Do fund investors destabilize the Chinese stock market?

Transcription

1 Do fund invesors desabilize he Chinese sock marke? Maozu LU a, Yun ZHU b Absrac In his paper, we sudy he relaion beween fund flow and marke volailiy a aggregae level using daily daa and provide empirical evidence on he impac of insiuional invesors on marke prices. On 9 h June 000, Muual Fund Index, which indicaes he aggregae price of muual funds lised in Shanghai Sock Exchange, was launched. Using he even sudy mehodology in GARCH framework we find ha robus empirical evidence ha he sock marke becomes significanly less volaile afer June 000. Furhermore, we find ha aggregae rading volume and urnover of muual fund are negaively correlaed wih subsequen marke volailiy, and are posiively correlaed wih subsequen marke performance. I implies ha muual fund rading conribues o he marke-wide volailiy, and heir rading sabilizes marke prices in Shanghai Sock Marke. Due o he fac ha fund inflow and ouflow informaion is no available for our daa se, we can examine wheher fund invesors follow he posiive-feedback rading sraegy. However, our resuls do no suppor he hypohesis ha insiuional rading desabilizes sock prices. Keywords: muual fund invesors, sock marke volailiy, GARCH model, Chinese sock marke JEL: G11; G14; G3 We hank Raymond O Brien and Jan Podivinsky for valuable commens. a Economics Division, School of Social Sciences, Universiy of Souhampon, UK; Guanghua School of Managemen, Peking Universiy, China; School of Economics, Shanghai Universiy of Finance and Economics, China; ml@soon.ac.uk b Economics Division, School of Social Sciences, Universiy of Souhampon, UK; yz@soon.ac.uk 1

2 1. Inroducion Due o he increase in he number of insiuional invesors rading on sock markes, financial researcher are ineresing in exploiing insiuions impac on sock prices. A main quesion concerning insiuion is wheher insiuions are a sabilizing or desabilizing force on equiy prices. Proponens of efficien marke heories sugges ha insiuion invesors arbirage irraionaliies in individual invesors responses o informaion and provide a sabilizing influences on sock prices. Compeing heories sugges ha insiuions herd ogeher and follow he posiive feedback sraegy or rade based on pas reurns ha may drive asse prices away from fundamenal values. These rading behavior of insiuions resul in desabilizing impac on sock prices. In general, insiuional invesors are more sophisicaed, manage a large fund and generae a large rading volume compared wih individual invesors. The microsrucure lieraure confirms ha subsanial rades can have a large impac on sock prices. Gabaix e al. (005) presen a model in which volailiy is caused by he rades of large insiuions. They expec ha a large fund can move he marke significanly. And hey show a simple case o suppor his idea ha large funds are indeed large compared o he liquidiy of he marke and ha price impac will be an imporan consideraion. In heir model, large invesors generae significan spikes in reurns and volume. They poin ou ha a small number of large insiuional invesors could cause exreme movemens in sock prices wihou news. The main argumen is ha insiuional invesors desabilize he sock prices and increase he volailiy of he marke because of he presence of posiive feedback rading and herding. A number of recen empirical sudies provide some evidences ha insiuional invesors (i.e. muual fund) have been engaged in posiive feedback rading- buying when he price increases and selling when i falls. If raional speculaors early buying rigger posiive feedback rading, hen an increase in he

3 number of forward-looking speculaors can increase volailiy abou fundamenal. Delong, Shleifer, summers and Waldmann (1990) summary a number of forms of behavior in financial markes which could be describes as posiive feedback rading, and menion ha he posiive feedback sraegies are common in financial markes. These same sudies sugges ha insiuional invesors exhibi herd behavior. However, he available evidence does no necessarily imply ha insiuions desabilize he marke. If he posiive feedback rading is a raional response o he marke signal, insiuions behavior will enhance he marke efficiency and reduce he possibiliy of sudden change in he marke. Of course even if insiuions affec prices, hey migh move hem oward raher han away from fundamenal. If prices move in one s favor afer rading, hen one has conribued o price sabilizaion. Similarly, if insiuions herd and response o he same fundamenal informaion, hey will speed up he adjusmen of sock prices o new informaion and enhance he marke efficiency. Moreover, hey will sabilize he sock prices insead of desabilizing hem. We can conclude from he shor discussion above ha empirical findings on insiuional invesors herding and posiive feedback rading behavior are no necessarily evidence in favor of he desabilizing effec on sock prices. As we know, he Chinese sock marke has grown rapidly during he las decade in erms of he number of companies lised and marke capializaion. However, as an emerging marke, i is more volaile and is dominaed by individual invesors compared wih maure markes. A he end of 1990s, governmen launched a se of plan and policy o encourage insiuional invesors i.e. muual fund, rade on socks. Hence, a number of muual funds were esablished afer ha. The major change in he invesors srucure on he Shanghai sock marke has is impac on sock prices. Finally fund index was launch on June 9, 000 in Shanghai Sock Exchange due o he significan growh of fund indusry. Before 000 he majoriy of raders in he marke were small, privae invesors and here were also some muual funds acive in he marke bu hey had relaively small amouns of capial under managemen. Afer ha 3

