U.S. and Domestic Market Gains and Asian Investors Overconfident. Trading Behavior

Transcription

1 U.S. and Domesic Marke Gains and sian Invesors Overconfiden Trading Behavior Wen-I Chuang 1, Bong-Soo Lee, and Kai-Li Wang 3 bsrac The overconfidence hypohesis pu forward by Gervais and Odean (001) predics ha if invesors are overconfiden, hey rade more aggressively subsequen o marke gains. To gain furher insigh ino he hypohesis, we examine wheher boh U.S. and domesic marke gains make sian invesors rade more aggressively in subsequen periods in heir domesic markes. Consisen wih he predicions of he heory, we find ha boh U.S. and domesic marke gains make sian invesors rade wih more overconfidence in bull markes, in periods of high invesor senimen, and in periods of exremely high marke reurns afer conrolling for he alernaive heories in explaining he reurn-volume relaion. Moreover, we find ha furher inegraion of sian sock markes wih U.S. sock markes afer he sian financial crisis in 1998 is an imporan reason for sian invesors response o U.S. marke gains by rading wih overconfidence. We also find ha sian invesors exhibi more significan overconfiden rading behavior in markes wih a shor-sale consrain han in markes wihou i. JEL classificaion: C3, G15 Keywords: Overconfidence; Marke gains; Trading volume; Senimen 1 Corresponding auhor. Deparmen of Finance, College of Managemen, Naional Taiwan Universiy, No. 1, Secion 4, Roosevel Road, Taipei 10617, Taiwan; phone: ; fax: ; wichuang@managemen.nu.edu.w Deparmen of Finance, College of Business, Florida Sae Universiy, 311 Rovee Building, Tallahassee, FL , US; phone: ; fax: ; blee@cob.fsu.edu 3 Deparmen of Finance, College of Managemen, Tunghai Universiy, No. 181, Secion 3, Taichung Por Road, Taichung 40704, Taiwan; phone: ex ; fax: ; kaiwang@hu.edu.w

2 U.S. and Domesic Marke Gains and sian Invesors Overconfiden Trading Behavior bsrac The overconfidence hypohesis pu forward by Gervais and Odean (001) predics ha if invesors are overconfiden, hey rade more aggressively subsequen o marke gains. To gain furher insigh ino he hypohesis, we examine wheher boh U.S. and domesic marke gains make sian invesors rade more aggressively in subsequen periods in heir domesic markes. Consisen wih he predicions of he heory, we find ha boh U.S. and domesic marke gains make sian invesors rade wih more overconfidence in bull markes, in periods of high invesor senimen, and in periods of exremely high marke reurns afer conrolling for he alernaive heories in explaining he reurn-volume relaion. Moreover, we find ha furher inegraion of sian sock markes wih U.S. sock markes afer he sian financial crisis in 1998 is an imporan reason for sian invesors response o U.S. marke gains by rading wih overconfidence. We also find ha sian invesors exhibi more significan overconfiden rading behavior in markes wih a shor-sale consrain han in markes wihou i. JEL classificaion: C3, G15 Keywords: Overconfidence; Marke gains; Trading volume; Senimen 1

3 1. Inroducion I has long been argued ha rading volume in speculaive markes is oo large o be usified on raional grounds (e.g., Dow and Gordon, 1997). 1 Excess rading volume herefore is a puzzle represening a grea challenge o he field of finance. De Bond and Thaler (1995, p.393) sae he key behavioral facor needed o undersand he rading puzzle is overconfidence. long his line, here are a growing number of heoreical models rooed in invesor overconfidence o accoun for he observed excess rading volume in securiies markes. Gervais and Odean (001) (henceforh GO) propose a self-learning model predicing ha biased invesors overesimae he degree o which hey conribue o reurns from general marke increases and exaggerae heir abiliy o pick up he winning socks, and herefore hey rade more aggressively following marke gains. Since invesors in aggregae hold long posiions in he equiy marke, he GO model provides a esable implicaion of he overconfidence hypohesis: high marke gains are followed by high marke rading volume. Several empirical sudies presen evidence ha overconfidence plays a pivoal role in explaining individual invesors propensiy o rade oo much and oo speculaively. For example, using a sample of discoun brokerage accouns, Odean (1999) and Barber and Odean (000) focus on he invesigaion of he irraional rading behavior of U.S. individual invesors and find evidence ha individual invesors appear overconfiden abou heir perceived informaion and abiliy o rade because hey rade oo much and oo speculaively, ye heir acive rading reduces heir performance. 3 Odean (1998b, 1999) and GO (001) argue ha people who are more overconfiden in heir invesmen abiliies are more likely o seek obs as raders or o acively rade on heir own accouns. If so, we can expec o observe ha financial markes are populaed by overconfiden invesors. Many researchers also argue ha overconfiden invesors can survive 1 From a heoreical viewpoin, rading volume can be generaed due o risk-sharing, hedging, speculaion, and liquidiy. Moreover, facors such as asymmeric informaion, ransacion coss, idiosyncraic risks, and oher forms of marke imperfecions are also likely o be relevan for deermining he level and variabiliy of rading aciviy. However, rading moivaed from hese heoreical argumens and from marke imperfecions seems o fail o suppor a subsanial amoun of rade observed in he real world. similar argumen ha overconfidence leads o greaer rading is also made in De Long, Shleifer, Summers, and Waldmann (1991), Kyle and Wang (1997), Benos (1998), Odean (1998b), Wang (001), Daniel, Hirshleifer, and Subrahmanyam (001), Hirshleifer and Luo (001), Caballé and Sákovics (003), and Scheinkman and Xiong (003). 3 The noion ha overconfidence leads individual invesors o rade oo much and oo speculaively is also empirically suppored by several experimenal sudies (e.g., Biais, Hilon, Mazurier, and Pouge, 005; Deaves, Lüders, and Luo, 009).

