Return-Volume Dynamics of Individual Stocks: Evidence from an Emerging Market

Transcription

1 Reurn-Volume Dynamics of Individual Socks: Evidence from an Emerging Marke Cein Ciner College of Business Adminisraion Norheasern Universiy 413 Hayden Hall Boson, MA Tel: Absrac: This paper invesigaes he linkage beween lagged volume and subsequen price movemens on he emerging equiy marke of Isanbul Sock Exchange. Recen heoreical models asser ha informaion asymmery deermines he cross-secional variaion across individual sock in he dynamic volume-reurn relaions. Using a sample of all socks raded on he Isanbul Sock Exchange for a leas six years, we find significan suppor for he heory. Socks of smaller firms, which end o suffer more from informaional asymmery, are indeed more likely o experience speculaive rades and hence, price coninuaions following high volume days. We also find ha he overall exen of speculaive rading on he ISE is higher ha ha on he NYSE, consisen wih he noion ha emerging markes end o me riskier han more developed markes. The resuls are robus o differen specificaions. Sepember 25, 2003

2 Reurn-Volume Dynamics of Individual Socks: Evidence from an Emerging Marke 1. Inroducion Equiy marke paricipans closely follow rading volume as a useful saisic. Saring wih he early models of Clark (1973) and Copeland (1976), financial researchers have argued for a close associaion beween price and volume. Many empirical sudies, such as Hiemsra and Jones (1994Conrad, Hameed and Niden (1994) and Cooper (1999), Gervais, Kaniel and Mingelgrin (2000), find ha rading volume conveys informaion abou price movemens. Recenly, Llorene, Michaely, Saar and Wang (LMSW, 2002) sugges ha volume could also be used o undersand he underlying srucure of price movemens on equiy markes. In an equilibrium framework, hey examine he ineracion beween volume and price changes and argue ha volume signals he informaion conen of a given price move. Specifically, hey show ha volume could be used o disinguish beween speculaive and hedging (risk allocaion) rades. Informaional asymmery plays a key role in heir model. Their cenral proposiion is ha price coninuaions will be observed following days wih high volume when speculaion on privae informaion is he main rading moive, while sock prices will exhibi reversals following high volume days if hedging is relaively more imporan. In his paper, we rely on he approach by LMSW and examine he dynamic linkage beween volume, curren reurn and he following reurn (volume-reurn auocorrelaion) of individual socks on an emerging marke, he Isanbul Sock Exchange (ISE). Several sudies sugges ha informaion asymmery problems are more acue on emerging markes, due o facors such as relaively hin rading and lack of insider'

3 rading rules. For example, Brennan and Cao (1997) argue ha a disadvanage in informaion migh explain home bias in inernaional invesmens. Bhaacharya and Daouk (2002) sugges ha lack of prosecuion of insider rading in emerging markes deers foreign invesmen. I is of ineres o es for he implicaions of he LMSW model using emerging marke daa, since differences in informaion asymmery dicae he volume-reurn dynamics of individual sock in heir model. Using a sample consising of all socks raded on he ISE for a leas six years, we repor ha majoriy of individual socks experience price coninuaions following high volume days, which suggess ha speculaive rades are more prevalen. Alhough his is similar o he resuls repored by LMSW, i is noeworhy ha percenage of firms experiencing speculaive rades is significanly higher on he ISE, which is consisen wih he noion ha emerging markes end o be riskier. We hen invesigae cross-secional variaions across individual firms in he dynamic volume-reurn linkage. The cenral implicaion of he LMSW model is ha differences in informaional asymmery explain cross-secional variaions. We use marke capializaion as a proxy for he exen of informaion asymmery. Evidence in he lieraure, such as Lo and McKinlay (1990), sugges ha large socks end o suffer less from privae informaion problems. Wihin he LMSW, his implies ha large firm socks will be less likely o experience price coninuaions following high volume days. We find ha he cross-secional behavior of he socks on he ISE is srongly consisen wih hese predicions. Small firm socks, which end o conain more informaional asymmery, are more likely o exhibi price coninuaions han hose of large firms. This resul lends suppor o LMSW as a universal approach.

