Discriminatory tariffs and international negotiations

Transcription

1 ournal of International Eonomis 61 (2003) loate/ eonbase Disriminator tariffs and international negotiations osh Ederington *, Phillip MCalman a, b a Department of Eonomis, Universit of Miami, P.O. Box , Coral Gables, FL 33124, USA b Universit of California, Santa Cruz, CA, USA Reeived 25 une 2002; reeived in revised form 11 ul 2002; aepted 16 September 2002 Abstrat Reent researh has highlighted the effiien of the MFN priniple within the GATT/ WTO struture. This paper analzes the exeption made to MFN within Artile III that allows disriminator punishment for deviations from the agreement. We argue that, in the absene of ollusion, the MFN exeption redues the severit of punishment and thus lowers the level of ooperation that an be ahieved b the agreement. However, disriminator punishment ma still be benefiial as we show that it redues the problems assoiated with the potential for renegotiation during the punishment phase. Finall, we argue that our results are also appliable to the question of whether to use trade poli santions as a means of enforing agreements overing domesti poliies Elsevier B.V. All rights reserved. Kewords: Trade agreements; Dispute settlement; Non-disrimination EL lassifiation: F02; F13; F15 1. Introdution Sine its ineption, the General Agreement on Tariffs and Trade (GATT) has been extremel suessful in liberalizing trade barriers. One of the pillars of the GATT sstem is the most-favored-nation (MFN) priniple that requires eah member of GATT to offer market aess on non-disriminator terms to all other *Corresponding author. Tel.: ; fax: address: ederington@miami.edu (. Ederington) / 03/ $ see front matter 2003 Elsevier B.V. All rights reserved. doi: / S (03)

2 398. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) members of GATT. Reent papers (e.g., Bagwell and Staiger, 1999a, 2002; MCalman, 2002; issimos and Vines, 1999) have made lear that this priniple of non-disrimination plas a ke role in the ahievement of effiient multilateral trade agreements. Given these effiien properties, it seems benefiial to analze the ases under whih exeptions to MFN are allowed in GATT. There are three main ases where exeptions to MFN are made within the GATT framework. The first involves safeguard ations under Artile I (the esape lause), although departures from MFN under Artile I are onl allowed on onsultation with the Safeguards Committee. The seond involves the signing of preferential ustoms unions or free trade agreements under Artile IV. The third main exeption involves retaliator ations under the dispute settlement mehanism of Artile III. If GATT authorizes a retaliator ation under Artile III, suh ation need not be taken on MFN basis (see akson, 1989). It is this last exeption that is the subjet of this paper. The exeption to MFN under Artile III allows ountries to disriminate in the dispute settlement phase of the agreement. Thus, under Artile III, in the event that a ountr deviates from the agreement, member ountries are allowed to punish onl the deviating ountr while maintaining ooperation among themselves. The question raised b this paper is whether allowing suh disriminator punishment an assist in enforing greater ooperation within the agreement. We adopt the view that enforement issues are entral to the design and understanding of international agreements. Speifiall we argue that, in the absene of an external enforement mehanism, international agreements are onl viable if member ountries view ontinued ooperation to be in their own self-interest. The GATT/ WTO dispute settlement mehanism ma pla an important role in oordinating multilateral enforement efforts (see, for example, Maggi, 1999). However, it has no independent abilit to punish violations on its own. Thus, international agreements like GATT are limited in their abilit to enfore ooperation b the severit of punishment whih member ountries an redibl threaten against potential heaters. In Setion 3 of the paper we argue that disriminator punishment atuall redues the severit of punishment that an be 1 threatened to potential heaters, and thus leads to a less ooperative agreement. Thus, ountries would be better off if the ould redibl ommit to nondisriminator punishment (smmetri punishment of all members of the agreement) sine suh punishment is a stronger deterrent to deviations. Suh a result asts doubts on the effiien of the Artile III exeption to MFN. However, in Setions 4 and 5 of the paper we provide two arguments that 1 This result is driven b a tariff omplementarit effet whereb a redution in the tariff against one ountr results in a redution in the optimal tariff against another ountr. Thus, our result is omparable to the reent theoretial literature on how the formation of trading blos affets the abilit to sustain multilateral agreements where tariff omplementarit also plas an important role (e.g., Bagwell and Staiger, 1999b; Bond et al., 2001).

3 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) suggest disriminator punishment ma be optimal within a self-enforing agreement. In Setion 4 we argue that disriminator punishment is optimal when member ountries are allowed to ollude in setting tariffs against the heating ountr during the punishment phase. In Setion 5, we argue that, even in the absene of ollusion, disriminator punishments ma be optimal when the self-enforing agreement is suseptible to renegotiation. Indeed, as mentioned b Ludema (2001), the GATT dispute settlement mehanism atuall enourages renegotiation after disputes have been triggered. As has been well established in the game theor literature, requiring punishments to be renegotiation-proof an redue the severit of the punishment that is threatened to potential deviators, reating problems for the enforement of agreements. In Setion 5, we show, using a standard definition of renegotiation-proof punishments suggested b Ludema (2001), that allowing disriminator punishments an redue the problems assoiated with the potential for renegotiation during the punishment phase of the agreement. Thus, we argue that the Artile III exeption to MFN is justified in agreements where punishments are sensitive to renegotiation. Finall, in Setion 6, we argue that our results are also relevant to the question of whether to use trade poli as a means of enforing international agreements overing domesti poliies (e.g., environmental poli). Speifiall, we argue that one of the inherent advantages of trade poli over domesti poli as an instrument of enforement is its disriminator potential (i.e., trade santions an be used to punish onl those members who heat on the agreement, maintaining ooperation among the rest). The results of our paper suggest that disriminator trade poli santions an be benefiial as a means of enforing international domesti poli agreements that are sensitive to the problems of renegotiation. 2. The model The analsis is onduted within a three-ountr (, and ), three-good (x, and z) ompeting exporters model of trade analagous to that emploed b Bagwell and Staiger (1999b). Demand for good i in ountr is given b a i i i i demand funtion D ( p ) where p is the loal prie of the good and D9( p ),0. i i Prodution of good i in ountr is defined b a suppl funtion Q ( p ) where i 2 Q9( p ). 0. These demand and suppl funtions are assumed to be defined so that eah ountr is the natural exporter of two of the goods and the unique importer of the third. For notational simpliit we let the lower ase letter denote the unique import good for eah ountr (e.g., ountr imports and exports x and z). 2 Note that demand and suppl are solel funtions of own prie effets and thus we are abstrating from ross-prie and inome effets. This framework an be rationalized in general equilibrium terms b assuming the presene of a numeraire good with quasi-linear utilit and with prodution assumed to be a funtion of labor where labor suppl is infinitel elasti at a unitar wage.

