Policy Choice: Theory and Evidence from Commitment via International Trade Agreements *

Transcription

1 Polcy Choce: Theory and Evdence from Commtment va Internatonal Trade Agreements * Nuno Lmão Unversty of Maryland, NBER and CEPR Patrca Tovar Brandes Unversty Ths verson: November 2008 (Frst verson: December 2007) Abstract Why do governments use relatvely neffcent polces to redstrbute ncome towards specal nterest groups (SIGs) when more effcent ones are avalable? To emprcally address ths longstandng puzzle we derve theoretcal predctons for a specfc set of polces: tarffs vs. non-tarff barrers. We show that a government barganng wth ts SIGs can gan by commttng to lmt tarffs through nternatonal trade agreements even f ths leads to the use of less effcent non-tarff barrers (NTBs). Such constrants mprove the government's barganng poston relatve to SIGs and thus ncrease ts share of the surplus. Moreover, we show that the NTB restrctveness ncreases wth the strngency of the tarff bndng. We also show how the decson to commt to constrants on any gven good depends on the government s barganng power relatve to SIGs. Usng detaled data, we test these predctons. In goods where Turkey commtted to tarff constrants n trade agreements we fnd a hgher lkelhood and restrctveness of NTBs. Furthermore, these effects are stronger for more strngent tarff constrants. Moreover, we fnd that the hgher the government barganng power relatve to a SIG the more relaxed the tarff constrant t chooses and the less lkely t s to constrant t at all, as predcted by the model. The man results are stronger for organzed ndustres, whch further supports the theory. The commtment channel hghlghted by the theory generates a quanttatvely large ncrease n the use of NTBs and we provde a structural estmate showng that these polcy nstruments are ndeed less effcent than the tarffs they replace. JEL classfcaton: C7; D7; F13; H2. Keywords: Transfers, lobbes, commtment, barganng, trade, tarff, non-tarff barrer * Contact nformaton: Lmao@econ.umd.edu, tovar@brandes.edu. Lmão gratefully acknowledges the research assstance of Antone Gervas and Shrayes Ramesh as well as the hosptalty of the Kel Insttute for World Economy where part of ths research was conducted. We thank Stephane Aaronson, Chad Bown, Peter Debaere, Allan Drazen, Andre Levchenko, Horst Raff, Robert Stager, Alan Sykes and partcpants at the Amercan Economc Assocaton Meetngs, Stanford Law and Economcs semnar, the George Washngton Unversty Semnar on Regonalsm, Kel Unversty and Internatonal Economc Insttutons Workshop n Seoul for useful comments and dscussons. Any remanng errors are ours.

2 1 INTRODUCTION Many economc polces are enacted as a form of redstrbuton towards specal nterest groups (SIGs). Ths s not puzzlng snce governments are not socal welfare maxmzers. What s puzzlng s that such redstrbuton s often done usng polces that appear to be neffcent,.e. polces that reduce the surplus that governments and SIGs can bargan over. Why are nstruments such as geographcally targeted publc projects or producton subsdes used for redstrbuton when, n the absence of specfc externaltes, lump-sum payments would be more effcent? Why are techncal regulatons used to restrct trade nstead of tarffs; or tarffs used nstead of producton subsdes? Most polcy models focus on a sngle polcy and gnore the exstence, choce and use of multple polces for a gven objectve. Ths use of partal poltcal equlbrum models can lead to erroneous normatve prescrptons and postve predctons. Although there s a growng theoretcal lterature dealng wth ths mportant puzzle there s almost no emprcal work addressng t. One lkely reason s that almost all theores addressng ths ssue model why a partcular agent (or group) prefers neffcent polces but they do not model the mechansm through whch those preferences are channeled. Thus, as we descrbe n the next secton, these theores generally do not provde postve predctons for whch polcy s actually used n equlbrum. The two man contrbutons of ths paper are the followng. Frst, we derve specfc predctons based on a model wth an explct mechansm for the choce of neffcent redstrbuton polces. Second, we test these predctons usng detaled product level data. More specfcally, our theoretcal model bulds on Drazen and Lmão (2008) who provde a theory of polcy choce where the government bargans wth SIGs over the level of a lump-sum transfer or producton subsdy. One of the key nsghts of ther model s that the government can beneft from constrants on the relatvely more effcent polcy (a lump-sum transfer) because although such constrants reduce the total surplus avalable they ncrease the government s share. Importantly, they also show how the constrants on the effcent polcy emerge under alternatve frst stage choce mechansms provded the government has some ablty to commt to such constrants. In order to tghtly lnk the theory and estmaton we must focus on a specfc set of polces. We do not know of any avalable detaled data on lump-sum transfers and producton subsdes to drectly test Drazen and Lmão. Thus, we alter t to examne a small country s choce of alternatve trade polces. Ths s a settng that often allows an effcency rankng of polces and one where the use of neffcent trade polces for redstrbuton s an mportant puzzle, as stressed by Rodrk (1995), for example. More mportantly perhaps, ths s a settng where governments have access to a ubqutous 1

