, 2003). The strkng characterstcs of these studes about the theoretcal foundatons are the ntroducton of prces or partcular forms of multlateral prce ndexes. Anderson and van Wncoop (2003) llustrate the omtted varables bas caused by gnorng prces n the crosssecton gravty equaton. They suggest theoretcally that the gravty model should be wrtten as:

6 5 ln( X / Y Y ) FTA CU ln D ln( Area Area ) ComCol Bothn Lang Onen 6 5 Border CurCol GS ln 7 6 Landl Colony 2 7 ln Island 8 ComNat 3 (2) where and are denoted as multlateral (prce) resstance terms. Anderson and van Wncoop then estmate ths system usng a customzed non-lnear least squares program, treatng all varables ( =...N countres) as endogenous. 2 However, Anderson and van Wncoop (2003) and Feenstra (2004) refer to an alternatve and computatonally easer way wth country-specfc fxed effects for estmatng the multlateral prce terms and. Ths method wll also generate unbased coeffcent estmates of the vector. Thus, we can wrte the theoretcally motvated cross-secton model as: ln( X / Y Y ) FTA CU ln D ln( Area Area ) ComCol Bothn Lang Onen 6 5 Border CurCol GS D 7 6 Landl Colony 2 7 D 2 2 ComNat Island 3 8 (3) where D s a dummy whch s unty f the exportng country s and D 2 s a dummy whch s unty f the mportng country s. For the two sets of country-specfc fxed effects, one for exporters and one for mporters, one set of dummes s enough (Rose, 2004). For ths reason, n ths paper, wth blateral trade I use the frst country fxed effects to control the prce effects. Furthermore, I also use specfcaton () wth country fxed effects to estmate the average treatment effect of FTAs, and call specfcatons (2) and (3) a unty ncome elastcty equaton. 2 The equaton needs equlbrum condton constrants, and n these condtons the multlateral prce terms become endogenous.

7 6 2.2 Data The data come from Rose s homepage 3 and trade data come from the Drecton of Trade (DoT) CD-ROM dataset developed by the IMF. The data cover blateral trade between 78 IMF tradng partners between 948 and 999 but wth gaps (Rose, 2004). FOB exports and CIF mports are recorded n US dollars. Rose has deflated trade by the US CI. opulaton and real GD data (n constant US dollars) have been obtaned from the enn World Table, the World Bank s World Development Indcators, and the IMF s Internatonal Fnancal Statstcs. Rose explots the CIA s World Factbook for a number of country-specfc varables ncludng: lattude and longtude, land area, land-locked and sland status, physcally contguous neghbours, language, colonzers, and dates of ndependence. I use these to create controls. Rose also adds nformaton on whether the par of countres was nvolved n a currency unon. FTA dummes are from the WTO, and WTO dummes were created by Rose. Usng North to denote hgh ncome countres and South to denote non-hgh ncome countres, I employ the World Bank s defnton of Hgh Income countres to dsaggregate the data nto three, North North, North South and South South FTAs. In 999, there are 223 FTAs of country pars n our dataset, 94 are North North FTAs, 3 are North South FTAs and the other 98 are South South FTAs. 3. Cross-secton Estmates I frst show the average treatment effect estmates of the FTA premum usng a crosssecton theory-motvated gravty equaton wth multple years (960, 970, 980, 990 and 999). 4 Table 2 provdes the results for these years estmated wth specfcaton () and 3 The webste s 4 The frst FTA was ntated n 958 by France, Netherlands, Belgum, Luxemburg, Italy and Germany accordng to the Treates of Rome, whch s why I have looked at the perod followng 958 here.

