A Structural VAR Analysis on The Linkages of China s Stock Market with Global Financial Markets

Transcription

1 A Srucural VAR Analysis on The Linkages of China s Sock Marke wih Global Financial Markes Yen-Hsiao Chen 1, Daniel Sanamaria 2, Yang Liu 3 Absrac Economic heories sugges ha financial variables are insananeously affeced by each oher. To empirically analyse heir conemporary relaionships, we specify a srucural vecor auoregressive model ha pairs China s equiy marke wih Asian and European equiy markes. The foreign exchange and money markes are also included for analysis since hey are risk sources o sock markes. The empirical resul indicaes an increasing inegraion of China s sock marke wih he Asian sock markes hrough ime, bu i is no symmerical; China s sock marke has become more influenced by he Asian sock markes, while is influence on he Asian sock markes has no equally increased. The resul also suggess ha he shock of foreign exchange markes is a common risk source o he Asian sock markes. While China s and he European sock markes have no conemporary relaionship once he common facor of US sock marke was considered, he resul suggess ha he exchange rae and ineres rae risks are cross-asses o equiy and cross-regional beween China and Europe in he global financial crisis period. Key words: impulse response; variance decomposiion; reurn spillover; volailiy spillover; wihin-asse behaviour; cross-asses behaviour; ransmission JEL Classificaion Codes: 1. Corresponding auhor: Dr Y.-H. Chen, yen.chen@canerbury.ac.uk. Telephone: +44(0) Address: Canerbury Chris Church Universiy Business School, Norh Holmes Road, Canerbury, Ken-CT1 1QU, Unied Kingdom. 2. Dr D. Sanamaria. daniel.sanamaria@canerbury.ac.uk. Address: Canerbury Chris Church Universiy Business School, Norh Holmes Road, Canerbury, Ken-CT1 1QU, Unied Kingdom. 3. Dr Y. Liu. yang.liu@canerbury.ac.uk. Address: Canerbury Chris Church Universiy, Augusine House, Canerbury, Ken CT1 1QU, Unied Kingdom. 1

2 1. Inroducion The seminal work of King and Wadhwani (1990) has led o a discussion in inernaional finance lieraure abou how shocks are ransmied across borders. A range of differen mehodologies has been proposed o assess he linkages of financial markes. Several erminologies, such as conagion, spillover, and inerdependence, have also been coined and defined o sudy he comovemen of financial asse prices in he inernaional scale (e.g., Engle, Io and Lin, 1994; Eichengreen e al., 1995; Forbes and Rigobon, 2002; Karolyi, 2003; Dungey and Marin, 2004; Boyer e al., 2006; Bekaer e al. 2009). Wih he rise of emerging economies, many subsequen sudies have devoed o invesigae how economic or financial shocks ransmi from developed markes o emerging markes, for example, Wei e al. (1995), Liu and Pan (1997), Masih and Masih (1999),, Ng (2000), Miyakoshi (2003), Lee, Rui and Wang (2004), Worhingon and Higgs (2004), Tai (2007), Kenourgios, Samias, and Palalidis (2011), Samarakoon (2011), Kokavuori-Örnberg, Nikkinen and Äijö (2013), Beirne e al. (2013), Bekaer e al. (2014), and Morales and Andreosso-O'Callaghan (2014). To furher undersand he issue, in his paper we analyse he linkage of China s sock marke wih oher sock markes across he globe. In addiion, we also look a he linkages of sock markes wih oher financial markes, i.e. foreign exchange markes and money markes. For doing his, we employ a wo-sage analysis ha was buil up on he framework of facor models. In analysing he inerdependence of asse reurns, he facor models based on arbirage pricing heory are usually applied, which sae ha he price of financial asses are deermined by boh common facors and idiosyncraic facors (Slonik, 1974; Mahieu and Schoman, 1994; Masson, 1999; Bekaer e al. 2005; Corsei e al. 2005; Dungey and Gajurel, 2014; Kaminsky, e al. 2016). Given he dominan impac of US sock marke on world s sock markes, on he firs sage we apply a world-facor model o sock 2

3 markes, wih he world facor proxied by he S&P 500 index, in order o obain he exra world-facor sock marke risk ha is no resuled from he common sock risk facor. On he second sage, we pair China s sock marke wih anoher sock marke while allowing for he relevan foreign exchange rae in currency marke and ineres rae differenial in money markes o be included in he analysis. We focus on China because he Chinese economy has been one of he fases growing economy since is shifing o marke-based economy in 1978, wih is sock exchange becoming one of he larges markes in he world in erms of marke capialisaion since is re-esablishmen in There are also some lieraures ha focused on China s sock marke, including, for example, Bailey (1994), Hu e al. (1997), Huang e al. (2000), Haemi-J and Roca (2004), Wang and Firh (2004), Li (2007), and Wang and Di Iorio (2007), in which he general finding is ha he US marke plays an imporan role in he Eas Asian sock markes. Noneheless, wheher China s sock marke was inerdependen wih inernaional sock markes is inconclusive. More recen sudies researched on he ineracions beween China s sock marke and oher sock markes sill generae differen resuls. Huyghebaer and Wang (2010), for example, find ha China s sock marke is neiher influenced by he USA s nor affeced by Hong Kong s and Singapore s sock markes afer he Asian financial crisis. Wang and Wang (2010) insead find ha, beween 1994 and 2002, he Chinese marke was conneced wih he US and Japan hrough volailiy spillovers, which were bi-direcional raher han uni-direcional from he developed o developing markes. Zhou, Zhang and Zhang (2012) find ha he US sock marke has dominan volailiy impacs on he world equiy markes, while China s sock marke also has had impacs on oher markes since They also find ha China s marke is more conneced o hose markes wih he same culural background, i.e., Hong Kong and Taiwan, han wih Wesern and oher Asian markes. Alhough he sudies on he linkages beween China s and oher major inernaional sock markes have increased generally due o 3

