A replication of "The role of intermediaries in facilitating trade" (Journal of International Economics, 2011)

Duan, Jianhua; Xue eng, Qian; Das, Kun al K.; Me iluo o, Lau a; Reed, W. Robe Wo king Pape A eplica ion o "The ole o in e media ies in acili a ing ade" (Jou nal o In e na ional Economics, 2011) Economics Discussion Pape s, No. 2019-66 P o ided in Coope a ion wi h: Kiel Ins i u e o he Wo ld Economy – Leibniz Cen e o Resea ch on Global Economic Challenges Sugges ed Ci a ion: Duan, Jianhua; Xue eng, Qian; Das, Kun al K.; Me iluo o, Lau a; Reed, W. Robe (2019) : A eplica ion o "The ole o in e media ies in acili a ing ade" (Jou nal o In e na ional Economics, 2011), Economics Discussion Pape s, No. 2019-66, Kiel Ins i u e o he Wo ld Economy (I W), Kiel This Ve sion is a ailable a : h ps://hdl.handle.ne /10419/209126 S anda d-Nu zungsbedingungen: Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen Zwecken und zum P i a geb auch gespeiche und kopie we den. Sie dü en die Dokumen e nich ü ö en liche ode komme zielle Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich machen, e eiben ode ande wei ig nu zen. So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen (insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en, gel en abweichend on diesen Nu zungsbedingungen die in de do genann en Lizenz gewäh en Nu zungs ech e. Te ms o use: Documen s in EconS o may be sa ed and copied o you pe sonal and schola ly pu poses. You a e no o copy documen s o public o comme cial pu poses, o exhibi he documen s publicly, o make hem publicly a ailable on he in e ne , o o dis ibu e o o he wise use he documen s in public. I he documen s ha e been made a ailable unde an Open Con en Licence (especially C ea i e Commons Licences), you may exe cise u he usage igh s as speci ied in he indica ed licence. h ps://c ea i ecommons.o g/licenses/by/4.0/ Discussion Pape No. 2019-66 | No embe 19, 2019 | h p://www.economics-ejou nal.o g/economics/discussionpape s/2019-66 A eplica ion o “The ole o in e media ies in acili a ing ade” (Jou nal o In e na ional Economics, 2011) Jianhua Duan, Xue eng Qian, Kun al K. Das, Lau a Me iluo o, and W. Robe Reed Abs ac This s udy eplica es Ahn, Khandelwal, and Wei’s (AKW´s) (2011) model o in e media y ade. The au ho s´ s udy p oduces wo main esul s. Fi s , he au ho s a e able o ep oduce empi ical e idence o AKW’s h ee main p edic ions o Chinese expo s. This is imp essi e because much o he da a o hei eplica ion a e independen ly sou ced. Howe e , when hey subjec AKW´s model o addi ional es s, hey ind ha he e idence is no obus . Using mo e ecen ly a ailable da a o es AKW´s i s p edic ion, he au ho s es ima e coe icien s ha a e w ong-signed and signi ican . When hey e-analyze he e idence suppo ing AKW´s second and hi d p edic ions, hey ind ha he ull sample esul s mask signi ican he e ogenei y ac oss Chinese egions. In many cases, key coe icien s a e insigni ican . In a ew cases, hey a e w ong-signed and signi ican . Finally, using mul iple e sions o a key a iable measu ing he numbe o equi ed impo documen s by coun y, hey disco e ha he esul s a e no obus ac oss e sions. The pape ´s da a se : h ps://doi.o g/10.7910/DVN/BT8JPN (Replica ion S udy) JEL F1 Keywo ds In e media ies; expo s; p oduc i i y; he e ogeneous i ms; China Au ho s Jianhua Duan, Uni e si y o Can e bu y, Ch is chu ch, New Zealand Xue eng Qian, Zhongnan Uni e si y o Economics and Law, Wuhan, China Kun al K. Das, Uni e si y o Can e bu y, Ch is chu ch, New Zealand Lau a Me iluo o, Uni e si y o Can e bu y, Ch is chu ch, New Zealand W. Robe Reed, Uni e si y o Can e bu y, Ch is chu ch, New Zealand, bob. eed@can e bu y.ac.nz The au ho s g a e ully acknowledge assis ance om JaiBin Ahn, Ami K. Khandelwal, and Shang-Jin Wei. They p o ided hei p og amming code, pa ien ly answe ed he au ho s´ que ies abou hei da a, and sha ed hei a i and impo p ocedu es da a wi h hem. Remaining e o s a e he au ho s own. Ci a ion Jianhua Duan, Xue eng Qian, Kun al K. Das, Lau a Me iluo o, and W. Robe Reed (2019). A eplica ion o “The ole o in e media ies in acili a ing ade” (Jou nal o In e na ional Economics, 2011). Economics Discussion Pape s, No 2019-66, Kiel Ins i u e o he Wo ld Economy. h p://www.economics-ejou nal.o g/economics/discussionpape s/2019-6 Recei ed Oc obe 10, 2019 Accep ed as Economics Discussion Pape Oc obe 10, 2019 Published No embe 19, 2011 © Au ho (s) 2019. Licensed unde he C ea i e Commons License - A ibu ion 4.0 In e na ional (CC BY 4.0) 1 I. INTRODUCTION In he widely acclaimed book, S a is ical In e ence as Se e e Tes ing, Debo ah Mayo (2018) a gues ha a oo cause o science’s ep oducibili y p oblem is ha scien i ic claims a e no s ongly es ed. She ad oca es o “se e e es ing”, whe e “A claim is se e ely es ed o he ex en i has been subjec ed o and passes a es ha p obably would ha e ound laws, we e hey p esen ” (Mayo, 2018, p. xii). Replica ion can play ha ole bo h by assessing he e idence in he o iginal s udy, and subjec ing i s claims o s ong es s. This pape eplica es Ahn, Khandelwal, and Wei (AKW), “The ole o in e media ies in acili a ing ade” (Jou nal o In e na ional Economics, 2011). AKW build on he seminal wo k o Meli z (2003) linking expo beha iou o he e ogeneous i m p oduc i i y. Whe eas he expo ing decision in Meli z (2003) consis s o o-expo o no - o-expo , AKW build in a hi d op ion, expo ing h ough in e media ies. AKW is impo an because i p o ides bo h a heo e ical ounda ion o he exis ence o ade in e media ies, and suppo ing empi ical e idence. While o he heo e ical models o in e media ion exis (c . Ake man, 2018; Ponce and Xu, 2018; Blum, Cla o, & Ho s mann, 2018; and Chan, 2018), AKW was he i s o be published in a majo jou nal and is he mos ci ed. A he ime o his w i ing, i has been ci ed 134 imes in Web o Science, and 541 imes in Google Schola . Two main esul s ollow om ou eplica ion. Fi s , we a e able o ep oduce he empi ical e idence ha AKW p o ide in suppo o hei model. This is imp essi e because we had o independen ly