4 muual funds became imporan players on he sock marke and he role of fund invesors in financial markes is likely o be more imporan. So i becomes a prevailing argumen for boh paricipans and researchers in Chinese sock marke wha he effec of insiuional rading on he prices is and wheher heir rading behavior sabilizes or desabilizes he financial marke. Since, muual fund grows o be a more and more imporan componen of he asses hold by households, cash flows ino differen ypes of muual fund could be a very good indicaor of changes in invesors demand for financial securiies. Sudying he relaionship beween marke volailiy and muual fund cash flows migh be able o reveal how aggregae invesors reac o he more and more volaile asse markes. By sudying he relaion beween fund flows and he marke volailiy we will have a beer undersanding of muual fund invesors behavior, no only heir reacion o pas marke performances, bu also heir aiude o invesmen risk a he aggregae level. In lieraure he previous resuls are only indirec empirical evidence on he desabilizing effecs of insiuional invesors rading behavior on sock prices. The exising lieraure on he insiuional rading behavior is predominanly based on quarerly ownership daa o compue changes in insiuional holdings and in urn draws conclusions abou he behavior of insiuional invesors, or use low frequency daa, i.e. monhly and quarerly daa, o examine he relaion beween fund flow and sock reurns. This paper exends his analysis o Chinese marke and conribues o he lieraure by hese issues a he aggregae level by using daily daa. The daily frequency of he daa allows for more efficien esimaes of ime variaion in sysemic risk han does lower frequency daa. I addresses he quesion of wheher muual fund invesors desabilize he marke and increase he volailiy of he sock reurns. In lieraure researchers regress volailiy on individual fund flows using OLS o check his issue 1. While, we consruc a GARCH (1,1) model o find ou wheher 1 For insance, Goezmann and Massa (003), and Luo (003) 4

5 muual fund conribues o desabilize he volailiy of he marke reurn. We fill his gap in he lieraure by providing evidence abou muual fund behavior, using muual fund index. Muual fund index was launce on June 9, 000 in Shanghai Sock Exchange. Firsly, a GARCH (1,1) model wih dummy variable is se up o examine he srucural change boh in mean reurn and condiional volailiy afer his even. Hence, he GARCH-ype even sudy may provide direc empirical evidence of wheher insiuions cause significan change in he volailiy of sock reurns. Secondly, we check he relaion beween fund flow and marke index. The aggregae rading volume and urnover of lised fund are used as proxy for fund flow. Finally, we examine wheher here is he presence of herding formaion by fund invesors in Chinese marke. This paper is organized as follows. We give a brief summary of he relaed lieraure in secion. In secion 3, we describe he daa ses used for he sudy. We inroduce he mehodologies adoped o examine he effec of fund rading on sock price and also in secion 4. The empirical findings are shown in secion 5. And secion 6 discusses wheher here is he presence of herding formaion. Finally, secion 7 summaries our findings.. Lieraure Review Researchers have long sudied he equiy rading process and is impac on sock prices. Much prior empirical research isolaes individual rades and analyzes he behavior of he sock price around each rade. They consider an individual rade as he basic uni of analysis in he sudy of rading aciviy and is effecs on prices. However, insiuional invesors generae large block rading and an invesmen manager s order is ofen broken up ino several rades. Hence, researchers ry o answer he quesion wha he effec of insiuional rading on he sock prices. Demand and supply heory provide he heoreical suppor o he relaion beween fund cash flow and price reurn. 5

6 The change in demand for sock represened by fund cash flows drive he marke price o he new equilibrium. Changes in fund cash flows reflec he shif in invesors demand for asses and hey are a poenial facor in deermining he marke movemen. Chan and Lakonishok (1995), for insance, uses he record of rades execued by 37 large invesmen managemen firms o sudy he price impac and execuion cos of he enire sequence of rades, and also examine he behavior of sock prices immediaely before and afer rade packages. Their findings are consisen previous research, which has documened ha large block rades have a subsanial price impac. And Sias and Sarks (1997) sudy in he relaion beween serial correlaion in daily reurn and he insiuional rading. They conclude ha insiuional rading reflecs informaion and increases he speed of price adjusmen, and insiuional invesors correlaed rading paern conribue o serial correlaion in daily reurn. The general conclusion is ha insiuional rading causes boh permanen and emporary daily price effec. While, Edelen and Warner (001) focus on he aggregae level and sugges ha aggregae flow can be use o sudy he aggregae price effec of insiuional rading. They sudy he relaion beween marke reurns and aggregae flow ino U.S. equiy funds using daily flow daa. Their ess show ha he concurren posiive relaion reflecs flow and insiuional rading affecing reurns. This daily relaion is similar in magniude o he price impac repored for an individual insiuion s rades in a sock. Aggregae flow also follows marke reurns wih a one-day lag. The lagged response of flow suggess eiher a common response of boh reurns and flow o new informaion, or posiive feedback rading. However, all he research menioned haven sudy any impac of insiuions on he second momen of sock prices. For researchers he main quesion concerning insiuion is wheher insiuions are a sabilizing or desabilizing force on securiy prices. The effec of raional speculaors See Chan and Lakonishok (1993) and Holhausen, Lefwich and Mayers (1987). 6

