Gold market price spillover between COMEX, LBMA and SGE

Qian, Xinyi A icle — Published Ve sion Gold ma ke p ice spillo e be ween COMEX, LBMA and SGE Jou nal o Economics and Finance P o ided in Coope a ion wi h: Sp inge Na u e Sugges ed Ci a ion: Qian, Xinyi (2020) : Gold ma ke p ice spillo e be ween COMEX, LBMA and SGE, Jou nal o Economics and Finance, ISSN 1938-9744, Sp inge US, New Yo k, NY, Vol. 44, Iss. 4, pp. 810-831, h ps://doi.o g/10.1007/s12197-020-09517-5 This Ve sion is a ailable a : h ps://hdl.handle.ne /10419/289036 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/ h ps://doi.o g/10.1007/s12197-020-09517-5 Gold ma ke p ice spillo e be ween COMEX, LBMA and SGE Xinyi Qian1 ©The Au ho (s) 2020 Abs ac In his pape , he au ho in es iga es spillo e be ween he main ma ke s om New Yo k, London and Shanghai. Speci ic con ac p ices om he Commodi y Exchange Inc. (COMEX), London Bullion Ma ke Associa ion (LBMA) and Shanghai Gold Exchange (SGE) we e u ilized. Resul s sugges ha e en wi h he inc easing ma ke in luence o SGE, i s ill emains an isola ed ma ke , COMEX and LBMA main- ain hei dominan posi ions and ac as he ne spillo e sp eade s in he wo ld gold ma ke wi h almos equally s ong ma ke impac s. Keywo ds Gold ·Vola ili y ·Spillo e ·COMEX ·LBMA ·SGE JEL Classifica ion C58 ·G14 ·G15 1 In oduc ion 1.1 Backg ound Gold is one o he mos homogeneous goods in he wo ld. Due o i ’s ai ly ideal p ese a ion o alue and easy s o age, gold has been aded globally among exchanges and banks, bo h as a spo and/o u u e, as a commodi y, and a inanced asse . Th ee majo gold ading cen es a e London, New Yo k and Shanghai. CME G oup (2017) The au ho is g a e ul o D . Guanpin Lu o ad ice and sugges ions, and o Jonas Klamka, D . Benjamin Sch¨ a e and all o he pa icipan s a he Uni-Siegen Doc o al Semina o help ul commen s. Special hanks shall be ex ended o Michael Wallace o co ec ing and p oo - eading my pape . Xinyi Qian [email protected] 1Uni e si y o Siegen,Facul y III, Un e es Schloss 3, Siegen, D-57072, Ge many Jou nal o Economics and Finance (2020) 44:810–831 Published online: 21 Augus 2020 Es ablished in 1933, Commodi y Exchange Inc. (COMEX) is he oldes among he h ee ma ke s s udied in his esea ch. COMEX me ged wi h he New Yo k Me - can ile Exchange (NYMEX) in 1996 and hen joined he CME G oup in 2008. Being he di ision esponsible o me als ading, COMEX is no longe a sepa a e ins i u- ion bu a p ima y u u es and op ions ma ke o ading me als such as gold, sil e , coppe and aluminium. Es ablished in 1987, London Bullion Ma ke Associa ion (LBMA) is a whole- sale o e - he-coun e ma ke o he ading o gold and sil e . LBMA ook o e he London Gold Fix ope a es by he ICE Benchma k Adminis a ion (IBA) since 19. Ma ch 2015 and se he benchma k p ice wice daily (a 10:30 and 15:00 London BST) in US Dolla s and is also a ailable in a u he six een cu encies (indica i e o se lemen s). The younges ma ke in his pape is he Shanghai Gold Exchange (SGE). Despi e i s sho his o y ( ounded in 2002), SGE has al eady become he la ges commodi y exchange in he People’s Republic o China o ading in p ecious me als (gold, sil e and pla inum). Fu he mo e, he daily ading olume o gold and sil e is a he second highes in he wo ld and in 2017, exceeded he Shanghai Fu u e Exchange (SHFE). 1.2 Mo i a ion and esea ch ques ion Take a pano amic iew o he o e all si ua ion o he global p ecious me al exchanges. COMEX is s ill i s wi h espec o ading olume in gold and sil- e . SHFE and Tokyo Commodi y Exchange (TOCOM) a e sh inking signi ican ly since 2016, while Shanghai Gold Exchange (SGE) has climbed o second posi- ion in he wo ld ankings. Wi h he eme ging ma ke powe in Asia, especially he in e na ionalisa ion p ocess o he Shanghai Gold Exchange (SGE), ading liquidi y du ing he Asia hou has been inc eased signi ican ly (CME G oup 2017). Mean- while, LBMA emains ela i ely s able as a domina ing wo ld OTC ma ke (Shanghai Gold Exchange 2018). This new wo ld anking inspi es he mo i a ion o esea ch- ing he la es ma ke s in e ac ion be ween he i s h ee anking exchanges, namely COMEX, LBMA and SGE. When SGE s a ed i s in e na ionalisa ion in 2015, expec a ions inc eased o e his newly eme ged ma ke . Chai man o Swiss-based e ining g oup MKS, Ma wan Shaka chi once p