Investor Sentiment and ETF Liquidity - Evidence from Asia Markets

Transcription

1 Advances in Managemen & Applied Economics, vol. 6, no.1, 2016, ISSN: (prin version), (online) Scienpress Ld, 2016 Invesor Senimen and ETF Liquidiy - Evidence from Asia Markes Yung-Ching Tseng 1 and Wo-Chiang Lee 2 Absrac This sudy aims o analyze he effec of invesor senimen on Exchange Traded Fund (ETF) liquidiy, and o capure he variaions in invesor senimen, mainly focusing on Asian ETF marke daa. We employ he Volailiy Index and GARCH model o capure he volailiy-clusering effec in he sudy. The empirical resul shows ha ETF has liquidiy, and he degree of invesor senimen plays an imporan role in ETF liquidiy wihin Asian counries. I indicaes a volailiy-clusering effec, dealing wih he difference of rading sysems, regulaions in he marke, and finds ha he relaionship beween VIX and ETF liquidiy is significanly differen. Considering hedging agains marke risk and porfolio invesmen, his paper suggess ha invesors should ake invesor senimen ino heir invesmen decisions, and re-adjus he invesmen weigh of ETF produc. JEL classificaion numbers: C02, G11, G15 Keywords: Invesor Senimen, ETF Liquidiy, Liquidiy-volailiy-clusering Effec, Volailiy Index 1 Ph.D. Candidae, Deparmen of Banking and Finance, Tamkang Universiy, Taiwan. 2 Professor, Deparmen of Banking and Finance, Tamkang Universiy, Taiwan Aricle Info: Received : November 15, Revised : December 8, 2015 Published online : January 30, 2016

2 90 Yung-Ching Tseng and Wo-Chiang Lee 1 Inroducion The global ETF Indusry experienced is bes growh ever pushing Asse Under Managemen, AUM) o $2.6 rillion by he end of 2014 seing a new record. ETF rading aciviy was up by 13% in 2014 reaching $18.7 rillion and will coninue o rise, and ETF markes are advancing globally showing no signs of slowing down. According o ETF flows, he researcher finds ha invesors preferred less risky asses. Deusche Bank expecs global ETF asses o pass $3 rillion in The rapid capial formaion of ETF during recen years has made i one of he favorie invesmen producs for reail invesors, as i has lower invesmen coss especially compared o index funds. Gradually, ETF has also become one of he mos imporan invesmen producs in cusomer invesmen porfolios as compiled by Bank wealh managemen schemes; he lower rading cos makes ETF one of he mos popular core asses for invesors. Many policy holders will also choose ETF o accumulae heir accoun value in life insurance and variable annuiy producs. The main reason why cusomer choose ETF is o be involved in marke growh, especially in an era of low ineres raes, while a he same ime avoiding decreasing wealh due o he effecs of inflaion. When hey are bullish abou he marke, hey can be involved wihou spending ime and effor o choose specific socks or equaliy producs. When hey choose ETF as heir invesmen produc, hey will consider he issuer, he rading plaform, he value of he produc, and liquidiy, however hey ofen ignore invesmen senimen. Invesor senimen may cause price volailiy, and herefore influence he liquidiy of he invesmen produc iself leading o a decrease in rading volume. A review of he lieraure on research ino ETF produc feaures mainly focuses on capuring he behavior of he ETF produc reurns. Fujiwara (2006) finds ha here is a correlaion beween he changes in discoun raes and he small capial sock index, bu hese phenomena were no observed in an ETF. And Li e al. (2012) observed a U-shaped and an L-shaped inraday paern for rading volume and reurn volailiy. They found a significan increase in rading volume and urnover raio for all ETFs boh during and afer he financial crisis. As here is a correlaion beween ETF and capial markes, a variey of ETF ypes in an invesmen porfolio can be means of hedging during a financial crisis. Boscaljon and Clark (2013) find ha during a financial crisis, here is a posiive abnormal reurn for equiies in SPDR Gold Share (GLD) exchange raded funds (ETF), if VIX increases 25%. Ivanova e al. (2013) observe ha price discovery is influenced in a differen manner by he emporal behavior of he exchange raded funds price discovery meric in he spo and fuures markes across indexes. The marke invesor s ransacion migh be influenced by heir individual senimen, especially when here is high uncerainy in he marke. Invesor rading behavior will change significanly, i.e. feedback rading will cause wildly flucuaing prices. Recen researchers, such as Chau e al. (2011), believe ha he presence of

3 Invesor Senimen and ETF Liquidiy senimen-driven noise rading will largely generae feedback rading aciviy. For regulaors and invesors, invesmen senimen and marke dynamics are direcly relevan, so when we sudy ETF, we need no only focus on ETF reurn, bu pay more aenion o invesor senimen. Besides senimen, liquidiy is also very imporan, especially facing markes a differen degrees of developmen, i.e. ETF produc volailiy in developed markes and significan variaion in newly emerging Asian markes(guierrez e al., 2009). The difference in volailiy migh be caused by he speed ha informaion spreads, he produc feaures and invesor s wihholding informaion (Chiu e al., 2012). Levels of volailiy will also cause changes in liquidiy. In his aricle, we assume ha he poor liquidiy of financial produc can lower he ransacion will of new financial produc, decrease he insiue invesor prof and hinder governmen o promoe new financial produc. Afer he 2008 financial crisis, invesors increased heir demand for muliple financial producs o reduce invesmen risk. In recen years, an increased emphasis on risk avoidance has mean ha various counries have begun o diversify he ypes of financial derivaives available. Generally speaking, invesor behavior impacs heir rading sraegy. Invesor senimen will influence his or her rading behavior. So ha when he marke liquidiy is measured by he rading volume, invesor senimen will of course become a criical facor. This paper aims o analyze wheher he liquidiy of ETF is influenced by invesor senimen. According o Chiu e al. (2012) research indicaes ha wih an increase in funding illiquidiy during he subprime crisis period, a corresponding increase in he bid ask spread and a decrease in marke deph was found, indicaing a general reducion in equiy liquidiy, According o he relaed inroducion on ETF liquidiy change (Chiu e al., 2012), previous findings demonsrae ha ETF liquidiy will be influenced by volailiy value, which is seldom discussed in curren lieraure. They menioned and believed, wih clear evidence, ha shocks o liquidiy and a coninuous flow of poor marke informaion will pu pressure on ETF redempion and herefore change financial liquidiy, which will influence he liquidiy of he ETF iself. However, hey did no explain he reason why invesor senimen migh be he cause of his ype of liquidiy change. Invesors are influenced by marke informaion, which will impac he flucuaion of invesor senimen, and possibly furher influences he liquidiy of ETF. This paper exends he research on ETF liquidiy, by considering he senimen facor. Chau e al. (2011) found saisically significan evidence suggesing ha a negaive relaionship beween auocorrelaion and volailiy, senimen influence seems o be sronger during a bullish marke. They find evidence of a direc impac of invesor senimen on he momenum-syle feedback rading sraegies, and hose resuls are very imporan