sou ce much o he da a used by AKW. Howe e , when he da a a e e-analysed and upda ed, we ind ha empi ical suppo o hei model is subs an ially diminished. We no e ha we see ou ole as independen audi o s, nei he p edisposed o p oduce e idence in a o o agains AKW’s model. One o he ad an ages o submi ing ou 2 eplica ion esea ch o Economics: The Open Access, Open Assessmen E-Jou nal is ha he jou nal s a es ha i “will publish bo h con i ma ions and discon i ma ions o o iginal s udies. The only conside a ion will be quali y o he eplica ing s udy.”1 Thus we do no ha e an incen i e o che y-pick e idence in suppo o a gi en esul . We also no e ha we ha e consul ed wi h he o iginal au ho s: JaeBin Ahn, Ami Khandelwal and Shang-Jin Wei. AKW we e e y g acious in sha ing hei da a and code wi h us. In addi ion, hey pa ien ly answe ed ou que ies and ga e us commen s on ou esea ch. While we ha e some disag eemen s wi h hem ega ding he in e p e a ion o ou esul s, we y ou bes o gi e oice o hei conce ns in ou discussion below. They will ha e he oppo uni y o aise addi ional conce ns in a eply. Ou s udy p oceeds as ollows. Sec ion II p o ides he empi ical and heo e ical con ex o AKW’s esea ch on in e media ion. Sec ion III b ie ly summa izes AKW’s model o ade in e media ion so ha he eade can unde s and how hei empi ical p edic ions de i e om he model. Sec ion IV desc ibes ou da a. Sec ion V epo s ou a emp o ep oduce he key e idence p o ided by AKW in suppo o hei model. Sec ions VI h ough IX p esen and discuss addi ional es esul s om e-analyzing and ex ending AKW’s da a. Sec ion X concludes. II. EMPIRICAL AND THEORETICAL CONTEXT In e media ion is widely ecognized as playing a p ominen ole in wo ld ade ma ke s. In he ea ly 1980’s, 300 Japanese ade s (non-manu ac u ing i ms) accoun ed o 80% o Japan’s ade (Rossman, 1984). Spulbe (1996) documen s ha in 1995, in e media ies accoun ed o abou a qua e o he GDP o he U.S., and close o wo million i ms ope a ed in he U.S. in e media ion indus y. In 2002, Ame ican in e media ies accoun ed o 44% and 56% o expo ing and impo ing i ms, and 11% and 24% o expo and impo 1 This s a emen is aken om he jou nal’s websi e: h p://www.economics-ejou nal.o g/special- a eas/ eplica ions-1. 3 alue, espec i ely (Be na d, Jensen, Redding, & Scho , 2010). The economies o Hong Kong, Singapo e, and Dubai ha e de eloped specialised expe ise in en epô ade, which has g ea ly con ibu ed o hei economic de elopmen (Feens a & Hanson, 2004). Fo hese easons and mo e, in e media ion has ecei ed much empi ical a en ion (An às & Cos ino , 2011; Blum, Cla o, & Ho s mann, 2018; Qu, Ra , & Schmi , 2014; Tang & Zhang, 2012). The li e a u e on in e media ies in in e na ional ade makes wo impo an obse a ions. Fi s , a signi ican ac ion o in e na ional ade is channelled h ough in e media ies. Second, he e a e sys ema ic a ia ions in he mode o expo s no only ac oss i ms wi hin an indus y, bu also ac oss indus ies. Fundamen al o he ole o in e media ies is connec ing buye s and selle s wo ldwide. A a ie y o easons has been sugges ed o why in e media ies exis . One eason is ha hey mi iga e unce ain y ega ding demand and supply, o when buye and selle cha ac e is ics a e unobse able (Spulbe , 1996). Ano he eason is ha ade in e media ies pe o m quali y assu ance. Unce ain y abou p oduc quali y c ea es he amilia p oblems o ad e se selec ion and mo al haza d. T ade in e media ies can alle ia e his p oblem by sc eening he quali y o he p oduc s and e ealing i o he cus ome s (Dasgup a & Mond ia, 2018). In e media ies can also acili a e expo ac i i y by p o iding a c edi -cons ained i m a channel h ough which capi al ma ke ic ions a e educed, hus enhancing he gains om ade (Chan, 2018). In e media ies a e no always seen as bene icial. An as and Cos ino (2011) p esen a model wi h sea ch ic ions whe e in e media ies p o ide ma ke access o a me s. They show ha depending on he kind o ma ke in eg a ion being conside ed, in e media ion can lowe he wel a e o a me s in de eloping coun ies. Simila ly, She ele a and K ishna (2017) show how a hold-up p oblem be ween a me s and in e media ies, a ising om con ac ual incomple eness, leads o he nonexis ence o ma ke s o ce ain ag icul u al 4 goods. While much o he li e a u e has ocussed on in e media ies as acili a ing in e na ional ma ches, AKW a gue ha in e media ies exis p ima ily o o e come ade cos s. Pa o he appeal o AKW’s model is ha i makes h ee s aigh o wa d p edic ions ha a e s ongly suppo ed in hei empi ical analysis. III. AKW’s MODEL OF TRADE INTERMEDIATION AND THREE PREDICTIONS The model. FIGURE 1 ep esen s he main ideas unde lying AKW’s heo y o in e media ed ade. The e ical axis measu es p o i s om expo ing (𝜋). The ho izon al axis measu es i m p oduc i i y (𝜑𝜎−1). No e ha 𝜋= 0 i he i m does no expo , and coincides wi h he ho izon al axis. The s eep do ed line (A) iden i ies he p o i -p oduc i i y nexus o he i m i i sells i s ou pu o an in e media y ha hen expo s i o e seas. The ini ial cos o con ac ing wi h he in e media y is 𝑓 𝑖. The in e media y hen epackages he i m’s ou pu and sells i in N o e seas ma ke s. The associa ed inc ease in p o i s esul ing om an inc ease in p oduc i i y is gi en by he slope o A. The poin a which A c osses he ho izon al axis iden i ies he minimal p oduc i i y (𝜑𝑖 𝜎−1) a which he i m will swi ch om selling only in he domes ic ma ke o also expo ing ia an in e media y. A i m can choose o di ec ly expo o a speci ic coun y j e en while selling o o he coun ies ia an in e media y. This decision is ep esen ed by lines B and C. Line B ep esen s he p o i -p oduc i i y nexus o indi ec ly expo ing o coun y j h ough an in e media y. Since he i m has al eady con ac ed wi h he in e media y, i does no need o pay any addi ional ixed cos o (indi ec ) expo s o j. B is la e han A because A cap u es p o i s om indi ec expo s o all coun ies, while B only cap u es p o i s om coun y j. I , ins ead, he i m chooses o di ec ly expo o j, i mus bea all he ixed cos s o en e ing ha ma ke and