esen ed his hypo hesis o Reu e s in ea ly 2016 when SGE launched a Yuan-denomina ed gold benchma k on 19 Ap il 2016:“(China) is a ma ke o 1.2 billion people and simply canno be neglec ed. I am con inced ha in he u u e we won’ say China is a a p emium o discoun o London, bu ice e sa”; howe e , e en hough he Chinese domes ic ma ke pa ic- ipan s can begin hei ading wi h local cu ency Chinese Yuan (CNY), he wo ld playe s may no accep o e en ake conside a ion o his newly-de eloping ma ke . Voice had been made om he side o SGE, exp essing he conce n ha an adjus - ing p ocess would be ime-consuming, and no going o happen in he nea u u e. Anan halakshmi (2015) Because o he sho ma ke his o y o SGE and he limi a ions o he exis ing econome ic model, no esea ch, o he bes o he au ho ’s knowledge, has been done abou he gold ma ke daily p ice ola ili y spillo e including SGE ill now. Hence, Jou nal o Economics and Finance (2020) 44:810–831 811 in his pape , he au ho would like o a) use a p ope me hod o in es iga e he ma ke impac s be ween he cu en op h ee ma ke s being in oduced in Sec ion 1.1 and b) examine he dynamic end o he in e ac ions be ween hese h ee ma ke s, check i he e has al eady been changes ha ha e aken place o whe he any signs can be de ined om he ecen obse a ions. The esul o his esea ch migh be help ul o he conce ning exchanges o ake a e iew o he s eng h and ma ke impac in he ecen yea s and de elop a quali a i e o e iew o he main ma ke si ua ion and end du ing sequen ial ini ia i es and changes, ei he made by hemsel es o by hei compe i o s and/o pa ne s. 1.3 Li e a u e e iews The a ibu es o gold ha e been s udied by a conside able amoun o li e a u e. Apa om hose which ocused on he indus ial sec o and jewelle y sec o (abou min- ing and ashioning echnique), we di ec ou a en ion o he mone a y a ibu es o gold in his subsec ion. Wo hing on and Pahla ani (2007) p o ided solid e idence o p o e he widely held iew ha an in es men in gold can se e as an e ec i e in la- iona y hedge. La e Cine e al. (2013) examined he co ela ion be ween s ocks, bonds, gold, oil and exchange a es using da a om Uni ed S a es and Uni ed King- dom. They ound ha gold can be ega ded as a sa e ha en agains exchange a es in bo h coun ies, which highligh ed i s ole in he mone a y asse s. Wi h he wo ld gold ma ke s becoming mo e open and easily accessible, he in e connec ion be ween ma ke s spu ed esea che s’ in e es o ind a p ope econo- me ic me hod o quan i y he connec ion s eng h as well as he di ec ion o he sp eading. Diebold and Yilmaz (2009) (hence o h DY09) i s examined he di - e en spillo e beha iou s be ween e u n and ola ili y using da a om equi y ma ke s and coined he e m connec edness using he Vec o Au o eg essi e Reg es- sion model. They exhibi ed di e ence be ween e u n and ola ili y by measu ing he ime- a ying and ime- a ia ion spillo e in ensi y. Acco ding o Diebold and Yilmaz (2009,2012,2014), such a ola ili y connec edness can be ea ed as “ ea connec ed- ness” exp essed by he ade s du ing di e en ma ke condi ions and is pa icula ly c isis sensi i e (Diebold and Yilmaz 2014). In hei 2012 esea ch (Diebold and Yil- maz 2012) (hence o h DY12), hey u he imp o ed he model om DY09, so ha i is no longe sensi i e o he VAR o de by eplacing he Cholesky decomposi ion, and is able o de ec he di ec ion o he connec edness low ( he so-called “di ec ional spillo e ”). In he echnical sense, DY12 can be ea ed as a obus e sion o DY09, which applied he decomposi ion app oach om Koop e al. (1996) and Pesa an and Shin (1998). Addi ionally, Ba un´ ık and Kˇ ehl´ ık (2018) ealised ha a long- e m- e ec shock has high powe a he low equencies, and hus hey sepa a ed he long- and sho - e m connec edness by applying he equency bands in o de o mimic he spillo e mo emen s be ween 1 o 4 days as well as 4 days o a longe pe iod. Acco ding o he ac ha mos o he gold ading olume is s ill se led in London, Lucey e al. (2014) applied he me hod om Diebold and Yilmaz (2009) and s udied he spillo e e ec o he spo gold p ices be ween ou ma ke s, namely: LBMA, COMEX, SHFE and TOCOM. Resul s showed ha SHFE as a newly eme ged ma ke has a ely any e ec on he o he h ee. Howe e , Lucey e al. (2014) only esea ched Jou nal o Economics and Finance (2020) 44:810–831 812 he u