Guglielmo Maria Caporale Brunel; University. Abstract

Transcription

1 Herding behaviour in exreme marke condiions: he case of he Ahens Sock Exchange Guglielmo Maria Caporale Brunel; Universiy Foini Economou Universiy of Piraeus Nikolaos Philippas Universiy of Piraeus Absrac This paper examines herd behaviour in exreme marke condiions using daa from he Ahens Sock Exchange. We es for he presence of herding as suggesed by Chrisie and Huang (1995) and Chang, Cheng, and Khorana (000). Resuls based on daily, weekly and monhly daa indicae he exisence of herd behaviour for he years Evidence of herd behaviour over daily ime inervals is much sronger, revealing he shor-erm naure of he phenomenon. When he esing period is broken ino semi-annual sub-periods, herding is found during he sock marke crisis of Invesor behaviour seems o have become more raional since 00, owing o he regulaory and insiuional reforms of he Greek equiy marke and he inense presence of foreign insiuional invesors. Ciaion: Caporale, Guglielmo Maria, Foini Economou, and Nikolaos Philippas, (008) "Herding behaviour in exreme marke condiions: he case of he Ahens Sock Exchange." Economics Bullein, Vol. 7, No. 17 pp Submied: November 1, 008. Acceped: November 16, 008. URL: hp://economicsbullein.vanderbil.edu/008/volume7/eb-08g1001a.pdf

2 HERD BEHAVIOUR IN EXTREME MARKET CONDITIONS: THE CASE OF THE ATHENS STOCK EXCHANGE Guglielmo Maria Caporale a*, Foini Economou b and Nikolaos Philippas b a Cenre for Empirical Finance, Brunel Universiy, Wes London, UB8 3PH, UK b Deparmen of Business Adminisraion, Universiy of Piraeus, 80 Karaoli and Dimiriou S., Piraeus, Greece *Corresponding auhor. Guglielmo-Maria.Caporale@brunel.ac.uk Ocober 008 Absrac This paper examines herd behaviour in exreme marke condiions using daa from he Ahens Sock Exchange. We es for he presence of herding as suggesed by Chrisie and Huang (1995) and Chang, Cheng, and Khorana (000). Resuls based on daily, weekly and monhly daa indicae he exisence of herd behaviour for he years Evidence of herd behaviour over daily ime inervals is much sronger, revealing he shor-erm naure of he phenomenon. When he esing period is broken ino semi-annual sub-periods, herding is found during he sock marke crisis of Invesor behaviour seems o have become more raional since 00, owing o he regulaory and insiuional reforms of he Greek equiy marke and he inense presence of foreign insiuional invesors. Keywords: Herd behaviour, Ahens Sock Exchange, Cross-secional dispersion of reurns JEL classificaion code: G15 1

3 I. Inroducion Herding can be defined as muual imiaion leading o a convergence of acion (Hirshleifer and Teoh, 003). A growing body of lieraure analyses herding in he sock marke using measures of dispersion around he marke reurn during periods of significan changes in sock prices (Chrisie e al., 1995; Chang e al., 000; Caparrelli e al., 004; Tan e al., 008 ec). The raionale is ha if during hese periods of marke pressure movemens of sock reurns have he endency o be more clusered, his is evidence ha here is co-movemen of sock prices which is independen of heir fundamenal characerisics. According o Chrisie and Huang (1995), hese periods are paricularly informaive because a herd is more likely o form under condiions of marke sress, when individual invesors end o suppress heir own beliefs and follow he marke consensus. Crosssecional dispersions of reurns are prediced o be low when herd behaviour is presen. II. The Case of he Ahens Sock Exchange The Ahens Sock Exchange (ASE) was founded in Unil 1987, he ASE was almos exclusively a domesic marke. In recen years he Greek equiy marke has undergone imporan regulaory and insiuional reforms wih significan growh in size and liquidiy. Moreover, here has been increased aenion from he inernaional invesors since 001, when Morgan Sanley gave ASE he saus of a developed marke. The Greek equiy marke provides an ineresing seing for he analysis of herd behaviour, especially during he period of crisis of he lae 1990 s. Beween 1/1/1998 and 17/9/1999 (hisoric high) sock prices increased by 33.69%. The sock marke crash ha followed resuled in a

