Entering Global Value Chains: Do Trade Missions Work?1

Working Paper 1/17

* Jann Milic, Alfons Palangkaraya† and Elizabeth Webster† * i3partners, †Centre for Transformative Innovation, Swinburne University of

Technology

March 2017

1Disclaimer:AnyopinionsandconclusionsexpressedhereinarethoseoftheauthorsanddonotnecessarilyrepresenttheviewsoftheDepartmentorABS.Allresultshavebeenreviewedtoensurethatnoconfidentialinformationisdisclosed.WewishtothankAnthonyJones,ConnieCrisafi,GonulSerbest,BruceLevett,MargaretBrettandRebeccaHallfromDEDJTRforsubstantialcommentsandsuggestionsandtheuseofVictoriaTradeMissionprogramparticipantsdatabase;DianeBraskic,DavidTaylorandTomPougherfromtheABSformakingtheanalysisoftheABSdatapossible;andHelenSilverandHowardRonaldsonfortheforesightandconfidencetofundtheresearch.

Abstract

Weusedmatchedfirm-leveldatafromthepopulationofbusinessesinAustraliatoevaluatetheimpactthattrademissions,operating2010to2013,hadonexportsusingadifference-in-differencemethod.Wefoundthat,onaverage,themissionsincreasedparticipatingfirms’exportswithin12monthsbyatleast172percent.Furthermore,trademissionparticipationappearedtoincreasetheprobabilityofbecominganexporterwithin12monthsby26percentagepoints.Theadditionaleffectofundertakinga subsequent trade mission was smaller than the first mission but still positive and statisticallysignificant.

1. Introduction

Trademissionsarestructured,personalvisitsbygroupsofproducersseekingnewtradeopportunitiesinforeigncountries.Foracountryontheperipheryofatradingbloc,remoteinculture,languageorlocation,missions can be effectiveway to initiate entry into global value chains.Notwithstandingevidence that it is only themost productive firms which export,2 trademissions can provide themarginaladvantageincustomisedtradeswherepersonalintroductions,productqualityandcompanyreputationarecritical.3Theaimofthisarticleistoevaluatewhethertrademissionsdoaffectexports,or,whethertheymerelysubsidisebusinessesthatwouldotherwiseexport.

Themostchallengingissueinanyevaluationistoestablishthecounterfactual–inourcase,thelevelofexportsifthesamefirmhadnotparticipatedintheprogram.Takenliterally,thiscounterfactualisnotpossible,butwecanmimicthecounterfactualbyconstructingacontrolgroupthatissimilartotheprogramparticipantswithrespecttothedeterminantsofexporting.

Constructingacontrolgrouprequiresaccesstovariablesatthefirmlevel.Priorto2008,thisdatawasnot available to most researchers evaluation had to resort to industry or country-level data.Accordingly, evaluations were only able to assess contemporaneous bundles of export supportprograms. These bundles of programs extends well beyond trade missions and includes export-orientated programs around information provision, coordination, R&D, training, feasibility tests,technicalandmarketingservices,technologyacquisition,loanguaranteesandinterestsubsidies,interalia.Wearguethatbecausetrademissionsarespecificallydesignedto insinuateafirmintoglobalvalue chains, and given the rising importance of these chains, these missions deserve separate,highlightedtreatment.

Thisevaluationdiffersfrompreviousevaluationsasitusesthefullcensusoffirm-leveldatarelevanttothejurisdictionandalsoseparatelyidentifiestrademissionparticipants.Focussingspecificallyontrademissionprogramsdoesnotallowustotestforsynergiesbetweenexportprograms(seeVanBiesebroeck, Konings, and VolpeMartincus 2016), but does give the policy decisionmakermoreconcreteinformationaboutwhethertocontinuetofundtheprogram.

Wefindthatonaverage,trademissionsincreasedparticipatingfirms’exportswithin12monthsbyatleast172percent.Furthermore,trademissionparticipationappearedtoincreasetheprobabilityofbecominganexporterwithin12monthsby26percentagepoints.Theadditionaleffectofundertakingasecondmissionwassmallerthanthefirst,butstillpositiveandstatisticallysignificant.

2Clerides,LachandTybout(1998);BernardandJensen(1995).3VolpeMartincusandCarballo(2012)findthatexportsupportprogramsaremosteffectiveforfirmsproducingdifferentiatedgoods.

2. Whyaretrademissionssopertinent?

Sincethe1970s,ithasbecomeincreasinglycommonforasingleproducttoconsistofmanyinputsfrom a series of specialised countries. An assortment of terms has appeared in the literature todescribe this phenomenon including: ‘slicing up the value chain’, ‘disintegration of production’,‘fragmentation’, ‘multi-stage production’, and ‘intra-product specialization’ (see Feenstra 1998;Hummels,IshiiandYi2001).Thisinternationalfragmentationofproductionisacorollaryofthedesirebycompaniestooutsourceparts,servicesorcomponentassemblytoexpertsuppliers.Fallingtradebarriers,mostnotablythereductionintariffsandreducedtransportandcommunicationcosts,havedivertedmuchofthisoutsourcingoffshore.