u e ma ke s and applied Ga man and Klass app oach (Ga man e al. 1980) app oach o he ola ili y spillo e es ima ion. E idence om Rosenbe g and T aub (2006) has shown, ha in he o eign exchange u u e and spo ma ke s, he la e one has he dominan in o ma ion sha e. On he o he hand, he Ga man and Klass (1980) app oach has been shown o unde es ima e he ola ili y because i igno es he o e nigh jump.1Fu he mo e, hei applica ion o DY09 only p o ided an o e - all insigh o he e u n o ola ili y o he mul iple ma ke s as a whole, lacking exac di ec ional spillo e pa e ns be ween he speci ic ma ke s due o echnical es ic- ions. Add essing he di ec ional spillo e pa e ns is exac ly he majo con ibu ion o his esea ch pape . 1.4 O ganisa ion o he pape The goal o his pape is o de e mine he s eng h o he ma ke impac s among h ee ma ke s by examining hei in e ac ions, i.e. ecei e/gi e spillo e om/ o each o he , in bo h quali a i e and quan i a i e senses. This pape has been o ganized as ollows: Sec ion 2p o ides de ail on he me hods we a e going o apply o es i- ma ing he spillo e , Sec ion 3in oduces he da a and he p ocessing p ocedu es o p o ide desc ip i e s a is ics o he sample da a, Sec ion 4p esen s he main esul s, and Sec ion 5o e s a conclusion. 2 Me hods DY09 is he i s esea ch de eloping a model based on VAR ( ec o au o eg es- si e model) amewo k and in es iga ing he ola ili y spillo e be ween di e en asse s by Diebold and Yilmaz. When se e al asse s (o equi ies) om di e en coun- ies a ec each o he (i.e., ha e spillo e e ec s on each o he ), DY09 can de ec he o al ola ili y among all hose asse s, de ine how much ola ili ies in he luc- ua ion a e ac ually caused by he spillo e s be ween hem. Howe e , as he au ho Diebold and Yilmaz la e in hei 2012 esea ch (Diebold and Yilmaz 2012) coined, he e a e wo main me hodological d awbacks. Namely ha a) DY09 depends ai ly s ong on he a iable o de ing because o he Cholesky iden i ica ion applied o a VAR decomposi ion, and b) DY09 can only add ess he o al spillo e among all he asse s being es ima ed. In p ac ice, one migh be mo e in e es ed in a sepa a ed di ec- ional spillo e om a speci ic asse o ano he speci ic one. In DY12, he au ho s ollowed he gene al idea o DY09 and compu ed he o al o e all spillo e index and an uncondi ional ull-sample s a ic a e age spillo e able. Thei inno a i e ini ia i e was using he decomposi ion om Koop e al. (1996) and Pesa an and Shin (1998) (hence o h KPPS) in o de o a oid he sensi i i y o he a iable o de ing caused by he Cholesky app oach hey p e iously used in DY09. Since he esul om DY12 no longe depends on he VAR o de ing, we can ea DY12 as a mo e obus e sion o 1See Yang and Zhang (2000). The Jou nal o Business, olume 73, numbe , p481, “The e o e. igno ing opening jumps will unde es ima e he ola ili y.” Jou nal o Economics and Finance (2020) 44:810–831 813 DY09 which ills he gap o DY09 and an ad anced model being able o de ec he di ec ion o he spillo e be ween e e y each single asse s. Thus, in his sec ion, DY09 is i s in oduced un il he “o e all spillo e index” is eached. The ea e , new imp o emen s ha e been con ibu ed by DY12, om whe e we u n o he me hod o DY12 and p esen all o mulae o he u he di ec ional spillo e es ima ion. DY09 in es iga ed he connec edness using VAR and i s c ea ed he concep o “spillo e index”. Fo a wo- a iable ec o o s a iona y i s -o de se ies x , i can be w i en as: x =Φx −1+ε (1) whe e x is a ec o o ei he e u ns o ola ili ies, he e x is a 2 ×1 ec o and Φis a2×2 pa ame e ma ix. Exp ession (1) can be ep esen ed in he ollowing o m: (I −ΦL)x =ε (2) Φ(L)x =ε (3) Φ(L) is a 2 ×2 ma ix polynomial in L. Then he equa ion can be w i en as: x =Θ(L)ε =A(L)μ (4) whe e Θ(L) =(I −ΦL)−1and A(L) =Θ(L)Q−1 .Q−1 s ands o a unique lowe iangula Cholesky ac o o ε , wi h μbeing de ined as μ=Q ε and E(μ μ )=I. One can show ha he one-s ep-ahead Wiene -Kolmogo o linea leas -squa e o ecas as: x +1, =Θx (5) wi h a co esponding one-s ep-ahead e o ec o : e +1, =x +1−x +1, =A0u +1=a0,11 a0,12 a0,21 a0,22 u1, +1 u2, +1(6) Thus he co a iance ma ix can be w i en as: E(e +1, e +1, )=A0A 0(7) The ad an age o using his app oach is being able o spli he o ecas e o a i- ances in o wo pa s, namely he one-s ep-ahead e o o x1, as a2 0,11 +a2 0,12,and a2 0,21 +a2 