4 and dominae markes in he long run (e.g., Kyle and Wang, 1997; Benos, 1998; Daniel, Hirshleifer, and Subrahmanyam, 1998; Gervais and Odean, 001; Hirshleifer and Luo, 001; Wang, 001). These argumens imply ha i is possible o deec invesors aggregae overconfiden rading behavior from he marke level if overconfidence is a sysemaic cogniive bias from which mos invesors suffer. Focusing on aggregae overconfiden rading behavior, Saman, Thorley, and Vorkink (006) and Chuang and Lee (006) find ha high U.S. marke-wide reurns are followed by high U.S. marke-wide rading volume, and hey inerpre heir finding as consisen wih he heoreical predicion of he GO model. Previous empirical research on he dynamic relaion beween invesors overconfidence and rading volume focuses primarily on U.S. invesors. However, overconfidence is no somehing unique o U.S. invesors. Psychologiss have demonsraed ha sians also exhibi overconfidence in general knowledge (e.g., Yaes, Lee, and Bush, 1997). This suggess sian invesors can be a good plaform upon which o es he overconfidence hypohesis. Indeed, Chen, Kim, Nofsinger, and Rui (007), Kim and Nofsinger (007), and Chuang and Susmel (011) find evidence ha Chinese, Japanese, and Taiwanese invesors display significanly overconfiden rading behavior, respecively. Prior sudies on invesors overconfiden rading end o focus on wheher domesic marke gains make invesors rade wih overconfidence in domesic markes. In his aricle, we conecure ha sian invesors overconfidence and overconfiden rading can be evoked by U.S. marke gains for hree reasons. Firs, for he purpose of inernaional diversificaion, sian invesors have a desire o inves in various U.S.-based invesmens (e.g., French and Poerba, 1991; Eun and Resnick, 1994; Bekaer and Urias, 1996; Driessen and Laeven, 007). We find ha he share of U.S. socks held by invesors in our sample sian counries seadily increases during our sample period. These invesors benefi from U.S. marke gains. Then, as argued by GO (001), he increase in wealh makes hem overconfiden and hey are encouraged o rade more acively in subsequen periods in domesic markes. Second, wih an increasing inegraion of inernaional equiy markes, any news abou U.S. sock markes may have imporan implicaions for oher sock markes. 4 This is why he financial media repor he performance of he U.S. sock markes on a regular basis in sian counries since his can help sian invesors infer fuure price movemens of heir domesic sock markes from he curren U.S. sock marke price movemens. sian invesors, 4 Numerous sudies on inernaional equiy markes presen evidence of he informaion spillover effec from U.S. sock markes o sian sock markes. mong hem are Eun and Shim (1989), Hamao, Masulis, and Ng (1990), Lin, Engle, and Io (1994), and Lee, Rui, and Wang (004). 3

5 for example, may expec ha heir domesic markes would go up in he near fuure afer observing U.S. marke gains. s a consequence, heir expecaions of gains from local equiy invesmens make sian invesors overconfiden and herefore hey rade more aggressively in domesic markes subsequen o U.S. marke gains. Third, Baker, Wurgler, and Yuan (010) show ha invesor senimen can be decomposed ino global and local componens. They find ha U.S. capial flows are a key mechanism by which senimen spreads across markes. They also argue ha word-of-mouh and he media can also have a similar effec. By he same oken, invesor overconfidence migh also have global and local componens. Given exensive evidence of he reurn spillover effec from U.S. sock markes o sian sock markes, U.S. marke gains can be a key facor conribuing o he global componen of sian invesors overconfidence and make hem rade more frequenly in heir domesic markes. Moreover, psychologiss have documened ha opimisic people end o be overconfiden. 5 s such, global bullish senimen may also nourish sian invesors overconfidence and make hem rade more aggressively in heir domesic markes. Based on he GO (001) overconfiden rading hypohesis, we examine empirically wheher invesors in he en sian sock markes rade wih overconfidence afer domesic and U.S. marke gains. Consisen wih our conecure, we find ha no only domesic marke gains bu also U.S. marke gains make sian invesors overconfiden and rade more aggressively in subsequen periods. Our Granger (1969) causaliy ess show ha U.S. marke reurns (posiively) Granger-cause sian marke reurns, excep for he China sock marke, even for hose markes in which U.S. marke reurns do no (direcly) Granger-cause domesic rading volume. This implies ha U.S. marke reurns may also indirecly affec he overconfiden rading of sian invesors. s for he impac of U.S. marke gains on sian invesors overconfiden rading, we find ha sian invesors rade more acively subsequen o U.S. marke gains when U.S. invesor senimen is high. This finding implies ha high U.S. invesor senimen leads sian invesors o expec o gain from heir domesic and U.S.-based equiy invesmens and herefore rade more acively in domesic markes. Our sub-period analysis also shows ha mos sian invesors rade more acively afer U.S. marke gains in he second sub-period in which he inegraion of sian sock markes wih he U.S. sock markes has become sronger and sian invesors rade more in U.S. socks han in he firs sub-period. s 5 See Odean (1998b) for a review of his line of he psychological lieraure and furher references. 4

