Essays on Stock Market Liquidity and Liquidity Risk Premium

Transcription

1 Universiy of New Orleans Universiy of New Orleans Theses and Disseraions Disseraions and Theses Essays on Sock Marke Liquidiy and Liquidiy Risk Premium Shu Tian Universiy of New Orleans Follow his and addiional works a: hps://scholarworks.uno.edu/d Recommended Ciaion Tian, Shu, "Essays on Sock Marke Liquidiy and Liquidiy Risk Premium" (2010). Universiy of New Orleans Theses and Disseraions hps://scholarworks.uno.edu/d/1153 This Disseraion is brough o you for free and open access by he Disseraions and Theses a ScholarWorks@UNO. I has been acceped for inclusion in Universiy of New Orleans Theses and Disseraions by an auhorized adminisraor of ScholarWorks@UNO. The auhor is solely responsible for ensuring compliance wih copyrigh. For more informaion, please conac scholarworks@uno.edu.

2 Essays on Sock Marke Liquidiy and Liquidiy Risk Premium A Disseraion Submied o he Graduae Faculy of he Universiy of New Orleans in parial fulfillmen of he requiremens for he degree of Docor of Philosophy in Financial Economics by Shu Tian B.S. Shandong Universiy, China, 1998 M.S. Sam Houson Sae Universiy, 2005 M.S. Universiy of New Orleans, 2008 May, 2010

3 Copyrigh 2010, Shu Tian ii

4 Table of Conens Absrac...v Chaper One: Inroducion...1 Chaper Two: Liquidiy, Macro Facors and he U.S. Equiy Flows o Emerging Markes 3 I. Inroducion...3 II. Relaed Lieraure...4 II. A. Possible Macro Facors Affecing Equiy Flows...4 II. A. 1. Counry Equiy Marke Reurn...4 II. A. 2. Marke Openness...5 II. A. 3. Counry Risk...5 II. A. 4. Trading Resricions...6 II. B. The Role of Liquidiy...7 III. Our Major Hypohesis...8 IV. Daa...8 IV. A. Daa on U.S. Equiy Flows. 8 IV. B. Daa on Various Macro Facors..8 IV. C. Measuremens of Liquidiy...10 IV. D. Summary Saisics.. 11 V. Mehodology and Empirical Resuls...16 V. A. Time-Series Cross-Secion Daa Analysis...16 V. A. 1. The Model...16 V. A. 2. Resuls on Conrol Variables...19 V. A. 3. The Impac of Liquidiy on Equiy Flows...21 V. A. 4. Does he Pas Liquidiy Maer?...28 V. B. A Comparison of Shor and Long-Horizon U.S. Equiy Flows...32 V. C. Dynamic Specificaions...46 VI. Conclusion and Conribuions...52 REFERENCES 54 Table I. Summary Saisics for Various Variables, Jan Dec Table II. The Impacs of Conrol Variables and Liquidiy on U.S. Equiy Flows Table III. Regressions wih Lagged Liquidiy.29 Table IV. Esimaion of he Average Twelve-Monh-Ahead U.S. Equiy Flows...36 Table V. Esimaion of he One-Monh-Ahead U.S. Equiy Flows...41 Table VI. Esimaion of a Dynamic Specificaion...49 Figure 1 Monhly Foreign Equiies Purchased by U.S. Invesors by counry...56 Figure 2 Monhly Foreign Equiies Purchased by U.S. Invesors by region...57 Figure 3 Monhly Foreign Marke Turnover...58 Figure 4 Monhly Foreign Marke Trading Volume...59 Figure 5 Monhly Foreign Marke Price Impac...60 Appendix...61 Table AI. Daa Sources...61 Table AII. Local Sock Markes of Fifeen Emerging Counries...62 Chaper Three: Liquidiy Risk Premium Puzzle and Possible Explanaions...63 I. Inroducion...63 II. Relaed Lieraure...64 iii

5 II. A. Liquidiy and Reurn...64 II. B. Volailiy of Liquidiy...65 II. C. Liquidiy Measures...66 II. D. Behavioral Finance Argumens as Possible Explanaions for he Puzzle...67 III. Mehodology and Daa...68 III. A. A Model for he Relaion beween Excess Reurn and Liquidiy Risk...68 III. B. Daa...71 III. B. 1. Daa Selecion...71 III. B. 2. Descripive Saisics...72 IV. Empirical Relaions beween Reurns and Liquidiy Risk...76 Sub-Period Resuls and Oher Robusness Checks...80 A Summary on Liquidiy Risk Puzzle...84 V. Poenial Facors ha Migh Influence he Reurn-Liquidiy Risk Relaion...84 V. A. The Impac of Marke Condiion on he Relaion beween Liquidiy Risk and Excess Reurn...84 V. B. Small vs. Large...89 V. C. The Role of Invesor Senimen...92 VI. Conclusions and Conribuions...97 REFERENCES...99 Table I. Descripive Saisics...74 Table II Correlaion beween Excess Reurn and Explanaory Variables...75 Table III. Regression of Excess Reurn on Liquidiy, Liquidiy Risk and Conrol Variables...79 Table IV. Regression of Excess Reurn on Explanaory Variables for Sub-periods...81 Table V. The Impac of Marke Condiion on he Puzzle...88 Table VI. The Impac of Firm Size on he Puzzle...91 Table VII. The Impac of Invesor Senimen...95 Figure I. Time Series Plos of Liquidiy Measures Appendix Table A1. Regression wihou Liquidiy Risk Measure Table A2. Regression of Excess Reurn on he Firs Difference of Non-Saionary Explanaory Variables Table A3. The Impac of Value Socks on he Puzzle Via iv