7 rades is o move prices in he direcion of fundamenals. While risk aversion keeps raional speculaors from aking large posiions, noise raders can affec prices. Raional invesors dampen noise rading and preven from preven noise raders from driving he prices beyond he fundamenal. Hence, raional speculaors mus sabilize asse prices. However, his is no always he case if noise raders follow feedback sraegies ha hey buy socks when prices rise and sell hem when prices fall. Delong, Shleifer, summers and Waldmann (1990) argue ha raional speculaion can desabilize he asse prices when here exiss he posiive feedback rading. Raional speculaors, for insance, rade on good news ha hey obain his period. Informed raional raders buy more his period because hey recognize ha he iniial price increase will simulae buying by posiive feedback raders nex period. And i drives prices up his period han fundamenal news warrans. So he rading generaed by posiive raders drive prices beyond fundamenal in nex period and desabilizes prices alhough raional speculaors dampen hem and sabilize prices. They conclude ha rading by raional speculaors desabilizes prices because i riggers posiive feedback rading by oher invesors. Their conclusion is consisen wih Har and Kreps (1986) s work 3. However, he phenomenon of he marke desabilized by posiive feedback rading is no always he case. If such ype of rading drives sock prices oward fundamenal raher han away from fundamenal, i may no desabilize he marke. Gabaix e al. (005) presen a model in which volailiy is caused by he rades of large insiuions. And hey analyze how rading by individual large invesors may creae price movemens ha are hard o explain by fundamenal news. Insiuional invesors appear o be imporan for he low-frequency movemens of equiy prices. In heir heory, spikes in rading volume and reurns are creaed by a combinaion of news and he rades by large invesors. Suppose news or proprieary analysis induces 3 However, in Har and Kreps s model, hey assume ha compeiive raional speculaors are he only invesors able o perform physical sorage. Speculaors can change commodiy suppliers in a way ha makes equilibrium prices more volaile. Hence, in heir model, price-desabilizing speculaion resuls from he effec of sorage on quaniies. 7

8 a large invesor o rade a paricular sock. Since his desired rading volume is hen a significan proporion of daily urnover, he will moderae his acual rading volume o avoid paying oo much in price impac. The opimal volume will noneheless remain large enough o induce a significan price change. They find ha price movemens reflec boh he inensiy of he perceived mispricing and he size of fund. A large price movemen can come from an exreme signal or he rade of a large fund. Alernaively, i implies ha he disribuion of price movemens reflecs boh he news, as well as he size of agens ha ac on news. However, when here is a very large movemen, i is more likely o come from he acions of a very large insiuion, raher han an objecively imporan news. This poenial imporan of a large insiuion may explain why prices moved in he absence of significan news iems. Therefore, in erm of his heory he exreme reurns occurred because some large insiuions wished o make subsanial rades in a shor ime period. Also here are some empirical works ha have been done on his opic. Empirical lieraure finds mixed resuls when invesigaing he presence of insiuional herding and posiive rading. Mos of his lieraure relies on aggregae quarerly or annual insiuional holding daa o infer insiuional rading. Nofsinger and Sias (1999) documen a srong posiive correlaion beween changes in insiuional ownership and lagged reurns. They conclude ha insiuional engage in posiive-feedback rading since ha insiuions purchase subsequenly ouperform hose hey sell. Also Dennis and Srickland (00) find srong evidence of herding. They examine he relaion beween quarerly ownership levels and he cross secional volailiy of sock reurns and urnover, and noe ha socks ha move he mos during periods of marke wide volailiy are hose ha have relaively larger insiuional holdings. Furhermore, he socks ha move he mos during hese periods experience subsequen price reversals. They conclude ha muual funds and pension plan sponsors are herding ogeher and rying o jump ino rising markes or ou of declining markes, hereby driving prices beyond fundamenal values. 8

9 While Lakonishok, Shleifer and Vishny (199) use quarerly daa on he holdings of 769 ax-exemp funds o evaluae he poenial effec of heir rading on sock prices. They find weak evidence of herding and some evidence of posiive-feed backing rading for smaller socks, while jus very lile evidence of eiher herding or posiive-feedback rading in larger socks. Overall, hey conclude ha here is no srong evidence ha insiuional invesors desabilize prices of individual socks in heir sample. However, hey don rule ou he possibiliy of wha insiuional invesors desabilize eiher aggregae sock prices or he price s of individual socks using more frequen daa. Warher (1995) examine he effec of money flow ino muual fund on aggregae securiy reurns. They check wo-direcion relaion beween fund flows and aggregae reurns, and find evidence of a posiive relaion beween flows and subsequen reurns and evidence of a negaive relaion beween reurns and subsequen flows. While, here is no evidence ha aggregae fund flows are posiively relaed o pas reurns in weekly, monhly, quarerly or yearly daa. This resul indicaes ha invesors don follow he posiive-feedback sraegy in aggregae level. In fac, hey find ha muual fund invesors appear o be somewha conrarian in erm of he evidence of a negaive relaion beween reurns and subsequen flows. Recenly, Lipson and Pucke (005) invesigae he effec of he rading behavior of muual funds and pension plan sponsors on prices during he period when here exiss he large marke movemens in he marke. They find ha insiuions rade in he opposie direcion of large moves. Specifically, boh money managers and pension plan sponsors are ne sellers (ne buyers) on days when markes experience large price increases (decreases). In oher words, insiuions are increasing heir buying when marke declines and increasing heir selling when marke rises. These resuls sugges ha insiuional invesors acually supply liquidiy o he marke a hese imes and dampen he effec of rading on marke volailiy. And also Goezmann and Massa (003) examine bi-direcional causaliy beween cash flows and he volailiy. However, Luo (003) poin ou ha muual fund invesors creae excess sock volailiy because fund flows have a posiive impac on he subsequen sock marke 9

