Private Information Trading and Corporate Governance In Emerging Markets

Transcription

1 Prvate Informaton Tradng and orporate Governance In Emergng Markets Olesya V. Grshchenko, Lubomr P. Ltov and Janpng Me 1 January 003 Abstract Ths paper develops a new approach to extract nformaton on corporate governance n emergng markets based on tradng data. We apply the theoretcal framework of Llorente, Mchaely, Saar, and Wang (00) to analyze the relaton between daly volume and frst-order return autocorrelaton for ndvdual stocks n emergng markets. We fnd strong evdence of return contnuaton followng hgh volume days, suggestng the presence of prvate nformaton tradng for many emergng market stocks. We dscover that prvate nformaton tradng s especally strong around major corporate event dates. In addton, we fnd stocks that provde better nvestor protecton and nformaton dsclosure exhbt less prvate nformaton tradng. These results suggest return autocorrelaton and tradng volume carry useful nformaton about corporate governance n emergng market. 1 Doctoral students and assocate professor of fnance at New York Unversty. Please send nqures to Janpng Me, Department of Internatonal Busness, Stern School, 44 West 4th Street, New York, NY Tel: , Fax: Emal: jme@stern.nyu.edu. We lke to thank John ampbell, Hyuk hoe, Robert Engle, Kent Hargs, Joel Hasbrouck, Stas Kolenkov, Gdeon Saar, Gl Schorr, Jang Wang, Robert Whtelaw, Danel Wolfenzon, partcpants at NYU Monday Fnance Workshop, Hong Kong UST semnar as well as the hna enter for Fnancal Research 00 conference, European Fnancal Management Assocaton Meetngs 00 at London and European Fnance Assocaton Meetngs 00 at Berln for helpful dscussons and comments. We are grateful to Rebecca Pe for her excellent research support. We also lke to acknowledge that Emergng Market Fnance Insttute has provded some fnancal support to ths research. 1

2 Prvate Informaton Tradng and orporate Governance In Emergng Markets Abstract Ths paper develops a new approach to extract nformaton on corporate governance n emergng markets based on tradng data. We apply the theoretcal framework of Llorente, Mchaely, Saar, and Wang (00) to analyze the relaton between daly volume and frst-order return autocorrelaton for ndvdual stocks n emergng markets. We fnd strong evdence of return contnuaton followng hgh volume days, suggestng the presence of prvate nformaton tradng for many emergng market stocks. We dscover that prvate nformaton tradng s especally strong around major corporate event dates. In addton, we fnd stocks that provde better nvestor protecton and nformaton dsclosure exhbt less prvate nformaton tradng. These results suggest return autocorrelaton and tradng volume carry useful nformaton about corporate governance n emergng market.

3 1. Introducton It s not enough for the playng feld to be level. Perhaps even more mportant, t must also be vsble. The late 1990s stock market boom and ts aftermath have shown that even n the Unted States - wth extensve nsder-tradng laws, outsde audtors, and a host of other regulatory features to protect nvestors - large parts of the playng feld are smply hdden from vew. Thus, for example, many brokerage frms mantaned publc buy recommendatons on companes ther analysts were quetly dsparagng to a favored subset of clents 3. Relatedly, there has been wdespread manpulaton of earnngs by corporatons. These factors have dealt a serous blow to nvestor trust n securtes markets n the Unted States and contrbuted to a bear market. In Southeast Asa, the durablty of crony captalsm contnues to plague the regon's markets. All ths suggests that the rule of law and nvestor rghts - whch we term "corporate governance" - s of prmary mportance to nvestment returns. A major dffculty n studyng corporate governance s the lack of qualty nformaton. Ths s especally true n emergng markets, where there are fewer dsclosure rules, much less enforcement of nsder tradng laws, poor protecton of mnorty shareholder rghts, unequal treatment of foregn and domestc shareholders, and a generally underdeveloped legal and regulatory envronment. Moreover, dsclosed nformaton, ncludng analyst reports, often s subject to manpulaton. orporate governance problems are even more pervasve n emergng markets due to poor protecton of the mnorty shareholder rghts, unequal treatment of foregn and domestc stockholders, and underdeveloped legal and regulatory envronment. To protect themselves aganst corporate predators, nvestors n emergng markets need to fnd alternatve source of nformaton on corporate governance. Recently the New York State attorney general has made numerous nvestgatons nto leadng Wall Street frms for ssung nflated nvestment ratngs and offerng hot IPOs n turn for nvestment bankng busness from telecom companes. Several EOs from once hgh-flyng companes have been ndcted or have pleaded gulty to outrght stealng from shareholders. 3 Hayes (1998), Lm (001), and Mchaely and Womack (1999) have shown that nformaton s manpulated n order to drum up nvestment bankng busness, to mantan access to nformaton, or to stmulate tradng by nvestors. (For example, numerous telecom analysts were alleged or found to have nflated earnngs forecasts for companes such as Global rossng, Quest, Worldom and Wnstar. Despte mountng loss and fallng revenues, many contnued ther buy recommendaton for nvestors untl ther bankruptcy.) 3