4 400. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) We assume that ountries hoose speifi import tariffs on their import goods whih reate a wedge between loal pries in the importing and exporting ountries. For example, define p as the loal prie (paid b onsumers) of good in the importing ountr. In that ase, the loal prie of in ountries and (the exporters of ) will be given (provided trade taxes are not prohibitive) b p ( p, t ) 5 p 2 t and p ( p, t ) 5 p 2 t where tij represents the tariff plaed b ountr i on the exports of ountr j. Pries of the other two goods are 3 defined aordingl. i i i Define exess demand for good i in ountr b M ( p ) where M9( p ), 0. i From the market learing ondition (i.e., o M 5 0) and the demand and suppl equations, one an derive the loal prie of eah good as a funtion of tariff poli. Impliitl differentiating the market learing ondition, one an derive that an inrease in tariffs against either of the two exporting ountries will raise the loal prie of the good in the importing ountr. For example, with respet to good : 9 9 p M p M ]] 5 ]].0 and ]] 5 ]].0 (1) t 9 t x9 O M O M From (1) and the definition of loal pries, one an derive that an inrease in the tariff against one exporting ountr will derease the loal prie in that ountr, but will inrease the loal prie in the alternate export ountr (i.e., the model exhibits trade diversion). For example, with respet to good : p 2 (M 1 M ) p M ]] 5 ]]]]],0 and ]] 5 ]].0 (2) t 9 t 9 O M O M Finall, governments are assumed to maximize the sum of onsumer surplus and produer surplus, and tax revenue. Thus welfare for ountr on its import 4 and export goods, respetivel, is given b: ` ˆp ˆ ˆ ˆ ˆ W (t, t ) ;E D(p )dp 1E Q(p )dp 2 t M 2 t M x ˆp 0 ˆp x ` W (t, t ) ;E D(p )dp 1E Q(p )dp z ˆ ˆ ˆ ˆ (3) ˆp x 0 ˆp x ` ˆ ˆ ˆ ˆ W (t, t ) ;E D(p )dp 1E Q(p )dp z 0 ˆp 3 x x x x x x x z z z Given p, p ( p, t ) 5 p2 t and p ( p, t ) 5 p2 t. Likewise, given p, p ( p, t ) 5 z z z z p2 t and p ( p, t ) 5 p2 t. 4 Welfare for ountries and are defined smmetriall.

5 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) In the absene of an international agreement, eah ountr sets trade taxes on its importing good to maximize national welfare, taking the poli hoies of its trading partner as given. Taking the derivatives of (3) with respet to t and t, and solving out the first-order onditions ields: 9 D M ]]] 9 9 M 1 M 9 D M ]]] 9 9 M 1 M t 5 e 1 (t 2 e ) t 5 e 1 (t 2 e ) (4) i i i9 j where e 5 M /M is inversel related to the trade elastiit of good i from ountr. The above two first-order onditions speif unilaterall optimal tariffs D D b ountr on imports from eah ountr (t and t ) and reflet familiar terms-of-trade inentives in whih ountries impose a positive tariff on imports so as to lower the world prie of the good. Thus, from (4), the higher the trade elastiit of the exporting ountr, the lower the optimal tariff on imports from that ountr. However, when importing from multiple soures the terms of trade argument needs to be augmented. Note from (4) that the optimal tariff against eah ountr is also a funtion of the tariff that is set against the alternate trading partner. This is due to the trade diversion exhibited in the model. For example, onsider an inrease b ountr of its tariff against ountr (t ). Due to trade diversion this leads to inreased demand for the output of ountr and hene raises the loal prie of the good from ountr. To the extent that ountr is setting low tariffs against ountr (i.e., t, e ) this inrease in the prie of the import good represents a terms of trade loss for ountr and thus redues the inentive to raise tariffs against ountr. Note that these terms-of-trade losses will be larger the more inelasti is trade with the alternate ountr (i.e., holding t onstant an inrease in e redues the inentive to raise t ). In the following Lemma we establish that, as a result of the above inentives, tariffs will be omplementar. That is, an inrease in the tariff against ountr will lead to an inrease in the optimal tariff against ountr : Lemma 1. Assuming demand and suppl funtions are linear, a ountr s tariffs on imports of the same good from different trading partners will be omplemen- D tar (i.e., t / t. 0). D Proof. Taking the derivative of t (defined b (4)) with respet to t, one 5 derives that: 5 Note that (5) is derived b either assuming that demand and suppl funtions are linear (so that seond derivatives an be suppressed) or b evaluating the derivative in the area of the Nash equilibrium (where t 5 e ).