3 commtment mechansm for whch detaled data s avalable: nternatonal agreements. The World Trade Organzaton (WTO) generally forbds producton subsdes; a relatvely more effcent redstrbuton polcy than tarffs, whch the WTO allows ts members to negotate and bnd. 1 The WTO s members have typcally placed even fewer constrants on varous non-tarff barrers (NTBs), whch are often even less effcent than tarffs. These NTBs can nclude quanttatve restrctons, product standards, mport surcharges, etc, and some argue that as tarffs have fallen they have been replaced wth NTBs. 2 In sum, ths s an nterestng settng for analyzng polcy choce n general and one that also allows us to address the mportant specfc queston of why small countres constran themselves n the use of tarffs and the effect ths has on the use of NTBs. We begn by showng that a self-nterested government n a small compettve economy can beneft from an nternatonal commtment to constran the most effcent avalable redstrbuton polcy or polces (e.g. producton subsdes and tarffs). Commttng to these constrants n any gven sector mproves the barganng poston of the government versus the SIG n that sector, as t lmts the maxmum redstrbuton t can make for a gven payment made by that SIG. However, these agreements do not constran all polces so lobbes can generally fnd some other one (e.g. an NTB) that s less effcent but stll allows them to explot any poltcal gans from trade. 3 We show that despte ths the government can stll beneft from the constrants. Gven the data avalablty on alternatve polces, we focus on the mpact of tarff constrants on NTBs and derve the precse structural relatonshp, whch we estmate. In our model, a bndng tarff cap ncreases the lkelhood of an NTB but ncreases n that cap reduce the trade restrctveness of the NTB. The structural approach can also provde estmates of the average neffcency of NTBs. We then show that governments choose not to constran products n whch they have suffcently hgh (Nash) barganng power relatve to lobbes. The ntuton s as follows: a government wth hgh barganng power captures most of the total surplus, so reductons n the surplus due to the constrants cannot be offset by an ncrease n the share the government captures. Moreover, f 1 In the last trade round for example, the percentage of ndustral tarff lnes subject to bndngs ncreased from 22 to 72 n developng countres, and 18 to 100% n agrculture (Martn and Francos [1997]). 2 Hllman (1989) states that ( ) because GATT negotatons have succeeded n securng substantal multlateral tarff reductons, non-tarff barrers have n many nstances come to replace tarffs as the means for protecton. (p. 76). Also, Marvel and Ray (1983) fnd that the tarff lberalzaton that resulted from the Kennedy Round led to substtuton of nontarff barrers for tarffs n the Unted States, and they argue that t may well be that n some ndustres overall protecton actually ncreased. Kee et al (2006) provde advalorem equvalent estmates of NTBs for several countres that show they are very trade restrctve. 3 There may be dfferent reasons for ths, one recently emphaszed by Magg, Horn and Stager (2006) s that t s costly to agree on any sngle polcy and thus trade agreements reman ncomplete contracts n order to save on such costs. 2

4 governments do choose to commt then we fnd less strngent caps on products where ther barganng power s hgher. We fnd support for several of the model's predctons by usng tarff and non-tarff barrers for about 5,000 goods. To explot varaton n tarff constrants across goods usng detaled data we focus on a sngle country, Turkey. We dscuss several reasons for ths choce n secton 4; one of them s that t allows us to analyze two of the most common types of commtment n tarffs: those mposed va multlateral agreements such as the WTO and va preferental trade agreements (PTAs). Goods wth tarff constrants set through the WTO and the PTA wth the European Unon ncrease the probablty and restrctveness of NTBs n Turkey. These effects are smaller when the tarff constrant s relaxed, whch s precsely what the theory predcts. We fnd stronger effects for the ndustres the theory apples to: those organzed n to SIGs. Moreover, the effects of caps are related to ther actual mplementaton dates, suggestng that the exstence of a cap s not smply a proxy for some other product characterstcs. Also, n unbound goods the appled tarffs are not negatvely correlated wth NTBs. The dfference n the tarff effect suggests that our fndng for the tarff effect n bound goods s not smply pckng up some substtuton between tarffs and NTBs n all goods, whch could arse due to some omtted factor. We also analyze whether governments choose constrants as predcted by the model. To focus on the model's central mechansm we construct a novel emprcal measure of government barganng power relatve to lobbes: ther relatve probablty of survval, as suggested by the theory. We fnd that the government s less lkely to constran tarffs n the WTO n products where t has hgh barganng power. Moreover, the government sets less strngent tarff constrants n goods where t has hgher barganng power. Ths result s partcularly strong for organzed ndustres, as the model predcts. We show the results are robust to varous ssues ncludng endogenety concerns, whch we address usng an IV approach. Moreover, the effects are also economcally sgnfcant. Turkey s government s more lkely to place tarff constrants on products from ndustres where t has low barganng power (60%) than n other ndustres (38%) and those constrants are about 20 percentage ponts tghter n low barganng power ndustres. We fnd that the tarff commtment effect due to the barganng power channels translates nto a large ncrease n the average NTB ad-valorem equvalent. Moreover, we provde a structural estmate of the NTB equaton that shows they are sgnfcantly less effcent than tarffs n organzed ndustres. In sum, the results support several predctons of the model. Ths strongly suggests that the model provdes a useful lens to analyze commtments of other countres va nternatonal agreements on trade and possbly other polces used for redstrbuton. 3

