NBER WORKING PAPER SERIESDYNAMIC PRICING IN THE PRESENCE OE ANTIDUMPING POLICY:THEORY AND EVIDENCEBruce A. BlonigenJee-Hyeong ParkWorking Paper 8477http://www.nber.org/papers/w8477NATIONAL BUREAU OE ECONOMIC RESEARCH1050 Massachusetts AvenueCambridge, MA 02138September 2001Preliminary draIt. Comments are very welcome. We thank Robert Feenstra, John Kennan, Scott Taylor, andparticipants oI seminars and conIerence presentations at the University oI CaliIornia-Davis, Dartmouth College,Midwest International Economics Meetings, the 99 International Economics Convention in Seoul, the internationalworkshop"WTOandWorldTradeII"inSeattle,andUniversityoIWisconsinIorhelpIulcommentsandsuggestions,withspecialthankstoMichaelKnetter,E.YoungSongandRobertStaigerIortheirdetailedcomments.Blonigen acknowledges support Irom NSF grant no. 9810706 and a Richard A. Bray Fellowship. Theviews expressed herein are those oI the authors and not necessarily those oI the National Bureau oI EconomicResearch. 2001 by Bruce A. Blonigen and Jee-Hyeong Park.All rights reserved.Short sections oI text, not to exceed twoparagraphs, may be quoted without explicit permission provided that Iull credit, including notice, is given to thesource.Dynamic Pricing in the Presence oI Antidumping Policy: Theory and EvidenceBruce A. Blonigen and Jee-Hyeong ParkNBER Working Paper No. 8477September 2001JEL No. E13, L11$%#%Antidumping (AD) duties are calculated as the diIIerence between the Ioreign Iirm`s product pricein the export market and some deIinition oI 'normal or 'Iair value, oIten the Ioreign Iirm`s productprice in its own market. Additionally, AD laws allow Ior recalculation oI these AD duties over time inwhat are known as administrative reviews.This paper examines Ior the Iirst time the resulting dynamicpricing problem oI a Ioreign Iirm that Iaces such an AD trade protection policy in its export market.When AD duties are certain Ior any dumping that occurs, we obtain the surprising result that dumpingand AD duties should increase over time toward a stationary equilibrium value. Adding uncertaintiesprevalent in AD enIorcement into our analysis changes these conclusions substantially and leads to morerealistic testable implications.Eirms with 0 ,390 expectations that the probability oI AD enIorcementis low, or with expectations that the probability oI a termination/VER (instead oI AD duties) is high, willdecrease their dumping and AD duties over time in the administrative review process once they Iace ADduties.UsingdetaileddataIromU.S.ADinvestigationsIiledIrom1980-1995,weIindevidenceconsistent with these hypotheses stemming Irom our analysis with uncertain AD enIorcement and provideempirical evidence consistent with James Anderson`s domino dumping hypothesis.Bruce A. Blonigen Jee-Hyeong ParkDepartment oI Economics Department oI EconomicsUniversity oI Oregon Wayne State UniversityEugene, OR 97403-1285 Detroit, MI 48202and NBER Tel: 313-577-2693Tel: 541-346-4680 Eax: 313-577-9564Eax: 541-346-1243 jhparkecon.wayne.eduEmail: bruceboregon.uoregon.edu1 1. Introduction Recentresearchonantidumping(AD)protectionpolicieshasrevealedthatthe institutionalstructuresurroundingthedeterminationandadministrationoIADdutiesmakes analysis oI this Iorm oI trade policy quite interesting and complicated.On the surIace, AD duties aresimpleadvaloremtariIIs,intendedtoreIlecttheextenttowhichtheIoreignIirmwas dumpingintheexportdestinationmarket.ThecalculateddegreeoIdumping(ordumping 'margin) is the diIIerence between a 'normal or 'Iair value Ior the product and the price it is charging in the export market, and there is a positive dumping margin any time the export price isbelowthe'normalvalue.TheU.S.DepartmentoICommerce(USDOC)primarilyusesthe price charged by the Ioreign Iirm in its own market as the estimate oI 'normal value in dumping calculations.In addition, aIter an AD duty comes into place there is the potential Ior continuous recalculationoItheADdutyovertimeinwhatarecalledadministrativereviews.These administrative reviews occur at the request oI the Ioreign Iirm or other interested parties and, in Iact, the majority oI U.S. AD duties are reviewed at least once subsequent to a case.The method oI AD duty calculation and administrative review process presents a Ioreign Iirm with a dynamic pricing problem that has not been explored beIore and will be the Iocus oI this paper. Whilepreviousresearchhasnotconsideredthisdynamicpricingproblem,therehave beenimportantexaminationsoIhowthestructureoIADinvestigationsaIIectseconomic outcomes,includingpricedetermination.StaigerandWolak(1994)andKruppandPollard (1996) Iocus on how investigation 'events (i.e., the Iiling oI the petition, public Iindings oI the USDOC,terminations,etc.)aIIectimportandpricingbehaviorbyIoreignIirms.Staigerand WolakIocusmainlyonimportvolumeandpricechangesduringU.S.ADinvestigationsIiled Irom1980-1986.They Iind that changes in prices and import volumes during the investigation dependonwhetherthedomesticindustryhasIiledthecasetomerelyharasstheIoreignIirm (theycallthesedomesticpetitioners'processIilers)ortogainactualtradeprotection ('outcome Iilers).To the extent that they examine changes in variables aIter the investigation, they do not consider the administrative review process.Krupp and Pollard use detailed data on chemical products subject to U.S. AD investigations to Iocus on changes in import volumes due toinvestigationevents,aswellaseventualADduties.DeVault(1996)examinesdataonthe changes in U.S. AD duties Irom the administrative review process.His empirical work uncovers IactorsthatleadtolowerADdutiesovertime,buthedoesnotintegratetheseobservations 2 withinaIormaltheoreticalIramework.Einally,BlonigenandHaynes(Iorthcoming)examine pass-through oI exchange rates and AD duties to U.S. import prices Ior products involved in U.S. AD investigations.They present a one-period model oI pricing Ior a Iirm subject to an AD duty tomotivatetheirempiricalwork,whichIindsthatbecauseoIthewayinwhichADdutiesare recalculated in U.S. administrative reviews, pass-through oI the AD duty is more than complete andexchangeratepass-throughissubstantiallyalteredIortheirsampleoICanadiansteel products. This paper's contribution is to more seriously consider the dynamic pricing problem Iirms Iace in the presence oI AD investigations and duties.We Iirst set up a Iormal theoretical model andthenusedataonADdutychangesIromU.S.administrativereviewstotestsomeoIthe model'simplications.Inthemodel,aIoreignIirmissolvingadynamicoptimizationproblem, where the current period price (p) aIIects the AD duty in the next period (T1).1SpeciIically, the lower the current period price is relative to the Iair value oI the product (pI), the higher the next period AD duty (T1 pI/p) will be.Given that the Ioreign Iirm has a static incentive to dump its product (the static proIit maximizing price is less than the Iair value), the Iirm needs to balance the loss in the current period Irom setting price higher than its static optimal against the gains in the Iuture discounted payoII Irom having a lower AD duty in the next period. UndercertainenIorcementoIADpolicy,thedynamicprogramminganalysisyieldsa surprising result with respect to the optimal pricing path.When a Ioreign Iirm dumps (setting its pricelowerthantheIairvalue)intheinitialperiod,thentheIirmwilldumpmoreovertime (settingthepricelowerovertime)untilthepricereachesasteadystatevalue.Dumpingduties will increase (or at least never decrease) through the review process!Knowing how an increase in the AD duty aIIects the balance between the loss in the current proIit and the gain in the Iuture discountedproIitassociatedwithcurrentpricingisthekeytounderstandingthisresult.In particular,whentheADdutyincreases,therateoIcurrent-proIitlossassociatedwithpricing higherthanthestaticoptimalpriceincreases,implyinghighercostsoIreducingthedumping marginIorahighercurrentADduty. Thus, once a Iirm dumps in the initial period Iree oI any AD duty, the Iirm will Iace a steeper trade-oII between the current proIit loss and the Iuture gain