inding buye s. This is ep esen ed by 𝑓 𝑥 𝑗. The ewa d o doing so is ha i enjoys highe ma ginal p o i s. The in e media y mus epackage and eb and he 5 p oduc , which aises ma ginal cos s, and hence, p ice, in he o eign ma ke . This highe p ice dec eases he quan i y demanded, which esul s in lowe p o i s o he i m. The i m ha di ec ly expo s i s p oduc a oids hese cos s. This enables i o sell i s p oduc a a lowe p ice o e seas, which inc eases p o i s. As a esul , he slope o line C is g ea e han he slope o line B. The p oduc i i y whe e he wo lines in e sec (𝜑𝑥 𝑗𝜎−1 ) iden i ies he h eshold whe e a i m ansi ions om indi ec ly selling i s p oduc o coun y j o di ec ly selling o j. Th ee p edic ions. AKW use he p eceding model o de i e he ollowing h ee p edic ions: P edic ion #1: “…we would expec a hockey s ick ela ionship be ween p oduc i i y and di ec expo s – only high p oduc i i y fi ms di ec ly expo while low and in e media e p oduc i i y fi ms do no – and an in e ed U-shape ela ionship wi h indi ec expo s” (page 78). As seen in FIGURE 1, he model p edic s ha inc eases in p oduc i i y will ha e no e ec on di ec expo s o low p oduc i i y i ms. Inc eases in p oduc i i y o high p oduc i i y i ms will gene a e linea inc eases in p o i s, gi ing a “hockey s ick” ela ionship. In con as , inc eases in p oduc i i y will i s gene a e an inc ease in indi ec expo s as i ms ansi ion om no expo s o expo ing ia an in e media y. Howe e , u he inc eases in p oduc i i y a e p edic ed o dec ease indi ec expo s as i ms ansi ion om expo ing ia an in e media y o di ec ly expo ing, gi ing an in e ed U-shape ela ionship. P edic ion #2: “Expo s by in e media ies should be mo e expensi e han di ec expo e s” (page 79). This p edic ion ollows di ec ly om he ollowing assump ion o he model: “In e media ies pu chase a ie ies om manu ac u e s a he same p ice as domes ic consume s ( he e is no p ice disc imina ion) and incu an addi ional ma ginal cos o selling hese a ie ies ab oad. This addi ional ma ginal cos cap u es e-labeling, packaging and o he pe -uni cos s associa ed wi h aking he i le o a ie ies om he manu ac u e s. The p ice o indi ec ly 6 expo ed a ie ies is he e o e highe han he p ice o di ec ly expo ed a ie ies by his ac o ” (page 75). P edic ion #3: “… he sha e o expo s h ough in e media ies is la ge in coun ies wi h (i) smalle ma ke size, (ii) highe a iable ade cos s, o (iii) highe fixed cos s o expo ing” (page 76). This p edic ion also ollows om FIGURE 1. Ce e is pa ibus, smalle des ina ion ma ke s o highe a iable ade cos s cause he slopes o lines C and B o swing down. Howe e , o any posi i e p oduc i i y le el, line C will swing down by mo e han line B. The esul is ha he h eshold om swi ching om indi ec o di ec expo ing (𝜑𝑥 𝑗𝜎−1) mo es o he igh , hus inc easing he sha e o expo s h ough in e media ies. Highe ixed cos s associa ed wi h expo ing shi s line C down, wi hou a ec ing line B. This again causes 𝜑𝑥 𝑗𝜎−1 o mo e igh , inc easing he sha e o indi ec expo s. IV. DATA Mos o he da a used o es AKW’s h ee p edic ions a e d awn om wo sou ces: China Cus oms Da a and En e p ise Su ey Da a (Chinese i ms only). The cus oms da a eco d in e na ional ade in o ma ion o indi idual Chinese i ms o e he 2000-2005 pe iod. They con ain de ailed in o ma ion o each i m-p oduc -pa ne ansac ion, including in o ma ion on p oduc ype, coun y des ina ion, and he p ice and quan i y o he ansac ions. The ex ensi e co e age o he China Cus oms Da a allows a ela i ely comple e eco d o ade ansac ions a he coun y le el. The China Cus oms Da a also eco ds in o ma ion abou he i m, including i s name. AKW exploi a Chinese naming con en ion o iden i y in e media y i ms. As hey desc ibe i , “We iden i y he se o in e media y fi ms based on Chinese cha ac e s ha ha e he English-equi alen meaning o “impo e ”, “expo e ”, and/o “ ading” in he fi m's name. A use ul ea u e abou fi m names in China is ha hey a e o en e y desc ip i e (a con en ion ha migh be aced o a ime when he coun y was unde cen al planning and he planne s a o ed desc ip i e company names). Many fi ms ounded du ing he pos -1980 e o m e a con inue o adop his naming con en ion” (pages 76 .). 7 While AKW’s app oach allows hem o dis inguish in e media y i ms om o he i ms in he China Cus oms Da a (“di ec i ms”), i does no allow hem o iden i y how much o di ec i ms’ ou pu is sold o in e media ies. A key aspec o AKW’s model is he decision by i ms o di ec ly expo hei p oduc o e seas e sus indi ec ly expo ing h ough in e media ies. A co e p edic ion o he model ela es how i m p oduc i i y a ec s he sha e o o al sales going o di ec o indi ec expo s. To add ess his issue, AKW u n o he Wo ld Bank’s En e p ise Su eys. The En e p ise Su ey Da a use “s anda d su ey ins umen s o collec i m-le el da a on he business en i onmen om business owne s and op manage s”.2 The su ey co e s a wide ange o subjec a eas, and includes da a om many coun ies. While esponses a e collec ed a he i m-le el, hey a e only a ailable o esea che s a he coun y-le el. A majo ad an age o he En e p ise Su ey Da a is ha i epo s he p opo ions o an es ablishmen ’s sales expo ed (i) di ec ly and (ii) h ough a dis ibu o . The su ey also eco ds a ious measu es o p oduc i i y. Su ey da a a e a ailable o he yea s 2002/2003 and 2012, he las o which was una ailable o AKW a he ime o hei w i ing. To es P edic ion #3, AKW employ a numbe o di e en a iables o cap u e ma ke cha ac e is ics. Dis ance om ading pa ne , Mos Fa ou ed Na ion (MFN) a i , size o e hnic Chinese popula ion in he des ina ion coun y, and ma ke size (measu ed by GDP) a e hypo hesized o p oxy a iable cos s. Inc eases in he i s wo a e assumed o aise a iable cos s, and inc eases in he la e wo a e assumed o lowe cos s. Impo ing p ocedu es, measu ed as he numbe o documen s equi ed o impo ing, a e assumed o be posi i ely associa ed wi h ixed cos s. AKW g aciously sha ed hei p og amming code o assis us in ou eplica ion p ojec . They we e unable o sha e hei China Cus oms Da a wi h us because i is p op ie a y. When 2 Taken om h ps://www.en e p isesu eys.o g/po al/login.aspx (accessed on 9 Janua y 2019). 