5 The structure of the paper s the followng. In secton 2, we dscuss the related lterature on polcy choce and commtment n trade agreements. In secton 3, we ntroduce the model and derve testable predctons. We test these predctons n secton 4 and then conclude. 2 LITERATURE Ths paper spans three topcs: the general polcy choce puzzle, the value of commtment va trade agreements and the relatve effcency of trade polces. We now brefly dscuss each. 2.1 Polcy Choce One general argument for the use of relatvely neffcent polces s that they make redstrbuton towards SIGs costler and thus act lke sand n the wheels of the redstrbutve process. Ths sand causes a reducton n the equlbrum amount of redstrbuton relatve to more effcent polces, whch can explan why relatvely neffcent polces may be preferred from a socal welfare perspectve. Ths type of mechansm s employed by Becker and Mullgan (2003), Rodrk (1986) and Wlson (1990) for example. However, these papers provde a normatve rather than a postve theory of neffcent transfers snce they leave the government n the background and do not model the polcy choce process. In contrast, n our approach, the government s an actve player and by modelng the frst-stage of polcymakng, we can provde a postve theory of neffcent transfers. Ths s partcularly mportant gven that our man goal s to test the model. 4 Other papers, such as Grossman and Helpman (1994) and Dxt, Grossman, and Helpman (1997), do not focus on the choce of redstrbuton polcy. But they do provde an argument as to why competton among SIGs for government transfers can mply that more dstortonary nstruments mprove the outcome for SIGs. These papers do not model the frst-stage polcy choce and also dffer from ours n other mportant ways. Competton among lobbes s not present n our model, so n our settng the SIGs generally prefer effcent polces whereas the government may prefer the opposte. Another key dstncton s that two basc modelng assumptons n those papers transferable utlty and no government barganng power actually mply that neffcent polces would not be adopted n our settng. The other promnent argument s the dsgused transfer dea put forward by Tullock (1983). Those who bear the costs of fundng a certan polcy may be gnorant of ts redstrbutve effects to SIGs and thus less lkely to oppose t f the polcy also has some socal beneft. Coate and Morrs 4 Moreover, n our model governments prefer the neffcent polces because they mprove ts barganng poston relatve to SIGs, whch s qute dstnct from the sand n the wheels argument. In fact, n our model, the decrease n barganng surplus from usng the neffcent polcy s costly for the government, so such polces are used n spte of actng lke sand n the wheels not because of t. 4

6 (1995) formalze ths dea and show that a bad poltcan one who values socal welfare and the utlty of the SIG drectly may choose the neffcent transfer (a one-off project that favors the SIG) nstead of a lump-sum transfer. They show that poltcan may be elected f there s asymmetrc nformaton relatve to the voters about the socal value of the project and the ams of poltcans. Ther model has the advantage of beng fully specfed n terms of the polcy choce. However, testng ts predctons s dffcult for another reason. As Coate and Morrs themselves note, the requrements that the project be socally benefcal n some states of nature and that voters have mperfect nformaton about ts effect (ex-ante and ex-post) mply that ther model s best suted to explan publc projects rather than tarffs, subsdes, etc (p. 1228). So when testng ther model one would need to (a) fnd systematc data on such projects and (b) determne f they were effcent. But f one s ndeed able to determne that effcency ex-post wth any certanty then the model would predct neffcent polces would not be used. Our model on the other hand does apply to polces such as subsdes, tarffs and NTBs whose relatve effcency s easer to determne and where the ablty to do so has no effect on the results snce we do not rely on asymmetrc nformaton. As we note n the Introducton we buld on Drazen and Lmão (2008). The key theoretcal dfference s that n ths paper we model tarffs and NTBs whereas they focus on a lump-sum transfer and a producton subsdy. Moreover, we analyze trade agreements as the polcy choce mechansm wth a vew to emprcal mplementaton. The sngle most mportant contrbuton relatve to ther work s that we derve and test several predctons. 5 The work on ths topc remans largely theoretcal. The excepton thus far s Ederngton and Mner (2006) who examne the determnants of tarffs as a share of tarff plus producton subsdy protecton. They use aggregate data for a panel of countres and mostly fnd support for the revenue generaton hypothess,.e. that because tarffs generate revenue they may be preferred to the typcally more effcent producton subsdes. The authors note n the concluson the dffculty n testng some theores snce (at that tme) none of the theoretcal models proposed a fully specfed equaton for the proper rato of tarffs to other polcy nstruments. Therefore, ther approach s to draw broad mplcatons from these models and then test them usng aggregate data, whch mples accordng to them that none of the results should be nterpreted as an outrght rejecton of any model. Our 5 Another broad argument that has been explored for the use of neffcent transfers s that they can gve poltcal benefts to the government or SIG that lump-sum transfers do not (c.f. Shepsle and Wengast [1981] and Wengast, Shepsle, and Johnsen [1981], Forna [1981] and Baron [1991], Dxt and Londregan [1996], Dxt and Londregan [1995], Acemoglu and Robnson [2001]). 5

7 approach tackles these ssues by specfcally dervng such equatons from a fully specfed model and testng them usng detaled product data Small Countres and the Value of Trade Agreements There s a long standng vew that trade agreements are valuable because they provde governments wth a commtment mechansm to better wthstand or mtgate mport competng pressures. 7 Somewhat surprsngly, ths vew has been formalzed almost exclusvely by appealng to specfc tme-nconsstency problems related to some form of nvestment. Stager and Tabelln (1987) consder a model n whch the reallocaton of labor after an adverse terms-of-trade shock s costly, and the government has an ncentve to provde unexpected protecton to redstrbute ncome after the shock. Commtment to free trade helps avod a tme-consstent equlbrum n whch labor reallocaton s reduced as people antcpate protecton and the government fulflls those expectatons wth socally excessve protecton levels. 8 Magg and Rodrguez-Clare (1998) extend Grossman and Helpman (1994) by allowng captal to be fxed n the short run but moble n the long run. They show that the government may beneft from commttng to free trade to avod a dstorton assocated wth the allocaton of resources for whch t may not receve compensaton by the lobbes. The dstorton can arse from overnvestment by a poltcally organzed ndustry, a bad equlbrum n whch an ndustry that would not produce under free trade becomes actve, or a slower than optmal rate of ext from a declnng ndustry. Mtra (2002) also models the nteracton between the government and the lobbes as a Nash barganng game, but he removes captal moblty and nstead ntroduces fxed costs of poltcal organzaton. Besdes fndng that governments wth low barganng power wll want to pre-commt to free trade agreements (as n Magg and Rodrguez-Clare [1998]), he shows that for suffcently hgh fxed costs of lobby formaton the government wll prefer not to commt. The strategc nteracton between nternatonal and domestc polcy negotatons has long been known (c.f. Putnam [1988]). Smlarly to the theory papers descrbed above we also explot the strategc nteracton due to a government s ablty to commt va trade agreements. But there are several key dfferences. Frst, the source of the gan from such commtment n our model s a government s mprovement n ts barganng poston relatve to the lobbes rather than a standard tme nconsstency 6 Chandra (2007) also studes the substtuton between subsdes and tarffs and fnds that Chna s tarff reductons upon enterng the WTO were smaller n products where t was most lkely to face retalaton f t used subsdes. 7 The other man argument s that trade agreements allow countres wth market power to reduce tarffs n a recprocal way and nternalzes terms-of trade effects (Bagwell and Stager [1999]). Magg and Rodrguez-Clare (2007) combne the commtment motve explored n ther 1998 paper wth the terms-of-trade motve. 8 Tornell (1991) also addresses tme-nconsstency problems n trade polcy. 6