4 92 Yung-Ching Tseng and Wo-Chiang Lee in conribuing o he curren debae on he role of invesor senimen in asse pricing and invesmen behavior. They focus on he evidence from research of a relaionship beween invesor senimen and rading behavior. Alhough here is a significanly negaive correlaion beween auocorrelaion and volailiy, which hey recognize, according o he saisical daa measuring invesor senimen, here is evidence ha invesor senimen will influence volailiy and herefore aler rading behavior. However, here is no furher explanaion or discussion abou he relaionship beween invesor senimen and volailiy change. The key poin in he above wo aricles is ha invesor senimen is an absrac qualiaive facor, which influences ETF volailiy o indicae is liquidiy. In addiion, our paper conribues o heir discussion on invesor senimen. Guierrez e al. (2009) finds ha he overnigh volailiy is higher han dayime volailiy, boh U.S. reurns and local Asian marke reurns explain he Asian ETF reurns. The rade locaion and invesor senimen effecs are furher suppored by he high reurn correlaion beween Asian and U.S. ETFs. The bi-direcional Granger causaliy in volailiy beween he U.S. and he six Asian markes analyzed are found in his aricle. Their findings demonsrae ha local marke informaion can be used o explain ETF volailiy and reurns in Asia, bu he discussion on he source of volailiy is no discussed in deail. On he oher hand, hey find ha ETF volailiy and reurn are influenced by each oher depending on differen rading regions, so ha he relaionship beween invesor senimen and volailiy migh be overlooked. This paper reviews he ineracion beween he findings of hese hree papers o observe how invesor senimen can influence ETF reurns and ETF volailiy using volailiy. This paper analyses he ETF samples from five main counries in he Asia-Pacific region, including Japan, Souh Korea, Taiwan, Malaysia and Singapore, from January 31s 2005 o January 30h This includes he period of he global financial crisis period, o provide evidence of he effec of he financial crisis on liquidiy. The empirical daa indicaes ha rading volume and invesmen senimen have a significan impac on he ETF liquidiy of he sample counries. We review previous daa, and find liquidiy feaures volailiy clusers, which means ha liquidiy will perform high or low in specified period. We adop he GARCH model o analyze he sample daa, and he empirical resuls prove ha invesor senimen has a significan impac on overall ETF volume. The empirical resuls will be significanly differen during a ime of crisis, and we believe ha his is caused by he differen financial environmen, sysem or invesmen senimen in hese counries, which is also proved by our resuls. We found ha he ETF liquidiy volailiy cluser phenomenon apparenly exiss in Malaysia and Souh Korea, bu i becomes less apparen in financial markes such as Japan, Singapore, and Taiwan, where here is a greaer mauriy in ETF producs. For counries wih inconsisen ETF liquidiy, we infer ha i is based on he counries

5 Invesor Senimen and ETF Liquidiy financial environmen; changes in ransacion sysems, mauriy of ETF producs and invesmen senimen. The remainder of he paper is organized as follows. Secion 2 briefly reviews he relaed lieraure and Secion 3 describes he variables and empirical research models used in our invesigaion. The daa and descripive saisics are provided in Secion 4 where he basic saisics applied o variables in he research are presened and which hen discusses he main empirical resuls and robusness checks. Finally, Secion 5 concludes he paper. 2 Lieraure review 2.1 Senimen and rading behavior Afer he 2008 financial crisis, invesors increased heir demand for muliple financial producs o reduce risk, causing various counries o diversify heir financial derivaives. ETF being one of he bigges. The recen research ino ETF volailiy poins ou ha invesor senimen, which migh be influenced by cerain excepional evens, impacing invesor rading behavior, depending on he invesor s posiive or negaive expecaion. Wha is more, his will furher affec he rading volume proporionally. In pas lieraures, Edelen e al, (2010) regarded senimen flucuaion in erms of risk olerance, or overly opimisic or pessimisic cash flow forecas invesmen environmen. In each case, he impac of he influence of senimen on asse pricing should be obvious from fundamenals. The research ino invesor senimen usually focuses on a discussion abou arge reurns, invesor senimen and rading. Glabadanidis (2014), has proven ha abnormal reurns are generaed by a moving average (MA) rading sraegy, bu invesor senimen canno fully explain is anomalies. This research shows ha i is impossible o use invesor senimen o explain abnormal reurns generaed by rading policies, as one empirical research shows ha abnormal reurns generaed from using rading sraegies migh exclude he influence on arge price performance caused by invesor senimen. However, anoher researcher believes ha invesor senimen can be used o improve he invesmen porfolio performance. According o Basu, Hung, Oomen, and Sremme (2006), senimen can improve he performance of dynamically managed porfolio sraegies for sandard marke-imers as well as for momenum-ype invesors. Recen research poins ou ha Invesor senimen will affec he rading behavior of he average invesor or insiuional invesors (Edelen e al., 2010). Feedback rading in he E-mini index fuures markes in microsrucure seing is examined by Kurov (2008), and he finds ha raders in index fuures markes are posiive feedback raders and heir feedback rading end o be more inense in periods where invesor senimen is high. There are normally hree ypes of invesor senimen, and he degree of each ype will have an impac on he arge price, he