6 Absrac This disseraion addresses issues concerning liquidiy and is volailiy. I consiss of wo essays. The firs essay, Liquidiy, Macro Facors and he U.S. Equiy Flows o Emerging Markes, examines he role of liquidiy on equiy flows from he U.S. o fifeen emerging markes around he world. Since liquidiy has many dimensions, an emphasis is placed on uilizing various measures of liquidiy. Moreover, boh saic and dynamic analyses, as well as shor and long-horizon regressions, are performed o invesigae he research quesions. The resuls sugges ha a liquid marke aracs flows, afer conrolling for marke size, poliical openness, exchange rae and oher macro facors. Addiionally, evidence indicaes ha he imporance of liquidiy varies across regions. For insance in he Asian region, he relaion beween equiy flows and volume-relaed liquidiy is weak while ha beween flows and price impacs of rading is srong. Evidence also suppors he relevance of macro facors such as a counry s economic freedom. The second essay, Liquidiy Risk Premium Puzzle and Possible Explanaions, aemps o resolve he liquidiy risk puzzle: a negaive relaion beween reurns and liquidiy risk, documened by Chordia, Subrahmanyam, and Anshuman (2001b), by employing alernaive liquidiy measures and by incorporaing facors ha migh poenially affec he relaion. The main findings are as follows. The relaion beween sock reurns and volailiy of liquidiy depends on he measure of liquidiy. When liquidiy measures are based on rading volume, he resuls are largely mixed, bu when liquidiy is measured based on price impac of rading, he relaion beween reurns and volailiy of price impacs is posiive, as expeced. The resuls are sensiive o ime periods examined. Moreover, during exreme down markes, he aversion o liquidiy volailiy is lower, suggesing behavioral bias migh poenially address he puzzle. Empirical findings also sugges ha liquidiy risk premium ends o be greaer for small socks. Finally, when he VIX index is included as a proxy for invesor senimen, he resuls indicae ha he relaion beween reurns and liquidiy risk is significanly posiive in four ou of five liquidiy measures. In sum, he empirical analysis parially bu no compleely addresses he puzzle. Keywords: Emerging Marke, Equiy Flows, Liquidiy, Liquidiy Risk Premium Puzzle, Marke Condiions, Invesor Senimen v

7 Chaper One: Inroducion This disseraion addresses issues concerning liquidiy and is volailiy. I consiss of wo essays. One of hem examines he relaion beween U.S. equiy flows o emerging markes and various macro facors, paricularly he liquidiy of emerging markes. The oher paper aemps o resolve he liquidiy risk puzzle: a negaive relaion beween reurns and liquidiy risk. The firs essay, presened in Chaper Two, is Liquidiy, Macro Facors and he U.S. Equiy Flows o Emerging Markes. Liquidiy in emerging sock markes is ypically low and may be a major concern o foreign invesors. This essay examines he role of liquidiy on equiy flows from he U.S. o fifeen emerging markes around he world, for he period of 1995 o Since liquidiy has many dimensions, an emphasis is placed on uilizing various measures of liquidiy. Moreover, boh saic and dynamic analyses, as well as shor and long-horizon regressions, are performed o invesigae he research quesions. The resuls sugges ha a liquid marke aracs flows, afer conrolling for marke size, poliical openness, exchange rae and oher macro facors. Addiionally, evidence indicaes ha he imporance of liquidiy varies across regions. For insance in he Asian region, he relaion beween equiy flows and volumerelaed liquidiy is weak while ha beween flows and price impacs of rading is srong, reinforcing he noion ha here are muliple dimensions in liquidiy. Evidence also suppors he relevance of macro facors such as a counry s economic freedom. The second essay is Liquidiy Risk Premium Puzzle and Possible Explanaions. This sudy aemps o resolve he liquidiy risk puzzle: a negaive relaion beween reurns and liquidiy risk, documened by Chordia, Subrahmanyam, and Anshuman (2001b), by employing alernaive liquidiy measures and by incorporaing facors ha migh poenially affec he relaion. The sample covers he period 1975 o 2008 and he oal number of firm-monh observaions is 415,403. The main findings are as follows. The relaion beween sock reurns and volailiy of liquidiy depends on he measure of liquidiy. When liquidiy measures are based on rading volume, he resuls are largely mixed, bu when liquidiy is measured based on price impac of rading, he relaion beween reurns and volailiy of price impacs is posiive, as expeced. Moreover, he resuls are sensiive o ime periods examined. The second par of he sudy incorporaes poenial facors ha migh affec he relaion beween reurns and liquidiy volailiy. The resuls indicae ha during exreme down markes, he aversion o liquidiy 1

8 volailiy is lower, suggesing behavioral bias migh poenially address he puzzle. Empirical findings also sugges a firm size effec; specifically, liquidiy risk premium ends o be greaer for small socks. Finally, when he VIX index is included as a proxy for invesor senimen, he resuls indicae ha he relaion beween reurns and liquidiy risk is significanly posiive in four ou of five liquidiy measures. In sum, he empirical analysis parially bu no compleely addresses he puzzle. 2

9 Chaper Two: Liquidiy, Macro Facors and he U.S. Equiy Flows o Emerging Markes I. Inroducion The financial crises during highligh he effecs and imporance of cross-border capial flows. Some sudies aemp o examine facors ha influence inernaional equiy invesmen flows. Aggarwal, Klapper, and Wysocki (2005) and Gelos and Wei (2005) find ha equiy flows and ransparency are correlaed, while Edison and Warnock (2003a) sugges ha flows are affeced by capial conrol and cross-border lising, conrolling for macro facors such as relaive sock performances. Kaminsky, Lyons, and Schmukler (2001) also provide an overview of muual fund invesmens in emerging markes. They find ha equiy flows o emerging markes are volaile. No emphasized in hese sudies is he poenially imporan role of liquidiy on equiy flows, which is somewha surprising given ha he lieraure on marke microsrucure generally indicaes ha liquidiy has imporan effecs on pricing, porfolio holdings, and rading 1. Emerging markes are largely illiquid because of heir limied access o world capial markes. Thus, marke liquidiy is likely o be imporan from he perspecive of foreign invesors, which moivaes his research. Specifically, we are raising he quesion wheher liquidiy also influences equiy flows. Mos of he exan empirical liquidiy lieraure focuses on he U.S. marke. Wih he increasing availabiliy of daa regarding rading on emerging markes in recen years, researchers are able o ake a closer look a he liquidiy of hese markes, where liquidiy may be a major concern. Indeed, Lesmond (2005) indicaes ha liquidiy varies considerably across emerging markes. Bekaer, Harvey, and Lundblad (2007) poin ou ha rading volume based liquidiy proxy in emerging markes does no predic fuure reurns bu some oher liquidiy measures do, while Jun, Marahe, and Shawky (2003) find a posiive correlaion beween reurns and urnover in emerging markes. These findings reinforce he idea ha liquidiy has many dimensions, including rading coss, rading frequency, rading inensiy, and price impacs. Wih his in mind, we employ several measures of liquidiy o deermine wheher foreign invesors desire liquidiy; if so, i should be refleced by heir invesmen flows o emerging markes. However, 1 These include, among many oher sudies, he seminal work by Kyle (1985) and Amihud and Mendelson (1986). More recen papers such as Acharya and Pedersen (2005) decompose liquidiy risk and furher confirm ha liquidiy risk affecs asse prices. 3