4 We conrol for he robusness of our resuls by conducing furher invesigaions. Firs, we examine wheher our resuls could be biased due o an economeric misspecificaion. Specifically, as LMSW also sae, he errors from individual regressions will be correlaed across socks and hence, he cross-secional analysis could produce biased sandard errors. Since i is plausible ha he cross-correlaion of errors arises from dependency of reurns o common facors, we model he facors direcly using a marke proxy, following Jorion (1990). We show ha he resuls are robus o his specificaion. Secondly, we recognize ha reurns and volume reflec boh marke-wide and firm-specific componens, while rades are conduced for firm-specific hedging needs and speculaion in he LMSW model. We eliminae marke-wide variaions in reurns and rading volume using marke model regressions and re-conduc he analysis using firmspecific componens of reurns and volume of individual socks. We repor ha size coninues o explain he cross-secional differences across socks. However, we also find ha he majoriy of socks experience price reversals (hedging rades) following days wih exensive rading volume when we use only firm-specific componens in reurns and volume. This suggess ha he relevan informaion for speculaion on he ISE is markewide informaion. We organize he res of he aricle as follows: In he nex secion, we discuss he heoreical background and hypoheses. We presen he daa se in secion hree and he empirical findings in secion five. We offer he concluding remarks of he sudy in he final secion.

5 2. Theory and Hypoheses LMSW propose a simple equilibrium model o examine he relaion beween rading volume and price movemens in equiy markes. Their model suggess ha reurns are generaed by hree separae sources: public informaion, hedging and speculaion. I is assumed ha public news causes only a whie noise componen, while reurns generaed by hedging and speculaion are serially correlaed. Hedging rades are due o risk allocaion decisions; because hey do no reflec new informaion, he expeced payoff from he asse remains he same and he asse mus be sold a a discoun o arac oher raders o ake he oher side of he ransacion. Price rises back o is original level in he nex period, since he fundamenal value is unchanged. In a hedging rade, herefore, an iniial negaive reurn is followed by a posiive reurn in he second period, generaing negaive reurn auocorrelaions. Speculaive rades, on he oher hand, are caused by he asymmeric informaion of informed raders. LMSW argue ha privae informaion will be only parially incorporaed ino prices in he curren period and herefore, prices will coninue o change in he same direcion in he nex period. Consequenly, speculaive rades generae posiive reurn auocorrelaions. Volume has a prominen role in he LMSW model. Specifically, LMSW argue ha volume can be used o disinguish beween price changes due o public informaion and hose due o hedging or speculaion. Public news is incorporaed ino prices via normal rading, while hedging and speculaive rades are characerized by exensive volume. Hence, as saed in he inroducion, he cenral implicaion of he LMSW approach is ha high volume days will be followed by price reversals, when hedging is

6 he primary moive o rade, however, price coninuaions will be observed when speculaion is he primary moive. This proposiion can be examined by esimaing he following equaion: R α + u (1) 2 1/ 2 = + ( β0 + β1v 1 + β 2V 1 + β3h 1 ) R 1 in which V denoes log volume series, R denoes reurns, calculaed as log price differences, and h -1 is he condiional volailiy series obained from he following GARCH model: R ε h R, = ε = α R, Ω 1 0, ~. d(0, h, v) + α ε 1 2 R, 1 + α h e in which he residual erm ε R, follows a condiional Suden s disribuion (.d) wih ν degrees of freedom and a condiional variance h. Ω -1 is he informaion se ha conains all relevan informaion a ime -1. The model in (1) is a modified version of he regressions in LSMW. 1 I measures he ineracion beween reurn auocorrelaion and lagged volume by β 1, lagged volume squared by β 2, and condiional volailiy by β 3. Squared volume series are included o accoun for nonlinear relaions beween reurn auocorrelaions and volume and β 3 examines linkages beween condiional variance and volume. Many sudies sugges ha volume and volailiy are closely relaed variables, as in he mixure of disribuions model of Clark (1973). Karpoff (1987) provides a review he lieraure on linkages beween volume and price variabiliy and documens ha generally a posiive 1 The regression esimaed by LMSW does no include a condiional volailiy erm.

7 simulaneous relaion is deeced beween he variables. Therefore, as Gagnon and Karolyi (1997) noe, he inclusion of a condiional volailiy variable could help o produce more powerful ess. However, he main coefficien ineres in he invesigaion is β 1, he measure of ineracion beween reurn auocorrelaion and lagged volume. If hedging is relaively more imporan han speculaion, high volume days will be followed by price reversals and β 1 will be negaive and saisically significan. On he oher hand, if speculaion is he primary rading moive, price coninuaions are expeced following high volume days and β 1 will be posiive and significan. The key asserion of LMSW is ha variaions in informaional asymmery deermine cross-secional differences in he volume-reurn linkage of individual socks. Socks wih a large degree of informaional asymmery are more likely o be exposed o speculaive rades and hence, o experience price coninuaions following high volume days. Therefore, a proxy for informaional asymmery, such as marke capializaion, should explain cross-secional variaion across firms in he volume-reurn dynamic. To es for his asserion, we use marke capializaion as a proxy and esimae he following regression: ß 1,i = a + b.mcap i +? i (2) in which MCAP denoes marke capializaion of he individual firm and ß 1 is colleced from regression in (1). If differences in informaional asymmery give rise o variaion in he volume-reurn linkage across socks, we expec b o be negaive and significan in his model.