4 Ths paper contrbutes to the lterature by establshng a relatonshp between return autocorrelaton and prvate nformaton tradng. Our paper s based on a dynamc volume-return model of Llorente, Mchaely, Saar, and Wang (00, LMSW hereafter). The essence of the model s that ntensve tradng volume together wth stock return autocorrelaton can help us dentfy frms wth a hgh degree of tradng based on prvate nformaton - that s, nformaton that has not been dssemnated to the publc. Prvate nformaton here ncludes both "nsder" nformaton and nformaton not avalable to the general publc. Usng US data, LMSW show that the dfferences n the dynamcs of returns and volume across stocks are closely assocated wth dfferent degrees of nformaton asymmetry. Ths paper extends ther work to measure the degree of nformaton asymmetry n emergng market stocks. We wll mplctly control for market mcrostructure effects such as bd-ask spread, poor lqudty, and nonsynchronous tradng. Another goal of ths study s to establsh a lnk between a measure of prvate nformaton tradng and varous ndcators of corporate governance qualty. Intutvely, nformatonal asymmetry s related to the qualty of corporate governance and nformaton dsclosure. Therefore, the ntensty of prvate nformaton tradng s a useful ndcator for dentfyng corporate governance problems at both macro- and mcro-levels. We examne how the regulatory envronment n the emergng markets countres affects nformatonal tradng. We also study the relaton between ntensty of prvate nformaton tradng and company-specfc ndcators of nvestor rghts protecton. We fnd that stocks n countres that enforce nsder-tradng law, have hgher standards of nformaton dsclosure, and provde better nvestor protecton exhbt less prvate nformaton tradng. By usng tradng data, our study provdes more rgorous results than f t had reled on accountng data, whch can vary greatly n qualty across dfferent markets. The paper s related to several studes that have nvestgated the effects of varous types of nformaton asymmetry on foregn equty holdngs. These asymmetres tend to scare away foregn nvestment and thus drve up the cost of captal. Bhattacharya, Daouk, and Welker (003) show that earnngs management n a country s lnked to a decrease n tradng n the stock market of that country, and s weakly lnked to an ncrease n US home bas toward that country. They also pont out that the practce of earnngs management makes t harder for foregn nvestors to obtan relevant nformaton about 4

5 emergng market frms. Brennan and ao (1997) also demonstrate that a dsadvantage n nformaton mght help explan nvestors home bas. Bhattacharya, Daouk, Jorgenson, and Kehr (000) show that the presence of unrestrcted nsder tradng has caused prces to fully ncorporate frm nformaton before ts publc release n Mexco. Bhattacharya and Daouk (00) dscuss how the lack of prosecuton of nsde tradng creates a hazardous nvestment envronment for foregn nvestors n emergng markets. Albuquerque, Bauer, and Schneder (001) also develop a framework for characterzng asymmetrc nformaton n nternatonal equty markets. Our work also bulds on a growng lterature on return autocorrelaton and turnover. Harvey (1995a) was frst to examne the determnants of return autocorrelaton n emergng market ndces. He documented that a lack of dversfcaton and tradng depth nduces spurous seral correlaton n some countres but noted that more needs to done to explan the return autocorrelaton. Rouwenhorst (1999) fnds that return factors and turnover are related n emergng markets. However, the focus of hs study s on lqudty ssues. We extend a large body of lterature on emergng-market asset prcng by studyng daly (hgh-frequency) tradng data at the frm level. Most prevous studes - ncludng Baley and Lm (199), Bekaert and Harvey (1995, 1997, 000), Bekaert, Harvey, and Ng (00), Harvey (1995a, 1995b), and Henry (1999, 000) - use monthly (low-frequency) data. Moreover, by relyng on turnover rather than accountng data, our data enjoy a hgh degree of transparency and unformty across dfferent markets, whch makes t easer for cross-country comparson. We measure turnover by the number of shares traded rather than by ther value. The paper s organzed as follows. Secton descrbes the methodology for measurng prvate-nformaton tradng after controllng for mcrostructure effects and the data. Emprcal results on the extent of nformatonal tradng are then presented n Secton 3. Secton 4 relates such tradng to aspects of corporate governance, wth prvatenformaton tradng around corporate events and corporate governance ssues n Russa recevng specal attenton. We conclude wth some caveats and topcs for further research n Secton 5. 5

6 . Methodology Ths secton brefly ntroduces the LMSW model, n whch nvestors trade for both hedgng and nformaton reasons. We use the model to demonstrate how the dynamc relaton between return and volume depends on the nformaton asymmetry between nvestors. Because the goal here s just to establsh the ntuton behnd our study, we wll smply descrbe the economy and provde the theoretcal results. Interested readers are referred to the orgnal paper of LMSW. The economy s defned on a dscrete tme sequence and there are two traded securtes, a rskless bond and a stock. The bond s n unlmted supply at a constant nonnegatve nterest rate. The stock s dvdend Ft t t+1 = Ft t D t+1 at the end of the tme perod s the sum of two components and G : D + G. There are two classes of nvestors. Investors are dentcal wthn each class, but are dfferent n ther endowments and nformaton. Both groups observe F, but group 1 has an nformaton advantage of also observng G. t The degree of nformatonal asymmetry s measured usng a standard devaton of second σ G sgnal. In addton, nvestors are also endowed wth a non-traded asset wth payoff, t N t and the random endowment Z t, = 1,. Investors maxmze ther expected utlty over the next perod wealth usng a common exponental utlty functon condtonal on ther respectve nformaton set. All shocks to the economy are normally dstrbuted wth zero mean and constant varances. In addton, they are assumed to be mutually ndependent, except for the payoff to non-traded asset and dvdend on the stock, whch are correlated. LMSW pont out that the model provdes two mportant motves for tradng: hedgng rsk and speculaton on future returns by nformed nvestors. All nvestors hold both stock and non-traded asset n ther portfolos. Because the returns on the two assets are correlated, as holdngs of the non-traded asset change, each nvestor wants to adjust hs stock postons to mantan an optmal rsk exposure. Ths generates hedgng trade n the model. On the other hand, some nvestors mght have prvate nformaton about future stock returns. As new prvate nformaton arrves, they take speculatve postons n the stock, whch generates nformatonal trade n the model. LMSW solve for the equlbrum stock prce and tradng volume by provdng the followng dynamc volume-return relatonshp: 6