6 94 Yung-Ching Tseng and Wo-Chiang Lee rading volume, or produc selecion. In he case of a posiive senimen, Chau e al. (2011) finds ha here is significan posiive feedback rading in he U.S. ETF markes, and he inensiy of which ends o increase when invesors are opimisic consisen wih he view ha he marke is less raional and inefficien during high-senimen periods due o a higher paricipaion by noise raders in such periods. In he case of negaive senimen, Chiu e al. (2014) shows ha when he fear-based marke senimen increases when here is a bearish insiuional invesor expecaion, here will be a significan decrease in ne buying volume and marke liquidiy han in normal imes in response o he ineracion beween fear-based senimen and insiuional invesor expecaion. Therefore, he variaion in Invesor senimen will impac invesmen producs and rading volume. This will especially impac produc selecion and herefore he invesmen porfolio. ETF can mee he invesor requiremen of reduced risk, and expeced reurn. From he above discussion i is clear from previous lieraure ha facors which impac rading such as senimen should no be ignored, especially he risk-averse invesor. A posiive senimen in he marke will increase he volailiy risk and affec ETF reurns. The implicaion is ha as marke rading volume is influenced, hen so is liquidiy. Invesor senimen does no only affec rading behavior, bu here is also a correlaion wih volailiy, which furher impacs reurns from invesmen producs. A conemporaneous relaion beween changes in invesor senimen and U.S. sock marke reurns is inroduced by Brown and Cliff (2004). Even if invesor senimen can be used o predic sock reurns, Lemmon and Porniaguina (2006) find he reurns on small size socks can be prediced by invesor senimen. In recen years, research on he correlaion beween invesor senimen and volailiy is expanding. The relaion beween he expeced reurns and volailiy in he U.S. sock marke hinges on invesor senimen were firs proposed by Yu and Yuan (2011). Furhermore, a posiive relaionship beween shifs in senimen and sock reurns is found by Li and Zhang (2008) in he Chinese sock marke and a negaive correlaion wih senimen during periods of high marke volailiy. This empirical research shows ha invesor senimen is obviously correlaed wih volailiy and reurn in a variey of ways. Such as, he asymmery found in he predicive power of invesor senimen in sock reurns in imes of flourishing economic environmens when invesors become more opimisic, and in imes of economic downurns when invesors are more pessimisic. This is capured by Chung e al. (2012).Furhermore, Baker and Wurgler (2006) prove ha invesor senimen is relaed o he expeced reurns and risks in he marke. Undervalued socks are likely o be undervalued more srongly when invesor senimen is low and vice versa when invesor senimen is high. However, some empirical resuls are differen wih invesor s idea. Schmeling (2009) finds ha in mos of he 18 indusrialized counries, fuure sock reurns end o be lower, when consumers have high confidence. According o Ho and Hung (2009), he explanaory power

7 Invesor Senimen and ETF Liquidiy of asse pricing models for sock reurns are enhanced by incorporaing invesor senimen in modeling he dynamics of risk exposure. In hese sudies, hey found ha alhough here are variaions in correlaion beween invesor senimen and volailiy, invesor senimen sill provides a powerful ool for explaining invesmen reurns. Chiu e al. (2012) recenly proposed ha as far as ETF liquidiy is concerned he fund flow and liquidiy will change in correlaion when big evens occur. The rading volume of ETF has increased significanly in recen years causing research ino ETF financial producs o arac more aenion. 2.2 Trading behavior and volailiy In order o discuss he correlaion beween invesor senimen and rading behavior, rading behavior and volailiy also needs o be included. The difference in ime zones or rading hours will cause produc price volailiy leading o dramaic increases or decreases. Masahiro (2008) proposes a hump-shaped relaion beween rading volume and informaion precision, and a posiive correlaion beween rading volume and absolue price changes. The volailiy and correlaion of sock reurns in he highly volaile and srongly correlaed equilibrium will increase when here is accurae informaion. Besides, many papers discuss he correlaion beween rading behavior and volailiy. According o Nielsen and Shimosu (2007), here is weak evidence of fracional coinegraion beween realized volailiy and rading volume for mos of he socks considered. Recen research ino he momenum effec has discovered ha behavior in he financial field may provide evidence of he exisence of a long-erm memory. Rossi and Magisris (2013) find ha in mos cases, volume and volailiy are characerized by a long memory bu no fracionally coinegraed. They also find righ ail dependence, which is indicaive of he behavior of volailiy and volume when surprising news impacs he marke. Almos all of his research demonsraes ha he correlaion beween rading behavior and volailiy are significan. 2.3 Volailiy and liquidiy As he expansion of he scale of he fund marke increases and he mauriy of he insiuional invesor coninues o grow, he requiremen of he insiuional invesor for marke liquidiy is increasing, and he risk managemen of his liquidiy receives increasing amouns of aenion. If he rading volume of he financial produc is no large, or he liquidiy is poor, i will cause concern for boh he insiuional invesors and he governmen financial regulaion deparmen, so ha liquidiy becomes an imporan research opic. In order o improve he rading diversiy in he fund marke, governmens need o pay more aenion o he liquidiy of newly promoed producs. Pas research has compared resuls where here is posiive and negaive senimen, and he research oucome can be used o mainain liquidiy in a low rading level during negaive invesor senimen o avoid poor liquidiy. The pas discussion on poor liquidiy has mainly focused on

