Volailiy and risk spillovers beween oil, gold, and Islamic and convenional GCC banks Walid Mensi a,b, Shawka Hammoudeh c,d, Idries Mohammad Wanas Al-Jarrah e, Khamis Hamed Al-Yahyaee b*, Sang Hoon Kang a Deparmen of Finance and Accouning, Universiy of Tunis El Manar, Tunis, Tunisia b Deparmen of Economics and Finance, College of Economics and Poliical Science, Sulan Qaboos Universiy, Musca, Oman Email: walid.mensi@fseg.rnu.n Email: yahyai@squ.edu.om c Lebow College of Business, Drexel Universiy, Philadelphia, Unied Saes d Energy and Susainable Developmen (ESD), Monpellier Business School, Monpellier, France Email address: shawka.hammoudeh@gmail.com e College of Business and Economic, Qaar Universiy, Qaar Email: idries@qu.edu.qa f Deparmen of Business Adminisraion, Pusan Naional Universiy, Busan 609-735, Republic of Korea Email: sanghoonkang@pusan.ac.kr Absrac This paper examines ime-varying risk spillovers and hedging effeciveness beween wo major commodiy markes (oil and gold) and boh he Islamic and convenional bank sock indices for five GCC counries (Bahrain, Kuwai, Qaar, Saudi Arabia and UAE), using he DECO- FIGARCH model and he spillover index of Diebold and Yilmaz (2012). The resuls of he DECO-FIGARCH model show evidence of a weak average condiional correlaion beween all he GCC bank sock indices and he wo commodiy markes. Moreover, we find significan risk spillovers beween hese Islamic and convenional GCC bank sock indices and he commodiy markes. The spillovers rise considerably during he 2008-2009 global financial crisis and he 2014-2015 oil price plunge periods Furher, oil, gold, and he convenional bank sock index of Saudi Arabia, Kuwai and Qaar are ne sources of volailiy spillovers ino he oher markes, while all he Islamic banks and convenional banks of UAE and Bahrain are ne volailiy recipiens of volailiy spillovers. Finally, we provide evidence assering ha including gold and oil in a GCC porfolio offers beer bu differen diversificaion benefis and hedging effeciveness for he GCC banks. Keywords: GCC, Islamic banking, Commodiy markes, Risk spillovers, Hedging effeciveness JEL classificaion codes: G14; G15 1
1. Inroducion The risk spillover analysis is an imporan ool for conrolling and managing bank risks. Significan effors have been dedicaed o he undersanding of excessive risk in financial sysems and idenificaion of hose policies ha may reduce such risk, paricularly following he 2007 2008 global financial crisis (e.g., he Lehman broher collapse on Sepember 15, 2008). The global financial crisis (GFC) has changed he global financial landscape. Indeed, he GFC had amplified he shocks and ransmied hem from one marke ino oher markes and hen involved he whole global financial sysem. The informaion se (including exreme movemens or major evens) for one bank may have a subsanial predicive power of oher banks and oher markes (e.g., foreign exchange markes, sock markes and commodiy markes). In addiion o marke fundamenals, many plausible causes or evens may jusify he presence of risk spillovers beween banks or markes including conagion, invesor senimens and reacion o news among ohers. Thus, deermining he sources and recipiens of risk spillovers is imporan for financial secors and markes. The presence of risk spillovers beween inernaional financial markes has moived individual and insiuional invesors o find new hedges and new havens o proec heir invesmens. The Islamic finance indusry is considered one of he new havens as refleced by he phenomenal growh of his indusry during he las decade, paricularly in he wake of he GFC. The indusry has grown in deph and breahs and offers alernaive ineresing insrumens, which make i a modern and viable financial sysem. The Islamic banking is he larges secor of he Islamic finance indusry and has asses in mos counries of he world. The Islamic banking follows he Shariah-compliance principles which screen invesmens for Islamic finance. In conras o convenional finance, Islamic banking is sricly prohibied from carrying ou business aciviies such as hose ha involve gambling, firearms, alcohol, 2
speculaion and ineres-gaining. The banking Shariah compliance is also based on risksharing and no deb accumulaion. Paricipaion in Islamic banking asses has also increased despie he major urbulences in he las few years. The core paricipaing counries include Bahrain, Qaar, Indonesia, Saudi Arabia, Malaysia, Unied Arab Emiraes, Turkey, Kuwai and Pakisan whose asses reached 93% of oal banking indusry asses, exceeding US$920 billion in 2015. An imporan subse of hose counries, known as QISMUT which includes Qaar, Indonesia, Saudi Arabia, Malaysia, UAE and Turkey, has a marke share sanding a 80% of he inernaional banking asses. The Islamic banking indusry in he Middle Easern counries, paricularly hose in he Gulf Cooperaion Council (GCC), has grown more rapidly han heir convenional counerpars in hose regions. Based on recen banking secor asse projecions, five counries including Saudi Arabia, Qaar, Pakisan, UAE and Turkey are idenified o be he key players by 2020 on he accouns of average annual growh rae and banking secor asses. In erms of banking paricipaion, Saudi Arabia, Kuwai, Bahrain and Qaar will be he major players in erms of having he highes banking marke shares by 2020. 1 We noe ha he risk spillovers are more visible during financial urmoil periods which increase asse volailiy. In fac, during he wo recen crisis periods, he cross-marke linkages among inernaional financial markes and beween financial and oil markes have increased, paricularly following he 2008-2009 GFC and he 2010-2012 European sovereign deb crisis (ESDC). The fligh-o-qualiy phenomenon has araced he aenion of individual invesors, insiuional invesors (e.g., banks) which are primarily ineresed in financial safey, and policymakers alike. In he safe haven lieraure, numerous sudies view gold as a good refuge asse (i.e., a srong hedge and a good safe haven asse) in financial 1 hp://www.ey.com/publicaion/vwluasses/ey-world-islamic-banking-compeiiveness-repor- 2016/$FILE/ey-world-islamic-banking-compeiiveness-repor-2016.pdf 3