10 volailiy. 3. Daa Descripion In his paper, we use daily daa o invesigae he effec of muual fund rading on sock prices a he aggregae level. Our empirical analysis relies on Shanghai A-Share Price Index and Fund Index. Shanghai A-Share Price Index is value-weighed marke index and is composed of all A-Share socks lised in Shanghai Exchange. And i is he mos imporan index in Shanghai Sock Exchange. The sample coves he period from January, 1996 o December 31, 004 4, oaling 110 observaions. Fund index was launched on June 9, 000, which is value-weighed index of all he open-fund lised in Shanghai Sock Exchange. And he sample of fund index covers he period from June 9, 000 unil December 31, 004. The index reurn is defined as he logarihmic difference R = 100(ln P ln P 1), where P and 1 P are daily closing prices of index a and 1. And we ake he naural logarihm for rading volume and urnover series. To check if reurns, rading volume and urnover series are non-saionary, we use Dickey-Fuller and Augmen Dickey-Fuller ess o es uni roos. The ess rejec a 99% confidence he exisence of uni roos for he considered series. Warher (1995) and Luo (003) consruc a measure of ne sales, which is new sales plus exchange sales minus redempions and exchange redempions, o idenify he ne cash flow of muual fund money ino he marke. However, individual fund cash flow is no available. In his paper, fund index, which is composed of all open-fund raded in Shanghai Sock Exchange, is employed o show he rading behavior of fund invesors a he aggregae level. We have daily aggregae rading volume and urnover 4 All he dada used in his paper are downloaded from 10

11 series of lised funds. While, he daa se doesn disinguish beween share purchase and share redempion. I means ha cash inflow and ouflow also are no available in he daa. Hence, we can follow Warher (1995) and Luo (003) s idea o consruc such kind of variable o describe fund flow. Busse (1999) menions ha fund manager expec invesors inves less ino he funds or even wihdraw money from he funds when he marke volailiy is higher ha average. If fund invesors negaively response o marke volailiy and fund manager able o predic his paern, fund managers would decrease marke exposure and hold more cash. Hence, he large change in he aggregae rading volume and urnover of lised funds indicaes large money flow in he marke. Then, he aggregae rading volume and urnover of lised muual fund are used as proxy for he aggregae fund cash flow, respecively. Figure1 and figure display he auocorrelaion funcion of he daily aggregae rading volume and urnover of muual funds raded in Shanghai Sock Exchange. They show ha here is a subsanial auocorrelaion of he rading volume and urnover series. And he lagged auocorrelaion in hese wo series implies ha hey are highly predicable. Hence, he predicable par componen is separaed from he unpredicable componen. Table1 presens esimaes of regression models of rading volume and urnover of muual fund on heir own lagged variables. Panel A and panel B are he resuls for rading volume and urnover series, respecively. The firs four lagged variables are saisically significan and he fifh lag is insignifican. F-saisics are used o es he null hypohesis ha here is he absence of auocorrelaion of he residuals. The resuls show ha residuals are significanly auo-correlaed in regression model I, II and III. And he null hypohesis of he absence of auocorrelaion of he residuals is acceped in he las wo regressions. Hence, I use an AR (4) model o esimae expeced rading volume and urnover of muual fund and he residuals are he unexpeced componens. 4. Mehodology 11

12 4.1 GARCH-ype even model As an emerging marke, Chinese sock marke is more volaile compared wih maure markes. A he end of 1990s, governmen launched a se of plan and policy o encourage insiuional invesors i.e. muual fund, rade on socks. Hence, muual fund indusry gradually grew and fund index was launched on June 000. Moreover, during he period from January 1996 unil May 000, lised open-fund raised 3.03 billion RMB and he fund raised increased o 9.5 billion RMB from June 000 o he end of 004. I implies ha fund invesors acively rade on he marke and muual funds become imporan player afer June 000. The appearance of large fund raders and he resuling increase in insiuional ownership allows us o invesigae he impac on he volailiy of sock reurns. Therefore, we employ GARCH (1,1) model o describe he marke volailiy paern and a ime dummy variable is involved also o model he change in sock reurns and he volailiy of sock reurns afer June 9, 000 when fund index was launched. Our empirical sudy on he fund invesors influence on sock marke volailiy relies on he following GARCH (1,1) model wih dummies: R = ω + θ R + θ D + ε (1) 1 1 h = α + αε + βh + γd () D = 1 afer June 9, 000; 0 oherwise where R is he reurn of A-Share Price Index and he error erm ε denoes he unpredicable componen of index reurn and is -disribued. And he dummy variable D is se a 0 before June 9, 000 and 1 afer June 9, 000, when he fund index was launched. The firs equaion is he mean equaion, which is a funcion of firs own lagged variable, dummy variable and an error erm. The second equaion 1

13 shows marke volailiy, which is a funcion of he lagged squared residual from mean equaion, pas condiional variance and dummy variable. In equaion (1), he dummy variable is included o model he srucural change in index reurn afer fund invesors significanly rading on he marke. Saisically significan coefficien θ indicaes a srucural change in he paer of sock reurns afer June 9, 000. In equaion (), he dummy variable is used o deec he impac of fund rading on he volailiy of he marke. If fund raders have an influence on he volailiy srucure of index reurn, he coefficien γ should be saisically significan. The esimaed coefficien γ provides a measure of he shif in he condiional volailiy process. If he coefficien γ is saisically posiively significan, i means ha muual fund rading desabilize he marke price and increases he marke volailiy. On he oher hand, a saisically significan and negaive γ implies ha fund rading conribues o marke volailiy and sabilize he marke prices. 4. Fund flows and Marke volailiy In order o ge a clear picure for he effec of fund rading on sock prices, we ry o deec he relaion beween pas fund flows and he marke volailiy. To examine he impac of sock fund flows on he subsequen sock marke volailiy, for insance, Luo (003) regress monhly volailiy series on curren and lagged ne flows of each fund group. While, we consruc a GARCH (1,1) model o check his issue. In he model, fund flow as one exogenous variable is included o gauge he impac of fund rading on changes in marke volailiy. Moreover, as discussed in secion 3, daily rading and urnover of muual fund are used as proxy for he daily aggregae fund flow. Hence, rading volume or urnover will be included in he model as one exogenous variable. The model is shown as following: R = ω + θ1r 1+ θv 1+ ε or R = ω + θ1r 1+ θt 1+ ε (3) h = α + αε + β h + γv or h = α + αε + βh + γt (4)