The result has been amajor transformation in world trade. Global grossmerchandise trade as apercentageofworldoutputrosefrom17.5in1960to45.0in2015.4Furthermore,thegrowinggapbetweengross and value-addedworld exports has implied that adisproportionate amountof theincreasedtradeisduetothetradeofinputsandcomponentsratherthanfinishedgoods(JohnsonandNoguera2014).

AccordingtoBernardetal.(2007),resultsfromvirtuallyeverystudyacrossindustriesfindthatonlythemostproductivefirmsexportwhichimpliesthepresenceofsunkentrycostsintoexportmarkets.Chen,LuandZhou(2015)havearguedthatmultinationalsaddressthesebarriersbyfirstoffshoringtotheirforeignsubsidiaries,andthenastheproductmatures,toexternalforeignsuppliers.Ifthetradedproduct iscustomised,thequalityoftherelationshipbetweenthebuyerandthesellercanbethedeterminingfactorinadeal.Personalvisits,mediatedbythewarmintroductionsofferedinatrademission,aredesignedtocreatetheserelationships.

3. Existingevidence

Althoughpopularwithgovernmentministersandpremiers,trademissionsdonothavetoberunorsubsidisedbythepublicsectorandcanbeoperatedbyindustryassociationsorfor-profitcompanies.Nonetheless,regardlessofwhichentityoperatesorpaysforamission,allparties,nottheleastthebusinessitself,haveaclearinterestinknowingwhetherornottheysucceed.Post-programsurveysofgratefulrecipientsofgovernmentlargessdonotconstituteacceptable,objectiveevidence.Rigorousevaluationsneedbebasedonbehaviourthatisrevealedthroughverifiablerecordssuchassalesoremployment.

Trademissionsaredistinctfromtradeshows.Tradeshowsaremassmany-to-manyexhibitions,wherethesponsorprovides,orsubsidises,boothspaceinahallforcomplementarybusinessestomeeteachother.Trademissions,incontrast,arebespokeorganisedtripstooverseasdestinations.Themissionwill introducebusinessestocustomersandbusinesscounterpartsone-on-one,often inthe latter’spremises.Bothshowsandmissionscanbeindustry-focusedandaspecificexportpromotionprogrammayexhibitthecharacteristicsofbothatradeshowandtrademission.

Theexistingempiricalevidenceontheeffectoftrademissionsisthinandmixedwithmoststudiesevaluating bundled export promotion programs. Nonetheless, the two developed economyevaluationsspecificallyontrademissionscompriseacountry-levelevaluation(HeadandRies,2010,whofindastatisticallyinsignificantbutsmallnegativeeffect)andasurveyof113participants(Spence,2003,whoreportspositiveeffects).

4Source:WorldDevelopmentIndicators,WorldBank,API_TG.VAL.TOTL.GD.ZS_DS2_en_csv_v2.zipWorldTradeOrganization,andWorldBankGDPestimates.Downloaded9/12/2016.Note,merchandisetradeonlyincludetradeinphysicalobjects,notservicesnorcapitaltransfersandforeigninvestments.

Asmentioned,themorecommonapproach,especiallyindevelopedcountryliterature,istoevaluatedexportsupportprogramsasabundle.Theolderstudiesalsousedaggregatednationalorregionaldataand again throw up inconclusive results. Gil, Llorca and Serrano (2007), for example, found thatregionalexportpromotion isassociatedwith74percenthigherexports;Lederman,OlarreagaandPayton(2010)findcountry-levelcorrelationsbetweenspendingonexportpromotionprogramsandexports,butBernardandJensen(2004)usingUSstate-leveldatadonot.Rose(2007)andCreusenandLejour(2012)bothfindapositiverelationbetweenthepresenceofaforeigntradeofficewithexportsto that destination. As correlations may indicate successful rent seeking by large exportercommunities, these aggregate studies rarely provide the hard evidence needed to convincemostpolicydecisionmakers.

Evaluationssince2008have takenadvantageof thebourgeoningavailabilityof firm-leveldatasets(see Van Biesebroeck, Konings, and VolpeMartincus 2016 for a review). Almost all studies find apositiveand significanteffectofexportpromotion supporton firm-level exports.However, as theabove authors note,most studies only havedata for firmswhich export, or are limited to a non-randomsampleoffirms(e.g.Gὂrg,Henry,andStrobl,2008;Lach2002;VanBiesebroeck,YuandChen2015; Van Biesebroeck, Konings and Volpe Martincus 2016; Mion and Muûls 2015; Lederman, Olarreaga and Zavala 2016).Althoughrelevantbackgroundforourquestion,thesestudies,someofwhich include trade mission programs, do not identify which types of export support are mosteffective.

Nonetheless,therearefivepublishedfirm-levelstudies,allfromSouthAmerica,specificallyevaluatingtrademissions. VolpeMartincus and Carballo (2008, 2010b, 2010c, 2012) and Álvarez and Crespi(2000) findconsistentlypositive results,especiallyalong theextensivemargin (newexportmarketentryornewproductintroductiontoexistingexportmarkets).Inthispaper,wetestforwhetherthesameresultcanbefoundforadevelopedeconomy,Australia.