markes (Baur and Lucey, 2010; Baur and McDermo, 2010; Mensi e al., 2014a, 2015b, 2016). Gold is an imporan asse for GCC banks for culural, financial and geopoliical reasons. Culurally, here is an obsession wih gold in regions ha populae he people and businesses of he Orien including India, China and he GCC areas. Moreover, financial reasons moivae us o include gold in our analysis because his shiny meal is used o hedge agains unanicipaed flucuaions and excessive variaions in oil prices, sock markes, ineres raes and foreign exchange markes (Allayannis and Ofek, 2001). Finally, he GCC banks are locaed in a region plagued wih high geopoliical risks and riven wih regional wars. Thus, he yellow meal should provide financial sabiliy o boh Islamic and convenional GCC banks. The associaion beween gold and he safe haven propery is also imporan for GCC banks o mainain liquidiy, credi risk managemen and porfolio risk assessmen. For he GCC cenral banks, gold reserves are also a sore of value, a way of fosering domesic currencies which are pegged o he dollar, and a guaranee of paymens o deposiors. On he oher hand, he oil marke has experienced srong insabiliy during he las en years. More explicily, he oil price exceeded US$145/barrel in summer 2008 and hen plunged o less han US$30/ barrel in June 2014. This insabiliy has affeced he financial secors especially hose of he oil-rich Gulf counries during he previous boom and bus oil cycles. Oil price shocks influenced he GCC bank profiabiliy hrough oil income (e.g., lending o privae secor), increased business aciviy and enhanced excess liquidiy in he banking sysem. Moivaed by hose consideraions, i is imporan o analyze he risk spillovers beween he Islamic and convenional GCC bank sock indices in Bahrain, Kuwai, Qaar, Saudi Arabia and UAE and wo sraegic commodiy markes including he gold and oil 4
markes. We have consruced hose Islamic and convenional indices from individual bank sock prices for each GCC counry o achieve his objecive. This sudy aims o examine he direcional spillovers and ne spillovers among he Wes Texas Inermediae (WTI) crude oil and gold wih boh he Islamic and convenional bank indices for he five GCC counries (Bahrain, Kuwai, Qaar, Saudi Arabia and UAE) over he period June 1, 2006 o Sepember 19, 2016. Furher, we conduc porfolio risk managemen by quanifying he Islamic and convenional GCC opimal porfolios weighs, hedge raios and hedging effeciveness for hose GCC markes. The choice of he Islamic GCC banks is moivaed by he large size of heir asses and heir imporan growh during he las few years, in addiion o heir risk exposure and business models ha are complian wih he Shariah rules ha govern he GCC economies. Regarding he commodiy markes, gold is widely known as a refuge asse and oil is seleced because he GCC counries (excep Bahrain) are heavily oil dependen as indicaed earlier. Consequenly, he GCC financial markes are sensiive o oil price shocks. Oil should also command more imporance in Islamic GCC markes because many of he major oilproducing GCC counries srongly follow he Islamic faih, and his underlines a risk feedback beween hese markes, paricularly during periods of economic downurns and financial urmoil. The increase in crude oil prices in he las few years has increased he growh rae of he Islamic bank asses in he Middle Eas (Khan, 2010). We noe ha hese wo imporan commodiies (oil and gold) offer differen volailiy and reurn behaviors during he financial urmoil periods. This is anoher reason ha moivaes us o selec hese wo imporan commodiies wih hose GCC markes. The sudy conribues o he exising lieraure in four aspecs. Firs, i uses he dynamic equicorrelaion (DECO)-FIGARCH model of Engle and Kelly (2012) o invesigae he dynamic condiional correlaions beween he commodiy markes wih Islamic and 5
convenional bank indices. The (DECO)-FIGARCH model ouperforms he sandard GARCH model because of is abiliy o deec he long-range memory process in he condiional variance of he financial ime series and assumes a ime-varying correlaion among financial asses. The DECO model provides an efficien way o assess in deph he variabiliy in correlaions during differen marke regimes (Hammoudeh e al., 2016). I is worh noing ha his model is a special case of he DCC model in which he correlaions across all pairs of he markes are equal, and he common equicorrelaion is ime-varying (Pan e al., 2016). Moreover, he DECO model provides consisen esimaes of he DCC parameers in large sysems, and in his sudy i allows one o quanify he linkages of he commodiy and bank markes as a common group for he purpose of porfolio diversificaion in asses issued by hese markes. Second, he paper analyzes he direcional spillovers and ne spillovers across he oil and gold commodiies and GCC bank markes, using he spillover index developed by Diebold and Yilmaz (2012). Wihin his framework, we use a rolling sample approach o deec he ime-varying dynamics of he volailiy spillover index, since he recen financial crises direcly affeced he volailiy srucures beween hese markes. Third, we assess he ne volailiy spillover index and he sensiiviy of his index as a robusness check. Finally, we quanify he opimal porfolio weighs, hedge raios, and hedging effeciveness for he major commodiy and bank sock asse porfolios. In fac, invesors aemping o offse heir risk exposures and risks agains downurn marke movemens should adjus heir hedge raios and rebalance sraegies in accordance wih he movemen of he bank marke (bear or bull markes) condiions. Our empirical resuls show ha he average condiional correlaion beween hese markes is close o zero. They also indicae ha oil, gold, and he convenional banks of Saudi Arabia, Kuwai and Qaar are a ne source of volailiy spillovers, while all he Islamic banks 6
