Preferential Trade Agreements as dynamic farsighted networks James Lake∗ Southern Methodist University September 15, 2012 Abstract In the presence of multilateral negotiations, are Preferential Trade Agreements (PTAs) necessary for, or will they prevent, global free trade? This question is ex- plored using a novel dynamic network theoretic model where countries are farsighted and asymmetric in terms of market size. I develop a new equilibrium concept that endogenizestheorderofnegotiations. WhentwocountrieshaveaPTA,onemember’s formationofanadditionalPTAmaycreateincentivesforitsoriginalpartnertoforma PTA with its new partner. When considering forming an additional PTA, the current member therefore recognizes the potential for erosion of its preferential access in both partnermarkets. Thisfearofpreferenceerosionunderminesitswillingnesstoformthe additionalPTA.Thus,globalfreetradeisattainedviaPTAformationwhencountries fearofpreferenceerosionissu(cid:30)cientlysmall. Looselyspeaking,PTAsarenecessaryfor global free trade when there are two small countries and one large country but PTAs preventglobalfreetradewhentherearetwolargecountriesandonesmallcountry. The model provides insights into the dynamics of recent trade negotiations involving the US and suggests ambiguities inherent in GATT Article XXIV may actually promote global free trade. JEL: C71, F12, F13 Keywords: PreferentialTradeAgreements,preferenceerosion,multilateralism,free trade, networks, farsighted ∗E-mail: [email protected]. I would like to especially thank Pravin Krishna and M. Ali Khan as well as the editor and three anonymous referees. I would also like to thank Kamal Saggi, Emanuel Ornelas, Ray Riezman,JonFaust,HulyaEraslan,HanOzsoylev,MattJackson,MartinRichardson,ReneeBowen,Kyle BagwellandSumitJoshiforusefulcommentsanddiscussionaswellasseminarandconferenceparticipants atthe2011SpringMidwestEconomicTradeMeetings,2011PublicEconomicTheoryConference,2011Asia Paci(cid:28)c Trade Seminar, 2011 International Trade, Development and Game Theory Workshop at the 22nd Stony Brook Game Theory Festival, the 26th Congress of the European Economic Association, Southern MethodistUniversity,CollegeofWilliamandMary,UniversityofMemphisandSAISBologna. 1 1 Introduction Inrecentdecades,thenumberofPreferentialTradeAgreements(PTAs)hasexpandedexpo- nentially. AlthoughsanctionedbyGATTArticleXXIV,suchagreementsarediscriminatory by construction and stand in contrast to the central principle of non(cid:21)discrimination artic- 1 ulated in the most favored nation (MFN) principle of GATT Article I. The proliferation of such discriminatory PTAs has stimulated a substantial academic and policy debate on whether PTAs hinder or facilitate greater liberalization, especially given the lack of multi- lateral liberalization since the 1994 Uruguay Round. In the phrasing of Bhagwati (1991, 1993) are PTAs (cid:16)building blocs(cid:17) or (cid:16)stumbling blocs(cid:17) on the path to global free trade? Various approaches in the academic literature have investigated the interplay between PTAs and non(cid:21)discriminatory multilateral liberalization. In essence, this is a dynamic question concerning the evolution of trade agreements over time, yet much of the previous analysis has used static three(cid:21)country models. These models ask whether an arbitrarily chosen pair of countries want to form a bilateral agreement (BA) and, if so, how this a(cid:27)ects the incentives for expansion of the agreement to include the third country, thus 2 achievingglobalfreetrade(see,forinstance,Levy(1997),Krishna(1998),Ornelas(2005b)). 