Invesor Herds in he Taiwanese Sock Marke Rıza Demirer *, Chun-Da Chen ** and Ali M. Kuan *** * Souhern Illinois Universiy Edwardsville ** Tennessee Sae Universiy *** Souhern Illinois Universiy Edwardsville, The William Davidson Insiue, Universiy of Michigan Business School, and The Emerging Markes Group, Cass Business School, London. January 2008 Corresponding auhor. E-mail: rdemire@siue.edu; Tel: 618 650 2939; Fax: 1 618 650-3047. Please do no quoe wihou permission.
Invesor Herds in he Taiwanese Sock Marke Asrac This paper compares four differen esing mehodologies designed o es he exisence of invesor herds. We use firm level daa on 689 firms raded in he Taiwan Sock Exchange, classified ino 18 differen secors. We find ha he cross secional sandard deviaion (CSSD) ased esing mehodology, which imposes a linear relaion eween reurn dispersions and marke reurn, fail o correcly es invesor herds and yields no significan evidence of herding among Taiwanese invesors. However, he non-linear model proposed y Chang e al. (2000) and he sae space ased models of Hwang and Salmon (2004) lead o consisen resuls indicaing srong evidence of herd formaion in all secors wih Elecronics displaying he greaes impac. The fac ha Elecronics happen o e he mos volaile secor among all and he mos heavily invesed one y foreign insiuional invesors is consisen wih he lieraure suggesing ha he rades of large insiuional invesors desailize marke prices and increase he volailiy of he marke due o posiive feedack and herding among invesors. We also find ha he herding effec is more prominen during down movemens of he marke, indicaing ha i is he prospec of a loss which riggers herd ehavior. JEL Classificaion Code: G14, G15 Keywords: Herd ehavior, Equiy reurn dispersion, Taiwan Sock Exchange. 2
1. Inroducion A major funcion, among ohers, of securiy markes is he process of price discovery. Consan research and profi maximizing ehavior of uyers and sellers in hese markes make sure ha every piece of informaion is refleced in he values of raded securiies, hence affecing he values of corporaions ha issue hem. Informaional efficiency of a securiy marke herefore is no only imporan from an academic poin of view, as many heories are ased on he concep of efficiency, u also crucial for invesors and policy makers who rely on he informaion ha hese markes signal aou he healh of corporaions as well as he overall economy. Clearly, he acquisiion and ransparency of informaion as well as how his informaion is processed y invesors is a major concern for academics as financial heories ha we have uil depend on he assumpion of invesor raionaliy and raional processing of informaion. Formaion of invesor herds has een proposed as an alernaive explanaion of how invesors process informaion and make invesmen choices. Herd ehavior is simply defined as a sraegy ased on mimicking oher invesors acions (Bikhchandani and Sharma, 2000). Prior sudies differ in heir explanaion o wha migh rigger such ehavior. Devenow and Welch (1996) use he argumens of invesor psychology where invesors feel a sense of securiy in following he crowd. Anoher view suggess ha he acions of more informed raders may reveal useful informaion which may no e accessile o individual invesors (Chari and Kehoe, 1999, Calvo and Mendoza, 1998, and Avery and Zemsky, 1998). Invesors perhaps canno idenify rue or false informaion rapidly especially if hey are in an unfamiliar marke. In such a siuaion, invesors migh follow oher invesors rades lindly suppressing heir own eliefs. Finally, a hird approach focuses on he principal-agen relaionship where fund managers migh wan o imiae ohers as a resul of he incenives provided y he compensaion scheme or in order o mainain heir repuaion (Scharfsein and Sein, 1990, Rajan, 1994, and Maug and Naik, 1996). 3
This sudy has wo main conriuions. Firs, we compare four differen mehodologies proposed in he lieraure ha are designed o es he exisence of invesor herds. The firs wo models we esimae are ased on reurn dispersions among individual firms; more specifically, cross secional sandard deviaions (CSSD) and cross secional asolue deviaions (CSAD) across a paricular secor. These mehodologies have een applied o differen markes in prior sudies y Chrisie and Huang (1995), Chang, Chen, and Khorana (2000) and Gleason, Lee and Mahur (2003), Gleason, Mahur and Peerson (2004) and Demirer and Kuan (2006). The las wo models are ased on a sae space model specificaion proposed y Hwang and Salmon (2004). These wo ses of models differ in he sense ha he firs wo focus on he cross-secional variailiy of reurns whereas he las wo focus on he cross-secional variailiy of facor sensiiviies. Our sudy, herefore, is he firs aemp o compare differen esing mehodologies on a large scale daa. The second conriuion of his sudy is o exend herding ess o he Taiwanese sock marke and provide insigh o his emerging marke. We apply he four esing mehodologies o firm level daa using a large numer of socks raded in he Taiwan Sock Exchange. As shown in Tale 1, domesic invesors, mosly individual, accoun for almos 80% of he invesmen amoun in his marke, however here has een increasing ineres y foreign invesors over he pas six years. Unlike invesmen russ, foreign invesors, and securiy dealers, mos individual invesors end o have less professional knowledge and canno access informaion accuraely and easily. This informaion asymmery may lead individuals o follow he acions of more informed insiuional and foreign invesors which may lead o herding formaion in his marke. For his purpose, i is especially ineresing o examine wheher herd formaion exiss in his emerging Taiwanese marke wih mosly domesic individual invesors. Our findings indicae ha he esing mehodology ased on CSSD as a measure of herding fail o idenify exisence of herds. This mehodology yields no significan evidence of herding among Taiwanese invesors. However, he non-linear model proposed y Chang e al. (2000) and he sae 4
