Eric Girardin. Dijun Tan

Transcription

1 Informaion onen of Order Flow and ross-marke Porfolio Rebalancing: Evidence for he hinese ock, reasury and orporae Bond Markes his version: February 2009 Eric Girardin GREQAM UMR NR 6579 EHE, Universiés d'aix-marseille II e III enre de la Vieille harié 2 rue de la harié, Marseille cedex 02 Dijun an GREQAM and chool of Economics and Managemen, Universiy of Elecronic cience and echnology of hina hengdu, hina, andj516@uesc.edu.cn Woon K. Wong Invesmen Managemen Research Uni ardiff Business chool ardiff, F10 3EU, Unied Kingdom. el: ; Fax: wongwk3@cardiff.ac.uk Absrac: his paper esablishes wihin-marke and cross-marke informaion conen of order flow for socks, corporae bonds and reasury bonds in hina. Wih ick-by-ick daa over hree years on he hanghai ecuriy Exchange marke, we show significan informaion conen of order flow and is relaionship wih ime-varying marke condiions. Bi-direcionally negaive cross-marke effecs of order flow on reurns boh beween socks and bonds and beween corporae and reasury bonds are observed afer he commonaliy and lead-lag relaionship of he hree markes are conrolled. Our resuls provide evidence ha cross-marke porfolio rebalancing or fligh produces cross-marke effec of order flow. In paricular, while reasury bonds are he main choice as safe asses in sock-bond fligh, corporae bonds are able o subsiue reasury bonds when eiher he sock or he bond marke is more volaile. Due o he small size of he bond markes in hina, sock-bond porfolio rebalancing will produce high price impac on he bond markes, which may incenives invesors o search oher subsiue safe asses, such as he high-qualiy corporae bonds for reasury bonds, or inversely. Keywords: Order flow, ross-marke hedging, Marke condiions, hinese capial marke

2 1 Inroducion ross-marke linkages involving boh reurns and order flow have become an acive area of sudy. Previous sudies have shown ha cross-marke porfolio rebalancing or fligh, paricularly fligh-o-qualiy or fligh-o-liquidiy under sress marke condiions, could be he main reason driving such linkages. However, hese sudies, which examine relaed issues joinly or in isolaion, usually resric hemselves o he ineracions beween only wo markes, ypically sock and reasury bond markes, and only focus on major OED counries where microsrucure daa is more easily available. his paper suggess enlarging he scope of his enquiry in hese wo direcions. We ake ino accoun he corporae bond marke in addiion o he reasury bond marke o examine cross-marke porfolio rebalancing or fligh boh beween sock and bond markes and beween corporae and reasury bond markes 1. We sudy hese hree markes in he larges emerging economy by far, hina. Our work hus firs provides an imporan addiional feaure compared o relaed empirical sudies which are only concerned wih fligh-o-qualiy beween socks and reasury bonds ha have a very differen fundamenal risk level (Underwood, 2008), fligh-o-liquidiy beween reasury and Refcorp bonds ha have a similar fundamenal risk level (Longsaff, 2004), or fligh-o-liquidiy accompanied by fligh-o-qualiy (Goyenko and Ukhov, 2008; and Brunnermeier and Pedersen, 2007). By naure, cross-marke asse allocaion or risk managemen ranges over all available asses, and he safe asses (bonds) in he porfolio could include no only reasury bonds bu also possible (high-grade) corporae bonds; herefore, leaving ou corporae bonds when examining cross-marke porfolio rebalancing probably resuls in biased resuls. hree feaures of he hinese financial markes boh disinguish hem from OED markes and lead us o expec specificiies in cross-marke porfolio rebalancing or fligh. Firsly, 1 In his paper, we view fligh-o-qualiy and fligh-o-liquidiy, which occurs paricularly under exreme marke condiions, as wo ypes of porfolio rebalancing. he general porfolio rebalancing could also include invesmen adjusmen due o reasons oher han liquidiy or qualiy problem of he asses hemselves, such as he liquidiy rading for reasons ouside he sock marke shown by Admai and Pfleiderer (1998). onnolly e al. (2005, 2007) and Underwood (2008) also refer o cross-marke rebalancing as cross-marke hedging, as reurn correlaions beween he sock and reasury bond markes likely o be negaive in crisis periods on he sock marke. 1

3 individual invesors predominae on he hinese sock marke, as opposed o he dominance of insiuional invesors in OED markes. Previous heoreical sudies abou fligh-o-qualiy or fligh-o-liquidiy behavior usually assume ha marke paricipans are fund managers or insiuions who usually allocae invesmen beween socks and bonds (Vayanos, 2004, Brunnermeier and Pedersen, 2007), and previous empirical sudies only focus on OED markes ha have more insiuional han individual invesors (Longsaff, 2004, Goyenko and Ukhov, 2008, Beber e al., 2008, Underwood, 2008). In hina, individuals, who represen more han 90% of invesors, held 51.29% of he oal marke value ousanding by he end of heir abiliy o underake cross-marke asse allocaion and hedging may be much weaker han ha of insiuional invesors, and mos individuals underake lile cross-marke asse allocaion beween socks and bonds, simply ransferring funds beween he sock marke and bank accouns. Accordingly, he very exisence of sock-bond marke fligh aciviies is quesionable. econdly, hinese markes are characerized by a lack of inernaional arbirage due o he sill-binding capial conrols. herefore, cross-marke effecs can only occur (officially) beween domesic markes. Previous sudies on OED markes suffer from missing inernaional cross-marke effecs, ignoring socks-o-foreign-socks and bonds-o-foreign-bonds effecs. However, he empirical resuls of onnolly e al. (2007) ec. show ha cross-counry hedging should no be ignored. Our sudy on he hinese marke hus focuses on pure domesic cross-marke effecs. hirdly, he hinese bond marke is a special case. he reasury bond marke is small and has no fixed ime and no fixed erm of bond issuing. he very sric condiions for he issuance of corporae bonds endow hem wih high qualiy, making hem close o reasury bonds. his makes he hinese bond markes ideal for invesigaing he preference of insiuional invesors beween reasury and corporae bonds in he sock-bond fligh and he sylized facs of fligh beween bond markes, which have no been graned much imporance. We are paricularly ineresed in he subsiue effec beween reasury and corporae bonds in sock-bond marke porfolio rebalancing. his sheds ligh on he possible role of each marke and he exen o which invesors bear credi risk in heir fligh. he daabase we sudy for he hinese financial markes covers ick-by-ick daa over a 2 Relaed saisics here come from he Repor of he hinese apial Marke ha is published by he hina ecuriies Regulaory ommission (hp:// 2