8 7 country fxed effects. Table 3 gves the coeffcents estmated wth specfcaton (3) wth unty ncome elastcty. Table 2: FTA remum: Cross-Secton Results wth Specfcaton () Global (North North) (North South) (South South) (0.48) (0.200) N/R / / *** *** (0.56) (0.97) (0.333) N/R / / / *** *** 2.080*** (0.244) (0.79) (0.276) (0.292) N/R / / / / *** *** 2.579*** (0.28) (0.6) (0.47) (0.358) N/R / / / / *** 0.454*** 2.064***.324*** (0.2) (0.08) (0.326) (0.80) N/R / / / /0.685 Robust standard errors n parentheses *** p<0.0, ** p<0.05, * p<0. Table 3: FTA remum: Cross-secton Results wth Specfcaton (3) Global (North North) (North South) (South South) *** 0.349* (0.54) (0.205) N/R / / *** *** (0.53) (0.72) (0.328) N/R / / / *** 0.269* 2.660*** 2.000*** (0.24) (0.60) (0.278) (0.298) N/R / / / / *** *** 2.640*** (0.202) (0.40) (0.408) (0.378) N/R / / / / *** 0.37*** 2.068***.300*** (0.4) (0.07) (0.326) (0.84) N/R / / / /0.584 Robust standard errors n parentheses *** p<0.0, ** p<0.05, * p<0.

9 8 From the tables we can see that some coeffcents are not stable from year to year, especally the coeffcents of the varable for North North blateral trade, whch means that the cross-secton estmates are not relable owng to omtted varables bas even though I try to control the possble effects (Baer and Bergstrand, 2007, 2009). As well, Novy (2008) and Stack (2009) made the pont that gravty relatons may change over tme. However, for North South and South South blateral trade, most of the FTA estmates are postve and sgnfcantly dfferent from zero. To test the robustness of the fndngs and to address the bas by usng a cross-secton equaton (Baldwn and Taglon, 2006), I now use the panel data to test the treatment effect of FTAs n dfferent trade groups. 4. FTA Treatment Effects wth anel Data Fxed-effects and frst dfferencng approaches wth panel data are effectve ways to control for tme-nvarant effects (Wooldrdge, 2002). Wth panel data, the gravty model s as follows: ln( X t ) FTA 0 Border 6 GS t t CurCol 7 t t ln ln D Landl 7 t Colony 2 2 t ln t ln( Y Y ) Island 8 3 ComNat 3 t t YY ln( ln( Area t CU 4 Tmedummy 9 4 t t t op Area op ) t Bothn 5 ) t Lang t t ComCol 0 5 t Onen 6 t (4) Furthermore, wth unty elastcty of ncome suggests estmatng: ln( X / Y Y ) t FTA ln( Area CU t t Area Bothn 5 Tmedummy ln D t ) t 2 t t t Lang ComCol 0 t Onen 6 5 t t Border CurCol GS 7 6 t t t Colony ln Landl 2 7 t ln t Island ComNat 3 8 t t (5) As the dataset covers the blateral trade of 78 countres coverng more than 50 years, there are extensve panel data but wth gaps. There are four ways to estmate the treatment effects of FTAs: pooled OLS, fxed effects, random effects and frst dfferencng. However,

10 9 for unbalanced panel data, the random effects approach requres much stronger assumptons compared wth fxed effects (Woodrdge, 2002). Therefore, I use pooled OLS, fxed effects and frst dfferencng to test the average treatment effects of FTAs. Secton 4. reports the pooled OLS and fxed effects regressons, and Secton 4.2 the frst dfferencng method. 4. ooled OLS and Fxed Effects Table 4 provdes the results of treatment effects of FTAs wth varous specfcatons by the pooled OLS method. We can see that most of the FTA coeffcents are sgnfcantly postve, whch means that FTAs have a sgnfcant premum,.e. countres wth FTAs have more blateral trade than countres wthout FTAs. However, the FTA premum s uneven across dfferent trade groups. For North South and South South trade, FTAs have a much hgher premum. The FTA coeffcents for North South FTAs are 0 tmes larger than for North North FTAs, and the coeffcents for South South FTAs are even larger. However, as OLS cannot completely control the tme-nvarant effects I use a fxed effects approach to rerun the regressons. Table 4: FTA Coeffcents Estmated Usng the ooled OLS Method Global (North North) (North South) (South South) OLS wth Country Effects 0.972*** 0.9***.482***.802*** (0.0334) (0.0230) (0.0994) (0.0535) N/R / / / /0.467 OLS wth Country and Tme Effects.29*** 0.96***.758*** 2.02*** (0.0337) (0.0220) (0.099) (0.059) N/R / / / /0.55 OLS wth Unty Income Elastcty and Country and Tme Effects.327*** 0.28***.923*** 2.003*** (0.0333) (0.022) (0.0987) (0.0520) N/R / / / /0.435 Robust standard errors n parentheses *** p<0.0, ** p<0.05, * p<0.