6 402. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) F S D G M 1 M M 1 M e M e t e e M 1 2]] 1]]] ]] ]] 5]] 1 1 2]] ]]] t t t t t S D (5) Simplifing using (1) and (2) and assuming that demand and suppl funtions are linear so that we an ignore seond derivatives, one derives that: j M (2O M 1 M ) t ]] 5]]]]]]]]]]]]]]] t M? M (M 1 M ) FO M 1 (M 1 M ) 1]]]] 9 9 G (M 1 M ) (6) It is diret to derive that, sine both the numerator and denominator of the above expression are unambiguousl positive, that tariffs are omplementar. h The non-ooperative (Nash) equilibrium is the intersetion of the above bestresponse funtions and is given b: N N t 5 e and t 5 e (7) Globall effiient trade poliies will be set to maximize world welfare (W 1 W1 W ), and will serve as the natural goals towards whih ountries strive when the ooperate. Taking the derivative of world welfare with respet to t and t solving out the first-order onditions, we derive that globall optimal poliies are defined b free trade (t 5 t 5 0). The setup of the model reflets the tpial prisoners dilemma, in whih ountries want to ooperate to free trade, but have a unilateral inentive to eret barriers to trade. Unfortunatel, the desire to eret trade barriers does not disappear one an agreement is in plae, and a ritial problem faed b an international agreement is the lak of an external enforement mehanism to ensure that the signatories uphold their obligations. In the absene of external enforement mehanisms, an agreement will onl be viable if it is self-enforing (i.e., member ountries must view ontinued ooperation to be in their own best interest). Dixit (1987), Bagwell and Staiger (1990) and Riezman (1991) show how ountries an support low tariffs within a repeated relationship b onfiguring the tariff agreement so that eah ountr fears that a deision to heat on the agreement would trigger a ostl retaliator episode in the future. In the following setion, we emplo this framework to determine if the disriminator power of tariffs an failitate ooperation.

7 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Self-enforing agreements An international agreement is defined b a set of binding ooperative poliies (denoted t ). However, given the lak of an external enforement mehanism, we require our agreement to be onfigured so that these ooperative trade barriers are self-enforing. To that end, we assume that an deviation from these ooperative poliies triggers a retaliator episode. Thus, we haraterize time periods as being either a period of ooperation, deviation or retaliation. The previous literature on self-enforing international trade agreements (suh as GATT) has foused on trigger strategies in whih retaliation entails reversion, for a fixed number of periods, to the stati Nash equilibrium (e.g., see Bagwell and Staiger, 1990; Riezman, 1991; Hungerford, 1991). The advantage of this approah is it ensures that the equilibrium of the repeated game is subgame perfet. Thus, as a benhmark ase, we fous on redible (subgame perfet) agreements in whih ountries threaten to revert to the stati Nash equilibrium in the event of a violation. For tratabilit we will assume that our model is smmetri whih allows us to ondut the analsis with respet to a single deviating ountr, with smmetri 6 onditions holding for the other two ountries. During periods of ooperation, ountries set a low, ommon trade barrier (i.e., tij 5 t for all ountries). Given the smmetri nature of the model, ommon ooperative trade barriers impl that ountries split the gains from ooperation equall. The level of welfare in a period of ooperation for ountr i is defined b: C x z W ; W (t 5 t 5 t ) 1 W (t 5 t 5 t ) 1 W (t 5 t 1 t ). (8) i i i i N However, provided that t, t, eah ountr will have an inentive to deviate from this ommon ooperative poli. In this paper, we onsider the ase where a 7 ountr deviates against both trading partners. When deviating, a ountr will impose unilaterall optimal trade taxes (defined b (7)) and the level of welfare in a period of deviation (where ountr has deviated on the agreement) for ountr i is given b: D x N z W ; W (t 5 t 5 t ) 1 W (t 5 t 5 t ) 1 W (t 5 t 5 t ). (9) i i i i One a ountr deviates from the agreement, it triggers retaliation b the 6 Smmetr an be imposed b assuming that ountries have idential demand and suppl funtions with respet to their respetive import goods, left-hand side export goods and right-hand side export goods. 7 This an be justified b assuming that a violation (raising of a tariff against either of the two exporting ountries) triggers multilateral retaliation b both ountries (see Maggi, 1999 for a disussion of the issues relating to multilateral retaliation within international trade agreements). Given that an deviation b the importing ountr will trigger multilateral retaliation, a ountr will alwas heat on both trading partners.

8 404. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) remaining ountries. If this retaliation is non-disriminator, then retaliation implies an abandonment of the entire agreement and a smmetri reversion to the Nash equilibrium on the part of all ountries. Thus, with non-disriminator N punishment, ountries will pla Nash tariffs (t given b (7)) during a period of retaliation. Welfare for ountr i in a retaliator period with nondisriminator punishment (denoted b NP) is given b: NP x N N z N W ; W (t 5 t 5 t ) 1 W (t 5 t 5 t ) 1 W (t 5 t 5 t ). i i i i (10) Alternativel, retaliation an be disriminator, in whih ase non-heating members of the agreement raise their tariffs against the heating member while keeping tariffs low against eah other. For example, assume ountr has heated on the agreement. Then, as a means of punishment, ountries and will onl revert to unilaterall optimal tariffs on trade with ountr. That is, during a period of retaliation, taxes set on goods from ountr are unilaterall optimal D taxes for ountries and (denoted t and defined b (4)), onditional on the fat that ountries and are ontinuing to ooperate among themselves (i.e., t 5 t 5 t )). We denote the lower trade barrier between ountries and b t sine there is no requirement that ountries and must neessaril set the same ooperative tariffs, t as in the multilateral agreement. The onl requirement N 8 we plae is that t, t. Therefore, provided that ountr has deviated from the agreement in a past period, welfare for ountr i in a period of retaliation with disriminator punishment (denoted b DP) is given b: W ; W (t 5 t 5 t ) 1 W (t 5 t (t ), t 5 t ) DP x N D i i i z D i 1 W (t 5 t (t ), t 5 t ). (11) An optimal international agreement results in the ountries jointl hoosing ooperative poliies (t ) to maximize the ooperative level of welfare subjet to the onstraint that no ountr has an inentive to defet from the agreement. The t that satisfies this onstrained maximization will be the smallest t that satisfies the self-enforement onstraint and will be referred to as the most-ooperative tariff (t ˆ ). Satisfing the self-enforement onstraint entails balaning the urrent gains to deviating from the agreement against the future longterm losses to retaliation. We assume that, in the ase of a violation of the agreement, ountries revert to retaliator tariffs for T periods followed b a reversion to most- 8 Intuitivel, we are assuming that the non-deviating ountries maintain some degree of ooperation N (i.e., low tariffs) among themselves. If t 5 t, then one would simpl have the ase of nondisriminator punishment.