8 problem. The underlyng barganng mechansm we explot s thought to be mportant n negotatons. Schellng (1960) states that The power of a negotator often rests on a manfest nablty to make concessons and meet demands. He goes on to argue ths s an advantage that domestc constrants can brng n an nternatonal negotaton but clearly, the effect can also run n the opposte drecton. Second, none of the papers above models the choce of polcy and, wth the excepton of Stager and Tabelln (1987), they do not even consder the possblty of alternatve polces. Ths s mportant because no nternatonal agreement allows commtment n all polces so to evaluate the value of such agreements we need to move away from partal poltcal equlbrum models. 2.3 The Relatve Effcency of Tarffs vs. Non-Tarff Barrers In the settng we focus on a small compettve economy wth no uncertanty a tarff s generally at least as effcent as an NTB that leads to the same traded quanttes. However, some theoretcal papers n trade provde counter examples where NTBs can be more effcent, economcally or poltcally. For example, the losses from some NTBs may be less transparent than those of a tarff snce the latter provdes an explct expresson of the ncrease n domestc prce relatve to the world (cf. Hllman [1989] for a dscusson of these arguments). A smlar argument has been appled to the use of tarffs vs. producton subsdes (Magee, Brock, and Young [1989]). Ths smply apples the dsgused transfer dea to ths settng and s thus subject to the same comments as above. Moreover, as Falvey and Lloyd (1991) argue, the mportance of the transparency motves has decreased over tme, as estmates of the costs of protecton have become more avalable (p. 463). Informaton asymmetres are not needed n our model. Most papers on ths topc examne the relatve effcency of tarffs vs. a specfc type of NTB: quotas; and provde a motve for whch the quota s ablty to control quanttes becomes an advantage. In Kaempfer et al. (1989) ths advantage s drven by the exstence of a domestc monopoly such that f quotas are used as an addtonal source of protecton they reduce the overproducton resultng from tarffs. Others have focused on the role of uncertanty about world prces under rsk averson where a quota may be preferred to a tarff because t stablzes the mport quanttes and thus the domestc prce (c.f. Young [1980], Young and Anderson [1982], Hllman [1989]). 9 We do not dspute that there are nstances when the ablty to precsely control quanttes generate an advantage of quotas over tarffs. We doubt that ths reversal of effcency between tarffs and quotas (or other NTBs) s the norm for most goods and f t were then our model would not apply and we should not fnd emprcal support for t. Moreover, snce the Uruguay Round many quanttatve 9 Falvey and Lloyd (1991) focus on the relatve effcency of quotas and tarffs under domestc demand vs. supply shocks. 7

9 restrctons were actually outlawed and NTBs now often take other forms such as product and techncal standards and varous forms of mport charges (Mchalopoulos [1999]). 10 Thus, the NTB we model can capture not only the effects of a quota but more generally of a measure that generates a wedge between domestc and world prces. More mportantly, our model also has the advantage that t can explan why tarffs nstead of even more effcent measures (e.g. producton subsdes) wll be used for redstrbuton and the way n whch the choce across dfferent polces occurs. Ultmately, whether appled NTBs are on average less effcent than tarffs s an emprcal queston and our model wll provde us wth an estmaton equaton that allows us to answer t. 3 THEORY 3.1 Setup We consder a small compettve economy that takes world prces as gven. Indvduals factor endowments may dffer but they have dentcal preferences descrbed by u = x n 0 + u ( x ) = 1 where x0 s consumpton of the numerare good; x denotes consumpton of good and the sub-utlty functons u ( ) are dfferentable, ncreasng and strctly concave. An ndvdual wth ncome E consumes 1 x = d ( p ) = [ u' ( p )] of each, and x = 0 E p d ( p ) of the numerare. The ndrect utlty s thus gven by v ( p, E) = E + s( p), where p s the vector of domestc prces, and the consumer surplus derved from the non-numerare goods s gven by ( d ( p )) s (p) = u p d ( p ). The numerare good s produced usng only labor wth a margnal product equal to one. Assumng the supply of labor s large enough ths good s always produced and thus the wage equals unty n equlbrum. Each of the non-numerare goods s produced usng labor and a sector-specfc factor, wth constant returns to scale. The supply of the specfc factors s fxed. Snce the wage s fxed, we can denote the return to the specfc factors as Π p ) a functon of domestc prces. By Hotellng s lemma, output s then gven by y = Π p ). ( ( 10 Whle determnng the relatve effcency of each type of NTB at each pont n tme relatve to tarffs s nearly mpossble, we now have evdence that the ones mposed are hghly trade restrctve n a large set of countres. Usng data and estmates from Kee et al (2006) we fnd that the overall trade restrctveness ndex for the typcal country s equvalent to a unform tarff of 14% f we gnore NTBs, but t jumps to 27% when NTBs are ncluded. Moreover, for about 35 countres that ndex doubles when NTBs are ncluded. 8