6 discussed above, his implies ha sian invesors overconfiden rading increases following U.S. marke gains due o he recen rend oward globalizaion of markes. The heoreical lieraure demonsraes ha invesors overconfidence can be affeced by shor-sale consrains. In Baker and Sein s (004) model, in he presence of shor-sale consrains, dumb invesors wih posiive senimen dominae he marke and smar invesors si on he sidelines, making dumb invesors become more overconfiden. Wihou shor-sale consrains, boh dumb and smar invesors acively paricipae in he marke. In his circumsance, he ineracion of heir senimen makes dumb invesors become overconfiden only o a lesser degree (see also Hong, Scheinkman, and Xiong, 006). In a similar vein, Scheinkman and Xiong (003) show ha when invesors face shor-sale consrains, he asse price bubble increases wih heir overconfidence and more rading is generaed during he bubble period. To es he impac of shor-sale consrains on sian invesors overconfiden rading, we divide our sample counries ino wo groups based on wheher shor sale is allowed and praciced in heir sock markes (see Bris, Goezmann, and Zhu, 007). 6 The heory predics ha invesors will rade more acively afer domesic marke gains in he presence of he shor-sale consrain han in he absence of i. Consisen wih he predicion, we find significan evidence ha sian invesors in markes wih shor-sale consrains, on average, end o rade more frequenly afer domesic marke gains han hose in markes wihou shor-sale consrains. To he bes of our knowledge, our paper is he firs o empirically address his issue in he conex of he overconfiden rading. In conras, we find lile difference in acive rading subsequen o U.S. marke gains beween sian invesors in markes wih and wihou shor-sale consrains. This is probably because shor sale is allowed and praciced in he U.S. sock marke in which socks are less overpriced incorporaing pessimisic invesors valuaion. To gain more insigh ino he overconfiden rading behavior of sian invesors, we furher analyze how sian invesors behave condiional on invesor senimen, marke condiions, and aenion-grabbing informaion proxied by exremely high marke reurns. These condiional evens are suggesed by he behavioral heory. Using his condiional framework, firs we find ha sian invesors end o rade more acively afer domesic (U.S.) 6 Bris e al. (007) find ha alhough shor sale is currenly legal in mos of heir sample counries, ax rules, ransfer laws, and oher marke regulaions make shor sale very difficul o pracice in hese counries. Consequenly, for he purpose of heir sudy, hey classify heir sample counries based on wheher shor sale is praciced raher han on wheher shor sale is legal. Their sample counries also include en sian counries in our sample, and herefore we use heir classificaion for our sample counries. 5

7 marke gains when hey (U.S. invesors) are more opimisic. Second, we find ha sian invesors end o rade more frequenly following domesic (U.S.) marke gains when heir domesic (U.S.) markes are in bull markes han when heir domesic (U.S.) markes are in non-bull markes. nd hird, we find ha sian invesors end o rade more aggressively when he domesic (U.S.) marke experiences exremely high reurns han when i does no. Overall, our sudy provides evidence ha sian invesors rade as overconfidenly as U.S. invesors, as documened in he prior lieraure, and ha no only domesic marke gains bu also U.S. marke gains make sian invesors rade wih overconfidence. Baker, Wurgler, and Yuan (010) show ha invesor senimen can be affeced by cross-marke facors. We conribue o he lieraure by showing ha cross-marke facors such as U.S. marke gains can affec sian invesors overconfidence, which furher affecs heir rading behavior in domesic markes. noher conribuion of our sudy is ha we provide evidence ha sian invesors overconfiden rading behavior increases wih shor-sale consrains. In addiion o reducing invesors wealh, heir overconfidence also has some negaive impacs on he marke. This finding has imporan policy implicaions. For example, Odean (1998b) heoreically shows ha invesors overconfidence makes he marke excessively volaile, and Chuang and Lee (006) offer evidence ha excessive rading of overconfiden invesors conribues o he observed excessive volailiy in securiies markes. Our finding suggess ha if marke policy makers aemp o proec invesors from succumbing o he overconfidence bias and o make he marke more sable, removing shor-sale consrains will help achieve hese goals. Indeed, we find ha our sample sock markes wihou shor-sale consrains, on average, are more sable han hose wih shor-sale consrains. However, here are alernaive heories ha predic he reurn-volume relaion we documen here. For example, o inerpre heir finding of he posiive relaion beween pas reurns and curren volume from 46 counries, Griffin, Nardari, and Sulz (007) review and discuss heories ha predic such a relaion. They noice ha in addiion o overconfidence, here are five alernaive heories ha provide poenial explanaions. They also noice ha heories do no provide clear disincions among hemselves and heir explanaions are mosly suggesive. Saman e al. (006) also recognize his dilemma. In spie of his dilemma, we also aemp o devise some empirical frameworks in order o es hese alernaive heories in explaining our findings. The resuls from hese ess show ha he reurn-volume relaion in hree of en sample counries can be explained by wo alernaive heories. Taking his ino accoun, we conrol for he impacs of hese wo alernaive heories on he reurn-volume relaion when we invesigae he overconfiden 6

8 rading behavior of invesors in hese hree counries. Compared o Saman e al. (006) and Griffin e al. (007), we ake a comprehensive approach o examining invesors overconfiden rading behavior based on he broad implicaions of he overconfidence hypohesis in an aemp o provide more convincing evidence on he issue. The advanage of doing so is ha exensive evidence obained from his approach also helps rule ou he possibiliy ha our empirical findings are us he spurious resuls of oher alernaive heories since none of alernaive heories individually can explain all of our findings as a whole. The remainder of he paper is organized as follows. Secion inroduces he daa, describes our mehod of derending he rading volume series o achieve is saionariy, and repors some descripive saisics of he daa. Secion 3 discusses our empirical frameworks ha are devised o invesigae wheher sian invesors rade wih overconfidence due o domesic and U.S. marke gains, and presens and discusses he empirical resuls. Secion 4 reviews he alernaive heoreical argumens ha imply a posiive reurn-volume relaion and discusses he es resuls of hese argumens, and conduc some robusness checks. Finally, Secion 5 offers some concluding remarks.. Daa and Derending Trading Volume Series.1 Daa Our daa se includes marke price indices and rading volumes for en sian sock markes: Hong Kong, Japan, Malaysia, Singapore, Thailand, China, Indonesia, Korea, he Philippines, and Taiwan. The sock indices for he en sian sock markes are he Hang Seng Index (HSI) for Hong Kong, he Tokyo Sock Exchange Price Index (TOPIX) for Japan, he Kuala Lumpur Composie Index (KLSE) for Malaysia, he Srais Times Index (STI) for Singapore, he Sock Exchange of Thailand Index (SET) for Thailand, he Shanghai SE Composie Index (EC) for China, he Jakara Composie Index (JKSE) for Indonesia, he Korea Sock Exchange Composie Index (KOSPI) for Korea, he Philippines Sock Exchange Composie Index (PSECI) for he Philippines, and he Taiwan Weighed Index (TWI) for Taiwan. The marke index for he U.S. sock marke is he S&P 500. To invesigae wheher sian invesors overconfiden rading is affeced by shor-sale consrains, we divide he sample sock markes ino wo groups based on wheher shor sale is allowed and praciced in heir sock markes. Following he Bris e al. (007) classificaion, 7