4. Evaluationmethodanddata

As mentioned, the problem confronting evaluations based on observational data is that thecounterfactuals, (whatwould have happened to the observed outcomes if the programwere notimplementedoriftheparticipantsdidnotparticipate),areneverobserved.Thebestwecandoistoinferthecounterfactualsfromobservingagroupofnon-participantswhicharesimilar,pre-program,inallthenon-programdeterminantsoftheoutcome.Ifprogramparticipationisnotrandom,weneedto account for the fact that themore capable, interested or talented businesses chose to do theprogram.

Once we have selected a control group from the pool of non-participants, we use difference-in-differenceanalysistoquantifytheeffectofprogramparticipationontheoutcome,inourcaseexports.Thisstandardapproachtoevaluationcombinestwomethods,first,theuseofamatchedcontrolgroupbasedonobservablecharacteristicssowecomparelike-firmswithlike;5andsecond,thedifference-in-difference estimator which eliminates time-invariant unobservables (Heckman et al. 1997).However,theestimatedprogrameffectcanbebiasedifthereremainunobservedtime-varyingfirm-related factors which affect both program participation and export outcomes. This could be thesudden identificationof anexportopportunityby the firms’managers,or a change in themarketcircumstancesforaspecificniche.

5Thismeanswedonothavetoextrapolateandinfer,say,theprogrameffectonalargefirmfromdataonsmallfirms.

Thedatabases

Twofirm-leveldatasets,linkedbyyearandbusinessidentifier,formsthebasisoftheevaluation:

(i) Thepopulationofall1192businesseswhichparticipatedinatrademissionwiththeVictorianGovernmentbetween1December2010to30June2013.VictoriaisastatewithinAustraliaandhasapopulationofabout6million.Eachtrademissioncomprises20-100businesses.

(ii) ThepopulationofallAustralianbusinessesthatcompletedaBusinessActivityStatementandBusiness Income Tax (BAS-BIT) database between 2001-02 and 2012-13 (over 19 millionrecords).6

TheBAS-BITdatabaseincludesanumberofindicatorsofbusinessperformanceincludingexportsofgoods and services; sales, turnover and effective full-time employment.7 Unlike most firm-leveldatasets,ourdatabasecontainsbusinessesofallsizes.However, thedatabaseonlyrecordsexportrevenuesiftherecipientofthegoodorserviceisoutsideAustralia.Thisincludesconsultancyservices,contractresearchorbusinessservicesundertakeninAustraliabutpaidforbyanoverseascompany.TourismandeducationservicesconsumedinAustraliabynon-residentsarenotrecordedintheBAS-BITdatabaseastheyarenottax-free8.Althoughthismeansserviceexportsalesareunderestimated,at least relative to measured goods exports, this will not bias our results if the extent ofunderestimation stays constant before and after theprogram, andbetween theparticipation andcontrolgroups.

In 2011-12, the BAS-BIT database contains records of 2.5 million businesses in Australia. Afterremoving records with zero values in sales, business income, total expenses, or salary and wageexpensesweareleftwith1.5millionAustralianand660thousandVictorianbusinesses.Ofthe1192businessesthatundertookatrademissionbetween2010and2013,wewereabletomatch843(ofapossible1192)uniquetrademissionbusinessestotheBAS-BITdatabase.ThismatchingrevealedthatManufacturing,Wholesale trade, Professional, scientific and technical services and Education andtrainingwereover-representedintheprogramcomparedwiththetotalpopulation.TheseindustriesrepresentVictoria’srelativecomparativeadvantageintermsofindustrialcapabilities.Threeinfourtrademissionbusinessesarefromservicesindustry.

WedonotexpectprogramparticipantstobearandomsampleofallVictoriafirms.Tobeeligibleforthetrademissionprogram,firmsmustbefinanciallyviable;beabletodemonstrateasoundcasefordoingbusiness in the targeted regions; andbe currently exportingor able to demonstrate exportreadiness.Table1showsthatprogramparticipantsaremuch larger (salesare70 times largerandemployment is 30 times larger) andmuchmore likely to be an exporter and exportmore. Thesecomparisonsindicatepotentialendogenousselectionintoprogramandaviolationofthe ‘commontrend’assumptionwhichonlyholdsifbothprogramandcontrolgroupexportswouldfollowthesametimetrendintheabsenceofthetrademissionprogram.

6NotethattheABSBAS-BITdatabaseislargeandcomplexandcanonlybeaccessedbyapprovedresearchersindirectlyviastafffromwithintheABS.Thedatabaseisconfidentialandnon-ABSanalystscannotseethedata.Resultsareonlyreleasedtonon-ABSpeopleaftercarefulscrutinyoftheoutputtoensurenobusinesscanbeidentified.Theseaccesslimitationsdonotaffectthequalityoftheempiricalanalysisduetoourdetailedandthoroughanalysis.7ExportedgoodsareGST-freeiftheyareexportedfromAustraliawithin60daysofoneofthefollowing,whicheveroccursfirst:thesupplierreceivespaymentforthegoodsorthesupplierissuesaninvoiceforthegoods.OtherexportsgenerallyincludesuppliesofthingsotherthangoodsorrealpropertyforconsumptionoutsideAustralia,suchasservices,variousrights,recreationalboats,financialsuppliesandotherprofessionalservices.8Goodsandservicestax.