8 96 Yung-Ching Tseng and Wo-Chiang Lee he pricing mechanism problem, i.e. excessive differences, opaque pricing, and a lack of marke makers. However, i is also very imporan o undersand he effec of changes in liquidiy on financial producs, because a lack of liquidiy will lower rading will which will resul in a sharp fall in insiuional invesor profi and hinder he governmen s abiliy o promoe new financial producs. In heir iniial research ino liquidiy, Pasor and Sambaugh (2003) firs repored on finding reurn sensiiviy in marke liquidiy. This highlighed his area of invesmen which led o more sudies on liquidiy and volailiy, Chordia e al. (2005) also found ha innovaions in he sock and bond marke are srongly correlaed o liquidiy and volailiy. From hese he common elemens which drive liquidiy and volailiy in sock and bond markes may be inferred. Consequenly, correlaion beween volailiy and liquidiy are receiving increasing aenion. In research by Karoly e al. (2012), hey inerpre he resuls as evidence for he demand-side heory, ha liquidiy commonaliy is greaer during imes of high marke volailiy in counries wih a greaer presence of inernaional invesors and more correlaed rading aciviy. Recenly, researchers have focused on developing measuremen ools for volailiy and oher facors. He e al. (2014) proposed ha all liquidiy measures of SEO (Seasoned Equiy Offering) firms show significan improvemens afer SEO evens. The relaive offer size, he change in sock price and volailiy wih corresponding signals are grealy associaed wih he magniudes of reducion in ransacion cos measures for illiquidiy. This research has revealed he imporance of liquidiy issues, which are no only faced by governmen financial supervision deparmens, bu are also an imporan global risk managemen opic afer he 2008 financial crisis. There is exensive research ino invesor senimen and rading behavior, rading behavior and volailiy, pas volailiy and liquidiy, however he correlaion beween invesor senimen and liquidiy are seldom examined. This paper will ry o make deducions based on pas heoreical foundaions, and analyze research daa using relaed models, o prove a correlaion beween invesor senimen and liquidiy. 3 Variables, informaion and research mehods This sudy explores wheher here is a significan correlaion beween invesor senimen and ETF liquidiy, a dummy variable is added o represen periods of pessimisic invesor senimen. In addiion, a second model is used o perform a paired observaion on he influence on change of fluidiy during a period of panic. We also use he GARCH model o capure wheher he liquidiy exhibis he volailiy cluser effec. In erms of informaion, we use he American panic index o represen invesor senimen, and for he ETF informaion, ishares by Black Rock (he world s larges ETF issuing plaform) is used. We seleced five

9 Invesor Senimen and ETF Liquidiy Asia-Pacific counries, namely Japan, Souh Korea, Taiwan, Malaysia and Singapore, o analyze and sudy Black Rock's ishare. The sudy period was from Jan. 31, 2005 o Jan. 30, 2015, covering he period of he financial sunami. The seings, definiions and verificaions which relae o he variables and models are described as follows: 3.1 Variables ETF liquidiy raio We seleced he ETF in Japan, Korea, Taiwan, Malaysia and Singapore from he ishare plaform, and we used Karolyi e al. (2012) o calculae ETF liquidiy, his paper calculaes he liquidiy raio as follows: 6 L log( 1 10 (1) V R where R and V are he reurns and rading volume for he counry ETFi on day, respecively. The liquidiy raio, L, is he increase in he liquidiy for counry ETFi Volume This paper uses rading volume in shares as found in Wang (2013) o calculae he liquidiy raio. In addiion, we also use he rading volume as an imporan variable, in order o observe he correlaion beween rading volume change and ETF liquidiy change Invesor senimen-vix index measures Marke Volailiy Index ("VIX") is a measure of he implied volailiy S&P 100 index opion. Ofen referred o as he "invesor fear index" (Whaley, 2000), we use his index as a proxy variable for invesor senimen. The VIX index was inroduced by CBOE (Chicago Board Opions Exchange) in 1993, i is an index obained afer weighing he average of index opions implying volailiy. The index reflecs he coss invesors are willing o pay and rea heir invesmen risk, i is widely used o reflec he invesor's panic degree regarding he afermarke, also known as he "fear index". When he index is higher, i means invesors are more anxious abou he sock marke saus; when he index is lower, i indicaes he sock index change for he marke will end o slow down. The calculaion of VIX is done by selecing a oal of eigh sequences from he previous-monh and he following-monh pu and call opions of S&P 100 index opion ha are closes o he a-he-money, and respecively calculaes is weighed average of implied volailiy o obain he index. Laer, he index was amended in 2003.The seleced

10 98 Yung-Ching Tseng and Wo-Chiang Lee subjec was changed from S&P 100 o S&P 500, and changes were made o he closes a-he-money pu and call opions sequences o all of he sequences. The broader subjec maer basis provides marke paricipans wih an indicaor ha can beer reflec he overall broader marke rend. The empirical period for his paper will use he new VIX index amended in 2003 for he esimae Dummy variable ( PESS Dummy i, ) When invesor senimen VIXR ) flucuaion is over (less han) a sandard ( deviaion (7.01%), he dummy variable of pessimisic (opimisic) senimen is expressed as 1, as opposed o 0. This sudy will be able o, hrough he seing of a cross-muliplying erm as a pessimisic dummy variable and invesor senimen VIXR PESS Dummy ),observes he influence of invesor senimen on ( liquidiy in he panic period. This will allow us o observe he degree of influence of he pessimisic marke-invesmen amosphere on each counry s ETF liquidiy. 3.2 Model specificaion Generalized auoregressive condiional heeroscedasiciy (GARCH) In order o idenify wheher he liquidiy exhibis volailiy clusers and oher characerisics, we added he deecion of he GARCH model o he model. The radiional economeric model and ime sequence model boh assume he variances of error erms are fixed in order o conduc relaed deducion and research. However, he raionaliy of his assumpion has been challenged by many scholars, because informaion he general financial ime sequence does no obey his assumpion, i.e. he presenaion of variances vary over ime. Therefore, Engle (1982) proposed he Auoregressive Condiional Heeroskedasiciy (ARCH) Model. Bollerslev (1986) amended he ARCH model and proposed he Generalized Auoregressive Condiional Heeroskedasiciy (GARCH) model, he hough ha he condiional variances are no only affeced by he previous ime periods' error squared erms, bu also affeced by he previous ime periods' condiional variances. Hence, he seing of he GARCH(p, q) model is as follows: Y (2) X 1 N 0, ) (3) ( C A B (4) In he above expression (3), 1 means all he informaion se can be obained in he -1 ime period. The Y and X represen he model s explained variable vecor and explanaory variable vecor, and includes he column vecors of is