8 3. Daa The daa se consiss of daily closing prices and rading volume of all socks raded on he ISE beween January 2, 1995 and May 1, There are a oal of 220 firms in he sample and all daa are from he ISE. The LMSW model requires ha he parameers in he model remain unchanged; hence, he daa are colleced for a six-year period. In Table 1, we provide some sample summary saisics. Our measure of rading volume is daily (log) urnover, he raio of daily number of shares raded divided by he number of shares ousanding. Campbell (2000) presens heoreical argumens o sugges ha urnover is a beer measure for rading volume. We also eliminae he deerminisic rends in rading volume using a 50-day moving average process. 4. Empirical Findings We begin he analysis by esimaing he regression in (1) for each sock in he sample. As menioned before, he main coefficien of ineres in his equaion is ß 1, which measures he linkage beween lagged rading volume and subsequen price movemens. If speculaion is more imporan hen we expec ß 1 o be saisically significan and posiive. If, however, hedging is he primary reason o rade he LMSW heory suggess ha ß 1 will negaive and significan. The resuls of individual regressions are repored in Table 1. I can be observed ha a significan majoriy of socks on he ISE experience price coninuaions following days wih high rading volume, i.e. ß 1 is posiive. I is noeworhy ha he percenage of socks ha experience speculaion on he ISE is higher he resul repored by LMSW on he NYSE. This is consisen wih he undersanding ha emerging equiy markes end o be more risky han more developed markes.

9 The hear of he LMSW analysis is he cross-secional relaion beween individual firms. The heory assers ha informaional asymmery explains differences across individual firms in he volume-reurn relaion. We run regression (2) o es for his implicaion and find ha here is indeed a negaive relaion beween our informaion asymmery proxy, marke capializaion, and he resuls for individual firms. The resuls from he regression analysis, repored also in Table 1, sugges ha socks of small firms are more likely o experience speculaive rades han hose of large firms, as implied by he LMSW heory and consisen wih resuls from he NYSE. We proceed o invesigae he robusness of our resuls. As menioned before, he findings could be biased due o an economeric misspecificaion. Specifically, as LMSW also sae, he errors from individual regressions will be correlaed across socks and hence, he cross-secional analysis could produce biased sandard errors. Since i is plausible ha he cross-correlaion of errors arises from dependency of reurns o common facors, we model he facors direcly using a marke proxy, following Jorion (1990). We consider ISE 100, he main index on he ISE, as represenaive of marke reurns and include reurns on he ISE 100 index in each individual regression o capure dependency o common facors. We reconduc he analysis and repor resuls in Table 2. The resuls are only slighly changed. Significan majoriy of socks coninue o experience price coninuaions following high volume days, indicaing he high exen of speculaion on he ISE. Perhaps more imporanly, he cross-secional relaion in he dynamic volume-reurn relaion is robus o his specificaion. There is a saisically significan negaive relaion beween marke size and price movemens following high volume days.

10 We also conduc a robusness analysis o invesigae wheher he resuls obain when only firm-specific componens of volume and reurns are considered. We recognize ha reurns and volume reflec boh marke-wide and firm-specific componens, while rades are conduced for firm-specific hedging needs and speculaion in he LMSW model. We eliminae marke-wide variaions in reurns and rading volume using marke model regressions and re-conduc he analysis using firm-specific componens of reurns and volume of individual socks. We repor in Table 3 ha size coninues o explain he cross-secional differences across socks. However, we also find ha he majoriy of socks experience price reversals (hedging rades) following days wih exensive rading volume when we use only firm-specific componens in reurns and volume. This suggess ha he relevan informaion for speculaion on he ISE is marke-wide informaion. This finding is no surprising considering ha Turkey experienced severe financial crises in he pas decade ha adversely affeced he performance of socks on he ISE. 5. Concluding Remarks In a recen conribuion LMSW asser ha informaion asymmery deermines he relaion beween equiy price movemens following days wih high rading volume. They assume ha invesors rade o adjus for risk, hedging, or on privae informaion, speculaion. Speculaive rades will be more imporan for firms wih high exen of informaion asymmery, while hedging rades will be more imporan for ohers. They sugges ha socks of large firms will be more likely o experience price reversals, hedging rades, while sock of small firms will be more likely o experience price coninuaions, speculaive rades. They find empirical evidence for heir model using a