4 he increasing inegraion of he Chinese economy wih he res of he world, he lieraure is sill relaively few. Apar from he wihin-asse behaviour beween sock markes, he exan lieraure also suggess some facors common o sock markes. For example, as argued by he Calvo (1996), herd menaliy of global invesors would lead o he simulaneous wihdrawals from a group of relaed markes, resuling in he comovemen of sock prices in emerging markes. Masson (1998) also argues ha invesor psychology and marke senimens can lead o he simulaneous movemen of sock markes. By conras, Eichengreen e al. (1996) sugges ha spillover effecs end o appear in counries wih close rade linkages, and Glick and Rose (1999) sugges ha currency crises affec geographically- proximal counries ied ogeher by inernaional rade. Since he Asian economies expor heir goods o he common desinaions, namely he US, Japan and China, he hird-marke compeiion would lead o he compeiive devaluaion of heir currencies. Furher, Hu e al. (1997) find ha he volailiy of foreign exchange marke can explain he momens of sock reurn series, while Van Rijckeghem and Weder (2001) find ha rade and financial linkages are highly correlaed. More recenly, Wali (2011) suggess ha apar from rade and financial inegraions, common moneary policy and he eliminaion of exchange rae volailiy can also lead o sronger sock marke synchronisaion. Furhermore, Jansen and Tsai (2010) find ha socks in bear sae respond more han socks in bull sae o a surprising moneary policy, while Baig and Goldfajn (1998) and Chrisopher e al. (2012) find ha sovereign spreads and sovereign credi raings influence regional sock and bond markes. The sudy of Panchenko and Wu (2009) finds he inerdependence of sock marke and credi marke in emerging economies, whereas oher sudies, for example, Elyasiani and Mansur (1998), Johnson and Soenen (2002), and Lusig e al. (2011), find ha ineres rae differenial beween counries is an imporan common risk facor in currency markes and sock markes. By following he 4

5 lieraure, we herefore also include wo inernaional facors, namely foreign exchange rae and ineres rae differenial, in our analysis. 1 2 Given he large marke value and he fas developmen pace of China s sock marke, much more researches are needed o undersanding is linkages wih oher financial markes. However, he previous sudies mainly focus on he ineracion of China s sock marke wih oher sock markes. Our firs conribuion o he lieraure is ha we exend he sudy on China s sock marke o include currency asse and money asse as well as equiy asse. 3 Our sudy herefore is an effor o undersand he wihin-asse behaviour of sock markes and cross-asse behaviour among sock marke, currency marke and money marke. Our second conribuion is ha we employ a Srucural Vecor Auoregressive model (SVAR). The SVAR models were originally applied in macroeconomical sudies. While hey have also been applied o financial economics, he sudies are relaively few. Some examples include Hess and Lee (1999), Gallagher and Taylor (2002), Binswanger (2004) and Fraser and Groenewold (2006), which all focus on he relaionship beween sock prices and real economic variables. There are also some SVAR sudies on he cross-asses behaviour and he wihin-asse behaviour of financial markes. For example, Ehrmann e al. (2011) apply a SVAR analysis o sudy he ineracions of he socks, bonds, money markes and exchange raes beween he USA and he Euro area in he period of 1989 o In general, hey find ha he inernaional cross-asses behaviours are significan, alhough he wihin-asse 1 China adoped a managed floaing exchange rae regime wih reference o a baske of currencies in July 2005, bu which currency is included in he Chinese yuan s baske and he weigh of each componen currency have never been revealed. 2 The People s Bank of China abolished he upper limis on iner-bank lending rae in June 1996, which since hen has been lef o deermine by he forces of ransacions. In January 2007, he Shanghai Inerbank Offered Rae was formally launched. 3 The Chinese yuan has been heavily pegged o he US dollar, especially in he Global Financial Crisis period. However, here have also been some reforms; he original daily rading band of he Chinese yuan agains he US dollar was ±0.3%, which was widened o ±0.5% in May 2007 and ±1.0% in April

6 behaviours unsurprisingly provide he sronges inernaional ransmission of financial shocks. Dungey e al. (2010) apply heir SGARCH model, which is a SVAR augmened wih he mulivariae GARCH(1,1) process for resolving he idenificaion problem, o sudy he four sock markes of Hong Kong, Korean, Indonesia, and Thailand. They find ha in he Asian crisis period Hong Kong and Korea became more conagious o ohers and Indonesia became more sensiive o ohers, while Thailand became more deached o ohers. However, heir sudy only focuses on he wihin-asse behaviours of inernaional equiy markes. In Dungey e al. (2015), hey exend he SGARCH model o embed smooh ransiion funcions so as o daing he phases of crisis and non-crisis periods. Their model allows hem o idenify he differen phases during he Global Financial Crisis as well as he ineracions beween financial markes. However, heir sudy only focuses on he cross-asses behaviours beween equiy, REIT and Treasury bond in he USA. By conras, our sudy focuses on he ineracions of China s sock marke wih oher inernaional sock markes, which, according o our survey, has never been analysed by he SVAR models. In oher words, we analyse no only he cross-asses behaviours of China s sock marke wih oher financial markes bu also is wihin-asse behaviours wih oher sock markes. Because he SVAR models are designed o resolve he simulaneiy issue beween endogenous variables, he mehodology allows us o look beyond he modelling of daa and look ino he ineracions of financial variables implied by economic heories. Furhermore, we also conribue o he esing of reurn spillover and volailiy spillover. To uilise he SVAR model, some assumpions mus be made for he insananeous relaionships beween financial asses. The classical economics explains ha governmen s economic behaviours will influence boh he values of currency asse and money asse. In an efficien sock marke, he invesors would respond o local news and global news fairly quickly and hence, he srucural shocks in foreign exchange marke, money marke as well as he sock 6