11 Invesor Senimen and ETF Liquidiy exogenous variables or lagged dependen variables. β is he o-be-esimaed parameer vecor. The parameers C, A and B are non-negaive real numbers, o ensure he variances is posiive, and mees he 1 condiion of he saionary sae. Meanwhile, i adops he maximum likelihood esimaion mehod o obain he esimaes of parameers C, A and B: Max 2 C, A, B L xp x (5) Take log of he above expression: x LL ln(2 ) ln 1 (6) Finally, adop he repeaed esimae algorihm o maximize he expression (5), o obain he esimaes of parameers C, A and B Empirical models This paper aims o explore if here is a significan correlaion beween invesor senimen and ETF liquidiy. We adoped ordinary leas squares (OLS) o se up. Model 1 adops he rading volume ( Vol i, ) and invesor senimen ( VIXR i, ) o observe he correlaion of each counry s ETF liquidiy ( L i, ); In Model 2, he pessimisic dummy variable was added ( PESS Dummy i, ) from model 1 and combined wih invesor senimen o form a cross-muliplying erm VIXR PESS Dummy ). This cross-muliplying erm was used o represen ( he marke panic period in order o observe when he marke was presening a pessimisic invesmen amosphere. This allowed us o observe wheher here was any difference in each counry s ETF liquidiy influence. Therefore, he model seings of his sudy are described as follows: Model 1 : Model 2: L a L a0 a1vol 1 a2vixri, 1 (7) 0 a1vol 1 a3vixr 1 PESS Dummy 1 (8) N 0, ) (9) ( h i, h C A Bh (10) 2 1 1

12 100 Yung-Ching Tseng and Wo-Chiang Lee for i=j, s, m, y, o be proxies as counries ETF In Model 1 (7), L represens he liquidiy of ETFi on day, L L log, Li, L represens he liquidiy flucuaions, aking firs difference 1 of liquidiy (L ) and hen ake log; Vol represens he rading volume of ETFi on day, Vol ( Vol Vol 1), Voli, represens he rading volume changes on VIX ha day and he day before; VIXR log, VIXRi, VIX represens he invesor 1 senimen flucuaion of ETFi on Day, aking he log of he firs difference in invesor senimen ( VIX, ). This model seing mainly observes each counry s i curren-period ETF liquidiy change, and herefore, a he righ side of he equaion, no maer wheher i is Vol i, or VIX i,,we ake boh of he previous period s daa, which means each counry's curren-period ETF liquidiy change is affeced by he previous period's rading volume change and invesor senimen flucuaion. In Model 2 (8), we add he ( PESS Dummy i, ), which is a dummy variable. When he invesor senimen flucuaion of ETFi on day is over a sandard deviaion, i means ha he pessimisic senimen has increased. This is hen deemed he panic period, and he dummy variable is expressed on he conrary as 0. Through forming a cross-muliplying erm VIXR PESS Dummy ) wih invesor senimen, when he marke presens a ( pessimisic invesmen amosphere, o observe any differences beween each counry s influence on ETF liquidiy. Equaions (9) and (10) are he condiional variance equaions. They are mainly o esimae he coefficiens of each counry's ETF ARCH effec (A) and GARCH effec (B), and o check if he ETF liquidiy has a volailiy-clusering (A+B<1, A and B >0) phenomenon. 4 Source and processing This paper focuses analysis on counry ETFs issued by ishares, which is he world's larges ETF issuer and marke leader owned by BlackRock. The sample period used in his paper is from Jan. 31, 2005 o Jan. 30, 2015 and he ETFs from 5 Asian counries wih enough hisorical daa and rading aciviy was adoped in his sudy o carry ou he ess. All he daa used in his sudy is obained from he Daa sream Inernaional daabase.

13 Invesor Senimen and ETF Liquidiy Counry Table 1: Daa Source and Descripion Ticker Underlying index Japan EWJ isharesmsci Japan Index Singapore EWS isharesmsci Singapore Index Malaysia EWM isharesmsci Malaysia ETF Souh Korea EWY isharesmsci Souh Korea Capped ETF Taiwan EWT isharesmsci Taiwan Index Noe: The able provides informaion on he sample of ETFs including he icker and underlying index. This paper focuses our analysis on counry ETFs issued by ishares, which is he world's larges ETF issuer and marke leader owned by BlackRock. 4.1 Basic saisics The influence of invesor senimen on liquidiy was explored in his sudy; he sudy period was from Jan. 31, 2005 o Jan. 30, In he sudy period, here was a major financial crisis affecs markes around he world. The counries included in his research were Japan, Souh Korea, Taiwan, Malaysia and Singapore were covering several major counries in Eas Asia. A oal of 2518 samples were aken from rading-days, due o he model adoping he esimaion from previous-day change; herefore, here were 2517 observaion samples. The flucuaion in he American panic index was used as he proxy variable for invesor senimen. In order o deec wheher he liquidiy exhibis volailiy clusers and oher characerisics, we included he GARCH in he model. Furhermore, variable sequence daa mus ake he uni roo es prior o each model esimaion, o deec wheher each variable obeys he assumpion of saionary sequence. This is necessary in order o avoid he problem of spurious regression. In he es, he ADF (Said and Dickey, 1984) and PP (Phillips and Perron, 1988) of he radiional linear uni roo es mehod were adoped o conduc he deecion. The resuls shows ha all he empirical variables aking he linear uni roo es were a he 1% significance level. They all rejec he null hypohesis of he uni roo, i.e. hey obey he assumpion of saionary sae demand. Table 1 is he descripion of he ransacion code; Table 2 shows he descripive saisics of he daa sample, and conains Jarque-Bera Normal Disribuion es resuls. In he ETF reurns par, we ake he log of he firs difference in he daily closing price for each counry's ETS index as is remuneraion, in order o check he flucuaion of daily price remuneraion. From he value of he sandard deviaion we can find ha from Japan's o Malaysia's , he daily price flucuaion of hese five Asian counries' ETF is quie large. The proxy variables VIX index of he relaed invesor senimen also used he aking he log