Table 1: Number of Victorian businesses and average firm characteristics 2001-02 to 2012-13, by trade mission participation status, (T = Trade mission participants; C = Control)

Number of businesses

Proportion of exporters (%)

Exports sales ($ thousands)

Total sales revenues

($ millions)

Employment (EFT persons)

Year T9 C T C T C T C T C

2001-02 424 397,189 41 3 20600 87 137.0 1.4 577 11 2002-03 459 440,022 43 3 15200 70 122.0 1.4 622 10 2003-04 501 488,299 41 3 15400 75 126.0 1.5 465 10 2004-05 525 493,570 43 3 17400 82 128.0 1.7 735 15 2005-06 552 548,418 42 3 16700 78 125.0 1.7 314 9 2006-07 589 613,271 42 2 11600 2 121.0 1.7 302 8 2007-08 646 666,195 43 2 14000 77 119.0 1.8 290 8 2008-09 657 676,267 40 2 13500 93 148.0 1.7 326 8 2009-10 713 626,120 43 2 7926 127 146.0 1.9 323 8 2010-11 772 646,030 44 2 8684 161 170.0 1.9 315 9 2011-12 821 661,278 44 2 7725 185 158.0 2.0 318 9 2012-13 795 656,152 45 2 6419 161 154.0 2.1 323 9

Notes: Constructed based on merged Victorian Government trade mission program administrative database and cleaned version of BAS-BIT database for the State of Victoria. The total number of businesses may not be identical to the official ABS’ estimate of number of businesses in Victoria in each financial year.

5. Themodel

Denoteprogramparticipation𝐷"#where𝐷"# = 1 if firm𝑖participates in theVictorianGovernmenttrade supported program in year 𝑡 and 𝐷"# = 0 otherwise. Denote 𝑋"# as a vector of observedcovariatescorrespondingtoobservablefirmandprogramcharacteristics.Denote𝑌"#,astheobservedoutcome(say,exportrevenues)and𝑌"#-astheunobserved(counterfactual)outcome.Hence,denote𝐸 𝑌"#,|𝑋"#, 𝐷"# = 1 as theobservedaverageoutcomeofparticipating firms conditionalon𝑋"# and𝐸 𝑌"#-|𝑋"#, 𝐷"# = 1 as the counterfactual average outcome of participating firms had they notparticipated.Notethat,foragivenfirmweeitherobserve𝑌"#,,or𝑌"#-,butnotbothvariablesatthesametime.

Then,theimpactoftradepromotionprogramismeasuredbytheaveragetreatmenteffectonthetreated(ATT)denotedby𝜏:

𝜏 = 𝐸 𝑌"#,|𝑋"#, 𝐷"# = 1 − 𝐸 𝑌"#-|𝑋"#, 𝐷"# = 1 (1)

Inequation(1)𝜏,measurestheaveragechangeintheoutcomesofparticipatingfirmsasthedifferencebetweenobservedaverageoutcomesaftertreatmentandcounterfactualaverageoutcomeshadthefir𝜏msnotreceivedthetreatments.Itisclearfromtheequationthattoobtainanunbiasedestimateofweneedanunbiasedestimatedof𝐸 𝑌"#-|𝑋"#, 𝐷"# = 1 ,thecounterfactual.Anobviouscandidateistousetheaverageoutcomeofaselectedgroupofnon-participants.Thiscontrolgroupwouldneedtobe identified by taking into account any potential non-randomness in program participation. The9Asmentionedintheprecedingparagraph,843businesswhichparticipatedintheTradeMissionsprogramandrecordedintheDEDJTRdatabasewerefoundintheABSBAS-BITdatabase.However,someofthesehavemissingvaluesintermsofthematchingvariablessuchsalesrevenues,wages/employmentorexportforvariousreasons.Forexample,someofthebusinessesmaynotexistpriorto2010-11ortheymayexistunderdifferentABNs.Asaresult,thefiguresreportedinthecolumnswiththe“P”heading(thatis,thenumberofparticipants)decreaseaswemoveawayfromtheVICTradeMissionYears(2010-11to2012-13).

descriptivestatisticspresentedinTable1showthatthedifferencesbetweenparticipantsandnon-participantsappeardonotappeartoberandom.10

Inotherwords,weneedtoselectthecontrolgroupsuchthatfirm’sheterogeneouscharacteristicsarecomparable inbothgroups.Asmentioned,wewilldo this firstby selectingacontrolgroupwhichmatchestheparticipationgrouponobservablefactorsthatwebelievedetermineexporting.Weusepropensityscorematchingwhichisestimatedasthepredictedprobabilityofafirmtoparticipateintheprogrambasedonobservedco-variates,𝑋.Foreachyear,theco-variatesvector𝑋"#consistsoftotalsalesrevenues,whetherornotanexporter, importvalues,totalwagespaid,shareofforeignownership and one-digit industry code. Thus, 𝑋"# measure size and the extent of internationalengagementofthefirmswithineachbroadindustry.

Usingonly theyearsbeforeVictorianTrade supportedprogrambegun (that is,data from2009orearlier),wecomputethepre-2009averagevaluesofeachcomponentsin𝑋"#acrosstheyearsforeachfirm.Denotethisaveragevaluesas𝑋"567;thiscovariatevectorsistheindependentvariablesforthepropensityscorematching.