10 due o he unavailabiliy of daa on rading coss, our liquidiy measures primarily capure rading frequency, rading inensiy, and price impacs of rading. Essenially, his sudy also invesigaes wheher or no he differen dimensions of liquidiy measures are consisenly influencing equiy flows. In addiion o he relaion beween equiy flows and liquidiy, his sudy considers he correlaions beween equiy flows and various macro facors including marke size, prior marke excess reurns, counry s economic freedom, capial conrol, changes in exchange rae, and rading resricions. Mos facors are shown by various papers o affec equiy flows o emerging markes. In brief, his sudy is comprehensive in erms of he number of facors being considered. We expec ha he above-menioned facors, as well as liquidiy, are correlaed wih he U.S. equiy flows o emerging markes. The differen measures of liquidiy may resul in differen impacs on equiy flows. The remainder of he paper is organized as follows. The nex secion discusses he mos relaed research, followed by a secion describing he hypoheses. Following ha, we describe he daa, variables, and various measures of liquidiy in Secion IV. Mehodology and empirical resuls are presened in Secions V. Secion VI conains concluding remarks. II. Relaed Lieraure II. A. Possible Macro Facors Affecing Equiy Flows II. A. 1. Counry Equiy Marke Reurn Earlier sudies on equiy flows o emerging markes focus on he relevance of macroeconomic variables especially sock marke reurns. In an early sudy ha covers boh emerging and developed counries, Bohn and Tesar (1996) documen no srong relaion beween sock reurns and equiy flows. Brennan and Cao (1997) heorize ha, when foreign invesors face an informaion disadvanage, hey end o chase foreign reurns; ha is, buy when foreign reurns are high and sell when reurns are low. Using daa on several emerging markes in early 1990s, 2 hey find evidence supporive of heir model. Using daily flow daa from one invesmen firm, Froo, O Connel, and Seasholes (2001) presen evidence also consisen wih flows being affeced by pas reurns. Addiionally, hey documen ha flows have predicive power for fuure reurns. More recenly, Griffin, Nardari, and Sulz (2002) provide an empirical invesigaion 2 Generally, early 1990s is when daa on emerging markes firs became available. 4

11 using daily fund flow daa, which hey obain direcly from exchanges from nine emerging markes. They find ha equiy flows are posiively correlaed o pas hos counry reurns as well as home counry reurns. Tha is, he evidence suggess ha equiy flows are boh pushed and pulled by relaed marke reurns; however, he effecs are relaively shor-lived presen only in daily daa bu no so wih he use of weekly daa. Overall, he evidence concerning he effecs of marke reurns are no conclusive. II. A. 2. Marke Openness Edison and Warnock (2003b) find ha flows are relaed o he degree of capial conrol, as well as economic condiions proxied by ineres raes. They poin ou ha he inensiy of capial conrol indicaes he openness o foreign invesmen, which in urn measures a governmen s commimen o free marke policies. Their evidence suggess ha markes are more likely o be viable if hey have less resricion imposed on foreign invesors. II. A. 3. Counry Risk Counry risk, such as poliical risk, exchange rae movemen, economic risk, ec, affecs he invesmen climae wihin a counry and he allocaion of foreign invesmen. Exan lieraure explores various indicaors for counry risk and heir impacs on inernaional rading. Here we highligh counry poliical risk, exchange rae movemen, sock marke ransparency, invesor proecion, and economic freedom as componens of counry risk. Erb, Harvey, and Viskana (1996) find a link beween cross-border invesmens and poliical risk and sugges ha poliical sabiliy is necessary o suppor free marke developmen and arac and reain long-erm sources of capial. Blonigen (1997) saes ha exchange raes movemens play an imporan role in influencing foreign direc invesmen. A counry s exchange rae regime developed by Reinhar and Rogoff (2003) is used as an indicaor of wheher he exchange rae is pegged, managed floa, or floaing. Gelos and Wei (2005) use he Global Compeiiveness Repor o measure ransparency 3 and poin ou ha fund inflows are on average greaer in counries and in firms wih a greaer 3 Financial ransparency refers o invesors access o ruhful, accurae and relevan informaion on he condiions in he counries and companies in which hey are invesing. Gelos and Wei (2005) develop a ransparency index for each counry and firm, based on several surveys conduced by Inernaional Moneary Fund and 5

12 degree of ransparency. Moreover, some sudies also use accouning sandards 4 as anoher proxy for ransparency. For example, Aggarwal e al. (2005) uilize boh counry-level and firm-level daa on corporae governance including accouning sandards. Based on U.S. fund holdings in 2002, hey find ha funds end o inves more in counries wih sronger accouning sandards and shareholder proecions. Parly in response o he crisis in confidence regarding corporae governance, a few recen papers examine he role of corporae governance in equiy flows across counries, saring wih he seminal work of La Pora, Lopez-de-Silanes, and Shleifer (1997) ha esablishes a link beween share holdings and shareholder proecion. 5 Wurgler (2000) furher finds ha capial is more efficienly allocaed in counries wih beer legal proecion for minoriy invesors and more firm-specific informaion in domesic sock reurns. Economic freedom measures counry s economic performance and he consisency of is insiuions and policies. Kim (2008) poins ou ha greaer economic freedom implies fewer barriers o economic aciviies and ends o generae more opporuniies for people and creae lasing prosperiy. II. A. 4. Trading Resricions Boh rading resricions and liquidiy are imporan consideraions for raders ha place a high value on liquidiy. We consider shor-selling resricions as anoher imporan aspec for he ease of rading. Bris, Goezmann, and Zhu (2007) presen evidence ha shor-selling consrains affec a marke s degree of efficiency. They find ha prices reflec negaive informaion faser in counries where shor-selling is praciced. In ha paper, hey carefully collec informaion abou shor-selling pracices in various counries and consruc a shor-selling indicaor, which we uilize here. Moreover, Ahearne, Griever, and Warnock (2004) find ha he porion of a PricewaerhouseCoopers from lae 1990s and early 2000s. Morck, Yeung and Yu (2000) 3 show ha R 2 and oher measures of sock marke synchroniciy, as alernaive measures of ransparency, are higher in counries wih less developed financial sysems and poorer corporae governance. They also poin ou ha R 2 is inversely relaed o he degree of invesor proecion. 4 Accouning sandards indicae he exen o which publicly raded companies in he counry uilize eiher US GAAP or IAS in financial reporing, and wheher he counry is a member of he Inernaional Accouning Sandards Council. 5 Daouk, Lee and Ng (2006) furher show ha good governance reduces cos of capial and increases liquidiy and pricing efficiency. 6