7 Here, E [ R, t + 1 R, t, V, t ] = 1R, t + R, tv, t. (1) 1 represents the uncondtonal return autocorrelaton provded the correlaton between volume and return s small. ndcates whether stocks are domnated by hedgng trades or trades generated by prvate nformaton. In prncple, all stocks have both nformatonal and hedgng trades. When nether domnates, s statstcally nsgnfcant. A statstcally postve coeffcent suggests sgnfcant nformatonal trades, whereas a statstcally negatve coeffcent ndcates domnatng hedgng trades. LMSW note that When all trades are hedgng trades, current returns together wth hgh volume predct strong reversals n future returns. Another way of understandng a negatve s that t captures the lqudty effect, whch measures the prce mpact of a gven sze trade. Thus, the more llqud the stock, the larger wll be the prce mpact, and the more negatve the coeffcent wll be. Pastor and Stambaugh (00) have used ths result to measure stock and constructed a market lqudty measure based on aggregatng 4 coeffcents across stocks. However, f prvate-nformaton trades are more mportant, the stock s future expected payoff changes. 5 Because LMSW adopt non-fully revealng sgnalng equlbrum model all the nformaton about future expected payoff s not ncorporated n the today s stock prce. Therefore t takes one or several more tradng perods n order for nformaton to be ncorporated fully n the stock prce. LMSW have shown that, ceters parbus, ncreases wth the degree of nformaton asymmetry. One apparent advantage of the LMSW model s that t can be easly extended to mplctly control for returns autocorrelaton resulted from mcrostructure effects such as bd-ask bounce (a stock alternately tradng on ts bd and ask prces) and non-synchronous tradng (stale prces). Accordng to Roll (1984), the bd-ask bounce would ntroduce a negatve seral correlaton for stock returns, whch can be captured by a negatve coeffcent. On the other hand, f non-synchronous tradng domnates, then we expect a 4 Also see ampbell, Grossman and Wang (1993). 5 Here we wll use prvate nformaton trades and speculatve trades nterchangeably. 6 To be more precse, ncreases wth the degree of nformaton asymmetry. 7 6 σ G 1

8 postve seral correlaton for stock returns, whch can be captured by a postve coeffcent. 1 could also be postve f there s short-term prce momentum or negatve f there s a reversal. In ths paper, we estmate equaton (1) and measure the ntensty of asymmetrc nformaton n emergng market stocks. Followng LMSW, we make several modfcatons of the theoretcal model for emprcal analyss. The theoretcal model uses dollar returns per share and normalzed volume, whle we use log returns (because stock prces n emergng markets are not statonary) and detrended log turnover as n LMSW. We begn by estmatng the followng regresson derved from equaton (1): 1 R, t + 1 = 0 + 1R, t + R, tv, t + ε, t + 1. () We use daly contnuously compounded return and tradng volume for the estmaton of (). Followng earler studes, such as LMSW and Stckel and Verreccha (1994), we defne contnuously compounded return as R log(( P, + D ) P, 1), and volume as, t = t t t V = VOL 1, t log(, t N, t ) log( VOL,, ), 0 t j N t j 0 j= 1 where P, t s the daly close prce, VOL, t s the daly number of shares traded, and N, t the total number of outstandng shares n day t for company. Detrended daly turnover s used as a measure of tradng volume for ndvdual stocks. Lo and Wang (000) provde theoretcal justfcaton for usng daly turnover as a proxy for the tradng volume of ndvdual stocks. We found that daly tme seres of turnover are nonstatonary, so we measure turnover n logs and detrend the seres. Followng LMSW, we change zero tradng volume to a small constant before takng logs and detrend resultng seres by subtractng 0-day movng average 7. We assume the error term n equaton () s uncorrelated over tme, but ts varance may vary. 8 We also examne the dynamc volume-return relatonshp by usng excess returns and excess turnover. Our objectve s to remove the market effect and concentrate 7 We also detrend seres by 60-day and 10-day movng average. The number of stocks robust to MA specfcaton n each country s reported n the last column of Table. 8 See Harvey (1995b), and Bekaert and Harvey (1995, 000) for tme-varyng volatlty n emergng markets. 8

9 on the dosyncratc component of ndvdual stock return and volume. It s concevable that the market component n returns and turnover s assocated wth allocatonal trades whle company-specfc (dosyncratc) component s assocated wth prvate nformatonal trade. Ths may make t easer to dscover the presence of nformatonal trades. Therefore, we run the followng regresson: e e e e R, t + 1 = 0 + 1R, t + R, tv, t + ε, t + 1, (3) e e where R R s return, and V = V s turnover, n excess of the market. R, t =, t m, t, t V, t m, t 3. Data There are 1001 companes from 19 emergng markets n the study. The sample perod s January 1, 1995 to November 1, 001 (except for Russa). Returns are converted to US dollars usng data from S&P DRI Pro. When the project started on January 11, 00, the smallest market was Sr Lanka, wth a market cap of $0.49 bllon, and the largest was South Korea wth $14 bllon. Basc statstcs are reported n Table 1. To the extent avalable from Datastream, the sample ncludes the stocks of each country's market ndex. We cross-checked the sample so obtaned wth the IF/S&P nvestble ndex for each market. We deleted from the sample any companes that are not consdered nvestment grade ones by the IF. Stock data are from Datastream, except for Russa, where data are from the Russan Tradng System (RTS) webste (but Gazprom, from AKM Russan consultng agency). For the selected stocks, our almost seven-year study perod has as many as 1785 daly observatons on three varables: closng prce, number of shares traded, and the total number of shares outstandng. Because some stocks are subject to mssng observatons, and dfferent countres observe dfferent closng days, the number of observatons actually used s less for many stocks. The Russan sample starts wth September 1, 1995, and ends November 1, 001. Appendx Table 1b reports the average number of tradng days for each country. The arthmetc mean, geometrc mean, and standard devaton of daly stock returns and daly turnover for each country n Table 1 were computed frst for each company and then averaged across stocks wthn a country. Mean and medan frst order 9