14 102 Yung-Ching Tseng and Wo-Chiang Lee of he firs difference mehod o observe he daily volailiy, he sandard deviaion of daily volailiy is In coefficiens of skewness and kurosis, i is found ha all variables showed in he resuls of non-normal disribuion. A he 5% significance level and above, he ETF reurns variable of all counries shows a posiively skewed lepokuric disribuion excep Singapore, which shows a negaively skewed lepokuric disribuion. VIXR daa also shows a posiively skewed lepokuric disribuion. In addiion, he rading volume and liquidiy proxy observaion also show a posiively skewed lepokuric disribuion. The model observes ha he rading volume adops daily differenial values. On he samples of observed counries, Japan has he larges rading volume, and is rading volume s daily average change is also he greaes. We used he mehod of formula (1) o calculae he daily liquidiy and adoped he same mehod of aking he log of firs difference o observe he daily liquidiy change. In view of he numerical values, he lowes sandard deviaion of daily liquidiy value is Japan's ; he highes liquidiy change is Malaysia's I also reflecs he larger rading volume, mean in smaller liquidiy flucuaion

15 Invesor Senimen and ETF Liquidiy Table 2: Summary saisics Variable Mean SD Minimum Maximum Skewness Kurosis Jarque-Bera Panel A : Reurn R j, ** *** *** R s, *** *** *** R m, *** *** *** R y, *** *** *** R, *** *** *** VIXR *** *** *** Panel B : Trading Volume ΔVol j, *** *** *** ΔVol s, *** *** *** ΔVol m, *** *** *** ΔVol y, *** *** *** ΔVol, *** *** *** Panel C : Liquidiy ΔL j, * *** *** ΔL s, *** *** *** ΔL m, *** *** *** ΔL y, ** *** *** ΔL, ** *** *** Noe: Table 2 shows he descripive saisics of daa sample, and conains Jarque-Bera Normal Disribuion es resuls. In he ETF reurns(r ) par, ake log of firs difference of daily closing price of each counry's index ETS as is remuneraion, in order o check he flucuaion of daily price remuneraion; L Vol ( Vol Vol 1), Voli, represens he rading volume(vol) changes on ha day and he day before;, L log, Li, represens he L 1 liquidiy (L ) flucuaions, aking firs difference of liquidiy and hen ake log; For all i=j, s, m, y, o be proxies as counries ETF, j=ewj(japan), s=ews(singapore), m=ewm(malaysia), y=ewy(souh Korea), =EWT(Taiwan). ***, **, and * indicae saisical significance a he 1%, 5%, and.10% levels.

16 104 Yung-Ching Tseng and Wo-Chiang Lee Variable Panel A : Mean Equaion Consan ΔVol -1 VIXR -1 VIXR -1 X Dum -1 Panel A : Variance Equaion C A B L j, Coefficien (Sd. Error) Model 1 Model 2 L s, Coefficien (Sd. Error) Model 1 Model 2 Table 3: Parameer esimae resuls Coefficien (Sd. Error) Model 1 Model 2 Coefficien (Sd. Error) Model 1 Model 2 L, Coefficien (Sd. Error) Model 1 Model (0.0264) (0.0255) (0.0254) (0.0253) (0.0266) (0.0263) (0.0267) (0.0263) (0.0262) ( ) *** *** *** *** *** *** *** *** *** *** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) *** ** *** *** *** *** *** *** * ** (0.3507) (0.4051) (0.3980) (0.4044) (0.3948) (0.3709) (0.3992) (0.4195) (0.4110) (0.4021) *** *** *** (0.0634) (0.0586) (0.0474) (0.0518) (0.0432) *** *** *** *** *** *** *** *** *** *** (0.1484) (0.1540) (0.1342) (0.1347) (0.1240) (0.1362) (0.1526) (0.1518) (0.1379) (0.1429) *** *** *** *** *** *** *** *** *** *** (0.0296) (0.0274) (0.0308) (0.0312) (0.0341) (0.0337) (0.0315) (0.0321) (0.0300) (0.0301) *** ** ** ** (0.0653) (0.0681) (0.0494) (0.0527) (0.0529) (0.0572) (0.0615) (0.0610) (0.0521) (0.0536) Log Likelihood Value L m, LR *** *** *** Noe: Model 1: L a0 a1vol 1 a2vixri, 1 ; Model 2: L a0 a1vol 1 a3vixr 1 PESS Dummy 1 ; For all i=j, s, m, y, o be proxies as counries ETF, j=ewj(japan), s=ews(singapore), m=ewm(malaysia), y=ewy(souh Korea), =EWT(Taiwan). LR = 2 (L R L U ) ~ χ2(m), L R = Model 1 L U = Model 2, m=1 ***, **, and * indicae saisical significance a he 1%, 5%, and.10% levels. L y,