13 counry s marke ha has a public U.S. lising is a major deerminan of a counry s weigh in U.S. invesors porfolio. II. B. The Role of Liquidiy Some lieraure provides evidence ha marke liquidiy has imporan effec on risk of emerging marke invesmen and, herefore, on he fuure reurns. As Bekaer e al. (2002) find, liquidiy measure significanly predics fuure reurns and equiy marke liberalizaion significanly improves he level of liquidiy. As a resul, he increasing invesmen ineres in emerging markes relaive o developed markes yields specacular reurns, which are subjec o increased risk and are significanly reduced by he increased illiquidiy of rading socks in emerging markes. As menioned in he inroducion, sudies such as Lesmond (2005) and Bekaer e al. (2007) sugges ha liquidiy has many dimensions. Lesmond (2005) indicaes ha liquidiy varies considerably across emerging markes. He finds ha price-based measures of liquidiy are more correlaed wih ransacion coss compared o hose based on rading volume. Similarly, Bekaer e al. (2007) presen evidence ha rading volume based liquidiy proxy in emerging markes does no predic fuure reurns while some oher liquidiy measures do. In addiion, Amihud (2002) measures marke illiquidiy as he daily raio of absolue sock reurn o is dollar volume averaged over some period and poins ou ha expeced marke illiquidiy posiively affecs ex ane sock excess reurn, whereas Jun e al. (2003) documen a posiive correlaion beween reurns and urnover in he markes for emerging counries, a puzzling resul conrary o he exisence of liquidiy premium. Lieraure indicaes he muliple dimensions of liquidiy measure: rading coss, rading frequency, rading inensiy, and price impacs. While some sudies on equiy flows include share urnover as a conrol variable, here is no sudy ha considers he possibiliy ha one single liquidiy measure migh no compleely capure a counry s sock marke liquidiy. Consequenly, we use hree differen measures of liquidiy in his sudy, deailed in Secion IV. In addiion, some sudies sugges ha he size of marke can be a proxy for liquidiy, hus we also include marke size as an alernaive measure of liquidiy. On he oher hand, marke size can also serve as a proxy for marke visibiliy or ransparency. Therefore, we consider he variable marke size. 7

14 III. Our Major Hypohesis If invesors demand liquidiy, hen markes/socks wih greaer liquidiy likely will arac more aenion and invesmens. Specifically, we hypohesize ha Emerging counries wih more liquid equiy marke and less rading resricions are likely o be associaed wih higher equiy inflows from he U.S. invesors. The resuls are expeced o be robus wih respec o muli-dimensions liquidiy measures. Based on previous lieraure, we expec he relaion beween equiy flows and prior counry marke excess reurns 6 o be posiive due o rading feedback. We also expec ha capial conrol is negaively correlaed wih he U.S. equiy flows ino emerging markes. Also, higher equiy flows from U.S. o emerging markes would be associaed wih higher local marke capializaion, greaer economic freedom, and favorable exchange rae movemens. IV. Daa IV. A. Daa on U.S. Equiy Flows The primary daa source for he U.S. equiy flows o emerging markes is he U.S. Treasury Inernaional Capial (TIC) Reporing Sysem, which repors porfolio equiy flows from he U.S. o emerging counries. The flow daa is available a he monhly frequency. In his sudy he U.S. equiy flows o a given emerging marke are he U.S. residens gross purchases of foreign socks from residens of ha counry scaled by he counry s GDP. Figure 1 presen he U.S. equiy inflows ino our fifeen sample emerging counries. Obviously, equiy flows vary considerably across counries. During he sample period he U.S. invesors purchase a larger amoun of equiies from Lain America, specifically Brazil and Mexico, han from he oher wo regions, as shown in Figure 2. IV. B. Daa on Various Macro Facors The daa on various explanaory variables are colleced from muliple sources. Table AI in Appendix provides deails of daa sources. Our sample 7 consiss of fifeen emerging markes 6 Edison and Warnock (2003a) find evidence ha invesors are chasing prospecive reurns, as proxied by dividend yields, bu no pas reurns. Moreover, he evidence on pas-reurns-chasing behavior is mixed. Using monhly daa, Bohn and Tesar (1996) find ha invesors chase pas reurns in 7 of 22 markes. 7 We also analyze annual daa. The empirical resuls are no repored. 8