and he convenional banks of UAE and Bahrain which house foreign banks are a ne recipien of volailiy spillovers. Regarding he opimal weighs for he commodiies, we show ha i is opimal for he GCC banks o hold more oil han gold in a diversified porfolio. The resuls for he average opimal hedge raios show ha all he raios are weak for gold and close o zero for oil. Finally, gold offers he bes hedging effeciveness for UAE, Qaar and Saudi Arabia banks, while oil provides he highes hedging effeciveness for he banks of Bahrain which is a minor oil producer. The remainder of his paper is organized as follows. Secion 2 develops he mehodology used in his sudy. Secion 3 describes he daa and conducs preliminary analyses. Secion 4 provides and discusses he empirical resuls. Secion 5 draws implicaions for risk managemen and porfolio diversificaion. Secion 6 provides concluding remarks. 2. Empirical mehod In his secion, we describe he empirical mehods which begin wih a descripion of he bivariae DECO-FIGARCH model ha assesses he dynamic correlaions beween he commodiies and banking sock indexes under consideraion. We also employ he volailiy spillover index of Diebold and Yilmaz (2012), which in his sudy idenifies he dynamics of direcional volailiy spillovers across he Islamic and convenional GCC banking sock indexes. 2.1. The DECO-FIGARCH model To explore he ime variaion in condiional correlaions beween he commodiy and GCC bank markes, his sudy presumes ha he reurn-generaing process r described by an auoregressive moving average (ARMA (1,1)) model as follows: can be 7
r 1r 1 1 1, wih z h, (1) where [0, ), 1 disribuion ~ ST0,1,, 1 and he innovaions { z } follow he Suden- z. 2 The condiional variance h is posiive wih probabiliy one and is a measurable funcion of he variance-covariance marix. 1 The FIGARCH p, d, q model of Baillie e al., (1996) is expressed as follows: h [1 ( L)] [1 [1 ( L)] ( L)(1 L) ], (2) 1 1 d 2 where is he mean of he process, is he GARCH parameer, is he finie order lag polynomials and d is he fracional differencing parameer capuring long memory o be esimaed where 0d 1, and L denoes he lag operaor. The fracional differencing operaor (1 L) d is defined as: (1 L) d k ( k d) L (3) ( d) ( k1) k0 The FIGARCH model provides a greaer flexibiliy for modeling he condiional variance and can disinguish beween he covariance saionary GARCH model for d 0 and he nonsaionary IGARCH model when d 1, while for 0d 1 he FIGARCH model is sufficienly flexible o allow for an inermediae range of persisence. In order o obain dynamic correlaions beween he variables under consideraion, we review he DCC model of Engle (2002). Assume ha he condiional expecaion E and he condiional variance-covariance marix E 1 H, where 1 0 E is he condiional expecaion for using he informaion se available a ime. The condiional variancecovariance marix H can be wrien as: H D R D, (4) 1/2 1/2 2 The Suden- disribuion is esimaed wih he parameer, which represens he number of degrees of freedom (df) and measures he degree of lepokurosis displayed by he densiy (Fiorenini e al., 2003). 8
where R i j, is he condiional correlaion marix, while he diagonal marix of he condiional variances is given by D diag hi,,, hn, side of Eq. (4), raher han srucure: * * 1/2 1/2. Engle (2002) models he righ-hand H, direcly by proposing he following dynamic correlaion R Q Q Q, (5) Q * diag Q, (6) Q q i j, 1 a b S au 1u 1 bq 1, (7) where u ui,,, u n, is he sandardized residuals (i.e. ui, i, / hi, ), S s E u u i j is he n n uncondiional covariance marix of u, and a and b are nonnegaive scalars saisfying ab 1). The resuling model is called he DCC model. In his conex, Aielli (2013) proves ha he esimaion of he covariance marix Q in his way is inconsisen because E R E Q, and suggess he following he consisen DCC model (cdcc model) for he correlaion-driving process: where * *1/2 *1/2 1 Q a b S a Q u u Q bq, (8) 1 1 1 1 1 * S is he uncondiional covariance marix of Q u. Engle and Kelly (2012) sugges ha be modelled by using he cdcc process o obain he condiional correlaion marix *1/2 Q and hen aking he mean of is off-diagonal elemens. This approach which reduces he esimaion ime is called he dynamic equicorrelaion (DECO) model. The scalar equicorrelaion is defined as: 1 2 n1 n qi j, J R J n n n n n q q DECO n n i j i cdcc 1 1 1 1 (9) ii, jj,, 9
where q DECO DECO,, 1, 1 DECO i j adeco ui u j bdeco qi j,, which is he ( i, j) h elemen of he marix Q from he cdcc model. We hen use his scalar equicorrelaion o esimae he condiional correlaion marix: R 1 I J, (10) n n where J n is he n n marix of ones and I n is he n -dimensional ideniy marix. This process allows one o represen he comovemen degree of a group of commodiy fuures wih a single ime-varying correlaion coefficien. Noe ha he esimaion of he DECO model is carried ou using a wo-sep maximum likelihood of he probabiliy densiy funcion of a bivariae Suden- disribuion expressed as: where v 2 l 2 v 2 log v / v v 2 1/ 2log H 1/ 2 1 v 2log 1 H / v 2, (11) is he Gama funcion, v is he degree of freedom for he Suden disribuion, H is a condiional variance-covariance marix. is a parameer vecor wih all of he coefficiens of he DECO-FIGARCH model. 2.2. Spillover Index framework We apply he generalized VAR mehodology, variance decomposiion, and he generalized volailiy spillover index of Diebold and Yilmaz (2012) o examine he direcional spillovers and ne spillovers across he Islamic and convenional GCC banking sock indexes. Following Diebold and Yilmaz (2012), we assume a covariance saionary n -variable VAR( p ): p i1 i 1, (12) y y 10
where y is he n1vecor of endogenous variables, are n n auoregressive coefficien marices, and is a vecor of error erms ha are assumed o be serially uncorrelaed. If he VAR sysem above is a covariance saionary, hen a moving average represenaion is wrien i as y A, where he n n coefficien marix A j obeys a recursion of he form j0 j Aj 1Aj 1 2Aj2 p Aj p, wih A 0 being he n n ideniy marix and A j 0 for j 0. The oal, direcional and ne spillovers are generaed by he generalized forecas-error variance decomposiions of he moving average represenaion of he VAR model. The framework of he generalized variance decomposiions eliminaes any dependence of he resuls on he ordering of he variables. Koop e al. (1996) and Pesaran and Shin (1998) propose he following H -sep-ahead generalized forecas-error variance decomposiion: ij H 1 H 1 jj e h 0 i Ah e j H 1 h0 ea Ae i h h i 2, (13) where is he variance marix of he vecor of errors, and jj is he sandard deviaion of he error erm of he j h equaion. Finally, e i is a selecion vecor wih a value of one for he i h elemen, and zero oherwise. The spillover index yields a n n marix H H, where each enry gives he conribuion of variable j o he forecas error variance of variable i. The own-variable and cross-variable conribuions are conained in he main diagonal and he off-diagonal elemens of he H marix, respecively. Because he own- and cross-variable variance conribuion shares do no sum o one n under he generalized decomposiion (i.e., H decomposiion marix is normalized by is row sum as follows: j1 ij 1 ), each enry of he variance ij 11
