Kent Academc Repostory Full text document (pdf) Ctaton for publshed verson Economou, Fotn and Katskas, Epamenondas and Vckers, Gregory (2016) Testng for herdng n the Athens Stock Exchange durng the crss perod. Fnance Research Letters, 18. pp. 334-341. ISSN 1544-6123. DOI https://do.org/10.1016/j.frl.2016.05.011 Lnk to record n KAR http://kar.kent.ac.uk/55491/ Document Verson Author's Accepted Manuscrpt Copyrght & reuse Content n the Kent Academc Repostory s made avalable for research purposes. Unless otherwse stated all content s protected by copyrght and n the absence of an open lcence (eg Creatve Commons), permssons for further reuse of content should be sought from the publsher, author or other copyrght holder. Versons of research The verson n the Kent Academc Repostory may dffer from the fnal publshed verson. Users are advsed to check http://kar.kent.ac.uk for the status of the paper. Users should always cte the publshed verson of record. Enqures For any further enqures regardng the lcence status of ths document, please contact: researchsupport@kent.ac.uk If you beleve ths document nfrnges copyrght then please contact the KAR admn team wth the take-down nformaton provded at http://kar.kent.ac.uk/contact.html
Testng for herdng n the Athens Stock Exchange durng the crss perod Fotn Economou a, Epamenondas Katskas b and Gregory Vckers c a Correspondng author. Centre of Plannng and Economc Research and Hellenc Open Unversty, 11, Amerks str. 106 72 Athens, Greece, Tel: (+30) 210 3676418, emal: feconom@kepe.gr. b Kent Busness School, Unversty of Kent, Parkwood Road, Canterbury, Kent, CT2 7PE, UK, emal: E.Katskas@kent.ac.uk. c Durham Unversty Busness School, Mll Hll Lane, Durham DH1 3LB, UK, emal: gregory.vckers@durham.ac.uk. Abstract Ths paper nvestgates herdng behavor n the Athens Stock Exchange focusng on the recent crss perod. We employ a survvor bas free dataset of all lsted stocks from 2007 to May 2015. We apply the cross sectonal dsperson approach and provde results that extend and are comparable wth prevous studes regardng the Greek stock market. The emprcal results ndcate the presence of herdng under dfferent market states. Employng the quantle regresson method, there s herdng n the hgh quantles of the cross sectonal return dsperson. Fnally, we document the mpact of sze effect on herdng estmatons. Keywords: herdng; cross sectonal dsperson; Athens Stock Exchange JEL Classfcaton: G10; G14; G15
1. Introducton Crses and perods of extreme market condtons facltate market anomales and devatons from the Effcent Market Hypothess. Under these crcumstances a herd,.e. a crowd convergng n ts actons and belefs (Hrshlefer and Teoh, 2003), s more lkely to form havng mportant mplcatons for portfolo dversfcaton and market stablty (Chang et al., 2000; Demrer and Kutan, 2006; Chang and Zheng, 2010; Economou et al., 2011). Despte the lack of conclusveness n the emprcal results both n emergng and developed markets, herdng s expected to be more pronounced under extreme market condtons (Chrste and Huang, 1995; Chang et al., 2000; Chang and Zheng, 2010; Economou et al., 2011) when ndvdual nvestors are more lkely to follow the crowd nstead of ther own belefs/knowledge (Chrste and Huang, 1995). Mobarek et al. (2014) provde evdence of sgnfcant herdng effects n varous European stock markets durng the global fnancal crss and the Eurozone crss, whle Peltomäk and Vähämaa (2015) document that herdng effects n the EMU markets affected herdng n the non-emu markets from September 2008 to January 2014. The Greek stock market provdes an nterestng settng for analyss due to the unprecedented debt crss that occurred n recent years and the potental spllover effects on other Eurozone markets. Ths paper nvestgates herdng behavor n the Athens Stock Exchange (ASE) focusng on the recent crss