9 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) ooperative tariffs (t ˆ ) thereafter, and refer to this sequene of periods as the punishment path. The self-enforement onstraint is satisfied provided that the disounted value of C ooperating in the agreement (1/(1 2 d )W ) is greater than the disounted value of deviating from the agreement and then triggering the punishment path. In the ase of non-disriminator punishment, the self-enforement onstraint for ountr is given b: NP 1 C D d NP C g(t ) ;]] W 2FW 1 ]] s(1 2 r)w 1 rw dg$ 0 (12) 1 2 d 1 2 d T where r 5 d represents the relative weight plaed on the ooperative versus non-ooperative welfare in the punishment path (and is, of ourse, a funtion of the duration of punishment, T ). In the ase of disriminator punishment, the self-enforement onstraint for ountr is given b: F DP 1 C D d DP C g(t ) ;]] W 2 W 1 ]] s(1 2 r)w 1 rw d $ 0. (13) 1 2 d 1 2 d If the threat of retaliation is not suffiient to support free trade as a selfenforing equilibrium (e.g., when ountries disount the future heavil) then the self-enforement onstraint will bind. In that ase, the question beomes whih form of punishment (i.e., disriminator or non-disriminator) an support a more ooperative agreement. One aspet of self-enforing agreements is that the degree of ooperation that an be sustained is tied to the degree of punishment that an be threatened in the ase of deviation. Speifiall, more stringent punishment lowers the gain to deviating from the agreement and thus allows the agreement to support a more ooperative outome. This is refleted in Lemma 2. Lemma 2. Given that the self-enforement onstraint binds, a more stringent punishment (i.e., a lower level of welfare for the deviating ountr in a period of retaliation) an support a lower ooperative tariff. Proof. In Appendix A. The above Lemma is refleted in Fig. 1. Rearranging (12) one derives that a ooperative tariff is supported within the non-disriminator agreement provided NP C that average welfare along the punishment path ((1 2 r)w 1 rw ) is less than C D C f(t ) ; W 2 [(1 2 d )/d](w 2 W ). If this ondition holds, then the gain to remaining in the agreement outweighs the potential gain to deviating on the agreement and then reverting to the punishment path. As is shown in Fig. 1, the range of enforeable tariffs, for whih no ountr would have an inentive to NP N deviate from the agreement, is given b the interval [t, t ]. Sine global welfare G

10 406. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Fig. 1. Determining the most-ooperative tariff. is monotoniall inreasing as the ooperative tariff is lowered towards free trade, NP t represents the most-ooperative tariff (t ˆ ). Note from Fig. 1 that, holding t onstant, as average welfare on the NP C punishment path falls (i.e., the line denoted (1 2 r)w 1 rw shifts down) the agreement will be able to support lower ooperative tariffs (i.e., ˆ t will derease). Lemma 2 implies that the optimal punishment sheme is the one that an threaten the most stringent punishment (i.e., the lowest level of welfare along the punishment path for the deviating ountr). In the analsis that follows, we determine whih punishment path (disriminator or non-disriminator) is assoiated with the most stringent punishment for the deviating ountr. Note, from (3), that foreign tariffs onl affet home ountr welfare in their effet on export pries, and that home welfare is monotoniall inreasing in the prie of its export good. Thus, the effet of disriminator punishment on the deviating ountr depends on the effet of disrimination on that ountr s export pries. For example, take the ase that ountr has deviated from the agreement. Then the effet of disriminator punishment an be redued to the effet of lowered tariffs between the non-deviating ountries (i.e., t, t ) on the export

11 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) pries of the deviating ountr (i.e., p, p ). For example, from the definition of p, one an derive that: z dp p p t ]] 5]] 1 ]]?]]. (14) dt t t t From (14) one sees that disriminator punishment has two related effets on the welfare of the deviating ountr. First, sine it allows the non-deviating ountries to maintain lower trade barriers against one another, the basi impat of disrimination is to divert trade from the deviating ountr to the non-deviating ountries. Thus, disriminating against the deviating ountr during periods of retaliation will diretl lower the welfare of the deviating ountr b diverting trade awa from that ountr. This an be seen in the first term of (14) where p / t. 0. On the other hand, as disussed in Setion 2, tariffs are omplementar aross ountries (i.e., t / t. 0). When the non-deviating ountries attempt to lower tariffs against one another, the are also lowering their unilaterall optimal tariff against the deviating ountr. Thus, nondisriminator punishment an redibl threaten higher tariffs against deviating ountries than an a disriminator punishment sheme. Aordingl, disriminating against the deviating ountr during periods of retaliation will indiretl raise the welfare of the deviating ountr sine it will fae lower retaliator tariffs. This an be seen in the seond term of (14) where p / t? t / t,0. As we establish in the following Lemma, this seond effet is the stronger of the two and thus: Lemma 3. Assuming that demand and suppl funtions are linear, non-disriminator punishment results in lower welfare for the deviating ountr than NP DP disriminator punishment in a period of retaliation (W, W ). Proof. From (14), (1) and (2) one an derive that lowering the tariff between non-deviating ountries will raise the welfare of the deviating ountr if: S 9 t M ]]. ]]] t 9 9 M 1 M D (15) Substituting (5), whih provides an expression for t / t into (15) we derive that p / t, 0 if: F 9 9 e ]] G p i9 M?O M. 0. (16) Sine M, 0 for an good i and an ountr and e / p. 0 for linear demand and suppl funtions, the above ondition is satisfied and a redution in