11 0 Table 5 shows the results of a fxed effects approach. We can see that most FTA coeffcents are postve and sgnfcant, whch confrms the exstence of an FTA premum. Whle the uneven roles of FTAs between dfferent trade groups stll exst, the coeffcents of North South FTAs lose ther sgnfcance and so we can expect that for North South FTAs, the premum s not robust wth the pooled OLS approach. South South FTAs stll have larger FTA coeffcents than North North blateral trade although these are largely decreased compared wth pooled OLS estmates, whch are about twce that of North North FTAs. Therefore, t can be concluded that the North South FTA premum s not robust and that South South FTAs have a much larger premum than North North FTAs. Table 5: FTA Coeffcents Estmated Usng Fxed Effects Global (North North) (North South) (South South) Fxed Effects wth Tme Effects 0.764*** 0.57*** *** (0.0732) (0.067) (0.93) (0.50) N/R / / / /0.07 Country ar Fxed Effects wth Unty Income Elastcty and Tme Effects 0.852*** 0.662*** 0.9.6*** (0.0720) (0.0679) (0.9) (0.46) N/R / / / /0.7 Country ar Robust standard errors n parentheses *** p<0.0, ** p<0.05, * p< Frst Dfferencng If we use the frst dfferencng method, the correspondng equatons to specfcatons (4) and (5) should be as follows:

12 ln( X, ) FTA, 2 ln( YY ) 3 ln( ) opop CU 4, Tmedummy Bothn 5,, Onen 6 Y Y, GS 7 CurCol,, (6) ln( X / Y Y ) FTA Bothn 5,, Tmedummy CurCol Onen 6,,, CU 4 GS 7,, (7) where. The fxed effect and frst dfferencng methods are both effectve ways to control the endogenety problems caused by omtted tme-nvarant varables. In contrast to prevous studes wth ntervals of the tme perod, n ths paper I employ the data of every year from 948 to 999, so the economc meanng of the estmated coeffcents wth frst dfferencng wll be dfferent from the fxed effects. Fxed effect coeffcents of FTAs can be regarded as the FTA premum of blateral trade, that s, countres wth FTAs have more blateral trade flows than those wthout FTAs. However, frst dfferencng coeffcents can be treated as the role of FTAs n promotng trade growth after ther formaton. Table 6: FTA Coeffcents Estmated Usng Frst Dfferencng Global (North North) (North South) (South South) Frst Dfferencng wth Tme Effects *** (0.0507) (0.0206) (0.82) (0.093) N/R / / / /0.03 Frst Dfferencng wth Unty Income Elastcty and Tme Effects *** (0.058) (0.0204) (0.90) (0.04) N/R / / / /0.08 Robust standard errors n parentheses *** p<0.0, ** p<0.05, * p<0.