space ased models of Hwang and Salmon (2004) lead o consisen resuls indicaing srong evidence of herd formaion. Therefore, regarding he comparison of esing mehodologies, we find ha CSSD ased mehodology, which assumes a linear relaion eween reurn dispersions and marke reurn, fail o correcly es invesor herds. Regarding es resuls using Taiwanese marke daa, we find srong evidence o herd formaion in all secors. Furher analysis of es resuls across secors indicaes ha herding has a igger impac paricularly on Elecronics secor. Considering he fac ha Elecronics happen o e he mos volaile secor among all and he mos heavily invesed one y foreign insiuional invesors, i is no surprising o find sronger evidence of invesor herds among invesors of securiies in his secor. This finding is also consisen wih he lieraure suggesing ha he rades of large insiuional invesors desailize marke prices and increase he volailiy of he marke due o posiive feedack and herding among invesors. Individual invesors in highly volaile markes are more likely o follow more informed invesors rades such as foreign invesors, and insiuional invesors. We also find ha he herding effec is more prominen during down movemens of he marke, indicaing ha i is he prospec of a loss which riggers herd ehavior. An ouline of he remainder of he paper is as follows. Secion 2 riefly summarizes he previous sudies on ess of invesor herds. Secion 3 provides he deails of differen esing mehodologies employed and daa descripion. Secion 4 presens empirical resuls and a comparison of he findings from he reurn dispersion ased models and sae space models. Finally, Secion 5 concludes he paper and discusses furher research. 2. Previous Sudies on Invesor Herds Differen mehodologies have een suggesed in he lieraure o es he exisence of invesor herds. Tesing mehodologies ased on reurn dispersions among a group of securiies focus on 5
cross secional sandard (or asolue) deviaions of reurns. Prior sudies include Chrisie and Huang (1995) on U.S. equiies, Chang, Cheng and Khorana (2000) on inernaional equiies, Gleason, Lee and Mahur (2003) on commodiy fuures raded on European exchanges, Gleason, Mahur and Peerson (2004) on Exchange Traded Funds, and Demirer and Kuan (2006) on Chinese socks. In general, hese sudies provide resuls in favor of he raional asse pricing heories and conclude ha herding is no an imporan facor in deermining securiy reurns during periods of marke sress. A differen esing mehodology ased on cross secional variailiy of facor sensiiviies, insead of reurns, is suggesed y Hwang and Salmon (2004). Their analysis of daily sock reurns in he Souh Korean marke provides suppor for herd formaion in his marke. Finally, Weiner and Green (2004) employed oh parameric and non-parameric mehodologies and found lile evidence of herding in heaing oil and crude-oil fuures. 3. Daa and Mehodology 3.1 Daa The daa se used in his sudy conains daily reurns for 689 Taiwanese socks raded on he Taiwan Sock Exchange over he January 1995 Decemer 2006 period. Daa are oained from he Taiwan Sock Exchange Corporaion (TSEC). Herding ess in he lieraure are ased on he suggesion ha a group is more likely o herd if i is sufficienly homogeneous, i.e. each memer faces a similar decision prolem, and each memer can oserve he rades of oher memers in he group (Bikhchandani and Sharma, 2000). Prior sudies have, herefore, applied he ess on groups of socks caegorized on he asis of indusry classificaion (e.g. Chrisie and Huang, 1995), exchange or counry assignmen (e.g. Gleason, Mahur and Peerson, 2004 and Chang, Chen, and Khorana, 2000), of course, afer he impac of fundamenals has een facored ou. Therefore, we assign each of he 689 firms o one of eigheen secor groups including Cemen, Food, Plasics, 6
Texile, Elecrical Appliances, Wire & Cale, Chemicals, Glass & Ceramics, Pulp & Paper, Seel, Ruer, Auomoile, Elecronics, Consrucion, Transporaion, Tourism, Banking & Securiies, and Reailing. We hen calculae porfolio reurns ased on an equally weighed porfolio of all firms in each secor classificaion. 3.2 Mehodology We employ four differen esing mehodologies; wo reurn dispersion ased models and wo sae space models. This secion summarizes he mehodology for each esing mehodology. 3.2.1 Reurn Dispersion Models The firs wo esing mehodologies employed are ased on cross secional sandard deviaions (CSSD) and cross-secional asolue sandard deviaions (CSAD) among individual firm reurns wihin a paricular group of securiies. Chrisie and Huang (1995) use CSSD as a measure of he average proximiy of individual asse reurns o he realized marke average in order o es herd ehavior. Chang e al. (2000) use CSAD in a non-linear regression specificaion in order o examine he relaion eween he level of equiy reurn dispersions and he overall marke reurn. The firs mehodology employed in his sudy is ased on reurn dispersions as measured y CSSD. This mehodology has een used y Chrisie and Huang (1995), Chang, Cheng, and Khorana (2000), Gleason, Lee and Mahur (2003), Gleason, Mahur and Peerson (2004), and Demirer and Kuan (2006) as explained in Secion 2. Cross-secional sandard deviaions (CSSD), used as a measure of reurn dispersion, is formulaed as follows: N ( ri, rp, ) 2 i= 1 CSSD = (1) N 1 where n is he numer of firms in he aggregae marke porfolio, r i, is he oserved sock reurn on firm i for day and r p, is he cross-secional average of he n reurns in he marke 7