perod. To ths end we employ a survvor bas free dataset from January 2007 to May 2015. We apply the cross sectonal dsperson approach and provde results that extend and are comparable wth prevous studes regardng the Greek stock market. Caporale et al. (2008) were the frst to nvestgate herdng n the ASE from 1998 to 2007. The authors dentfed evdence of herdng whch s much stronger usng daly nstead of weekly or monthly data. Moreover, herdng was more pronounced durng rsng market days beng also present durng the stock market bubble of 1999. Tessaromats and Thomas (2009) also confrmed strong evdence of herdng for the perod 1998-2004. Herdng n the ASE has been extensvely examned by Economou et al. (2011) for the perod 1998-2008, testng for potental herdng asymmetres wth reference to dfferent market states as well as for cross market effects n four South European stock markets,.e. Greece, Italy, 1
Span and Portugal. The authors provde evdence of herdng that s more pronounced on days wth postve market returns, whle there s no evdence of asymmetres regardng tradng volume and stock market volatlty. Mobarek et al. (2014) examned a large number of European stock markets from 2001 to 2012 and dentfed herdng n Greece durng the Eurozone crss (from May 2010 to February 2012). Ther dataset dffers from prevous studes snce t only ncludes the ATHEX Composte consttuent stocks nstead of all lsted stocks n the ASE. In ths paper we extend the work of Economou et al. (2011) for the recent Greek debt crss perod. Our emprcal results ndcate the presence of herdng under dfferent market states. These fndngs provde nsght nto nvestors behavor, especally n the lght of the unprecedented events of the Greek crss and are n lne wth the man fndngs of prevous studes that dentfy herdng n the ASE. The rest of the paper s organzed as follows: Secton 2 presents the dataset and methodology employed, Secton 3 reports the emprcal results and Secton 4 concludes. 2. Methodology and Data Chrste and Huang (1995) and Chang et al. (2000) proposed a cross sectonal dsperson approach to capture herdng, employng the cross sectonal dsperson of ndvdual asset returns as follows: (1) where s the return of stock on day t, s the stock market return on day t and N s the number of all lsted stocks n the stock market on day t. The non-lnear model proposed by Chang et al. (2000) estmates the relatonshp between the CSAD and the stock market return n order to capture herdng as follows: (2) Under ratonal asset prcng models, ths relatonshp s expected to be postve and lnear,.e. under extreme market condtons the CSAD s expected to ncrease snce the ndvdual stocks dffer n senstvty to the stock market returns. If herdng effects are present ths relatonshp s non lnear and coeffcent s expected to be negatve and statstcally sgnfcant. The Chang et 2
al. (2000) model s qute nfluental n the aggregate data studes of herd behavor. Chang and Zheng (2010) proposed an adaptaton of ths model addng to the standard equaton, whch permts the nterpretaton of asymmetrc effects by estmatng a sngle model, whch s more streamlned than the ntal regresson of Chang et al. (2000). It also permts greater analyss of the asymmetres present n up and down markets and t s specfed as follows: (3) In equaton (3), the relatonshp between return dsperson and stock market return s captured by when market returns are postve, and by when they are negatve or zero. Thus, the asymmetrc relatonshp between stock return dsperson and stock market return can be presented by the rato (Duffee, 2001). Followng Chang et al. (2000), s used to dentfy a non-lnear relatonshp and a negatve and statstcally sgnfcant coeffcent wll ndcate the presence of herdng. Apart from the tradtonal OLS method, we also employ the quantle regresson method followng Chang et al. (2010) and Zhou and Anderson (2010). Ths s a popular approach, orgnally ntroduced by Koenker and Bassett (1978). In ths case we examne the coeffcents of model (3) for dfferent quantles of the dependent varable. 