0,22 o x2, . By doing his, one can be in o med on which p opo ion o he e o a iance has been con ibu ed by he shock o i sel (in a 2- a iable case, his can be x1), he so-called own a iance sha es, o by he shock o he o he s (x2in his case), he c oss a iance sha es, which is also he spillo e we wan o examine. The o al o ecas e o a ia ion is de ined as: a2 0,11 +a2 0,12 +a2 0,21 +a2 0,22 = ace(A0A 0)(8) The spillo e as a pe cen age a io o he o al o ecas e o a ia ion leads o he o mula o he spillo e index o he basic i s -o de wo- a iable case: S=a2 0,12 +a2 0,21 ace(A0A 0)×100 (9) Jou nal o Economics and Finance (2020) 44:810–831 814 Analogously he case o p h-o de N- a iable VAR wi h H-s ep-ahead o ecas spillo e index can be immedia ely deduced as: S=H−1 h=0N i,j=1 i=j a2 h,ij H−1 h=0 ace(AhA h)×100 (10) Howe e , he e a e ce ain limi a ions o he use o his me hod om DY09. Fi s ly, he a iance decomposi ion esul ing om he Cholesky ac o isa ion leads o a s ong dependency on he o de ing o he a iables. The second me hodologi- cal limi a ion comes when one has mo e han 2 ma ke s and would like o de ec he di ec ional spillo e s. Wi h DY09, only a o al spillo e is iden i ied. By exploi ing an o de -in a ian a iance decomposi ion aised by KPPS, DY12 p oduced a new app oach based on he gene al idea om DY09. An H-s ep-ahead o ecas e o a i- ance decomposi ion being deno ed by θg ij (H) o H=1,2,... has he ollowing ep esen a ion: θg ij (H) =σ−1 jj H−1 h=0(e iAhΛej)2 H−1 h=0(e iAhΛA hei)(11) whe e ei=1i h elemen 0o he wise eiis a selec ion ec o which equals 1 o he i h i ems and 0 o he wise. Λs ands o he a iance ma ix o he e o ec o εand σjj ep esen s he s anda d de ia- ion o he e o e m o he j h equa ion. Realising ha each en y o he a iance decomposi ion ma ix can be no malised by he ow sum as: ˜ θg ij (H) =θg ij (H) N j=1θg ij (H) wi h j=1 N˜ θg ij (H) =1andi,j=1 N˜ θg ij (H) =Nby cons uc ion, we hen ha e he exp ession o o al spillo e , which is he KPPS analogue based on he DY09 o al spillo e as o mula (10): Sg(H) =N i,j=1 i=j ˜ θg ij (H) N i,j=1˜ θg ij (H) ×100 =N i,j=1 i=j ˜ θg ij (H) N×100 (12) Thus in DY12, he ollowing exclusi ely u he spillo e s ha e been deduc ed: Di ec ional ola ili y spillo e ecei ed by ma ke i om all o he ma ke s j: Sg i.(H) =N j=1 j=i ˜ θg ij (H) N i,j=1˜ θg ij (H) ×100 =N j=1 j=1 ˜ θg ij (H) N×100 (13) Di ec ional ola ili y spillo e ansmi ed by ma ke i o all o he ma ke s j: Sg .i(H) =N j=1 j=i ˜ θg ji(H ) N i,j=1˜ θg ji(H ) ×100 =N j=1 j=1 ˜ θg ij (H) N×100 (14) Jou nal o Economics and Finance (2020) 44:810–831 815 Ne ola ili y spillo e s om ma ke i o all o he ma ke s j: Sg i(H) =Sg .i(H) −Sg i.(H) (15) Ne pai wise spillo e s be ween ma ke s i and j ( om i o j): Sg ij (H) =˜ θg ji(H ) N i,k=1˜ θg ik(H ) − ˜ θg ij (H) N j,k=1˜ θg jk(H)×100 =˜ θg ji(H ) −˜ θg ij (H) N×100 (16) In he ollowing sec ions, hose o mulae will be applied o he da a in oduced in Sec ion 3 o e u n and ola ili y speci ically. The esul s will be shown in Sec ion 4 in he o m o ables as well as olling-window dynamic plo s. 3Da a Daily da a which includes open, close, high and low in o ma ion o COMEX, LBMA and SGE has been applied om 23. No embe 2012 o 15. Augus 2018. Wi hin COMEX, gold (p oduc icke GC) has been aded wi h 100 oy ounces con ac size. The minimum ick is $0.10 pe oy ounce. Fo one ick, he dolla alue is hence 10$. T ading hou s a e om Sunday o F iday, 5:00 pm ill 4:00 pm he nex day depends on he No h Ame ican Cen al Time Zone (CT). Fo each ading day, he e is only one hou b eak om 4:00 pm o 5:00 pm. Con ac s a e only mon hly signed wi hin 23- o 72-mon hly pe iod o Feb, Ap , Aug & Oc o Jun & Dec espec i ely. In o de o compa e he gold u u e p ices om COMEX wi h o he spo p ice se ies, “ o wa ds Panama adjus ed p ice oll on las ading day” (GC1) (S e ens Analy ics 2019) has been used om Quandl.2 LBMA daily gold p ice da a is p o ided by Bloombe g Te minal. ICE Benchma k Adminis a ion (IBA) is he cu en ope a o o he LBMA p ice se ing. The ading is 24 hou s and he p ices a e se wice daily a 10:30 am and 3:00 pm London BST in US Dolla and 16 o he cu encies adjus ed a he spo exchange a e as indica i e p ices o se lemen s only. In his pape we use he a e noon se ing benchma k in US Dolla as he closing p ice o he node o one day. Fo