4 hree-year period saring in January 2004 for all socks, reasury bonds and corporae bonds. his enables us o examine cross-marke effecs no only among reurns bu also beween and hrough order flow. Order flow here is defined as he excess of buying pressure, ha is, he rading volume ha buyers iniiae over ha iniiaed by sellers. Explanaions for such a role sem from he informaion revealed by aggregae order flow on preferences, endowmens and he projecion of news by marke paricipans (Lyons, 2001, Evans and Lyons, 2002, Underwood, 2008), and is suppored by he empirical evidence of Hasbrouck and eppi (2001), hordia e al (2002), Evans and Lyons (2002), Brand and Kavajecz (2004), and Harford and Kaul (2005) ec., who show ha order flow presen imporan informaion on variaion of asse reurns on he sock, bond and foreign exchange markes. here are wo major public securiy exchange markes in hina, he hanghai ecuriy Exchange (HE) and he henzhen ecuriy Exchange (ZE) marke. his paper examines he informaion conen of order flow for socks, reasury bonds and corporae bonds on HE, which has more socks and bonds han in ZE. By employing ick-by-ick daa over a hree-year sample for all socks, reasury bonds and corporae bonds on HE, we deermine boh he relaive role of each marke in cross-marke fligh and is precise feaures. Firsly, we find ha he reasury bond marke serves more for sock invesor hedging han he corporae bond marke, and, in addiion o he sock-bond fligh, fligh beween corporae and reasury bond markes is also significan. Underwood (2008) provides evidence of a (unidirecional) negaive correlaion beween reasury bond marke order flow and sock marke reurns, and aribues i o cross-marke hedging. We find bi-direcionally negaively cross-marke effec of order flow on reurns boh beween socks and reasury bonds and beween corporae and reasury bonds. Namely, a rise in order flow on one marke means a fall in reurns on anoher marke. As aggregae order flow reveals informaion on preferences, endowmens and he projecion of news by marke paricipans, he bi-direcionally negaive effec of order flow on reurns beween wo markes imply widespread cross-marke porfolio rebalancing or fligh aciviies. Alhough boh he hinese reasury and corporae bond markes are relaive small and individual invesor are predominan on he hinese sock marke, our resuls also provide evidence of cross-marke porfolio rebalancing boh beween socks and reasury bonds and beween corporae and reasury bonds. 3

5 econdly, when he sock marke is more volaile and during exreme rising or falling, or when he reasury bond marke is more volaile or falling, corporae bond order flow also presen significanly negaive effec on sock reurns. Furhermore, under exreme sock marke condiions, order flow in boh he corporae and reasury bond markes has much sronger effec on sock marke reurns. his means ha, wih respec o porfolio rebalancing or fligh-o-qualiy beween he sock and bond markes in hina, corporae bonds are able o subsiue reasury bonds as he heaven of sock invesors, eiher when he demand of safe asses rises during sock marke sress or when he cos of porfolio rebalancing during volaile reasury bond marke. ross-marke porfolio rebalancing or fligh may have high price impac on he objec marke, even hough such aciviies are good news for he objec marke due o more invesmen coming. his is paricularly rue in hina since boh he reasury and corporae bond markes are relaively small compared wih bond markes in oher developed counries. Under his circumsance, he high price impac on he reasury bonds during fligh beween he sock and reasury bond markes should be an imporan reason ha incenives invesors o search oher subsiue safe asses, such as he high-qualiy corporae bonds for reasury bonds, or inversely. Finally, wih respec o he wihin-marke effec of order flow, in line wih previous sudies in relaionship beween liquidiy and price discovery, e.g. hordia and waminahan (2000) and Underwood (2008), we find ha he informaion conen of order flow differs significanly boh beween index and non-index socks and beween liquid and illiquid bonds. However, on he hinese sock marke, he order flow of non-index socks has sronger effecs on marke reurns han index socks. Alhough his conrass wih he resuls of Harford and Kaul (2005) and Underwood (2008) ha index sock order flow explains marke movemens more han non-index sock order flow, our resuls suppor he findings of Bailey e al. (2008) ha he order flow of individual invesors explains more flucuaion in sock reurns han insiuional invesors in hina, and are in line wih he fac ha he individual invesor who are predominan on he hinese sock marke prefer small non-index socks. Furhermore, we also find evidence of wihin-marke porfolio rebalancing beween index and non-index socks when he sock marke is more volaile or under exreme falling, which reduce he informaion of index sock order flow on sock marke movemens. On he bond markes, he order flow of acively raded bonds wih high liquidiy has 4