14 di ec ading igh s. In he second yea a e accession, he minimum capi al equi emen o di ec ading was RMB 3 million, and his ell o RMB 1 million by 2004. Howe e , da a om he Wo ld Bank's En e p ise Su ey o China ha co e s 2002 and 2003 indica e ha fi ms below his cu o epo ed di ec expo s. This could be because expo licenses we e only equi ed o a limi ed se o p oduc s and/o because hese cu o s we e no s ingen ly applied, a leas o expo s. By 2005, any fi m ha wished o di ec ly ade wi h o eign pa ne s was ee o do so. So while we a e confiden ha he licenses will no a ec he in e p e a ion o ou esul s, he main analysis uses da a o 2005 when he licenses had been emo ed” (page 77). In o he wo ds, p io o 2005, i ms we e equi ed o ha e go e nmen -issued licenses in o de o expo . The e is e idence ha his equi emen was no s ic ly en o ced. Fu he , i only applied o a subse o p oduc s.4 Thus, AKW a e com o able using he 2002/2003 En e p ise Su ey Da a. Ne e heless, expo da a a e 2004 is p e e able because by 2005 i ms we e ee o expo wi hou license es ic ions. While we ecognize ha he p e-2005 da a is in e io o he 2005 da a, we belie e he e is alue in in es iga ing hese ea lie da a. We base his on he ac ha AKW also use da a om hese ea lie yea s as e idence in suppo o hei heo y (see TABLE 2). Expanding he analysis o es P edic ion #2 using p e-2005 da a allows addi ional es s o AKW’s model. The las i e ows o TABLE 5 epo he esul s. The e a e i e coe icien s co esponding o he in e media y dummy a iable o he yea 2000 o 2004 eg essions. Two p o ide suppo o AKW’s p edic ion, being posi i e and signi ican (2001 and 2002). O he emaining h ee, one is w ong-signed and insigni ican (2004), and wo a e w ong-signed and signi ican a he 1 pe cen le el (2000 and 2003). Summa y. TABLE 5 p o ides eigh “addi ional” es s o AKW’s second p edic ion. “Addi ional” is in quo a ions because he h ee 2005 subsamples a e no independen o he ull 2005 sample epo ed in TABLE 2. Ne e heless, only h ee o he eigh eg essions p oduce esul s ha con i m AKW’s model. Fou o he eigh p oduce esul s ha con adic AKW’s model. The emaining es is inconclusi e, since he associa ed coe icien is s a is ically 4 AKW no e in Foo no e 17, page 78: “While he e we e some es ic ions o ading du ing his pe iod, hey we e limi ed o only a subse o p oduc s 15 insigni ican . Quali ica ions. One conce n wi h ou obus ness checks ela es o ou decision o di ide he sample in o h ee geog aphical samples: Eas , Cen al, and Wes . To add ess conce ns o che y-picking esul s, we should ha e comple ed a p e-analysis plan. We did no do ha . In ou de ense, we a e s ill lea ning Open Science me hods and a e only now beginning he p ac ice o de eloping, and publicly pos ing, p e-analysis plans in ou esea ch. We again no e ha we did no app oach ou eplica ion wi h he goal o ying o e u e AKW’s indings. Ra he , we a emp ed o be objec i e, independen audi o s. Fu he , we no e ha i is qui e common o di ide samples by geog aphical egion. Fo example, g ow h s udies will o en epo esul s o subsamples by con inen (e.g., A ica, Sou h Ame ica, Asia); o by s a e o economic de elopmen (e.g., OECD e sus non-OECD coun ies; de eloped e sus de eloping coun ies). Thus, i seemed na u al o us o di ide he Chinese sample in o Eas , Cen al, and Wes egions. AKW’s analysis comp ised all h ee subsamples, indica ing ha hei model was applicable o i ms in all h ee egions. Sepa a ing ou hei sample in o h ee subsamples hus should no be a p oblem. Ano he conce n is ou decision o use p e-2005 da a China Cus oms Da a. AKW make a con incing case ha he 2005 da a is be e sui ed o es ing hei heo y because o he exis ence o licensing equi emen s. As hey no e, licensing equi emen s we e educed in s eps o e he pe iod 2003-2005. To he ex en licensing es ic ions wo ked agains inding suppo o AKW’s heo y, he esul s in a o o hei heo y should be s onge as one mo es om 2002 o 2004. Howe e , ha is no he pa e n e iden om TABLE 5. The “bes ” yea s in e ms o suppo o P edic ion #2 a e 2001 and 2002. Coe icien es ima es om 2003 and 2004 a e w ong-signed, wi h he es ima e om 2003 being signi ican a he 1-pe cen le el. VIII. ROBUSTNESS CHECK: P edic ion #3 Desc ip ion. The hi d and mos impo an o AKW’s p edic ions is ha he sha e o ade 16 ca ied by in e media ies is sys ema ically ela ed o coun y cha ac e is ics. As e idence in a o o hei model, AKW epo ha he sha e o ade accoun ed by in e media ies is posi i ely ela ed o dis ance, a i s, and numbe o equi ed impo documen s; and nega i ely ela ed o GDP and he size o he Chinese popula ion in he des ina ion coun y. Resul s. The i s se o addi ional es s o P edic ion #3 epea s AKW’s analysis o TABLE 3, his ime di iding he ull sample in o Eas , Cen al, and Wes subsamples. The esul s a e epo ed in Panel A o TABLE 6. Ou benchma k speci ica ion is AKW’s ull model wi h Log Dis ance, Log GDP, Log Chinese Popula ion, # o Impo ing P ocs, MFN Ta i , and HS6 ixed e ec s all included in he same eg ession (c . Column 4.a in TABLE 3). The es ima ed coe icien s o he coun y cha ac e is ics a e epo ed ow-wise, wi h he a iable p edic ions lis ed a he op o he columns. Fo each o he i e coun y cha ac e is ic a iables, AKW’s p edic ion is es ed by h ee es ima es, one o each subsample. Fo dis ance and GDP, AKW’s p edic ion is success ul in h ee ou o h ee es s. In con as , AKW’s p edic ion is success ul in only one ou o h ee es s o he o he coun y cha ac e is ics. This is a esul we will encoun e equen ly in he es s ahead: s ong suppo o AKW’s P edic ion #3 o dis ance and GDP, bu weak suppo o Chinese popula ion, numbe o impo ing p ocedu es, and a i s. Ou nex se o es s ocuses on di e en measu es o he numbe o equi ed impo ed p ocedu es a iable. AKW epo ha hey sou ced hese da a om he Wo ld Bank's Doing Business Repo 2006, and ha he a iable measu es “ he numbe o p ocedu es equi ed o impo ing a con aine ” (page 80). Ou own esea ch iden i ied wo e sions o he Doing Business Repo 2006, a p in ed book e sion and an online e sion. These we e bo h di e en om each o he , and di e en om AKW’s da a. TABLE 7 epo s summa y s a is ics o he h ee e sions o he numbe o equi ed impo documen s a iable. While he h ee e sions a e simila , he summa y s a is ics e eal 17 signi ican di e ences. This is mos appa en in he pai wise compa isons epo ed in he las wo columns. A compa ison o he AKW and Book da a e eal ha 61.6% o he co esponding alues in he wo da a se s a e di e en om each o he . 