9 Hong Kong, Japan, Malaysia, Singapore, and Thailand are classified as he sock markes wihou shor-sales consrain since shor sales are allowed and praciced, while China, Indonesia, Korea, he Philippines, and Taiwan are classified as hose wih shor-sales consrain since shor sale is prohibied or is allowed bu no praciced. Hereafer, we call he markes and invesors in he former counries as he (shor-sale) markes and invesors, respecively, and he markes and invesors in he laer counries as he (shor-sale consrains) markes and invesors, respecively. For our analysis, we use weekly observaions ha are consruced from he daily daa exraced from he Daasream Inernaional daabase, and he sample period is from January 1995 o December 010. The weekly reurn of each index is compued as he reurn from Wednesday s closing price o he following Wednesday s closing price (see also Griffin e al., 007). 7 The Wednesday-o-Wednesday close-o-close reurns of each marke index are measured as he log-difference of he index prices and are expressed in percen. We use he oal number of shares raded in a rading day as a measure of raw (or underended) rading volume. Following Lo and Wang (000), he weekly raw rading volume is compued as he log of a summaion from he oal number of shares raded on Thursday o ha raded on he following Wednesday. We mach all series of sock reurns and raw rading volume because our empirical analyses employ Zellner s (196) Seemingly Unrelaed Regression (SUR) model ha requires all variables o have he same number of observaions.. Derending Trading Volume Series Previous work finds significan evidence of boh linear and nonlinear ime rends in he rading volume series (e.g., Gallan, Rossi, and Tauchen, 199). s a resul, many empirical sudies on rading volume use some form of derending o achieve saionariy. In he spiri of Gallan e al. (199), we derend he raw rading volume series, RV, for each sian sock marke by aking ino accoun he calendar effec on rading volume as follows (see also Lo and Wang, 000): 7 Previous empirical sudies documen ha non-synchronous rading is a more serious problem in daily daa han in weekly daa (e.g., Kadlec and Paerson, 1999) and ha Wednesday rading volume is higher relaive o oher weekdays rading volume (e.g., Barclay, Lizenberger, and Warner, 1990). The use of he Wednesday-o-Wednesday week will alleviae he concerns of he non-synchronous rading and non-rading problems (see also Chordia and Swaminahan, 000). 8

10 V RV ( ˆ ˆ ˆ ˆ DEC1 ˆ DEC ˆ DEC3 ˆ DEC ˆ JN1 ˆ JN ˆ JN3 ˆ JN L ˆ 1MR 13PR 0NOV lrv l l1 ˆ ˆ... ˆ ), (1) where he regressors DEC1,, DEC4 and JN1,, JN4 denoe weekly indicaor variables for he weeks in December and January, respecively; and MR,, NOV denoe monhly indicaor variables for he monhs of March hrough November, respecively. February is omied o avoid perfec collineariy. The number of lags in he auoregressive erms, RV l, is deermined by he Lung-Box Q-saisic; ha is, we add lags unil he Lung-Box saisic shows no auocorrelaion of he residual erms for each derended rading volume series. Throughou our analysis, he resuling derended rading volume, V, is used as a measure of rading volume for sian sock marke. To faciliae reporing of he esimaed coefficiens in our ess, he esimaed derended rading volume, V, is muliplied by Summary Saisics Table I presens descripive saisics on he weekly reurn and rading volume series for en sian marke indices and he weekly reurn series for he S&P 500 for he sample period from January 4, 1995 o December 9, 010. Specifically, he able repors he mean, sandard deviaion (SD), minimum and maximum values, he es saisic (-saisic) of he augmened Dickey-Fuller es (DF, Dickey and Fuller, 1979), wo sums of he correlaions, and he resuls of hree Granger-causaliy ess. Over he enire sample period, Indonesia has he highes mean reurn of 0.30%, while Japan has he lowes, %. Korea is he mos volaile marke wih a sandard deviaion of 4.654%, whereas he S&P 500 is he mos sable index wih a sandard deviaion of.848%. I is worh noing here ha Hong Kong, Japan, Malaysia, and Singapore have a smaller sandard deviaion han China, Indonesia, Korea, and he Philippines. This implies ha he sock markes wihou shor-sale consrains, on average, are more sable han hose wih shor-sale consrains. The rading volume of all sian sock markes exhibis a mean equal o zero afer derending. This is because derended rading volume is he residual erm in 8 We hereafer refer o he derended rading volume as rading volume or volume. 9

11 Equaion (1). Trading volume appears o be more sable relaive o sock reurns for all sian sock markes; he Philippines and Japan have he highes and lowes volume volailiies of and 0.169, respecively. The resuls of he DF es show ha he null hypohesis of a uni roo can be reeced for all sian sock marke reurns and rading volume, indicaing ha hey are saionary ime series. 1 and are used o measure he sum of he correlaions beween curren volume and lagged domesic reurns up o hree lags, and beween curren volume and lagged U.S. reurns up o hree lags, respecively. The able shows ha 1 are posiive for all sian counries and are posiive for all sian counries bu China, suggesing a posiive relaion beween lagged domesic reurns and curren volume, and beween lagged U.S. reurns and curren volume for hese counries. The formal es of hese relaions can be found in causaliy ess 1,, and 3, which are he es resuls of he rivariae Granger-causaliy ess of domesic rading volume, domesic marke reurns, and U.S. marke reurns. 9 Specifically, causaliy es 1 is he es resul of wheher domesic marke reurns posiively Granger-cause domesic rading volume, causaliy es wheher U.S. marke reurns posiively Granger-cause domesic rading volume, and causaliy es 3 wheher U.S. marke reurns posiively Granger-cause domesic marke reurns. The resuls of he causaliy es 1 show ha domesic marke reurns posiively Granger-cause domesic rading volume for all sian counries, excep for Singapore. This implies ha sian invesors, excep for Singapore invesors, end o rade more acively afer domesic marke gains. The resuls of he causaliy es show ha U.S. marke reurns posiively Granger-cause domesic rading volume for Hong Kong, Singapore, Korea, and Taiwan. This implies ha invesors in hese four counries end o rade more aggressively afer U.S. marke gains. The resuls of he causaliy es 3 show ha, excep for China, U.S. marke reurns posiively Granger-cause domesic marke reurns for all he oher sian counries. This indicaes ha he sock price movemens in he U.S. sock markes help predic he sock price movemens in domesic sock markes for all counries, excep for China. Moreover, his finding provides a basis for our conecure of why sian invesors rade more acively subsequen o U.S. marke gains. 9 In our rivariae Granger-causaliy ess, we allow for he differen lag lenghs of he righ-hand side variables in each equaion. Specifically, he ess of wheher domesic and U.S. marke reurns Granger-cause domesic rading volume are performed afer conrolling for he impacs of informaion flows emanaing from domesic and U.S. sock markes on domesic rading volume (see our discussion in Secion 3.1 for deails). Following Chuang and Lee (006) and Chuang and Susmel (011), our Granger-causaliy ess examine no only for causal relaions bu also for heir signs. Specifically, he examinaion of wheher he sum of esimaed lagged coefficiens is greaer han zero is used o idenify he sign of he causal relaions. The causaliy ess and he sign of he causaliy are performed using a Wald es. 10