UsingonlythedatafromtheparticipationandcontrolgroupsweestimatetheDIDestimatoras:

𝑌"# = 𝑋"#𝛽 + 𝜏𝐷"# + 𝜇" + 𝜆# + 𝜀"# (2)

Note that in specifyingequation (2),weassume theconditionalexpectation function𝐸 𝑌|𝑋, 𝐷 islinearandanyunobservedfirmcharacteristicsisdecomposableintoatime-invariantfirmspecificfixedeffects(𝜇"),commonacrossfirmsyeareffect(𝜆#)andarandomcomponent(𝜀"#).Theintroductionofthecovariates (𝑋) linearlymay leadto inconsistentestimateof𝜏duetopotentialmisspecification(Meyer,1995;Abadie,2005)ifwehadnotlimitedourestimationsamplewiththematchinganalysis.

Thematchingdifference-in-differencemethodcanestimatetreatmenteffectswithoutimposingthelinearfunctionalformrestrictionintheconditionalexpectationoftheoutcomevariableis(ArnoldandJavorcik,2005;Gorgetal2008).Thematchingmethodpartcontrolsforendogenousselectionintoprogramsbasedonobservables(HeckmanandRobb,1985;Heckmanetal1998).Thedifference-in-difference part controls for endogenous selection into programs based on time invariantunobservables.Notehoweverthatwestillneedtoassumethatthereisnotimevaryingunobservedeffectsinfluencingselectionandexports(seeHeckmanetal.,1997;BlundellandCostaDias,2002).

6. EvaluationFindings

Impactsonexportrevenues

WeobtainedeightsetsofDIDimpactestimatesbycomparingVictoriaTradeMissionsparticipantstodifferentsetsofnon-participantsproducedbydifferentmatchingmethods.WerefertotheseeightsetsofimpactestimatesasModel1toModel8estimates.

InModel1,wedidnotperformanymatching.Allavailablenon-participatingfirmswereusedasthecontrolgroup.Intherestofthemodelsweusedmatching.11InModel2weusedthenearestneighbour10Asastylisedfact,exporters,fromallcountries,arelarger(onthebasisofemploymentandtangibleassets);employmoreskilledandwell-paidworkersandaremorelikelytobeforeignownedandpartofamulti-plantenterprise(BernardandJensen,2004;RobertsandTybout,1997;Wagner2007;Bernardetal.,2007;Gὂrg,Henry,andStrobl,2008).Moreover,exportingisapersistentprocess:today’sexportersaremorelikelytoexporttomorrow,whichsuggeststheinfluenceofthefixedandsunkcostsofexporting;andthemanagerialorproductorientationtowardsexportmarkets(BernardandJensen,2004;Timoshenko2015,KaiserandKongsted2008,Padmaja,andSasidharan,2016).11SeethediscussionsinAppendix1and2formoredetails.

based on estimated propensity scores. In Model 3 we used five nearest neighbours based onestimatedpropensityscores.InModel4weusedoneCoarsenedExactMatching(CEM)matchednon-participantforeachparticipant.InModel5usedallCEMmatchednon-participatingfirms.Models6-8aresimilartoModels2-4respectively,exceptfortheadditionoftwotime-varyingcontrolvariables(firmageand sizeof employment). Theseeight setsof estimatesof the impactsofVictoria TradeMissionprogramontheparticipants’exportsalesaresummarisedinTable2.

Table2:Average increase inexportsalesofVictoriaTradeMissionsparticipants,2010-2013,percent.

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 0-12 months Average 135 219 192 186 138 172 161 157Lower 95%-CI 117 117 141 103 120 60 85 51Upper 95%-CI 152 321 244 269 156 284 237 2630-24 months Average 165 345 226 291 174 343 224 332Lower 95%-CI 139 198 170 172 147 151 131 142Upper 95%-CI 190 491 281 409 200 535 316 522

Notes: Estimates are based on difference-in-difference analysis of participating Victorian firms compared to different sets of non-participating Victorian firms Model 1 uses all non-participating firms as control group. Model 2 uses one propensity score matched non-participating firm for each treated firm as control. Model 3 uses five propensity score matched non-participating firms. Model 4 uses one Coarsened Exact Matching (CEM) matched non participant. Model 5 uses all CEM matched non-participating firms. Models 6-8 are similar to Models 2-4 respectively, except for the addition of two time-varying control variables (firm age and size of employment). Lower and upper bounds (Lower 95%-CI and Upper 95%-CI) are approximated 95% confidence intervals.

Table2showsthatregardlessofthemethoduse,theaverageimpactofthetrademissionprogramonexport revenue is positive and significant both in terms ofmagnitude and statistical significance.Beforecontrollingforselectiononobservables,participantshadonaverage135percent(seeModel1)higherexportrevenuewithin12monthscomparedwiththecontrolgroup.Thecorresponding95%confidenceintervalwas117to152percent.Theestimatedimpactwithin24monthswashigheratanaverageof165percent.However,movingfromaone-yeartoatwo-yearperiodonlyaddedaround30percentagepointstotheimpactwhichislessthanthe135percentinitialimpactinthefirstyear.

Model2(anditsmorerobustversionModel6)shouldprovidethemostreliableimpactestimatesasthecontrolgroupshowednostatisticallysignificantdifferencetotheprogramparticipantsintermsofpre-programexportperformance.Onaverage,theimpactestimatesproducedbyModels2and6were186and172percentrespectively.However,their95%confidence intervalswerealsowider,suggestingthatweneedtotakeintoaccountoftherangeoftheimpactestimates.Nevertheless,eventhemost conservative estimates summarised in Table 2 above (which is 51 per cent according toModel8’slowerbound)suggeststhatthetrademissionparticipationhadasignificantpositiveimpact.