11 sample of socks raded on he NYSE. Hence, heir analysis reconciles differen findings in he lieraure regarding large and small firms in he dynamics volume-reurn relaion. Our sudy examines he LMSW heory in an emerging marke seing. Emerging markes provide an opporuniy o invesigae he LMSW heory since informaion asymmery is he key variable in he model and emerging markes suffer from his problem more severely han more advanced markes. Or resuls provide srong suppor for he LMSW model. Socks of small firms on he ISE are more likely o experience price coninuaions following days wih high volume han large firms. We also show ha he resuls are robus o differen specificaions. Hence, he findings of he sudy offer supporive evidence o he LMSW heory as a universal approach. In addiion o he cross-secional analysis, we also show ha a significan majoriy of individual socks on he ISE demonsrae price coninuaions following days wih high volume. This indicaes ha speculaive rades are much more prevalen on he ISE han he NYSE.

12 References Blume, L., D. Easley and M. O Hara, 1994, Marke saisics and echnical analysis: The role of volume. Journal of Finance 49, Chang, E., R. Y. Chou and E. F. Nelling, 2000, Marke volailiy and demand for hedging in sock index fuures, Journal of Fuures Markes 20, Clark, P., 1973, A subordinaed sochasic process model wih finie variances for speculaive prices, Economerica 41, Gagnon, L. and G. A. Karolyi, 1997, Informaion, Trading Volume and Inernaional Sock Marke Comovemens, Working Paper, Ohio Sae Universiy. Godfrey, L. G., 1978, Tesing agains general auoregressive and moving average error models when he regressors include lagged dependen variables, Economerica 46, Karpoff, J., 1987, The relaion beween price changes and rading volume: A survey, Journal of Financial and Quaniaive Analysis 22, Llorene G., R. Michaely, G. Saar and J. Wang, 2002, Dynamic Volume-Reurn Relaion of Individual Socks, Review of Financial Sudies 15, Moosa, A. I. And E. Al-Loughani, 1994, Unbiasedness and ime varying risk premia in he crude oil fuures marke, Energy Economics 16, Suominen, M., 2001, Trading volume and informaion revelaion in sock markes, Journal of Financial and Quaniaive Analysis 36, Wang, C., 2003, The behavior and performance of major ypes of fuures raders, Journal of Fuures Markes 23, 1-31.

13 Table 1. Volume and Serial Correlaion: Basic Regression Analysis Panel A: Individual Socks β 1 β 2 β 3 # (posiive) # (significan) # (negaive) #(significan) Panel B: Cross-Secional Analysis β 1, = a + bmcap + ξ a.128 (.00) b (.00) R-square.026 Noe- This able provides he resuls of esing for linkages beween volume and reurn auocorrelaions on energy fuures markes. The following equaion is esimaed by he OLS: R + u 2 1/ 2 = α + ( β0 + β1v 1 + β 2V 1 + β3h 1 ) R 1

14 Table 2. Volume and Serial Correlaion: Sensiiviy o Common Facors Panel A: Individual Socks β 1 β 2 β 3 # (posiive) # (significan) # (negaive) #(significan) Panel B: Cross-Secional Analysis β 1, = a + bmcap + ξ a.160 (.00) b (.00) R-square.06 Noe- This able provides he resuls of esing for linkages beween volume and reurn auocorrelaions on energy fuures markes. The following equaion is esimaed by he OLS: R + R + u 2 1/ 2 = α + ( β0 + β1v 1 + β 2V 1 + β3h 1 ) R 1 where R m represens reurns on he ISE 100. m

15 Table 3. Volume and Serial Correlaion: Firm-Specific Effecs Panel A: Individual Socks β 1 β 2 β 3 # (posiive) # (significan) # (negaive) #(significan) Panel B: Cross-Secional Analysis β 1, = a + bmcap + ξ a.160 (.00) b (.02) R-square.06 Noe- This able provides he resuls of esing for linkages beween volume and reurn auocorrelaions on energy fuures markes. The following equaion is esimaed by he OLS: R + u 2 1/ 2 = α + ( β0 + β1v 1 + β 2V 1 + β3h 1 ) R 1 where only firm-specific reurns are used by exracing marke reurns.