1 The -th condtonal quantle functon of the dependent varable dstrbuton s defned as follows: QY ( / x) x (4) where Y s a dependent varable, x s a vector of ndependent varables and s a vector of coeffcents. The ˆ ( quantle ) estmator results from the followng weghted mnmzaton: n ˆ ( ) arg mn ( y x ) (5) quantle 1 where s a weghtng factor, also called check functon. For any (0,1 ) a weghtng functon s defned as follows: 1 See Koenker (2005) for a more techncal presentaton of the method. 3
( quantle u ) { u ( 1) u f u 0 f u 0 (6) where u y x. From equatons (5) and (6) we get the quantle regresson estmator by mnmzng the weghted sum of absolute errors, where the weghts depend on the quantle under examnaton as follows: ˆ ( quantle ) arg mn y x (1 ) y x (7) : y x : y x Furthermore, consderng the evdence avalable regardng herdng asymmetres we examne the relatonshp more formally, through the mplementaton of a seres of dummy varables n lne wth both Chang and Zheng (2010) and Economou et al. (2011). Ths method s more robust compared to examnng the relatonshp usng two dfferent regressons, as n prevous studes (see Tan et al., 2008 among others). In ths case the model s structured as follows: (8) where s negatve, and, otherwse. The hypothess of asymmetrc herdng s examned usng equalty tests of pars of up and down market coeffcents (.e. and and and ) by subtractng the coeffcent of the down markets from up markets and testng f the result s equal to zero. If herdng s present then we expect coeffcents and to be negatve. The relatve magntudes of coeffcents and wll demonstrate any asymmetrc herdng effects. If herdng s more pronounced on days when the market s down, then we expect <. Moreover, dummy varables are assgned to days of hgh/low market tradng volume. A day of hgh (low) tradng volume s when the value of the traded stocks on that day s above (below) the prevous 30-day movng average. The respectve model specfcaton follows: (9) where f hgh tradng volume on that day, and, otherwse. If herdng s present then we expect coeffcents and to be negatve. The relatve magntudes of coeffcents 4
and wll demonstrate any asymmetrc herdng effects. If herdng s more pronounced on days wth hgh average value of total tradng volume, then we expect <. Moreover, accordng to Chrste and Huang (1995), herdng s more lkely to appear durng perods of extreme market movements beng obvously more prevalent durng market crss perods. Economou et al. (2011) also address the potental ssue of hgh market volatlty employng a dummy varable determned by the relatonshp of the day s market volatlty ( ) relatve to the prevous 30-day movng average. The examned regresson s the followng: (10) where f hgh market volatlty that day,, otherwse. If herdng s present then we expect coeffcents and to be negatve. The relatve magntudes of coeffcents and wll demonstrate any asymmetrc herdng effects. If herdng s more pronounced on days wth hgh volatlty, then we expect <. Fnally, we test for possble asymmetrc herdng effects relatve to the soveregn bond spreads. Gven that the euro area soveregn bond yeld dfferentals can be explaned by general rsk averson and ts nteracton wth macroeconomc fundamentals, as well as by domestc factors, especally durng tmes of fnancal stress (Barros et al., 2009), we examne herdng under dfferent market states wth reference to the 10-year Greek bond spread over the German. To ths end we employ a dummy varable that equals to 1 when the value of the spread on day t s above the prevous 30-day movng average. The model s structured as follows: CSAD D R D R D R D R (11) where,, f spread s hgher than the 30-day movng average that day,, otherwse. If herdng s present then we expect coeffcents and to be negatve. We expect that hgh spreads, reflectng greater rsk averson and negatve country-specfc factors, facltate herdng behavor. The relatve magntudes of coeffcents and wll demonstrate any asymmetrc herdng effects. If herdng s more pronounced on days wth hgh spreads, then <. 5