1Notethatthis(one-period)laginadjustingtheADdutyiscrucialinraisingthe"dynamic"pricingissue.Thisis becauseinstantaneousadjustmentintheADdutywillsimplysettheconsumerpriceasfixedatthe"fairvalue", regardless of foreign firms' pricing (below the fair value), and, thus, eliminates the incentive to dump. As discussed in Section 2, it is reasonable to assume that there exist significant lags in adjusting AD duties, given the practices of AD policy.3 with an AD duty being imposed in the next period, inducing the Iirm to choose a price that is at leastaslowastheinitialprice.AsthesameprocesscontinuesintheIollowingperiods,the IoreignIirmwillloweritspriceovertimetowardastationaryequilibriumvalue,implyinga risingADdutythroughthereviewprocess.ThisprocessmayevenleadtogreaterlevelsoI dumpingthanwhensuchpolicyisabsent,thusADpolicywithlaggedadministrativereview processcanperverselycontributetomoreaggressivedumpingbehaviors.2Eirmswithhigh enoughdiscountIactorswillIollowthispathoIincreasingdumping,whereasIirmswith relatively low discount Iactors will never dump, despite economic conditions that make dumping optimal in absence oI the AD policy.While the benchmark case oI certain AD enIorcement is important Ior establishing basic Ieatures oI our model, it is unlikely that enIorcement oI AD policy is certain.3 In our U.S. data oIADcasesIrom1980through1995,41oItheADcaseswereruledaIIirmative,48 negativeand11wereterminatedbecauseoIapublicly-announcedsettlement(oItenaVER).Inaddition,whilethebenchmarkcasepredictsrisingADdutiesovertimeaItertheinitialAD dutyisimposed,thedatarevealthatthemajorityoIchangesinADdutiesduringthe administrative review process are declines, not increases. ToanalyzedynamicpricingbehaviorwithuncertainADenIorcement,wemodiIythe modelsothatIoreignIirmshaveexanteexpectedprobabilitiesoIthepossibleADcase outcomes: AIIirmative with AD duties, negative, or termination with a settlement such as a VER.WiththisuncertainenIorcement,ouranalysisshowsthattheIoreignIirmmayreducetheAD dutythroughtheadministrativereviewprocess(settinghigherpriceovertimeoncetheIirm needstopaythedumpingduties)whentheprobabilityoIgettingtheinitialADdutyislow enough. We can explain this result as Iollows.As the probability oI getting the initial AD duty becomes smaller, the Ioreign Iirm has an incentive to set its price closer to its static optimum.II

2Section3.2providesanumericalexampleofthiscase.Previouspapers,includingAnderson(1992;1993)and BlonigenandOhno(1998),alsoshowhowthemerepresenceofADpolicycanleadtogreaterlevelsofdumping thanwhenADpolicyisabsent.Incontrasttotheseearlierstudies,wherestrategicincentivesfortheforeignand domesticfirmstouseADpolicyasmeansofachievinganti-competitiveoutcomedrivethiskindofdumping behavior,ouranalysisshowsthatthedynamicstructureofADpolicyitselfmayinducetheforeignfirmsto excessive dumping over time even in the absence of any strategic incentives. 3Oneimportantexception,suggested by Robert Eeenstra, was the Trigger Price Mechanism (TPM) adopted by the U.S.withrespecttosteelproductsinthelate1970`s.UndertheTPM,foreignfirmsautomaticallyfacedduties whenever their price fell below a trigger level, which fits our model of certain enforcement well.The program was unsuccessfulduetotheproblemsofcalculatingtriggerpriceadjustmentsduetoexchangeratemovements (Eichengreen and van den Ven, 1984), thus making it difficult to compare pricing behavior by firms under the TPM with our model`s predictions. 4 it Iinds itselI subject to paying an AD duty despite its low probability, the initial dumping price canbelowerthanthestationaryequilibriumpriceundercertainenIorcement(notethatcertain enIorcement is assumed in the administrative review process).Thus, the Ioreign Iirm will set its pricehigherovertimeuntilthepricereachesthestationaryvalue,implyingreductionoIAD duties through the review process.We also show that when Iirms believe greater dumping will lead to greater beneIits under a terminated/VER outcome (as in Anderson, 1992 and 1993) a high probability oI a terminated/VER agreement will lead to AD duty reductions in the event that the Iirm receives an aIIirmative outcome.4 The latter halI oI the paper tests the hypotheses Irom our model oI dynamic pricing under uncertainADenIorcementthatlinkstheexanteprobabilitiesoIcaseoutcomeswiththe reductioninADdutiesthroughthereviewprocess.Theempiricalanalysisadoptsatwo-step estimationmethod.WeIirstestimatedeterminantsoItheprobabilitiesoIADcaseoutcomes.WethenusetheparameterestimatestoIormexanteoutcomeprobabilitiesandestimatethe eIIect oI these ex ante probabilities on changes in AD duties in the administrative review process aIter the initial AD duty is imposed. Using a sample oI all Iirm-product combinations subject to U.S.aIIirmativeADdecisionsandADdutiesIorcasesIiledbetween1980and1995,the empiricalresultssupportthehypotheses.Consistentwithourtheory,weIindthatalowerex ante aIIirmative probability leads to signiIicantly greater ex post reductions in the AD duty in the administrativereviewprocess.WealsoIindthatahigherexanteprobabilityoITERleadsto signiIicantlygreaterexpostreductioninADduties,whichisconsistentwithAnderson`s (1992;1993) 'domino dumping phenomenon. TherestoIthepaperproceedsasIollows.Thenextsectionprovidessomerelevant detailsoIU.S.ADtradeprotectionandtheadministrativereviewprocess.Section3provides Iormaldynamicpricingmodelsandourmainpropositions.Section4thendescribesour empirical analysis and a Iinal section concludes. 2. Salient Features of U.S. AD Law and Administration. ThissectionprovidesabrieIoverviewoItherelevantdetailsconnectedwithU.S.AD investigationsandadministrativereviews.TheU.S.ADlawsareadministeredbytheUSDOC

4Anderson(1992;1993)showsthatifanagreementafterterminationtakestheformofaVERwheremoreexport licensesaregiventoaforeignfirmwithhighermarketshare,thenforeignfirmswillmoreaggressivelydumpto 5 andU.S.InternationalTradeCommission(USITC),eachwithdistinctrolesintheprocess.When an AD petition is Iiled, the USDOC determines whether the subject product is being sold at'lessthanIairvalueintheUnitedStates.Incontrast,theUSITCdetermineswhetherthe relevant U.S. domestic industry has been materially injured, or is threatened with material injury, by reason oI the imports subject to its investigation.The calculation oI the dumping margin by the USDOC is usually not straightIorward and revolves around how the USDOC measures what should be the 'Iair value oI the product sold in the United States. Selling a product in the United States at less than 'Iair value is the deIinition oI dumping and the diIIerence between the U.S. price and 'Iair value is the dumping margin.In theory, the USDOC deIines 'Iair value as the exporting Iirm`s price Ior the same product in its ownhomemarket.However,iItheIirm`shomemarketsalesaredeemedinadequate,thenthe USDOCmaybase'IairvalueontheexportingIirm`spricesinthirdcountrymarketsorona constructedvalueIortheproductusingmanuIacturingcosts,selling, general and administrative costs,proIitsandpackagingcosts.Thesecalculationsobviouslyinvolvehighlydetailedand conIidentialdataonthetransactionsoItheinvestigatedIirm,whicharerequestedbyUSDOC IromtheinvestigatedIirm.IItheinvestigatedIirmdoesnotcomplysuIIiciently,theUSDOC willturntousingthe'bestinIormationavailable,whichisoIteninIormationsuppliedbythe U.S. Iirms that Iiled the petition. II an aIIirmative preliminary determination is made by both the USDOC and the USITC (whichmaytakeupto160days),thentheimportermustpostacashdeposit,abondorother securityIoreachentryequaltothepreliminarymargindeterminedbytheUSDOC.This requirementstaysineIIectuntileithertheUSDOCortheUSITCmakesanegativeIinal determination,whichmaytakeanother120days.IIanaIIirmativeIinaldeterminationismade byboththeUSITCandUSDOC,thenUSDOCissuesanADordertolevyadutyequaltothe estimated dumping margin on the subject product. When a subject Ioreign product enters the United States, the importer must pay Customs acashdepositequal to the margin times the value oI the subject product.However, these cash depositsdonotnecessarilyrepresenttheIinalamountoIdutiestobeassessedonthesubject imports.Rather,themargindeterminedinUSDOC'sIinalinvestigationisonlyusedasabasis IorestimatingthedutyliabilityoItheimporter.TheactualliabilityoItheimportermaybe compete for market share. 6 determinedinsubsequentyearsbytheUSDOC.BeIore1984,thiswasaccomplishedby automaticyearlyadministrativereviewsbytheUSDOC.However,since1984,suchreviews havebecomevoluntary;thatis, unless an interested party requests a review, the duties assessed arethoseIoundinUSDOC'sIinaldetermination(ormostrecentadministrativereview).An administrativereviewservestwopurposes.Eirst,itadjuststhemarginonsubjectimportsto reIlect changes in the diIIerence between the Ioreign Iirm`s U.S. price and the Iair value so that theimporterpaysanadjustedcashdepositbasedonthereviewedmarginuntilthenext administrativereview.Second,itmakesADdutiesretroactive:IIareviewdeterminesthatthe marginduringthereviewperiodisdiIIerentIromthepreviousmarginusedasabasisIorthe importer's cash deposit, a bill (or reIund) in the amount oI the diIIerence plus interest is assessed (or rebated). 3. Model of Dynamic Pricing in the Presence of AD Policy. This section develops models oI dynamic pricing oI a Ioreign Iirm in the presence oI AD policydescribedintheprecedingsection.WewillIirstanalyzethebenchmarkcasewherethe ADpolicyisenIorcedwithcertaintyandthenanalyzethecaseoIuncertainADenIorcement.While we think the implications oI the benchmark model may be unrealistic Ior the large part oI pricing behavior we observe with respect to U.S. AD cases, it establishes a number oI important Ieatures oI the model that are important Ior understanding the case oI uncertain AD enIorcement. 3.1. Benchmark Model: Certain AD Enforcement.We assume that a Ioreign Iirm may sell its products both in Ioreign and domestic markets andthesemarketsareseparatedIromeachothersothatpricediscriminationispossibleacross the markets. However, when the Ioreign Iirm chooses its export price in the domestic market, p, lower than its price in its own Ioreign market, pI, the Ioreign Iirm Iaces an AD duty, T (1) such thatpTpIinthedomesticmarket.5WhentheIoreignIirmchangesporpI,thisADdutyis subject to change through administrative review process, according to pT pI.