3 Building a dynamic model to analyze the building bloc(cid:21)stumbling bloc issue faces the importantchallengethatadynamicframeworkrequiresspecifyingtheorderofnegotiations. That is, which country is initially left out and how, if at all, will it be included later? Specifyinganarbitraryorderofnegotiations(asinMukunokiandTachi(2006)andAghion et al. (2007)) creates di(cid:30)culties because the equilibrium of sequential move games is often sensitive to the order of negotiations (Ludema (1991), Ray and Vohra (1997), Jackson (2008)) especially since asymmetries in country characteristics are likely to endogenously 4 5 determine the order. To alleviate this problem, this paper develops a novel dynamic network model in which 6 countriesarefarsightedandasymmetricintermsofmarketsize. Thenoveltyarisesbecause 1WhileGATTArticleIrequiresanytari(cid:27)reductionsa(cid:27)ordedtoonecountryarea(cid:27)ordedtoall,GATT ArticleXXIVprovidesanescapeclausewherebyPTAmemberscaneliminatetari(cid:27)sbetweenthemselvesas longastheydonotraisetari(cid:27)sorothernontari(cid:27)barriersonothercountries. 2Alternatively, Ornelas (2005a) views multilateral liberalization as tari(cid:27) reductions on non(cid:21)PTA mem- bers. BagwellandStaiger(1999),Freund(2000),Bondetal.(2001)andSaggi(2006)alsotakethisapproach butviewdeterminationoftari(cid:27)sonnon(cid:21)PTAmembersasaselfenforcingoutcomeofarepeatedgame. In contrast,takingthestallednatureofMFNliberalizationasexogenous,Baldwin(1996)showsthepossibility ofaPTA(cid:16)dominoe(cid:27)ect(cid:17) inastaticncountrymodel. 3This two(cid:21)step approach is e(cid:27)ectively static because countries do not consider that the initial BA may createincentivesforformationoffurtherBAs. 4Thein(cid:29)uentialworkofAghionetal.(2007)assumesaleadercountrymakessequentialtakeitorleaveit o(cid:27)erstotwofollowingcountrieswhonevernegotiatebetweenthemselves. Incontrast,MukunokiandTachi (2006)assumeanorderinginwhicheachpairofcountrieshasanopportunitytonegotiate. 5Jackson (2008, p. 372) states that (cid:16)... the ordering of links can have a substantial impact on which networksemerge,anditisnotclearwhatanaturalorderingis.(cid:17) 6SeeGoyalandJoshi(2006)andFurusawaandKonishi(2007)forthe(cid:28)rstapplicationsofnetworktheory to the PTA literature. They use the Jackson and Wolinsky (1996) static network equilibrium concept of pairwise stability which is a myopic best response concept. For alternative network theoretic approaches 2 the model embeds a coalitional simultaneous move game in each period of the dynamic game to endogenously determine the order of negotiations. The new equilibrium concept thatemerges,calledafarsighteddynamicnetworkequilibrium(FDNE),isintuitivelysimilar to a subgame perfect equilibrium. However, rather than requiring a Nash equilibrium in every subgame, an FDNE requires an Equilibrium Binding Agreement, a solution concept introduced by Ray and Vohra (1997) and Diamantoudi (2003), in every subgame. The 7 dynamics arise because countries form agreements sequentially, at most one per period. Countries are farsighted because they anticipate the equilibrium evolution of the global tradenetworkconditionaluponformingandnotformingapotentialagreement,andcompare the two discounted payo(cid:27)s. Because countries are farsighted, the presence of market size asymmetryandanendogenousorderofnegotiationsleadstoarichenvironmentofstrategic interaction and equilibrium dynamics. The dynamic network model developed provides an opportunity to capture the equilib- rium e(cid:27)ect of BAs in a three country setting. Loosely, this e(cid:27)ect translates into the idea of whether BAs are building blocs or stumbling blocs. However, the important work of Saggi andYildiz(2010,2011)emphasizesthatcapturingtheequilibriume(cid:27)ectrequirescomparing the equilibrium in which only the MFN principle exists to the equilibrium in which BAs 8 also exist. In contrast, much of the earlier literature draws implications from formation of a single BA. By comparing the di(cid:27)erent equilibria, Saggi and Yildiz (2011) de(cid:28)ne BAs as (cid:16)strong building blocs(cid:17) if global free trade is only attained in the presence of BAs. In contrast, this paper refers to BAs as (cid:16)strong stumbling blocs(cid:17) if global free trade is only 9 10 