12 408. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) the tariff between the nondeviating ountries will raise the export pries and hene the welfare of the deviating ountr. h The result that tariff omplementarit dominates trade diversion is due to the i effet that trade diversion has on the trade elastiities (i.e., e ). Intuitivel, a redution in the tariff between non-deviating ountries (e.g., t ) leads to inreased trade between the non-deviating ountries and redued trade between the deviating ountr and the non-deviating ountries. Fousing on the ase where demand and suppl funtions are suffiientl linear, the redued trade between the deviating ountr and non-deviating ountr leads to a lower optimal tariff against the 9 non-deviating ountr (i.e., e / t. 0). Likewise, the inreased trade between the non-deviating ountries leads to a higher optimal tariff between the non deviating ountries (i.e., e / t, 0) and thus, b (4), redues the inentives to raise the tariff against the deviating ountr. Sine both of these effets strengthen the tariff omplementarit effet (b enouraging a lower tariff against the deviating ountr) it dominates the trade diversion effet and thus leads to non-disrimination being a superior punishment strateg. However, if higher-order derivatives are relativel large, then the opposite result is possible. For example, if M 5 (M )/M then e / t 5 0 and, as an be seen from (16), the tariff-omplementarit effet would equal the trade diversion effet and disrimination and nondisrimination would be equivalent. B the same 0 rationale, if the seond derivatives dominate the first derivatives (i.e., M, 9 2 (M )/M, 0) then e / t, 0 and disrimination ould atuall be preferred. Intuitivel, if the exess demand urve is suffiientl onave, then a redution in the tariff between the non-deviating ountries inreases the elastiit of trade between the non-deviating ountries thus weakening the tariff-omplementarit effet so that it no longer dominates the trade diversion effet. While the above qualifiation should be kept in mind and the limitations of the partial equilibrium model are aknowledged, nevertheless similar results are also present in general equilibrium models. For example, Kennan and Riezman (1990) show (in a general-equilibrium model of trade), that the reation of a free-trade agreement will shift the terms-of-trade in favor of and raise the welfare of non-member ountries. The impliation of this analsis is that maintaining ooperation among non-deviating ountries will atuall raise the welfare of the deviating ountr and thus lessen the threat of punishment. Thus, it is the ase that within a standard model of self-enforing agreements, the disriminator power of trade poli ould atuall redue the enforement power of the agreement. NP DP This result is refleted in Fig. 1, where, for a given T and t, sine W, W, average welfare along the non-disriminator punishment path is less than average welfare with disriminator punishment. As an be seen in Fig. 1, the stronger 9 B suffiientl linear we mean that the first derivatives are large relative to the seond derivatives.

13 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) punishment of non-disriminator tariffs implies that the non-disriminator agreement an support lower ooperative tariffs. Thus, from Lemma 2 and Lemma 3 we derive our first Proposition: Proposition 1. Within a self-enforing international agreement, holding T onstant, non-disriminator punishment will be preferred to disriminator punishment when demand and suppl funtions are linear. The above proposition implies that disriminating against heating members in the punishment phase does not inrease the severit of the punishment threatened to potential deviators. Surprisingl, disriminator punishment atuall lowers the punishment threat of the agreement, and thus non-disrimination would be the preferred punishment sheme. This result is a funtion of the fat that, when the non-deviating ountries lower their bilateral trade barriers during periods of retaliation, the are providing a net terms-of-trade benefit to the deviating 10 ountr. 4. Collusive punishment In the previous setion it was shown that allowing disriminator tariffs during the punishment phase atuall redues the punishment threat of the agreement. However, this result was derived under the assumption that member ountries ( and ) do not attempt to exploit their joint market-power against the deviating ountr () in the punishment phase. In this setion we show that the opposite is possible if we allow member ountries to ollude against deviators b setting tariffs jointl (i.e., if ountries and set tariffs against ountr whih maximize their joint-welfare during the punishment phase). To see this, we first derive optimal (ollusive) tariffs against the deviating DC ountr under onditions of both disriminator (t ) and non-disriminator NC (t ) punishment. Under non-disriminator punishment, ountries and will set non-disriminator tariffs against both ountr and eah other (i.e., t 5 NC t 5 t ) to maximize their joint-welfare (i.e., W (t,t ) 1 W (t,t )). Taking the derivatives of this joint-welfare funtion with respet to the non- NC disriminator trade barrier (t ) and solving out the first-order ondition one derives that: 10 The above analsis does assume that retaliation entails reversion to an interior Nash equilibrium. In the ase that ountries have aess to export-setor poliies and autark is used as the Nash threatpoint, then disriminator and non-disriminator punishments are equivalent. Thus, in this setion we an onl state that non-disrimination is weakl preferred to disrimination as punishment sheme.