5 Once a product is subject to a AD duty, it is not the foreign exporter but the domestic importer who is supposed to paytheADdutyinpractice,potentiallycreatingaconflictofinterestsamongthemtowardpricingdecision,thus a dynamicgametobeplayedbetweenthem.However,wedonotanalyzethisissueinthefollowinganalysis,by assuming away the existence of the domestic importer who has autonomy in setting consumer price, or alternatively by assuming that the foreign exporter pursues joint profit maximization with the domestic importer.7 AstheinitialdumpingdeterminationandtheIollowingadministrativereviewareIar Irominstantaneoustochangesintheprices,theIoreignIirmIacesadynamicpricingproblem. To analyze this problem, we use an inIinite horizon model in which the consumers' tastes are the same in each period. Eor simplicity, we assume that the change in the AD duty lags one period to changes in the price, 1 if1 i ip / p T =where the subscripts i and i-1 denote period i and period i-1, respectively,withT01(theinitialperiodisIreeIromdumpingduties).Inaddition,weset f0f1 ip p = (~0) Ior all i 1 in determining Ti, because there are compelling reasons and empirical evidencethattheoverwhelmingpartoIthepricechangeaIIectingthedumpingdutiesinthe review process occurs with the domestic (U.S.) price, not the Ioreign price.6 LetcbetheIoreignIirm'sconstantmarginalproductioncostandq(pc)beitsper-period domesticmarketdemandIunctionwithpc(pT)representingtheconsumerprice.7Then,the discounted proIit Irom the domestic market Ior the Ioreign Iirm in period k is given by (1)( )= + + + + 0 i1 i k i k i k i ki) p ( T p q ) c p ( where ) 1 , 0 ( is the discount Iactor, 1 i kf0 1 i k i kp / p ) p ( T + + +=iI f0 1 i kp p 0 < +,1 ) p ( T1 i k i k= + + iI f0 1 i kp p > +,and = + +) p ( T1 i k i kiI0 p1 i k= +,meaningthattheimportswillbeprohibited Irom k1 period on.8

6 As Gallaway et al. (pp. 219-220) discuss, the USDOC has wide discretion to disallow a foreign firm`s prices in its own market when determining 'normal value, either deeming them as 'below cost, or by finding insufficient sales andturningtoaconstructedcostmeasureorpricestoathirdmarket.ThismakesthestrategyofreducingtheAD dutybyloweringthepriceintheforeignmarket tenuous at best, and limited empirical evidence shows that foreign firmsreduceADdutiesprimarilythroughincreasesinthepricetothehomemarket,notdecreasesintheforeign price. 7Wedonotanalyzestrategicinteractionsbetweenfirmsinsettingprices,focusingonhowthestructureofAD policy influences a foreign firm`s dynamic pricing (dumping) behavior in the domestic market. This simple demand functioncanbeconsideredasanindividualfirm`sdemandinamonopolisticallycompetitiveindustry.Onemay pursueanalternativepathtowardmodelingoligopolyfirmsandanalyzingtheirdynamicpricinginteractionsinthe presence of AD policy, but we do not pursue this avenue in this paper. 8 As discussed earlier, the U.S. applies AD duties retroactively so that foreign firms pay estimated AD duties at the beginningoftheperiod,andthenwhenADdutiesarere-estimatedattheendoftheperiod,theypayadditional duties or receive a refund, depending on whether the AD duty increased or decreased over the period.Denoting the refund duty rate for sales during period i by ir , with 5 % 57f0 i i i =if f0 i5 5 and 5 % 57f0 if0 i =if f0 i5 5 > , wecananalyzetheeffectofintroducing'retro-activenessintheadministrativereviewprocessbyanalyzingthe following problem:8 The dynamic pricing problem oI the Ioreign Iirm would involve choosing the sequence oI itsIoreignprices,aswellasthesequenceoIitsdomesticprices.Tocharacterizetheoptimal sequenceIorthedomesticprices,however,wecanIocusontheproblemoImaximizingthe discounted proIit Irom the domestic market by choosing the optimal domestic price sequence Ior a given f0p(optimally chosen in the initial period).This is because only the initial Ioreign price will matter in determining all subsequent AD duties Ior the domestic market.ThereIore, we analyze the Iollowing optimization problem, taking f0p and) , 0 | p1 as given: (2)( ) ..., 2, 1, 0, i ), , 0 | pwith) p ( T p q ) c p ( Supi0 i1 i i i ii p'' (c, f0p].(See appendix for proof) 11 Lemma3impliesthattheIoreignIirm'soptimalpriceinthecurrentperioddoesnot increasewhenthecurrentADdutyincreases.Eigure1ishelpIulinunderstandingthisresult. Eigure1adepictsproIitIunctions((p;T'),(p;T''))withdiIIerentADdutiesoIthecurrent period (T' T''), and Eigure 1b depicts the increasing value Iunction as a Iunction oI the current periodprice(V(p)).BydeIinition,settingpl(f0p/T'p')maximizesthecurrentdiscounted payoIIwhenthecurrentADdutyisT';(l(p');T') V(l(p')) (p; T') V(p) Ior all p (c, f0p|. II the current period AD duty increases to T'', the proIit Iunction shiIts down in such a way that (p; T')/p ~ (p; T'')/p Ior all p due to (A4). This shiIt in proIit Iunction in response to anincreaseintheADdutymakessettingthecurrentperiodpricehigherthanl(p')evenless attractive than beIore. Note that the Ioreign Iirm tries to balance between the current period proIit lossagainsttheIuturediscountedproIitgaininsettingthecurrentpricehigherthanitsstatic optimum.BecauseahighercurrentADdutywillraisethecurrentperiod proIit loss associated with higher prices, it induces the Ioreign Iirm to choose a lower current price than beIore, iI not an equal price. BasedonLemma1-3,wecannowcharacterizethedynamicpricingbehavioroIthe Ioreign Iirm in the presence oI AD policy as Iollows: Proposition 1.When q(p.) satisfies (A1)-(A4), there exist . (0,1] such thati)for > ., the foreign firm will set p

f0p, thus T

1 for i 0, 1, 2, . . ii)for < ., the foreign firm will set 0p < f0p, thus T

>1 with 0p 1p 2p ., thusT

T

T . . (See appendix for proof) Asmentionedearlier,theoptimalpriceissettobalancethecurrentperiodproIitloss against the discounted proIit gain in the next period.ThereIore, we will observe the Ioreign Iirm dumpingitsproductsonlywhentheIoreignIirmvaluesthecurrentperiodproIithighenough relativetoitsIuturediscountedproIit(c).IItheIoreignIirmdumpsintheinitialperiod, then rather surprisingly the Ioreign Iirm will never try to reduce its dumping margin through the administrative review process, according to Proposition 1. ThisisshowninEigure2a,whichdepictspossibleoptimalpolicycorrespondencesIor Iirms with c and those with > c.Eocusing on G(p; c), suppose we begin with no AD 12 duty,thushavingp(thelastperiodprice)being f0p.Then,itscurrentperiodoptimalpriceis given by the policy correspondence, G(p f0p; c), which is represented by the vertical height atpointA.Thispriceisclearlylessthan f0pbecauseitisbelowthe45line.InIact,by translatingittothe45lineatpointB,wecanseeitcorrespondstoP0.Butthentheoptimal policycorrespondenceG(p;c)givesustheoptimalpriceinthenextperiod,whichis representedbypointCandtranslatestopriceoIP1.Thisprocesscontinuesuntilwereachthe steady state price oI pS.Thus, Ior Iirms that do not value Iuture period`s proIits very highly, c,optimalpricesIalland,henceADdutiesriseovertimeuntilreachingasteadystate equilibrium.Incontrast,theoptimalpolicycorrespondenceIunctionIorIirmsthatvaluethe Iuture relatively highly, > c, crosses the 45 line at f0p, indicating that this price is the steady state equilibrium price. The optimal policy correspondence may not behave smoothly as illustrated in Eigure 2b, butwecaneasilyveriIy that the Ioreign Iirm will never reduce its dumping margin through the review process.12The intuition is that iI a Ioreign Iirm dumps in the initial period, this behavior revealsthattheIirmcaresitscurrentperiodproIitlossassociatedwithpricinghigherthanits staticoptimumrelativelymorethanthegainintheIuturediscountedproIitIromavoidinga positiveADdutyinthenextperiod.Then,inthenextperiodwithapositiveADduty,(A4) assuresthatthe'currentproIitlossassociatewithhigherpricinggetssteeperthanbeIore, inducing the Iirm to reduce its price even Iurther or to set the same price.13Thus, the Iact that a Ioreign Iirm dumps in the initial period despite the dynamic costs Irom Iuture AD duties, reveals the Iirm's dynamic preIerence toward dumping.Proposition1generatesaspeciIicempiricalprediction:IoreignIirmswillnevertryto reducetheirADduties through the administrative review process. Even a brieI look at the U.S.