attained in their absence. The farsighted and dynamic aspects of the FDNE shed new light on the strong building bloc(cid:21)strong stumbling bloc question with the notion of preference erosion central to the analysis. Suppose two countries have the sole BA. These countries are (cid:16)insiders(cid:17) and the other country is the (cid:16)outsider(cid:17). If an insider forms an additional BA, it becomes the (cid:16)hub(cid:17) while the other countries become (cid:16)spokes(cid:17). Now the hub has sole preferential access in each spoke market. However, when contemplating such a move, the insider recognizes the see Zhang et al. (2009) for a farsighted static framework and Zhang et al. (2011) for implementation in a symmetricsequentialmovesetting. 7DespitetherecentdramaticaccelerationinPTAformation,individualtradeagreementsstillformslowly overtime. Forexample,NAFTAwasnotimplementeduntil1994despitenegotiationsbeginningasearlyas 1986, andtwotothreeyearsbetweencommencementofnegotiationsandimplementationofanagreement iscommon. 8Thisapproachwas(cid:28)rstadoptedbyRiezman(1999)andAghionetal.(2007). However,unlikeAghion etal.(2007)whouseanextensiveformgamewherealeadermakessequentialtakeitorleaveito(cid:27)erstotwo followercountries,SaggiandYildiz(2010,2011)allowthefollowercountriestoformanagreementbetween themselves. UnlikeRiezman(1999),SaggiandYildiz(2010,2011)solvethemodelanalytically. 9ThenewterminologyofSaggiandYildiz(2011)re(cid:29)ectsthevariedusageofthebuildingbloc(cid:21)stumbling bloc terminology (e.g. Bhagwati and Panagariya (1996), Saggi (2006), Aghion et al. (2007) and Furusawa andKonishi(2007)). 10WhenglobalfreetradeisonlyattainedintheabsenceofBAs,SaggiandYildiz(2011)refertoBAsas weak stumbling blocs if some BAs form but they refer to BAs as strong stumbling blocs if no BAs form. Thispapermakesnosuchdistinction,referringtoBAsasstrongstumblingblocsinbothsituations. 3 spokes may form their own subsequent BA, which erodes the value of preferential access in each spoke market. When the discount factor is su(cid:30)ciently large, the e(cid:27)ect of preference erosion dominates the immediate hub bene(cid:28)ts and the insiders (cid:28)nd it Pareto dominant to remain insiders. Thus, the fear of preference erosion prevents BA formation leading 11 to global free trade. Because sequential BA formation creates rents on the equilibrium path, sequential BA formation arises in equilibrium even when countries have the option to formMFNagreements. Moreover, thepreferenceerosionlogicappliesregardlessofwhether MFN agreements, by themselves, lead to global free trade. Thus, BAs can be either strong building blocs or strong stumbling blocs. Marketsizeasymmetryandthediscountfactordrivethefearofpreferenceerosion. Even whencountriesaresymmetric,asu(cid:30)cientlylargediscountfactormeansthefearofpreference erosionpreventsadditionalBAformationandunderminesglobalfreetrade. Initially,greater asymmetry between the insiders and the outsider increases insider rents by increasing the valueofpreferentialaccessprotectedbyinsiders. Thismakespreferenceerosionmorecostly. Inturn,sincepreferenceerosionarisesinthefuture,attainmentofglobalfreetraderequires greater insider impatience. Thus, the extent to which BAs lead to global free trade falls. In terms of the equilibrium when only MFN agreements exist, the largest country blocks the move to global free trade when it is su(cid:30)ciently large relative to the medium sized country. This has two implications for the strong building bloc(cid:21)strong stumbling bloc issue. First, looselyspeaking,BAsarestrongbuildingblocswhenthereisone(cid:16)larger(cid:17) andtwo(cid:16)smaller(cid:17) countries. When only MFN agreements exist, the largest country blocks a direct move to global free trade because the medium country is su(cid:30)ciently small. But, unless the discount factor is very high, the relatively small insider rents protected by the largest and medium countries mean