porfolio for day. This measure can e regarded as a proxy o individual securiy reurn dispersion around he marke average. The main idea in his mehodology is ased on he argumen ha he presence of herd ehavior would lead securiy reurns no o deviae far from he overall marke reurn. The raionale ehind his argumen is he assumpion ha individuals suppress heir own eliefs and make invesmen decisions ased solely on he collecive acions of he marke. On he oher hand, raional asse pricing models offer a conflicing predicion suggesing ha dispersions will increase wih he asolue value of marke reurn since each asse differs in is sensiiviy o he marke reurn. This mehodology also suggess ha he presence of herd ehavior is mos likely o occur during periods of exreme marke movemens, as hey would mos likely end o go wih he marke consensus during such periods. Hence, we examine he ehavior of he dispersion measure in (1) during periods of marke sress and esimae he following linear regression model: CSSD = α + β D + β D + ε (2) D L U U where L D = 1, if he reurn on he aggregae marke porfolio on day lies in he lower ail of he reurn disriuion; 0 oherwise, and U D = 1, if he reurn on he aggregae marke porfolio on day lies in he upper ail of he reurn disriuion; 0 oherwise. Alhough somewha arirary, in he lieraure, an exreme marke reurn is defined as one ha lies in he one (and five) percen lower or upper ail of he reurn disriuion. The dummies in equaion (2) aim o capure differences in reurn dispersions during periods of exreme marke movemens. As herd formaion indicaes conformiy wih marke consensus, he presence of negaive and saisically significan β D (for down markes) and β U (for up markes) coefficiens would indicae herd formaion y marke paricipans. The second mehodology employed in his paper is suggesed y Chang e al. (2000) and uses he 8
cross-secional asolue deviaion of reurns (CSAD) as a measure of reurn dispersion. CSAD is expressed as CSAD = 1 N N i= 1 ri, rm, (3) Chang e al. (2000) challenge he CAPM assumpion ha reurn dispersions are an increasing funcion of he marke reurn and ha his relaion is linear. They sugges ha during periods of marke sress, one would expec he relaion eween reurn dispersion and marke reurn o e non-linearly increasing or even decreasing. Therefore, hey propose a esing mehodology ased on a general quadraic relaionship eween CSAD and r m, of he form: CSAD = α + γ + 2 1 rm. + γ 2rm. ε (4) According o his mehodology, herding would e evidenced y a lower or less han proporional increase in he cross-secional asolue deviaion (CSAD) during periods of exreme marke movemens. As a resul, if herding is presen, hen he non-linear coefficien, γ 2 will e negaive and saisically significan; oherwise a saisically posiive γ 2 would indicae no evidence of herding. 3.2.2 Sae Space Models The nex wo esing mehodologies we employ in his sudy are suggesed y Hwang and Salmon (2004). Raher han reurns, Hwang and Salmon (2004) focus on he cross-secional variailiy of facor sensiiviies. Considering a one facor model wih he facor eing he marke reurn, hey formulae a herding measure ased on he relaive dispersion of he eas for all asses in he marke. Nex, we riefly explain hese wo mehodologies. Consider he following CAPM in equilirium, E r ( ri i ) = β E ( r ) (5) where r and r are he excess reurns on asse i and he marke a ime, respecively, i 9
β is he sysemaic risk measure, and E ( ) is condiional expecaion a ime. When i herding ehavior is presen, invesors disregard he equilirium relaionship of Equaion (5) and rade in such a way ha maches individual asse reurns wih ha of he marke. When ha happens, he β and expeced rae of reurn presens a ias in a way ha would reflec his maching of individual asse reurns wih ha of he marke. So, considering CAPM again, when herding ehavior is presen, real β coefficien oeys he following relaion which replaces equaion (5): E ( r E ( r i ) ) β = β h ( β 1), (6) = i i i where E r ) and are he marke s iased shor run condiional expecaion on he excess ( i β i reurns of asse i and is ea a ime, and h is a laen herding parameer ha changes over ime, h 1, and condiional on marke fundamenals. In general, when ( 0 < h < 1), one could argue ha some degree of herding exiss in he marke deermined y he magniude of. h Since he form of herding we discuss represens marke-wide ehavior and Equaion (6) is assumed o hold for all asses in he marke, we should calculae he level of herding using all asses in he marke raher han a single asse, herey removing he effecs of idiosyncraic β i movemens in any individual. The sandard deviaion of can e formulaed as β i Sd ( β c i ) = Ec (( β i h ( β i 1) 1) 2 ) = E (( β c i 2 1) )(1 h ) = Sd β )(1 h ), (7) c ( i where E C ( ) represens he cross-secional expecaion. Taking he logarihm of Equaion (7) and assuming ha Sd c ( β i ) can e ime-varying, we rewrie Equaion (7)as 10