14 410. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) NC NC M ]]] 9 9 M 1 M t 5 t 5 NC (17) Note that the non-disriminator ollusive tariff (t ) given in (17) is lower N than the nondisriminator Nash tariff (t ) given in (7). Intuitivel, with ollusion, MFN punishment tariffs are low beause member ountries are unwilling to raise tariffs against their olluding partners. Sine these low tariffs derease the punishment threat of the agreement, it should be apparent that when punishment is onstrained to be non-disriminator, the agreement should disourage ollusion in the punishment phase. With disriminator punishment, tariffs set against ountr are optimal DC ollusive taxes for ountries and (denoted t ) onditional on the fat that ountries and are ontinuing to ooperate among themselves (i.e., t 5 t 5 t )). One again, optimal ollusive taxes are set to maximize the joint welfare of ountries and. Taking the derivative of the joint-welfare funtion with respet to t and solving out the first-order ondition ields: 9 DC M ]]] 9 9 M 1 M t 5 e 1 t. (18) Note that for an t. 0 the disriminator ollusive tariff (t ) is greater than D the unilaterall optimal tariff (t ) given in (4). Intuitivel, with ollusive taxes member ountries an internalize the trade diversion effets of their tariffs. Thus, disriminator tariffs will be high sine member ountries will be seeking to divert trade to their olluding partners, leading to lower welfare for deviating ountries in the punishment phase of the agreement. As we show in the following proposition, this ollusive effet will outweigh the tariff omplementarit effet for small deviations from non-disriminator Nash punishment (i.e., if t is suffiientl N lose to t ): DC Proposition 2. For linear demand and suppl funtions, there exists a disriminator punishment path, involving ollusion among the non-deviating ountries, whih is preferred to non-disriminator punishment in a self-enforing agreement. Proof. In Appendix B. 5. Renegotiation-proof agreements In line with previous literature, the analsis of the previous setion foused on trigger strategies in whih violations are automatiall punished for a fixed number of periods. However, one of the noteworth features of the dispute settlement

15 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) proedures (DSP) within the GATT sstem is the opportunities it provides for onsultation and negotiation. Indeed the reent Urugua round expliitl provides a forum in whih member ountries an disuss both pereived deviations and potential santions. As noted b Ludema (2001), if there is a mehanism through whih ountries an resolve their differenes after deviation, then the potential severit of punishment ma be limited. Speifiall, one ountries have reverted to the punishment phase, the will have an inentive to renegotiate so as to ahieve some additional gains from trade. Unfortunatel, suh renegotiation redues the threat of being punished and ma undermine the previous ooperative agreement. The question raised in this setion is whether the disriminator power of tariffs provides a means of minimizing this potential problem. To address this question, we fous our attention on international agreements that are renegotiation-proof. There are man definitions of renegotiation proofness as well as punishment paths that satisf ertain definitions of renegotiation proofness. In this paper, we adopt the punishment path proposed b Ludema (2001) in whih the time spent retaliating (defined in the previous setion and denoted b T )is limited to ensure that the punishment is redible and is not suseptible to renegotiation. The advantage of adopting Ludema s proposed punishment sheme is that it is designed to mimi elements of the GATT dispute settlement proedure Non-disriminator punishment In this setion, we haraterize a renegotiation-proof agreement that emplos non-disriminator retaliation. As before, a ommon ooperative tariff (t ) is negotiated in the initial phase to maximize ooperative welfare, subjet to the provision that it is self-enforing. In the event of a violation of the agreement, a punishment path is triggered. However, in this setion, we are onerned with the possibilit that, one ountries have entered the punishment path, the will attempt to renegotiate the original agreement (and thereb avoid the threatened santions). To this end we follow Ludema (2001) b requiring that our punishment path has the following properties: (1) subgame perfetion: no ountr has an inentive to unilaterall deviate; (2) reiproit: eah ountr reeives equal welfare along the path; (3) onsisten: the same path is speified for an deviation; and (4) unanimit: the path is Pareto effiient. As shown b Ludema (2001), the above properties an be satisfied b a N punishment path in whih ountries pla smmetri Nash tariffs (i.e., t defined b (7)) for T periods, followed b a return to the ooperative equilibrium (i.e, the lowest self-enforing ooperative tariff). It should be apparent that the nondisriminator punishment path speified above satisfies the first three properties. In addition, the fourth propert an be satisfied b simpl restriting the time spent

16 412. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) retaliating along the punishment path (T ). Speifiall, b adjusting T, the agreement an be onfigured so that the proposed punishment-path is pareto effiient (and thus will satisf the unanimit requirement and be immune to renegotiation). Note that for the agreement to be redible the proposed ooperative tariff (t ˆ (r)) must be self-enforing. As a result, (12) must be satisfied to ensure that no 11 ountr would deviate from the ooperative tariff. From (12) one derives ˆ t as a funtion of T (or equivalentl, r [ [0,1]). Note that the longer the punishment phase remains at the non-ooperative equilibrium (i.e., a lower r), the more severe the punishment and the lower the t the agreement an support (i.e., t / r. 0). Thus, T must be suffiientl large so as to satisf the self-enforement onstraint and support the proposed most-ooperative tariff. The basi result of introduing renegotiation into the analsis is that the length of time spent retaliating is now endogenous. Speifiall, a renegotiation-proof agreement will hoose T (or r) to maximize global welfare along the punishment path: NP GNP GC max v 5 (1 2 r)w 1 rw (19) r GNP where W represents global welfare in periods of non-disriminator punish- GNP NP NP NP GC ment (i.e., W 5 W 1 W 1 W ) and W represents global welfare in GC C C C periods of ooperation (i.e., W 5 W 1 W 1 W ). The solution to the above C onstrained maximization is demonstrated b Fig. 2. The paoff, W (t ˆ ), is the C ooperative level of welfare from a given most-ooperative tariff. Thus, W is an inreasing funtion of T (sine greater punishment implies that a lower ooperative tariff an be supported). The average welfare along the punishment path (v) isa C NP weighted average of W and welfare in the punishment phase (W ) and is equal NP to W at T 5 0 and at the limit as T approahes infinit. Thus, as an be seen in Fig. 2, b restriting the time spent in the punishment phase to T*, the agreement maximizes average welfare along the punishment path. Solving (19) we find that the optimal r* is given b the following first-order ondition:) GNP GC ((1 2 r)w ) (rw ) ]]]]] 1]]] 50 (20) r r Of ourse reduing the time spent in the punishment phase limits the amount of ooperation that the agreement an ahieve, but is neessar in order to make the required santions renegotiation-proof. Likewise, given r*, the most-ooperative tariff (t ˆ ( r)) is given b the self-enforement onstraint (12). Assuming that the 11 There also exists a ondition that the ountries would be willing to enter the punishment phase (as opposed to reverting indefinitel to the Nash equilibrium) but this ondition is satisfied triviall.