the fear of preference erosion does not prevent BA formation leading to 12 global free trade. Second, loosely speaking, BAs are strong stumbling blocs when there is one (cid:16)smaller(cid:17) and two (cid:16)larger(cid:17) countries. When only MFN agreements exist, the largest country does not block a direct move to global free trade because the medium country is su(cid:30)cientlylarge. But,unlessthediscountfactorisverylow,thelargeinsiderrentsprotected by the largest and medium countries means the fear of preference erosion prevents further BA formation. While greater asymmetry initially makes preference erosion more costly, the fear of preference erosion eventually disappears for some countries. When su(cid:30)ciently large, the largest country no longer forms the (cid:28)nal spoke(cid:21)spoke BA with the smallest country. Thus, the medium country no longer fears preference erosion and, thus, no longer hesitates in becoming the hub. In turn, the largest country exercises far less restraint in becoming the 11BecauseleastdevelopedcountrieshaveprivilegedaccesstotheEUmarketthroughtheEverythingBut Arms agreement, recent work in the policy literature has investigated the cost of preference erosion they faceiftheEUgrantslargescaleMFNtari(cid:27)reductions(seeHoekman(2006),LawrenceandRosito(2006), LimªoandOlarreaga(2006)). 12Thedependenceonthediscountfactorwhichunderliesthe(cid:16)looselyspeaking(cid:17) quali(cid:28)cation. 4 hub. Hence, regardless of whether global free trade is attained when only MFN agreements exist, the role of BAs becomes much more constructive. The key implication is a non monotonic relationship between market size asymmetry and the role of BAs: greater levels of asymmetry initially decrease (increase) the extent to which BAs are strong building (stumbling)blocsbutcrossingthethresholdlevelofasymmetryprovidesadramaticreversal. WhilethequestionaskedinthispaperisthesameasAghionetal.(2007)andSaggiand Yildiz (2010, 2011), the answers are di(cid:27)erent. In addition to developing a dynamic model that endogenizes the order of negotiations, the results di(cid:27)er from Aghion et al. (2007) because BAs can be strong stumbling blocs even if the aggregate world payo(cid:27) is maximized 13 underglobalfreetrade. RelativetoSaggiandYildiz(2010),themostimportantdi(cid:27)erence is that BAs can be strong stumbling blocs. The presence of dynamics and farsightedness underlie this di(cid:27)erence. Even when all countries would willingly move to global free trade if MFN agreements were the only form of liberalization, a su(cid:30)ciently large discount factor meansfearsofpreferenceerosiondominatetemporaryhubrents. Inthiscase,thepresenceof BAs prevent global free trade. However, this dynamic farsighted logic of preference erosion is nonexistent for Saggi and Yildiz (2010). There, an insider does not contemplate that the spokes may subsequently form their own BA if it becomes the hub. Since the spoke(cid:21)spoke BA underlies the fear of preference erosion, Saggi and Yildiz (2010) (cid:28)nd that BAs cannot be strong stumbling blocs. An obvious corollary is that the rationale behind BAs being strong building blocs, i.e. necessary for global free trade, is di(cid:27)erent to Saggi and Yildiz (2010, 2011). There, the discrimination faced by an outsider when insiders form a BA can induce the outsider’s participation in an MFN agreement leading directly to global free trade. Here, in contrast, insider incentives drive the strong building bloc result. When the fear of preference erosion is su(cid:30)ciently small, BAs lead to global free trade because an insider willingly forms a BA with the outsider to become the hub. Moreover, for Saggi and Yildiz (2010, footnote 24), the outsider’s discrimination fear is general enough that analogous results hold under the caseofonelargerandtwosmallercountriesinadditiontothecasetheypresentwhichisone smaller and two larger countries. However, in this