10 The government has tarffs and NTBs at ts dsposal. 11 For concreteness, we consder NTBs that generate a wedge between the domestc and foregn prce, as a tarff does, and can also generate rents. We model NTBs as less effcent than tarffs n a sngle dmenson: a fracton φ of those rents s dsspated whereas n the tarff case they are avalable n the form of tarff revenues that can be consumed n the mportng country. A specfc example of such NTBs are quanttatve restrctons where some of the lcenses are gven to foregners or to resdents that must burn resources n some rent seekng process. By allowng φ to range between zero and one we can capture dfferent degrees of the neffcency n a smple but clear way. 12 So, for the model s purposes the key requrement s that the NTB s less effcent than the tarff and t generates a percentage ncrease n the domestc prce, whch we refer to as the NTB s advalorem tarff equvalent and denote by τ. The total per capta rents avalable domestcally from usng τ and an advalorem tarff t s then gven by r( p ) = tp + (1 φτ ) p d( p) y( p) N [ ] where p s the world prce and tp measures the ncrease n the domestc prce due to the tarff whereas τ p s the equvalent wedge due to the NTB. The second term n brackets represents mport quantty (N measures the total populaton). All the tarff rents (.e., ts revenue) are avalable for domestc consumpton, but for the NTB that s the case for only a fracton 1 φ [0,1]. We assume that the government redstrbutes these rents unformly to all ndvduals. In an exogenously gven set of sectors L, the specfc factors are able to organze nto SIGs that lobby the government. The jont gross welfare of lobby s: W =Π ( p ) + α N[1 + r( p) + s( p )] where α s the fracton of the populaton that owns some of the specfc factor n ths ndustry and the terms n brackets are those owners wage rents and consumer surplus. We analyze the case of hghly 11 Ths polcy set s determned by the data avalable to test the model n the emprcal secton. We could have nstead ncluded producton subsdes and tarffs and the same mechansm we explore would mply that the government would gan by restrctng subsdes and usng tarffs. Alternatvely, f we allowed subsdes, tarffs and NTBs we conjecture that the qualtatve results would be smlar to the ones we obtan as long as governments had a commtment technology to constran subsdes and tarffs as occurs n the WTO. 12 If there s revenue-seekng behavor then we can renterpret φ as the dfferental amount that s wasted n NTBs relatve to tarffs. It s reasonable to consder φ >0 gven that NTBs tend to generate lumpy amounts of rents that can be dscretely allocated. For addtonal arguments and evdence on the relatve neffcency of quotas see Anderson (1988). Ths does not mply that the model captures all the dfferent types of NTBs, whch would be mpossble. As dscussed n the lterature revew, certan NTBs have the same effects as tarffs n some economc envronments but not n others. Nonetheless, the key nsght of the model should apply to varous NTBs as long as they generate a wedge relatve to the world prce and are less effcent than the tarff n maxmzng the poltcal surplus, as defned below. 9

11 concentrated factor ownershp, α 0, so each ndustry lobbes only for ts own product. Ths allows us to focus on the nteracton between the government and each SIG and abstract from lobby competton. Each SIG offers the government a lobby good, represented by C and descrbed below, n order to obtan an ncrease n the level of protecton t receves. Thus, we denote the net welfare of the members of lobby byv = W C. The government maxmzes a weghted sum of lobby goods and socal welfare: G = Ψ ( C ) + aw ( p ), a 0 (1) L where socal welfare s gven by the sum of ndrect utltes over all ndvduals, whch ncludes wage and specfc factor ncome plus net taxes (or rents) from polcy and consumer surplus: n W( p) = N + Π ( p ) + N[ r( p) + s( p )]. = 1 Several models of SIGs, e.g. Grossman and Helpman (1994), assume that the lobby good s equally valued by the government and the lobby,.e. that Ψ s lnear and thus utlty s transferable. Ths s a useful smplfyng assumpton that may be reasonable when that lobby good s cash contrbutons and there are no lmts on them. However, n several countres ncludng the one we analyze n the emprcal secton there are constrants on such contrbutons. Thus SIGs can and do resort to other goods and servces, whch are not necessarly perfect substtutes. Moreover, as Drazen and Lmão (2008) argue, poltcans may have dmnshng margnal utlty for lobby goods such as gettng out the vote n a dstrct where a lobby's membershp s concentrated; provdng nformaton about an ssue; lendng jets for campagnng or vacatonng; etc. Thus we thnk that n several settngs t s reasonable to assume, as we do, that Ψ s strctly concave. As we wll see the resultng nontransferablty of utlty between government and lobbes s key n generatng the use of neffcent polces. 13 There are two stages n the game. In the frst stage, the government may set a cap on the tarffs t can choose. In the second stage, the government (Nash) bargans wth each of the SIGs over the level of lobby goods and trade polces that wll be mplemented subject to the frst stage constrants. 3.2 Absence of Commtment and the Use of the Most Effcent Avalable Polcy We frst show that f the government does not set a cap on the most effcent avalable polcy, then ths s the only polcy used n equlbrum. Ths wll hghlght the role of commtment by showng that n ts 13 Alternatvely, we could model non-transferablty by allowng C to enter lnearly n G but requre t to be produced by each lobby usng the numerare as the nput nto a dmnshng returns producton process. 10