7 marke should all be he sources of risk o he sock marke. Wih his economic reasoning, we specify a cause-effec relaionship in our SVAR model ha incorporaes his view. In addiion, he specificaion is over-idenified ha allows us o furher es he significance of facor loadings in sock marke. On one hand, we herefore can es he causaliy beween sock markes, i.e., wheher he sock reurns relaionship is unrelaed, uni-direcional or bidirecional. In oher words, he srucural form in our model can idenify he conemporary reurn spillover beween financial markes. On he oher hand, if one is willing o view he srucural shocks of financial markes as risk sources o one anoher, hen he immediae impulse response funcion of our SVAR model can be hough of as he measuremen of volailiy spillover from one marke o he oher, while he insananeous relaionship beween hese financial variables can be hough of as he risk ransmission channel. In addiion o he es of he significance of risk sources, he forecas error variance decomposiion funcion of he SVAR model can measure he relaive magniudes of hese risk sources. Our analysis herefore no only examines he exisence of volailiy spillover, bu i also provides informaion on he relaive imporance of financial variables in affecing each oher. Finally, for esing he general applicabiliy of our model, he empirical analysis looks a he daa range from he beginning of 2005 o he end of 2014, during which period he global financial environmen has changed dramaically due o he Global Financial Crisis and he Eurozone Deb Crisis. Specifically, we apply he model o hree subperiods; he pre-crisis period, he Global Financial Crisis period and he Eurozone Deb Crisis period. We also apply he empirical framework o a range of developed and developing markes across he globe. Our analysis is emporal hrough several global evens and spaial in inernaional scale, which makes conribuion o beer undersanding he cross-asses and wihin-asse behaviours of financial variables. 7

8 The res of he sudy is srucured as follows. In secion 2 we presen he srucural form of conemporary reurn relaionship, and how i is o be esed based on a SVAR model. Secion 3 discusses he daa source and heir preliminary saisics. Secion 4 presens and discusses he empirical resuls, and in paricular, we also discuss how he over-idenified SVAR is used o deermine he risk ransmission channel based on a sequence of likelihood raio ess. Secion 5 hen offers he concluding remarks. 2. Empirical Mehodology 2.1 The simulaneous srucural form We begin he analysis by assuming ha all sock markes share a common world facor ha deermines he average level of sock marke reurns across inernaional markes. The proxy for he world facor in our model is he S&P500 index of he USA and herefore, he worldfacor model can be wrien as: R i i i 2 i, yi ~ 0, y w y (1) where R i are he sock marke reurns of he sample counry i and i = CHN, JPN, KOR, TWN, HKG, SGP, AUS, RUS, DEU, FRA or GBR, and he variable w represens he common world facor ha impacs upon all sock markes wih he loading i for he counry i. The reurn of sock marke no relaed o he common world shock is herefore spli ino wo pars, i represens he expeced excess reurn of sock marke i above he reurn of he world marke w, and y i represens he exra world-facor risk. One advanage of using he single world-facor model is ha he exra world-facor covariance is no zero because he 8

9 model makes no assumpion abou he wihin-asse and cross-asses covariance, i.e., y, y 0,i j cov. i j To analyse he conemporary effecs of he cross-asses behaviour amongs sock marke, foreign exchange marke and money marke as well as he wihin-asse behaviour beween sock markes, we assume ha he financial variables in ineres have he following insananeous relaionship, y (2) 1 12y2 11 y (3) 2 21y1 23 fx 24id 2 2 fx 34id 3 fx (4) id 4 id (5) where y 1 is he sock marke reurns of China and y is he sock marke reurn of one of he 2 res 10 counries acquired from he world facor model (1), fx is he depreciaion rae of he Chinese currency wih respec o counry 2 s currency, and id is he ineres rae differenial of China wih respec o he corresponding counry. The coefficien allows he model o cach he insananeous response of a variable o anoher variable. For example, he significance of 12 would indicae ha China s sock marke is conemporarily affeced by he price change of counry 2 s sock marke, while he significance of 23 would indicae ha counry 2 s sock marke is conemporarily affeced by he change of China-counry 2 foreign exchange rae. The erms in he equaions are idiosyncraic facors ha are unique o he specific asse markes. The conribuions of idiosyncraic shocks o he volailiy of 9

10 asse prices are deermined by he loadings 0. These facors are assumed o be sochasic processes wih zero mean and uni variance, i.e., ~ 0,1. To explain, noice from equaion (5) ha he variable id is assumed o be compleely auonomous in he srucure, while he variable fx is causally prior o he sock markes. The variable fx is se o be responsive o he change of he variable id in equaion (4) because in macroeconomics he Mundell-Fleming model dicaes ha boh variables would respond o governmen s economic policies, and in ineres rae pariy condiion he change of ineres rae differenial would change he expecaion of exchange rae. In equaion (2) and (3), he seing of y being responsive o he change of he variables id and 2 fx suggess ha counry 2 s sock marke is more developed and opened and hence, invesors respond more quickly o news in foreign exchange and money markes. The above insananeous relaionship beween he financial asses can be wrien in a srucural form, Ay B (6) if we define, y y y fx id 1 2 (7) 1 fx id (8) A (9)

11 B (10) To complee he specificaion of he srucural model, he idiosyncraic shocks should have he cross-secional and emporal properies, E E 0,i j (11) i, j, 0, i j (12) i, j,, 2.2 Srucural Vecor Auoregression To saisfy he properies of idiosyncraic shocks (11) and (12), we wrie he equaion (6) ino a Srucural Vecor Auoregression (SVAR) model ha caches he dynamics of he own auocorrelaions and cross auocorrelaions wihin he vecor y, Ay A y A y B 1 1 p p (13) where he coefficien marices, A i for srucural errors are whie noise. i 1,, p, are srucural coefficiens, and he To ransform he SVAR ino a reduced form VAR for esimaion, one can pre-muliply he above equaion (13) by he inverse of A o ge, y A v A A y A A y A B (14) p p 11