10 autocorrelaton n daly returns were smlarly computed. The standard devaton, however, was computed as the cross-sectonal varaton of autocorrelatons. Average daly returns are mostly negatve durng the sample perod. Ths s not surprsng, as most countres studed suffered through a seres of fnancal crses. Daly-return volatlty ranged from.% n Portugal to 6.% n Russa. Average turnover vares dramatcally across countres, wth a sluggsh 0.06% for olomba and an extremely heavy 3.% n South Korea. There s also a large varaton of turnover volatlty across markets, rangng from 0.1% n Russa to 6.% n South Korea. The nnth to eleventh column presents the mean, medan, and standard devaton of frst order autocorrelaton n daly returns. For the mean and medan, we compute them for each company frst and then average them across stocks wthn a country. For the standard devaton, however, t s computed as the cross-sectonal varaton of autocorrelatons. We can see from the nnth column that the mean autocorrelaton of stocks tend to be postve n most countres but t could be negatve such as n Pakstan. From the varaton of autocorrelatons, we can see that even for countes lke Russa, where the mean autocorrelaton s negatve (-0.074), the large standard devaton (0.154) suggests that there are stll a large number of stocks wth postve autocorrelaton. These results are smlar n sprt to those found n Harvey (1995a), who dscovers postve frst order autocorrelatons for almost all emergng market ndces but they vary greatly across markets. He further fnds that the varaton s partly explaned by dfference n market cap, tradng volume, and asset concentraton. However, he found there s much varaton left to be explaned. Our next objectve s to further explore return autocorrelaton and turnover and to extract some useful nformaton on speculatve tradng and corporate governance. 4. Emprcal Results on Prvate Informaton Tradng Ths secton presents the results of regressons on equatons () and (3). Our results suggest notceable nformaton asymmetry n emergng market stocks, wth partcularly pervasve levels n some countres. 10

11 4.1 Estmatng Equaton (): Panel A We start by estmatng equaton () for all emergng market stocks n our sample. For smplcty, we aggregate the parameter estmates by countres and the results are presented n Table, Panel A. The mean 1 coeffcent for most countres s postve (column ), suggestng the presence of prce momentum or the effects of non-synchronous tradng. These results are consstent wth those of Harvey (1995a), who fnds that emergng market ndces generally exhbt short-term postve seral correlaton. olumn 3 reports the cross-sectonal averages of coeffcents for each country n the sample and column 4 reports the percentage of postve coeffcents. The mean coeffcent s postve for 10 of the 19 countres (column 3). The strongest evdence of nformatonal tradng s n Brazl ( > 0 for 7% of the stocks). Argentna, Indonesa, Pakstan, and Russa also have partcular strong evdence for such tradng. South Korea generally s perceved to have the most developed fnancal market among the countres n our study, and t shows the least prvate-nformaton tradng n the tests: > 0 for 15.5% of the stocks. hle, olomba, Greece, Inda, Malaysa, Phlppnes, and Thaland, also have less than half of postve. The total number of postve coeffcents could be msleadng because some mght be not statstcally sgnfcant. Thus, we also report the percentage of companes wth sgnfcantly postve coeffcents (at 5% level) n column 5. The t-statstc s computed based on the Whte heteroscedastcty robust standard errors. The coeffcent s statstcally sgnfcant and postve for about 14% of the sample (140 stocks). Indonesa, perceved to have some of the worst crony captalsm n the world, has the hghest percentage, 4%, of sgnfcantly postve coeffcents. coeffcent s determned by a balance between lqudty and nformaton tradng, so t s remarkable that many Russan stocks stll show a presence of nformaton tradng despte ther low turnover. To domnate the effects of lqudty, prvatenformaton tradng must be qute extreme. The sgn of coeffcent from regresson () s robust to dfferent specfcaton of turnover measures. We consder log turnover detrended wth a 0-, 60-, or 10-day movng average. Table, Panel A, olumn 7 reports the number of stocks whose 11

12 coeffcent does not change sgn under the three dfferent specfcatons. The percentage of robust stocks vares between 71% (for Peru) and 93% (for Inda). It s worth notng that although our study shows a sgnfcant presence of prvatenformaton tradng n emergng markets, t does not necessarly mean the strong presence of nsder tradng. However, the data do ndcate a great deal of nformaton asymmetry stacked up aganst less nformed local and foregn nvestors Estmatng Equaton (): Panel B It s concevable that the market component n return and turnover reflects overall market nformaton whle the dosyncratc part corresponds to trades based on prvate nformaton related to the company. Thus, we would expect more to be postve n a modfed regresson (3). Table Panel B reports regresson results. The total number of postve coeffcents has ncreased from 414 n Panel A to 497 n Panel B. The ncrease n postve coeffcents s found at the country level for 16 out of 19 countres. Unfortunately, due to the ncrease n measurement errors, we have a drop n the percentage of companes wth sgnfcantly postve coeffcents. However, the robustness check of coeffcents wth respect to the three dfferent turnover measures ndcates that results are qute robust to the alternatve specfcatons of tradng volume. Percentage of robust stocks vares between 66% (for Greece) and 90% (for Venezuela). The presence of prvate-nformaton trades n emergng markets s further supported when we decompose return and turnover nto market and dosyncratc components. An equalty of means test between coeffcents from regressons () and (3) has a t-statstc of 3.34, sgnfcant at the 5% level. 5. Relatng Prvate-Informaton Tradng to orporate Governance Ths secton examnes the relatonshp between prvate nformaton tradng and our 9 onceptually, prvate nformaton may come from two man sources, nsde nformaton and nformaton derved from research by securty analysts. As Banbrdge (000) notes, nsder tradng s hard to measure because t s usually llegal. The only relable, but not necessarly tmely, source of data concernng legal trades s the reports fled by corporate nsders n the Unted States and Unted Kngdom, but such strngent reportng seldom exsts n emergng markets. 1