17 Invesor Senimen and ETF Liquidiy Empirical resuls analysis Wih consideraion of each counry s ETF and individual ime effec, Table 3 liss he regression resuls of Model 1 and Model The regression resuls of Model 1 In Model 1, he previous period's rading volume ( ( i Vol i, ) and invesor senimen VIXR, ) was used o observe he effec of each counry's ETF liquidiy, ( ), and he Model s coefficiens of variaion were adoped o deec if each counry s ETF liquidiy possesses volailiy-clusering. From he empirical resuls from Model 1, we can see ha hese five counries' daily rading volume and liquidiy changes show significan posiive resuls. Alhough he coefficien is very small, i also shows an increase (decrease) of he previous period's rading volume which will make he liquidiy increase (decrease). In he coefficiens of invesor senimen ( VIXR, ), each counry's numerical values are a he 10% significance level and above, all showing negaive resuls, which is consisen wih our general undersanding. When he financial marke is full of uncerainy and he change of invesor senimen volailiy is large, i will affec he liquidiy of he invesmen subjec maer; i.e. when he invesor senimen volailiy is large, he ETF subjec maer liquidiy will deeriorae, in which Japan is , Singapore of , Malaysia of , Souh Korea of , Taiwan of From each counry's empirical values, we can find ha when here is a change in invesor senimen hen volailiy becomes larger. The ETF liquidiy of Malaysia and Souh Korea is worse han he oher hree counries. I can be inferred from his ha hese wo counries' financial markes have a much bigger delayed reacion o messages when compared o he oher hree counries; he liquidiy is easily affeced by inernaional siuaions and invesor senimen. In erms of he condiional variance equaions in Model 1, he esimaed coefficiens for each counry s ETF ARCH effec (A) and GARCH effec (B) are: Japan: and , Singapore: and , Malaysia: and , Souh Korea: and , Taiwan: and respecively. A he 1% significance level, only he coefficiens of he ARCH effec show significan resuls. However, a he 5% significance level and above, he coefficiens of Malaysia and Souh Korea, he GARCH effec show significan resuls, and hese wo counries' esimaed coefficiens are non-negaive real numbers, meeing he posiive defined condiion assumpion. In addiion, he volailiy-clusering esimaed coefficiens (A+ B) namely Malaysia , Souh Korea , are boh less han 1; hese also mee he GARCH model's condiion for sabiliy. Therefore, i shows ha in Malaysia and Souh Korea ETF liquidiy hey exhibi he liquidiy-volailiy-clusering phenomenon, namely Malaysia and i L i,

18 106 Yung-Ching Tseng and Wo-Chiang Lee Souh Korea ETF liquidiy has a significan GARCH effec. From hese resuls i seems we can infer ha if he financial marke's ETF produc developmen level is more maure in counries such as Japan, Singapore, Taiwan, hen he GARCH effec will be less significan, i.e. he counry wih a more maure financial marke's ETF produc developmen level can reac o he financial marke informaion quickly and compleely The regression resuls of Model 2 In Model 2, we added he cross-muliplying erm VIXR PESS Dummy ) as ( he pessimisic dummy variable for invesor senimen under he framework foundaion of Model 1. When invesor senimen ( VIXR, ) flucuaion is over a sandard deviaion (7.01%), we define he dummy variable pessimisic senimen as 1, oherwise i would be 0. This sudy uses he seing of cross-muliplying erms for he pessimisic dummy variable and invesor senimen o magnify he effec of invesor senimen on liquidiy, in order o observe he degree of influence of he pessimisic marke-invesmen amosphere on each counry s ETF liquidiy. Through Table 3, we can see ha he likelihood esimaes for Model 2 are all larger han model 1, so Model 2 is a beer fi han model 1. The resuls from Model 2 show hese five counries' ETF daily rading volume and liquidiy changes are consisen wih he model showing significan posiive resuls, is coefficien is very small. In he pessimisic senimen period, he increase (decrease) of he previous period's rading volume will sill make he liquidiy increase (decrease). In he coefficiens of invesor senimen ( VIXR, ), each counry's numerical values are a he 5% significance level and above, all showing negaive resuls. The variables in he added cross-muliplying erms of he pessimisic dummy variable and invesor senimen, we find ha he coefficiens of he variables all show negaive resuls. This is consisen wih our general undersanding. When he financial marke is full of uncerainy, he volailiy in invesor senimen change will be bigger; also, he effec on he liquidiy of invesmen subjec maer will be deeper. I is worhwhile o noe ha he variables in he cross-muliplying erm, in which Japan is , Singapore of , Taiwan of , all show he 1% significance level. However, Malaysia and Souh Korea negaive coefficiens do no show a significan level. This shows ha he increasing uncerainy of financial marke will cause more inense invesor senimen volailiy. Especially, when he marke is facing a pull-up panic index. ETF liquidiy in Japan, Singapore and Taiwan will quickly reac, showing a negaive correlaion, i.e. in he panic period, hese hree counries' ETF liquidiy will fall significanly, while Malaysia and Souh Korea will no have significanly increased change of liquidiy. The i i

19 Invesor Senimen and ETF Liquidiy resuls of he empirical model show ha he liquidiy difference of he various counries during he panic period is easily affeced by he characerisics of he world's financial markes, such as he differences in markup-markdown resricions and shor sale consrains; herefore, i shows hese characerisics are inconsisen. In erms of he condiional variance equaions in Model 2, he esimaed coefficiens for each counry s ETF ARCH effec (A)and GARCH effec (B)are Japan and , Singapore and , Malaysia and , Souh Korea and , Taiwan and respecively. I is consisen wih Model 1. A he 1% significance level, only he coefficiens of he ARCH effec show significan resuls. Similarly, a he 5% significance level and above, he coefficiens of Malaysia and Souh Korea GARCH effec show significan resuls, and hese wo counries' esimaed coefficiens are non-negaive real numbers, meeing he posiive defined condiion assumpion. In addiion, he volailiy-clusering esimaed coefficiens (A+B), are namely Malaysia , Souh Korea , boh are less han 1; hey also mee he GARCH model's condiion for sabiliy. I shows ha in he panic period, Malaysia and Souh Korea ETF liquidiy sill has he liquidiy-volailiy-clusering phenomenon. To summarize he above resuls, rading volume and invesor senimen has a significan influence on he sample counries ETF liquidiy. When rading volume increases (decreases), he liquidiy of he subjec maer also increases (decreases), when invesor senimen volailiy ( VIXR, ) increases (decreases), he liquidiy shows a worse (beer) performance. I shows ha he invesor senimen does affec he ETF liquidiy; especially in he panic period In Malaysia and Souh Korea ETF, here is no significan evidence o show a srong relaionship beween invesor senimens and will be inensely performed on poor liquidiy. In addiion, we found ha wheher during he panic period or no, Malaysia and Souh Korea ETF liquidiy has a liquidiy-volailiy-clusering phenomenon, while his phenomenon is no significan in counries wih more maure financial marke ETF produc developmen level, such as Japan, Singapore, and Taiwan. i 5 Conclusion In he pas, lieraure invesigaed invesor senimen and ransacion behavior, analysis ino he correlaion beween liquidiy and reurns volailiy, as well as he relaionship beween invesor senimen and reurns. However, despie all his research, he relaionship beween invesor senimen and liquidiy was rarely discussed. The effec of invesor Senimen on each counry s ETF liquidiy was explored in his sudy.