ij H j1 ij ij H, (14) n H n n wih ij H 1 and ij H j1 n by consrucion. j1 This allows one o define a oal spillover index as: TS H n i, j1 ij H H i, j 1, i j i, j 1, i j ij H 100 100. n n n (15) This index measures he average conribuion of he spillovers of a volailiy shock from one marke o all (oher) markes o he oal forecas error variance. Addiionally, his index is flexible and enables an idenificaion of he direcional spillovers among all markes. Specifically, he direcional spillovers received by marke i from all oher markes j are defined as: DS n H ij H 100 100. H n j1, ji ij j1, ji i j H n i, j1 ij n (16) Similarly, he direcional spillovers ransmied by marke i o all oher markes j are defined as: DS n H ji H 100 100. H n j1, ji ji j1, ji i j H n i, j1 ji n (17) The se of direcional spillovers provides a decomposiion of oal spillovers ino hose coming from (or o) a paricular marke. In he presen applicaion, his means ha his spillover marix H consiss of he main diagonal elemens reflecing own-marke spillovers, and he off-diagonal elemens reflecing cross-marke spillovers. Finally, by subracing Eq. (17) from Eq. (16), we compue he ne volailiy spillovers from each marke o all oher markes as: NS i H DS H DS H. (18) i j i j 12
The ne spillovers demonsrae wheher a marke is a receiver or a ransmier of spillovers in ne erms. I is also our ineres o examine he ne pairwise spillovers (NPS) as following: NPS H ji H ij H 100. ik H jk H i, k 1 j, k 1 ij n n (19) The ne pairwise spillovers beween markes i and j is simply he difference beween he gross shocks ransmied from marke i o marke j and hose ransmied from j o i. 3. Daa and preliminary analysis 3.1. Daa Our daa se consiss of daily closing spo prices for he Islamic and convenional bank sock price indexes for he five GCC counries (Bahrain, Kuwai, Qaar, Saudi Arabia and UAE). We exclude Oman because of a lack of sufficien daa. We use he daily closing spo price daa for he hree-monh fuures conrac of he Wes Texas Inermediae (WTI) crude oil, which is he benchmark or he reference price for he U.S. crude oil. These fuures conacs are raded on he NYMEX. The daa for he closing oil price, which is expressed in USD/barrel, is exraced from he US Energy Informaion Adminisraion (EIA) websie (www.eia.gov). The gold fuures price, expressed in USD/roy ounce, is raded on COMEX in New York. The daa for gold and bank sock price series are exraced respecively from he World Gold Council and Bloomberg daabase. The sample period runs from June 1, 2006 hrough Sepember 19, 2016 (a oal of 2221 observaions) for he wo commodiy markes and all Islamic and convenional GCC counries. The beginning of he sample period is dedicaed by he daa availabiliy for he Gulf banks. This period covers several episodes of wide insabiliies and crises including he 13
dramaic increases in oil prices hroughou 2007 and early 2008, he 2008-2009 GFC, he 2010-2012 ESDC, he 2011-2013 Arab spring, he drasic oil price plunge since June 2004 and he gradual recovery of he global sock markes. To analyze he risk spillovers beween he wo commodiy and five bank sock markes, we consruc wo banking sock price indices for he Islamic banks and convenional banks for each of he five GCC counries. Following Fakhfekh e al. (2016), each consruced banking sock index is defined as a weighed average of he bank sock prices for each GCC counry. Formally, each Islamic or convenional GCC banking sock price index is defined n CBi as i 1 CBT Price, wih CBi denoing he sock marke capializaion of he bank i, n i CB T CB, Price denoes he daily observed ih bank sock price and n is he number 1 i of banks in a specific GCC marke. We calculae he coninuously compounded daily reurns by aking he difference in he logarihms of wo consecuive GCC bank sock indices or commodiy prices. Figure 1 displays he dynamic reurns of he oil, gold, Islamic and convenional GCC bank sock index reurns. The figure shows some periods of significan flucuaions and ha all commodiies and GCC banks exhibi a volailiy clusering. We apply he Markov-swiching-dynamic regression (MS-DR) o deec wo ranquil and volaile regimes in he reurn series ha allows one o idenify he beginning and end of each phase of he financial crises, which is of grea imporance when one deals wih he cross-marke spillovers issue. 3 The shaded regions highligh he regimes of excess volailiy according o he MS-DR and show he effecs of GFC on hese banks. In fac, he oil and gold prices and all of he Islamic and convenional GCC bank sock index reurns exhibi significan abrup variaions 3 The MS-DR model idenifies he exisence of wo regimes: regime zero (he sable regime) and regime one (he volaile/crisis regime). For furher deails of he MS-DR model, see Hamilon (1988) and Hamilon and Susmel (1994). 14
during he crisis period 2008-2009 bu wih differen magniudes. Concerning he wo commodiy markes, he figure reveals ha oil is more volaile han gold. This graphical analysis will be confirmed by he descripive saisics below. Among he Islamic GCC bank sock indices, we noe ha he Islamic bank index for Bahrain (Saudi Arabia) is less (more) affeced by he GFC due o he concurren collapse in oil prices which is he main revenue source for he Saudi economy. Bahrain is a minor oil producer. Furher, he Saudi banks are more segmened han he Bahrain banks and do no receive much help from inernaional banks bu are sill affeced by global oil shocks. As for he convenional GCC bank sock price indices, he Kuwai bank index is more affeced by he GFC. Anoher volailiy clusering can be observed during he period 2014-2016 which corresponds o he recen oil price plunge. In fac, he economies of hese Gulf counries, paricularly ha of Saudi Arabia, are highly oil dependen as indicaed. 15
Fig. 1. Dynamics of he commodiy and GCC index reurns Noe: The shaded areas highligh he regimes of excess volailiy according o he Markov swiching-dynamic regression (MS-DR). 16