paper, these two cases deliver di(cid:27)erent results in terms of the strong building bloc(cid:21)strong stumbling bloc dichotomy because the value of insider rents, and thus the fear of preference erosion, depend on the joint size of insider markets. The dynamic role of preference erosion underlying the results of this paper is identi(cid:28)ed 14 byMukunokiandTachi(2006). TheirmainresultshowsaconstructiveroleforBAs: when insiders have a BA, sequential BA formation can yield global free trade when a Customs 13More speci(cid:28)cally, given government payo(cid:27) functions, BAs can be strong stumbling blocs if the world payo(cid:27)ismaximizedunderglobalfreetrade. Themaintextconsiders(cid:28)rmpro(cid:28)tsasthegovernmentpayo(cid:27) functionbutAppendixDshowsthemainresultsarerobusttousingnationalwelfareinstead. 14Indeed, Proposition 1 closely resembles the equilibrium characterization described by Mukunoki and Tachi(2006). 5 Union (CU) would not. Thus, they do not focus on the destructive impact that preference erosion can have on BA formation which is the focus of this paper. Moreover, since their focus is the relative merits of BAs versus CUs, they do not consider MFN agreements and thus they do not consider the question of this paper: the strong building bloc(cid:21)strong stumbling bloc question. Additionally, they do not endogenize the order of negotiations or consider country asymmetries which are important features of this paper. In a related strand of the literature, Krueger (1999) describes how the MFN principle emerged in the 1800s and early 1900s as a mechanism to protect the value of preexisting trade concessions. Indeed, Schwartz and Sykes (1997) argue this is the main bene(cid:28)t of the MFN principle. This may appear to contrast with the result that, despite the presence of the MFN principle, preference erosion allows a destructive role for BAs as strong stumbling blocs. However, this result merely emphasizes a key point of Bagwell and Staiger (2005): whiletheMFNprinciplecanhelpprotectpreexistingMFNconcessions,itcannot,ingeneral, protect preexisting discriminatory concessions (i.e those won via PTAs). Insummary,thispaperdevelopsadynamicfarsightednetworkmodelthatendogenously determines the order in which countries negotiate. The dynamic and farsighted behavior of countriesimplythat,unlikeearlierwork,BAscanbestrongstumblingblocs,thuspreventing global free trade, even if the world payo(cid:27) is maximized under global free trade. Moreover, asymmetry and an endogenous order of negotiations lead to rich equilibrium dynamics that haveobservableimplicationsforPTAnegotiations. Indeed,theseimplicationsareconsistent with actual PTA negotiations. The model predicts a relationship between the order that negotiations commence and theordertheyconclude: thesmallerinsiderbeginsnegotiationswiththeoutsiderbeforethe larger insider but, in equilibrium, this induces the larger insider to become the hub. Recent BAnegotiationsbetweentheUS,CanadaandColombiaaswellastheUS,CanadaandKorea 15 and,toalesserextent,theUS,AustraliaandKoreasupportthissequenceofevents. While sequential negotiations like these are the most common type of equilibrium that leads to global free trade, the model does predict that NAFTA style negotiations, where a single BA expands into a three country PTA, can occur. The model also suggests that preference erosion underlies why BAs do not form. While the reason behind BA non(cid:21)formation is inherently unobservable, spoke(cid:21)spoke BAs do not su(cid:27)er from the fear of preference erosion that insider(cid:21)outsider BAs su(cid:27)er. Indeed, the observable implication that spoke(cid:21)spoke BAs should have a higher probability of formation than insider(cid:21)outsider BAs receives empirical 16 support from Chen and Joshi (2010). While the theoretical literature abstracts away from the ambiguities inherent in GATT ArticleXXIV,theseambiguitiesaregenerallyseenasproblematic. Theambiguitieslieinthe 15TheexacttimingoftheseeventsisdiscussedinSection5.3. 