12 absence our model predcts that only the most effcent avalable polcy s used. The absence of a cap can ether be due to no access to a commtment technology or a choce not to use t. 14 Before showng the relatvely neffcent polces are not used f no constrant s set on tarffs we brefly explan the ntuton. It s obvous that a tarff s more effcent from a socal welfare perspectve than the alternatve polcy n our setup. Ths s smply because f both polces lead to the same level of mports and prces then under the NTB a fracton of the rents are dsspated. But the relevant defnton of effcency n the context of the polcy choce puzzle s poltcal effcency,.e. whch polcy maxmzes the jont payoff to the government and lobby for any gven level of the lobby good. It s straghtforward to see that n our model the two neffcency defntons (socal and poltcal) exactly match, snce the government objectve s a weghted value of socal welfare and lobby goods. Therefore for a gven level of mports and contrbutons, the lobby payoff, V s dentcal under t or τ, but the government payoff s lower under τ. So t s both socally and poltcally more effcent n ths setup. Ths mples that n equlbrum t s costler for the lobby to compensate the government for an ncrease n τ that leads to the same change n mports as an equvalent change n t, and thus only the latter s used. To analyze the specfc case of no commtment we employ the general Nash barganng problem solved n the second stage and assume the frst-stage tarff constrant s not bndng. In the followng secton we relax ths. Formally, we wrte the maxmzaton as follows: max t, τ, C U = 0 ( ) γ 0 1 G( t, τ, C) G ( V ( t, τ, C) V ) γ 0 0 s.t. t t c ; G G ; V V (2) Lettng λ be the multpler assocated wth the tarff constrant, the frst order condtons for t, τ and C when tc>, 0 and V V G G 0 0 >, > are gven respectvely by: γ G G γ G G γ G G G G G 1 γ + V V t V 0 t τ 1 γ + V 0 V V 1 γ + V V τ C V 0 C λ = 0 0 = 0 (3) (4) (5) c t t 0 (6) and the (omtted) correspondng complementary slackness condtons for (4) and (6); here a subscrpt denotes a partal dervatve. To ensure an nteror soluton we assume throughout that Ψ (0). 14 In our emprcal work the goods where such a commtment s absent wll be used as the counterfactual to test the model s predcton that constrants on a polcy lead to the use of relatvely less effcent ones. 11

13 When the tarff cap s absent or not bndng, then λ equals 0 and from (3) and (5) we get: * 2 * G t Vt at ( ) ( ) = p m y p p = Gc Vc Ψ ( C ) 1 Subsequently we wll determne the optmal constrant and whether t bnds relatve to the unconstraned, so t s useful to derve the (mplct) value of the unconstraned tarff from (7) as: t = y ( p ) Ψ ( C ) u * apm ( p) Thus, organzed ndustres receve tarff protecton. Note that f C entered the government s objectve lnearly then ths expresson s smlar to the well-known expresson obtaned by Grossman and Helpman (1994) for an organzed ndustry when factor ownershp s concentrated. To show that only the tarff s used n the absence of a commtment note that τ = 0 f (4) s negatve. Ths nequalty must hold whenever λ = 0 and (3) holds wth equalty,.e. whenever there s an unconstraned postve tarff. Ths s straghtforward to show because Vτ = Vt both polces have a smlar effect on the domestc prce and thus proft and G than the tarff. τ t (7) (8) < G snce the NTB generates fewer rents 3.3 Commtment Tarff Caps n the Absence of NTBs We now allow government access to commtment so t can choose whether t prefers to set a maxmum cap on the tarff pror to negotatng wth each SIG. To clearly llustrate the government s ncentve to do so we frst assume that no other redstrbuton polces can be used. In the next secton we show the government s ncentve s stll present when less effcent polces are avalable. The government sets the cap, t c, n the frst stage by maxmzng ts objectve, n (1), takng nto account the effect on the equlbrum tarff and contrbutons, whch are found by solvng the Nash problem prevously defned but wth τ = 0 as an addtonal constrant. The frst order condton for t c s C W Ψ ( C ) + a = 0 (9) t t whch we solve to obtan t C t = Ψ ( C ) * 2 a( p ) m c If t c t u (10) then the constrant does not bnd; otherwse t bnds and ths would prove the government s beneft from constranng tarffs ex-ante. Gven the ndependence of rrelevant alternatves n Nash barganng, the government would actually be ndfferent between t c = t u and any 12

14 hgher constrant, so we can focus on determnng f (10) s equal to (8) or lower. Omttng the product subscrpts the condton for a non-bndng constran s u C t c yt ( ) u Ψ ( Ct ( )) = Ψ ( Ct ( )) * 2 c * u a( p ) m ( t ) ap m ( t ) If t c = t u then the equlbrum values of C, p, and thus y and m' would be dentcal n (8) and (10). Replacng these above we should then obtan: C t = Π t In the Theory Appendx we derve C t by mplctly dfferentatng (5) and show that the equalty above holds f and only f ether (a) the government has all the barganng power or (b) lobby goods are valued lnearly so utlty would be transferable. If the government does not have all the barganng power and lobby goods have dmnshng margnal utlty, then C t < Π, t.e. we have a contradcton that shows the government prefers a constrant. The ntuton s the followng: f the constrant bnds then nstead of the equalty above we have ( Π C) t > 0, whch means that relaxng the constrant,.e. ncreasng the tarff, would ncrease the payoff to lobbes wth no frst order cost to the government (snce t s optmally settng t). The resultng ncrease n jont surplus could be collected va the barganng n the second stage f the government has all the barganng power. Alternatvely, t would also be collected f contrbutons enter lnearly so that they are used to share the jont surplus. But f contrbutons have dmnshng margnal utlty and the government cannot obtan the entre jont surplus then the ncrease n jont surplus from relaxng the cap s offset by a smaller government share of t. Ths decrease n the share s due to the deteroraton n the government s poltcal terms-of-trade. In other words, a bndng cap mproves the government s barganng poston thus generatng a beneft for t that offsets the loss due to the decrease n jont surplus. One pont to note n the precedng analyss s that the government s able to extract contrbutons from the lobby under a tarff cap even n the absence of NTBs as long as t can credbly threaten to set a zero tarff. Ths s certanly the case wth WTO commtments snce they are defned as a maxmum tarff so the analyss apples drectly to ths case. However, n the case of the customs unon the government threat of a zero tarff s less credble snce the customs unon partner may not allow t to fulfll t. In practce, customs unons probably allow some flexblty for members to threaten to devate down from the exact external tarff. However, n the extreme case where they do not the only threat governments can use n barganng wth SIGs s the cap tself n whch case the government can only extract contrbutons f addtonal unconstraned polces are avalable, as we now analyze. 13