72.2% o he alues in he AKW and Online da a se s a e di e en , and 85.4% o he alues in he Book and Online e sions a e di e en . Acco dingly, we in es iga e whe he hese di e ences a ec he es s o AKW’s model. Panel B o TABLE 6 epo s he esul s o subs i u ing he book and online e sions o he numbe o equi ed impo documen s o AKW’s e sion in he ull sample. Focusing i s on he numbe o p ocedu es a iable, he es ima ed coe icien o he p in ed book e sion is co ec ly signed, bu insigni ican . The co esponding es ima e o he online e sion is i ually iden ical o he es ima e using AKW’s da a (see Column 4.b in TABLE 3). The es s o he o he coun y cha ac e is ics a e quali a i ely una ec ed by using al e na i e e sions o he equi ed documen s a iables, wi h all es ima es con i ming AKW’s model. Panel C epea s he p eceding analysis, excep ha i di ides he ull sample in o egional subsamples. This p oduces a o al o six es s o each a iable. As we ha e consis en ly seen p e iously, he es ima es o dis ance and GDP con inue o suppo AKW’s model. In con as , he es ima es o Chinese popula ion, numbe o p ocedu es, and a i s is subs an ially weake , wi h only wo ou o six, one ou o six, and ou ou o six o he es ima es p o iding e idence in a o o AKW’s p edic ions, espec i ely. The las panel exploi s he ac ha we ha e Chinese Cus oms Da a o he yea s 2000 o 2004. Un o una ely, we do no ha e co esponding annual da a o Chinese popula ion, numbe o equi ed impo documen s, o MFN a i s. The a iable measu ing Chinese popula ion is ob ained om he Ohio Uni e si y Shao Cen e. This a iable is collec ed in di e en egions in di e en ime pe iods, and because we do no ha e ime se ies da a o his a iable, we a e o ced o assume he popula ion da a a e equally applicable o all yea s in he 18 2000-2005 pe iod. We a e con on ed wi h simila da a p oblems o he o he wo a iables. The numbe o equi ed impo documen s is only a ailable in he 2006 edi ion o Doing Business Repo s, which epo s in o ma ion as i was in Janua y 2005. Likewise, we only ha e MFN a i da a a he coun y, HS6 le el o 2005. Wi h espec o he la e , we no e ha China’s admission o he WTO in 2001 would no ha e had an impac on Chinese a i s. China had al eady gained access o MFN a i s in 1980. Admission o he WTO gua an eed pe manen MFN s a us, bu i did no by i sel a ec he a i s paid by Chinese expo e s du ing ou sample pe iod. In any case, in es ing AKW’s heo y o P edic ion #3, we a e o ced o assume ha he impo p ocedu es da a and MFN a i da a ha we ha e o 2005 a e applicable o he ea lie yea s as well. Wi h hese ca ea s in mind, Panel D epo s he esul s o es ima ing he benchma k speci ica ion o he yea s 2000-2004, using he h ee di e en measu es o numbe o impo documen s (AKW, Book, and Online). The esul is a o al o 15 es s o each a iable. Once again, he coe icien s o dis ance and GDP ha e he p edic ed sign and a e signi ican in e e y eg ession. Wi h espec o he Chinese popula ion coe icien , he esul s con inue o be a o able o AKW’s model when using hei measu e o equi ed documen s. The espec i e coe icien is co ec ly signed in ou o he i e eg essions (2000-2003). When al e na i e e sions o he equi ed documen s a e subs i u ed o AKW’s e sion, his alls o wo ou o i e (2000 and 2003 o bo h he Book and Online e sions). Fo he equi ed documen s coe icien , bo h AKW’s and he Online da a p o ide s ong suppo o AKW’s model o he yea s 2000-2004, p oducing co ec ly-signed, signi ican coe icien s in ou ou o i e and i e ou o i e cases. In con as , when he da a o numbe o equi ed impo documen s come om he p in ed book e sion, none o he i e es ima es p o ides suppo ing e idence because all a e 19 s a is ically insigni ican . Las ly, he esul s o he a i a iable a e easy o summa ize: None o he 15 es ima es p o ide suppo o AKW’s model. In ac , wo o he 15 es ima es a e w ong-signed and signi ican a he 5-pe cen le el (2002/AKW and 2002/Online). Summa y. The las wo ows o he able summa ize he esul s om he addi ional es s o AKW’s P edic ion #3. The eco d wi h espec o dis ance and GDP is pe ec . In e e y case, he es ima ed coe icien s line up wi h AKW’s p edic ions and a e s a is ically signi ican . The eco d wi h espec o he o he h ee coun y cha ac e is ics is gene ally no suppo i e: In hal o less o he es s, he es ima ed coe icien s do no suppo AKW’s p edic ion. Howe e , i can be easonably a gued ha one should no place oo much weigh on he 2000-2004 esul s o hese a iables because hese assume ha he 2005 alues a e also applicable o ea lie yea s. Acco dingly, we ecalcula e he success a es, emo ing he esul s om Panel D. Ra he han 13/26, 12/26, and 7/26 o he es ima es suppo ing he p edic ions o Chinese popula ion, numbe o equi ed documen s, and a i s; he numbe s a e 5/11, 3/11, and 7/11. In o he wo ds, e en wi hou he es esul s om Panel D, he o e all esul s a e gene ally unsuppo i e o AKW’s hi d p edic ion. Quali ica ions. As no ed abo e, ou analysis was o ced o subs i u e 2005 alues o Chinese popula ion, numbe o equi ed impo documen s, and MFN a i s in he annual eg essions o ea lie yea s. This in oduces measu emen e o in ou analysis, which would bias he coe icien s owa ds ze o. This could explains why he e a e so ew, co ec ly signed es ima es o hese a iables. Consis en wi h his is ha he e a e no inco ec ly signed, signi ican es ima es o #Chinese and #P ocedu es in any o he annual eg essions om 2000- 2004. On he o he hand, his does no explain he wo inco ec ly signed, signi ican es ima es o Ta i s in hese ea lie yea s (Panel D/2002/AKW and Panel D/2002/Online). No does i explain he insigni ican es ima es o all h ee a iables in he geog aphical subsample eg essions o Panel A. 20 Wi h espec o he h ee di e en measu es o numbe o equi ed impo documen s (AKW/Book/Online), he e is some eason o p e e he esul s om he Online measu e o his a iable, because he Wo ld Bank Doing Business Da abase s a es ha he online da a a e con inuously e ised as new in o ma ion becomes a ailable.5 On he o he hand, he esul s using he Online measu e a e only ma ginally be e han hose using he Book measu e. IX. OVERALL ASSESSMENT OF THE ROBUSTNESS CHECKS TABLE 8 pulls oge he all he es esul s o AKW’s h ee p edic ions, bo h (i) ep oduc ion and (ii) e-analysis and ex ension. Ou ep oduc ion o AKW con i ms AKW’s conclusion ha he empi ical analysis p o ides s ong suppo o hei model. 