13 broadly defned corporate governance measures. Our ntuton s that nformatonal asymmetry should be related to the qualty of corporate governance and nformaton dsclosure. Therefore, ntensty of prvate nformaton tradng could be a useful ndcator for dentfyng corporate governance problems n emergng markets. For smplcty, we classfy ndcators of corporate governance nto two broad categores: the rule of law and nvestor rghts. We frst relate the ntensty of prvate nformaton tradng to the rule of law and nformaton dsclosure, whch together defne the macro-envronment for corporate governance. 5.1 The Data Here we ntroduce our data source and the defnton of the ndcators. The varable used for the enforcement of nsder tradng laws s from Bhattacharya and Daouk (00). It takes value of 1 for the correspondng country f the frst prosecuton under these laws was pror to or durng the sample perod of our study. 10 All countres n our sample had establshed nsder tradng law pror to the begnnng of the sample perod, but enforcement vares across countres. Other ndcators (as used by LaPorta et al (1998)) are the qualty of accountng standards, the rsk of expropraton, and the effcency of judcal system Rule of Law All companes n our sample are pooled to run the followng cross-sectonal regresson usng the above ndcators: = a + a ITLE + a EJS + a RExp + a QAS + error, (4), See Bhattacharya and Daouk (00) for the constructon of the dummy varable. 11 The ndex of accountng standards s created by examnng and ratng the 1990 annual reports on ncluson or omsson of 90 tems n seven categores (general nformaton, ncome statements, funds flow statement, balance sheets, accountng standards, stock data, and specal tems). Expropraton rsk s for outrght confscaton as estmated n the Internatonal ountry Rsk gude. The scale s from 0 to 10, wth hgher scores representng lower rsks. The value s an average of Aprl and October monthly data durng , representng 8 data ponts. Judcal effcency s a 0 to 10 ratng of how the legal system affects busness, partcularly foregn companes, produced by the country-rsk ratng agency Busness Internatonal orp. A hgher score ndcates hgher effcency. The value s an average ndex from 1980 to 1983, whch are the only data avalable. We feel justfed n usng them because judcal systems have not changed much, at least n practce. 13

14 where ITLE s the nsder tradng law enforcement ndex, EJS s the effcency of the judcal system, RE s the rsk of expropraton, and QAS s the qualty of accountng standards for country. Results of cross-sectonal regresson (4) are n Table 3 panel A. The dependent varable for each company s obtaned from regresson (). s nversely related to nsder-tradng laws enforcement, the effcency of the judcal system, and the qualty of the accountng standards; t s postvely related to the rsk of expropraton. Thus, countres wth no or lttle enforcement of ther nsder tradng laws, neffcent judcal system, hgh rsk of expropraton, and low qualty of the accountng standards are expected to have more prvate-nformaton tradng. Results are unchanged f we regress coeffcents on each ndvdual ndctor. 1 As a robustness check, we conducted a cross-sectonal regresson analyss for estmates obtaned from the modfed regresson (3). Our results reman largely unchanged, especally for the sngle-ndcator regresson. The only dfference s that EJS and QAS become nsgnfcant n the jont estmaton. Note that the rsk of expropraton ndex has the hghest adjusted R-square value and t-statstcs n the ndvdual regressons. 5.3 Shareholders rghts To explore the relatonshp across countres between prvate-nformaton tradng and the protecton of the shareholder rghts, we regress coeffcents on another set of ndcators, lsted n Appendx Table 1. Two ndcators, shares not blocked before meetng and proportonal representaton of cumulatve votng, are procedural varables whose mpact on nvestor protecton s somewhat ambguous. The others are clearly related to mnorty nvestor protecton. We now proceed wth the regresson: 1 orrupton, rule of law, and rsk of contract repudaton varables are not ncluded n the regresson specfcaton due to ther hgh correlaton wth the nsder-tradng law enforcement and the rsk-ofexpropraton varables. 14

15 , 5 = a 0 + a OSOV + a OMN + a PRI + a ESM a O + a SNB + a umvot + error (5) Table 4 reports results for cross-sectonal regresson (5). Panel A uses estmates obtaned from regresson () for 978 stocks. 13 Panel B uses estmates from regresson (3). In Panel A all of the coeffcents n the ndvdual regressons are sgnfcant at the 5% level. The one share-one vote varable s nversely related to. Ths mples that the presence of the one share-one vote rule s assocated wth a lower degree of prvate nformaton tradng. As expected, the oppressed mnorty mechansm varable s nversely related to. The ntuton s that the mechansm for protecton of oppressed mnorty nterest allows the latter to dspute the decsons of the management or the assembly. A hgher percentage of share captal needed to call extraordnary shareholders meetng s postvely related to. Moreover, hgher degree of ownershp concentraton seems to gve rse to more nformatonal tradng. However, our results on the preemptve rghts to ssue are ambguous, snce the results are not dfferent n the sngle and jont estmaton. Shares not blocked before meetng varable s postvely related to. One possble explanaton here s that, f shares are not blocked before the shareholders meetng then they can be sold just before the meetng to take advantage of nsder nformaton. As a result, prvate-nformaton tradng could be hgh. The cumulatve votng/ proportonal representaton dummy varable s postvely related to. Most estmates from (5) do not change ther sgn rrespectve of whether the regresson s run ndvdually or jontly. All ndcators, except for preemptve rghts to ssue, are statstcally sgnfcant and preserve ther sgns n the ndvdual regressons n Panel B. 13 Russa s excluded from ths part of the study because we are not aware of macro-ndcators to allow ncludng t. 15