6 higher informaion conen on marke movemens. In paricular, off-he-run reasury bonds in hina are more liquid and have higher wihin-marke informaion conen of order flow han on-he-run reasury bonds. he res of his paper is organized as follows. ecion 2 offers a review of he relaed lieraure. ecion 3 provides a briefly inroducion o he hinese capial markes and secion 4 describes he daa. In secion 5 we presen he empirical resuls and analyze he cross-marke effecs of order flow, he influence of marke condiions on order flow informaion conen. ecion 6 concludes. 2 Lieraure Review Our work is closely relaed o he previous sudies on he informaion conen of order flow, including boh wihin-marke and cross-marke effecs of order flow. For he wihin-marke effecs, here is evidence from sock, bond and foreign exchange markes. On he sock marke, Hasbrouck and eppi (2001) find commonaliy of order flow beween socks. By means of principal componen and canonical correlaion analyses, he common informaion of order flow for he 30 Dow-Jones index socks is exraced. he empirical resuls of Hasbrouck and eppi show ha he common movemens of marke order flow can explain more han wo-hirds of he common movemens of marke reurns, bu reurns of individual socks are mosly affeced by heir own order flow. imilarly, Harford and Kaul (2005) suppose ha he commonaliy of marke order flow comes from he operaion of index funds. hey find ha he commonaliy of order flow for &P500 socks is sronger han ha for non-&p500 socks. In addiion, hordia e al. (2002) find a significan correlaion beween order flow and conemporaneous or lagged reurns, and reurns may reverse in he fuure when curren order flow is oo high (posiive) or oo low (negaive) 3. Wih a unique daase of raders accouns, Bailey e al. (2008) have examined commonaliy in order flow across individual, insiuional, and proprieary invesors on he hanghai sock marke. heir resuls show ha he marginal explanaory power of order flow for individual invesors on sock reurns is higher han ha for 3 In hordia e al. (2002), he order flow is called order imbalance. Acually, boh expressions have he same definiion in his paper, as well as in oher lieraure. 5

7 insiuional and proprieary invesors, and commonaliy of order flow for individual invesors is also sronger. Order flow on he bond and foreign exchange markes is also informaive. Brand and Kavajecz (2004) apply principal componen and vecor auoregressive (VAR) models o he U. reasury bond marke and find ha order flow affecs bond yields due o liquidiy premium bu no invenory premium. onrolling for major macroeconomic news or policies, order flow can explain abou 20% of variaion in reasury bond yields, and he informaion conen of order flow is higher when marke liquidiy is lower. In he foreign exchange marke, Evans and Lyons (2002) find ha order flow explains more ha 60% of he variaion in exchange raes, which is much higher han he radiional macro-economic variables. Wih respec o cross-marke hedging or fligh aciviies, our work is closely relaed o Underwood (2008), who sudies he effec of order flow on cross-marke reurns beween U. sock and reasury bond markes. His work shows ha aggregae order flow plays a srong role in explaining he links beween cross-marke reurns. he order flow of he reasury bond marke significanly negaively affecs sock marke reurns. Moreover, he effec of reasury bond marke order flow is sronger when sock marke volailiy rises, and he effec of order flow for insiuional invesors or acively raded 5-year reasury noes is also sronger han individual invesors or reasury bonds wih oher mauriies. Underwood aribues such a cross-marke effec of order flow o cross-marke hedging. hordia e al. (2005) also find ha no only order flow affecs is own-marke reurns, volailiy and liquidiy in boh sock and bond markes, bu order flow in one marke may also affec reurns, volailiy and liquidiy in anoher marke. Moreover, our work is also relaed o he lieraure which provides evidence abou he effec of marke uncerainy on marke correlaions and draws implicaions for cross-marke hedging. onnolly e al. (2005, 2007) find ha sock and bond marke reurns are posiively correlaed in he long run, bu when uncerainy on he sock marke rises, he araciveness of he sock marke is reduced and more invesmen will be ransferred o he safer bond marke, which leads o shor-erm negaive sock-bond correlaions. By using hicago Board Opions Exchange s Volailiy Index (VIX) and he urnover of he sock marke as he proxies of marke uncerainy, onnolly e al. (2005) find ha he higher sock marke uncerainy, he higher probabiliy of 6

8 fuure negaive sock-reasury bond correlaions. onnolly e al. (2007) examine sock and bond marke reurn correlaions for he hree counries, U., UK and Germany, and find ha no only a rise, bu a sharp change (rise or fall), in sock marke uncerainy may also lead o negaive sock-bond correlaions. onnolly e al. (2005, 2007) show ha cross-marke hedging should be responsible for he shor-erm negaive correlaions beween he sock and he reasury bond markes. his paper examines porfolio rebalancing or fligh on he hinese sock, reasury and corporae bond markes by examining boh wihin-marke and cross-marke informaion conen of order flow. hese previous sudies shown above ensure ha such an approach is effecive as order flow provide direc on informaion on invesors rading aciviies and explain asse reurns very well. Deparing from hordia e al. (2005) and Underwood (2008), who also examine cross-marke effecs of order flow, his paper akes ino accoun a hird marke, he corporae bond marke, in order o exend invesigaions on invesors porfolio rebalancing o include all possible asses. In line wih onnolly e al. (2005), we also examine he informaion conen of order flow on imes of differen marke condiions o furher es wheher porfolio rebalancing or fligh is he main source of cross-marke effecs of order flow. 3 Marke background General iuaions and Marke egmenaion A well-running capial marke is usually viewed as one of he mos imporan requiremens for a srong economy. hina has he larges and fases growing economy in he world. In U dollar erm, he size of is economy sands a $3.4 rillion in 2007, ranked afer U, Japan and Germany. In parallel wih he fas-growing economy, is capial marke also develops very fas. Righ row, ocks, reasury bonds and corporae bonds are raded on he wo public securiy exchange markes, he hanghai ecuriy Exchange (HE) and he henzhen ecuriy Exchange (ZE). However, as shown by Nefci e al. (2007), Fan and Zhang (2007), Qiao e al. (2008) and Wong e al. (2008), boh he hinese sock and bond markes are marked as being segmened. Firsly, he hinese sock marke includes boh an A-share marke ha is only available for 7