The average exports sale of participants in the base year (that is pre-program participation) was$809,662.Basedonthemostconservativemodelspecification,Model6(whichisthemorerestrictiveversion of the preferred Model 2), in monetary terms trade mission participation increasedparticipants’ exports sales by at least 60% x $809,662 = $485,797within 12months and 151% x$809,662=$1,222,590within24months.

Table3comparestheestimatesfoundbythisstudywiththeself-reportedestimatesfromasurveyadministeredbytheVictorianGovernment.Itshowsthatthereportedincreaseinexportsiswithinthe rangeofourestimates (closer to the lowerboundsof theDID impact estimates). This findingsupportsthenotionthattheself-evaluationdatareportedbyparticipantscanbevaluable.Table3:AverageincreaseinthevalueofexportsalesofVictoriantrademissionparticipants,2010-

2013,asreportedbyparticipantsandestimatedbythisevaluation Averageincreaseinexportsales

Reportedbyparticipants

Thisevaluation’smost-conservativeestimates

ImmediateExportSales $212,476 NotestimatedWithin1-12Months $565,592 60.0%x$809,662=$485,797Within13-24Month $1,116,893 NotestimatedWithin0-24Month $1,317,355 151%x$809,662=$1,222,590

Notes:Estimatesarebasedondifference-in-differenceanalysisofparticipatingVictorianfirmscomparedtodifferentsetsofnon-participatingVictorianfirms(seethenotesforTable5.1).Theimpactelasticitiesusedinthethirdcolumn(117.4%and139.4%)correspondtothesmallest95%confidenceintervallowerboundssummarisedinTable5.1.

7. Impactsontheprobabilityofexporting

Approximatelyhalfofprogramparticipantswerenotexporters inthebaseyear.Usingthisnaturalvariationinthedata,wehavederivedDIDimpactestimatesusingtheprobabilityofbeinganexporterastheexportperformancemeasure(insteadofthevalueofexports).

The results, summarised in Table 4, presents five sets of estimates corresponding toModels 1-5discussed above.12 Based on the preferred specification of Model 2, trade mission participationincreased the probability of becoming an exporter by 26 percentage points within 12 months(approximately53percentincrease)and35percentagepointswithin24months(approximately71percentincrease).

Table4:IncreaseinprobabilityofexportofVictoriantrademissionparticipants,2010-2013,byempiricalmodelspecification,percentagepoints.

Model1 Model2 Model3 Model4 Model50-12months Average 21 26 26 24 20Lower95%-CI 15 17 18 15 18Upper95%-CI 26 35 34 33 21 0-24months Average 26 35 32 34 25Lower95%-CI 18 26 24 24 18Upper95%-CI 33 45 39 43 32

Notes:Estimatesarebasedondifference-in-differenceanalysisofparticipatingVictorianfirmscomparedtodifferentsetsofnon-participatingVictorianfirms(seethenotesforTable5.1).NoresultsforModel6-8duetonon-convergenceissues.Lowerandupperbounds(Lower95%-CIandUpper95%-CI)areapproximated95%confidenceintervals.

8. Repeatandmulti-yearparticipations

Somebusinessesparticipatedinmorethanonemission.Overtheevaluationperiod,442outof1192participating businesses participated more than once, with the average number of missions perparticipatingbusinessbeing1.7.Thus,itisofaparticularinteresttoknowifthoserepeatparticipantsexperiencehigherimpactstoone-offparticipants.Itisnotpossibletoseparatetheimpactsofrepeatparticipationswithin the sameyear,13 however, formulti-yearparticipation (regardlesshowmany

12Models6-8estimatesareunavailableduetoconvergenceissuesinestimatingtheconditionallogitmodelwhenthetwotimevaryingvariables(ageandemployment).13Technicallyspeaking,thetimeinvariantindicatorstatusofparticipantswithandwithoutrepeatparticipationisdifferencedoutbytheDIDanalysis.

trademissions attendedwithin each year)we can obtain separate estimates for the first year ofparticipationandthesecondyearofparticipation.

Theestimates for first yearparticipation is summarised in Table5below.14 Theseestimates showdiminishingreturnstotrademissionparticipation.Theincreaseinexportsalesfromparticipationinthe second year (or more) was on average around 50 per cent smaller than the increase fromparticipatingonlyinoneyear.

Table5:Average increase inexport salesofVictoriaTradeMissionsparticipants in the first andsecond(ormore)yearofparticipation,2010-2013,percent.

Model1Firstyearparticipation Average 248Lower95%-CI 136Upper95%-CI 359 Second(ormore)yearofparticipation Average 110Lower95%-CI 2Upper95%-CI 218

Notes: Estimates are based on difference-in-difference analysis of participating Victorian firms compared to different sets of non-participating Victorian firms. Lower and upper bounds (Lower 95%-CI and Upper 95%-CI) are approximated 95% confidence intervals.