SGE, we a e going o use he spo de e ed con ac Au(T+D) i s in oduced by Shanghai Gold Exchange in 2004. Au(T+D) allows he ade s o pos pone hei con ac made on h day by d(d≥0) day(s) h ough a de e al paymen , which only 2This me hod is some imes also called “ i s - ue me hod”. By olling he p ice o wa dly om he oldes his o y con ac o he la es one, one achie es a smoo hed con ac p ice se ies wi hou jumps due o he ma u i ies. The p ice o he oldes con ac will he e o e be he “ ue” as he olling base o he se ies, hus he name “ i s - ue me hod”. O iginal da a and his me hod a e p o ided by S e ens Analy ics. Jou nal o Economics and Finance (2020) 44:810–831 816 amoun s o 10%(as ma gin le el) o he con ac alue. We also no e ha Au(T+D) is no a u u e con ac , bu a special ype o spo con ac . Gold u u es ha e ixed deli - e y da es, while a posi ion o Au(T+D) can always be held wi hou a ixed deli e y da e. Fu he mo e, Au(T+D) has nigh ading hou s, which b ings ano he ad an- age o he ading lexibili y. The eason o choosing his con ac ins ead o using he pu e spo con ac AU9999 om SGE is o i s con ac a ibu es, i can a oid he incen i e o man-made manipula ion. Ha ing no bounda y o daily ola ili y, he p ice o AU(T+D) con ac e lec s mo e abou he ue spo alue balanced by eal demand and supply om he ma ke , unlike he se lemen p ice o AU9999, which is ypically ancho ed o he daily LBMA ixing. Besides, Au(T+D) also has he la ges ading olume among all gold con ac s in SGE du ing ou sample pe iod.3Da a is aced om Wind Financial Te minal. The ading uni is 1kg wi h Yuan/g am as he uni o quo a ion and he minimum ick is 0.01 Yuan/g am. T ading hou s ha e been sepa a ed in o h ee pe iods: 9:00 o 11:30 am, 1:30 o 3:30 pm, 8:00 o 2:30 am (nex day).4 The con inuously compounded e u n o log e u n will be calcula ed using daily close- o-close p ice h ough he ollowing app oach. =ln(1+R )=ln P P −1 =p −p −1 whe e p =ln(P ) By using he log e u n, one can easily achie e a no malisa ion and a oid he o igina ing om p ice se ies o unequal uni s and/o quo ing cu encies. Gold being aded in COMEX and LBMA a e bo h quo ed in US Dolla , he only omi ed ex e - nal changing a iable is he exchange a e be ween US Dolla and Chinese Yuan. Since gold is highly homogenous, he possibili y o a bi age does no incen i ise he in es o s due o he law o one p ice. Vola ili ies ha e been ob ained by applying he Yang and Zhang (2000) measu e- men ully using daily open, close, high, low in o ma ion. Yang-Zhang measu emen also akes he o e nigh jumps and ola ili y d i in o conside a ions. An once- di e ence ha e been aken on i a e wa ds o each a s a iona y ime se ies.5The o mula is as ollow, whe e no a ions h,l,oand cs and o logged daily high, low, open and close espec i ely: Vola ili yYang−Zhang =σ2 YZ =σ2 o e nigh ola ili y +kσ2 open o close ola ili y +(1−k)σ2 RS 3Au(T+D) ading olume amoun s o 34.43% o he o al gold ading olume in 2017. 4The e ha e been sequen ial changes in ma gin le el and ading hou s. The la es in o ma ion in Decembe 2019 indica es a 6% ma gin a io and a longe ading hou wi h wo pe iods: 9:00 - 15:30 and 19:50 - 2:30 (nex day) 5The e is no o e nigh jump o he COMEX since i uns 24 hou s a ound he clock. LBMA closes o only one hou wi h sligh jumps mos o he ime. In con as , he e is always a gap o he SGE since he closing ime is much longe han he o he wo. Jou nal o Economics and Finance (2020) 44:810–831 817 ill he end o ou sample pe iod, he alue emained be ween 52% and 54% mos o he ime. The ola ili y plo shown in Fig. 4has a much smoo he luc ua ion han he e u n plo wi h simila mo emen pa e ns. The o e all ola ili y spillo e was sligh ly abo e 50% a he beginning o he sample pe iod, which indica es, in he ea ly hal o 2013, mo e han hal o he a iance o ecas e o comes om he spillo e . Then i suddenly slid o 40%. A e a deep sink a mid-2014, i climbed back o app ox- ima ely 45% and emained a ha le el o mos o he ime un il end-2015. The second d op appea ed a ea ly 2016, o e all ola ili y spillo e ell unde nea h 30% le el and bounced back again o a sho ime a e wa ds. F om ea ly-2016 ill mid- 2018, he o e all end o ola ili y spillo e was a downwa d mo emen . The nadi ook place in 2018 wi h a alue e en lowe han 25%. Finally, a small upwa d end can be obse ed be o e he end o ou da a sample. 