16Speci(cid:28)cally,they(cid:28)ndthattheconditionalprobabilityofaspoke(cid:21)spokeBAexceedsthatofaninsider(cid:21) outsiderBAbyafactoroffour. 6 realitythatGATTArticleXXIVonlyrequiresPTAmembersremovetari(cid:27)son(cid:16)substantially alltrade(cid:17) andallowsremovaltobephasedoutovertime. However,themodelsuggeststhese ambiguities may actually promote achievement of global free trade. Taking advantage of these ambiguities increases hub bene(cid:28)ts. Indeed, they may rise su(cid:30)ciently that the fear of preference erosion no longer prevents the insider from becoming the hub. In this sense, the ambiguities are productive in promoting global free trade. The rest of the paper proceeds as follows. Section 2 develops the network terminol- ogy and equilibrium concepts. Section 3 presents the underlying trade model. Section 4 begins by analyzing the equilibrium when symmetric countries choose between BAs and MFN agreements. It then compares the resulting equilibrium with that when only MFN agreements exist, which underlies the strong building bloc(cid:21)strong stumbling bloc question. Sections 5 and 6 perform similar exercises under asymmetric market size, (cid:28)rst assuming all countries fear preference erosion and then relaxing this assumption. Finally, Section 7 concludes. Appendix B collects proofs not given in the text. 2 Network formation games and equilibrium 2.1 Overview Following Saggi and Yildiz (2010, 2011), this paper considers two three(cid:21)player games: a multilateralism game where only MFN agreements exist and a bilateralism game where countries choose between MFN agreements and BAs. However, unlike Saggi and Yildiz (2010, 2011), each game is a farsighted dynamic in(cid:28)nite horizon game where at most one agreement can form per period. Figure 1 depicts the possible networks where the countries are generically labeled i, j and k. An edge between two countries represents a BA. The free trade network can arise because of the three country MFN agreement or three individual BAs. Three features of the model warrant explicit discussion. First, the assumption of one agreement per period is not crucial to the strong building bloc(cid:21)strong stumbling bloc results. Because a direct move to global free trade is always an option, the assumption only rules out a direct move from the empty network to a hub(cid:21) spoke network. BAs can be strong stumbling blocs when the insiders prefer to remain insiders rather than become the hub on the path to free trade. But then, by construction, they prefer a permanent move to the insider(cid:21)outsider network rather than jumping to the hub(cid:21)spoke network from the empty network. Conversely, when BAs are strong building blocs, global free trade arises because the spokes bene(cid:28)t from forming their own BA. But, this does not depend on how the hub(cid:21)spoke network emerged. Despite the non(cid:21)essential nature of the assumption, it increases analytical tractability of the model. Indeed, de(cid:28)ning a period as the time required to complete a PTA is not inconsistent with history in that PTA formation takes a number of years (see footnote 7). 7 Figure 1: Network positions Second, BAs formed in previous periods are assumed binding. Nevertheless, naturally, when checking whether an agreement formed in the current period is an equilibrium, de- viations that involve severing this agreement are considered. Imagining networks as nodes of a game tree, binding agreements mean players cannot move backwards in the game tree. Together with an assumption of Markov behavior, this implies the status quo remains for- 17 ever once global free trade is reached or no BA forms in a given period. Since there are onlythreepossibleBAs, thenetworkremainsunchangedafter, atmost, threeperiods. This property preserves analytical tractability. Ornelas (2008) explains