20 108 Yung-Ching Tseng and Wo-Chiang Lee Through rading daa abou each counry's ETF financial producs, liquidiy models were esablished o represen capial marke liquidiy for various counries in order o analyze and research he changes in invesor senimen and liquidiy. In addiion, a dummy variable was added in he empirical model o reflec he panic period in which he liquidiy was observed. The sudy period was from Jan 31, 2005 o Jan 30, The observaion period was 10 years, and he sample period covered he financial sunami period. This helped us o deec he effec of he crisis on liquidiy. In addiion, he ETFs of five main counries in he Asia-Pacific region were used in he sudy as research samples. These included Japan, Souh Korea, Taiwan, Malaysia and Singapore. Resuls from he empirical daa indicaed ha rading volume and invesor senimen have a significan effec on he liquidiy of he ETFs in hese counries. The increase (decrease) in he previous period's rading volume would also make he ETF liquidiy increase (decrease). In he panic period, here is no significan evidence ha he ETFs of Malaysia and Souh Korea invesor senimen would be refleced in he liquidiy performance. We reviewed hisorical daa and found ha liquidiy exhibis Volailiy-clusering characerisics, ha is, in a specific period, liquidiy has beer or worse volailiy-clusering, so we adoped he GARCH model o capure i. The empirical resuls show ha he overall rading volume of ETFs show a significan correlaion wih invesor senimen. However, during he panic period, he resuls showed much more significan differences. We believe ha his is caused by he differences in financial environmens, sysems or invesor senimens wihin individual counries. Such resuls are also confirmed by our resuls. In paricular, we found ha wheher i is in he panic period or no, he liquidiy of he ETFs in Malaysia and Souh Korea have a significan Liquidiy-volailiy-clusering phenomenon. However, his phenomenon in he counries wih ETF producs wih a higher developmen level in financial markes, such as in Japan, Singapore, Taiwan, hen i becomes insignifican. For he inconsisencies in marke liquidiy, i was inferred in his paper ha i is caused by he financial environmen, rading sysem changes, mauriy of he developmen of ETF financial producs and invesor's rading resricions in a counry. For example, in he financial sunami period, Taiwan implemened a comprehensive shrinkage limi on shor sales rading, and his kind of resricion will have a significan effec on liquidiy. Wih he empirical resuls in his paper, we indeed confirmed ha invesor senimen has a significan effec on he liquidiy of ETF. The financial environmenal differences among differen counries include rading sysems, he mauriy of ETF financial commodiy developmenal level, and he specific supporing policies implemened by governmens when invesors face specific major marke messages, such as he resricions on shor sales. However, we did carry ou an in-deph discussion abou wha he relaed effecs beween he said

21 Invesor Senimen and ETF Liquidiy differences and liquidiy are. We also recommend ha researchers should analyze and research ino some of he oher characerisics ha impac liquidiy such as price limis and rading volume resricions in he fuure. References [1] M. Baker and J. Wurgler, Invesor senimen and he cross-secion of sock reurns, Journal of Finance, 61, (2006), [2] M.J. Barclay, Bid ask spreads and he avoidance of odd-eighh quoes on he Nasdaq: an examinaion of exchange lisings, Journal of Financial Economics, 45, (1997), [3] D. Basu, C.H. Hung, R. Oomen, and A. Tremme, When o pick he losers: Do senimen indicaors improve dynamic asse allocaion?. SSRN working paper, (2006). [4] H. Berkman and N.H. Nguyen, Domesic liquidiy and cross-lising in he Unied Saes, Journal of Banking and Finance, 34, (2010), [5] T. Bollerslev, Generalized Auoregressive Condiional Heeroskedasiciy, Journal of Economerics, 31, (1986), [6] B. Boscaljon and J. Clark, Do Large Shocks in VIX Signal a Fligh-o-Safey in he Gold Marke?, Journal of Applied Finance, 2, (2013), [7] M. Brien-Jones and A. Neuberger,Opion prices, implied price processes, and sochasic volailiy, Journal of Finance, 55, (2000), [8] G. Brown and M. Cliff, Invesor senimen and he near-erm sock marke, Journal of Empirical Finance, 11, (2004), [9] F. Chau, R. Deesomsak, and Marco C.K. Lau, Invesor senimen and feedback rading: Evidence from he exchange-raded fund markes, Inernaional Review of Financial Analysis, 20, (2011), [10] J. Chiu, H. Chung, K.Y. Ho, and George H.K. Wang, Funding liquidiy and equiy liquidiy in he subprime crisis period: Evidence from he ETF marke, Journal of Banking & Finance, 36, (2012), [11] J. Chiu, H. Chung, and K.Y. Ho, Fear Senimen, Liquidiy, and Trading Behavior: Evidence from he Index ETF Marke, Review of Pacific Basin Financial Markes and Policies, 17(3), (2014), [12] T. Chordia, A. Sarkar, and A. Subrahmanyam, An empirical analysis of sock and bond marke liquidiy, Review of Financial Sudies, 18, (2005), [13] S.L. Chung, C.H. Hung, and Chung-Ying Yeh, When does invesor senimen predic sock reurns?, Journal of Empirical Finance, 19, (2012), [14] T.E. Copeland and D. Gala Informaion effecs on he bid ask spread, Journal of Finance, 38, (1983), [15] J.C. Cox and M. Rubinsein, Opions Markes, Prenice Hall, New Jersey, 1985.