3.2. Preliminary analysis Table 1 provides he descripive saisics of he daily oil, gold and Islamic and convenional bank sock index reurns for he GCC banks. This able shows ha he average daily (price) index reurns are negaive for oil and all GCC bank index reurn series excep for he Bahrain convenional bank index and he gold price. For he case of he convenional and Islamic GCC bank indices, he (uncondiional) volailiy as measured by he sandard deviaions is he highes for he Islamic GCC banks han for heir convenional counerpars. Among he Islamic (convenional) GCC bank sock indexes, ha of UAE (Saudi Arabia) is he mos volaile, while ha for Bahrain (UAE) is he leas volaile. The UAE banks may be affeced by he heavy borrowing of Dubai and he afermah of he 2006 Dubai deb crisis. On he oher hand, he Bahrain banks have a life line suppor wih major American and European banks since Bahrain is an inernaional financial cener in ha region. Fig. 1 plos he ime variaions of he commodiy (oil and gold) price reurns and he GCC bank share index reurns. This figure shows evidence of sylized facs for all of he reurn series, such as volailiy clusering and asymmeric volailiy. This graph also shows evidence of a presence of nonlineariy and srucural breaks. Thus, a nonlinear model (such our GARCH model) fis he daa. The shaded regions underline he regimes of excess volailiy according o he Markov-swiching dynamic regression (MS-DR) and provide evidence of srong volailiy clusering during he GFC period for all he considered markes bu wih differen magniudes. Despie he similariy of he Islamic GCC banks, hose banks presen differen magniudes of sylized facs. Furher, he effec of he oil price plunge on he bank reurns is observed beween years 2014-2015. Among he commodiy markes, he oil marke is more volaile han he gold marke and is also more volaile han he GCC bank markes. These descripive saisics are in concordance wih he dynamics of he commodiy and sock index reurns displayed in Fig. 1. 17
The skewness coefficiens are negaive for he majoriy of he GCC bank sock index reurn series (excep for he Qaari Islamic bank sock index). The kurosis coefficiens are above hree for all he index reurn series, which indicaes ha he probabiliy disribuions of he commodiy price reurns, Islamic and convenional GCC banking index reurns are skewed and lepokuric, hus rejecing normaliy. This finding is also confirmed by he Jarque-Bera saisics. To examine he saionariy process, we use wo popular convenional uni roo ess, he augmened Dickey and Fuller (1979) and Phillips and Perron (1988) ess, and he saionariy es of he Kwiakowski e al. (1992) es o check saionary of he series. The resuls indicae ha all he index reurn series are saionary a he convenional levels. Furher, we examine he exisence of he ARCH effec using he LaGrange muliplier es of Engle (1982), and he resuls show ha all he reurn series exhibi an ARCH behavior. Therefore, he esimaion of a GARCH model is appropriae for modeling sylized facs such as fa-ails, clusering volailiy and persisence for he commodiy price and bank index reurns. According o he Ljung-Box es resuls, we provide evidence for serial correlaions for he residual and squared residual a he convenional level of 1%. The presence of a long memory process in he financial markes is essenial for analyzing he ime-varying risk spillovers. For his purpose, we jusify he use of he FIGARCH model. However, we use four popular convenional long memory (LM) ess namely he Hurs-Mandelbro R/S es, Lo s modified R/S es, he Gaussian semi-parameric (GSP) es of Robinson and Henry (1999), and he GPH es of Geweke and Porer-Hudak (1983). 4 Tables 2 and 3 presen he LM ess resuls for he reurn and he squared reurn (as a proxy of volailiy) series for he commodiy (oil and gold) prices and he Islamic and convenional bank indexes for he GCC counries. The resuls do no rejec he null hypohesis 4 For furher informaion on he LM ess, he reader can read he aricle of Mensi e al. (2014b, 2014c). 18
of no LM propery when examining he reurn series for socks and commodiies, wih he excepion of he UAE convenional bank sock index when we consider he GSP es. This resul is consisen wih he LM lieraure. However, he LM resuls for he sock index reurns change radically when we consider volailiy. In fac, we srongly rejec he null of no LM propery for all commodiy and GCC bank squared reurns a he 1% significance level, regardless of he applied LM ess. In sum, he squared reurns may be governed by a fracionally inegraed model for all cases. These resuls also jusify he use of he FIGARCH model. 19
Table 1: Descripive saisics of he pairs of he daily oil price, gold price, and he Islamic and convenional GCC bank sock index reurns Gold WTI SAI SAC QAI QAC UAI UAC BAI BAC KUI KUC Mean 0.0328-0.0233-0.0522-0.0347-0.0051-0.0002-0.0121-0.0188-0.0639 0.0007-0.0114-0.0187 Max. 8.6249 16.414 11.208 21.284 13.644 10.011 14.353 11.645 12.548 5.9954 13.351 9.1586 Min. -9.8104-19.164-20.749-22.564-14.252-15.131-11.372-15.348-15.934-7.7241-14.734-6.5142 Sd. dev. 1.3681 2.6971 2.0145 1.9846 1.9233 1.6406 2.0367 1.3952 1.5634 1.0076 2.0204 1.0811 Skewness -0.3736-0.1939-0.8969-0.3656 0.0281-0.7132-0.0282-0.3574-0.5238-0.4423-0.2489-0.0605 Kurosis 8.9056 9.9403 15.406 30.084 13.815 15.740 11.602 25.250 19.799 9.6833 10.650 10.161 J-B 3277. *** 3278. *** 14535. *** 67904. *** 10819. *** 15202. *** 6844. *** 45839. *** 26206. *** 4204. *** 5436. *** 4745. *** Q(30) 41.102 58.246 ** 56.743 *** 61.969 *** 90.842 *** 109.79 *** 92.137 *** 65.814 *** 63.585 *** 74.581 *** 71.125 *** 69.236 *** Q 2 (30) 395.08 *** 1952. *** 371.05 *** 467.44 *** 1294.9 *** 1449.8 *** 1708.7 *** 177.88 *** 126.86 *** 462.07 *** 3650. *** 1912. *** ADF -48.26 *** -31.74 *** -44.27 *** -50.10 *** -40.67 *** -40.87 *** -42.85 *** -30.02 *** -43.86 *** -46.82 *** -48.31 *** -45.83 *** PP -48.27 *** -49.33 *** -44.33 *** -50.08 *** -40.58 *** -40.83 *** -42.81 *** -44.16 *** -43.90 *** -47.38 *** -48.40 *** -45.82 *** KPSS 0.2458 0.1244 0.1987 0.1195 0.0787 0.0963 0.3256 0.1762 0.0695 0.1508 0.1606 0.1851 LM- ARCH(10) 10.876 *** 35.259 *** 13.777 *** 68.843 *** 31.540 *** 34.564 *** 54.238 *** 14.602 *** 5.4886 *** 16.158 *** 56.025 *** 44.023 *** Noes: SAI, SAC, QAI, QAC, UAI, UAC, BAI, BAC, KUI and KUC are respecively he Saudi Arabia Islamic bank index, Saudi Arabia convenional bank index, UAE Islamic bank index, UAE convenional bank index, Bahrain Islamic bank index, Bahrain convenional bank index, Kuwai Islamic bank index and Kuwai convenional bank index. J-B denoes he empirical saisics of he Jarque-Bera es for normaliy. The Ljung-Box Q(30) and Q 2 (30) ess for no auocorrelaion of residuals and square residuals, respecively. ADF, PP and KPSS are he empirical saisics of he Augmened Dickey and Fuller (1979), and he Phillips and Perron (1988) uni roo ess and he Kwiakowski e al. (1992) saionariy es, respecively. The ARCH- LM(10) es of Engle (1982) checks he presence of ARCH effecs. The aserisk *** denoes he rejecion of he null hypoheses of normaliy, no auocorrelaion, uni roo, saionariy, and condiional homoscedasiciy a he 1% significance level. 20