12 3. Empirical Frameworks and Tess of sian Invesors Overconfiden Trading Behavior 3.1 Overconfiden Trading across sian Sock Markes s noiced by Griffin e al. (007), in addiion o invesor overconfidence, several alernaive heories have been proposed for a posiive relaion beween lagged reurns and curren volume. We also conduc he empirical ess o examine wheher hese heories provide imporan deerminans of he posiive reurn-volume relaion in Secion 4. Our resuls sugges ha he reurn-volume relaion needs o be conrolled for he paricipaion effec and informed rading effec for Hong Kong and Singapore and for Taiwan, respecively. 10 Since we find some evidence for hese alernaive heories, we conrol for hese facors when we invesigae sian invesors overconfiden rading behavior. Before inroducing our empirical models, we discuss wo conrol variables used in all of our ess, including he Granger-causaliy ess in Table I. Ross (1989) shows ha in a fricionless marke characerized by an absence of arbirage opporuniies, he rae of informaion flow is revealed by he degree of he volailiy of asse reurns. Based on his, previous sudies employ he absolue value of sock reurns as a proxy for informaion flow o he sock markes (e.g., Bessembinder, Chan, and Seguin, 1996; Chuang and Lee, 006; Chuang and Susmel, 011). Following previous sudies, we use he absolue value of he reurns of he sian sock marke, R, and ha of he reurns of he U.S. sock marke, US R, as proxies for informaion flows emanaing from sian sock marke and from he U.S. sock marke, respecively, o accoun for informaional rades. These wo conrol variables play an imporan role in our empirical analysis because he heory suggess ha overconfiden invesors rade on non-informaional moives (i.e., an overconfidence bias). Following Chuang and Susmel (011), we work wih he derended R and R US US series since i is well-known ha R and R are serially correlaed (e.g., Ding, Granger, and Engle, 1993). We follow Pagan and Schwer s (1990) mehod o filer R US and R using he following wo models: 10 We discuss he alernaive heories on he posiive relaion beween lagged reurns and curren volume and he es resuls of hese heories in Secion 4. 11

13 P p p p1 R R e, () Q US US US US US q q q1 R R e. (3) We ake he esimaed residual in Equaion (), e, as he derended absolue value DR, US and similarly he esimaed residual in Equaion (3), e, as he derended absolue value US DR, and use hem as conrol variables in our empirical models. 11 We uilize he SUR model in mos of our ess, including he Granger-causaliy ess in Table I. This is because he mulivariae SUR model allows us o es he cross-equaion resricions when we examine he impac of shor-sale consrains on he overconfiden rading of versus invesors. 1 To compare he relaive degree of he overconfiden rading of versus invesors, we also sandardized he variables in our empirical models. We esimae he following SUR model o invesigae sian invesors overconfiden rading behavior across counries: J K US US 1 3 k k 1 k1 V DR DR R R, (4) where he superscrip represens he cross-secional uni of he en sian sock markes ; V is he rading volume of he marke on week ; US R ( R ) is he sock reurn of he marke (U.S. sock marke) on week ; US DR ( DR ) is he derended absolue value of US R ( R ) on week. In Equaion (4), he ( ) coefficiens measure he causal k relaion from lagged domesic (U.S.) reurns o curren volume for he marke. If he ( ) coefficiens are posiive and significan as a group a convenional levels, domesic k (U.S.) marke gains help predic invesors in he marke o rade wih overconfidence in subsequen periods. To conrol for he paricipaion effec on he domesic reurn-volume relaion for Hong Kong and Singapore, we modify Equaion (4) in such a way ha domesic reurns are decomposed ino wo componens for hese wo counries: domesic reurns condiional on 11 US We also reesimae all our empirical models using non-derended R and R and find ha he resuls are similar o wha we repor in he paper. 1 One advanage of using he SUR model is ha i accouns for he cross-equaion correlaions of he conemporaneous residuals in drawing inferences concerning he regression parameers. 1