9 domesic invesors (recenly for some Qualified Foreign Insiuional Invesors, QFII) and a B-share marke ha is available only for foreign invesors. he B-share marke has been opening o domesic invesors only afer June By he end of 2005, 1240 companies had issued A-share, and 23 companies had issued B-share, and 86 companies were issuing boh A and B shares 4. he price of socks on he A-share marke is denominaed in Renminbi (hinese Yuan, NY), bu ha on he B-share marke is denominaed in U dollars (for HE marke) or Hongkong dollars (for ZE marke). econdly, because one of he main objecives o seup he sock marke in hina a he beginning of 1990 s is o find a new source of financing for large sae owned enerprises (OEs), which were heavily dependen on bank loans. Before 2004, for he sake of socialis economic and poliical sysem, a large proporion of shares, abou wo-hirds, are held by hese OEs hemselves and can no be raded publicly 5. As a resul, he A-share marke was segmened as radable and non-radable shares. A new reform adoped by he hina ecuriy Regulaory ommission (R) required all hese non-publicly exchanged shares o be placed on he marke afer 2005, and mos firms had complied wih he reform by he end of hirdly, in addiion o he exchange bond marke, in he middle of 1997, an iner-bank bond marke was esablished, and originally only commercial banks, laer oher financial insiuions, are allowed o rade in his marke. Righ now, hree segmened markes, he iner-bank bond marke, he exchange marke (boh on HE or ZE) and he over-he-couner (O) marke (for individuals o buy bonds on he couner of commercial banks), consiues he secondary bond marke. Wih respec o he four ypes of bonds on he hinese domesic bond marke, i.e. reasury bonds, cenral bank noes, financial bonds and corporae bonds 6, all cenral bank noes and financial bonds are only raded on he iner-bank bond marke. reasury bonds and corporae bonds are raded on boh on he iner-bank and exchange markes, bu a larger proporion is only lised and raded on he iner-bank bond marke. he 2007 Annual Repor of he hinese Bond Marke show ha he rading volume on he iner-bank marke in 2007 is abou rillion 4 Relaed saisics here come from he Repor of he hinese apial Marke ha is published by he hina ecuriies Regulaory ommission (hp:// 5 By he end of 2005, he oal number of share of all hese lised companies had reached billion NY, bu he radable shares were only billion NY, accouning for 38.2% of he oal shares. 6 More informaion of he four ypes of bonds will be provided laer his paper. 8

10 (NY), bu ha on he exchange marke and O marke is respecively only 1.75 rillion and 3.5 billion. Moreover, wih respec o he rading eiher on he iner-bank or exchange marke, he volume of repo rading is more han spo rading. For insance, in 2007, he spo rading on he hanghai and henzhen exchange markes is abou billion NY, bu ha for he repo rading on he wo markes is abou billion NY. ecuriy Issuing Before he 1990 s, bank loans were almos he only source of financing for all companies in hina. his led o a large amoun of non-performing loans from he banks and increased bank risk. A he beginning of he 1990 s, wih he furher reform of he hinese economy, he sock marke was creaed o provide anoher source of company financing. By he end of 2007, he capializaion of he hinese sock marke had grown o rillion NY, he larges among he emerging markes, and he hird larges in he world. In 2007, new issue of public equiy reached abou billion NY, he highes in he world. By he end of 2007, a oal of 1,616 companies had been lised on HE and ZE, ranging over almos every secor in he hinese economy. he value ousanding of he four ypes of bonds in hinese domesic bond marke, reasury bonds, cenral bank noes, financial bonds and corporae bonds, is respecively 2,149 billion NY (27.4% of oal ousanding bonds), 2,931 billion (37.4%), 2,097 billion (26.8%) and 170 billion (2.2%) a he end of Ocober, 2006 (Huang and Zhu, 2007). reasury bonds are issued by he Minisry of Finance o he public and are mosly medium and long erm bonds, bu here is no fixed ime and specified erm for he issuance of reasury bonds. enral bank noes are issued by he People s Bank of hina for he implemenaion of moneary policies. Financial bonds are issued by hinese policy banks (ae Developmen bank, hina Impor and Expor Bank, and hina Agriculure Developmen Bank) o suppor he developmen projecs in special economic secors such as agriculure, impor and expor, and in special regions, especially rural and underdeveloped areas (Fan and Zhang, 2007). In paricular, he hinese corporae bond marke is composed of hree pars. he firs par includes all converible bonds issued by public firms, mos of which had evenually been 9

11 convered o equiy 7. he second par is corporae bonds ha are issued by public firms wihou embedded opions. However, he firs bond of his ype was issued on 24 h epember 2007, and here are only wo on he marke by he end of he hird par ranges over all he bonds issued by large sae-owned enerprises, which have no public equiy. In his paper, we only focus on he hird par, he corporae bonds wihou embedded opion and issued by non-public sae-owned firms. We don no consider he firs or he second ype of corporae bonds because of uncerainies opion properies for converible bonds or he limied daa for corporae bonds issued by public firms. All corporae bonds issued by non-public firms have high credi qualiy due o sae-ownership of hese firms and sric issuing requiremens. hese requiremens include a deb o asse raio of less han 0.4, an average disposable profi in he pas hree years of higher han he one-year ineres paymen, a raing of higher han A, a qualified guaranor ec. hese requiremens limi he size of he corporae bond marke bu grealy improve bond qualiy. econdary Marke Paricipans On he hinese sock marke, by he end of 2007, more han 90% invesors are individual ones, who have hold 51.29% of he oal sock shares. Wih a unique daase of rader accouns on HE, Bailey e al (2008) find ha for a ypical sock on a ypical rading day, 91.76% of all rades were iniiaed by individual invesors. When i comes o insiuional invesors, here are no real invesmen banks because no hinese financial insiuion is permied o provide he full range of financial services currenly provided by inernaional invesmen banks. here were 133 securiy companies in hina by he end of 2003 wih oal asses of only U$67.87 billion, merely 20.22% of he oal asses of U$335.6 billion of one single U invesmen bank, Goldman achs (Zhang, 2004). Excep commercial banks and rural credi cooperaives, all ypes of invesors (including insiuional and individual invesors) can paricipae in he exchange bond marke. hough he rading volume on he iner-bank bond marke, as menioned above, is more han ha on he exchange bond marke, saisics show ha rading is more frequenly on he exchange bond 7 ome very sric limiaions are regulaed by he hina ecuriy Regulaory ommission (R) for easoned Equiy Offering (EO), which is even sopped a long ime around 2004, so i s more easily o issue converible bonds han new shares for he public firms. However, because he original incenive is no for issuing equiy bu deb, so mos of he converible bonds are evenually ransferred o shares. 10