12 Wih he operaion, we hus have a sable four-dimensional VAR(p) process for empirical esimaion. To see ha equaion (14) is a reduced form VAR ha allows for esimaion, we can define 1 c A v, A j A 1 A j 1, and u A B, so ha, j 1, p y c A y A y u 1 1 p p (15) where y is a (4 1) vecor of he observed variables defined in equaion (7), he Aj s (j=1,, p) are he (4 4) reduced form coefficien marices and c is a (4 1) vecor of inercep erms allowing for he possibiliy of nonzero mean E ( y ). Finally, u is 4-dimensional innovaion process wih E ( u ) 0, E( u u ) u and E( u u s ) 0, s. The resricions imposed on he marices A and B in he SVAR equaion (13) can be used o idenify shocks and race hese ou by employing impulse response funcions and forecas error variance decomposiion. For such purposes, i should be noed ha he insananeous relaionship specified in equaions (6) o (10) can be shifed o he inerpreaion of he unexpeced par of heir shocks, since in our case we have Au B. More specifically, he reduced form residuals required for he esimaion of impulse responses can be rerieved from he SVAR model by u A 1 B, and he variance-covariance marix required for he esimaion of forecas error variance decomposiion can be calculaed based on 1 1 A B BA u. To allow he for he idenificaion of he marix B, we furher assume ha he srucural innovaions are orhonormal, i.e., is covariance marix is an ideniy marix E I, and herefore, 1 1 A A u (16) 12

13 Due o he symmery of he covariance marix, he minimum number of resricions for 2 idenificaion is K K( K 1) 2, where K is he dimension of he VAR sysem, and herefore we need o impose 22 resricions in he SVAR for jus-idenificaion. In he specificaion of he marix B in equaion (10), here are 12 resricions and hence, he marix A requires anoher 10 resricions for jus-idenificaion. The specificaion of he marix A has 11 resricions, and herefore we have an over-idenified SVAR wih one degree of freedom. To esimae he parameers specified in he marices A and B, one hen can minimise he negaive of he concenraed log-likelihood funcion, ln L c KT T 2 T 2 T 1 ~ 1 A, B ln2 ln A ln B rab B A (17) u where K is he dimension of he sysem, T is he sample size, A and B are he deerminans of he marices, ~ u is he sample esimae of he variance-covariance marix for he reduced formed error process u, and r. is he race of he indicaed marix. 3. Daa and Preliminary Saisics For he analysis, we collec from he Daasream hree daa ses wih daily frequency for he period from he 31 s of December 2004 o he 31 s of December The firs daa se comprises of he daily closing price index for 12 sock exchanges across he globe, including SSE Composie Index of China, TOPIX of Japan, KOSPI of Korea, TAIEX of Taiwan, HANG SENG Index of Hong Kong, STI of Singapore, ASX200 of Ausralia, MICEX Index of Russia, DAX30 of Germany, CAC40 of France, FTSE100 of he UK, and S&P500 of he USA. The daily sock marke reurns, R i, of each counry are he logarihmic firs difference of he sock price index muliplied by 100. The second daa se are he foreign exchange indices of he Chinese currency wih respec o counry 2 s currency, where counry 2 is one 13

14 of he 11 counries lised above. 4 The depreciaion rae of he Chinese currency wih respec o counry 2 s currency, fx, is calculaed as he logarihmic firs difference of he foreign exchange index muliplied by 100. The hird daa se are he shor-erm ineres raes of each counry, including he inerbank overnigh of China, he uncollaeral overnigh of Japan, he overnigh call rae of Korea, he overnigh repo rae of Taiwan, he overnigh deposi of Hong Kong, he inerbank call of Singapore, he overnigh deposi of Ausralia, he overnigh deposi of Russia, he overnigh inerbank of Germany, he Eu-Franc shor-erm deposi of he France, he LIBID/LIBOR overnigh of he UK, and he overnigh repo of he USA. 5 The ineres rae differenial of China wih respec o counry 2 id is calculaed as he difference beween he Chinese ineres rae and counry 2 s ineres rae muliplied by 100. For he problem of non-synchronous rading hours, whereby he opening and closing hours wih reference o he Greenwich Mean Time of various sock markes are differen, we adop a simple mehod similar o Cai e al. (2009) by lagging one day on he US and European daa ses. Wih regard o he differen sample observaions in differen counries due o each counry s specific holiday and insiuional arrangemens, we follow he exising lieraure, for example, Wang and Firh (2004) and Wang and Wang (2010), by excluding he daes when any one of he sock markes was closed. Because here have been several dramaic changes in he financial environmen faced by inernaional markes during he full period, we spli he full sample ino hree sub-periods; namely, he pre-crisis period from he 31 s of December 2004 o he 8 h of Augus 2007, he Global Financial Crisis (GFC) period from he 9 h of Augus 2007 o he 8 h of May 2010, and 4 Due o he daa availabiliy, we apply he cross rae calculaions o acquire he exchange rae of he Chinese currency wih respec o he Russian, Souh Korean or Taiwanese currency, whereby he common currency for calculaion is he US dollar. 5 Mos of he overnigh raes are middle raes, excep for Germany and China where he raes are offered raes. 14