12 Eor the case where 2(p;T)/pT ~ 0 instead of having (A4), it can be shown that G(p) is decreasing in p rather thanincreasinginp,usingasimilarargumentasinLemma3.Thiscanleadtothedynamicpricingsequence through the administrative review process where AD duties under the review will move up and down, but eventually decreasetoalowervaluethantheinitialADduty.Thisistheoreticallyaninterestingphenomenon,butisnot something we generally observe in our data for the U.S. AD cases from 1980 through 1995. 13Because(A4)isacrucialassumptionfor this result, it may be worthwhile to identify the conditions under which (A4)issatisfied,otherthanthecurvaturesofdemandcurvesalreadymentionedinEootnote9.Morespecifically, one may try to relate (A4) with the price elasticity of demand, p(p) (q(p)/p)(p/q(p)). Using p(p)/phas the signsameasthatof|(q(p)/p)(1p)p(2q(p)/p2)],itiseasytoshowthat(A4)willbe satisfied as long as p(p)/p ~ 0 ,3/ p (~1) is close enough to 1. This result is not too surprising because p(p)/p ~0 can be linked to 13 ADadministrativereviewdatacantellusthatthispredictionislargelywrong;approximately 45oItheinitialdumpingdutieshavebeenreducedthroughthereviewprocessduring1980-1995period.Thus,wenextintroduceuncertaintyintheenIorcementoIADpolicyintothe model, which may induce the Ioreign Iirms to reduce their dumping duties through the review. 3.2. Model with Uncertain Enforcement of AD PolicyAsmentionedearlier,manyU.S.ADcasesarenotruledaIIirmativeand,hence,donot leadtoADduties.ThisisdespitetheIactthattheUSDOCIindsdumpinginover95oIthe investigations!AbouthalIthecasesareruled negative, almost exclusively because they do not meet the injury criteria oI the USITC, and a signiIicant amount lead to terminations, suspensions or withdrawn cases based on agreements between domestic and Ioreign Iirms, oIten VERs (e.g., see Prusa, 1992 and Anderson, 1992;1993). In addition, the AD investigation starts only aIter the domestic Iirms or the USDOC Iile an AD petition against potential dumping activities. Thus, it is obvious that there exist uncertainties in the outcomes and enIorcement oI AD policy, especially in the initial AD investigation.Erom the Ioreign Iirm's point oI view, this implies that setting 0p f0p, does not necessarily incur T1 f0p/p0 ~ 1.TomodelthedynamicpricingprobleminthepresenceoIuncertainenIorcement,we classiIydiIIerentcontingenciesoIADenIorcementintotheIollowingthreecategories:cases with Iinal aIIirmative dumping determinations (ADD), cases being ended with negative injury or dumpingdeterminations(NEG)includingthecaseoInoADpetitionIiling,andcasesbeing terminated, suspended, or withdrawn in lieu oI some settlement, such as a VER (TER).Note that Irom a Ioreign exporting Iirm's point oI view, NEG cases will require no adjustment in its pricing behavior,butTERcaseswillrequiresomeadjustmentswhichcaneitherbeIavorableor unIavorable ones depending on the nature oI settlements it can reach. InthepresenceoIuncertaintiesintheenIorcementoIADpolicy,aIoreignIirm's expectationonprobabilitiesoIdiIIerentcontingencieswillplayanimportantroleinitsinitial pricing (dumping) decision and the Iollowing pricing decisions once it is subject to an AD duty.Given that a Ioreign Iirm dumps (pi f0p) in the absence oI any current AD duty (Ti 1) at period i,letPr(ADD),Pr(NEG),andPr(TER)|0,1|denotetheexanteprobabilitiesoIgettingan ADD,NEG,orTERcaseinthenext(i1)period,respectively.BydeIinition,Pr(ADD) anincreasingcost(consumersbeingmoresensitive)forsettinghigherpriceasthepricegetshigher.However, 14 Pr(NEG)Pr(TER)1.Notethattheseareexanteprobabilities,sothat,Iorexample,aIirm havingaverylowvalueIorPr(ADD)mayIinditselItobesubjecttoanADdutyinthenext period,despiteitslowexpectationoIsuchaduty.Asdiscussedlater,thiskindoIpotential "shock" to expectations will generate an interesting pattern in dynamic pricing behaviors oI Iirms subject to AD duties, which will be the basis oI the empirical analysis.Eoranalyticalsimplicity,weintroducetheIollowingassumptionsonthenatureoI uncertaintiessurroundingtheADinvestigation.Eirst,onceanADdutyisimposed,theinitial ADdutyandtheIollowingadministrativereviewswillbeperIectlyenIorced,applyingTi1 f0p/pi.ThisIitswiththeobservationthatnoinjurydeterminationbytheUSITCoccursduring administrativereviews,whichisthemainsourceoIuncertaintyintheinitialADinvestigation.Second,onceaTERcasearisesduetoanagreement among interested parties, we assume such anagreementwillbeperIectlyenIorced,eliminatingpossibilitiesIoranyIutureAD investigation.Einally, we assume that these ex ante probabilities, Pr()s are not Iunctions oI the pricechosenintheabsenceanADdutyoranagreement,eventhoughPr()smaydiIIeracross diIIerent Ioreign Iirms belong to diIIerent industries.14

Given the uncertain enIorcement oI AD policy described above, the discounted expected payoII in the initial period, EV(f0p) will be given by: ( ) ( ) [ ]( ) ( ) [ ]( ) ( ) [ ]( ) ( ) [ ]( ) ( ) [ ]( ) ( ) [ ],) p / p ( p q ) c p ( ) p / p ( p q ) c p ( ) ADD Pr(p q ) c p ( p q ) c p ( ) TER Pr( ) p ( q ) c p () NEG Pr() p / p ( p q ) c p ( ) p / p ( p q ) c p ( ) ADD Pr(p q ) c p ( p q ) c p ( ) TER Pr( ) p ( q ) c p () NEG Pr() p / p ( p q ) c p ( ) p / p ( p q ) c p ( ) ADD Pr(p q ) c p ( p q ) c p ( ) TER Pr( ) p ( q ) c p () p ( EV2 d0f02 d12 d14 e2f02 d02 d032 t12 t14 2 t02 t03 e2e2221 d0f01 d11 d13 e1f01 d01 d021 t11 t13 1 t01 t02 e1e10 d0f00 d10 d12 e0f00 d00 d00 t10 t12 0 t00 t0e0e0 f0 + + + + + + + + + + + + + + + + + + + + + = increasing price elasticity w.r.t. p , is neither a sufficient nor a necessary condition for (A4).14Itisconceivablethatpricing(dumping)behavioraffectstheUSITCinjurydetermination,whichlargely determinesoutcomes.However,givenweassumethateachindividualfirmisasmallplayerinamonopolistically competitiveindustry(theimplicitassumptionbehindthemodelinginthispaper),Pr(ADD)andPr(TER)canbe consideredasexogenousvariablesfromanindividualfirm`spointofviewandmainlydependontheindustry characteristics. 15 where eip represents the price to be set in period i when neither an ADD case nor a TER case has everoccurreduntilperiodi,{ } ... , p , p , pdi2di1di0denotesthepricesequenceIromperiodi1in whichaninitialADdutyisimposedbasedon eip,and{ } ... , p , p , pti2ti1ti0denotestheprice sequence Irom period i1 in which a TER occurs with a speciIic value Ioreip.Because the AD duty is determined by applying Ti1 f0p/pi, no previous periods' pricing but only eip will matter indetermining{ } ... , p , p , pdi2di1di0.GiventhataTERoutcomeoccursinthepresenceoIa possible ADD case, once again only eip (among all pervious periods' pricing) will play a role in determining{ } ... , p , p , pti2ti1ti0.15

II neither an ADD nor a TER outcome has ever occurred, then the Ioreign Iirm will Iace the same problem as its initial problem regarding its choice oI eip Ior i 1, 2, 3, . . Thus, the optimalpricewillbeidenticaltotheinitialoptimalpriceaslongastheIoreignIirmhas continuedtohavenegativecases.Denotethis"initial"optimalchoicetobe E0p. IIaADduty (D0T f0p/E0p )isever-imposedIortheIirsttime,thentheIoreignIirmIaces the same certainty problem described in Section 3.1 Irom that period on.Thus, we can use the same value Iunction as in Section 3.1 to represent the discounted proIit Irom that period on, V(E0p ): the solution to the problemin(3).IIaTERcaseoccursinperiodi1,then E0p willdeterminetheIollowing periods'pricingpath,{ } ... , p , p , pti2ti1ti0,yieldingthediscountedproIitIromthatperiodon, ) p ( q ) c p ( ) p ( q ) c p (ti1ti1ti0ti0 + + + ) p ( q ) c p (ti2ti22.DenotethisdiscountedproIitIora TER case by) p ( VE0T, using the Iact that E0pis a major variable that determines the pricing path Ior a TER case.Then, we can rewrite the discounted expected payoII in the initial period as:

15 In determining the outcome of an agreement for a TER case, it is obvious that the level of ei5relative to f05will playacrucialrole,becauseaVERoftenassignsexportlicensesbasedontheperviousperiodmarketshareof relevantfirmsandthepossiblebargainingforanagreementwilltakeplacewithanADdutybeingconsideredasa possible alternative to an agreement.16 (5) [ ][ ][ ][ ] { } [ ]. ) NEG Pr( 1 / ) p ( V ) TER Pr( ) p ( V ) ADD Pr( ) p ( q ) c p () NEG Pr( ) NEG Pr( ) NEG Pr( 1 ) p ( V ) TER Pr( ) NEG Pr( ) NEG Pr( ) NEG Pr( 1 ) p ( V ) ADD Pr( ) NEG Pr( ) NEG Pr( ) NEG Pr( 1 ) p ( q ) c p ( ) p ( EVE0T E0E0E03 3 2 2 E0T3 3 2 2 E03 3 2 2 E0E0f0 + + = + + + + + + + + + + + + + + = Becausetheexanteprobabilities,Pr()sarenotIunctionsoI E0p, thedynamicpricing problem under uncertain enIorcement is reduced to Iind out E0psatisIying:

(6)[ ] { } ) p ( V ) TER Pr( ) p ( V ) ADD Pr( ) p ( q ) c p ( SupE0T E0E0E0]f0p , c (E0p + + , andthentheoptimalpricingpathundertheadministrativereviewprocesswillIollowG(E0p ) deIinedin(4)oISection3.1.16ThereIore,toanalyzetheeIIectoIintroducinguncertain enIorcement on the Ioreign Iirm`s dynamic pricing path under the administrative review process, weshouldIirstIocusonhowchangesinPr(ADD)andPr(TER)aIIect E0p.Proposition2 provides results on this issue. Proposition 2.i) E0pis increasing in Pr(ADD),ii) E0pis decreasing in Pr(TER) if) p ( VT is strictly decreasing in p (c, f0p], andE0pis increasing in Pr(TER) if) p ( VT is strictly increasing in p (c, f0p].(See appendix for proof) ItiseasytounderstandwhyanincreaseinPr(ADD)willinducetheIoreignIirmto choose a higher (at least not lower) value Ior its initial price (E0p ).17Due to the same reason as

16InthepresenceofalargevalueforPr(TER)togetherwith 5'Tbeingstrictlydecreasinginp,wecannot eliminate the possibility of having E05 c. Because we have characterized G(p) and V(p) on p (c, f0p ] ,we will focus on the case where E05~ c. 17 If V(p) is differentiable at E05 *E05 , and *E05is an interior solution that satisfies the first and the second order conditions for (6), then E05is strictly increasing in Pr(ADD) at E05 *E05in the sense that E05 /Pr(ADD) ~ 0. This canbeeasilyshownbyapplyingtheenvelopetheoremtothef.o.c.for(6).However,V(p)isonlyshowntobe 17 under certain enIorcement, the optimal price (in the absence oI the AD duty) E0pis set to balance thecurrentperiodproIitlossIromsettingthepricehigherthanitsstaticoptimumagainstthe Iuture expected discounted proIit gain Irom reducing the AD duty that the Iirm would incur once itbecomessubjecttoaADduty.AhighervalueIorPr(ADD)raisesthecostIorchoosinga lower value Ior E0p , because it implies a higher probability oI getting a costly (note that V(p) is strictlyincreasinginp)ADduty,reducingtheincentivetoset E0p tobeclosertoitsstatic optimum price, p*. The eIIect oI an increased value Ior Pr(TER) on E0pdepends on whether) p ( VT is strictly decreasingorincreasinginp.II) p ( VTisstrictlyincreasinginp,ahighervalueIorPr(TER) inIlicts a higher cost Ior choosing a lower value Ior E0pbecause a lower value Ior E0pwill work as a disadvantage under a TER case.II a lower value Ior E0pworks as an advantage under a TER case( ) p ( VTisstrictlydecreasinginp),thenahighervalueIorPr(TER)willprovideanextra incentive to choose a lower value Ior E0p .When an agreement under a TER case takes a Iorm oI VERwheremoreexportlicensesaregiventoaIirmwithhighermarketshare,liketheone considered in Anderson (1992:1993), then a lower E0pwill work as an advantage under such an agreement.18IIaTERagreementisreachedthroughabargainingprocessbetweenIoreignand domestic Iirms with the ADD case as a threat point, however, it is possible to have a case where alower E0p worksasadisadvantageIortheIoreignIirmunderaTERcase.ThereIore,which type oI an agreement prevails under a TER case is an empirical question and will be analyzed in the Iollowing empirical section. differentiablealmosteverywhere,implyingthattheremayexistaset(withmeasurezero)whereV(p)failstobe differentiable. Even in that case, E0pis non-decreasing in Pr(ADD). 18Inthiscase,themaximizationin(6)canbeusedtodemonstratearesult(Proposition1)ofJamesE.Anderson (1993) that an increased enforcement () may intensify dumping behaviors for a sufficiently large probability of an AD VER (). Using that Pr(ADD) (1-) and Pr(TER) , (6) can be rewritten as (6`) [ ] { } ) p ( V ) p ( V ) 1 ( ) p ( q ) c p ( SupE0T E0E0E0]f0p , c (E0p + + .When is close to 1, then an increase in will raise Pr(TER) without really affecting Pr(ADD), thus strengthening the incentive to dump (setting a lower E05 ) as 5'E0T is strictly decreasing in E05 .18 GiventheanalysisoIeIIectsoIPr(ADD)andPr(TER)on E0p ,wecannowdiscussthe eIIectoIuncertaintiesintheenIorcementoIADpolicyonthedynamicpricing(ADduty)path underanadministrativereviewprocess.IncontrasttothecertainenIorcementcasewherethe ADdutycanonlyincrease(orstaythesame)throughthereviewprocess,thepresenceoI uncertaintyinenIorcementgeneratesthepossibilityIordiminishingADdutiesthroughthe review process.The Iollowing Corollary 1 demonstrates this.19

Corollary1.Assumethatthereexistsauniquestationaryequilibriumundercertain enforcement, denoted by p$.

i)If p$ f0p, there exists a Pr.(ADD) (0,1] such that the foreign firm sets E0p< f0p for Pr(ADD) .Considersettingpp'with( ) 0 ) p ( T ' p q1= .Given(A3)andp-1 |c, f0p|,( ) 0 ) p ( T ' p q1= 34 impliesthatp'~ f0p.Because( ) ( )f0 1f0f0p g ) p ( T p q ) c p ( + ~( ) ' p g ,pp'with ( ) 0 ) p ( T ' p q1= will not be chosen as p f0p is a strictly dominating choice.Also note that p c will not be chosen.Consider setting p p' c, thus having( ) 0 ) p ( T ' p q ) c ' p (1 .Given (A4) and p-1 |c, f0p|, it is possible to set p p" c with ~0, so that( ) 0 ) p ( T " p q ) c " p (1> .Asp"~p'andgC',thisimpliesthat( ) ( ) " p g ) p ( T " p q ) c " p (1 + ~( )+ ) p ( T ' p q ) c ' p (1 ( ) ' p g .Havingestablishedthatp~cand( ) 0 ) p ( T p q1> insolving(Sg)(p-1) ( ) () [ ] p g ) p ( T p q ) c p ( Max1) p ( p

+ ,wecaneasilyshowthatS(C')C".Eorpsolving(Sg)(p-1),( ) () = + 1 1p / | p g ) p ( T p q ) c p |( ( ) ) p / p p ( ) p / ) p ( T p q ( ) c p (21f0c1 ~ 0 Iorp-1|c, f0p)becausep~ c and( ) 0 p / ) p ( T p qc1< with( ) 0 ) p ( T p q1> .This implies that S(C') C" Ior p-1 |c, f0p). Now,wecannarrowdowntheeconomicallyrelevantdomainandimageoIG(p)as Iollows.Eirst, it is easy to establish that we can Iocus on the optimal pricing path where pi ~ c Iori0,1,2,.w.l.o.g.,iItheIoreignIirmchoosesitsinitialprice 0p intheabsenceoIAD duties(T01).Asshownabove,anypp'cisstrictlydominatedbysomepp"~cin solving(Sg)(p-1) ( ) () [ ] p g ) p ( T p q ) c p ( Max1) p ( p