4.3 Condi ional di ec ional e u n spillo e dynamic olling window plo s 4.3.1 Di ec ional om e u n spillo e The di ec ional om e u n spillo e plo (Fig. 5) desc ibes he e u n o ecas a i- ance e o esul ing om ecei ing he spillo e s om o he ma ke s. Among he whole da a pe iod, he spillo e ecei ed by SGE is always la ge han o he o he wo ma ke s, e en i s lowes alue (a ound 18%) in ea ly o middle o 2017 was jus abou he uppe bounda ies o he pe cen age spillo e s ecei ed by COMEX and LBMA du ing he i e yea s. Ye he o e all end demons a es a slowly dec easing ecei ing o SGE. The same downwa d ends also appea ed in COMEX and LBMA and became mo e ob ious since 2016. Fig. 5 Di ec ional om Re u n Spillo e , h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 Jou nal o Economics and Finance (2020) 44:810–831 824 4.3.2 Di ec ional o e u n spillo e The di ec ional o spillo e plo is illus a ed in Fig. 6. Wi h he same e ical axis anging om app oxima ely 15% o 18.5% o COMEX and LBMA in he om plo , hey ha e in his case bo h s onge spillo e sp eading abili ies, which we e mainly be ween an in e al om 20% o 30%. Two s ages can be easily iden i ied om he plo , namely be o e and a e 2016. The e u n spillo e s be o e 2016 we e mos o he ime abo e 24% o COMEX and 26% o LBMA. Then a down u n occu ed subsequen ly. Bo h o hem had an all- ime-low concu en ly du ing he middle o 2016. SGE had a posi i e spillo e jump up o 10% om ime o ime. Ne e heless, his in luence om SGE is no a all compa able o ei he COMEX o LBMA. 4.3.3 Ne e u n spillo e Sub ac ing he om alues om he o esul s leads o Fig. 7o ne e u n spillo e . Value in e als om he o dina e axis p o ide he oles all h ee ma ke s a e play- ing, i.e. COMEX and LBMA a e he spillo e ne gi e s while SGE is a spillo e ne ecei e . Bu since he dynamic plo ing was ge ing less nega i e wi h ime, he ma - ke in luence o SGE among he h ee was also sligh ly g owing. We migh expec a posi i e ne spillo e soon. 4.3.4 Pai wise e u n spillo e In addi ion o ea ing all h ee ma ke s as a whole, spillo e be ween he pai s i.e. COMEXandLBMA,LBMAandSGEaswellasCOMEXandSGEwasalsocon- side ed. The pai wise spillo e plo enables us o in es iga e he ela ionship be ween Fig. 6 Di ec ional To Re u n Spillo e , h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 Jou nal o Economics and Finance (2020) 44:810–831 825 Fig. 7 Ne Re u n Spillo e o h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 only wo ma ke s as i all he o he s do no exis . The i s ow in Fig. 8shows he spillo e s eng h om LBMA o COMEX, which was mainly close o ze o wi h sligh luc ua ion be o e mid-2016. F om la e-2016 ill almos end-2017, LBMA ans e ed a se ies o spillo e o COMEX. The eason migh be B exi in 2016. Apa om ha occasion, he spillo e be ween COMEX and LBMA we e e enly ma ched. Howe e , he ela ionship be ween SGE and ei he COMEX o LBMA is Fig. 8 Pai wise Re u n Spillo e o h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 Jou nal o Economics and Finance (2020) 44:810–831 826 en i ely dispa a e. The second ow p esen s he ne spillo e om SGE o COMEX. Rema kably, he whole dynamic olling es ima ion esul s a e lying in he nega i e dimension, om -4% o -12%. I began wi h a 6% e u n spillo e om COMEX o SGE, hen his spillo e e ec u ned s onge in a s epwise ashion un il ea ly- 2015. A e ha , he e ec i eness o e u n spillo e o SGE pushed he pe cen age back o app oxima ely -5% and emained a ha le el wi hou o he la ge change. Till he end o he sample pe iod, he e is no sign o a posi i e spillo e om SGE o COMEX. Finally, he las ow desc ibes he ela ionship be ween SGE and LBMA, which is also an one-sided spillo e ansmission om London o Shanghai, he pe - cen age scale was loa ing -6% o -14%. The pa e n was simila o he one be ween SGE and COMEX be o e mid-2017. F om hen on, COMEX-SGE pai wise spillo e emained a ound -6% while a s onge endency o a pe cen age spillo e highe han 10% om LBMA o SGE appea ed in he LBMA-SGE pai wise pa e n. 4.4 Condi ional di ec ional ola ili y spillo e dynamic olling window plo s As Diebold and Yilmaz (2009) men ioned, spillo e in ensi y is indeed ime- a ying and he na u e o he ime- a ia ion is s ikingly di e en o e u n and ola ili y. Using a dynamic olling window plo , i was al eady shown ha s a ic ables canno ully summa ise he dynamic o spillo e pa e n. Now we mo e o he spillo e plo s o ola ili y and examine he la e a gumen . 