15 he Eurozone Deb Crisis (EDC) period from he 9 h of May 2010 o he 31 s of December The daes chosen for spliing he full sample are based on he informaion acquired from he Fullime of he Financial Crisis released by he Federal Reserve Bank of S. Louis. The 9 h of Augus 2007 was he key dae ha marked he beginning of he Global Financial Crisis when he banking sysem was seized by he informaion ha he derivaives hey held were acually worh a lo less han hey previously imagined, and consequenly he banks sopped doing business wih each oher. The 9 h of May 2010 was he key dae ha marked he beginning of he Eurozone Deb Crisis when he global economy was hreaened by he possible solvency of he governmens in he Eurozone. Wih he division, he sample sizes for hese hree sub-periods are 495, 525, and 884 respecively. Table 1 repors he preliminary saisics of each period for he 12 sampled sock markes. The sample mean of he pre-crisis period for each of he sock markes is posiive and relaively large, while he sample means of he wo crises periods are generally smaller or even non-profiable. The sandard deviaion is generally highes in he GFC period and lowes in he pre-crisis period. Only Russia s sock marke has sandard deviaions higher han hose of China s sock marke, and ineresingly, he sandard deviaions of boh markes in he precrisis period are higher han hose in he EDC period. The res of he basic saisics also indicae ha he impacs of boh crises on he global sock markes are subsanial, and he invesmen in China s sock marke is generally risker han in oher sock markes. Table 2 repors he correlae of China s sock marke wih he global financial markes. The upper of he able displays he correlaion beween China s sock marke and one of he oher 11 sock markes. Overall, he correlaions of sock markes are highly significan, alhough in he pre-crisis period he correlaions beween China and he Wesern markes are no significan. The middle of he able displays he correlaion beween China s sock marke 15

16 and he relevan foreign exchange marke. The correlaion generally has become highly significan afer he ouburs of he global financial crisis. The ineresing case is Hong Kong where none of he correlaions in hree periods is significan. The lower of he able displays he correlaion beween China s sock marke and he relevan ineres rae differenial in money markes. There is some evidence ha China s sock marke has become more correlaed o money markes afer he global financial crisis. 4. Empirical Evidence 4.1 The world-facor model The esimaion of he world-facor model in equaion (1) is repored in Table 3. The nonsignificance of indicaes ha here is no expeced excess reurn agains he reurn of he US sock marke for each of he sock markes. 6 The significance and he magniude of sugges ha China s sock marke is less inegraed o USA s sock marke, while he Japan s and all he European sock markes are more inegraed wih he USA s han hose of oher Asian sock markes. The magniude of he loading of exra world-facor risk, i.e., suggess ha he unique risk of Russia s sock marke is higher han ha of China s sock marke. y, Afer acquiring he exra world-facor risk y for each of he sock marke from he residual series of he world-facor model, we proceed o esimae he VAR process in equaion (15). 4.2 SVAR esimaes 6 The world-facor model is esimaed wih he full sample size. The basic assumpion is ha he inerdependence beween China and world s financial markes is consan and herefore, any change of parameer esimaes hroughou hree sub-periods in he subsequen esimaion of marix A should ascribe o he conagion phenomena of financial markes. 16

17 To choose he appropriae lag lengh p for he VAR process, we use he log-likelihood raio es wih he maximum leg lengh being 50, hereby roughly caching one season s dynamic in he sysem. For he VAR process o saisfy he sabiliy condiion, we make sure ha all of he characerisic roos lies wihin he uni circle. We also es he auocorrelaion LM es for up o 50 lags, and make sure ha he null of no serial correlaion is adhered o. 7 As explained in he previous secion, he marix A in equaion (9) is over-idenified wih one degree of freedom, and hence we can es he possibiliy ha eiher 13 or 14 is non-zero. In oher words, he over-idenified SVAR sysem allows us o es wheher China s sock marke is conemporarily affeced by he relevan foreign exchange or money marke. For esing he possibiliy, we apply he likelihood raio es wih is saisic, LR1 2 ln c R1 A, B ln L A B L, (18) c u where ln L c A, B u and ln c A, BR 1 L are he concenraed log-likelihood funcions for he jusidenified SVAR and he over-idenified SVAR wih one degree of freedom, wih heir 2 esimaes compued by he equaion (17). I can be shown ha LR1 has an asympoic - disribuion wih one degree of freedom, and is criical value of 5% significance level is 2 roughly equal o , i.e., We herefore rejec he null of he resriced model and hen proceed o esimae he unresriced model if 2 1. LR1 X 0.05 For he sock markes in he European zone, because oday s price in China canno possibly affec yeserday s price in Europe, we impose he resricion As a resul, we have a differen A marix for he European counries, ha is, 7 The resul is available upon reques. 17

18 A (19) The marix in equaion (19) is obviously over-idenified wih 2 degrees of freedom. The specificaion herefore allows us o es he possibiliy ha eiher 13 or 14 is non-zero or boh of hem are non-zero. We sar he es by applying he following likelihood raio es wih is saisic, A, B ln L A B LR2 2 ln Lc R2 c, R1 (20) ln L A, B is he concenraed log-likelihood funcion for he over-idenified SVAR where c R 2 wih wo degree of freedoms. The variable LR2 has he 2 -disribuion wih one degree of freedom, so we rejec he null of wo resricions SVAR and hen proceed o esimae he one resricion SVAR if 2 LR2 X Following he same logic, we hen compare he one resricion SVAR wih he jus-idenified SVAR by applying equaion (18). If he procedure furher akes us o selec he jus-idenified SVAR, hen his would sugges ha China s sock marke responds insananeously o he changes of exchange rae and ineres rae differenial. Table 4 repors he parameer esimaes of he marix A in he pre-crisis period for each of he 10 counries paired wih China. The non-significance of for mos of he esimaions 12 indicaes ha in general China s sock marke was no conemporarily affeced by oher sock markes in he pre-crisis period. The general non-significance of also suggess ha 21 China s sock marke did no conemporarily affec oher sock markes. Noneheless, he non-zero of 13 and 14 as well as he high significance of 23 and 24 for several Asian counries sugges ha he Asian sock markes were inerdependen wih each oher hrough 18