4 fall in he ASE general index o in 31/3/003, i.e. a 77.88% loss. The crash shook invesor confidence and had major economic implicaions, especially for individual invesors. Even now, some domesic invesors coninue o be uncerain abou growh prospecs for he ASE. The impressive rise of he sock marke ha ensued araced housands of new invesors, especially domesic and small ones (Fig. 1). In Sepember 1999 here were new invesor shares. The sharp upward and downward movemens in reurns could hen be aribued o herd behaviour, reflecing he massive rading of new, inexperienced and uninformed individual invesors. The presen paper examines he sock marke crisis of 1999 in he ASE providing a herding explanaion, which has no been done before. III. Mehodology and Daa Chrisie and Huang (1995) and Chang, Cheng and Khorana (000) argued ha herding can be analysed using cross-secional mehods for asse reurns, since a smaller cross-secional dispersion of reurns indicaes parallel movemen wih he cross-secional mean reurn, ha is o say movemen owards some ype of marke consensus. They used such measures o deec herding in special periods of exreme upward or downward movemens in reurns. However, i is no necessarily he case ha herding is only presen during such periods of urmoil, and resricing he analysis o hem does no allow one o sudy how his phenomenon migh evolve over ime. The ype of herding ha hese paricular measures examine is markebased and is quie differen from he usual definiion of herding, which refers o subgroups of invesors behaving similarly, buying or selling he same asses 3

5 simulaneously. In he laer case individual sock reurns follow he marke reurn. Boh definiions are valid, hough, and can lead o asse mispricing. Chrisie and Huang (1995) esimaed he cross-secional sandard deviaion (CSSD) of single sock reurns wih respec o marke reurns, which is expressed as: N ( Ri, Rm, ) i= 1 CSSD = N 1 (1) where R i, is he observed sock reurn of firm i a ime, R m, is he crosssecional average reurn of he N reurns in he marke porfolio a ime, and N is he number of socks in he marke porfolio. The CSSD of reurns was hen regressed agains a consan and wo dummies in order o idenify he exreme marke phases, wih D L = 1 if he marke reurn on day lies in he exreme 1% and 5% lower ail of he disribuion of marke reurns (and zero oherwise), and D U = 1 if i lies in he exreme 1% and 5% upper ail of he same disribuion (and zero oherwise): CSSD D + e L = a + b1 D + b U () where he α coefficien denoes he average dispersion of he sample excluding he regions corresponding o he wo dummy variables. According o his model, saisically significan negaive values for b 1 and b indicae he presence of herd behaviour. When individual reurns herd around he marke consensus, dispersions are prediced o be relaively low. By conras, raional asse pricing models predic an increase in dispersion because individual asses differ in heir sensiiviy o he marke reurn. 4

6 Alhough he cross-secional sandard deviaion of reurns is an inuiive measure for capuring herding, i can be considerably affeced by he exisence of ouliers. Tha is why Chrisie and Huang (1995) as well as Chang, Cheng and Khorana (000) proposed he use of he cross-secional absolue deviaion, (CSAD), as a beer measure of dispersion: CSAD N R N R i, m, i= = 1 (3) where R i, is he observed sock reurn of firm i a ime, R m, is he crosssecional average reurn of N socks in he porfolio a ime, and N is he number of socks in he porfolio. The equaion for he CSAD corresponding o Equ. () is he following: CSAD + L U = a + b1 D + bd e (4) Chang, Cheng and Khorana (000) proposed an alernaive approach o he one suggesed by Chrisie and Huang (1995), using he enire disribuion of marke reurns, as in he following equaion: CSAD = a + γ + 1 Rm, + γ Rm, ε (5) The relaionship beween CSAD and R m, is used o deec herd behaviour. According o Chang, Cheng, and Khorana (000), in he presence of herding he relaionship beween CSAD and he average marke reurn is nonlinear. A significanly negaive coefficien γ implies he presence of herd behaviour. This is likely o increase he correlaion among individual asse reurns, and he dispersion among asse reurns will eiher decrease or increase a a decreasing rae. If marke paricipans are more likely o herd during periods of large price movemens, hen here should be a less han proporional increase (or decrease) in he CSAD measure. In he absence of herding, he relaionship is 5