The data employed n ths paper conssts of daly stock prce, market value, and trade value data for the ASE, obtaned from Thomson Reuters Datastream. The stocks ncluded are those wthn the Worldscope Greece stock ndex, whch also ncludes dead stocks. Ths helps us elmnate survvorshp bas. Thus, the number of stocks n the sample ranges from 188 to 309. The date range for the data used s 02/01/2007 to 29/05/2015. Days n whch no tradng was recorded have been manually elmnated. Return s calculated as ln ln (12) and CSAD s calculated as reported n the methodology secton employng both equally and value weghted market 2 returns n the estmatons to account for sze effect n the stock market. 3. Emprcal Results Table 1 presents the descrptve statstcs for the calculated CSAD and market return, both equal and value weghted. A frst pont of nterest s that the mean for both market return varables s negatve as a result of the poor performance of the ASE over the perod under examnaton. The mean return for the equally weghted market return s more negatve than that of the value weghted one suggestng that smaller market value stocks have suffered greater losses. The same holds for CSAD, wth the value of equally weghted CSAD beng much greater than that of the value weghted one suggestng that the dspersons from the market return are lkely to be more prevalent n smaller stocks. The data presents hgh levels of leptokurtoss wth ths close clusterng around the mean and thcker tals meanng that there s a hgh probablty for extreme values. Ths s consstent wth theory, as a large number of extreme values are to be expected durng perods of fnancal nstablty. The decrease n Kurtoss when comparng value weghted to equally weghted returns also ndcates that these extreme results are more lkely to appear n smaller stocks. Table 1. Descrptve statstcs for CSAD and stock market returns Equal Weghted Market Returns Value Weghted Market Returns CSAD R m CSAD R m Mean 1.0991-0.0441 0.1924-0.0043 Medan 1.0494-0.0226 0.1779 0.0009 2 We employ daly data of each stock s market value n order to assgn the weghts to estmate the value weghted market return. These weghts are re-adjusted on a daly bass. 6
Maxmum 3.1332 2.9124 0.7912 0.9979 Mnmum 0.5440-4.8087 0.0360-0.8480 Std. Dev. 0.3114 0.5240 0.0898 0.1682 Skewness 1.1161-0.5972 1.7083-0.0778 Kurtoss 5.4158 9.6710 8.3409 6.4800 Jarque-Bera 942.61 4001.54 3502.27 1057.25 Sum 2298.1850-92.1493 402.3974-9.0277 Sum Sq. Dev. 202.7282 573.850 16.8600 59.1594 Observatons 2091 2091 The results of the emprcal analyss begn wth the standard model (3) n order to test for the presence of herdng effects (usng both equal weghted and value weghted methods of calculatng market returns), and examne for dfferences n herdng behavor between up and down markets. All the results are derved usng a Newey-West (1987) consstent estmator to correct for autocorrelaton and heteroskedastcty. Table 2 presents the results of the basc model employng both equal and value weghted returns. Followng Chang and Zheng (2010), the coeffcent on ( ) detects the presence of nonlnearty n the relatonshp between CSAD and stock market returns. The estmate for coeffcent s negatve and statstcally sgnfcant at the 5% level, ndcatng herdng towards the market return. Coeffcents and are also mportant n the analyss of the model, as the rato s a measure of the relatve amount of asymmetry n the relatonshp. Gven the nsgnfcance of the coeffcent at the 5% level, the value of ths rato s 1, although coeffcent s sgnfcant at the 10% level, where by the rato would be 1.108, showng large, but weakly sgnfcant asymmetry. The adjusted R-squared value ndcates that ths regresson captures 34.09% of the CSAD devaton through these ndependent varables. Table 2. Herdng estmatons Standard model 7