19 heir insananeous connecions in he foreign exchange and money markes. The loadings of idiosyncraic shocks sugges ha he invesmen in sock marke is much riskier han he invesmens in foreign exchange and money markes. The resul also suggess ha China s sock marke has higher unique risk han oher sock marke does, excep for Russia s. Table 5 repors he parameer esimaes of he marix A in he GFC period. The esimaes of 12 and 21 sugges ha China s sock marke has become conemporarily affeced by oher sock markes, and in a lesser degree, China s sock marke also has had more significan influences on oher Asian sock markes. The non-zero of 13 and 14 as well as he high significance of 23 and 24 for he European counries sugges ha boh China s and he European s sock markes have become more sensiive o he risk in he relevan foreign exchange and money markes in he GFC period. In addiion, he high significance of he esimaes of 23 for he Asian counries suggess ha he Asian sock markes in general were sensiive o he change of foreign exchange raes. The loadings of idiosyncraic shocks obviously show ha invesmens in any of he financial marke, including he sock exchange, foreign exchange and money markes, have become much riskier in he GFC period. China s sock marke in his period again has a higher unique risk when compared o oher sock markes, excep for Russia s. Table 6 repors he parameer esimaes of he marix A in he EDC period. The esimae of 12 suggess ha China s sock marke has become even more sensiive o oher Asian sock markes, while he role of he European sock markes on China s is minor. However, China s sock marke is indirecly conneced o he European sock markes hrough he conemporary connecions of sock marke wih foreign exchange and money markes, as indicaed by he relevance of 13, 14, 23 and 24. The general significance of 23 again suggess ha he 19

20 Asian sock markes were sensiive o he change of foreign exchange raes. The esimae of suggess ha he unique risk of sock markes as well as ha of foreign exchange markes have become smaller, while he unique risk of money markes has become larger in he EDC period, when compared o GFC period. Alogeher, he hree ables sugges ha afer he GFC period China s sock marke generally has become more inegraed wih he Asian sock markes. In he pre-crisis period, only Korea s sock marke has bi-direcional reurn spillover effec wih China s, and only Ausralia s sock marke has uni-direcional reurn spillover effec o China s. The ineresing case is Hong Kong where none of he parameers is significan, indicaing he isolaion beween China s and Hong s sock markes in his period. Our resul is consisen wih ha of Huyghebaer and Wang (2010) who find no evidence of sock marke inegraion beween China and oher Asian markes afer he Asian financial crisis. 8 In he GFC period, however, he bi-direcional reurn spillover effec is found for he cases of Korea and Ausralia, while he uni-direcional reurn spillover effec is found from Japan s and Singapore s sock markes o China s. In he EDC period, he bi-direcional reurn spillover effec is found o exis in hree cases, namely Taiwan, Singapore and Ausralia, whereas he uni-direcional reurn spillover effec is found o spread from he sock markes of Japan, Korea and Hong Kong o China s sock marke. The hree ables sugges ha China s sock marke and he European sock markes have been generally non-sensiive o each oher hrough ime, bu hey also sugges ha foreign exchange and money markes have been he imporan channels in ransmiing risk beween China s and he European sock markes afer he GFC period. 4.3 Impulse response funcion 8 Their empirical evidence is for he sudy beween 1 s of July, 1992 and 30s of June,

21 As can be seen in he previous secion, he unexpeced par of he SVAR sysem, i.e., B in equaion (13), will reach is impacs on he observed vecor y via he insananeous relaion Au B hrough he VAR sysem (14) o he regressand. The srucure of he marix A herefore can be hough of as he risk ransmission channel hrough which he magniude of idiosyncraic shocks B cass is immediae impacs on he curren prices of financial markes, and hen wih he dynamics of he VAR sysem is subsequen impacs on he fuure prices of financial markes. The dynamic sysem of ransmiing risk can be expressed more clearly if we rewrie he VAR sysem (15) ino a Wold moving average represenaion, y u u u (21) where, s s s j j1 A j (22) wih 0 I K Accordingly, he impulse response funcion ha can be calculaed based on he Wold MV decomposiion for a sable VAR(p) process has he immediae impulse response equal o u 1 A B. The produc of marices A 1 B herefore can be hough of as he measuremen of wheher each elemen of he idiosyncraic shocks has significan immediae impacs on he financial markes. In oher words, he immediae impulse response funcions allow us o differeniae he sources of risk o sock markes. Table 7 repors he immediae impulse response funcions of China s sock marke o four idiosyncraic shocks, namely he shocks of China s sock marke, counry 2 s sock marke, 21

22 he relevan foreign exchange marke and ineres rae differenial. Since 1 is he idiosyncraic shock of China s sock marke, i is no surprising ha he parameer esimaes of immediae impulse response of China s sock marke o is own shock, i.e., he numbers in he wo columns beneah 1, are all highly significan. The esimaes of immediae impulse response of China s sock marke o counry 2 s sock marke shock, i.e., he numbers in he wo columns beneah 2, sugges ha in general China s sock marke responded immediaely o he Asian sock marke shocks, especially in he wo crises period. The esimaes of immediae impulse response of China s sock marke o he European sock marke shocks, however, are no significan excep for Russia s and UK s in he GFC period. Wih respec o he foreign exchange shocks, he esimaes of immediae impulse response of China s sock marke are generally significan, especially in he wo crises periods. As for he money marke, here is also some evidence ha China s sock marke responded immediaely o he shock of ineres rae differenial. Table 8 repors he immediae impulse response funcions of global sock markes o he four idiosyncraic shocks in he hree sub-periods. Since 2 is he idiosyncraic shock of counry 2 s sock marke iself, i is no surprising ha he parameer esimaes of immediae impulse response of counry 2 s sock marke o is own shock, i.e., he numbers in he wo columns beneah 2, are all highly significan. The esimaes of immediae impulse response of counry 2 s sock marke o China s sock marke shock, i.e., he numbers in he wo columns beneah 1, sugges ha in general he Asian sock markes were no insananeously responsive o China s sock marke shock, alhough Ausralia s sock marke was a special case ha responded immediaely o China s in all hree sub-periods. Noice ha he esimaes of immediae impulse response of he European sock markes o China s sock marke shock are zero. The reason is we impose in he marix A in equaion (19) he elemen