12 marke. For insance, in 2007, he number of rades on he exchange marke is nearly 798,600, and ha on he iner-bank marke is only 188,600. ha is, mos of he rading on he iner-bank marke is wholesale of big insiuions such as he large sae-owned commercial banks, insurance companies, bu he paricipans in he exchange bond marke are mainly he relaive small insiuions or individuals, who are also he main paricipans in he sock marke 8. Definiely, even individual invesors on he exchange bond markes can no be negleced in erms of numbers, bu in fac nonbank financial insiuions play he dominan role (Nefci e al., 2007). Insurance companies, invesmen funds and oher banks are he chief holders on he exchange bond marke. he commercial banks, which are he chief paricipans on he iner-bank marke, are no allowed o buy and sell socks, and, a he same ime, individuals or small insiuional invesors canno paricipae in he iner-bank bond marke. his paper discusses he informaion conen of order flow for A-share socks, corporae bonds and reasury bonds on HE. Only securiies on HE are considered because boh he capializaion and number of securiies, paricularly bonds, ousanding on HE is more han ha on ZE, and B-share socks are excluded here o eliminae he effec of foreign exchange rae movemens and because hey accouns for only 2% of marke shares. By he end of 2007, on HE, here are 850 A-share and 54 B-share firms are lised wih oal capializaion of rillion NY, 61 reasury bonds wih oal issued value of billion NY, and 63 corporae bonds wih oal issued value of billion NY. In examining cross-marke fligh aciviies, we do no direcly consider he rading on he iner-bank bond marke bu only include iner-bank repo raes in our model, because 1) high-frequency rading daa on he iner-bank marke is no available; 2) he small insiuional and individual invesors who are he main paricipans on he sock marke can no paricipae in he iner-bank marke 9. his naurally excludes he cenral bank noes and financial bonds ha are 8 aisics of he 2007 Annual Repor of hinese Bond Marke (hp://zsfx.chinabond.com.cn/chinabond/aricle/ fj_conen. jsp?slmid=113&sfjid=18847&sype=2) show ha he main paricipans in he iner-bank bond marke are sae-own commercial banks, insurance companies, muual funds. Acually, hese hree ypes of insiuions respecively hold 67.7%, 7.6% and 4.2% of he oal value of all he bonds in he iner-bank marke. he share for non-bank financial insiuions and non-financial insiuions is respecively 0.6% and 14.3%, and ha for individuals (hrough he O marke) is less han 1%. hus, for hese small insiuional invesors, exchange bond marke should be an imporan choice even hough he exchange bond marke is smaller han he iner-bank bond marke. 9 o include he possible effec of iner-bank marke rading on he exchange marke rading, we also es he effec of iner-bank repo rae on he spo sock and bond marke reurns laer his paper. 11

13 only raded on he iner-bank marke. Definiely, we do no discuss he O marke because is rading volume is oo small. In our sample, he corporae bond marke is smaller han he reasury bond marke for lower value ousanding and lower urnover. However, here are more corporae bonds ousanding in he exchange marke, and saisics show ha while repo rading is much more han he spo rading on he reasury bond marke, spo rading dominaes he corporae bond marke because only a few of corporae bonds are relaed o repo rading (Nefci e al., 2007). his can, o some exen, reduce he low liquidiy problem of he corporae bond marke compared wih he reasury bond marke. 4 Daa his paper uses inraday ick-by-ick daa and daily daa for all A-share socks, reasury bonds and corporae bonds from 2004 o 2006 on HE, wih a oal of 724 rading days. he daa are provided by MAR orporaion, one of he chief daabase providers for he hinese financial markes. he daa are processed as follows. Wih he ick-by-ick high frequency daa, order flow for each sock, reasury bond and corporae bond are aggregaed every day. MAR daabase provides records ha direcly idenify each rade as buyer- or seller-iniiaed afer Wih hese records, he oal buyer-iniiaed rading volume in value over he oal seller-iniiaed rading volume in value over one day is aggregaed as daily order flow for each sock and bond 10. he oal number of observaions of daily aggregaed order flow for all bonds and socks is , including for socks, and respecively for reasury and corporae bonds. Daily reurns of socks are compued as he firs difference of he logarihm of daily close price. o dampen exreme price movemens and provide a cool-off period in he evens of overreacion in order o proec he public invesors, HE currenly ses he daily price limi a 10%. In addiion o daa errors, hese daa wih daily absolue reurns higher han 10% are chiefly conneced o he socks which were goen rid of he special rea () or paricular ransfer (P) 10 ome lieraures, such as Brand and Kavajecz (2004), define he difference beween buyer-iniiaed and seller-iniiaed rading volume in dollar as order imbalance, which, obviously, is he same as he order flow defined here. 12