Equal Weghted Market Returns Value Weghted Market Returns 0 0.9048 0.1202 1 0.0285 0.0072 (p-value) 0.0585* 0.2876 2 0.5567 0.5628 3-0.0360 0.1449 (p-value) 0.0250** 0.0050*** Adj-R 2 0.3409 0.6850 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD R R R respectvely.. Daly data from January 2007 to May 2015. ***,**,* statstcally sgnfcant at 1%, 5% and 10% level, The second column n Table 2 tests the same model as the frst column employng the value weghted method to calculate the stock market return n order to elmnate any potental sze bas n the dataset, as smaller frms stocks are known to have greater herdng effects (Lakonshok et al., 1992). As a result, the mpact of these frms s overstated n an equal weghted market return specfcaton. In ths specfcaton, whlst the fnal value for coeffcent s even more sgnfcant than the same value n the equally weghted model, ts value s postve ndcatng absence of herdng. Smaller frms stocks are expected to be more susceptble to herdng due to poorer nformaton flow, and equal weghtng of market returns wll over-estmate the mpact of these effects. Thus, ths emprcal evdence s n lne wth theory as t suggests greater levels of herdng n smaller stocks. In order to evaluate the herdng effect on small captalzaton stocks 3 we reestmate model (3) employng a small captalzaton equty portfolo. To ths end we created 5 quntles based on market value and employed the smallest sze one for our estmatons. The emprcal results presented n equaton (13) confrm our assumptons snce coeffcent s negatve and hgher compared to the results presented n Table 2. All coeffcents apart from are statstcally sgnfcant at 1% level., Adj-R 2 26.52% (13) 3 We would lke to thank an anonymous revewer for makng ths suggeston. 8
Table 3 reports the quantle regresson results. The results employng the equally weghted returns (Panel A) ndcate that herdng s present only n the hgh quantles of the cross sectonal return dsperson. The sgn and statstcal sgnfcance of coeffcent 3 change across quantles, from postve for =10% to negatve for =10%, =25%, =50%, =75% and =90%, wth the results beng statstcally sgnfcant only for =75% and =90%. Ths fndng s n lne wth Zhou and Anderson (2010), who document herdng n the US REITS only n the hgh quantles and attrbute ths behavor to hgh-quantle dsperson beng typcally assocated wth large market prce movements and volatle market condtons. However, when employng the value weghted approach (Table 3, Panel B) there s no evdence of herdng, consstent wth the results of Table 2. In ths case, coeffcent 3 s postve for all quantles and statstcally sgnfcant for =50%, =75% and =90%. Table 3. Quantle Regresson Results for Model 3 =10% =25% =50% =75% =90% Panel A. Equal Weghted Market Returns 0 0.6735 0.7553 0.8728 1.0150 1.1818 0.0000*** 0.0000*** 0.0000*** 1 0.0310 0.0138 0.0520 0.0484 0.0122 (p-value) 0.0837* 0.4488 0.0006*** 0.1580 0.6210 2 0.3156 0.4776 0.5806 0.7113 0.7420 0.0000*** 0.0000*** 0.0000*** 3 0.0167-0.0378-0.0541-0.0557-0.0608 (p-value) 0.4926 0.3902 0.1356 0.0001*** 0.0966* Adj-R 2 0.0922 0.1256 0.1682 0.2024 0.2573 Panel B. Value Weghted Market Returns 0 0.0584 0.0779 0.1103 0.1540 0.1923 0.0000*** 0.0000*** 0.0000*** 1 0.0106 0.0108 0.0086 0.0132 0.0070 (p-value) 0.0178** 0.0713* 0.1340 0.1731 0.6706 2 0.6713 0.6135 0.5509 0.4757 0.4568 0.0000*** 0.0000*** 0.0000*** 3 0.0139 0.0948 0.1879 0.2624 0.3024 (p-value) 0.4508 0.1851 0.0001*** 0.0009*** 0.0001*** Adj-R 2 0.4508 0.4352 0.4183 0.4225 0.4570 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD R R R respectvely.. Daly data from January 2007 to May 2015. ***,**,* statstcally sgnfcant at 1%, 5% and 10% level, 9