4.4.1 Di ec ional om ola ili y spillo e As o he e u n es ima ions, one i s gene a es he di ec ional om ola ili y spillo e in Fig. 9, which illus a es he spillo e ecei ed by he espec i e ma ke on he o dina e om he o he wo ma ke s. In gene al, COMEX and LBMA ecei e mo e ola ili y spillo e (bo h scales a e om a ound 10% o abo e 16%, and lie abo e 15% o mos o en) han SGE (scale om 5% o app oxima ely 20%, bu he majo pa lies unde nea h he 15% line). Di e en om he p e ious cases, ola ili y dynamics a e no as ola ile as he e u ns’. Apa om mino -inconsis encies a some speci ic ime poin s, i s ays ai ly cons an du ing mos o he ime. This is he i s poin which dis inguishes om he e u n spillo e . The big change was qui e simila o he e u n es ima ion, namely a lowe s age a e he beginning high le el spillo e om la e-2013 ill mid-2014. Then, COMEX jumped back o i s p e ious high-le el o app oxima ely 15%, LBMA also wen back o a la le el sligh ly lowe han he i s high s age. In con as o hem, SGE didn’ bounce back, he spillo e s eng h dec eased slowly e en when he e was ano he low a mid-2014 and small spike mid-2016. Finally, a e se e al s eps, SGE eached i s nadi in 2017, he spillo e was e en lowe han 5% du ing la e-2016 and he i s wo- hi ds o 2017. On he o he hand, COMEX and LBMA did no ha e an ob ious medium end. A e hei second spillo e pla eaux om mid-2014 o la e-2015, ano he la ge dec ease ol- lowed in ea ly-2016, which las ed only o a sho pe iod and bo h spillo e s eng hs jumped back o hei p e ious le el di ec ly a e wa ds. The hi d d op in COMEX and LBMA appea ed in 2017 ( om app oxima ely 15% o 11%) while o SGE i was Jou nal o Economics and Finance (2020) 44:810–831 827 Fig. 9 Di ec ional F om Vola ili y Spillo e , h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 an accompanying uphea al. The bigges eason o his change was a sudden ising exchange a e o he Chinese Yuan du ing he end o Augus ill mid-Sep embe 2017. 4.4.2 Di ec ional o ola ili y spillo e Figu e 10 illus a es he di ec ional o ola ili y spillo e . One s ill obse es simila pa e n om COMEX and LBMA, in which bo h ha e spillo e ange be ween 12% Fig. 10 Di ec ional To Vola ili y Spillo e o h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 Jou nal o Economics and Finance (2020) 44:810–831 828 o 24%, whe eas he SGE’s spillo e keeps in single digi s, e en close o ze o o mos o he ime. Fo COMEX and LBMA, a sligh ly dec easing end can be iden i ied, bu o SGE, he e is no clea mo emen endency. 4.4.3 Ne ola ili y spillo e Again we use he di e ence be ween “di ec ional o” and “di ec ional om” o com- pu e he “ne ola ili y spillo e ”, which is shown in Fig. 11. He e we obse e a mi o spillo e pa e n be ween SGE and he o he wo ma ke s. While bo h COMEX and LBMA expe ience a dec easing phase be ween mid-2015 and ea ly-2018, he e was an in e sely inc easing s age o SGE. Howe e , his “inc easing s age” in he ne spillo e o SGE is ac ually a “less nega i e s age” a e all. The o al ne ola il- i y spillo e o SGE s ays en i ely in he nega i e dimension. Jus as in he case o e u n spillo e , COMEX and LBMA a e he ne sp eade s among he h ee ma ke s while SGE plays he ole o a ne ecei e . 4.4.4 Pai wise ola ili y spillo e Finally, Fig. 12 examines he pai wise spillo e s be ween any wo o he h ee ma - ke s. As can be obse ed om he i s plo window, COMEX and LBMA we e again mos ly e en wi h he ola ili y spillo e s aying a ound 0 wi hin mos o he ime pe iod. The spillo e s om COMEX and LBMA o SGE ha e simila pa e ns. How- e e , he pai wise spillo e be ween LBMA and SGE is ma ginally s onge han he one be ween COMEX and SGE. Du ing mid-2015 and la e-2017, SGE pushed bo h spillo e s om COMEX and LBMA ai ly close o ze o, p obably due o he in e na ionalisa ion o SGE du ing ha pe iod. Fig. 11 Ne Vola ili y Spillo e o h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 Jou nal o Economics and Finance (2020) 44:810–831 829 Fig. 12 Pai wise Vola ili y Spillo e o h ee ma ke s, 23. No . 