9. Robustnessandlimitations

Ingeneral,programimpactevaluationwithobservationaldata(thatis,wheretheanalysthadnodirectcontrol on the data generation process or on how the sampleswhose data being observedwereselected) suffers frompotential selectionbiasdue toobservedandunobserved factors thataffectboth decision to participate in the program and the intended outcomes from the program. Forexample,programeligibility,incentivesandexpectationsmayleadtoparticipantsbeingsystematicallydifferent from non-participants in such a way that a naïve comparison of the performance ofparticipants and non-participants would lead to biased estimates of the program’s impact. Asmentioned above, in order to be eligible for the trademissionprogram, firmsmust be financiallyviable; be able to demonstrate a sound case for doing business in the targeted regions; and becurrently exporting or able to demonstrate export readiness. These characteristics were notobservableinourdatabase,butastheydetermineprogramparticipation,theyarelikelybecorrelatedwithoutcomes.

In this evaluation, we implemented difference-in-difference analysis in order to eliminate theinfluence of unobserved and time-invariant factors (factors which do not change over time butdeterminewhetherornotafirmparticipatedintheprogramandarecorrelatedwiththeoutcomesbeingevaluated)bycomparingthechangeintheperformanceoftheparticipantbeforeandaftertheprogram to the change in theperformanceof non-participants. Effectively,wedifferenceout anytime-invariantconfoundingeffectsthatcouldleadtobiasedestimates.

However,we still had to dealwith potential bias caused by unobserved but time-varying factors.Furthermore,implicitinthedifference-in-differenceanalysisisacommontrendassumption:thatthechangesintheperformanceofbothparticipantsandnon-participantsarethesameintheabsenceoftheprogramintervention.Inpractice,weensurethatthecommontrendassumptionwasnotviolated

14TheseestimatesarebasedonthepreferredModel2specification.

byselectingonly‘similar’non-participantsasthecontrolgroup.Todothis,weappliedtwodifferentmatching techniques (propensity scorematchingandcoarsenedexactmatching)onobservedpre-program businesses characteristics thatwere likely to be related to decision to participate in theprogram.Tohandlethefirstproblemofunobservedtime-varyingconfoundingeffects,weestimatedtheimpactsoftheprogramconditionalontwoobservedtimevaryingvariableswhicharelikelytobecorrelatedwiththeunobservedtime-varyingfactors:businessageandemploymentsize.

Therefore,we believe our estimateswere robust to different potential bias sources: observed orunobservedandtime-varyingortime-invariant.Therobustnessofourfindingswasfurtherevidencedbytherelativelysimilarresultsexhibitbyouruseofdifferentmodelspecificationstocontrolthesesourcesofbias(Model1–Model8)anddifferentmeasurestoderiveimpactestimates(exportsalesand export probabilities, 0-12 and 0-24months, Year 1 and Year 2+, and the approximated 95%confidenceinterval).

Therearesomelimitationstothisevaluation,mostlyrelatedtodataavailability.First,whileweknewthedestinationcountriesoftrademissions,wedidnotknowtheexportdestination.OnemayexpectthataparticipationinatrademissiontoChinawouldbemorelikelytoincreaseexporttoChinathantoothercountries.Globalisation invaluechainsofproductionmay temper thisdirect relationshippartly,butitremainsthatifweknewexportdestinationwemightbeabletoobtainamorepreciseestimate (in terms of its causality relationship) of the program impact. To address this limitationrequires theBAS-BITdatabase tobe supplementedwithdetailedcustomsdata.Webelieve this isfeasible since the information is collectedbyAustralianCustomsoffice and theABShas aplan tomergethecustomsdatabasewiththeBAS-BITdatabase.15

Anotherlimitationofthecurrentevaluationthatisrelatedtodataavailabilityisthesmallsamplesizeof program participants (relative to the sample size of non-participants). There are potentiallyinterestingaspectsofdifferent trademissionssuchasdestinationcountriesmentionedaboveandcharacteristics of the tradeevents themselves (which industry, regional or country specific,whichdelegatesfromothercountriesparticipate,whichcountryofficialsweremet,andmanyothers).Ananalysis of the roles of these factors on the impact of trade missions would yield interestingimplicationtoimproveprogramdesignandtargeting.However,suchanalysisisomittedduetolimitedsamplesizeandinformation.

10. Conclusion

Thisevaluationsupportsthecasefortrademissionsrunthroughindustryassociationsorfor-profitorganisations:onaveragebusinessesthatparticipateinamissionmorethandoubletheirexports.Webelievethatthisoccursbecausepersonalcontactsmadeduringthemissionsovercomesomeoftheadditionalinformationcostsassociatedwithdoingbusinessinforeignmarkets.Thesecostsincludeestablishingarelationshipoftrust;identifyingpotentialcustomersandunderstandingtheirnuancedrequirements.

However,thequestionforpolicymakersis:shouldthesevisitsbesubsidisedbythepublicpurse?Forpublicsupporttobejustified,wehavetomakeacasethattherearepositivespilloversfromtheserelationships.Thesespilloversmayoccurwhenademonstrationbyonefirmiscopiedbyitspeersorwhenforeigncontactsareshared.Wehavenotestablishedthepresenceofspilloversandthisremainsasubjectforfurtherresearch.

References

15However,thereisnoannounceddatewhenthiswillhappen.