13 2 Table 6 reports the results usng the frst dfferencng method. Most of the coeffcents estmated are not statstcally sgnfcant. North North FTAs have a strong effect n promotng trade growth after the FTA s formed. Although South South FTAs have a much hgher FTA premum, they do lttle to promote trade growth after the formaton of the FTA. Smlar to the export and productvty lterature, ther premum may be caused by a selecton effect (larger blateral trade partners form FTAs) but the learnng effect (after FTA formaton) s not sgnfcant. For North North FTAs, although there s only a mnor FTA premum but a sgnfcant learnng effect, after the FTA trade growth s sgnfcant. In the next secton I test the hypothess. 5. Selecton Effect or Learnng Effect In ths secton, I use standard econometrcs to test the mechansm of the FTA premum: selecton effect or learnng effect? To test the selecton effect and learnng effect, I choose 960, 970, 980, 990 and 999 as the tme perods to study. Secton 5. looks at the selecton effect and secton 5.2 dscusses the learnng effect. 5. Selecton Effect Test Frst, I test the selecton effect. As wth the self-selecton effect n export and productvty shown n the frm-level lterature, I use the above mentoned hypothess namely, that the more developed s blateral trade, the easer t s to form an FTA to nvestgate the pre-fta dfferences n trade volume between countres wth FTAs and countres wthout FTAs. If countres wth greater trade volume fnd t easer to form FTAs, we should expect to fnd sgnfcant dfferences n trade volume between countres wth future FTAs and countres that never form FTAs several years before the formaton of FTAs. To test whether countres that now have FTAs had more developed blateral trade before enterng these agreements than countres that have not entered FTAs I select all countres

14 3 that dd not begn a FTA between years t- and t- and estmate the average dfference n blateral trade volume n year t- between those countres that formed the FTAs n year t and those that dd not, controllng for other factors n year t-. More formally, we estmate the gravty model: ln( X, ) FTA Lang 5, ComCol 0 Bothn 5 0,,, t ln D Border 6, CurCol Onen 6 2,, Landl, ln( Y Y ) 7 Colony GS 7 3 2,t-n, Island 8, t YY ln( 4,t-n ComNat 3 op op ln( Area 9, ) Area CU 4 ), (9) Specfcally, I frst let t = 999, n =, 2, 3, 4, and keep countres that dd not have FTAs from 960 to 990. Thus, the pre-entry premum that can be computed from the estmated coeffcent, shows the average percentle dfference between 999 s FTA countres and 999 s non-fta countres n 960, 970 and 980, controllng for the characterstcs some years before. Because the number of FTAs ncreased rapdly after 999, 5 I gnore the pre-entry premum n the nearest year pror to 999, whch s 990 n the case of based estmates. Smlarly, I also let t = 990, n =,2,3 and keep countres that dd not have FTAs from 960 to 980. I then compute the pre-fta premum between 990 s FTA countres and 990 s non-fta countres n 960, 970 and 980, controllng for the characterstcs some years before; and let t = 980, n =,2 and keep countres that dd not have FTAs from 960 to 970. I then compute the pre-fta premum between 980 s FTA countres and 980 s non-fta countres n 960 and 970, controllng for the characterstcs some years before. Because most FTAs were formed after 970, especally North South FTAs and South South FTAs, and some North North FTAs were formed before 960, there 5 In our sample, n 999 the number of FTAs (RTAs) s 223, and accordng to WTO statstcs, close to 400 RTAs are scheduled to be mplemented by 200. See the webste:

15 4 s lttle varaton from 960 to 970 6, and thus I do not estmate the pre-fta premum between 970 s FTA countres and 970 s non-fta countres n 960. Table 7 shows the results of the selecton effect. From the coeffcents of, we can see that most of the estmates about North South and South South FTAs are sgnfcant and there are large postve values but that there are also sgnfcant and large negatve values for North North FTAs. That s, the selecton effects for North South and South South FTAs are much more sgnfcant than for North North FTAs and there seems lttle relaton between the selecton effects and North North FTAs. The selecton effects are the most sgnfcant for South South FTAs; blateral trade n 970 for South South FTAs formed n 980 was 229.2% hgher than n South South blateral trade n countres wthout FTAs throughout the perod, controllng for other possble effects. Thus, we can conclude that the premum of South South FTAs s caused by the selecton effect; North South FTAs also dsplay ths selecton effect, but the premum has lttle relaton wth the selecton effect n North North FTAs. Table 7: Selecton Effect Blateral Trade FTA(999) FTA(990) FTA(980) North-North * 0.683*** *** 0.790*** * 0.602*** North South ***.348*** ***.750** *** South South * 2.076***.285*** * 2.292*** 6 In our dataset, the number of FTAs was only 5 n 970, and n 999 t was 223. North North FTAs were frst formed n 958, and between 958 and 970 no further North North FTAs were formed; the number of North North FTAs n 999 was 94.