Table 2. Resuls of he long memory ess for he reurns of oil price, gold price, and he Islamic and convenional GCC banking sock indexes Gold WTI SAI SAC QAI QAC UAI UAC BAI BAC KUI KUC Panel A: Hurs-Mandelbro R/S es R/S saisic 1.413 1.217 1.639 1.395 1.240 1.755 1.777 1.689 1.135 1.794 1.331 1.407 Panel B: Lo s modified R/S es ( q 1) 1.431 1.224 1.591 1.441 1.159 1.643 1.695 1.632 1.101 1.789 1.349 1.389 ( q 5) 1.473 1.226 1.528 1.444 1.099 1.619 1.657 1.509 1.063 1.703 1.375 1.414 Panel C: GSP es -0.029 0.0015 0.029-0.028-0.008-0.004 0.041 0.062 dm T / 4 *** -0.002 0.069 *** -0.031-0.0002-0.070 dm T /16 dm T /32 0.0612 0.113 dm T /64 Panel D: GPH es 0.45 0.192 dm T 0.5 0.034 dm T 0.55 0.129 dm T 0.6 0.024 dm T 0.106 0.195-0.166-0.141 0.095 0.220 0.170-0.018-0.011-0.019 0.007-0.004-0.026-0.003-0.044-0.012 0.033 0.019 0.076 0.037 0.022-0.074 0.038 0.133 0.044 0.031 0.021 0.090 0.0387-0.041 0.050 0.026 0.102 0.037 0.083 0.241 *** 0.122 0.074 0.151 0.150 0.193 0.289 *** 0.183 *** 0.345 *** 0.270 *** 0.209 0.244 0.281 *** 0.190 *** -0.089-0.001 0.012 0.243 0.198 0.051-0.040 0.229 0.124 0.124 0.054 0.037 0.170 0.231 *** 0.026 0.184 *** 0.169 0.078 0.144 0.225 0.118 0.030 0.238 *** 0.265 *** 0.181 0.196 0.188 0.165 Noes: The criical values of he Hurs-Mandelbro R/S es and Lo s modified R/S analysis are 1.862 and 2.098 a he 5% and 1% significance levels, respecively. The numbers in he parenheses are he sandard deviaions of he esimaes. q in Lo s modified R/S es is he number of lags of auocorrelaion. mdenoes he bandwidh for he GSP and GPH ess. ** and *** indicae significance a he 5% and 1% levels, respecively. 21
Table 3. Resuls of he long memory ess for he squared reurns of oil price, gold price, and he Islamic and convenional GCC banking sock indexes Gold WTI SAI SAC QAI QAC UAI UAC BAI BAC KUI KUC Panel A: Hurs-Mandelbro R/S es R/S saisic 3.190 *** 4.459 *** 4.566 *** 2.433 *** 4.938 *** 5.019 *** 4.466 *** 2.799 *** 3.422 *** 4.343 *** 5.894 *** 5.189 *** Panel B: Lo s modified R/S es ( q 1) 3.093 *** 4.038 *** 4.247 *** 2.232 *** 4.542 *** 4.690 *** 3.871 *** 2.515 *** 3.300 *** 4.018 *** 5.250 *** 4.753 *** ( q 5) 3.190 *** 3.011 *** 3.654 *** 2.120 *** 3.617 *** 3.665 *** 2.818 *** 2.265 *** 2.933 *** 3.478 *** 3.937 *** 3.626 *** Panel C: GSP es 0.234 dm T / 4 *** 0.415 dm T /16 *** 0.541 dm T /32 *** 0.555 dm T /64 *** Panel D: GPH es 0.45 0.583 dm T *** 0.5 0.607 dm T *** 0.55 0.603 dm T *** 0.6 0.549 dm T *** Noes: See he noes of Table 2. 0.374 *** 0.417 *** 0.830 *** 0.574 *** 0.576 *** 0.828 *** 0.871 *** 0.529 *** 0.206 *** 0.353 *** 0.268 *** 0.587 *** 0.727 *** 0.592 *** 0.339 *** 0.413 *** 0.100 *** 0.175 *** 0.257 *** 0.275 *** 0.307 *** 0.276 *** 0.279 *** 0.230 *** 0.266 *** 0.438 *** 0.402 *** 0.494 *** 0.496 *** 0.425 *** 0.423 *** 0.506 *** 0.296 *** 0.481 *** 0.390 *** 0.474 *** 0.605 *** 0.475 *** 0.489 *** 0.442 *** 0.373 *** 0.378 *** 0.399 *** 0.074 *** 0.358 *** 0.435 *** 0.461 *** 0.470 *** 0.115 *** 0.196 *** 0.327 *** 0.431 *** 0.501 *** 0.417 *** 0.374 *** 0.256 *** 0.181 *** 0.238 *** 0.282 *** 0.448 *** 0.557 *** 0.475 *** 0.375 *** 0.340 *** 0.201 *** 0.355 *** 0.454 *** 0.413 *** 0.371 *** 0.508 *** 0.448 *** 0.384 *** 0.344 *** 0.625 *** 0.768 *** 0.527 *** 0.493 *** 0.689 *** 0.845 *** 0.828 *** 0.308 *** 0.431 *** 0.558 *** 0.566 *** 0.585 *** 0.579 *** 0.642 *** 0.573 *** 22
4. Empirical resuls 4.1. Esimaion of marginal model To avoid he non-synchronous rading effec in he world s financial markes, we have followed Forbes and Rigobon (2002) o employ wo-day rolling reurns based on each aggregaed marke index. Furhermore, we have applied he wo-day rolling reurns o he mulivariae DECO-FIGARCH model for modelling all sylized facs for sock reurns such as volailiy clusering, volailiy persisence (long memory), and ime-variaions in condiional volailiy and correlaion. The resuls of he esimaion of he mulivariae DECO-FIGARCH (1, d, 1) model beween he commodiy (oil and gold) and GCC bank sock (Islamic and convenional) markes are summarized in Table 4. Panels A, B and C summarize he mean and variance equaions, and he average correlaion for each pair and he diagnosic ess, respecively. 5 Looking a Panel A, we find ha he auoregressive (AR(1)) parameer of he mean equaion is posiive and saisically significan a he 1% level for all cases (excep for he convenional bank sock index of UAE). This resul indicaes ha he pas reurns are insananeously and rapidly included in he curren reurns for hese banks. Moreover, he fracional inegraed coefficien (d) is highly significan for all he reurn series, suggesing a high level of persisence. Among all series, he Islamic bank sock index for Saudi Arabia exhibis he highes parameer, while he convenional bank sock index of UAE presens he lowes long memory parameer. We also noe ha he Islamic bank sock indices for Saudi Arabia, Qaar and Kuwai are more persisen han heir convenional counerpars. Panel B shows ha he adeco coefficien is posiive and saisically significan a he convenional level, underlying he imporance of shocks beween he commodiy and boh he 5 One should noe ha he lag order (1, d, 1) is seleced by using he Akaike Informaion Crieria (AIC) and he Schwarz Informaion Crieria (SIC). The resuls are no repored here o honor space bu hey are available upon reques. 23