6/9 es esul s suppo P edic ion #1, 3/3 es esul s suppo P edic ion #2, and 13/14 es esul s suppo P edic ion #3. Howe e , when we e-analyze and ex end he analysis, di iding he da a by geog aphical egions, using di e en e sions o numbe o equi ed impo documen s, and ex ending he da a o addi ional yea s, he e idence becomes much weake . Wi h espec o P edic ion #1, less han hal o he p edic ions a e suppo ed by he da a (4/9). Likewise o P edic ion #2: only h ee ou o eigh p edic ions a e suppo ed. AKW iden i y P edic ion #3 as hei “cen al p edic ion”. He e he esul s a e somewha be e , wi h 84 ou o 130 p edic ions suppo ing he heo y (37 ou o 55 i one excludes he es esul s om Panel D o TABLE 6). Howe e , hese la e esul s mask an impo an di e ence wi hin he i e coun y cha ac e is ics examined by AKW. While he es esul s o dis ance and GDP s ongly suppo AKW’s heo y, hose o he emaining h ee coun y cha ac e is ics do no . Fo Chinese popula ion, numbe o impo p ocedu es, and a i s, he esul s gene ally do no suppo hei p edic ions. X. CONCLUSION Mayo (2018) a gues ha one eason science has been ulne able o a ep oducibili y c isis is 5 See h p://www.doingbusiness.o g/en/da a/da a- e isions 21 because heo ies a e weakly es ed. Acco dingly, she ad oca es o “se e e es ing”. Only when a heo y has success ully su i ed an app op ia e numbe o es s should ha heo y be iewed as c edible. We apply his app oach in ou eplica ion o Ahn, Khandelwal, and Wei’s (2011) model o in e media e ade. AKW has been in luen ial in he li e a u e because i p o ides bo h a heo e ical amewo k o explaining he exis ence o ade in e media ies, and an accompanying empi ical con i ma ion o he model’s p edic ions. We employ a wo-p onged app oach o e-examining hei empi ical analysis. Fi s , we ep oduce he main esul s om hei empi ical analysis, applying he same empi ical p ocedu es o (mos ly) independen ly sou ced da a. Ou ep oduc ion esul s s ongly con i m AKW’s conclusions. We hen unde ake a se ies o e-analyses and ex ensions, some imes e-wo king he ep oduc ion da a, o he imes u ning o al e na i e da a sou ces o ex ending he ime ames o hei analysis. When we do ha , we ind ha empi ical suppo o he AKW model o in e media ed ade is g ea ly diminished. Fo example, when we di ide he da a acco ding o he h ee majo geog aphical egions o China, we con i m he p edic ed esul s o g ea e uni alues o i ms in he eas e n p o inces in China, bu we ob ain he opposi e esul s o i ms in he cen al and wes e n p o inces o China. Ano he example a ises when we ex end AKW’s use o En e p ise Su ey Da a o new da a om 2012. When we apply AKW’s empi ical me hodology o hese da a, we es ima e a U- shaped ela ionship be ween p oduc i i y and labo p oduc i i y, a he han he in e ed U- shaped ela ionship p edic ed by hei heo y. In summa y, while ou addi ional es s p oduce some successes o he AKW model, a holis ic assessmen leads o he conclusion ha he da a do no p o ide s a is ical e idence o suppo hei heo y. Tha being said, he esul s epo ed he e should no be in e p e ed as solely nega i e. They poin o possible a enues o u u e esea ch. The con lic ing uni alue 22 esul s o in e media y i ms by egion sugges a ole o loca ional ac o s no iden i ied by AKW’s model. Fu he , a obus esul in ou analysis is ha dis ance is posi i ely co ela ed wi h he use o in e media ies o expo , and GDP (size o ma ke ) is nega i ely co ela ed. This highligh s he need o a heo e ical explana ion o why hese wo coun y cha ac e is ics in pa icula should be associa ed wi h in e media ed ade. We hope ha his esea ch s imula es e o s in his di ec ion. 23 REFERENCES Ahn, J., Khandelwal, A. K., & Wei, S.-J. (2011). The ole o in e media ies in acili a ing ade. Jou nal o In e na ional Economics, 84(1), 73-85. Ake man, A., 2018. A heo y on he ole o wholesale s in in e na ional ade based on economies o scope. Canadian Jou nal o Economics/Re ue canadienne d'économique, 51(1), 156-185. An às, P., & Cos ino , A. (2011). In e media ed ade. The Qua e ly Jou nal o Economics, 126(3), 1319-1374. Be na d, A. B., Jensen, J. B., Redding, S. J., & Scho , P. K. (2010). Wholesale s and e aile s in US ade. Ame ican Economic Re iew, 100(2), 408-413. Blum, B.S., Cla o, S. and Ho s mann, I.J., 2018. T ade cos s and he ole o in e na ional ade in e media ies. In Handbook o In e na ional T ade and T anspo a ion. Edwa d Elga Publishing. Chan, J.M., 2018. Financial F ic ions and T ade In e media ion: Theo y and E idence. Eu opean Economic Re iew, h ps://doi.o g/10.1016/j.eu oeco e .2018.04.002. Dasgup a, K. and Mond ia, J. (2018). Quali y unce ain y and in e media ion in in e na ional ade, Eu opean Economic Re iew, 104, 68-91. Feens a, R. C., & Hanson, G. H. (2004). In e media ies in En epo T ade: Hong Kong Re ‐ Expo s o Chinese Goods. Jou nal o Economics & Managemen S a egy, 13(1), 3-35. Mayo, D. G. (2018). S a is ical in e ence as se e e es ing. Camb idge: Camb idge Uni e si y P ess. Meli z, M. J. (2003). The impac o ade on in a ‐ indus y eal indus y p oduc i i y. Econome ica, 71(6), 1695-1725. Ponce , S. and Xu, M., 2018. Quali y sc eening and ade in e media ies: E idence om China. Re iew o In e na ional Economics, 26(1), 223-256. Qu, Z., Ra , H., & Schmi , N. (2014). In en o y and In e na ional T ade Vola ili y: The Role o Expo and Impo Wholesale s. Wo king pape , Uni e si y o Kiel. Rossman, M. L. (1984). Expo ading company legisla ion: US esponse o Japanese o eign ma ke pene a ion. Jou nal o Small Business Managemen (p e-1986), 22(000004), 62. She ele a, Y., and K ishna, K. (2017). Whea o s awbe ies? In e media ed ade wi h limi ed con ac ing. Ame ican Economic Jou nal: Mic oeconomics, 9(3), 28-62. Spulbe , D.F. (1996). Ma ke mic os uc u e and in e media ion. Jou nal o Economic Pe spec i es, 10(3), 135–152. 