16 5.4 Market aptalzaton and Prvate Informaton Tradng Here we analyze the relatonshp between the average market captalzaton of the companes n the sample and the ntensty of prvate-nformaton tradng. We estmate the followng regresson: 19, = α j D j + α 0 log( Mktap ) + error, (6) j= 1 where D j s the country dummy and log( Mktap ) s the log of company s average market captalzaton n US dollars durng the sample perod. We use the country dummy varables to control for dfferences n the qualty of corporate governance and nformaton dsclosure. Table 5 reports results of the regresson. In regresson (6) I, the dependent varable s the coeffcent from regresson (), whle n regresson (6) II t s from regresson (3). We obtan a statstcally sgnfcant nverse relatonshp between log-average market captalzaton and n both cases. Thus, the ntensty of prvate-nformaton tradng s hgher for small companes n emergng markets. Ths s n lne wth the orgnal results of LMSW, who fnd that smallcap stocks exhbt more nformatonal tradng than large-cap stocks. 6. Specal ase of the Russan Market In ths secton, we explctly lnk the extent of prvate nformaton tradng to corporate governance rankngs. The Russan market s of partcular nterest, because t s often consdered one of the most opaque and hazardous markets n the world. Untl a few years ago ts legal envronment was so murky that Russa was not even rated by many nternatonal ratng agences. It s a market where undsclosed nsder tradng s a real possblty and where superor nformaton of nsders and local nvestors could be ncorporated n stock prces through ther trades. In addton, we were able to obtan corporate governance rankngs only for Russan companes. 16

17 6.1 orporate events The LMSW theoretcal model suggests that prce changes generated by nformatonal trade tend to contnue f tradng volume remans hgh. Therefore, we examne the relatonshp between volume and return around major corporate events. Our conjecture s that corporate-event perods are assocated wth days of hgh tradng volume. We focus on the announcement and holdng of corporate meetngs. These are mportant n Russa, because corporate ownershp structure, board structure, control rghts, and asset dsposton are determned n those meetngs. Press conferences also are ncluded because they often nvolve news of corporate scandals and thus can have a large mpact on share prces. We conjecture that prvate nformaton tradng tends to be most heavy when some mportant news s expected. Insders already know what the news are and may start tradng on t n advance of publc release of the nformaton. The event wndow, durng whch we expect tradng volume to be hgh, s defned as the perod from 10 days pror to the announcement of the corporate event, to 10 days after the actual (effectve) date of the event. The mnmum wndow s thus 0 days, but can be longer f an event s held more than 10 days after t s announced. For each stock we create a corporate dummy varable, whch s one for days n an event wndow for stock, and zero otherwse. To capture addtonal nformaton asymmetry durng the event tradng perods we estmate the followng regresson: c D, t R ε (7) c., t+ 1 = 0 + 1R, t + R, tv, t + 3R, tv, t D, t +, t+ 1 Here, shows what s the usual nformatonal tradng for stock, and shows the 3 addtonal nformaton asymmetry around corporate events days. We have data about corporate announcements and ther dates from Bloomberg (see Appendx Table 3). The database ncludes corporate meetngs, and captal changes, as well as all press conferences on company-related news. Table 6 reports results for regresson (7). Although many coeffcents are nsgnfcant, almost all are postve. Ths supports our story of more nformatonal tradng around corporate meetngs, even for blue chps. For example, coeffcent for Gazprom, Russa s largest company and ts natural gas monopoly s around event perods

18 Our results are consstent wth several papers nvestgatng the relaton between the dynamcs of return and tradng volume usng US data. For example, Stckel and Verreccha (1994) fnd that when earnngs announcements are accompaned by hgher volume, returns tend to be postvely correlated. Ther results ndcate that earnngs announcements may generate a large amount of prvate nformaton that lead to actve prvate-nformaton tradng and return contnuaton orporate Governance Issues The relaton between the ntensty of prvate-nformaton tradng and Russan corporate governance s our next topc. We use the corporate governance rsk rankng for Russan companes computed by the nvestment bank Brunswck UBS Warburg (see Appendx Table ). Although we lmt our cross-sectonal analyss to only 15 companes, ther stocks account for some 90% of Russan market captalzaton. onsder the followng regresson specfcaton: = a + a GovernanceRsk + error, (8), 0 1 We ran OLS regresson (8) and corrected errors for heteroskedastcty. The results are n Fgure 1 and. The governance-rsk rankng s clearly postvely related to the ntensty of prvate-nformaton tradng. The relatonshp s statstcally sgnfcant. The results are smlar usng obtaned from regresson (3). Whle we are aware that the sample s relatvely small, our results suggest a possble relatonshp between corporate governance and prvate-nformaton tradng. 7. oncluson Ths paper develops a new approach to extract nformaton on corporate governance n emergng markets based on tradng data. It measures the ntensty of prvate 14 We also looked how the nature of trades changes wth the ntroducton of Amercan Depostary Recepts (ADRs) on a Russan stock. The hypothess was ths facltates hdng nformatonal trades. Ths means that the nformatonal-allocatonal tradng pattern should change as a result of nternatonal cross-lstng. However, we dd not fnd any change. Ths s not surprsng, as the ADR market for Russan stocks s very thn and one generally s tradng aganst large nsttutonal market makers. (see Domowtz, Glen, and Madhavan (1998) for a dscusson on nternatonal cross-lstng and segmentaton of stock markets). 18