14 classes 11. hus, daa of ceiling or floor hi as well as wih absolue reurn higher han 10% are excluded here. Daily reurns of socks are compued as he firs difference of he logarihm of daily close invoice price, and daa on he ineres paymen days are excluded. Harford and Kaul (2005) and Underwood (2008) show ha index and non-index socks have differen informaion conen of order flow, or pricing discovery efficiency. Here we also spli all socks on he marke ino index and non-index sock porfolio. hanghai ecuriy 180 (180) index was creaed in Jul.2002 and composed of he ypical 180 A-share socks on HE. Afer ha, he index is also adjused wih he change of he price and shares ousanding for all index socks. We consruc index and non-index porfolios wih he sample and is change informaion abou he 180 index. Bonds are usually classified as on-he-run and off-he-run porfolio due o differen liquidiy. Here, we use his disincion since some sudies pricing discovery efficiency beween hese wo ypes of bond may also be differen. For insance, Goyenko e al (2008) show ha shor-erm off-he-run bonds reflec macro shocks firs. he basic saisics for differen sock and bond porfolios is shown in able 1. In able 1, he daily average number of rades and rading volume in value for differen sock and bond porfolios are provided. In addiion o hese wo indirec proxies, we also provide saisics of anoher wo direc proxies of liquidiy. One is he relaive spread compued from ick-by-ick rading daa, and he oher is he price impac coefficien, or illiquidiy, of Amihud (2002) compued from daily daa. he relaive spread is defined as, Ask1 Bid1 Pspread = ( Ask Bid1 ) / 2 1 Here, Ask1 and Bid1 are respecively he bes ask and bid prices for each rade. he relaive spread is compued for each rade and he daily relaive spread is averaged over all rades every day for each sock and bond. In line wih Amihud (2002), his paper define he price impac coefficiens, or illiquidiy, of he securiy i on day as follows, i R i ILIQ = M i 11 More informaion abou he or P can be achieved in he web hp:// 13

15 i R i Here, is he reurn and M is he rading volume in value for securiy i on day 12. Liquidiy for he marke or a paricular porfolio is equally weighed by individual i Pspread or i ILIQ. he liquidiy provided in able 1 is he average daily liquidiy over he oal sample. As shown in able 1, he number of rades and rading volume in value for 180-index socks are much higher han non-180-index socks. Moreover, spread ( Pspread ) and illiquidiy ( ILIQ ) are lower for 180-index socks. herefore, 180-index socks have higher liquidiy. imilarly, on he corporae bond marke, he number of rades and rading volume in value is higher, and Pspread and ILIQ are lower for on-he-run bonds han off-he-run bonds. herefore, on-he-run bond liquidiy is higher han off-he-run bond liquidiy on he hinese corporae bond marke. However, on he reasury bond marke, we can observe ha, while he number of rades for on-he-run bonds is higher han off-he-run bonds, rading volume in value is lower, and Pspread and ILIQ are higher han off-he-run bonds. his is conras o he common facs ha on-he-run bonds would be more liquid han off-he-run bonds. For insance, arig and Warga (1989), Houweling e al (2005) ec. find ha off-he-run bonds are more likely o be included in he buy-and-hold porfolios of invesors and may even have higher liquidiy premium. wo reasons could be responsible for such a differen resul on he hinese reasury bond marke. Firs, he small size of bond issuing in hina is unable o saisfy invesor demand, paricularly for insiuions. his easily leads o over-compeiion and biased price when a new bond is jus issued (chulz, 2001), and hus giving rise o high price impac and low liquidiy for on-he-run bonds even hey are raded more frequenly (having higher number of rades as shown in able 1). econdly, in line wih Goyenko e al (2008), off-he-run reasury bonds may response macroeconomic shocks faser han on-he-run bonds. herefore, off-he-run bonds may have higher informaion conen of bond valuaion and, hus, have higher liquidiy han on-he-run bonds. 12 he logarihm of he rading volume in value is applied o reduce he daa dimension problem. 14

16 able 1 Basic Informaion of daily rading for he sock and bond porfolios sock Porfolios Mean of daily rades Mean of daily rading volume in value (Million Yuan) Mean of daily illiquidiy (Price impac) Mean of daily relaive pread ock marke (1) 180 index sock (2) Non-180 index sock Difference (1)-(2) orporae bond marke (3) On-he-run corporae bond (4) Off-he-run corporae bond Difference (3)-(4) reasury bond marke (5) On-he-run reasury bond (6) Off-he-run reasury bond Difference (5)-(6) (0.01) (0.17) Here, saisics abou he rading aciviy and liquidiy for differen porfolios are provided, including he number of rades, rading volume in value, illiquidiy (price impac of rading; Amihud, 2002) and relaive spread. Difference is he difference es on mean and is p-value for specified variable of wo specified porfolios in a marke. Average rades, rading volume in value and liquidiy for each porfolio are compued firsly, and hen he mean of each variable for each porfolio and is difference saisics can be compued over he whole sample (724 days). 15

17 When comparing wih rading aciviies beween he sock and bond markes, resuls in able 1 show ha, while he number of rades on he sock marke is much higher han ha on he bond marke, he average rading value of each rade is higher for bonds (boh for reasury and corporae bonds) han socks. his is reasonable since more individual invesors are paricipaed in he sock marke and have smaller rades, and more insiuional invesors are paricipaed on he bond marke and have larger rades. his is also he possible reason lead o much lower price impac or illiquidiy on he bond marke han on he sock marke. 5 Empirical Resuls Descripive saisics of order flow and reurns for he hree markes are shown in able 2, in which r and OF are he daily equally weighed reurns and aggregaed order flow for all A-share socks on HE. r and r are respecively he daily equally weighed reurns on he reasury and corporae bond marke, and OF and OF are respecively he aggregaed order flow on he reasury and corporae bond marke. As shown in able 2, he sock marke has he highes reurn and highes volailiy (sandard deviaion of reurns), and he reasury bond marke has he lowes reurns and lowes volailiy. However, he harp-raio of mean reurns divided by sandard deviaion is highes on he corporae bond marke, and lowes on he sock marke, which indicaes ha reurns for bearing one uni of risk is much higher on he corporae bond marke han on he reasury bond marke in spie of he close credi qualiy beween he wo bond markes. Mean and median of order flow for he sock marke are negaive, and ha for he wo bond markes are posiive. Kurosis of order flow on he corporae bond marke is highes among he hree markes, indicaing ha order flow in his marke is more easily o have exreme values. orrelaions of order flow and reurns among he hree markes are provided in able 3. As shown by Forbes and Rigobon (2002), reurn correlaions beween asses or markes are grealy affeced by heeroskedasiciy. hus, o es he significance of correlaions, we compue p-value for all correlaions by he approach of boosrapping his is paricularly imporan, as resuls in able 3 show ha of he correlaion beween sock and 16