The frst column n Table 4 provdes a more n depth examnaton of the asymmetrc relatonshp between the CSAD and the equal weghted stock market return. In ths case, only coeffcent s negatve and statstcally sgnfcant, thus ndcatve of herdng n down markets. Even though Economou et al. (2011) dentfed herdng n up markets n the ASE for the perod 1998-2008, the respectve coeffcent obtaned from our study for days wth negatve market returns ndcates sgnfcant rse n herdng on down market days, Ths fndng could be related to the prolonged exposure to negatve market returns n the ASE over the perod under examnaton. The second column n Table 4 presents the same model as the frst employng the value weghted method to calculate the stock market return. The coeffcents of nterest, and, are 0.1204 and 0.1774 respectvely wthout beng statstcally sgnfcant dfferent from each other (Wald coeffcent test p=0.35). Both are statstcally sgnfcant and postve, ndcatng a lack of evdence of herdng n the value weghted sample. Table 4. Herdng estmatons Market asymmetry Equal Weghted Market Returns Value Weghted Market Returns 0 0.9054 0.1203 1 0.5764 0.5785 2 0.5295 0.5434 3-0.0287 0.1204 (p-value) 0.4265 0.0187** 4-0.0374 0.1774 (p-value) 0.0336** 0.0055*** Adj-R 2 0.3406 0.6850 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD DR D R D R D R, D f R, and D otherwse. Daly data from January 2007 to May 2015. ***,**, statstcally sgnfcant at 1% and 5% level, respectvely. Table 5 presents the herdng behavor estmates usng a dummy varable whch s based on the prevous 30-day movng average of total market tradng volume of the frms lsted n our dataset. Usng the equal weghted market return results for coeffcents and, only coeffcent s sgnfcant at the 1% level, and negatve, ndcatng the presence of herdng durng days where the value of the traded stocks was greater than the movng average. Employng value weghted 10
approach to calculate market returns, the coeffcents of nterest, and change dramatcally, wth coeffcent plummetng to 0.3196 and coeffcent to 0.0872, beng statstcally sgnfcant dfferent from each other (Wald coeffcent test p=0.00). These results are statstcally sgnfcant at 1% and 10% level respectvely demonstratng a lack of herdng n the tradng volume model specfcaton f market weghted returns are used. Table 5. Regresson Results for Tradng Volume Dummy Based Model Equal Weghted Market Returns Value Weghted Market Returns 0 0.9200 0.1216 1 0.4397 0.4846 2 0.59455 0.5954 3-0.0048 0.3196 (p-value) 0.8834 0.0000*** 4-0.0543 0.0872 (p-value) 0.0064*** 0.0762* Adj-R 2 0.3446 0.6885 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD D R D R D R D R, D f hgh tradng volume that day, D otherwse. Daly data from January 2007 to May 2015. ***,* statstcally sgnfcant at 1% and 10% level, respectvely. Table 6 shows the mpact of market return volatlty on the relatonshp between the CSAD and the stock market return n equal weghted and value weghted terms. The results demonstrate sgnfcance of above average daly market return volatlty n the relatonshp between the CSAD and the squared market return. Ths fndng dffers from the ones reported by Economou et al. (2011) that dd not document asymmetrc herd behavor wth reference to market volatlty for the perod 1998-2008. Ths could be attrbuted to the large rse market return volatlty gven the economc turbulence n the Greek market over recent years. As a result, there s potental for herdng to be caused by ths mechansm. Ths ssue certanly needs further examnaton and understandng, especally gven the current Greek soveregn debt crss and the rsks that s exposes the whole European Unon to. However, the asymmetrc herdng behavor dsappears when we employ the value weghted approach. The second column n Table 6 presents a dstnct dfference 11