Bernard,A.B.,Jensen,J.B.,Redding,S.J.andSchott,P.K.,(2007),Firmsininternationaltrade.TheJournalofEconomicPerspectives,21(3),pp.105-130.

Clerides,S.,Lach,S.,&Tybout,J.(1998).‘IsLearningbyExportingImportant?Micro-DynamicEvidencefromColombia,Mexico,andMorocco’,TheQuarterlyJournalofEconomics,113(3),903-947.

Creusen,H.,andLejour,A.(2012)“Marketentryandeconomicdiplomacy,”AppliedEconomicLetters20(5),504–7

DanielLederman,MarceloOlarreaga&LucasZavala(2016)Exportpromotionandfirmentryintoandsurvivalinexportmarkets,CanadianJournalofDevelopmentStudies/Revuecanadienned'étudesdudéveloppement,37:2,142-158,DOI:10.1080/02255189.2016.1131671

Feenstra,RobertC."Integrationoftradeanddisintegrationofproductionintheglobaleconomy."TheJournalofEconomicPerspectives,12.4(1998):31-50.

Gil,S.,Llorca,R.,Serrano,J.A.M.(2007)“MeasuringtheImpactofRegionalExportPromotion:theSpanishCase”,PapersinRegionalScience,87(1),139-146.

Heckman,J.,Ichimura,H.,Todd,P.,1997.‘Matchingasaneconometricevaluationestimator:evidencefromevaluatingajobtrainingprogramme’,ReviewofEconomicStudies,64,605–654.

Hummels,D.,Ishii,J.andYi,K-M.(2001)"Thenatureandgrowthofverticalspecializationinworldtrade."JournalofinternationalEconomics,54.1(2001):75-96.

Johnson,R.C.“FiveFactsaboutValue-AddedExportsandImplicationsforMacroeconomicsandTradeResearch.”TheJournalofEconomicPerspectives,vol.28,no.2,2014,pp.119–142.

Johnson,R.C.,andGuillermoNoguera."Aportraitoftradeinvalueaddedoverfourdecades."UnpublishedManuscript,DartmouthCollege(2014).

Johnson,R.C.,andGuillermoNoguera."Accountingforintermediates:Productionsharingandtradeinvalueadded."JournalofinternationalEconomics,86.2(2012):224-236.

Kaiser,U.andKongsted,H.C.,2008.Trueversusspuriousstatedependenceinfirmperformance.EmpiricalEconomics,35(2),pp.207-228.

Lach,S.(2002).DoR&DSubsidiesStimulateorDisplacePrivateR&D?EvidencefromIsrael.TheJournalofIndustrialEconomics,50(4),369-390.

Lederman,D.,M.OlarreagaandL.Payton(2010).‘Exportpromotionagencies:dotheywork?’,JournalofDevelopmentEconomics,91,257–65.

Padmaja,M.andSasidharan,S.,2016.SunkCosts,FirmHeterogeneity,ExportMarketEntryandExit:EvidencefromIndia.JournalofQuantitativeEconomics,pp.1-27.

Spence,M.M.(2003)“EvaluatingExportPromotionProgrammes:U.K.OverseasTradeMissionsandExportPerformance”,SmallBusinessEconomics,20(1),83-103.

Timoshenko,OlgaA."Learningversussunkcostsexplanationsofexportpersistence."EuropeanEconomicReview79(2015):113-128.

VanBiesebroeck,J.,Konings,J.andVolpeMartincus,C.(2016)‘Didexportpromotionhelpfirmsweatherthecrisis?’EconomicPolicy,31,653-702.

VolpeMartincus,C.andCarballo,J.(2008)“IsExportPromotionEffectiveinDevelopingCountries?Firm-levelEvidenceontheIntensiveandExtensiveMarginsofExports”,JournalofInternationalEconomics,76(1),89-106.

VolpeMartincus,C.andCarballo,J.(2010a)“BeyondtheAverageEffects:TheDistributionalImpactsofExportPromotionProgramsinDevelopingCountries”,JournalofDevelopmentEconomics,92(2),201-214.

VolpeMartincus,C.andCarballo,J.(2010b)“ExportPromotion:BundledServicesWorkBetter”,TheWorldEconomy,33(12),1718-1756.

VolpeMartincus,C.andCarballo,J.(2010c)“EnteringNewCountryandProductMarkets:DoesExportPromotionHelp?”,ReviewofWorldEconomy,146,437-467.

VolpeMartincus,C.andCarballo,J.(2012)“ExportPromotionActivitiesinDevelopingCountries:WhatKindofTradeDoTheyPromote”,TheJournalofInternationalTrade&EconomicDevelopment,21(4),539-578.

VolpeMartincus,C.,Carballo,J.andGallo,A.(2011)“TheImpactofExportPromotionInstitutionsonTrade:IsIttheIntensiveortheExtensiveMargin?”,AppliedEconomicsLetter,18,127-132.

VolpeMartincus,C.,Estevadeordal,A.,Gallo,A.andLuna,J.(2010a)“InformationBarriers,ExportPromotionInstitutionsandtheExtensiveMarginofTrade”,ReviewofWorldEconomy,146,91-111.

Wagner,J.(2007).Exportsandproductivity:Asurveyoftheevidencefromfirm‐leveldata.TheWorldEconomy,30(1),60-82.