23 since here exiss he problem of non-synchronous rading hours beween China s and he European sock markes. The esimaes of immediae impulse response of he Asian sock markes o he foreign exchange shocks are overall highly significan, while he immediae impulse response of he European sock markes o he foreign exchange shocks is generally weak. Finally, alhough he immediae impulse response of sock markes o he money marke shocks is no srong, he European sock markes seemingly are more responsive o he money marke shocks han he foreign exchange shocks. Alogeher, he wo ables sugges ha he volailiy of financial asses provides an increasing imporan role in connecing China s sock marke wih oher Asian sock markes. Before he GFC period, here were wo cases of bi-direcional volailiy spillover effec wih China, namely he sock markes of Korea and Ausralia. Since he GFC period, however, here has been more cases of bi-direcional volailiy spillover effec wih China, including Korea, Ausralia, and Singapore in he GFC period, and Ausralia, Singapore and Taiwan in he EDC period. While here was no uni-direcional volailiy spillover effec from China s sock marke o ohers in any sub-period, here were some cases of uni-direcional volailiy spillover effec from ohers o China s sock marke. These include he case of Japan in he GFC period, and he cases of Japan, Hong Kong, and Korea in he EDC period. There were also he uni-direcional volailiy spillover effec from Russia and UK s sock markes o China s in he GFC period. The empirical evidence overall suggess ha he linkage of China s sock marke wih he Asian sock markes has become sronger hrough ime, bu he volailiy spillover effec is no symmerical; China s sock marke has become more influenced by he volailiy of oher sock markes, while is influence on ohers has no increased as equally. 23

24 As menioned previously, he srucure of he marix A can be hough of as he risk ransmission channel, and correspondingly he immediae impulse response funcions can be used o idenify he sources of risk o sock markes. From his perspecive, Table 7 and Table 8 ogeher sugges ha he mos imporan source of risk o sock marke is is own idiosyncraic shock. The resuls also sugges ha he idiosyncraic shocks of he Asian sock markes are imporan sources of risk o China s sock marke. In conrary, he idiosyncraic shock of China s sock marke seems no o be as imporan as a source of risk o he Asian sock markes. Ineresingly, he resul also srongly suggess ha he idiosyncraic shock of foreign exchange markes is a common source of risk o boh China s and he Asian sock markes. Wih regard o he European financial markes, here is no evidence ha he idiosyncraic shock of he European sock markes is a source of risk o China s sock marke. There is some evidence, however, ha he idiosyncraic shock of he relevan foreign exchange markes is a source of risk o China s sock marke. Finally, here is also some weak evidence ha he idiosyncraic shock of he relevan money markes is a source of risk o he European sock markes. 4.3 Forecas error variance decomposiion Afer he idenificaion of he sources of risk o sock markes, a complemenary quesion one can ask is how relaively imporan each of hese sources of risk is o he sock markes. To answer his quesion, we employ he forecas error variance decomposiion since i separaes he variaion in an endogenous variable ino he componen shocks o he VAR sysem. In our sudy, we display he relaive imporance of each idiosyncraic shock o sock marke for roughly up o one-monh ime, i.e., 20 periods, since he relaive conribuion of each idiosyncraic shock o sock marke has become quie sable hereafer. 24

25 Table 9 displays he variance decomposiion of China s sock marke in he pre-crisis period. As expeced, he mos imporan source of risk o China s sock marke is is own idiosyncraic shock, i.e., 1. Moreover, by looking a he columns beneah 2, he variance decomposiion recognises ha he idiosyncraic shocks of Ausralia s, Korea s and Japan s sock markes are imporan sources of risk o China s sock marke, while he idiosyncraic shocks of he European sock markes and oher Asian sock markes are much less imporan. Table 10 displays he variance decomposiion of China s sock marke in he GFC period. The resul suggess ha he idiosyncraic shocks of Ausralia s, Singapore s, Korea s and Hong Kong s sock markes are imporan sources of risk o China s sock marke, apar from he own idiosyncraic shock of China s sock marke. Table 11 displays he variance decomposiion of China s sock marke in he EDC period. The major sources of risk o China s sock marke are sill is own idiosyncraic shock and he idiosyncraic shocks of Asian sock markes, excep for ha of Japan s sock marke. The hree ables alogeher sugges ha China s sock marke has become more influenced by he idiosyncraic shocks of Asian sock markes hrough ime. The inspecion of individual counry furher suggess ha he sock markes of Hong Kong, Singapore and Taiwan have become more imporan in conribuing o he variaion of China s sock marke hrough ime, while he sock markes of Japan and Korea have become less imporan. Our resul herefore is consisen o he resuls of Haemi-J and Roca (2004) and Zhou, Zhang and Zhang (2012) ha he Chinese culural background provides a risk facor in China s sock marke. Wih regard o he European sock markes, he resul suggess ha he idiosyncraic shocks of he European sock markes generally play very minor roles in deermining he variaion of China s sock marke. The resul is consisen wih ha of impulse response funcions in Table 7. Furhermore, he resuls from he columns beneah fx and id in he hree ables 25