15 3.4 Commtment and the Co-exstence of Effcent and Ineffcent Polces SIGs are notorously creatve n fndng alternatve redstrbuton polces and the government s not able to constran all of them. 15 Therefore, we now show how constrants on tarffs lead to the emergence of less effcent polces. We frst take these constrants as gven, as they would be n the second stage, and derve ther mpact on NTBs. Ths s one of the relatonshps we estmate n the emprcal secton. In the next secton we provde testable predctons for whch goods the government chooses to commt to such constrants. c Let us frst consder under what stuatons a tarff cap t leads to the use of NTBs n the second stage. Clearly f that cap s equal to zero then f the NTB were also set to zero there would be gans from trade between the lobby and government. Ths occurs snce when t=τ=0 the lobby offers C=0 and the margnal beneft of ncreasng C s suffcently large to the government. Ths s smple to verfy usng the frst order condtons n (4) and (5). Gven the strctly postve and large gans from poltcal tradng at t = τ = 0, t s also straghtforward to show that an NTB wll also be used for some strctly postve cap level. However, as c u the cap ncreases the NTB value must eventually declne snce, as we have shown earler, when t = t we have τ = 0. So our model predcts that: (1) a good wth a suffcently low tarff cap mples an NTB wll be used and (2) the lkelhood and the value of the NTB are eventually decreasng n the value of that cap, or more precsely the dfference between that cap and the unconstraned tarff value. Alternatve models could predct smlar relatonshps between tarffs and NTBs n all goods. One specfc feature of our model, and one we wll test, s that t predcts these relatonshps only n goods where a bndng tarff cap exsts. Our model also provdes specfc predctons about whch goods the government chooses to commt n. Before dervng these addtonal predctons, we derve the NTB level for a gven bndng tarff to show the results mentoned above. The nteror NTB soluton when there s a bndng cap requres the margnal rate of the substtuton across polces for the government to be equal to the lobbes, as we can see from solvng (4) and (5) to obtan: G G V τ τ = c V c c * 2 * ( τ φ + ) φ a (1 ) t ( p ) m pm y ( p ) p = Ψ ( C) 1 * (11) 15 Ths can be ether because of nsttutonal constrants of the agreement (e.g. the tradtonal focus of WTO and PTAs to restrct certan producton subsdes and tarffs but not other polces). We take the avalablty of commtment technology across polces as gven but conjecture that the model can be extended to predct that the governments would optmally choose to frst commt n the most effcent polces f there s some fxed cost to commtng. 14

16 whch we can re-arrange to obtan c 1 y( p ) c φ 1 τ = Ψ ( C ) * t (12) 1 φ apm 1 φε * ε = m ( p ) p m ( p ) s a measure of the absolute value of mport demand elastcty. We where ( ) mmedately see that f the tarff constrant was zero and the NTB was almost as effcent as the tarff,.e. φ 0, then the NTB level would equal the frst term n the brackets, whch s exactly the unconstraned tarff level n (8), t u. More generally, the NTB level s ncreasng n the gap between the unconstraned level and the cap. Ths s one of the central predctons we wll test Commtment Decson We now proceed to the ntal stage and ask under what condtons the government chooses to commt. More specfcally, we derve predctons about whch goods the government chooses to commt n and, when possble, the level of that commtment. The government commtment decson n the frst perod can be modeled n two alternatve ways. Frst, allow t to choose whether or not to commt to an exogenously gven cap. Second, allow t to choose both whether to commt and the optmal level of the cap that maxmzes ts objectve n the frst stage. Clearly the condtons for the frst alternatve are more strngent snce the government cannot optmally choose the cap. It s mportant to show the result for the frst alternatve snce t s often the only avalable one. One example s when a country must adopt another s common external tarff and t can at most decde f a gven good s subject to an exogenous cap level. Another example s f the government has some nfluence over the level of the cap but s unable to choose t to exactly maxmze ts objectve. Such a stuaton could occur when there are other constrants (unobservable to us) that prevent t from choosng the optmal cap level (e.g. a sectoral agreement of all WTO members on agrculture or Informaton Technology). Thus we derve suffcent condtons such that a government can beneft from a commtment to cap even f t s not necessarly able to set ts level optmally. As wll be clear, such condtons wll also be suffcent for the government to commt f t could choose the optmal cap and wll also allow us to generate a predcton about the level of that cap. 16 Note that when t c and φ are postve the equlbrum value of the expresson that captures the unconstraned level motves s dfferent from the equlbrum value of t u n (8). Moreover, the equlbrum value of the NTB can dffer for a gven postve commtment level t c n the WTO vs. a customs unon snce n the latter the government generally cannot threaten to set the tarff below the commtment. Thus n a good bound under the customs unon the government wll typcally collect fewer (or no) contrbutons for the tarff although t wll stll be able to collect them for NTBs, whch t can always threaten to set to zero. 15