14 where V 1 and T 1 denoe rading volume and urnover of muual fund a ime 1, respecively. Regression (3) is used o sudy he relaion beween pas fund flow and marke reurn. And Regression (4) looks more closely a he quesion of wheher fund raders desabilize he marke. A significan and posiive γ implies ha aggregae fund flow conribues o marke volailiy and desabilizes he marke prices. And if he coefficien γ is saisically negaive significan, hen we conclude ha fund flow reduces marke volailiy and sabilizes he marke. 5. Empirical resuls There are 1075 observaions from January, 1996 unil June 8, 000 and are 110 observaions from June 9, 000 hrough December 31, 004. Figure 3 illusraes he reurn of A-Share Price Index. I is obvious ha marke index reurns flucuae in a narrow band afer June 000 compared o he period before. I shows ha before June 000 he marke is more volaile han i afer ha. Table repors he empirical findings for A-Share Price Index relying on he June 9, 000 dummy. A ha day fund index was launched in Shanghai Sock Exchange, ha even delivers such kind of informaion o he public ha muual fund has been becoming an imporan player in he marke. Our concern is wheher his even of launching fund index affecs he sock prices and resul in srucural change in marke reurn and he volailiy of marke reurn. The dummy variable is saisically significan in boh mean and condiional volailiy equaion. The coefficien θ is negaively significan a 95% confidence level. Tha means ha afer June 000 he marke reurn is lower han i before June 000. Figure 4 is he ime series plos of 14

15 naural logarihm of closing price of A-Share Price Index during he period from January,1996 o December 31, 004. I is obvious ha he marke experience an increasing rend from January 1999 o July 001. Hence, he upward ime rend may conribues he high index reurn before June 000. And i may no conclude ha fund flow negaively correlaed marke reurn. When looking a he esimaed coefficiens describing condiional volailiy process of he GARCH (1,1) model wih dummy variable, all he coefficiens of GARCH model are saisically significan. More imporanly he parameer of he dummy variable in he condiional volailiy process ˆ γ is saisically significan a he 95% level and negaive. I indicaes ha he volailiy of he marke reurn is reduced afer June 006. We may conclude ha he even of launching fund index significanly affecs sock marke volailiy based on our findings, and he marke-wider volailiy is reduced when fund invesors play imporan role in he marke. Also we do he robusness check in erms of differen cu-off daes for he dummy variable D. 5 And he findings for he differen dummy variables confirm he empirical resuls we repor in able. These resuls suppor ha he behavior of muual fund rading do sabilize sock prices. To ge a more direc evidence of he relaion beween fund flow and sock prices, we reconsruc GARCH (1,1) model. Daily aggregae rading volume and daily aggregae urnover of open-fund lised in Shanghai Sock Exchange are used as proxy for he daily aggregae fund flow. Hence, one day lagged rading volume and urnover are included in GARCH (1,1) model as exogenous variable, respecively. The empirical resuls are shown in Table 3 and Table 4, respecively. From pane A of able 3, we find ha he coefficien θ is posiively significan a he 99% level in mean equaion. This provides he direc evidence of he posiive relaion beween he lagged rading volume of lised fund and marke index reurn. Also all he 5 The resuls for robus check aren shown in he paper and hey are available for reques. 15

16 coefficiens in condiional volailiy process of marke reurn are saisically significan. The coefficien γ is negaively saisically significan a a 90% confidence level. Tha means ha pas aggregae rading volume has negaive impac on he sock marke volailiy. Furhermore, in erms of he analysis above in secion 3, here exiss a subsanial auocorrelaion of he rading volume series. And he lagged auocorrelaion in he series implies ha i is highly predicable. Hence, panel B and panel C show he resuls of he GARCH (1,1) model wih exogenous variable when we replace he rading volume by he expeced rading volume and unexpeced componen. Panel B of able 3 shows ha pas expeced componen of rading volume sill is posiively correlaed wih marke reurn and i conribues o he marke volailiy and reduces he volailiy of marke reurn a a 90% confidence level. The resuls in panel C are similar wih he findings we have from panel A and B excep for one ha unexpeced pas rading volume is no significanly correlaed wih marke reurn. Table 4 presens he resuls of GARCH (1,1) model in which daily aggregae urnover is included as proxy for fund flow. And all findings are he same as he resuls we have in able 3. The coefficiens of lagged urnover of muual fund are negaively saisically significan and hey show ha urnover of lised muual funds has negaive impac on he sock marke volailiy. Hence, we find ha fund flow does affec marke volailiy in Shanghai sock marke and he marke volailiy negaively response o pas fund flow. However, hese findings are no consisen wih he previous research. Luo (003), for insance, sudies he relaion beween fund flows and sock marke volailiy o examine how invesors reac o marke volailiy. And he finds a srong posiive impac of fund flow on he subsequen sock marke volailiy. Goezmann and Massa (003) check he impac of fund inflows and ouflows on he sock marke volailiy, respecively. In summary, he sudy of fund flow in he period of June 9, 000 hrough December 16