14 he sae of posiive marke reurns wih high volailiy and hose condiional on he oher sae. Under his specificaion, he posiive relaion beween curren volume and pas reurns condiional on he sae of posiive reurns wih high volailiy is explained by he paricipaion effec, while he relaion condiional on he oher sae is explained by invesor overconfidence. Similarly, o conrol for he informed rading effec on rading volume for Taiwan, we modify Equaion (4) in such a way ha U.S. reurns are decomposed ino wo componens for Taiwan: U.S. reurns associaed wih he high volailiy of U.S. reurns and hose no associaed. posiive relaion beween curren volume and pas U.S. reurns relaed o high U.S. marke volailiy is inerpreed by he informed rading effec, while he relaion unrelaed o high U.S. marke volailiy is inerpreed by invesor overconfidence. We always use he similar specificaions o conrol for he paricipaion and informed rading effecs for Hong Kong and Singapore and for Taiwan, respecively, in our following examinaions of sian invesors overconfiden rading behavior. We use he sum of he coefficiens on lagged domesic (U.S.) reurns as he measure of cumulaive overconfiden rading caused by domesic (U.S.) marke gains. The greaer he sum is, he higher he degree of overconfiden rading. However, some of he coefficiens may no be significan a convenional levels. To avoid he poenial problem of over-parameerizaion and o obain a parsimonious model, financial economiss end o adop a general-o-simple sraegy, in which one begins wih a general specificaion and hen simplifies i afer esing parameer esimaes (e.g., Hendry and Ericsson, 1991; Ericsson and Marquez, 1993). Based on his empirical sraegy, we use a wo-sep procedure o esimae he SUR model of Equaion (4) and ohers. For insance, in he firs sep, he SUR model of Equaion (4) is esimaed as usual. In he second sep, he SUR model esimaed from he firs-sep procedure is re-esimaed by dropping all insignifican variables a he 10% significance level. 13 The resuling SUR model conains he variables, excep for he consan erms, ha are significan a leas a he 10% level. This assures ha our measure of (cumulaive) overconfiden rading is free from he concern ha he influence of he insignifican coefficiens on he measure may disor and blur he resuls. Then, we compare he relaive degrees of overconfiden rading of versus invesors based on he relaive magniude of he measure of heir overconfiden rading. If, for example, we find ha he sum of he 13 s a robusness check, we also esimae he models wihou deleing he insignifican esimaed coefficiens and conduc he ess wih only he significan esimaed coefficiens. Overall, we find ha evidence on sian invesors overconfiden rading are even slighly sronger han wha we presen in he paper. To conserve space, we do no repor he resuls of his robusness check, bu hey are available from he auhors. 13

15 significan coefficiens divided by he number of he markes is significanly smaller han ha of he significan coefficiens divided by he number of he markes, i provides evidence ha he degree of overconfiden rading of invesors is less han ha of invesors. Table II repors he esimaion resuls of Equaion (4). 14 The resuls show ha he sum of he bea coefficiens ha are individually significan a leas a he 10% level is posiive for hree counries of Japan (J), Malaysia (M), and Thailand (TH) and for all counries. This implies ha boh and invesors rade more frequenly afer domesic marke gains. We furher compare he relaive degree of he overconfiden rading of versus invesors. This is esed based on he chi-squared saisic wih one degree of freedom under he null hypohesis ha he average of he sum of he coefficiens across hree markes for = HK, M, and TH (i.e., / 3) is equal o ha of he coefficiens across all markes (i.e., / 5). The observaion ha /3 /5 and he reecion of he saisic a he 1% level oinly indicae ha invesors, on average, rade wih more overconfidence afer domesic marke gains han invesors. Looking a he impac of U.S. marke gains on sian invesors overconfiden rading, Table II shows ha he sum of he gamma coefficiens ha are individually significan a leas a he 10% level is posiive for hree counries of Hong Kong, Malaysia, and Singapore and for one counry of Korea. This implies ha invesors in Hong Kong, Malaysia, and Singapore and invesors in Korea rade more frequenly afer U.S. marke gains. The saisic wih one degree of freedom is used o es he null hypohesis ha he average of he sum of he k coefficiens across hree markes for = HK, M, and SI is equal o ha of he k coefficiens for = KO. The observaion ha he saisic is no significan a convenional levels indicaes ha he degree of overconfiden rading due o U.S. marke gains exhibis no significan difference across invesors in Hong Kong, Malaysia, and Singapore and invesors in Korea. Lasly, i is noed from Table II (and he following ables) ha on average, he response 14 For ease of discussion, in Table II and he following ables we do no repor he resuls of he bea coefficiens on lagged domesic reurns condiional on posiive marke reurns wih high volailiy for Hong Kong and Singapore and hose on lagged U.S. reurns associaed wih high marke volailiy for Taiwan since hey can no inerpreed as invesors overconfiden rading. Bu if we find he resuls of he reurn-volume relaion ha can be inerpreed as invesors overconfiden rading, we will repor hem in he ables. 14

16 of volume o U.S. reurns is smaller han ha o domesic reurns. This makes sense because local marke gains should lead sian invesors o make more overconfiden rading in heir domesic markes han U.S. marke gains. 3. The Effec of Marke Inegraion sian financial markes have become increasingly inegraed over ime wih global markes and each oher. For example, Kim (005) and Jeon, Oh, and Yang (006) find ha he inegraion of sian sock markes wih U.S. sock markes has become increasingly sronger in recen years, paricularly afer he sian financial crisis in Such inegraion would make sian invesors rade more in U.S.-based equiy invesmens and pay more aenion o U.S. sock markes. We collec he monhly daa of he dollar amoun of U.S. socks raded by invesors in each of our sample counries from he U.S. Treasury Inernaional Capial (TIC) reporing sysem. For all sample counries, we find ha he ime series of he dollar amoun of U.S. socks raded by sian invesors exhibis a significan linear ime rend and ha he ime series mean of he dollar amoun is significanly higher afer year 1998 han ha before i. 15 More imporanly, he increasing inegraion also makes he price movemens in he U.S. sock markes more useful in predicing he price movemens in sian sock markes, and, as a consequence, U.S. marke gains migh conribue more o he global componen of sian invesors overconfidence. Therefore, we conecure ha he degree of sian invesors overconfiden rading due o U.S. marke gains would be more significan afer he sian financial crisis han before. I is no clear, however, wheher such inegraion would exer is impac on sian invesors overconfiden rading promped by domesic marke gains. 16 Some researchers also find ha he inegraion of sian sock markes wih U.S. sock markes increases during he global financial marke urmoil caused by he subprime morgage crisis in July 007 (e.g., Diebold and Yilmaz, 009). lhough his suggess ha sian invesors overconfiden rading due o U.S. marke gains migh also be affeced by his episode, in he ime-series framework he exen o which he sian financial crisis affecs sian invesors overconfiden rading would be more pronounced han ha he global 15 We do no abulae he es resuls of he linear ime rend in he dollar amoun of U.S. socks raded by sian invesors in our sample counries and he ime series difference in he dollar amoun beween before and afer year 1998, bu hey are available upon reques from he auhors. 16 If sian invesors learn o be less overconfiden from heir rading, heir overconfiden rading will decrease over ime. However, Gervais and Odean (001) argue ha less experienced invesors end o learn o be overconfiden. They also argue ha overconfidence will always flourish in a marke in which new raders are born every minue. If so, sian invesors overconfiden rading will no decline over ime. 15