+ whenp-1 c,andthesameis true when () ) p ( V p g = .This implies that G(p) ~ c Ior p c, thus optimal pricing path will involve prices strictly greater than c.Weshowthattheoptimalpricingpathinvolvespricesequalorlessthan f0pasIollows.Notethatweonlyneedtoconsidertheoptimalpricingpathwithpositivesales.Considerthe casewheretheIirmchooses 0p~ f0pintheinitialperiod.ThiscannotbeapartoIoptimal pricingpathbecausesetting /0p f0p will strictly raise the discounted payoII, (/0p - c) q(/0p) V(/0p) ~ (0p - c) q(0p) V(0p).Because (A4) is a suIIicient condition Ior 2q(pc)/pc (pc-c) (2q(pc)/pc2) 0 (the second order condition Ior static proIit maximization with T1) as long aspc ~ c and q(pc) ~ 0, (A1) and (A4) together imply that q(pc) (pc-c)(q(pc)/pc) 0 Ior pc f0pwithq(pc)~0,whichinturnimplies(0p-c)q(0p)(/0p-c)q(/0p).V(/0p)V(0p) because T1 1 Ior both /0p and 0p.Now, consider that the Iirm chooses p1 ~ f0p with T1 f0p/c.ThiscannotbeapartoIoptimalpricingpathsincesettingp'1 f0pwillstrictlyraisethe discounted payoII, (p1' - c)q(p1'T1) V(p1') ~ (p1 - c)q(p1T1) V(p1).(A4) implies that q(pT1) (p-c)T1(q(pT1)/pc) is a decreasing Iunction in T1 with q(pc) (pc-c) (q(pc)/pc) 0 Ior pc f0pIrom(A1).Thus,q(pT1)(p-c)T1(q(pT1)/pc)0Iorp f0p,implyingthat(p1-c) q(p1T1)(p1'-c)q(p1'T1).V(p1')V(p1)becauseT2 1Iorbothp'1andp1.Eorthenext period on, we can apply the same logic that we use Ior period 1. Proof for Lemma 3Eirst, note that p' in G(p') deIines the current period AD duty by T' f0p/p', since we can Iocus on p' (c, f0p| due to Lemma 2 .By deIinition oI l(p'), 35 (a1)( ) ( ) ( ) ( ) ( ) () | p , c ( p all Iorp V ' T p q c p ) ' p ( l V ' T ) ' p ( l q c ) ' p ( lf0 + + . Consider p'' p', thus T'' ~ T'.Then, there exists 0 such that (l(p') c)q(l(p') T'') V(l(p')) (l(p') c)q(l(p') T') V(l(p')) because q(pc)/pc 0.We can show that (a2) (p c)q(pT') V(p) ~ (p c)q(pT'') V(p) Ior all p (l(p'), f0p| DeIineA(p)|(pc)q(pT')V(p)||(pc)q(pT'')V(p)|.BydeIinitionoI, A(l(p)) 0 and A(p)/p |q(pT') q(pT'')| (p c)| T'(q(pT')/pc) - T''(q(pT'')/pc)| ~ 0 Ior p |l(p'), f0p|.A(p)/p ~ 0 because q(pT) (p c)T(q(pT)/pc) is strictly decreasing in T by (A4): (q(pc)/pc)p (p c)|(q(pc)/pc) (2q(pc)/pc2)pc| 0.Erom (a1), (a2), and (l(p')c)q(l(p')T'')V(l(p'))(l(p')c)q(l(p')T')V(l(p')),wehave(l(p')c) q(l(p')T'')V(l(p'))~(pc)q(pT'') V(p) Ior all p (l(p'), f0p|.Thus, p G(p'') Ior p (l(p'), f0p|, implying h(p'') l(p') Ior p'' p'. Proof for Proposition 1 Eirst, we can easily show that there exists a critical level oI , denoted by c, such that the Ioreign Iirm will dump in the initial period, p0 f0p iI c and will not dump, p0 f0p iI ~ c.TheIoreignIirmwillnotdumpintheinitialperiodiI(f0p;T01)V(f0p)~(p;T01) V(p), or equivalently (p;T0 1) (f0p;T0 1) |V(f0p) V(p)| Ior all p (c, f0p).DeIine a set |0,1)]f0p , c | pMax|(p;T0) (f0p;T0)| |V(f0p) V(p)|} 0}.Then, because thereexistsp'(c, f0p)suchthat(p';T01)(f0p;T01)~0Irom(A1)and(A4).Now, deIine c Sup(), then the Ioreign Iirm will set p0 f0p iI c.II c 1, then the Ioreign Iirm will dump Ior all (0,1). II c 1, the Ioreign will not dump in the initial period iI ~ c. Toprove(ii),wecanusethesameargumentasinLemma3.Withc,theIoreign Iirm will set its initial optimal price, 0p f0p, having (a3)( ) ( ) ( ) ( ) ( ) () | p , c ( p all Iorp V T p q c p p V T p q c pf0 0 0 0 0 0 + + , where T0 1.In the next period, the Iirm will try to choose p1 to maximize its discounted proIit given that T1 f0p/0p ~ 1.Let's consider p1 0p as such a choice, then there exists 0 such that (0p c)q(0p T1) V(0p) (0p c)q(0pT0) V(0p)because q(pc)/pc 0 and T1 ~ T0.We can show that (a4) (p c)q(pT0) V(p) ~ (p c)q(pT1) V(p) Ior all p (0p, f0p| DeIine A(p) |(p c)q(pT0) V(p)| |(p c)q(pT1) V(p) |.By deIinition oI , A(p0) 0 and A(p)/p |q(pT0) q(pT1)| (p c)| T0(q(pT0)/pc) T1(q(pT1)/pc)| ~ 0 Ior p |p0, 36 f0p| by the same reason as in Lemma 3.Erom (a3), (a4), and (0p c) q(0p T1) V(0p) (0p c)q(0p T0) V(0p), we have (0p c) q(0p T1) V(0p) ~ (p c)q(pT1) V(p) Ior all p (0p, f0p|.This implies p1 (0p, f0p|, thus p1 0p.II p1 0p, then we can repeat the same argument as above to show that p2 p1 by replacing T0 and T1 (~T0) with T1 and T2 (~T1). II p1 0p, then we can use the exactly same argument as above to show that p2 p1 0p. Proof for Proposition 2 (i) To prove E0pis increasing in Pr(ADD), we use a similar argument as in Lemma 3.Let's assume that the Ioreign Iirm sets its initial optimal price, E0pIor a given Pr(ADD), having (a5) ( ) ( ) ( ) [ ]( ) () () [ ] |. p , c ( p all Ior) p ( V ) TER Pr( p V ) ADD Pr( p q c p) p ( V ) TER Pr( p V ) ADD Pr( p q c pf0TE0T E0E0E0 + + + + Now consider Pr'(ADD) ~ Pr(ADD), and the Ioreign Iirm tries to set its initial optimal price E0p ' Ior the given Pr'(ADD).There exists ~ 0 such that (E0p c)q(E0p ) Pr'(ADD)V(E0p ) (E0p c)q(E0p ) Pr(ADD)V(E0p ) because Pr'(ADD) ~ Pr(ADD).We can show that (a6)( ) () () [ ]( ) () () [ ] ). p , c ( p all Ior, ) p ( V ) TER Pr( p V ) ADD ( Pr' p q c p) p ( V ) TER Pr( p V ) ADD Pr( p q c pE0TT + + > + + DeIineA(p)|(pc)q(p)Pr(ADD)V(p)||(pc)q(p)Pr'(ADD)V(p)|.By deIinitionoI,A(E0p )0andA(p)|Pr(ADD)Pr'(ADD)|V(p)~0Iorp(c, E0p ) becauseV(p)isstrictlyincreasinginp(c,| pf0.Erom(a5),(a6),and(E0p c)q(E0p ) Pr'(ADD)V(E0p )(E0p c)q(E0p )Pr(ADD)V(E0p ),wehave(E0p c)q(E0p ) Pr'(ADD)V(E0p ) ~ (p c)q(p) Pr'(ADD)V(p) Ior all p (c,E0p ).This implies E0p ' E0pIor all Pr'(ADD) ~ Pr(ADD). (ii)WecanusethesameargumentasintheprooIIorProposition2(i),exceptusingthat thereexists~0suchthat(E0p c)q(E0p )Pr'(TER)V(E0p )(E0p c)q(E0p ) Pr(TER)V(E0p ) and (a6')( ) () () [ ]( ) () () [ ], ) p ( V ) TER ( Pr' p V ) ADD Pr( p q c p) p ( V ) TER Pr( p V ) ADD Pr( p q c pTT + + > + + Ior all p | p , p (f0E0iI) p ( VT is strictly decreasing in p, orIor all p ) p , c (E0iI) p ( VT is strictly increasing in p. 37 Proof for Corollary 1.(i)DeIineasetPr(ADD)|0,1|]f0p , c | pmax Aug|(pc)q(p)Pr(ADD)V(p)| f0p}.Then,because[ ] p ) p ( q ) c p ( max Arg| p , c | p1