30 TABLE 5 Robus ness Check o AKW’s Resul s: P edic ion #2 Sample {In e media y} (Expec = Pos) Obse a ions 2005/Eas 0.022*** (0.004) 4,815,809 2005/Cen al -0.058*** (0.013) 224,526 2005/Wes -0.057*** (0.013) 166,825 2000 -0.003 (0.004) 1,731,580 2001 0.026*** (0.004) 2,019,600 2002 0.022*** (0.004) 2,688,711 2003 -0.023*** (0.003) 3,358,635 2004 -0.003 (0.003) 4,166,559 NOTES: The i s h ee ows o he able (“2005/Eas ”, “2005/Cen al”, “2005/Wes ”) e- es ima e speci ica ion (3.a/3.b) om TABLE 2 using he same da a, excep ha he ull sample is di ided in o h ee subsamples. “Eas ” includes Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang, Fujian, Shandong, Guangdong, Hainan. “Cen al” includes Shanxi, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hunan, Hubei. “Wes ” includes Chongqing, Sichuan, Guizhou, Yunnan, Tibe , Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, Inne Mongolia, Guangxi. The bo om i e ows “(2000”-“2004”) e-es ima e speci ica ion (3.a/3.b) om TABLE 2 using ea lie yea s o he China Cus oms Da a. The dependen a iable is he log uni alue om indi idual ansac ions. Reg essions a e es ima ed using OLS wi h clus e obus s anda d e o s g ouped by p oduc (HS8). In addi ion o he in e media y dummy a iable, all eg essions include iple ixed e ec s o p oduc , owne ship, and coun y. “Pos” e e s o posi i e. *, **, and *** indica e s a is ical signi icance a he 10-, 5-, and 1-pe cen le el, espec i ely. The ose-shaded cells indica e ha he espec i e es ima es ha e he w ong sign, bu a e no s a is ically signi ican a he 5- pe cen le el. The ed-shaded cells indica e ha he espec i e es ima es ha e he w ong sign and a e s a is ically signi ican a he 5-pe cen le el. 31 TABLE 6 Robus ness Check o AKW’s Resul s: P edic ion #3 Sample Dis ance (Expec = Pos) GDP (Expec = Neg) Chinese (Expec = Neg) # P ocedu es (Expec = Pos) Ta i s (Expec = Pos) Obse a ions R-sq A) 2005: Geog aphical subsamples Eas 0.035*** (0.007) -0.016*** (0.002) -0.004*** (0.001) 0.005*** (0.001) 0.027 (0.021) 139,430 0.18 Cen al 0.037** (0.016) -0.011** (0.005) -0.001 (0.003) 0.006 (0.004) 0.066 (0.059) 44,765 0.28 Wes 0.034*** (0.011) -0.018*** (0.005) 0.000 (0.003) 0.001 (0.002) 0.164** (0.073) 33,835 0.32 B) 2005: Al e na i e da a sou ces o numbe o equi ed p ocedu es Book 0.026*** (0.009) -0.018*** (0.003) -0.004*** (0.001) 0.001 (0.001) 0.060** (0.023) 188,120 0.17 Online 0.029*** (0.009) -0.016*** (0.002) -0.004*** (0.001) 0.004*** (0.001) 0.048** (0.023) 189,030 0.17 C) 2005: Al e na i e da a sou ces o numbe o equi ed p ocedu es by geog aphical subsample Book/Eas 0.030*** (0.007) -0.017*** (0.003) -0.005*** (0.001) 0.002 (0.001) 0.057** (0.023) 139,430 0.18 Book/Cen al 0.034** (0.015) -0.011** (0.005) -0.001 (0.002) 0.003 (0.003) 0.092 (0.069) 44,765 0.28 32 Sample Dis ance (Expec = Pos) GDP (Expec = Neg) Chinese (Expec = Neg) # P ocedu es (Expec = Pos) Ta i s (Expec = Pos) Obse a ions R-sq Book/Wes 0.032*** (0.011) -0.020*** (0.005) 0.000 (0.003) -0.003 (0.002) 0.195** (0.079) 33,835 0.32 Online/Eas 0.034*** (0.008) -0.014*** (0.002) -0.004*** (0.001) 0.004*** (0.001) 0.048** (0.024) 140,297 0.18 Online/Cen al 0.034** (0.015) -0.012** (0.005) -0.001 (0.003) 0.002 (0.003) 0.106* (0.061) 44,839 0.28 Online/Wes 0.033*** (0.010) -0.017*** (0.005) 0.000 (0.003) 0.001 (0.002) 0.168** (0.070) 33,862 0.32 D) 2000-2004: Di e en equi ed p ocedu es a iables 2004/AKW 0.039*** (0.009) -0.023*** (0.004) -0.002 (0.002) 0.002* (0.001) -0.007 (0.007) 106,033 0.23 2003/AKW 0.047*** (0.006) -0.021*** (0.004) -0.006*** (0.002) 0.005*** (0.001) 0.008 (0.009) 99,608 0.24 2002/AKW 0.046*** (0.008) -0.020*** (0.004) -0.004** (0.002) 0.005*** (0.002) -0.006*** (0.002) 96,785 0.24 2001/AKW 0.046*** (0.007) -0.017*** (0.003) -0.005*** (0.002) 0.006*** (0.002) -0.008 (0.009) 91,491 0.24 2000/AKW 0.042*** (0.008) -0.022*** (0.004) -0.008*** (0.002) 0.006*** (0.001) -0.004 (0.012) 82,396 0.26 33 Sample Dis ance (Expec = Pos) GDP (Expec = Neg) Chinese (Expec = Neg) # P ocedu es (Expec = Pos) Ta i s (Expec = Pos) Obse a ions R-sq 2004/Book 0.037*** (0.009) -0.025*** (0.004) -0.002 (0.002) -0.000 (0.001) -0.004 (0.007) 106,033 0.23 2003/Book 0.045*** (0.007) -0.022*** (0.004) -0.005** (0.002) 0.002* (0.001) 0.014 (0.010) 99,608 0.24 2002/Book 0.043*** (0.009) -0.023*** (0.004) -0.004 (0.002) 0.002 (0.001) -0.004* (0.002) 96,785 0.23 2001/Book 0.043*** (0.008) -0.020*** (0.004) -0.004* (0.002) 0.001 (0.002) 0.009 (0.014) 91,491 0.24 2000/Book 0.040*** (0.008) -0.024*** (0.005) -0.008*** (0.002) 0.003* (0.001) 0.006 (0.013) 82,396 0.26 2004/Online 0.040*** (0.009) -0.022*** (0.003) -0.002 (0.002) 0.003** (0.001) -0.005 (0.007) 107,110 0.23 2003/Online 0.047*** (0.008) -0.020*** (0.004) -0.005** (0.002) 0.004** (0.002) 0.018 (0.011) 99,677 0.24 2002/Online 0.045*** (0.010) -0.020*** (0.004) -0.003 (0.002) 0.005*** (0.002) -0.005** (0.002) 96,907 0.23 2001/Online 0.045*** (0.008) -0.017*** (0.004) -0.004* (0.002) 0.005*** (0.002) 0.010 (0.014) 92,197 0.24 2000/Online 0.041*** (0.009) -0.022*** (0.004) -0.007*** (0.002) 0.005*** (0.001) 0.009 (0.012) 82,443 0.26 34 Sample Dis ance (Expec = Pos) GDP (Expec = Neg) Chinese (Expec = Neg) # P ocedu es (Expec = Pos) Ta i s (Expec = Pos) Obse a ions R-sq Co ec &Signi ican /To al (All) 26/26 26/26 13/26 12/26 7/26 ---- ---- Co ec &Signi ican /To al (Excluding Panel D) 11/11 11/11 5/11 3/11 7/11 ---- ---- NOTES: The h ee ows o Panel