n the sgn of the varables of nterest, wth coeffcent ( ) beng postve and statstcally nsgnfcant (sgnfcant). As a result, herdng could be attrbuted mostly to small captalzaton stocks snce the phenomenon dsappears takng market value nto consderaton. Table 6. Regresson results for Volatlty Dummy Based Model1 Equal Weghted Market Returns Value Weghted Market Returns 0 0.9067 0.1138 1 0.5928 0.7101 2 0.5543 0.5516 3-0.0946 0.0008 (p-value) 0.2071 0.9978 4-0.0398 0.1819 (p-value) 0.0140** 0.0008*** Adj-R 2 0.3375 0.6919 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD D R D R D R D R, D f hgh market volatlty that day, D otherwse. Daly data from January 2007 to May 2015. ***,** statstcally sgnfcant at 1% and 5% level, respectvely. Fnally, Table 7 reports the results testng for asymmetres relatve to the 10 year 10-year Greek Government bond spread for both the equal weghted and the value weghted samples as n equaton (11). The results document evdence of herdng on days wth hgh as well as low spread compared to the 30-day movng average wth coeffcents and beng both negatve and statstcally sgnfcant. However, there s an asymmetrc mpact on herdng snce coeffcents and are statstcally sgnfcant dfferent from each other (Wald coeffcent test p=0.02). As a result, herdng s more pronounced on days when the 10-year Greek bonds dsplay low spread. Even though ths fndng does not confrm our ntal hypothess of ncreased herdng on days wth hgh spreads, whch s qute common durng crss perods, t s consstent wth studes that ndcate reduced herdng durng crss perods (Bowe and Domuta, 2004) as well as greater mpact of sentment durng non-crss perods (Chung et al., 2012; Hudson and Green, 2015). Table 7. Regresson results for Spread Dummy Based Model2 Equal Weghted Market Returns Value Weghted Market Returns 0 0.8982 0.1203 12
1 0.7735 0.5239 2 0.5272 0.5815 3-0.2257 0.2160 (p-value) 0.0071*** 0.1275 4-0.0276 0.1165 (p-value) 0.0674* 0.0293** Adj-R 2 0.3451 0.6867 Notes: Ths table presents the estmated coeffcents of the followng model: CSAD D R D R D R D R, D f hgh spread, D otherwse. Daly data from January 2007 to May 2015. ***,**,* statstcally sgnfcant at 1%, 5% and 10% level, respectvely. 4. Conclusons Ths study s n lne wth the aggregate-data models of Chang and Zheng (2010) and Economou et al. (2011), provdng further nsght nto the recent developments of herdng behavor n the the Greek stock market,.e. n an economy undergong a sgnfcant soveregn debt crss. In order to test for herdng towards the market consensus, we employ a survvorshp bas free dataset, usng the Worldscope Greece lst of stocks from January 2007 to May 2015. Herdng asymmetry has been tested for dfferent market states regardng market return, tradng volume, volatlty and 10- year government bond spread alongsde the basc model. The emprcal results are very conclusve, demonstratng herdng n the case of the equal weghted market returns, beng stronger n down markets, hgh volume and hgh market volatlty days. Ths s consstent wth prevous studes about the ASE, as well as other studes that examne less developed economes, or economes undergong extreme prce movements (Chang et al., 2000; Chang and Zheng, 2010; Economou et al., 2011). Moreover, testng for the mpact of soveregn bond spreads herdng behavor s more pronounced on days wth low spreads. Fnally, employng the quantle regresson method, we document herdng only n the hgh quantles of the cross sectonal return dsperson. 13