19 nformaton tradng by examnng the dynamc relaton between return and volume of ndvdual stocks n Russa and other emergng markets. It then relates the measure of prvate-nformaton tradng to legal envronment and corporate governance. Our analyss s based on a smple theoretcal model of Llorente, Mchaely, Saar, and Wang (001), whch shows that returns generated by rsk-sharng trades tend to reverse themselves whle returns generated by prvate nformaton trades tend to contnue themselves. Our emprcal study fnds strong evdence of prvate-nformaton tradng n emergng markets. Usng corporate announcement data from Russa, we fnd prvatenformaton tradng s especally strong around major corporate event dates. We also fnd such tradng n Russan stocks s related to poor corporate governance. Stocks n countres that enforce nsder-tradng laws and provde better nvestor protecton exhbt less return contnuaton followng hgh volume days. Moreover, ntense prvate-nformaton tradng reflects a hgh degree of expropraton rsk and poor mnorty shareholder protecton. Thus, the ntensty of prvate-nformaton tradng can be used as a possble measure of corporate governance (broadly defned) for rankng emergng market stocks. The robustness of our results s examned along several dmensons. Frst, we decompose both the volume and return seres nto systematc and dosyncratc components. Informaton asymmetry remans when we remove the market-wde varatons from the analyss. Second, our fndngs are not senstve to alternatve defntons of tradng volume. We have made a methodologcal contrbuton to the prvate nformaton lterature by combnng the tradtonal event study approach wth LMSW regressons. Prevously, studes of prvate-nformaton (nsder) tradng used cumulatve abnormal returns around event wndows to measure the mpact of prvate nformaton on stock returns (see, for example, Bhattacharya, Daouk, Jorgenson, and Kehr (000) and Banerjee and Eckard (001)). The ntuton behnd our approach s that nsders and others wth materal nformaton related to corporate event explot ther nformaton advantage by tradng aganst unnformed outsders. Thus, condtonal on corporate event and hgh tradng volume, we are more lkely to observe return contnuaton when there s nformaton asymmetry. 19

20 However, there are several caveats for our measure of nformaton asymmetry. Frst, the measure s based on hstorcal data. There s no guarantee that ths asymmetry wll persst n the future. Ths s especally the case f nvestors use t to avod stocks that have a hgh degree of bad corporate governance. Ths may cause corporate nsders to change ther behavor n order to attract lqudty traders or unnformed nvestors. Second, prvate nformaton tradng may happen nfrequently and vary n ts ntensty and tradng frequency. For example, f nsder tradng s conducted wthn a few hours or over a few weeks rather than over several days, t would be hard for an econometrcan to detect t usng only daly return and turnover data. Moreover, although we provde some measure for the ntensty of nformaton asymmetry and ts relatonshp to poor corporate governance, t s not a drect measure of fnancal loss lkely to be ncurred by unnformed nvestors. Although the measures developed n the paper are certanly qute crude nstruments, they do provde some ndependent nformaton on the status of corporate governance of emergng market companes. A byproduct of our nformaton measure s that t may sometmes serve as a lqudty proxy when s negatve. Pastor and Stambaugh (00) demonstrated that aggregated could be used as a proxy for the lqudty factor n the US market. It would be nterestng to see whether smlar results can be obtaned n emergng markets as well. There are several ssues that reman to be examned. It s nterestng to know whether the measure s persstent. Gven that nsders face lttle rsk of prosecuton n many markets, we conjecture that they may contnue to explot ther nformaton advantage by tradng aganst unnformed nvestors. Thus, return contnuaton on hgh tradng volume could persst and be reflected n our measure. As a result, we conjecture that could provde an ex ante as well as ex post measure of nformaton asymmetry aganst mnorty shareholders. However, we need to confrm ths ntuton wth more emprcal work. It s nterestng to know how our measure of nformaton asymmetry (and corporate governance) s related to the cost of captal. Presumably, unnformed nvestors may stay away from stocks n whch others have a dstnct nformaton advantage. Ths may ncrease cost of captal as found for nsder tradng (Bhattacharya and Daouk (00)). 0

21 How nformaton asymmetry and poor corporate governance n emergng market stocks affects the nvestment strateges of unnformed (or poorly nformed) global nvestors also s an mportant ssue n need of further research. 1

22 References Albuquerque, R., G. Bauer, and M. Schneder, 00, haracterzng Asymmetrc Informaton n Internatonal Equty Markets, Workng paper. Antonewcz, R.L., 1993, Relatve Volume and Subsequent Stock Prce Movements, Workng paper, Board of Governors of the Federal Reserve System. Baley, W., and. X. Mao, 001, Investment Restrctons and the ross-border Flow of Informaton: Some Emprcal Evdence, Workng paper, ornell Unversty. Baley, W. and J. Lm, 199, Evaluatng the Dversfcaton Benefts of the New ountry Funds, Journal of Portfolo Management, 17, Banbrdge, Stephen, 000, Insder tradng, n Encyclopeda of Law and Economcs III (Edward Elgar Publshng, heltenham, U.K.). Banerjee, A. and W. Eckard, 001, Why Regulate Insder Tradng? Evdence from the Great Merger Wave ( ), Amercan Economc Revew, Bekaert, G., and. Harvey, 1995, Tme-Varyng World Market Integraton, Journal of Fnance, Vol. 50, Bekaert, G., and. Harvey, 1997, Emergng Equty Market Volatlty, Journal of Fnancal Economcs, 43:1, Bekaert, G. and. Harvey, 000, Foregn Speculators and Emergng Equty Markets, Journal of Fnance, Bekaert, G.,. Harvey, and A. Ng, Market Integraton and ontagon, Journal of Busness, forthcomng. Bhattacharya, U., and H. Daouk, 00, The world prce of nsder tradng, Journal of Fnance, 57, Bhattacharya, U., and H. Daouk, Bran Jorgenson, and arl-henrch Kehr, 000, When an event s not an event: The curous case of an emergng market, Journal of Fnancal Economcs 55, Bhattacharya, U., H. Daouk, and M. Welker, 003, The World Prce of Earnngs Opacty, Accountng Revew, forthcomng. Black, Bernard, 001a, Does orporate Governance Matter? A rude test Usng Russan Data, Unversty of Pennsylvana Law Revew, Vol. 149, Black, Bernard, 001b, The orporate Governance and Market Value of Russan Frms,