15 for he period beween January 1995 and December These emerging counries are classified ino hree subgroups based on regions: Lain America Argenina, Brazil, Chile, Mexico, and Peru, Europe, Africa, and Middle Eas Israel, Poland, Souh Africa, and Turkey, and Asia Indonesia, Malaysia, Philippines, Taiwan, Thailand, and Souh Korea. Due o daa unavailabiliy, daa is incomplee for some variables under invesigaion. The monhly daa analysis includes he following conrol variables: counry sock marke size, he pas excess counry marke reurn, counries economic freedom, capial conrol, exchange rae movemen, and shor-selling resricion. These variables are seleced based on he earlier discussions of lieraure and described as follows: 1) Marke Size: I is measured by counry equiy marke capializaion as a percenage of counry s GDP, boh denominaed in curren U.S. dollars. As menioned earlier, marke size can be a proxy for liquidiy. 2) Prior Excess Local Sock Marke Reurns 8 : The excess reurn is compued by subracing he value-weighed reurn on all NYSE, AMEX and NASDAQ socks from he local sock marke reurn. Table AII in Appendix conains he informaion regarding he local sock markes of he fifeen emerging counries. 3) Counry s Economic Freedom 9 : The index of economic freedom is based on en componens of economic freedom which represens openness o he global commerce, ransparency, and he rule of law, and raed on a scale of zero (leas free) o 100 (mos free). Also, a higher score implies a lower level of governmen inerference in he markeplace. The naural logarihm of he index is used for regression analysis. 4) Capial Conrol 10 : I represens changes in foreign ownership resricion or he inensiy of capial conrols, which is presened by Edison and Warnock (2003b). The measure of capial conrol can vary from zero o one, wih zero represening a compleely open marke wih no resricions, and a value of one indicaing ha he marke is compleely closed. However, heir daa only covers he period from January 1989 o December Some sudies use dividend yield o esimae prior equiy marke performance. We use i in our annual daa analysis. Bekaer e al. (2002) use lagged excess reurns (excess of he foreign marke over he U.S. marke) in heir sudy. 9 Our sudy excludes he measure of ransparency because he daa on ransparency is fairly limied and seldom undaed in emerging markes. Because currenly he Inernaional Accouning Sandards are undergoing significan revisions, we do no analyze he impac of accouning sandard eiher. 10 Because he daa used o compue capial conrol is propriey daa which is no accessible o he general public, we ake he daa from Edison and Warnock (2003b). 9

16 5) Changes in Exchange Rae 11 : Presumably exchange rae movemens affec U.S. capial flows; herefore, we compue he changes in he exchange raes relaive o he U.S. dollar over he pas monh by aking naural logarihm of he curren exchange rae over he pas rae, where exchange rae is saed as he number of foreign currencies per U.S. dollar. 6) Trading Resricion 12 : The dummy variable used here akes on he value of one when shor-selling is praciced in counry and zero oherwise (eiher shor sales are no allowed or no praciced). IV. C. Measuremens of Liquidiy To accoun for various liquidiy dimensions, hree alernaive measures of liquidiy are aken ino accoun: 1) Trading Frequency 13 : I is denoed by marke urnover, TURNOVER, which is he raio of he value of shares raded o he value of shares ousanding. 2) Trading Dollar Volume: We use wo ways o calculae rading volume: aking naural logarihm of he value of share raded 14, denoed by TRADVOL_1; he share rading value as a percenage of counry s GDP, denoed by TRADVOL_ ) Price Impac: I represens he absolue monhly price change per dollar of monhly rading volume, or say, he monhly price impac of he order flow, which serves as a rough measure of price impac and reflecs sock illiquidiy. Following Amihud (2002), we calculae he monhly price impac for each emerging counry marke by dividing he absolue value of monhly sock marke price changes by he dollar value raded for ha monh. I can be wrien as: Price Impac i, ln( pi, / pi, 1 ), where i denoes individual counry and denoes monh. Dollar Traded We use hree differen price indexes o calculae price impac: i, 11 Daa on exchange rae regime developed by Reinhar and Rogoff (2003) is no accessible o he public. 12 The number of ADR or ADR issuance can be anoher proxy of rading resricion, since firms in emerging counries may be more inclined o choose o issue ADRs or GDRs. According o Aggarwal e al. (2005), we can calculae he percenage of marke value of firms ha issue ADRs relaive o he oal marke value of all lised firms, or he percenage of lised firms ha issue ADRs. This is lef for furher research if he daa is accessible. 13 I can also be measured by he raio of he number of shares raded o shares ousanding. However, we do no obain monhly daa on hese wo variables. 14 Share rading value is he oal number of shares raded muliplied by heir respecive prices. 15 This measure is on an economy-wide basis since i gauges he posiive effecs of liquidiy as a share of naional oupu. 10

17 PRIIMPACT_1 i, is he absolue value of he Inernaional Finance Corporaion s Global (IFCG) price change over dollar value raded on monh, where IFCG index capures he reurn of mosly large firms, specifically i covers all publicly lised equiies wih floa-adjused marke values of US $100 million or more and annual dollar value raded of a leas US $50 million; PRIIMPACT_2 i, is he absolue value of he Inernaional Finance Corporaion s Invesable (IFCI) reurn over dollar value raded on monh, where IFCI represen he invesable index available o foreign invesors 16 ; The difference beween PRIIMPACT_1 and 2 is he number of resriced shares (resriced from foreign invesors). PRIIMPACT_3 i, is he absolue value of reurn of he value-weighed composie (including all socks) marke index over dollar value raded on monh. The difference beween his and measures 1 and 2 is he represenaion of smaller firms socks. The graphs on our measures of hree dimensions of liquidiy by regions are presened in Figures 3-5. The figures show ha Asia has he highes TURNOVER and Lain America has he lowes among he hree regions during our sample period. Furhermore, Lain America has larger PRIIMPACT han he oher wo across he sample period. These resuls poin ou ha Asian marke is more liquid han Lain America and Europe, Africa, and Middle Eas markes. As hypohesized, he coefficiens of TURNOVER, TRADVOL_1, and TRADVOL_2 are expeced o be posiive, and he coefficiens of PRIIMPACT_1 i,, PRIIMPACT_2 i,, and PRIIMPACT_3 i, expeced o be negaive. IV. D. Summary Saisics Table I Panel A repors descripive saisics for equiy flows, various conrol variables, and six measures of liquidiy employed by his sudy. The average gross equiy flow from he U.S. invesors o he sample emerging markes is $ million and he median flow is subsanially lower a $114 million, suggesing ha he disribuion of equiy flows is posiively skewed. The average equiy flow is percen of counry s GDP. The mean values of onemonh ahead and he average welve-monh-ahead equiy flows are percen and percen of GDP, respecively. The average marke size scaled by GDP is Panel B of Table 16 Edison and Warnock (2003b) sugges ha for a given counry he IFCG index is designed o represen he overall marke porfolio, while he IFCI index represen a porfolio available o foreign invesors. The laer excludes from he IFCG hose socks no available o foreigners due o eiher legal resricions or low liquidiy. 11