7 linear and increasing, ha is he dispersion increases proporionaely wih he increasing reurns of he marke. Moreover, he relaionship beween CSAD and marke reurns may be asymmeric. This hypohesis can be esed using wo differen models: UP UP UP UP UP CSAD α + γ1 R m, + γ (R m, ) + = ε, if R m, >0 (6) DOWN DOWN DOWN DOWN DOWN CSAD α + γ1 R m, + γ (R m, ) + = ε, if R m, <0 (7) UP DOWN The absolue values R and R are used because we are m, m, concerned wih he size of he reurn, no wih is sign. This also makes a UP DOWN comparison beween and possible. γ γ In order o examine herd behaviour in he ASE, we use daily, weekly, and monhly sock percenage log-differenced reurns in our herding ess from January 1, 1998, o December 31, 007. The daa include all he socks available on day (minimum 15 socks, maximum 337 socks). IV. Empirical resuls Table 1 conains summary saisics for CSSD, CSAD and he equally weighed marke reurn R m,. Tables and 3 repor he regression resuls for CSSD and CSAD respecively. For all he crieria used he esimaed coefficiens for b L and b U are posiive and saisically significan, suggesing he absence of herding during he period under invesigaion. However, using he mehodology of Chang, Cheng, and Khorana (000), resuls based on daily, weekly and monhly daa indicae he exisence of herd behaviour over he whole period. Table 4 repors he oal marke regression resuls. The coefficien γ 1 is posiive and saisically significan for all he ime inervals, confirming he resuls of he es wih he dummy (CSAD 6

8 increases wih R m, ). The coefficien γ is negaive and saisically significan, suggesing he exisence of herding, i.e. as he average marke reurn becomes large in absolue erms, cross-secional reurn dispersion increases a a decreasing rae (Fig. ). The evidence of herding over daily ime inervals is much sronger, revealing he shor-erm naure of he phenomenon. Table 5 repors he herding regression resuls under asymmeric marke condiions. The resuls sugges ha herding is sronger during periods of a rising marke. In fac, when using weekly and monhly ime inervals, here is evidence of herding only during such periods. The coefficiens indicae ha, beyond a cerain hreshold poin, he CSAD may decline as R m, increases. When subsiuing he esimaed coefficiens for he rising marke ( UP γ UP γ 1 = and = using daily daa) ino Equ. (6), i is calculaed ha CSAD reaches a maximum when R m, =1,791%. Similarly, using he values for he down marke DOWN DOWN ( = and γ = ) and daily daa, i is found ha CSAD γ 1 reaches a maximum when R m, = -3,407%. This suggess ha for large swings in he marke reurn, above hese hreshold poins, CSAD has a endency o decrease. To invesigae he presence of herding during he sock marke crisis of 1999, we break he esing period ino semi-annual sub-periods, using daily daa. Table 6 repors he resuls. There is evidence of herding during he sock marke crisis of 1999, since he coefficien γ is negaive and saisically significan during he whole 1999 and he firs semesers of 000 and 001. However, in several semesers he sign of he coefficien γ is posiive and saisically significan (1 s semeser 00, 004, nd semeser 005, 006). In his case individual reurns end o diverge from he marke reurns (ani-herding). 7