log[ Sd c ( β )] = μ + υ i m 2 where μ m = E[log[ Sd c ( β i )]] and υ ~ iid(0, σ ). Sae Space Model 1 hen can e esimaed as c i m mυ log[ Sd ( β )] = μ + H + υ, (8) H = φ H 1 + η, (9) m 2 where H = log( 1 h ) and η ~ iid(0, σ ). Equaions (8) and (9) are he sandard mη sae-space model wih Kalman filer esimaion. In his mehodology, we only focus on he dynamic paern of movemens in he laen sae variale, H, he sae equaion. When 2 σ m = 0, here is no herding, i.e., H = 0 for all. We allow herding,, o evolve over η H ime and follow a dynamic process. A significan value of 2 σ mη can e inerpreed as he exisence of herding and a significan φ suppors his paricular auoregressive srucure. An alernaive model can e formulaed when we add o Equaion (8) marke volailiy, log σ, and reurn, r, as independen variales. This leads o Sae Space Model 2 formulaed as log[ Sd ( β )] r + υ. (10) c i = μm + H + cm 1 logσ + cm2 4. Empirical resuls 4.1 Descripive saisics Tale 2 provides summary saisics for average daily log reurns, reurn dispersions, and he average numer of firms used o compue hese saisics for each secor. Since he numer of socks in a secor does no say consan over ime, we repor he average numer of firms over he sample period in he second column of Tale 2. Examining Tale 2, we oserve ha he average daily reurns for all secors are posiive while elecronics and auomoile secors have he highes average daily reurns. In erms of volailiy, 11
consrucion and elecronics secors are ranked as he mos volaile secors. Over he pas few years, he elecronics secor has shown rapid growh and socks in his secor have een among he mos acively raded socks in Taiwan. According o he annual repor of Taiwan Sock Exchange Corporaion, he rading volume and value of securiies in his secor accoun for 60% and 70% of he oal volume and value of he Taiwan sock marke respecively. Heavy rading volume and invesor ineres in his secor may e a reason for he high reurn volailiy ha we oserve in he elecronics secor. Considering he high volailiy, one migh expec herding ehavior o e more likely in his secor relaive o oher secors. Due o heavy rading volume and high uncerainy in he ehavior of socks, invesors in his secor may find i raional o follow insiuional invesors rading sraegies as well as each oher. Furhermore, foreign shareholding percenage of elecronic secor is aou 35% which may furher affec invesor ehavior in erms of heir rades, making i more likely o oserve herd ehavior. The foreign shareholding percenage of socks in he anking and securiies secor is aou 25%, ranked second place in he sock marke; however, we oserve ha he average and he sandard deviaion of daily sock reurns in his secor are no as high as oher secors. Moreover, during he sample period, he governmen of Taiwan has acively execued he so-called Bank Merger Policy. Having his in mind, one migh argue ha invesors would no e inclined o follow ohers sraegies promply during he peak of a ank merger. Due o hese wo facors, one may expec herd ehavior in he anking and securiies secor o e less likely. Compared o elecronics (Hi-Tech) secor, he remaining more radiional secors end o have a lower foreign shareholding percenage. However, despie eing a radiional secor, we oserve a higher average reurn and sandard deviaion for consrucion secor. One reason for his oservaion may e due o he recession in he housing marke during he sample period. In order o simulae growh in he housing marke, he governmen of Taiwan has execued differen policies frequenly, governmenal inervenion policy migh have led o he higher reurns and 12
corresponding higher volailiy in his secor. Panel B in Tale 2 repors summary saisics for daily cross secional sandard deviaions wihin each secor. Consisen wih he findings from Panel A, we oserve he highes cross secional volailiy in Elecronics followed y Consrucion. 4.2 Resuls of reurn dispersion models Tale 3 presens esimaion resuls for he CSSD ased model in Equaion 2. Given he significan variaion in dispersions and srong correlaion, all esimaions are done using he Newey-Wes heeroskedasiciy and auocorrelaion consisen sandard errors 1. We use he Taiwan Sock Exchange Composie Index o represen he marke and use he upper and lower one and five perceniles of he marke reurn o represen periods of marke sress. For a majoriy of he secors analyzed, we do no find any evidence in favor of herd formaion during periods of large marke swings. The regressions yield saisically significan and posiive β esimaes indicaing ha equiy reurn dispersions increase during periods of large price changes as prediced y CAPM. The only excepion o his is Elecronics where we oserve significanly lower reurn dispersions when marke is in he upper or lower one percenile, indicaing herd formaion during exreme moves of he marke index. Laer in his secion, we will provide explanaions as o why i is more likely o oserve herd ehavior in Elecronics. However, for he momen, keep in mind ha Elecronics happen o e he mos volaile secor among all and he mos heavily invesed one y foreign invesors. Tale 4 presens esimaions resuls for he non-linear CSAD ased model in Equaion 4. Following Chang e al. (2000), we run hree separae regressions for each secor: one using he whole sample, and wo resricing he daa o up (or down) movemens of he marke index. Running separae models in his manner allows us o examine wheher here is any asymmeric effec of herd ehavior. Our findings wih he non-linear model lead o compleely differen resuls 1 We also esimaed he models using GARCH models; he resuls were qualiaively he same. 13