16 *** Note: * p<0., ** p<0.05, *** p< Learnng Effect Test Now we turn to the learnng effect trade growth after the formaton of FTAs. The learnng effect s much more mportant from the polcy perspectve, and the polcy mplcaton s qute dfferent from the selecton effect. To do the formal estmate, I choose countres that form FTAs at year t and those that never form FTAs to estmate how FTAs ncrease blateral trade after the formaton of FTAs compared wth those that never form FTAs. I do not nclude countres wth FTAs over the entre sample perod but use countres that begn the FTA process after 960. Because most FTAs formed after 970, I do not estmate the promotng effect of 970 FTAs but let t be the years 980, 990, 999, and t n 999. Formally, I run the followng regresson to estmate the learnng effect of FTAs roles n promotng nternatonal trade growth after the formaton of FTAs compared wth those countres wthout FTAs. ln( X, ) ln( X, Lang 5, ComCol 0 Bothn 5 ) FTA Border,, 0 6, t, CurCol Onen 6, ln D Landl 7 2,, Colony GS 7 2,t-, ln( Y Y ) 3 Island 8,,,t- YY ln( ln( Area ComNat 3 9 4, op op Area CU 4 ) ), (0) The learnng or promotng effect can be estmated by the coeffcent. Ths coeffcent shows the promotng effect of trade growth at varous years compared to the year before the FTA s formaton ndcatng percentle dfference. Table 8 reports the results of. From the estmated coeffcents, we can clearly see that the learnng effect s sgnfcant for North North FTAs, but not for North South and South South FTAs. For North North

17 6 FTAs formed n 980, the trade growth between 970 and 980 s on average 93.5% faster than n those countres wthout FTAs throughout the perod, controllng for other possble effects. Thus, we can conclude that the learnng effect of North North FTAs s the man cause of North North FTAs premum, but for North South and South South FTAs, the FTA premum has lttle relaton wth the learnng effect. Table 8: Learnng Effect FTAs North North FTAs FTA(980) 0.389** 0.782*** 0.935*** FTA(990) 0.666*** 0.56*** FTA(999) North South FTAs FTA(980) FTA(990) FTA(999) 0.23 South South FTAs FTA(980) 0.832* FTA(990) FTA(999) Note: * p<0., ** p<0.05, *** p<0.0.the trade growth s as follows: for FTA (980), the trade growth s between 980 and 970, 990 and 970, 999 and 970; for FTA (990), the trade growth s between 990 and 980, 999 and 980; for FTA (999), the trade growth s between 990 and 999. From above analyss, we can see that both the selecton effect and the learnng effect are strongly uneven across dfferent trade groups. There s an uneven premum of dfferent FTAs n selecton 3 and selecton 4, wth the South South FTAs havng a much more sgnfcant premum than North North FTAs, whch shows the mechansm more clearly. The premum of North North FTAs s manly caused by learnng effects whle South South FTA premums are much more related to selecton effects. For North South FTAs, some level of selecton effect domnates the nsgnfcant learnng effect and causes the premum of North South FTAs not to be robust. Thus, North North FTAs have real roles n promotng members blateral trade. However, f we care only about the tradtonal FTA premum, we wll be msled on the roles of FTAs, whch s extremely mportant for polcy