9 V. Conclusions This paper has examined he exisence of herd behaviour in he ASE. Resuls based on daily, weekly and monhly daa indicae ha i is indeed presen. Evidence of herding over weekly and monhly ime inervals is much weaker, suggesing ha i is a shor-erm phenomenon. These resuls are similar o hose repored by Tan e. al. (008) for he Chinese sock markes. Moreover, herding is found o be sronger during periods of rising markes. Anoher ineresing aspec is he idenificaion of hreshold marke reurns, above which herd behaviour is more likely o occur. We show he exisence of significan herd behaviour boh during and afer he sock marke crisis of However, invesor behaviour seems o have become more raional since 00, as he esimaes for several subsequen semesers are consisen wih raional behaviour. This change in invesmen behaviour can be aribued o he regulaory and insiuional reforms of he Greek equiy marke and he diffuse presence of foreign insiuional invesors (Fig. 3). These findings have imporan implicaions for sock marke efficiency and offer an ineresing insigh ino he behaviour of Greek invesors. Regarding fuure research, i would be ineresing o adjus he individual sock reurns for hin rading, since he rading volume is exremely low in many cases. Acknowledgemens Foini Economou wishes o acknowledge financial suppor from he Alexander S. Onassis Public Benefi Foundaion, under he Scholarships Programme. 8

10 References Bikhchandani, S. and Sharma, S. (001) Herd Behavior in Financial Markes, IMF Saff Papers, Inernaional Moneary Fund, 47, Caparrelli, F., D Arcangelis, A.M., Cassuo, A. (004) Herding in he Ialian Sock Marke: A Case of Behavioral Finance, The Journal of Behavioral Finance, 5, 30. Chang, E. C., Cheng, J. W. and Khorana, A. (000) An examinaion of herd behavior in equiy markes: an inernaional perspecive, Journal of Banking and Finance, 4, Chrisie, W. G. and Huang, R. D. (1995) Following he pied piper: do individual reurns herd around he marke?, Financial Analyss Journal, 51, Hirshleifer, D. and Teoh, S. H. (003) Herd Behaviour and Cascading in Capial Markes: a Review and Synhesis, European Financial Managemen, 9, Kallinerakis, V. and Kraunova, T. (007) Does Thin Trading Impac Upon he Measuremen of Herding? Evidence from Bulgaria, Social Science Research Nework. Newey, W. K. and Wes, K. (1987) A simple posiive semi-definie, heeroskedasiciy and auocorrelaion consisen covariance marix, Economerica, 55, Tan, L., Chiang, T. C., Mason, J. R. and Nelling, E. (008) Herding behavior in Chinese sock markes: An examinaion of A and B shares, Pacific-Basin Finance Journal, 16,

11 Table 1. Descripive saisics Daily daa Weekly daa Monhly daa ( ) CSSD CSAD R m, CSSD CSAD R m, CSSD CSAD R m, Observaions,501,501, Minimum Maximum Mean Median Sandard deviaion Table. Regression resuls for CSSD - Chrisie and Huang (1995) L U CSSD a + β D + D + e = 1 β Daily daa Weekly daa Monhly daa 1% crierion 5% crierion 1% crierion 5% crierion 1% crierion 5% crierion α -saisic (75.08)* (76.47)*.817 (35.13)*.738 (37.04)* (0.99)* (0.0)* β 1 -saisic (6.09)* (10.57)* (4.35)* (5.06)* (.18)** (4.70)* β -saisic (1.01) (5.80)* (5.7)* (9.1)* (11.59)*.9367 (9.30)* Adjused R Noes: Numbers in parenheses are -saisics based on Newey Wes (1987) consisen sandard errors. *, ** and *** sand for saisical significance a he 1%, 5%, and 10% levels, respecively. Table 3. Regression resuls for CSAD - Chrisie and Huang (1995) L U CSAD a + β D + D + e α -saisic β 1 -saisic β = 1 β Daily daa Weekly daa Monhly daa 1% crierion 5% crierion 1% crierion 5% crierion 1% crierion 5% crierion (66.88)* (69.58)* (3.43)* (34.85)* (19.07)* (18.78)* (4.59)* (15.7)* (4.66)* (6.01)* (3.1)* (6.33)* (1.78)*** (11.08)* (4.70)* (11.8)* (11.0)* (10.07)* -saisic Adjused R Noe: See noes for Table. 10