han he firs mehodology. Examining Tale 4, we find evidence o herd formaion in all secors excep for Tourism and Auomoile. The regressions yield saisically significan and negaive γ 2 esimaes indicaing a non-linear and decreasing relaion eween equiy reurn dispersions and he marke reurn. However, when we examine regression resuls run wih daa resriced o up and down markes separaely, we oserve ha herding effec is mosly prominen during marke losses. I is more likely o oserve herd formaion during periods of marke losses. Furhermore, when we compare he asolue values of γ 2 esimaes, we see ha reurn dispersions during downside movemens of he marke are much lower han hose for upside movemens. Therefore, one can conclude ha i is he prospec of losing which pushes invesors o follow ohers rades and display herd ehavior. However, such ehavior is no he case during up markes indicaing invesor confidence when invesors are confiden in heir decisions. In he nex susecion, we summarize our findings from he sae space models esimaed. 4.3 Resuls of sae space models Tale 5 presens esimaion resuls for Sae Space Model 1. The resuls indicae srong evidence of herding hrough H. We oserve ha H is highly persisen wih large and significan values of φˆm. More imporanly, he esimaes for σ m η are highly significan providing suppor for herd ehavior. In paricular, Elecronics, Texile, and Banking & Securiies display higher H esimaes implying ha invesors may e following insiuional invesors rading ehavior. The findings make more sense considering he fac ha Elecronic and Banking & Securiies secors are wo of he mos volaile secors and he wo favorie rading arges for foreign insiuional invesors. In a separae u relaed sudy Gaaix e al. (2005) presen a model in which he rades of large insiuional invesors desailize marke prices and increase he volailiy of he marke due o posiive feedack and herding among invesors. Unlike invesmen russ, foreign invesors, and 14
securiy dealers (informed raders), mos individuals have less professional knowledge and canno access informaion accuraely as easily. This informaion asymmery would furher lead individuals o perform momenum sraegies which migh lead o herding effecs. Similarly, Luo (2003) find ha muual fund invesors creae excess volailiy as fund flows lead o higher susequen marke volailiy. In anoher relaed sudy, Dennis and Srickland (2002) find ha socks ha move he mos during large moves of he marke index happen o e hose ha have relaively larger insiuional holdings. They sugges ha his finding is due o herd formaion among muual funds and pension plan sponsors. Oher secors we analyze also have significan and highly persisen H esimaes and exhii herding ehavior; however, Securiies secors. H values are lower han hose for Elecronics and Banking & Tale 6 presens esimaion resuls for Sae Space Model 2. Our findings are similar o hose we oserve wih he firs model resuls. Once again, aking ino accoun he level of marke volailiy and reurn his ime, we find srong evidence of herding hrough H as he sandard deviaion of η is significanly differen from zero and H is highly persisen wih he φˆ esimae m eing significan. Please noe he difference eween he wo sae space models. The second model conains wo marke variales in he measuremen equaion allowing us o analyze he degree of herding given he sae of he marke. Furhermore, we focus on he coefficien of marke volailiy and marke reurns. The findings for Plasics, Elecrical App., Chemicals, Pulp, Seel, Auomoile, Tourism, and Reailing secors indicae ha Sd c ( β i ) values decrease wih marke volailiy, since log σ values have significan and negaive coefficiens. These resuls are consisen wih previous sudies which sugges ha herding is more likely o occur during periods of marke sress. 15
However, he resuls show ha Sd c ( β i ) values increase as marke volailiy rises for Cemen, Food, Texile, Wire & Cale, Glass, Elecronics, and Banking & Securiies since log σ have significan and posiive coefficiens. I is ineresing o noe ha hese secors wih posiive log σ coefficien esimaes have higher daily rading aciviies and higher foreign shareholding percenage. Thus, we elieve ha invesors in hese secors usually exhii herding ehavior regardless of marke siuaion. Furhermore, comparing he esimaes for φˆm across secors we noice ha he effec of herd ehavior in Elecronics is greaer compared o financial and radiional secors. 5. Conclusions Herd formaion has een proposed as an alernaive explanaion of how invesors make invesmen choices. The exisence of invesor herds is a major concern for academics as financial heories ha we have uil depend on he assumpion of invesor raionaliy as well as raional processing of informaion. Such ehavior also presens a concern o policymakers as such ehavior migh aggravae volailiy of reurns and hence desailize financial markes, especially in crisis condiions. In his paper, we compare four differen mehodologies o es herd formaion in Taiwan Sock Exchange. We use firm level daa on 689 firms and classify hem ino 18 differen secors. We hen esimae four differen models o es he exisence of invesor herds in hese markes. The firs wo models we esimae are ased on reurn dispersions among individual firms; more specifically, cross secional sandard deviaions (CSSD) and cross secional asolue deviaions (CSAD) across a paricular secor. The las wo models are ased on a sae space model specificaion. These wo ses of models differ in he sense ha he firs wo focus on he cross-secional variailiy of reurns whereas he las wo focus on he cross-secional variailiy of facor sensiiviies. Our sudy is he firs aemp o compare differen esing mehodologies ha 16