that the implicit or explicit assumption of binding PTAs is very common in the literature (see, for example, Bagwell and Staiger (1997), Bond and Syropoulos (1996), Maggi and Rodriguez-Clare (1998), Bond et al. (2001), Mitra (2002), MukunokiandTachi(2006),Ornelas(2007)andOrnelas(2008))andconsistentwithhistory. Indeed, Ornelas and Liu (2012) note that the the only PTA violating the assumption is the 2005 Venezuelan withdrawal from the Andean Pact. McLaren (2002) provides strong theoretic justi(cid:28)cation for the assumption. In his model, initial expenditure on sunk costs ensureaPTAremainsinplaceeventhough,absentsunkcostexpenditures,countrieswould prefer an MFN agreement. The sunk costs justi(cid:28)cation also has empirical support (e.g. RobertsandTybout(1997),EichengreenandIrwin(1998)andFreundandMcLaren(1999)). Additionally, Baldwin(2008)explainsthatthecostsofseveringBAsfarexceedlostexports 18 because tari(cid:27)s are merely one element of a large and varied set of retaliation instruments. Third,backwardinductioncanbeusedtosolvetheequilibrium. Tobegin,considereach hub(cid:21)spoke network and solve for the Equilibrium Binding Agreement (EBA; explained in Section 2.2.3), given global free trade remains forever once attained. Then, consider each insider(cid:21)outsider network and solve for the EBA given how the network evolves from any hub(cid:21)spoke network. Finally, consider the empty network and solve for the EBA given how the network evolves from any insider(cid:21)outsider network. The sequence of networks on the 17Suppose no link forms in the current period. Markov behavior then rules out the possibility that a playerwouldformalinkinthefollowingperiodgivenrefusaltoformthislinkinthecurrentperiod. 18Retaliation instruments include the whole host of non tari(cid:27) trade barriers as well as development aid, military aid and political support in various international political issues. A breakdown in international cooperation could imply costs via less progress in issues such as climate change, human rights, money launderingorillegaldrugs. 8 equilibrium path is the farsighted dynamic network equilibrium (FDNE). Thus, an FDNE is intuitively similar to a subgame perfect equilibrium. However, it requires an Equilibrium Binding Agreement, rather than a Nash equilibrium, in every subgame. 2.2 Network formation games 2.2.1 Networks and payo(cid:27)s ThesetofplayersisN ={s,m,l};thesmall,mediumandlargecountries. Generically,they aredenotedi,j andk. Thegameisanin(cid:28)nitehorizongamewhereanetworklinkrepresents a trade agreement. Agreements formed in previous periods cannot be severed and, at most, one agreement can form per period. Thus, denoting the trade agreement formed in period t by (cid:96) , the network in existence at the end of period t is g =((cid:96) ,(cid:96) ,...,(cid:96) ). t t 1 2 t Given a network g, the one period payo(cid:27) to player i is π (g). Given a path of networks i (Fgotr0,agnt0y+1c,o.a..l)itiforonmSp⊆erioNd,t(0gon,wgards,,..p.)la(cid:31)yer i(cid:0)’gs(cid:48)c,ogn(cid:48)tinu,a.t..i(cid:1)ondpenaoytoe(cid:27)sitsh(cid:80)at∞t=tht0eβcto−atl0iπtiio(ngtS). t0 t0+1 S t0 t0+1 prefers the former path over the latter. When the context makes clear the network path following from g and g(cid:48) then g (cid:31) g(cid:48) and (g ,g ,...) (cid:31) (cid:0)g(cid:48) ,g(cid:48) ,...(cid:1) are used t0 t0 t0 S t0 t0 t0+1 S t0 t0+1 interchangeably. 