However, these emprcal results are derved usng an equal weghted market return measure to compute CSAD, and are not robust when sze effect s accounted for ndcatng the mpact of sze effect on herdng estmatons n a thnly traded market. Accordng to ths fndng, herdng n the ASE can be attrbuted to small captalzaton stocks. The emprcal fndngs are of sgnfcant mportance, especally gven the current economc stuaton n Greece and the ongong soveregn debt crss. A better understandng of the market partcpants decsons could provde valuable nsght for portfolo management and tradng strateges formaton. Investors should take nto consderaton the mpact of herdng n the asset allocaton process, especally on small captalzaton stocks, snce correlated tradng patterns reduce dversfcaton benefts, exposng at the same tme market partcpants to addtonal rsk. References Barros, S., Iversen, P., Lewandowska, M., Setzer, R. (2009). Determnants of ntra-euro area government bond spreads durng the fnancal crss (No. 388). (DG ECFIN), EC. Bowe, M., Domuta, D. (2004). Investor herdng durng fnancal crss: A clncal study of the Jakarta Stock Exchange. Pacfc-Basn Fnance Journal, 12(4), 387-418. Caporale, G.M., Economou, F., Phlppas, N., 2008. Herdng behavor n extreme market condtons: the case of the Athens stock exchange. Economcs Bulletn 7, 1-13. Chrste, W.G., Huang, R.D., 1995. Followng the ped pper: do ndvdual returns herd around the market? Fnancal Analysts Journal 51, 31-37. Chang, T.C., Zheng, D., 2010. An emprcal analyss of herd behavor n global stock markets. Journal of Bankng and Fnance 34, 1911 1921. Chang, E.C., Cheng, J.W., Khorana, A., 2000. An examnaton of herd behavor n equty markets: An nternatonal perspectve. Journal of Bankng and Fnance 24, 1651-1679. Chang, T. C., L, J., Tan, L., 2010. Emprcal nvestgaton of herdng behavor n Chnese stock markets: Evdence from quantle regresson analyss. Global Fnance Journal 21(1), 111-124. 14
Chung, S.L., Hung, C.H., Yeh, C.Y., 2012. When does nvestor sentment predct stock returns? Journal of Emprcal Fnance 19(2): 217-240. Demrer, R., Kutan, A., 2006. Does herdng behavor exst n Chnese stock markets? Journal of Internatonal Fnancal Markets, Insttutons and Money 16, 123-142. Duffee, G., 2001. Asymmetrc cross-sectonal dsperson n stock returns: evdence and mplcatons. Federal Reserve Bank of San Francsco Workng Papers n Appled Economc Theory, 2000-18. Economou, F., Kostaks, A., Phlppas, N., 2011. Cross-country effects n herdng behavor: Evdence from four south European markets. Journal of Internatonal Fnancal Markets, Insttutons and Money 21, 443-460. Hrshlefer, D., Hong Teoh, S., 2003. Herd Behavour and Cascadng n Captal Markets: a Revew and Synthess. European Fnancal Management 9, 25-66. Hudson, Y., Green, C.J., 2015. Is nvestor sentment contagous? Internatonal sentment and UK equty returns. Journal of Behavoral and Expermental Fnance 5: 46-59. Koenker, R.W., 2005. Quantle regresson. Cambrdge Unversty Press. Koenker, R.W., Bassett, G., Jr., 1978. Regresson quantles. Econometrca 46, 33 50. Lakonshok, J., Shlefer, A., Vshny, R., 1992. The mpact of nsttutonal tradng on stock prces. Journal of Fnancal Economcs 32, 23-43. Mobarek, A., Mollah, S., Keasey, K., 2014. A cross-country analyss of herd behavor n Europe. Journal of Internatonal Fnancal Markets Insttutons and Money 32, 107 127. Newey, W., West, K., 1987. A smple, postve sem-defnte, heteroskedastcty and autocorrelaton consstent covarance matrx. Econometrca 55, 703 708. Peltomäk, J., Vähämaa, E., 2015. Investor Attenton to the Eurozone Crss and Herdng Effects n Natonal Bank Stock Indexes. Fnance Research Letters 14, 111-116. Tan, L., Chang, T.C., Mason, J.R., Nellng, E., 2008. Herdng behavor n Chnese stock markets: An examnaton of A and B shares. Pacfc-Basn Fnance Journal 16, 61 77. 15
Tessaromats, N., Thomas, V., 2009. Herdng behavor n the Athens stock exchange. Investment Management and Fnancal Innovatons 6, 156-164. Zhou, J., Anderson, R. I., 2013. An emprcal nvestgaton of herdng behavor n the US REIT market. The Journal of Real Estate Fnance and Economcs 47(1), 83-108. 16