26 sugges ha he idiosyncraic shocks of foreign exchange and money markes, wheher from Asia or from Eurpore, are no imporan sources of risk o China s sock marke. To invesigae he influences of China s sock marke shock on he global sock markes, we now urn o Tables 12 o 14. Table 12 shows he variance decomposiion of he global sock markes in he pre-crisis period. As expeced, he mos imporan source of risk o any counry s sock marke is is own idiosyncraic shock, represened by 2. The resul also suggess ha he idiosyncraic shock of China s sock marke is generally an imporan source of risk o oher Asian sock markes, while i plays a very minor role in conribuing o he variaion of European sock markes. Table 13 shows he variance decomposiion of he global sock markes in he GFC period. The own idiosyncraic shock is unsurprisingly he mos imporan source of risk o a counry s sock marke. The resul also suggess ha he idiosyncraic shock of China s sock marke is an imporan source of risk o Ausralia s, Singapore s and Hong Kong s sock markes. Table 14 shows he variance decomposiion of he global sock markes in he EDC period. The role of China s sock marke as he risk source o he Asian sock markes has generally diminished, alhough i is sill imporan for Ausralia s and Hong Kong s sock markes. The hree ables also sugges ha China s sock marke is no an imporan source of risk o he European sock markes. Alogeher we herefore find no empirical evidence ha China s sock marke has become more influenial o he global sock markes hrough ime. Furher inspecion of he influence of China on oher Asian counries suggess ha China s sock marke is no an imporan risk source o Taiwan s and Singapore s sock markes, despie in he GFC period Singapore s sock marke showed vulnerabiliy o he idiosyncraic shock of China s. By conras, Hong Kong s sock marke has been grealy influenced by he variaion of China s sock marke hroughou he samples. Combined wih he resuls in 26

27 Tables 9 o 11, our sudy suggess ha here are differen degrees of inegraion wih he Chinese culure cluser. Ineresingly, Ausralia s sock marke also shows high inegraion wih China s sock marke, which may be due o China being he larges rade parner of Ausralia. Finally, he role of China s sock marke in deermining he variaion of Japan s and Korea s sock markes is diminishing. The overall resuls from Table 9 o 14 herefore sugges ha only Ausralia s and Hong Kong s sock markes show evidence of inegraion wih China s sock marke. In paricular, he resuls from he columns beneah fx in Tables 12 o 14 sugges ha mos of he Asian sock markes are grealy influenced by he idiosyncraic shocks of foreign exchange markes, excep for Hong Kong. The reason for he minor influence of China-Hong Kong exchange rae on Hong Kong s sock marke migh be because China and Hong Kong boh heavily peg heir currencies o he U.S. dollar. Furher inspecion suggess ha he idiosyncraic shock of foreign exchange marke in some cases is even more imporan han he shock of China s sock marke. For example, Taiwan s sock marke has been more influenced by he shock of China-Taiwan exchange rae han by he shock of China s sock marke hroughou he sample, and Japan s and Korea s sock markes are also more vulnerable o he shocks in currency markes han he shock of China s sock marke in boh crises period. By conras, for he European sock markes he risk emanaing from currency marke is much less imporan; only Russia s sock marke in he GFC period is criically influenced by he shock of he relevan foreign exchange marke. In addiion, despie he impulse response funcion suggess ha some European sock markes, for example, Germany, France and he U.K. in he GFC period, and Russia in he EDC period, are sensiive o idiosyncraic shocks in money markes, he variance decomposiion suggess ha he money markes are no imporan sources of risk. 27

28 In essence, he resul of forecas error variance decomposiion suggess ha he degree of China s sock marke being influenced by Asian sock markes is greaer han he degree of influence i has on he Asian sock markes. I also suggess ha he foreign exchange and money markes are no risk sources o China s sock marke, while he foreign exchange markes are he imporan sources of risk o oher Asian sock markes. In addiion, he resul suggess ha he European sock markes are quie disconneced from China s sock marke even in he crisis period. Alhough here is some evidence from he impulse response funcions ha foreign exchange and money markes are he sources of risk o he China s and European s sock markes in crises period, he magniudes of hese risks arrived in he sock markes are acually minor. In shor, he resul of variance decomposiion no only is consisen wih ha of he impulse response funcions, bu i also provides some furher evidence for inspecing he exen o which he idiosyncraic shock of financial markes ransmis o each oher. 5. Conclusion Finding ou he causaliy of a naional equiy marke wih inernaional financial markes is an empirical quesion of grea imporance o boh policy makers and invesors. Wih he increasing liberalisaion of China s capial markes, many researchers have devoed o he sudy of he spillover effec beween China s sock marke and inernaional sock markes. In his paper, we specified a srucural form ha deermines he insananeous relaionship among equiy, currency and money markes, which was hen analysed by a SVAR model. In oher words, we used he SVAR model o es he reurn spillover as well as volailiy spillover effecs beween sock markes, wih he preview ha currency and money markes may funcion as he linkages beween sock markes. 28

29 The empirical evidence from he es of reurn spillover suggesed ha China s sock marke has become more inegraed wih he Asian sock markes since he global financial crisis. The evidence also showed ha here was no conemporary reurn spillover effec beween China s and he European sock markes, alhough he currency and money markes may have funcioned as he linkages beween hem since he global financial crisis. The empirical evidence from he es of volaliy spillover also indicaed ha he inegraion of China s sock marke wih he Asian sock markes has become sronger, wih furher suggesion ha he effec is no symmerical; China s sock marke has become more influenced by he volailiy of he Asian sock markes hrough ime, bu China s influence on Asia has no increased as equally. In addiion, he volailiy of China s sock marke is generally unrelaed o he volailiy of he European sock markes, alhough in he global financial crisis here was he uni-direcional volailiy spillover from Russia and UK o China. Furhermore, he SVAR model allowed us o idenify he idiosyncraic shocks specified in he srucural form sysem. Wih his perspecive, he immediae impulse response funcions can be viewed as a es on he saisical significance of he risk sources of financial markes and he forecas error variance decomposiions can be viewed as an examinaion on he relaive imporance of he risk sources, while he srucural relaionship among financial markes can be hough of as he risk ransmission channel. The mos imporan risk source of China s sock marke, as suggesed by he immediae impulse response funcions, is he own idiosyncraic shock, followed by he shocks in he Asian sock markes. By conras, he idiosyncraic shock of China s sock marke is no an imporan source of risk o boh he Asian and European sock markes. Ineresingly, he impulse response funcions also suggesed ha currency markes were a common source of risk o boh China s and he Asian sock markes. The forecas error variance decomposiions furher suggesed ha China s sock marke has become more influenced by he sock markes wih he Chinese culure 29