17 31, 004 shows ha he aggregae rading volume and urnover of muual fund have a significan negaive impac on he subsequen marke volailiy. This finding doesn suppor he null hypohesis ha insiuional rading desabilizes he marke prices. Also we examine he relaion beween fund flow and marke index reurn wihou he GARCH (1,1) model. We find ha boh rading volume and urnover of muual fund lised in Shanghai Sock Exchange in our sample are posiively correlaed wih curren and pas index reurn 6. Due o he fac ha rading volume and urnover are aggregae daa and we don disinguish hese wo series generaed by buyer from hose generaed by seller. Hence, we may no conclude ha in Chinese marke aggregae fund invesors follow posiive fee-back sraegy or negaive feed-back sraegy, buying shares when he marke is down and selling shares when he marke is up, based on hese empirical resuls. A more appropriae way of examining he impac of fund flows on he marke volailiy is o decompose flow daa ino inflows and ouflows and o sudy heir relaion wih marke volailiy separaely. Moreover, Demirer and Kuan (004) examine he presence of herd formaion in Chinese markes using individual firm level as well as secor level daa and find ha herd formaion does no exis in Chinese markes. Their evidence suppors our findings. 6. Herding behavior The empirical resuls show ha pas fund flow is negaively correlaed wih he volailiy of he sock reurn and sabilize he marke prices in Shanghai Sock marke. In lieraure i is argued ha herding or posiive-feedback rading migh be he wo resources o desabilize he marke prices. However, he conclusion is no exclusive. Demirer and Kuan (004) conclude ha here is no evidence o suppor he presence of herd formaion in firm level, secor-level and marke level daa from he Shanghai and Shenzhen Sock Exchanges, while we focus on he rading generaed by fund invesors. Hence, we check wheher here is he presence of herding formaion by 6 Table isn shown in his paper and resuls are available for reques. 17

18 fund invesors in Shanghai Sock marke. We follow he mehodology used by Chang, cheng and Khorana (000). They menion herd behavior as an obvious inen by invesors o copy he behavior of oher invesors. The esing mehod is based on he idea ha invesors are more likely o suppress heir own beliefs in favor of he marke consensus during large price changes, so herd behavior is mos likely o emerge during such periods 7. Hence, equiy reurn dispersions around aggregae marke reurn are used o es formaion of herds during periods of marke sress. Then, one should expec significanly lower dispersions in individual securiy reurns as invesors are drawn o consensus of he marke. The squared differences beween fund index reurn and marke index reurn are used as a measure of reurn dispersion as follows: Df R R = ( f m) (5) where R f and R m are he reurns of fund index and A-share price index a ime, respecively. This mehodology suggess ha he presence of herd behavior is mos likely o occur during periods of exreme marke movemens, as hey would mos likely end o go wih he marke consensus. Hence we esimae he following linear regression model: Df = µ + γ D + γ D + ε (6) L U 1 where D =1, if he marke index reurn on ime lies in he 1%(5%) lower ail of he L reurn disribuion, 0, oherwise. 7 Chang, Cheng and Khorana (000) follow he same idea o es herd behavior in inernaional equiies. 18

19 U D =1, if he marke index reurn on ime lies in he 1%(5%) upper ail of he reurn disribuion, 0, oherwise. We define an exreme marke reurn he same as Chang, cheng and Khorana (000) do as one ha lies in he 1%(5%) lower or upper ail of he reurn disribuion. If here is he presence of herding formaion by fund invesors, he coefficiens γ 1 and γ are expeced o be negaive and saisically significan. Table 5 presens he empirical resuls of regression model (6). All he coefficiens are posiive and significanly significan a he 99% level. Our resuls are consisen wih prior research ha we do no find any evidence in favor of herding formaion during period of large marke movemens. The posiive coefficien γ 1 and γ indicae ha reurn dispersion increases when he marke experiences large marke change. This findings suppors he raional asse pricing models ha suggess ha periods of marke sress induce increased levels of dispersion as individual reurns differ in heir sensiiviy o he marke reurn. Based on he daa of changes in quarerly holding of individual socks, Lakonishok, Shleifer and Vishny (199) find ha here only exis very lile herding by insiuional invesors in heir sample. However, i doesn suppor he hypohesis of he presence of herding formaion. They poin ou ha insiuions are no desabilizing he prices of he individual socks hey rade under he mos common noaion of desabilizing speculaion Finally, Perhaps we should no be surprised by his resul. In he marke as a whole, aggregaing across all raders, here can be no herding, since for every share bough here is a share sold. The money manager do no seem o herd very much and hey use various rading syles ha resul in uncorrelaed rading decision on average. 7. Conclusion 19