17 financial marke urmoil. To examine our conecure, we divide our sample ino hree sub-periods: pre-crisis period (January 1995 ~ December 1998), pos-crisis period (January 1999 ~ June 007), and periods during and afer he global financial marke urmoil (July 007 ~ December 010). We esimae he following mulivariae SUR model over he full sample period: V DR (1S )(1S3 ) DR S DR S US US US 5 DR (1S )(1S3 ) 6 DR S 7 DR S3 J J J 1R(1S )(1S3 ) RS 3RS K K K US US US 1 k R k S k S k k R k S k 3 k R k S k k1 k1 k1 (1 )(1 3 ) 3, (5) where he variables are defined as before, and he dummy variable S ( S 3 ) akes on a value of one for he sub-sample period from January 1999 o June 007 (from July 007 o December 010), and zero oherwise. In Equaion (5), he 1,, and 3 (,, and ) coefficiens measure he 1k k 3k relaion beween lagged domesic (U.S.) reurns and curren domesic volume in pre-crisis period, pos-crisis period, and periods during and afer he global financial marke urmoil, respecively, for he marke. sharp decline in U.S. and sian sock markes during he global financial sunami period is no a good environmen for sian invesors o nourish heir overconfidence. 17 Consequenly, i is reasonable o conecure ha sian invesors overconfiden rading migh no be so eviden during his period, hough he marke linkages beween sian and U.S. sock markes increase. s such, we focus on examining wheher sian invesors overconfiden rading due o U.S. marke gains is sronger afer he sian financial crisis han before. If our conecure is correc, we expec o see ha he sum of he 1k coefficiens is smaller han ha of he k and 3 coefficiens for he marke. To faciliae discussion, he period beween January 1999 and December 010 is simply referred o as he pos-crisis period. Table III repors he esimaion resuls of Equaion (5). Some imporan observaions are noed as follows. Firs, he k (1 vs. 3 ) saisic is used o es he null hypohesis ha he sum of he coefficiens is equal o ha of he and 3 coefficiens for he 1 17 In unrepored resuls of Table III, we find no evidence of sian invesors overconfiden rading behavior due o U.S. marke gains during and afer he global financial sunami. 16

18 marke. Tha is, we es wheher here is a significan difference in sian invesors overconfiden rading due o domesic marke gains beween pre- and pos-crisis periods. We find ha local marke gains make sian invesors in each counry rade wih overconfidence afer domesic marke gains in a leas one sub-period. Based on he magniude of he sum of he bea coefficiens and he (1 vs. 3 ) saisic, invesors in Hong Kong and Singapore display more overconfiden rading behavior in he pre-crisis period, while invesors in Japan, Thailand, Indonesia, he Philippines, and Taiwan display more overconfiden rading behavior in he pos-crisis period. I is noed ha alhough we find no evidence on Hong Kong and Singapore invesors overconfiden rading behavior due o domesic marke gains afer conrolling for he paricipaion effec in Table II, hese invesors do rade wih overconfidence in he sub-period analysis. Second, he sum of he gamma coefficiens reveals ha invesors in Hong Kong, Malaysia, Singapore, Thailand, and Korea rade wih overconfidence afer U.S. marke gains only in he pos-crisis period, which is consisen wih our conecure. However, invesors in Japan rade wih overconfidence afer U.S. marke gains only in he pre-crisis period. I is worh poining ou here ha we find evidence from Table III ha U.S. marke gains make invesors in Japan and Thailand rade wih overconfidence in he pre- and pos-crisis period, respecively, which is no found from he resuls of Table II for he whole sample period. Third, we compare he degrees of overconfiden rading of versus invesors in pre- and pos-crisis periods. In he pre-crisis period, invesors in Hong Kong, Malaysia, and Singapore and invesors in China, Korea, and Taiwan rade wih overconfidence afer domesic marke gains. The average of he sum of he 1 coefficiens across hree markes is smaller han ha of he 1 coefficiens across hree markes. We use he (1 ) saisic o es he null hypohesis ha he average of he sum of he 1 coefficiens across hree markes is equal o ha of he 1 coefficiens across hree markes. The resuls show ha he null hypohesis is no reeced, signifying ha he degrees of versus invesors overconfiden rading associaed wih domesic marke gains in he pre-crisis period are no significanly differen. In he pos-crisis period, invesors in Japan, Malaysia, and Thailand and invesors in all markes rade wih overconfidence afer domesic marke gains. The average of he sum of he and 3 coefficiens across hree markes is smaller han ha of he and 3 coefficiens across all markes. The 17

19 saisic, which is used o es he null hypohesis ha he average of he sum of he ( 3 ) and 3 coefficiens across hree markes is equal o ha of he and 3 coefficiens across all markes, is reeced a he 5% level. This provides evidence ha invesors, on average, end o rade wih more overconfidence afer domesic marke gains han invesors in he pos-crisis period. Fourh, in he pre-crisis period, only invesors in Japan rade wih overconfidence afer U.S. marke gains, which implies ha invesors rade more overconfidenly afer U.S. marke gains han invesors in his period. We find ha Japanese invesors rade in U.S. socks in he pre-crisis period much more han invesors in he oher sian counries. nd his may be one reason for he finding ha only Japanese invesors overconfiden rading is affeced by U.S. marke gains in his period. In he pos-crisis period, invesors in all markes, excep for Japan, and invesors in Korea rade wih overconfidence afer U.S. marke gains. The average of he sum of he k and 3 k coefficiens across four markes is slighly larger han ha of he k and 3 k coefficiens for Korea. However, a formal es based on he saisics, which is used o es he null hypohesis ha he (k 3 k) average of he sum of he k and 3 k coefficiens across four markes is equal o ha of he k and 3 k coefficiens for Korea, shows ha here is no significan difference beween he degrees of versus invesors overconfiden rading promped by U.S. marke gains in he pos-crisis period. 3.3 The Impac of U.S. Invesor Senimen on sian Invesors Trading Opimisic invesors end o be overconfiden (e.g., Odean, 1998b; Hirshleifer, 001; Wang, 001). Baker and Sein (004) heoreically show ha when shoring is relaively cosly, senimenal invesors are inclined o become overconfiden and rade more acively when hey are opimisic. Baker, Wurgler, and Yuan (010) presen evidence ha senimen is conagious across markes and herefore has local and global componens. If high senimen spreads from U.S. o sian invesors, hen sian invesors will feel more opimisic abou heir domesic and U.S. sock markes, and consequenly become overconfiden in heir abiliy o profi from heir rading in domesic markes. This suggess ha sian invesors will rade more aggressively afer domesic and U.S. marke gains when global senimen is high. Unforunaely, we can no explore his issue because we are unable o consruc a global 18