f0pIrom(A1).Now,deIinePrc(ADD) Sup(), then the Ioreign Iirm will set E0p f0piI Pr(ADD) Prc(ADD) because E0pis increasing in Pr(ADD) Irom Proposition 2.II an ADD case occurs with E0f0di1 ip / p T =+ ~ 1 at period i, then the Iollowing pricing path{ } ... , p , p , pdi2di1di0 under the administrative review will be determined byG(E0p );notethattheadministrativereviewisenIorcedwithcertainty.BecausepSisthe uniquestationaryequilibriumundercertainenIorcementandG(p)isanon-decreasingupper-hemicontinuouscorrespondence, dijp withji1willincreasetowardpS f0p.Thisinturn implies that dijTwith j i1 will decrease toward 1. (ii)Eor the Ioreign Iirm with pS f0p,[ ] p ) p ( q ) c p ( max Arg| p , c | p1

pS is a necessary condition IortheexistenceoIPrc(ADD)|0,1|suchthattheIoreignIirmsets E0p pSIorPr(ADD) Prc(ADD).Note that = p pE0 when Pr(ADD) 0 (and Pr(TER) 0).Because E0pis increasing inPr(ADD)asshowninProposition2, Sp p willimply S E0p p IorallPr(ADD)|0,1|.DeIine a set Pr(ADD) |0,1|]f0p , c | pmax Aug|(p c)q(p) Pr(ADD)V(p)| pS}.Then, because[ ] p ) p ( q ) c p ( max Arg| p , c | p1

pS.Now,deIinePrc(ADD)Sup(),thentheIoreign Iirmwillset E0p pSiIPr(ADD)Prc(ADD)because E0p isincreasinginPr(ADD)Irom Proposition 2.II an ADD case occurs with S f0E0f0di1 ip / p p / p T > =+ at period i, then the Iollowing pricingpath{ } ... , p , p , pdi2di1di0undertheadministrativereviewwillbedeterminedbyG(E0p ).Because pS (~E0p ) is the unique stationary equilibrium under certain enIorcement and G(p) is an increasing upper-hemi continuous correspondence, dijpwith j i1 will increase toward pS.This implies that dijTwith j i1 will decrease toward S f0p / p . Proof for Corollary 2. Eirst, note that the stationary equilibrium price under the administrative review process, pS is not aIIected by Pr(ADD) and Pr(TER).ThereIore, Pr(ADD) and Pr(TER) aIIect the increase in the dumping duty through the administrative review process, S S E0p / ) p p ( , only through changing the initial price choice, E0p .Then, Corollary 2 (i), (ii) and (iii) are direct results Irom corresponding results in Proposition 2. 38 References Anderson,JamesE.(1992)'DominoDumpingI:CompetitiveExporters,AmericanEconomic Review. Vol. 82: 65-83. Anderson,JamesE.(1993)'DominoDumpingII:Anti-dumping,JournalofInternational Economics. Vol. 35: 133-150. Baldwin,RobertE.(1985)ThePoliticalEconomvofU.S.ImportPolicv.Cambridge,MA:The MIT Press. Blonigen, Bruce A. and Stephen E. Haynes. (Iorthcoming) 'Antidumping Investigations and the Pass-Through oI Antidumping Duties and Exchange Rates, American Economic Review. Blonigen, Bruce A. and Yuka Ohno. (1998) 'Endogenous Protection, Eoreign Direct Investment and Protection-building Trade, Journal of International Economics, Vol. 46(2): 205-227. DeVault,JamesM.(1993).'EconomicsandtheInternationalTradeCommission,Southern Economic Journal, Vol. 60(2): 463-478.

DeVault,JamesM.(1996)'U.S.AntidumpingAdministrativeReviews,InternationalTrade Journal.Vol. 10(2): 247-67. Eichengreen, Barry J., and Hans van der Ven. (1984) 'U.S. Antidumping Policies: The Case oI Steel, in R.E. Baldwin and A.O. Krueger (Eds.), The Structure and Evolution of Recent U.S. Trade Policv.Chicago: University oI Chicago Press Ior NBER. Eeinberg,RobertM.andBarryT.Hirsch.(1989).'IndustryRentSeekingandtheEilingoI UnIairTrade`Complaints,InternationalJournalofIndustrialOrgani:ation,Vol.7(3):325-340. Einger,J.Michael.(1981).'TheIndustry-CountryIncidenceoILessthanEairValue` Cases in US Import Trade, Quarterlv Review of Economics and Business, Vol. 21(2): 260-279. Einger,J.Michael,H.KeithHall,andDouglasR.Nelson.(1982).'ThePoliticalEconomyoI Administered Protection, American Economic Review, Vol. 72(3): 452-466. Eurusawa,Taiji,andThomasJ.Prusa.(1996).'AntidumpingEnIorcementinaReciprocal Model oI Dumping: Theory and Evidence, Mimeo. Gallaway,MichaelP.,BruceA.BlonigenandJosephE.Elynn.(1999)'WelIareCostsoIthe U.S. Antidumping and Countervailing Duty Laws, Journal of International Economics. Vol. 49: 211-244. Hansen,WendyL.(1990).'TheInternationalTradeCommissionandthePoliticsoI Protectionism, American Political Science Review. Vol. 84(1): 21-46. 39 Hansen,WendyL.andThomasJ.Prusa.(1996).'CumulationandITCDecisionMaking:The Sum oI the Parts is Greater than the Whole, Economic Inquirv, Vol. 34(4): 746-769. Hansen,WendyL.andThomasJ. Prusa. (1997). 'The Economics and Politics oI Trade Policy: An Empirical Analysis oI ITC Decision Making, Review of International Economics, Vol. 5(2): 230-245. Herander, Mark G. and J. Brad Schwartz. (1984). 'An Empirical Test oI the Impact oI the Threat oI US Trade Policy: The Case oI Antidumping Duties, Southern Economic Journal, Vol. 51(1): 59-79. Knetter,MichaelM.andThomasJ.Prusa.(2000).'MacroeconomicEactorsandAnti-dumping Eilings: Evidence Irom Eour Countries, NBER Working Paper No. 8010. Krupp,Corinne.(1994).'AntidumpingCasesintheUSChemicalIndustry:APanelData Approach, Journal of Industrial Economics. Vol. 42 (3): 299-311. Krupp,CorinneM.,andPatriciaS.Pollard.(1996)'MarketResponsestoAntidumpingLaws: SomeEvidenceIrom the U.S. Chemical Industry, Canadian Journal of Economics. Vol 29(1): 199-227. Lichtenberg,ErankandHongTan.(1994).'AnIndustry-LevelAnalysisoIImportRelieI PetitionsEiledbyUSManuIacturers,1958-1985,inHongTanandHaruoShimada(eds.), Troubled Industries in the United States and Japan.New York: St. Martin`s Press, 161-88. Moore,MichaelO.(1992).'RulesorPolitics?AnEmpiricalAnalysisoIITCAnti-dumping Decisions, Economic Inquirv, Vol. 30(3): 449-466. Pagan,Adrian.(1984).'EconometricIssuesintheAnalysisoIRegressionswithGenerated Regressors, International Economic Review. Vol. 25( 1): 221-47. Prusa,ThomasJ.(1992)'WhyAreSoManyAntidumpingPetitionsWithdrawn?Journalof International Economics. Vol. 33(1-2): 1-20. Staiger,RobertW.andErankA.Wolak.(1994)'MeasuringIndustrySpeciIicProtection: AntidumpingintheUnitedStates,"BrookingsPapersonEconomicActivitv.Microeconomics, 51-118. Stokey, Nancy L. and Robert E. Lucas, Jr. with Edward C. Prescott. (1989) 'Recursive Methods inEconomicDynamics,HarvardUniversityPress,Cambridge,Massachusetts,andLondon, England. Taylor,ChristopherT.(2001)'TheEconomicEIIectsoIWithdrawnAntidumping Investigations: Is There Evidence oI Collusive Settlements? Mimeo. Zanardi,Maurizio(2000)'AntidumpingLawasaCollusiveDevice,BostonCollegeWorking Papers No. 487. 40 ProIit (p; T') (p-c)q(pT') (p; T'') (p-c)q(pT'') l( ) ' T / pf0 f0p p Figure 1a V(p) cl( ) ' T / pf0 f0p p Figure 1b 41 G (p) G(p; ~ c) BA G(p; c) C 450 cPSP1P0

f0pp Figure 2a G (p)