20 I s sill an argumen in Chinese sock marke wha is he impac of insiuional rading on sock prices. If insiuions herd ogeher and follow posiively feedback sraegy, he rading generaed by insiuions may desabilize he marke. However, empirical findings on insiuional invesors herding and posiive feedback rading behavior are no necessary evidence in favor of he desabilizing effec on sock prices. By sudying he relaion beween fund flows and he marke volailiy we have a beer undersanding of muual fund invesors behavior, no only heir reacion o pas marke performances, bu also heir aiude o invesmen risk a he aggregae level. In his paper, we show ha a daily frequency he behavior for invesors in muual funds urns ou o be somewha differen. The invesors behavior appears o relae o measures of marke volailiy. We examine he impac of fund raders on sock prices using daily daa a he aggregae level. Shanghai A-share price index and fund index are employed and cover he period of January, 1996 o December 31, 004. Fund index was launched in Shanghai Sock Exchange on June 000 and i could be regard as an indicaor ha muual fund plays an imporan role. We fill his gap in he lieraure by providing evidence abou muual fund behavior, using muual index. We consruc a GARCH (1,1) model wih a ime dummy variable o deec he srucural change in mean and volailiy of marke reurn. We find ha he marke-wider volailiy is reduced afer June 000 and i provides direc evidence of fund rading sabilizing financial marke. Also we check he relaion beween fund flow and marke price index o deec he impac of fund flow on sock prices. Using he GARCH (1,1) model, he resuls show ha aggregae rading volume and urnover of lised funds are posiively correlaed wih subsequen index reurn and negaively correlaed wih subsequen condiional volailiy. I seems ha he behavior of invesors in muual funds conribues o he marke volailiy in Chinese marke. Due o he lack of fund inflow and ouflow daa, we can conclude wheher fund invesors follow posiive feedback sraegy or negaive feedback sraegy in Chinese marke. While, he absence of herding 0

21 formaion by fund invesors suppors our empirical finding ha rading generaed by fund invesors doesn desabilize he marke. 1

22 Figure 1. The Auocorrelaion Funcion of he series of daily aggregae rading volume of Open-Fund lised in Shanghai Sock Exchange during he period from June 9, 000 unil December 31, Figure. The Auocorrelaion Funcion of he series of daily aggregae urnover of Open-Fund lised in Shanghai Sock Exchange during he period from June 9, 000 unil December 31, 004.

23 Figure 3. Daily reurn of Shanghai A-share Price Index over he period January, 1996 o December 31, Figure 4. The ime series plos of closing price of Shanghai A-share price index during he period from January, 1996 o December 31, 004 3

24 Table 1 Table 1 presens ime series regressions of daily aggregae rading volume and urnover of lised open-fund on heir own lagged variables for he period June 9, 000 hrough December 31, 004. T-saisics are in parenheses. Independen Regression variables I II III IV V Panel A Trading Volume Conac (11.9) 3.714(8.67).5194(6.6).169(5.75).1718(5.54) Lag (40.) (19.9) 0.556(17.9) (16.6) (16.4) Lag 0.469(8.44) (3.45) (3.03) (.89) Lag 3 0.7(7.7) (4.98) (4.91) Lag (3.93) (3.1) Lag (0.649) R F -saisics ( p value) a (0.0000) (0.0000) (0.0004) (0.3149) (0.576) Panel B Turnover Conac.564(8.8) 1.950(6.68) 1.500(5.) (4.55) 1.70(4.34) Lag (57.5) (.6) (0.8) 0.588(19.6) (19.3) Lag 0.351(8.01) (.63) (.31) (.13) Lag (7.3) 0.144(4.16) 0.14(4.09) Lag (3.93) (.79) Lag (1.14) R F -saisics ( p value) a (0.0000) (0.0000) (0.000) (0.3331) (0.159) a. The es for he presence of firs-order and second-order auocorrelaion in he residuals. 4

25 Table Table presens he resul of he following GARCH (1,1) model: R = ω + θ R + θ D + ε 1 1 h = α + αε + βh + γd D = 1 afer June 9, 000; 0 oherwise where R is he reurn of A-Share Price Index and D is he ime dummy variable. The above model is used o examine he impac of fund rading on he sock prices. The ime dummy variable is included in boh mean equaion and condiional volailiy equaion. The sample sars January, 1996 and ends December 31, 004 ha amouns o 179 observaions. The esimaors are repored and Sandard Errors are in parenheses. Coefficien p-value ω (0.0439) ** (0.016) θ 1 θ α 0 α 1 β ( ) ** (0.0690) *** (0.0300) *** (0.0355) *** γ (0.0453) ** 0.03 Log-likelihood * denoes significance a he 10% level ** denoes significance a he 5% level *** denoes significance a he 1% level 5

26 Table 3 Table 3 repors he resuls of he following GARCH (1,1) model: R = ω + θ R + θ V + ε h = α + αε + β h + γv where R is he reurn of A-Share Price Index and 1 V is he daily aggregae rading volume of open-fund lised in Shanghai Sock Exchange a ime 1. And he aggregae daily volume is used as proxy for he daily aggregae fund flow. This model is employed o invesigae he effec of fund flow on sock prices during he period of June 9, 000 o December 31, 004. Panel A shows he resuls of he above model. Panel B and C show he resuls of he model when he exogenous variable, daily aggregae rading volume, is replaced by he expeced daily aggregae rading volume and unexpeced daily aggregae rading volume. Sandard Errors are in parenheses. Panel A: Panel B: Panel C: Coefficien Coefficien Coefficien ω (0.607) *** -.064(0.7498) *** (0.030) (0.098) (0.095) (0.098) θ 1 θ α 0 α 1 β (0.0316) *** (0.039) *** (0.0584) (0.471) * (0.445) * (0.078) *** 0.157(0.099) *** 0.14(0.095) *** (0.070) *** 0.80(0.0394) *** 0.87(0.0370) *** (0.031) *** γ (0.009) * (0.050) * (0.0400) *** Log-likelihood * denoes significance a he 10% level ** denoes significance a he 5% level *** denoes significance a he 1% level 6