20 senimen index as Baker e al. (010) due o daa limiaion. 18 n alernaive approach we can ake is o see how U.S. senimen affecs sian invesors overconfiden rading. We argue ha even hough senimen does no necessarily spillover from U.S. o sian invesors, high U.S. senimen migh also make sian invesors rade more acively afer U.S. marke gains. This is because, as argued by Baker e al. (010) ha opimism leads socks o be overvalued a leas in he shor run, high U.S. senimen implies ha U.S. invesors feel opimisic abou fuure price movemens in U.S. sock markes. Given he predicive power of U.S. sock reurns for sian domesic sock reurns, sian invesors migh also have he same feeling and expec ha sock prices go up in heir domesic markes when hey are aware of U.S. invesors opimism. This would lead hem o furher expec ha hey can gain from heir domesic and U.S.-based equiy invesmens and herefore rade more aggressively in domesic markes. To evaluae he above argumen, we use he SUR model o examine he relaion beween pas U.S. reurns and curren sian domesic volume condiional on U.S. invesor senimen, proxied by Baker and Wurgler s (006) monhly U.S. orhogonalized senimen index ha filers ou business cycle effecs. 19 In his model specificaion, we define high U.S. invesor senimen as he senimen index included in he op 30% of is disribuion. To save space, we do no repor he esimaion resuls of his es, bu hey are available from he auhors upon reques. The mos imporan observaion from his es is ha consisen wih our expecaion, we find evidence ha invesors in Malaysia, Thailand, and Taiwan rade more aggressively afer U.S. marke gains only when U.S. invesor senimen is high. We also find ha Hong Kong invesors rade more frequenly subsequen o U.S. marke gains no maer wheher U.S. invesor senimen is high or no. Sill, we find no evidence ha here is a significan difference beween he degrees of versus invesors in heir overconfiden rading induced by U.S. marke gains when U.S. invesor senimen is high. 3.4 The Impac of Domesic Invesor Senimen on sian Invesors Trading The consumer confidence index is commonly used as a measure of invesor senimen in he 18 We also ry o use Baker and Wurgler s (006) monhly U.S. orhogonalized senimen index as a proxy for global invesor senimen since Baker e al. (010) find ha boh U.S. and U.K. invesor senimen conribue mos o heir global senimen index. However, we do no find evidence ha sian invesors rade more acively afer domesic marke gains when he index value is high han when i is low. 19 The daa on he monhly U.S. orhogonalized senimen index are obained from Jeffrey Wurgler s websie a Since he senimen daa are available up o December 007, he sample period of our analysis here is from January 1995 o December

21 lieraure (e.g., Jansen and Nahuis, 003; Brown and Cliff, 005; Lemmon and Porniaguina, 006; noniou, Doukas, and Subrahmanyam, 011). long he line of he above argumen, we would expec o see ha sian invesors will rade more overconfidenly when invesor senimen proxied by he consumer confidence index is high. To explore his issue, we collec he daa on he consumer confidence index for China, Hong Kong, Indonesia, Japan, Korea, and Thailand from he Daasream Inernaional daabase and for Taiwan from he Taiwan Economic Journal (TEJ) daabase. 0 Because only seven sample counries have he daa on he consumer confidence index and heir sample periods are differen from each oher, we use he regression model, insead of he SUR model, for our analysis. Tha is, we regress domesic volume on pas domesic reurns condiional on domesic invesor senimen for each of hese counries. In his regression, he condiional variable is a dummy variable ha akes on a value of one if he consumer confidence index is included in he op 30% of is disribuion, and zero oherwise. The regression resuls, as expeced, show ha invesors in Hong Kong, Indonesia, Japan, and Thailand rade wih more overconfidence afer domesic marke gains when domesic invesor senimen is high han i is no. These resuls provide corroboraing evidence for our earlier argumen ha opimisic invesors end o rade wih overconfidence. Bu he resuls also show ha here is no significan difference in he overconfiden rading of invesors in China, Korea, and Taiwan across invesor senimen saes. We do no repor hese resuls o conserve space, bu hey are available upon reques from he auhors. 3.5 sian Invesors Overconfiden Trading Condiional on Marke Condiions n old Wall Sree adage, Don confuse brains wih a bull marke, provides invesors wih he bes warning agains becoming overconfiden during a bull marke. GO argue ha overconfiden invesors are more likely o rade aggressively and speculaively righ afer a bull marke (see also Odean, 1998b; Daniel e al., 001). 1 This implies ha if invesors rade wih overconfidence, hey rade more during bull markes han during non-bull markes, which implies ha he posiive relaion beween lagged sock reurns and curren rading 0 The sample period sars from January 1995 for China, from he firs quarer of 000 for Hong Kong, from June 000 for Indonesia, from November 001 for Japan, from December 1998 for Korea, from July 1999 for Thailand, and from January 1999 for Taiwan. ll sample periods end in December I should be noed ha he GO argumen means ha invesors rade overconfidenly boh in bull and non-bull markes, bu hey rade more overconfidenly in bull markes han in non-bull markes (see also Chuang and Lee, 006; Chuang and Susmel, 011) 0