18 1 repors he average U.S equiy flows and marke capializaions by counry and regions. Alhough Asia has he larges marke size, he U.S. invesors prefer Lain American marke. The average local sock marke reurn, he U.S. value-weighed marke reurn on NYSE/AMES/NASDAQ, and he excess reurn are 1.4 percen, 0.9 percen, and 0.5 percen, respecively. The disribuion of he excess reurn is posiively skewed, and he middle 80 percen of he excess reurn lie beween percen and 10.3 percen, represening a spread of 20.5 percen during January 1995 and December Regarding he liquidiy measures, he mean, median, and 10h and 90h perceniles are of similar magniude for all measures of liquidiy excep TRADVOL_1. The correlaions 17 among he variables are shown in Panel C of Table 1. Mos correlaion coefficiens confirm our expecaions: he U.S. equiy flows are posiively correlaed wih marke size, counry sock marke excess reurn, deregulaions of shor-selling resricion, urnover, and rading volumes; and negaively correlaed wih capial conrol, changes in exchange rae, and price impac. Panel D presens he correlaions among he alernaive measures of liquidiy. As expeced, TURNOVER and TRADVOL_1 and TRADVOL_2 are highly correlaed bu far from perfec, and hese hree are negaively relaed wih PRICEIMPACT_1, PRICEIMPACT_2, and PRICEIMPACT_3. 17 The correlaion ables for each region are also made bu no repored. Wih few excepions, he correlaions are of similar magniude. 12

19 Table I. Summary Saisics for Various Variables, Jan Dec EQUITYFLOWS ($ millions) is he gross equiy flows from U.S. invesors o foreign markes. EQUITYFLOW, EQUITYFLOW (+1)/1, and EQUITYFLOW (+12)/12 are he gross equiy flows scaled by counry s GDP in a specific monh, one monh ahead, and he average welve-monh ahead, respecively. SIZE is he marke capializaion as a percenage of GDP. RETURN, VWUSRET, and EXCESSRET are counry local sock marke reurn, value-weigh U.S. marke reurn on NYSE/AMEX/NASDAQ, and he excess reurn of local sock marke, respecively. ECONFREE is he naural logarihm of he index of counry economic freedom. CAPLCONS is he inensiy of capial conrols. EXCHG is he change in exchange rae. SSDUMMY is shor-selling resricion dummy variable. TURNOVER is he raio of he value of share raded o marke capializaion. TRADVOL_1 is he naural logarihm of he value of share raded. TRADVOL_2 is he value of share raded as a percenage of GDP. PRIIMPACT_1 is he absolue value of he IFCG price change over dollar value raded. PRIIMPACT_2 is he absolue value of he IFCI price change over dollar value raded. PRIIMPACT_3 is he absolue value of local sock marke price change over dollar value raded. Table I. Panel A repors he descripive saisics for U.S. equiy flows, various conrol variables, and liquidiy measures. The sample period is from January 1995 o December Panel A. Descripive Saisics Obs Mean Median 10 h Percenile 90 h Percenile EQUITYFLOWS EQUITYFLOW EQUITYFLOW (+1)/ EQUITYFLOW (+12)/ SIZE RETURN VWUSRET EXCESSRET ECONFREE CAPLCONS EXCHG SSDUMMY LIQUIDITY TURNOVER TRADVOL_ TRADVOL_ PRIIMPACT_ PRIIMPACT_ PRIIMPACT_ Sources: Daa sources are repored in Table A1 of Appendix. 13

20 Table I. Panel B presens he average U.S. equiy flows in millions of dollars o each emerging marke and region, and he average marke capializaion in millions of dollars of each counry and region during January 1995 and December 2002 Panel B. U.S. Equiy Flows vs. Counry Marke Cap. Counry U.S. Equiy Ouflows Marke Capializaion ($ millions) ($ millions) Lain America ,177 Argenina ,389 Brazil 1, ,780 Chile 81 64,112 Mexico ,946 Peru 45 11,788 Europe, Africa, Middle Eas ,555 Israel ,024 Poland 23 18,199 Souh Africa ,854 Turkey ,144 Asia, Pacific ,963 Indonesia 84 45,651 Malaysia ,340 Philippine 63 40,584 Souh Korea ,121 Taiwan ,668 Thailand 62 64,367 Sources: Daa sources are repored in Table A1 of Appendix. 14

21 Table I. Panel C repors he panel daa correlaion coefficiens beween equiy flows and he various explanaory variables. Panel C. Correlaions EQUITYFLOW EQUITYFLOW (+12)/12 EQUITYFLOW (+1)/1 SIZE EXCESSRET ECONFREE CAPLCONS EXCHG SSDUMMY EQUITYFLOW (+12)/ EQUITYFLOW (+1)/ SIZE EXCESSRET ECONFREE CAPLCONS EXCHG SSDUMMY TURNOVER TRADVOL_ TRADVOL_ PRIIMPACT_ PRIIMPACT_ PRIIMPACT_ Sources: Daa sources are repored in Table A1 of Appendix. Table I. Panel D repors he correlaion coefficiens among he alernaive measures of liquidiy. TURNOVER 1 Panel D. Correlaion among Liquidiy Measures TURNOVER TRADVOL_1 TRADVOL_2 PRIIMPACT_1 PRIIMPACT_2 PRIIMPACT_3 TRADVOL_ TRADVOL_ PRIIMPACT_ PRIIMPACT_ PRIIMPACT_ Sources: Daa sources are repored in Table A1 of Appendix. 15