Islamic and convenional bank GCC sock index reurn. The b DECO parameer is significan and very close o one for all pairs. This resul corroboraes he resuls of higher persisence of volailiy across he considered markes as deermined by he variance equaion, paricularly he GARCH parameer and he d-figarch parameer. We show ha he average correlaion is close o zero (0.083) bu is saisically significan a he 1% level. This resul exhibis he presence of diversificaion invesmen opporuniies beween he markes. Addiionally, he evidence on he degrees of freedom of he Suden- disribuion (df) indicaes ha fa-ailedness characerizes he disribuion of all reurn series. Taking ogeher, he significance of he parameers a DECO, bdeco and df demonsraes he appropriaeness of using he DECO-FIGARCH model wih Suden- disribuions. The resuls of he diagnosic ess using he Ljung Box ess for serial correlaion in he sandardized residuals and he squared sandardized residuals resuls do no rejec he null hypohesis of no serial correlaion in all pairs. This finding provides no evidence of misspecificaion in our model. Fig. 3 plos he dynamic equicorrelaion for he group of he commodiy and GCC bank markes. From his figure, we can draw several ineresing findings. Firs, we observe ime-varying correlaions over he sample period, suggesing ha insiuional invesors do or should frequenly change heir porfolio srucure. Second, he correlaion is posiive and weak for all sample period, wih a higher level of correlaion during he GFC wih a value equals 0.35. This rise in correlaion beween he markes decreases he poenial of diversificaion benefis during crises. Finally, he correlaion level increases during he more recen period 2014 2015, which corresponds o he oil price collapse. This resul suppors he hypohesis of financial conagion beween he commodiy and GCC sock reurns under consideraion. Thus, he rajecories of he ime-varying condiional correlaions show ha he share prices of he Islamic and convenional GCC banks are no immune agains inernaional 24
facors and oil price shocks. Following he recen oil price decline, he condiional correlaions exhibi a gradual decrease, indicaing a gradual recovery of hese markes. Fig. 2. Dynamic equicorrelaion among he commodiies and GCC banks Noe: The dynamic equicorrelaion beween Gold, WTI and he five GCC bank indices is esimaed from he mulivariae ARMA-FIGARCH(1,d,1)-DECO model. 25
Table 4. Esimaion of he mulivariae ARMA-FIGARCH(1,d,1)-DECO model Saudi Arabia Qaar UAE Bahrain Kuwai Gold WTI SAI SAC QAI QAC UAI UAC BAI BAC KUI KUC Panel A: Esimaes of ARMA-FIGARCH(1,d,1) model Cons. 0.0304 0.0365 0.0243-0.0701 0.0127 0.0373 0.0106-0.0361-0.0274 *** 0.0163 0.0008-0.0080 (0.0264) (0.0426) (0.0290) (0.0770) (0.0253) (0.0255) (0.0299) (0.0235) (0.0081) (0.0169) (0.0255) (0.0152) AR(1) -0.0181 (0.0208) -0.0175 (0.0244) 0.0849 *** (0.0263) 0.1009 ** (0.0503) 0.0858 *** (0.0303) 0.0819 *** (0.0268) 0.0475 (0.0254) 0.0069 (0.0548) -0.0140 *** (0.0293) -0.0274 (0.0266) -0.1242 *** (0.0247) -0.0547 ** (0.0241) MA(1) 0.9812 *** (0.0032) 0.9773 *** (0.0036) 0.9647 *** (0.0133) 0.9636 *** (0.0158) 0.9478 *** (0.0088) 0.9543 *** (0.0079) 0.9571 *** (0.0055) 0.9685 *** (0.0059) 0.9182 *** (0.0094) 0.9862 *** (0.0025) 0.9800 *** (0.0030) 0.9806 *** (0.0043) Cons. 1.1145 1.7433 ** 2.0059 0.1318 3.7552 0.0070 6.6453 0.4249 *** 4.0441 0.2806 ** 0.7906 ** 0.3539 ** (0.8869) (0.7276) (1.5310) (0.0854) (3.7946) (0.0036) (4.3303) (0.1303) (3.2239) (0.1188) (0.3862) (0.1627) d-figarch 0.4823 *** (0.1324) 0.4971 *** (0.0683) 0.7240 *** (0.0752) 0.3189 ** (0.1623) 0.6628 *** (0.0908) 0.5266 *** (0.0700) 0.6126 *** (0.0609) 0.1859 *** (0.0706) 0.7097 *** (0.0924) 0.3615 *** (0.0894) 0.4542 *** (0.0662) 0.4339 *** (0.0652) ARCH 0.2442 *** 0.2956 *** 0.0072 0.2331 *** 0.2956 *** 0.3354 *** 0.0546 0.8563 *** 0.3686 ** 0.4126 *** 0.4296 *** 0.1818 (0.0990) (0.0771) GARCH 0.6718 *** 0.6844 *** (0.1265) (0.0902) Panel B: Esimaes of he DCC model DECO 0.10059 *** (0.0098) a 0.0584 *** DECO (0.0155) b 0.8929 *** DECO (0.0327) df 6.1278 *** (0.1884) AIC 27.86028 SIC 28.02476 Panel C: Diagnosic ess Q(30) 25.804 [0.6850] Q 2 (30) 19.458 [0.9298] 19.099 [0.9378] 19.963 [0.9174] (0.0894) 0.6116 *** (0.0847) 38.392 [0.1399] 7.0039 [0.9999] (0.2497) 0.1291 *** (0.1818) 37.093 [0.1744] 8.4525 [0.9999] (0.0982) 0.7155 *** (0.0981) 31.792 [0.3772] 27.581 [0.5925] (0.0683) 0.7197 *** (0.0596) 20.926 [0.8901] 11.093 [0.9993] (0.0915) 0.5268 *** (0.0836) 40.661 [0.0926] 23.626 [0.6235] (0.0981) 0.8851 *** (0.0831) 32.991 [0.3229] 12.126 [0.9984] (0.1676) 0.7183 *** (0.1279) 18.941 [0.9412] 0.2375 [1.0000] 0.0801) 0.6749 *** (0.0868) 36.962 [0.1782] 20.394 [0.9058] (0.0761) 0.6735 *** (0.0759) 35.309 [0.2315] 35.922 [0.2106] (0.1076) 0.4409 *** (0.1223) 28.963 [0.5195] 29.802 [0.4758] Noes: Q(30) and Q 2 (30) are he Ljung-Box es saisic applied o he sandard residuals and he squared sandardized residuals, respecively. The P-values are repored in brackes [.]. The sandard error values are repored in parenheses (.). The aserisks ** and *** indicae significance a he 5% and 1% levels, respecively. 26
4.2. Toal volailiy spillover index and rolling-sample spillover analysis Table 5 summarizes he esimaed resuls of he oal volailiy spillover marix. We noe ha he (i, j)h enry in each panel is he esimaed conribuion o he forecas-error variance of variable i coming from innovaions o marke j. The row sums excluding he main diagonal elemens (ermed From ohers ) and he column sums (ermed To ohers ) repor he oal spillovers o (received by) and from (ransmied by) each volailiy. The oal volailiy spillovers value is 29%, indicaing diversificaion gains. Le us firs focus on he direcional spillovers ransmied To ohers. Gold has a lower impac on he all Islamic and convenional GCC bank sock indices excep for he bank sock index of Saudi Arabia han oil. In conras, he WTI crude oil price conribues more significanly o he convenional banks han he Islamic counerpars for Bahrain, Saudi Arabia and Qaar. The crude oil acs as he price discovery ool for he convenional bank of hese hree counries. Furher, he risk spillovers beween he GCC bank markes hemselves are low because hose banks are isolaed according o counry lines and heavily involved wih he naional governmens. For insance, he Islamic Saudi banks conribue 3.8%, 2.3%, 0% and 0.2% o he forecasing variance of he Islamic banks of Qaar, UAE, Bahrain and Kuwai, respecively. Moreover, he volailiy ransmission from he Islamic and convenional GCC bank o oil and he yellow meal is weak. Taking for example he gold marke, he risk spillover coefficien is close o zero from he Bahrain Islamic and UAE convenional banks o he shiny meal, and is less han 1% for he spillovers from Saudi Arabia and Qaar. The Kuwai Islamic (convenional) bank sock index conribues o 0.98% (2.4%) o he forecasing variance of gold. 27