are designed o es he exisence of invesor herds. We apply hese ess o firm level daa using a large numer of socks raded in he Taiwan Sock Exchange. Our findings indicae ha Chrisie and Huang (1995) mehodology ased on CSSD as a measure of herding fail o idenify exisence of herds. This mehodology yields no significan evidence of herding among Taiwanese invesors. However, he non-linear model proposed y Chang e al. (2000) and he sae space ased models of Hwang and Salmon (2004) lead o consisen resuls indicaing srong evidence of herd formaion. Therefore, regarding he comparison of esing mehodologies, we find ha CSSD ased mehodology, which assumes a linear relaion eween reurn dispersions and marke reurn, fail o correcly es invesor herds. One of he implicaions of he resuls is ha fuure sudies should capure such non-linear relaionships in he daa. Regarding es resuls using Taiwanese marke daa, we find srong evidence o herd formaion in all secors. Comparison of es resuls across secors, we see ha herding has a igger impac paricularly on Elecronics secor. Considering he fac ha Elecronics happen o e he mos volaile secor among all and he mos heavily invesed one y foreign insiuional invesors, i is no surprising o find sronger evidence of invesor herds among invesors of securiies in his secor. This finding is also consisen wih he lieraure suggesing ha he rades of large insiuional invesors desailize marke prices and increase he volailiy of he marke due o posiive feedack and herding among invesors. Individual invesors in highly volaile markes are more likely o follow more informed invesors rades such as foreign invesors, and insiuional invesors. We also find ha he herding effec is more prominen during down movemens of he marke, indicaing ha i is he prospec of a loss which riggers herd ehavior. Our sudy has several limiaions. We rely on only a sock marke o provide inferences aou he rousness of differen esing mehodologies on herding ehavior. In addiion, we use daa on an emerging marke. The resuls could e sensiive o using daa from advanced markes. Hence, our 17
resuls are preliminary u could e used as a yardsick for fuure sudies. Furher evidence from several oh emerging and advanced markes is necessary o generalize he resuls of he paper. 18
References Avery, C., Zemsky, P., 1998. Mulidimensional Uncerainy and Herd Behavior in Financial Markes. American Economic Review 88, 724-748. Bikhchandani, S., Sharma, S., 2000. Herd Behavior in Financial Markes: A Review. IMF Working Paper WP/00/48 (Washingon: Inernaional Moneary Fund). Calvo, G., Mendoza, E., 1998. Raional Herd Behavior and Gloalizaion of Securiies Markes. Mimeo, Universiy of Maryland. Chang, E. C., Cheng, J. W., Khorana, A., 2000. An Examinaion of Herd Behavior in Equiy Markes: An Inernaional Perspecive. Journal of Banking and Finance 24, No. 10, 1651-1699. Chari, V. V., Kehoe, P., 1999. Financial Crises as Herds. Mimeo, Federal Reserve Bank of Minneapolis. Chrisie, W.G., Huang, R. D., 1995. Following he Pied Piper: Do individual Reurns Herd around he Marke? Financial Analys Journal, July-Augus 1995, 31-37. Demirer, R. and Kuan, A. M., 2006, Does herding ehavior exis in Chinese sock marke? Journal of Inernaional Financial Markes, Insiuions and Money 16, 123-142. Dennis, P. and D. Srickland, 2002. Who links in volaile markes, individuals or insiuions?, Journal of Finance 51, 111-135. Devenow, A., Welch, I., 1996. Raional Herding in Financial Economics, European Economic Review 40, 603-615. Gaaix, X., P. Gopikerishman, V. Plerou and H. E. Sanley, 2005, Insiuional invesors and sock marke volailiy, Quarerly Journal of Economics, forhcoming. Gleason, K. C., Lee, C. I., Mahur, I., 2003. Herding Behavior in European Fuures Markes. Finance Leers 1, 5-8 Gleason, K. C., Mahur, I., Peerson, M. A., 2004. Analysis of Inraday Herding Behavior among he Secor ETFs. Journal of Empirical Finance 11, 681 694 Hirshleifer, D., Teoh, S. H., 2001. Herd ehavior and Cascading in Capial Markes: A Review and Synhesis. Working Paper. Hwang, S., and Salmon, M., 2004, Marke sress and herding, Journal of Empirical Finance 11, 585-616 19
Luo, D., 2003. Marke volailiy and muual funds cash flows. Working paper, Yale Inernaional Cener for Finance. Maug, E., Naik, N., 1996. Herding and Delegaed Porfolio Managemen. Mimeo, London Business School. Rajan, R. G., 1994. Why Credi Policies Flucuae: A Theory and Some Evidence. Quarerly Journal of Economics 436, 399-442. Scharfsein, D., Sein, J., 1990. Herd Behavior and Invesmen. American Economic Review 80, 465-479. Shiller, R. J., 1981. Do Sock Prices Move oo Much o e Jusified y Susequen Changes in Dividends? American Economic Review, 71, 421-436. Summers, L. H., 1986. Does he Sock Marke Raionally Reflec Fundamenal Values? Journal of Finance 41, 591-601. Weiner, R. J., and Green, M. A., 2004. Do irds of a feaher flock ogeher? Speculaor herding in dervaives markes. Working Paper George Washingon Universiy 20
Tale 1. Securiies Trading Value Percenage y Invesors Type (%) Year Domesic Domesic Juridical Foreign Juridical Individual Person Foreign Individual Person 1996 89.25 8.62 0.01 2.12 1997 90.73 7.55 0.01 1.71 1998 89.73 8.63 0.02 1.62 1999 88.23 9.36 0.01 2.40 2000 86.10 10.27 0.01 3.62 2001 84.41 9.69 0.01 5.89 2002 82.30 10.05 0.97 6.68 2003 77.84 11.51 1.24 9.41 2004 75.94 11.56 1.63 10.87 2005 68.84 13.29 2.41 15.46 2006 70.56 11.04 2.25 16.15 21