18 7 makers. It seems that for South South and North South FTAs, trade development has a sgnfcant nfluence on the choce of formaton. Whle FTAs usually do not ncrease blateral trade once they are formed, North North FTAs acheved deep ntegraton after ther formaton and greatly ncreased blateral trade. 6. Concluson The roles of FTAs can be analysed by both the selecton effect and the learnng effect, however, prevous studes were concerned only wth the FTA premum. The more developed blateral trade s between two countres, the easer t s to form FTAs. Ths s the selecton effect. The learnng effect, whch s borrowed from export and productvty n the frm-level lterature, shows that blateral trade growth s faster for those countres that formed FTAs compared wth those countres that dd not. I use a large blateral trade panel comprsng data coverng more than 50 years from 78 countres to reconsder the roles of FTAs based on the above two effects. The data cover most global trade, so there s lttle selecton problem. As I also control other varables that may have an mpact on blateral trade, there should be lttle omtted varable bas. An extensve nvestgaton reveals some new fndngs. Although South South FTAs had a much hgher FTA premum than North North FTAs, South South FTAs domnated n the selecton effect and North North FTAs domnated n the learnng effect. The North South FTA premum was offset by a sgnfcant selecton effect and an nsgnfcant learnng effect. It can be nferred that after FTA formaton, North North countres acheve deep ntegraton but North South and South South countres do not dsplay any sgnfcant development n blateral trade growth. Ths fndng suggests that the functon of FTAs related to South developng countres should be reconsdered.

19 8 REFERENCES Anderson, J.E. (979), A Theoretcal Foundaton for the Gravty Equaton, Amercan Economc Revew, 69(), Anderson, J.E. and E. van Wncoop (2003), Gravty wth Gravtas: A Soluton to the Border uzzle, Amercan Economc Revew, 93(), Baer, S.L. and J.H. Bergstrand (200), The Growth of World Trade: Tarffs, Transport Costs and Income Smlarty, Journal of Internatonal Economcs, 53(), 27. Baer, S.L. and J.H. Bergstrand (2007), Do Free Trade Agreements Actually Increase Members' Internatonal Trade?, Journal of Internatonal Economcs, 7(), Baer, S.L. and J.H. Bergstrand (2009), Estmatng the Effects of Free Trade Agreements on Internatonal Trade Flows Usng Matchng Econometrcs, Journal of Internatonal Economcs, 77(), Baldwn, R. and D. Taglon (2006), Gravty for Dummes and Dummes for Gravty Equatons, NBER Workng aper No. 256 (Cambrdge, MA: Natonal Bureau of Economc Research). Bergstrand, J.H. (985), The Gravty Equaton n Internatonal Trade: Some Mcroeconomc Foundatons and Emprcal Evdence, Revew of Economcs and Statstcs, 67(3), Brada, J.C. and J.A. Mendez (985), Economc Integraton among Developed, Developng and Centrally lanned Economes: A Comparatve Analyss, Revew of Economcs and Statstcs, 67(4), Deardorff, A.V. (998), Determnants of Blateral Trade: Does Gravty Work n a Neoclasscal World?, n J.A. Frankel (ed.), The Regonalzaton of the World Economy (Chcago: The Unversty of Chcago ress).

20 9 Eaton, J. and S. Kortum (2002), Technology, Geography, and Trade, Econometrca, 70(5), Feenstra, R.C. (2004), Advanced Internatonal Trade: Theory and Evdence (rnceton, NJ: rnceton Unversty ress). Frankel, J.A., E. Sten and S.J. We (995), Tradng Blocs and the Amercas: The Natural, the Unnatural, and the Super-natural, Journal of Development Economcs, 47(), Novy, D. (2008), Gravty Redux: Measurng Trade Costs wth anel Data, Warwck. Economc Research apers No. 86(UK: Unversty of Warwck). Rose, A.K. (2004), Do We Really Know that the WTO Increases Trade?, Amercan Economc Revew, 94(), Rose, A.K. (2004), Response to Subramanan and We, Resource Document (See Stack, M. (2009), Regonal Integraton and Trade: Controllng for Varyng Degrees of Heterogenety n the Gravty Model, The World Economy, 32(5), Tnbergen, J. (962), Shapng the World Economy (New York: The Twenteth Century Fund). Wooldrdge, J.M. (2002), Econometrc Analyss of Cross-Secton and anel Data (Cambrdge, MA: MIT ress).