12 Table 4. Toal marke regression resuls - Chang, Cheng & Khorana (000) CSAD = a + γ R + γ R + ε 1 Daily daa Weekly daa Monhly daa α -saisic (57.60)* (5.84)*.8468 (5.07)* γ 1 -saisic (13.64)* (8.73)* (6.8)* γ -saisic (-6.43)* (-.01)** (-1.76)*** Adjused R Noe: See noes for Table. m, m, Table 5. Up & Down marke regression resuls - Chang, Cheng & Khorana (000) Panel A: Up marke regression resuls UP UP UP UP UP CSAD = a + γ R + γ (R ) + ε,r m, >0 α -saisic UP γ 1 -saisic Daily daa (1,306 observaions) (45.97)* (13.51)* 1 m, m, Weekly daa ( 66 observaions) (19.85)* (8.85)* Monhly daa (64 observaions).834 (0.17)* (6.05)* UP γ saisic (-8.81)* (-3.71)* (-.45)** Adjused R Panel B: Down marke regression resuls DOWN DOWN DOWN DOWN DOWN CSAD = a + γ R + γ (R ) + ε, R m, <0 α -saisic DOWN γ 1 -saisic 1 m, Daily daa (1,195 observaions) (44.13)* (9.96)* m, Weekly daa (55 observaions) (4.15)* (7.9)* Monhly daa (56 observaions).7814 (15.41)* (5.4)* DOWN γ saisic (-3.6)* (0.09) (-1.43) Adjused R UP DOWN Panel C: Wald Tes H 0 : γ - γ = 0 UP DOWN γ - γ saisic (-5.45)* Chi-square [9.65]* Noe: See noes for Table (-3.35)* [10.47]* (-1.11) [1.4] 11

13 Table 6. Semi-annual regression resuls - Chang, Cheng & Khorana (000) Adjused α saisic γ 1 saisic γ saisic R-squared Α * * % Β * * % Α * * * 17.34% Β * * * 5.7% Α * * * 18.87% Β * * % Α * * ** 39.63% Β * * % Α * * 76.05% Β * * % Α * * % Β * * % Α * * ** 69.6% Β * *** * 59.61% Α * * % Β * * * 5.39% Α * * ** 69.75% Β * * 48.5% Α * * % Β * * % Noe: See noes for Table. 160,000 7,000 New invesor shares 140,000 10, ,000 80,000 60,000 40,000 0,000 6,000 5,000 4,000 3,000,000 1,000 ASE General Index /1/ /11/ /9/ /7/000 31/5/001 31/3/00 31/1/003 30/11/003 30/9/004 31/7/005 31/5/006 31/3/007 New invesor shares ASE General Index Fig. 1. ASE General Index and new invesor shares (monhly daa, ). Sources: Thomson Daasream, ASE Monhly Saisics Bulleins 1

14 .5 CSAD (%) Rm (%) Fig.. Relaionship beween he daily cross-secional absolue deviaion (CSAD ) and he corresponding equally-weighed absolue marke reurn ( R m. ) for he ASE (January 1998-December 007). (%) Jan-0 May-0 Sep-0 Dec-0 May-03 Jul-03 Sep-03 Dec-03 Mar-04 Jun-04 Sep-04 Dec-04 Mar-05 Jun-05 Sep-05 Dec-05 Mar-06 Jun-06 Sep-06 Dec-06 Mar-07 Jun-07 Sep-07 Dec-07 Foreign Insiuional Invesors Percenage Upon Capializaion Toal Capializaion 50,000 00, , ,000 50,000 Fig. 3. Toal marke capializaion in he ASE and foreign insiuional invesors percenage upon capialisaion (quarerly daa, ). Source: ASE Monhly Saisics Bulleins 0 Toal Capializaion (in millions ) 13