Table 5. Toal volailiy spillovers for he commodiy and GCC bank. To (i) From(j) Gold WTI SAI SAC QAI QAC UAI UAC BAI BAC KUI KUC From ohers Gold 87.7 2.18 0.9 0.7 0.7 0.5 0.3 0.07 0.05 0.59 0.98 2.4 9.3 WTI 5.44 67.9 0 0.21 2.1 0.5 0.9 0.03 0.23 1.04 7.81 14 32.1 SAI 0.16 0.02 76 4.48 1.8 3.5 5.2 0.03 0.6 0.05 1.13 6.9 23.8 SAC 0.69 0.58 1.4 90.8 0.7 0.6 0.4 0.06 0.56 0.25 0.58 2.4 8.2 QAI 1.57 3.29 3.8 0.73 48 21 4.8 0.09 0.08 2.67 4.69 9.9 52.1 QAC 3.35 3.68 4.9 0.78 18 46 5.6 0.03 0.01 4.9 3.52 8.9 54.1 UAI 0.31 3.88 2.3 1.29 2.4 5.7 71 0.05 1.29 0.2 1.82 9.6 28.9 UAC 0.03 0.13 0.4 0.17 0.1 0.1 0 96.9 0.53 0.75 0.09 0.6 3.1 BAI 0.01 0.1 0 0.11 0.6 0.4 0.8 0.43 91.6 1.96 3.54 0.4 8.4 BAC 0.63 2.59 0 0.06 2.7 6.9 1 0.14 0.55 74.5 5.91 5.1 25.5 KUI 2.13 15 0.2 0.2 5.3 3.7 2.7 0 0.9 6.32 46.9 17 53.1 KUC 1.47 13.2 0.3 0.95 4.9 4.8 2.9 0.14 0.04 3.61 17.6 52 49.9 To ohers 15.8 44.7 14 9.7 40 47 25 1.1 4.8 22.3 47.6 76 348 All 104 113 91 101 87 93 96 98 96.5 96.8 94.5 128 Toal: 29% Noes: The underlying variance decomposiion is based on a daily VAR of order 4 (as deermined by he Schwarz informaion crierion) using he generalized VAR spillover index of Diebold and Yilmaz (2012). The (i,j)h elemen of he able shows he esimaed conribuion o he variance of he 10-sep-ahead forecas error of i coming from innovaion shocks o variable j. The diagonal elemens (i=j) are he own variance share esimaes, which show he fracion of he forecas error variance of marke i ha is due o is own shocks. The las column From ohers shows he oal spillovers received by a paricular marke from all oher markes, while he row To ohers shows he spillover effec direced by a paricular marke o all oher markes. The lower righ corner Toal indicaes he level of oal spillovers. 28
Similar resuls can be seen for he crude oil marke. This resul demonsraes he role played by he precious meal marke as a refuge asse during he marke urmoil, and his resul is in line wih previous sudies including, among ohers, Baur and Lucey (2010), Baur and McDermo (2010), Bredin e al. (2015), Mensi e al. (2015a). For he graphical evidence, we plo he ime-varying volailiy spillover index in Fig. 3. A close inspecion of his figure, we show ha he volailiy spillovers aain heir maximum level during 2008 2009 and 2015 2016, which as indicaed earlier corresponds o he GFC and he oil price plunge periods. Moreover, we see he commodiy-bank linkages are highly influenced by he poliical and economic evens as illusraed in Figure 3. The 2007 2008 commodiy crisis, he 2011 Arab spring revoluion and he changes of raes by he U.S. Federal Reserve beween 2009-2013 and he Chinese sock markes crash in summer 2015 increase he spillovers beween hese markes, which reduces invesmen diversificaion opporuniies for he considered markes. I is worh noing ha in July 2015 he Shanghai sock marke had fallen 30% over hree weeks, as more han half of he lised companies filed for a rading hal in an aemp o preven furher losses. Again, he Shanghai index fell in Augus by 8.48%, which is he larges fall since 2007. 29
Fig. 3. The dynamics of he oal volailiy spillover index Noes: The dynamics of oal volailiy spillovers are calculaed from he forecas error variance decomposiions of 10-sep-ahead forecass wih 200-day rolling windows. 4.3. Ne volailiy spillover We deepen our analysis by examining he ime-varying behavior of he volailiy spillovers. More specifically, we sudy he ne pairwise volailiy spillovers, which provide a fruiful informaion abou he direcional volailiy spillovers among he bank-commodiy fuures markes. We divide he oal volailiy spillover index ino wo direcional spillovers: i) he receivers of volailiy spillovers, ermed direcionally as from, and ii) he ransmiers of volailiy spillovers, ermed direcionally as o. The ne dynamic volailiy spillover index is hen compued by subracing he direcional o spillovers from he direcional from spillovers. The posiive (negaive) values indicae a source (a recipien) of reurn and volailiy o (from) ohers. Table 6 repors he ne direcional pairwise index o idenify he main ne recipiens an d conribuors o he volailiy spillovers. Regarding he direcional spillovers received From ohers, he resuls exhibi ha he WTI oil receives more shocks from he res of he markes 30
han gold. Also, he Islamic GCC bank sock indices receive more risk han he convenional b ank indices for all he GCC markes. In fac, he Qaar Islamic bank sock index receives more risk from he remaining markes (commodiies and he res of banks) han any of he oher Isl amic bank sock indices, while he Islamic bank sock index of Bahrain receives less risk han he oher markes. The Qaar convenional bank index receives 6.73% and 47.3% of he risk s pillovers from gold, oil and oher Islamic and convenional GCC bank sock indexes. From his able, we can conclude ha boh he gold and oil markes are ne conribuors of risk. More precisely, gold receives risk spillovers from he Saudi Islamic and convenional bank index, he UAE convenional bank index, he Bahrain Islamic bank index and he Kuwai convenional bank index, while gold is a ne conribuor o he remaining markes. In fac, gold conribues o he error forecas variance of he convenional bank index of Qaar a meager of 2.88%, he convenional bank index of he Bahrain 0.04%, and he Islamic bank index for Kuwai 1.15%. Again, his resul is in line wih previous works (see Baur and Lucey, 2010; Mensi e al., 2015b; Mensi e al., 2016) on he abiliy of gold o be a good hedge and/or a safe haven asse, no only for he sock markes bu also for he bank markes of he GCC economies. The WTI crude oil receives risk spillovers from gold (3.26%), he Islamic banks of Bahrain (0.1%) and he convenional banks of Kuwai (0.6%) On he oher hand, oil conribues o he oher markes wih risk spillovers ranging from 0.1% for he UAE convenional banks o 7.21% for he Islamic banks of Kuwai. On he oher hand, we find ha he risk spillovers beween GCC bank sock markes is weak since hese markes are well capialized, srongly supervised by heir respecive cenral banks and domesically isolaed from oher GCC banks. The GCC governmens also deposi heir oil revenues in hose banks and borrow from hem, and hus here is no room for conagion o ake place among hem. 31