17 In the frst stage, the government chooses to commt to an exogenously gven cap n a product f ts payoff evaluated at the constraned equlbrum exceeds the unconstraned: 17 c c c c c u u u Gt (, τ ( t, γ ), C( t, γ ),.) Gt ( ( γ ), τ ( γ ), C( γ ),.) (13) The NTB and contrbuton values under the tarff constrant are the equlbrum ones determned n the second stage as explaned n the prevous secton. The unconstraned values are the ones determned n u secton 3.2, whch we showed mply τ ( γ ) = 0. A suffcent condton for (13) to hold also mples that the government prefers to commt and choose the optmal cap, snce that amounts to allowng t to choose the value of c t that maxmzes the left-hand sde of (13), whch confrms our asserton n the prevous paragraph. We llustrate the result graphcally n Fgure 1, whch depcts payoffs n G-V space. The bold lne represents the Pareto fronter n the absence of commtment. It yelds a hgher jont payoff than the alternatve where the tarff s constraned, whch we have seen n the prevous secton mples the less effcent NTB s used. Note frst that for a large enough barganng power (13) never holds, that s the government always chooses not to commt to a cap. Ths s obvous for γ = 1, snce then the government obtans the entre surplus and never wants t reduced. It s also smple to llustrate that the same s true for other suffcently hgh γ that are stll lower than 1. When the cap s suffcently low the maxmum possble government payoff s strctly lower than wth no constrant. We can thus defne u h cm h as the level at whch G ( γ = γ ) = G ( γ = γ ), as shown n fgure 1. Therefore, governments wth suffcently hgh barganng power do not commt to a strngent cap. Ths s true even though we focus on a case where the commtment does not worsen the government s threat pont. 18 Governments wth no barganng power are ndfferent between polces. When γ s zero, equaton (13) must hold wth equalty snce the government obtans ts reservaton payoff, whch s the free trade equlbrum n good and thus t s dentcal wth or wthout commtment. Therefore, (13) holds wth a strct nequalty f as we ncrease γ from zero the government payoff ncreases faster under commtment than n ts absence. When ths s the case the government has a strctly hgher payoff h γ 17 Gven the separablty of G over goods, we can treat the choce over each good ndependently of the values for other goods. If the government had to choose between enterng an agreement wth an exogenously gven set of caps on a set of goods and could not opt out of any gven one then we would need to consder the aggregate effect and (13) would not necessarly have to hold for each. In the WTO and n some PTAs the government has some dscreton to opt out, whch s why we focus on ths formulaton. 18 As we noted before the threat pont of zero tarffs s credble after a commtment to a tarff n the WTO snce t s mplemented as a maxmum. In the case of the customs unon the threat pont may no longer be a zero tarff n whch case there s an addtonal cost to the government from commttng. Snce the Turksh government had lttle choce over the bndng tarffs n the EU we do not analyze ths choce n the emprcal work and thus do not extend the theoretcal model to nclude t here ether. 16

18 * * under commtment for some γ (0, γ ), where γ s defned as the lowest postve γ at whch (13) holds wth equalty. The formal suffcent condton for ths s d d lm Gt (, τ ( t, γ ), C( t, γ ),.) > lm Gt ( ( γ ), τ ( γ ), C( γ ),.) (14) c c c c c u u u γ 0 γ 0 dγ dγ Ths condton can be smplfed and nterpreted n an ntutve way that n our context requres the mprovement n the government's barganng poston from commttng to a tarff cap to exceed the loss arsng from the reducton n barganng surplus due to the constrant and subsequent use of the NTB. Drazen and Lmão (2008) show ths for dfferent polces and note that ths condton need not always hold and must be checked n each polcy settng. Thus n the Appendx we provde a numercal smulaton showng ths condton holds n our model f the cap s not too strngent relatve to the unconstraned value. Very strngent caps make the suffcent condton less lkely to hold because they destroy too much surplus, whch cannot be offset by the government s ncreased share n t. Fgure 1 llustrates the role of barganng power n the government s decson to commt. * Consder a value ofγ (0, γ ) so that pont U s the soluton under no commtment. The slope of the Pareto fronter reflects the rate at whch government s payoffs can be traded for those of the lobby. The steeper t s, the more costly a gven ncrease n V s and thus t becomes more effcent (n terms of the Nash product beng maxmzed) to ncrease the government share of the payoffs. To see ths consder the ray from the orgn that mantans the same rato of payoffs at C as at U. The dotted Nash so-value lne at C has the same slope as U U at U (snce the Nash product s log lnear) ndcatng an unchanged margnal rate of substtuton of payoffs. But the Pareto fronter under commtment s steeper due to the neffcency of the NTB and the fact that the government has dmnshng utlty for lobby goods. Thus the equlbrum under commtment entals a value of G above that n C, whch at the crtcal value γ * s equal to the unconstraned. For lower γ the government payoff s hgher when t commts to a constrant. 17

19 G U c ( γ h ) um G cm G γ G C U c U ( γ ) u U ( γ ) C g 0 v 0 V Fgure 1 In sum, the model predcts that a government s less lkely to commt to a cap n products where ts barganng power s hgher. If the government prefers to commt to an exogenous cap for some γ * (0, γ ) then t also prefers to do so f t can optmally choose the cap level. Moreover, when the barganng power s suffcently hgh t chooses not to commt even f t can optmally choose the cap level snce, n the lmt, when t has all the barganng power any commtment constrans the overall surplus t receves. When we examne the data, we wll consder the role of barganng power n determnng f the government constrans the tarff n a gven good. However, from that data alone t s not always obvous that the constrant wll bnd. In fact, n the absence of any other cost from commttng, the government could always choose to commt to a tarff constrant for any barganng power f that constrant was not bndng. Thus barganng power may not affect the choce of goods a government bnds, partcularly f the government has some nfluence on the cap level. But n that case we can explore a related predcton: a postve relatonshp between barganng power and the cap to unconstraned tarff dfference. As noted before, when γ<γ *, equaton (13) holds strctly for an exogenous cap and thus t must also hold f that cap s optmally chosen. Therefore t c* (γ) - t u (γ) s negatve for at least some γ<γ * but t must be non-negatve for suffcently large γ. 18