18 able 2 Descripive saisics of marke reurns and order flow Reurns Order flow (100 Million) r r r OF OF OF Mean 0.019% 0.021% 0.011% Median 0.120% 0.033% 0.019% Maximum 7.528% 0.674% 0.905% Minimum % % % d. Dev % 0.155% 0.145% kewness Kurosis harp-raio Here, r and OF are he daily average reurns and aggregaed order flow on he hanghai sock marke; r ( r ) and OF ( OF ) are he daily average reurns and aggregaed order flow, in 100 million, for all he reasury (corporae) bonds on HE; harp-raio are mean reurns divided by heir sandard deviaion. able 3 orrelaion of order flow and reurns among he sock, corporae and reasury bond markes r r r ln OF ln OF ln OF r * r *** r ln OF *** * ** ln OF *** ln OF *** *** Here, r, r and r are respecively daily average reurns on he sock, corporae and reasury bond markes; ln OF, ln OF and ln OF are respecively daily aggregaed order flow on he sock, corporae and reasury bond markes; he p-value of correlaions are compued by boosraps, *, ** and *** respecively represen significan a 10%, 5% and 1% level. reasury bond marke order flow is significan bu of he correlaion beween sock and corporae bond marke order flow is insignifican. hus, difference in heeroskedasiciy on each marke could have high effec on significance of correlaion ess. he deail of boosrapping could be ge upon reques. 17

19 In able 3, regarding reurn or order flow correlaions among he hree markes (lef-upper and righ-lower pars of able 3), resuls in able 3 show ha reurns beween he sock and corporae bond markes and order flow beween he sock and reasury bond markes are significanly posiive. he correlaion of reurns beween he wo bond markes is relaive high, abou 0.5, bu he correlaion of heir order flow is marginal. he correlaions beween reurn and order flow in each marke (lef-lower par of able 3), i.e. wihin-marke correlaions beween order flow and reurns, are high on all hree markes. In paricular, i is highly posiive for he correlaion beween order flow and reurns on he sock marke, abou 0.79, and ha on he reasury bond marke is also high, abou On he corporae bond marke, however, he correlaion beween reurns and order flow is lower, only 0.14, even less han he correlaion beween corporae bond reurns and reasury bonds order flow (abou 0.20). Furhermore, sock marke order flow is also significanly posiively correlaed wih reasury bond marke reurns. hese cross-marke correlaions beween order flow on one marke and reurns or order flow on anoher markes indicaes ha, on he whole, buying or selling of socks could imply buying or selling of boh socks and reasury bonds, and buying or selling of reasury bonds could imply buying or selling of reasury and corporae bonds as well as socks. In line wih onnolly e al (2005) and Underwood (2008), his indicaes ha here are long erm posiive correlaions among he hree markes since hey are affeced by many common economic facors. 5.1 Reurn correlaions and marke uncerainy hough he sock, corporae and reasury bond markes play is own role and have differen fundamenal facors of asse valuaion, more and more sudies find ha reurn correlaions among he hree markes are varying wih marke condiions or uncerainies. Generally, a leas wo ypes of facors may affec correlaions among he hree markes. he firs ype includes common facors such as ineres rae, inflaion, macroeconomic condiions and polices, ec., and he second ype is cross marke porfolio rebalancing, including fligh-o-qualiy, which ransfers invesmen o marke wih lower risk, and he fligh-o-liquidiy, which ransfers invesmen o marke wih higher liquidiy. Wih hese facors, in he shor erm, he weak posiive sock-bond correlaions may become 18

20 negaive 14, as shown by onnolly e al. (2005, 2007), and he highly posiive reasury-corporae bond correlaions may become sronger or weaker. For insance, sock-bond marke reurn correlaions will fall when he sock marke is under sress and more invesmen is ransferred from he sock o he bond marke, or when boh markes are affeced by macroeconomic common facors ha have inversely effecs on he wo markes. By conras, he correlaions will rise when boh markes are srongly affeced by common facors ha have similar effecs on hem. Reurn correlaions among he hree markes on differen marke condiions are shown in able 4. Here, we simply consider ime-varying correlaions condiional on he rise or fall of he marke. hus, for reurn correlaions beween marke A and B, we spli he whole sample ino 5 subsamples respecively by reurns on each marke, and hen correlaions are compued and heir significance are esed by boosrapping in each subsample. For example, wih respec o reurn correlaions beween he sock and corporae bond markes, ( r r ) cor,, we firsly spli he whole sample respecively by he 20%, 40%, 60% and 80% quanile of sock marke reurns quanile of corporae bond marke reurns and r, and hen cor ( r, r ) and is significance level (p-value) by boosrapping in hese subsamples are compued. here are abou 145 observaions in each subsample. r 14 Wihou specified explain, he sock-bond correlaions in his paper include he correlaions beween he sock and reasury bond marke, he correlaions beween he sock and corporae bond marke, and he bond markes may denoe boh he corporae and reasury bond marke. 19