Tale 2. Summary Saisics: Average Daily Reurns and Cross-Secional Sandard Deviaions. Indusry # Firms # Os. Mean Sd. Dev. Min. Max. Panel A: Average Daily Reurns Cemen 8 3154 0.016% 2.457% -7.000% 7.000% Food 18 3154 0.026 2.404-7.440 7.000 Plasics 20 3154 0.028 2.735-7.000 7.000 Texiles 47 3154 0.014 2.773-8.090 14.070 Elecrical App 36 3154 0.040 2.506-7.600 10.880 Wire & Cale 14 3154 0.021 2.670-7.000 14.250 Chemicals 35 3154 0.035 2.476-7.000 9.980 Glass and Ceramics 7 3154 0.003 3.019-19.680 7.000 Pulp & Paper 7 3154 0.017 2.707-7.000 7.000 Seel 24 3154 0.043 2.827-7.000 14.980 Ruer 9 3154 0.044 2.723-7.000 16.720 Auomoile 5 3154 0.056 2.212-6.990 7.000 Elecronics 307 3154 0.056 3.106-38.240 277.000 Consrucion 34 3154 0.051 3.183-7.560 13.680 Transporaion 18 3154 0.049 2.641-7.550 7.000 Tourism 6 3154 0.039 2.486-7.000 7.000 Banking & Securiies 45 3154 0.019 2.426-7.770 11.470 Reailing 10 3154 0.040 2.433-7.000 7.000 Ohers 39 3154 0.049 2.498-7.200 11.650 Panel B: Cross-Secional Sandard Deviaions Cemen 1.558% 0.811% 0.106% 5.004% Foods 1.868 0.705 0.384 5.193 Plasics 1.895 0.725 0.104 4.883 Texile 2.119 0.608 0.278 4.462 Elecrical App 2.015 0.648 0.230 4.479 Wire and Cale 1.956 0.790 0.102 4.989 Chemicals 1.923 0.638 0.254 4.619 Glass and Ceramics 2.299 1.084 0.100 8.545 Pulp and Paper 1.690 0.815 0.068 5.042 Seel 1.917 0.736 0.108 5.358 Ruer 1.914 0.843 0.124 5.698 Auomoile 1.405 0.888 0.000 6.398 Elecronics 2.433 0.920 0.151 22.296 Consrucion 2.324 0.730 0.492 5.089 Transporaion 1.868 0.731 0.100 4.745 Tourism 1.655 0.843 0.036 5.795 Banking & Securiies 1.644 0.639 0.176 4.613 Reailing 1.845 0.778 0.108 5.108 Ohers 2.073 0.590 0.447 4.938 22
Tale 3. Regression Coefficiens for CSSD = α + β D + β D + ε D L U U (-raios in parenheses). Reurn Dispersions Marke reurn in he exreme upper/lower 1% of he reurn disriuion Marke reurn in he exreme upper/lower 5% of he reurn disriuion Indusry α β D β U α β D β U Cemen 1.553% 0.245% 0.294% ** 1.527% 0.353% *** 0.269% *** (1.532) (2.053) (5.057) (4.563) Food 1.857 0.699 *** 0.442 *** 1.824 0.541 *** 0.351 *** (4.812) (4.004) (8.979) (7.005) Plasics 1.893 0.168 0.026 1.857 0.498 *** 0.260 *** (1.019) (0.314) (7.580) (5.704) Texile 2.116 0.146 0.149 * 2.090 0.339 *** 0.242 *** (1.014) (1.840) (6.507) (6.433) Elecrical App. 2.009 0.243 * 0.304 *** 1.980 0.402 *** 0.285 *** (1.646) (3.449) (6.727) (6.954) Wire and Cale 1.955 0.089 0.016 1.928 0.322 *** 0.238 *** (0.512) (0.117) (4.530) (4.001) Chemicals 1.919 0.379 *** 0.043 1.890 0.476 *** 0.193 *** (2.649) (0.629) (8.739) (4.764) Glass and Ceramics 2.296 0.089 0.186 2.262 0.331 *** 0.403 *** (0.334) (0.899) (3.394) (4.313) Pulp and Paper 1.691-0.0002-0.073 1.670 0.307 *** 0.099 (-0.001) (-0.523) (4.245) (1.612) Seel 1.914 0.086 0.165 1.881 0.404 *** 0.313 *** (0.724) (1.488) (6.770) (5.814) Ruer 1.916 0.061-0.219 * 1.886 0.456 *** 0.113 * (0.277) (-1.661) (5.214) (1.902) Auomoile 1.395 0.392 * 0.569 *** 1.367 0.479 *** 0.285 *** (1.947) (3.423) (5.777) (3.920) Elecronics 2.330-0.389 *** -0.485 *** 2.319 0.069-0.012 (-3.111) (-5.340) (1.245) (-0.260) Consrucion 2.319 0.217 0.274 ** 2.299 0.257 *** 0.257 *** (1.424) (2.388) (4.441) (5.262) Transporaion 1.862 0.426 ** 0.197 * 1.828 0.514 *** 0.283 *** (2.541) (1.945) (8.813) (5.951) Tourism 1.641 0.838 *** 0.542 *** 1.609 0.564 *** 0.344 *** (3.574) (4.162) (6.480) (5.384) Banking & Securiies 1.639 0.320 ** 0.135 1.610 0.359 *** 0.326 *** (2.559) (1.090) (6.306) (7.047) Reailing 1.841 0.350 ** 0.100 1.811 0.448 *** 0.249 *** (2.497) (1.177) (7.214) (5.435) Ohers 2.067 0.348 *** 0.210 *** 2.041 0.418 *** 0.221 *** (3.019) (2.418) (9.346) (5.410) (***, **, and * denoe saisical significance a 1%, 5%, and 10%, respecively) 23
Tale 4. Regression Coefficiens forcsad = α + γ 1 r + γ 2r + ε (-raios in parenheses). m. 2 m. Asolue Deviaion Whole Sample Down Marke (R m <0) Up Marke (R m >0) Indusry α γ 1 γ 2 α γ 1 γ 2 α γ 1 γ 2 Cemen 1.016 0.162 *** -0.018 *** 1.002 0.217 *** -0.029 *** 1.036 0.095 ** -0.003 (5.993) (-2.691) (6.050) (-3.637) (2.373) (-0.263) Food 1.196 0.193 *** -0.013 ** 1.173 0.265 *** -0.024 *** 1.221 0.117 *** 0.0004 (7.913) (-2.050) (8.051) (-3.030) (3.240) (0.044) Plasics 1.211 0.304 *** -0.043 *** 1.188 0.381 *** -0.056 *** 1.236 0.221 *** -0.029 *** (11.281) (-6.643) (10.103) (-6.143) (6.373) (-3.728) Texile 1.393 0.229 *** -0.030 *** 1.370 0.281 *** -0.039 *** 1.421 0.169 *** -0.018 *** (9.322) (-4.774) (8.065) (-4.392) (5.593) (-2.606) Elecrical App. 1.324 0.206 *** -0.022 *** 1.308 0.258 *** -0.032 *** 1.345 0.144 *** -0.009 (7.944) (-3.291) (7.018) (-3.393) (4.843) (-1.298) Wire & Cale 1.296 0.195 *** -0.027 *** 1.277 0.258 *** -0.039 *** 1.321 0.122 *** -0.013 (7.033) (-4.059) (6.836) (-4.573) (2.993) (-1.195) Chemicals 1.234 0.242 *** -0.030 *** 1.209 0.320 *** -0.042 *** 1.260 0.161 *** -0.018 *** (10.459) (-5.398) (9.789) (-5.291) (5.364) (-2.604) Glass & Ceramics 1.545 0.201 *** -0.023 ** 1.551 0.246 *** -0.034 ** 1.544 0.145 *** -0.009 (4.725) (-2.073) (4.025) (-2.078) (2.712) (-0.692) Pulp & Paper 1.123 0.203 *** -0.036 *** 1.105 0.257 *** -0.044 *** 1.143 0.146 *** -0.027 *** (7.887) (-6.058) (7.301) (-5.561) (3.954) (-3.156) Seel 1.253 0.231 *** -0.029 *** 1.255 0.302 *** -0.043 *** 1.255 0.149 *** -0.011 (8.772) (-4.593) (8.505) (-5.418) (3.855) (-1.148) Ruer 1.248 0.271 *** -0.045 *** 1.231 0.347 *** -0.057 *** 1.266 0.194 *** -0.033 *** (9.607) (-6.644) (8.809) (-6.383) (4.731) (-3.147) Auomoile 0.923 0.125 *** -0.008 0.881 0.224 *** -0.028 *** 0.978 0.002 0.020 ** (4.028) (-0.954) (5.470) (-2.854) (0.051) (1.984) Elecronics 