A in he able (“Eas ”, “Cen al”, “Wes ”) e-es ima e speci ica ion (4.a/4.b) om TABLE 3 using he same da a, excep ha he ull sample is di ided in o h ee subsamples. The wo ows o Panel B (“Book”, “Online”) e-es ima e speci ica ion (4.a/4.b) om TABLE 3 using he same da a excep AKW’s da a o numbe o equi ed impo documen s (“# o P ocedu es”) is eplaced wi h he p in ed book and online e sions, espec i ely, o he same a iable om Doing Business Repo 2006. The six ows o Panel C in he able ake he ull samples o he “Book” and “Online” eg essions o Panel B, di ide each in o h ee subsamples based on he h ee egions o China, and e-es ima e speci ica ion (4.a/4.b) om TABLE 3. The i een ows o Panel D e-es ima e speci ica ion (4.a/4.b) om TABLE 3 using ea lie yea s (2000-2004) o he China Cus oms Da a. I does his o each o he h ee e sions o he “# o P ocedu es” a iable (“AKW”, “Book”, “Online”). No e ha he same alues ha we e used o he a iables “Chinese”, “# o P ocedu es”, and “Ta i s” in he 2005 eg essions a e also used o he ea lie yea s because da a o 2000-2004 o hese a iables a e una ailable. The dependen a iable in each eg ession is he sha e o in e media y expo s o o al coun y HS6 expo s. Reg essions a e es ima ed using OLS wi h clus e obus s anda d e o s g ouped by coun y. All eg essions also include ixed e ec s o p oduc ype (HS6). “Pos” and “Neg” e e o posi i e and nega i e. *, **, and *** indica e s a is ical signi icance a he 10-, 5-, and 1-pe cen le el, espec i ely. The g ey-shaded cells indica e ha he espec i e es ima es ha e he p edic ed sign, bu a e no s a is ically signi ican a he 5-pe cen le el. The ose-shaded cells indica e ha he espec i e es ima es ha e he w ong sign, bu a e no s a is ically signi ican a he 5-pe cen le el. The ed-shaded cells indica e ha he espec i e es ima es ha e he w ong sign and a e s a is ically signi ican a he 5-pe cen le el. 35 TABLE 7 Summa y S a is ics o Th ee Da a Sou ces o he Numbe o Requi ed Impo Documen s Va iable Sou ce Obse a ions Mean Median Min Max % Di e en : AKW % Di e en : Book AKW 151 9.10 9 3 19 ---- ---- Book 151 10.77 11 3 19 61.6 ---- Online 171 8.08 7 2 21 72.2 85.4 NOTES: This able compa es he h ee e sions o he a iable “# o P ocedu es” a ailable om AKW, he p in ed book e sion o he Wo ld Bank’s Doing Business Repo 2006, and he online e sion o he Wo ld Bank’s Doing Business Repo 2006. The las wo columns compa e he pe cen o o al numbe o obse a ions ha a e di e en om each pai o e sions (AKW/Book, AKW/Online, Book/Online). Pe cen ages a e calcula ed om he o al numbe o pai ed obse a ions in which bo h e sions epo non-missing alues. 36 TABLE 8 Summa y o Resul s om Tes s o AKW’s Model P edic ion Rep oduc ion Re-analysis and Ex ension P edic ion #1 6/9 4/9 P edic ion #2 3/3 3/8 P edic ion #3 – To al 13/14 84/130 37/55 P edic ion #3 – Dis ance 4/4 26/26 11/11 P edic ion #3 – GDP 4/4 26/26 11/11 P edic ion #3 – Chinese 3/3 13/26 5/11 P edic ion #3 – # P ocedu es 2/2 12/26 3/11 P edic ion #3 - Ta i 0/1 7/26 7/11 NOTES: The alues in he able epo he numbe o es s ha suppo he espec i e p edic ion o e he o al numbe o es s. The esul s in he “Rep oduc ion” column collec he es esul s om TABLES 1 o 3. The esul s in he “Re-analysis and Ex ension” column collec he esul s om TABLES 4 o 6. A es was judged o suppo AKW’s p edic ions i i had he p edic ed sign and was s a is ically signi ican . A coe icien ha had he p edic ed sign bu was s a is ically insigni ican was in e p e ed as no p o iding s a is ical suppo o AKW’s model. “P edic ion #3 – To al” combines he esul s om p edic ions o he a iables Dis ance, GDP, Chinese, # o P ocedu es, and Ta i . “P edic ion #3 – Dis ance” o “P edic ion #3 – Ta i ” b eak ou he esul s o es ing he p edic ions o he indi idual a iables. 37 FIGURE 1 AKW’s Model o In e media y T ade NOTE: The e ical axis measu es p o i s om expo ing (𝜋). The ho izon al axis measu es i m p oduc i i y (𝜑𝜎−1). The ho izon al axis ep esen s he p o i s om expo ing i he i m does no expo (=0). Line A ep esen s he p o i s o he i m i i sells i s ou pu o an in e media y ha hen epackages he ou pu and sells i in N o e seas ma ke s. The poin a which A c osses he ho izon al axis iden i ies he minimal p oduc i i y (𝜑𝑖 𝜎−1) a which he i m will swi ch om selling only in he domes ic ma ke o also expo ing ia an in e media y. Line B ep esen s he p o i -p oduc i i y nexus o indi ec ly expo ing o coun y j h ough an in e media y. Line C indica es he associa ed p o i s o di ec ly expo ing o j. The p oduc i i y whe e he wo lines in e sec (𝜑𝑥 𝑗𝜎−1) iden i ies he h eshold whe e a i m ansi ions om indi ec ly selling i s p oduc o coun y j o di ec ly selling o j. A C B 38 APPENDIX Desc ip ion o Va iables and Da a Sou ces Used in he Rep oduc ion Replica ion o AKW Va iable De ini ion Sou ce Di ec expo sha e Pe cen age o i m’s sales expo ed di ec ly Wo ld Bank’s En e p ise Su ey Da a Indi ec expo sha e Pe cen age o i m’s sales expo ed indi ec ly ( h ough a dis ibu o ) Wo ld Bank’s En e p ise Su ey Da a Sales To al sales 1 yea ago Wo ld Bank’s En e p ise Su ey Da a Employmen A e age numbe o pe manen wo ke s plus empo a y wo ke s 1 yea ago Wo ld Bank’s En e p ise Su ey Da a Labou p oduc i i y 𝑠𝑠𝑠𝑠𝑠 𝑠𝑒𝑒𝑠𝑒𝑒𝑒𝑠𝑒𝑒 Wo ld Bank’s En e p ise Su ey Da a Uni alue 𝑣𝑠𝑠𝑣𝑠 𝑒𝑓 𝑠𝑠𝑠𝑠𝑠 𝑞𝑣𝑠𝑒𝑒𝑞𝑒𝑒 𝑠𝑒𝑠𝑠 China Cus oms Da a In e media y = 1 i he i m is an in e media y and = 0 o he wise China Cus oms Da a Fi m size Fi m’s o al expo alue China Cus oms Da a In e media y expo sha e Sha e o in e media y expo s o o al coun y-HS6 expo s China Cus oms Da a Dis ance Ai dis ance in nau ical miles be ween he ading coun y pai s h ps://www. imeandda e. com/wo ldclock/dis ance.h ml GDP G oss domes ic p oduc Wo ld Bank’s Wo ld De elopmen Indica o s 39 Va iable De ini ion Sou ce Chinese popula ion Size o he e hnic Chinese popula ion pe coun y Ohio Uni e si y Shao Cen e Impo ing p ocedu es Numbe o documen s equi ed o impo in China Wo ld Bank’s Doing Business Repo (supplied by AKW) MFN a i Mos Fa ou ed Na ion du y a e ea men Wo ld Bank WITS (supplied by AKW)