G(p; c) 450 c PS

f0pp Figure 2b 42 5

5 /

5/

5 /

5$5345678 3 4 5 6 7 8 Figure 3a. (q(p)10-p, c 2, pI 8, 0.9) 5

5 /

5 /

5$51

5345678 3 4 5 6 7 8 Figure 3b. (q(p)10-p, c 2, pI 8, 0.5) 43 Table 1: Descriptive Statistics for Entire First-stage Sample of Variables, 1980-1994. Variables Mean Standard Deviation Minimum Maximum Random-eIIects logit estimates oI petition probabilities Dependent variable: '1 petition, '0no petition 0.038 0.190 0.000 1.000 Independent Variables: Import penetration share lagged. 13.484 15.669 0.000 95.801 Import penetration share lagged and squared. 427.29 1017.78 0.000 9177.87 Industry employment lagged. (in 000s)40.10561.5440.000694.000 Real GDP growth rate2.7472.197-2.0237.263 Exchange rate index (1981100)94.80513.07281.500121.100 Previous AD petitions.0.1500.3570.0001.000 Multinomial logit estimates oI case outcomes Dependent variable: '1 aIIirmative, '2negative, '3terminated/VER 1.699 0.661 1.000 3.000 Independent Variables: Import penetration share lagged. 15.076 8.806 1.027 66.494 Import penetration share lagged andsquared. 304.71 462.64 1.055 4421.48 Industry employment lagged. (in 000s)117.71128.691.300632.400 Real GDP growth rate2.6832.310-2.0237.263 Exchange rate index (1981100)96.27114.60681.500121.100 Import value oI investigated product.33826833510.000859800 Import value oI investigated product squared. 8.08E09 4.82E10 0.000 7.39E11 Previous aIIirmative decision.0.2170.4120.0001.000 Previous negative decision.0.2980.4580.0001.000 Previous terminated/VER decision.0.1650.3710.0001.000 4-Iirm concentration ratio.37.53816.2129.000100.000 Iron and steel products.0.3660.4420.0001.000 Steel pipe products.0.0830.2750.0001.000 Non-market economy.0.1080.3100.0001.000 Less-developed country.0.4990.5000.0001.000 European union.0.2880.4530.0001.000 Japan.0.1220.3270.0001.000 Asian NICs.0.1580.3650.0001.000 44 Table 2: Random-effects logit estimation of first-stage determinants of U.S. AD petition probability. Dependent Variable: '1 iI petition; '0 otherwise. Explanatory Variables Estimation Ior (using 1980-1984 data)Estimation Ior (using 1980-1989 data) Estimation Ior (using 1980-1994 data) Import penetration share lagged. 0.052* (0.031) 0.073*** (0.023) 0.093*** (0.022) Import penetration lagged and squared. - 0.001 (0.001) - 0.001*** (0.0005) - 0.002*** (0.0005) Industry employment lagged. 0.005*** (0.001) 0.005*** (0.001) 0.005*** (0.001) Real GDP growth rate - 0.040 (0.047) - 0.065* (0.036) - 0.061* (0.033) Exchange rate index (1981100) - 0.014 (0.017) 0.017** (0.007) 0.022*** (0.006) Previous AD petitions 2.506*** (0.285) 1.653*** (0.305) 1.105*** (0.294) Chi-squared statistic 129.99*** 109.85*** 86.73*** Number oI observations21594324 6471 NOTES: Standard errors are in parentheses, with ***, ** and * denoting statistical signiIicance (two-tailed test) at the 1, 5 and 10 percent levels, respectively.45 Table 3: Multinomial logit estimation of first-stage determinants of U.S. AD case outcome probabilities. Dependent Variable:'1 iI aIIirmative; '2 iI negative; '3 iI terminated/VER. Explanatory Variables Estimation Ior (using 1980-1984 data)Estimation Ior (using 1980-1989 data) Estimation Ior (using 1980-1994 data) Determinants oI aIIirmative outcomes Industry import penetration share lagged. - 0.089 (0.064) - 0.060 (0.041) - 0.087** (0.030) Industry import penetration lagged and squared. 0.002 (0.001) 0.001 (0.001) 0.002*** (0.001) Industry employment lagged. 0.001 (0.002) 0.002 (0.002) 0.0002 (0.001) Real GDP growth rate 0.051 (0.078) 0.050 (0.055) 0.119*** (0.045) Exchange rate index (1981100) 0.0003 (0.030) - 0.005 (0.011) - 0.014* (0.007) Import value oI investigated product- 6.30e-07 (9.46e-06) 4.44e-06 (4.11e-06) 6.13e-06** (2.83e-06) Import value oI investigated product squared 2.57e-12 (3.68e-11) - 5.30e-12 (7.37e-12) - 7.71e-12* (4.59e-12) Previous aIIirmative decision in investigated product - 1.127 (1.379) 1.848*** (0.520) 0.820*** (0.313) Previous negative decision in investigated product - 0.097 (0.802) - 0.037 (0.404) - 0.638** (0.292) Previous terminated/suspended decision in investigated product - 0.509 (1.083) - 1.536** (0.612) - 1.021*** (0.385) Industry 4-Iirm concentration ratio- 0.024 (0.014) - 0.011 (0.007) - 0.009 (0.006) Iron and steel products.0.702 (0.944) - 0.356 (0.626) 0.578 (0.379) Steel pipe products 1.001 (0.899) 0.428 (0.528) 1.139*** (0.397) Non-market economy 0.595 (0.925) 0.722 (0.550) 0.886** (0.357) Less-developed country 1.464* (0.768) 0.676 (0.429) 0.625* (0.323) European union 0.551 (0.648) 0.003 (0.394) 0.254 (0.314) Japan 1.952** (0.765) 0.981** (0.471) 1.088*** (0.378) Asian NICs - 0.825 (0.787) - 0.305 (0.399) - 0.180 (0.285) Determinants oI terminated outcomes Industry import penetration share lagged. 1.708*** (0.522) - 0.058 (0.088) 0.093* (0.056) 46 Industry import penetration lagged and squared. - 0.081*** (0.023) 0.0003 (0.003) - 0.002 (0.001) Industry employment lagged. - 0.005 (0.004) 0.006*** (0.002) 0.005*** (0.001) Real GDP growth rate 0.091 (0.103) 0.153** (0.070) 0.127** (0.057) Exchange rate index (1981100) - 0.003 (0.047) 0.030* (0.017) 0.035*** (0.009) Import value oI investigated product4.27e-05*** (1.23e-05) 1.55e-05*** (5.37e-06) 9.12e-06** (3.63e-06) Import value oI investigated product squared - 6.56e-11** (3.17e-11) - 2.05e-11* (1.07e-11) - 1.07e-11* (6.30e-12) Previous aIIirmative decision in investigated product 1.656 (1.149) 2.196*** (0.571) 0.851** (0.393) Previous negative decision in investigated product 0.180 (0.759) - 0.559 (0.493) - 0.538 (0.371) Previous terminated/suspended decision in investigated product - 0.246 (0.914) - 0.590 (0.532) - 0.794* (0.427) Industry 4-Iirm concentration ratio - 0.043 (0.027) - 0.033** (0.016) - 0.017* (0.010) Iron and steel products. 3.895*** (1.385) 1.369* (0.710) 0.555 (0.442) Steel pipe products 1.673 (1.132) 1.521** (0.658) 0.682 (0.513) Non-market economy 2.215** (1.128) 1.149* (0.618) 0.774* (0.436) Less-developed country 0.552 (0.952) 1.057** (0.531) 1.181*** (0.427) European union 0.732 (0.796) 0.933* (0.507) 1.043** (0.412) Japan 0.170 (1.119) 0.540 (0.727) 0.149 (0.610) Asian NICs - 3.042** (1.353) - 1.473** (0.640) - 1.019** (0.437) Chi-squared statistic151.88***218.48***229.40*** Pseudo R20.320.230.15 Number oI observations 214 448 715 NOTES: Standard errors are in parentheses, with ***, ** and * denoting statistical signiIicance (two-tailed test) at the 1, 5 and 10 percent levels, respectively. 47 Table 4: Estimation of second-stage determinants of AD margin changes after affirmative AD outcome. Tobit Estimation Probit Estimation Regressors No Controls Controls Broader DeIinition oI Term/Susp. Controls Eocus Variables Ex ante probability oI AEEIRMATIVE outcome.

- 0.642(0.494)

0.802** (0.409)

0.716** (0.340) 2.799*** (0.901) Ex ante probability oI TERMINATION/VER outcome. - 3.282*** (0.901) - 1.716* (0.934) - 0.406 (0.679) - 2.799(3.301) Controls Ior DOC methodology: Change Irom BIA to no BIA.

0.108* (0.060)

0.116* (0.060)

0.627*** (0.137) Change Irom no BIA to BIA. 1.244*** (0.127)

1.263*** (0.128)

1.314*** (0.402) Period oI automatic administrative reviews.

- 0.481*** (0.091)- 0.547*** (0.093) - 0.819*** (0.198)Non-market economy.

0.138* (0.080)0.137 (0.080) 0.831*** (0.207)Revocation oI AD duty.

- 1.047*** (0.162)- 1.107*** (0.157) - 1.325*** (0.306) Chi-squared statistic. 64.59*** 260.12***249.92*** 152.97*** Pseudo R2.0.05 0.220.210.21 Number oI observations. 541541541541 NOTES: Dependent variable Ior tobit estimation is the 4-year percentage change in the AD duty, while the dependent variable Ior the probit estimation a dummy variable that takes the value oI '1 iI AD duty does not decrease.Robust standard errors Ior all variables except the ex ante probabilities, Ior which the standard errors are bootstrapped.***, ** and * denote statistical signiIicance (two-tailed test) at the 1, 5 and 10 percent levels, respectively.