17 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Fig. 2. Renegotiation proof punishment. self-enforement onstraint binds (and thus (12) holds with equalit), b impliitl differentiating (12) one derives that at ˆ t (r) F C D NP C s W W (1 2 r)w (rw ) ]] 5 (1 2 d )]] 1 d ]]]] 1 ]]] (21) r r r r Sine non-disrimination results in smmetri welfare along the punishment C G NP GNP path (i.e., 3W 5 W C and 3W 5 W ), substituting (20) into (21) one derives that at the ˆ t ( r) for non-disriminator punishment: d G C D W W ]] 5 (1 2 d )]]. (22) r r As an be seen in Fig. 3, Eq. (22) impliitl defines the most-ooperative tariff

18 414. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Fig. 3. Determining the most-ooperative tariff in a renegotiation proof agreement. NP 12 for non-disriminator punishment (t ) as a funtion of d and r. Note that as D ountries plae higher weight on the future (i.e., d 1), the line (1 2 d )dw shifts upward and the agreement an ahieve a lower most-ooperative tariff (i.e., NP t 0). The question we investigate in the following setion is whether allowing disriminator punishment will assist in ahieving an even lower most-ooperative tariff. 12 C D Fig. 3 is drawn to reflet the fat that W / r 5 0. (1 2 d ) W / r at t 5 0 and that C D N W / r, (1 2 d ) W / r at t 5 t for d [ (0,1) thus ensuring the existene of an equilibrium. C 2 Assuming well-behaved welfare funtions (satisfing the seond-order onditions that W / (t ˆ ), and ˆ t / r, 0) this equilibrium will be unique. If there are multiple equilibrium, we assume that ountries oordinate to hoose the lowest renegotiation-proof tariff and the analsis of this setion goes through.

19 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Disriminator punishment In this setion, we speif a punishment path that satisfies Ludema s properties for a renegotiation proof agreement with the exeption that we modif the reiproit ondition to aommodate the possibilit of disriminator punishment. In Ludema (2001), reiproit required that ountries reeive equal welfare from the punishment path, and was intended to mimi the GATT s emphasis on restoring a balane of onessions in negotiations. With disriminator punishment we maintain a balane of onessions sine ountries ontinue to set a ommon ooperative tariff (t ) in periods of ooperation, and deviators and non-deviators ontinue to pla Nash tariffs against one another in periods of retaliation. We adopt the same punishment path as before with the exeption that, in a period of retaliation, the non-deviating ountries maintain a lower tariff against eah other (t ) than the Nash tariffs the pla against the deviator. As we establish in the following Lemma, allowing non-deviating ountries to maintain lower tariffs against eah other (t ) in periods of retaliation means that the deviator experienes lower welfare along the punishment path than do the nondeviators: Lemma 4. A small derease in t from non-disriminator punishment results in higher welfare for the non-deviating ountries relative to the deviating ountries. Proof. In Appendix C We argue in this setion that this relaxation of the reiproit requirement an assist in solving the problem of renegotiation-proofness and thereb lead to a more ooperative agreement. Speifiall, we argue that beause the non-deviating ountries are reeiving higher welfare in periods of retaliation, disriminator punishment an sustain punishment for a longer period of time, and thus an support a lower most-ooperative tariff. As before, to satisf the unanimit ondition, an agreement with disriminator punishment must maximize global welfare along the punishment path: DP GDP GC max v 5 (1 2 r)w 1 rw (23) r GDP where W represents global welfare during a period of disriminator retaliation GDP DP DP DP (i.e., W 5 W 1 W 1 W ). In this setion we maintain the assumption that ountr is the deviating ountr, while ountries and are the non-deviating ountries. The most-ooperative tariff (t ˆ ) set in periods of ooperation is determined b the self-enforement ondition for disriminator punishment. Impliitl differentiating the self-enforement onstraint for disriminator punishment (13), one derives that at ˆ t (r):

20 416. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) F C D DP C s W W (1 2 r)w (rw ) ]] 5 (1 2 d )]] 1 d ]]]] 1 ]]]. (24) r r r r Eq. (24) impliitl defines the most-ooperative tariff for disriminator DP punishment (t ) as a funtion of d and r. Note, from Fig. 3, that if disriminator DP punishment does assist in ahieving a lower most-ooperative tariff (i.e., t, NP DP C t ), then it must be the ase that t is in the area where W / r. (12 D d ) W / r. Thus, we an derive Lemma 5: Lemma 5. Within a renegotiation-proof agreement, disriminator punishment assists in ahieving a lower most-ooperative trade barrier if: DP GDP 1 W 1 W t GDP DP ] W 2 W 1 (1 2 r) F]] 2 ]]]G].0. (25) 3 t 3 t r Proof. From (24), if DP C C D [(1 2 r)w ] [rw ] W W ]]]] 1 ]]].0 then ]]. (1 2 d )]] r r r r DP DP NP at t. As an be seen from Fig. 3, this implies that t, t. It is diret to derive that: s d s d F G d DP C C DP C DP (1 2 r)w rw W W d ]]]] 1]]] 5W 2 W 1 r]] 1 (1 2 r) ]] r r r r G (26) Likewise, from (23), the first-order ondition for the optimal r* for disriminator punishment redues to: GC GDP GC GDP W W W 2 W 1 r]] 1 (1 2 r) ]] 5 0 (27) r r Sine disrimination results in smmetri welfare onl during the ooperative GC C phase of the agreement (i.e., W 5 3W ), one an substitute (27) into (26) to DP C DP NP derive that d [ (1 2 r)w / r 1 rw / r]. 0 and thus t, t if (25) is satisfied. h Aording to Lemma 5, if (25) is satisfied, then disriminator punishment is benefiial in minimizing the problems of renegotiation and ahieving a lower most-ooperative tariff. Note that the sole differene between disriminator and non-disriminator punishment is that nondeviating ountries pla Nash tariffs N against eah other with non-disrimination (i.e., t 5 t 5 t ) while maintaining some degree of ooperation with disriminator punishment (i.e., t 5 t 5 t ).