1.597 0.275 *** -0.058 *** 1.577 0.293 *** -0.060 *** 1.617 0.258 *** -0.057 *** (12.534) (-11.953) (9.692) (-9.514) (7.737) (-7.063) Consrucion 1.588 0.196 *** -0.025 *** 1.570 0.238 *** -0.033 *** 1.610 0.145 *** -0.014 (7.027) (-3.635) (6.103) (-3.552) (3.847) (-1.525) Transporaion 1.205 0.233 *** -0.027 *** 1.187 0.310 *** -0.039 *** 1.227 0.150 *** -0.012 (8.669) (-3.929) (8.258) (-4.146) (4.244) (-1.439) Tourism 1.096 0.137 *** 0.000 1.095 0.178 *** -0.006 1.097 0.094 ** 0.008 (4.042) (0.042) (3.662) (-0.467) (2.339) (0.805) Banking & Securiies 1.023 0.218 *** -0.025 *** 0.999 0.257 *** -0.032 *** 1.050 0.175 *** -0.017 * (9.637) (-4.496) (8.310) (-4.576) (4.958) (-1.774) Reailing 1.218 0.188 *** -0.021 *** 1.174 0.257 *** -0.029 *** 1.263 0.120 *** -0.012 * (7.105) (-3.262) (6.718) (-3.142) (3.519) (-1.656) Ohers 1.346 0.192 *** -0.018 *** 1.326 0.237 *** -0.024 *** 1.365 0.147 *** -0.012 * (8.020) (-2.907) (6.818) (-2.648) (4.957) (-1.664) (***, **, and * denoe saisical significance a 1%, 5%, and 10%, respecively)
Tale 5. Sae Space Model 1 (-raios in parenheses) log[ Sd c ( β i )] = μm + H + υ and H = φ mh 1 + η Indusry μ φ m σ m υ σ m η Cemen -0.685 *** 0.940 *** 0.085 *** 0.099 *** (-20.667) (125.420) (20.057) (20.354) Food -0.635 *** 0.930 *** 0.019 *** 0.091 *** (-26.858) (133.023) ( 3.173) (35.052) Plasics -0.580 *** 0.945 *** 0.048 *** 0.091 *** (-19.496) (153.061) (11.752) (27.420) Texile -0.647 *** 0.958 *** 0.040 *** 0.068 *** (-20.535) (144.139) (10.893) (18.509) Elecrical App. -0.698 *** 0.913 *** 0.053 *** 0.113 *** (-28.868) (116.803) (11.126) (28.401) Wire and Cale -1.456 *** 0.927 *** 0.015 *** 0.064 *** (-25.168) (708.846) ( 2.655) (21.435) Chemicals -0.657 *** 0.954 *** 0.025 *** 0.081 *** (-20.784) (143.880) ( 3.276) (17.644) Glass and Ceramics -0.577 *** 0.919 *** 0.082 *** 0.142 *** (-14.992) (100.782) (11.846) (21.611) Pulp and Paper -0.809 *** 0.892 *** 0.070 *** 0.165 *** (-22.822) ( 78.357) ( 7.603) (23.430) Seel -0.707 *** 0.949 *** 0.047 *** 0.103 *** (-21.097) (149.692) ( 6.952) (20.184) Ruer -0.828 *** 0.905 *** 0.100 *** 0.165 *** (-18.120) ( 77.216) (11.559) (20.181) Auomoile -0.457 *** 0.920 *** 0.022 *** 0.082 *** (-21.511) (106.424) ( 4.300) (29.282) Elecronics -1.282 *** 0.997 *** 0.038 *** 0.054 *** (-21.152) (624.326) (18.075) (23.420) Consrucion -0.622 *** 0.942 *** 0.014 0.070 *** (-29.220) (131.795) ( 1.536) (17.702) Transporaion -0.690 *** 0.942 *** 0.080 *** 0.105 *** (-21.305) (156.531) (12.039) (16.675) Tourism -0.555 *** 0.932 *** 0.033 ** 0.113 *** (-11.245) ( 76.183) ( 2.383) (14.307) Banking & Securiies -0.757 *** 0.956 *** 0.039 *** 0.085 *** (-22.293) (171.375) (11.205) (28.180) Reailing -0.606 *** 0.933 *** 0.039 *** 0.101 *** (-21.166) (121.470) ( 4.920) (17.390) Ohers -1.374 *** 0.916 *** 0.085 *** 0.091 *** (-10.797) (460.285) (21.045) (19.421) (***, **, and * denoe saisical significance a 1%, 5%, and 10%, respecively) 25
Tale 6. Sae Space Model 2 (-raios in parenheses) log[ Sd c ( β i )] = μm + H + cm 1 logσ + cm2r + υ and H = φ m H 1 + η Indusry μ φ m σ m υ σ m η σ m log r m Cemen 0.451 *** 0.941 *** 0.085 *** 0.098 *** 0.659 *** -0.005 (13.803) (131.070) (18.904) (19.428) (34.854) (-0.040) Food 0.589 *** 0.931 *** 0.019 *** 0.091 *** 0.682 *** 0.005 (24.607) (138.585) ( 3.376) (34.885) (51.060) (0.066) Plasics -0.848 *** 0.944 *** 0.048 *** 0.091 *** -0.151 *** 0.046 (-32.457) (137.411) (9.191) (18.795) (-10.268) (0.510) Texile 0.897 *** 0.957 *** 0.040 *** 0.068 *** 0.876 *** 0.027 (32.465) (172.134) (15.143) (26.050) (56.141) (0.374) Elecrical App. -0.806 *** 0.913 *** 0.053 *** 0.113 *** -0.058 *** -0.069 (-39.118) (113.037) (9.079) (22.353) (-4.683) (-0.520) Wire and Cale 1.437 *** 0.917 *** 0.015 *** 0.063 *** 1.658 *** 0.037 (22.304) (677.640) (4.315) (33.953) (45.111) (0.820) Chemicals -1.628 *** 0.950 *** 0.024 *** 0.082 *** -0.545 *** 0.026 (-56.555) (157.520) (5.994) (32.524) (-34.230) (0.336) Glass and Ceramics 1.330 *** 0.917 *** 0.082 *** 0.142 *** 1.092 *** -0.068 (36.300) (99.504) (10.205) (21.199) (52.221) (-0.415) Pulp and Paper -2.479 *** 0.892 *** 0.070 *** 0.164 *** -0.970 *** -0.092 (-70.245) (78.008) (7.548) (23.473) (-47.373) (-0.503) Seel -1.470 *** 0.945 *** 0.047 *** 0.104 *** -0.429 *** 0.029 (-42.510) (150.194) (10.907) (28.278) (-22.200) (0.266) Ruer -0.471 *** 0.907 *** 0.100 *** 0.165 *** 0.211 *** 0.006 (-10.372) (79.038) (10.540) (18.077) (7.917) (0.026) Auomoile -0.541 *** 0.919 *** 0.021 *** 0.082 *** -0.047 *** 0.017 (-25.690) (107.362) (4.277) (29.303) (-4.000) (0.205) Elecronics 0.481 0.997 *** 0.038 *** 0.053 *** 1.022 * 0.007 (0.527) (604.024) (13.811) (17.754) (1.933) (0.118) Consrucion 0.998 *** 0.937 *** 0.015 *** 0.069 *** 0.947 *** -0.03 (50.873) (150.635) (2.584) (35.092) (82.499) (-0.587) Transporaion -0.179 *** 0.941 *** 0.080 *** 0.105 *** 0.293 *** -0.028 (-5.391) (142.554) (17.725) (21.771) (15.301) (-0.222) Tourism -6.787 *** 0.926 *** 0.032 *** 0.111 *** -3.535 *** -0.228 * (-226.046) (100.682) (4.499) (23.539) (-208.998) (-1.823) Banking & Securiies -0.293 *** 0.957 *** 0.040 *** 0.085 *** 0.264 *** -0.058 (-8.487) (169.286) (11.260) (28.132) (13.559) (-0.687) Reailing -2.139 *** 0.926 *** 0.038 *** 0.102 *** -0.833 *** 0.009 (-90.847) (128.728) (8.629) (31.852) (-65.823) (0.074) Ohers -0.547 *** 0.954 *** 0.084 *** 0.095 *** 0.013 0.137 (-14.520) (159.310) (20.375) (20.390) (0.631) (0.870) (***, **, and * denoe saisical significance a 1%, 5%, and 10%, respecively) 26