22 V. Mehodology and Empirical Resuls V. A. Time-Series Cross-Secion Daa Analysis V. A. 1. The Model As menioned earlier, his sudy focuses on he impac of various macro facors, especially liquidiy, on he U.S. equiy flows ino emerging markes. We conduc a ime seriescross secion daa 18 framework which allows us o analyze ime series properies of he daa and idenify individual counry effecs. The counry effec model ha we esimae is of he following ype: where o o y i M k 1 k X, LIQUID (1) k i y i denoes he U.S. equiy flows of counry i a ime ; i=1, 2, N is a counry index; =1,2,,T is a ime index. X k,i denoes he explanaory variables including conrol variables of counry i a ime ; k = 1, 2, M he number of explanaory variables. The conrol variables include various macro facors: marke size SIZE i,, prior foreign sock marke excess reurns EXCESSRET i, -1, counry economic freedom ECONFREE i,, capial conrol CAPLCONS i,, change of exchange rae EXCHG i,, and he dummy variable on shorselling resricion SSDUMMY i, o LIQUID i denoes he alernaive measures of liquidiy of counry i a ime ; Alernaive measures of liquidiy are rading frequency TURNOVER i,, dollar rading volume TRADVOL_1 i, and TRADVOL_2 i,, and price impac PRIIMPACT_1 i,, PRIIMPACT_2 i,, and PRIIMPACT_3 i, o o o β k and γ denoe he coefficiens of X k,i and LIQUID i, respecively. We assume ha β k and γ are he same from one counry o he nex. α is he inercep i is he combined ime series and cross-counry error componen, or he observaionspecific errors. i i i 18 Our panel daa is unbalanced (some counries do no have daa for one or more monhs, or no all counries have daa for all monhs) 16

23 o i is an unobserved variable and varies across counries bu no over ime. The counryspecific effecs can be reaed as fixed or as random. Wih he fixed-effec model, assumed o be correlaed wih he regressors and represens he counry-specific characerisics 19, and is he inercep specific for each counry. Wih he randomeffec model, i i is assumed o be uncorrelaed wih he regressors and represens he counry-specific error componen i.e., differences across counries are capured hrough i is he disurbance erm, and 2 is he error erm, where ~ IID (0, ) i i i and 2 i ~ IID (0, v ). Since each counry has is own individual characerisics ha may influence or bias some or all he explanaory variables and/or he prediced equiy inflows, he pooled OLS regression model may yield biased coefficiens because i does no accoun for hose unobservable imeinvarian 20 counry-specific characerisics. We use a Fixed-Effec (FE) esimaion 21 approach o verify wheher he inclusion of counry-specific characerisics in he model is valid. F-ess of he presence of counry-specific effecs rejec he null hypohesis ha all i shown in Table II. Therefore, he ime series-cross secion daa are no poolable wih respec o counry. To decide which esimaion, fixed or random effecs, is superior o he oher, we furher perform a Hausman es 22. The resuls are repored in Table II, indicaing ha he fixed-effec is an appropriae esimaor for all regression models in he full sample and Lain America, he models (1) and (7) in Europe, Africa, and Middle Eas sample, and he models (1) (4) and (7) in Asia samples. Thus, we use random-effec esimaion for he oher models. However, we use he OLS esimaion for he models (5) and (6) in Europe, Africa, and Middle Eas sample, because he daa on he liquidiy measures of PRIIMPACT_1 and PRIIMPACT_2 are only 19 Time-invarian counry characerisics (fixed effecs), such as culure, language, geography and demographics, may be correlaed wih he explanaory variables. Naional policies, federal regulaions, and inernaional agreemens, may change over ime bu no across counries. 20 We assume ha he counry-specific characerisics are ime-invarian. 21 I is equivalen o he leas square dummy variable (LSDV) esimaion, where he es associaed wih individual group is he es ha all of he dummy variables are zero 22 Hausman es: he null hypohesis is ha he preferred model is random effecs and he alernaive is he fixed effecs (Greene, 2000) or he unique errors (u i ) are correlaed wih he regressors. If he null is rejeced, hen he RE esimaors are biased. Random-Effec is usually preferred when he panel has large number of eniies. 17

24 available for one counry. We also es if cross-secional variances 23 equal o zero for hose models esimaed by random-effecs. The p-values rejec he null for models (5) and (6) in Asia sample, meaning ha here are counry-specific error componen. The p-values are no rejeced for models (2) o (4) in Europe, Africa, and Middle Eas sample, indicaing ha here is no individual effec for hese hree models and he pooled regression model is appropriae. Modified Wald es for groupwise heeroskedasiciy 24 in he residuals of he fixed-effec model is also performed repored in Table II. The null hypohesis is consan variance for all counries. The rejeced null indicaes ha he error variances are specific o he individual counry. Therefore, we need o conrol for he effecs of he unobserved heeroskedasiciy for boh fixed and random-effecs by adjusing he regression sandard errors for clusering on counry. In addiion, we perform Wooldridge es 25 for auocorrelaion in our ime series-cross secional daa. The null hypohesis of no serial correlaion for models is rejeced in he full sample and Lain America sample, as well as models (3), (4), and (7) in Europe, Africa, and Middle Eas sample and models (5) and (6) in Asia sample a he 5 percen significan level. The p-value is repored in Table II where we can see ha wih very few excepions, mos es saisics are no highly significan. Basically, here is no serial correlaion in models (1) and (2) for Europe, Africa, and Middle Eas sample. For Asia sample, here is no serial correlaion for all models excep models (5) and (6). The serial correlaion es indicaes ha for some of he sample counries he errors wihin counries are correlaed across ime. Therefore, we also compue Newey-Wes (1987) sandard errors o conrol for boh heeroskedasiciy and serial correlaion. The resuls indicae ha here is a very sligh difference beween he robus sandard errors menioned earlier and he Newey-Wes sandard errors, and he difference does no affec he saisical significance of each coefficien; consequenly we only repor robus sandard errors in Table II. 23 The Breusch-Pagan Lagrange mulipliers es helps us on he selecion of he appropriae model (pooled vs. individual effecs). If he null of cross-secional variance of 18 equals o zero: 2 0 i is no rejeced, he pooled regression model is appropriae. We also es if he variance componens for ime are zero. The null is rejeced a 5 o 10 percen significance levels. 24 The modified Wald es allows for unbalanced panels (as ours). 25 I is a es for serial correlaion in linear panel daa models. The null hypohesis is he residuals from he regression of he firs-differenced variables should have an auocorrelaion of -0.5.