21 able 4 orrelaions of reurns among he sock, corporae and reasury bond markes or( r, r ) or( r, r ) or( r, r ) ondiional variables r r r r r r All * * *** *** 0-20h pcl ** ** *** *** 20-40h pcl ** ** *** 40-60h pcl *** h pcl * h pcl *** ** *** *** In able 4, r, r and r are respecively he weighed average reurns on he sock, corporae and reasury bond markes; For correlaions beween marke A and B reurns, we spli he whole sample ino 5 subsamples by reurns on each marke, hen correlaions and heir p-value in hese subsamples are compued; For example, wih respec o reurn correlaions beween he sock and corporae bond markes ( r r ) sample respecively by he 20%, 40%, 60% and 80% quanile of sock marke reurns marke reurns cor,, we firsly spli he whole and corporae bond r, and hen we compue cor ( r, r ) and is significance level (p-value) by boosrapping in hese subsamples. here are abou 145 observaions in each subsample; Pcl is he quanile of reurns on each marke; *, ** and *** respecively represen significan a 10%, 5% and 1% level. r Resuls in able 4 show ha boh sock-bond and corporae-reasury bond correlaions are varied on differen marke condiions. In paricular, we can observe ha sock-bond correlaions are significanly posiive when reurns on eiher bond marke is low, bu hey are significanly negaive when reurns on he sock marke are high or reurns on he (corporae) bond marke are high. In line wih onnelly e al. (2005, 2007), our resuls may imply ha porfolio rebalancing beween socks and bonds happens when sock marke reurns are high, and no only fligh-o-qualiy or fligh-o-liquidiy bu also fligh-from-qualiy or fligh-from-liquidiy may reduce marke correlaions, as sock-bond marke correlaions follow an decreasing paer wih bond reurns. Due o he small size of he bond markes, such aciviies may have high impac on bond prices. ha is, bond price may have a large rise during fligh from sock o bond marke since he small bond marke should face a large hedging demand and coming invesmen. Furhermore, he differen rend beween sock-corporae bond correlaions and sock-reasury 20

22 bond correlaions may indicae ha he wo bond markes may play a differen role in sock-bond marke hedging. Wih respec o corporae-reasury bond reurn correlaions, i is ineresing ha, while hey are highly posiive in he whole sample, hey largely varied in differen subsamples. Generally, correlaions beween he wo bond markes are high when eiher marke has a fall, bu become more weakly, bu sill significanly, correlaed when eiher bond marke has a rise. However, he correlaions are insignifican, or negaive, when boh markes have neiher a sharp rise nor a sharp fall. his means ime-varying correlaions beween bond markes and he behind reasons also need o be examined more even hough mos of previous sudies only examine correlaions beween sock and bond markes. 5.2 he cross-marke informaion conen of order flow As above secions show evidence of ime-varying correlaions beween markes condiional on marke rising and falling ha may be resuled from cross-marke porfolio rebalancing or fligh, here we examine such an issue by discussing he informaion conen of order flow, because aggregaed marke order flow is able o reveal informaion on risk preferences, beliefs and endowmens of invesors ha is relevan o pricing securiies (Underwood, 2008), or reflec he differen projecion of public or macroeconomic informaion by invesors even when here is no privae informaion in he marke (Lyons, 2001). Previous empirical sudies have showed ha order flow grealy explains variaion in asse prices, such as he findings of Hasbrouck and eppi (2001), Harford and Kaul (2005), hordia e al. (2002) from he sock marke, he findings of Brand and Kavajecz (2004) from he reasury bond marke, and he findings of Evans and Lyons (2002) from he foreign exchange marke. In addiion o wihin-marke effecs, order flow may also have cross-marke effecs due o invesors cross-marke invesing and hedging aciviies. he order flow on he sock marke will rise when more is invesed in socks. Inversely, sock order flow may fall and bond order flow may rise when sock marke risk rises and more invesmen is rebalanced o he bond marke. For insance, because of cross-marke hedging, hordia e al. (2005) show ha order flow in one marke may be relaed o reurn, volailiy or liquidiy on is own marke and on anoher marke, 21

23 and Underwood (2008) finds significan cross-marke effec of reasury bond marke order flow on sock marke reurns. hus, in addiion o he wihin-marke effec of order flow on asse reurns, he cross-marke effec of order flow by naure provides direc evidence of asse allocaion and porfolios rebalancing. Here, o discuss cross-marke porfolio rebalancing or fligh boh beween socks and bonds and beween corporae and reasury bonds, we examine he conemporaneous effec of order flow for one marke on reurns for anoher marke by an UR (ysem of eemingly Unrelaed Regression) model as follows, r = a 0 + a OF 1 index + a pread 7 + a OF 2 + a non index ommon 8r + a OF + a 9 3 Repo, on + ε + a OF 4, off + a OF 5, off + a OF 6, on (1) r = b + b OF 0 1 index + b pread 7 + b OF 2 + b non index ommon 8r + b OF, on + b Repo + ε b OF 4, off + b OF 5, off + b OF 6, on (2) r = c + c OF 0 1 index + c pread 7 + c OF 2 + c non index ommon 8r + c OF, on + c Repo + ε c OF 4, off + c OF 5, off + c OF 6, on (3) [ ] = 0 E ε, =, K M [ ε ] E[ ε ] 0 [ ] = 0 E ε ; σ If = s E ε s =, K, M =,, 0 Oherwise r r r Here,, and are respecively he equally weighed reurns on he hanghai sock, corporae and reasury bond marke on day ; OF and OF are respecively he aggregaed order flow for he 180-index and he non-180-index sock porfolio on day ; index non Index OF, on and OF, Off are respecively he aggregaed order flow for he on-he-run and he on off-he-run corporae bond porfolio on day ; OF, and OF, off are respecively he aggregaed order flow for he on-he-run and he off-he-run reasury bond porfolio on day. pread, pread and pread are respecively liquidiy variables (Relaive pread) on he sock, corporae and reasury bond markes; ommon r is he reurn principal componen decomposed by he sample correlaion marix for he vecor of marke reurn series ( r, r, r ) ; Repo is he 7-day repo raes of he iner-bank reasury bond marke. Reurns or liquidiy is 22