2.2.2 Actions, strategies and equilibrium Given a network g, A (g) and a (g) ∈ A (g) denote, respectively, the action space and an i i i (cid:81) action for player i. Similarly, A (g) = A (g) and a (g) ∈ A (g) denote the action S i∈S i S S space and an action for a coalition S ⊆ N. A (g) is the set of announcements i can make i given the network at the beginning of the period is g. When i does not have BAs with j and k in the bilateralism game, i can make the following announcements: i) it wants to form no agreement, denoted φ, ii) it wants to move to free trade, denoted FT, or iii) the country name (j or k) with whom it wants to form a BA but has not yet done so. Table 1 summarizes the action space for each player and network type where (cid:216) and FT denote, respectively, the empty and free trade networks. A proposed agreement forms when all members of the proposed agreement announce in favor. In the multilateralism game, only MFN consistent agreements are allowed meaning any tari(cid:27) concessions must be given to all countries. This requirement rules out BAs. In this paper, subject to members WTO commitment of not raising tari(cid:27)s, the two country jointly 19 optimal MFN tari(cid:27) is the initial common tari(cid:27). Thus, only the three country MFN agreement is considered. Moreover, since the three country joint optimal tari(cid:27) is zero, this 20 agreement is equivalent to a direct move to free trade. Hence, the only possible networks 19Saggi and Yildiz (2010) allow two country MFN agreements. There, the jointly optimal tari(cid:27) di(cid:27)ers fromtheinitialtari(cid:27). 20Seefootnote31 9 Player action space Network A (g) A (g) A (g) i j k (cid:216) {φ,j,k,FT} {φ,i,k,FT} {φ,i,j,FT} (ij) {φ,k,FT} {φ,k,FT} {φ,i,j,FT} (ij,ik) Empty {φ,k} {φ,j} FT Empty Empty Empty Table 1: Action space for the bilateralism game inthemultilateralismgamearetheemptynetwork,withA ((cid:216))={φ,FT}foreachcountry i i, and the free trade network. Given the action space for each network, a Markov strategy for player i is a function σ i that assigns an action a (g) for every network g. Since a strategy pro(cid:28)le σ = (σ ,σ ,σ ) i i j k induces a unique network path (g ,g ,...) from any initial network g , σ induces the t0 t0+1 t0 continuation payo(cid:27) for player i of (cid:80)∞t=t0βt−t0πi(gt). A farsighted dynamic network equi- librium (FDNE) is simply de(cid:28)ned as the equilibrium path of networks that emerges by using backward induction to solve for a strategy pro(cid:28)le σ such that the action pro(cid:28)le a(g) = (a (g),a (g),a (g)) is an EBA for every network g or, equivalently, is an EBA i j k 21 22 in every subgame. 2.2.3 Equilibrium Binding Agreements RayandVohra(1997)introducedtheequilibriumconceptofanEquilibriumBindingAgree- ment (EBA). This paper uses a slight variation introduced by Diamantoudi (2003) and discusses this variation below. AnEBAissimilartoacoalitionproofNashequilibrium(CPNE),recentlyusedbySaggi and Yildiz (2010, 2011) and Saggi et al. (2011). Each concept begins by only considering Pareto optimal action pro(cid:28)les and restricts attention to pro(cid:28)table deviations that are (cid:16)self enforcing(cid:17). A deviation is self enforcing if the deviation results in an action pro(cid:28)le that is itself an equilibrium. Thus, both concepts require coalition deviations be robust to subse- quentdeviationbycoalitionmembers. Hence,bothconceptsare,tosomeextent,farsighted. However, the key di(cid:27)erence is that deviating players in a CPNE assume other players ac- tionsremain(cid:28)xedfollowingtheirdeviationwhiledeviatingplayersinanEBAassumeother players actions adjust in response to their own deviation. CentraltoanEBAistheideaofacoalitionstructure. Acoalitionstructure, denotedP, 21An FDNE can be seen as an extension of the Dutta et al. (2005) dynamic network formation model. Unlikehere,theDuttaetal.(2005)notionofanequilibriumprocessofnetworkformation(EPNF)assumes the pair of players who negotiate in any given period, the (cid:16)active pair(cid:17), is exogenous. However, unlike an FDNE, an EPNF allows each player in the active pair to sever any of its agreements. Nevertheless, unfortunately from the perspective of BAs, an EPNF does not allow players other than the active pair to severexistingagreements. 22Indeed,Bernheimetal.(1987)notonlyintroducecoalitionproofNashequilibrium(CPNE)tothegame theoreticliteraturetheyalsointroduceperfectlyCPNEwhichsolvesforaCPNEineachsubgame. 10