25 V. A. 2. Resuls on Conrol Variables Table II four panels shows, separaely, he regression resuls for full sample, Lain America sample, Europe, Africa, and Middle Eas sample, and Asia sample. For each individual panel, we perform seven regression models, each wih a differen measure of liquidiy excep for model (1) which only includes conrol variables. The coefficien of SIZE is significanly posiive in all models excep models (3) and (4) in he Lain America sample. I is consisen wih he noion ha a larger marke likely aracs more invesors. As menioned earlier, SIZE may proxy for liquidiy o some exen, since larger markes also end o be more liquid. To he exen his argumen is rue, he resuls on SIZE is in agreemen wih he hypohesis ha a more liquid marke aracs more equiy flows. Based on he resul on EXCESSRET -1, here appears o be no evidence supporing reurn chasing behavior: among all four samples, only six models in Asia sample indicae a posiive bu insignifican coefficien. Ineresingly, he coefficien of EXCESSRET -1 is negaive and saisically significan in all models of he full sample and Lain America sample, as well as in models (3) and (4) of Europe, Africa, and Middle Eas sample. Also, he coefficiens are 26 remarkably close across samples. EXCESSRET -1 here is he pas excess counry sock marke reurn over he U.S. value-weighed sock marke reurn. The empirical evidence indicaes ha a decrease in he pas excess reurn of emerging sock marke, especially in Lain America marke, leads o an increase in immediae purchases of foreign equiies by he U.S. invesors. This suggess ha some invesors behave more like conrarians han momenum chasers. While his appears o be inconsisen wih previous sudies, invesors sraegies migh depend on he ime horizon since Griffin e al. (2002) indicae ha reurn-chasing behavior is shor-lived. I should be noed ha he sandard deviaion, skewness, and kurosis 27 of EXCESSRET for Lain America sample are much larger han hose for he oher wo regions. Also, we suspec ha some unobserved variables, such as geographic preference, may correlae wih one or some of explanaory variables, e.g., EXCESSRET. The impac of counry marke economic freedom, ECONFREE, on equiy flows is posiive and saisically significan in all models of he full sample and boh Lain America and 26 Analysis was also performed in sub-periods, bu no repored. Excep he coefficiens on excess reurn, ohers remain largely unchanged for sub-periods. 27 No repored. 19

26 Europe, Africa, and Middle Eas samples. This resul is consisen wih our expecaion ha equiy flows are larger for he counry wih more economic freedom. However, for he Asian sample, he coefficien of ECONFREE is negaive bu insignifican. We check he correlaions beween ECONFREE and oher explanaory variables for he Asian sample and do no see any correlaion more han 0.3. Thus, he couner-inuiive resul on ECONFREE in he Asian marke is unlikely o be caused by mulicollineariy. Overall, economic freedom of emerging counries, paricularly hose in Lain America and Europe, Africa, and Middle Eas, plays a significan role in aracing foreign equiy inflows. As for CAPLCONS, is coefficien is significanly negaive in all models for he full sample, Lain America, and Asia samples, which is consisen wih our expecaion. However, in models (1) o (4) and (7) of Europe, Africa, and Middle Eas sample, he coefficien of CAPLCONS has a posiive and saisically significan value. I should be noed ha he region of Europe, Africa, and Middle Eas includes relaively fewer counries and some daa is lacking for a couple of counries here. These may affec he esimaes for CAPLCONS in ha region. Wih he excepion of his region, our evidence in general indicaes ha a loosening of foreign ownership resricion expands he invesmen opporuniy and leads o an increased purchases of foreign equiy by U.S. invesors. If he value of U.S. dollar is increasing (decreasing) hen he U.S. invesors can afford o inves more (less). However, if he rend is fairly predicable, an increasing (decreasing) dollar would encourage he U.S. invesors o purchase less (more) foreign securiies. Consequenly, we make no specific predicion regarding he coefficien of he variable represening exchange rae movemens. Indeed, he resuls are mixed concerning exchange rae movemens: all models in he full sample and he Asian sample, and models (1) and (5) o (7) in Europe, Africa, and Middle Eas sample show a posiive impac. In Lain America, he impac is negaive bu no saisically significan. Moreover, he resuls indicae a posiive and saisically significan impac in models (2) o (4) for Europe, Africa, and Middle Eas sample. On he oher hand, he empirical resuls in he nex secion show ha here is saisically significan and negaive relaion beween he changes in exchange raes and he long-horizon equiy flows in he full sample and mos models in Lain America sample. We elaborae on his issue furher in nex secion. 20

27 As menioned earlier, foreign invesors should prefer an emerging marke where shor selling is allowed; hence a posiive relaion beween shor selling dummy and equiy flows is expeced. Our evidence indicaes ha he impac of shor-selling pracice varies across regions: he coefficien for SSDUMMY is posiive for all models in he full sample and Lain America, and is saisically significan in Lain America sample. For boh Europe, Africa, and Middle Eas sample and he Asian sample, he impac is negaive and saisically significan for some models. Since he daa on shor-selling is very limied, we should no surprise for he mixed resuls. V. A. 3. The Impac of Liquidiy on Equiy Flows We now urn o he focus of our paper, he relaion beween liquidiy and equiy flows in emerging markes. As a measure of rading frequency, TURNOVER in model (2) has a posiive sign in boh full sample and hree regional samples, and is coefficiens are saisically significan a 1 percen level, excep in Asia sample. As an illusraion, wih every 1 percen increase in TURNOVER, he equiies purchased by he U.S. invesors increased abou percen of a counry s GDP in Lain America and 3.49 percen in Europe, Africa, and Middle Eas during January 1995 and December Figure 3 also displays ha excep Souh Korea and Taiwan, oher sample counries mainain a seady TURNOVER raio across he sample ime. Inuiively, a higher TURNOVER raio signals lower ransacions coss of he local equiy marke. Tha is, U.S. invesors prefer a more sable, cheaper, and easier marke. Models (3) and (4) use dollar rading volume as he liquidiy measure. TRADVOL_1 is he naural logarihm of he value of share raded and TRADVOL_2 is he raio of he value of share raded over a counry s GDP. Our empirical resuls indicae ha he coefficiens on TRADVOL_1 are posiive and saisically significan for all models and in all four samples, which is consisen wih our hypohesis. The coefficiens on TRADVOL_2 are also posiive for all samples and are saisically significan for all excep he Asian sample. A counry s sock marke wih a higher dollar volume of rading indicaes an acive marke, mos likely associaed wih lower spreads, and ha a high number of invesors paricipae in ha marke. I migh also reflec invesors confidence on ha marke. If so, he evidence suggess ha U.S. invesors 21