27 Table 4 Table 4 repors he resuls of he following GARCH (1,1) model: R = ω + θ R + θ T + ε h = α + αε + βh + γt where R is he reurn of A-Share Price Index and 1 T is he daily aggregae urnover of open-fund lised in Shanghai Sock Exchange a ime 1. And he daily aggregae urnover is used as proxy for he daily aggregae fund flow. This model is employed o invesigae he effec of fund flow on sock prices during he period of June 9, 000 o December 31, 004. Panel A shows he resuls of he above model. Panel B and C show he resuls of he model when he exogenous variable, daily aggregae urnover, is replaced by he expeced daily aggregae urnover and unexpeced daily aggregae urnover. Sandard Errors are in parenheses. Panel A: Panel B: Panel C: Coefficien Coefficien Coefficien ω (0.5379) *** ) *** (0.0301) (0.098) (0.095) 0.01(0.0300) θ 1 θ α 0 α 1 β (0.081) *** (0.0318) *** 0.003(0.0649) (0.3700) * 0.676(0.3379) * (0.030) *** 0.164(0.0301) *** 0.155(0.097) *** 0.131(0.083) *** 0.803(0.0396) *** 0.868(0.0368) *** (0.0340) *** γ (0.0178) * (0.0165) -0.33(0.0649) *** Log-likelihood * denoes significance a he 10% level ** denoes significance a he 5% level *** denoes significance a he 1% level 7

28 Table 5 Table 5 presens he resuls of regression of reurn dispersion on exreme marke movemens. Df = µ + γ D + γ D + ε L U 1 where Df is reurn dispersion, and D =1, if he marke index reurn on ime lies in L he 1%(5%) lower ail of he reurn disribuion, 0, oherwise; D =1, if he marke index reurn on ime lies in he 1%(5%) upper ail of he reurn disribuion, 0, oherwise. Sandard Errors are in parenheses. Marke index reurn in exreme 1% lower and upper of he reurn disribuion U Marke index reurn in exreme 5% lower and upper of he reurn disribuion µ 0.513(0.0339) *** (0.0341) *** γ 1 γ (0.3376) *** (0.1488) ***.715(0.3376) *** (0.1488) *** R F-es (,1098) 65.78[0.000] *** 11[0.000] *** * denoes significance a he 10% level ** denoes significance a he 5% level *** denoes significance a he 1% level 8

29 Reference: Busse, Jeffrey, 1999, Volailiy iming in muual funds: evidence from daily reurns, The Review of Financial Sudies 1, P Chan, Louis K.C. and Josef Lakonishok, 1993, Insiuional rades and inra-day sock price behavior, Journal of Financial Economics 33, P Chan, Louis K.C. and Josef Lakonishok, 1995, The behavior of sock prices around insiuional rades, Journal of Finance L (4), P Chang, E.C., J.W. Cheng and A. Khorana, 000, An examinaion of herd behavior in equiy markes: An inernaional perspecive, Journal of Banking and Finance 4(10), P Delong, J.Bradford, Andrei shleifer, Lawrence H. Summers and Rober J. Waldmann, 1990, Posiive feedback invesmen sraegies and desabilizing raional speculaion, Journal of Finance, XLV (), P Demirer, Riza and Ali M. Kuan, 004, Does herding behavior exis in Chinese Sock Markes?, Working paper, Souhern Illinois Universiy Edwardsville. Dennis, Parick and Deon Srickland, 00, Who blinks in volaile markes, individuals or insiuions?, Journal of Finance 51, P Edelen, Roger M., 1999, Invesor flows and he assessed performance of open-end muual funds, Journal of Financial Economics 53, P Edelen, Roger M. and Jerold B. Warner, 001, Aggregae price effecs of insiuional 9

30 rading: a sudy of muual fund flow and marke reurns, Journal of Financial Economics 59, P Faugere, Chrisophe and Hany A. Shawky, 003, Volailiy and Insiuional invesor holdings in a declining marke: A sudy of Nasdaq marke during he year 000, working paper, Cener for insiuional invesmen managemen, Universiy a Albany, SUNY. Gabaix, Xavier, Parameswaran Gopikerishnan, Vasiliki Plerou and H. Eugene Sanley, 005, Insiuional invesors and sock marke volailiy, Quarerly Journal of Economics, forhcoming. Goezmann, William N. and Massimo Massa, 003, Index funds and sock marke growh, Journal of Business 76(1), P1-8. Har, Oliver and David Kreps, 1986, Price desabilizing speculaion, Journal of Poliical Economy 94, P Holhausen, Rober, Richard Lefwich and David Mayers, 1987, The effec of large block ransacions on securiy prices: A cross-secional analysis, Journal of Financial Economics 19, P Lakonishok, Josef, Andrei Shleifer and Rober W. Vishny, 199, The impac of insiuional rading on sock prices, Journal of Financial Economics 3, P3-43. Lipson, Marc and Andy Pucke, 005, Volaile markes and insiuional rading, working paper, Terry college of Business of Universiy of Georgia. Luo, Dengpan, 003, Marke volailiy and muual fund cash flows, working paper, Yale inernaional cener for finance working paper. 30

31 Nofsinger, John and Richard Sias, 1999, Herding and feedback rading by insiuional and individual invesors, Journal of Finance 54, P Sias, Richard W. and Laura T. Sarks, 1997, Reurn auocorrelaion and insiuional invesors, Journal of Financial Economics 46, P Warher, V.A., 1995, Aggregae muual fund flows and securiy reurns, Journal of Financial Economics 39, P Wermers, Russ, 1999, Muual fund herding and he impac on sock prices, Journal of Finance 54, P