23 Emergng Markets Revew,. Brennan, M, and H. ao, Internatonal Portfolo Investment Flows, Journal of Fnance, Vol. 5, No. 5. (Dec., 1997), Brennan, M. and A. Subrahmanyam, 1995, Investment Analyss and Prce Formaton n Securtes Markets, Journal of Fnancal Economcs 38, ampbell, J.Y., S.J. Grossman, and J. Wang, 1993, Tradng Volume and Seral orrelaton n Stock Returns, Quarterly Journal of Economcs 108, onrad, J., A. Hameed, and.m. Nden, 199, Volume and Autocovarances n Short- Horzon Indvdual Securty Returns, Journal of Fnance 49, Domowtz I., J. Glen, and A. Madhavan, 1998, Internatonal ross-lstng and Order Flow Mgraton: Evdence From An Emergng Market, Journal of Fnance 53, Easley, D., and M. O'Hara, 1987, Prce, Trade Sze, and Informaton In Securtes Markets, Journal of Fnancal Economcs 18, Gallant, R., P. Ross, and Tauchen, G., 199, Stock Prces and Volume, Revew of Fnancal Studes 5, Glosten, L. and P. Mlgrom, 1985, Bd, Ask and Transacton Prces n a Market-Maker Market wth Heterogeneously Informed Traders, Journal of Fnancal Economcs 14, Harvey,., 1991, The world prce of covarance rsk, Journal of Fnance 46, Harvey,., 1995a, Predctable rsk and returns n emergng markets, Revew of Fnancal Studes 8, Harvey,., 1995b, The ross-secton of Volatlty and Autocorrelaton n Emergng Markets, Fnanzmarkt und Portfolo Management 1, Hayes, R.M., 1998, The mpact of tradng commsson ncentves on analysts stock coverage decsons and earnngs forecasts, Journal of Accountng Research, 36, Henry, Peter, 000, Stock Market Lberalzaton, Economc Reform, and Emergng Market Equty Prces, Journal of Fnance, Vol. LV, No., Henry, Peter, 000, Do Stock Market Lberalzatons ause Investment Booms? Journal of Fnancal Economcs, Vol. 58, No.1-, La Porta, R., F. Lopez-de-Slanes, A. Shlefer, and R. Vshny, 1997, Legal determnants of external fnance, Journal of Fnance 5,

24 La Porta, R., F. Lopez-de-Slanes, A. Shlefer, and R. Vshny, 1998, Law and fnance, Journal of Poltcal Economy 106, Lm, T., 001, Ratonalty and analysts forecast bas, Journal of Fnance, 56, Llorente G., R. Mchaely, G. Saar, and J. Wang, 00, Dynamc Volume-return Relaton of Indvdual Stocks, Revew of Fnancal Studes, Vol. 15, Lo, A.W., and A.. MacKnlay, 1990, An Econometrc Analyss of Nonsynchronous Tradng, Journal of Econometrcs 45, Lo, A. and J. Wang, 000, Tradng Volume: Defntons, Data Analyss, and Implcatons of Portfolo Theory, Revew of Fnancal Studes 13, Madhavan, A., M. Rchardson, and M. Roomans, 1997, Why Do Securty Prces hange? A Transacton-Level Analyss of NYSE Stocks, Revew of Fnancal Studes 10, Mchaely and Womack, 1999, onflct of Interest and The redblty of Underwrter Analyst Recommendatons, The Revew of Fnancal Studes, 1999, 1, Pastor and Stambaugh, Lqudty Rsk and Expected Stock Returns, Journal of Poltcal Economy, forthcomng. Roll, R, 1984, A Smple Implct Measure of the Effectve Bd-Ask Spread n an Effcent Market, Journal of Fnance, 39, Stckel, S.E. and R.E. Verreccha, 1994, Evdence that Volume Sustans Prce hanges, Fnancal Analyst Journal (November-December),

25 Table 1. Descrptve Statstcs for Emergng Markets The sample covers Dec 30, 1994, to Nov 1, 001. ountry market captalzaton s as of Jan 11, 00, based on the sample. For Argentna, foregn companes are excluded because no data are reported for them n Datastream fles. ountry Mkt ap Stocks Daly Mean Daly Std. Average (BN $US) Dev. Turnover Autocorrelaton Obs Arthmetc % Geometrc % % Mean % Std. Dev % Mean ρ 1,cs Medan ρ 1,cs Std. Dev. ρ 1,cs Argentna Brazl hle olomba Greece Inda Indonesa South Korea Malaysa Mexco Pakstan Peru Phlppnes Portugal Russa Sr Lanka Thaland Turkey Venezuela

26 Table. Dynamc Volume-Return Relatonshp, Emergng-Markets Sample Panel A: Results of the regresson analyss for the model R, t+ 1 = 0 + 1R, t +, R, tv, t + ε, t+ 1, where ndexes the companes n the correspondng country from our sample. s the contnuously compounded daly return, and denotes the company turnover (tradng volume). The fracton of postve coeffcents (column four) and V, t the fracton of 5%-level-sgnfcant coeffcents (column fve) are reported n percentage ponts. Stocks robust to MA specfcaton (column seven) reports the number of stocks whose coeffcents do not change sgn under alternatve specfcatons for turnover (n partcular when we detrend the turnover measure by a 60-day and 10-day movng average.) ountry Total number of stocks Mean 1 Mean % > 0 out of total % t( ) > 1.95 out of total R, t Mean adj. R Stocks robust to MA (1) () (3) (4) (5) (6) (7) Argentna Brazl hle olomba Greece Inda Indonesa South Korea Malaysa Mexco Pakstan Peru Phlppnes Portugal Russa Sr Lanka Thaland Turkey Venezuela