21 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Assume that the tariff set between non-deviating ountries (t ) is given exogenousl, and is independent of the time spent retaliating r. Under this assumption, welfare during periods of retaliation is independent of r and (25) redues to the ondition that: 1 GDP DP ]W. W 3 (28) Using (28), Lemma 5 states that disriminator punishment will be benefiial in supporting a lower most-ooperative tariff if welfare for the deviating ountr is less than average welfare along the punishment path. However, we have alread N established in Lemma 4 that this is the ase for t lose to t. Thus, we an state Proposition 3: Proposition 3. A disriminator punishment path exists whih will be preferred to nondisriminator punishment in a renegotiation-proof agreement. Proposition 3 is based on the fat that a small movement towards disriminator punishment will benefit the non-deviating ountries more than the deviating ountr. Even though average welfare along the punishment path ma be the same with both disriminator and nondisriminator punishment (sine the higher welfare in periods of retaliation with disriminator punishment is balaned b a longer amount of time spent retaliating), the deviating ountr will reeive less than average welfare and thus more severe punishment with disriminator punishment. However, not all disriminator shemes are superior to non-disrimination. It is N possible for ertain values of t, t, that welfare for the deviating ountr is higher than average global welfare along the punishment path. To show this, we parameterize our model b assuming that demand for eah ount on eah good i i i i is given b D ( p ) 5 a 2 b? p. Similarl, suppl for eah ountr on its import j j j good is given b Q ( p ) 5? p, while the suppl funtions for the export goods i i j are given b Q j( p j) 5 d? p (with d. ). Solving out the parameterized model and plotting welfare for the deviating ountr versus average welfare along the 13 punishment phase, one derives Fig. 4. As an be seen, while small deviations from the Nash equilibrium result in relativel lower welfare for the deviating ountr, it is possible that large deviations from non-disrimination will result in the deviating ountr reeiving higher than average welfare in the punishment 13 Fig. 4 is drawn to orrespond with the parameter values b and d 5 2. If either b or is suffiientl large, or d is suffiientl small, then t, 0 and thus disrimination would be preferred for N the entire range of t [ [0,t ].

22 418. Ederington, P. MCalman / ournal of International Eonomis 61 (2003) Fig. 4. Disrimination vs. non-disrimination with renegotiation. 14 path. Thus, for t, t in Fig. 4, non-disriminator punishment will be preferred in a renegotiation-proof agreement. In Fig. 4, t divides t values into a region where disrimination would be preferred and a region where non-disrimination would be preferred. The fration N N t/t represents the proportion of t [ [0,t ] suh that non-disrimination is N preferred. Solving out for t and t in terms of parameter values, it is diret to N derive that t/t is monotoniall inreasing in d and dereasing in and b (for an positive values of these parameters where d. ). This is due to the fat the tariff-omplementarit effet (given b (6)) is also monotoniall inreasing in d and dereasing in and b. Intuitivel, for ertain parameter values, the tariff omplementarit effet is ver large and the deviating ountr is more likel to 14 This ounter-intuitive result an also be seen, for ertain parameter values, in Kennan and Riezman (1990) where the ountr outside the free-trade agreement atuall reeives higher welfare than the ountries that have signed the free-trade agreement.

23 . Ederington, P. MCalman / ournal of International Eonomis 61 (2003) benefit from a large redution in the tariff between non-deviating ountries. Thus, non-disrimination is more likel to be preferred when omparative advantage (i.e., d 2 ) is large sine a greater omparative advantage implies a larger tariff-omplementarit effet. Likewise, disrimination is more likel to be preferred when the demand urve is steep (i.e., b is large) sine a steep demand urve implies a smaller tariff-omplementarit effet. Another qualifiation to Proposition 3 is the possibilit that t will be a funtion of r. For example, assume that the non-deviating ountries simpl maintain the most-ooperative tariff in periods of retaliation (i.e., t 5 ˆ t ( r)). In this ase, disriminator punishment will be preferred onl if (25) is satisfied (where t / r. 0). This assumption will atuall have the effet of expanding the parameter values for whih non-disrimination would be preferred, sine (as an be seen in Fig. 4) there is a wide range of t for whih DP GDP W 1 W ]] 2 ]]],0. t 3 t However, given Lemma 4, it is still the ase that a small movement awa from non-disriminator punishment will be benefiial as DP GDP W 1 W ]] 2 ]]].0. t 3 t in the area around the Nash equilibrium (i.e., when t 5 ˆ t (r) is lose to t ). N Conlusion In this paper we are onerned with the question of whether exeptions to the MFN priniple should be made within the dispute phase of international agreements. We show that, within a standard model of a self-enforing agreement, disriminator punishment atuall redues the severit of punishment that an be threatened to potential heaters, and thus leads to a less ooperative agreement. Therefore, ountries would be better off if the ould redibl ommit to non-disriminator punishment (smmetri punishment of all members of the agreement) sine suh punishment is a stronger deterrent to deviations. However, we also derive two ases under whih disriminator punishment is benefiial. First, disriminator punishment is benefiial when member ountries an ollude against heaters in the punishment phase. Seond, disriminator punishment is benefiial sine it minimizes the enforement problems reated b 15 Note that, when t 5 ˆ t (r,d ), the relative benefits of disrimination will be tied to the disount fator as, when ountries disount the future heavil, ˆ t (r,d ) will be lose to the Nash equilibrium and thus disrimination will be preferred.