2012 - 15. Aug. 2018 5 Conclusions In his pape , he spillo e s eng h be ween COMEX, LBMA and SGE was exam- ined using he me hod om Diebold and Yilmaz (2012). Bo h s a ic, as well as dynamic esul s, p o e ha he spillo e s eng h o SGE is s ill compa ably mino , and SGE as an eme ging exchange emains isola ed as compa ed o COMEX and LBMA. Ne e heless, h ough he dynamic olling-window plo , an inc easing spillo e can be obse ed since mid-2015, when SGE s a ed i s in e na ionaliza- ion: howe e , he e does no appea o be any de ini e signs o SGE becoming as s ong as he o he wo olde ma ke s in he sho e m, and he cu en si ua ion is a om building a ipa i e con on a ion be ween hose h ee exchanges. COMEX and LBMA a e s ill he dominan spillo e playe s in he gold ma ke s p esum- ably because o hei s onge in es con idence which is based on hei mo e solid ounda ions, longe ading windows as well as la ge ading olume. Acknowledgemen s The au ho is g a e ul o he edi o o he Jou nal o Economics and Finance, M . Payne, o his ime and e o o conside ing he pape and gi ing me his oppo uni y o e ising i . I would also like o ex end my deepes g a i ude o my e iewe s, who p o ided me a lo o use ul commen s and c i iques, which poin ed ou he p oblems I ha e missed, ga e me a di ec ion o imp o ing my wo k, and inspi i ed me he ollowing esea ch a ge s. Thank you! Funding In o ma ion Open Access unding p o ided by P ojek DEAL. Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a c edi line o he Jou nal o Economics and Finance (2020) 44:810–831 830 ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o his licence, isi h p://c ea i ecommonsho g/licenses/by/4.0/. Re e ences Anan halakshmi A (2015) Shanghai gold exchange in alks o lis p oduc s on CME. h ps://www. eu e s. com/a icle/us-gold-china-sge/shanghai-gold-exchange-in- alks- o-lis -p oduc s-on-cme-idUSKBN0 P610220150626. (accessed: 26.06.2015) Ba un´ ık J, Kˇ ehl´ ık T (2018) Measu ing he equency dynamics o inancial connec edness and sys emic isk. J Financ Econ 16(2):271–296. h ps://doi.o g/10.1093/jj inec/nby001 Cine C, Gu dgie C, Lucey BM (2013) Hedges and sa e ha ens: an examina ion o s ocks, bonds, gold, oil and exchange a es. In Re Financial Anal. h ps://doi.o g/10.1016/j.i a.2012.12.001 CME G oup (2017) T ading COMEX gold and sil e - COMEX s SHFE/TOCOM. CME G oup Diebold F, Yilmaz K (2009) Measu ing inancial asse e u n and ola ili y spillo e s, wi h applica ion o global equi y ma ke s. Econ J 119(Janua y):158–171. h ps://doi.o g/10.1111/j.1468-0297.2008. 02208.x Diebold F, Yilmaz K (2012) Be e o gi e han o ecei e: P edic i e di ec ional measu emen o ola ili y spillo e s. In J Fo ecas 28(1):57–66. h ps://doi.o g/10.1016/j.ij o ecas .2011.02.006 Diebold F, Yilmaz K (2014) On he ne wo k opology o a iance decomposi ions: measu ing he connec edness o inancial i ms. J Econom. h ps://doi.o g/10.1016/j.jeconom.2014.04.012 Ga man M, Klass M (1980) On he es ima ion o secu i y p ice ola ili ies om his o ical da a. J Bus. h ps://doi.o g/10.1086/296072 Koop G, Pesa an MH, Po e SM (1996) Impulse esponse analysis in nonlinea mul i a ia e models. J Econom Lucey BM, La kin C, O’Conno F (2014) Gold ma ke s a ound he wo ld - who spills o e wha , o whom, when? Appl Econ Le 21(13):887–892. h ps://doi.o g/10.1080/13504851.2014.896974 Mandelb o B (1963) The a ia ion o ce ain specula i e p ices. J Bus. h ps://doi.o g/10.1086/294632 Pesa an HH, Shin Y (1998) Gene alized impulse esponse analysis in linea mul i a ia e models. Econ Le . h ps://doi.o g/10.1016/s0165-1765(97)00214-0 Roge s LCG, Sa chell SE (1991) Es ima ing a iance om high, low and closing p ices. Ann Appl P obab. h ps://doi.o g/10.1214/aoap/1177005835 Rosenbe g JV, T aub LG (2006) P ice disco e y in he o eign cu ency u u es and spo ma ke . h ps://doi.o g/10.2139/ss n.922317 Shanghai Gold Exchange (2018) China gold ma ke epo 2018 S e ens Analy ics (2019) Con inuous u u es. h ps://www.quandl.com/da abases/SCF. ( isi ed on Feb. 9, 2019) Wo hing on AC, Pahla ani M (2007) Gold in es men as an in la iona y hedge: coin eg a ion e idence wi h allowance o endogenous s uc u al b eaks. Appl Financ Econ Le . h ps://doi.o g/10.1080/ 17446540601118301 Yang D, Zhang Q (2000) D i Independen Vola ili y Es ima ion Based on High, Low, Open, and Close P ices. J Bus. h ps://doi.o g/10.1086/209650 Publishe ’s no e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. Jou nal o Economics and Finance (2020) 44:810–831 831